Complete Two-Loop Dominant Corrections To The Mass of The Lightest in The Minimal Supersymmetric Standard Model

Nuclear Physics B 586 (2000) 338
www.elsevier.nl/locate/npe
Complete two-loop dominant corrections

to the mass of the lightest CP-even Higgs boson
in the minimal supersymmetric standard model
Jose Ramn Espinosa a , Ren-Jie Zhang b,1
a Instituto de Matemticas y Fsica Fundamental (CSIC), Serrano 113 bis, 28006 Madrid, Spain
b Department of Physics, University of Wisconsin, 1150 University Avenue, Madison WI 53706, USA
Received 29 March 2000; accepted 26 June 2000
Abstract
Using an effective potential approach, we compute two-loop radiative corrections to the MSSM
lightest CP-even Higgs boson mass Mh0 to O(t2 ) for arbitrary left-right top-squark mixing and
tan . We find that these corrections can increase Mh0 by as much as 5 GeV; assuming a SUSY scale
of 1 TeV, the upper bound on the Higgs boson mass is Mh0 129 5 GeV for the top quark pole
mass 175 5 GeV. We also derive an analytical approximation formula for Mh0 which approximates
our numerical results to a precision of . 0.5 GeV for most of the parameter space and suitable
to be further improved by including renormalization group resummation of leading and next-toleading order logarithmic terms. Our final compact formula admits a clear physical interpretation:
radiative corrections up to the two-loop level can be well approximated by a one-loop expression with
parameters evaluated at the appropriate scales, plus a smaller finite two-loop threshold correction
term. 2000 Elsevier Science B.V. All rights reserved.
PACS: 12.60.Jv; 14.80.Cp; 11.10.Gh
Keywords: Supersymmetric standard model; Higgs physics
1. Introduction
It is difficult to overestimate the importance of the experimental discovery of the Higgs
boson. It would not only help us to elucidate the dynamics responsible for electroweak
symmetry breaking but it will most probably offer also an important clue as to the nature
of the Physics beyond the Standard Model (SM).
The paradigm for this new Physics, granted that a fundamental scalar drives electroweak
symmetry breaking, is the Minimal Supersymmetric Standard Model (MSSM) (for reviews
1 After September 2000: School of Natural Science, Institute for Advanced Study, Princeton, NJ 08540, USA.
0550-3213/00/$ see front matter 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 2 1 - 1
J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338
see, e.g., [1,2]): the most economical extension of the Standard Model that incorporates
(softly-broken) Supersymmetry (SUSY). In spite of the uncertainties related to the
origin of supersymmetry breaking (and therefore of the masses of the so far undetected
supersymmetric particles), it is well known that the MSSM predicts the existence of a
light Higgs particle with mass below about 135 GeV (this bound depends sensitively on
the top quark mass one uses; our present calculation intends to set a precise and firm
bound). Unlike the case of the Standard Model (in which the mass of the Higgs boson
is an unknown parameter), the mass of the light CP-even Higgs boson of the MSSM is
calculable as a function of other masses of the model. A precise calculation of that mass is
of prime importance for Higgs searches at LEP, Tevatron and the LHC, and is the topic of
this paper.
We recall at this point that the Higgs sector of the MSSM consists of two SU(2) doublets,
H1 (which gives mass to down-type quarks and charged leptons) and H2 (which gives
mass to up-type quarks). The vacuum expectation values (v1,2 ) of these doublets break
the electroweak symmetry, after which, the Higgs spectrum contains two CP-even scalars
(h0 and H 0 ; with mh0 6 mH 0 ), one CP-odd pseudoscalar (A0 ) and a pair of charged
Higgses (H ). At tree-level, the masses, couplings and mixing angles of these particles
are determined by one unknown mass parameter (say, mA0 ) and the parameter , which
measures the ratio v2 /v1 ( tan ). In the limit mA0 MZ all the Higgs particles except
h0 have masses mA0 and rearrange in a complete SU(2) doublet almost decoupled from
electroweak symmetry breaking, while h0 remains light with m2h0 6 MZ2 cos2 2 and has
SM properties. This bound (which applies for any value of mA0 and is saturated for mA0
MZ ) is extremely important: it represents the limit that experimental bounds should reach
to falsify the MSSM. In fact, the present experimental bound [3] from LEP, including the
latest data with up to s = 202 GeV, is mh0 & 107.7 GeV (for large mA0 case in which the
SM limit is applicable; the limit falls to 91 GeV for smaller mA0 ), which is well above
this bound. This is not yet conclusive evidence against the MSSM because it does not take
into account the radiatively corrected form of the mass bound.
Radiative corrections to m2h0 have been computed using three different techniques (or
combinations of them): effective potential method [413], direct diagrammatic calculation
[1422] and effective theory (or renormalization group) approach [10,13,2330]. The full
one-loop radiative corrections to mh0 have been computed diagrammatically. The most
important of these corrections come from top quark/squark loops and are given by
1m2h0 =
m2t
3m4t
ln
,
2 2 v 2 m2t
(1)
where mt is the top quark mass, mt an average top-squark mass and v 2 v12 + v22 =
(246 GeV)2 . This correction can be very large if mt mt , and in such case mh0 can evade
easily the current experimental lower bound. This important O(t ) logarithmic correction
to the dimensionless ratio m2h0 /m2t [here t h2t /(4), where ht is the top-quark Yukawa
coupling] can be most easily reproduced using renormalization group (RG) techniques. In
addition, there is a finite (non-logarithmic) correction which may also be important, and
which depends on the details of the top-squark spectrum. This correction is (assuming
again for simplicity degenerate soft masses for the top-squarks)
!
3m4t
Xt4
Xt2
2
,
(2)
1mh0 =
2 2 v 2 m2
12m4t
t
where Xt = At + cot is the top-squark mixing parameter, At the soft trilinear coupling
associated to the top-Yukawa term in the superpotential and the supersymmetric Higgs
mass parameter. Correction (2) is maximized for Xt2 = 6m2t (the so-called maximalmixing case). When using one-loop equations like (1) and (2) to compute the Higgs mass
one has to decide whether to use on-shell (OS) or running values for the mass parameters
that enter such formulae (and if the latter, at which scale to evaluate them). The difference
between two such choices is of higher order, but can be non-negligible, especially because
of the m4t -dependence of 1m2h0 . Although RG techniques can be used to make an educated
guess of the scale at which those mass parameters should be evaluated (see, e.g., [28,29]),
a precise answer to such questions could only be unambiguously given by a two-loop
calculation like the one we perform in this paper.
At two loops, radiative corrections to m2h0 depend not only on the large top-Yukawa
coupling but also on the QCD coupling g3 . It is reasonable to expect that the dominant
two-loop corrections will be of order O(s t [ln(m2t /m2t )]k ) and O(t2 [ln(m2t /m2t )]k ), k =
0, 1, 2. Terms with k = 2 are the two-loop leading logarithmic contributions and can be
obtained by RG techniques using one-loop RG equations; no true two-loop calculation
is required and RG resummation will take into account such leading-logarithmic (LL)
corrections to all loops. The k = 1 terms are the two-loop next-to-leading-logarithmic
(NTLL) corrections, which can be obtained (and resummed to all loops) with two-loop RG
equations. Finally, the two-loop non-logarithmic terms (k = 0) can be interpreted in the
effective theory language as threshold corrections (at the supersymmetric scale set by the
mass of the top-squarks) and require a genuine two-loop calculation; they simply cannot
be obtained from RG arguments.
The status of these higher-loop calculations of the radiatively corrected m2h0 is the
following. Higher-order logarithmic corrections were included in studies which used
RG techniques almost since the dramatic impact of radiative corrections on mh0 was
first recognized. Hempfling and Hoang [11] were the first to perform a genuine twoloop calculation of mh0 which also included non-logarithmic terms. They computed the
dominant two-loop radiative corrections [to O(s t ) and O(t2 )] in the case tan 1 and
zero top-squark mixing. Their computation also included the most important logarithmic
corrections, which could be alternatively incorporated by RG resummation from one-loop
results, as done, e.g., in Ref. [10]. In this last paper it was also pointed out that by a
judicious choice of the renormalization scale at which to evaluate one-loop corrections, the
higher order logarithmic corrections could be automatically taken into account. A similar
idea was later implemented in [2830] to write down simple analytical approximations for
the radiatively corrected m2h0 , obtained by iterative integration of RG equations.
Besides being limited to a particularly simple value of tan , the calculation in Ref. [11]
missed the sizable impact of non-zero top-squark mixing in two-loop effects, that is, higher
order corrections to the contribution written down in Eq. (2). Such corrections were first
included to order O(s t ) in the diagrammatic calculation [21], and by the effective
potential method in Ref. [12]. The effect of these corrections is to shift the values of
Xt that give maximal mixing, change the corresponding Higgs mass by up to 10
GeV 2 and introduce an asymmetry in the dependence of mh0 with the sign of Xt . This
two-loop top-squark mixing dependent correction was also explicitly isolated recently
by the present authors in Ref. [13], which uses effective potential plus RG techniques.
Besides confirming the independent diagrammatic results of Ref. [21] we clarified the
relation of these calculations to previous ones (in particular matching results expressed
in different renormalization schemes; see also [31]). We also derived a compact formula
for the Higgs mass (in the spirit of [2830]) which took into account the most important
radiative corrections, and used RG techniques to include in a compact way two-loop LL
and NTLL corrections. With the O(s t ) radiative corrections organized in this way, we
find that the mixing-dependent genuine two-loop threshold corrections are generally small
<
3 GeV).
(
Nevertheless, the large computing effort just reviewed did not exhaust the potentially
important radiative corrections: the two-loop O(t2 ) top-squark-mixing-dependent corrections to mh0 remained unknown to this day, while it is clear that they could compete in
principle with the O(s t ) contributions. The purpose of this paper is to complete the calculation performed in [1113] by using effective potential techniques (plus RG techniques)
to compute such O(t2 ) contributions for general top-squark mixing parameters and any
value of tan . The results in this paper can be considered the most complete and accurate
approximation to mh0 presented in the literature.
The structure of the paper is the following: Section 2 describes the strategy of our
calculation and presents some analytical formulae for mh0 , obtained in the limit of
mt mt . Section 3 goes one step ahead implementing the RG-improvement of such
approximations and, in doing so, clarifies the organization of the higher order radiative
corrections. This procedure is not only important to provide a clearer physical picture
in connection with the effective field theory but also to classify those corrections
calculated in Section 2 into a numerically dominant and compact part plus smaller finite
threshold correction terms. In Section 4 we present our numerical results for the Higgs
mass, illustrate the size of the new corrections and check the validity of our analytical
approximation formulae. We draw some conclusions in Section 5.
Several appendices are devoted to technical details of different aspects of the calculation.
Appendix D is worth special mention as it contains the two-loop O(t2 ) MSSM effective
potential used as starting point of our calculation and first computed in this paper.
2 This correction is relative to the one-loop mass using on-shell parameters. The size of the correction would
be much smaller if running parameters are used in the one-loop formula (1).
2. CP-even Higgs boson masses to two-loop order

The momentum-dependent mass-squared matrix for the CP-even Higgs bosons of the
MSSM in the interaction eigenstate basis h1 , h2 is
"

#
m2Z c2 + m2A0 s2 + 1M211 p2
m2Z + m2A0 s c + 1M212 p2

2 2
, (3)
Mh p =

m2Z s2 + m2A0 c2 + 1M222 p2
m2Z + m2A0 s c + 1M221 p2
where s sin and c cos . The mass parameters mZ and mA0 are the (scaledependent) running masses of the Z-boson and CP-odd Higgs boson A0 ; they are related to
the on-shell masses MZ and MA0 (we use capital letters to distinguish on-shell parameters
from running ones) by

T
MZ2 ,
m2Z = MZ2 + ReZZ

T1
T2
(4)
m2A0 = MA2 0 + ReAA MA2 0 s2 c2 ,
v1
v2
T is the transverse part of the Z-boson self-energy and
0
where ZZ
AA the A -boson selfenergy, T1 , T2 are the tadpoles of the CP-even (real) fields h1 , h2 . Their explicit one-loop
expressions can be found, e.g., in Ref. [22].
In (3), 1M2 stands for the contributions from radiative corrections. They are

Ti
(5)
1M2ij p2 = ij p2 + ij , i, j = 1, 2,
vi
where ij is the self-energy matrix of the Higgs fields h1 and h2 . The masses, mh0 , mH 0 ,
of the two CP-even Higgs bosons are then obtained from the real part of the poles of the
propagator matrix,

(6)
Det m2h0 ,H 0 1 M2h m2h0 ,H 0 = 0.
The radiatively corrected mixing angle is obtained as the angle of that rotation which
diagonalizes M2h (for some choice of p2 , say p2 = m2h0 ):

2 M2h 12

.
(7)
tan 2 =
M2h 11 M2h 22
Computing Higgs boson masses to a certain order of perturbation theory then requires
calculating the self-energies and tadpoles in Eqs. (4) and (5) to that order.
In the effective potential approach [413], self-energies and tadpoles can be calculated
as derivatives of the Higgs potential V according to:

2

V (h1 , h2 )
V (h1 , h2 )
,
ij (0) =
.
(8)
Ti =

h1 =v1
h1 =v1
hi
hi hj
h2 =v2
h2 =v2
Note that the s obtained from derivatives of V have zero external momentum.
In the limit MA0 MZ , the lightest CP-even Higgs state lies along the direction of
the breaking in field space [32], that is, /2 + O(m2Z /m2A ), and its radiatively
corrected mass has a very simple expression
Mh20

4m4t
d 2
= 2
V Rehh m2h0 + Rehh (0),
2
v
dmt
(9)
which is exact up to corrections of order O(m4Z /m2A0 ). 3

In Eq. (9), V is the projection of V (h1 , h2 ) along the light Higgs h = h1 c + h2 s :
hc , h2 hs ). Then V (h) can be expressed as a function of mt
V (h) = V (h1
using h mt 2/(ht s ). We decompose V in its nth-loop pieces Vn (explicitly given
in Appendix D) as V = V0 + V1 + V2 . The tree-level part V0 is the only one in which we
keep non-zero electroweak gauge couplings. We approximate the one-loop part V1 by its
O(t ) piece coming from top quark/squark loops. The two-loop part V2 is approximated
by V2s + V2t , where V2s is the O(s t ) part and V2t the O(t2 ) one.
Next, hh (p2 ) is the light Higgs self-energy at external momentum p, related to the
self-energies of h1,2 by
hh (p2 ) 11 (p2 )s2 + 22 (p2 )c2 212 (p2 )s c .
(10)
Notice that the self-energy difference in (9) involves non-zero external momentum and
would require a diagrammatic two-loop calculation. However, throughout this paper we
work in the approximation of neglecting in the radiative corrections all couplings other
than ht or g3 . In that case, realizing that at tree level mh0 depends only on electroweak
gauge couplings while its dependence on ht appears only at one-loop, we can write

d
,
(11)
hh m2h0 hh (0) ' m2h0 2 hh (p2 )
dp
p 2 =0
which gives rise to O(t2 ) contributions from the one-loop O(t ) self-energy hh .
In Section 4 we present the numerical results of such procedure for the two-loop
potential with O(s t ) and O(t2 ) corrections included. The general expression for the
O(s t ) potential was first computed in [12] while the complete O(t2 ) terms were still
missing. Both contributions to V are given in Appendix D.
It is useful, both for a better understanding of the numerical results and for practical
applications, to derive an analytical expression for the light Higgs mass in the case of a
large hierarchy between the supersymmetric scale and the electroweak scale (say when
the SUSY scale is of order 1 TeV). Such limit is interesting because it maximizes the
radiative corrections to m2h0 (so that it corresponds to the most pessimistic scenario for
Higgs searches; the case one should be able to discard to rule out the MSSM), and at the
same time simplifies the structure of the radiative corrections, avoiding the proliferation of
a multitude of different supersymmetric thresholds.
We consequently assume now that all supersymmetric particles have roughly the same
mass MS MZ . In more detail, focusing on the particles relevant for the radiative
3 This formula can be proved as follows: If m2 m2 , /2, and we can therefore use the
Z
A0
approximation 1m2 0 ' 1M211 c2 + 1M222 s2 + 21M212 s c , up to higher order terms in m4Z /m2 0 . Observing
h
A
that the potential V depends on the fields h1 and h2 only through (field-dependent) top quark mass and the
off-diagonal elements of the top-squark mass-squared matrix and using (8), we can easily express the partial
derivatives of V in terms of the total derivative in (9). A similar formula was already used in [11].
corrections to mh0 , we take equal soft masses MQ

e = MU
e = MS for the top-squarks (with
diagonal masses m2t ' MS2 + m2t ). The two eigenvalues and mixing angle of the top-squark
squared-mass matrix are then
m2t = m2t + mt Xt ,
1
m2t = m2t mt Xt ,
2
1
st2 = ct2 = .
2
(12)
We also take the same mass MS for the gluino and the pseudoscalar Higgs [this means
in particular that we can use Eq. (9) for the light Higgs boson]. In principle we admit the
possibility that the parameter could be smaller than MS , in which case we expect that one
chargino and two neutralinos will have masses || below the common supersymmetric
threshold. In this situation, which broadly corresponds to the case of a common heavy
SUSY scale, we find that, using the operator

4m4
d 2
2
2t
,
(13)
Dm
v
dm2t
the different parts entering (9) are
2
V0 = m2Z cos2 2,
Dm

MS2
3m4t
X t4
2
2
,
Dm V 1 =
ln 2 + Xt
2 2 v 2
12
mt
(
M2
M2
s m4
m2
m2
2
V2s = 3 2t ln2 2S 2 ln2 S2 + 2 ln2 t2 + ln t2 1
Dm
v
Q
Q
Q
mt

M2
m2
+ 1 + 2 ln S2 + 2 ln t2 X t
Q
Q
)

t3 X t4
MS2
X
2
+ 1 2 ln 2
,
X t +
Q
3
12
(
2
4
2
MS2
3
m2t
m
t
t
2
2 MS
2 mt
V2t =
6
ln
ln
3
ln
9
ln
Dm
16 3 v 2
Q2
Q2 Q2
Q2

M2
+ 2 3f2 ()
3f1 ()
8 ln 2S
mt

2
M
2
+ 6
2 1 ln S2 2(4 + 2 )f1 ()
+ 4f3 ()

Q
3

2
MS
2
2
2
+ (4 6 )f1 ()
X t2
+ (33 + 6
) ln 2 10 6 4f2 ()
Q

4
M2
Xt
+ 4(7 + 2 ) ln S2 + 23 + 4 2 + 2f2 ()
2(1 2 2 )f1 ()
4
Q

2
M
1
+ s2 X t6 ln S2 1
2
Q
(14)
(15)
(16)
10
"
MS2
m2t
4 2
4
ln
3
+
60K
+
Q2
Q2
3
m2t

2

MS2
MS
2
+ 12 24K 18 ln 2 Xt 3 + 16K 3 ln 2 4X t Yt + Yt2

Q
Q

2
2
M
11 MS 4
ln 2 Xt + 4 + 16K 2 ln S2 X t3 Yt
+ 6 +
2
Q
Q
#)

2
MS
MS2
1
14
19
2
2
4
2
+ 24K 3 ln 2 X t Yt
+ 8K ln 2 X t Yt
+
.
3
Q
12
2 Q
+ c2
3 ln2
MS2
+ 7 ln
(17)
The notations used are X t = A t + cot , Yt = A t tan , with reduced parameters z

z/MS , and (see Appendix A) K ' 0.1953256. We also use the following non-singular
functions of
2
ln 2 ,
1 2

1
2
2
=
ln ,
1+
f2 ()
1 2
1 2

(1 + 2 2 + 2
2
4)
2
2
2
2
2
ln ,
=
ln ln(1 ) + Li2 ( )
f3 ()
6
(1 2 )2
=
f1 ()
(18)
with f1 (0) = 0, f2 (0) = 1, f3 (0) = 2 /6 and f1 (1) = 1, f2 (1) = 1/2, f3 (1) = 9/4.
Finally, the correction for non-zero external momentum in Eq. (9) is given by (see
Appendix C)

h2t
X t2
m2t
2
2 2
.
(19)
m 0 s 3 ln 2 + 2
Re hh (mh0 ) + hh (0) =
2
16 2 h
Q
The parameters that appear in these expressions are running parameters, evaluated in the
satisfy MSSM RG equations. In fact, it can be checked that the (physical)
Higgs mass, given by Eq. (9), is renormalization-scale independent (up to two-loop order),
as it should. This scale independence is at the root of the RG-resummation procedure
discussed in the next section. It is evident that for different values of the renormalization
scale, the magnitude of the two-loop corrections will change, so that it should be possible to
choose the scale in such a way that the bulk of the corrections is transferred to the one-loop
terms (which depend on the scale implicitly).
Therefore, the magnitude and relevance of the two-loop corrections depends on the
definition of the mass parameters that enter the one-loop corrections. It is in this respect
convenient to write down the two-loop expressions just obtained in the particular case in
which all mass parameters in the one-loop correction are the OS ones. This is also useful to
compare with explicit diagrammatic calculations. The relationships between running and
OS parameters are listed in Appendix C. Using them, we obtain for the two-loop correction
to m2h0 :
DR -scheme and
1m2h0
(
)
2
MS2
MS2 2 3 4
s m4t
2 MS
= 3 2 3 ln
6 ln 2 + 6Xt 3 ln 2 Xt Xt
v
4
m2t
mt
mt
(

2
MS2 2 7 2
3t m4t
2 MS
3 2
+
13
ln
3
ln
s
+
16 3 v 2
2
3
m2t
m2t

2

11 2 + 3 4 f1 ()
3 1 2 ln 1 2 + 3f2 ()
+ 4f3 ()

2
13 4
+
+ c2 60K +
2
3

2
M
73
69
+ 9
2 + f1 ()
+ 24K X t2
7f2 ()
c2
+ 3s2 ln 2S +
2
2
mt

s2
1
61
+ 26 9 2 + 3f1 ()
+ 3f2 ()
+ c2 X t4 + X t6
6
2
2

X 4
+ 3(1 2 ) (2 3 2 )f1 ()
(1 + 3 2 ) ln(1 2 ) X t2 t
6

2
+ c2 (3 16K 3)(4X t Yt + Yt2 ) + 16K + X t3 Yt
3

)
4
7
2
2
4
2
+ 8K + X t Yt
+ + 24K + 3 X t Yt
.
3
12
2 3
11
(20)
We emphasize that this expression gives the two-loop corrections when the one-loop
contribution (15) is expressed in terms of OS parameters, that is,
"
OS 2

#
Mt2

3g 2 Mt2
Xt
1 XtOS 4
2 OS
ln 2 +
.
(21)
1mh0 1-loop =
2
Mt
12 Mt
8 2 MW
Mt
Several features of Eq. (20) are worth commenting. First, if we restrict Eq. (20) to
tan 1 and zero At , to compare with the result of Ref. [11], we find the same logarithmic
terms. However, the O(t2 ) finite term is different. In particular, that term is sensitive to the
value of the parameter , contrary to what is stated in Ref. [11]. Nevertheless, the result
quoted for that finite term in Ref. [11] is inside the range we would find by varying 2
from 0 to 1, and the impact of this -dependence on the final Higgs mass is quite small.
Second, we see that radiative corrections no longer depend on At and in the
combination Xt that appears through the off-diagonal entry of the top-squark mass matrix:
besides the explicit dependence on the parameter already noticed, the quantity Yt also
introduces a different combination of At and . This dependence on Yt originates from the
Htb and Htt diagrams of Fig. 8.
Third, although roughly speaking the top-Yukawa correction has a small pre-factor 3/16
in comparison with the QCD correction, this does not guarantee that the new contributions
will be negligible compared to the QCD one. In fact, we will see that for two-loop topsquark-mixing-dependent corrections of (20), the top Yukawa contributions have opposite
12
signs as that of the QCD corrections and could be as much as 60% of the latter (see Fig. 6).
In the next section, we will follow RG methods and reorganize these corrections in the
effective theory language, with the most important corrections of Eq. (20) reshuffled in a
RG-motivated one-loop formula.
3. Renormalization group resummation

Before illustrating in Section 4 the impact of the newly computed corrections on the
Higgs mass, we show in the following how the use of renormalization group techniques
[2330] allows us to write the previous complicated corrections [see Eq. (20)] in a
simpler and more transparent way, while at the same time it clarifies the connection to
the RG programme, which can be used to improve the precision of the mass formula by
resummation of higher order corrections.
We already applied this idea in Ref. [13] to the O(s t ) two-loop corrections. By a
convenient (and physically well motivated) choice of the scale at which to evaluate running
parameters in the one-loop mass correction one can absorb large logarithms in Eq. (20).
The RG evolution of the parameters is given by the corresponding one-loop RG functions
listed in Appendix B.
We use the following equations to relate supersymmetric running parameters at different
scales [cf. Eqs. (B.17) and (B.18)]:
m2t (Q)
)

02

1
3
16
Q
g 2 h2 X 2 s 2 + Yt2 c2 + c2 + 2 2 2 ln 2 ,
= m2t (Q0 ) 1 +
16 2 3 3 2 t t
Q

1
Q02
16 2
2
2
2
g
M
+
3h
(X
+
X
s
+
Y
c
)
ln
,
Xt (Q) = Xt (Q )
S
t
t
t
t
16 2 3 3
Q2
0
(22)
(23)
where we have used At = Xt s2 + Yt c2 , A2t + 2 = Xt2 s2 + Yt2 c2 and m2H2 + 2 = m2A0 c2 .

Notice that, to the order we work, it is sufficient to use these one-loop LL approximations to
the full RG evolution because we are concerned with parameters that appear in a one-loop
order term.
The Standard Model MS top quark mass mt and the Higgs VEV v are related to the
on-shell mass Mt and MSSM VEV v by [cf. Eqs. (C.2) and (C.10), from which relevant
terms can be easily identified]
"

#
h2t s2
g32
m2t
m2t
2
2
m t (Q) = Mt 1
4 3 ln 2 +
8 3 ln 2 ,
(24)
6 2
Q
32 2
Q
"
#
h2t s2 2
2
2
X .
v (Q) = v (Q) 1 +
(25)
32 2 t
We also use one-loop LL solutions of the SM RG equations to relate these parameters at
different scales
13
"
#

1
3 2 2
Q02
2
1+
8g3 ht s ln 2 ,
16 2
2
Q
"
#
3h2t s2 Q02
v 2 (Q) = v 2 (Q0 ) 1 +
ln 2 .
16 2
Q
m 2t (Q) = m 2t (Q0 )
(26)
(27)
Using the above equations, we find the following compact expression for the Higgs
boson mass, which is one of the main results of this paper
Mh20 = MZ2 cos2 2 +
3m 4t (Qt )
2 2 v 2 (Q1 )
ln
m2t (Qt)
m 2t (Q0t )
(2) 2
2
+ 1(1)
th mh0 + 1th mh0 .
The one-loop threshold correction is

3m 4t (Qth ) 2
X t4 (Qth )
(1) 2
,
Xt (Qth )
1th mh0 =
12
2 2 v 2 (Q2 )
(28)
(29)
and the two-loop threshold correction reads

s m4
7
1
1
(2)
1th m2h0 = 3 2t 2X t X t2 + X t3 + X t4 X t5
v
3
12
6
(
4
1
3t mt
+ R2 ()
X t2 + R4 ()
X t4 s2 X t6
R0 ()
+
3
2
2
16 v

2

9 4
t2
(3 + 16K) 4X t Yt + Yt2 + (15 24K)X
+ c2 60K +
2
3

25 4
14
3
+ 24K X t2 Yt2
Xt + (4 + 16K)Xt Yt +
4
3

)
19
+ 8K X t4 Yt2 .
(30)
12
We have used the short-hand notation

9 2
+ 6 2 11 + 2
=
2 f1 ()
+ 3f2 ()
+ 4f3 (),
R0 ()
2
3

= 11 2 6 + 6f1 ()
+ 10f2 ()
,
R2 ()

2
= 6 + 1 + f1 ()
+ f2 ()
.
R4 ()
The scales required in (28,29) are
Q0t = (mt m2t )1/3 ,

Qt = mt mt,
Q1 = e1/3 mt ' 0.7 mt ,
(31)
Qt = Qth = mt,
Q2 = e1/3mt ' 1.4 mt .
(32)
It is a non-trivial check of our calculation that the values of the scales (32) required to
re-absorb the large ln(MS2 /m2t ) logarithms in the two-loop corrections are consistent with
the ones obtained in [13] for the QCD corrections alone. We see, in particular, that the
uncertainty found there in the definition of the scales Q0t and Qt is here resolved by the
need of absorbing the new radiative corrections.
14
2
We still find a somewhat complicated expression for the threshold correction 1(2)
th mh0 ,
due to the fact that we have kept free the parameter. Expressions for the two limiting
cases of heavy ( ' MS ) and light ( Ms ) can be readily derived. In both cases,
the resulting threshold correction is much simpler than the general case (30) and contains
no more logarithms. Explicitly, for Ms we find
2 3
,
3
2
and for ' MS :
R0 (0) =
R2 (0) = 11,
R4 (0) = 6,
(33)
2
13
,
R2 (1) = 16,
(34)
R4 (1) = .
3
2
It is perhaps convenient to make more explicit the connection between our results and
those obtained in the RG approach (see, e.g., Ref. [30]). To be concrete, let us assume || =
MS so that all supersymmetric particles (including charginos and neutralinos) have masses
of order MS ; below that scale, the effective theory is the SM. The light Higgs quartic
coupling at MS consists of a tree-level part plus higher-order threshold corrections
which arise from the heavy decoupling supersymmetric particles, it can be evolved down
to the electroweak scale, say Q = mt , using the SM RGEs; at that scale is related to the
physical Higgs mass. This procedure should reproduce all the logarithmic corrections we
have found.
More explicitly, defining d/d ln Q2 , we can write
R0 (1) = 7
ZQt
(Qt ) = (Qt)
d ln Q2 .
(35)
Q=Qt
We use a special notation for the high and low scales between which we run to distinguish
them from other definitions of mt and mt that appear in the paper. These quantities are
defined by:
Qt mt (Qt ),
Qt mt(Qt),
(36)
i.e., they are the running masses evaluated at a scale equal to the corresponding mass. This
is the natural definition in the RG approach.
Making a loop expansion of in (35) and a further expansion around a particular value
of Q (say the low energy limit of the running interval, Qt ), we obtain to the two-loop order

Q 1 d(1)
Q
(Qt ) ln2 t + , (37)
(Qt ) ' (Qt) (1) (Qt ) + (2)(Qt ) ln t
2
Qt
2 d ln Q
Qt
where the one- and two-loop contributions to are approximated by [neglected all
couplings other than the strong gauge coupling g3 and the SM top Yukawa coupling gt
( ht s )]

3gt2
gt2 + ,
2
8

2gt4
(2)
15gt2 16g32 .
=
2
2
(16 )
(1)
(38)
15
We note that for a correct two-loop computation it is necessary to retain also the gt2 term
(1)
(1)
in because gets one-loop contributions proportional to gt4 . d /d ln Q2 can be
calculated from the one-loop RG evolution of gt

gt2
dgt2
=
9gt2 16g32 .
(39)
d ln Q2 32 2
Once (Qt ) is obtained from (37), we extract the physical Higgs mass using the SM
relation [33]:

g2
(40)
(Qt )v 2 (Qt ) = Mh20 1 t 2 .
8
This correction arises from wave-function renormalization and takes into account the fact
that the physical mass is defined on-shell, and not at zero external momentum. Its physical
content is therefore similar to the correction (11) in our effective potential approach.
According to (37), the large LL and NTLL corrections to Mh20 arise solely from (Qt ).
Additional radiative contributions in (40), coming from v 2 (Qt ) and the wave-function
correction factor, affect the large logarithmic terms only through multiplication of (Qt ).
It is therefore clear that it is enough for our purposes to know these correction factors at
one-loop order. Based on this observation, we can combine both factors together using (25)
to write the simpler formula
Mh20 = (Qt )v 2 (Q1 ),
Q1
(41)
= e1/3 Q
with
t , in accordance with (32).
It is now straightforward to show perfect agreement of Mh0 as obtained from the above
expression with our results (28)(30). All logarithmic corrections up to two-loops are
exactly reproduced while the finite part agrees if one uses as boundary condition at the
SUSY scale

1
(1)
(2)
+ th
,
(42)
(Qt) = g12 + g22 cos2 2 + th
4
with

3g 4 (Qth ) 2
X 4 (Qth )
(1)
,
= t 2
Xt (Qth ) t
th
12
8
(2)
=
th
2
1(2)
th mh0
.
(43)
v2
To summarize, we find full agreement between our approximate formula (28) for the
Higgs boson mass and the RG-improved mass calculated in the RG (or effective theory)
approach, to two-loop order. The connection to the effective theory language clarifies the
origin of the different terms in (28), and rewrites them in a very convenient way, absorbing
the large (logarithmic) two-loop effects in the one-loop correction and leaving behind
two-loop threshold corrections which are numerically small, as we will see in the next
Section. Note that this applies in particular to the sizable top-squark-mixing-dependent
corrections of Eq. (20), the bulk of which is transferred to the RG-reshuffled one-loop
threshold correction of Eq. (29).
16
Knowing the boundary condition, (Qt), one can integrate (35) numerically by solving
a coupled set of differential equations (describing the two-loop evolution of , g3 , gt ), find
(Qt ) and use (41) to get the Higgs mass. The final result will be the full RG-improved
value of Mh0 and will resum LL and NTLL corrections to all loops [numerical integration
includes all the terms from the expansion around Qt which were neglected in (37)]. In this
respect, note that our compact formula, Eq. (28), which has been found by requiring that
logarithmic contributions are correctly reproduced up to two-loops only, contains in fact
logarithmic corrections of higher order. It can be shown that these higher order logarithmic
corrections do not match exactly the correct ones (obtained by the RG method) if we
use simple one-loop approximations [like those given in Eqs. (24), (25)] to evaluate the
parameters in (28) at their corresponding scales. However, evaluation of those parameters
by means of a full numerical integration [similar to that in Eq. (35) for ] would correctly
take into account the LL (but not the NTLL) terms to all loops. Nevertheless, as we will
see in the next Section, the error made in neglecting logarithmic corrections of higher order
is very small for SUSY scales of interest [below MS O(1) TeV]. If MS turns out to be
significantly larger than that (starting to be in conflict with naturalness criteria), then one
should revert to the numerical RG integration of to get a reliable estimate of the Higgs
mass. Our results for the boundary condition (Qt) will still be useful in such a case.
4. Numerical results
In this section we present numerical results from our two-loop study. For the one-loop
analysis we closely follow Ref. [22], which has included complete radiative corrections
from the dominant top quark/squark sector and the sub-dominant gauge/Higgs boson and
neutralino/chargino sectors. In what follows, we shall concentrate on two-loop radiative
corrections.
We start by sketching the procedure for this analysis, which is the following: we first take
OS
OS
OS
as inputs the on-shell mass parameters 4 MA0 , Mt , MQ
e , MU
e and At . From them we
can determine the values of the corresponding running parameters at any renormalization
scale Q. To do this, we have to calculate the one-loop self-energy diagrams for Higgses and
top-squarks (the latter are collected in Appendix C). We also input tan and parameters,
and convert s (MZ ) = 0.118 to the MSSM DR running value. Next, we calculate in the
MSSM the two-loop corrections to the CP-even Higgs mass matrix, 1M2ij , from the twoloop potential (D.5) and (D.6) using Eq. (8). Numerically, the partial derivatives in these
equations are replaced by finite differences in h1 , h2 , i.e., we vary the values of these fields
by a finite amount and recalculate the field-dependent top-quark mass m2t = 12 h2t h22 and topsquark masses mt1 , mt2 , mixing angle t from Eq. (B.5). With these new parameters, the
two-loop potential is reevaluated and their variations from the reference values [calculated
at h21 + h22 = (246 GeV)2 ] are found. Finally, equipped with the corrections 1M2ij , we
compute the lightest CP-even Higgs boson mass by solving Eq. (6).
4 For the top-squark sector, we can alternatively take as inputs the on-shell top-squark masses and mixing angle.
17
Fig. 1. Higgs boson mass Mh0 vs. the on-shell top-squark mixing parameter X tOS . Dotted, dot-dashed
lines show one-loop and two-loop O(s t ) results from the program FeynHiggs, corresponding
results from our numerical analyses are shown in dashed and solid lines, respectively.
Several approximations have been made to quantities in Eqs. (3)(5), in particular we

neglect all dimensionless couplings other than the top-Yukawa coupling ht and the QCD
gauge coupling g3 . In this way we pick up the dominant radiative effects only, what we
term throughout leading corrections. We notice that the two-loop self-energy of the Zboson and the non-zero external momentum corrections to two-loop Higgs boson selfenergies can be neglected in our calculation since all these corrections are higher order
effects in the leading approximation. However, we need to calculate AA to the two-loop
level since it has O(s t ) and O(t2 ) corrections and in principle could contribute to (4)
at the same order as 1M2ij . It is not possible to obtain these self-energies in our current
approach, and explicit two-loop calculation of the corresponding two-point functions are
needed. Fortunately, the correction to mh0 from AA is always numerically negligible for
large mA0 as can be easily seen from the structure of the Higgs mass matrix (3). That is, (9)
is correct for large mA0 and we can safely neglect the AA corrections. (For mA0 mZ , a
complete two-loop calculation of mh0 would need AA .)
Fig. 1 is used as calibration: we compare in it our numerical results for Mh0 including
only up to two-loop O(s t ) corrections with the mass obtained by the program FeynHiggs
[34] which uses the explicit two-loop diagrammatic results of Ref. [21]. We choose two sets
of parameters:
OS = M OS = M = 500 GeV, = 200 GeV,
(a) MA0 = M3 = MQ
S
e
e
U
OS = M OS = M = 1 TeV, = 500 GeV.
(b) MA0 = M3 = MQ
S
e
e
U
For each case, results for two values of tan (1.6 and 20) are plotted. We find good
agreement (given the fact that they are two independent programs) between both one-loop
(shown in dotted and dashed lines) and two-loop QCD corrected (shown in dot-dashed and
solid lines) masses; this shows numerically that the two approaches are equivalent to that
order. This equivalence is easily understood since the effective potential, as a generating
18
Fig. 2. Higgs boson mass Mh0 vs. the on-shell SUSY scale MS , for two top-squark mixing
parameters X tOS = 0 and 2. One-loop mass, two-loop masses to O(s t ) and O(s t + t2 ) are
shown in dashed, dot-dashed and solid lines, respectively.
functional [35,36], encompasses all tadpole and self-energy diagrams (as well as all other
multi-point functions) which are calculated in [21]. The effective potential approach is
more efficient for the purpose of calculating Mh0 and much simpler to implement in a
Fortran program since it requires evaluating only one set of two-loop functions.
In Fig. 2 we show the Higgs boson mass Mh0 vs. the (on-shell) SUSY scale MS , for two
values of the top-squark mixing parameters X tOS (0 and 2). All the physical masses MA0 ,
OS
OS
MQ
e , MU
e and M3 have been set to MS (and the same will be done for all the following
plots). The dashed, dot-dashed and solid lines in this figure correspond to masses Mh0
corrected to one-loop, two-loop O(s t ) and two-loop O(s t + t2 ) order. 5 Fig. 2(a)
(X tOS = 0) corresponds to the case of minimal left-right top-squark mixing, and the two OS
loop O(t2 ) corrections are generally small, <
2 GeV. For Fig. 2(b) (Xt = 2), which
roughly corresponds to the maximal left-right top-squark mixing case, we find that the
two-loop O(t2 ) corrections are sizable (' 5 GeV).
In Fig. 3 we examine the upper limit on the Higgs boson mass Mh0 by including the
dominant two-loop corrections. We show corrected massed to the two-loop O(s t ) and
O(s t + t2 ) orders in dot-dashed and solid lines for X tOS = 0, 2 and the top quark pole
mass Mt = 175 GeV. We see that maximal values for Mh0 of ' 129 GeV can be reached for
large tan and left-right top-squark mixing parameter X tOS ' 2. Without two-loop O(t2 )
corrections, the upper bound of Mh0 would be at ' 124 GeV. We also show the Higgs boson
masses for Mt = 180 and 170 GeV (including all two-loop dominant corrections) in dashed
and dot-dashed lines; the masses are increased or decreased by 5 GeV, respectively.
We remark that this upper bound on Mh0 is asymmetric with respect to X tOS . For X tOS =
2 and Mt = 175 GeV, we find the bound is about 5 GeV lower. As is well know, this
asymmetry arises from the two-loop O(s t ) corrections [12,21].
5 In general, we try to follow the rule that denser lines correspond to more precise approximations.
19
Fig. 3. Higgs boson mass Mh0 vs. tan for the top-squark mixing parameters X tOS = 0
and 2. Dot-dashed and solid lines correspond to two-loop Higgs boson masses to O(s t ) and
O(s t + t2 ) respectively for Mt = 175 GeV. Two-loop masses to O(s t + t2 ) for Mt = 170
and 180 GeV are also shown in dotted and dashed lines.
Fig. 4. Higgs boson masses Mh0 vs. X tOS . One-loop masses, two-loop masses to O(s t + t2 ) and
their approximations are shown in dashed, solid and dot-dashed lines.
In Fig. 4 we compare results from our analytical approximation formula for Mh0 in
Section 2 with those obtained by full numerical evaluations. They are shown in dot-dashed
and solid lines respectively. The analytical approximation formula works remarkably well:
it approximates our numerical results to a precision of <
0.5 GeV for almost all the
parameter space. The analytical approximation has a complicated dependence on the parameter. Numerically this dependence is quite weak: varying from 100 GeV to 1 TeV
20
Fig. 5. Higgs boson masses Mh0 vs. X tOS . Two-loop masses to O(s t ) and O(s t + t2 ) are
shown in dotted and dashed lines, their corresponding RG-corrected masses are shown in dot-dashed
and solid lines.
for a fixed X tOS changes the Higgs boson mass by less than 1 GeV. We emphasize that the
analytical formula is useful for several reasons: (1) the logarithmic and finite corrections
can be easily separated, and one can weight the relative importance of these terms; (2) all
terms can be traced back to the potential, so one can easily locate the particles giving the
biggest contributions; (3) the formula can significantly simplify the numerical evaluations
of Mh0 to a good precision.
In Fig. 5 we further compare the results for our RG-corrected Higgs boson masses,
Eqs. (28)(30), with those of the full numerical evaluation. For comparison, we have
also shown two-loop O(s t ) corrections and their RG-corrected results; they have been
studied previously in [13]. As mentioned in Section 3, the good agreement between these
curves is an indication of the smallness of the logarithmic corrections beyond two-loops
and illustrates the accuracy of our results.
Finally in Figs. 6 and 7 we detail the size of the two-loop top-squark-mixing-dependent
corrections in the OS-scheme and their corresponding finite threshold corrections in the
RG approach. Fig. 6 shows in dotted lines two-loop masses without including the topsquark-mixing-dependent corrections of Eq. (20). Refs. [12,13,21] have already calculated
the QCD corrections, and they are depicted in dashed lines. The difference of the solid and
dashed lines is the two-loop O(t2 ) terms which are calculated in this paper. We see clearly
that these terms are sizable: for large mixing parameters, they increase Mh0 by about 4 GeV
and 23 GeV for small and large tan , respectively.
2
Fig. 7 shows the effect of two-loop threshold corrections 1(2)
th mh0 evaluated following
the RG-inspired analysis of Section 3. The dotted lines show the Higgs boson mass
neglecting these corrections; this would have been obtained by integrating two-loop
RG equations (with the two-loop boundary threshold correction being set to zero), as
we have shown in the second part of Section 3. Two-loop masses without the O(t2 )
threshold correction and the complete two-loop results are shown in dashed and solid lines,
21
Fig. 6. Higgs boson masses Mh0 vs. X tOS . Two-loop masses without the top-squark-mixingdependent correction terms of Eq. (20) are shown in dotted lines. The corresponding masses without
the O(t2 ) corrections only and the full numerical results are shown in dashed and solid lines,
respectively.
Fig. 7. Higgs boson masses Mh0 vs. X tOS . Two-loop masses without the threshold correction
(2)
term 1th m2 0 are shown in dotted lines. The corresponding masses without the O(t2 ) threshold
h
corrections only and with the full numerical results are shown in dashed and solid lines, respectively.
respectively. The RG reshuffling of radiative corrections has absorbed the main part of the
two-loop top-squark-mixing-dependent terms of Eq. (20) into the RG-corrected one-loop
(1)
term 1th m2h0 ; the remaining genuine two-loop threshold corrections (in the sense of the
effective field theory) are generally small, <
3 GeV.
22
5. Conclusions
In this paper we calculate radiative corrections to the lightest MSSM CP-even

Higgs boson mass to the two-loop O(t2 ) order. Our analysis extends existing two-loop
diagrammatic results [1113,21] using a simpler effective potential approach and provides
the most complete and accurate calculation presented in the literature. We also derive
useful analytical approximation formulae, applicable when the supersymmetric particles
are heavy, which accurately reproduce results from the full numerical study.
Our calculation includes effects which can have an impact on the final Higgs mass but
were neglected by previous studies. In particular, the two-loop O(t2 ) top-squark-mixingdependent corrections to Mh20 [see Eq. (20)] are calculated for the first term in this paper
and are numerically important.
We further simplify our analytical formula by reshuffling higher order logarithmic corrections (using RG techniques) in a compact one-loop expression [Eq. (28)]. In that expression all mass parameters are evaluated at appropriate renormalization scales chosen to
reproduce the numerically most important leading and next-to-leading logarithmic corrections. The remaining two-loop finite terms can be interpreted as threshold corrections, and
are numerically less important. This RG rewriting clarifies the structure of the two-loop
corrections to Mh20 , identifies the most important contributions and links our work to the
effective theory or RG approach, as we have shown in detail in Section 3.
To summarize our numerical results, we have shown that two-loop top Yukawa
corrections to Mh0 are sizable for the maximal top-squark mixing case. They can increase
the Higgs boson mass Mh0 by as much as 5 GeV (among which the top-squark-mixingdependent corrections account for about 4 GeV) for small tan where ht is large. The
upper bound on Mh0 is 129 5 GeV for Mt = 175 5 GeV. Our final approximation
formulae (20)(21) and (28)(30) have been shown to excellently agree with the full
numerical results and can be easily implemented in precision numerical studies.
Although we have focussed in this paper on the Higgs mass, it is worth mentioning that
we have also presented in Appendix D the MSSM two-loop effective potential including
top-quark Yukawa contributions (for general top-squark mixing parameters and any tan ).
This knowledge may well prove useful for other studies.
Acknowledgements
We thank Andr Hoang for correspondence. R.-J.Z. was supported in part by a
DOE grant No. DE-FG02-95ER40896 and in part by the Wisconsin Alumni Research
Foundation.
23
Appendix A. One- and two-loop scalar functions

A.1. One-loop scalar functions
In this subsection we define the scalar functions A0 , B0 , B1 , B22 and G, which appear
in one-loop self-energy calculations.
The A0 function is defined by the following momentum integral in d = 4 2
dimensions

Z
m2
dd p
1
2 1
=
m
+
1
ln
,
(A.1)
A0 (m2 ) = 16 2 4d
i(2)d p2 m2 + i

Q2
where Q2 = 42 eE is the renormalization scale, with E the Euler constant.
The B0 function is
Z

1
dd q
2
2
2
2 4d
B0 p , m1 , m2 = 16
i(2)d [q 2 m21 + i][(q p)2 m22 + i]
1
=

Z1
dx ln
(1 x)m21 + xm22 x(1 x)p2 i

.
Q2
(A.2)
The remaining functions can be related to A0 and B0 as follows

i
1 h
B1 p2 , m21 , m22 = 2 A0 m22 A0 m21 + p2 + m21 m22 B0 p2 , m21 , m22 ,
2p

2
B22 p2 , m21 , m2
(A.3)

1
= A0 m22 + 2m21 B0 p2 , m21 , m22
6

p2
B1 p
,
p
3

G p2 , m21 , m22 = p2 m21 m22 B0 p2 , m21 , m22 A0 m21 A0 m22 .
2
+ m21
m22
, m21 , m22
+ m21
+ m22
(A.4)
(A.5)
In all one-loop expressions of radiative corrections, we adopt a (modified) minimal

subtraction procedure to remove poles in and keep only finite (real) parts of the above
functions.
Some useful expressions for these functions in limiting cases are (after minimal
subtraction)

m2
m2
m2
B0 0, m21 , m22 = 1 ln 12 + 2 2 2 ln 22 ,
Q
m1 m2 m1

m21
m22
m22
m2 m2
2
2
B0 m1 , m2 , 0 = 2 ln 2 1 2 ln 1 2 22 ln 22 ,
Q
m1
m1
m1 m1

1
d
B0 p2 , m2 , m2 p2 =0 =
,
2
dp
6m2
(A.6)
(A.7)
(A.8)
24
m2
B0 m2 , m2 , m2 = ln 2 + 2 .
Q
3
(A.9)
A.2. Two-loop scalar functions

In this subsection we collect some useful formulae of zero-point two-loop scalar functions. They have been studied extensively by several groups using two different methods: a
differential equation method [3739] and an integral Mellin-Barnes transformation method
[4042]; their results all agree. Here we mainly follow Ref. [38].
The momentum integrals appearing in a two-loop effective potential calculation can be
reduced to the following two types of scalar functions [corresponding to the topologies of
two distinct zero-point two-loop irreducible Feynman diagrams (the figure-8 and sunset
diagrams)]:
Z d d

1
d pd q
2
2
2 4d 2
,
(A.10)
J m1 , m2 = 16
(2)2d [p2 m21 + i][q 2 m22 + i]
and

2
I m21 , m22 , m23 = 16 24d
[p2
m21
dd p dd q
(2)2d
+ i][q 2
m22
1
.
+ i][(p + q)2 m23 + i]
The function J is symmetric in m1 , m2 and I symmetric in m1 , m2 and m3 .

The function J can be reduced to the product of one-loop scalar functions as

J m21 , m22 = A0 m21 A0 m22 .
(A.11)
(A.12)
The function I satisfies the following first-order partial differential equation [39]
R2

I m21 , m22 , m23 = (d 3) m23 m21 m22 I m21 , m22 , m23

2
m3
2

m m21 + m22
J m21 , m23
+ (d 2) 3
2
2m3
+
m23 + m21 m22

2m23
m22 , m23
m21 , m22

(A.13)
where
R 2 = m41 + m42 + m43 2m21 m22 2m21 m23 2m22 m23 .
(A.14)
This differential equation can be used to solve for the I function. The initial value of this
function can be evaluated from (A.13) which reduces to a simple algebraic equation when
m3 = m1 + m2 , i.e., R = 0.
In our calculation, any Feynman diagram in the two-loop effective potential is subtracted
by all its possible one-loop sub-diagrams; this is done by replacing the I and J functions
as follows [38]:
25

I m21 , m22 , m23 I m21 , m22 , m23
1h

i
= I m21 , m22 , m23 A0 m21 + A0 m22 + A0 m23 ,

1h

i
2
2
2
2
J m1 , m2 J m1 , m2 = J m21 , m22 + m21 A0 m22 + m22 A0 m21 . (A.15)

It is then straightforward to show

m21
m22
m21 m22
2
2
2 2
1 ln 2 ,
(A.16)
J m1 , m2 = 2 + m1 m2 1 ln 2

Q
Q
and with some effort

1
1
m21 + m22 + m23
I m21 , m22 , m23 = 2 m21 + m22 + m23
2
2

m2 m2
1
m21 + m22 + m23 ln 22 ln 32
2
Q
Q
m2 m2
+ m21 m22 + m23 ln 12 ln 32
Q
Q
m2 m2
+ m21 + m22 m23 ln 12 ln 22
Q
Q

m2
m2
m2
4 m21 ln 12 + m22 ln 22 + m23 ln 32
Q
Q
Q

+ m21 , m22 , m23 + 5 m21 + m22 + m23 ,
where (for R 2 > 0) is given by [40]

m2 + m21 m22 R m23 m21 + m22 R
ln
m21 , m22 , m23 = R 2 ln 3
2m23
2m23
2

m3 + m21 m22 R
m2 m2
ln 12 ln 22 2 Li2
m3 m3
2m23

2

m3 m21 + m22 R
2
,
+
2 Li2
3
2m23
(A.17)
(A.18)
where Li2 (x) is the dilogarithm function

Z1
Li2 (x) =
dy
ln(1 xy)
.
y
(A.19)
In the region where R 2 < 0, (A.18) should be replaced by its analytical continuation.
Equivalent expressions for also appear in [37] and [39]; we find that (A.18) is most
convenient for series expansions. We also define a function L for future use
26

L m21 , m22 , m23 = J m22 , m23 J m21 , m22 J m21 , m23

m21 m22 m23 I m21 , m22 , m23 .
(A.20)
Performing a (modified) minimal subtraction (by removing the single and double poles
in ), it is the finite (real) parts of (A.16) and (A.17) that we use in our two-loop effective
potential expressions. We will also omit the carets of I and J to simplify the notation.
When computing the two-loop potential, some argument of the I function, e.g., the
bottom-quark mass mb , tree-level Higgs boson mass mh0 , can be taken to be zero. The
function I is well-behaved in these limiting cases:

m2 m2 m2
m2 m2
I m21 , m22 , 0 = m21 ln 12 ln 22 m21 m22 ln 1 2 2 ln 12
Q
Q
Q
m2

m2
m2
m2
1
5
+ m21 m22 ln2 12 + 2 m21 ln 12 + m22 ln 22 m21 + m22
2
2
Q
Q
Q

2
m22
+ Li2
,
(A.21)
+ m21 m22
6
m21

m2 5 2
1 2 m2
ln
+
,
(A.22)
2
ln
+
I (m2 , 0, 0) = m2
2
6
Q2
Q2 2
where we have kept only the finite terms as explained before. In (A.21) we have implicitly
assumed m1 > m2 . The symmetry of the above expression for I (m21 , m22 , 0) in m1 and m2
[which obviously follows from the definition (A.15) of I ] can be explicitly checked by
using the identity
Li2 (x) = Li2 (x 1 )
1 2
2
ln (x)
.
2
6
(A.23)
Finally, we collect expansion formulae for the function which we use in the derivation
of an analytical approximation formula for the two-loop Higgs boson mass corrections.
The functions we find can be reduced to one of the different types we list below using
the relation

(A.24)
m21 , m22 , m23 = m21 1, m22 /m21 , m23 /m21 .
(1) For 0 6 r 6 1 and 0 6 1:
)
(

2
+ ln 2 ln(1 r) ln r 2 Li2 (r)
(1, r, ) = (1 r)
3
(
2 2 ln + ln r

)

1 + r 2
+ ln 2 ln(1 r) 1 ln r 2 Li2 (r)
+
1r 3
2
+
(1 r)3
(
27

3
2 2
ln (1 r 2 )
r
2
3
)

2 ln + r 4 ln(1 r) r ln r + 4r Li2 (r) + O( 3 ).
(A.25)
If r > 1, one uses (1, r, ) = r(1, 1/r, /r) and the above expression.
Two particular cases of the previous expansion are:
(1a) For 0 6 1 , 2 1:

2
2
+ ln 1 ln 2 2 1 +
+ ln 1 ln 2 1 2
(1, 1 , 2 ) =
3
3

2
+ 2 ln 1 ln 1 ln 2 1
+ 2
3

3
2
ln 1 1 + (1 2 ) + O 1m 2n ,
+
2
with m + n = 3, and
(1b) For 0 6 |1 |, 2 1:

8 1
ln 2 22
(1, 1 + 1 , 2 ) = 2(4 + 1 2 ln 2 )2 +
9 3

1
7
+ ln 2 2 12
+ 2 ln 2 +
18 6

1 1
2 1
3
ln 2 14 + O 1m 2n ,
+ + ln 2 1 +
2 2
9 3
with m + 2n > 5.
Finally we also give
(2) For |1 |, |2 | 1:
(A.26)
(A.27)

8
5
K 12 22
(1, 1 + 1 , 1 + 2 ) = 36K + (8K 1)1 2 +
36 3

2 3
16
8
1
2
K
K 14
+ 12K1 + (1 8K)1 +
+
3
9 1
108
9

2

11 8
+ K 13 2 + (1 2 ) + O 1m 2n ,
+ 1 +
6
54 9
(A.28)

with m + n = 5. In this expansion the constant number K is given by

1
K =
3
Z/6
dx ln(2 cos x) ' 0.1953256.
0
(A.29)
28
Appendix B. MSSM in the leading approximation

The general structure of the MSSM is quite complicated, with many different fields
and field mixings. This makes the computation of the complete potential prohibitive at
two-loops. However, it is a good approximation to keep only those terms of the MSSM
Lagrangian which depend of the SU(3) gauge coupling g3 and the top Yukawa ht (and
neglect the electroweak gauge couplings g1 , g2 and the rest of the Yukawa couplings). We
call this the leading approximation and it greatly simplifies our two-loop effective potential
calculation. In this appendix, we summarize the necessary Feynman rules for computing
the two-loop potential in this leading approximation and also some MSSM renormalization
group equations, useful to check the scale invariance of the potential.
B.1. Masses and Feynman rules
The Higgs sector scalar potential in the leading approximation is

VHiggs = m2H1 + 2 |H1 |2 + m2H2 + 2 |H2 |2 + B (H1 H2 + H.c.),
(B.1)
where mH1 , mH2 and B [with dimensions of (mass)2 ] are the soft-breaking Higgs mass
parameters, and the supersymmetric Higgs-boson mass parameter. Although we do
not write the quartic Higgs couplings, which depend on the electroweak gauge coupling
constants, they are responsible for the tree-level mass of the lightest Higgs boson, which
we of course include in our calculations.
The SU(2) doublet Higgs fields H1 and H2 can be written as follows:
#
"
"
#
h+
(h1 + ia1 )/ 2
2
(B.2)
,
H2 =
H1 =
.
(h2 + ia2 )/ 2
h
1
In our approximation, the mass-squared matrices for CP-even and odd Higgs fields are
!
m2H1 + 2 B
2
,
(B.3)
M =
B m2H2 + 2
where the positive (negative) sign applies to the CP-even (odd) fields, respectively. The
charged Higgs fields have the same mass-squared matrix M2 as the CP-odd Higgses.
The CP-even interaction eigenstates h1 , h2 are rotated by the angle into the mass
+
eigenstates H 0 and h0 . Similarly, the CP-odd states a1 , a2 (charged states h+
1 , h2 ) are
0
0
+
+
rotated into mass eigenstates G and A (G and H ) by the angle . This angle is
conventionally defined in terms of the CP-even Higgs field VEVs, hh1,2 i = v1,2 , by tan =
v2 /v1 . The fact that diagonalizes M2 is obvious when the minimization conditions of
the potential (B.1), m2H1 + 2 = B tan and m2H2 + 2 = B cot , are imposed and
the soft parameters in the matrix are replaced by tan and m2A0 = B (tan + cot ).
Since we have neglected all g1 , g2 related terms in (B.1), in our approximation (we use
shorthand notations c = cos , s = sin , etc.)
c = s , and s = c .
(B.4)
29
This approximation is excellent when MA0 MZ but would fail for MA0 MZ . The
effect is numerically relevant for the tree level masses and we take it into account, but it
may be consistently neglected in the two-loop corrections.
The (field-dependent) top squark mass-squared matrix in the leading approximation is
1 2 2
1
2
h
M
h
+
h
(A
h
+
h
)
t
t
2
1
e
Q
2 t 2
2
,
(B.5)
M2t =
1
1
2
2 2
MUe + ht h2
ht (At h2 + h1 )
2
2
where MQ
e , MU
e are soft-breaking mass parameters of the left-handed and right-handed
e
e, and At the usual trilinear soft-breaking parameter. Denoting the
top-squarks Q and U
mass eigenvalues of the matrix (B.5) by mt1 , mt2 and the mixing angle by t, the Feynman
rules for Higgs/Goldstone-boson-top squark trilinear coupling are simply i, with as
listed below:
H + t1 bL = ht (ct mt + st Yt )c ,
H + t2 bL = ht (st mt ct Yt )c ,
G+ t1 bL = ht (ct mt + st Xt )s ,
H 0 t1 t1 = 2 ht (mt + st ct Yt )c ,
h0 t1 t1 = 2 ht (mt + st ct Xt )s ,
1
H 0 t1 t2 = ht c2t Yt c ,
2
1
A0 t1 t2 = A0 t2 t1 = ht Yt c ,
2
G+ t2 bL = ht (st mt ct Xt )s ,
H 0 t2 t2 = 2 ht (mt st ct Yt )c ,
h0 t2 t2 = 2 ht (mt st ct Xt )s ,
1
h0 t1 t2 = ht c2t Xt s ,
2
1
G0 t1 t2 = G0 t2 t1 = ht Xt s ,
2
(B.6)
where ct = cos t, st = sin t, c2t = cos 2t and

Xt = At + cot ,
Yt = At tan .
(B.7)
The couplings of squarks to neutralinos and charginos are very simple in the leading
approximation, since the gaugino-Higgsino mixing can be neglected and the only
interactions are Higgsino-squark interactions. The Feynman rules for the h 0i t tj couplings
can be written as i(aPL + bPR ) and that of h + t bL as iC 1 (aPL + bPR ) (PL,R are chiral
projectors and C the charge-conjugation matrix), with
ht
ah 0 t t1 = iah 0 t t1 = bh 0 t t2 = ibh 0 t t2 = ct ,
1
2
1
2
2
ht
ah 0 t t2 = iah 0 t t2 = bh 0 t t1 = ibh 0 t t1 = st ,
1
2
1
2
2
ah + bt1 = ht st ,
ah + bt2 = ht ct ,
ah + t b L = ht ,
(B.8)
when > 0; for < 0, we only need to interchange a h 0 t ti and a h 0 t ti , as well as bh 0 t ti and
1
2
1
bh 0 t ti .
2
Other Feynman rules of O(g3 ) and O(ht ) vertices are exactly the same as in the general
MSSM and we do not present them explicitly.
30
B.2. Renormalization group equations

The MSSM RGEs [4649] that we will use to check the invariance of the potential to
two-loop order under renormalization scale transformations are the following. First, we
need the two-loop RGEs for those parameters entering in the tree-level potential (B.1)
m2H2
ln Q2
ln
ln Q2
B
ln Q2

16g32 h2t
3h2t
18h4t
2
M
+
M2t + 2M32 2M3 At
M2t + A2t ,
t
2
2
2
2
2
16
(16 )
(16 )
(B.9)
2
2
2
4
8g3 ht
ln h2
3ht
9 ht
=
=
+
,
(B.10)
ln Q2
32 2 (16 2 )2 2 (16 2 )2

16g32 h2t B
3h2t
B
+
A
+ At M3
=
+
t
16 2 2
(16 2)2 2

B
9h4t
+
2A
,
(B.11)
t
(16 2 )2 2
=
2 + M 2 + A2 . Then we need one-loop RGEs for those masses

where M2t = m2H2 + MQ
t
e
e
U
entering in the one-loop potential

16g32
m2t
2
2
+
3h
=
(B.12)
16 2
t mt ,
ln Q2
3

m2t
16g32 2
Xt
2
1
m
=
+
M
s
m
M
16 2
2t t
3
t
3
3
2
ln Q2

3Xt
,
(B.13)
+ h2t 3m2t + (1 + st2 )M2t + 3s2t mt At +
2

m2t
16g32 2
Xt
2
2
m
=
+
M
+
s
m
M
16 2
2t t
3
t
3
3
2
ln Q2

3Xt
,
(B.14)
+ h2t 3m2t + (1 + ct2 )M2t 3s2t mt At +
2

m2H 0
B
2
2
2
n
+
A
=
3h
+
D
M
+
E
,
(B.15)
16 2
n
n
t
t
t
ln Q2
2

m2H +
B
n
2
2
2
2
+ At ,
= 3ht + Dn+2 Mt + En+2
(B.16)
16
2
ln Q2
where Dn = s2 , c2 , s2 , c2 and En = s2 , s2 , s2 , s2 for n = 1, 2, 3, 4. [Here we use

the angle to keep track the H 0 and h0 contributions; it can be replaced by the angle
as in (B.4) in the leading approximation.] The ordering of the Higgs/Goldstone bosons are
Hn0 = H 0 , h0 , G0 and A0 for n = 1, 2, 3, 4 and Hn+ = G+ , H + for n = 1, 2. Eqs. (B.13)
(B.16) seem unfamiliar, but they follow directly from (B.3), (B.5) and the one-loop MSSM
RGEs of soft parameters entering those equations.
Using (B.13) and (B.14), we find one-loop RGEs for Xt and m2t , the (arithmetic) average
of the (squared) top squark masses. They are
31
16
Xt
= g32 M3 + 3h2t (At + Xt ).
2
ln Q
3

2
m

16 2 2
3 2
2
2
2
t
g
M
=
+
M
+
m
+
h
16 2
3m
,
t
t
3
ln Q2
3 3 t
2 t
16 2
(B.17)
(B.18)
these two equations are used in Section 3 for the RG discussion of the formula for the
Higgs boson mass Mh0 . Eq. (B.17) can also be derived from (B.7) and one-loop RGEs of
At , and tan .
Appendix C. One-loop self-energies

In this appendix, we collect formulae for those MSSM one-loop self-energies which are
necessary for our analysis. We present these self-energies in the leading approximation of
keeping only ht and g3 -dependent terms, as explained in Appendix B; their full form can
be found in Ref. [22], which we follow for notation. (See also [50] for top quark/squark
self-energies.)
Top quark
16 2 t (p2 ) =

4g32 n
m2
mt B1 p2 , m2g , m2t + B1 p2 , m2g , m2t mt 5 3 ln t2
1
2
3
Q
o

s2t mg B0 p2 , m2g , m2t B0 p2 , m2g , m2t
1
n

+ mt c2 2B1 p2 , m2t , m2A0 + B1 p2 , m2b , m2A0
2

+ s2 2B1 p2 , m2t , m2Z + B1 p2 , m2b , m2Z

o
+ B1 p2 , 2 , m2t + B1 p2 , 2 , m2t + B1 p2 , 2 , m2b ,
h2t
(C.1)
where we have assumed all heavy Higgs bosons have mass mA0 much larger than the
masses of the light Higgs and W -boson, taken to be mZ .
From (C.1) we find the running top-quark mass at the scale Q (under the simplified
assumptions of a common heavy SUSY scale MS while the parameter is left free, see
Section 2)
(

g2
M2
m2
2
2
(C.2)
mt (Q) = Mt 1 32 5 3 ln t2 + ln S2 X t
6
Q
Q

1
MS2
m2t
2 8
ln 2 + s
ln 2
2
Q
3
Q
)

2
2
2
.
ln
1
+
1 2
1 2
3h2t
+
32 2
1 + c2
32
In this equation we have neglected the external momentum and used (A.3) and (A.6). We
have used the reduced parameters X t Xt /MS , /MS and Mt is the top quark pole
mass (we use capital letters to denote on-shell mass parameters).
Top squarks
16 2 tL tL (p2 )

8g 2 n
= 3 G p2 , m2g , m2t + ct2 A0 m2t p2 + m2t B0 p2 , m2t , 0
1
1
1
3
o

2
2
2
2
2
2
+ st A0 mt p + mt B0 p , mt , 0
2

1
+ h2t st2 A0 m2t + ct2 A0 m2t +
1
2
2
+
2
4 X
X
4
X

Dn A0 m2H 0 + G p2 , 2 , m2t
n
n=1
2
X
2H 0 t t B0 p2 , m2H 0 , m2t +
2H + t
n=1 i=1
n L i
i,n=1

p2 , m2H + , m2b ,
B0
L b L
(C.3)
16 2 tR tR (p2 )

8g 2 n
= 3 G p2 , m2g , m2t + st2 A0 m2t p2 + m2t B0 p2 , m2t , 0
1
1
1
3
o

+ ct2 A0 m2t p2 + m2t B0 p2 , m2t , 0
2
2
2
"
4

1X

Dn A0 m2H 0
+ h2t ct2 A0 m2t + st2 A0 m2t + A0 m2b +
1
2
L
n
2
n=1
#
2
X

Dn+2 A0 m2H + + G p2 , 2 , m2t + G p2 , 2 , m2b
+
n
n=1
2
4 X
X
2
X

2H 0 t t B0 p2 , m2H 0 , m2t +
2H + t
n=1 i=1
16 2 tL tR (p2 ) =
n R i
i,n=1

p2 , m2H + , m2b , (C.4)
B0
R b L

4g32
2 p2 + m2t B0 p2 , m2t , 0 +2 p2 + m2t B0 p2 , m2t , 0
1
1
2
2
3
3

+ 4mg mt B0 p2 , m2g , m2t + h2t s2t A0 m2t A0 m2t
1
2
2
2
4 X
X

Hn0 tL ti Hn0 tR ti B0 p2 , m2H 0 , m2t
+
n=1 i=1
2
X

Hn+ tL bL Hn+ tR bL B0 p2 , m2H + , m2b ,
n
i,n=1
(C.5)
where H 0 tL t1 = ct H 0 t1 t1 st H 0 t2 t1 , H 0 tR t1 = st H 0 t1 t1 + ct H 0 t2 t1 , etc. The symbols

Dn are defined after (B.7).
33
From (C.3)(C.5) we derive relations between running and on-shell top-squark masses
and mixing parameters using the following one-loop relationships (for ct2 = st2 = 1/2):

1
Re tL tL Mt2 + tR tR Mt2 Re tL tR Mt2 ,
1
1
1
2

1
2
2
2
+ mt mt Xt Re tL tL Mt2 + tR tR Mt2 + Re tL tR Mt2 , (C.6)
Mt = MQ
e
2
2
2
2
2
we obtain (assuming again a common heavy SUSY scale MS and leaving free the parameter):
(

g32
MS2
2
2
2 ln 2
mt (Q) = Mt 1
3 2
Q

2

MS2
M2
2
3ht
t2 s2 + Yt2 c2 2 ln S + c2 1
Y
X
ln
+
t
32 2
Q2
Q2
3
)

MS2
4
2
2
2 2
2
ln 1
3 2 ln 2 1
, (C.7)
+ ln + 1
Q
2
2
Mt2 = MQ
e + mt + mt Xt
1

g32
MS2
MS2
+
mt MS 4 2 ln 2 + 2Xt ln 2
12 2
Q
Q
(

2
2

M
3ht
mt Xt s2 + Yt c2 2 ln S2 Yt c2
+
16 2
Q
3
)

2

3 MS
1
2
4
2
4
2
+ Xt 1 ln 2 1 + ln + 1 ln 1 Xt ,
2 Q
2
mt Xt (Q) = Mt XtOS
(C.8)
where we have used (A.7), (A.9) and the definition Yt (At tan )/MS .
W boson
16
1

2B22 p2 , m2t , m2b + G p2 , m2t , m2b
2

1

2
2
2
2
2
2ct B22 p , mt , mb A0 mt
1
1
L
4
#
)

1

1
2
2
2
2
2
2
+ A0 mb
.
2st B22 p , mt , mb A0 mt
2
2
L
L
4
2
T
2
2
W
W (p ) = 3g
(C.9)
This gives (under the assumption of the simplified SUSY spectrum of Section 2, described
already for previous self-energies)

2
h2t s2
4MW
4 2
m2t
T
2
2
+ Re W
+
3
+
X
1
6
ln
,
M
=
v 2 (Q) = 2 MW
t
W
W
g
g2
32 2
Q2
(C.10)
where we have neglected the external momentum in (C.9) and used (A.3)(A.6).
34
Higgs boson
We need only the difference

16 2 hh m2h0 hh (0)
"
#

d
2
2
2
B0 p , mt , mt
dp2
p 2 =0

X

d
2h0 t t
B0 p2 , m2t , m2t
,
+ 3m2h0
i j dp 2
i
j
p 2 =0
= 3h2t m2h0 s2
B0 0, m2t , m2t
4m2t
(C.11)
i,j
where hti tj are defined in (B.6). Using (A.6) and (A.7) we get (19).
Appendix D. MSSM effective potential to the two-loop order

In this appendix we present the MSSM effective potential for the (real) neutral
components of the Higgs fields up to the two-loop level in the leading approximation
(which neglects all dimensionless couplings except ht and g3 ). We first write the potential
as
V (h1 , h2 ) = Vvac + V0 (h1 , h2 ) + V1 (h1 , h2 ) + V2 (h1 , h2 ),
(D.1)
where Vvac is a field-independent vacuum energy term. 6 The tree-level potential V0 is

1 2
1
(m + 2 )h21 + (m2H2 + 2 )h22 + B h1 h2 ,
(D.2)
2 H1
2
which simply follows from substituting Eq. (B.2) into the MSSM Higgs sector scalar
potential (B.1).
The one-loop potential is well known and the O(s t ) part of the two-loop potential
was computed in [12]; we list them here for completeness and for future reference. The
complete one-loop potential in Landau gauge is 7
#
"
X f X

3

2
2
2
16 V1 (h1 , h2 ) =
Nc
H mf 2H mf + 3H m2W + H m2Z
i
2
V0 (h1 , h2 ) =
1
+
2
i=1,2
4
X
n=1
m2H 0
n
2
X
m2H +
n
n=1
2
X
i=1
m2 +
i
4
X
i=1

H m2 0 ,
i
(D.3)
f
where f sums over all the (s)quarks and (s)leptons, Nc is the color factor, 3 for (s)quarks
and 1 for (s)leptons. Following the leading approximation, we keep only the numerically
6 This term is a function of the soft-breaking parameters; it is needed for the invariance of the potential under a
RG transformation.
7 We adopt the (modified) DR -scheme of Refs. [4345].
35
important parts, i.e., those from top (s)quarks. In Eq. (D.3), i+ (i = 1, 2) and i0 (i =
1, 2, 3, 4) represent charginos and neutralinos, and the function H is

m2 3
m4
ln 2
.
(D.4)
H (m2 ) =
2
Q
2
The QCD contribution to the two-loop effective potential in the MSSM is
(

2 2
2
V2s (h1 , h2 ) = 8g3 J m2t , m2t 2m2t I m2t , m2t , 0
16

X

X 2
1 4
ct + st4
J m2t , m2t + 2st2 ct2 J m2t , m2t +
mt I m2t , m2t , 0
i
i
i
i
i
1
2
2
2
X
i=1
i=1
m2t , m2g , m2t

i
)

2
2
2
2
2
2
4mg mt st ct I mt , mg , mt I mt , mg , mt
, (D.5)
1
i=1
where g is the gluino, with tree-level mass given by the SU(3) gaugino soft mass M3 . The
two-loop scalar functions I , J and L in Eq. (D.5) are given in Appendix A, Eqs. (A.17),
(A.16) and (A.20). 8
The top Yukawa contribution to the two-loop potential is a new result of this paper. The
relevant Feynman diagrams are shown in Fig. 8. To simplify the final result, we neglect
left-right mixings in the bottom-squark sector and the gauginoHiggsino mixings in the
neutralinochargino sector (under this assumption, the Higgsino masses are simply ||);
Fig. 8. Feynman diagrams for the two-loop effective potential of order O(t2 ) in the MSSM. Hn0
represent H 0 , h0 , G0 and A0 for n = 1, 2, 3, 4. Hn+ represent G+ , H + for n = 1, 2. The neutral and
0 ) and e
charged Higgsinos e
h01,2 ( 3,4
h+ ( 2+ ) have degenerate mass of ||.
8 The procedure we have followed of subtracting all possible one-loop subdivergences to define these functions
is an alternative to the direct way used in Ref. [11]. The direct way requires the computation of some one-loop
quantities to order O(); perhaps we find the subtraction method simpler. We have explicitly checked that, in the
particular limit studied in [11], we exactly reproduce their unexpanded mass formula, Eq. (11) of [11], which
shows the equivalence of both methods.
36
these simplifications are valid in the leading approximation. Using the Feynman rules given
in Appendix B, we find (the last diagram of Fig. 8 is of order h2t but does not contribute to
mh0 )
16 2
2
V2t (h1 , h2 )
"
#
( 4
2
X Dn

X

2
2
2
2
2
2
2
2
2
2
L mH 0 , mt , mt 2mt I mH 0 , mt , mt +
J mt , mH 0
= 3ht
i
n
n
n
2
n=1
2
X

Dn+2 L m2H + , m2t , m2b + st2 J m2t , m2H +
1
n=1
+ ct2 J
2
+s2t
i=1
m2t , m2H + + J m2b , m2H +
2
X
+ st2 J m2t , m2b

1

L
+ ct2 J m2t , m2b

2

L
2

X

J m2t , m2t + c4t J m2t , m2t +
L m2t , 2 , m2t
i
i=1
+L
i=1
m2b , 2 , m2t
L
st2 L
m2t , 2 , m2b
1
+ ct2 L
m2t , 2 , m2b
2
4
2
2
2 X
X

3 XX 2
2
2
2
H 0 t t I mH 0 , mt , mt 3
2H + t b I m2H + , m2t , m2b ,
i
j
i
n
L
n i j
n
n i L
2
i,j =1 n=1
i,j =1 n=1
(D.6)
where in the first line of Eq. (D.6), positive and negative signs apply to CP-even (H 0 , h0 )
and odd (A0 , G0 ) Higgs/Goldstone bosons respectively. The Higgs/Goldstone-squark
couplings are defined in Eq. (B.6) while the Dn symbols are defined after Eq. (B.7).
Two tests can be applied to check the correctness of the effective potential V (h1 , h2 ).
First, the potential should vanish in the supersymmetric limit (i.e., when all soft-breaking
parameters are taken to be zero), and second, the potential V (h1 , h2 ) should be invariant
under changes of the renormalization scale, up to the order of our perturbative calculation.
The vanishing of the potential in the supersymmetric limit is proved by simple algebra. In
the following we show the invariance of the two-loop potential under a RG transformation.
Using the derivatives of I , J and L functions with respect to the renormalization scale Q
X

I (m21 , m22 , m23 )
=
A0 m2i + m2i ,
2
ln Q
3
(D.7)
i=1

J (m21 , m22 )
= m21 A0 m22 + m22 A0 m21 ,
2
ln Q

L(m21 , m22 , m23 )
= m21 2m22 2m23 A0 m21 m22 A0 m22 m23 A0 m23
ln Q2
2
+ m41 m22 + m23 ,
(D.8)
(D.9)
37
and the one-loop MSSM RGEs for top-(s)quark and Higgs boson masses, Eqs. (B.12)
(B.16), we find that the RG variation

8g32 h2t h22
V2
(1)
2
2
2
2
D
V
=
MQ
1
e + MU
e + 2M3 + Xt 2M3 Xt
ln Q2
(16 2)2

9h4t h22
1 2
2
2
2
m + (At + Xt )
MQ
e + MU
e+
(16 2 )2
2 H2
(D.10)
modulo terms independent of the Higgs field h2 . Here D(1) V1 represents the one-loop RG
variation of the one-loop potential Eq. (D.3). This result agrees exactly with the two-loop
RG variation of the tree-level potential D(2) V0 [cf. Eqs. (D.2) and (B.9)(B.11)], so that
V2
d
(V0 + V1 + V2 ) D(2) V0 + D(1) V1 +
= 0.
d ln Q2
ln Q2
(D.11)
Note that this is a nontrivial check that all ln Q2 terms cancel with each other between
Eq. (D.10) and D(2) V0 ; this cancellation guarantees the correct leading and next-to-leading
order logarithmic behavior of the effective potential.
To derive the analytical expression of 1m2h0 in Section 2, we need to expand the twoloop potential V2 in powers of mt /MS and mt Xt /MS2 ; besides many straightforward
expansions, we have used (A.25)(A.28) for the t q h and t q h diagrams of
(D.6), with q = t or b.
References
[1] H.P. Nilles, Phys. Rep. 110 (1984) 1.
[2] H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75.
[3] LEP Experiments Committee Meeting, March 7th, 2000:
http://alephwww.cern.ch/ALPUB/seminar/lepc mar2000/LEPC2000files/v3 document.htm;
http://delphiwww.cern.ch/offline/physics links/lepc.html;
http://l3www.cern.ch/analysis/latestresults.html; http://www.cern.ch/Opal/PPwelcome.html.
[4] S.P. Li, M. Sher, Phys. Lett. B 140 (1984) 339.
[5] J. Ellis, G. Ridolfi, F. Zwirner, Phys. Lett. B 257 (1991) 83; Phys. Lett. B 262 (1991) 477.
[6] Y. Okada, M. Yamaguchi, T. Yanagida, Prog. Theor. Phys. 85 (1991) 1.
[7] D.M. Pierce, A. Papadopoulos, S.B. Johnson, Phys. Rev. Lett. 68 (1992) 3678.
[8] M. Drees, M.M. Nojiri, Phys. Rev. D 45 (1992) 2482.
[9] A.V. Gladyshev, D.I. Kazakov, W. de Boer, G. Burkart, R. Ehret, Nucl. Phys. B 498 (1997) 3.
[10] J.A. Casas, J.R. Espinosa, M. Quirs, A. Riotto, Nucl. Phys. B 436 (1995) 3.
[11] R. Hempfling, A.H. Hoang, Phys. Lett. B 331 (1994) 99.
[12] R.-J. Zhang, Phys. Lett. B 447 (1999) 89.
[13] J.R. Espinosa, R.-J. Zhang, JHEP 0003 (2000) 026.
[14] M.S. Berger, Phys. Rev. D 41 (1990) 225.
[15] H.E. Haber, R. Hempfling, Phys. Rev. Lett. 66 (1991) 1815.
[16] M.A. Daz, H.E. Haber, Phys. Rev. D 46 (1992) 3086.
[17] P.H. Chankowski, S. Pokorski, J. Rosiek, Phys. Lett. B 274 (1992) 191; Nucl. Phys. B 423
(1994) 437.
[18] A. Yamada, Phys. Lett. B 263 (1991) 233; Z. Phys. C 61 (1994) 247.
[19] A. Brignole, Phys. Lett. B 281 (1992) 284.
38
[20] A. Dabelstein, Z. Phys. C 67 (1995) 495.

[21] S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Rev. D 58 (1998) 091701; Phys. Lett. B 440
(1998) 296; Eur. Phys. J. C 9 (1999) 343.
[22] D.M. Pierce, J.A. Bagger, K.T. Matchev, R.-J. Zhang, Nucl. Phys. B 491 (1997) 3.
[23] R. Barbieri, M. Frigeni, M. Caravaglios, Phys. Lett. B 258 (1991) 167.
[24] Y. Okada, M. Yamaguchi, T. Yanagida, Phys. Lett. B 262 (1991) 54.
[25] J.R. Espinosa, M. Quirs, Phys. Lett. B 266 (1991) 389.
[26] K. Sasaki, M. Carena, C.E.M. Wagner, Nucl. Phys. B 381 (1992) 66.
[27] H.E. Haber, R. Hempfling, Phys. Rev. D 48 (1993) 4280.
[28] M. Carena, J.R. Espinosa, M. Quirs, C.E.M. Wagner, Phys. Lett. B 355 (1995) 209.
[29] M. Carena, M. Quirs, C.E.M. Wagner, Nucl. Phys. B 461 (1996) 407.
[30] H.E. Haber, R. Hempfling, A.H. Hoang, Z. Phys. C 75 (1997) 539.
[31] M. Carena, H.E. Haber, S. Heinemeyer, W. Hollik, C.E.M. Wagner, G. Weiglein, Nucl. Phys.
B 580 (2000) 29.
[32] D. Comelli, J.R. Espinosa, Phys. Lett. B 388 (1996) 793.
[33] A. Sirlin, R. Zucchini, Nucl. Phys. B 266 (1986) 389.
[34] S. Heinemeyer, W. Hollik, G. Weiglein, Comp. Phys. Commun. 124 (2000) 76.
[35] S. Coleman, E. Weinberg, Phys. Rev. D 7 (1973) 1888.
[36] R. Jackiw, Phys. Rev. D 9 (1974) 1686.
[37] C. Ford, D.R.T. Jones, Phys. Lett. B 274 (1992) 409; Phys. Lett. B 285 (1992) 399 (Erratum).
[38] C. Ford, I. Jack, D.R.T. Jones, Nucl. Phys. B 387 (1992) 373; Nucl. Phys. B 504 (1997) 551
(Erratum).
[39] M. Caffo, H. Czyz, S. Laporta, E. Remiddi, Nuovo Cimento A 111 (1998) 365.
[40] A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 (1993) 123.
[41] A.I. Davydychev, V.A. Smirnov, J.B. Tausk, Nucl. Phys. B 410 (1993) 325.
[42] F.A. Berends, J.B. Tausk, Nucl. Phys. B 421 (1994) 456.
[43] W. Siegel, Phys. Lett. B 84 (1979) 19.
[44] D.M. Capper, D.R.T. Jones, P. van Nieuwenhuizen, Nucl. Phys. B 167 (1980) 479.
[45] I. Jack, D.R.T. Jones, S.P. Martin, M.T. Vaughn, Y. Yamada, Phys. Rev. D 50 (1994) 5481.
[46] K. Inoue, A. Kakuto, H. Komatsu, S. Takeshita, Prog. Theor. Phys. 68 (1982) 927;
Prog. Theor. Phys. 71 (1984) 413.
[47] S.P. Martin, M.T. Vaughn, Phys. Rev. D 50 (1994) 2282.
[48] Y. Yamada, Phys. Rev. D 50 (1994) 3537.
[49] I. Jack, D.R.T. Jones, Phys. Lett. B 333 (1994) 372.
[50] A. Donini, Nucl. Phys. B 467 (1996) 3.

Bs,d `+` in a two-Higgs-doublet model

Heather E. Logan , Ulrich Nierste 1
Theoretical Physics Department, Fermilab, Batavia, IL 60510-0500, USA 2
Received 17 April 2000; revised 9 May 2000; accepted 26 June 2000
Abstract
We compute the branching fractions of Bs,d `+ ` in the type-II two-Higgs-doublet model
with large tan . We find that the parameters of the neutral Higgs sector of the two-Higgs-doublet
model cancel in the result, so that the branching fractions depend only on the charged Higgs mass
and tan . For large values of tan and a charged Higgs mass above the bound from b s , we
find that the branching fractions can be enhanced by up to an order of magnitude or suppressed by
up to a factor of two compared to the Standard Model result. We point out that previous calculations
in the literature are gauge-dependent due to the omission of an important diagram, which gives the
dominant contribution in the t HooftFeynman gauge. We have analyzed in detail the region of the
(MH + , tan ) plane to be probed by searches for Bs + in Run II of the Tevatron. Since the
branching fraction increases like tan4 , this decay mode is complementary to b s and efficiently
probes the large tan region. For tan = 60, an integrated luminosity of 20 fb1 in an extended Run
II will probe charged Higgs masses up to 260 GeV, if the background to Bs + is small. For
the same value of tan , the LHC may be able to explore charged Higgs masses up to 1 TeV using
this decay. 2000 Elsevier Science B.V. All rights reserved.
PACS: 13.20.He; 12.60.Fr; 14.80.Cp
Keywords: B, leptonic decay; Higgs particle, multiplet; Higgs particle, mass
1. Introduction
The ongoing and forthcoming high-statistics B-physics experiments at BaBar, BELLE,
HERA-B, the Tevatron, and the LHC experiments ATLAS, CMS and LHCb [17] will
probe the flavor sector of the Standard Model (SM) with high precision. These experiments
may reveal physics beyond the SM, and a substantial theoretical effort is devoted to
calculating the observables that will be tested in various scenarios of new physics.
Corresponding author. E-mail: logan@fnal.gov
1 nierste@fnal.gov
2 Fermilab is operated by Universities Research Association Inc. under contract no. DE-AC02-76CH03000
with the US Department of Energy.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 1 7 - X
40
H.E. Logan, U. Nierste / Nuclear Physics B 586 (2000) 3955
A common feature of all popular weakly-coupled extensions of the SM is an enlarged

Higgs sector. In this paper we study the type-II two-Higgs-doublet model (2HDM),
which has the same particle content and tree-level Yukawa couplings as the Higgs sector
of the Minimal Supersymmetric Model (MSSM). If the ratio tan of the two Higgs
vacuum expectation values is large, the Yukawa coupling to b quarks is of order one and
large effects on B decays are possible. Direct searches for the lightest neutral MSSM
Higgs particle have begun to constrain the low tan region in the MSSM, because the
theoretically predicted mass range increases with tan . Hence observables allowing us
to study the complementary region of large tan are increasingly interesting. A further
theoretical motivation to study the large tan case is SO(10) grand unified theories [8
10]: they unify the top and bottom Yukawa couplings at high energies, corresponding to
tan of order 50.
The leptonic decay Bd 0 `+ ` , where d 0 = d or s and ` = e, or , is especially
well suited to the study of an enlarged Higgs sector with large tan . In the SM the decay
amplitude suffers from a helicity-suppression factor of m` /mb , which is absent in the
Higgs-mediated contribution. This helicity suppression factor numerically competes with
the suppression factor of (m` /MW ) tan stemming from the HiggsYukawa couplings to
the final state leptons, so that one expects the new contributions in the 2HDM to be similar
in size to those of the SM.
Earlier papers have already addressed the decay Bd 0 `+ ` in the 2HDM or the full
MSSM [1117]. Yet the presented results differ analytically and numerically substantially
from each other, so that we have decided to perform a new analysis.
This paper is organized as follows. In Section 2 we give a brief review of the type-II
2HDM. In Section 3 we review the SM calculation of the decay Bd 0 `+ ` and describe
our calculation of the relevant 2HDM diagrams. We finish Section 3 by combining the
results for the 2HDM diagrams and giving compact expressions for the branching fractions.
In Section 4 we compare our result with previous calculations. In Section 5 we present a
numerical analysis of our result and estimate the reach of future experiments. We present
our conclusions in Section 6. Finally, Appendix A contains a discussion of trilinear Higgs
couplings.
2. The two-Higgs-doublet model

In this paper we study the type-II 2HDM. The model is reviewed in detail in Ref. [18].
The 2HDM contains two complex SU(2) doublet scalar fields,
!
!
1+
2+
,
2 =
,
(1)
1 =
10
20
which acquire the vacuum expectation values (vevs) hi0 i = vi and break the electroweak
2 =
symmetry. The Higgs vevs v1 and v2 are constrained by the W boson mass, MW
1 2 2
1 2 2
2
2 g (v1 + v2 ) = 2 g vSM , where vSM = 174 GeV is the SM Higgs vev. Their ratio is
parameterized by tan = v2 /v1 .
41
Since in this paper we are not interested in CP-violating quantities, we assume CP

is conserved by the Higgs sector for simplicity. The mass eigenstates are then given as
follows. The charged Higgs states are
G+ = 1+ cos + 2+ sin ,
H + = 1+ sin + 2+ cos ,
(2)
and their Hermitian conjugates. The CP-odd states are

G0 = 10,i cos + 20,i sin ,
A0 = 10,i sin + 20,i cos ,
where we use the notation i0 = vi + 1 i0,r
2
i0 . The would-be Goldstone bosons G and

+ ii0,i for the real
G0 are eaten by the
(3)
and imaginary parts of
W and Z bosons. The
CP-even states mix by an angle giving two states,

H 0 = 10,r cos + 20,r sin ,
h0 = 10,r sin + 20,r cos .
(4)
In order to avoid large flavor-changing neutral Higgs interactions we require natural

flavor conservation [19,20]. We impose the discrete symmetry 1 1 , 2 2
(which is softly broken by dimension-two terms in the Higgs potential), with the SU(2)
singlet fermion fields transforming as d d, u u, e e. These transformation
rules define the type-II 2HDM and determine the Higgs-fermion Yukawa couplings. The
Yukawa terms in the Lagrangian are:
1 d Yu Q
c u Yl L
1 e + h.c.,
LYuk = Yd Q
2
(5)
where c = i2 . Down-type quarks and charged leptons (up-type quarks) are given
mass by their couplings to 1 (2 ).
Most of the Higgs couplings needed in our calculation are given in Ref. [18]. In addition
we must consider the trilinear H + H H couplings (H = h0 , H 0 ) which were first given in
Ref. [13] and are discussed in Appendix A.
3. Effective Hamiltonian for B `+ `
The decay Bd 0 `+ ` proceeds through loop diagrams and is of fourth order in the
weak coupling. In both the SM and 2HDM, the contributions with a top quark in the loop
are dominant, so that one may describe the decay at low energies of order mb by a local
coupling via the effective Hamiltonian,
0 ``
bd
EM
GF
t [CS QS + CP QP + CA QA ].
(6)
H=
2 2 sin2 W
Here GF is the Fermi constant, EM is the electromagnetic fine structure constant and W
is the Weinberg angle. The CKM elements are contained in t = Vtb Vt d 0 . The operators in
(6) are
L d 0 ``,
QS = mb bP
L d 0 `
5 `,
QP = mb bP
PL d 0 `
5 `,
QA = b
(7)
42
where PL = (1 5 )/2 is the left-handed projection operator. We have neglected the righthanded scalar quark operators because they give contributions only proportional to the d 0
` does not contribute to Bd 0 `+ ` , because it
mass. The vector leptonic operator `
gives zero when contracted with the Bd 0 momentum. Finally, no operators involving =
i [ , ]/2 contribute to Bd 0 `+ ` .
Because mb MW , mt , MH + , there are highly separated mass scales in the decay
Bd 0 `+ ` . Short-distance QCD corrections can therefore contain large logarithms like
log(mb /MW ), which must be summed to all orders in perturbation theory with the help of
renormalization group techniques. The calculation of the diagrams in the full high-energy
theory gives the initial condition for the Wilson coefficients at a high renormalization
scale on the order of the heavy masses in the loops. The hadronic matrix elements,
however, are calculated at a low scale = O(mb ) characteristic of the Bd 0 decay. The
renormalization group evolution down to this low scale requires the solution of the
renormalization group equations of the operators QA , QS and QP . Yet the operator QA has
zero anomalous dimension because it is a (V A) quark current, which is conserved in the
limit of vanishing quark masses. Similarly, the operators QS and QP have zero anomalous
dimension because the anomalous dimensions of the quark mass mb () and the (chiral)
L d 0 () add to zero. Hence the renormalization group evolution is trivial:
scalar current bP
if the bottom quark mass in QS and QP is normalized at a low scale = O(mb ), then no
large logarithms appear in the effective Hamiltonian or in the decay rate.
In the SM, the dominant contributions to this decay come from the W box and Z penguin
diagrams shown in Fig. 1. These diagrams were first calculated in [21] and give a nonnegligible contribution only to the Wilson coefficient CA . There is no contribution from
a photonic penguin because of the photons pure vector coupling to leptons. There are
Fig. 1. Dominant SM diagrams.
43
also contributions to the Wilson coefficient CS from a SM Higgs penguin [22] and to
the Wilson coefficient CP from the would-be neutral Goldstone boson penguin [23], but
2 relative to the
these contributions to the amplitude are suppressed by a factor of m2b /MW
dominant contributions and can be ignored.
The SM decay amplitude is then given by the Wilson coefficient [21]
CA = 2Y (xt ),
(8)
2 = 4.27 0.26 and m is evaluated in the MS scheme at = m ,

where xt = m2t (mt )/MW
t
t
s
Y1 (xt ),
giving mt (mt ) = 166 GeV. The function Y (xt ) is given by Y (xt ) = Y0 (xt ) + 4
where Y0 (xt ) gives the leading order (LO) contribution calculated in [21] and Y1 (xt )
incorporates the next-to-leading (NLO) QCD corrections and is given in [24,25]. The NLO
corrections increase Y (xt ) by about 3%, if mt is normalized at = mt . Numerically,

Y (xt ) = 0.997
mt (mt )
166 GeV
1.55
,
(9)
where we have parameterized the dependence on the running top quark mass in the MS
scheme.
We limit our consideration to the case of large tan , for which the 2HDM contributions
to this decay are significant. In the large tan limit, the Wilson coefficients CP and CS
receive sizeable contributions from the box diagram involving W and H + and the penguins
and fermion self-energy diagrams with neutral Higgs boson exchange shown in Fig. 2.
We have calculated the individual diagrams explicitly in a general R gauge, keeping
only the terms proportional to tan2 . Although each diagram that involves a W or G
boson is gauge-dependent, their sum is gauge-independent. For compactness, we give the
results of the individual diagrams below in the t HooftFeynman gauge.
There are no new contributions to CA in the 2HDM, which therefore retains its SM
value.
Fig. 2. Dominant diagrams in the 2HDM with large tan .
44
3.1. Box diagram

The box diagram in Fig. 2 gives the following contribution to CS and CP :
m`
tan2 B+ (xH + , xt ) .
CSbox = CPbox =
2
2MW
(10)
2 /M 2 and x was defined after (8) in terms of m (m ). Strictly speaking,

Here xH + = MH
t
t
t
+
W
in a LO calculation like ours, one is not sensitive to the renormalization scheme and we
could equally well use the top quark pole mass in xt . However, experience with NLO
calculations in the SM [24,25] shows that with the definition of mt adopted here, higherorder QCD corrections are small in leptonic decays. Finally the loop function B+ in (10)
reads

log x
y
log y
.
(11)
B+ (x, y) =
xy y 1 x 1
B+ (xH + , xt ) also contains the contribution from internal up and charm quarks with mc =
mu = 0 from the implementation of the GIM mechanism. The effect of a nonzero charm
quark mass is negligibly small.
3.2. Penguins
The penguin diagram with H + and W + in the loop (see Fig. 2) contributes
2

m`
sin cos2
peng,1
2
tan P+(xH + , xt )
=
+
,
CS
2
2
Mh20
MH
0
peng,1
CP
m`
1
tan2 P+(xH + , xt ) 2 .
2
MA
(12)
Here again all three quark flavors enter the result from the GIM mechanism, and the effect
of nonzero charm quark mass is negligible. By contrast, in the penguin diagram involving
H + and G+ in the loop only the internal top quark contribution is relevant, because the
coupling of G+ to quarks is proportional to either of the quark masses and we neglect ms .
This diagram gives
m`
peng,2
tan2 P+(xH + , xt )
=
CS
2
2
2
2
2
2
sin (MH + Mh0 ) cos2 (MH + MH 0 )
+
,
2
2
2
Mh20
MW
MH
MW
0
peng,2
CP
m`
1
tan2 P+ (xH + , xt ) 2
=
2
M 0
2 M2
MH
+
A0
The results in (12) and (13) involve the loop function

x log x y log y
y
.
P+ (x, y) =
xy x 1
y 1
2
MW

.
(13)
(14)
45
3.3. Self-energies
Before we write down the result from the diagrams with self-energies in the external
quark lines, we discuss how these contributions come into play. A treatment of flavorchanging self-energies has been proposed in [26,27] and analyzed in some detail in [28].
In these works flavor-changing self-energies have been discussed in a different context, the
renormalization of the W -boson coupling to quarks. In [2628] counterterms have been
chosen in such a way that the flavor-changing self-energies vanish if either of the involved
quarks is on-shell. For the down-type quarks these counterterms form two 3 3 matrices
in flavor space, one for the left-handed quark fields and one for the right-handed ones, and
similarly for the up-type quarks. It was argued in [27,28] that the anti-Hermitian parts of
the counterterm matrices for the left-handed fields can be absorbed into a renormalization
of the CKM matrix, and the Hermitian parts of the matrices can be interpreted as wave
function renormalization matrices ZijL and ZijR with i, j = d, s, b for our case of external
down-type quarks.
However, it has also been argued [29] that the on-shell scheme of [27,28] is not gauge
invariant. In addition we find that the approach of [2628] leads to an inconsistency in our
calculation. We cannot cancel the anti-Hermitian parts of the self-energies in the external
lines with the counterterms for the CKM matrix, because unlike in the case of the W
coupling renormalization there is no tree-level coupling of a neutral scalar or vector boson
0 to be renormalized. Hence we cannot absorb the anti-Hermitian parts of the flavorto bd
changing self-energy matrices into counterterms and they do contribute to our calculation.
The absorption of the Hermitian parts into wave function counterterm matrices is
optional, because the introduction of wave function counterterms only trivially shuffles
self-energy contributions into vertex counterterms. It is most straightforward then to
avoid the issue of counterterms altogether by simply calculating the fermion self-energy
diagrams as they are shown in Fig. 2. This calculation is straightforward because the
internal b quark line is off-shell and therefore it does not contribute to the 1-particle pole
of the s quark and needs not be truncated. It is crucial to note that one must start with
ms 6= mb in the diagrams with external self-energies in Fig. 2 to properly account for the
quark propagator 1/(mb ms ), and then take the limit mb , ms 0 (except in the tan enhanced Higgs couplings, of course) at the end. FCNC transitions become meaningless
for degenerate quark masses, and one obtains an incorrect result if one starts with ms =
mb and regulates the propagator pole with an off-shell momentum p with p 0. In this
respect the Higgs exchange diagrams in Fig. 2 differ from the situation with - or Zexchanges, where both methods give the same correct result. Further we note that for ms 6=
mb one must include flavor-changing self-energies in external lines with a factor of 1 rather
than of 1/2 as in the flavor-conserving case. This is due to the fact that flavor-conserving
self-energies come from the residue of the one-particle pole, while in our approach flavorchanging self-energies are part of the non-truncated Greens function. By close inspection
of the formulae in [28] one also recovers this factor of 1 rule from the expressions for
the wave function renormalization matrices derived in [28].
46
There are two fermion self-energy diagrams that contribute in the 2HDM, one with a
would-be Goldstone boson G+ and one with the physical charged Higgs H + . Their sum
is ultraviolet-finite. This is different from the SM case, in which the UV divergence of the
G+ diagram cancels with the UV divergence of a SM Higgs vertex diagram involving a
G+ and a top quark in the loop. As in the penguin diagrams involving H + and G+ in the
loop, only the internal top quark contributions to the self-energy are relevant here, because
the coupling of H + or G+ to quarks is proportional to either of the quark masses and we
neglect ms . The self-energy diagrams add the following term to the Wilson coefficients:
2

m`
sin cos2
tan2 (xH + 1) P+ (xH + , xt )
+
,
CSself =
2
2
Mh20
MH
0
CPself =
m`
1
tan2 (xH + 1) P+ (xH + , xt ) 2 .
2
M 0
(15)
3.4. 2HDM Wilson coefficients and branching ratios

Adding (10), (12), (13) and (15) we obtain the 2HDM Wilson coefficients in (6):
CS = CP =
m`
log r
,
tan2
2
r 1
2MW
(16)
2 /m2 (m ). Note that the dependence on the masses of the neutral

where r xH + /xt = MH
t
+
t
Higgs bosons from the penguin and fermion self-energy diagrams has dropped out in their
sum without invoking any relation between the mixing angle and the Higgs masses. The
result depends on only two of the 2HDM parameters: tan and MH + .
The two hadronic matrix elements involved are related by the field equation of motion

5 d 0 (x)Bd 0 PB 0 = ifB 0 P ei PBd 0 x ,
0b
B 0
d
d
d

5 d 0 (x)Bd 0 PB 0 = ifB 0
0b
d
d
m2B
d0
mb + md 0
i PB 0 x
d
The resulting decay amplitude is

2m`
GF EM mBd 0 fBd 0
.
C
C
C
`
m
``
+
m
`
|A| =
t
B
S
B
P
A
5
0
0

d
d
m Bd 0
2 4 sin2 W
(17)
(18)
Here fBd 0 is the Bd 0 decay constant, normalized according to f = 132 MeV. The
corresponding branching ratio is
v
u
2
2 2
m3B 0 Bd 0 fB2 0
u
G
d
d
EM
+
2
F
t1 4m`
|
|
B(Bd 0 ` ` ) =
t
8
m2B 0
32 2 sin4 W
d
"
!

2 #
2
4m`
2m`
2
2
1 2
CA
mBd 0 CS + mBd 0 CP
,
(19)
m Bd 0
mB 0
d
where Bd 0 is the Bd 0 lifetime.

Numerically, the branching fractions are given by
47
v

2
2 2 u
u

4m2`
m
|
f
|V
B
B
t
d
d
d
` t
B(Bd `+ ` ) = 3.0 107
1
1.54 ps 210 MeV

0.008 m2B
m2Bd
d
"
! 2

2
2 #
mBd tan2 log r
mBd tan2 log r 2
4m2`
Y (xt )
1 2
+
,
2
2
r 1
r 1
m Bd
8MW
8MW
v

2
2 2 u

u

4m2`
m
fBs
B s
|Vt s |
` t
1
B(Bs `+ ` ) = 1.1 105

1.54 ps 245 MeV
0.040 m2B
m2Bs
s
"
! 2
2
2 #

mBs tan2 log r
mBs tan2 log r 2
4m2`
Y (xt )
1 2
+
.
(20)
2
2
r 1
r 1
mB
8MW
8MW
s
It is a well known property of the 2HDM that there exists a limit in which the particles
H 0 , and H + become very heavy and decouple from processes at the electroweak
energy scale while h0 remains light and its couplings approach those of the SM Higgs
2
particle [30,31]. In the limit of large MH + , CP and CS fall as MH
+ . Thus the deviation of
2
the branching fractions from their SM prediction falls as MH + in the large MH + limit, and
the effects of the enlarged Higgs sector decouple.
Next we discuss the accuracy of the large tan approximation. Subleading terms in tan
could be enhanced by powers of mt /mb compared to our result in (16), as is the case for
the SM contribution. Such terms indeed occur, but they are suppressed by two powers of
cot compared to the SM terms in (16). Hence the formulae above are sufficient for all
purposes; e.g., if tan = 50 the terms subleading in tan give a correction only of O(2%).
If tan is between a few and 15 the 2HDM corrections are small and experimentally hard
to resolve, so that an error of order cot is tolerable as well.
A0 ,
4. Comparison with other calculations

4.1. The analyses of He et al. and of Savage
In the paper of He, Nguyen and Volkas [11], the decays B `+ ` , B K`+ ` and
b s`+ ` are analyzed in both type-I and type-II 2HDMs. In [11], the only diagrams
considered are the box diagram involving two charged Higgs bosons and the A0 penguin
involving H + and W + in the loop. Although the calculations of [11] are performed in
the t HooftFeynman gauge, the A0 penguin involving H + and G+ in the loop is not
considered. Similarly, in the paper of Savage [12], the decay B + is considered in
the 2HDM, with and without tree-level flavor-changing neutral Higgs couplings. Only the
contribution of the A0 penguin is considered. In both [11] and [12], several diagrams that
are important at large tan and required in order to obtain a gauge-independent result are
neglected.
48
4.2. The analysis of Skiba and Kalinowski

In the paper of Skiba and Kalinowski [13], the decay Bs + is analyzed in
the type-II 2HDM. In [13] additional penguin diagrams are considered that are not
proportional to tan2 , but rather contain one or no powers of tan . We have neglected
these contributions in our analysis, because they are not relevant for the interesting case of
large tan . These additional penguin diagrams can be important for small values of tan
and in regions of the parameter space where some of the Higgs quartic couplings are very
large resulting in large trilinear H + H H couplings (H = h0 , H 0 , A0 ).
Considering only terms proportional to tan2 , our results for the individual diagrams
agree with those of [13], with two important exceptions. First, the authors of [13]
incorrectly conclude that the box diagram involving H + and W + is negligible and
therefore neglect it. If we neglect the box diagram, we find that the sum of the remaining
contributions proportional to tan2 is gauge-dependent. In the t HooftFeynman gauge
employed in [13] the omitted diagram gives the dominant contribution, affecting the
numerical result substantially. Second, our result for the A0 penguin diagram involving
H + and G+ in the loop differs from that of [13] by a sign. Our sign is required for the
gauge-independence of CP .
4.3. The analyses of Huang et al. and Choudhury et al.
In the paper of Dai, Huang and Huang [14], the Wilson coefficients in (6) are calculated
in the type-II 2HDM at large tan . As in our calculation, only the diagrams proportional to
tan2 are considered. However, the authors of [14] consider only the penguin and fermion
self-energy diagrams with neutral Higgs boson exchange and neglect the box diagram with
a W and charged Higgs boson in the loop. Still, after leaving out the box diagram, our
results for CS and CP in the t HooftFeynman gauge do not agree with those of [14]. This
is partly due to a typographical error in [14], which is corrected in [1517]. There are two
remaining discrepancies. First, our result for the A0 penguin diagram involving H + and
W + in the loop differs from that of [14] by a sign. Second, in [14] a contribution from the
h0 and H 0 penguin diagrams with two H + bosons in the loop is included. This diagram is
included in [14] because it apparently receives a factor of tan from the trilinear H + H H
couplings (H = h0 , H 0 ). We argue in Appendix A that the trilinear couplings should not
be considered tan enhanced. Therefore we conclude that the penguin diagram with two
H + bosons in the loop should not be included in the O(tan2 ) calculation because it is of
subleading order in tan .
In [1517] the Wilson coefficients in (6) are calculated for supersymmetric models
with large tan . If the diagrams involving supersymmetric particles are neglected,
this calculation reduces to that for the type-II 2HDM with parameters constrained by
supersymmetric relations. Again, in [1517] only the diagrams with neutral Higgs boson
exchange are considered, and the box diagram with a charged Higgs boson and W
boson is not included. Leaving out the box diagram, our result for CS and CP in the
t HooftFeynman gauge does not agree with the non-SUSY part of that of [1517]. This
49
discrepancy arises because our result for the penguin diagrams involving H + and W + in
the loop differs from that of [1517] by a sign. Once SUSY relations are imposed on the
Higgs sector, it is clear that the penguin diagrams with two H + bosons in the loop are not
of order tan2 , and the authors of [1517] have omitted these diagrams, as we did.
A final critical remark concerns the treatment of the renormalization group in the paper
by Choudhury and Gaur [17]. They include an additional renormalization group factor to
account for the running of the Wilson coefficients. Yet these authors have overlooked that
the running of the (chiral) scalar quark current in QS and QP (see (7)) is compensated by
the running of the b-quark mass multiplying the currents as explained in Section 3. This
leads to an underestimate of the Wilson coefficients by roughly 23%.
In conclusion, the papers in [1113] and [1417] disagree with each other, and our
calculation does not agree with any of them. None of the results in [1117] passes the
test of gauge-independence and, in our opinion, each contains mistakes.
5. Phenomenology
As can be seen from the numerical coefficients in (20), B(Bs `+ ` ) is significantly
larger than the corresponding branching fraction for Bd . This is due primarily to the relative
sizes of |Vt s | and |Vt d |. As a result, even though the production rate of Bs is three times
smaller than that of Bd at the Tevatron, the bounds on the leptonic branching fractions
of Bs are much closer to the SM predictions than those of Bd [32]. For this reason we
concentrate on the decays of Bs . Because of the suppression of the branching fractions by
m2` /m2B , clearly the decay to pairs is the largest of the leptonic branching fractions in both
the SM and the 2HDM. However, this decay is very difficult to reconstruct experimentally
(due to the two missing neutrinos), and as a result the experimental limits on B decays to
pairs are very weak. Therefore, in our numerical analysis we focus on the decay Bs
+ , for which the experimental bound is the closest to the SM prediction. The best
experimental bound comes from CDF [32], where one candidate event for B +
has been reported; this event was consistent with the expected background and lay in the
overlapping part of the search windows for Bd and Bs . The corresponding 95% confidence
level upper bound on the Bs + branching fraction is [32]
B(Bs + ) < 2.6 106
(expt).
(21)
The SM prediction for the branching fraction is

B(Bs + ) = 4.1 109
(SM),
(22)
where we have taken the central values for all inputs in (20) and ignored the 2HDM
contributions as well as the errors in the hadronic parameters.
Because the 2HDM Wilson coefficients in (16) depend on only two of the parameters
of the 2HDM, tan and MH + , the behavior of the result in different parts of parameter
space is easy to understand. In regions of the parameter space with a large 2HDM contribution to B(Bd 0 `+ ` ) compared to the SM contribution, we may neglect Y (xt ) in
50
(20). Then the result is particularly simple: the branching fractions are proportional to
tan4 log2 r/(r 1)2 .
We can see from (20) and the value of Y (xt ) given in (9) that the interference between
the SM and 2HDM contributions to the branching fractions is destructive. The effect of
the destructive interference can clearly be seen in Fig. 3. In Fig. 3 we plot the predicted
value of B(Bs + ) in the 2HDM as a function of MH + , for various values of tan .
For comparison we show the constraint on MH + from b s [3337], obtained from the
current 95% confidence level experimental upper bound of B(b s ) < 4.5 104 from
the CLEO experiment. For very large tan and relatively light H + , the 2HDM contribution dominates and the branching fraction is significantly enhanced compared to its SM
value. As the 2HDM contribution becomes smaller due to increasing MH + or decreasing
tan , the branching fraction drops, eventually falling below the SM prediction due to the

destructive interference. If we ignore the kinematical factor of 1 4m2` /m2B 0 in front of
d
CS in (19) (which is a good approximation for Bd 0 + but not for Bd 0 + ) then
we can easily show that the branching fraction in the 2HDM crosses the SM value when
m2Bs tan2 log r
= Y (xt ),
2
r 1
8MW
(23)
and reaches a minimum of half the SM value when
Fig. 3. B(Bs + ) in the 2HDM as a function of MH + for tan = 100, 75, 60, 50 and 25. For
comparison we show the current experimental bound [32] and the SM prediction for the branching
fraction. The vertical line is the lower bound on MH + in the type-II 2HDM from b s [3337].
m2Bs tan2 log r

1
= Y (xt ).
2
8MW r 1 2
51
(24)
These correspond to tan2 log r/(r 1) = 1790 and 893, respectively. As a numerical example, taking tan = 60 and MH + = 175 GeV (500 GeV), we find B(Bs + ) =
1.8 108 (2.1 109 ).
In Fig. 4 we plot the regions of MH + and tan parameter space that will be probed as the
sensitivity to the decay Bs + improves at the Tevatron Run II. The contours shown
(from left to right) were chosen as follows. The current upper bound on B(Bs + )
is 2.6 106 from CDF [32] with about 100 pb1 of integrated luminosity. This bound
excludes a small region of parameter space with very high tan and very light H + , shown
by the solid line at the far left of Fig. 4. Such low H + masses are already excluded by the
constraint from b s [3337]. The rest of the contours in Fig. 4 show the regions that
we expect to be probed at the Tevatron Run II and extended Run II with various amounts of
integrated luminosity, assuming that the background for this process remains negligible. In
this case the sensitivity to the branching fraction should scale with the luminosity. If there
is background, however, then the sensitivity will scale only with the square root of the
luminosity. With 2 fb1 from each of the two detectors, the sensitivity should improve by
a factor of 40 over the current sensitivity, to 6.5 108, shown by the short dashes in Fig. 4.
For the values of tan that we consider, this sensitivity will still only probe values of MH +
already excluded by b s . We also show two contours for the expected sensitivity with
Fig. 4. Regions of MH + and tan parameter space probed by measurements of B(Bs + ).

Contours are chosen as described in the text. The vertical dashed line is the present lower bound on
MH + in the type-II 2HDM from b s [3337].
52
10 fb1 and 20 fb1 of integrated luminosity per detector (dotted and dot-dashed lines in
Fig. 4, respectively). These correspond to an extended Run II of the Tevatron. With 10 fb1
we expect the sensitivity to reach a branching fraction of 1.3 108 , allowing one to begin
to probe H + masses above the current bound from b s for tan > 54. With 20 fb1
we expect a reach of 6.5 109 , less than a factor of two above the predicted SM branching
fraction. This would allow one to probe H + masses above the current bound from b s
for tan > 47. In particular, for tan = 60, a non-observation of Bs + at this
sensitivity would rule out H + lighter than 260 GeV.
Looking farther into the future, the experiments at the CERN LHC expect to observe
the following numbers of signal (background) events for Bs + after three years
of running at low luminosity, assuming the SM branching fraction [38]: ATLAS: 27 (93);
CMS: 21 (3); and LHCb: 33 (10). Since the suppression of this branching fraction in the
2HDM is at most a factor of two, the LHC experiments will be able to observe this decay
for any configuration of the 2HDM at large tan . For, e.g., tan = 60 and MH + < 285
GeV, the branching fraction in the 2HDM is enhanced by 30% or more compared to the
SM. Similarly, for tan = 60 and 375 GeV < MH + < 1 TeV, the branching fraction is
suppressed by 30% or more compared to the SM. 3 In these regions, we expect the LHC to
be able to distinguish the 2HDM from the SM. In the region of large MH + , the dependence
on MH + is very weak; hence the LHC measurement will give powerful constraints on tan
in the large tan region.
We have made no attempt to simulate the experimental background for this decay in
order to obtain an accurate estimate of the reach of the Tevatron Run II. Neither have we
taken into account the theoretical uncertainty. We expect the largest theoretical uncertainty
to come from uncertainties in the input parameters, primarily the B meson decay constants
and CKM matrix elements in (20). These uncertainties will be reduced as the B physics
experiments progress and lattice calculations improve. Also QCD corrections to the 2HDM
contribution will arise at NLO and require the calculation of two-loop diagrams. In the SM,
the NLO corrections increase the decay amplitude by about 3%, and therefore increase the
branching fraction by about 6%. We expect the NLO corrections to the 2HDM contribution
to be of the same order, in which case our conclusions are not significantly modified.
In order to evaluate the usefulness of Bs + as a probe of the 2HDM, we must
compare it to other measurements that constrain the 2HDM in the large tan regime. As the
statistics of B physics experiments improve, the measurement of b s will improve as
well. If Bs + is to be a useful probe of the 2HDM, it must be sensitive to a range of
parameter space not already explored by b s at each integrated luminosity. Fortunately,
Bs + is complementary to b s because of the different dependence on tan .
While the 2HDM contributions to b s are independent of tan for tan larger than
a few, the 2HDM contributions to Bs + depend sensitively on tan . This makes
Bs + an especially sensitive probe of the 2HDM in the large tan regime, while
b s will remain more sensitive for moderate and small tan . Finally, a fit to the Z
decay data in the 2HDM [39] puts weak constraints on the H + mass for very large tan :
3 We do not consider charged Higgs masses larger than 1 TeV for naturalness reasons.
53
MH + > 40 GeV at 95% confidence level for tan = 100. The fit gives no constraint for
tan < 94.
6. Conclusions
In this paper we have analyzed the decays Bd 0 `+ ` in the type-II 2HDM with large
tan . Although these decays have been studied in a 2HDM before, the previous analyses
omitted the box diagram involving W and H + , which is the dominant contribution at large
tan in the t HooftFeynman gauge and is needed for gauge independence. We showed
that when all the contributions are properly included in the large tan limit, the resulting
expressions for the branching fractions are quite simple and depend only on tan and the
charged Higgs mass. These 2HDM contributions can enhance or suppress the branching
fractions by a significant amount compared to their SM values, providing tantalizing search
possibilities with the potential to probe large parts of the large tan parameter space of the
2HDM. We have focused in our numerical analysis on Bs + , for which the the
experimental sensitivity is best. We find that for reasonable values of tan up to 60 and
charged Higgs masses above the lower bound set by b s , B(Bs + ) can be
increased by up to a factor of five above its SM expectation or suppressed by up to a factor
of two, depending on the charged Higgs mass. Although very high statistics will be needed
to observe this decay, it promises to be an experimentally and theoretically clean probe of
new Higgs physics.
Acknowledgements
We are grateful to Jan Kalinowski and Witold Skiba for helpful discussions, and to
Jonathan Lewis for discussions on the reach of the Tevatron in Run II. We would also like
to thank the conveners of the B-Physics at Run II Workshop at Fermilab for facilitating
useful interaction among theorists and experimentalists. Finally we owe a debt of gratitude
to Piotr Chankowski and ucja Sawianowska for pointing out an error in the relative sign
between the SM- and 2HDM-induced contributions to the decay rate in an earlier version
of this manuscript, and for confirming our result for the 2HDM Wilson coefficients.
Appendix A. Trilinear Higgs couplings

The trilinear H + H H couplings (H = h0 , H 0 ) for the non-supersymmetric 2HDM
were first presented in [13]. These couplings depend strongly on the model considered.
For the most general CP-conserving 2HDM with natural flavor conservation, the H + H H
couplings are given by igQH /MW , where

2
sin( )(23 + 4 ) sin cos cos( + )5
Qh0 = vSM

+ 2 sin cos ( sin sin 1 + cos cos 2 ) ,
54

2
QH 0 = vSM
cos( )(23 + 4 ) sin cos sin( + )5

+ 2 sin cos (cos sin 1 + sin cos 2 ) .
(25)
The H + H A0 coupling is zero. Here vSM = 174 GeV is the SM Higgs vev, and the i are
the scalar quartic couplings of the Higgs potential given in [18]. To write these couplings
in terms of Higgs masses and mixing angles, one must make an assumption to eliminate
one of the independent i . In [13], formulae are presented for the two cases 1 = 2 and
5 = 6 . The formulae in (25) agree with [13] in these two cases.
At large tan , (25) reduces to

2
cos (23 + 4 ) 1 + O(cot ) ,
Qh0 ' vSM

2
sin (23 + 4 ) 1 + O(cot ) .
(26)
QH 0 ' vSM
These couplings are not enhanced at large tan .
Considering instead the case 1 = 2 and writing the trilinear couplings in terms of
Higgs masses and mixing angles, we find at large tan ,

1 2
2
Qh0 ' MH
0 tan cos sin 1 + O(cot ) ,
2

1 2
(27)
QH 0 ' Mh0 tan cos sin2 1 + O(cot ) .
2
Naively, one would conclude that these couplings are enhanced at large tan . This is
incorrect because the angle depends on tan . At large tan we have
tan 2 =

2(43 + 5 )
cot 1 + O(cot2 ) .
5 4(2 + 3 )
(28)
Thus for generic values of the i , sin cot , and the tan enhancement in (27) is
cancelled.
References
[1] D. Boutigny et al., BaBar technical design report, SLAC-R-0457.
[2] M.T. Cheng et al., Belle Collaboration, A study of CP violation in B meson decays: Technical
design report, BELLE-TDR-3-95.
[3] K. Pitts, for Fermilab D0 and CDF Collaborations, in: Proceedings 4th Workshop on Heavy
Quarks at Fixed Target (HQ 98), Batavia, USA, 1998.
[4] P. Krizan et al., HERA-B, an experiment to study CP violation at the HERA proton ring using
an internal target, Nucl. Instrum. Meth. A 351 (1994) 111.
[5] The ATLAS Collaboration, ATLAS Technical Proposal, CERN/LHCC/94-43.
[6] The CMS Collaboration, CMS Technical Proposal, CERN/LHCC/94-38.
[7] The LHCb Collaboration, LHCb Technical Proposal, CERN/LHCC/98-4.
[8] H. Fritzsch, P. Minkowski, Ann. Phys. 93 (1975) 193.
[9] M.S. Chanowitz, J. Ellis, M.K. Gaillard, Nucl. Phys. B 128 (1977) 506.
[10] H. Georgi, D.V. Nanopoulos, Nucl. Phys. B 155 (1979) 52.
[11] X.G. He, T.D. Nguyen, R.R. Volkas, Phys. Rev. D 38 (1988) 814.
[12] M.J. Savage, Phys. Lett. B 266 (1991) 135.
[13] W. Skiba, J. Kalinowski, Nucl. Phys. B 404 (1993) 3.
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
55
Y.-B. Dai, C.-S. Huang, H.-W. Huang, Phys. Lett. B 390 (1997) 257.
C.-S. Huang, Q.-S. Yan, Phys. Lett. B 442 (1998) 209.
C.-S. Huang, W. Liao, Q.-S. Yan, Phys. Rev. D 59 (1999) 011701.
S.R. Choudhury, N. Gaur, Phys. Lett. B 451 (1999) 86.
J.F. Gunion, H.E. Haber, G. Kane, S. Dawson, The Higgs Hunters Guide, Addison-Wesley,
Reading, MA, 1990; hep-ph/9302272 (Errata).
S.L. Glashow, S. Weinberg, Phys. Rev. D 15 (1977) 1958.
E.A. Paschos, Phys. Rev. D 15 (1977) 1966.
T. Inami, C.S. Lim, Prog. Theor. Phys. 65 (1981) 297; Prog. Theor. Phys. 65 (1981) 1772,
(Erratum).
B. Grzadkowski, P. Krawczyk, Z. Phys. C 18 (1983) 43.
P. Krawczyk, Z. Phys. C 44 (1989) 509.
G. Buchalla, A.J. Buras, Nucl. Phys. B 400 (1993) 225.
M. Misiak, J. Urban, Phys. Lett. B 451 (1999) 161.
K.I. Aoki, Z. Hioki, M. Konuma, R. Kawabe, T. Muta, Prog. Theor. Phys. Suppl. 73 (1982) 1.
W.J. Marciano, A. Sirlin, Nucl. Phys. B 93 (1975) 303.
A. Denner, T. Sack, Nucl. Phys. B 347 (1990) 203.
P. Gambino, P.A. Grassi, F. Madricardo, Phys. Lett. B 454 (1999) 98.
H.E. Haber, Y. Nir, Nucl. Phys. B 335 (1990) 363.
H.E. Haber, in: F. Csikor, G. Pocsik (Eds.), Budapest Electroweak 1994, World Scientific,
Singapore, 1995, p. 1, hep-ph/9501320.
F. Abe et al., CDF collaboration, Phys. Rev. D 57 (1998) 3811.
J.L. Hewett, Phys. Rev. Lett. 70 (1993) 1045.
V. Barger, M. Berger, R.J.N. Phillips, Phys. Rev. Lett. 70 (1993) 1368.
S. Bertolini, F. Borzumati, A. Masiero, Probing new physics in FCNC B decays and oscillations,
in: S. Stone (Ed.), B Decays, 1st edn., World Scientific, Singapore, 1992, pp. 458478, and 1994,
2nd edn., pp. 620643.
M. S. Alam et al., CLEO collaboration, CLEO CONF 98-17, talk presented at XXIX
International Conference on High Energy Physics, Vancouver, Canada, July 2329, 1998.
F.M. Borzumati, C. Greub, hep-ph/9810240.
P. Ball et al., hep-ph/0003238.
O. Lebedev, W. Loinaz, T. Takeuchi, hep-ph/0002106.

Quartic mass corrections to Rhad at O(s3)

K.G. Chetyrkin a , R.V. Harlander b, , J.H. Khn a
a Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe, D-76128 Karlsruhe, Germany
b Brookhaven National Laboratory, Upton, NY 11973, USA
Received 16 May 2000; accepted 20 June 2000
Abstract
The total cross section for the production of massive quarks in electronpositron annihilation can
be predicted in perturbative QCD. After expansion in m2 /s the quartic terms, i.e., those proportional
to m4 /s 2 , are calculated up to order s3 for vector and axial current induced rates. Predictions
relevant for charm, bottom and top quarks production are presented. The s3 corrections are shown
to be comparable to terms of order s and s2 . As a consequence, the predictions exhibit a sizeable
dependence on the renormalization scale. The stability of the prediction is improved and, at the same
time, the relative size of the large order terms decreases by replacing the running mass m() with
the scheme independent invariant one m
b. By combining these results with the prediction for massless
case and the quadratic mass terms the cross section for massive quark production at electronpositron
colliders is put under control in order s3 from the high energy region down to fairly low energies.
2000 Elsevier Science B.V. All rights reserved.
PACS: 12.38.-t; 12.38.Bx; 13.65.+i
The total cross section for hadron production in electronpositron annihilation is one of
the most fundamental observables in particle physics. For energies sufficiently far above
threshold it can be predicted by perturbative QCD, and it is well accessible experimentally
from threshold up to the highest energies of LEP and a future linear collider. It allows
for a precise determination of the strong coupling s and, once precision measurements at
different energies are available, for a test of its evolution dictated by the renormalization
group equation. In many cases the cms energy is far larger than the quark masses which
motivated the original calculations to be performed in the idealized case with the masses
set to zero from the start. In this limit, the results of O(s2 ) [13] and O(s3 ) [46] were
obtained more than two and nearly one decade ago, respectively (for a review see [7]).
However, in a number of interesting cases quark masses do play an important role [7].
For Z decays into bottom quarks the mass effects have to be included as a consequence of
Corresponding author. E-mail: rharlan@bnl.gov
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 3 - X
K.G. Chetyrkin et al. / Nuclear Physics B 586 (2000) 5672
57
the extremely precise measurements. Bottom quark production at lower energies is affected
from production threshold up to a few tens of GeV. Other cases of interest [8,9] are charm
production between roughly 5 and 10 GeV and, last but not least, top quark production at
a future linear collider [10].
In two-loop approximation the full mass dependence of the cross section has been
evaluated since long and, exploiting the optical theorem, both real and imaginary parts
are available [11]. In three-loop approximation (O(s2 )) the corresponding results were
obtained only recently. Two methods have been used for this purpose: the first one is
based on the evaluation of a large number of terms for the Taylor series of the polarization
function (q 2 ) at q 2 = 0 and an appropriately chosen analytic continuation [12]. The
second one is based on the application of the large momentum expansion (see [13,14] and
references therein), which provides an expansion of (q 2 ) in powers (m2 /q 2 )n , modulo
logarithms [15]. From the comparison between full result and expansion one learns that
the first few terms of the high energy expansion provide a remarkably good description of
the full result, from high energies down to values of 2m/ s = 0.65 0.75. The existence
of the four quark threshold at s = 4m and of a corresponding branching point for the
polarization function suggests that the high energy expansion diverges for s below 4m.
Nevertheless, numerical studies [15] as well as qualitative arguments demonstrate that the
sum of the first two or three terms of the expansion can be trusted even down to 3m or
even, with some optimism, 2.5m (where m stands for the pole mass).
These considerations pave the way to a prediction of R(s) including the quark mass
dependence to O(s3 ) along the following route: In addition to the massless result the m2 /s
terms of O(s3 ) have been calculated for the absorptive part nearly a decade ago [16]. They
were obtained by reconstructing the logarithmic s3 m2 /s terms of (q 2 ) from the full
three-loop O(s2 m2 /s) result of [17] with the help of the renormalization group equations.
These are sufficient to calculate the m2 /s terms of the imaginary part in the time-like
region. The result is cast into a particularly compact form once expressed in terms of the
running mass m(2 ), with the t Hooft scale set to s throughout [18], since this choice
eliminates all terms ln(s/m2 ). A generalization of this approach has been formulated for
the quartic mass terms in [19,20] and was originally adopted for the calculation of s2 m4 /s 2
terms. It is based on the operator product expansion and the usage of the renormalization
group equation to again construct the logarithmic terms of the polarization function. In
addition to the anomalous mass dimension and the -function the anomalous dimensions
of the operators of dimension four are required in appropriate order.
This method also allows to determine the s3 m4 /s 2 terms. The calculation can be
reduced to the evaluation of massless propagators and massive tadpole integrals, both at
most in three-loop approximation. Details of the calculation will be given elsewhere [21].
Compared to the case of the m2 /s terms an important difference arises: Even adopting the
MS definition (at scale 2 = s) for the quark mass, logarithmic terms ln(s/m2 ) remain
in order s2 and above. The power of these logarithms, however, is reduced by one. In fact,
these logs may be also summed up (see Refs. [19,22]).
To predict the R ratio, it is well justified to consider only one quark as massive and
to neglect the masses of the lighter quarks. The heavier quarks decouple (apart from the
58
tiny axial-vector singlet contribution see below) and can therefore be omitted in the
calculations. The massive quark will generically be denoted by Q in the following, while
q refers to all the lighter quarks.
In the energy region where quark mass effects are relevant, charm and bottom production
is solely induced by the electromagnetic vector current. Top quark production, however,
which sets in above 350 GeV, receives additional contributions from the axial current.
Both vector and axial-vector will be considered separately in the following. The prediction
for both of them is conveniently split up as follows:
X

(v)
2 (v)
vq2 rq + vQ
rQ + rsing ,
(1)
R (v) = 3
q
and similarly for the axial-vector part (v a). The sum runs over all massless quark
flavors. vq/Q (aq/Q ) is the coupling constant to the light/heavy quarks in the vector (axialvector) case. For low energies only the electromagnetic current contributes, v e and
a 0. For high energies, both the electromagnetic and neutral current pieces are relevant
(v/a)
represent
and have to be included with appropriate weights (see, e.g., [23]). rq and rQ
the non-singlet contributions arising from diagrams where the external currents are linked
by a common quark line. They originate from two different types of diagrams: For rq the
external current couples to massless quarks; the massive quark then only appears through
its coupling to virtual gluons. rq is the same for external vector and axial-vector currents.
(v/a)
On the other hand, rQ corresponds to diagrams where the external current couples to the
(v/a)
massive quark. rsing , finally, comprises massless and massive singlet contributions, where
either of the external currents is coupled to a separate closed quark line.
(v/a)
are written as series in m2Q /s, where s is the cms energy and mQ is
Both rq and rQ
the MS mass of the quark Q:
rq = r0 + rq,2 + rq,4 + ,
(v/a)
rQ
(v/a)
(v/a)
= r0 + rQ,2 + rQ,4 + .
(v/a)
(2)
r0 denotes the massless approximation, while rq,n and rQ,n are the mass terms of order
mnQ . If not stated otherwise, the renormalization scale 2 = s is adopted below.
Denoting nf the number of active flavors, the massless terms are given by (all the
following formul are valid up to O(s3 ) unless indicated otherwise)
2

s
s
365
11 2
+
113 + nf
+ 3
r0 = 1 +
24
12 3
3
1103
275
87029 121
s
2
3 +
5
+
288
8
4
6

1
262
25
19
7847 11
2 151
+ 2 +
3 5 + nf
2 3
+ nf
216
6
9
9
162 18
27
2
s
s
+
(1.98571 0.115295nf )
1+
3

s
6.63694 1.20013nf 0.00517836n2f .
(3)
+
The quadratic mass corrections, separated according to (1), read [16,24]:

m2Q s 3
32 8
3
80 + 603 + nf
rq,2 =
s
9
3

m2Q s 3
[7.87659 + 0.35007nf ],
2

m2Q
s
s
253 13
(v)
nf
12 +
rQ,2 =
s
2
3
3
490
5225
855
s
2 +
3
5
2442
+
2
3
6

2
466
1045
4846
2 125
+ 342
3 +
5 + nf
2
+ nf
27
27
27
54
3
2

m2Q
s
s
(126.5 4.33333nf )
12 +

3

s
1032.14 104.167nf + 1.21819n2f ,
+

2
2
mQ
s
s
8221
(a)
6 22 +
+ 572 + 1173
rQ,2 =
s
24

151
22 43
+ nf
12
3
121075
4613165
s
+ 13402 +
3 12705
864
36

656
72197 209
2
3 + 54 + 555
+ nf
162
2
3

26
13171 16
+ 2 + 3
+ n2f
1944
9
9
2

2
mQ
s
s
(108.140 + 4.48524nf )
6 22 +

3

s
2
409.247 + 73.3578nf 0.378270nf .
+
59
(4)
(5)
(6)
(7)
The quadratic mass terms in rq contribute in O(s3 ) and higher only [16]. Finally, we
present the quartic mass corrections up to O(s3 ):
2 2 2

mQ
s
13
lms 43
rq,4 =
s
3

3
50
9707
43
s
2
+ 2lms + 152 + 253 + 5 + lms
223
144
3
12

22
457 2
13 4
2 3 + lms + 3
+ nf
108 3
9
18 3
60
m2Q
s

2
3
2
(0.474894 lms )
2
4.59784 22.8619lms + 2lms

+ (0.196497 + 0.88052lms )nf ,
2
2
s
13
s
2977
+ 1622 + 1083 lms
6 22 +
12
2

8
143 1
+ lms 42 3
+ nf
18
3
3
3
64123
13285
1274461 12099
s
+
2 +
3
5
432
4
18
9

1672
1309
130009 574
2
223 + 13lms
2
3
+ lms
+ nf
6
648
3
9

440
199 4
2 2
5 + lms
+ 3 lms
+ 104 +
9
12
3
3

28
5
463 23
+ 2 + 3 lms
+ n2f
972
9
27
27
2 2
mQ
s
6 22
2

s
148.218 6.5lms + (1.84078 + 0.333333lms )nf
+
3

s
2
4776.95 244.612lms + 13lms
+

2
nf
+ 275.898 + 18.1861lms 0.666667lms

+(4.97396 0.185185lms )n2f ,

(v)
rQ,4 =
m2Q
2
2
s
75
s
1147
1622 2243 + lms
6 + 10 +
s
4
2

16
7
41
+ nf + 42 + 3 lms
6
3
3
3
20147
2225
222421 10881
s
2
3
5
+
48
4
6
3

4385
2
223 75lms
+lms
6

5002
1040
200923
+ 1962 +
3 104 +
5
+ nf
648
27
27

(a)
rQ,4 =
(8)
m2Q
(9)

2323 4
14 2
+ 3 + lms
+ lms
36
3
3

29
20
23
2995
2 3 + lms
+ n2f
972
9
27
27
2 2
mQ
s
6 + 10
2

s
248.99 + 37.5lms + (6.15737 2.33333lms )nf
+
3
s
2
4646.22 + 704.388lms 75lms
+

2
+ 264.151 62.9250lms + 4.66667lms
nf

+ (3.10948 + 0.851852lms )n2f .
61
(10)
The singlet contributions in the vector case are numerically small:

(v)
3

X 2 55
5
3
vq
vQ +
216 9
q

X 20 50
m2Q 2
vQ vQ +
vq + 3
+
s
9
3
q

3
2
X
s
vq
0.413180 vQ +
q

2 3
X m2Q 2
mQ
+ 17.8121vQ vQ +
vq
+O
.
s
s
q
rsing =
(11)
They do not receive m2 corrections [16].

The singlet contributions in the axial-vector case are exceptional in the sense that for
bottom quark production the top quark does not decouple, i.e., the contributions where the
top quark couples to one of the external currents are not suppressed by powers of 1/m2t .
The corresponding corrections up to O(m2b s3 ) have been computed in [2529]. They turn
out to be small, so we will neglect them in the following.
Considering the axial vector singlet contribution to top quark production, on the other
hand, one should take into account the full topbottom doublet, because the axial anomaly
cancels in this combination. However, this strategy induces completely massless final
states without any top quarks. At order s2 the expansion for the contributions from the
full topbottom doublet starts with the m6 terms [10]. The separate piece from massless
cuts is known in analytical form [25] and has to be subtracted if one wants to obtain the
contribution from the final states with top quarks. The complete result is given in Ref. [10].
Thus the singlet contribution to t t production is given in expanded form by
62

4 20
1 s 2 5 2
m2
2 2
2 +
+ 2 + 4lms lms
rsing =
4
4 3
s
3
3
3

2 2
16
2 3
13 2
4
1 2
m
lms
+ 2 3 + lms 5 + 2 lms
+
s
2
3
3
3
3
9

3
+ O s

m2
1 s 2
2
9.63289 + 4lms 0.666667lms
0.153377 +
=
4
s

2 2

m
2
3
0.222222lms
1.18565 2.80675lms 0.333333lms
+
s

+ O s3 ,
(12)
where the weak coupling at2 = ab2 = at ab = 1/4 has been pulled out for clarity. The
separate contribution from the massless cuts is not yet available in order s3 . In principle it
should be subtracted from the following result in order to arrive at the production rate for
top quarks:

1 s 3 m2t 2 380
1520
160
40
0
3 +
5 + nf
3
5
rsing =
4
s
3
3
9
9

1 s 3 m2t 2
(373.116 13.0918nf ).
(13)
4
s
The prime indicates that this expression still contains the contributions from purely
massless final states, as mentioned before. We explicitly denote the mass by mt here in
Table 1
Running coupling and masses at different scales. Matching for s is
performed at the values m(m) given in (14)
5.0
6.0
7.0
8.0
9.0
10.0
10.5
s (s)
(4)
mc ( s )
0.212
0.829
0.2
0.803
0.192
0.784
0.185
0.769
0.179
0.756
0.174
0.745
0.172
0.74
11.0
12.0
14.0
16.0
20.0
s (s)
(5)
mb ( s )
0.174
3.61
0.171
3.57
0.165
3.5
0.161
3.44
0.153
3.35
420.0
460.0
500.0
0.097
154
0.0961
153
0.0952
152
(4)
(5)
(6)
s (s)
(6)
mt ( s )
63
order to recall that, on the one hand, this result applies only to top production, and on the
other hand, the full (t, b) doublet has been taken into account.
Let us now investigate the numerical significance of mass effects for the charm and
bottom case. If not stated otherwise, we will adopt the following input data:
MZ = 91.19 GeV
s (MZ2 ) = 0.118,
mc (mc ) = 1.2 GeV,
mb (mb ) = 4.2 GeV,
mt (mt ) = 165 GeV.

(14)
The running of the quark masses and the coupling constant to the scale s is performed
with three-loop accuracy. Since we are working in the MS scheme, we have to take into
(n )
account the matching conditions for s f when going from nf to nf 1. We perform
the matching from nf = 5 to nf = 4 at mb (mb ) and from nf = 5 to nf = 6 at mt (mt ).
For illustration, some values for s (s), mc ( s ), mb ( s ), and mt ( s ) are displayed in

Table 1. The charm, bottom, and top masses are defined for nf = 4, 5, and 6, respectively.
As stated above, the approximation is expected to become unreliable in the threshold
region, that means below 5.56 GeV for charm, 1212.5 GeV for bottom, and 420
450 GeV for top contributions, which justifies the choice for the cms energies in the figures
presented below.
Fig. 1. The massless approximation r0 in three different energy ranges, relevant for (a) charm,
(b) bottom and (c) top production.
64
Fig. 2. Mass corrections to the non-singlet contribution of rq for c, b, and t production (1st, 2nd
and 3rd row, respectively), arising from diagrams where the external current couples to massless
quarks only. Left column: quadratic, starting to be non-zero in O(s3 ); right column: quartic mass
corrections, starting to be non-zero in O(s2 ).
65
Fig. 3. Quadratic mass corrections ( m2 ) to the non-singlet contribution of rQ for Q = c, b, t ,

arising from diagrams where the external current couples directly to the massive quark. Upper row:
vector current for c and b quarks; lower row: vector and axial currents for t quark.
In Figs. 14 we display separately the contributions for the massless case (Fig. 1) and
for the m2 and m4 terms, including successively higher orders in s . While Fig. 2 shows
the effects of the diagrams with light quarks coupling to the external current, Figs. 3 and
4 correspond to the case where the massive quark couples directly to the external current.
The axial contribution is presented for the top quark only.
The variation of the prediction with the renormalization scale is shown in Figs. 5 to 8.
As discussed already in earlier papers, the size of the higher order corrections decreases
quickly with increasing order in s , as far as the massless approximation is concerned
(Fig. 1). This is reflected in the stability of the result under variation of the renormalization
scale by a factor between 1/2 and 2 as displayed in Fig. 5. The m2 terms in rq (Figs. 2 (a),
(c), (e)) arise for the first time in order s3 . As a consequence this prediction exhibits a
strong dependence as displayed in Figs. 6 (a), (c), and (e). However, this term is typically
around 104 and thus irrelevant for all practical purposes.
The quartic terms in rq which are displayed in Figs. 2 (b), (d), and (f) contribute in
O(s2 ) and higher. Terms of order m4 s3 and m4 s2 are of comparable magnitude, not only
for the low energy case but even at the highest energy, whence the prediction for the m4
66
Fig. 4. Quartic mass corrections ( m4 ) to the non-singlet contribution of rQ for Q = c, b, t , arising

from diagrams where the external current couples directly to the massive quark. Upper row: vector
currents for c and b quarks; lower row: vector and axial currents for t quark.
terms in rq must be considered as uncertain within a factor two. This is reflected in the
strong dependence of the result shown in Figs. 6 (b), (d), and (e).
No reduction of this variation is observed when moving from the s2 to the s3
calculation. Nevertheless, the prediction for the total cross section is not seriously affected
by this instability since these terms are again of order 104 only and thus far below the
foreseeable experimental precision.
Table 2
(v)
rQ for s = 10.5 GeV
m0
m2
m4
s0
s1
s2
s3
1
0
0.0001477
0.05482
0.003264
0.00002969
0.004581
0.001628
0.00001245
0.001898
0.000519
0.00002016
1.058
1.063
1.063
67
Fig. 5. Variation of r0 with . s is fixed to values relevant

for the production
of (a) charm, (b) bottom,
and (c) top quarks. The abscissa ranges from = s/2 to = 2 s.
The dominant quartic mass terms are obviously expected from rQ , the part where the
massive quark is coupled to the external current. For comparison we also discuss the
quadratic terms which are known since long and are shown in Fig. 3. The vector current
induced piece is displayed in Fig. 3 for c, b, and t quarks, the axial piece is shown
for t quarks only. Terms of increasing order in s decrease in magnitude and apparent
convergence is observed. This welcome behavior is reflected by the improved stability of
higher orders under variations of (Fig. 7). The behavior of the quartic terms which are
the theme of this paper is more problematic (Fig. 4). The bulk of the result is given by the
Born approximation. The order s -, s2 -, and s3 -corrections are significantly smaller than
the leading terms. However, with increasing order they do not decrease but remain roughly
comparable in magnitude. This is reflected in the strong dependence displayed in Fig. 8.
Depending on the choice of , the relative size of the three corrections varies drastically.
Nevertheless, the higher orders are small compared to the m4 Born terms and, even
more important, their instability does not affect the stability of the overall prediction for R
resulting from the sum of the different terms. Accepting as an estimate of the uncertainty
of rQ,4 its variation with between s/2 and 2 s, rc,4 at 6 GeV varies by 0.0005, rb,4
at 14 GeV by 0.0016. For rt,4 the variation is negligible.
68
Fig. 6. Variation of rq,2 (left column) and rq,4 (right column) with . The first, second, and third
row correspond to charm, bottom, and top production, respectively.
These considerations demonstrate that a prediction for R has been obtained which is
valid in order s3 and includes mass terms in an expansion up to order m4 .
From a pragmatic point of view the smallness of the m4 terms, in particular of their QCD
corrections, allows to ignore their apparent instability. Nevertheless, one may also attempt
to arrive at a more stable result for the quartic terms by replacing the running mass by the
69
(v/a)
Fig. 7. Variation of rQ,2 with . First row: (a) charm, (b) bottom production for vector case;
second row: (c) vector and (d) axial-vector case for top production.
invariant one (b
m) through
Z
m (a)
,
m() = m
b exp da
a(a)
(15)
where a s (2 )/ . (a) and m (a) are the renormalization group functions governing
the running of the strong coupling constant and the quark mass:
2
d
a = a(a),
d2
d
m = mm (a).
d2
(16)
Their perturbative expansion is known up to O(s4 ) both for (a) [30] and m (a) [31,32].
The integral in (15) is solved perturbatively, and the resulting expression for R is reexpanded in s up to the third power. For the scale invariant mass we assume the values
bb = 15.3 GeV, and m
bt = 1076 GeV which corresponds to solving (15)
m
bc = 2.8 GeV, m
w.r.t. m
b for = m at three-loop order and using the numerical values of Eq. (14). The
(v)
result for the dependence of rQ,4 in this approach is shown in Fig. 9.
The variation of the O(s3 ) prediction with is reduced. For example, for s = 6 GeV
the O(s3 ) prediction in Fig. 8 varies between 0.10 102 and 0.20 102 with a
central value of 0.18 102 at = s, compared to a range from 0.13 102 to
70
(v/a)
Fig. 8. Variation of rQ,4 with . Conventions as in Fig. 7.
0.20102 with a central value of 0.19102 in Fig. 9. At the same time one observes
a reduction of the higher order terms compared to the Born approximation and the O(s )
prediction. For = s the results within the two approaches are quite close, which gives
additional confidence in the reliability of these numbers.
Summary
The total cross section for the production of massive quarks in electronpositron
annihilation can be predicted in perturbative QCD. After expansion in m2 /s the quartic
terms, i.e., those proportional to m4 /s 2 , were calculated up to order s3 for vector
and axial current induced rates. Predictions relevant for charm, bottom and top quark
production were presented. The s3 corrections were shown to be comparable to terms
of order s and s2 . As a consequence, the predictions exhibit a sizeable dependence on the
renormalization scale. Adopting instead of the MS scheme a framework where the running
b, the stability of the prediction is improved
mass m() is replaced by the invariant mass m
and, at the same time, the relative size of the large order terms decreases. Obviously, an
improved understanding of the origin of these large corrections would be highly desirable.
71
(v/a)
Fig. 9. Variation of rQ,4 with , using the invariant mass m

b. The conventions are the same as in
Figs. 7 and 8.
Combining these results with the massless prediction and the quadratic mass terms we
have demonstrated that the cross section for massive quark production at electronpositron
colliders is under control in order s3 from the high energy region down to fairly low
energies.
The results of this paper are available in M ATHEMATICA format at http://wwwttp.physik.uni-karlsruhe.de/Progdata/ttp00/ttp00-09/.
Acknowledgements
This work was supported by DFG-Forschergruppe Quantenfeldtheorie, Computeralgebra und Monte-Carlo-Simulation. R.H. acknowledges support by Landesgraduiertenfrderung at the University of Karlsruhe and by Deutsche Forschungsgemeinschaft.
References
[1] K.G. Chetyrkin, A.L. Kataev, F.V. Tkachov, Phys. Lett. B 85 (1979) 277.
[2] M. Dine, J. Sapirstein, Phys. Rev. Lett. 43 (1979) 668.
72
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
W. Celmaster, R.J. Gonsalves, Phys. Rev. Lett. 44 (1980) 560.

S.G. Gorishny, A.L. Kataev, S.A. Larin, Phys. Lett. B 259 (1991) 144.
L.R. Surguladze, M.A. Samuel, Phys. Rev. Lett. 66 (1991) 560; Phys. Rev. Lett. 66 (1991) 2416.
K.G. Chetyrkin, Phys. Lett. B 391 (1997) 402.
K.G. Chetyrkin, J.H. Khn, A. Kwiatkowski, Phys. Rep. 277 (1997) 189.
K.G. Chetyrkin, J.H. Khn, Phys. Lett. B 342 (1995) 356.
K.G. Chetyrkin, J.H. Khn, T. Teubner, Phys. Rev. D 56 (1997) 3011.
R. Harlander, M. Steinhauser, Eur. Phys. J. C 2 (1998) 151.
G. Kllen, A. Sabry, K. Dan. Videnk. Selsk. Mat.-Fys. Medd. 29 (17) (1955).
K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Phys. Lett. B 371 (1996) 93; Nucl. Phys. B 482
(1996) 213; Nucl. Phys. B 505 (1997) 40.
V.A. Smirnov, Mod. Phys. Lett. A 10 (1995) 1485.
V.A. Smirnov, Renormalization and Asymptotic Expansion, Birkhuser, Basel, 1991.
K.G. Chetyrkin, R. Harlander, J.H. Khn, M. Steinhauser, Nucl. Phys. B 503 (1997) 339.
S.G. Gorishny, A.L. Kataev, S.A. Larin, Nuovo Cimento A 92 (1986) 119.
G. t Hooft, Nucl. Phys. B 61 (1973) 455.
K.G. Chetyrkin, V.P. Spiridonov, Yad. Fiz. 47 (1988) 818; Sov. J. Nucl. Phys. 47 (1988) 522.
K.G. Chetyrkin, J.H. Khn, Nucl. Phys. B 432 (1994) 337.
K.G. Chetyrkin, R. Harlander, J.H. Khn, in preparation.
D.J. Broadhurst, S.C. Generalis, OUT-4102-12, Open University preprint, 1984.
J.H. Khn, P. Zerwas, in: A.J. Buras, M. Lindner (Eds.), Advanced Series on Directions in High
Energy Physics, Vol. 15, Heavy Flavours II, World Scientific, 1997, p. 493.
B.A. Kniehl, J.H. Khn, Nucl. Phys. B 329 (1990) 547; Phys. Lett. B 224 (1990) 229.
K.G. Chetyrkin, O.V. Tarasov, Phys. Lett. B 327 (1994) 114.
S.A. Larin, T. van Ritbergen, J.A.M. Vermaseren, Phys. Lett. B 320 (1994) 159.
K.G. Chetyrkin, A. Kwiatkowski, Phys. Lett. B 319 (1993) 307; Phys. Lett. B 305 (1993) 285.
T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379.
K.G. Chetyrkin, Phys. Lett. B 404 (1997) 161.
J.A.M. Vermaseren, S.A. Larin, T. van Ritbergen, Phys. Lett. B 404 (1997) 153.

Chiral models of weak scale supersymmetry

Jens Erler
Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104-6396, USA
Received 6 June 2000; accepted 3 July 2000
Abstract
I discuss supersymmetric extensions of the Standard Model containing an extra U (1)0 gauge
symmetry which provide a solution to the -problem and at the same time protect the proton
from decaying via dimension 4 operators. Moreover, all fields are protected by chirality and
supersymmetry from acquiring high scale masses. The additional requirements of anomaly
cancellation and gauge coupling unification imply the existence of extra matter multiplets and that
several fields participate in U (1)0 symmetry breaking simultaneously. While there are several studies
addressing subsets of these requirements, this work uncovers simultaneous solutions to all of them.
It is surprising and encouraging that extending the Minimal Supersymmetric Standard Model by
a simple U (1) factor solves its major drawbacks with respect to the non-supersymmetric Standard
Model, especially as current precision data seem to offer a hint to the existence of its corresponding
Z 0 boson. It is also remarkable that there are many solutions where the U (1)0 charges of the known
quarks and leptons are predicted to be identical to those E6 motivated Z 0 bosons which give the
best fit to the data. I discuss the solutions to these constraints including some phenomenological
implications. 2000 Elsevier Science B.V. All rights reserved.
PACS: 12.60.Cn; 12.60.Jv; 12.10.Kt; 12.10.Dm
Keywords: Supersymmetry; Proton decay
1. Introduction
Besides its conceptual elegance, the prime motivation to consider low energy supersymmetry (SUSY) (for reviews see [1,2]) is the stabilization of the electroweak scale under
radiative corrections. However, the Minimal Supersymmetric Standard Model (MSSM)
provides no explanation to the -problem [3], i.e., to why the scales of SUSY breaking
and of the supersymmetric bilinear Higgs -term,
hd hu ,
(1)
are of comparable magnitude. The MSSM should therefore be extended by at least some
sector solving the -problem. In this sense, the term MSSM is a misnomer.
E-mail address: erler@langacker.hep.upenn.edu (J. Erler).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 2 7 - 2
74
J. Erler / Nuclear Physics B 586 (2000) 7391
Moreover, in sharp contrast to the non-supersymmetric Standard Model (SM) where the
stability of the proton can be understood entirely by gauge invariance, the MSSM predicts
rapid proton decay via dimension 4 operators, unless an additional discrete symmetry, such
as R-parity, is imposed ad hoc. While this is a logical possibility, it severely compromises
the initial elegance and one prefers to save the successful features of the SM when
exploring its alternatives. It has also been argued [4] that those discrete symmetries 1 which
survive as remnants of some underlying string model, might either be unsuited to forbid (at
least) the dangerous dimension 4 operators, or arise only along some fine-tuned flat scalar
field direction. Indeed, baryon number conservation by the principles of gauge invariance
(and renormalizability) has been on the wish list of SUSY practitioners for almost two
decades [68].
It is well known that either of the problems addressed in the previous two paragraphs
can be solved by the introduction of an extra U (1)0 symmetry. Examples for either case are
reviewed in the next section using U (1) groups motivated by E6 Grand Unified Theories
(GUTs). Additional U (1)s have long been considered as very well motivated extensions
of the MSSM (for a recent review see [9]); they are predicted in most GUTs and appear
copiously in superstring theories. In this article, I address the question whether it is possible
to find models in which an extra U (1)0 solves both problems simultaneously.
If such an extension is free of gauge and gravitational anomalies, one can allow a grand
desert scenario between the electroweak scale and a more fundamental high mass scale.
This is desirable since our confidence in the MSSM is boosted by the observation of
approximate gauge coupling unification [1019] (as predicted by the simplest and most
economic GUT and string models) at a scale somewhat below the Planck scale. An extra
U (1)0 does not affect the renormalization group equations at the one-loop level, and it is
conceivable that the quality of unification even improves at two-loop precision. In order
to rigorously protect the electroweak scale from higher mass scales, I also demand that
the field content of the models be completely chiral before gauge and supersymmetry
breaking. I.e., all superfields should transform non-trivially with respect to the low-energy
SU(3) SU(2) U (1)Y U (1)0 gauge symmetry, and all bilinear (mass) terms must
be forbidden by gauge invariance. Note, that these requirements present a strong set of
constraints, and it is not clear a priori, whether any solutions exist. The key ingredient
necessary to arrive at a solution is that at least two MSSM singlet fields which are charged
under the U (1)0 have to acquire vacuum expectation values (VEVs). A similar approach
in the context of gauge mediated supersymmetry breaking, but without consideration of
gauge coupling unification, can be found in Refs. [20,21].
Electroweak precision data imply further hints and constraints. The predictions of the
SM are generally in very good agreement with experiments provided the Higgs boson
mass, MH , is smaller than about 250 GeV [22,23]. This is also predicted by the MSSM and
many extensions (MH . 150 GeV) [2532]. Moreover, the superpartners and extra Higgs
fields decouple from the precision observables over large parts of the MSSM parameter
space, so that the MSSM is likewise in agreement with observations [33,34]. Addition of
1 Continuous global symmetries are not expected to arise from string theory [5].
75
an extra Z 0 boson can even improve the global fit to all data [35], driven mostly by the
parity violation experiments in Cs [3642], 2 but the results in Tl [43,44], and from the
Z lineshape measurements at LEP 1 [45] also play a role. Note, that these fits suggest
a Z 0 boson mass MZ 0 . O (1 TeV) and continue to yield results on MH consistent with
SUSY [35].
Fits to the oblique parameters, S, T , and U [46] (or 1 , 2 , and 3 [47]), describing new
physics contributions to vector boson self-energies, yield results consistent with zero [23]
(for Higgs masses in the MSSM range). This implies restrictions on extra matter fields
which can contribute significantly to S and T , but if they are non-chiral (with respect
to the SM gauge group) and approximately mass degenerate their effects are rendered
small. If they arise in complete representations of SU(5), they do not affect gauge coupling
unification at the one-loop level, nor its scale. However, in the bottom-up approach of
this paper I do not assume an SU(5)-GUT symmetry. I will show that other matter
configurations can preserve unification, as well.
To reiterate, I will assume that there is an additional U (1)0 symmetry with a
mechanism built in to solve the SUSY -problem. I will classify the solutions to the
conditions of anomaly cancellation, gauge coupling unification, and chirality. To avoid
potential problems with fractionally charged states, I will also assume SU(5)-type charge
quantization. About half of the resulting models predict a sufficiently long proton lifetime.
In the next section, I review the situation for the most popular U (1)0 extensions arising
from SO(10) and E6 GUTs [48]. In Section 3, I introduce the anomaly conditions and show
that it is not possible to find chiral models if only one MSSM singlet field develops a VEV.
Section 4 shows that there are 33 solutions (family universal with respect to the ordinary
fermions) if there are two singlets acquiring VEVs in the course of U (1)0 breaking. Five
of these models involve a number of 5 + 5 representations and singlets of SU(5), but no
extra matter beyond that. Allowing incomplete representations of SU(5) and the weaker
requirements of gauge unification and absence of fractionally charged states yield the
remaining 28 solutions. I will also briefly address the prospects to derive these models
from the E8 E8 heterotic string theory. Some of the phenomenological implications of
the type of models obtained in Section 4 are discussed in Section 5. My conclusions are
presented in Section 6.
2. E6 inspired U (1)0 models

Consider, as an illustrative example, the case of an E6 gauge group which is free of
anomalies in all its representations. A fundamental 27 representation contains a pair of
SU(2) doublets, hd and hu , carrying the appropriate U (1)Y hypercharges to serve as the
MSSM Higgs doublets, i.e., allowing the Yukawa couplings,
qhd d,
qhu u,
lhd e+ ,
lhu .
2 Implications of a Z 0 for Q have been emphasized in Refs. [12, 3640].

W
(2)
76
and u denote the SU(2)

Here l and q are the lepton and quark doublet superfields 3 , e+ , d,
singlets, and refers to the right-handed neutrino superfield. If one identifies the U (1)0
with the U (1) defined by E6 SO(10) U (1) , then hd and hu have equal U (1)0
charges, excluding an elementary -term. A 27 also includes an MSSM singlet superfield,
S, which has the right U (1)0 charge to allow a trilinear Yukawa term,
Shd hu ,
(3)
in the superpotential. If the scalar component of S develops a VEV triggered by SUSY

breaking, it breaks the U (1)0 gauge symmetry and induces an effective -term, linking
these scales. Since it is assumed that electroweak symmetry breaking is triggered by SUSY
breaking, as well, all scales are related and the -problem is solved [49,50]. Moreover,
one predicts MZ 0 . O (1 TeV) if one wants to avoid excessive fine-tuning. The U (1)
symmetry also forbids all baryon number, B, and lepton number, L, violating terms in the
renormalizable superpotential of the MSSM,
lhu ,
lle+ ,
lq d,
ddu.
(4)
S Upon
However, each 27 also contains an extra pair of SU(2) singlet quarks, D and D.
U (1)0 breaking they are expected to receive TeV scale masses through terms of the form, 4
S
SD D,
(5)
but they also allow new B and L violating (though BL conserving) terms,
qqD,
Su.
dD
(6)
Together with the B and L conserving operators,

S
lq D,
e+ D u,
D d,
(7)
S quarks.
dimension 4 proton decay is reintroduced through the exchange of D and D
Clearly, one can avoid this problem by restricting oneself to SO(10) and identifying the
U (1)0 with the U (1) appearing in SO(10) SU(5) U (1) , which likewise forbids the
terms in Eq. (4). The ordinary fermions and the complete anomaly free 16 representations
(avoiding the extra singlet quarks), and one has to add a pair of Higgs doublets which by
anomaly cancellation are required to have equal and opposite U (1)0 charges. As a result,
a primordial -term is allowed, and any term of the form (3) can only include a U (1)0
neutral singlet field, S. Therefore, both aspects of the U (1)0 solution to the -problem
(exclusion of an elementary -term and generation of an effective -term at the TeV scale)
are lost, while there is no problem with too rapid proton decay. This is the reversed situation
compared to the U (1) case.
One possibility is to ignore the -problem. Models containing an anomaly free U (1)0
symmetry designed to stabilize the proton have been discussed by various authors [4,6
8,51,52]. Conversely, non-anomalous U (1)0 s addressing the -problem without reference
3 I denote the MSSM multiplets by lower case letters, while additional fields will be capitalized.
4 Terms of this type can also trigger radiative U (1)0 symmetry breaking in analogy to electroweak symmetry
breaking in the MSSM, provided the coupling strength is comparable to the top quark Yukawa coupling [50].
77
Table 1
U (1)0 charges of a 27 representation for an arbitrary E6 boson [35]. The upper
half of the table with b = 0 corresponds to a family in SO(10) GUT. The doublets
on the left-hand side are the fields l, q, hd , and hu , respectively. The special cases
= (0, 0), (0, 4/3), and (5/3, 0) correspond, respectively, to the U (1) ,
(a,
b)
the U (1) , and the U (1)Y (up to normalization)

e

u
d
N
E
+1/3
+
E
S
N
e+
+5/3
+1/3
+a
a
+2b
+b
+b
u
d
+1/3
1
+a
a
+b
+2/3
+a
D
S
D
2/3
+2/3
2b
b
2/3
2b
+3b
to proton decay are discussed in Ref. [53,54]. An interesting solution to both problems
S quarks relative to the
is to flip the signs of the hypercharge assignments of the D and D
E6 case [55]. This does not alter the SM anomaly conditions, and forbids the dangerous
terms in Eq. (6). However, it implies the existence of stable fractionally charged baryons
with TeV scale masses. Such states have not been observed, yet, and lead to serious
cosmological problems. 5 If the reheating temperature after inflation is of the order of the
mass of these exotic particles or above, their hadronic interactions give rise to rather large
number densities of these stable hadrons [56,57], orders of magnitude above experimental
bounds [58,59]. Lowering the reheating temperature does not solve the problem either,
since successful baryogenesis probably requires a reheating temperature after inflation at
least of the order of the critical temperature of the electroweak phase transition of about
100 GeV (for a recent review and references see [60]). Then a significant number density
of TeV scale particles can still be produced by the high-energy tail of thermal particle
distributions, or during reheating itself [61]. The present work is therefore devoted to
models without fractionally charged states, i.e., SU(5) charge quantization is imposed,
C2 C3
C3
= QY +
+
Z,
(8)
3
2
3
where Q is electric charge, QY is hypercharge, and where C2 is an even (odd) integer for
vector (spinor) representations of SU(2). Similarly, C3 is the triality class of SU(3) (e.g.,
1 for quarks and antiquarks).
All of the E6 inspired models discussed above have another unwanted feature in the
context of gauge coupling unification, which requires the addition of an extra pair of
SU(2) doublets. By virtue of anomaly cancellation these must be non-chiral and one
would therefore need a mechanism preventing them from receiving high scale tree-level
Q+
5 I am indebted to Michael Plmacher for discussions on this point.
78
masses. This is not a problem in the context of string models which predict the absence
of fundamental mass parameters in the low energy effective Lagrangian by conformal
invariance. However, as these models stand, the extra doublets would then be massless.
One could couple them to an overall singlet receiving a VEV, but in general this is not
protected to be of electroweak size. Alternatively, one can assume that the extra doublets
develop VEVs, as well, but it seems difficult to arrange that all chargino and neutralino
masses are compatible with the experimental lower mass limits as some masses would be
predicted to be strictly of order the Z boson mass, . MZ . This paper is therefore devoted
to stabilize electroweak scale physics by the principles of gauge and supersymmetry, while
assuring that all fields receive large enough masses after symmetry breaking.
An alternative way to remove the dangerous terms in Eq. (6) would be to choose
a particular linear combination of U (1) , U (1) , and U (1)Y . Table 1 shows that the
condition,
a + 2b +
5
= 0,
3
(9)
so that D and d can be removed from the spectrum. Baryon

allows a Dirac mass term, D d,
number is now conserved because the trilinear terms in Eq. (4) are still forbidden, and
S but since
effects from the ones in Eq. (6) decouple. The role of d is now played by D,
S
lq D is the only type of operator involving it, l must act as a Higgs superfield, and trade its
role with hd . However, Eq. (9) also implies that l and hu have equal and opposite U (1)0
charges and furthermore allows to decouple from the spectrum. Hence, we recover the
SO(10) case (with replaced by S) and the -problem which comes with it. Eq. (9) defines
therefore an alternative set of SO(10) models within E6 , in analogy with the alternative
left-right model [6264].
3. Anomaly cancellation in the presence of an extra U (1)0
Consider the U (1)0 charge assignment in Table 2. Indicated are three generations of
ordinary matter fields, one pair of Higgs doublets, hd and hu , appropriate to break
electroweak symmetry, one singlet Higgs, S, for U (1)0 symmetry breaking, and further
singlet fields 6 , Ti , which are (initially) assumed not to receive VEVs. Shown are also
examples of extra matter fields, where E are charged lepton singlets, L are lepton
doublets, U and D are up-type and down-type quark singlets, Q are quark doublets, and
S U
S, D,
S and Q
S are their mirror partners. W and G are neutral fields under
where E + , L,
the U (1)Y transforming in the adjoint representations of SU(2) and SU(3), respectively.
These are the fields which will play a role in the models surviving the analysis in Section 4,
but initially any field respecting Eq. (8) is allowed. I assume family universality for the
ordinary fermions and also the exotic fields, i.e., the new fields have identical U (1)0 charges
in the cases of multiplicities greater than one, but this assumption is not crucial and can be
relaxed easily.
6 I will refer to a singlet as a right-handed neutrino, , if it admits the Yukawa coupling in Eq. (2).
79
Table 2
U (1)Y and U (1)0 charge assignments for the MSSM matter fields including righthanded neutrinos and the two Higgs doublets (upper half), a number of singlets S
and Ti , and extra exotic fields. Three generations of ordinary fermions are assumed
and generation indices suppressed. Table 1 corresponds to a = 1 (normalization),
c1 + c2 = s = (a + 3b), and d1 = 2b
QY
Q0
QY
Q0
1/2
e+
0
+1
(a + c2 )
(a + c1 )
+1/6
u
d
2/3
+1/3
(b + c2 )
(b + c1 )
hd
1/2
c1
hu
+1/2
c2
D
L
Q
U
E
W
1/3
1/2
+1/6
+2/3
1
0
d1
d2
d3
d4
d5
s/2
S
D
S
L
S
Q
S
U
E+
G
+1/3
+1/2
1/6
2/3
+1
0
(s + d1 )
(s + d2 )
(s + d3 )
(s + d4 )
(s + d5 )
s/2
Ti
ti
e

u
d
As usual, all ordinary charged leptons and quarks are expected to become massive
through the Yukawa couplings 7 in Eq. (2). On the other hand, the extra fields, such as
D and E, are assumed to get vector-like (with respect to the SM group) masses induced
by trilinear couplings with S and their respective SU(5) mirror partners, as for example
in Eq. (5). This way their contributions to oblique parameters decouple. If an irreducible
representation is real, r = r , and has Q0 = s/2, it can acquire a supersymmetric Majorana
mass through a trilinear coupling to S. Examples are the fields W and G, and in this case
it suffices to add a single copy. Finally, I demand,
s = (c1 + c2 ) 6= 0,
(10)
to guarantee a solution to both aspects of the -problem.

Since the exotic matter is vector-like with respect to the SM gauge group it does not spoil
the anomaly cancellation present there. Triangle diagrams with two SM (i.e., either SU(3),
or SU(2), or U (1)Y ) and one U (1)0 gauge fields as external legs yield three conditions
which have to be satisfied. At first sight one would expect that there are many solutions
given the many free parameters in Table 2, but as it will turn out there are none. For
example, if the exotic matter consists of k5 copies of 5 + 5 representations of SU(5), the
mixed SU(3)/U (1)0 anomaly condition reads,
(k5 3)(c1 + c2 ) = 0,
(11)
7 While m = 0 is strongly disfavored, it is not firmly ruled out. I will comment on this case later in this section.
u
80
and together with Eq. (10) fixes k5 = 3. Eq. (11) is independent of the di , and so are the
mixed SU(2)/U (1)0 and U (1)2Y /U (1)0 conditions, which are, respectively,
(k5 + 1)(c1 + c2 ) + 3(a + 3b) = 0,

10
3 k5 14 (c1 + c2 ) 6(a + 3b) = 0.
(12)
(13)
This is solved only by the SO(10)-type relation, a = 3b, and for equal and opposite Higgs
charges, c1 = c2 .
This conclusion cannot be avoided even if we generalize to arbitrary extra matter
fields of the form N + S
N, where N is any (in general reducible) representation of SU(5).
Eqs. (11)(13) still apply, but k5 is now defined as the index of N,
X nr dim r
Cr ,
(14)
k5 = 2
dim SU(5)
r
where the sum is over the irreducible representations, r N. nr are the multiplicities (nr =
1/2 for Majorana types) and Cr is the second-order Casimir invariant of representation r.
One can also allow extra matter fields in incomplete representations of SU(5). However,
preservation of gauge coupling unification enforces the differences of the one-loop
-function coefficients to be the same as in the MSSM. 8 This leaves only one adjustable
parameter, the index k5 , and the anomaly conditions remain unchanged.
The conclusion, c1 +c2 = 0, is also independent of the assumption of family universality.
If, for example, the U (1)0 charges of the third generation were different from the two lighter
ones and associated with parameters a 0 and b0 (cf. Table 2), then this would merely result
in the replacement,
3(a + 3b) 2a + a 0 + 3(2b + b0 ),
(15)
in Eqs. (12) and (13).

A more interesting case [65] can be constructed from the one in the previous paragraph
and conversely
by replacing (b + c2 ) by (b0 + c2 ) for the quark singlets u and c,
for t. This results in a zero eigenvalue for the up-type quark mass matrix, and would
either require mu = 0, or that a non-vanishing mu can be generated radiatively, nonperturbatively, or by some other mechanism. Family non-universal U (1)0 charges can
also induce significant flavor changing neutral current effects [66]. If one allows for this
possibility, Eqs. (11)(13) are replaced by,
2a + a 0 + 9b
0 +1
k5 3
(16)
k5 + 1
b b0
= 0.
+1 3
2
34
3
10
3 k5
14
c1 + c2
However, the determinant of this matrix is non-vanishing (= 4) independently of k5 , and

there is only the trivial solution implying b = b 0 , i.e., nothing new.
The setup discussed in this section does not seem to provide solutions to the -problem.
It should be stressed, however, that this conclusion depends crucially on the additional
8 I will refer to such configurations as quasi-complete SU(5) representations.
81
assumptions of gauge coupling unification and chirality. For example, the model in
Ref. [55] is chiral, but fails to produce unification; alternatively, if unification is enforced by
adding an extra pair of doublets it turns non-chiral. Another example can be constructed
with the assignments in Table 2. Suppose the exotic matter consists of the fields (Q +
S + (U + U
S). Eq. (11) is then satisfied (k5 = 3), and Eq. (13) becomes equivalent to
Q)
Eq. (12), which are solved by, 4(c1 + c2 ) = 3(a + 3b). The quadratic anomaly condition
from triangle graphs with one hypercharge and two U (1)0 gauge bosons in the external
legs depends on d3 and d4 and can easily be solved. Extra singlets, Ti , can be added to
satisfy the cubic (pure U (1)0 ) and trace (mixed gravitational/U (1)0) anomaly conditions.
This model is chiral but the gauge couplings do not unify. Unification can be arranged
S representation of SU(5), but it would have
by adding an E pair to complete a 10 + 10
to carry equal and opposite U (1)0 charges, spoiling the chirality of the model. Thus, this
model mirrors the situation in Ref. [55] except that it has no fractionally charged baryons.
4. Two Higgs singlet solutions

Suppose now that besides S another MSSM singlet field, T = T0 , with Q0 = t = t0 ,
S N+S
develops a VEV. A subset of the exotic matter fields, M + M
N, could then acquire
TeV scale masses through trilinear couplings to T , provided the U (1)0 charge assignments
S are changed 9 from di and (s + di ) to ei and (t + ei ),
of the fields in M and M
respectively. The anomaly conditions in Eqs. (11)(13) now change to,
k3
3(a + 3b)
0
k5 k3 3
(17)
k5 k2 + 1
k2
s
1
= 0,
+2
10
3 (k5
k1 ) 14
10
3 k1
P
where k3 is the SU(3) index of M, k3 dim SU(3) = 2 r mr dim r Cr . Similarly, k2 is the
SU(2) index of M, and k1 = 6/5 TrM Q2Y . The SU(5) charge quantization condition (8)
implies,
1
1
5
k1 k2 k3 k0 Z.
(18)
6
2
3
Notice, that k3 , k2 , and k0 are integers, and that by definition, 0 6 kj 6 k5 for 1 6 j 6 3,
which implies the bound |k0 | 6 5k5/6. If the determinant of the matrix in Eq. (17) is nonvanishing, there is only the unwanted solution, implying c1 + c2 = 0. The matrix becomes
singular for,
k3 = (k5 3)(k0 + k2 k3 ).
(19)
Using the definitions,

t
(a + 3b)
,
,
s
s
the solutions can be classified as follows:
(20)
9 In general, this explicitly violates the assumption of universality w.r.t. the exotic fields which was made in
Section 3.
82
The case k3 = k0 + k2 = 0 implies 5k1 = 3k2 , and therefore k0 = k1 = k2 = =

= 0 independently of k5 , which will be rejected because it corresponds essentially
to the non-chiral situation in Section 3. This is the only possibility for k5 = 1.
If k5 > 5, Eq. (19) implies that k3 is an integer multiple of k5 3, and kj 6 k5 requires
k3 = k5 3 and k0 + k2 = k5 2. One has 3 = (k5 k2 + 1) and = 0, i.e.,
S representation
these models are also non-chiral since the fields within the M + M
are not tied to U (1)0 symmetry breaking. Such solutions exist for k5 = 3 (e.g., the
three generation E6 model with an extra pair of doublets) and k5 = 4, as well, but
they will be discarded.
k5 = 2 implies k0 +k2 = 0, k3 > k2 , 3 = 3 +k2/k3 and = 1 +1/k3 . There are three
solutions, but the case (k3 , k2 , k1 , k0 ) = (2, 1, 1/5, 1) cannot be made consistent
with the SU(5) quantization condition.
k5 = 3 implies k3 = 0, 3 = 3 + k0 /(k0 + k2 ) and = 1 1/(k0 + k2 ). There are
eight solutions of this type.
k5 = 4 implies k3 = 2, k0 + k2 = 4, = 5/3 k2 /6 and = 1/2. There are two
solutions, but the case (k3 , k2 , k1 , k0 ) = (2, 3, 19/5, 1) cannot be made consistent with
the SU(5) quantization condition.
Thus, there are an infinite number of non-chiral solutions, but only 11 chiral ones, which
are collected in Table 3. All of them predict = t/s > 0, implying that there are no D-flat
Table 3
Classification of solutions to the conditions of anomaly cancellation, chirality, SU(5) charge
quantization, and gauge coupling unification. It is assumed that two fields participate in U (1)0
gauge symmetry breaking. For comparison, the non-chiral three generation E6 solution with an
extra neutral (t = 0) Higgs singlet giving mass to an extra vector-like pair of doublets (k2 = 1) is
also shown. 1 and 3 are shown for 5 + 5 representations only, in which case they are defined
in Eqs. (21) and (26), respectively. The column before these specifies the fields receiving masses
S fields can be replaced
through trilinear couplings to T (cf. Table 4 below). In addition, 2(L + L)
by the combination W + E , where applicable
Solution
k5
k3
k2
k1
k0
I
II
III
IV
V
2
2
3
3
4
1
2
0
0
2
0
0
2
3
4
2/5
4/5
6/5
9/5
16/5
0
0
0
0
0
1
1
1
1
1
2
3/2
1/2
2/3
1/2
E6
3/5
VI
VII
VIII
IX
X
XI
3
3
3
3
3
3
0
0
0
0
0
0
3
1
2
3
0
1
3/5
9/5
12/5
3
12/5
3
1
1
1
1
2
2
5/6
7/6
10/9
13/12
4/3
11/9
1/2
1/2
2/3
3/4
1/2
2/3
Yukawa couplings to T
3max
S
(D + D)
S
2(D + D)
2(L + S
L)
3(L + S
L)
S
all but 2(D + D)
0
1/2
3/2
4/3
3/2
3/2
23/16
3/2
31/27
9/16
(L + S
L)
W + (L + S
L)
S + E
(L + L)
S + E
2(L + L)
S + E
3(L + L)
2E
S
(L + L) + 2E
83
Table 4
Quasi-complete SU(5) representations (preserving gauge coupling unification) which can be used to
realize the solutions in Table 3. There are other configurations with k5 = 4, but solution V requires
S representation of SU(5) (k5 = 3) cannot
S quarks. Similarly, the 10 + 10
at least two pairs of D + D
be used for any of the solutions in Table 3 and has been omitted
Representation
k5
S
D+D
S
L+L
S
Q+Q
S
U +U
Solution(s)
1
2
3
2
2
2
1
2
2
1
1
I
I, II
I, II
4
5
6
7
8
9
3
3
3
3
3
3
1
2
3
3
1
3
1
1
3
2
1
2
1
1
1
1
III, IV, VIXI

III, IV, VIIIX
X
VI
III, IV, VI, VII
III, IV
10
11
12
13
14
15
16
4
4
4
4
4
4
4
2
2
3
3
4
4
4
2
4
1
1
2
1
2
1
2
1
2
1
V
V
V
V
V
V
V
scalar field directions. Furthermore, all of them predict k5 6 4, which guarantees that the
gauge couplings remain perturbative up to the unification scale, while k5 = 5 would yield
non-perturbative unification (for a discussion see [67]).
In general, each solution in Table 3 can be realized with various exotic matter contents.
Observe that the configurations of ki in the table restrict the possible matter content to
the fields displayed in Table 2. These fields can be combined into quasi-complete SU(5)
representations shown in Table 4.
In the remainder of this section and in the next section I will focus on exotic matter in
complete 5 + 5 representations of SU(5), unless noted otherwise. In this case one can write
the sum over the other singlet charges, which is fixed by the trace condition, Tr Q0 = 0, as
1X
ti = 5k5 + ( 1)(3k3 + 2k2 1) 12.
(21)
1
s
i=1
Interestingly, when 3 right-handed neutrinos are included, 1 is completely determined

within each model. For 5 + 5 representations one has k0 = 0, and then Eqs. (10), (17),
and (18) imply the relation,
s = (c1 + c2 ) = (a + 3b),
(22)
which is familiar from the E6 assignment in Table 1. Thus, most of the phenomenological
studies of E6 inspired Z 0 bosons apply here, and also to the other solutions with k0 = 0.
84
To simplify the algebra, I impose the orthogonality condition, Tr QY Q0 = 0, which can

be relaxed at a later stage. Combined with the anomaly condition quadratic in the U (1)0
charges 10 , Tr QY Q0 2 = 0, I find,
(a b) + (c1 c2 )
(k3 + k2 )s,
1
(6 + 3)(a b) + (7 + 2)(c1 c2 )
2[(k3 k5 )d1 + (k2 k5 )d2 ] =
1

+ 2k5 (k3 + k2 ) s.
2(k3 e1 + k2 e2 ) = 9
(23)
(24)
Similarly, the cubic condition, Tr Q0 3 = 0, reads,

3

1 X 3
ti = 5k5 + 3 1 (3k3 + 2k2 1)
(25)
3
s
i=1

c1 c2 2
ab 2
3
+4
3+9
4
s
s

e1 e1
e2 e2
+ + 2 k2
+
+ 3 3 k3
s s
s s

d1 d1
d2 d2
+ 1 + 2(k5 k2 )
+1 ,
+ 3(k5 k3 )
s s
s s
and one can show that,

3 6

1
5(k5 1) + 3 1 (3k3 + 2k2) 3 .
4
(26)
The inequality (26) is saturated when,

s
a=b= ,
4
s
c1 = c2 = d 1 = d 2 = ,
2
e1 = e2 = s,
2
(27)
which yields a class of Z 0 bosons which couple like the Z to the ordinary fermions and the
Higgs doublets. This is encouraging as the recent analysis [35] of precision data revealed
an excellent fit to the Z assuming some amount of mixing with hypercharge. Recall
that U (1)0 subgroups from E6 are frequently discussed because they represent manifestly
anomaly free examples (and because they arise in popular GUT and string models), but the
E6 gauge symmetry and Yukawa coupling relations are usually explicitly broken to avoid
serious conflicts with observation (such as the proton lifetime). Here we recover U (1)0
U (1) symmetries which are very similar to the U (1) from a very different bottom-up
approach.
Relaxing the orthogonality condition, Tr QY Q0 = 0, and using the trivial solution to
Eq. (17), a + 3b = c1 + c2 = s = t = 0, one can construct another family of special
solutions, U (1) , where the MSSM fields have charges as in the U (1) case. While they
would yield non-chiral pairs of Higgs doublets, these solutions could be interesting if they
10 Tr Q Q0 2 also vanishes if s = 0.
Y
85
enter linear combinations with the U (1) or if the corresponding Z coexists with the Z .
The quadratic anomaly condition now reduces to
(k3 k5 )d1 + (k2 k5 )d2 (k3 e1 + k2 e2 ) + 2(a + c1 ) = 0,
(28)
where solutions with c1 = a correspond to hypercharge and c1 = 2a/3 to the U (1) .

Note, that there are other solutions to the anomaly constraints with s = 0, but only those
satisfying Eq. (28) can mix with the chiral U (1)0 s.
The encounter of the U (1) and U (1) groups is also encouraging from a top-down
perspective. There are classes of string models where gauge coupling unification is
predicted without the appearance of a full GUT group in the zero-slope limit. Examples
are heterotic string constructions where the gauge group factors considered are realized at
the same KacMoody level, k, or open string models with the group factors arising from
the same D-brane. The E8 E8 heterotic string is particularly suited here, because it offers
the prerequisite E6 subgroups. The reference E8 symmetry implies the relevant matching
(unification) conditions, as well as quantization conditions for its U (1) subgroups. The
latter are given by
6
Q QY Z,
5
5
Q Q Z,
4
4
Q Q Z,
3
(29)
where the last relation refers to the U (1) defined by E7 E6 U (1) . The
normalizations are chosen such that QY = 1 for e+ (a = 1/2), Q = 1 for (a = 3/5),
Q = 1 for S (s = 1), and Q = 1 for one of the E6 singlets within the 56 of E7 . Eq. (28)
implies for the U (1) of solution II, 2d2 + 3e1 + 2/5 = 0, or when combined with the
first Eq. (29), e1 + 2/5 2Z. With the exception of two fields, the second Eq. (29) is also
satisfied. But the D quarks have the sign of their U (1) charges reversed and the T field
(Q = 0) has half-integer instead of integer U (1) charge. It is not clear at present 11
whether E8 charge quantization, i.e., Eqs. (29), are predicted in string models with gauge
coupling unification and SU(5) charge quantization. If so one would have to impose the
condition Z, and only solution I would survive.
The E8 embedding also allows one to fix the overall normalization (coupling strength)
of the U (1)0 . In units in which an E8 root vector P has length P 2 = 1, one has P 2 = 3/5
for a 10 of SU(5), P 2 = 5/8 for a 16 of SO(10), P 2 = 2/3 for a 27 of E6 , and P 2 = 3/4 for
a 56 of E7 . The 10 of SU(5) is relevant here, because it contains the SU(3) SU(2) singlet
with unit charge (the e+ ) and can therefore serve to properly normalize the hypercharge
gauge coupling, gY , in terms of one of the non-Abelian gauge couplings, g. Assuming the
normalization in Eq. (29) one finds the well-known result, gY = 3/5 g. Notice, that this
argument needs no reference to a complete GUT group like SU(5), but it applies only to
string constructions in which the embedding into E8 is still traceable. Referring again to the
normalization fixed by Eq. (29) one finds in a similar way, g = 5/8 g and g = 2/3 g.
11 In KacMoody level 1 models SU(5) charge quantization can never be obtained unless a complete SU(5)
GUT group survives [68]. On the other hand in the level 2 models or Ref. [69] the third Eq. (29) is respected by
all fields.
86
The reference E8 root vector also allows to set lower limits on the possible KacMoody
levels, k. For example, the singlet T of solution II has Q = 3s/2, which implies PT2 =
3/2 > 1 and means that T cannot be obtained in the massless sector if k = 1, while k >
2 would be allowed by this consideration. Interestingly, the theorem about fractionally
charged states in level k = 1 models [68] implies the same conclusion.
5. Phenomenological aspects
The one-loop -function coefficient for the U (1)0 is given by,

10
ab 2
c1 c2 2
2
=
+
5k
+
1
(3k
+
2k
+
1)
+
6
+
9
+
7
(30)
2
5
3
2
s2
s
s

e1 e1
e2 e2
+ + 2k2
+
+ 2 3k3
s s
s s

d1 d1
d2 d2
+ 1 + 2(k5 k2 )
+1 ,
+ 3(k5 k3 )
s s
s s
P
2
2
0
where 2 = s
t =i ti . 1 is minimized when Eqs. (27) are satisfied, i.e., for the U (1) ,
in which case one finds,

1
(31)
10 = 5(k5 + 2) + 2 1 (3k3 + 2k2 ) + 2 2 + 2 .
3
In Eq. (31) I have chosen the E6 inspired normalization, s 2 = 2/3. For comparison, the
SU(3) SU(2) U (1)Y one-loop -function coefficients are 3 = k5 3, 2 = k5 + 1,
and 1 = k5 + 33/5 (using SU(5) normalization, a 2 = 3/20), respectively.
For concreteness, I now focus on the Z of solution II which is reminiscent of an E6
S quarks removed and the U (1)0
model with three families of 27, but with one copy of D + D
charges of the other two copies altered. The trace and cubic anomaly conditions, 1 = 1/2
and 3 = 3max = 23/16, can be satisfied by adding, for example, SU(5) singlets with
U (1)0 charges, t1 = t2 = s, t3 = 3s/4, t2i+2 = s/2, and t2i+3 = s/4, where 1 6 i 6 9, or
in short-hand notation, ti [12 , 3/4, (1/2)9, (1/4)9 ]. There are no singlets with charges
1 or 3/2 in this choice, assuring that the crucial fields, S and T , cannot acquire high
, are summarized in Table 5.
scale masses. The quantum numbers for this case, Z (II)
There are a total of 12 fields with quantum numbers like the right-handed neutrinos.
Since these fields are protected from acquiring high scale Majorana masses, one predicts
a Dirac mass matrix with three electroweak scale eigenvalues. I.e., there are unacceptably
large neutrino masses, unless the Yukawa couplings are chosen zero or tiny. A clean way to
cure this problem is to add the set of singlets 12 , ti 3[1, (3/4)4, (1/2)6 , (1/4)4]. Now
the fields with ti = s/4 (and some others) can form bilinear Dirac mass terms, resulting in
a neutrino mass matrix with three zero eigenvalues, which correspond predominantly to the
left-handed neutrinos provided the bilinear masses are much larger than the Yukawa mass
12 By virtue of the identity, n3 4(n 1)3 + 6(n 2)3 4(n 3)3 + (n 4)3 = 0, such a configuration does
not alter 1 or 3 .
87
Table 5
Charge assignment for the Z of solution II realized with a 5 + 5 of SU(5).
The U (1)0 normalization corresponds to s = 1
QY

e

u
d
(II)
(II)
QY
1/2
+1/4
e+
0
+1
+1/4
+1/4
+1/6
+1/4
u
d
2/3
+1/3
+1/4
+1/4
hd
1/2
1/2
hu
+1/2
1/2
D1,2
L1,2
S
1/3
1/2
0
0
0
3/4
1/2
+1
+1
1/2
+1/3
+1/2
0
0
0
3/4
1/2
+3/2
+3/4
+1/4
+1
T1,2
1/2
T2i+2
S1,2
D
S1,2
L
T
+3/4
T3
+1/4
T2i+3
terms. Clearly, the bilinear masses can be taken to infinity and the corresponding fields can
be smoothly removed from the spectrum, yielding the residual chiral set of singlets (now
including S, T , and the three ), [3/2, 16, (3/4)11, (1/2)9 ].
There is another interesting solution which can be obtained by adding the fields, ti
3[(5/4), 14, (3/4)6, (1/2)4 , (1/4)] to the ones in Table 5. This results in the chiral
singlet set, [3/2, (5/4)3, 115 , (3/4)17, (1/2)3 , (1/4)9], where the fields with Q0 =
5s/4 allow trilinear couplings with S and the . This once again results in three zero
eigenvalues, and if the masses generated by S are an order of magnitude larger than the ones
from the standard Yukawa terms, these also correspond predominantly to the left-handed
neutrinos, but the mixing with the massive states could induce a small amount of missing
invisible Z width, inv . Indeed, the LEP Collaborations [45] currently report a shortage of
about 0.5% in inv . Furthermore, the Fermi constant, GF , extracted from decays could
increase, significantly increasing the extracted Higgs boson mass from electroweak fits
(the central value of which is currently below the lower search limit [23]) but potentially
deteriorating the quality of the global fit to all data. There can be other effects associated
with this scenario, such as a violation of weak charged current universality. There are
strong limits on the non-universality between the first two families, however, driven by the
B( )
well measured branching ratio, B(ee) . Thus the effect in inv should be expected to be
dominated by the sector, in agreement with the experience that the third family Yukawa
couplings are typically the largest.
Yet another solution to the problem of neutrino mass would be to implement the see-saw
mechanism which can be achieved by choosing a linear combination of the U (1) and
the U (1) , such that the are U (1)0 neutral and allowed to receive high scale Majorana
masses. As is well known, this also produces small but non-vanishing Majorana masses
88
for the left-handed neutrinos, which is not the case for the scenarios in the previous
two paragraphs. However, one should generally expect non-vanishing masses for the lefthanded neutrinos originating from non-renormalizable terms suppressed by powers of the
high mass scale [70]. Dimension 5 operators would yield the same kind of suppression
as in the see-saw mechanism, but higher order suppressions (by intermediate scales) are
conceivable. 13
As for baryon number violating interactions, the U (1) symmetry of solution II forbids
the terms in Eq. (6), thus assuring proton stability at the renormalizable level. Dimension 5
operators can also lead to proton decay at rates conflicting with the experimental limits.
However, the D-terms,
qq d ,
S ,
qq D
DD u ,
,
u dD
S ,
u DD
(32)
are also absent by virtue of the U (1) , and likewise for the F -terms,
lqqq,
e+ DDD,
Hd qqq,
Lqqq,
Hu qDD,
S
LqDD,
e+ u u d.
(33)
On the other hand, the F -terms,

qqD,
S
u d D,
3/4
Ti
u dd,
+5/4
Ti
SD,
S
u D
(34)
Q0
= 5s/4, s/4,
are guaranteed to be absent only when there are no MSSM singlets with
or 3s/4. While this can be arranged, there is another dimension 5 F -term,
S
e+ u u D,
(35)
present in the R-parity conserving version of

which is similar to the operator e+ u u d,
the MSSM. There one has to assume that its corresponding coupling strength is small,
and taking into account suppressions from first generation Yukawa couplings, Cabibbo
mixing, and phase space, the resulting proton lifetime can be acceptable. In comparison,
the term (35) is less concerning, because at the renormalizable level the D quarks couple
+1/2
S
dD-term) via gauge interactions and the T D Donly (ignoring the possibility of a Ti
term, both conserving D-number. Therefore, observable proton decay is not expected,
S mixing effect can be generated. Alternatively, the critical dimension 5
unless a large dD
F -term in Eq. (35) can be removed if there is an admixture of the U (1) . Thus, choosing
k3 = k5 (or more generally coupling all of the exotic quarks to singlets other than S) is an
efficient strategy to achieve a realistic proton lifetime.
6. Conclusions
I have presented a class of models with an additional non-anomalous U (1)0 in which
gauge couplings unify, the -problem can be solved, and electric charge is quantized.
13 Employing the Z 0 scale, M 0 800 GeV [35], and the unification scale, M 21016 GeV, suggests neutrino
Z
2 /M 0.03 eV, in excellent agreement with the mass scale inferred from atmospheric
masses of order MZ
0
neutrino oscillations [71]. One might wonder whether this agreement not only signals the presence of a new
physics scale close to the unification scale, but possibly even suggests the absence of further scales below that.
89
These models are chiral, protecting a large desert between the electroweak (SUSY
breaking) and unification (string) scales. Moreover, in some of these models the U (1)0
is sufficient to avoid conflicts with experimental bounds on the proton lifetime. Within a
deliberately traditional and conservative framework the analysis presented is rather general,
but there are various directions which can be chosen to go beyond it.
For example, I assumed that two MSSM singlet fields participate in U (1)0 symmetry
breaking. Only a finite number of solutions exist since the anomaly matrix in Eq. (17) has
to be singular. The analysis generalizes straightforwardly if one allows three (or more)
singlets to acquire VEVs. In this case an infinite number of solutions exist, a large (but
finite) number of which being consistent with perturbative gauge coupling unification.
A preliminary study of this class reveals that unlike the solutions in Table 3 there are also
models with D-flat scalar field directions, a property which can be phenomenologically
advantageous. There are also solutions simultaneously in accordance with E8 charge
quantization (see Eq. (29)) and the proton lifetime.
I have further assumed that all exotic fields receive their masses through couplings to
MSSM singlets acquiring VEVs upon U (1)0 breaking. But exotic leptons participating in
SU(2) U (1)Y symmetry breaking will in general receive mass contributions of O(MZ )
which could conceivably by themselves be large enough to be in agreement with current
limits. This seems difficult but is a possible loophole which I have ignored in this work.
An implicit assumption has been made concerning the precise form of gauge coupling
unification. It is the often made assertion that gauge coupling unification works satisfactorily in the MSSM and should not be altered in leading (one-loop) order. Extra matter multiplets are then usually restricted to appear exclusively in complete SU(5) representations.
In the present work, I have already relaxed this restriction and included quasi-complete
SU(5) representations. However, gauge coupling unification works only approximately,
with the prediction for the strong coupling constant, s (MZ ), significantly higher than
most observations and the unification scale at least one order of magnitude lower than the
string scale. It is fair to ask whether some arbitrary (incomplete from the perspective of
SU(5)) representation could improve the quality of unification. Indeed, by including a set
of fields which contributes stronger to the SU(3) -function than to the one of SU(2) improves the prediction for s (MZ ). Furthermore, by having the smallest contribution to the
U (1)Y -function, one can increase the unification scale. The minimal choice is the set
G + W + E , contributing 13 = 3, 12 = 2, and 11 = 6/5, respectively. Alternatively,
one could add another 5 + 5 of SU(5) to this set, or choose a family plus a mirror-family
S U +U
S, and D + D.
S Both cases correspond to 13 = 4, 12 = 3, and
of quarks, Q + Q,
11 = 11/5. The case 13 = 5 (and the same -function differences) would yield unification in the non-perturbative domain, but is also conceivable. In any case, these choices yield
SU(2) U (1)Y unification slightly above the reduced Planck scale of 2.4 1018 GeV. The
prediction for s (MZ ) is now below typical observed values but the discrepancy is in general neither much better nor much worse than in the MSSM. If it is found that higher order
or threshold effects tend to raise the prediction for s , these scenarios may be of potential
interest, but I have not considered them in the analysis of this paper. In this context it is
amusing to note that one would obtain a very good prediction for s , as well as unification
90
very close to the string scale, if these extra fields have intermediate scale masses near the
geometric mean of the string and electroweak scales.
Even given these possible directions it is possible to draw some general conclusions
from the kind of analysis introduced in this work. Having left the framework of E6 inspired
U (1)0 models, anomaly cancellation tends to provide U (1) groups which have charges to
the ordinary fermions just as the U (1) subgroups of E6 . The charges of some of the extra
fields typically differ, sometimes forbidding fast proton decay mediating couplings. A more
thorough phenomenological investigation will be presented elsewhere.
Acknowledgement
It is a pleasure to thank Paul Langacker and Michael Plmacher for many fruitful
discussions. This work was supported in part by U.S. Department of Energy Grant EY76-02-3071.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
H.P. Nilles, Phys. Rep. 110 (1984) 1.

H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75.
J.E. Kim, H.P. Nilles, Phys. Lett. 138B (1984) 150.
A. Font, L.E. Ibez, F. Quevedo, Phys. Lett. 228B (1989) 79.
T. Banks, L. Dixon, Nucl. Phys. B 307 (1988) 93.
S. Weinberg, Phys. Rev. D 26 (1982) 287.
N. Sakai, T. Yanagida, Nucl. Phys. B 197 (1982) 533.
L.J. Hall, I. Hinchliffe, Phys. Lett. 112B (1982) 351.
M. Cvetic, P. Langacker, Z 0 Physics and Supersymmetry, in: G.L. Kane (Ed.), Perspectives in
Supersymmetry, World Scientific, Singapore, 1998, p. 312, hep-ph/9707451.
S. Dimopoulos, S. Raby, F. Wilczek, Phys. Rev. D 24 (1981) 1681.
L.E. Ibez, G.G. Ross, Phys. Lett. 105B (1981) 439.
U. Amaldi et al., Phys. Rev. D 36 (1987) 1385.
J. Ellis, S. Kelley, D.V. Nanopoulos, Phys. Lett. 249B (1990) 441.
P. Langacker, M.-X. Luo, Phys. Rev. D 44 (1991) 817.
U. Amaldi, W. de Boer, H. Frstenau, Phys. Lett. 260B (1991) 447.
P. Langacker, N. Polonsky, Phys. Rev. D 47 (1993) 4028.
P.H. Chankowski, Z. Puciennik, S. Pokorski, Nucl. Phys. B 439 (1995) 23.
J. Bagger, K. Matchev, D. Pierce, Phys. Lett. 348B (1995) 443.
H.C. Cheng, B.A. Dobrescu, K.T. Matchev, Phys. Lett. B 439 (1998) 301.
H.C. Cheng, B.A. Dobrescu, K.T. Matchev, Nucl. Phys. B 543 (1999) 47.
J. Erler, Implications of precision electroweak measurements for the Standard Model Higgs
boson, in: Proceedings of the 17th International Workshop on Weak Interactions and Neutrinos,
Cape Town, South Africa, WIN 99, January 1999, p. 40, hep-ph/9904235.
J. Erler, P. Langacker, Electroweak Model and Constraints on New Physics, in Ref. [24].
D.E. Groom et al., Eur. Phys. J. C 15 (2000) 1.
M. Carena, M. Quiros, C.E.M. Wagner, Nucl. Phys. B 461 (1996) 407.
D.M. Pierce, J.A. Bagger, K. Matchev, R.J. Zhang, Nucl. Phys. B 491 (1997) 3.
H.E. Haber, R. Hempfling, A.H. Hoang, Z. Phys. C 75 (1997) 539.
[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]
[70]
[71]
91
M. Masip, R. Muoz-Tapia, A. Pomarol, Phys. Rev. D 57 (1998) 5340.

S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Rev. D 58 (1998) 091701.
S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Lett. 440B (1998) 296.
S. Heinemeyer, W. Hollik, G. Weiglein, Eur. Phys. J. C 9 (1999) 343.
R.J. Zhang, Phys. Lett. 447B (1999) 89.
J. Erler, D.M. Pierce, Nucl. Phys. B 526 (1998) 53.
G.C. Cho, K. Hagiwara, Nucl. Phys. B 574 (2000) 623.
J. Erler, P. Langacker, Phys. Rev. Lett. 84 (2000) 212.
C.S. Wood et al., Science 275 (1997) 1759.
S.C. Bennett, C.E. Wieman, Phys. Rev. Lett. 82 (1999) 2484.
W.J. Marciano, J.L. Rosner, Phys. Rev. Lett. 65 (1990) 2963.
P. Langacker, Phys. Lett. 256B (1991) 277.
K.T. Mahanthappa, P.K. Mohapatra, Phys. Rev. D 43 (1991) 3093.
R. Casalbuoni, S. De Curtis, D. Dominici, R. Gatto, Phys. Lett. 460B (1999) 135.
J.L. Rosner, Phys. Rev. D 61 (2000) 016006.
N.H. Edwards et al., Phys. Rev. Lett. 74 (1995) 2654.
P.A. Vetter et al., Phys. Rev. Lett. 74 (1995) 2658.
The LEP Collaborations, ALEPH, DELPHI, L3, OPAL, the LEP Electroweak Working
Group, and the SLD Heavy Flavour and Electroweak Groups, A Combination of Preliminary
Electroweak Measurements and Constraints on the Standard Model, preprint CERN-EP-2000016.
M. Peskin, T. Takeuchi, Phys. Rev. Lett. 65 (1990) 964.
G. Altarelli, R. Barbieri, Phys. Lett. 253B (1990) 161.
J.L. Hewett, T.G. Rizzo, Phys. Rep. 183 (1989) 193.
D. Suematsu, Y. Yamagishi, Int. J. Mod. Phys. A 10 (1995) 4521.
M. Cvetic, P. Langacker, Phys. Rev. D 54 (1996) 3570.
L.E. Ibez, G.G. Ross, Nucl. Phys. B 368 (1992) 3.
A.H. Chamseddine, H. Dreiner, Nucl. Phys. B 447 (1995) 195.
M. Cvetic, D.A. Demir, J.R. Espinosa, L. Everett, P. Langacker, Phys. Rev. D 56 (1997) 2861.
M. Cvetic, D.A. Demir, J.R. Espinosa, L. Everett, P. Langacker, Phys. Rev. D 58 (1998) 119905.
M. Aoki, N. Oshimo, Phys. Rev. Lett. 84 (2000) 5269.
S. Wolfram, Phys. Lett. 82B (1979) 65.
E. Nardi, E. Roulet, Phys. Lett. 245B (1990) 105.
P.F. Smith, J.R. Bennett, G.J. Homer, J.D. Lewin, H.E. Walford, W.A. Smith, Nucl. Phys. B 206
(1982) 333.
T.K. Hemmick et al., Phys. Rev. D 41 (1990) 2074.
A. Riotto, M. Trodden, Ann. Rev. Nucl. Part. Sci. 49 (1999) 35.
D.J. Chung, E.W. Kolb, A. Riotto, Phys. Rev. D 60 (1999) 063504.
E. Ma, Phys. Rev. D 36 (1987) 274.
K.S. Babu et al., Phys. Rev. D 36 (1987) 878.
J.F. Gunion et al., Int. J. Mod. Phys. A 2 (1987) 1199.
P. Langacker, private communication.
P. Langacker, M. Plmacher, Phys. Rev. D 62 (2000) 013006.
R. Hempfling, Phys. Lett. 351B (1995) 206.
A.N. Schellekens, Phys. Lett. 237B (1990) 363.
J. Erler, Nucl. Phys. B 475 (1996) 597.
P. Langacker, Phys. Rev. D 58 (1998) 093017.
Super-Kamiokande Collaboration, Y. Fukuda et al., Phys. Rev. Lett. 81 (1998) 1562.

Renormalization-group-improved effective
potential for the MSSM Higgs sector with explicit
CP violation
M. Carena a,b , J. Ellis a , A. Pilaftsis a,b, , C.E.M. Wagner a,c,d
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b Fermilab, P.O. Box 500, Batavia, IL 60510, USA
c High Energy Physics Division, Argonne National Lab., Argonne, IL 60439, USA
d Enrico Fermi Institute, University of Chicago, 5640 Ellis Ave., Chicago, IL 60637, USA
Abstract
We perform a systematic study of the one-loop renormalization-group-improved effective potential
of the minimal supersymmetric extension of the Standard Model (MSSM), including CP violation
induced radiatively by soft trilinear interactions related to squarks of the third generation. We
calculate the charged and neutral Higgs-boson masses and couplings, including the two-loop
logarithmic corrections that arise from QCD effects, as well as those associated with the top- and
bottom-quark Yukawa couplings. We also include the potentially large two-loop non-logarithmic
corrections induced by one-loop threshold effects on the top- and bottom-quark Yukawa couplings,
due to the decoupling of the third-generation squarks. Within this minimal CP-violating framework,
the charged and neutral Higgs sectors become intimately related to one another and therefore require
a unified treatment. In the limit of a large charged Higgs-boson mass, MH MZ , the lightest
neutral Higgs boson resembles that in the Standard Model (SM), and CP violation occurs only in
the heavy Higgs sector. Our analysis shows that sizeable radiative effects of CP violation in the
Higgs sector of the MSSM may lead to significant modifications of previous studies for Higgs-boson
searches at LEP 2, the Tevatron and the LHC. In particular, CP violation could enable a relatively
light Higgs boson to escape detection at LEP 2. 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
The violation of CP was first observed in the neutral kaon system [16], where a
deviation from the superweak theory has recently been confirmed [79], and CP violation
in B-meson decays 1 is strongly suggested by recent experiments [12]. In addition
apostolos.pilaftsis@cern.ch
1 For pedagogical introductions to CP violation in the B-meson system, see Refs. [10,11].
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 5 8 - 8
M. Carena et al. / Nuclear Physics B 586 (2000) 92140
93
to its interest for particle physics, CP non-conservation provides a key ingredient for
cosmological baryogenesis, namely for explaining the underlying mechanism which
caused matter to dominate over anti-matter in our observable universe [13]. Although
a fundamental understanding of the origin of CP violation is still lacking, most of the
scenarios proposed in the existing literature indicate that Higgs interactions play a key
role in mediating CP violation. For instance, CP violation is broken explicitly in the
Standard Model (SM) by complex Yukawa couplings of the Higgs boson to quarks.
Another appealing scheme of CP violation occurs in models with an extended Higgs sector,
in which the CP symmetry of the theory is broken spontaneously by the ground state
of the Higgs potential [1418]. Supersymmetric (SUSY) theories, including the minimal
supersymmetric extension of the Standard Model (MSSM), predict an extended Higgs
sector, and therefore may realize either or both the above two schemes of explicit and
spontaneous CP violation. However, the Higgs potential of the MSSM is invariant under
CP at the tree level, and any explicit or spontaneous breakdown of CP symmetry can arise
only via radiative corrections. The case of purely spontaneous CP violation in the MSSM
[19] leads to an unacceptably light CP-odd scalar [2022], as a result of the GeorgiPais
theorem [23], and hence such a scenario is ruled out experimentally.
It has recently been shown [24] that the tree-level CP invariance of the MSSM Higgs
potential may be violated sizeably by loop effects involving soft CP-violating trilinear
interactions of the Higgs bosons to top and bottom squarks. A detailed study [25] has
shown that significant CP-violating effects of level crossing in the Higgs sector can take
place in such a minimal SUSY scenario of explicit CP violation, which may lead to
drastic modifications of the tree-level couplings of the Higgs particles to fermions [25,
26] and to the W and Z bosons [25]. The latter can have important phenomenological
consequences on the production rates of the lightest Higgs particle, even though the upper
bound on its mass was found [25] to be very similar to that found previously in the CPconserving case [2749,5153]. The MSSM predicts an upper bound on the lightest Higgs
boson mass of approximately 110 (130) GeV for small (large) values of the ratio of Higgs
vacuum expectation values tan 2 (> 15). On the other hand, experiments at LEP 2,
running at center-of-mass energies s = 196202 GeV, have placed a severe lower bound
of approximately 108 GeV on the mass of the SM Higgs boson [54]. LEP 2 is expected to
run at center-of-mass energies up to s = 206 GeV during the year 2000. Consequently,
for stop masses smaller than 1 TeV, a significant portion of the parameter space spanned
by tan and the CP-odd scalar mass MA can be tested for the CP-conserving case in this
next round of experiments at LEP 2 [25]. However, the explorable region of parameters
is smaller for larger amounts of stop mixing and/or larger CP-violating phases. A decisive
test of such a scenario can only be provided by the upgraded Tevatron collider and the
LHC.
The earlier study of the renormalization-group (RG) improved effective potential of the
MSSM with explicit CP violation in [25] was based on an expansion of the Higgs quartic
couplings in inverse powers of the arithmetic average of stop and sbottom masses, under
the assumption that the mass splittings of the left- and right-handed stop and sbottom
masses are small. The mass expansion of the one-loop effective potential was truncated
94
up to renormalizable operators of dimension four. Although the above approach captures

the basic qualitative features of the underlying dynamics under study, it is known [25,
48,49] that such a mass expansion has limitations when the third-generation squark
mixing is large. Since the dominant CP-violating loop contributions to the effective Higgsboson masses and mixing angles occur for large values of the third-generation squarkmixing parameters, it is necessary to provide a more complete one-loop computation of
the effective MSSM Higgs potential with explicit radiative breaking of CP invariance,
including non-renormalizable operators below the heavy scalar-quark scale and without
resorting to any other kinematic approximations. 2
To this end, we consider here the two-loop leading logarithms induced by top- and
bottom-quark Yukawa couplings as well as those associated with QCD corrections, by
means of RG methods. In the calculation of the RG-improved effective potential, we
also include the leading logarithms generated by one-loop gaugino and higgsino quantum
effects [44]. Finally, we implement the potentially large two-loop corrections that are
induced by the one-loop stop and sbottom thresholds in the top- and bottom-quark Yukawa
couplings, which may become particularly relevant in the large-tan regime. On the basis
of the RG-improved effective potential, we present predictions for the Higgs-boson mass
spectrum and the effective Higgs couplings to fermions and gauge bosons. The results of
the analysis are compared with those obtained in the CP-conserving case [48,50] and also
with the results obtained by truncating the one-loop RG-improved effective potential up to
renormalizable operators [25].
In this analysis, it is important to consider the constraints on the low-energy CP-violating
parameters of the MSSM that originate from experimental upper limits on the electron [55]
and neutron [56] electric dipole moments (EDMs) [5784]. Most of the EDM constraints
affect the CP-violating couplings of the first two generations [5765]. Thus, making the
first two generations of squarks rather heavy, much above the TeV scale [6671], is a
possibility that can drastically reduce one-loop contributions to the neutron EDM, without
suppressing the CP-violating phases of the theory. Another interesting possibility for
avoiding any possible CP crisis in the MSSM is to arrange for cancellations among the
different EDM terms, either at the level of short-distance diagrams [7274] or (for the
neutron EDM) at the level of the strong-interaction matrix elements for operators with s,
u and d quark flavours [75,76]. Alternatively, one might require a specific form of nonuniversality in the soft trilinear Yukawa couplings [77,8083].
However, third-generation squarks can also give rise by themselves to observable effects
on the electron and neutron EDMs through the three-gluon operator [78,79], through
the effective coupling of the CP-odd Higgs boson to the gauge bosons [8083], and
through two-loop gaugino/higgsino-mediated EDM graphs [84]. These different EDM
contributions of the third generation can also have different signs and add destructively
to the electron and neutron EDMs. In our phenomenological discussion, we take into
2 We recall that the diagrammatic computation of scalarpseudoscalar transitions is the focus of [24], whilst
sbottom contributions and relevant D-term effects on the effective potential are not considered in [26].
95
account the relevant EDM constraints related to the CP-violating parameters of the stop
and sbottom sectors.
This paper is organized as follows: in Section 2 we calculate the complete one-loop CPviolating effective potential, and derive the analytic expressions for the effective charged
and neutral Higgs-boson mass matrices. Technical details are given in Appendices A and B.
Section 3 describes our approach to determining the RG-improved Higgs-boson mass
matrices, after including the leading two-loop logarithms associated with Yukawa and
QCD corrections. Section 4 is devoted to the calculation of the effective top- and bottomquark Yukawa couplings, in which one-loop threshold effects of the third-generation
squarks are implemented. In Section 5, we discuss the phenomenological implications of
representative CP-violating scenarios compatible with EDM constraints for direct Higgs
searches at LEP 2 and the upgraded Tevatron collider. We also compare the results of our
analysis with those obtained using the mass-expansion method [25]. Finally, in Section 6
we summarize our conclusions.
2. CP-violating one-loop effective potential

In this section, we first describe the basic low-energy structure of the MSSM that
contains explicit CP-violating sources, such as soft CP-violating trilinear interactions.
Then we calculate the general one-loop CP-violating effective potential. Finally, after
implementing the minimization tadpole conditions related to the Higgs ground state, we
derive the effective charged and neutral Higgs-boson mass matrices.
CP violation is introduced into the MSSM through the Higgs superpotential and the soft
supersymmetry-breaking Lagrangian:
b1T i2 L E
b + hd H
b1T i2 Q
bD
b + hu Q
bT i2 H
b2 U
b H
b1T i2 H
b2 ,
W = hl H

1
i i
eL2 L L
Lsoft = mg ag ag + mW
e W
eB
eB
e + h.c. + M
e W
e + mB
2
2 e e
2 e e
eU2 U
e U
e+M
eD
eE2 E
e E
e + m21
eQ
e
e
Q Q+M
D D+M
+M
1 1

e1T i2 2 + h.c.
+ m22 2 2 B

T
e + hd Ad Q
ee
ee
+ hl Al 1 L E
1 D hu Au 2 i2 QU + h.c. ,
(2.1)
(2.2)
b1 and 2 is
e1 = i2 is the scalar component of the Higgs chiral superfield H
where
1
the usual Pauli matrix. The conventions followed throughout this paper, including the
quantum-number assignments of the fields under the SM gauge group, are displayed in
Table 1.
As can be seen from (2.1) and (2.2), the MSSM includes additional complex parameters
with new CP-odd phases that are absent in the SM. These new CP-odd phases may reside
in the following parameters: (i) the mass parameter describing the bilinear mixing
of the two Higgs chiral superfields in the superpotential; (ii) the soft supersymetrybreaking gaugino masses mg , mW
e and mB
e of the gauge groups SU(3)c , SU(2)L and
U(1)Y , respectively; (iii) the soft bilinear Higgs-mixing mass B; and (iv) the soft
96
Table 1
The field content of the MSSM
Superfields
Bosons
Fermions
SU(3)c SU(2)L U(1)Y
Gauge multiplets
ba
G
b
W
Ga 21 a
g a
ei
W
(8, 1, 0)
e
B
(1, 1, 0)
b
B
Matter multiplets
L
b
E
b
Q
b
U
b
D
b1
H
b2
H
Wi 21 i
B
L
L T = (l , l)
e = l
E
R
eT = (u,
L
Q
d)
e = u
U
R
e = d
D
R
eT = ( 0 , )
2T = (2+ , 20 )
(1, 3, 0)
LT = (l , l)L
(eR )C = (eC )L
QT = (u, d)L
(uR )C = (uC )L
(dR )C = (d C )L
0 , )
( H
H1
1
+
0 )
(H
, H
2
(1, 2, 1)
(1, 1, 2)
(3, 2, 13 )
(3, 1, 43 )
(3, 1, 23 )
(1, 2, 1)
(1, 2, 1)
supersymmetry-breaking trilinear couplings Af of the Higgs bosons to sfermions. In

addition, there may exist other large CP-odd phases, associated with flavor off-diagonal
soft supersymmetry-breaking masses of squarks and sleptons. We assume that these offdiagonal masses are small and therefore do not give sizeable contributions to the effective
Higgs potential [85,86]. The number of independent CP-odd phases may be reduced if one
prescribes a universality condition for all gaugino masses at the unification scale MX ; the
gaugino masses will then have a common phase. Correspondingly, the different trilinear
couplings Af may be considered to be all equal at MX , i.e., Af A. In this case, however,
because of the different RG running of the phases of the trilinear couplings, their values
at low energies will be different. 3 Two CP-odd phases may further be eliminated by
employing the following two global symmetries that govern the dimension-four operators
in the MSSM Lagrangian:
= 0,
b1 ) = 1, Q(H
b2 ) = 2, Q(Q)
b = Q(L)
(i) The U(1)Q symmetry specified by Q(H
b) = 2 and Q(D)
b = Q(E)
b = 1. This U(1)Q symmetry is broken by the
Q(U
parameter and the respective soft supersymmetry-breaking one, B.
(ii) The U(1)R symmetry acting on the Grassmann-valued coordinates, i.e., the
coordinate of superspace carries charge 1. Under the R transformation, the matter
superfields and gaugino fields carry charge 1, whilst the Higgs superfields are
R-neutral. The R symmetry is violated by the gaugino masses, the trilinear
couplings Af and the parameter .
We concentrate on the parameters which may have a dominant CP-violating effect on
the MSSM Higgs potential, under the assumption of a common phase for the gauginos;
the latter is made less important by the fact that the one-loop gaugino corrections are
3 For a discussion of the RG effects, see [87].
97
subdominant compared to the ones induced by the third-generation squarks. As has been
mentioned above, two CP-odd phases of the complex parameters {, m212 , m , A} may
be removed by employing the global symmetries (i) and (ii). Specifically, one of the Higgs
doublets and the common phase of the gaugino fields can be rephased in a way such that
the gaugino masses and B become real numbers. As a consequence, arg() and arg(At,b )
are the only physical CP-violating phases in the MSSM which affect the Higgs sector in a
relevant way.
It is obvious from (2.1) and (2.2) that the Yukawa interactions of the third-generation
eT =
quarks, QT = (tL , bL ) and tR , bR , as well as their SUSY bosonic counterparts, Q
(tL , bL ) and tR , bR , play the most significant role in radiative corrections to the Higgs
sector. Therefore, it is useful to give the interaction Lagrangians related to the F and D
terms of the third generation:

e2 + |ht |2 2T i2 Q
e2
LF = |hb |2 1+ Q

e 2 bR + ht Q
e i2 1 tR + h.c.
hb Q

ht 2 tR ,
(2.3)
hb bR 1T i2 + ht tR 2 hb i2 1 bR
2

g
e Q
e2 + 2 Q
e2 Q
e 1 + 2
LD = w 21T i2 Q
2
1
2
4

02
1 4

g
eQ
e tR tR + 2 bR bR
2 2 1 1
Q
,
(2.4)
+
4
3
3
3
where gw and g 0 are the usual SU(2)L and U(1)Y gauge couplings. Further, the interaction
Lagrangian of the Higgs bosons to the top and bottom quarks is given by

(2.5)
Lfermions = hb bR QT 1 + h.c. + ht tR QT i2 2 + h.c. ,
where ht and hb are the top and bottom Yukawa couplings, respectively.
With the help of the Lagrangians (2.3)(2.5), we now proceed with the calculation of the
one-loop effective potential. More explicitly, in the MS scheme, the one-loop CP-violating
effective potential is determined by
"

#

m
e2qi
m
2q
3 X X 4
3
3
0
4
m
eqi ln 2
2m
q ln 2
.
LV = LV +
32 2
Q
2
Q
2
q=t,b
i=1,2
(2.6)
In (2.6), L0V is the tree-level Lagrangian of the MSSM Higgs potential

2
L0V = 21 1 1 + 22 2 2 + m212 1 2 + m2
12 2 1 + 1 1 1
2

+ 2 2 2 + 3 1 1 2 2 + 4 1 2 2 1 ,
(2.7)
with
22 = m22 ||2 ,
m212 = B,
21 = m21 ||2 ,

1 2
1 2
1 2
+ g 02 ,
g 02 ,
.
3 = gw
4 = gw
1 = 2 = gw
8
4
2
(2.8)
98
Further, in (2.6), m
2i (with i = t, b) and m
e2qk (with qk = t1 , b1 , t2 , b2 ) denote the eigenvalues
f2 , respectively, which depend on the
of the quark and squark mass matrices M M and M
Higgs background fields. Specifically, M M reads

|ht |2 |2 |2
ht hb 1T i2 2
,
(2.9)
M M=
ht hb 1 i2 2
|hb |2 |1 |2
with eigenvalues
1h
m
2t (b) = |hb |2 |1 |2 + |ht |2 |2 |2
2
q
i
2
+ () |hb |2 |1 |2 + |ht |2 |2 |2 4|ht |2 |hb |2 |1 2 |2 .
(2.10)
= 0, (2.10) simplifies to the known expressions: m

2t =
It is easy to see that, for 1,2
0 2
0 2
2
2
2
b = |hb | |1 | .
|ht | |2 | and m
f2 is more complicated. In the weak basis {Q
eT =
The (4 4) squark mass matrix M
e
e
(tL , bL ), U = tR , D = bR }, M may be cast in the form:

f2
f2 ) e e (M
f2 ) e e
(M )Q
e Q
e (M
Q U
Q D
f2 =
f2 ) e e (M
f2 ) e e (M
f2 ) e e
(2.11)
M
(M
UQ
UU
UD ,
2
2
2
f ) e e (M
f ) e e (M
f ) e e
(M
DQ
with
DD

+ |hb |2 1 1 + |ht |2 2 2 12 2 2

1 02
1 2
1 2

g g
1 1 2 2 12 ,
gw 1 1 2 2 +
2
4 w 12

f2 e e = ht At T i2 + ht T i2 ,
f2 e e = M
M
2
1
UQ
Q U

f2 e e = hb Ab hb ,
f2 e e = M
M
1
2
DQ
Q D
f2
M
f2
M
f2
M
DU
e2
e Q
e = MQ 12
Q
eU
e
U
eD
e
D

f2 e e
M
UD

et2 + |ht |2 2 + 1 g 02 1 2 ,
=M
2
1
2
3

1
eb2 + |hb |2 1 g 02 1 2 ,
=M
1
1
2
6

f2 e e = ht h T i2 2 .
= M
b 1
DU
e2 ,
M
Q
et2
M
e2
M
b
(2.12)
and
the soft supersymmetryHere and in the following, we denote by
breaking masses of the third generation of squarks. It is rather difficult to express the four
f2 in a simple form. However, as we detail in
eigenvalues m
e2qk (qk = t1 , b1 , t2 , b2 ) of M
Appendix A.3, it is not necessary to know the analytic form of m
e2qk in order to evaluate the
Higgs-boson masses and mixing angles [88]. Of course, for 1,2 = 0, the field-dependent
squark eigenvalues simplify to

2
02

1 e2
2
et2 + 2|ht |2 0 2 + gw + g 0 2 0 2
m
et1 (t2 ) = MQ + M
2
1
2
2
4

q

0 2
0 2 2
0
0 2
2
2
2

e
e
+ 4|ht | At 2 1 ,
+ () MQ Mt + xt |1 | |2 |
m
e2b1 (b2 ) =
99

2
02

1 e2
eb2 + 2|hb |2 0 2 gw + g 0 2 0 2
MQ + M
1
1
2
2
4

q

e 2 xb | 0 |2 | 0 |2 2 + 4|hb |2 A 0 0 2 ,
e2 M
+ () M
Q
b 1
(2.13)
2 5 g 02 ) and x = 1 (g 2 1 g 02 ).
where xt = 14 (gw
b
3
4 w
3
After having set the stage, we now derive the minimization conditions governing the
ground state of the MSSM one-loop effective potential and then determine the Higgsboson mass matrices. As usual, we consider the following linear expansion of the Higgs
doublets 1 and 2 around the ground state:

1+
2+
,
2 = ei
,
(2.14)
1 = 1
(v1 + 1 + ia1 )
1 (v2 + 2 + ia2 )
2
where v1 and v2 are the moduli of the vacuum expectation values (VEVs) of the Higgs
doublets and is their relative phase. Following [24,25], we require the vanishing of the
total tadpole contributions

LV
T1 (2 )
1(2)

v1 v2
1
2
2 i
2
2
= v1(2) 1(2) + 2 Re m12 e + 1(2) v1(2) + (3 + 4 )v2(1)
2
v1(2)
"

m2

m
2q
e2qi
3 X X m
q i
2
2
m
ln
m2
1(2) qi
1(2) q
16 2
Q2
q=t,b i=1,2

#
m2q
ln 2 1 ,
(2.15)
Q

LV
Ta1 (a2 )
a1(2)

m2
e2qi

3 X X m
q i
2
1
,
m
ln
= +()v2(1) Im m212 ei
a1(2) qi
16 2
Q2
q=t,b i=1,2
(2.16)
2q /1(2)i, h m
e2qk /1(2)i and h m
e2qk /
where he
m2qi i = m2qi , and the tadpole derivatives h m
a1(2)i are given in (A.1) and (A.3). Moreover, from (A.3), we readily see that Ta1 =
tan Ta2 , with tan = v2 /v1 . This last fact [24,25] allows us to perform an orthogonal
rotation in the space spanned by the CP-odd scalars a1 and a2 ,
0

cos sin
G
a1
.
(2.17)
=
a2
a
sin
cos
The Higgs potential then has a flat direction with respect to the G0 field, i.e., hLV /
G0 i = 0, and the G0 field becomes the true would-be Goldstone boson eaten by the
longitudinal component of the Z boson.
100
We observe from (2.16) that a relative phase between the two Higgs vacuum
expectation values is induced radiatively in the MS scheme [24,25]. However, we should
stress that the phase is renormalization-scheme dependent. For example, one may adopt
a renormalization scheme, slightly different from the MS one, in which is set to zero
order-by-order in perturbation theory [24]. This can be achieved by requiring the bilinear
Higgs-mixing mass m212 to be real at the tree level, but to receive an imaginary counter-term
(CT), Im m212 , at higher orders, which is determined by the vanishing of the CP-odd tadpole
parameters Ta1 and Ta2 for = 0. As we detail below, the scheme of renormalization of
Im(m212 ei ) does not directly affect the renormalization scheme of other physical kinematic
parameters of the theory to one loop, such as Higgs-boson masses and tan . In fact, it
has been explicitly demonstrated in [24] that physical CP-violating transition amplitudes,
such as scalar-pseudoscalar transitions, are independent of the renormalization subtraction
point Q2 and the choice of phase . In the following, we adopt the = 0 scheme of
renormalization, as irrelevant -dependent phases in the effective chargino and neutralino
mass matrices can thereby be completely avoided.
In the remaining part of this section, we evaluate the one-loop effective Higgs-boson
mass matrices. Employing the tadpole conditions T1 = T2 = 0 and Ta1 = Ta2 = 0 allows
one to substitute the mass parameters 21 and 22 , and Im m212 into the effective potential
(2.6). After performing the above substitutions, we can express the charged-Higgs-boson
mass matrix as follows:

1 2
v1 v2
v1 v2
Re m212 + gw
M2 ij = (1)i+j
vi vj
4
"

2
e2qk
2m
eqk
ij m
3 X X
+
+
2
vi j
16
i j
q=t,b
k=1,2

m2
e2qk
i(1 ij ) m
q
m2qk ln 2k 1
vi
aj
Q
#
2
2 2
q
q
m2q
m
ij m
2
mq ln 2 1 .
2
vi j
Q
i+ j

(2.18)
2 may be
Since det M2 = 0 in (2.18), the square of the charged Higgs-boson mass MH
+
2
determined by the matrix element (M )12 :
(

v2
1 2
2
v1 v2
Re m212 + gw
MH + =
v1 v2
4
"

2 2
2
m2
eqk
m
eqk
i m
3 X X
q k
2
mqk ln 2 1
16 2
v1 a2
Q
1+ 2
q=t,b k=1,2
#)
2 2
q
m2q
m
2
.
(2.19)
mq ln 2 1
2
Q
1+ 2
2q /1+ 2 i and h 2 m
e2qk /
The self-energy derivatives appearing in (2.18) and (2.19): h 2 m
+
2
2
2
qk /j i, h m
eqk /j i and h m
eqk /aj i, are
1 2 i, as well as the tadpole terms h m
101
exhibited in Appendix A.
By analogy, in the weak basis {1 , 2 , a1 , a2 }, the neutral-Higgs-boson mass matrix
takes on the form 4

M2S
M2SP

M20 =
,
(2.20)
T
M2P
M2SP
where M2S , M2P and M2SP denote the two-by-two matrices of the scalar, pseudoscalar
and scalarpseudoscalar squared mass terms of the neutral Higgs bosons. Observe that the
presence of CP-violating self-energy terms leads to mass eigenstates with no well-defined
CP quantum numbers. Therefore the CP-odd Higgs-boson mass MA cannot be identified
with any of the neutral Higgs-boson masses. The individual matrix elements of M20 are
given by
M2S

ij

1 2
v1 v2
Re m212 + gw
+ g 02 vi vj
vi vj
4
(

2
m2
X
e2qk
2m
eqk
ij m
3 X
q k
2
+
mqk ln 2 1
i j
vi j
16 2
Q
= (1)i+j

+
q=t,b
i

2
+
k=1,2
2
eqk
m
e2qk m
j

2
2q
m
i j
2 2
q
m
q m
i
ln
ln
m2qk
Q2
2
q
ij m
vi

m2q
Q2
"
j
)
m2q

m2q
ln 2 1
Q
(2.21)
2
eqk
(1 ij ) m
i aj
vi
aj
q=t,b k=1,2
2 2 m2 #
m2
m
eqk m
eqk
q k
q
2
ln 2k ,
mqk ln 2 1 +
Q
i
aj
Q

v1 v2
Re m212
M2P ij = (1)i+j
vi vj
(

2
X 2 m
e2qk
eqk
ij m
3 X
+
2
16
ai aj
vi j
M2SP ij
3 X X
=
16 2

m2qk
q=t,b
ln
m2qk
Q2
2m
e2qk
k=1,2
1 +
m
e2qk

m
e2qk
(2.22)
m2qk
ln 2
ai
aj
Q

2
2 2
2
q
q
mq
m
ij m
m2q ln 2 1 .
2
ai aj
vi j
Q
(2.23)
4 Notice that our convention differs from that given in [25], as the neutral-Higgs-boson mass matrix M2 in that
0
work is expressed in the weak basis {a1 , a2 , 1 , 2 }.
102
Again, the analytic expressions for the self-energy and tadpole derivatives with respect to
the background Higgs fields are given in Appendix A.
Since G0 does not mix with the other neutral fields, the (4 4) matrix M20 reduces to a
(3 3) matrix, which we denote by M2N . In the weak basis {1 , 2 , a}, the reduced neutral
mass-squared matrix M2N may be expressed by

1
M2
M2
M2
M2N =
S 11
S 12

2
MS
1
cos
21
M2SP
cos

2

12
MS
22
sin1 M2SP

21
SP 12
sin1 M2SP
21

2
sin 1cos MP
(2.24)
12
in (2.24), we have used the properties of the matrix elements of M2SP :

In writing
2
(MSP )11 = tan (M2SP )12 and (M2SP )22 = cot (M2SP )21 , and likewise for M2P .
Using the expressions (2.19) and (2.24), we determine the analytic forms of the RGimproved charged and neutral Higgs-boson masses in the next section.
M2N
3. RG-improved Higgs-boson mass matrices

In this section, we perform a one-loop RG improvement of the squared charged Higgs2
2
boson mass MH
+ and of the squared neutral Higgs-boson mass matrix MN . The RG
improvement incorporates all leading two-loop logarithmic corrections to the Higgs-boson
mass-matrix elements, which were already found in the CP-conserving case to give rise
to significant contributions to the Higgs-boson masses and couplings. In particular, the
upper bound on the lightest CP-even Higgs mass was found to be strongly affected by
the two-loop logarithmic corrections [4549]. In carrying out the RG improvement, we
follow the procedure outlined in [48,49], in which the improvement of the Higgs-boson
mass-matrix elements was performed by carefully applying the process of decoupling of
the third-generation squarks.
Within the framework of the RG approach, the dominant contributions to the Higgsboson mass matrix M2 may be written conceptually as a sum of two terms:
M2 (mt ) = M 2 (mt ) + M2,th (mt ).
(3.1)
The first term, M 2 (mt ), contains the genuine logarithmic contributions which determine
the whole scale dependence of the one-loop effective potential. These contributions would
be present, even if the left-right mixing of the stop and sbottom states were absent. The
second term, M2,th(mt ), describes the threshold effect of the decoupling of the heavier
stop and sbottom squarks and their respective mixing with the lighter states. At the oneloop level, the second term is manifestly scale independent. In (3.1), we are interested in
evaluating the effective potential at mt , since it has been shown [46] that this is the scale
at which two-loop corrections are minimized. As we explain below, the renormalization of
the above two contributions must proceed in different ways.
Let us denote by Qt b the scale of the heaviest third-generation squark, which we assume
to be higher than the electroweak scale. In the language of the RG approach, we have
103
first to consider that the aforementioned threshold contribution is frozen at the scale
Qt b , M2,th(Qt b ) 1M2 (Qt b ), with all the involved kinematic parameters defined at this
particular scale. Then, we have to rescale the threshold contribution with the anomalous
dimension factors of the relevant Higgs fields:
2,th
(mt ) = 1M2ij (Qt b )i1 (mt )j1 (mt ),
Mij
(3.2)
where i (mt ) is the anomalous dimension factor of the Hi state to be determined below.
The one-loop matrix elements 1M2ij (Qt b ) depend on the running quark masses at the
scale Qt b , which have to be conveniently re-expressed as functions of the corresponding
running masses at mt . Thus, the anomalous-dimension factors in combination with the
one-loop relation between the quark masses at scales Qt b and mt yield sizeable two-loop
corrections to the mass-matrix elements originating from the one-loop threshold effects.
As was already mentioned, the contribution M 2 (mt ) of the third-generation squarks to
the effective potential describes the genuine one- and two-loop leading-logarithmic running
of the Higgs quartic couplings. In this context, there are two important technical details
that should be mentioned. First, we notice that, in the MSSM, the tree-level Higgs quartic
2
couplings i , with i = 1, 2, 3, 4, are all proportional to the squared gauge couplings gw
and g 02 (cf. (2.8)). However, the one-loop functions of i can generally have appreciable
values, as they are proportional to the fourth power of the top- and bottom-quark Yukawa
couplings. As a result, the low-energy values of i differ significantly from their tree-level
ones. The RG-improved approach followed here is crucial for implementing properly the
potentially large logarithmic corrections to the Higgs quartic couplings.
The second technical remark pertains to the RG evolution of the Higgs quartic
couplings i , with i = 5, 6, 7 (for the notation, see [25,44]), which are absent in the Born
approximation to the MSSM Higgs potential. On field-theoretic grounds, these quartic
couplings must have vanishing one-loop functions, and cannot be generated by RG
running. However, these quartic couplings are radiatively induced by threshold effects,
and have already been taken into account in 1M2 (Qt b ), given by (3.2).
Following the above discussion, we now proceed with the RG improvement of the
Higgs-boson mass-matrix elements. To this end, we first need to compute the one-loop
values of the quartic couplings i , with i = 1, 2, 3, 4, where the decoupling of the stop
and sbottom contributions at their appropriate thresholds is properly taken into account.
The best way to calculate the latter effects is to consider the logarithmic part of the
effective potential (2.6) in the limit where the squark mixing parameters vanish, i.e.,
(1)
= At = Ab = 0. The pertinent one-loop running quartic couplings, denoted by i , may
then be obtained by
"
e2

2
MQ + m2t
g 02 2
3
gw
(1)
ln
1 =
4
12
32 2
Q2
e2

MQ + m2b
g 02 2
g2
ln
+ |hb |2 w
4
12
Q2
2
e2

#
et + m2t
Mb + m2b
M
g 02 2
g 04
2
ln
,
(3.3)
ln
+ |hb |
+
9
Q2
6
Q2
104
"
e2

2
MQ + m2t
g 02 2
3
gw
2
+
=
ln
|ht |
32 2
4
12
Q2
e2

2
MQ + m2b
g 02 2
g
ln
+ w+
4
12
Q2
e2
e2

#

Mb + m2b
g 02 2
Mt + m2t
g 04
2
ln
ln
+
,
(3.4)
+ |ht |
3
36
Q2
Q2
(
e2

e 2 + m2 )
et2 + m2t , M
MQ + m2t
max(M
3
(1)
b
b
2
2
|
|h
|
|h
ln
+
ln
3 =
t
b
16 2
Q2
Q2

2
e2

2
2
MQ + m2t
g 02 gw
g 02
gw
gw g 02
2
+
|ht |
ln
4
12
4
12 4
12
Q2

2
e 2 + m2
2
2
M
g 02
gw
g 02 gw
gw g 02
Q
b
2
+
|hb |
+
ln
4
12
4
12 4
12
Q2

2

et + m2t
g 02
M
g 02
2
|ht |
ln
+
3
3
Q2
)

02
e 2 + m2
M
g
g 02
b
b
|hb |2
ln
,
(3.5)
+
6
6
Q2
(
e2

e 2 + m2 )
et2 + m2t , M
MQ + m2t
max(M
3
(1)
b
b
2
2
|
|h
|
|h
ln
+
ln
4 =
t
b
16 2
Q2
Q2
)
e2 + m2t g 2
e 2 + m2
2
2 M
2 M
g
g
gw
Q
Q
b
|ht |2 w ln
w |hb |2 w ln
.
2
4
2
4
Q2
Q2
(1)
2
(3.6)
Moreover, we need to know the one-loop running of the soft supersymmetry-breaking
parameter Re m212 . Gathering the relevant logarithmic terms present in the effective
potential (2.6), we find

e 2 + m2t , M
et2 + m2t )
max(M
3
Q
2(1)
2
|ht | Re(At ) ln
Re m12 =
16 2
Q2

2
2
e2 + m2 )
e +m , M
max(M
Q
b
b
b
.
(3.7)
+ |hb |2 Re(Ab ) ln
Q2
The analytic form of 1M2 (Qt b ) in (3.2) can now be obtained by subtracting the oneloop Born-improved mass matrix M2(0) from its total one-loop contribution M2(1) . Here,
2(1)
2(1)
we have in mind the charged and neutral Higgs-boson mass matrices M and MN
2
calculated in Section 2. More explicitly, 1M (Qt b ) is given by

(1)
(3.8)
1M2 (Qt b ) = M2(1)(Qt b ) M2(0) Re m2(1)
12 (Qt b ), i (Qt b ) ,
where M2(0) represents the tree-level functional form of M2 , expressed in terms of (1)
i
and Re m2(1)
.
Furthermore,
it
is
essential
to
stress
again
that
the
kinematic
parameters
12
105
involved in (3.8), such as masses and couplings, are evaluated at the scale Qt b .
Another important ingredient in the RG improvement of the Higgs-boson mass
matrices is the analytic two-loop result for the Higgs quartic couplings 1 , . . . , 4 .
As has been done in (3.3)(3.6), we have to include two-loop leading logarithms,
by appropriately considering the stop and sbottom thresholds. These two-loop leading
(2)
logarithmic contributions to the Higgs quartic couplings, which we denote by i , can
be determined by solving iteratively the RG equations [48,49]. In this way, we obtain

6|hb |4 3
1
(2)
2
2
2
|hb | + |ht | 8gs
1 =
(32 2 )2 2
2
e2
e2
2
2
+
m
M
Q
b
2
2 Mb + mb
+ ln
,
(3.9)
ln
Q2
Q2

6|ht |4 3
1
|ht |2 + |hb |2 8gs2
=
(2)
2
2
2
2
(32 ) 2

e2
e2
2
2
MQ + mt
2
2 Mt + mt
+ ln
,
(3.10)
ln
Q2
Q2

3|ht |2 |hb |2
=
|ht |2 + |hb |2 8gs2

(2)
3
2
2
(16 )

e2
2 e2
2
e2
MQ + m2t
2
2 max(Mt + mt , Mb + mb )
+ ln
,
(3.11)
ln
Q2
Q2
(2)
(2)
4 = 3 .
(3.12)
For later convenience, we define collectively the sum of the tree, one-loop and two-loop
quartic couplings as follows:
(2)
i = i + (1)
i + i ,
(3.13)
with i = 1, 2, 3, 4. Similarly, the sum of the tree, one-loop and two-loop contributions to
the soft-bilinear Higgs mixing may be defined as
2(2)
Re m
212 = Re m212 + Re m2(1)
12 + Re m12 .
(3.14)
As we see below, knowledge of the two-loop contribution Re m2(2)

12 is not required in the
one-loop RG improvement of the MSSM Higgs potential.
Given the above definitions of the quartic couplings and the soft Higgs-mixing
parameter in (3.13) and (3.14), we can express the one- and two-loop leading logarithmic
contributions M 2 (mt ) to M2 (mt ) by means of the two-loop Born-improved mass matrix:

M 2 (mt ) = M2(0) Re m
(3.15)
212 (mt ), i (mt ) .
Note that M 2 (mt ) also includes the tree-level terms. As has also been stated explicitly in
(3.15), M 2 (mt ) is expressed in terms of mass and coupling parameters evaluated at the
top-quark-mass scale.
The last ingredient for completing the programme of RG improvement of the Higgsboson mass matrices is knowledge of the analytic expressions for the anomalous dimension
106
factors that occur in (3.2). These analytic expressions are given below for the charged and
neutral Higgs-boson cases separately.
Adopting the framework outlined above [48,49], it is not difficult to compute the RG2 (m ) at the top-mass scale through the relation:
improved charged Higgs-boson mass MH
t
+
+
1
tb
2
2
2
(Qt b )
1MH
MH
+ (mt ) = M H + (mt ) + 1 (mt )2 (mt )
+

2(1) t b
(mt ),
+ MH
+
(3.16)
e 2 + m2t , M
et2 + m2t , M
e 2 + m2 ), and + (mt ) and (mt ) are the
where Q2t b = max(M
Q
b
b
1
2
anomalous dimension factors of the charged Higgs fields 1+ and 2 , respectively:
1+ (mt ) = 1 +
3|hb |2 Q2t b
ln 2 ,
32 2
mt
2 (mt ) = 1 +
3|ht |2 Q2t b
ln 2 .
32 2
mt
(3.17)
Further, MH2 + (mt ) is the squared two-loop Born-improved charged Higgs-boson mass
given by
MH2 + (mt ) =
Re m
212 (mt )
1
+ 4 (mt )v 2 (mt )
sin (mt ) cos (mt )
2
(3.18)
2 )tb is the one-loop scale-invariant part that contains the stop and sbottom
and (1MH
+
contributions:

tb
tb
1
2
M2(1)
(Qt b ) + Re m2(1)
1MH + (Qt b ) =
12 (Qt b )
12
sin (mt ) cos (mt )

1 (1)
(3.19)
+ 4 (Qt b )v1 (Qt b )v2 (Qt b ) ,
2
where the scale at which the kinematic parameters are to be evaluated has been indicated
explicitly. In (3.19), the term between brackets is the threshold contribution to the off2(1) b
diagonal matrix element of the charged-Higgs-boson mass matrix, where (M )t12
2
denotes the one-loop contribution of the third-generation squarks to (M )12 . Finally,
2(1) t b
2 (see (3.16)).
describes the one-loop quark contribution to MH
(MH
+ )
+
We remark that the charged-Higgs-boson mass matrix receives the common anomalous
dimension factor 1+ (mt )2 (mt ), even though different matrix elements of M2 are
involved. This is because M2 must possess a vanishing determinant at any RG scale
Q2 , as one of its mass eigenstates must correspond to the massless would-be Goldstone
boson G+ that forms the longitudinal component of the W + boson. As a consequence,
the following relations among the matrix elements of the RG-frozen part 1M2 of the
charged-Higgs-boson mass matrix are obtained:

1M2 11 (Qt b ) = tan (Qt b ) 1M2 12 (Qt b ),

(3.20)
1M2 22 (Qt b ) = cot (Qt b ) 1M2 21 (Qt b ).
After including the RG running due to the Higgs-boson anomalous dimensions, we find
(1M2 )11 (Qt b )
[1+ (mt )]2

(1M2 )22 (Qt b )
[2 (mt )]2
= tan (mt )
= cot (mt )
(1M2 )12 (Qt b )
(1M2 )12 (Qt b )
1+ (mt )2 (mt )
1+ (mt )2 (mt )
107
(3.21)
where we have used the RG relation:

tan (Qt b ) =
1+ (mt )
2 (mt )
tan (mt ).
(3.22)
As a consequence of this last relation, it is evident that the RG running of the different
matrix elements of 1M2 may be expressed in terms of the running of (1M2 )12 and the
value of tan at the scale mt .
Correspondingly, the RG-improved neutral-Higgs-boson mass matrix M2N may be
computed by

1

t
1M2N ij (Qt )
M2N ij (mt ) = MN2 ij (mt ) + ijt (mt )
1
b

1M2N ij (Qb )
+ ijb (mt )
2(1) t b
(3.23)
+ MN ij (mt ),
e2 + m2t , M
et2 + m2t ) and Q2 = max(M
e 2 + m2 , M
e2 + m2 ).
with i, j = 1, 2, 3, Q2t = max(M
Q
b
Q
b
b
b
Notice that, unlike the charged-Higgs-boson case, one has to introduce here two decoupling
scales Qt and Qb , as the stop/top and sbottom/bottom loop effects occur separately in the
q
threshold contributions. The parameters ij (with q = t, b) are the anomalous-dimension
factors related to the neutral Higgs-boson fields
( q
q
(mt )j (mt ), for i, j = 1, 2,
q
q
(3.24)
ij (mt ) = j i (mt ) = iq
q
1 (mt )2 (mt ), for i = 1, 2, 3 and j = 3,
with
1t(mt ) = 1 +
1b (mt ) = 1 +
3|hb |2 Q2t
ln 2 ,
32 2
mt
2t(mt ) = 1 +
3|hb |2 Q2b
ln 2 ,
32 2
mt
2b (mt ) = 1 +
3|ht |2 Q2t
ln 2 ,
32 2
mt
3|ht |2 Q2b
ln
.
32 2 m2t
(3.25)
We should observe that, for the very same reasons as in the charged-Higgs-boson case, the
vanishing of the determinants of M2P and M2SP at any Q2 scale leads to the common
anomalous dimension factor i3 (mt ) = 3i (mt ) = 1 (mt )2 (mt ) in the calculation of
M2N (mt ) in (3.23). Since this last fact involves the matrix elements (M2P )12 , (M2SP )12
q
and (M2SP )21 , the corresponding matrix elements of (1M2N )i3 in (3.23) are given by
1M2N
1M2N
q
(Qq ) =
13
q
(M2SP )12 (Qq )

,
cos (mt )
(Qq ) =
33
1M2N
q
(Qq ) =
23
(M2SP )21 (Qq )

,
sin (mt )
(1M2P )12 (Qq )

,
sin (mt ) cos (mt )
(3.26)
108
where the RG-scale dependence of the involved quantities has been displayed explicitly.
In (3.23), the (3 3) matrices MN2 and (1M2N )q (with q = t, b) describe, respectively,
the two-loop Born-improved effects and the one-loop threshold contributions associated
with the decoupling of the heavy squark states:

2(0)
MN2 (mt ) = MN Re m
(3.27)
212 (mt ), 1 (mt ), 2 (mt ), 34 (mt ) ,

2(1) q
2 q
(Qq )
1MN (Qq ) = MN

2(1),q
(1),q
(1),q
(1),q
Re m12 (Qq ), 1 (Qq ), 2 (Qq ), 34 (Qq ) , (3.28)
M2(0)
N
(1)
(1)
(1)
2(0)
where 34 = 3 + 4 (likewise 34 = 3 + 4 ) and MN is the tree-level functional
2
form of MN :
sin2
sin cos 0
2
Re
m
2(0)
12
0
sin cos
cos2
MN =
sin cos
0
0
1
2
34 sin cos 0
21 cos
(3.29)
v 2 34 sin cos
22 sin2
0,
0
0
0
q
with 34 = 3 + 4 . As in the charged-Higgs-boson case, we write (M2(1)
N ) to denote
2
the one-loop part of MN containing the contributions of the third-generation squarks, and
tb
(M2(1)
N ) to denote its fermionic one-loop counterpart.
The resulting RG-improved Higgs-boson mass matrix M2N (mt ) is a symmetric,
positive-definite (3 3) matrix, and can therefore be diagonalized by an orthogonal
transformation as follows:
2

2
2
(mt ), MH
(mt ), MH
(mt ) ,
(3.30)
O T M2N (mt )O = diag MH
1
2
3
where we have defined the Higgs fields such that their RG-improved masses satisfy the
inequality:
MH1 (mt ) 6 MH2 (mt ) 6 MH3 (mt ).
(3.31)
Notice that our convention in (3.30) differs from that chosen in [25], as we assign the
Higgs fields in the reversed order. Analytic expressions for MHi (mt ) and O are presented
in Appendix B.
Before closing this section, two important remarks are in order. First, we observe that
the free kinematic parameters of the MSSM Higgs sector are
MH + (mt ),
2
eQ
(Qt b ),
M
tan (mt ), (Qt b ), At (Qt b ),

et2 (Qt b ), M
eb2 (Qt b ).
M
Ab (Qt b ),
(3.32)
In fact, the soft Higgs-mixing parameter Re m

212 (mt ) may be substituted by the squared
2
RG-improved mass MH + (mt ) of the charged Higgs boson (cf. (3.16) and (3.18)) in the
neutral Higgs-boson mass matrix M2N (mt ) in (3.23).
Secondly, we reiterate the fact that Im m212 can be renormalized independently, without
affecting the renormalization of the physical parameters of the theory [24]. As was
109
stressed in Section 3, the = 0 scheme of renormalization gives rise to a considerable

simplification, since we can get rid of the radiatively-induced phase between the two
Higgs vacuum expectation values in the analytic expressions of the Higgs-boson masses
and mixing angles. For example, within the above = 0 scheme, the mass renormalization
of H + may be entirely reabsorbed by a corresponding renormalization of Re m212 and 4 . In
other words, it can be shown that MH + is Q2 -independent, after including the RG running
of Re m212 and 4 , denoted as Re m2 and 4 . For simplicity, we assume that only the third
12
generation of squarks contributes to Re m2 , since fermions do not contribute to the RG
12
running of Im m212 . The analytic forms of Re m2 and 4 are given by

12

3 2
|ht | Re(At ) + |hb |2 Re(Ab ) ,
Re m2 =
2
12
16

(1)
2
4
gw
d4
3
gw
2
2
2
2
|ht | + |hb | +
=
.
2|ht | |hb |
4
d ln Q2
16 2
2
4
(3.33)
(3.34)
Obviously, the RG running of Re m212 due to t and b is only relevant for non-zero values
of At and Ab . Employing (2.19) and examining only the ln Q2 -dependent part, one can
verify that
2
dMH
+
d ln Q2
1
Re m2 4 v1 v2
12
2
2 f4

f4
3
Tr M
i Tr M
+
= 0,
32 2 1+ 2
v1
a2
(3.35)
as it should be. As can also be seen from

an important role in this proof is played
(3.35),
f4 /a2 [24].
by the necessary CP-odd tadpole term Tr M
4. Effective top and bottom Yukawa couplings

In addition to the RG improvement of the Higgs-boson mass matrices discussed in the
previous section, we consider here a further improvement related to the non-logarithmic
threshold corrections to the top- and bottom-quark Yukawa couplings. Specifically, apart
from the usual RG running, the effective top- and bottom-quark Yukawa couplings obtain
additional non-logarithmic threshold contributions, which are induced by the decoupling
of the heavy SUSY states at a high scale, e.g., Qt b . For the bottom-quark Yukawa case, the
one-loop RG relation between the bottom mass and the bottom-quark Yukawa coupling
at the scale Qt b receives quantum corrections that also include terms proportional to
tan [8998]. Since these last terms can be significant for large values of tan , 5 we must
resum them within the RG approach, so that the actual size of the radiative corrections to
the Higgs-boson masses and couplings can properly be extracted [50,99]. For the top-quark
case, instead, although one-loop suppressed, the respective corrections can still give rise to
5 For the -lepton Yukawa coupling, the corresponding enhanced tan terms are much smaller, because they
are proportional to the weak gauge couplings [50,9093].
110
an enhancement of up to 4 GeV in the prediction for the lightest Higgs-boson mass [100,
101], and therefore should be included in the computation.
There may also be important CP-violating one-loop corrections to the bottom- and topquark Yukawa couplings, in addition to the CP-violating effects induced by the radiative
mixing of the Higgs states, which were considered in Sections 2 and 3 in detail. In the
leptonic sector, these CP-violating vertex corrections are generally small [102]. However,
the CP-violating radiative corrections to the couplings of the Higgs bosons to b quarks are
significant [25], because of the large Yukawa and colour-enhanced QCD interactions [89].
In particular, the radiative effects of the Higgs-boson couplings to the bottom quarks can
be further enhanced, if the respective Higgs-mass eigenstate couples predominantly to the
Higgs doublet 2 [9098], as the tree-level b-quark Yukawa coupling is suppressed in this
case. For a general discussion of the form and the origin of these finite Yukawa corrections
to the third-generation quark masses, the reader is referred to the original literature [90
98]. In the following, we give a brief discussion of the non-logarithmic corrections to the
top and bottom Yukawa couplings, and pay special attention to the CP-violating vertex
effects.
We start our discussion by considering the effective Lagrangian of the b-quark Yukawa
coupling [25,50]:
0
0
L 0 bb
= (hb + hb )1 b R bL + 1hb 2 b R bL + h.c.,
(4.1)
with
|ht |2

2s
hb
mg Ab I m2b , m2b , |mg |2
=
||2 I m2t , m2t , ||2 ,
2
1
2
1
2
hb
3
16
2

|h
1hb 2s
t|
mg I m2b , m2b , |mg |2 +
=
At I m2t , m2t , ||2 ,
2
1
2
1
2
hb
3
16
(4.2)
(4.3)
where s = gs2 /(4) is the SU(3)c coupling strength, and I (a, b, c) is the one-loop
function
ab ln(a/b) + bc ln(b/c) + ac ln(c/a)
.
(4.4)
I (a, b, c) =
(a b)(b c)(a c)
The b-quark Yukawa coupling hb (Qt b ) is then related to the running b-quark mass
mb (Qt b ) by
gw mb
,
(4.5)
hb =
2 MW cos [1 + hb / hb + (1hb / hb ) tan ]
where hb / hb and 1hb / hb are given in (4.2) and (4.3), respectively. The running b-quark
mass mb (Qt b ) is obtained by means of the RG running of the b-quark mass from the scale
mb . In (4.5), we have redefined the right-handed b-quark superfield, so that the physical
b-quark mass is positive. Under such a field redefinition, only the Yukawa coupling hb
becomes complex, while the phases of hb / hb and 1hb / hb as well as those of the
supersymmetry-breaking parameters do not change. Moreover, since only the moduli of
the Yukawa couplings hb and ht enter the field-dependent quark and squark masses in
(2.10) and (2.13), the neutral Higgs-boson mass matrices remain unaffected by the above
111
field redefinition. Also, we have checked that the very same property of invariance under
rephasings of ht and hb persists for the charged Higgs-boson mass matrix as well. At this
point, it is interesting to observe that the sum hb + 1hb tan in (4.5) receives two sorts
of quantum corrections, one originating from QCD effects and another from a charginomediated graph. The QCD correction is proportional to the hermitean conjugate of the
sbottom-mixing parameter Xb = Ab tan , whilst the chargino-induced diagram [90
93] depends linearly on the stop-mixing parameter Xt = At cot .
The effective Lagrangian describing the t-quark Yukawa coupling is given by
L 0 tt = 1ht 10 tR tL + (ht + ht )20 tR tL + h.c.
(4.6)
The corresponding relation for ht as a function of mt may easily be determined analogously

by the effective Lagrangian (4.6), and reads
gw mt
,
(4.7)
ht =
2 MW sin [1 + ht / ht + (1ht / ht ) cot ]
with
|hb |2

2s
1ht
mg I m2t , m2t , |mg |2 +
=
Ab I m2b , m2b , ||2 ,
(4.8)
2
1
2
1
2
ht
3
16
|hb |2 2

2s
ht
mg At I m2t , m2t , |mg |2
=
|| I m2b , m2b , ||2 .
(4.9)
2
1
2
1
2
ht
3
16
As in the case of the b-quark Yukawa coupling, we have to make a judicious phase rotation
of the right-handed t-quark superfield, such that the physical top-quark mass becomes
positive. Again, one can show that such a field redefinition does not change the analytic
results of the RG analysis.
At this stage, it is important to remark that, within the RG-resummation approach
described in Section 3, the non-logarithmic corrections must be treated as threshold effects
and hence they should only contribute to the RG-frozen part of the Higgs-boson mass
matrices, generically denoted as 1M2 (Qt b ). Therefore, the decoupling procedure for the
heavy squark states requires that the effective b- and t-quark Yukawa couplings given by
(4.5) and (4.7) are evaluated at the scale Qt b . As we discuss in Section 5, these additional
Yukawa corrections can lead to observable effects in Higgs-boson searches.
It is now straightforward to obtain the interaction Lagrangians of the Higgs-boson mass
eigenstates Hi to the up- and down-type quarks, collectively denoted as u and d. Taking
into account both CP-violating self-energy and vertex effects, we find

3
X

gw mu
gw md S
P
S
P
d gHi dd + igHi dd 5 d +
Hi
u gHi uu + igHi uu 5 u ,
LH ff =
2MW
2MW
i=1
(4.10)
with
S
gH
i dd

1
O1i
O2i
Re(hd + hd )
+ Re(1hd )
=
hd + hd + 1hd tan
cos
cos

Im(hd + hd ) tan Im(1hd ) O3i ,
(4.11)
112

1
Re(1hd ) Re(hd + hd ) tan O3i
hd + hd + 1hd tan

O1i
O2i
Im(1hd )
,
Im(hd + hd )
cos
cos

1
O2i
O1i
S
=
Re(hu + hu )
+ Re(1hu )
gH
i uu
hu + hu + 1hu cot
sin
sin

Im(hu + hu ) cot Im(1hu ) O3i ,

1
P
Re(1hu ) Re(hu + hu ) cot O3i
gHi uu =
hu + hu + 1hu cot

O2i
O1i
Im(1hu )
Im(hu + hu )
,
sin
sin
P
gH
=
i dd
(4.12)
(4.13)
(4.14)
where the Higgs scalar and pseudoscalar couplings are normalized with respect to their
SM values.
Finally, it is interesting to investigate the behaviour of self-energy- and vertex-type CP
violation in the decoupling limit of a heavy charged Higgs boson in the MSSM. Thus, for
values of the charged Higgs mass MH + MZ , one has O31 0, while O11 cos and
O21 sin . In this limit, the scalar components of the H1 dd and H1 uu couplings acquire
the known SM form, given by gw md /(2MW ) and gw mu /(2MW ), respectively, where

1
md = hd + hd + 1hd tan v1 ,
2

1
(4.15)
mu = hu + hu + 1hu cot v2
2
have already been defined to be positive in (4.5) and (4.7). For similar reasons, the
pseudoscalar parts of the H1 dd and H1 uu couplings vanish, as they are proportional to
O31 and Im md,u = 0. On the other hand, in the same large MH + limit, the scalar and
pseudoscalar couplings of both the two heaviest Higgs bosons H2 and H3 to the up and
down fermions do not vanish. We can therefore conclude that CP-violating self-energy and
vertex effects do not decouple in the heavy neutral Higgs sector. In the next section, we
demonstrate explicitly the aforementioned (non-)decoupling features of CP violation by
analyzing specific phenomenological examples.
5. Phenomenological discussion
In this section we discuss the phenomenological implications of radiative Higgs-sector
CP violation in the MSSM for Higgs-boson searches at high-energy colliders. We focus
our attention on the physics potential for discovering Higgs bosons with mixed CP parities
at LEP 2 and the upgraded Tevatron collider, and also comment on the enhanced search
capabilities offered by the LHC.
At the LEP 2 and Tevatron colliders, neutral Higgs bosons are predominantly produced
via the Higgs-strahlung processes in e+ e and q q collisions, such as e+ e Z
113
ZHi [103105], q q Z ZHi and q q W W Hi [106], with i = 1, 2, 3.

If the next-to-lightest Higgs boson is not too heavy, Higgs bosons can also be produced
copiously in pairs through the reactions: e+ e Z Hi Hj and q q Z Hi Hj .
In addition to the Higgs-boson masses, the Higgs-boson couplings to the gauge fields play
an essential role in our forthcoming discussion. The effective Lagrangians governing the
interactions of the Higgs bosons with the W and Z bosons are given by [25]

3
X
1
gHi V V Hi W+ W , +
H
Z
Z
,
(5.1)
LH V V = gw MW
i
2 cos2 w
i=1
LH H W =
LH H Z =
gw
2
3
X

gHi H W + Hi i H W +, + h.c.,
(5.2)
i=1
3
X

gw
gHi Hj Z Hi Hj Z ,
2 cos w
(5.3)
j >i=1
where
cos w = MW /MZ ,
, and
gHi V V = cos O1i + sin O2i ,

gHi Hj Z = O3i (cos O2j sin O1j ) O3j (cos O2i sin O1i ),
gHi H W + = cos O2i sin O1i + iO3i .
(5.4)
(5.5)
(5.6)
For completeness, we have included in (5.2) the interactions of the charged Higgs bosons
H with the neutral Higgs and W bosons. Note that the couplings Hi ZZ and Hi W + W
are related to the Hi Hj Z couplings through
gHk V V = ij k gHi Hj Z .
(5.7)
Moreover, unitarity provides the constraint

3
X
2
gH
= 1.
iV V
(5.8)
i=1
Evidently, if two Higgs-boson couplings to gauge bosons are known, this is sufficient to
determine the complete set of the couplings gHi V V and gHi Hj Z [107,108].
In the above calculation of the effective Higgs-gauge-boson couplings, we have assumed
that the dominant contributions arise from Higgs-mixing effects. As opposed to the b-quark
Yukawa case discussed in Section 4, proper vertex corrections to the Hi ZZ and Hi Hj Z
couplings do not contain strong-coupling- or tan -enhanced diagrams. Therefore, naive dimensional analysis suggests that these corrections are suppressed relative to their tree-level
values by loop factors of the kind: (3w /4)(|At |/m2t ), (3w /4)(|At |2 v 2 /m6t ) .
1
1
102 , where t1 is the heaviest stop squark (see (2.13)). In the following, we neglect proper
vertex corrections to the Hi ZZ and Hi Hj Z couplings.
For our phenomenological discussion of CP violation, we consider the following two
representative values for tan : (i) tan = 4 and (ii) tan = 20. For definiteness, unless
stated otherwise, the soft supersymmetry-breaking and parameters are set to the values
114
eQ = M
et = M
eb = 0.5 TeV,
MSUSY = M
= 2 TeV,
|At | = |Ab | = 1 TeV,
|mBe| = |mW
e | = 0.3 TeV,
|mg | = 1 TeV,
(5.9)
As can be seen in (5.9), we have chosen relatively large values for the stop and sbottom
mixing parameters At , Ab and , as well as a common left- and right-handed squark mass
MSUSY , which leads to enhanced CP-violating effects of the CP-odd phases arg(At,b) on
the Higgs sector.
As was mentioned in the introduction, large CP-odd phases may lead to rather large
EDM contributions, thereby violating the known upper bounds on the electron and neutron
EDMs de and dn : de /e < 0.5 1026 cm [55] and dn /e < 0.6 1025 cm [56] at the
2- level. One phenomenologically interesting possibility for avoiding the possible CP
crisis is to make the first two generations of squarks rather heavy with masses much above
the TeV scale [6671], keeping the third generation relatively light with masses of order
0.5 TeV. In such a scenario, CP violation may only reside in the third generation. For
our illustrations, we shall take the parameter to be real and assume that the only CPodd phases in the theory are arg(At ) = arg(Ab ) and arg(mg ). Again, as one way to avoid
the one-loop EDM constraints [7274], we have taken a gluino mass of 1 TeV in (5.9).
However, in such a scheme, one has to worry about the fact that Higgs-boson two-loop
contributions to the EDMs [8083] might still become sizeable. For the low-tan scenario
in (5.9), the two-loop EDM contributions are of the order of the experimental EDM upper
bounds mentioned above. Since these two-loop EDM effects depend almost linearly on
tan , one might then need to arrange for cancellations among the different one- and twoloop EDM terms [7274] at the level of 10% for the scenario with tan = 20. We believe
that this can be achieved without excessive fine-tuning of the CP-violating parameters of
the theory.
As was already noticed in [25,44,48,49], the radiative corrections to the lightest
Higgs boson H1 depend crucially on the stop mixing parameter |Xt | = |At cot |.
Specifically, the radiatively-corrected H1 -boson mass increases as |Xt | increases, reaching
a maximum when |Xt |/MSUSY 2.45. Then, as |Xt | further increases, the radiative
corrections to H1 -boson mass decrease and may even become negative, driving the latter to
very small, experimentally excluded values. A distinctive feature of the CP-violating SUSY
scenario compared to the CP-conserving one is that |Xt | can be increased by varying only
the phase arg(At ) from zero to higher values, but holding fixed |At | and ||. For similar
reasons, high values of |Xt | induced by large values of arg(At ) can make the mass of
the lightest stop squark t2 very low, so as to violate present experimental constraints, i.e.,
mt2 & 100 GeV. Furthermore, light stop quarks, with mt2 . 300 GeV and large |Xt | values,
can give rise to observably large contributions to the electroweak precision parameter 1
[109,110]. For the scenarios under discussion, mt2 is always larger than about 300 GeV, for
all the parameter space for which the H1 -boson mass acquires. Therefore, apart from the
bounds derived by EDM constraints, the requirement that the lightest Higgs-boson mass is
positive can be used naively to set an upper bound on the phase of At .
115
2
Fig. 1. Numerical estimates of (a) MH1 and MH2 and (b) gH
as functions of arg(At ), for the
i ZZ
indicated choices of MSSM parameters. Solid lines correspond to arg(mg ) = 0, dashed lines to
arg(mg ) = 90 .
In Fig. 1 we give numerical predictions for the two lightest Higgs-boson masses MH1
and MH2 , and for the three relevant Hi ZZ couplings squared as a function of arg(At ),
for two different values of arg(mg ): arg(mg ) = 0 (solid lines) and arg(mg ) = /2 (dashed
lines). We first discuss the scenario with tan = 4, for which the values of the remaining
116
soft supersymmetry-breaking parameters and are given in (5.9). Since our interest is
to analyze dominant CP-violating effects for a light Higgs sector, we present predictions
for a relatively small charged-Higgs-boson mass, MH + = 150 GeV. In the CP-conserving
limit of the theory (arg(At ) = 0), the mass MH1 of the lightest neutral Higgs boson
2
, is approximately
is close to 85 GeV, while the square of the H1 ZZ coupling, gH
1 ZZ
equal to 0.8. These values of masses and couplings are now excluded by Higgs-boson
searches at LEP 2 [54]. However, this situation changes crucially once CP-violating
phases become relevant. As the phase of At increases, two important effects take place.
First, as was mentioned above, the stop mixing parameter |Xt | becomes larger, giving
rise to larger H1 -boson masses. Second, the mass-matrix terms describing the scalar
pseudoscalar mixing are enhanced, thereby effectively leading to large modifications in
the couplings of the Higgs bosons to gauge bosons. This second effect can be attributed
entirely to CP violation. In fact, as can be seen from Fig. 1(b), for arg(At ) 80 degrees,
2
gets very suppressed, implying that LEP 2 cannot detect the Higgs boson H1 via
gH
1 ZZ
+
e e Z ZH1 . On the other hand, for the same range of values of arg(At ), i.e.,
arg(At ) = 80 95 , the H2 and H3 bosons have significant couplings to the Z bosons.
Although the H3 boson is too heavy to be detected at LEP 2 in this case, the H2 boson has
2
0.80.6, which may be probed at LEP 2 in this years
a mass of 105110 GeV and gH
2 ZZ
run. For larger values of the phase of At , 95 < arg(At ) < 110 , the discovery of a Higgs
boson at LEP 2 is more challenging. The lightest Higgs boson H1 acquires a mass below
90 GeV, but the H1 ZZ coupling is too small to allow experimental detection through the
reaction e+ e Z ZH1 . In addition, the H2 boson becomes too heavy to allow for
2
. 0.7.
discovery via e+ e Z ZH2 , with gH
2 ZZ
As was discussed in [25], the H1 and H2 bosons may also be searched for in the
2
2
= gH
0.2 almost
channel e+ e Z H1 H2 . Since the squared coupling gH
1 H2 Z
3 ZZ
independently of arg(At ), a careful experimental analysis of the parameter region of
interest will be necessary to determine whether the two lightest Higgs bosons can be
observed for such large mass differences (MH1 MH2 > 40 GeV) and such small gH1 H2 Z
couplings.
In Fig. 1 we also present predictions for a gluino phase of 90 degrees (dashed lines).
Since the vertex corrections are generally small for low or moderate values of tan , they
are expected to induce only small corrections to the Higgs-boson masses and mixings. This
2
2
(gH
) gets slightly smaller
last fact is reflected in Fig. 1, even though the coupling gH
1 ZZ
2 ZZ
(larger) for larger values of the phase of At .
Fig. 2 shows the changes in the predictions for the same choice of parameters as in
Fig. 1, but with MSUSY , and |At | = |Ab | rescaled by a factor of 2. This rescaling leads
2
exhibits
to a slight increase (decrease) of the Higgs-boson mass MH1 (MH2 ), while gH
1 ZZ
a slightly different quantitative dependence on the phase of At , especially for the region in
2
is very small. Thus, although MH1 becomes small for arg(At ) > 115 degrees,
which gH
1 ZZ
the H1 ZZ coupling gets sizeable again, well within the capabilities of LEP 2 to test.
We now investigate more quantitatively the predictions of a large-tan scenario for the
Higgs-boson masses and couplings. We adopt the scenario given in (5.9), with tan = 20
and MH + = 150 GeV. In this large-tan scenario, one has || cot |At |, and the
117
Fig. 2. As Fig. 1, but with = 4 TeV, |At | = |Ab | = 2 TeV and MSUSY = 1 TeV.
effective stop mixing parameter |Xt | |At | is almost independent of arg(At ). Therefore,
as is seen in Fig. 3(a), the Higgs-boson masses MH1 and MH2 do not exhibit any significant
variation as a function of arg(At ). In contrast to the Higgs-boson masses, Fig. 3(b)
2
on arg(At ).
shows that there is a non-trivial dependence of the squared couplings gH
i ZZ
Furthermore, the next-to-lightest Higgs boson H2 is heavy enough to render its search
118
Fig. 3. As in Fig. 1, but with tan = 20.
through the e+ e Z ZH2 reaction kinematically inaccessible at LEP 2. For similar

reasons, we find that, for all values of arg(At ), the H1 and H2 bosons are rather too heavy
to be produced via the e+ e Z H1 H2 channel at LEP 2. As a result, Higgs-boson
searches at LEP 2 tend to be more efficient for small values of the At phases, for which
2
the lightest neutral Higgs-boson mass is close to 100 GeV and gH
is non-negligible
1 ZZ
119
2
(gH
0.3). In addition, for large At phases, arg(At ) & 80 , the H1 V V coupling (with
1 ZZ
V = Z, W ) is rather suppressed, so that the H1 Higgs boson, although it becomes lighter
with a mass in the range 9095 GeV, will be elusive at LEP 2, and may also escape
detection via the corresponding channel at the upgraded Tevatron. However, the next-tolightest Higgs boson H2 has couplings of order unity to the Z and W bosons. Present
simulations show that a neutral Higgs boson, such as H2 , with MH2 180 GeV and a
SM-like coupling strength to vector gauge bosons can be tested at the Tevatron collider
with a total integrated luminosity of 10 fb1 . However, discovery of such a Higgs boson
at the 5- level would demand a total integrated luminosity of 30 fb1 , and would have a
reach up to MH2 120 GeV [111]. Finally, even though the H1 H3 Z coupling is close to
2
2
= gH
) for arg(At ) > 100 , the Higgs-pair production of H1 and H3 is
unity (gH
1 H3 Z
2 ZZ
not kinematically allowed at LEP 2, since MH3 MH + . Further studies will be necessary
to investigate the potential of this production mechanism at the Tevatron.
It is interesting to present predictions for the neutral Higgs-boson masses and their
couplings to gauge bosons for lower values of the charged Higgs-boson mass MH + in the
above large-tan scenario. In Fig. 4, we plot numerical estimates for the same kinematic
parameters as in Fig. 3, but with MH + = 135 GeV. In this case, the H1 -boson mass varies
approximately between 80 and 65 GeV, and the H1 ZZ coupling rapidly decreases as the
phase of At increases. The two heaviest neutral Higgs bosons H2 and H3 have masses in
the range between 120 and 130 GeV. Hence, these two Higgs bosons cannot be produced
via e+ e Z ZH2 or e+ e Z ZH3 at LEP 2. However, the H2 and H3
bosons may still be accessed via e+ e Z H1 H2 or H1 H3 . Interestingly, the squared
2
2
2
2
= gH
and gH
= gH
exhibit a cross-over as a function of
couplings gH
1 H2 Z
3 ZZ
1 H3 Z
2 ZZ
arg(At ). The crossing point of the two squared couplings is when arg(At ) 90 . For
arg(At ) = 180 , one of the squared couplings goes to 0 and the other to 1, depending on
the phase of the gluino mass. In this case, the two heaviest neutral Higgs bosons become
almost degenerate in mass. For the whole range of values of arg(At ), either the H2 or H3
Higgs boson can be tested at the upgraded Tevatron collider provided a total integrated
luminosity of 10 fb1 per detector is available [111].
Figs. 5(a) and (b) show the degree of mass splitting between the two heaviest neutral
Higgs bosons H2 and H3 , for the same choice of parameters as in Figs. 1 and 2, but for
MH = 200, 300, 400, 500 GeV. As was already observed in [24,25], even though the H2
and H3 bosons are almost degenerate in the CP-conserving limit of the theory, they can
have a degree of splitting up to 30% for a maximal CP-violating phase arg(At ) 90 . The
comparison of the Fig. 5(a) with (b) reveals that this last result is almost independent of
the common scale factor of MSUSY , and |At |. Also, the degree of mass splitting is not
much affected by the value of the gluino phase arg(mg ): the results for arg(mg ) = 90 are
slightly higher than those of arg(mg ) = 0. In this vein, it is interesting to mention that large
CP-violating scalar-pseudoscalar mixings can lead to observable phenomena of resonant
CP violation at high-energy colliders [112125].
In Figs. 6 and 7, we examine the behaviours of the scalar and pseudoscalar parts of
the H1 bb coupling as functions of the CP-odd phase arg(At ), for two different charged
Higgs-boson masses, MH + = 150 and 300 GeV. As was done in [25], we find that the
120
Fig. 4. As in Fig. 3, but with MH + = 135 GeV.
best way of analyzing such a behaviour is in terms of the CP-even and CP-odd quantities:
S
S
S
P
P
)2 + (gH
)2 ] and 2gH
g P /[(gH
)2 + (gH
)2 ]. For example, Higgs-boson
[(gH
1 bb
1 bb
1 bb H1 bb
1 bb
1 bb
branching ratios are proportional to the first quantity, while the second one will only
S
S
P
g P /[(gH
)2 + (gH
)2 ]
occur in CP-violating observables. In other words, 2gH
1 bb H1 bb
1 bb
1 bb
gives a measure of the CP-violating component in the H1 bb coupling. If we compare the
121
Fig. 5. Numerical estimates of (a) MH1 and (b) 2|MH2 MH3 |/(MH2 + MH3 ) as functions of the
CP-violating phase arg(At ). Lower values of the same line type correspond to arg(mg ) = 0, the
higher ones to arg(mg ) = 90 .
predictions for MH + = 150 GeV with those for MH + = 300 GeV in Figs. 6 and 7, we
find that the CP-violating component of the H1 bb coupling reduces in magnitude, for large
values of the charged Higgs-boson mass. Such a decoupling behaviour of the CP-violating
H1 bb component is in agreement with our observation, which we already made at the end
122
S
S
S
P
P
Fig. 6. Numerical estimates of (a) (gH
)2 + (gH
)2 and (b) 2|(gH
)(gH
)|/[(gH
)2 +
1 bb
1 bb
1 bb
1 bb
1 bb
P
2
(gH bb ) ] versus arg(At ). Solid lines correspond to arg(mg ) = 0, dashed ones to arg(mg ) = 90 .
1
of Section 4. From Figs. 6(b) and 7(b), we see that the impact of the gluino phase on
the CP-violating component of the H1 bb coupling is more important for the large-tan
scenario. This may be attributed to the fact that the radiatively-induced term 1hb tan ,
which crucially depends on the gluino phase and tan , has a dominant contribution to the
H1 bb coupling.
123
Fig. 7. As in Fig. 6, but with tan = 20.
There can be a cancellation or a strong suppression of the coupling of the lightest Higgs
boson H1 to the bottom quarks, depending on the magnitude of the CP-violating phases
and of the products At , Ab and mg . This cancellation usually takes place for moderate
values of the charged Higgs mass and large values of tan . Such an effect is also present in
the CP-conserving case, for specific signs and magnitudes of the above products involving
124
S
P
Fig. 8. Numerical estimates of (a) MH1 and MH2 and (b) [(gH
)2 + (gH
)2 ] versus arg(At ),
1 bb
1 bb
e
e
e
with soft squark masses: MQ = Mb = 0.6 TeV and Mt = 0. Solid lines are for arg(mg ) = 90 and
dashed lines for arg(mg ) = 0 .
the trilinear terms At,b and the gluino mass, and has been discussed in detail in [126,127].
Fig. 8(b) illustrates such a cancellation for the CP-violating SUSY model under discussion.
For example, we observe that for the set of SUSY parameters considered in Fig. 8, the
H1 bb coupling can be strongly suppressed for arg(At ) = arg(Ab ) 15 . Moreover, we
125
have checked that for this same set of parameters the H1 ZZ coupling is almost SM-like.
In addition, Fig. 8(a) shows that the mass of the lightest Higgs boson is of order 105
GeV, practically independent of the CP-violating phase. This is therefore an extremely
interesting example, since the lightest Higgs mass is in the mass range that may be within
the reach of LEP 2, and its production cross section will be SM-like. However, the main
can be strongly suppressed if the CP-violating phases arg(At )
decay channel, H1 b b,
and arg(mg ) lie in a specific range. Therefore, in such a scenario, the detection of the H1
boson may in principle be impossible, even in the final run of LEP 2. To make a conclusive
statement on this possibility, one should study in detail the capability of LEP 2 to detect
such a light H1 boson via its decays into pairs, or into other hadronic modes. Most
intriguingly, the set of parameters considered in this example also allow for a light righteQ , of the type necessary to
handed stop squark and moderate mixing parameter, |Xt |/M
allow the possibility of electroweak baryogenesis [128144]. Hence, if such a Higgs boson
cannot be discovered at LEP 2 via other decay channels, the final phase of LEP 2 will
leave an open window for electroweak baryogenesis. A careful study of the CP-violating
phases required for electroweak baryogenesis and the detection capabilities of LEP 2 for
alternative decay modes becomes essential for testing this exciting scenario.
We shall briefly comment on the enhanced LHC capabilities for Higgs-boson searches
[145153]. The LHC has a considerably higher reach than LEP 2 and the upgraded
Tevatron collider in the search for heavier Higgs bosons, and hence has more chances
to unravel the complete Higgs-boson spectrum of the MSSM with explicit CP violation.
At the LHC, Higgs bosons may be copiously produced via a wide variety of processes
which depend in many different ways on the couplings of the neutral and charged Higgs
bosons both to gauge bosons and fermions [145153]. In the case of the CP-violating
version of the MSSM under study, we have shown that mixing between states with different
CP parities can dramatically modify those couplings and, hence, importantly affect the
associated production and decay mechanisms. Studies including CP-violating effects on
the gluon-fusion production of Higgs bosons at the LHC have already been considered in
the literature [154,155]. Our work provide the basic tool to improve further those studies,
and to perform a complete analysis of the CP-violating effects on the many other Higgsboson search mechanisms available at LHC. The LHC, together with the information
gathered from experiments at LEP 2 and the upgraded Tevatron collider, will be capable
of providing a thorough test of the MSSM Higgs sector and shed light on the possibility of
explicit radiative breaking of CP invariance in supersymmetry.
Finally, it is interesting to make a comparative analysis between our results and those
obtained previously in [25]. In the latter work, the effective RG-improved potential was
expanded up to renormalizable operators of dimension 4. The expansion was performed
in powers of the stop-mass splitting, m2t m2t , relative to the arithmetic average of the
1
2
= 12 (m2t + m2t ). Moreover, the two-loop effect originating
squared stop masses, MSUSY
1
2
from the one-loop radiative corrections to the Yukawa couplings of the top and bottom
quarks was not taken into account in the computation of the effective potential in [25].
Nevertheless, for moderate values of tan and for all soft supersymmetry-breaking masses
equal to MSUSY , the deviations of the results presented in [25] with our results are expected
126
2
Fig. 9. Numerical predictions for MH1 and gH
as obtained by the present complete RG approach
1 ZZ
(solid lines) and the operator-expansion method of [25] (dashed lines).
to be small, for small values of the stop-mixing parameter Xt = |At cot |; the
deviations will only grow for increasing values of Xt . For larger values of tan , instead, the
impact of the bottom-mass quantum corrections, which were omitted in [25], is significant.
In fact, only in the limit of small values of ||, in which case the bottom-mass corrections
are small, are both approaches guaranteed to give comparable numerical estimates.
127
In Fig. 9, we show the predictions for the mass of the lightest neutral Higgs boson
H1 and its coupling to Z bosons, as obtained obtained by our RG approach (solid lines)
and the operator-expansion method of [25] (dashed lines). For the sake of comparison,
we consider the same input parameters as those chosen in Fig. 1, for vanishing gluino
phase and three different values of the charged Higgs-boson mass: MH + = 150, 200
and 500 GeV. As was discussed above, we find that the predictions of the two works
are in excellent agreement with one another for small values of Xt . For large values of
Xt , instead, we observe larger quantitative differences in the results obtained by the two
approaches, even though the qualitative behaviour of the two predictions exhibits a quite
analogous functional dependence.
Our one-loop RG-improved approach overcomes the limitations present in earlier
analyses. In particular, our RG approach allows for a rather precise determination of the
radiative effects on a generic Higgs-boson mass spectrum, even in cases of large stop
mixings and/or large hierarchies between the left- and right-handed stop masses. Also,
within our RG approach, the important effects of the one-loop corrections to the quark
Yukawa couplings are incorporated in the computation of the Higgs-boson masses and in
their respective Higgs-boson couplings to gauge and fermion fields. A Fortran code that
computes the Higgs-boson masses and couplings, including all the CP-violating effects as
presented in this work, may be found in [156].
6. Conclusions
We have performed a complete one-loop RG improvement of the effective Higgs
potential in the MSSM, in which CP violation is induced radiatively by soft CP-violating
trilinear interactions that involve the Higgs fields and the stop and sbottom squarks. Earlier
studies [2426] of the neutral Higgs-boson mass spectrum were based on a number of
particular assumptions and/or kinematic approximations. The present work goes well
beyond those studies, and extends the most detailed analysis [25], in which the oneloop RG-improved effective potential was expanded up to renormalizable operators of
dimension 4, assuming a moderate mass splitting among the stop squarks. This assumption
seemed to impose a serious limitation, given that CP-violating effects exhibit an enhanced
behaviour for large values of the stop-mixing parameter Xt . The results obtained using
the present RG approach confirm, however, the qualitative phenomenological features
found in [25], for the Higgs-boson masses and their couplings to fermions and to the W
and Z bosons. It offers very accurate predictions, at the same level as the most accurate
calculations in the CP-conserving case (implying an uncertainty of order 3 GeV) [51,100,
101], for the Higgs-boson masses and for the whole range of the MSSM parameter space
in the presence of non-trivial CP-violating phases. More specifically, the present study
also includes two-loop leading logarithms associated with QCD effects and t- and b-quark
Yukawa couplings. It also contains all dominant two-loop non-logarithmic contributions
to the one-loop effective potential, which are induced by one-loop threshold effects on the
t- and b-quark Yukawa couplings due to the decoupling of the third-generation squarks
128
(as considered in [100,101] in the CP-conserving limit). These one-loop threshold terms,
hu,d and 1hu,d , strongly depend on the phase of the gluino mass and so introduce new
CP violation into the MSSM effective potential at the two-loop level.
Large radiative effects of CP violation in the Higgs sector of the MSSM can
have important phenomenological consequences on Higgs-boson searches at LEP 2, the
Tevatron and the LHC. We have explicitly demonstrated that the radiatively-induced CP
violation in the MSSM Higgs potential can lead to important effects of mass and coupling
level crossing among the three neutral Higgs particles. These CP-violating effects of level
crossing in the Higgs sector modify drastically the Higgs-boson couplings to the up- and
down-type quarks, and to the W and Z bosons. In particular, CP violation in the lightest
Higgs sector becomes relevant for a relatively light charged Higgs boson, with MH + . 160
GeV. For instance, for MH + = 150 GeV and tan = 4, even a neutral Higgs boson as light
as 60 GeV may escape detection at LEP 2. However, the upgraded Tevatron may have the
physics potential to explore such CP-violating scenarios at low tan values, which may
remain at the edge of accessibility even during the final LEP 2 run.
We have also studied the effects induced by a non-trivial CP-odd gluino phase, which
enters the effective potential at the two-loop level. The presence of a gluino phase gives rise
to small but non-negligible changes in the Higgs-boson mass spectrum and in the couplings
of the Higgs fields to the W and Z bosons. In this context, we have also found that the
product of the scalar times the pseudoscalar coupling of the lightest Higgs boson H1 to the
bottom quarks has a very strong dependence on the gluino phase. This product of couplings
gives a measure of CP violation in the H1 bb coupling, which can even be of order unity
for relatively small charged Higgs-boson masses.
Finally, it is worth stressing that CP violation decouples from the lightest Higgs sector in
the large-mass limit of a heavy charged Higgs boson. This decoupling property, which was
known to hold for the CP-violating self-energy effect [24,25], has now been shown to be
valid for the CP-violating vertex effects as well. As a result, the predictions for the lightest
Higgs-boson mass and its couplings to gauge bosons in the above decoupling regime of the
theory will be practically identical to the corresponding predictions in the CP-conserving
case. However, unlike the lightest Higgs sector, CP violation does not decouple from the
heaviest Higgs sector in the MSSM, opening up new possibilities for studying enhanced
effects in CP-violating Higgs scalar-pseudoscalar transitions at the LHC or future muon
colliders [112125], where the heavy MSSM Higgs bosons can be resonantly produced.
In conclusion: the present analysis has shown that the MSSM with explicit radiative
breaking of CP invariance constitutes a very rich theoretical framework, introducing new
challenges in the search for fundamental Higgs scalars at LEP 2, the Tevatron and the LHC.
Acknowledgements
We thank Manuel Drees for useful discussions. A.P. thanks the theory group of Fermilab
for the kind hospitality extended to him while part of this work was done. Work supported
in part by the U.S. Department of Energy, High Energy Physics Division, under Contract
W-31-109-Eng-38.
129
Note added
After completion of the work described here, we saw [157], which also computes
the one-loop RG-improved effective potential of the MSSM with explicit CP violation.
Here we further (i) perform a more complete RG improvement of the effective potential,
(ii) develop a self-consistent treatment of the whole MSSM Higgs sector, taking into
account the crucial one-loop relation between the charged Higgs-boson and neutral Higgsboson mass matrices, and (iii) include the dominant two-loop non-logarithmic corrections
to the effective potential, which may also have an important impact on the H1 bb coupling.
In the limits where a comparison between the results was possible, we find reasonable
agreement with [157].
Appendix A. Derivatives of background-field-dependent masses
In this appendix, we list analytic expressions pertaining to derivatives of quark and
squark masses with respect to their background Higgs fields. These derivative expressions
are very useful, as they constitute the building blocks of the general one-loop effective
potential presented in Section 2. We have divided the appendix into three subsections. In
the first subsection, we list the derivatives of quark masses with respect to Higgs fields,
while the next two subsections contain the corresponding expressions for derivatives of
squark masses with respect to neutral and charged Higgs fields, respectively.
A.1. Quark derivatives
First, we give the derivatives related to non-vanishing tadpole contributions:
2
2
b
1 m
t
1 m
2
(A.1)
= |ht | ,
= |hb |2 ,
v2 2
v1 1

where the operation denotes that the above expressions should be evaluated in the
ground state of the Higgs potential.
Then, the self-energy-type derivatives involving neutral and charged Higgs bosons may
be listed as follows:
2 2 2 2
2 2 2 2
m
b
b
m
t
t
m
m
2
=
= |ht | ,
=
= |hb |2 ,
2
2
2
2
a2
1
a12
2 2
2 2
m
b
|hb |2 m2b
t
|hb |2 m2t
m
,
,
=
=
1+ 1
1+ 1
m2t m2b
m2b m2t
2 2
2 2
m
|ht |2 m2b
b
t
|ht |2 m2t
m
,
,
=
=
2+ 2
2+ 2
m2t m2b
m2b m2t
2 2
2 2 2 2
b
m
t
t
m
m
=
=
1+ 2
2+ 1
1+ 2
2 2
b
m
|ht hb |mt mb
.
(A.2)
=
=
+
2 1
m2b m2t
130
Note that we have not listed derivatives that vanish.

A.2. Derivatives of squark masses with respect to neutral Higgs fields
e2b1,2
This section contains the derivatives of the field-dependent squark masses m
e2t1,2 and m
with respect to neutral Higgs fields 1,2 and a1,2 . We first give the tadpole terms:
2
et1 (t2 )
g 2 + g 02
1 m
= w
v1 1
8

1
2
et2 + 1 xt v 2 cos 2
eQ
xt M
M
+ ()
2
2
2
2 mt mt
1
2

2|ht |2 Re(At ) tan ||2 ,
2

eb1 (b2 )
g 2 + g 02
1 m
= |hb |2 w
v1
1
8

1
2
eb2 1 xb v 2 cos 2
eQ
xb M
M
(+)
2
2
2
2 m m
b1
b2

+ 2|hb |2 Re(Ab ) tan |Ab |2 ,
2
et1 (t2 )
g 2 + g 02
1 m
= |ht |2 w
v2 2
8

1
2
eQ
xt M
M
(+)
2
2
2
2 mt mt
1
2

+ 2|ht |2 Re(At ) cot |At |2 ,
2

eb1 (b2 )
g 2 + g 02
1 m
= w
v2
2
8

1
2
eb2 1 xb v 2 cos 2
eQ
xb M
M
+ ()
2
2
2
2 m m
b1
b2

2|hb |2 Re(Ab ) cot ||2 ,
2
2
et1 (t2 )
et1 (t2 )
1 m
|ht |2 Im(At )
1 m
,
=
= (+)
v2
a1
v1
a2
m2t m2t
1
2
2

2

eb1 (b2 )
eb1 (b2 )
1 m
|hb |2 Im(Ab )
1 m
,
(A.3)
=
= (+)
v2
a1
v1
a2
m2 m2
b1
b2
where the coupling parameters xt and xb are defined after (2.13). Then, the self-energye2qk /ai aj i, with qk = t1 , b1 , t2 , b2 and i, j = 1, 2, are found to be
type terms h 2 m
2 2
2 + g 02
et1 (t2 )
m
gw
=
8
a12

1
2
et2 + 1 xt v 2 cos 2 + 2|ht |2 ||2 ,
eQ
xt M
M
+ ()
2
2
2
2 mt mt
1
e2b1 (b2 )
2m
2 + g 02
gw
8

1
2
eb2 1 xb v 2 cos 2 2|hb |2 |Ab |2 ,
eQ
xb M
M
(+)
2
2
2
2 m m
= |hb |2
a12

e2t1 (t2 )
2m
= |ht |2
a22
e2b1 (b2 )
2m
a22
b1
e2t1 (t2 )
2m
e2b1 (b2 )
2m
= (+)
1
m2
b2
m2
b1
= (+)

2
eb2 1 xb v 2 cos 2 + 2|hb |2 ||2 ,
eQ
xb M
M
2
2v1 v2 |ht |4 Im2 (At )

|ht |2 Re(At )
+ ()
,
3
2
2
mt mt
m2 m2
1
a1 a2
b2
+ g 02
g 2 + g 02
= w
8
a1 a2
2
gw
8

1
2
et2 + 1 xt v 2 cos 2 2|ht |2 |At |2 ,
eQ
xt M
M
(+)
2
2
2
2 mt mt
+ ()

131
|hb
t1
|2 Re(A
b)
m2
b2
m2
b1
+ ()
2v1 v2 |hb
t2
|4 Im2 (A
3
m2
b2
m2
b1
b)
(A.4)
e2qk /i aj i are given

In addition, the non-vanishing CP-violating self-energy terms h 2 m
by

e2t1 (t2 )
2m

= +()
1 a2
|ht |2 Im(At )
m2t m2t

1
v12

2
eQ
M
x
M
t

2
2 2
m2t mt
1
2|ht | Re(At ) tan ||

2
e2b1 (b2 )
2m

= +()
1 a2
|hb |2 Im(Ab )
m2 m2

1+
b1
b2
v12
m2 m2
b1

2
eb2 1 xb v 2 cos 2
eQ
M
2 xb M
2
b2
+ 2|hb |2 Re(Ab ) tan |Ab |2

e2t1 (t2 )
2m
2 a1

= (+)

,
|ht |2 Im(At )
m2t m2t
1

,
132

1+

2
eQ
M
x
M
t

2
2 2
v22
m2t mt
1
+ 2|ht | Re(At ) cot |At |

2
e2b1 (b2 )
2m

= (+)
2 a1

,
|hb |2 Im(Ab )
m2 m2
b1

1
b2

2
eb2 1 xb v 2 cos 2
eQ
M
2 xb M
2
v22
m2 m2
b1
b2
2|hb |2 Re(Ab ) cot ||2

(A.5)
e2qk /i j i have been calcuFinally, the CP-conserving self-energy-type derivatives h 2 m

lated to be

e2t1 (t2 )
2m
12

=
2 + g 02
gw
8
+ ()
1
2
m2t
2
m2t
1

2
et2 + 1 xt v 2 (1 + 2 cos 2)
eQ
xt M
M
2
+ 2|ht |2 ||2
(+)
v12

2m
e2b1 (b2 )
= |hb |2
12

2
eQ
M
xt M

2
3
2
2 m2t mt
1
2|ht |2 Re(At ) tan ||2
(+)
2
gw
8
1
m2
b1
m2
b2
v12
2 m2 m2
b1

e2t1 (t2 )
2m
= |ht |2
22
+ g 02

2
eb2 1 xb v 2 (1 + 2 cos 2)
eQ
xb M
M
2
2|hb |2 |Ab |2
(+)
2

2
eb2 1 xb v 2 cos 2 ,
eQ
M
3 xb M
2
b2
+ 2|hb |2 Re(Ab ) tan |Ab |2
2
gw
2
+ g 02
8

1
2
et2 + 1 xt v 2 (2 cos 2 1)
eQ
xt M
M
(+)
2
2
2
2 mt mt
1
2

2|ht |2 |At |2
(+)
v22

e2b1 (b2 )
2m
=
22
v22
2 m2 m2
2m
e2b1 (b2 )
1 2

2
eb2 1 xb v 2 cos 2
eQ
M
3 xb M
2
b2
2|hb |2 Re(Ab ) cot ||2

1
12 xt2 v 2 sin 2 + 2|ht |2 Re(At )
2
2
mt
2

v1 v2
2
eQ
M
+ ()
3 xt M
2
2
2
2 mt mt
= (+)
2
m2t
1

1
2
eb2 1 xb v 2 (2 cos 2 1)
eQ
xb M
M
2
2 m2 m2
b1
b2

+ 2|hb |2 ||2
b1
1 2
2
2 + g 02
gw
8
(+)

e2t1 (t2 )
2m
+ 2|ht |2 Re(At ) cot |At |2
+ ()

2
eQ
M
x
M
t

2
2 3
2 m2t mt
1
133
2|ht |2 Re(At ) tan ||2

2
eQ
M
xt M
2

+ 2|ht |2 Re(At ) cot |At |2 ,

1
1 xb2 v 2 sin 2 + 2|hb |2 Re(Ab )
2 m2 m2 2
b1
b2

v1 v2
2
eb2 1 xb v 2 cos 2
eQ
M
+ ()
3 xb M
2
2
2
2 m m
= (+)
b1
b2
+ 2|hb |2 Re(Ab ) tan |Ab |2

2
eb2 1 xb v 2 cos 2
eQ
M
xb M
2

2|hb |2 Re(Ab ) cot ||2 .
(A.6)
A.3. Derivatives of squark masses with respect to charged Higgs fields

Here we evaluate the derivatives of the field-dependent squark masses with respect to
e2qk /i+ j i directly turns out to
charged Higgs fields. To calculate the expressions h 2 m
be a formidable task. The reason is that m
e2qk are the eigenvalues of a non-trivial (4 4)
2
f (cf. (2.11)), and their analytic form is very complicated. Therefore,
squark mass matrix M
we proceed differently, using a mathematical trick which was first applied in [88].
First, we notice that h m
e2qk /i i = 0, as a consequence of the fact that the true ground
state of the effective potential should conserve charge. Then, one may make use of the
134
eigenvalue equation:

f2 m
e2qk 14 = m
m6qk + B m
e4qk + C m
e2qk + D = 0,
e8qk + Ae
det M
(A.7)
with
f2 ,
A = Tr M

1
f2 Tr M
f4 ,
B = Tr2 M
2

1
f2 Tr2 M
f6 1 Tr M
f4 ,
f2 Tr M
f2 Tr M
C = Tr3 M
3
2

1
2
8
f
f6 + BTr M
f4 + CTr M
f2 ,
f
D = det M = Tr M + ATr M
4
to obtain

2 2
eqk
e6qk + Bij m
e4qk + Cij m
e2qk + Dij
Aij m
m

,
=
Q
e2ql
m
e2qk m
i+ j
(A.8)
(A.9)
ql 6=qk
with ql , qk = t1 , b1 , t2 , b2 , and Aij = 2 A/i+ j , Bij = 2 B/i+ j , etc. For our

purposes, it is sufficient to calculate the derivatives with respect to 1+ and 2 . In particular,
it proves convenient to use a representation in which the columns and rows 2 and 3 of
f2 have been interchanged. With such a reordering, we find
the (4 4) matrix M

2

f2

f
f+
2
M
0M
f Mt 0 2 ,
,
M
fb
0 0
0 M
1+

f2
M
0 0
,
(A.10)
f
M 0
2
f2 are the usual t- and b-(2
f2t and M
2)-mass matrices, respectively, and
where M
b

1
1 2
2

g
|h
|
h
A
v
b
1
b b
2 w
,
f+ = 2
M
1
ht
ht hb v2
2

1 2
1
2

g
|h
h
A
v
t
2
t t
2 w
.
f = 2
(A.11)
M
1
hb
ht hb v1
2
Then, the relevant coefficients A12 , B12 , C12 and D12 may be expressed in a compact form
as follows:
hA12 i = 0,

f ,
f+ M
hB12 i = Tr M

f Tr M
f+ M
f
f2t + Tr M
f2b Tr M
f+ M
f2t M
hC12 i = Tr M

f M
f+ ,
f2 M
Tr M
b
135

f+ M
f2 M
f Tr M
f+ M
f Tr M
f M
f+
f2t M
f4t M
f4 M
hD12 i = Tr M
b
b

f2 Tr M
f+ M
f + Tr M
f M
f+
f2t M
f2 M
f2t + Tr M
Tr M
b

1 f4
f .
f4
f2
f2 2 Tr M
f+ M
+ Tr M
t + Tr Mb Tr Mt + Tr Mb
2
(A.12)
e2qk /1+ 2 .
With the help of (A.12), it is straightforward to obtain the derivatives 2 m
More explicitly, we have
2 2
hB12 im4t + hC12 im2t + hD12 i
et1
m
1
1
,
=
1+ 2
m2t m2 m2t m2t m2t m2

e2t2
2m
1+ 2
b1
b2
hB12 im4t + hC12 im2t + hD12 i

2
2
,
= 2
mt m2 m2t m2t m2t m2

e2b1
2m
=
1+ 2

e2b2
2m
=
1+ 2
b1
b2
hB12 im4 + hC12 im2 + hD12 i

b1
b1

,
m2 m2t m2 m2t m2 m2
b1
b1
2
4
hB12 im + hC12 im2
b2
b2

m2 m2 m2 m2t
b2
b1
b2
2
1
b1
b2
+ hD12 i
m2 m2t
b2
.
(A.13)
Appendix B. Higgs-boson masses and mixing angles

Here we present analytic expressions for the Higgs-boson masses MHi (mt ) (i = 1, 2, 3)
and the corresponding (3 3) orthogonal matrix O, after diagonalizing the RG-improved
Higgs-boson mass matrix (M2N )(mt ).
For notational simplicity, we do not display explicitly the functional dependence of M2N
on mt . The mass eigenvalues of the (3 3) matrix M2N are then obtained by solving the
characteristic equation of cubic order:
with
x 3 + rx 2 + sx + t = 0,
(B.1)

r = Tr M2N ,

1
s = Tr2 M2N Tr M4N ,
2

t = det M2N .
(B.2)
To this end, it proves useful to define the following auxiliary parameters:

3s r 2
,
3
2r 3 rs
+ t,
q=
27
3
p3 q 2
+ .
D=
27
4
p=
(B.3)
136
To ensure that the three eigenvalues are positive, it is necessary and sufficient to require
that
D < 0,
r < 0,
s > 0,
t < 0.
(B.4)
Imposing these inequalities on the kinematic parameters of the theory, we may express the
three mass eigenvalues of M2N as

p
1
2
2
+
,
MH1 = r + 2 p/3 cos
3
3
3

p
1
2
2
= r + 2 p/3 cos
,
MH
2
3
3
3

p
1
2
,
(B.5)
= r + 2 p/3 cos
MH
3
3
3
with

q
p
and 0 6 6 .
= arccos
2 p3 /27
(B.6)
Since the Higgs-boson mass matrix M2N is symmetric, we can diagonalize it by means
of an orthogonal rotation O as stated in (3.30). Furthermore, one can show [158] that the
Higgs-boson mass eigenvalues in (B.5) satisfy the desired mass hierarchy in accordance
with the inequality of (3.31).
If Mij2 , with i, j = 1, 2, 3, denote the matrix elements of M2N , the elements Oij can then
be obtained by appropriately solving the underdetermined coupled system of equations,
P
2
2
k Mik Okj = MHj Oij :

2
2
2
2
MH
O2i + M13
O3i = 0,
O1i + M12
M11
i

2
2
2
2
M21 O1i + M22 MHi O2i + M23 O3i = 0,

2
2
2
2
O1i + M32
O2i + M33
MH
O3i = 0.
M31
i
More explicitly, we have
|x1 |/11 x2 /12

O = y1 /11 |y2 |/12
z1 /11
z2 /12
where
1i =
and
(B.7)
x3 /13
y3 /13 ,
|z3 |/13
(B.8)
q
xi2 + yi2 + zi2

M 2 M 2
M2

|x1 | = 22 2 H1 2 23 2

M32
M33 MH1
2

2
2
M M22
MH
1 ,
z1 = sx1 21
2
2

M31
M32
(B.9)

,

y1 = sx1

2
2
M23
M21

2
2
2 ,
M33
MH
M31
1

2
2

M13
M12

x2 = sy2 2
2
2 ,
M33 MH
M32
2
2

2
2
M12 M11
MH

,
2
z2 = sy2 2
2

M32
M31

2

M2
M13

x3 = sz3 2 12 2
2 ,
M22 MH3 M23

M 2 M 2
M2

|z3 | = 11 2 H3 2 12 2

M21
M22 MH3
137

M 2 M 2
M2

|y2 | = 11 2 H2 2 13 2

M31
M33 MH2
2
2
2
M M11
MH
3
y3 = sz3 13
2
2
M23
M21

,

(B.10)
In (B.10), the abbreviation sx sign(x) is an operation that simply gives the sign of a real
expression x.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
J.H. Christenson, J.W. Cronin, V.L. Fitch, R. Turlay, Phys. Rev. Lett. 13 (1964) 138.
P.K. Kabir, The CP Puzzle, Academic Press, London, 1968.
W. Grimus, Fortschr. Phys. 36 (1988) 201.
R. Decker, Fortschr. Phys. 37 (1989) 657.
E.A. Paschos, U. Trke, Phys. Rep. 178 (1989) 147.
B. Winstein, L. Wolfenstein, Rev. Mod. Phys. 65 (1993) 1113.
G.D. Barr et al., NA31 Collaboration, Phys. Lett. B 317 (1993) 233.
A. Alavi-Harati et al., KTeV Collaboration, Phys. Rev. Lett. 83 (1999) 22.
V. Fanti et al., NA48 Collaboration, Phys. Lett. B 465 (1999) 335.
M. Neubert, Int. J. Mod. Phys. A 11 (1996) 4173.
A.J. Buras, hep-ph/9806471.
E. Berberio, Report on the Heavy Flavour Working Group (LEP/SLD/CDF) to the Open
Session of the LEP Experiments Committee on March 7th 2000 available at http://delphiwww.
cern.ch/offline/physics_links/lepc.html.
A.D. Sakharov, Pisma Zh. Eksp. Teor. Fiz. 5 (1967) 32, JETP Lett. 5 (1967) 24.
T.D. Lee, Phys. Rev. D 8 (1973) 1226.
S. Weinberg, Phys. Rev. Lett. 37 (1976) 657.
G.C. Branco, Phys. Rev. Lett. 44 (1980) 504.
G.C. Branco, M.N. Rebelo, Phys. Lett. B 160 (1985) 117.
J. Liu, L. Wolfenstein, Nucl. Phys. B 289 (1987) 1.
N. Maekawa, Phys. Lett. B 282 (1992) 387.
A. Pomarol, Phys. Lett. B 287 (1992) 331.
N. Haba, Phys. Lett. B 398 (1997) 305.
O.C.W. Kong, F.-L. Lin, Phys. Lett. B 419 (1998) 217.
H. Georgi, A. Pais, Phys. Rev. D 10 (1974) 1246.
A. Pilaftsis, Phys. Rev. D 58 (1998) 096010; Phys. Lett. B 435 (1998) 88.
A. Pilaftsis, C.E.M. Wagner, Nucl. Phys. B 553 (1999) 3.
D.A. Demir, Phys. Rev. D 60 (1999) 055006.
J. Ellis, G. Ridolfi, F. Zwirner, Phys. Lett. B 257 (1991) 83.
M.S. Berger, Phys. Rev. D 41 (1990) 225.
Y. Okada, M. Yamaguchi, T. Yanagida, Prog. Theor. Phys. 85 (1991) 1; Phys. Lett. B 262
(1991) 54.
H.E. Haber, R. Hempfling, Phys. Rev. Lett. 66 (1991) 1815.
138
[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]
[70]
[71]
[72]
R. Barbieri, M. Frigeni, F. Caravaglios, Phys. Lett. B 258 (1991) 167.

P.H. Chankowski, S. Pokorski, J. Rosiek, Phys. Lett. B 274 (1992) 191.
J.R. Espinosa, M. Quirs, Phys. Lett. B 266 (1991) 389.
J.L. Lopez, D.V. Nanopoulos, Phys. Lett. B 266 (1991) 397.
M. Carena, K. Sasaki, C.E.M. Wagner, Nucl. Phys. B 381 (1992) 66.
P.H. Chankowski, S. Pokorski, J. Rosiek, Phys. Lett. B 281 (1992) 100; Nucl. Phys. B 423
(1994) 437.
D.M. Pierce, A. Papadopoulos, S.B. Johnson, Phys. Rev. Lett. 68 (1992) 3678.
A. Brignole, Phys. Lett. B 281 (1992) 284.
M. Drees, M.M. Nojiri, Phys. Rev. D 45 (1992) 2482.
V. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 49 (1994) 4908.
G.L. Kane, C. Kolda, L. Roszkowski, J.D. Wells, Phys. Rev. D 49 (1994) 6173.
R. Hempfling, A.H. Hoang, Phys. Lett. B 331 (1994) 99.
H.E. Haber, R. Hempfling, Phys. Rev. D 48 (1993) 4280.
J. Kodaira, Y. Yasui, K. Sasaki, Phys. Rev. D 50 (1994) 7035.
J.A. Casas, J.R. Espinosa, M. Quirs, A. Riotto, Nucl. Phys. B 436 (1995) 3; Nucl. Phys. B 439
(1995) 466 (Erratum).
M. Carena, J.R. Espinosa, M. Quirs, C.E.M. Wagner, Phys. Lett. B 355 (1995) 209.
M. Carena, M. Quiros, C.E.M. Wagner, Nucl. Phys. B 461 (1996) 407.
H.E. Haber, R. Hempfling, A.H. Hoang, Z. Phys. C 75 (1997) 539.
M. Carena, S. Mrenna, C.E.M. Wagner, Phys. Rev. D 60 (1999) 075010, hep-ph/9907422.
S. Heinemeyer, W. Hollik, G. Weiglein, Phys. Lett. B 440 (1998) 96; Phys. Rev. D 58 (1998)
091701.
R.-J. Zhang, Phys. Lett. B 447 (1999) 89.
J.R. Espinosa, R.-J. Zhang, hep-ph/9912236.
P.J. Dornan (ALEPH Collaboration), M. Grnewald (L3 Collaboration, C. Mariotti (DELPHI
Collaboration), R. McPherson (OPAL Collaboration), Reports to the Open Session of the LEP
Experiments Committee on March 7th 2000, available from http://delphiwww.cern.ch/offline/
physics_links/lepc.html; LEP working group for Higgs boson searches, P. Bock et al., Searches
for Higgs bosons: Preliminary combined results using LEP data collected at energies up to 202
GeV, available from http://www.cern.ch/LEPHIGGS/papers/index.html.
E.D. Commins, S.B. Ross, D. DeMille, B.C. Regan, Phys. Rev. A 50 (1994) 2960.
P.G. Harris et al., Phys. Rev. Lett. 82 (1999) 904.
J. Ellis, S. Ferrara, D.V. Nanopoulos, Phys. Lett. B 114 (1982) 231.
W. Buchmller, D. Wyler, Phys. Lett. B 121 (1983) 321.
J. Polchinski, M. Wise, Phys. Lett. B 125 (1983) 393.
F. del Aguila, M. Gavela, J. Grifols, A. Mendez, Phys. Lett. B 126 (1983) 71.
D.V. Nanopoulos, M. Srednicki, Phys. Lett. B 128 (1983) 61.
T. Falk, K.A. Olive, M. Srednicki, Phys. Lett. B 354 (1995) 99.
S. Pokorski, J. Rosiek, C.A. Savoy, hep-ph/9906206.
E. Accomando, R. Arnowitt, B. Dutta, hep-ph/9907446.
M. Dugan, B. Grinstein, L. Hall, Nucl. Phys. B 255 (1985) 413.
P. Nath, Phys. Rev. Lett. 66 (1991) 2565.
Y. Kizukuri, N. Oshimo, Phys. Rev. D 46 (1992) 3025.
G.F. Giudice, S. Dimopoulos, Phys. Lett. B 357 (1995) 573.
G. Dvali, A. Pomarol, Phys. Rev. Lett. 77 (1996) 3728.
A.G. Cohen, D.B. Kaplan, A.E. Nelson, Phys. Lett. B 388 (1996) 588.
P. Bintruy, E. Dudas, Phys. Lett. B 389 (1996) 503.
T. Ibrahim, P. Nath, Phys. Lett. B 418 (1998) 98; Phys. Rev. D 57 (1998) 478; Phys. Rev. D 58
(1998) 019901 (E); Phys. Rev. D 58 (1998) 111301, hep-ph/9910553.
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
[88]
[89]
[90]
[91]
[92]
[93]
[94]
[95]
[96]
[97]
[98]
[99]
[100]
[101]
[102]
[103]
[104]
[105]
[106]
[107]
[108]
[109]
[110]
[111]
[112]
[113]
[114]
[115]
[116]
139
M. Brhlik, G.J. Good, G.L. Kane, Phys. Rev. D 59 (1999) 115004.

M. Brhlik, L. Everett, G.L. Kane, J. Lykken, Phys. Rev. Lett. 83 (1999) 2124, hep-ph/9908326.
J. Ellis, R.A. Flores, Phys. Lett. B 377 (1996) 83.
A. Bartl, T. Gajdosik, W. Porod, P. Stockinger, H. Stremnitzer, Phys. Rev. D 60 (1999) 073003.
S.A. Abel, J.-M. Frre, Phys. Rev. D 55 (1997) 1623, and references therein.
J. Dai, H. Dykstra, R.G. Leigh, S. Paban, D.A. Dicus, Phys. Lett. B 237 (1990) 216; Phys.
Lett. B 242 (1990) 547 (E).
S. Weinberg, Phys. Rev. Lett. 63 (1989) 2333.
D. Chang, W.-Y. Keung, A. Pilaftsis, Phys. Rev. Lett. 82 (1999) 900; Phys. Rev. Lett. 83 (1999)
3972 (E).
A. Pilaftsis, Phys. Lett. B 471 (1999) 174.
D. Chang, W.-F. Chang, W.-Y. Keung, hep-ph/9910465.
S.M. Barr, A. Zee, Phys. Rev. Lett. 65 (1990) 21.
A. Pilaftsis, hep-ph/9912253, to appear in Phys. Rev. D.
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321.
M. Misiak, S. Pokorski, J. Rosiek, hep-ph/970344, to appear in Heavy Flavors II, A.J. Buras,
M. Lindner (Eds).
R. Garisto, J.D. Wells, Phys. Rev. D 55 (1997) 1611.
A. Brignole, J. Ellis, G. Ridolfi, F. Zwirner, Phys. Lett. B 271 (1991) 123.
E. Ma, Phys. Rev. D 39 (1989) 1922.
R. Hempfling, Phys. Rev. D 49 (1994) 6168.
L. Hall, R. Rattazzi, U. Sarid, Phys. Rev. D 50 (1994) 7048.
M. Carena, M. Olechowski, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 426 (1994) 269.
D. Pierce, J. Bagger, K. Matchev, R. Zhang, Nucl. Phys. B 491 (1997) 3.
J.A. Coarasa, R.A. Jimenez, J. Sola, Phys. Lett. B 389 (1996) 312.
R.A. Jimenez, J. Sola, Phys. Lett. B 389 (1996) 53.
K.T. Matchev, D.M. Pierce, Phys. Lett. B 445 (1999) 331.
P.H. Chankowski, J. Ellis, M. Olechowski, S. Pokorski, Nucl. Phys. B 544 (1999) 39.
K.S. Babu, C. Kolda, hep-ph/9811308.
M. Carena, D. Garcia, U. Nierste, C.E.M. Wagner, hep-ph/9912516, Nucl. Phys. B, in press.
M. Carena, H.E. Haber, S. Heinemeyer, W. Hollik, C.E.M. Wagner, G. Weiglein, hepph/0001002.
M. Carena, S. Heinemeyer, C.E.M. Wagner, G. Weiglein, hep-ph/9912223.
K.S. Babu, C. Kolda, J. March-Russell, F. Wilczek, Phys. Rev. D 59 (1999) 016004.
J. Ellis, M.K. Gaillard, D.V. Nanopoulos, Nucl. Phys. B 106 (1976) 292.
B.L. Ioffe, V.A. Khoze, Sov. J. Part. Nucl. 9 (1978) 50.
B.W. Lee, C. Quigg, H.B. Thacker, Phys. Rev. D 16 (1977) 1519.
S.L. Glashow, D.V. Nanopoulos, A. Yildiz, Phys. Rev. D 18 (1978) 1724.
A. Mndez, A. Pomarol, Phys. Lett. B 272 (1991) 313.
J.F. Gunion, B. Grzadkowski, H.E. Haber, J. Kalinowski, Phys. Rev. Lett. 79 (1997) 982.
P. Chankowski, hep-ph/9711470, Proceedings of the International Workshop on Quantum
Effects in the Minimal Supersymmetric Standard Model, pp. 87102, Barcelona, Spain, Sep.
1997.
M. Carena, D. Choudhury, S. Raychaudhuri, C.E.M. Wagner, Phys. Lett. B 414 (1997) 92.
M. Carena, J. Lykken (Eds.), Report of the Physics at Run II Supersymmetry/Higgs Workshop,
Fermilab 1999, in preparation. Results available at http://fnth37.fnal.gov/higgs.html.
A. Pilaftsis, Phys. Rev. Lett. 77 (1996) 4996; Nucl. Phys. B 504 (1997) 61.
S.-Y. Choi, M. Drees, Phys. Rev. Lett. 81 (1998) 5509.
B. Grzadkowski, J.F. Gunion, J. Kalinowski, Phys. Rev. D 60 (1999) 075011, hep-ph/0001093.
A. Pilaftsis, M. Nowakowski, Int. J. Mod. Phys. A 9 (1994) 1097.
G. Cvetic, Phys. Rev. D 48 (1993) 5280.
140
[117]
[118]
[119]
[120]
[121]
[122]
[123]
[124]
[125]
[126]
[127]
[128]
[129]
[130]
[131]
[132]
[133]
[134]
[135]
[136]
[137]
[138]
[139]
[140]
[141]
[142]
[143]
[144]
[145]
[146]
[147]
[148]
[149]
[150]
[151]
[152]
[153]
[154]
[155]
[156]
[157]
[158]
B. Grzadkowski, Phys. Lett. B 338 (1994) 71.

B. Grzadkowski, J.F. Gunion, Phys. Lett. B 350 (1995) 218.
A. Skjold, P. Osland, Nucl. Phys. B 453 (1995) 3.
C.A. Boe, O.M. Ogreid, P. Osland, J.-Z. Zhang, Eur. Phys. J. C 9 (1999) 413.
W. Bernreuther, A. Brandenburg, M. Flesch, Phys. Rev. D 56 (1997) 90, hep-ph/9812387.
T. Han, T. Huang, Z.H. Lin, J.X. Wang, X. Zhang, Phys. Rev. D 61 (2000) 015006.
B. Grzadkowski, J. Pliszka, Phys. Rev. D 60 (1999) 115018.
S.Y. Choi, J.S. Lee, Phys. Rev. D 61 (2000) 015003, hep-ph/9909315.
B. Grzadkowski, J.F. Gunion, J. Pliszka, hep-ph/0003091.
H. Baer, J. Wells, Phys. Rev. D 57 (1998) 4446.
W. Loinaz, J. Wells, Phys. Lett. B 445 (1998) 178, See also Ref. [50].
M. Carena, M. Quirs, C.E.M. Wagner, Phys. Lett. B 380 (1996) 81; Nucl. Phys. B 524
(1998) 3.
J.R. Espinosa, Nucl. Phys. B 475 (1996) 273.
D. Delepine, J.M. Grard, R. Gonzalez-Felipe, J. Weyers, Phys. Lett. B 386 (1996) 183.
A. Riotto, Phys. Rev. D 53 (1996) 5834.
J.R. Espinosa, B. De Carlos, Nucl. Phys. B 503 (1997) 24.
D. Bdeker, P. John, M. Laine, M.G. Schmidt, Nucl. Phys. B 497 (1997) 387.
M. Carena, M. Quirs, A. Riotto, I. Vilja, C.E.M. Wagner, Nucl. Phys. B 503 (1997) 387.
J.M. Cline, M. Joyce, M. Kainulainen, Phys. Lett. B 417 (1998) 79.
M. Laine, K. Rummukainen, Phys. Rev. Lett. 80 (1998) 5259; Nucl. Phys. B 535 (1998) 423.
J.M. Cline, G.D. Moore, Phys. Rev. Lett. 81 (1998) 3317.
M. Losada, Nucl. Phys. B 537 (1999) 3.
K. Funakubo, Prog. Theor. Phys. 101 (1999) 415; Prog. Theor. Phys. 102 (1999) 389.
J. Grant, M. Hindmarsh, Phys. Rev. D 59 (1999) 116014.
M. Laine, K. Rummukainen, Nucl. Phys. B 545 (1999) 141.
A.B. Lahanas, V.C. Spanos, V. Zarikas, Phys. Lett. B 472 (2000) 119.
S. Davidson, M. Losada, A. Riotto, hep-ph/0001301.
S.J. Huber, M.G. Schmidt, hep-ph/0003122.
E. Richter-Was, D. Froidevaux, F. Gianotti, L. Poggioli, D. Cavalli, S. Resconi, Int. J. Mod.
Phys. 13 (1998) 1371.
G. Acquistapace et al., CMS Collaboration, CERN-LHCC-9710.
R. Kinnunen, D. Denegri, CMS-NOTE 1997-057.
K. Lassila-Perini, CMS Thesis-1998-147.
D. Rainwater, D. Zeppenfeld, Phys. Rev. D 60 (1999) 113004; Phys. Rev. D 61 (2000) 099901
(E).
T. Plehn, D. Rainwater, D. Zeppenfeld, Phys. Lett. B 454 (1999) 297, hep-ph/9911385.
G. Belanger, S. Boudjema, K. Sridhar, hep-ph/9904348.
A. Dedes, S. Moretti, hep-ph/9904491.
M. Carena, S. Mrenna, C. Wagner, hep-ph/9907422.
A. Dedes, S. Moretti, hep-ph/9908516; hep-ph/9909418.
S.Y. Choi, J.S. Lee, hep-ph/9910557.
The Fortran code cph.f is available at http://home.cern.ch/p/pilaftsi/www/.
S.Y. Choi, M. Drees, J.S. Lee, hep-ph/0002287.
C. Panagiotakopoulos, private communication.

Bulk fields and supersymmetry in a slice of AdS

Tony Gherghetta a,b, , Alex Pomarol a,1
a Theory Division, CERN, CH-1211 Geneva 23, Switzerland
b IPT, University of Lausanne, CH-1015 Lausanne, Switzerland
Received 21 March 2000; revised 23 May 2000; accepted 20 June 2000
Abstract
Five-dimensional models where the bulk is a slice of AdS have the virtue of solving the hierarchy
problem. The electroweak scale is generated by a warp factor of the induced metric on the brane
where the standard model fields live. However, it is not necessary to confine the standard model fields
on the brane and we analyze the possibility of having the fields actually living in the slice of AdS.
Specifically, we study the behaviour of fermions, gauge bosons and scalars in this geometry and their
implications on electroweak physics. These scenarios may provide an explanation of the fermion
mass hierarchy by warp factors. We also consider the case of supersymmetry in the bulk, and analyze
the conditions on the mass spectrum. Finally, a model is proposed where the warp factor generates
a small (TeV) supersymmetry-breaking scale, with the gauge interactions mediating the breaking to
the scalar sector. 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
One of the major puzzles in particle physics is the large hierarchy of scales that appear
in the standard model (SM). The Planck scale, MP 1018 GeV, is much larger than the
electroweak scale ( 100 GeV), and even larger than the electron or neutrino mass. It
has recently been realized that the origin of these hierarchies in scale can be related to
the presence of extra dimensions with nontrivial spacetime geometries [1,2]. An explicit
example of this is the RandallSundrum model [2]. This consists of a five-dimensional
theory where the extra dimension is compactified on an orbifold, and the bulk geometry is
a slice of anti-de Sitter (AdS5 ) space of length R. Hence, in this theory there are two fourdimensional boundaries. The induced metric on these boundaries differs by an exponential
(warp) factor, and generates two effective scales, MP and MP ekR (where k is the AdS
curvature scale of order MP ). Thus, for a moderately large extra dimension, kR & 1, one
can obtain an exponentially small scale, MP ekR , on one of the boundaries. If the SM
Corresponding author. E-mail: tony.gherghetta@cern.ch
1 On leave of absence from IFAE, Universitat Autnoma de Barcelona, E-08193 Bellaterra, Barcelona.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 2 - 8
142
T. Gherghetta, A. Pomarol / Nuclear Physics B 586 (2000) 141162
fields live on this boundary, then the electroweak scale can be associated with the effective
scale on this boundary. In particular for kR 12, this will be O(TeV) and the brane is
referred to as the TeV-brane [3].
In the model of Ref. [2] only gravity propagates in the 5D bulk, while the SM fields are
on the TeV-brane. It is interesting to study other possibilities, such as having all the SM
fields in the 5D bulk. This is the subject of this article, where we will study the behaviour
of various spin fields in a slice of AdS5 . We will first analyze the non-supersymmetric
case. Part of this analysis has been carried out previously in the literature. The role of
scalar fields in the bulk was first considered in Ref. [4]. Subsequent studies considered
the SM gauge bosons [5,6], and fermions [7] in the bulk. The complete SM in the bulk
was considered in Refs. [8,9]. We will present here a more compact and general analysis,
including for example, the case when scalar fields have boundary mass terms. As an
important new ingredient, we will consider supersymmetry in the AdS5 slice. We will
derive the conditions necessary for supersymmetry and the implications for the particle
spectrum.
Armed with the above, we will then study the phenomenological consequences of having
the SM fields in the 5D bulk. We will show that SM fermions and gauge bosons can
propagate in the AdS5 bulk without conflicting with any experimental data. However,
the Higgs field must arise from a KaluzaKlein excitation that is localized by the AdS
metric on the TeV-brane [4,8]. We will show that having the SM fermions in the bulk
can explain the fermion mass hierarchy by means of the metric warp factor. We will also
study the magnitude of higher-dimensional operators. These operators are known to be
a problem in theories with a low (TeV) cutoff scale (including in the original Randall
Sundrum setup [10]), since they can induce large proton decay rates or large flavour
violating interactions. These operators can be suppressed by allowing the fermions to be
off the TeV-brane. This leads to partial success because flavour violation can be suppressed
to the desirable level, but this is not the case for proton decay.
The inclusion of supersymmetry allows for new alternatives. The Higgs field can now
be delocalized from the TeV-brane, together with the fermions. This delocalization is
possible thanks to supersymmetry that protects the Higgs massless mode far from the
TeV-brane. If fermions are also distant from the TeV-brane, then proton decay can be
sufficiently suppressed, while at the same time allowing for the fermion masses to be
generated. In fact, this model at low energy resembles the ordinary MSSM. Nevertheless,
it does have important differences. For instance, the boundary TeV-brane can now be
the source of supersymmetry breaking and a window to Planckian physics. Gauge
superfields living in the bulk will be responsible for mediating TeV-brane physics to fields
that are far away from it. For example, the gauge interactions can be the messengers of
supersymmetry breaking to the quark and lepton sector, guaranteeing a universal scalar
mass spectrum. Similar scenarios were considered in the case of a flat extra dimension.
This was done either by breaking supersymmetry in the bulk [1114] or on a distant
brane [1518]. In Refs. [1114], a large (TeV) extra dimension was required that implied
a cutoff (string) scale of O(TeV). In Refs. [1518], a small (Planckian) extra dimension
was advocated, but in this case the resultant phenomenology was similar to the gravity-
143
mediated models of supersymmetry-breaking proposed earlier with no new low-energy

implications. Nevertheless, the possibility here of mediating supersymmetry-breaking by
five-dimensional gauge bosons leads to a very different scenario with respect to the
previous examples. As we will see, we can combine the good implications of having a
large ( MP ) cutoff scale (such as small proton decay, small neutrino masses), with the
new phenomenology of having an extra dimension (KaluzaKlein states at collider physics,
quantum gravity effects at the TeV).
It is important to point out why the slice of AdS5 is interesting and different
from other compactifications [1921]. Theories that solve the hierarchy problem with
other compactifications [1,1921] usually require a large compactification radius (or,
equivalently, a large volume in the extra dimensions). Therefore, if SM fields propagate
in these extra dimensions, the couplings become too weak.
Finally, three important remarks are in order. Even though we will be considering fields
other than the graviton living in the 5D bulk, we will assume that the AdS metric is not
modified by the presence of these bulk fields. In other words, the back-reaction from the
bulk fields is neglected. Second, there have been recent unsuccessful attempts at deriving
the scenario of Ref. [2] from a five-dimensional supergravity Lagrangian [2226]. While
there is no general proof that a description from a more fundamental theory does not exist,
we will not have anything more to add here. Further studies must still be carried out to settle
this issue. We will assume that the solution of Ref. [2] exists and study its compatibility
with supersymmetry. Finally, to obtain the AdS slice, the TeV-brane in Ref. [2] requires
a negative tension. This will violate the weak-energy condition [27]. Nevertheless, it is
conceivable that a setup, like we are considering, is possible without violating the weakenergy condition [27], especially given the fact that the fundamental theory is most likely
to be supersymmetric.
The article is organized in the following way. In Section 2, we introduce the
compactification scenario of Ref. [2], and derive the KaluzaKlein decomposition of fields
with different spins. In Section 3, we study the conditions necessary for supersymmetry
in a slice of AdS5 . In particular, we derive the mass spectrum and wavefunctions of
the gauge supermultiplet and hypermultiplet. The possibility of having the SM fermions
and gauge bosons in the 5D bulk is studied in Section 4, where particular attention is
paid to constraints from electroweak data. This will lead us to consider supersymmetric
models, together with their resulting phenomenological predictions. Our conclusions and
final comments appear in Section 5.
2. Bulk fields in a slice of AdS5

We will consider the scenario of Ref. [2], based on a non-factorizable geometry with
one extra dimension. In this scenario, the fifth dimension y is compactified on an orbifold,
S 1 /Z2 of radius R, with R 6 y 6 R. The orbifold fixed points at, y = 0 and
y = R are also the locations of two 3-branes, which form the boundary of the five-
144
dimensional spacetime. Consequently, the classical action for this configuration is given
by
ZR
Z
S=
d x
dy

1
+ M53 R + (y) (0) + L(0) + (y R) (R) + L(R) ,
2
(1)
where M5 is the five-dimensional Planck mass scale and R is the five-dimensional Ricci
scalar, constructed from the five-dimensional metric gMN , with the 5D coordinates labelled
by capital Latin letters, M = (, 5). The cosmological constants in the bulk and boundary
are , and (y ) , respectively.
A solution to the five-dimensional Einsteins equations, which respects four-dimensional
Poincare invariance in the x directions, is given by [2]
ds 2 = e2 dx dx + dy 2 ,
(2)
where
= k|y| ,
(3)
and 1/k is the AdS curvature radius. The four-dimensional metric is = diag(1, 1,
1, 1) with = 1, . . . , 4. The metric solution (2) is valid provided that the bulk and
boundary cosmological constants are related by
= 6M53 k 2 ,
(0) = (R) =
.
k
(4)
The four-dimensional reduced Planck scale MP is related to M5 in the following way:

MP2 =

M53
1 e2kR .
k
(5)
From the form of the metric solution (2), the spacetime between the two branes located
at y = 0 and y = R is simply a slice of AdS5 geometry. The effective mass scale
on the brane at y = 0 is the Planck scale MP , and we will refer to this 3-brane as the
Planck-brane. Similarly, at y = R, the effective mass scale is MP ekR , which will be
associated with the TeV scale provided kR ' 12. This 3-brane will be referred to as the
TeV-brane. It is also interesting to note that the relations (4) arise in the effective fivedimensional HoravaWitten theory [28] when the CalabiYau moduli are fixed.
So far we have only assumed gravity to be present in the five-dimensional bulk. We will
now study a U(1) gauge field, VM , a complex scalar, , and a Dirac fermion, , living in
a slice of AdS5 given by the metric Eq. (2). The five-dimensional (5D) bulk action, S5 , has
kinetic energy and mass terms given by
Z
S5 =
d4 x

Z
dy
145

1 2
2
M
2
2
,
F
+
|
|
+
i
+
m
||
+
im
M
M
MN
4g52
(6)
where g = det(gMN ). The U(1) gauge field strength is FMN = M VN N VM and in

curved space the covariant derivative is DM = M + M , where M is the spin connection.
In particular, for the metric defined by Eq. (2), we have
1
d
and 5 = 0 .
(7)
= 5
2
dy
, where e
The gamma matrices, M = ( , 5 ) are defined in curved space as M = eM
M
is the funfbein and are the Dirac matrices in flat space.
Under the Z2 parity the boson fields can be defined either odd or even (depending
on their interactions). For the fermion , the Z2 transformation is given by (y) =
5 (y), where the arbitrariness of the sign can only be determined by the fermion
interactions. We then have that is odd under the Z2 symmetry, and consequently the
Dirac mass parameter m must also be odd. This nontrivial transformation of the fermion
mass parameter implies that m must arise from an underlying scalar field that receives a
vacuum expectation value (VEV) with an odd kink profile. In other words, the vacuum
configuration of the scalar field can be thought of as an infinitely thin domain wall. This
is very similar to the background 3-form field in the dimensional reduction of Horava
Witten theory [28]. Furthermore, note that a Dirac mass term cannot be induced on the
boundaries, since vanishes there. On the other hand, a Majorana mass term for the
fermion can be added, either in the bulk or on the boundaries. This term will necessarily
break supersymmetry, and we will comment on this in Section 4. The scalar mass term is
even under the Z2 symmetry and can be either a bulk or boundary term. Therefore, the
mass parameters of the scalar and fermion fields can be parametrized as 2
m2 = ak 2 + b 00 ,
m = c 0 ,
(8)
where a, b and c are arbitrary dimensionless parameters, and the derivatives are defined as
d
= k(y),
dy

d2
(9)
00 = 2 = 2k (y) (y R) .
dy
The step function (y) is defined as being 1 (1) for positive (negative) y, and (y) is the
Dirac delta-function.
The equations of motion for the gauge, scalar and fermion fields are, respectively, given
by
0 =
2 We are assuming that the magnitude of the boundary mass for the scalar is the same on the two boundaries,
but with opposite sign. As we will see, this is required by supersymmetry. The generalization to different masses
for different boundaries can be easily obtained from the analysis here.
146
1
M
M g g MN g RS FNS = 0,
g g MN N m2 = 0,

g MN M DN + m = 0.
(10)
Using the metric of Eq. (2), one can write a general second-order differential equation 3
2

(11)
e + es 5 (es 5 ) M2 (x , y) = 0,
where = {V , , e2 L,R }, s = {2, 4, 1} and M2 = {0, ak 2 + b 00 , c(c 1)k 2 c 00 }.
For the fermion, we have introduced the exponential factor, e2 , which takes into account
the spin connection, and we have explicitly separated the left and right components (defined
by L,R = 5 L,R ).
2.1. KaluzaKlein decomposition
Let us decompose the 5D fields as
1 X (n)
(x )fn (y),
(x , y) =
2R n=0
(12)
where the KaluzaKlein modes fn (y) obey the orthonormal condition

1
2R
ZR
dy e(2s) fn (y)fm (y) = nm .
(13)
Thus from Eq. (11), the KaluzaKlein eigenmodes fn (y) satisfy the differential equation

s
b2 fn = e2 m2n fn .
(14)
e 5 (es 5 ) + M
b2 = {0, ak 2, c(c 1)k 2 },
In Eq. (14), we have defined the mass-squared parameter, M
00
without the boundary mass terms proportional to . These terms will be considered
later when we impose the boundary conditions on the solutions. The corresponding
eigenfunctions fn , obtained by solving Eq. (14), are

es/2
mn
mn
e + b (mn ) Y
e
J
,
(15)
fn (y) =
Nn
k
k
where mn is the mass of the KaluzaKlein excitation (n) and the normalization factor Nn
and coefficients b (mn ) are constants. The Bessel functions J and Y are of order
q
b 2 /k 2 .
= (s/2)2 + M
Using the orthonormal condition Eq. (13), an expression for the normalization constant Nn
is

ZR
1
mn
mn 2
2
e + b (mn )Y
e
dy e2 J
.
(16)
Nn =
R
k
k
0
3 For the gauge field we work, as in the graviton case [2], in the gauge where V = 0 together with the constraint
5
V = 0.
147
In the limit mn k and kR 1 an approximate expression for the normalization constant

is

ekR
mn kR
ekR/2
J
e
.
(17)
'p
Nn '
k
2kR
2 Rmn
Note that in this limit the normalization constant, Nn , does not depend on the particular
type of field, i.e., the parameter s. The KaluzaKlein masses mn and the coefficients
b (mn ) are determined by imposing the boundary conditions on the solution (15). The
appropriate boundary condition depends on whether the field is odd or even under the
orbifold Z2 symmetry.
2.1.1. Even fields
If the 5D field is even under the Z2 symmetry, then we have

fn (y) fn |y| .
(18)
The derivative of fn must be either discontinuous (proportional to (y)) or continuous on

the boundaries depending on whether or not the mass-squared parameter, M2 in Eq. (11),
has delta function terms, 00 . Thus we must impose

dfn
r 0 fn
= 0,
(19)
dy
0,R
where the parameter r has the values r = {0, b, c} for = {V , , e2 L,R },
respectively. Imposing Eq. (19) gives rise to the two equations
b (mn ) =
(r + s/2)J ( mkn ) +
(r + s/2)Y ( mkn ) +
mn 0 mn
k J ( k )
mn 0 mn ,
k Y ( k )
b (mn ) = b (mn ekR ).
(20)
(21)
These two conditions determine the values of b and mn . In the limit that mn k and
kR 1 the KaluzaKlein mass solutions for n = 1, 2, . . . , and > 0 are

3
kekR .
(22)
mn ' n +
2 4
The approximate mass formula (22) becomes more exact, the larger the value of n.
2.1.2. Odd fields
If the 5D field is odd under the Z2 symmetry, then we have
fn (y)

0
fn |y| .
k
The continuity of fn at the boundaries implies that

fn 0,R = 0,
and consequently
(23)
(24)
148
b (mn ) =
J ( mkn )
,
Y ( mkn )
b (mn ) = b (mn ekR ).
(25)
(26)
In this case one can check that the derivative of fn is continuous on the boundaries and
does not lead to further conditions. As in the even case, an approximate solution for the
KaluzaKlein tower in the limit that mn k and kR 1 is

1
kekR ,
(27)
mn ' n +
2 4
where n = 1, 2, . . . , and > 0. Note that the difference in the approximate mass formulae
between the even and odd mass solutions is /2.
In the case of fermion fields one should note that the even and odd functions are related
by coupled first-order differential equations
bR + 5
bL + m
bL = 0,

(28)
bL 5
bR + m
bR = 0,

(29)
b = e2 . Thus, the even and odd fermion fields must also satisfy these equations.
where
3. Supersymmetry in a slice of AdS5

An AdS space is compatible with supersymmetry [29,30]. However, in contrast to the
flat space case, AdS supersymmetry requires that different fields belonging to the same
supersymmetric multiplet have different masses. This is because the momentum operator,
P , in AdS space does not commute with the supersymmetric charges and P 2 is not a
Casimir operator.
The supermultiplet mass-spectrum has been derived in Ref. [31] for a five-dimensional
AdS space. Here we will extend the analysis to the case where the fifth dimension is
compactified on an orbifold S 1 /Z2 . In the next section, we will see that the supermultiplet
mass-spectrum in a slice of AdS5 will have interesting phenomenological implications.
3.1. Supergravity multiplet
, the graviphoton B and
The on-shell supergravity multiplet consists of the funfbein eM
M
i
(i = 1, 2). The index i labels the fundamental
two symplectic-Majorana 4 gravitinos M
representation of the SU(2) automorphism group of the N = 1 supersymmetry algebra
in five dimensions. In a slice of AdS5 , the supergravity Lagrangian has extra terms
proportional to the cosmological constants:
4 We will follow the conventions of Ref. [32] (except for our metric convention
= (1, 1, 1, 1)), where the
two symplectic-Majorana spinors, i and i , satisfy i M P j = ik j l l P M k .
149
1
d4 x dy g
2

3 0 i MN
00
3
i MNR
i
ij j
DN R i M
(3 ) N + 2 2 ,
M5 R + iM
2
k
S5 =
(30)
P
where MNR perm (1)p M N R /3! and MN = [ M , N ]/2. In Eq. (30) we do not
show the dependence on BM , since in the AdS5 background we have BM = 0. In order to
respect supersymmetry in (30), the supersymmetric transformation of the gravitino must
be changed, with respect to the supergravity case with no cosmological constant, in the
following way
0
M (3 )ij j ,
(31)
2
where 3 = diag(1, 1) and the symplectic-Majorana spinor i is the supersymmetric
parameter. Without loss of generality, we have defined the Z2 transformation of the
symplectic-Majorana spinor as
i
= DM i +
M
i (y) = (3 )ij 5 j (y) .
(32)
i
= 0,
The condition that the AdS5 background does not break supersymmetry is M
and using Eq. (31) this leads to the Killing spinor equation
0
M (3 )ij j .
(33)
2
In a non-compact five-dimensional AdS space this condition is always fulfilled. However,
in the orbifold compactification, the boundary terms require an extra condition to be
satisfied, namely
DM i =
5 i = (3 )ij j .
(34)
This condition implies that only half of the 5D supersymmetric charges are preserved.
Therefore, after compactification, one has in 4D a N = 1 supersymmetric theory instead
of N = 2.
3.2. Vector supermultiplet
The on-shell field content of the vector supermultiplet is V = (VM , i , ) where VM is
the gauge field, i is a symplectic-Majorana spinor, and is a real scalar in the adjoint
representation. For simplicity we will consider a U(1) gauge group. The action has the
form
Z
Z
1
4
d x dy g
S5 =
2

1 2
2
i M
i
2
2
i
ij j
F
+
(
)
+
i
+
m
+
im
(
)
. (35)
M
M
3
MN
2g52
In flat space, supersymmetry requires that m = m = 0. This is to be contrasted with an
AdS5 background, where fields in the same supermultiplet must have different masses [31].
By requiring Eq. (35) to be invariant under the supersymmetric transformations
150
VM = i i M i ,
= i i ,

i = MN FMN + i M M i 2i 0 (3 )ij j ,
(36)
where i satisfies the Killing equation (33) and the condition (34), one finds that in a
slice of AdS5 , the five-dimensional masses of the scalar and spinor fields in the vector
supermultiplet must be
m2 = 4k 2 + 2 00 ,
1
(37)
m = 0 .
2
The KaluzaKlein decomposition can be obtained easily from the analysis of the previous
sections. From Eq. (37) and Eq. (8) we have
1
a = 4, b = 2, and c = .
2
(38)
Using Eq. (38), we find that = 1 for V and 1L , while = 0 for and 2L . If we assume
that V and 1L are even, while and 2L are odd, then the KaluzaKlein masses are
determined by the equation
J0 ( mkn ) J0 ( mkn ekR )
=
.
Y0 ( mkn ) Y0 ( mkn ekR )
(39)
Thus, even though the values of are different, we find as expected that all fields of the
supermultiplet have identical KaluzaKlein masses. In fact, the approximate solution for
the mass of the KaluzaKlein modes with n = 1, 2, . . . , is given by [6]

1
kekR ,
(40)
mn ' n
4
and is consistent with the mass formulae (22) and (27). The even fields V and 1L will
have a massless mode with the following y dependence:
1
V(0) (x) + ,
V (x, y) =
2R
e3/2 1 (0)
(41)
L (x) + .
1L (x, y) =
2R
Notice that the zero modes have the proper y dependence to make the 4D kinetic terms of
these modes in the action
ZR
1
dy g
d x
2R
R

1 (0)
1 (0)
(0)
3 1 (0)
V
(x)
V
(x)
+
ie
(x)
(x)
,
L
L
2g52
Z
(42)
invariant under the conformal transformation g e2 g . This implies that the action
(42) in the background metric (2) is independent of the y coordinate. Consequently, the
151
zero modes are not localized by the AdS spacetime and they behave like zero modes in
flat space. Furthermore, their couplings to the two boundaries are of equal strength.
The odd fields and 2L do not have massless modes because this is not consistent with
the orbifold condition. Therefore, the massless sector from V and 1L forms an N = 1
supersymmetric vector multiplet.
3.3. Hypermultiplet
The hypermultiplet consists of H = (H i , ) where H i are two complex scalars and
is a Dirac fermion. The action has the form

Z
Z

2
S5 = d4 x dy g M H i +i M DM + m2H i |H i |2 + im . (43)
Invariance under the supersymmetric transformations
H i = 2 i ij j ,

3
= 2 M M H i ij 0 H i (3 )ij m H i ij j ,
2
(44)
where i satisfies the Killing equation (33) and the condition (34), requires that the fivedimensional masses of the scalars and fermion satisfy

15 2
3
2
2
c 00 ,
k +
mH 1,2 = c c
4
2
(45)
m = c 0 ,
where c remains an arbitrary dimensionless parameter. Using Eq. (8) we can identify
3
15
and b = c .
(46)
a = c2 c
4
2
Thus, we find that = |c + 1/2| for H 1 and L , and = |c 1/2| for H 2 and R .
Assuming that H 1 and L are even, while H 2 and R are odd, then the KaluzaKlein
masses (identical for both the even and odd modes) are determined by the equation
J|c+1/2| ( mkn ) J|c+1/2| ( mkn ekR )
=
.
Y|c+1/2| ( mkn ) Y|c+1/2| ( mkn ekR )
(47)
In fact using Eq. (22) for the even modes, and Eq. (27) for the odd modes, we can again
see that the odd and even KaluzaKlein modes, n = 1, 2, . . . have identical masses which
are approximately given by

c 1
(48)
kekR .
mn ' n +
2 2
The massless modes of H 1 and L are given by
e(3/2c) 1 (0)
H
(x) + ,
H 1 (x, y) =
2R N0
e(2c)
(0)
L (x) + ,
L (x, y) =
2R N0
(49)
152
where
N02 =
e2kR(1/2c) 1
.
2kR(1/2 c)
(50)
When c = 1/2, we have N0 = 1 which signals the conformal limit, where the kinetic terms
are again independent of the y coordinate. Thus, as explained in the vector multiplet
case, the massless modes are not localized by the AdS space and will couple to the
two boundaries with equal strength. For values of c < 1/2, the AdS space will localize
the massless modes towards the boundary y = R. As c becomes more negative, the
localization at the y = R boundary becomes increasingly more effective. On the other
hand when c > 1/2, the massless modes will be localized towards the y = 0 boundary. In
this case the localization becomes more effective the larger the value of c.
The odd fields do not have a massless mode, since this is not consistent with the orbifold
condition. Therefore, as in the case of the vector multiplet, the massless sector of the
hypermultiplet (from H 1 and L ), forms an N = 1 supersymmetric chiral multiplet.
4. Phenomenological implications
In the original proposal in Ref. [2], the SM fields were assumed to be confined on the
TeV-brane. The question naturally arises then of whether the SM fields can actually live
in the bulk. We will first consider models with bulk fields but without supersymmetry. A
non-supersymmetric model can be constructed, but eventually we will see that many more
interesting possibilities exist for supersymmetric models.
4.1. The standard model in the bulk
We have seen that the massless modes of a 5D scalar or fermion field in the AdS
slice can be localized at either of the two boundaries, depending on the value of the
dimensionless mass parameter c. This continues to be true for massless fermions, even
without supersymmetry, due to the fact that chiral symmetry protects the massless mode.
However, without supersymmetry, the massless mode of the bulk scalar field will receive
radiative corrections of order the Planck scale. Thus, it would appear that the hierarchy
problem is again reintroduced, as in flat space. This dire conclusion is not in fact realized,
because the nonzero KaluzaKlein scalar modes are localized near the TeV-brane and
consequently have TeV scale masses. Thus, even though the scalar zero mode becomes
heavy, the mass of the lightest nonzero KaluzaKlein mode is still of order the TeV scale,
and therefore the Higgs can be associated with this lightest KaluzaKlein state. This is
effectively like having the Higgs field on the TeV-brane.
If the fermion fields are located in the bulk, then the gauge bosons must necessarily
reside in the bulk. In this case, the KaluzaKlein excitations of the gauge bosons can induce
four-fermion interactions that are severely constrained by the experimental data [3342].
To study these constraints in our case, let us consider the five-dimensional gauge coupling
between the bulk gauge boson and fermion. It is given by
Z
d4 x
dy
g g5 (x, y)i A (x, y) (x, y),
153
(51)
where g5 is the five-dimensional gauge coupling. Using the expression for the zero-mode
fermion (49), the gauge coupling of a gauge boson KaluzaKlein mode n to the zero-mode
fermions is

ZR
k
dy e(12c)
g = g (12c)kR
e
1 Nn
0

mn
mn
e + b1 (mn )Y1
e
J1
,
(52)
k
k
where g = g5 / 2R is the four-dimensional gauge coupling, and c is defined in Eq. (8).

In Fig. 1, we show the ratio g (n) /g, for n = 1, 2, 3, as a function of c. When c is large and
negative, the fermion is localized
near the TeV-brane and the ratio g (1) /g approaches the
(1)
asymptotic limit g /g ' 2kR ' 8.4, which corresponds to a TeV-brane fermion [5,6].
In this region the first excited KaluzaKlein mode of the gauge boson couples strongly to
the fermion zero-mode, leading to a restrictive lower bound on the first excited Kaluza
Klein mass scale m1 & 20 TeV [5]. At c = 0, we recover the massless fermion case
considered in Ref. [8]. As the value c of the bulk fermion mass increases, we find that
the bound considerably weakens, because the fermion zero mode couples less strongly. At
the conformal limit c = 1/2, the coupling completely vanishes due to the fact that the fivemomentum is conserved in this limit. For c > 1/2, the coupling quickly becomes universal
for all fermion masses. This is because the fermions are now localized near the Planckbrane, where the wavefunction of the KaluzaKlein gauge bosons is constant. In this region
the asymptotic limit g (1) /g ' 0.2, agrees with the case of a Planck-brane fermion [6].
If all fermions have equal five-dimensional masses, the bound from electroweak precision
experiments on the first excited KaluzaKlein mass is
(n)
1 2c
Fig. 1. The ratio of the gauge couplings, g (n) /g, for n = 1 (dotted line), n = 2 (solid line) and n = 3
(dashed-dotted line), as a function of the dimensionless fermion mass parameter c.
154

m1 & 2.1
g (1)
g

TeV,
(53)
where m1 ' 2.5 kekR . Thus we see that in the region where c & 1/2, the lower bound is
fairly innocuous. The coupling of the higher-order KaluzaKlein modes (n > 1) is always
smaller than the coupling of the first excited state, and thus gives negligible contributions
to any four-fermion operator (unless the fermion is on the TeV-brane). As can be seen in
Fig. 1, the ratio of the couplings, g (2) /g, and g (3) /g, continue to remain c-independent for
c & 1/2.
4.2. Fermion mass hierarchy
In the original scenario [2], the warp factor e , was used to solve the gauge hierarchy
problem by generating the TeV scale from the Planck scale. The fermion mass hierarchy
problem was not addressed there because the fermions were confined to the brane and
the Yukawa interactions were conformally invariant. However, if we allow the fermions to
live in the bulk, a similar use of the warp factor can be used to explain the fermion mass
hierarchies. For neutrino masses this idea has been considered in Ref. [7]. Here we will
analyze it for the quarks and leptons. Our setup is as follows. For each fermion flavour i, we
have two five-dimensional Dirac fermions iL (x, y), and iR (x, y), while for simplicity
the Higgs H (x) will be localized on the TeV-brane. Thus, the five-dimensional action will
have the Yukawa coupling term
Z
Z
(5)
d4 x dy g ij H (x) iL (x, y)j R (x, y) + h.c. (y R)
Z

(0)
(0)
(54)
d4 x ij H (x) iL (x)j R (x) + h.c. + ,
(5)
where ij are the five-dimensional Yukawa couplings and ij define the effective four(0)
(0)
dimensional Yukawa couplings of the zero modes, iL and j R . If each fermion field has
a bulk mass term parametrized by ciL and cj R , then the fermion zero modes will develop an
exponential profile which gives rise to the following four-dimensional Yukawa couplings
(5)
ij =
ij k
NiL Nj R
e(1ciL cjR )kR ,
(55)
where
1/2 ciL
1
,
(12c )kR
2
iL
e
1
NiL
(56)
and similarly for Nj R . Note that in deriving (55), the Higgs must be rescaled by an amount
H (x) ekR H (x), in order to obtain a canonically normalized kinetic term. We see then
from Eq. (55), that exponentially-small Yukawa couplings can be generated for values of
ciL and cj R slightly larger than 1/2.
The precise flavour structure of the Yukawa coupling matrix, ij , will depend on the
values of ciL and cj R . We can consider two simple limits, which in some sense represents
155
the two possible extremes for the flavour structure. Suppose first that we have a leftright
symmetric theory such that ciL = ciR . In Fig. 2 we have plotted the diagonal element
of the Yukawa matrix, Eq. (55), in this leftright symmetric limit. We have assumed, for
simplicity, that (5)
ij k 1. For ciL > 1/2, we clearly see in Fig. 2 the exponential damping
of Eq. (55), where the diagonal elements of Eq. (55) simplifies to (for kR 1)
2(ciL1/2)kR
.
ii ' (5)
ii k (ciL 1/2) e
(57)
An electron Yukawa coupling e 106 can be generated for ceL ' 0.64 (assuming kR
12.46). On the other hand, a top Yukawa coupling t 1 can be obtained for ct L = 1/2,
since the exponential factor in (55), disappears for ciL < 1/2.
Alternatively, we can suppose that the right-handed fermions all have cj R = 1/2. Now,
the effective Yukawa couplings, in the limit that kR 1 and ciL > 1/2 become
k p
ciL 1/2 e(ciL 1/2)kR .
(58)
ii ' ii
2kR
In this case the exponential factor is more dominant compared to the leftright symmetric
assumption. This can clearly be seen in Fig. 2.
It is easy to understand the behaviour of the curves in Fig. 2. For ciL > 1/2, the
Yukawa couplings quickly become exponentially-small since fermions are localized near
the Planck-brane and have an exponentially small overlap with the Higgs, which is located
on the TeV-brane. In this way we see that the mass of the lightest fermion states of the SM
can be naturally generated for values of ciL > 1/2. The heavier fermions (c, , b) require
a larger overlap with the Higgs and cannot be confined near the Planck-brane. In this case
the values of the mass parameters must be close to the conformal limit (ciL ' 1/2), where
fermions are not localized on either of the two branes. The top quark having a comparable
mass to the VEV of the Higgs, needs the largest overlap of all, and must be localized
on the TeV-brane (ct L . 1/2). Again we must stress that this scenario does not predict
the fermion mass spectrum (which would correspond to predicting the values of ciL and
(5)
Fig. 2. The effective four-dimensional Yukawa coupling ii as a function of ciL , for the case
ciL = ciR (solid line) and ciR = 1/2 (dotted line) .
156
cj R ). However, it does offer a possible explanation for the mass hierarchy between fermion
families. This idea was first discussed in the context of string theory by Ref. [43], and is
also similar to Ref. [4446], where the exponential overlap between the Higgs and fermion
wavefunctions also generated fermion mass hierarchies.
It is important to realize, however, that if the fermions are localized at different points (by
assuming that they have different 5D masses ci ), then they will couple differently to the
KaluzaKlein gauge bosons. This can give rise to large flavour-changing neutral current
(FCNC) processes [42]. For example, in the model of Ref. [44], a lower bound on the
compactification scale of 25300 TeV is obtained [42] (the range in the bound depends
on whether or not the induced FCNC processes violate the CP symmetry). This would also
seem to be a problem for our scenario. However, in our scenario the FCNC constraints lead
to a much smaller lower bound on the first excited KaluzaKlein mass m1 . For the case
represented by Eq. (57) we have m1 & 230 TeV, while m1 & 0.22.5 TeV in the case
of Eq. (58). The lower bounds are much reduced because, as shown in Fig. 1, the Kaluza
Klein gauge-bosons couple almost universally to any fermion with ciL & 1/2. This is the
case for the first and second family. The third-family fermions do not have a universal
coupling, since ciL < 1/2, but in this case the FCNC constraints are much weaker [42].
4.3. Higher-dimensional operators
When fermions are confined on the TeV-brane, as in the original setup [2], higherdimensional operators are suppressed by the TeV scale and not the Planck-scale [10]. In
the absence of any discrete symmetries, this of course leads to proton decay problems. In
addition, constraints on KK mixing requires that the dimension-six operator, (d s)2 /M42
needs to be suppressed by at least an effective four-dimensional mass scale M4 1000
TeV. There is also the problem of generating large neutrino masses.
If the fermion fields are instead in the bulk, then we can ask whether effective higherdimensional operators are suppressed. As we saw in Section 3, each fermion zero mode
has a wavefunction proportional to e(2ci ) , and depending on the values of ci can lead
to an exponential suppression of mass scales. Let us consider first the following generic
four-fermion operators which are relevant for proton decay and KK mixing:
Z
Z
Z
1
1
(0) (0) (0) (0)
(59)
d4 x dy g 3 i j k l d4 x 2 i j k l ,
M4
M5
where the effective four-dimensional mass scale M4 is
2k
1
e(4ci cj ck cl )kR 1
1
,
=
M42 M53 Ni Nj Nk Nl (4 ci cj ck cl )
(60)
and Ni are defined in Eq. (56). For the values 1/2 . ci . 1, the four-dimensional
suppression scale, M4 ranges from the TeV scale (for ci ' 1/2) to MP (for ci ' 1).
Therefore, we can only obtain the necessary suppression for proton decay if the mass
parameters satisfy ci & 1. This, of course, means that the light fermions are localized
on the Planck-brane. Unfortunately, for ci & 1 fermion masses are much too small (see
Eq. (55)). This is because (in non-supersymmetric models), the Higgs field must live on
157
the TeV-brane and will have a very small overlap with the fermions. Thus, it is impossible
to suppress proton decay operators and have fermion masses. On the other hand, we find
that the constraints on KK mixing, that require M4 & 1000 TeV, can be satisfied for
values of ci compatible with generating fermion masses.
Finally, let us consider the dimension-five operator responsible for generating Majorana
neutrino masses. Again assuming that the left-handed fields are in the bulk we obtain
Z
Z
1 T
C5 iL H (x)H (x)(y R)
d4 x dy g 2 iL
M5
Z
1 (0) T
(0)
C5 iL
H H,
(61)
d4 x
M4 iL
where C5 is the charge-conjugation operator. For ciL > 1/2, M4 is given by
M4 '
M52 2(1ciL )kR

e
.
k
(62)
Using the relation (58), we can relate the neutrino masses to the fermion masses:
mi ' 1022ii GeV,
(63)
and leads to neutrino mass values of (1, 103, 107 ) eV for (e , , ). While these neutrino
masses are not ruled out by collider experiments, they can be problematic for the recent
SuperKamiokande results and cosmology. In particular, if the neutrinos are stable, then the
masses do not satisfy the bound from overclosing the universe. This constraint is easily
avoided if the neutrinos decay, but this now requires a more detailed analysis.
In conclusion, the requirement that the Higgs field is localized on the TeV-brane, in order
to solve the gauge-hierarchy problem, is in direct conflict with proton decay, although
FCNC constraints can be safely avoided if fermions live in the bulk. To delocalize the
Higgs from the TeV-brane and allow new scenarios we need supersymmetry.
4.4. Supersymmetric models
In the case of non-supersymmetric theories we were forced to have the Higgs field
localized on the TeV-brane, with a small overlapping on the Planck-brane. However, if
supersymmetry is realized, we can relax the above assumption. The Higgs fields can now
live anywhere in the bulk since their masslessness will be stable from radiative corrections.
We will consider two new possibilities. The Higgs can be either
(a) not localized at all, c = 1/2 in Eq. (49) (i.e., conformally flat limit), or
(b) localized on the Planck-brane, c 1/2 in Eq. (49).
These two possibilities, (a) and (b), allow the fermions (and their superpartners) to be
placed on the Planck-brane. The main motivation for having the SM fermion living on
the Planck-brane is to avoid dangerous higher-dimensional operators that, as discussed
in the previous section, could induce large proton decay. By placing the fermions on
the Planck-brane, any higher-dimensional operator will be suppressed by MP [6] (or the
slightly smaller GUT scale of 1016 GeV). This occurs if either the SM fermions are actually
158
confined to the Planck-brane as four-dimensional fields, or they are bulk fields with fivedimensional masses c 1 (see discussion below Eq. (60)). In the latter case, the fermions
form part of the five-dimensional hypermultiplets that, as shown in Section 3, are localized
by the AdS metric on the Planck-brane. Unfortunately with the Higgs off the TeV-brane, the
Yukawa couplings are no longer exponentially suppressed, and the fermion mass hierarchy
cannot be explained by the mechanism outlined in Section 4.2.
Finally, we will consider the SM gauge-boson as a five-dimensional bulk field. This is a
necessary consequence if some of the SM fermions (or the Higgs) live in the bulk. There
is also another motivation here. If the SM gauge-bosons live in the bulk, they will behave
as messenger fields between the two branes, communicating any physics on the TeV-brane
to the Planck-brane. In particular, if supersymmetry is assumed to be broken on the TeVbrane, they will communicate it to the Planck-brane. But also Planckian (stringy) physics
can be communicated to the Planck-brane. Note that at the TeV-brane, Planckian physics
is rescaled down to TeV energies. Therefore a gauge boson can be produced at the Planckbrane with a TeV energy and then propagate to the TeV-brane. At the TeV-brane, it will
interact with Planck scale excitations of the gauge boson, which now have TeV masses.
Thus, we expect that in this scenario we will be able to test GUT or theories of quantum
gravity at the TeV.
4.4.1. Supersymmetry breaking on the TeV-brane
Let us assume that supersymmetry is broken on the brane at y = R. Since
all mass scales at y = R, are rescaled to MP ekR , one can generate small
supersymmetry-breaking soft masses. This presents a new alternative to models with
gaugino condensation where the small scale of supersymmetry breaking is generated by
dimensional transmutation.
Since we assume that the SM gauge superfields live in the 5D bulk, they couple to the
TeV-brane and can feel the breaking of supersymmetry at tree-level. In particular, gaugino
mass terms can be generated from the TeV-brane spoiling the supersymmetric condition
Eq. (37). This will modify the KaluzaKlein decomposition of the gaugino fields given
in Section 3. For example, the full KaluzaKlein mass spectrum will be shifted up with
respect to the boson sector. There will no longer be a massless gaugino mode in the
spectrum. Although the precise value of the KaluzaKlein masses and eigenfunctions will
be left for future work, we can infer here some of the consequences.
The breaking of supersymmetry on the y = R brane will be communicated to the
rest of SM particle superfields by the gauge superfields. Squarks and sleptons on the
Planck-brane will receive masses at the quantum level. Since gauge interactions are flavourindependent, the scalar masses will be flavour diagonal, solving the supersymmetric flavour
problem. Notice that not only the massless mode but all the KaluzaKlein tower will
mediate the breaking of supersymmetry similarly as in Ref. [1114]. Therefore we expect
that the sparticle spectrum will be very similar to that in theories of supersymmetrybreaking by TeV compactifications [1114]. In particular we expect a mass gap between
the scalar masses and the gaugino masses. This is because the scalars will receive masses
at the one-loop level
m2i
i
(TeV)2 ,
4
159
(64)
where i is the gauge coupling of the gauge group under which the scalar i transforms. The
scalar masses will be an order of magnitude smaller than the gaugino masses and fulfill the
relation
m2i
m2j
i
.
j
(65)
The right-handed slepton will be the lightest scalar. However, the lightest supersymmetric
particle (LSP) will be the gravitino. Since it couples to the TeV-brane by 1/MP -suppressed
interactions, its mass will be of order TeV2 /MP . This represents a very light gravitino
with mass m3/2 104 eV and satisfies the usual constraints from cosmology and collider
experiments [47,48].
What about the Higgs? In model (a) the Higgs will also couple to the TeV-brane and
Higgsino masses can be induced at tree-level. In general, the scalar Higgs will also get
masses at tree-level. This can be problematic since the electroweak scale will be of order
TeV. 5 Therefore, we must require that only the Higgsino mass is induced at the tree-level.
The origin of such a pattern of supersymmetry-breaking masses must be addressed by the
fundamental (string) theory. Of course, even if the Higgs soft masses are zero at tree-level,
they will be induced at the loop level. As in the case of the squark and slepton masses,
we expect that the soft Higgs masses will be positive at the one-loop level. Nevertheless,
for the Higgs coupled to the top, there are also negative two-loop effects coming from the
top/stop that can dominate the one-loop contribution and make the Higgs mass negative
[1114]. This can lead to electroweak breaking at a scale an order of magnitude smaller
than the supersymmetry-breaking scale. For model (b), however, the Higgsino mass is zero
since the Higgs superfields do not couple to the TeV-brane. This gives rise to the usual
-problem. The simple solution that we envisage is to introduce a singlet field that couples
to the Higgsinos. This singlet can get a VEV at the electroweak scale and induce Higgsino
masses.
Finally, we want to comment on a different alternative to the models considered here,
where only gravity lives in the five-dimensional bulk, while the complete SM, including
the gauge fields, lives on the Planck-brane. Supersymmetry breaking arising from the
brane at y = R, would then be communicated by gravity to the SM fields. In this
case, soft masses will be of order (MP ekR )2 /MP , and implies that the radius of the
extra dimension satisfies kR ' 6. This model is very similar to gravity-mediated models
of supersymmetry-breaking and we do not expect any new interesting phenomenology.
5 We could lower the supersymmetry breaking scale from a TeV to 100 GeV, but in this case the squark and
slepton masses, that arise at the loop level, will be too small. Another alternative would be to reduce the couplings
of the Higgs to the TeV-brane by increasing its 5D mass, c & 1/2, and having the Higgs slightly localized towards
the Planck-brane.
160
5. Conclusion
Bulk fields living in a slice of AdS have different behaviour, depending on their spin. We
have studied the KaluzaKlein mass spectrum and wavefunctions of scalars, gauge bosons
and fermions and analyzed the conditions for supersymmetry. In the massless sector we
find that the scalar and fermion fields can be localized by the AdS geometry on either
brane (depending on their 5D masses), while the gauge bosons are always nonlocalized.
Furthermore, the gauge bosons couple to each brane with equal strength, and therefore
provide a window for physics of the distant brane.
We have also studied the phenomenological implications of having the SM fields
propagating in a slice of AdS. In non-supersymmetric theories, only the Higgs is required
to be localized on the TeV-brane. The precise scale of the TeV-brane is constrained by
electroweak precision data, and strongly depends on the 5D fermion masses as shown in
Eq. (53) and Fig. 1. Remarkably, there is no significant bound if the 5D fermion masses
satisfy m > 0 /2. For these values of m , the four-dimensional massless fermions are
slightly localized near the Planck-brane and have an exponentially small coupling to the
Higgs. We have shown that this could explain the fermion mass hierarchy by confining the
light fermions to the Planck-brane, and the heavy fermions to the TeV-brane. Similarly,
by studying higher-dimensional operators, we have shown that constraints from proton
decay forces the fermions to be strongly confined towards the Planck-brane. Unfortunately,
these constraints forbid the generation of fermion masses, because now there is very little
wavefunction overlap with the Higgs.
If supersymmetry is present, the massless Higgs mode is protected from radiative
corrections and the Higgs can now live off the TeV-brane. Thus, higher-dimensional
proton decay operators can be suppressed, while at the same time having fermion masses.
However, the warp factor needed to explain the hierarchy becomes impotent, since the
Higgs and fermions are located near the Planck-brane. Assuming that the source of
supersymmetry-breaking is on the TeV-brane, the scale of supersymmetry-breaking will
be of O(TeV). This provides a new alternative to gaugino condensation models. If the
gauge fields live in the bulk, then gaugino mass terms will be generated at tree-level. The
gauge fields will communicate the supersymmetry breaking to the Planck-brane where the
squarks and sleptons live. Thus, the supersymmetric flavour problem is naturally solved.
This model leads to a supersymmetric mass spectrum where the right-handed slepton is
the lightest scalar and the LSP is a very light gravitino. One particularly favourable model,
is when the Higgs field is conformally flat (c = 1/2) and nonlocalized. In this case the
Higgsino mass could also be induced at tree-level.
Finally, we reiterate that the whole scenario, including the details of the source branes,
must eventually be embedded into some underlying theory (such as string theory). In
particular, such a fundamental theory must provide a tree-level supersymmetry-breaking
mechanism on the TeV-brane. It is appealing that the hierarchies in the fundamental scales
of physics can be directly related to the geometry of spacetime. This fact alone, warrants
further investigation of these scenarios.
161
Acknowledgements
We wish to thank Emilian Dudas, Oriol Pujols, and Daniel Waldram for helpful
discussions. One of us (AP) acknowledges the ITP in Santa Barbara for hospitality during
the initial stages of this work. The work of TG is supported by the FNRS, contract no. 2155560.98, while that of AP is partially supported by the CICYT Research Project AEN990766.
References
[1] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263; Phys. Rev. D 59
(1999) 086004.
[2] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370.
[3] J. Lykken, L. Randall, hep-th/9908076.
[4] W.D. Goldberger, M.B. Wise, Phys. Rev. D 60 (1999) 107505; Phys. Rev. Lett. 83 (1999) 4922.
[5] H. Davoudiasl, J.L. Hewett, T.G. Rizzo, hep-ph/9911262.
[6] A. Pomarol, hep-ph/9911294.
[7] Y. Grossman, M. Neubert, hep-ph/9912408.
[8] S. Chang, J. Hisano, H. Nakano, N. Okada, M. Yamaguchi, hep-ph/9912498.
[9] B. Bajc, G. Gabadadze, hep-th/9912232.
[10] K.R. Dienes, E. Dudas, T. Gherghetta, hep-ph/9908530.
[11] I. Antoniadis, C. Munoz, M. Quiros, Nucl. Phys. B 397 (1993) 515.
[12] A. Pomarol, M. Quiros, Phys. Lett. B 438 (1998) 255.
[13] I. Antoniadis, S. Dimopoulos, A. Pomarol, M. Quiros, Nucl. Phys. B 544 (1999) 503.
[14] A. Delgado, A. Pomarol, M. Quiros, Phys. Rev. D 60 (1999) 095008.
[15] D.E. Kaplan, G.D. Kribs, M. Schmaltz, hep-ph/9911293.
[16] Z. Chacko, M.A. Luty, A.E. Nelson, E. Ponton, JHEP 0001 (2000) 003.
[17] K. Benakli, hep-ph/9911517.
[18] M. Schmaltz, W. Skiba, hep-ph/0001172.
[19] A. Kehagias, Phys. Lett. B 469 (1999) 123, hep-th/9911134.
[20] A. Brandhuber, K. Sfetsos, JHEP 9910 (1999) 013.
[21] C. Grojean, J. Cline, G. Servant, hep-th/9910081.
[22] H. Verlinde, hep-th/9906182.
[23] R. Kallosh, A. Linde, JHEP 0002 (2000) 005.
[24] D. Youm, hep-th/9911218; hep-th/0001166.
[25] K. Behrndt, M. Cvetic, hep-th/0001159.
[26] M. Cvetic, H. Lu, C.N. Pope, hep-th/0002054.
[27] E. Witten, hep-ph/0002297.
[28] A. Lukas, B.A. Ovrut, K.S. Stelle, D. Waldram, Phys. Rev. D 59 (1999) 086001.
[29] P.K. Townsend, Phys. Rev. D 15 (1977) 2802.
[30] S. Deser, B. Zumino, Phys. Rev. Lett. 38 (1977) 1433.
[31] E. Shuster, Nucl. Phys. B 554 (1999) 198.
[32] E.A. Mirabelli, M.E. Peskin, Phys. Rev. D 58 (1998) 065002.
[33] I. Antoniadis, K. Benakli, Phys. Lett. B 326 (1994) 69.
[34] P. Nath, M. Yamaguchi, Phys. Rev. D 60 (1999) 116004.
[35] M. Masip, A. Pomarol, Phys. Rev. D 60 (1999) 096005.
[36] T.G. Rizzo, J.D. Wells, Phys. Rev. D 61 (2000) 016007.
[37] W.J. Marciano, Phys. Rev. D 60 (1999) 093006.
[38] A. Strumia, Phys. Lett. B 466 (1999) 107.
162
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
R. Casalbuoni, S. De Curtis, D. Dominici, R. Gatto, Phys. Lett. B 462 (1999) 48.

C.D. Carone, Phys. Rev. D 61 (2000) 015008.
F. Cornet, M. Relano, J. Rico, Phys. Rev. D 61 (2000) 037701.
A. Delgado, A. Pomarol, M. Quiros, JHEP 0001 (2000) 030.
L.E. Ibanez, Phys. Lett. B 181 (1986) 269.
N. Arkani-Hamed, M. Schmaltz, Phys. Rev. D 61 (2000) 033005.
E.A. Mirabelli, M. Schmaltz, hep-ph/9912265.
G. Dvali, M. Shifman, Phys. Lett. B 475 (2000) 295.
T. Gherghetta, Phys. Lett. B 423 (1998) 311.
A. Brignole, F. Feruglio, F. Zwirner, Nucl. Phys. B 516 (1998) 13.

Minimal surfaces and reggeization in the AdS/CFT

correspondence
R.A. Janik a,b, , R. Peschanski a
a Service de Physique Theorique CEA-Saclay F-91191 Gif-sur-Yvette Cedex, France
b M. Smoluchowski Institute of Physics, Jagellonian University Reymonta 4, 30-059 Cracow, Poland
Received 10 March 2000; revised 14 June 2000; accepted 21 June 2000
Abstract
We address the problem of computing scattering amplitudes related to the correlation function
of two Wilson lines and/or loops elongated along light-cone directions in strongly coupled gauge
theories. Using the AdS/CFT correspondence in the classical approximation, the amplitudes are
shown to be related to minimal surfaces generalizing the helicoid in various AdS5 backgrounds.
Infra-red divergences appearing for Wilson lines can be factorized out or can be cured by considering
the IR finite case of correlation functions of two Wilson loops. In non-conformal cases related
to confining theories, reggeized amplitudes with linear trajectories and unit intercept are obtained
and shown to come from the approximately flat metrics near the horizon, which sets the scale for
the Regge slope. In the conformal case the absence of confinement leads to a different solution.
A transition between both regimes appears, in a confining theory, when varying impact parameter.
PACS: 11.15-q; 11.25.-w; 12.38.-t
Keywords: Wilson loops; AdS/CFT correspondence; High energy amplitudes
1. Introduction
The theoretical calculation from first principles of high energy scattering amplitudes
in the so-called soft regime of QCD is among the oldest and yet unsolved problem of
strong interaction physics. The main reason is that it requires a good understanding of 4dimensional gauge field theories at strong coupling which we do not possess till now. In
view of the recent developments of the AdS/CFT correspondence [14] it is thus natural to
address this problem in the new setting proposed in this way. An exact correspondence
for QCD is not yet known, however useful information can be obtained from known
realizations for confining theories.
Corresponding author. E-mail: janik@spht.saclay.cea.fr
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 0 5 - 3
164
R.A. Janik, R. Peschanski / Nuclear Physics B 586 (2000) 163182
We would like to discuss relevant physical properties of scattering amplitudes at

high energy expected from the S-matrix theory of strong interactions [5]. In particular,
Reggeization of scattering amplitudes is expected to occur, i.e., high-energy two-body
amplitudes behaving as A(s, t) = s (t ) (prefactors), where s, t are the well-known
Mandelstam variables. (t) is the Regge trajectory corresponding to singularities of partial
waves at j = (t) in the t-channel. Unitarity, analyticity and crossing relations implied
by the S-matrix theory impose constraints on (t). In particular the Froissart bound [6]
implies that (t = 0) 6 1 and the prefactors of the amplitude are at most like log2 s. Note
that the Froissart bound assumes an underlying confining field theory, or at least a mass
gap, since the scale of the bound is fixed by the particle of smallest mass (e.g., the pion).
In [7,8], we considered large impact parameter and high energy scattering of colourless
states for SU(N) supersymmetric gauge theories in the strong coupling, large N limit
using the AdS/CFT correspondence. The gauge theory scattering amplitude is linked with
a correlation function of tilted Wilson loops elongated along the light-cone directions [9
12]. In the AdS/CFT correspondence, these correlation functions are related to minimal
surfaces in the AdS5 geometry which have the Wilson loops as boundaries. The case
considered in our previous paper Ref. [7,8] involved disjoint minimal surfaces and thus the
necessity of including supergravity field exchanges between the two corresponding string
worldsheets. The dominant contributions were identified and all correspond to real phase
shifts, i.e., purely elastic scattering. In particular, the contribution of the bulk graviton gives
an unexpected gravity-like s 1 behaviour of the gauge theory phase shift in a specific
range of energies and (very) large impact parameters.
The main but stringent difficulty which limited the scope of Refs. [7,8] was that the weak
field approximation in supergravity was shown to be broken unless the impact parameter
L was sufficiently large, namely L/a s 2/7 , where a is the transverse extension of the
Wilson loop. If the above condition is not met, the produced gravitational field in the
dual AdS theory becomes strong, preventing perturbative calculations to be done in this
background.
We will concentrate on a situation where the difficulty with supergravity field exchanges
does not arise, since there exists a single connected minimal surface which gives the
dominant contribution to the scattering amplitude in the strong coupling regime, i.e., when
0 0. This will allow us to extend our study to small impact parameters, where inelastic
channels are expected to play an important rle.
In this approach we will start by considering the correlation function of two Wilson
lines elongated along the two light-cone directions, a configuration which can be used
for the description of high-energy quarkquark or quarkantiquark amplitudes in gauge
theories [911]. The rle of the quarks in the AdS/CFT correspondence will be played, as
in [13,14], by the massive W bosons arising from breaking U (N + 1) U (N) U (1).
The case of IR finite correlators of Wilson loops will be dealt with in a second stage.
The plan of our paper is as follows: in Section 2, we will analyze the correlation function
of Wilson lines leading to an evaluation of q q and qq scattering amplitudes at high energy.
This will be done in the context of the black hole geometry in AdS space [15] (static
Wilson loops were first studied in this background in [16,17]), where one can use a flat
165
metric as a good approximation scheme near the horizon. We analyze the factorizable
structure of the IR divergences and isolate a cutoff independent inelastic amplitude leading
to reggeization. In Section 3, we consider the so-called conformal case of the AdS/CFT
correspondence for N = 4 supersymmetric SU (N) gauge theory, where the AdS5 metric
gives rise to a different minimal surface solution. The problem of the cancelation of the
infra-red divergences is analyzed by considering Wilson loop correlators in Section 4,
leading to the (approximate) derivation of scattering amplitudes between colourless states,
while the conclusions and open problems are pointed out in the final section.
2. Wilson lines and minimal surfaces in quasi-flat geometry

Let us start by defining an appropriate gauge theory observable for q q scattering
amplitudes A(s, t). It is convenient to pass from transverse momentum t = q 2 to impact
parameter space
Z
i
1
l),
A(s, t) =
(1)
d2 l eiql A(s,
s
2
where l is the 2-dimensional impact parameter (in the following we will denote its modulus
by L), and A is the amplitude in the impact parameter space.
In the eikonal approximation the impact parameter space amplitude for q q scattering
is given by a correlation function of two Wilson lines [911] which follow the classical
straight line quark trajectories W1 x1 = p1 and W2 x2 = x + p2 , with |x | =

L, see Fig. 1. The IR cutoff will correspond to a fixed temporal extent of the lines T <
< +T .
The AdS/CFT correspondence gives a recipe [13,14,18] for calculating this correlation
function through
1
l) = e 2 0 Aminimal ,
hW1 W2 i A(s,
Fig. 1. Geometry of the Wilson lines in Euclidean space.
(2)
166
q
2 N in units of the AdS radius,
where hW1 W2 i is the Wilson line correlator 1, 0 = 1/ 2gYM
and Aminimal is the area of the minimal surface in the appropriate background geometry
(e.g., AdS5 S 5 for the conformal N = 4 SYM, an AdS black hole [1517] among other
geometries [19] for confining theories) bounded by the Wilson line segments limited by
the cutoff T . A different approach to discuss the minimal surface problem in the conformal
AdS5 was considered in [20], which concentrated on the elastic part of the amplitude.
Since the disjoint contour formed by the two Wilson line segments is not closed, the
procedure for finding a minimal surface is ambiguous. We will adopt a prescription for
finding the minimal surface for infinitely long lines and then truncating it to a finite
temporal extent parameterized by the IR cutoff T . This implicitly consists of forming
a big Wilson loop closed at large temporal distance by curves drawn on the infinite
minimal surface.
In turn, this procedure defines the appropriate colour decomposition of the associa = (1/2N) +
ated amplitude. Using the well known colour decomposition tija tkl
ij kl
(1/2)il j k , we have

1
(3)
A(s, l) N A0 (s, l) + AN 2 1 (s, l) ,
2
where A 0 (resp. A N 2 1 ) are the amplitudes in the singlet (resp. adjoint) representations.
Using the same strategy as in our first paper [7,8], we will perform the calculation with
Euclidean signature for Wilson lines in the boundary R4 forming a relative angle in the
longitudinal plane and then we will make an analytical continuation into Minkowski space
by rotating the euclidean time coordinate clockwise and the angle anticlockwise (see [21]
in this context):
s
i i log 2 ,
m
T iT .
(4)
Note that a priori there is an ambiguity in making the analytical continuation depending
on the precise choice of the path. This phenomenon did not appear in the context of
large impact parameter near forward scattering discussed in [7,8] since there, the hW W i
correlation function had only simple poles in the complex plane. In the case considered
in this paper the analyticity structure contains branch cuts in the complex plane which have
to be taken into account.
2.1. AdS black hole solution and its flat space approximation
In [15] a proposal was made that a confining gauge theory is dual to string theory in an
AdS black hole (BH) background the relevant part of which can be written as
2
=
dsBH
16 1 dz2 dx dx
+
+ ,
9 f (z) z2
z2
(5)
1 The free propagation of the q and q states is not included in the correlator hW W i, which is thus implicitly
1 2
normalized by 1/ hW1 i hW2 i.
167
where f (z) = z2/3 (1 (z/R0 )4 ) and R0 is the position of the horizon. 2 Although it was
later found that the S 1 KK states do not strictly decouple in the interesting limits [22],
we will use this background to study the interplay between the confining nature of
gauge theory and its reggeization properties. Actually the qualitative arguments and
approximations should be generic for most confining backgrounds 3, as already discussed
in Refs. [19,25].
In order to calculate the scattering amplitude, we have to evaluate the correlation
function (2). Therefore we put the two tilted lines depicted in Fig. 1 on the boundary at
z = 0. Next we have to find the minimal surface in the appropriate geometry which has the
two lines as its boundaries. The relative angle (tilt) in the ty plane and the separation in
the transverse direction x (impact parameter) therefore define the boundary conditions for
the geodesic equations for the string.
As is well known for the Plateau problem of minimal surfaces [26] the boundary
conditions determine the solutions. Although an exact solution for the minimal surface
spanned by the tilted Wilson lines is unknown for the metric (5), the properties of the black
hole (BH) geometry allow for quite a good approximation scheme.
Two salient features of the metric (5) are (i) the standard AdS prefactor 1/z2 close to
the boundary (z = 0), (ii) the existence of a horizon which limits from above the values
of z. A consequence of (i) is that it is most efficient for a minimal surface to perform
the twisting between the two Wilson lines as far away from the boundary as possible.
Property (ii) effectively induces this twisting to occur near the horizon as we shall show
below.
The appropriate minimal surface in the BH geometry will look as follows. Due to
property (i), the minimal surface between well separated lines rises vertically in the
z direction up to the horizon without sizable motion in the other R4 coordinates (see a
schematic representation in Fig. 2). The metric at the horizon is effectively flat
1
2
2 dx dx ,
(6)
dshorizon
R0
and the motion in the z direction is frozen out. Now near the horizon, following property
(ii), the minimal surface performs the twisting (not displayed in Fig. 2) corresponding to
the tilt angle between the initial Wilson lines. At this stage we thus have to find a minimal
surface between the lines at an angle in the flat space metric (6). Finally the surface falls
off again vertically towards the boundary. The area of the vertical pieces is removed by
the standard subtractions [13,14], so the resulting area which enters the formula for the
amplitude (2) may be approximated by the area of the flat space piece.
We will now substantiate this intuitive picture with a more quantitative study of
the geodesic equation for the string. Let us determine under what conditions the
2 Compared to standard coordinates [17] we used U = z4/3 and U = R 4/3 .
T
0
3 Two other geometries for (supersymmetric) confining theories have been discussed recently [23,24]. They
have the property that for small z, i.e., close to the boundary, the geometry looks like AdS5 S 5 (in [24] up to
logarithmic corrections related to asymptotic freedom) giving a coulombic q q potential. For large z the geometry
is effectively flat. In all cases there is a scale, similar to R0 above, which marks a transition between the small z
and large z regimes.
168
Fig. 2. The minimal surface in the black hole geometry. The Wilson lines are drawn here with
vanishing angle of tilt = 0.
minimal surface spanned by the two tilted Wilson lines is indeed predominantly flat and
concentrated near the horizon following the general line of discussion of [19].
The minimal surface equations follow from the NambuGoto action:
1
S=
2 0
ZT
l(
Z )/2
d
T
det hab ,
(7)
l( )/2
where the induced metric on the worldsheet is

hab Gij
Xi (, ) Xj (, )
.
v a
v b
(8)
coordinates (z, x ) in (5), Gij is the background metric, v 0 , v 1

Xi stands for general
, and l( ) = L2 + 2 2 is the Euclidean distance between points on the two Wilson

lines with the same value of the time coordinate .
As a first remark we note that using the background metric (5) the terms in the induced
metric h corresponding to the twisting are of the form
2 2
t
y
1
+
.
(9)
z2
Hence, near the boundary (z 0), the minimization will not change noticeably the twist
angle. Thus the boundary conditions are frozen and transported to the vicinity of the
horizon.
For further discussion we shall make an approximation (similar to [20]) of neglecting
explicit dependence in the EulerLagrange equations following from (7) and leaving
it only in the implicit dependence on the boundary conditions through l( ). Within this
approximation the estimate of [19], made for the case of the static q q potential may be
directly applied to our problem.
In Ref. [19] a distance d is defined, for all metrics giving confinement, which measures
the transverse distance (on the boundary) over which the string worldsheet significantly
deviates from being flat. In all cases the ratio d/ l( ) 0 when l( ) . Depending on
169
the confining metric considered, d behaves as a logarithm or a power of l( ) smaller than

one. It is interesting to note that the condition d/ l( ) 1 leads to a lower bound on the
impact parameter L since the above condition is most restrictive for the smallest value of
l( ), which is equal to L.
The precise dependence of d(l( )) on the horizon scale R0 depends on the metric
considered. For instance for the metric (5) rescaling arguments lead to a dependence
2/3
d(l( )) R0 log l( ). Therefore, as long as the impact parameter is large with respect
to R0 the approximations considered in this section should be valid.
However, it may of course happen that the impact parameter distance between the two
Wilson lines becomes much smaller than R0 . In this case (see Fig. 2) the minimal surface
problem becomes less affected by the black hole geometry (or the large z behaviour of the
different metrics [23,24]) and will just probe the small z region of the geometry.
The precise behaviour at these shorter distances will depend on the type of gauge theory
and, in particular, on the small z limit of the appropriate metric. In this paper we will
consider the generic case (from the 4D (S)YM point of view) when this limit resembles
the original AdS5 S 5 geometry [13]. We will consider this conformal (non-confining)
regime in detail in a further section. We note that the same behaviour can be equivalently
obtained through rescaling, by keeping the impact parameter fixed and putting the scale
R0 .
Let us concentrate in the following on the case when the impact parameter is larger than
the scale R0 . To summarize the discussion, the string is then to a large degree concentrated
in the region near the horizon (6) with the boundary conditions essentially transported
from z = 0. We are thus led first to calculate the area of the minimal surface bounded by
the tilted lines in the flat geometry (6) at the horizon. We will first perform the calculation
in euclidean signature and then perform the analytical continuation (4).
2.2. Helicoid geometry
The basic building block of our construction is a minimal surface spanned by two
straight line segments of length 2T , corresponding to the two Wilson lines separated by a
distance L in the transverse direction x and with a relative angle in the longitudinal
plane:
L1 : (, 0, 0, 0),
L2 : ( cos , sin , 0, L).
(10)
It is well-known that in the flat R4 geometry the minimal surface with infinite boudaries
= + is a helicoid. We will also be interested by the truncated helicoid 4
where = T . . . T .
Let us recall the minimal surface solution in flat space. The helicoid is the only regulated
(spanned by straight lines) minimal surface. The truncated helicoid solution may be
parametrized by
4 For finite cutoff T in flat space, the truncated helicoid obviously remains a solution if one adds the boundary
helices at = T , T as new boundaries. Note, however, that with these boundaries the helicoid may be an
unstable [27] minimum for a too large value of the cutoff. We will not consider this problem in the present paper.
170
,
y = sin
,
x = ,
L
L
where = T . . . T and = 0 . . . L and is the total twisting angle.
Its area is given by the formula
s
ZL
ZT
2 2
d 1 + 2
Area S(T ) = d
L
0
T
q
s
2 2
2
2
2
1 + TL2 + TL
T
L
q
log
= LT 1 +
+
.
2 2
L2
2
1+ T T
t = cos
L2
(11)
(12)
Let us now perform the analytical continuation (4), which links euclidean correlation functions in gauge theories with minkowskian ones directly related to scattering amplitudes. A
naive continuation of the area formula (12) leads to a pure phase factor in (2):
q
q
s
2 2
2 N
2gYM
2
2 2
1 + T L2 + TL
T
L
,
log q
i LT 1 +
+
(13)
exp
2R02
2
L2
T 2 2
T
1 + L2 L
q
2 N/(2R 2 ) coming from the
where 1/(2 0 ) in (2) has been replaced by the factor 2gYM
0
flat metric (6).
However, the analytic structure of the euclidean area (12) involves cuts in the complex
T , planes and thus leads to an ambiguity coming from the branch cut of the logarithm. In
fact when performing the analytical continuation we have to specify the Riemann sheet of
the logarithm (i.e., log log +2in). This leads to an additional real multiplicative factor
in (2):
q
2 N
2gYM
L2
,
(14)
exp n
2R02
the form of which is uniquely fixed by the euclidean expression (12) up to a choice of the
integer n. Within the classical approximation which we have been using it is not possible
to determine the value of n. On a more physical ground, in Section 4, we will relate
the analogue of the label n which appears in the calculation of Wilson loop correlators
with multivalued saddlepoint minima of a minimization equation and thus to different
classical solutions. The determination of the relative weights of the various contributions
goes beyond the classical approximation used throughout this paper. 5
As can be seen the contribution (14) is cutoff-independent.
Another useful way of deriving the above factor (14) can be directly obtained from the
integral leading to (12). This method can be generalized to more complicated background
5 We also note the close similarity of the n, L and dependence in (14) with an analogous factor
exp(nL2 /( )) in the imaginary part of the D-brane scattering amplitude [28] where n labels the poles of
the appropriate string partition function between the branes.
171
geometries, for instance to the conformal case, which we will consider later, for which we
lack an exact expression of the form (12).
Let us perform only the first part of the analytical continuation (4) i but
otherwise remain with the time variable T in Euclidean space. This procedure yields the
expression:
s
L/
Z
ZL
2 2 L2
.
(15)
d d 1 2 =
L
2
L/
We see that the imaginary part may be obtained by integrating (n times) around the branch
cut of the square root. A convenient reinterpretation
of the above formula follows from
q
performing the change of variables 0 = 1
r
+L/
Z
2in
L/
2 2
L2
2 2
.
L2
Then we get
d 0 = ni
L2
(16)
which is effectively twice (in) the area of a minimal surface bounded by a half-elipse
of radii L and L/. This T -independent imaginary part is unaffected by the second part of
the analytical continuation (4), and leads directly to the factor (14).
2.3. Reggeization in quark(anti)quark scattering
Our result for the Wilson line correlation function for the AdS BH geometry gives rise
to the following contributions
q
q
s
T 2 2
2 N
T
2gYM
2
2
2
1
+
+
T
L
L
L2
log q
i LT 1 +
+
A n = exp
2R02
L2
2
T 2 2
T
1 + L2 L
q
2 N
2gYM
L2
.
(17)
exp n
2R02
There is a divergent phase in the above amplitude when the temporal length of the lines T
goes to infinity. We interpret this divergence as reflecting the expected IR divergence of the
qq scattering amplitude [10,11]. A consistent way to eliminate this cutoff dependence is
to consider an IR finite physical quantity like scattering of two q q pairs (see Section 4).
In the present case of Wilson lines, the specific factorized form of (17) allows for a
determination of an IR finite contribution, which can be interpreted as an effect of inelastic
channels on the Wilson line correlator.
It is known since a very long time that the superposition of long range and short
range potentials in the Schroedinger equation leads to a factorization formula for the
relevant S matrix elements for each partial wave [29]. For instance in nuclear physics, the
superposition of long range coulombic and short range interactions leads to a factorization
172
into the elastic coulombic S matrix element and a short range amplitude modified by
the long-range background. The elastic S matrix may be treated as a redefinition of the
asymptotic initial and final states. The amplitude reads
A(l, s) = e2i(l,s) T (l, s),
(18)
where is the real phase shift due to the elastic long range interactions and T is the short
range part of the amplitude. For instance, in the QED result for electron scattering [30,31],
the real phase shift exhibits a divergence which can be written as
2
2
L
e
coth log
,
(19)
e2i(l,s) exp i
4
4T 2
where 1/T has been substituted for a fictitious photon mass (IR regulator).
In hadronic interaction physics [32], a similar factorization appears for the S matrix
elements for 2-body channels in terms of an elastic contribution and an amplitude T (l, s)
which, by unitarity of the S matrix, arises from the contribution of many inelastic channels
to the 2-body S matrix [33]. In this context the amplitude (18) can be related to the
inelasticity (overlap matrix) in the scattering namely
1 1 2f (l, s)
,
(20)
T (l, s) =
2
where the overlap matrix elements f (l, s) are defined from the 2-body S matrix
contribution to unitarity |S(l, s)|2 1 2f (l, s).
We are led to interpret our resulting amplitude (17) in the same way. The factor (13) can
be treated as redefining the initial and final q q states due to long range interactions. 6
Naturally, this phase is IR-divergent. The analogous inelastic contribution Tn (l, s) is
obtained to be the cutoff independent factors (14). Note that this physical interpretation
requires the integer n to be positive. We will return to the discussion of the n dependence
in a further section.
Let us discuss both factors of the amplitude (17). The contribution of the real phase
shifts behaves in the large T limit like
q

2 N
2gYM
2

2 eT
L
2
2
log
+
O
1/T
i
T
+
.
(21)
exp
2R02
L
The appearance of the IR divergent T 2 and L2 log T terms in the phase shift can be linked
with the linear confining potential of the theory.
The effect of the confining potential is expected to generate inelastic channels through
the phenomenon of string breaking and/or closed string emission. Within the above
framework, where we select initial and final q q states, this contribution is expected to
appear as an inelastic real factor in the amplitude, while the phase factor diverges with
T .
6 It is clear that there remains a freedom in attributing a finite real phase shift either to the redefinition of the
states or to the interaction. Here we adopt the convention that T (l, s) is purely real and thus contains information
only on the inelasticity.
The inelastic q q interaction amplitude at level n is

q
2 N
2gYM
L2
,
Tn (l, s) = exp n
2R02
173
(22)
where the initial and final states are both q q.

It can be easily fourier transformed into
transverse momentum space giving
R2
0
iR 2 ln s 1+ 2n2g2 N t
YM
s
.
Tn (s, t) = q 0
2 N
n 2gYM
(23)
This contribution is thus reggeized with a linear Regge trajectory with unit intercept and
the slope given by the string tension related to the horizon distance R02 .
It is worthwhile to consider what changes in the preceeding discussion if we go from
qq scattering to qq scattering. In geometric terms, this corresponds to changing the
orientation of one of the lines, and since the string worldsheet spanned on the Wilson lines
is oriented, the twisting angle of the helicoid changes as
.
(24)
Upon analytical continuation this means that log s changes to i log sei ,
as required by crossing properties, which are seen to have here a very simple geometric
interpretation. We note that in the asymptotically high energy limit log s 1, one
obtains the same factors (17) for both qq and q q channels. Keeping the next to leading
correction corresponding to log s log sei preserves the crossing relations between
those channels.
Finally let us compare our result with the general structure of Wilson line correlators
at weak gauge coupling. Indeed, the large T dependence of the q q amplitudes we discuss
reflects IR divergences which appear already in perturbative (weak coupling) calculations
of the same quantities.
For instance in the case of QED the whole dynamics is contained in the infinite phase
factor (19) and the divergence is logarithmic.
The (renormalon improved) 1-loop QCD result [11,34] for q q scattering is

T
L2
1 s
log
2
,
(25)
exp

L
where is an undetermined nonperturbative parameter. We note the compatibility between

the nonperturbative cutoff independent piece in (25) and an analogous term in our result
(22). Our nonperturbative result gives a hint on the scale and coupling dependence.
If the intercept one common to all contributions Tn (s, t) (see formula (23)) is not spoiled
by the different weights corresponding to fluctuations of the worldsheet around classical
solutions, it would be a candidate for the intercept one trajectories (the so-called pomeron
and odderon) which are expected to emerge from a confining strongly interacting gauge
theory.
174
3. Conformal case
The flat metric approximation which we have used to derive the resulting area (12)
assumed that the impact parameter L is sufficiently large with respect to the scale set by
the horizon radius R0 (or a similar scale in the backgrounds [23,24] interpolating between
a confining geometry at large z and approximately AdS5 S 5 near the boundary z = 0).
In this regime the dominant contribution to the amplitudes came from the part of the string
worldsheet stretched near the horizon.
If we go to smaller impact parameters L < R0 (and also for T < R0 ), the minimal
surface would only penetrate into a limited region near the boundary z = 0, see Fig. 2.
In the scenarios which behave better at short distances than the original BH proposal, the
metric becomes closer and closer to the conformal AdS5 case. We note that the AdS5 S 5
setting is directly related to scattering in the N = 4 SYM. This different geometry leads to
a qualitatively new behaviour which we now analyze.
3.1. The conformal AdS5 case
In the case of N = 4 SYM corresponding to the AdS5 S 5 background we do not know
yet the exact generalization of the helicoid, and some approximation scheme is needed. As
in the previous case, we will concentrate on extracting the inelastic contribution which
appears also here to be independent of the IR temporal cutoff T . We use the method
outlined in Section 2 leading to formulae (15)(16).
Within a variational approximation approach, we will look for a minimal solution in a
restricted set of surfaces (generalized helicoids) parameterized by
,
L
,
y = sin
L
x = ,
t = cos
(26)
(27)
(28)
z = z(, ).
(29)
Evaluation of the induced metric gives rise to the following area functional:
s

ZT
ZL

1
2 2
1 + z2 + z2 .
d d 2
1+ 2
z
L
T
(30)
p
We perform a change of variables 0 = 1 + 2 2 /L2 which yields
r
1
2 0
L 1+
ZT
d
T
2 2
L2
d 0
1
z2
1 + z2 + z2 0 .
(31)
As in the previous section the cutoff independent part is obtained from the branch cut
structure of the area functional (31). The analytic continuation i changes the
175
boundary conditions for the minimal surface to be a half-elipse

of width L/ and height
p
L (the upper integration limit in (31) then becomes L 1 2 2 /L2 ). Due to conformal
invariance we know that the minimal area has the following form:
Aminimal = f (L/, ) + g(),
(32)
where is the 5th AdS coordinate where we put the D3 brane probe. translates directly
into the mass of the W bosons which play here the role of quarks. We do not expect higher
poles in than first order, which are in the standard way subtracted out [13,14,35,36], so
we have at most a logarithmic behaviour in L/.
It is possible to obtain an approximate result in the high energy limit from
known properties of Wilson loop expectation values [13,14,35,36]. The half-elipse has two
cusps each with an angle /2, whose contribution to Aminimal can be obtained from the
results of [35,36]. This leads to the following logarithmic terms:
2
L
1
F (/2) log
,
2

(33)
where F () is a complicated function calculated in [35,36] (F (/2) 0.3 ). The

independent term g() in (32) can be approximated by noting that at high energies the
half-elipse is very much elongated and looks like parallel lines of length L, roughly 2L/
apart. An approximate evaluation is then given by integrating the coulombic potential [13,
14]:
ZL
c
0
d 0
= c ,
2
2
02
4
L
(34)
where c = 8 3 / 4 (1/4) is the coefficient in front of the (screened) coulombic potential.

So we get
2
2
N

n F (/2) 2gYM
4
2gYM N
2
L
n 2 4
2
s (1/4)
.
(35)
Tn (l, s)
log s
Here, as in the case of the confining theory, the values of n and the weights of the different
components Tn (l, s) are not specified.
Let us comment on the behaviour of the various components. In all cases we obtain
a factorized energy behaviour with no moving Regge trajectories. We note that for n
positive a similar energy dependence (i.e., with intercept greater than 1 and a (nearly)
flat Regge trajectory) is obtained by resumming the leading log s terms in the perturbative
expansion at weak coupling ([3741]; for a discussion in impact parameter space see [41])
for the singlet exchange amplitude. In the conformal case, there remains a nonperturbative
screening effect (already
present for the static q q potential [13,14]) which appears as the
q
2 N g 2 N in the exponent of s.
change gYM
YM
Considering the impact parameter dependence and its Fourier transform to momentum
space, in the window of convergence (the exponent of L in (35) between 2 and 3/2),
we get
176
Tn (s, t) is
4
1+n 2 4
(1/4)
2 N
2gYM
2
1+n F (/2)
2
1
t
2 N
2gYM
2
(36)
Otherwise one observes either an UV divergence (for exponent values less than 2) or an
IR one (for values larger than 3/2). For positive values of n the IR divergence requires a
careful treatment which is beyond the scope of this paper. Note that, in the case of N = 4
SYM, it can lead to infra-red divergent pieces also in the inelastic amplitude, as is the case
already in the perturbative limit [11,42].
3.2. Conformal/non-conformal transition
As already mentioned, the result (36) obtained for the pure AdS5 S 5 case should give
the dominant behaviour also for the confining theory for impact parameters small with
respect to the horizon scale 7 R0 (or more generally an analogous transition scale in the
geometries [23,24]). Indeed this R0 provides a natural value of the impact parameter cutoff
L0 . Thus even in the confining theory, when the impact parameter is decreased and gets
smaller than R0 , we expect a transition from the set of components (22) to the results (35),
as long as the relevant geometry for small z is similar to AdS5 S 5 .
This process can be observed by noting that both (22) and (35) were derived from a
minimal surface spanned on a semielipse. The result for impact parameters L > O(R0 ) was
obtained by using the area law for Wilson loops, while the conformal case (corresponding
here to L R0 ) used an approximation using coulombic potential. The solution of the
appropriate minimal surface problem in the full geometry would lead to an interpolation
between the two extreme cases.
4. Wilson loop correlators and scattering amplitudes

We saw that an inherent feature of the qq scattering amplitude is its IR divergence. In
order to remedy this, and also to show a context where the finite behaviour of the inelastic
amplitudes calculated in the previous section appears directly without the infinite phases,
we are led to consider the scattering of two q q pairs of transverse size a, and impact
parameter distance L. This process is interesting to study in itself, since it gives some
information on the scattering amplitudes between colourless states in gauge theories at
strong coupling.
For this setup we have to calculate the correlation function of two Wilson loops [12],
where the loops are choosen to be elongated along the time direction and have a large
but arbitrary temporal length T (the exact analogue for Wilson loops of T considered in
the previous section). However, the cutoff dependence on T is expected to be removed
together with the related IR divergence which was present for the case of Wilson lines.
For large positive and negative times the minimal surface will be well approximated
by two seperate copies of the standard minimal surfaces for each loop separately. When
7 We need also a sufficiently small T parameter.
177
we come to the interaction region, and for L sufficiently small, one can lower the area by
forming a tube joining the two worldsheets. Since we want to calculate the normalized
correlator hW1 W2 i / hW1 i hW2 i, the contributions of the regions outside the tube will cancel
out (in a first approximation neglecting deformations near the tube). Therefore we have just
to find the area of the tube, and subtract from it the area of the two independent worldsheets.
It is at this stage that we see that the result does not depend on the maximal length of the
Wilson loops T , and hence is IR finite. The whole contribution to the amplitude will just
come from the area of the tube.
Since we cannot obtain an exact minimal surface for these boundary conditions, let
us perform a variational approximation. Namely, we will consider a family of surfaces
forming the tube, parameterized by Ttube , which has the interpretation of an effective
time of interaction. Then we will make a saddle point minimization of the area as a function
of this parameter.
Suppose that the tube linking the two Wilson lines is formed in the region of the time
parameter t (Ttube, Ttube ). In our approximation its two sides are formed by sheets
of the helicoid solution (of area S(Ttube ), see (12) for the euclidean case and (13) for
theqminkowskian one). The front and back will be each approximated by strips of area
2 2 /L2 (we assume a, L > R ).
aL 1 + Ttube
0
The total area corresponding to the two Wilson loops is then given by
s
s
T
Ztube
T 2 2
2 2
d 1 + 2 + 2aL 1 + tube2 4a Ttube,
Area(Ttube ) = 2L
L
L
(37)
Ttube
where 2aTtube is the contribution of each individual Wilson loop to the normalization
1/(hW1 i hW2 i) of the Wilson loop correlation function.
Analytically continuing the area formula (37) to the Minkowskian case and using a
convenient change of variables, the Minkowskian area can be put in the following simple
form

2L2
sin 2
+ cos 2 sin ,
(38)
+
Area(Ttube ) =
2
the new variational parameter.
where a/L and sin = iTtube /L isq
2 N 0) the parameter is dynamically

In the strong coupling limit ( 0 = 1/ 2gYM
determined from the saddle point equation:
0=
Area()
= cos (cos )
sin .
(39)
It is easy to realize that for large enough energy, there exists a solution with n .
Inserting this solution into the area (38) we find
Area() =
2L2
n + 2aL(1)n ,
(40)
where we retain the physical solutions with n positive integer. We thus find a set of
solutions very similar to the inelastic factor obtained in Section 2. The modification due to
178
the frontback contribution 2aL is negligible in the Fourier transformed amplitude for
momentum transfer t a/R02 . Also this term is probably more dependent on the
treatment of the frontback parts of the tube in our approximation.
It is interesting to note that the minimization (39) gives rise in a natural way to a similar
set of solutions parameterized by integers as found from the branch cut arguments in
Section 2. Each value of n corresponds to a saddle point, i.e., a classical solution. The
determination of the weights of each component to the total scattering amplitude is beyond
the reach of the classical approximation.
For completeness, let us briefly discuss the general saddle point solution. For lower
energies, there are families of solutions also leading to reggeized behaviour but with
distorted trajectories. For small there exist solutions with imaginary and thus leading
to elastic parts of the amplitude. The study of these solutions is beyond the scope of the
present paper. For too large impact parameters we may enter the purely elastic regime
found in [7,8] which does not correspond to connected minimal surfaces (the GrossOoguri
transition [18]).
As a word of caution (and incentive for further study) we note that the saddle point in
terms of Ttube is mainly driven to complex values. This indicates that a complete treatment
and an investigation of the GrossOoguri transition requires a more refined study of the
tube minimal surface.
Let us analyze the properties of the resulting amplitude. Recalling that charge
conjugation acting on one of the q q pairs is equivalent to considering the transformation
i , it is convenient to analyze the components of definite signature [5] with the
even and odd contributions given by
2
2
Tn (l, s) = e
2gYM N L2
R02
2gYM N L2
i
R02
(41)
Note the relative factor of 2 in the exponent in comparison with (22) due to the two sheet
structure of the minimal surface.
Using the Fourier transform (1) we finally get
Tn (s, t) =
iR02 ln s
iR02 ln(s)
q
s n (t ) q
(s)n (t ) ,
2
2
2n 2gYM N
2n 2gYM N
(42)
where
n (t) = 1 +
R2
q 0
t.
2 N
4n 2gYM
(43)
Let us consider the contribution with n = 1 which is dominant at large L. It is easy to

realize that the amplitude (42) corresponds to specific Regge singularities in the S-matrix
framework, namely double Regge poles whose trajectory is given by 1 (t). Indeed, using
R s j dj
the usual Mellin transform s 2i(j
) , it can be written in the following equivalent
forms:

iR 2
1 (t )
s
T1 (s, t) = q 0
(s)1 (t )
2 N
2 2gYM

Z
R02
dj eij/2 s j
i sin j/2
= q
,
(j 1 (t))2 cos j/2
2 N
2 2gYM
C
179
(44)
where the complex contour C can be taken around the Regge (di)pole trajectory 1 (t) and
the signature factors are either sin(j/2) or i cos(j/2) depending on the positive or
negative signature.
Let us discuss the contributions Tn to the amplitude with n > 1. In the absence of a
direct determination of their relative weights, it is interesting to note that unitarization
of Regge amplitudes in the S matrix framework [32,33] leads to a similar decomposition
where the Tn correspond to Regge pole/cut singularities. In particular, the overlap matrix
formalism [33], see (20), leads to a specific model forqthe relative weights of the Tn s, if

2 N
2gYM
L2
for the inelasticity. In
we assume a gaussian distribution f (l, s) f0 exp
2
2R0
this framework [33] unitarity is fulfilled whenever 0 < f0 < 1/2. However, the derivation
of the Wilson line/loop correlation function does not allow us to give model-independent
predictions for these weights in the total amplitude.
Finally let us comment on the relation of our results on the trajectory (t) with the
glueball spectrum calculations [4346]. An extrapolation of the trajectory (43) to positive
t leads to masses of the form
q
M 2 = 4n(J 1)
2 N
2gYM
R02
(45)
where J is the spin and n labels the different trajectories. Because of the appearance of
coupling constant dependence it is easy to see that these states correspond to massive
string states and not to supergravity fields associated with the glueballs found in [4346].
Indeed, the latter states have masses proportional just to 1/R02 and spin limited by J 6 2.
The appearence of massive string states is not surprising in our case as we consider an
extended string worldsheet between the two Wilson loops instead of a supergravity field
exchange. The transition between both situations and thus the relation between both sets
of states remains an open problem.
We should note that our approximations for calculating the Wilson loop correlator
(which is the channel relevant for glueballs) are rather crude and become problematic
at small t (consider the discussion after (40)). Therefore the extrapolation of the linear
trajectory into the glueball regime can easily break down. Unfortunately the complexity of
the minimal surface problem with the Wilson loop boundary conditions does not allow us
to make more quantitative estimates.
180
5. Conclusions and outlook

Let us give our main conclusions. By computing Wilson line and Wilson loop correlation
functions in the framework of the AdS/CFT correspondence we show a relation between
minimal surface problems in AdS5 metrics and reggeization in gauge field theory at strong
coupling.
For Wilson line correlators, we isolate in certain cases IR finite inelastic amplitudes
coming from the branch cut structure of the analytical continuation of helicoid-like
surfaces, i.e., minimal surfaces with straight line boundary conditions corresponding to
classical trajectories in Minkowski space.
We considered three cases: (i) flat metric approximation of an AdS black hole metric
giving rise to Regge amplitudes with linear trajectories, (ii) an approximate evaluation for
the conformal AdS5 S 5 geometry leading to flat Regge trajectories 8 and (iii) evidence
for a transition, in a confining theory, from behaviour of type (i) to (ii) when the impact
parameter decreases below the interpolation scale set by the horizon radius. In this case,
confinement provides a natural IR cutoff scale.
In a second stage we considered the correlation function of two Wilson loops elongated
along the light cone directions for the confining geometry. This configuration corresponds
to a high energy scattering amplitude between colourless q q states. We use a variational
approximation where the minimal surface is constructed from two helicoidal sheets. As
expected, the obtained amplitude is free from IR divergences and gives rise to reggeization
with a linear trajectory with unit intercept. For high energies the amplitude is imaginary
and thus mainly reflects the inelasticity of the process.
These results call for some comments.
We note that the structure of our resulting amplitudes for the confining case (in particular
the n, and L dependence) matches the calculations of the imaginary part of flat space
D-brane scattering amplitudes [28]and some specific Wilson loop correlators 9 [47,48],
when the effective string length 0 is taken to be set by the horizon radius in our case.
It is interesting to note that the imaginary part in those calculations is generated from the
singularities of the string amplitudes which are an infinite set of poles. The slopes of the
trajectories are the same as in our case (43), while the intercepts are different. However
the geometrical configurations in [47,48] is quite different from the one we considered in
Section 4. Even in the flat space approximation it would be useful to have a direct string
calculation of the tube configuration.
Beyond the flat space approximation, we want to emphasize the interest of solving
exactly the well defined mathematical problem of finding the generalization of the helicoid
for various AdS metrics, i.e., the minimal surface spanned between infinite lines forming
an angle at the boundary. Another goal is to go beyond the classical approximation in
order to derive the n-dependent weights to the scattering amplitudes.
8 A remaining IR divergence in the inelastic amplitude is still present in the absence of confinement.
9 We have extracted the imaginary part from the formulae in Ref. [47,48], along the lines of [28].
181
Indeed the generalization of the helicoidal geometry in AdS space seems to be a building
block for high energy scattering amplitudes in gauge theories at strong coupling.
Acknowledgements
R.J. was partially supported by KBN grants 2P03B00814, 2P03B08614. We thank T.
Garel and B. Giraud for useful remarks.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231.

S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253.
O. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory
and gravity, hep-th/9905111.
M. Froissart, R. Omnes, Mandelstam Theory and Regge Poles, Frontiers in Physics, Benjamin,
1963.
M. Froissart, Phys. Rev. 123 (1961) 1053.
R.A. Janik, R. Peschanski, Nucl. Phys. B 565 (2000) 193.
R.A. Janik, Gauge theory scattering from the AdS/CFT correspondence, Cargese Summer
School 1999, hep-th/9909124.
O. Nachtmann, Ann. Phys. 209 (1991) 436.
H. Verlinde, E. Verlinde, QCD at high energies and two-dimensional field theory, hepth/9302104.
G.P. Korchemsky, Phys. Lett. B 325 (1994) 459.
O. Nachtmann, High energy collisions and nonperturbative QCD, hep-ph/9609365 (see, e.g.,
Eq. (3.87) for colourless state scattering).
J. Maldacena, Phys. Rev. Lett. 80 (1998) 4859.
S.-J. Rey, J. Yee, Macroscopic strings as heavy quarks in large N gauge theory and anti-de Sitter
supergravity, hep-th/9803001.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 505.
S.-J. Rey, S. Theisen, J.-T. Yee, Nucl. Phys. B 527 (1998) 171.
A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Yankielowicz, Phys. Lett. B 434 (1998) 36.
D.J. Gross, H. Ooguri, Phys. Rev. D 58 (1998) 106002.
Y. Kinar, E. Schreiber, J. Sonnenschein, Nucl. Phys. B 566 (2000) 103; see also Section 11 in
[25].
M. Rho, S.-J. Sin, I. Zahed, Phys. Lett. B 466 (1999) 199.
E. Meggiolaro, Z. Phys. C 76 (1997) 523; Eur. Phys. J. C 4 (1998) 101; Phys. Rev. D 53 (1996)
3835.
H. Ooguri, H. Robins, J. Tannenhauser, Phys. Lett. B 437 (1998) 77.
A. Kehigas, K. Sfetsos, Phys. Lett. B 456 (1999) 22.
C. Angelantonj, A. Armoni, RG flow, Wilson loops and the dilaton tadpole, hep-th/0003050.
Y. Kinar, E. Schreiber, J. Sonnenschein, N. Weiss, Quantum fluctuations of Wilson loops from
string models, hep-th/9911123.
A.T. Fomenko, The Plateau Problem, Gordon & Breach, 1989.
A. Boudaoud, P. Patrcio, M. Ben Amar, Phys. Rev. Lett. 83 (1999) 3836.
C. Bachas, Phys. Lett. B 374 (1996) 37.
A. Messiah, Quantum Mechanics, North-Holland, 1961.
182
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
H. Cheng, T.T. Wu, Phys. Rev. Lett. 22 (1969) 666.

H. Abarbanel, C. Itzykson, Phys. Rev. Lett. 23 (1969) 53.
A. Biaas, L. van Hove, Nuovo Cimento 38 (1965) 1385.
L. van Hove, Rev. Mod. Phys. 36 (1964) 655.
I.A. Korchemskaya, Nucl. Phys. B 490 (1997) 306.
N. Drukker, D.J. Gross, H. Ooguri, Phys. Rev. D 60 (1999) 125006.
H. Ooguri, Prog. Theor. Phys. Suppl. 134 (1999) 153.
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; Sov. Phys. JETP 45
(1977) 199.
I.I. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 (1978) 822.
G.P. Salam, Quarkonium Scattering at High Energies, PhD thesis, University of Cambridge,
1996.
Z. Bern, J. Rozowsky, B. Yan, Two loop N = 4 supersymmetric amplitudes and QCD, hepph/9706392, talk given at DIS 97.
C. Csaki, H. Ooguri, Y. Oz, J. Terning, JHEP 9901 (1999) 017.
R. de Mello Koch, A. Jevicky, M. Mihailescu, J.P. Nunes, Phys. Rev. D 58 (1998) 105009.
R.C. Brower, S.D. Mathur, C.-I. Tan, Glueball spectrum for QCD from AdS supergravity
duality, hep-th/0003115, talk given at DIS 97.
N.R. Constable, R.C. Myers, JHEP 9910 (1999) 037.
S. Chaudhuri, Y. Chen, E. Novak, Pair correlation function of Wilson loops, hep-th/9910183.
S. Chaudhuri, E. Novak, Supersymmetric pair correlation function of Wilson loops, hepth/0002046.

How massless are massless fields in AdSd

L. Brink a , R.R. Metsaev b,c, , M.A. Vasiliev d,1
a Department of Theoretical Physics, Chalmers University of Technology,
S-412 96 Gteborg, Sweden

b Department of Theoretical Physics, P.N. Lebedev Physical Institute,
Leninsky prospect 53, 117924 Moscow, Russia

c Department of Physics, The Ohio State University, Columbus, OH 43210-1106, USA
d Albert-Einstein-Institut, Max-Planck-Institut fr Gravitationsphysik, Am Mhlenberg 1,
D-14476 Golm, Germany

Abstract
Massless fields of generic Young symmetry type in AdSd space are analyzed. It is demonstrated
that in contrast to massless fields in Minkowski space whose physical degrees of freedom transform
in irreps of o(d 2) algebra, AdS massless mixed symmetry fields reduce to a number of irreps
of o(d 2) algebra. From the field theory perspective this means that not every massless field in
flat space admits a deformation to AdSd with the same number of degrees of freedom, because it is
impossible to keep all of the flat space gauge symmetries unbroken in the AdS space. An equivalent
statement is that, generic irreducible AdS massless fields reduce to certain reducible sets of massless
fields in the flat limit. A conjecture on the general pattern of the flat space limit of a general AdSd
massless field is made. The example of the three-cell hook Young diagram is discussed in detail.
In particular, it is shown that only a combination of the three-cell flat-space field with a graviton-like
field admits a smooth deformation to AdSd . 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.10.Kk; 11.15.-q
Keywords: Massless fields; Gauge invariance; AdS spacetime
1. Introduction
The aim of this paper is to demonstrate some peculiarities of generic free massless fields
in anti-de Sitter (AdS) space of an arbitrary spacetime dimension d. The main conclusion
will be that, in general, an irreducible AdSd massless field does not classify according to
irreducible representations of the flat space massless little algebra o(d 2), but reduces
to a certain set of irreducible flat space massless fields. The pattern of necessary flat-space
Corresponding author. E-mail: metsaev@td.lpi.ac.ru
1 On leave of absence from Lebedev Physical Institute, Moscow.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 0 2 - 8
184
L. Brink et al. / Nuclear Physics B 586 (2000) 183205
massless fields will be given. Another (rather unexpected) manifestation of this fact is
that not every massless field in flat space admits a deformation to AdSd with the same
number of degrees of freedom, since it is impossible to keep all the flat space gauge
symmetries unbroken in the AdSd space. This phenomenon does not take place, though,
for all those types of massless fields that appear in the usual low-energy massless sectors
of the superstring models and supergravities, because it holds only for the representations
of the spacetime symmetries described by non-rectangular Young diagrams. For the same
reason it cannot be observed in AdS4 higher spin gauge theories [13] (all massless fields in
AdS4 are described by one-row Young diagrams). The effect discussed in this paper takes
place for d > 6.
In superstring theory all types of representations appear at the higher massive levels. The study of higher spin gauge theory has two main motivations (see, e.g.,
[3,4]): firstly to overcome the well-known barrier of N 6 8 in d = 4 supergravity models and, secondly, to investigate if there is a most symmetric phase of superstring theory that leads to the usual string theory as a result of a certain spontaneous breakdown of higher spin gauge symmetries. These two motivations lead
in fact into the same direction because, as shown for the d = 4 case [5,6] higher
spin gauge theories require infinite collections of higher spin gauge fields with infinitely increasing spins. Another important feature discovered in [7] is that gauge
invariant higher spin interactions require the cosmological constant 2 of the background AdS space to be non-zero to compensate the extra length dimensions carried
by the higher derivative interactions required by the higher spin gauge symmetries.
(In this perspective plays the rle analogous to 0 in superstring theory.) The fact
that higher spin theories require an AdS background was regarded as rather surprising until it was realized that it plays a distinguished rle in the superstring theory as
well [810].
To investigate a possible relationship between the superstring theory and higher spin
theories one has to build the higher spin gauge theory in higher dimensions, d > 4 (e.g.,
d = 10, 11, . . .). A conjecture on the possible form of the higher spin symmetries and
equations of motion for higher spin spin gauge fields was made in [11,12] as a certain
generalization of the d = 4 results [13,14] which were proved to describe interactions of
all d = 4 massless fields.
A generalization like this to higher dimensions is not straightforward because of the use
of certain auxiliary twistor type variables. As a starting point, it is therefore important to
analyze more carefully the notion of a general massless field in AdSd . This is the main goal
of this paper.
Another motivation comes from the flat space analysis of certain massless (nonsupersymmetric) triplets in d = 11 in [15,16], the dimension of M-theory, where it was shown
that there exists an infinite collection of triplets of higher spin fields having equal numbers
of bosonic and fermionic degrees of freedom. These triplets show some remarkable properties. For the first four Dynkin indices the bosonic and the fermionic numbers match up.
This phenomenon is rather special for 11 dimensions and follows from the fact that the
little group SO(9) is an equal-rank subgroup of F4 , bringing in exceptional groups into the
185
picture. If these triplets have anything to do with higher spin gauge field theories and/or
M-theory it is an interesting question whether it is possible to extend the analysis of [16]
to the AdS case. That question in fact triggered this investigation.
2. Massless unitary representations in AdSd

AdSd is a d-dimensional spacetime with signature (d 1, 1) and the group of motions
SO(d 1, 2). It is most useful to identify AdSd with the universal covering space of the
appropriate hyperboloid embedded into Minkowski spacetime with signature (d 1, 2).
Physically meaningful relativistic fields in AdSd are classified according to the lowest
weight unitary representations of o(d 1, 2). Unitarity implies compatibility with quantum
mechanics, while lowest weight of a unitary representation guarantees that the energy is
bounded from below.
2.1. General facts
The commutation relations of o(d 1, 2) are
[Mm n , Mk l] = i(n k Mm l m k Mn l n lMm k + m lMn k ),
(1)
l = 0 d.
n,
k,
where n m = (, +, +, ) is the flat metric in the (d 1, 2) space, m,
The generators Mm n are Hermitian. Let us choose the following basis in the algebra:
a
=
t

1
M0 a iMd a ,
2
E = M0d ,
Lab = iM ab ,
where a, b = 1 d 1. The commutation relations (1) take the form

a
a
,
= t
E, t
(2)
(3)
(4)
1

a b
, t+ = E ab Lab ,
t
2
(5)
[Lab , tc ] = bc ta ac tb ,
(6)
[Lab , Lce ] = bc Lae ac Lbe be Lac + ae Lbc
(7)
with all other commutators vanishing. The hermiticity conditions are

a
a
t
= t
,
Lab = Lab .
E = E,
(8)
The generators E and Lab can be identified with the energy and angular momenta,
respectively, and span the Lie algebra o(2) o(d 1) of the maximal compact subgroup
a are combinations of AdS
of the AdS group SO(2, d 1). The non-compact generators t
translations and Lorentz boosts. The commutation relations are explicitly Z-graded with
186
a having grade 1. E is the grading operator. The lowest weight unitary representations
t
are now constructed in the standard fashion (for a review of the AdS4 case see, e.g., [17]
a
and [18]) starting with the vacuum space |E0 , si that is annihilated by t
a
t
|E0 , si = 0
(9)
and forms a unitary representation of the compact subalgebra o(2) o(d 1) that means
in particular that
E|E0 , si = E0 |E0 , si.
(10)
Here s denotes the type of representation of o(d 1) carried by the vacuum space: s =
(s1 , . . . , s ) with = [(d 1)/2]. s is a generalized spin characterizing the representation.
In terms of Young tableaux si is the number of cells in the ith row of the Young tableaux.
Since we are talking about representations of orthogonal algebras, the corresponding
tensors are traceless. We will here not discuss the self-dual representations that can be
singled out with the aid of the Levi-Civita symbol. 2 Note that s describes a finitedimensional representation of o(d 1). Field-theoretically this corresponds to a finitecomponent field carrying a finite spin.
The full representation of the Lie algebra o(d 1, 2), denoted in [19] D(E0 , s), is
spanned by the vectors of the form
a1
ak
t+
|E0 , si
t+
(11)
for all k. The states with fixed k are called level-k states. Note that states with pairwise
different k are orthogonal as a consequence of (4). As a result, the analysis of unitarity
can be reformulated as the check of the positivity of the norms of the finite-dimensional
subspaces at every level. For sufficiently large generic E0 it is intuitively clear that the
representation D(E0 , s) is irreducible and unitary, i.e., all norms are strictly positive. Such
representations are identified with massive representations of AdSd .
Let us emphasize that the elements of the module (11) can be identified with the modes
of a one-particle state in the corresponding free quantum field theory. The elements of the
AdSd algebra o(d 1, 2) are then realized as bilinears in the quantum fields. Note that at
the level of equations of motion the light-cone field-theoretical realization of generic AdSd
massive representations for arbitrary E0 and s has been developed recently in [20].
We see that massive states are classified by the parameter E0 (which is the analog of
mass) and a representation of o(d 1). This picture is in agreement with the standard
description of massive relativistic fields in flat spacetime in terms of the Wigner little
group SO(d 1).
If E0 gets sufficiently small, the norm cannot stay positive as is most obvious from the
following consequence of (5)
d 1
a
E,
=
ta , t+
2
(12)
2 The self-dual and antiself-dual representations which appear for odd d are usually distinguished by a sign of
the s(d1)/2 .
187
which implies that for negative E0 some level 1 states cannot have positive norm for a
positive-definite vacuum subspace. There is therefore a boundary of the unitarity region
E0 = E0 (s) > 0 such that some states acquire negative norm for E0 < E0 (s). Obviously,
these states should have zero norm for E0 = E0 (s).
Starting from the inside of the unitarity region decreasing E0 for some fixed s one
approaches the boundary of the unitarity region, E0 = E0 (s). Some zero-norm vectors
then appear for E0 = E0 (s). These necessarily should have vanishing scalar product to
any other state. (Otherwise, one can build a negative norm state that is in contradiction
with the assumption that we are at the boundary of the unitarity region.) Therefore, the
zero-norm states form an invariant subspace called a singular submodule. By factoring out
this subspace one is left with a unitary representation which is shorter than the generic
massive representation.
The resulting shortened unitary representations correspond either to massless fields
[21] or to singletons [22] and doubletons [23,24] identified with the conformal fields at
the boundary of the AdS space [22]. The fact that a singular submodule can be factored
out admits an interpretation as some sort of a gauge symmetry (true gauge symmetry for
the case of massless fields, or independence of bulk degrees of freedom for singletons and
doubletons). For this reason we choose this definition of masslessness for all fields in AdSd
except for the scalar and spinor massless matter fields which are not associated with any
gauge symmetry principle and singletons. 3
The analysis of positive definiteness of the scalar product of level-1 states was done
in [25] for an arbitrary even d and an arbitrary type of representation of s carried by the
vacuum state. The final result is
E0 > E0 (s),
E0 (s) = s1 + d t1 2,
(13)
(14)
where t1 is the number of rows of the maximal length s1 , i.e.,

s1 = = st1 1 = st1 > st1 +1 > st1 +2 > > s .
It can be shown that the same result is true for odd d (provided that s is replaced by |s |).
Note that this bound for d = 4 was originally found in [21]. For the case d = 5 see [26]
and references therein.
Massless representations are shorter than massive ones classified according to the
parameter E0 (equivalent of mass) and a representation of o(d 1). In flat space, massless
fields are classified according to the representation of the massless little group SO(d 2).
The question we address here is whether or not the shortening in AdSd can be interpreted
in terms of irreducible representations of SO(d 2). We will show that the answer is no
for a generic representation. This will be demonstrated both at the algebraic level using
3 Singleton-type fields live at the boundary of AdS and cannot be interpreted as bulk massless fields.
d
Presumably, all singletons except for scalar and spinor correspond to maximally antisymmetrized representations
of the AdSd algebra equivalent to their duals. In particular, this is true for the second rank antisymmetric tensor
representation in the AdS5 case that can be identified with the field strength of the YangMills fields of the N = 4
SYM at the boundary of AdS5 .
188
the language of singular vectors and at the field-theoretical level focusing on the simplest
nontrivial massless field with s = (2, 1, 0, . . . , 0). The most important field-theoretical
conclusion is that a generic irreducible massless field in AdSd decomposes into a collection
of massless fields in the flat limit. In that sense, a massless field in the AdSd is generically
less massless than elementary massless field in flat space. An important consequence
of this fact is that not every massless field in flat space can be deformed into AdS
geometry.
2.2. Singular vectors
It is useful to reformulate the problem in terms of singular vectors. Since the energy E
is bounded from below for the whole representation, the singular submodule spanned by
zero-norm states is itself a lowest weight representation. Therefore it contains at least one
nontrivial subspace |E00 , s0 i that has the properties analogous to (9) and (10),
a
|E00 , s0 i = 0
t
(15)
E|E00 , s0 i = E00 |E00 , s0 i,
(16)
and
i.e., it forms some irreducible representation s0 of o(d 1). Obviously,

E00 = E0 + k,
(17)
if |E00 , s0 i belongs to the level-k subspace.

Such spaces |E00 , s0 i we will call singular vacuum spaces while any of their elements
will be called a singular vector. 4 Clearly, singular vacuum spaces form representations
of the algebra o(2) o(d 1) and therefore decompose into a direct sum of irreducible
representations of o(d 1) on different levels. The standard situation is with a single
irreducible singular vacuum space. The singular module as a whole then has the structure
a1
ak
t+
|E00 , s0 i.
t+
(18)
The factorization to a unitary irreducible representation is equivalent to identifying all these

vectors to zero.
Let us now consider the example of a vacuum state |E0 , si with s = (s1 , s2 , 0, . . . , 0)
corresponding to a two-row Young diagram,
s1
s2
(19)
with 0 6 s2 6 s1 .
4 This terminology is very closely although not exactly coinciding with that used for the Verma module
construction when irreducible vacuum subspaces are one-dimensional because the grade zero subalgebra, namely
the Cartan subalgebra, is Abelian.
189
This means that |E0 , si can be realized as a tensor va1 ...as2 ,b1 ...bs1 (E0 ) which is symmetric
both in the indices a and in the indices b, satisfying the antisymmetry property
va2 ...as2 {bs1 +1 ,b1 ...bs1 } (E0 ) = 0,
(20)
which implies that symmetrization over any s1 + 1 indices a and/or b gives zero. The tensor
is traceless, which means that contraction of any two indices with the o(d 1) invariant
flat metric ab gives zero. Taking (20) into account it is enough to require
b1 b2 va1 ...as2 ,b1 ...bs1 (E0 ) = 0.
(21)
The level one states are

c
va1 ...as2 ,b1 ...bs1 (E0 ).
t+
(22)
These states form a reducible representation of o(d 1) (the tensor product of the vector
representation with the representation (19)). For generic d, s2 and s1 it contains five
irreducible components: two Young diagrams with one cell less (index c is contracted to
either one of the indices a or one of the indices b) and three diagrams with one cell more:
adding one cell to the first, second or an (additional) third row. Our problem therefore is
to check whether any of these irreducible representations can be a singular vacuum space,
i.e.,

e
c
t+
va1 ...as2 ,b1 ...bs1 (E0 ) 0
(23)
t
for some E0 ( is a projector to one or another irreducible component in (22)). Let us
consider the two representations with cells cut. The appropriate projections are given by
the following formulae describing irreducible o(d 1) tensors
va11 ...as
,b1 ...bs1 1

c
= t+
va1 ...as2 ,b1 ...bs1 1 c (E0 ) +
2
s2
vca ...a ,b ...b a (E0 )
s1 s2 + 1 1 s2 1 1 s1 1 s2

(24)
(symmetrizations within each of the groups of indices a and b are assumed) and
va21 ...as
2 1
,b1 ...bs1
c
= t+
vca1 ...as2 1 ,b1 ...bs1 (E0 ).
(25)
Elementary computations give that

tc va11 ...as ,b1 ...bs 1
2
1

1
= E0 (d + s1 3) va1 ...as2 ,b1 ...bs1 1 c (E0 )
2

s2
va ...a c,b ...b a (E0 )
+
s1 s2 + 1 1 s2 1 1 s1 1 s2
(26)
and

1
E0 (d + s2 4) va1 ...as2 1 c,b1 ...bs1 (E0 ).
2
As a result, singular vectors appear at
tc va21 ...as
2 1
,b1 ...bs1
E01 = d + s1 3
and
(27)
(28)
190
E02 = d + s2 4.
(29)
A few comments are now in order.

The cases s1 = s2 and s1 > s2 are different because v 1 0 for s1 = s2 as a consequence
of the antisymmetry property (20). For s1 > s2 both v 1 and v 2 are nontrivial.
As expected, the values of the singular energies (28) and (29) are in agreement with
the general analysis of [25], where it was also shown that only the representation resulting
from the factorization of the singular submodule with the highest energy of a singular
vector is unitary (this fact is natural from the singular vector description: unitarity can be
preserved only when the boundary of the unitarity region is approached; this implies the
highest E0 ). We therefore conclude that unitary massless particles appear for
E0 = d + s1 3,
for s1 > s2
(30)
E0 = d + s1 4,
for s1 = s2 .
(31)
and
The analysis in terms of singular vectors is simple enough but can be simplified further
with the aid of the technique proposed in [27] with the tensors corresponding to various
Young diagrams realized as certain subspaces of an appropriate Fock space. This technique
is explained in Section 4.1 since we will use it in the field-theoretical part of the paper.
We here use the tensor language to make most clear the interpretation in terms of the
representations of the massless little algebra o(d 2). One can analogously investigate
singular spaces with the boxes added to make sure that they have negative energies and
therefore do not play a rle in our analysis.
We expect that the analysis of higher levels does not affect our conclusions. One reason
is that the appearance of singular vectors at higher levels within the unitarity region would
imply existence of higher spin gauge fields with gauge transformations having more than
one derivative acting on a gauge parameter. The general analysis of massless fields in flat
spacetime of an arbitrary dimension [28] shows that this does not take place.
3. Flat space pattern of AdS massless fields

Let Aa1 ...as1 ,b1 ...bs2 ,... be an irreducible tensor of o(d 1) (a, b, . . . = 1 d 1) of a
specific symmetry type. Let na be a nonzero vector of o(d 1). o(d 2) can then be
identified with the stability subalgebra of o(d 1) that leaves na invariant. If the tensor
Aa1 ...as1 ,b1 ...bs2 ,... is orthogonal to na with respect to all possible contractions of indices
na1 Aa1 ...as1 ,b1 ...bs2 ,... = 0,
nb1 Aa1 ...as1 ,b1 ...bs2 ,... = 0 . . .
(32)
then it describes a representation of o(d 2) of the same symmetry pattern. If some

contractions with na are nonzero, one can decompose the o(d 1) tensor Aa1 ...as1 ,b1 ...bs2 ,...
into irreducible representations of o(d 2) with the aid of the projection operators
constructed from na or, in other words, performing dimensional reduction. For example,
for a vector,
191
na nb b
na nb b
ka
A
,
A
=
A .
(33)
nc nc
nc nc
The analysis of singular vectors in Section 2.2 admits a similar interpretation. Indeed,
a as a vector na analogous to the momentum operator in flat-space field
let us interpret t+
a
commute with themselves). The fact that a singular
theory (note that the operators t+
a with the vacuum vector decouple
vector appears means that some contractions of t+
a
from the spectrum and therefore are equivalent to zero. If all possible contractions of t+
would decouple this would mean that the o(d 1) vacuum representation would reduce
to the o(d 2) of the same symmetry. Since the energies (28) and (29) are different this
cannot be true simultaneously. Therefore, when a singular vector is present, the reduced
representation is effectively smaller than an irreducible representation of o(d 1) but
may be larger than the corresponding irreducible representation of o(d 2), containing a
number of irreducible representations of o(d 2).
Let us note that the energy in AdSd is measured in units of the inverse AdS radius
that was set equal to unity in our analysis. Reintroducing and taking the flat limit
0, all energies of singular vectors tend to zero. This means that in the flat limit
different singular vectors may decouple simultaneously and therefore a natural possibility
a orthogonal) representation of
consists of the flat space reduction to one (totally t+
o(d 2), in agreement with the standard analysis [29] of massless representations of
the Poincar algebra. However such massless representations of the Poincar algebra
may not admit a deformation to a representation of the AdS algebra with 6= 0. At the
field-theoretical level this means that it will not be possible to preserve all necessary
gauge symmetries for 6= 0. This phenomenon is demonstrated in the field theoretical
example in Section 4.2. The deformation will be possible however, if one starts with
an appropriate collection of massless fields in flat space dictated by the incomplete
dimensional reduction via decoupling of singular vectors. The main aim of this section
is to formulate a conjecture on the pattern of massless fields in flat space compatible with
the deformation to AdSd .
Let us consider an arbitrary Young diagram with row lengths s1 > s2 > s3 > . For
our analysis it is more convenient to build Young diagrams not from rows as elementary
entities but from rectangular blocks of an arbitrary height t and length s
Aa = Aka + Aa ,
Aa = Aa
(34)
In other words, a block is a Young diagram composed of t rows of equal length s
(equivalently, from s columns of equal height t). A general Young diagram is a combination of blocks with decreasing lengths (equivalently, heights)
192
s1
t1
s2
t2
q q q q q q
q q q q q
q q q q
sp1
tp1
sp
tp
(35)
In these terms, a Young diagram Y [(si , ti )] is described by a set of pairs of positive
integers (si , ti ) with s1 > s2 > s3 > > sp > 0 and arbitrary ti such that
p
X
i=1
1
ti 6 (d 1).
2
(36)
In other words, s1 is the maximal row length in the Young diagram, while t1 is the number
of rows of the length s1 . s2 is the maximal row length of the remaining rows and t2 is the
number of rows of length s2 . Note that an elementary block Y [(s, t)] is described in these
terms by a single pair of integers (s, t).
Let us now address the question what is the result of a dimensional reduction to one
dimension less of a general diagram Y [(si , ti )]. Every cell can be identified with some
vector index. It can either be aligned along na or along the d 2 perpendicular directions.
In the first case we cancel a cell, while in the second case we keep it. There cannot be more
than s1 indices along na because symmetrization with respect to more than s indices gives
identically zero by the definition of a Young diagram. But any number of indices from 0
to s1 can be chosen to take the extra value d 1. Therefore any number of cells from 0
193
to s1 can be canceled. Of course only such cancelings are allowed that result in a Young
diagram. (If not, any resulting tensor is identically zero.)
Let us consider some examples.
First consider a representation (e.g., tensor T ) described by the Young diagram
Y [(s, t)] which is itself an elementary block. Since components of tensors along na
are automatically symmetrized because the tensor N a1 a2 ...ak = na1 na2 . . . nak is totally
symmetric, we can use the symmetry properties of the Young diagrams to reduce any
contraction with N a1 a2 ...ak of the original tensor to a contraction of N a1 a2 ...ak with the
bottom row of the block. As a result, dimensional reduction will lead to a number of tensors
resulting from cutting an arbitrary number of cells in the bottom row of the block; every
tensor appears once (note that the condition that the tensor is traceless does not affect this
analysis since the reduced o(d 2) tensors are also assumed to be traceless).
s1
t1 1
n1
(37)
In other words, the dimensional reduction of the block Y [(s, t)] gives rise to the following
representations of o(d 2): Y [(s, t)] (all indices are orthogonal to na ), Y [(s, t 1)] (a maximal possible number s of indices is contracted) and all diagrams Y [(s, t 1); (s1 , 1)]
which consist of two blocks with the bottom block having an arbitrary length 0 < s1 < s
and height 1.
Now, consider a representation T described by a Young diagram Y [(s1 , t1 ); (s2 , t2 )]
composed from two blocks.
s1
t1
s2
t2
(38)
194
Again, dimensional reduction means that one can cut some cells from the bottom rows
of the upper and lower blocks (all cuts inside a block are equivalent by the properties
of the Young diagrams to cutting its bottom line). But now, one cannot cut an arbitrary
number of boxes in the top block because the cut line cannot be shorter than the length
of the second block s2 . (Such tensors vanish identically.) One can, however, take away an
arbitrary number of cells from the bottom line of the second box. The rule therefore is: take
away an arbitrary number n1 such that s1 s2 > n1 > 0 from the bottom line of the top
block and take away an arbitrary number n2 such that s2 > n2 > 0 of cells from the bottom
line of the bottom block
s1
t1 1
n1
s2
t2 1
n2
(39)
Note that with this prescription we have n1 + n2 6 s1 in accordance with the general
argument that one cannot cut a number of cells exceeding the maximal row length in the
Young diagram. The pattern of the dimensionally reduced representation therefore consists
of four-block diagrams Y [(s1 , t1 1); (s10 , 1); (s2 , t2 1); (s20 , 1)] with arbitrary integers s10
and s20 such that
s1 > s10 > s2 > s20 > 0
(40)
and their degenerate versions described by the three-block diagrams

Y (s1 , t1 ); (s2 , t2 1); (s20 , 1) ,
Y (s1 , t1 1); (s2, t2 ); (s20 , 1) ,

Y (s1 , t1 1); (s10 , 1); (s2 , t2 1)
Y (s1 , t1 1); (s10 , 1); (s2 , t2 ) ,
and two-block diagrams

Y (s1 , t1 1); (s2 , t2 ) ,
Y (s1 , t1 ); (s2 , t2 ) ,

Y (s1 , t1 1); (s2 , t2 + 1) .
Y (s1 , t1 ); (s2 , t2 1) ,
Analogously one proceeds for Young diagrams built from a larger number of blocks. The
final result is that the dimensional reduction of a general Young diagram to one dimension
less consists of the Young diagrams of the form (every diagram appears once; see, e.g.,
[29] and references therein):
195
s1
t1 1
n1
s2
t2 1
n2
q q q q q q
q q q q q
q q q q
sp1
tp1 1
sp
np1
tp 1
np
(41)
One is allowed to take away any numbers ni of cells from the ith block provided that
0 6 ni 6 si si+1
(42)
(with the convention that sj corresponding to the next to last block equals zero).
Let us now formulate the final result concerning a pattern of flat space massless fields
that admit a unitary deformation to AdSd .
Conjecture. Consider a AdSd massless field characterized by a diagram (35). Take away
any number of cells ni satisfying the conditions (42) of the bottom lines of all blocks except
for the top one, i.e., require n1 = 0. Any diagram that appears as a result describes some
irreducible representation of the massless little algebra o(d 2) corresponding to some
flat space massless field that should be present in the full set compatible with the AdSd
geometry and unitarity.
Note that massless fields corresponding to arbitrary irreducible representations of flat
space massless little algebra o(d 2) were considered in [28].
A few comments are now in order.
196
The role of the upper block is singled out by the unitarity condition: only singular
a to the upper block have maximal energies and
vectors corresponding to contractions of t+
describe unitary representations [25]. Therefore canceling out a box from the upper block
corresponds to pure gauge (i.e., singular vector) components that decouple. Non-unitary
(i.e., ghost containing) sets of fields can be obtained by a similar procedure with one of the
lower blocks remaining untouched instead of the top one as in the unitary case.
For Young diagrams being themselves elementary blocks (i.e., ti = 0 for i > 0) only one
o(d 2) representation appears, described by the same block. This means that elementary
block massless fields in AdSd classify according to irreducible representations of o(d 2)
as in the flat case (i.e., no additional massless fields should be added to deform to AdSd ).
In fact all examples of massless fields that appear in supergravity and low energy string
theory are described by elementary blocks (specifically, either by single rows, or by single
columns). That is why the phenomenon discussed in this paper was not observed before.
Note also that for the well studied case of lower dimensions d 6 4, only block-type
massless representations are nontrivial (propagating) and therefore this phenomenon does
not occur either.
The spectrum of flat-space massless fields to which an elementary AdS massless field
decomposes is non-degenerate, i.e., all the representations of o(d 2) are pairwise
different.
Some standard massless (gauge) fields may be needed as ingredients of AdS massless
fields with nontrivial diagrams. For example, for the representation Y [(2, 1), (1, 1)]
a graviton-type flat space massless field Y [(2, 1)] will be present (see example in
Section 4.2). It is tempting to speculate that this may correspond to a nontrivial deformation
of gravity to the AdS geometry in the presence of other fields. Analogously one can find
a totally symmetric spin s1 > 2 field corresponding to the diagram Y [(s1 , 1)] among the
fields resulting from the decomposition of the AdSd field Y [(s1 , 1); (s2 , 1)], 0 6 s2 6 s1 .
This is to say that AdSd massless field corresponding to Y [(s1 , 1); (s2 , 1)] decomposes into
the following irreps of o(d 2) algebra
s2

X

Y (s1 , 1); (s, 1) ,
Y (s1 , 1); (s2, 1) AdS Y (s1 , 1)
(43)
s=1
where each term under summation appears just once.

An important consequence of the analysis of this section is that totally antisymmetric
gauge tensors (i.e., differential forms) corresponding to the diagrams Y [(1, t)] can never
appear as a result of a decomposition of a certain irreducible (unitary) AdSd massless
field in the flat limit. The space of differential forms (including the spin one gauge fields)
therefore is closed with respect to the deformation to AdSd .
4. Field theoretical example

Now let us explain what happens from the field-theoretic perspective. In fact, it has been
observed already in [25] at the level of equations of motion in the Lorentz gauge that some
197
of the redundant gauge symmetries expected in the flat-space description are absent in the
AdS case. Here we analyze the problem at the Lagrangian level focusing on the explicit
comparison with the flat-space limit.
4.1. Flat space
Let us consider the simplest nontrivial example of a massless field having the symmetry
properties of a non-block diagram with three cells Y [(2, 1); (1, 1)]
(44)
In flat space the massless field of this symmetry type was described in [30]. To begin
with, let us reformulate the results of these authors in a somewhat different, although
equivalent way. We take the representation with the field m1 m2 ,n being a symmetric tensor
in m1 and m2 , satisfying the condition that full symmetrization with respect to all three
indices gives zero
{m1 m2 ,m3 } = 0.
(45)
(The authors of [30] used an equivalent representation with explicit antisymmetry in two
indices.) The Lagrangian can be chosen to be of the form
1
L = m1 m2 ,n 2 m1 m2 ,n mm1 ,n m1 m2 mm2 ,n
2
1
3
m1 m2 ,n1 n1 n2 m1 m2 ,n2 m1 m1 ,n 2m2 m2 ,n
2
4
3 m
3 m1 m2 ,n
m1 m2 ,n
+ m,n m1 m2
+
m1 m2 m m,n
4
4
3
+ m1 m1 ,n1 n1 n2 m2 m2 ,n2 .
4
The corresponding action is invariant under the gauge transformations
(46)

1 m1 m2 n
m1 n
as + m2 as
,
(47)
2

1
m2 n
m1 n
1 m2
+ m2 sym
(48)
n m
sym m1 m2 ,n = m1 sym
sym
2
mn
with antisymmetric gauge parameter mn
as (x) and symmetric gauge parameter sym (x),
as m1 m2 ,n =
nm
mn
as (x) = as (x),
nm
mn
sym (x) = sym (x).
(49)
In [30] it was proved that the Lagrangian (46) describes a physical massless field in flat
space corresponding to the irreducible representation Y [(2, 1); (1, 1)] of the massless little
algebra o(d 2). Because this Lagrangian is fixed by the gauge transformations one can
say that it is the gauge invariance with respect to the both gauge transformations (47) and
(48) that ensures irreducibility of the massless field.
198
Let us now introduce a notation that simplifies computation. In curved spacetime it is

convenient to use fiberwise fields
m1 m2 m3 em1 m1 em2 m2 em3 m3 m1 m2 m3 ,
(50)
where em m is the vielbein of an appropriate spacetime (e.g., Minkowski or AdS). 5

It is most convenient to formulate the action in terms of the following Fock-type
generating function [27]
1
|i = m1 m2 ,n 1m1 1m2 2n |0i,
2
(51)
m
and Bn are auxiliary creation and annihilation operators
where A
m n
m n
m n
A , B = 0,
A , B = 0,
A , B = mn AB ,
(52)
and |0i is a Fock vacuum

m
|0i = 0.
A
(53)
The indices A, B, C, E = 1, 2 label two sets of oscillators.

The fact that we deal with the Young diagram (44) is equivalent to imposing the
following constraints on the generating function |i
N11 |i = 2|i,
(54)
N22 |i = |i,
(55)
N12 |i = 0,
(56)
where we use the notation

m
Bm ,
NAB A
m
PAB A
Bm ,
m
PSAB A
Bm .
1m
(57)
2m
and
occur twice and once,
The constraints (54) and (55) tell us that the oscillators
respectively, on the right-hand side of Eq. (51). The constraint (56) is equivalent to the
condition (45).
The Lorentz covariant derivative for the representation |i takes the form
X

1
m n
n m
M mn =
A A
A ,
A
(58)
Dm m + mmn M mn ,
2
A=1,2
mn
where m is the Lorentz connection of space, while M mn forms a representation of the

Lorentz algebra so(d 1, 1). In the sequel we will often use the notation
DA A m Dm ,
SA A m Dm ,
D
Dm em m Dm .
(59)
The flat-space Lagrangian (46) now takes the form

1
S1 D2 D
S2 3 P11 2PS11
L = h|2 D1 D
4
2

2
2
3
S1 + D1 PS11 + P11 D2 D
S2 PS11 |i,
+ 4 P11 D
5 Tangent space (fiber) indices m, n and target space (base) indices m, n take the values 0, 1, . . . , d 1.
(60)
199
SB and 2 = D m Dm are defined via (58) and (59) with the flat space vielbein
where DA , D
and Lorentz connection
n
,
em n = m
m kn = 0.
The Lagrangian (60) is invariant under two gauge symmetries generated by the antisymmn
metric gauge parameter mn
as and the symmetric gauge parameter sym which can be conveniently described as Fock vectors
1
|as i = as mn 1m 2n |0i,
2
1
|sym i = sym mn 1m 1n |0i.
2
Now the gauge transformations (47), (48) take the form

as |i = D1 |as i,
sym |i =
1
2 D1 N21

D2 |sym i.
(61)
(62)
4.2. Gauge symmetries in AdSd

Let us now analyze the situation in the AdS case. The Lorentz covariant derivatives (58)
are no longer commuting but satisfy the commutation relationships
[Dm , Dn ] = 2 Mmn ,
Dm nA Dn mA = Dm nA Dn mA = 0
(63)
(64)
with the convention

nA = en n nA ,
nA = en n nA .
(65)
The condition (64) is just the standard zero torsion condition

Dm en a Dn em a = 0
(66)
while (63) is the equation of the AdS space. The covariant DAlembertian is
2
+ m mn Dn ,
D2 Dm
(67)
where the second term accounts for Dm being rotated as a tangent vector. With these
SB satisfy a number of useful relationships
conventions the covariant derivatives DA and D
summarized in the appendix.
As in the flat case we will analyze gauge symmetries with totally symmetric and totally
mn
antisymmetric gauge parameters mn
sym and as . It is sometimes convenient to combine
them into a gauge parameter m,n having no definite symmetry properties
|i = m,n 1m 2n |0i.
The symmetric and antisymmetric parts can be singled out as
|sym i |Si = N12 |i,
|as i 1
1
2 N12 N21
(68)
|i.
(69)
200
The gauge transformation |i which respects the constraints (54)(56) is

|i = D1 |i D2 |sym i.
(70)
It can equivalently be written as a combination of the gauge transformations with

symmetric and antisymmetric gauge parameters
|i = as |i + sym |i
(71)
with the gauge transformations of the form (61) and (62) but now with the derivatives D1
and D2 as in AdSd .
Next we analyze whether there exists a Lagrangian that generalizes (46) (equivalently,
(60)) to AdSd . The most general deformation of (60) to the AdS case without higher
derivatives is of the form

1
S1 D2 D
S2 + 3 P11 D2 + g2 PS11
L = h|D2 f 2 D1 D
4
2

S12 + D12 PS11 + P11 D2 D
S2 PS11 |i,
+ 34 P11 D
(72)
where f and g are arbitrary parameters. A straightforward but rather tedious computation
with the use of the identities collected in the appendix leads to the following result
Z

S1 |i
e h|(f + 3)D1 + 32 (d 5 g)P11 D
S = 2
AdSd
S1
+ h|(6 3d f )D2 + 3(d 3)P21 D

+ 3 (1 d g)P21 D1 PS11 |Si .
2
(73)
From this expression it is clear that the freedom in the parameters f and g is not enough to
warrant an action invariant under both types of symmetries. The best one can do is to find
a Lagrangian invariant either with respect to the gauge symmetry with the parameter |as i
or the one with |sym i. Note that at the level of equations of motion in the Lorentz gauge
an analogous phenomenon was observed for redundant gauge symmetries in [25]. Since,
according to the general analysis of unitary representations of AdSd in [25], the case with
gauge invariance with respect to |sym i does not lead to unitary dynamics, we focus on the
Lagrangian possessing the |as i invariance. From (73) it is obvious that this is achieved
by setting
f = 3,
g = d 5,
(74)
since the antisymmetric part of the gauge parameter enters only via the first term. Thus, we
set

1
S1 D2 D
S2 + 3 P11 D2 + 2 (d 5) PS11
L = h|D2 + 32 D1 D
4
2

3
S12 + D12 PS11 + P11 D2 D
S2 PS11 |i.
+ P11 D
(75)
4
Because one of the gauge symmetries is lost, the Lagrangian (75) describes more
degrees of freedom than the original flat-space Lagrangian we started with. This is in
201
agreement with the general conclusion of Section 3 that physical d.o.f. of massless AdS
fields may not be described by an irreducible representation of o(d 2). The conjecture
of Section 3 suggests that the flat space and the AdS dynamics can match only once one
starts with specific (reducible) collections of fields in flat space. From the general analysis
of Section 3 it follows that, in order to make the AdS deformation consistent for the case
under consideration, one has to add a massless spin two field analogous to a graviton field.
Let us therefore introduce the field mn symmetric in indices m, n, described by the
Fock vector
|i mn 1m 1n |0i.
Since this field should describe a massless spin 2 field in the flat limit, it has its own gauge
symmetry with the gauge parameter
| i m 1m |0i.
(76)
The idea is that starting from the sum of the free Lagrangians L + L one should add
cross terms L which (i) reestablish all (appropriately deformed by -dependent terms)
gauge symmetries with the parameters |i and | i and (ii) tend to zero in the flat limit. It
turns out that this is indeed possible. The final result is that the action
Z

e L + L + L ,
(77)
S=
AdSd
where
3
S1 + P11 D2 PS11 |i,
L = (d 3)h| 2D2 + 2P12 D
2

3
S1 1 P11 D2 + 2 PS11
L = (d 3)h|D2 + d2 D1 D
2
2

S12 + D12 PS11 |i
+ 12 P11 D
(78)
(79)
is invariant under the gauge transformations of the form

|i = D1 |i D2 |Si + (P12 P11 N21 )| i,
(80)
|i = D1 | i + |Si.
(81)
Note that the Lagrangian L is proportional to and tends to zero in the flat limit.
Therefore, as expected, the action reduces in the flat limit to the sum of two actions for the
irreducible fields. In the AdS case, however, the cross term L becomes nontrivial so that
the system does not decompose into a sum of elementary subsystems. Another comment is
that according to (81) the field |i becomes a Stueckelberg field that can be gauged away
for 6= 0. The resulting gauge fixed action is nothing but the action (75) invariant under the
gauge symmetry with antisymmetric gauge parameter. Therefore it describes properly the
irreducible AdS representation. The gauge fixing |i = 0 is impossible however for = 0.
This is why the naive flat limit of the action (75) describes not two fields but only one in
agreement with [30]. This phenomenon can be interpreted as some sort of nonanalyticity
of the flat limit exhibited already at the free field level.
202
We expect that one can analogously find a deformation of the flat space Lagrangian (60)
to AdSd by adding an antisymmetric second rank gauge tensor. This would correspond to
keeping the symmetry with symmetric parameters. However, because the corresponding
representation of the AdS algebra is not unitary, the resulting gauge invariant Lagrangian
is expected to have a wrong relative sign of the kinetic terms of the elementary flat-space
Lagrangian (i.e., incompatible with unitarity). A similar phenomenon is expected to be true
for more complicated Young diagrams: only the sets of fields predicted in Section 3 will
have all signs of kinetic terms of the flat-space Lagrangians correct.
The equations of motion for the fields |i and |i that follow from the Lagrangian (77)
can be reduced to the form
S1 D2 D
S2 + 1 D12 PS11 + D2 D1 PS12
D2 D1 D
2

2 P11 PS11 22 P12 PS12 + 32 |i
S1 + P11 N21 D
S1
+ (d 3)(D1 N21 2D2 ) P12 D

+ (P12 D1 P11 D2 )PS11 |i = 0,
(82)

S2 D1 PS12 )|i = 0.
S1 + 1 D12 PS11 1 P11 PS11 + d2 |i + (D
D2 D1 D
2
2
(83)
Si |i = 0, the tracelessness condition PSij |i = 0 and the

By imposing the Lorentz gauge D
condition |i = 0, we are left with

(84)
D2 + 32 |i = 0.
There is a leftover symmetry with the parameter |Si satisfying certain differential
conditions. Taking into account that one can identify the Stueckelberg field |i with |Si one
Si |i = 0,
can derive these conditions from the equations of motion for |i in the gauge D
S
Pij |i = 0

(85)
D2 + d2 |Si = 0.
Let us compare these results with the equations for the gauge field corresponding to the
AdSd massless representation D(E0 , s)
!
X
2
2
sA |i = 0,
(86)
D E0 (E0 + d) +
A=1,2
and the conditions on the leftover gauge parameter |Si

D2 2 (s2 2)(s2 3 + d) + 2
sA 2 |Si = 0
A=1,2
found in [25]. For the case under consideration the E0 and s are
E0 = (d 1),
s = (2, 1, 0, . . . , 0).
Plugging these values into (86), (87) we indeed arrive at the equations (84) and (85).
(87)
203
5. Conclusions
We have shown that generic irreducible massless (gauge) fields in AdSd in the flat limit
decompose into nontrivial sets of irreducible flat space massless fields. These sets are,
however, smaller than the result of a dimensional reduction to one less dimension of a
corresponding massive field. In that sense AdSd massless fields are less massless" than
flat space massless fields. We made a conjecture on the pattern of the flat-space reduction
of a generic AdSd massless field. From this conjecture it follows that there is a unique
nontrivial situation when a flat space spin two massless field appears as a result of a
nontrivial reduction of AdSd massless field with mixed symmetry properties. This example
has been considered in detail. It is tempting to speculate that there may exist some new
version of gravity associated with this type of field.
On the other hand we have argued that totally antisymmetric tensors can never result
from the flat limit decomposition of other types of AdSd unitary representations. In
other words, the space of differential forms is closed with respect to the flat space limit
decomposition.
An interesting problem for the future is to generalize these results to the supersymmetric
cases to analyze generic AdS supermultiplets in higher dimensions and, in particular,
in AdS11 . Another problem is to consider the multiplets occurring in [15] to see how they
group themselves in the case of AdS.
Acknowledgements
The work of R. Metsaev and M. Vasiliev is supported in part by INTAS, Grant No. 960538 and Grant No. 99-0590 and by the RFBR Grant No. 99-02-16207. R. Metsaev is also
supported by DOE/ER/01545-787. M. Vasiliev was also supported by NFR F-FU 08115347. The work of L. Brink is supported by NFR F 650-19981268/2000. R. Metsaev and
M. Vasiliev would like to thank for hospitality at Chalmers University. M. Vasiliev would
also like to express his gratitude to Prof. H. Nicolai for the hospitality at the MPI fr
Gravitationsphysik, Albert Einstein Institute where some part of this work was done. The
authors are grateful to O. Gelfond for the help in LATEX drawing of Young diagrams.
Appendix A. Algebra of commutators

In this appendix we collect some formulas that are used in the computations of the
Section 4.2.
!
X

2
2
S
NCC + 2 (1 d)NBA
DA , DB = AB D +
+
X
C

PBC PSAC NCA NBC ,
(88)
204
[DA , DB ] = 2
SB =
SA , D
D

X
(PBC NAC PAC NBC ),
X

l NCB PSAC NCA PSBC ,
D , DA = (1 d)DA + 22
2
SA = (d 1)D
SA + 2
D ,D
2

SC DC NAC ,
PAC D

SC DC PSAC ,
NCA D
X
C
D2
in (88) is a covariant DAlembertian operator (67). These formulas can be

where
derived by straightforward but sometimes lengthy calculation. The derivation of the
following relationships:
[NAB , NCE ] = BC NAE AD NCE ,

SC ,
SA , NBC = AB D
[NAB , DC ] = BC DA ,
D

SA , PBC = AB DC + AC DB ,
D

SA + AC D
SB ,
PSAB , DC = BC D

PSAB , PCE = BC NEA + BE NCA + AC NEB
+ AE NCB + d(BC AE + BE AC )
is elementary.
References
[1] C. Fronsdal, Phys. Rev. D 18 (1978) 3624; Phys. Rev. D 20 (1979) 848.
[2] J. Fang, C. Fronsdal, Phys. Rev. D 18 (1978) 3630; Phys. Rev. D 22 (1980) 1361.
[3] M.A. Vasiliev, Fortschr. Phys. 35 (1987) 741; M. Shifman (Ed.), Higher spin gauge theories:
star-product and AdS space, Contributed article to Golfands Memorial Volume, World
Scientific, hep-th/9910096.
[4] R.R. Metsaev, IIB supergravity and various aspects of light-cone formalism in AdS spacetime,
Based on talk given at International Workshop Supersymmetries and Quantum Symmetries,
Dubna, Russia, July 2731, 1999, hep-th/0002008.
[5] E.S. Fradkin, M.A. Vasiliev, Ann. Phys. 177 (1987) 63.
[6] M.A. Vasiliev, Ann. Phys. 190 (1989) 59.
[7] E.S. Fradkin, M.A. Vasiliev, Phys. Lett. B 189 (1987) 89; Nucl. Phys. B 291 (1987) 141.
[8] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[9] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[10] E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[11] M.A. Vasiliev, Nucl. Phys. B 301 (1988) 26.
[12] M.A. Vasiliev, Phys. Lett. B 257 (1991) 111.
[13] M.A. Vasiliev, Fortschr. Phys. 36 (1988) 33.
[14] M.A. Vasiliev, Phys. Lett. B 243 (1990) 378; Class. Quant. Grav. 8 (1991) 1387; Phys. Lett.
B 285 (1992) 225.
[15] T. Pengpan, P. Ramond, Phys. Rep. 315 (1999) 137, hep-th/9808190.
[16] L. Brink, P. Ramond, in: M. Shifman (Ed.), Dirac equations, light cone supersymmetry, and
superconformal algebras, Contributed article to Golfands Memorial Volume, World Scientific,
hep-th/9908208.
205
[17] H. Nicolai, Representations of supersymmetry in anti-de Sitter space, in: B. de Wit, P. Fayet,
P. van Nieuwenhuizen (Eds.), Supersymmetry and supergravity, 1984, World Scientific,
Singapore, 1984.
[18] B. de Wit, I. Herger, Anti-de Sitter Supersymmetry, Lectures given at 35th Winter School of
Theoretical Physics: from Cosmology to Quantum Gravity, Polanica, Poland, 212 February
1999, hep-th/9908005.
[19] C. Fronsdal, Phys. Rep. D 12 (1975) 3819.
[20] R.R. Metsaev, Nucl. Phys. B 563 (1999) 295, hep-th/9906217.
[21] N.T. Evans, J. Math. Phys. 8 (1967) 170.
[22] P.A.M. Dirac, J. Math. Phys. 4 (1963) 901.
[23] M. Gunaydin, Singleton and doubleton supermultiplets of spacetime supergroups and infinite
spin superalgebras, Invited talk given at Trieste Conf. on Supermembranes and Physics in (2+1)
Dimensions, Trieste, Italy, July 1721, 1989.
[24] M. Gunaydin, D. Minic, Nucl. Phys. B 523 (1998) 145, hep-th/9802047.
[25] R.R. Metsaev, Phys. Lett. B 354 (1995) 78; Arbitrary spin massless bosonic fields in
d-dimensional anti-de Sitter space, Talk given at Dubna International Seminar Supersymmetries and Quantum Symmetries dedicated to the memory of Victor I. Ogievetsky, Dubna, 2226
July, 1997, hep-th/9810231.
[26] G. Mack, Comm. Math. Phys. 55 (1977) 1.
[27] V.E. Lopatin, M.A. Vasiliev, Mod. Phys. Lett. A 3 (1988) 257.
[28] J.M.F. Labastida, Nucl. Phys. B 322 (1989) 185.
[29] A.O. Barut, R. Raczka, Theory of Group Representations and Applications, PWN, Warszawa,
1977.
[30] C.S. Aulakh, I.G. Koh, S. Ouvry, Phys. Lett. B 173 (1989) 284.

Gauge and gravitational interactions

of non-BPS D-particles
L. Gallot a , A. Lerda a,b, , P. Strigazzi a
a Dipartimento di Fisica Teorica, Universit di Torino and INFN, Sezione di Torino, Via P. Giuria 1,
I-10125 Torino, Italy

b Dipartimento di Scienze e Tecnologie Avanzate, Universit del Piemonte Orientale,
I-15100 Alessandria, Italy

Abstract
We study the gauge and gravitational interactions of the stable non-BPS D-particles of the type I
string theory. The gravitational interactions are obtained using the boundary state formalism while
the SO(32) gauge interactions are determined by evaluating disk diagrams with suitable insertions of
boundary changing (or twist) operators. In particular the gauge coupling of a D-particle is obtained
from a disk with two boundary components produced by the insertion of two twist operators. We
also compare our results with the amplitudes among the non-BPS states of the heterotic string
which are dual to the D-particles. After taking into account the known duality and renormalization
effects, we find perfect agreement, thus confirming at a non-BPS level the expectations based on the
heterotic/type I duality. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Hf
Keywords: String theory; D-branes; Boundary states
1. Introduction
The spectrum of supersymmetric string theories usually contains a special class of
states known as BPS states, which are characterized by the property that their mass is
completely determined by their charge under some gauge field. They form short (or ultra
short) supersymmetric multiplets and, because of this fact, are stable and protected from
quantum radiative corrections. A well-known example of such BPS states is provided by
Work partially supported by the European Commission TMR programme ERBFMRX-CT96-0045, by

MURST and by the Programme Emergence de la rgion Rhne-Alpes (France).
Corresponding author.
E-mail addresses: gallot@to.infn.it (L. Gallot), lerda@to.infn.it (A. Lerda), strigazzi@to.infn.it (P. Strigazzi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 0 6 - 5
L. Gallot et al. / Nuclear Physics B 586 (2000) 206230
207
the supersymmetric Dp-branes of the type II theories (with p even in type IIA and p
odd in type IIB) [1]. However, supersymmetric string theories quite often contain states
that are stable without being BPS. These are in general the lightest states which carry
some conserved quantum numbers. For them there is no particular relation between their
mass and their charge; they form long multiplets of the supersymmetry algebra and receive
quantum radiative corrections. However, being the lightest states with a given set of
conserved quantum numbers, they are stable since they cannot decay into anything else.
Usually, it is not difficult to find such non-BPS states with the standard string perturbative
methods and analyze their properties at weak coupling; but, since they cannot decay,
they should be present also in the strong coupling regime, or equivalently they should
appear as non-perturbative (D-brane type) configurations in the weakly coupled dual
theory. To verify the existence of these non-BPS states is therefore a very strong test on
the duality relations between two string theories which does not rely on supersymmetry
arguments. The study of the stable non-BPS D-branes in string theory, pioneered by Sen
in a remarkable series of papers [26], has attracted a lot of interest during the last year
(for reviews see Refs. [79]) also for several other reasons; among them we recall the fact
the non-BPS D-branes might be useful for analyzing the non-perturbative properties of the
non-supersymmetric field theories that live on their world-volumes, or the fact that they
may lead to novel types of relations among string theories [7].
One of the most notable examples of stable non-BPS configurations is provided by the
perturbative states at the first excited level of the SO(32) heterotic string [10] which carry
the spinor representation of the gauge group and whose mass is given by
2
gYM
.
Mh = =
0
10
(1.1)
In the last equality we have introduced the low-energy gauge and gravitational couplings
gYM and 10 of the heterotic string following the conventions of Ref. [11] which are also
reviewed in Appendix B. Being at the first massive level, these states are non-BPS, but
being the lightest ones carrying the spinor representation of SO(32), they are stable and
should be present also when one increases the heterotic string coupling constant gh . In
this process however, the mass Mh gets renormalized since there are no constraints on it
coming from supersymmetry. Thus we can write
Mh =
gYM
f,
10
(1.2)
where the renormalization function f can in principle be computed perturbatively in the

heterotic string and is such that f 1 for gh 0.
If the heterotic/type I duality [12] is correct, also in the type I theory there should exist
stable non-BPS configurations that are spinors of SO(32). Such states do indeed exist and
were identified by Sen as the non-BPS D-particles of type I [4,5]; then, an explicit boundary
state description for them was provided in Ref. [13]. 1 The mass of these D-particles turns
out to be
1 For the description of other non-BPS D-branes using the boundary state formalism see Refs. [1418].
208
Me0 =
1
0 gI
gYM 3/4 1/2

2
gI
,
10
(1.3)
where gI , gYM and 10 are, respectively, the string, the gauge and the gravitational coupling
constants of the type I theory in the conventions of Ref. [11] (see also Appendix B).
Comparing Eqs. (1.2) and (1.3), and remembering that under the duality map the heterotic
gauge and gravitational couplings turn into the corresponding ones of type I, we can deduce
that the renormalization function f must be such that f 23/4 gI 1/2 for gI 0 in
order for the masses to agree on both sides. Clearly, this result cannot be obtained using
perturbative methods, but is a prediction of the heterotic/type I duality. 2
In this paper we elaborate further on these stable non-BPS particles and study in
detail their gravitational and gauge interactions. On the heterotic side, these can be
easily obtained using standard perturbative techniques from correlation functions of vertex
operators. In this way one can show, for example, that, at the lowest order in the heterotic
string coupling constant, the gravitational and gauge potential energies of two such
particles at large distance are given, respectively, by the Newtons law and the Coulombs
law for massive and charged point-like objects in ten dimensions. On the type I side,
instead, the interactions of the non-BPS D-particles must be obtained using less standard
methods and have not been fully investigated so far; indeed, only the general rules for
computing string amplitudes with these D-particles have been given in the literature [5,20].
It is the purpose of this paper to fill this gap.
In particular, we will concentrate on processes involving massless string modes that are
responsible for the long range interactions among D-particles. To study the gravitational
interactions we adopt the boundary state formalism [2124] and obtain the energy due to
the exchange of closed string states between two D-particles by simply computing the
diffusion amplitude between the two corresponding boundary states in relative motion
[25,26] (for recent reviews on the boundary state formalism and its applications see
Ref. [27]). Then, by taking the large distance limit to which only graviton and dilaton
exchanges contribute, we find that the gravitational potential energy of two D-particles
exactly agrees with the one of their heterotic duals, provided that the duality relations and
the mass renormalization previously discussed are taken into account.
For the gauge interactions, instead, the situation is a bit more involved. In fact, we cannot
use any more the boundary state formalism since this accounts only for the couplings
of the D-particles with the closed strings that live in the bulk, but is completely blind
to the other bulk sector of the type I theory consisting of open strings with Neumann
boundary conditions in all directions to which the SO(32) gauge fields belong. On the
other hand, the open strings attached to the D-particles have Neumann boundary conditions
only along the time direction. Therefore, to study the gauge interactions of our D-particles
we should consider scattering amplitudes involving open strings with mixed boundary
conditions in an odd number of dimensions. Calculations of open string amplitudes with
mixed boundary conditions have already appeared in the analyses of systems of several
D-branes with different dimensionality (see for instance Ref. [28]), and require the use
2 The T-dual heterotic/type I correspondence has been analyzed at the non-BPS level in Ref. [19].
209
of twist operators to produce mixed boundary conditions in certain directions. These twist
operators were used in the past to study strings on orbifolds [29,30], and have been recently
reconsidered from an abstract conformal field theory point of view [31]. Using such twist
operators and applying the rules of Refs. [5,20], we will describe how to compute scattering
amplitudes involving non-BPS D-particles and bulk open strings of type I. Special care is
required in these calculations because the twist operators that we use change the boundary
conditions in an odd number of directions. In particular, we will explicitly determine the
gauge coupling of the D-particles by evaluating a correlation function on a disk with two
boundary components produced by the insertion of two twist operators. The result of this
calculation is extremely simple, namely the non-BPS D-particles couple minimally to the
gauge field. Exploiting this fact, we then determine the gauge potential energy of a pair
of D-particles at large distance and see that after taking into account the duality map, this
exactly agrees with the corresponding energy computed in the heterotic theory.
This paper is organized as follows: in Section 2 we compute the gauge coupling of a
non-BPS D-particle of type I by evaluating a disk diagram with two twist insertions, and
then determine the gauge potential energy between two D-particles. In Section 3 we use the
boundary state formalism to compute the gravitational contribution to the potential energy
of two (moving) D-particles. In Section 4 we study the gauge and gravitational interactions
of the non-BPS heterotic states that are dual to the D-particles. In Section 5 we compare
the results for the non-BPS D-particles and for their dual heterotic states, and discuss their
relations. In Appendix A we show how to compute the gauge interactions between two BPS
D-strings of type I by extending the method of Section 2 and verify the no-force condition.
Finally, Appendices B and C contain our conventions and a list of more technical formulas.
2. Type I D-particle interactions: the gauge amplitude

As we mentioned in the introduction, an important check of the heterotic/type I duality
has been the discovery by Sen [35] that the stable non-BPS heterotic states carrying
the spinor representation of SO(32) at the first massive level are dual to the non-BPS
D-particles of type I. Specific rules for computing amplitudes involving such D-particles
have been given by Sen [5] and Witten [20] in two different ways which we briefly recall
here. Sens approach heavily relies of the use of ChanPaton factors to distinguish the
various kinds of open strings. The 00 strings, whose end-points lie on the non-BPS
D-particle, contain both states that are even and states that are odd under (1)F ; the former
carry a ChanPaton factor 1, the latter a ChanPaton factor 1 . The 90 strings stretching
between one of the 32 D9-branes of the type I background and a D-particle contain only
(1)F even states but, due to the existence of an odd number of fermionic zero modes,
their vertex operators comprise the standard GSO-even
as well as the corresponding
part
1
GSO-odd part, weighted by ChanPaton factors 0 and 01 , respectively. Besides these
factors, the 90 strings also carry a ChanPaton factor A (A = 1, . . . , 32) labeling the
fundamental representation of the SO(32) gauge group. Finally, the 99 strings are the
210
usual open strings of the type I theory which are GSO projected and carry only the standard
ChanPaton factors of the gauge group.

The presence of the unusual ChanPaton factors 1, 1 , 10 or 01 shows that the states
of the 00 and 90 sectors have a non-trivial structure which is really due to the presence
of an odd number of fermionic zero modes. In order to remedy to this oddity, in Ref. [20]
Witten has proposed to introduce an extra one-dimensional fermion on each boundary of
the string world-sheet lying on a D-particle. In this way, in the 90 sector one recovers an
even number of fermionic zero modes and can perform the usual GSO projection. Also in
the 00 sector one performs a (generalized) GSO projection to obtain physical states, but
since the extra fermion is odd under this GSO parity, one obtains two types of 00 states,
similarly to what found by Sen.
Let us now give some details on how to construct the massless states in the various open
string sectors using Wittens rules. We start with the NS sector of the 00 strings where at
the massless level there are nine scalars x i (i = 1, . . . , 9) corresponding to the freedom of
moving the D-particle in its nine tranverse directions. These modes, which are present also
on the BPS D0 brane of the type IIA theory, correspond to vertex operators Vx i that do not
depend on the boundary fermion . In the (1) superghost picture, these vertex operators
are simply
= i e .
Vx(1)
i
(2.1)
Notice that there is no factor of eikX in (2.1) because massless states of 0p strings have no
momentum. Let us now consider the R sector of the 00 strings. Here both the ten worldsheet fermions and the boundary fermion possess zero modes so that the massless R
states form a GSO-even spinor of SO(1, 10). Note that in this case the GSO projection is
simply the ten-dimensional chirality projection which is natural when one extends SO(1, 9)
to SO(1, 10) by adding . Thus, in the (1/2) superghost picture the vertex operator for
the massless R states reads
1 + /2
S e
,
(2.2)
2
where S is the spin field of conformal dimension 10/16 associated to the ten world-sheet
fermions. Upon quantization, the 16 massless fermionic modes described by (2.2) account
for the 216/2 = 256 degeneracy of the non-BPS D-particle.
We now turn to the 90 strings which are more relevant for our purposes. Since the NS
sector does not contain massless states, we just consider the R sector. In this case, the only
world-sheet fermion to have a zero mode is 0 so that the ground state is a GSO-even
(chiral) spinor of the algebra SO(1, 1) generated by 00 and . Hence, the vertex operator
describing the massless modes of the 90 sector should contain
a spin field S associated to the fermion 0 , of conformal dimension 1/16;
a boundary changing operator for the nine space directions transverse to the
D-particle, of conformal dimension 9/16;
a GSO (or chirality) projector for the Clifford algebra of SO(1, 1) (1 + 00 )/2, of
conformal dimension zero;
(1/2)
211
a superghost contribution in the 1/2 picture e/2 , of conformal dimension 3/8;

a gauge ChanPaton factor A to specify which of the 32 D9-branes one is
considering.
Thus, we have
1 + 00
S e/2 .
(2.3)
2
It is easy to check that the operator (2.3) has indeed conformal dimension 1 as it should
be for a physical vertex operator. Notice that the GSO projection in (2.3) keeps only
one fermionic degree of freedom for each value of the index A of the fundamental
representation of SO(32). Upon quantization, the states described by V90 form a spinorial
representation of SO(32), and hence we can conclude that the marginal operator (2.3)
accounts for the SO(32) degeneracy of the non-BPS D-particle. Since in type I the strings
are unoriented, we should consider also the 09 sector. This is merely related to the 90
sector through the action of the world-sheet parity . Recalling that simply acts by
transposition on the ChanPaton factors without changing the physical content of the vertex
operators, we have
(1/2)
V90
= A
1 + 00
S e/2 .
(2.4)
2
Notice in particular that the SO(1, 1) GSO projection is the same in both vertices (2.3) and
(2.4).
Finally, there are the 99 strings which, as we mentioned above, are the usual open
strings of the type I theory; in particular in the NS sector at the massless level we find the
SO(32) gauge bosons which are described by the following vertex operators in the (1)
superghost picture
(1/2)
V09
(1/2)
= V90
= t
(1)
= AB A eikX e ,
Vgauge
(2.5)
where AB are the generators of SO(32) in the fundamental representation (see

Appendix B for our conventions) and A is the polarization vector.
We now face the problem of finding the coupling between the non-BPS D-particle and
the gauge field. Since the latter belongs to the 99 massless sector, the diagram we have
to compute corresponds to a disk with a part of its boundary on the D-particle and a
part on the D9-branes from which the gauge boson is emitted. This is represented in the
Fig. 1. We thus have to insert two vertices containing the boundary changing operator
which turns a boundary of type 0 into one of type 9 (or viceversa). The obvious choice
is then to make insertions of the vertices V90 and V09 given in (2.3) and (2.4) which
correspond to the SO(32) degeneracy of the D-particle. Note that, although no momentum
is carried by these vertices, the emitted gauge boson may have non-zero space momentum.
Indeed, the twist operators are reservoirs of transverse momentum [29,30] which allow
emissions with non-zero momentum in the transverse directions; on the other hand, this
is to be expected because the presence of a D-brane breaks the translational invariance in
transverse space. Note also that, due to the insertion of the two vertices V90 and V09 , the
boundary component associated with the D-particle carries indices in the bi-fundamental
212
Fig. 1. The disk diagram describing the gauge coupling of a type I D-particle.
representation of SO(32). This is consistent with the fact that a D-particle should emit all,
both massless and massive, perturbative open string states which group in the adjoint or in
the symmetric representation of SO(32).
The gauge coupling of a (static) D-particle is then given by the expectation value of the
gauge boson emission vertex (2.5) in the vacuum representing the D-particle. Thus, the
diagram of Fig. 1 corresponds to
(1)
|90 i,
F gauge = h09 |Vgauge
(2.6)
where
(1/2)
|90 i = lim V09

z0
(1/2)
(z) |0i and h09 | = lim h0| V09

z
(z).
(2.7)
Note that due to the presence of the twist operators in V09 and V90 , the expectation value
of the gauge emission vertex is not vanishing, as we will explicitly see in the following.
After including the normalization factor Cdisk appropriate of any disk amplitude, the
normalization factors NR for the R vertices (2.3) and (2.4), and NNS for the NS vertex
(2.5), 3 F gauge may be reexpressed as a 3-point function on the world-sheet and reads
Z
(1/2)

(1/2)
gauge
2
(1)
= Cdisk NR NNS d(zi ) c V09
(z1 ) c Vgauge
(z2 ) c V90
(z3 ) , (2.8)
F
where we have also added a ghost c in each vertex operator. The notation h i means that
the correlator must be evaluated by including the action for the boundary fermion as
explained in Ref. [20].
The correlation function in (2.8) may be decomposed into a longitudinal and a transverse
piece. The latter vanishes because

(2.9)
(z1 ) i (z2 ) eikX(z2) (z3 ) = 0
for i = 1, . . . , 9. Thus, there is no emission of gauge bosons with polarization Ai along
the transverse directions, as it should be for a minimally coupled particle at rest. We then
consider the longitudinal part for which the basic correlators are
3 We refer to Appendix B for the explicit expression of these normalization factors and to Refs. [11,32] for their
derivation.
hc(z1 ) c(z2 ) c(z3 )i = z12 z13 z23 ,

1/2 1/4 1/2
(z1 )/2 (z2 ) (z3 )/2
e
e
= z12 z13 z23 ,
e

0 k 2 0 k 2 9/8 0 k 2
z13
z23 .
(z1 ) eikX(z2) (z3 ) = z12
213
(2.10)
(2.11)
(2.12)
Notice that in (2.12) the transverse momentum k i of the emitted gauge boson is not subject
to any constraint, as we have anticipated. The remaining correlator to be considered is

1 + 00
1 + 00
S(z1 ) 0 (z2 )
S(z3 )
.
(2.13)
2
2
This splits into four pieces, two of which vanish. Indeed, according to Ref. [20] the only
non-vanishing correlation functions are those containing one factor of . In particular one
has
h1i = 0.
(2.14)
hi = 2,
Finally, we have

0
1/2 3/8 1/2
0 S(z1 ) 0 (z2 ) S(z3 ) = S(z1 ) 0 (z2 ) 00 S(z3 ) = z12 z13 z23 .
(2.15)
Notice that a correlation function similar to (2.15) appears in the 2D Ising model. Indeed,
the spin field S may be identified with the order parameter (i.e., the magnetization) while
the other spin field 00 S plays the role of the disorder parameter [33].
Inserting Eqs. (2.10)(2.15) into (2.8) and exploiting the projective invariance to fix the
position of the three punctures at arbitrary values, we easily get

0 2
Z

Cdisk NR2 NNS
z13 k
gauge
t A BC D
Tr A0 d(zi )
=
F
z12 z23
2

Cdisk NR2 NNS
A
Tr t BC D A0 .
=
(2.16)
2
Then, using the explicit expressions of the normalization coefficients and ChanPaton
factors given in Appendix B, we can rewrite F gauge as follows

gYM
(2.17)
F gauge = i AB CD AC BD A0 ,
2
where gYM is the gauge coupling constant of the type I theory.
Eq. (2.17) represents the amplitude for the emission of a gauge boson with longitudinal
polarization 0 and color index (BC) from a 0-boundary in the bi-fundamental of SO(32).
The appearance of this representation is a direct consequence of our construction in which
the D-particle is represented by the 0-component of a disk boundary produced by the
insertion of the vertex operators V90 and V09 . On the other hand, the SO(32) spinor
degeneracy of the non-BPS D-particle of type I arises from the (second) quantization of
the 32 massless fermionic zero-modes of the 09 open strings, and thus it is clear that such
a degeneracy cannot be seen in our operator formalism. This fact should not be surprising
because a completely analogous situation occurs in the familiar description of Dp-branes
using boundary states. Indeed, a boundary state is a single state that correctly represents
214
a Dp-brane and its couplings to the bulk closed strings, even if it does not account for
the degeneracy of the Dp-brane under the supersymmetry algebra. Similarly, in our case
the 0-component of the disk boundary produced by the insertion of V90 and V09 correctly
describes a D-particle and allows to obtain its coupling with the bulk 99 open strings,
even if does not account for its degeneracy under the gauge group. In fact, as we will
see later and in the following sections, using this construction we are able to obtain nontrivial information about the gauge interactions between two D-particles at large distance.
Moreover, after taking into account the known duality relations, we will show that the
results obtained in this way exactly agree with those in the heterotic theory, as required by
the heterotic/type I duality, thus confirming the validity of our construction.
gauge
Using the result (2.17) we can now easily compute the gauge potential energy VI
due to the exchange of the SO(32) gauge bosons between two D-particles. As indicated
in Fig. 2, this can be obtained simply by sewing two emission amplitudes F gauge with the
gauge boson propagator
P=
BB 0 CC 0
0
0
BC CB ,

2
q
(2.18)
yielding
gauge
VI
2
1
gYM
AE F D AF DE 2 .
2
q
(2.19)
Performing a Fourier transform, we get the following (static) gauge potential in configuration space
gauge
VI
(r) =
2
1
gYM
,
AE F D AF DE
2
78 r 7
(2.20)
where q = 2 (q+1)/2/ ((q + 1)/2) is the area of a unit q-dimensional sphere. Eq. (2.20)
clearly represents a Coulomb-like potential energy for point particles at a distance r in
ten dimensions.
We conclude this section by mentioning that the same results (2.17) and (2.20) can be
obtained also using the rules given by Sen in Ref. [5] for computing amplitudes with nonBPS D-particles.
Fig. 2. The diagram describing the gauge amplitude between two D-particles.
215
3. Type I D-particle interactions: the gravitational amplitude

The gravitational contribution to the scattering of two non-BPS D-particles of type I
can be calculated, at the leading order in the string coupling constant, from the diffusion
amplitude between two corresponding boundary states. The boundary state description
of the non-BPS D-particles has already been given in Refs. [5,13] from which we recall
the results that are relevant in the forthcoming analysis. For details and conventions on
boundary states, we refer the reader to Refs. [13,23,24,26].
In the closed string operator formalism, one describes a Dp-brane by means of a
boundary state |Dpi [21,22]. This is a closed string state which inserts a boundary on the
world-sheet, enforces on it the appropriate boundary conditions and represents the source
for the closed strings emitted by the brane. As an example, the boundary state for a BPS
D-particle of type IIA may formally be written as 4
|D0iIIA = |D0iNS + |D0iR ,
(3.1)
where the NSNS and the RR components are both proportional to T0 which is the tension
of the D-particle in units of the gravitational coupling constant, namely
T0 = 8 7/2 0 3/2 .
(3.2)
The presence of both the NSNS and the RR components implies that the spectrum of the
open strings living on the D-particle is GSO-projected. The partition function of such open
strings may be obtained by evaluating the cylinder/annulus amplitude in the closed string
channel which is given by
IIA hD0|P |D0iIIA ,
(3.3)
and then performing a modular transformation. In Eq. (3.3), P denotes the closed string
propagator
P=
1
0
.
e0 a e
2 L0 + L
a
(3.4)
where a (e
a ) is the left (right) intercept (aNS = 1/2, aR = 0).
The boundary state for the non-BPS D-particle of the IIB theory [3,13] has
instead only
a component along the NSNS sector and a tension Te0 greater by a factor of 2 than T0 .
Thus, we can write
(3.5)
|De
0iIIB = 2 |D0iNS .
As a consequence, there is no GSO-projection in the spectrum of the open strings lying
on the non-BPS D-particle and the presence of a tachyon in the NS sector renders it
unstable. However, if we consider the type I theory [3437], the tachyon is removed by
the projection onto states invariant under the world-sheet parity [3,13]. In the boundary
state formalism, the projection is implemented by adding the so-called crosscap state
4 In order to avoid clutter, we shall denote the NSNS (respectively, RR) component of a boundary state with
the simplified subscript NS (respectively, R)
216
|Ci [22], which corresponds to inserting on the closed string world-sheet a boundary with
opposite points identified. The negative (32) charge for the non-propagating RR 10form that the crosscap generates in the background, must be canceled by the introduction
of 32 D9-branes. Hence, the type I theory possesses a background boundary state given
by [22]

1
(3.6)
|Ci + 32|D9i ,
2
where the factor of 1/ 2 has been introduced to obtain the right normalization of the
various spectra. Then, the partition function for unoriented open 99 strings, given by the
sum of the annulus and the Mbius strip contributions, is

1 10
(3.7)
2 hD9|P |D9i + 25 hD9|P |Ci + 25 hC|P |D9i ,
2
while the contribution of the Klein bottle
1
hC|P |Ci
(3.8)
2
added to the torus contribution gives the partition function for unoriented closed strings.
The boundary state of the non-BPS D-particle of type I reads

1
2 |D0iNS = |D0iNS ,
(3.9)
|De
0iI =
2
where we have added the same factor of 1/ 2 for consistency with (3.6). The mass of the
D-particle is then given by
1 Te0
T0
1
=
=
,
(3.10)
Me0 =
2 10
0 gI
10
where 10 is the ten-dimensional gravitational coupling constant of the type I theory (see
Appendix B). The partition function for open 00 strings living on the D-particle, obtained
by summing the contributions from the annulus and the Mbius strip, is

1
2 hC|P |D0iNS + 2 NS hD0|P |Ci .
(3.11)
NS hD0|P |D0iNS +
2
In this theory, there are also 09 and 90 open strings with one end on the D-particle and
the other on one of the 32 D9-branes of the type I background. The world-sheet parity
exchanges the two sectors 09 and 90 so that we only retain symmetric combinations
corresponding to the partition function

32 2
(3.12)
NS hD0|P |D9i + hD9|P |D0iNS .
2
The spectrum of open strings stretching between two different (distant) D-particles at rest,
one labeled with a prime, has a partition function given by

1 2
( 2 ) NS hD0|P |D00 iNS + ( 2 )2 NS hD00 |P |D0iNS ,
(3.13)
2
where the factor of one-half indicates that, compared to the IIB case, only the symmetric
combinations are retained. Notice that, at sufficiently small distance, a tachyon develops in
217
this open string spectrum signaling the instability of the configuration which decays into
the vacuum [5].
Our aim is to study the diffusion of a moving D-particle with a velocity v along one
space direction, say X1 , on another D-particle at rest at the origin. Such an interaction may
be evaluated analyzing the spectrum of the open strings stretching between the two objects
with modified boundary conditions in the 0 and 1 directions. This can be done generalizing
the treatment for the BPS D-branes presented in Ref. [38], but we find it simpler to use the
method of the boosted boundary state [25,26]. Indeed, the interaction amplitude just reads
AI (v) = NS hD00 |P |D0iNS + NS hD0| P |D00 iNS ,
(3.14)
where is the boost operator

= eiJ
01
(3.15)
is the
acting on the boundary state of a particle at rest. Here we have v = th() and
generator of the Lorentz transformations. Notice that the amplitude (3.14) reduces to the
static one (3.13) in the limit of vanishing velocity. The boosted boundary state 5 reads
"
#
X
T0 1 (8)
0
1
(x) (x v + x ) exp
an S e
an
|D0, iNS =
2
n=1
# 9
"
X
Y
er
r S
|k = 0i,
(3.16)
exp i
J
r=1/2
=0
where the boundary conditions are encoded in the matrix S = (V01 , 18 ) with

cosh(2) sinh(2)
.
V01 =
sinh(2) cosh(2)
(3.17)
Note that cosh() is the Lorentz factor.

The interaction amplitude AI (v) can be evaluated using standard techniques [25,26] and
explicitly reads 6
1
2 0 2
AI (v) = 8
Z
d
2 sinh()
ds s 9/2 e
b2 + 2 v 2 2
2 0 s
0

6
f3 (q) 3(i|is) f46 (q) 4(i|is)
f16 (q) 1(i|is)
(3.18)
in which q = es , is the proper time of the moving particle and b is the impact
parameter. We are now in a position to extract the long range interaction potential energy
grav
due to gravitational exchange between the two particles. To do so we have to perform
VI
the limit s in the integrand of Eq. (3.18), then integrate on the variable s and finally
identify the potential energy according to
5 The signs correspond to the two possible implementations of boundary conditions for world-sheet fermions.
In a physical (GSO projected) boundary state, only a suitable linear combination of them is retained.
6 See for instance Ref. [39] for definitions and conventions about the modular functions f and .
k
k
218
Z
AI (v) =
grav
d VI
(3.19)
In the non-relativistic limit, we obtain the Newtons law with its first correction

Me02
1 2
grav
v
1
+
+ o(v 2 ), v 0,
VI (r) = (210 )2
78 r 7
2
(3.20)
where we have introduced the radial coordinate r 2 = b2 + v 2 2 2 . Thus, in the nonrelativistic limit the boundary state calculation reproduces correctly the gravitational
potential energy that we expect for a pair of D-particles in relative motion.
4. Interactions of the heterotic non-BPS states

The non-BPS D-particles described in the previous sections account for the presence in
the spectrum of the type I theory of long super-multiplets of states carrying the spinorial
representation of SO(32). These non-perturbative states are dual to those appearing at
the first massive level in the heterotic theory. Carrying the same quantum numbers, one
naturally expects that these heterotic states have the same kind of interactions as the
D-particles of type I. In this section we will check this idea and investigate the gauge
and gravitational interactions of the non-BPS heterotic states using standard tools of
perturbative string theory. In doing so, we will adopt the bosonized formulation of the
heterotic string in which the gauge degrees of freedom are described by sixteen chiral
eI (I = 1, . . . , 16) appropriately compactified [10].
bosons X
The long super-multiplet of the stable heterotic states appears at the first massive level
(Mh2 = 4/ 0 ), and contains the following bosonic states
1 1/2 |ki |K I i,
1/2 1/2
1/2 |ki |K I i.
3/2 |ki |K I i,
]
|ki |K I i,
1 1/2
(4.1)
(4.2)
with , , . . . = 0, . . . , 9. In these formulas k denotes the spacetime momentum

(k 2 = Mh2 ) while K I is the adimensional momentum associated to the sixteen internal
eI . The states of Eq. (4.1) describe massive degrees of freedom which
coordinates X
transform in the 44 representation of the Lorentz group, whereas those of Eq. (4.2)
transform in the 84 7 . The level matching condition requires that K 2 = 4. This may be
realized for example by taking K I to be of the form ( 12 , 12 , . . . , 12 ) with an even
number of + signs, thus obtaining the spinorial representation of SO(32) with positive
chirality. The vertex operators for the states (4.1) and (4.2) will be denoted by V and can
be found in Appendix C both in the (1) and in the (0) superghost pictures.
We now study the interactions of these states with the massless gauge bosons of SO(32).
In the bosonized formulation of the heterotic string we must distinguish between the states
associated to the 16 Cartan generators that are given by
7 The fermionic states that complete this long multiplet transform in the 128 representation of the Lorentz
group.
I
A 1/2 |qi e
1
|Q = 0i,
with q 2 = 0 and I = 1, . . . , 16,
219
(4.3)
and those associated to the remaining 480 generators which are instead given by
A 1/2 |qi |Qi,
with q 2 = 0
(4.4)
and the internal momentum Q of the form (0, . . . , 0, 1, 0, . . ., 0, 1, 0, . . . , 0). Also the
vertex operators for the states (4.3) and (4.4), which we denote collectively by Vgauge , can
be found in Appendix C in the (1) and (0) superghost pictures.
The gauge coupling of the states (4.1) and (4.2) is obtained by simply computing the
3-point function on the sphere among two vertex operators V and one vertex operator
Vgauge (see Fig. 3). Including the normalization factor C0 appropriate of any tree-level
b for each vertex operator, we have
closed string amplitude and a normalization factor N
Z
gauge
b 3 d 2 (zi , z i )
= C0 N
F3
1 2

(0)
2(1)(z2 , z 2 ) ccV
gauge
(z3 , z 3 ) ,
(4.5)
ccV
1(1) (z1 , z 1 ) ccV
where 1 and 2 label the spinor representation of SO(32) carried by the non-BPS states
and a ghost factor cc has been added in each puncture. Actually, we are not interested in the
complete expression of this correlation function but only in the scalar part of it, namely in
the terms where the polarizations 1 and 2 of the two spinor states are contracted between
themselves. 8 This is because we want to compare our results with those of the non-BPS
D-particles of type I obtained in the previous sections in which the Lorentz group structure
was not manifest.
Using the explicit expression of the vertex operators reported in Appendix C, we find
that the terms of (4.5) proportional to 1 2 are given by
b 3 2 0
C0 N
gauge I

(1 2 ) A k1 k2 (K1 K2 )I K1 +K2 ,0 , (4.6)
=
F3
1 2
4
when Vgauge corresponds to the gauge bosons associated to the Cartan generators, and by
b 3 2 0
C0 N
gauge Q

(1 2 ) A k1 k2 CQ (K1 ) K1 +K2 +Q,0 , (4.7)
=
F3
1 2
2
Fig. 3. The 3-point function on the sphere.

8 The polarization can be either a vector, a symmetric or antisymmetric two-index tensor or an antisymmetric
i
three-index tensor depending on which particular states (4.1) and (4.2) are considered.
220
when Vgauge corresponds to the gauge bosons associated to the remaining generators (here
K denotes the lattice vector corresponding to the spinorial state and CQ (K1 ) is a
cocycle factor; see, e.g., Refs. [10,40] for details). The dependence on the internal momenta
looks different in the two expressions (4.6) and (4.7), but a moment thought reveals that it
is actually of the same form as required by gauge invariance. Indeed, we can rewrite both
equations in the following form
gYM
gauge AB

= (1 2 ) A k1 k2 AB
,
(4.8)
F3
1 2
1 2
2
where we have used the definitions of the normalization coefficients to express the prefactor
are the
in terms of the YangMills coupling constant of the heterotic theory. Here AB
1 2
matrix elements of the antisymmetrized product of two -matrices of SO(32) which
represent the fusion coefficients among the adjoint and two spinor representations of
SO(32). 9
We are now in the position of evaluating the contribution to the diffusion amplitude
among four particles due to the exchange of gauge bosons at tree level. In fact, this can
gauge
with the massless propagator
be simply obtained by sewing two 3-point functions F3
(2.18); in this way we obtain
gauge
1 4 ; 2 3
F4
2
gYM
us
(1 4 ) (2 3 )
2
t
1
2

AB
AB ,
1 4 2 3
(4.9)
where s, t and u are the usual Mandelstam variables which satisfy s + t + u = 16/ 0 . For
later convenience, we introduce the adimensional variable = u 0 /4, which in the limit
t 0 is related to the Lorentz parameter according to = 2( 1). Then, for t 0
Eq. (4.9) becomes
gauge
1 4 ; 2 3
F4
2
= gYM
(1 4 ) (2 3 )
Mh2 (2 + )
t
1
2

AB
AB .
1 4 2 3
(4.10)
Reverting
to the standard field theory normalization by multiplying each external leg by
1/ 2E, and removing for simplicity the polarization factors, we finally obtain the gauge
potential energy
gauge
1 4 ; 2 3
Vh
2
gYM
1
2 t
1
2

AB
AB ,
1 4 2 3
(4.11)
which in configuration space becomes

gauge
(r) =
1 4 ; 2 3
Vh
2
gYM
2
1
2
AB
AB
1 4 2 3
1
.
78 r 7
(4.12)
Notice that this potential does not depend on the relative velocity of the particles involved
in the interaction; moreover, as expected, it is a Coulomb-like potential for point particles
in ten dimensions carrying the spinor representation of the gauge group.
9 See for instance Ref. [41] for the expression of these -matrices in terms of the internal momenta and cocycle
factors.
221
We now turn to the gravitational interactions of the non-BPS heterotic particles (4.1) and
(4.2) following the same steps we have described for the gauge interactions. Let us recall
that the massless bosonic states of the graviton multiplet of the heterotic theory are
1
|qi |Q = 0i,
1/2 e
with q 2 = 0,
(4.13)
where the polarization is

= h = h ,
q h = 0
(4.14)
for the graviton,
= ( q ` q ` ),
8
q ` = 1,
`2 = 0
(4.15)
for the dilaton, and

1
1
= B = B ,
2
2
q B = 0
(4.16)
for the antisymmetric KalbRamond field. The vertex operators corresponding to these
states are written in Appendix C in the (1) and (0) superghost pictures, and will be
denoted generically by Vgrav .
The gravitational coupling of the non-BPS particles can be determined by evaluating the
correlation function among two vertex operators V and one vertex operator Vgrav , namely
Z
grav
b 3 d 2 (zi , z i )
F 3 = C0 N
1 2

(1)
(1)
(0)
2 (z2 , z 2 ) ccV
grav
(z3 , z 3 ) .
(4.17)
ccV
1 (z1 , z 1 ) ccV
As before, also now we are interested only in the scalar part of this expression which is
proportional to 1 2 , since we want to compare it with the boundary state calculation of
Section 3. Using the expression of the vertex operators reported in Appendix C, it is not
difficult to find that
grav
b 3 0 (1 2 ) k k ,
(4.18)
F 3 = C0 N
2 2
1 2
from which we read that the couplings of the non-BPS states with the graviton, the dilaton
and the antisymmetric tensor field are
grav (h)
F3 = 410 (1 2 ) h k2 k2 ,
1 2
grav ()
(4.19)
F3 = 2 10 Mh2 (1 2 ) ,
1 2

grav (B)
F3 = 0.
1 2
Note that the vanishing of the heterotic non-BPS states coupling with the antisymmetric
KalbRamond field is consistent with the fact that the type I D-particle does not couple to
the RR 2-form.
Now we can evaluate the diffusion amplitude of the non-BPS particles due to
gravitational exchanges by gluing two 3-point functions (4.19) with the appropriate
222
massless propagators. Summing over graviton and dilaton exchanges and introducing the
same notation adopted for the gauge interactions, we obtain
grav
1 4 ; 2 3
F4
= (210)2 (1 4 )(2 3 )
Mh4 (2 + )2
.
t
(4.20)
Notice that, as expected, neither the 3-point functions (4.19) nor the 4-point function
(4.20) depend on the indices i that span the SO(32) spinor representations carried by
the non-BPS
particles; therefore they can be suppressed. Normalizing each external leg by
a factor of 1/ 2E and removing for simplicity the polarization terms, we can obtain from
Eq. (4.20) the following gravitational potential energy
grav
Vh
(r) = (210 )2
Mh2
,
78 r 7
(4.21)
which in the small velocity limit becomes

grav
Vh
(r) = (210 )2

Mh2
1 2
2
v
1
+
+
o
v
.
2
78 r 7
(4.22)
In Eq. (4.21) we recognize Newtons law for point particles of mass Mh separated by a
distance r in ten dimensions with the appropriate relativistic correction.
We conclude this section by mentioning that the same results (4.10) and (4.20) can be
directly obtained by evaluating a 4-point function of non BPS states on the sphere, or more
precisely its universal part in the t-channel which is proportional to (1 4 )(2 3 ). In
fact, using standard techniques, one can show that this part of the 4-point amplitude is
A4 =
4
b 4 (1 4 )(2 3 )A(s, t, u; S, T , U ),
C0 N
2
(4.23)
where
A(s, t, u; S, T , U )
0 0 0
t
u
s
sin
sin
= (K) sin
4
4
4

t 0
u 0
s 0

3
3
4
4
4

t 0 T
u 0 U
s 0 S
.
1
4
2
4
2
4
2
(4.24)
In this expression S, T and U are the Mandelstam variables for the internal momenta
which obey S + T + U = 16, and (K) cK3 (K1 + K2 ) cK2 (K1 )()U/2 is a cocycle
factor whose values are 1 (see for instance Ref. [40]). Since we are interested only in
the contributions due to exchanges of massless states in the t channel, we must look for
the poles of A4 with respect to t. Inspection of Eq. (4.24) shows that these occur only for
T = 0 and T = 2. The poles for T = 0 correspond to exchanges of gravitons, dilatons
and gauge bosons associated to the Cartan generators of SO(32), while those for T = 2
correspond to exchanges of the remaining 480 gauge bosons. In the limit t 0, we have
223
b4
1
16C0N
(4.25)
(4 + S/2)(2 + ) + (2 + )2 ,
0
t
b4
16C0N
1
(4.26)
(2 + ) .
A4 | T =2 '
0
t
We can disentangle the gauge and gravity pieces of Eq. (4.25) by observing that, because
of gauge invariance, the gauge part at T = 0 should have the same dynamical dependence
as the amplitude (4.26) for T = 2. Hence, we can conclude that the term of Eq. (4.25)
linear in (2 + ) is due to gauge interactions, while the term quadratic in (2 + ) comes
from gravity. Inspection of the coefficients and a little algebra show that these expressions
indeed match with Eqs. (4.10) and (4.20), thus providing a strong check on our previous
calculations and on their interpretation.
A4 | T =0 '
5. Conclusions
We now compare the results obtained in the previous sections and discuss their relation
in the light of the heterotic/type I duality. For the gravitational interactions, the comparison
is quite simple since in both theories we have found a potential energy of the form

M2
(5.1)
1 + 12 v 2 + o(v 2 )
78 r 7
in the non-relativistic limit (see Eqs. (3.20) and (4.22)). The only thing that one has to do
to have complete agreement is to change the values of the gravitational coupling constant
10 and of the mass M according to the duality map as we discussed in the introduction.
What is nice to observe is that these changes make the two gravitational potential energies
agree not only at the static level but also at the first non-trivial order in the velocity v.
For the gauge interactions the situation is a bit different. Both in the type I theory and
in the heterotic string we have found that the gauge potential energy of the stable non-BPS
states is in the form of Coulombs law (see Eqs. (2.20) and (4.12)). However, the detailed
gauge group structure is not the same in the two cases. The reason for this is quite simple.
In the heterotic theory one is able to describe the non-BPS particles in a complete way
because they are perturbative configurations of the heterotic string, and in particular one
can fully specify the polarizations of these states also with respect to the gauge group. This
is why the gauge amplitudes involving these particles explicitly depend on the indices of
the spinorial representation of SO(32) (see Eqs. (4.8) and (4.10)). On the other hand, in
the type I theory the non-BPS particles are non-perturbative configurations of the type I
string, and thus the description one is able to provide for them using perturbative methods
is necessarily incomplete. This fact should not be surprising, because also in the case of
the supersymmetric BPS D-branes one is not able to account for their degeneracy (with
respect to both the Lorentz group and the gauge group) using open strings with Dirichlet
boundary conditions or equivalently boundary states. Indeed with these methods one can
compute only the universal parts of the interactions involving D-branes.
In Section 2 we have introduced a method to describe the emission of a colored gauge
boson from a non-BPS D-particle viewed as a source carrying not the spinorial indices of
V grav (r) = (210 )2
224
SO(32), but rather those of the bi-fundamental representation formed with the ChanPaton
factors of the boundary changing vertex operators V09 and V90 (see Eqs. (2.4) and (2.3)).
In this framework, using the various kinds of open strings of type I we have been able to
account for the gauge interactions of the non-BPS D-particles, but then the comparison
with the heterotic theory is not immediate. In order to do such a comparison, we must
reduce the heterotic gauge potential energy by taking into account the contribution
of all pairs of states compatible with the emission of a gauge boson of definite color.
From the group theory point of view, this amounts to transform the spinorial indices of
gauge
)1 4 ; 2 3 given in Eq. (4.12) into those of the bi-fundamental representation. This
(Vh
can be easily done by noting that
AD
BC

= Tr A D BC = Tr(1) BD AC CD AB .
(5.2)
Then, using this identity and Eq. (4.12), we obtain the following reduced gauge potential
energy for the heterotic non-BPS particles
gauge
Vh
(r) Tr(1)2 A D
=
4 1
EF
3 2
gauge
Vh
2
1
gYM
AE F D AF DE
.
2
78 r 7
1 4 ; 2 3
(r)
(5.3)
This expression exactly agrees with the corresponding one for the type I theory given in
Eq. (2.20).
We remark that it is not meaningful to perform this reduction directly on the heterotic
3-point function (4.8) and then compare it with the type I amplitude (2.17) describing the
emission of a gauge boson from a D-particle. In fact, in the perturbative type I theory the
D-particle is an infinitely massive object which acts as a reservoir of momentum, and thus
the spacetime structure of its amplitudes cannot match with that of the heterotic scattering
amplitudes. In other words, the diagram represented in Fig. 1 describing the gauge emission
from a D-particle of type I must not be considered as a vertex or a 3-point function in the
field theory sense, but rather as a 1-point function in some definite background. A similar
situation occurs also in the gravitational sector where the boundary state representing the
D-particle generates all its 1-point functions, i.e., all its couplings with the closed string
states of the bulk. In this sense, what we have done in Section 2 is to find the 1-point
function of the non-BPS D-particle with the massless states of the other sector of the bulk,
namely the open strings of type I. It would be nice to extend these results to all states of
this open string sector.
In conclusion, in this paper we have described how to compute the gauge and
gravitational potential energies of the non-BPS D-particles of type I and shown that these
agree with the corresponding ones computed for the dual heterotic states provided that one
uses the known duality and renormalization effects. Our results thus provide a dynamical
test of the heterotic/type I duality at the non-BPS level.
225
Acknowledgements
We would like to thank C. Bachas, M. Bill, P. Di Vecchia, M. Frau, B. Pioline,
C. Schweigert and R. Russo for several useful discussions. We especially thank B. Pioline
for valuable discussions and remarks.
Appendix A
In this appendix we describe the gauge emission from a D-string of type I, following the
scheme we proposed in Section 2.
As a consequence of the BPS condition, two parallel BPS Dp-branes of type I or type II
do not exert any force on each other. In the type II theories, the interaction between two
branes is mediated only by the exchange of closed strings and the vanishing of the force is
easily seen using boundary states. Indeed one finds that
hDp|P |Dpi = 0
(A.1)
at the leading order in the string coupling constant. In the limit of large distance between
the branes, when only massless closed string states are exchanged, this means that the
attraction due to gravitons and dilatons is compensated by the repulsion due to the p + 1
RR form under which the Dp-branes are charged. In the case of two parallel D-strings
the interaction (A.1) is globally invariant under the world-sheet parity so that it vanishes
also in type I theory. However, in this theory one has to consider also the exchange of open
99 strings which are present in the bulk and whose first contribution associated to a disk
with two boundary components on the D-strings and two on the D9-branes appears at the
next-to-leading order in the string coupling constant. For the no force condition to be true,
this disk amplitude has thus to vanish identically. At first sight, this seems striking since
the D-string is charged under the gauge potential (in fact it carries the SO(32) spinorial
representation). However, one must recall that, being an extended object, the D-string
cannot be minimally coupled to the gauge potential and thus the naive conclusion does
not apply. The aim of this appendix is to evaluate the gauge emission from a D-string
of type I using the same methods applied in the case of the D-particle (but without the
technicalities due to the boundary fermion ), and then to compute its contribution to the
diffusion process between two D-strings.
We first briefly discuss the spectrum of the 19 open strings stretching between a
D-string and a D9-brane. As for the 09 strings, also here the NS sector is massive and does
not represent any degeneracy of the D-string; thus we do not consider it. In the R sector,
instead, there are massless states. Since the world-sheet fermions 0 , 1 have zero modes,
the massless R ground state is a GSO even (chiral) spinor of SO(1, 1). The corresponding
vertex operator reads
(1/2)
V91
= A S + 0 e/2 ,
(A.2)
where S + is a positive chirality spin field of conformal dimension 1/8, and 0 is a

boundary changing operator for the eight space directions transverse to the D-string of
226
conformal dimension 1/2. The vertex operator for the massless R states of the 19 sector
is obtained by acting with on V91 ; thus
(1/2)
V19
(1/2)
= V91
= t S + 0 e/2 .
A
(A.3)
We now evaluate the coupling of the D-string with the gauge bosons of type I. As in the
case of the D-particle, this is determined by the amplitude on a disk with one boundary
component on the D-string and one component on one of the 32 D9-branes from which a
gauge boson is emitted, and is represented by the 1-point function of the gauge boson in
the vacuum representing the D-string, i.e.,
(1)
(1)|91i,
F gauge = h19 |Vgauge
(A.4)
where
(1/2)
|19 i = lim V19

z0
(1/2)
(z)|0i and h19 | = lim h0| V19

z
(z).
(A.5)
After introducing the appropriate normalization factors, we can express F gauge as

Z
(1/2)

(1/2)
(1)
(z1 ) c Vgauge
(z2 ) c V91
(z3 ) .
F gauge = Cdisk NR2 NNS d(zi ) c V19
(A.6)
For the same reasons discussed in the case of the D-particle, also here there is no emission
of gauge fields with polarizations along the directions transverse to the D-string; thus we
have emissions only in the two longitudinal directions = 0, 1, but these can occur with
arbitrary transverse momentum. To evaluate (A.6) we need the following basic correlators
[41]

0
0 k 2 0 k 2 1 0 k 2
z13
z23 ,
(z1 ) eikX(z2) 0 (z3 ) = z12

+
++
1/2 1/4 1/2
z12 z13 z23
(A.7)
S (z1 ) (z2 ) S + (z3 ) = C 1
for = 0, 1. Inserting them into Eq. (A.6), we obtain

++
A
F gauge = Cdisk NR2 NNS Tr t BC D A C 1

= i gYM AB CD AC BD (A0 A1 ),
Z
d(zi )
z13
z12 z23
0 k 2
(A.8)
where we have used that ( 0 C 1 )++ = ( 1 C 1 )++ = 1. As anticipated, this is not a

minimal gauge coupling because the D-string is an extended object.
Now, using this coupling and the propagator (2.18) for the massless gauge bosons, we
gauge
gauge
between two D-strings given by VI
can obtain the gauge potential energy VI

gauge
gauge
PF
. Inserting the explicit values, we find
F
gauge
VI
2
= gYM
(AE DF AF DE )
(1 + 1)
= 0.
q2
(A.9)
The vanishing of this contribution confirms our previous statements about the no force
condition.
227
Appendix B
In this appendix we present the definitions of the various normalization factors that are
needed for the calculations presented in Sections 2, 3 and 4.
The gravitational coupling constant 10 , the normalization C0 of the closed-string treeb of the closed string vertex
level diagrams (sphere diagrams) and the normalization factor N
operators have the same expressions for both the SO(32) heterotic string and the type I
theory, and are given by
10 = 8 7/2 0 2 g,
4 0 5 2
C0 = (2) g
b = 8 5/2 0 2 g,
N
(B.1)
,
(B.2)
(B.3)
where g is the string coupling constant (gh for the heterotic string and gI for the type I
theory).
In the heterotic string, the gauge coupling constant gYM is related to 10 as follows [11]
4 2
= 28 7 0 3 gh2 ,
0 10
while in the type I theory the relation between gYM and 10 is [11]
2
=
gYM
2
=
gYM
23/2 2
= 215/2 7 0 3 gI .
0 gI 10
(B.4)
(B.5)
In type I string theory one must consider also diagrams involving open strings. The
normalization of the disk diagrams is [11,32]
2
(2 0 )2 ,
Cdisk = gYM
(B.6)
while the normalization factors of the open string vertex operators in the NS and R sectors
are
NR = gYM (2 0 )3/4 .
(B.7)
NNS = gYM 2 0 ,
We now list the expression of the various ChanPaton factors that were used in Section 2.
The factor AB carried by the gauge boson has indices in the adjoint representation of
SO(32). Its matrix elements explicitly read

B
A B
D
C .
(B.8)
AB CD = i CA D
The ChanPaton factor A associated to 09 strings is a column vector with an index in
the fundamental representation of SO(32). It reads

(B.9)
A B = BA .
With these expressions it is easy to see that

Tr AB CD = 2 AC BD AD BC ,

A
Tr t BC D = i AB CD AC BD .
(B.10)
228
Appendix C
In this appendix we write the vertex operators of the non-BPS heterotic states (4.1) and
(4.2), of the gauge bosons of SO(32) given in Eqs. (4.3) and (4.4), and of the bosonic states
(4.13) of the graviton multiplet.
In the (1) superghost picture, the vertices associated to the non-BPS states (4.1) and
(4.2) are
(1)
(z, z ) = A (z) eikX(z,z) CK ei(K/ 2 )X(z) ,
VA
1
0 e
(1)
B (z) eikX(z,z) CK ei(K/ 2 )X(z) ,
VB (z, z ) =
2 0
1
0 e
VC(1) (z, z ) = C (z) eikX(z,z) CK ei(K/ 2 )X(z) ,
3!
where
(C.1)
(C.2)
(C.3)
A (z) = (z) e(z),

B (z) = (z) x (z) e(z),
C (z) = (z) (z) (z) e(z) .
(C.4)
In Eq. (C.2) the polarization tensor is symmetric or antisymmetric depending on

whether one considers a state in the 44 or in the 84 representation of the Lorentz group.
Moreover, in all vertex operators we have introduced suitable cocycle factors CK , which
depend only on the internal momenta and satisfy [40]
CK (P ) CK 0 (P ) = CK+K 0 (P ).
(C.5)
Applying the picture-changing operator to V (1) we can obtain the vertices V (0) in the (0)
superghost picture. They are given by the same expressions (C.1)(C.3) with A , B and
C replaced, respectively, by

b (z) = i 2 X (z) i 0 (k ) (z) ,
A
2 0

b (z) = i 2 0 (z) (z) (i/2)(k ) (z)X (z)

B

1
+ 0 X (z)X (z) ,
2

i
i 0 (k ) (z) (z) (z) + X (z) (z) (z)
Cb (z) =
2 0

(z)X (z) (z) + (z) (z)X (z) .
(C.6)
The vertex operators for the gauge bosons (4.3) associated to the 16 Cartan generators of
SO(32) in the (1) superghost picture are
i
eI (z),
A (z) e(z) eiqX(z,z) X
(C.7)
2 0
while those for the gauge bosons (4.4) associated to the 480 remaining generators are
(1)
(z, z ) =
Vgauge

(1)
Vgauge
(z, z ) = A (z) e(z) eiqX(z,z) CQ ei(Q/
e z)
2 0 )X(
229
(C.8)
In the (0) superghost picture these vertices become, respectively,

(0)
(z, z ) =
Vgauge
1
eI (z),
A [X (z) i 0 (q ) (z)] eiqX(z,z) X
2 0
(C.9)
and
(0)
(z, z ) =
Vgauge
i
2 0
A [X (z) i 0 (q ) (z)] eiqX(z,z) CQ ei(Q/
e z)
2 0 )X(
(C.10)
Finally, the vertices for the bosonic states (4.13) of the graviton multiplet are
i
(1)
e (z)eiqX(z,z)
(z, z ) =
(z) e(z) X
Vgrav
2 0
in the (1) superghost picture, and
1
e (z) eiqX(z,z)
[X (z) i 0 (q ) (z)] X
2 0
in the (0) superghost picture.
(0)
(z, z ) =
Vgrav
(C.11)
(C.12)
References
[1] J. Polchinski, Phys. Rev. Lett. 75 (1995) 184, hep-th/9510169; TASI lectures on D-branes, hepth/9611050.
[2] A. Sen, JHEP 9806 (1998) 007, hep-th/9803194.
[3] A. Sen, JHEP 9808 (1998) 010, hep-th/9805019.
[4] A. Sen, JHEP 9809 (1998) 023, hep-th/9808141.
[5] A. Sen, JHEP 9810 (1998) 021, hep-th/9809111.
[6] A. Sen, JHEP 9812 (1998) 021, hep-th/9812031.
[7] A. Sen, Non-BPS states and branes in string theory, hep-th/9904201.
[8] A. Lerda, R. Russo, Stable non-BPS states in string theory: a pedagogical review, hepth/9905006.
[9] J.H. Schwarz, TASI lectures on non-BPS brane systems, hep-th/9908144.
[10] D.J. Gross, J.A. Harvey, E. Martinec, R. Rohm, Nucl. Phys. B 256 (1985) 253; Nucl. Phys.
B 267 (1986) 75.
[11] J. Polchinski, String Theory, Vol. II, Cambridge Univ. Press, 1998.
[12] J. Polchinski, E. Witten, Nucl. Phys. B 460 (1996) 525.
[13] M. Frau, L. Gallot, A. Lerda, P. Strigazzi, Nucl. Phys. B 564 (2000) 60, hep-th/9903123.
[14] O. Bergman, M.R. Gaberdiel, Phys. Lett. B 441 (1998) 133, hep-th/ 9806155.
[15] O. Bergman, M.R. Gaberdiel, JHEP 9903 (1999) 013, hep-th/9901014.
[16] M.R. Gaberdiel, A. Sen, JHEP 9911 (1999) 008, hep-th/9908060.
[17] M.R. Gaberdiel, B.K. Stefanski, Dirichlet branes on orbifolds, hep-th/ 9910109.
[18] M. Frau, L. Gallot, A. Lerda, P. Strigazzi, Stable non-BPS D-branes of type I, hep-th/0003022.
[19] T. Dasgupta, B. Stefanski, Nucl. Phys. B 572 (2000) 95, hep-th/9910217.
[20] E. Witten, D-branes and K-theory, hep-th/9810188.
[21] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 293 (1987) 83; Nucl. Phys.
B 308 (1988) 221.
[22] J. Polchinski, Y. Cai, Nucl. Phys. B 286 (1988) 91.
230
[23] P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Nucl. Phys. B 507 (1997)
259, hep-th/9707068.
[24] M. Bill, P. Di Vecchia, M. Frau, A. Lerda, I. Pesando, R. Russo, S. Sciuto, Nucl. Phys. B 526
(1998) 199, hep-th/9802088.
[25] M. Bill, D. Cangemi, P. Di Vecchia, Phys. Lett. B 400 (1997) 63, hep-th/9701190.
[26] M. Bill, P. Di Vecchia, M. Frau, A. Lerda, R. Russo, S. Sciuto, Mod. Phys. Lett. A 13 (1998)
2977, hep-th/9805091.
[27] P. Di Vecchia, A. Liccardo, D branes in string theories, I, hep-th/9912161; D branes in string
theories, II, hep-th/9912275.
[28] A. Hashimoto, Nucl. Phys. B 496 (1997) 243, hep-th/9608127.
[29] S. Hamidi, C. Vafa, Nucl. Phys. B 279 (1987) 465.
[30] L. Dixon, D. Friedan, E. Martinec, S. Shenker, Nucl. Phys. B 282 (1987) 13.
[31] J. Frhlich, O. Grandjean, A. Recknagel, V. Schomerus, Fundamental strings in DpDq brane
systems, hep-th/9912079.
[32] P. Di Vecchia, A. Lerda, L. Magnea, R. Marotta, R. Russo, Nucl. Phys. B 469 (1996) 235,
hep-th/9601143.
[33] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
[34] A. Sagnotti, in: G. Mack et al. (Eds.), Non-Perturbative Quantum Field Theory, Pergamon Press,
1988, p. 521.
[35] N. Ishibashi, T. Onogi, Nucl. Phys. B 310 (1989) 239.
[36] G. Pradisi, A. Sagnotti, Phys. Lett. B 216 (1989) 59.
[37] P. Horava, Nucl. Phys. B 327 (1989) 461.
[38] C. Bachas, Phys. Lett. B 374 (1996) 37, hep-th/9511043.
[39] E. Kiritsis, Introduction to superstring theory, hep-th/9709062.
[40] M.B. Green, J.H. Schwarz, E. Witten, Superstring theory, Vol. I, Cambridge Univ. Press, 1987.
[41] V.A. Kostelecky, O. Lechtenfeld, W. Lerche, S. Samuel, S. Watamura, Nucl. Phys. B 288 (1987)
173232.

Nonnormalizable zero modes on BPS junctions

Kenji Ito, Masashi Naganuma , Hodaka Oda, Norisuke Sakai
Department of Physics, Tokyo Institute of Technology, Oh-okayama, Meguro, Tokyo 152-8551, Japan
Received 28 April 2000; accepted 5 July 2000
Abstract
Using an exact solution as a concrete example, NambuGoldstone modes on the BPS domain wall
junction are worked out for N = 1 supersymmetric theories in four dimensions. Their wave functions
extend along the wall to infinity (not localized) and are not normalizable. It is argued that this feature
is a generic phenomenon of NambuGoldstone modes on domain wall junctions in the bulk flat
space in any dimensions. We formulate mode equations and show that fermion and boson with the
same mass come in pairs except massless modes which can appear singly, in accordance with unitary
representations of (1, 0) supersymmetry. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.27.+d; 11.30.P; 11.15.Kc; 11.10.-z
1. Introduction
In recent years an interesting idea has been advocated to regard our world as a
domain wall embedded in higher dimensional spacetime [1,2]. Most of the particles in
the standard model should be realized as modes localized on the wall. Phenomenological
implications of the idea have been extensively studied from many aspects. Another
fascinating possibility has also been proposed to consider walls in the bulk spacetime which
has negative cosmological constant [3,4]. The model can give large mass hierarchy or can
give massless graviton localized on the wall. Subsequently a great deal of research activity
has been performed to study and extend the proposal [811].
Since walls typically have co-dimension one, it is desirable to consider intersections
and/or junctions of walls in order to obtain our four dimensional world from a spacetime
with much higher dimensions. The model with the bulk cosmological constant has been
extended to produce an intersection of walls [57].
E-mail addresses: kito@th.phys.titech.ac.jp (K. Ito), naganuma@th.phys.titech.ac.jp (M. Naganuma),

hoda@th.phys.titech.ac.jp (H. Oda), nsakai@th.phys.titech.ac.jp (N. Sakai).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 3 6 - 3
232
K. Ito et al. / Nuclear Physics B 586 (2000) 231260
Supersymmetry has been useful to achieve stability of solitonic solutions such as domain
walls. Domain walls in supersymmetric theories can saturate the Bogomolnyi bound [12,
13]. Such a domain wall preserves half of the original supersymmetry and is called a
1/2 BPS state [14]. It has also been noted that these BPS states possess a topological
charge which becomes a central charge Z of the supersymmetry algebra [1517]. Thanks
to the topological charge, these BPS states are guaranteed to be stable under arbitrary local
fluctuations. Various properties of domain walls in N = 1 supersymmetric field theories
in four dimensions have been extensively studied [1822]. In particular the modes on the
domain wall background have been worked out and are found to contain fermions and/or
bosons localized on the wall in many cases [23,24].
Recently domain wall junctions have attracted much attention as another interesting
possibility for BPS states [2527]. Domain walls occur in interpolating two discrete
degenerate vacua in separate region of space. If three or more different discrete vacua
occur in separate region of space, segments of domain walls separate each pair of the
neighboring vacua. If the two spatial dimensions of all of these domain walls have one
dimension in common, these domain walls meet at a one-dimensional junction. The
solitonic configuration for the junction can preserve a quarter of supersymmetry and is
called a 1/4 BPS state. There has been progress to study general properties of such
domain wall junctions. For instance a new topological charge Y is found to appear for
such a 1/4 BPS state [2527]. If we start from N = 1 four dimensional supersymmetric
field theories, the domain wall junction preserves only one supercharge. Consequently the
resulting theory was expected to be a (1, 0) supersymmetric theory in 1 + 1 dimensions
[25] which offers an intriguing possibility of chiral fermions. Moreover, there have been a
number of numerical simulations which indicate the existence of the domain wall junction
solutions [28,29]. In spite of all these efforts, it has been difficult to obtain an explicit
solution and to prove the existence of a BPS domain wall junction.
Recently we have succeeded to work out an exact solution for the BPS domain wall
junction for the first time [30]. The exact solution allows a thorough study of the properties
of the BPS domain wall junction. Consequently several misconceptions can be pointed out
and rectified. One such point is the sign and meaning of the new central charge Y which
arises when walls form a junction. Our exact solution showed that the central charge Y
contributes negatively to the mass of the domain wall junction configuration. Therefore we
should not consider the central charge Y alone as the mass of the junction. Various other
aspects of the domain wall junctions are also studied recently [3447].
The purpose of the present paper is to give a more detailed study of the properties of the
BPS domain wall junction in N = 1 supersymmetric field theories. We study the modes on
the background of the domain wall junction, especially the NambuGoldstone modes. We
will use our exact solution as a concrete example and will extract the generic properties
of the BPS domain wall junctions. We define mode equations and demonstrate explicitly
that fermion and boson with the same mass have to come in pairs except massless modes.
Massless modes can appear singly without accompanying fields with opposite statistics.
We also show that unitary representations of the surviving (1, 0) supersymmetry are
classified into doublets for massive modes and singlets for massless modes. We work out
233
explicitly massless NambuGoldstone modes associated with the broken supersymmetry

and translational invariance. We find that the NambuGoldstone fermions exhibit an
interesting chiral structure in accordance with the surviving (1, 0) supersymmetry algebra.
However, we also find that any linear combinations of the NambuGoldstone modes
associated with the junctions become a linear combination of zero modes on at least one of
the domain walls asymptotically along these walls. Since their wave functions are extended
along these walls without damping, they are not localized states on the junction. Therefore
they are not normalizable, contrary to a previous expectation [25]. This indicates that the
resulting theory cannot be regarded as a genuine (1 + 1)-dimensional field theory with
discrete particle spectrum even at zero energy. Although the remaining supersymmetry is
just (1, 0) which is characteristic to 1 + 1 dimensions, we have to keep in mind that the
domain wall junction configuration is actually living in one more dimensions similarly to
the domain wall itself. Zero modes on the junction turn out to have properties quite similar
to those on the domain wall. The non-normalizability of NambuGoldstone modes on the
junction configuration is not an accident in this particular model. We observe that the origin
of this property can be traced back to the fact that the supersymmetry is broken by the
coexistence of nonparallel walls. Therefore the fact that the NambuGoldstone modes on
the BPS domain wall junction are not normalizable is a generic feature of supersymmetric
field theories in the bulk flat space.
One should note that our conclusion need not apply to the case with negative
cosmological constant in the bulk. In the presence of a bulk negative cosmological constant
in six dimensions, five-dimensional walls can intersect in anti-de Sitter space. If one
demands a flat space at the four-dimensional intersection, one has an anti-de Sitter space
not only in the bulk but also even on the walls [57]. Since anti-de Sitter space does not
have translational invariance, the wave function of the zero mode does not become constant
along the wall asymptotically, contrary to our situation. If one approaches the intersection
along the wall, one meets precisely the same situation as the wall in the five-dimensional
anti-de Sitter space. For instance graviton zero mode is exponentially suppressed away
from the intersection along the wall direction to produce a normalizable wave function.
Therefore the anti-de Sitter geometry along the wall plays an essential role to achieve the
localization of the wave function on the intersection in models with cosmological constant.
In Section 2, we introduce BPS equations and the exact solution for the domain wall
junction and discuss representations of the surviving (1, 0) supersymmetry algebra. In
Section 3, we present mode equations which define the fluctuations on the background
of domain wall junction. We work out the NambuGoldstone mode explicitly and show
that they are not normalizable. Physical origin of the nonnormalizability is clarified and the
general validity of this phenomenon is argued. In Section 4, the relation between the choice
of BPS equations and the boundary condition is discussed for a general WessZumino
model. In Section 5, the central charge density and the energy density are examined and
an interesting behavior is observed. The fermionic contributions to the central charges and
mode equations in a convenient gamma matrix representation are given in the appendices.
234
2. BPS equations and the (1, 0) supersymmetry algebra

2.1. Two 1/4 BPS states and two BPS equations
It is known that if the translational invariance is broken as is the case for domain walls
and/or junctions, the N = 1 superalgebra in general receives contributions from central
charges [1632]. The anti-commutator between two left-handed supercharges has central
charges Zk , k = 1, 2, 3

(2.1)
{Q , Q } = 2i k 0 Zk .
Here and the following we use two-component spinors following the convention of
Ref. [33] except that the four-dimensional indices are denoted by Greek letters , =
0, 1, 2, 3 instead of roman letters m, n. The anti-commutator between left- and righthanded supercharges receives a contribution from central charges Yk , k = 1, 2, 3 besides
the energymomentum four-vector P , = 0, . . . , 3 of the system

S } = 2 P + k Yk .
(2.2)
{Q , Q

One may call Zk and Yk as (1, 0) and (1/2, 1/2) central charges in accordance with the
transformation properties under the Lorentz group. Central charges, Zk and Yk , come from
the total divergence, and they are non-vanishing when there are nontrivial differences in
asymptotic behavior of fields in different region of spatial infinity as is the case of domain
walls and junctions [26]. Therefore these charges are topological in the sense that they
are determined completely by the boundary conditions at infinity. For instance, we can
take a general WessZumino model with an arbitrary number of chiral superfields i , an
arbitrary superpotential W and an arbitrary Khler potential K( i , j )
Z

Z

d 2 W i + h.c. ,
(2.3)
L = d 2 d 2 K i , j +
and compute the anticommutators (2.1), (2.2) to find the central charges. The contributions
to these central charges from bosonic components of chiral superfields are given by [26] 1
Z
Zk = 2 d 3 x k W (A ),
(2.4)
Z

(2.5)
Yk = i knm d 3 x Kij n Aj m Ai , 123 = 1,
where Ai is the scalar component of the ith chiral superfield i and Kij =
2 K(A , A)/Ai Aj is the Khler metric.
BPS domain wall is a 1/2 BPS state [16,17] and BPS domain wall junction is a 1/4
BPS state [25,26]. To find the BPS equations satisfied by these BPS states, we consider a
S with an arbitrary complex two-vector
hermitian linear combination of operators Q and Q
and its complex conjugate = ( ) as coefficients

S .
K = Q + Q
(2.6)
1 The central charge Y also receives contributions from fermionic components of chiral superfields which is
k
given in Appendix A.
235
We treat as c-numbers rather than the Grassmann numbers. Since K is hermitian, the
expectation value of the square of K over any state is non-negative definite, hS|K 2 |Si > 0.
The field configuration of static junction must be at least two-dimensional. If we assume,
for simplicity, that it depends on x 1 , x 2 then we obtain hZ3 i = hY1 i = hY2 i = 0 from
Eqs. (2.4) and (2.5), and the inequality implies in this case
n

1
1 2 2 2 hY3 i + Re 1 2 hZ2 iZ1 i
hH i > 1 2
2
2
| | + | |
o
2
(2.7)
+ Re 2 hZ2 + iZ1 i ,
for any and for any state. The equality holds if and only if the linear combination of
supercharges, K, is preserved by the state |Si
K |Si = 0.
(2.8)
In this case, the state |Si saturates the energy bound and is called a BPS state. We find that
there are two candidates for the saturation of the energy bound [30]:
H = HI |hiZ1 Z2 i| hY3 i,
hiZ1 + Z2 i
, 2 = 2 = 0,
when 1 = 1
|hiZ1 + Z2 i|
H = HII |hiZ1 Z2 i| + hY3 i,
hiZ1 + Z2 i
.
when 1 = 1 = 0, 2 = 2
|hiZ1 + Z2 i|
(2.9)
(2.10)
In the case of HI 6= HII , the BPS bound becomes hH i > max{HI , HII }. If HI > HII ,
then supersymmetry can only be preserved at hH i = HI and the only one combination of
supercharges is conserved

hiZ1 + Z2 i S
(2.11)
Q1 |hH i = HI i = 0.
Q1 +
|hiZ1 + Z2 i|
If HII > HI , then supersymmetry can only be preserved at hH i = HII and the only one
combination of supercharges is conserved

hiZ1 + Z2 i S
Q |hH i = HII i = 0.
(2.12)
Q2 +
|hiZ1 + Z2 i| 2
In the case of HI = HII , two candidates of BPS bounds coincide and BPS state conserves
both of two supercharges, (2.11) and (2.12); this is a 1/2 BPS state.
For the general WessZumino model in Eq. (2.3), the condition of supercharge
conservation (2.11) for H = HI applied to chiral superfield i = (Ai , i , F i ) gives after
eliminating the auxiliary field F i
2
Ai
W
= + F i = + K 1ij
,
z
Aj
+ i
hiZ1 + Z2 i
,
|hiZ1 + Z2 i|
(2.13)
where complex coordinates z = x 1 + ix 2 , z = x 1 ix 2 , and the inverse of the Khler
metric K 1ij are introduced. We can also consider gauge interactions. For simplicity we
236
take only the U (1) gauge interaction. Then the derivative Ai / z in the above Eq. (2.13)
should be replaced by the gauge covariant derivative Dz Ai
1
W
2Dz Ai = + K 1ij
,
Dz = (D1 + iD2 ),
j
A
2

ei
i
i
+ i v A .
(2.14)
D A =
x
2
Moreover the same BPS condition (2.11) applied to vector superfield in the WessZumino
gauge V = (v , , D) gives after eliminating the auxiliary field D
1 X j
A ej A j ,
v03 = 0,
v01 = v31 ,
v23 = v02 , (2.15)
v12 = D =
2
j
where v v v and ej is the charge of the field Aj . Here we assume for

simplicity the minimal kinetic term both for the chiral superfield Kij = ij and for the
vector superfield.
Similarly the condition of supercharge conservation (2.12) for H = HII applied to chiral
superfield in the WessZumino model gives after eliminating the auxiliary field
2
Ai
W
= F i = K 1ij
,
j
z
A
hiZ1 Z2 i
.
|hiZ1 Z2 i|
(2.16)
If U (1) gauge interaction is present, the derivative Ai /z should be replaced by the

covariant derivative Dz Ai = 12 (D1 iD2 )Ai . In this case the BPS condition applied to
U (1) vector superfield in the WessZumino gauge becomes in the case of minimal kinetic
terms
1 X j
A ej A j ,
v03 = 0,
v01 = v31 ,
v23 = v02 . (2.17)
v12 = D =
2
j
In Section 4, we shall present a simple way to find the correspondence between the choice
of boundary conditions and the choice of BPS equations (2.13) and (2.15) or (2.16) and
(2.17).
2.2. The exact solution of BPS domain wall junction
In a previous article [30], we have found an exact solution of BPS domain wall junction
in a model motivated by the N = 2 supersymmetric SU(2) gauge theory with one flavor
broken to N = 1 by the mass of the adjoint chiral superfield. This model has the following
chiral superfields with the charge assignment for the U (1) U (1)0 gauge group
f D D
e Q Q
e T
M M
(2.18)
U (1) 0
0
1 1 1 1 0
0
1 1 1 1 0
0
0
U (1)
interacting with a superpotential
f + (T + )DD
e + (T m)QQ
e h2 T ,
W = (T )MM
(2.19)
where parameters and h can be made real positive and a parameter m is complex [30].
In this model there are three discrete vacua,
237
Fig. 1. Boundary condition of the model in [30].
Vac.1 : T = ,
Vac.2 : T = m,
f = h, Q = Q
e= D = D
e = 0,
M=M
e = h, M = M
f= D = D
e = 0,
Q=Q
W1 = h2 ,
W2 = h2 m,
e = h, Q = Q
e= M = M
f = 0, W3 = h2 ,
(2.20)
Vac.3 : T = , D = D
and when m = i 3, this model becomes Z3 symmetric. Thus three half walls are
expected to connect at the junction with relative angles of 2/3. For definiteness, we
specify the boundary condition where the wall 1 extends along the negative x 2 axis
separating the vacuum 1 (x 1 > 0) and 3 (x 1 < 0) as shown in Fig. 1. If we have only
the wall 1, we obtain the central charge Zk (vanishing Yk ) and find the two conserved
supercharges from Eqs. (2.11) and (2.12) as
1
S ,
Q(1) = ei 4 Q2 + ei 4 Q
2
2
1
S .
Q(2) = ei 4 Q1 + ei 4 Q
1
2
(2.21)
The other two walls have also two conserved supercharges

1
S ,
at wall 2 Q(3) = ei 12 Q1 + ei 12 Q
1
2
1
S ,
besides Q(1) = ei 4 Q2 + ei 4 Q
2
2

5
5
1
S ,
at wall 3 Q(4) = ei 12 Q1 + ei 12 Q
1
2
1
S .
besides Q(1) = ei 4 Q2 + ei 4 Q
2
2
(2.22)
When these three half walls coexist,

we can have only one common conserved supercharge
S )/ 2. In fact we find that the domain wall junction configuration

Q(1) = (ei 4 Q2 + ei 4 Q
2
conserves precisely this single combination of supercharges, even though it has also
another central charge Yk contributing. Correspondingly we obtain the BPS equations
(2.16) and (2.17) for H = HII with = 1. The BPS equations (2.17) for the vector
238
superfield can be trivially satisfied by v = 0 and D = 0. The BPS equations (2.16) for
chiral superfields become in this case
2
W
Ai
= i ,
z
A
(2.23)
assuming the minimal kinetic term. The solution for these BPS equations is given by [30],
2 s
f
,
M(z, z ) = M(z, z ) =
s+t +u
2 t
e z ) =
,
D(z, z ) = D(z,
s +t +u
2 u
e z ) =
,
Q(z, z ) = Q(z,
s+t +u
i
2 ei 6 s + ei 6 t + ei 2 u
+ ,
T (z, z ) =
s+t +u
3
3
1

1
2
s = exp Re ei 6 z ,
3

1
2
u = exp Re ei 2 z .
3
(2.24)

5
2
t = exp Re ei 6 z ,
3
(2.25)
This model is motivated by the softly broken N = 2 SU(2) gauge theory with
one flavor. However, we can simplify the model without spoiling the solvability to
obtain a WessZumino model consisting of purely chiral superfields by the following
procedure. The vector superfields actually serve to constrain chiral superfields to have the
identical magnitude pairwise through D = 0 to satisfy the BPS equation (2.17) for vector
f = |M|, |D|
e = |D|, |Q|
e = |Q|. Therefore we can eliminate the vector
superfields: |M|
f = M,
superfields and reduce the number of chiral superfields by identifying pairwise M
e
e
D = D, Q = Q. Correspondingly we should take the superpotential as

1
1
h2
1
W = (T )M2 + (T + )D2 + T i 3 Q2 T .
2
2
2
2
(2.26)
This WessZumino model has the same solution as ours by changing h2 h2 /2,
3/2. A similar observation has also been made in Ref. [36].

2.3. Unitary representations of (1, 0) supersymmetry algebra
Let us examine states on the background of a domain wall junction from the point of
view of surviving symmetry. In the case of the BPS states satisfying the BPS equation
(2.16) corresponding to H = HII , we have only one surviving supersymmetry charge
Q(1) , two translation generators H , P 3 , and one Lorentz generator J 03 , out of the
N = 1 four-dimensional super Poincar generators. Since we are interested in excitation
modes on the background of the domain wall junction, we define the hamiltonian H 0 =
H hH i measured from the energy hH i of the background configuration. By projecting
239
from the supersymmetry algebra (2.1), (2.2) with central charges in four dimensions, we
immediately find
2
(2.27)
Q(1) = H 0 P 3 .
We also obtain the Poincar algebra in 1 + 1 dimensions
03 0
i

J ,H P3 = i H0 P3 ,
J 03 , Q(1) = Q(1) ,
2

03 0
J , H + P 3 = i H 0 + P 3 .
(2.28)
Other commutation relations are trivial

0
H P 3 , H 0 + P 3 = H 0 P 3 , Q(1) = H 0 + P 3 , Q(1) = 0.
(2.29)
This is precisely the (1, 0) supersymmetry algebra on the domain wall junction as
anticipated [25].
To obtain unitary representations, we can diagonalize H 0 and P 3

H 0 E, p3 = E E, p3 ,
(2.30)
P 3 E, p3 = p3 E, p3 , E > |p3 |,
and combine them by means of Q(1) . If E p3 > 0, we can construct bosonic state from
fermionic state and vice versa by operating Q(1) on the state:
|Bi = p
1
E p3
Q(1) |F i,
|F i = p
1
E p3
Q(1) |Bi.
(2.31)
Therefore we obtain a doublet representation (|Bi, |F i). If E p3 = 0, operating by Q(1)

on the state gives an unphysical zero norm state
(1)

Q E, p3 2 = E, p3 Q(1) 2 E, p3 = E, p3 H 0 P 3 E, p3 = E p3 = 0.
(2.32)
Then the massless right-moving state |E, p3 = Ei is a singlet representation. This singlet
state can either be boson or fermion. Thus we find that there are only two types of
representations of the (1, 0) supersymmetry algebra, doublet and singlet. We also find that
massive modes should appear in pairs of boson and fermion, whereas the massless rightmoving mode can appear singly without accompanying a state with opposite statistics. This
provides an interesting possibility of a chiral structure for fermions.
If another BPS equation (2.13) corresponding to H = HI is satisfied instead of Eq.
(2.16), we have (0, 1) supersymmetry and the left-moving massless states can appear as
singlets.
3. NambuGoldstone and other modes on the junction

3.1. Mode equation on the junction
Since the vector superfields have no nontrivial field configurations, NambuGoldstone
modes have no component of vector superfield. Moreover we can replace our model, if
240
we wish, by another model with purely chiral superfields without spoiling the essential
features including the solvability. Consequently we shall neglect vector superfields and
consider the general WessZumino model in Eq. (2.3) in the following. For simplicity we
assume the minimal kinetic term here Kij = ij .
Let us consider quantum fluctuations A0i , i around a classical solution Aicl which
satisfies the BPS equations (2.13) and (2.15) for H = HI or (2.16) and (2.17) for H = HII .
Ai = Aicl + A0i .
(3.33)
We retain the part of the Lagrangian quadratic in fluctuations and eliminate the auxiliary
fields F i to obtain the linearized equation for the scalar fluctuations
A0i +
2W
3W
2W
W 0j
0j
A
+
A = 0.
j
j
k
Aicl Akcl Ak
Aicl Akcl Acl Acl
cl Acl
(3.34)
In order to separate variables in x 0 , x 3 and x 1 , x 2 we have to define mode equations on

the background which has a nontrivial dependence in two dimensions, x 1 , x 2 . The bosonic
modes A0in (x 1 , x 2 ) can easily be defined in terms of a differential operator OB in x 1 , x 2
space

2W
3W
W
2W
OB i j
12 + 22 i j +
Aicl Akcl Ak Aj
cl
cl
3W
j Ak
k
Ai
cl
cl Acl Acl

OB i j
0j
An
0j
An

= Mn2
A0i
n
A0in
j
k
Aicl Akcl Acl Acl

2W
2W
12 + 22 i j +
i
k
Acl Acl Ak Aj
cl
cl
(3.35)

,
(3.36)
where the eigenvalue Mn2 has to be real from Majorana condition. The quantum fluctuation
for scalar can be expanded in terms of these mode functions to obtain a real scalar field
equation with the mass Mn for the coefficient bosonic field an (x 0 , x 3 )
X

an x 0 , x 3 A0in x 1 , x 2 ,
(3.37)
A0i x 0 , x 1 , x 2 , x 3 =
n
02
32
+ Mn2

an x 0 , x 3 = 0.
(3.38)
Similarly the linearized equation for fermions is given by

i i
i i
2W
j
Ai
cl Acl
2W
j
Aicl Acl
j = 0,
j = 0.
(3.39)
(3.40)
To separate variables for fermion equations, it is more convenient to use a gamma matrix
representation where direct product structure of 2 2 matrices for (x 0 , x 3 ) and (x 1 , x 2 )
space is manifest. We shall describe one such representation in Appendix B. Transforming
from such a representation to the Weyl representation which we are using, we can define
241
i ,
ni combining components of left-handed and right-handed
the fermionic modes n
spinors by means of the following operators
2W
i
i(
+
i
)
1
2 j
j
Ai
cl Acl
(3.41)
O1 i j
,
2W
i
i(1 + i2 )j
j
AiclAcl
2W
i
i(1 i2 )j
i
j
Acl Acl
i
(3.42)
O2 j
,
2W
i(1 i2 )ji
j
Ai
cl Acl
" i #
nj 1
n1
= im(1)
(3.43)
O1 i j
n
,
j
ni 2
n2
" #
j
i1
n1
i
(2) n
=
im
,
(3.44)
O2 j
n
j 2
i
n2
n
(2)
where the mass eigenvalues m(1)
n , mn are real. Please note a peculiar combination of leftand right-handed spinor components to define eigenfunctions. We can expand i in terms
of these mode functions

i (x 1 , x 2 )
X bn (x 0 , x 3 )n1
i
0 1 2 3
(3.45)
x , x , x , x =
i (x 1 , x 2 ) .
cn (x 0 , x 3 )n2
n
Since (x 0 , x 1 , x 2 , x 3 ) is a Majorana spinor, the coefficient fermionic fields bn , cn are

real. The linearized equations (3.39) (3.40) for the fermion gives a Dirac equation in
1 + 1 dimensions for the coefficient fermionic fields (cn , ibn ) with two mass parameters
(1)
(2)
m n , mn

1 + 3
cn (x 0 , x 3 )
(2) 1 3
m
= 0,
(3.46)
i 1 0 + i2 3 m(1)
n
n
ibn (x 0 , x 3 )
2
2
where we use Pauli matrices a , a = 1, 2, 3 to construct the 2 2 gamma matrices 1 , i2
in 1 + 1 dimensions. Since we have a Majorana spinor in 1 + 1 dimensions which does not
(1)
(2)
allow chiral rotations, we have two distinct real mass parameters mn , mn .
To relate the mass eigenvalues of fermions and bosons, let us multiply two differential
operators for fermions O2 to O1 . In this ordering, we can use the BPS equation (2.16)
corresponding to H = HII to find the differential operator for bosons OB
i/4 1/2

i/4 1/2

e
0
e
i
O2i k O1k j =
O
j
1/2
1/2 . (3.47)
B
0
ei/4
0
ei/4
Therefore the BPS equation (2.16) corresponding to H = HII guarantees that the existence
i of fermionic mode equations implies the existence of a solution of

of a solution ni 1 , n2
(1) (2)
bosonic mode equations with the mass squared Mn2 = mn mn
1/2
i/4
A0i
n =e
ni 1 ,
i
A0in = ei/4 n2
.
1/2
(3.48)
242
If another BPS equation (2.13) corresponding to H = HI is valid, operator multiplication

with different ordering gives the same bosonic operator whose rows and columns are
interchanged

1/2
1/2
0
ei/4 +
0
ei/4 +
i
k
i
OB j i/4 1/2
.
O1 k O2 j =
1/2
ei/4 +
0
e
+
0
(3.49)
Therefore the BPS equation (2.13) corresponding to H = HI guarantees that the existence
i
of fermionic mode equations implies the existence of a solution of
of a solution ni 2 , n1
(2)
bosonic mode equations with the mass squared Mn2 = m(1)
n mn
1/2
i/4
A0i
+
n = e
ni 2 ,
i
A0in = ei/4 + n1
.
1/2
(3.50)
Therefore we find that all massive states come in pairs of boson and fermion with
(1) (2)
the same mass squared Mn2 = mn mn in accordance with the result of the unitary
representation of the (1, 0) supersymmetry algebra.
3.2. NambuGoldstone modes
Since we are usually most interested in a low energy effective field theory, we wish to
study massless modes here. If global continuous symmetries are broken spontaneously,
there occur associated massless modes which are called the NambuGoldstone modes.
To find the wave functions of the NambuGoldstone modes, we perform the associated
global transformations and evaluate the transformed configuration by substituting the
classical field. For supersymmetry we obtain nontrivial wave function by substituting
the classical field Aicl (x 1 , x 2 ) and Fcli (x 1 , x 2 ) to the transformation of fermions by a
Grassmann parameter , since classical field configuration of fermion vanishes cli = 0
(3.51)
i = i 2 Aicl + 2 Fcli .
If the BPS equation (2.16) for the junction background is valid, we obtain

1 Aicl + i 2 + i
2 Aicl .
i = 2 i 1
(3.52)
We see that there is one conserved direction in the Grassmann parameter:
i 1 =
and 2 =
.
(3.53)
The other three real Grassmann parameters correspond to broken supercharges. For our
exact solution, for instance, we find it convenient to choose the three broken supercharges
as the following real supercharges

1
S ,
QI = ei/4 Q2 + ei/4 Q
2
2

1 i/4
S .
Q1 + ei/4 Q
QIII = e
1
2

1
S ,
QII = ei/4 Q1 + ei/4 Q
1
2
Then the corresponding massless mode functions are given by
(3.54)

(I)i
(II)i
0(III)i

4z Aicl x 1 , x 2 ei/4
,
0

0

x 1, x 2 =
,
21 Aicl x 1 , x 2 ei/4

0
1 2

.
x ,x =
22 Aicl x 1 , x 2 ei/4

x 1, x 2 =
243
(3.55)
(3.56)
(3.57)
Since the transformation parameter should correspond to the NambuGoldstone field with
zero momentum and energy, the three transformation parameters should be promoted to
(I)
(II)
(III)
three real fermionic fields in x 0 , x 3 space, b0 (x 0 , x 3 ), c0 (x 0 , x 3 ), and c0 (x 0 , x 3 ), to
obtain the NambuGoldstone component of the mode expansion

i x 0 , x 1 , x 2 , x 3 = b0(I) x 0 , x 3 0(I)i x 1 , x 2 + c0(II) x 0 , x 3 0(II)i x 1 , x 2
(III)i 1 2
(III)
x ,x
+ c0 x 0 , x 3 0
!
0
3
i
X bn (x , x )n1 (x 1 , x 2 )
.
(3.58)
+
0 3
i
1 2
n>0 cn (x , x )n2 (x , x )
We have explicitly displayed three massless NambuGoldstone fermion components
distinguishing from the massive ones (n > 0). The Dirac equation for the coefficient
fermionic fields (3.46) shows that b0(I) (x 0 x 3 ) is a right-moving massless mode, and
(II)
(III)
c0 (x 0 + x 3 ), and c0 (x 0 + x 3 ) are left-moving modes. We plot the absolute values of
(a)i=T
| of the i = T component of the wave function of the NambuGoldstone fermions
|0
a = I, II, III in Fig. 2. We can see that NambuGoldstone fermions have wave functions
which extend to infinity along three walls. They become identical to fermion zero modes on
at least two of the walls asymptotically and hence they are not localized around the center
of the junction. We can construct a linear combination of the NambuGoldstone fermions
to have no support along one out of the three walls. However, no linear combination of
these NambuGoldstone fermions can be formed which does not have support extended
along any of the wall.
Therefore these wave functions are not localized and are not normalizable. This fact
means that the low energy dynamics of BPS junction cannot be described by a (1 + 1)dimensional effective field theory with a discrete particle spectrum.
Similarly the NambuGoldstone bosons corresponding to the broken translation P a ,
a = 1, 2, are given by

(a)
(3.59)
A0 x 1 , x 2 = a Aicl x 1 , x 2 , a = 1, 2.
These two bosonic massless modes consist of two left-moving modes and two right-moving
modes. On the other hand, we have seen already that there are two left-moving massless
NambuGoldstone fermions and one right-moving massless NambuGoldstone fermion.
These two left-moving NambuGoldstone bosons and fermions form two doublets of the
(1, 0) supersymmetry algebra. The right-moving modes are asymmetric in bosons and
fermions: two NambuGoldstone bosons and a single NambuGoldstone fermion. These
three states are all singlets of the (1, 0) supersymmetry algebra in accordance with our
244

(I)T
The wave function 0

(II)T
The wave function 0

(III)T
The wave function 0

Fig. 2. The birds eye view of the absolute value of the i = T component of the wave functions of
the NambuGoldstone fermions on the junction in the (x 1 , x 2 ) space.
analysis in Section 2.3. Therefore we obtained a chiral structure of NambuGoldstone

fermions on the junction background configuration.
3.3. Non-normalizability of the NambuGoldstone fermions
We would like to argue that our observation is a generic feature of the NambuGoldstone
fermions on the domain wall junction in a flat space in the bulk: NambuGoldstone
fermions are not localized at the junction and hence are not normalizable, if they are
associated with the supersymmetry breaking due to the coexistence of nonparallel domain
walls. The following observation is behind this assertion. A single domain wall breaks only
a half of supercharges. Nonparallel wall also breaks half of supercharges, some of which
may be linear combinations of the supercharges already broken by the first wall. If the
junction configuration is a 1/4 BPS state, linearly independent ones among these two sets
of broken supercharges of nonparallel walls become 3/4 of the original supercharges.
245
To see in more detail, let us first note that the junction configuration reduces
asymptotically to a wall if one goes along the wall, say the wall 1. On this first wall,
a half of the original supersymmetry (Q(1) , . . . , Q(N) ) is broken. Denoting the number
of original supercharges to be N , we call these broken supercharges as Q(1) , . . . , Q(N/2) .
Consequently we have NambuGoldstone fermions localized around the core of the wall
and is constant along the wall. In the junction configuration, we have other walls which
are not parallel to the first wall. Asymptotically far away along one of such walls, say
wall 2, another half of the supersymmetry Q0(1) , . . . , Q0(N/2) is broken. If the junction
is a 1/4 BPS state, a half of these, say Q0(1) , . . . , Q0(N/4) , is a linear combination of
N
N
Q(1) , . . . , Q(N/2) broken already on the wall 1. The other half, Q0( 4 +1) , . . . , Q0( 2 ) are
unbroken on the wall 1. Altogether a quarter of the original supercharges remain unbroken.
Consequently the NambuGoldstone fermions corresponding to Q0(1) , . . . , Q0(N/4) have
a wave function which extends to infinity and approaches a constant profile along both
N
N
the walls 1 and 2. Those modes corresponding to Q0( 4 +1) , . . . , Q0( 2 ) have support only
along the wall 2, and those corresponding to the linear combinations of Q(1) , . . . , Q(N/2)
orthogonal to Q0(1) , . . . , Q0(N/4) have support only along the wall 1. Thus we find that any
linear combinations of the NambuGoldstone fermions have to be infinitely extended along
at least one of the walls which form the junction configuration. Therefore the Nambu
Goldstone fermions associated with the coexistence of nonparallel domain walls are not
localized at the junction and are not normalizable.
In our exact solution, domain wall junction configuration reduces asymptotically to the
wall 1 at x 2 with fixed x 1 . On the wall, only two supercharges in Eq. (3.54) are
broken

1
1
S ,
S ,
QII = ei/4 Q1 + ei/4 Q
(3.60)
QI = ei/4 Q2 + ei/4 Q
2
1
2
2
and there are two corresponding NambuGoldstone fermions which become domain wall
zero modes asymptotically

4z Aicl (x 1 , x 2 ) ei/4
(x 1 ) ei/4
21 Aiwall
(I)i 1 2
cl
,
0 x , x =
0
0

0
0
(3.61)
0(II)i x 1 , x 2 =
1 i/4 .
21 Aiwall
21 Aicl (x 1 , x 2 ) ei/4
cl (x ) e
These wave functions are localized on the core of the wall 1 in the x 1 direction and
are constant along the wall. Along the other walls we find two broken supercharges
one of which is identical to one of the broken supercharges, QI . The other broken
supercharge is Q0II on the wall 2 and Q00II on the wall 3. There are only two independent
supercharges among QII , Q0II , and Q00II . Together with QI we obtain three independent
broken supercharges. We can construct a linear combination of the NambuGoldstone
fermions to have no support along one out of the three walls. However, any linear
combination has nonvanishing wave function which becomes fermion zero mode on at
least one of the wall asymptotically. Therefore the associated NambuGoldstone fermions
have support which is infinitely extended at least along two of the walls.
246
If a single wall is present, we can explicitly construct a plane wave solution propagating
along the wall, which may be called a spin wave and is among massive modes on the wall
background. Even if there are several walls forming a junction configuration, we can consider excitation modes which reduce to the spin wave modes along each wall. They should
be a massive mode on the domain wall junction background. The NambuGoldstone mode
on the domain wall junction is the zero wave number limit of such a spin wave mode. This
physical consideration suggests that the massless NambuGoldstone fermion is precisely
the vanishing wave number (along the wall) limit of the massive spin wave mode.
Let us note that our argument does not apply to models with the bulk cosmological
constant. In such models, massless graviton is localized on the background of intersection
of walls [57]. In that case, massless mode is a distinct mode different from the massless
limit of the massive continuum, although the massless mode is buried at the tip of the
continuum of massive modes. The normalizability of the massless graviton is guaranteed
by the anti-de Sitter geometry away from the junction or intersection including the direction
along the wall.
4. Boundary conditions and central charges

For a 1/4 BPS state, there are two sets of BPS equations, Eqs. (2.13)(2.15) and (2.16)
(2.17), corresponding to the two kinds of BPS domain wall junctions. In this section we
make explicit the relation between the boundary conditions and the choice of these BPS
equations.
BPS domain wall junction is formed when nonparallel BPS walls meet at a junction.
In regions far away from the junction, the configuration approaches to isolated walls
asymptotically. BPS domain wall is a 1/2 BPS state and conserves two supercharges. These
two supercharges are given, from Eqs. (2.11) and (2.12), in terms of central charges Z1
and Z2 for the wall. Let us take a general WessZumino model in Eq. (2.3) and examine
if a domain wall junction can be formed where N different vacua appear in asymptotic
regions. These N vacua correspond to N points in the complex plane of superpotential W.
The field configuration of the junction at infinity is mapped to a straight line connecting
these N vertices [26,28,29]. In order to have a balance of force, this polygon has to be
convex [26]. We set the origin of the W space at an arbitrary point inside this BPS polygon
and denote the value of the superpotential at the I th vacuum as WI , for I = 1, . . . , N , as
illustrated in Fig. 3(b). Let us take the origin in x 1 , x 2 space as the junction point of these
BPS walls. If we denote I J the angle of the half wall separating two vacua, I and J , as
illustrated in Fig. 3(a), the central charges Z1 and Z2 of this wall are given by Eq. (2.4) as

E I J (Z1 , Z2 )I J = 2 WJ WI
(4.62)
E I J (Area),
Z

(4.63)
E I J cos(I J + /2), sin(I J + /2) .

Thus two supercharges conserved at this wall are
S ,
Q1 + ei(I J I J ) Q
1
where I J = arg(WJ WI ).
S ,
Q2 + ei(I J +I J ) Q
2
(4.64)
(a)
247
(b)
Fig. 3. Walls in x 1 , x 2 space and BPS polygon in W space. (a) Walls separating vacua I, J, K in
x 1 , x 2 space; (b) BPS polygon in W space.
BPS domain wall junction is a 1/4 BPS state and conserves only one supercharge. Let
S is conserved as
us consider the case of H = HII where a linear combination of Q2 and Q
2
shown in Eq. (2.12). This must be the common conserved supercharge for all the walls
S = Q2 + ei(J K +J K ) Q
S = .
= Q2 + ei(I J +I J ) Q
2
2
(4.65)
Then the relative angle of the two neighboring walls must be equal to the difference of two
phases of the differences 1W of the superpotentials for the two walls
. . . , J K I J = J K I J , . . . .
(4.66)
Moreover the field configuration at infinity should move counterclockwise in W space, as

we go around the origin counterclockwise in x 1 , x 2 space.
S is the common conserved supercharge in
Similarly, a linear combination of Q1 and Q
1
the case of H = HI . We obtain in this case
. . . , J K I J = (J K I J ), . . .
(4.67)
and that the field configuration at infinity should move clockwise in W space, as we go
around the origin counterclockwise in x 1 , x 2 space.
Therefore we find that the BPS equations (2.16)(2.17) for the case H = HII should
be used if the phase of the superpotential W increases as we go around the origin
counterclockwise in x 1 , x 2 space. If the phase of the superpotential W decreases as we
go around the origin counterclockwise in x 1 , x 2 space, the other BPS equations (2.13)
(2.15) for H = HI should be used.
Next we discuss the sign of the contribution of the central charge Y3 to the mass of
the junction configuration. We can use the Stokes theorem to obtain an expression for the
central charge Y3 as a contour integral [26,30]
Z
Z
I
Z

(4.68)
Y3 = dx 3 i d 2 x 1 Ki 2 Ai 2 Ki 1 Ai = dx 3 i Ki dAi ,
where Ki K/Ai is a derivative of the Khler potential K. This contour integral in
the field space should be done as a map from a counterclockwise contour in the infinity of
248
z = x 1 + ix 2 plane. Only complex fields can contribute to Y3 . Let us assume for simplicity
that there is only one field which can contribute to Y3 as in our exact solution.
Eq. (4.68) shows that the central charge Y3 becomes negative (positive), if the asymptotic
counterclockwise contour in x 1 , x 2 is mapped into a counterclockwise (clockwise) contour
in field space. On the other hand, the sign of the contribution of the central charge Y3 to
the mass of the junction configuration is determined by the formula H = HII = |hiZ1
Z2 i| + hY3 i, or H = HI = |hiZ1 Z2 i| hY3 i. The choice of these mass formulas are in
turn determined by the map of the asymptotic counterclockwise contour in x 1 , x 2 space to
a counterclockwise or clockwise contour in the superpotential space W. Combining these
two observations, we conclude that the contribution of the central charge Y3 to the mass of
the junction configuration is negative if the sign of rotations is the same in field space Ai
and in superpotential space W, and positive if the sign of rotations is opposite.
The field configuration moves counterclockwise in field space in our exact solution in
(2.24) and then the central charge is negative in this solution. Since the exact solution
satisfies the BPS equation for the case H = HII , the central charge contributes to the
mass of the junction configuration negatively. Therefore we should not consider the
central charge Y3 alone as the physical mass of the junction at the center. In the junction
configuration, the junction at the center cannot be separated from the walls. We also
can find a solution for the other case of H = HI in our model. The solution is just a
configuration obtained by a reflection x 1 x 1 . Then the central charge is positive, but
the contribution to the mass H = HI becomes again negative. In either solution, the rotation
in field T space has the same sign as the rotation in superpotential W space. Therefore
central charge Y3 contributes negatively to the mass of the junction, irrespective of the
choice of H = HI or H = HII .
More recently this feature of negative contribution of Y3 to the junction mass is studied
from a different viewpoint and it is argued that this feature is valid in most situations
except possibly in contrived models [36]. These models, if they exist, should correspond
to the case of opposite sign of rotations in W space and field space.
5. Energy density and central charges

5.1. Charge densities
Our exact solution is useful to examine how the topological charges Zk , Yk and energy
of the domain wall junction are distributed in x 1 , x 2 space. We shall study their densities
and integrated quantities in finite regions in this section.
5.1.1. Y charge density
The Y3 charge density Y3 = i 3nm n (T m T ) is given in our exact solution by
Y3 = 24
4
e
3 x 2
3 x 2
+2 cosh(x 1 )
3
249
Fig. 4. A birds eye view of Y3 .
= 244
1
e
2r
sin
+e
2r
sin(+ 2
3 )
+e
2r

sin( 2
3 ) 3
(5.69)
where the cylindrical coordinates r and is used to make Z3 symmetry explicit. A birds
eye view of the Y3 is given in Fig. 4. Here and the following, we shall take the unit of
1 in drawing figures. The density is localized near the origin and the Z3 symmetry is
manifest.
5.1.2. Z charge density
We obtain the superpotential as the function of x 1 and x 2 , by inserting the solution (2.24)

3 x 2 cosh(x 1 ) + cosh(2x 1 ) sinh(x 1 )
3 2 + 3e
,
Re W = 8
3

e 3 x 2 +2 cosh(x 1 )

3x 2 2 + e2 3x 2 +6 e 3 x 2 cosh(x 1 )
3
.
(5.70)
Im W = 2 3
3

e 3 x 2 +2 cosh(x 1 )
The Z charge densities are given by Zk = 2k W (k = 1, 2) and are found to be
2 + e2
Re Z1 = 484
3 x 2
3 x 2
cosh(2x 1 ) + 3 e
cosh(x 1 )
,

4
e 3 x 2 +2 cosh(x 1 )

2
2
4 e 3 x sinh(x 1 ) 1 + 2 e 3 x cosh(x 1 )
,
Im Z1 = Re Z2 = 48 3
4

e 3 x 2 +2 cosh(x 1 )

3 x 2 cosh(x 1 ) + e 3 x 2 2 + 3 cosh(2x 1 )
4e
.
(5.71)
Im Z2 = 48
4

e 3 x 2 +2 cosh(x 1 )
We can define the effective value of the Z charge which contributes to the energy of the
junction as Zeff = Re Z1 + Im Z2 . Corresponding effective charge density is given by
250
3 x 2
41 + 2e
Zeff = 96

2
cosh(x 1 ) + e2 3 x 1 + 2 cosh(2x 1 )
.
4

e 3 x 2 +2 cosh(x 1 )
(5.72)
Let us note that the effective charge density is Z3 symmetric, whereas individual charges
Z1 , Z2 are not.
5.1.3. Energy density
Adding Zeff and Y3 together, the energy density of the junction is obtained,

3 x 2 cosh(x 1 ) + e2 3 x 2 3 + 8 cosh(2x 1 )
44 + 6e
.
H = 24
4

e 3 x 2 +2 cosh(x 1 )
(5.73)
A birds eye view of H is shown in Fig. 5. The energy density is Z3 symmetric as expected.
A cross section of the densities, H, Zeff and Y3 along one of the walls (e.g., negative x 2
direction) is shown in Fig. 6. The Zeff charge contributes to the energy positively while Y3
does negatively. Since the decrease of Zeff is faster than the increase of Y3 , a small dip is
found around R 1. Far from the origin there is practically no difference between Zeff
and H because Y3 is localized near the origin.
5.2. Charge densities integrated over a region of finite radius
In this subsection we shall evaluate the central charge densities integrated over a
triangular or circular region depicted in Fig. 7.
5.2.1. Y3 charge
Integrating Y3 over a triangle whose inscribed circle has a radius R as shown in Fig. 7,
we obtain for large R (R 1)

2

3 3R
triangle
2 3R
e
(R) = 2 3 L3 1
+O e
,
(5.74)
Y3
4
Fig. 5. A birds eye view of the energy density of the junction.
251
Fig. 6. A cross section of the densities H (solid line), Zeff (short dashed line) and Y3 (long dashed
line) along the negative x 2 direction.
Fig. 7. The domains of integration: triangle (solid line), inscribed circle (short dashed line) and
circumscribed circle (long dashed line). The bold lines denote the domain walls forming a junction.
where L3 denotes the length along the x 3 direction. The leading term agrees with our
previous evaluation by the step function approximation [30] and the subleading terms
vanish exponentially as R .
On the other hand, for small R (R 1), we obtain the Y3 charge

8 2
4 3 5 5
triangle
2 2
4 4
6 6
R +O R .
(R) = L3 R R
(5.75)
Y3
45
3
252
Notice that the leading term comes from the density at the origin multiplied by the area
of the circle. Although there are no terms of the first or third degree in R, there is a fifth
degree term.
We can also integrate the Y3 over a circle of radius R. In this case it is more convenient
to use cylindrical coordinates (r, , and x 3 ), and the following surface integral formula
obtained from the Stokes theorem,
L
Z3 /2
Y3circle (R) =
Z2
dx
L3 /2
8R3 L3

0
0
d iT T
Z2
d
0
sin
2
3

2R cos
2
e2R cos + sin e2R cos + 3 + sin + 2
e
3
2
2R
2
2R
2R

sin(+
sin(
sin 3
3 )
3 )
e 3
+e 3
+e 3
2
3

,
(5.76)
where the Z3 symmetry is manifest. Expanding the integrand for small R (R 1), we
obtain

8
1
(5.77)
Y3circle (R) = 2 L3 R 2 2 R 4 4 + O R 6 6 .
9
2
The leading term is again the density at the origin multiplied by the area of the circle. In
contrast to the triangle case, there is no term of odd degree in R.
For large R (R 1), we have to perform numerical integration to evaluate the
circle
Y3circle (R). We compare the Y3 (R) evaluated for triangle, inscribed and circumscribed
in Fig. 8. In the limit of R , Y3 (R) for all the regions converge to 2 3 2 L3 as

expected.
5.2.2. Z charges
Since the Z1 charge is given by a total derivative in x 1 , we can rewrite the Z1 charge as
Z
Z
Z1 (R) = 2 dx 2 dx 1 1 W (A )
Zx 2+
h

i
dx 2 W x 1+ x 2 , x 2 W x 1 x 2 , x 2 ,
=2
(5.78)
x 2
where x i (i = 1, 2) denote the upper and lower bound of the domain of integration. Since
Eq. (5.70) shows that Im W (x 1 , x 2 ) is even and Re W (x 1 , x 2 ) is odd in x 1 , we obtain
Im Z1 (R) = 0 and Re Z2 (R) = 0 for an integration region symmetric in x 1 which we shall
use. Let us note that Re Z1 (Im Z2 ) is negative (positive) definite.
Firstly we choose as a domain of integration the triangle region whose inscribed circle
has a radius R. For large R (R 1), we obtain
253
Fig. 8. Y3 (R) evaluated for the triangle (solid line), inscribed circle of radius R (short dashed line)
and circumscribed circle (long dashed line).

3 3 R
e
+O e2 3 R ,
R +
4

4 3 R
3
triangle
e
(R) = 122 L3 R +
+O e2 3 R .
Im Z2
3 8
3
triangle
Re Z1
(R) = 122 L3
(5.79)
The leading linear term represents the contribution of charge density per unit length of
the wall. It is interesting to observe that there are no constant terms. The exponentially
suppressed terms represent the way the domain wall junction configuration converges to
isolated walls as R . The effective value of the Z charge becomes
triangle
Zeff
triangle
(R) = Re Z1
triangle
(R) + Im Z2
(R)

7 16
2 L3 e 3 R +O e2 3 R .
= 243 L3 R + 3
2
3
(5.80)
For small R (R 1), Z charges become

32 2
1 4 4 2 3 5 5
triangle
2 2
R + O R 6 6 ,
(R) = L3 R + R
Re Z1
2
9
3

32 2
1 4 4 2 3 5 5
triangle
2 2
6 6
R + O R . (5.81)
(R) = L3 R R +
Im Z2
2
9
3
Notice that the leading term represents the density at the origin multiplied by the area of
the triangle, and that there is no term of the first or third degree in R. The effective value
of Z charge is
triangle
Zeff

64
(R) = R 2 4 L3 + O R 6 6 .
3
(5.82)
We can also choose a circle of radius R as a domain of integration. For small R

(R 1) we obtain
254
Fig. 9. Zeff (R) evaluated for the triangle (solid line), the inscribed circle of the radius R (short dashed
line), and circumscribed circle (long dashed line).

32 2
1
L3 R 2 2 R 4 4 + O R 6 6 ,
9
4

32
1
2 L3 R 2 2 R 4 4 + O R 6 6 .
Im Z2circle (R) =
9
4
Re Z1circle (R) =
(5.83)
The leading term is again given by the densities at the origin multiplied by the area of the
circle. The effective value of the Z charge is

64 2
1 4 4
circle
2 2
6 6
L3 R R + O R .
(5.84)
Zeff (R) =
9
4
A numerical evaluation is needed for large R. We compare the effective Z value
evaluated for triangle, inscribed circle and circumscribed circle in Fig. 9. As R ,
ins.circle
(R) for inscribed circle converges to the same value as
the asymptotic slope of Zeff
that for the triangle. In the case of the circumscribed circle, the Zeff becomes twice as large
as those of the other cases for large R, since the total length of the walls is twice as long as
those of the other cases.
5.2.3. Energy of the junction
Since our exact solution satisfies the BPS equation corresponding to H = HII , the energy
of the junction is obtained by adding Y3 and Zeff together H = Zeff + Y3 = Re Z1 +
Im Z2 + Y3 .
Firstly we choose the triangle region whose inscribed circle has radius R. For large R
(R 1), the energy is

16
3
2 L3 e 3 R
Htriangle(R) = 24 L3 R 2 3 L3 + 3 5
3

2 3 R
+O e
.
(5.85)
255
Fig. 10. The energy of the junction configuration H (R) evaluated for the triangle (solid), inscribed
circle of radius R (short dashed line) and circumscribed circle (long dashed line).
The first linear term can be regarded as the contribution from the walls and the second
constant term can be regarded as the contribution from the junction at the center. For small
R (R 1), the energy is

56 2
1 4 4 4 3 5 5
2 2
6 6
R +O R .
(5.86)
Htriangle(R) = L3 R + R +
7
315
3
In the case of the circle of radius R, the energy is given for small R (R 1) as

56 2
3 4 4
2 2
6 6
L3 R R + O R .
(5.87)
Hcircle (R) =
9
14
The energy of the triangle region is compared to those of inscribed and circumscribed
circles in Fig. 10. For large R (R 1), the energy H reduces to Zeff .
Finally we plot the -dependence of the energy and charges that are obtained by
integrating the densities from r = 0 to r = R with fixed (see Fig. 11). In each figure
the energy H (solid line) is the sum of the Zeff (short dashed line) and Y3 (long dashed
line) and Z3 symmetry is manifest in their -dependence. In Fig. 11(a), all the quantities
are almost uniform in near the junction at the center, reflecting the fact that the junction
is a string-like object and symmetric around the x 3 axis. As we move away from the origin,
main contribution comes from the direction of the walls (in our case /2, /6, and 5/6)
(see Fig. 11(b) and (c)). As R grows, Y3 disappears and the energy H approaches Zeff (see
Fig. 11(d)).
Acknowledgements
One of the authors (H.O.) gratefully acknowledges support from the Iwanami Fujukai
Foundation. This work is supported in part by Grant-in-Aid for Scientific Research from
the Japan Ministry of Education, Science and Culture for the Priority Area 291 and 707.
256
(a)
(b)
(c)
(d)
Fig. 11. The -dependence of the energy and charges that are obtained by integrating the densities
over the radial direction up to R with fixed. In each figure horizontal axis denotes , and solid,
short dashed and long dashed lines correspond to H , Zeff and Y3 , respectively. (a) R = 0.5 case;
(b) R = 1 case; (c) R = 2 case; (d) R = 3 case.
Appendix A. Fermionic contributions to central charges

We shall derive the central charges including fermionic contributions in the case of a
general WessZumino model with an arbitrary superpotential W. For simplicity, Khler
metric is assumed to be minimal Kij = ij :
i
i
L = Aj Aj + F j F j + j j j j
2
2
1 i j W
1 i j W
W
j W
+Fj
+
F
.
(A.1)

Aj
2
Ai Aj
Aj
2
Ai Aj
We have added a surface term to Eq. (2.3) to make the variational principle meaningful.
This is the starting Lagrangian to derive central charges and we will not neglect any total
divergences from now on. The canonical supercurrent is found to be

J = 2 i Ai + i 2 i F i

W
= 2 i Ai i 2 i
,
(A.2)
Ai

J = 2 i Ai + i 2 i F i

W
= 2 i Ai i 2 i
.
(A.3)
Ai
The canonical energy momentum tensor is given by
i
i
T = Aj Aj + Aj Aj + j j + j j
2
2

2

W
i
i
+ g Aj Aj j + j j j j
A
2
2

1 i j W
1 i j W

.
2
Ai Aj
2
Ai Aj
Canonical quantization gives (anti-) commutation relations

i
A (x), 0 Aj (y) x 0 =y 0 = i 3 (x y) ij ,
o
n
j
i (x), (y) 0 0 = 3 (x y) ij 0 .
x =y
257
(A.4)
(A.5)
The anticommutator between supercharges of the same chirality gives the supersymmetry
algebra (2.1) with the central charge Zk in Eq. (2.4)
Z
(A.6)
Zk = 2 d 3 x k W (A )
which turns out to have only bosonic contributions. The anticommutator between
supercharges of the opposite chirality gives the supersymmetry algebra (2.2) with the
central charge Yk

Z
1 j
knm
3
j
j
j
Yk = i
(A.7)
d x n A m A m , 123 = 1,
2
which has both bosonic and fermionic contributions.
Appendix B. Gamma matrices and fermion mode equations

In order to separate variables (x 1 , x 2 ) and (x 0 , x 3 ) for spinors, it is most convenient to
use a gamma matrix representation where the direct product structure of 2 2 matrices in
(x 1 , x 2 ) and (x 0 , x 3 ) becomes manifest. One such representation is
0 = 1,
3 = i 2 ,
1 = i 1 3 ,
2 = i 2 3 ,
(B.1)
are Pauli matrices acting on 2 2 matrices and

acting on indices of blocks
where
of these 2 2 matrices. The four component spinor can be decomposed into a pair of two
component spinors and in 0 + 2 dimensions

=
.
(B.2)
i
a
The B matrix for 1 + 3 dimensions can be defined as a product of B matrices B (1) for 1 + 1
dimensions and B (2) for 0 + 2 dimensions
B = B (1) B (2) ,
B (1) = 3 ,
B (2) = i 1 .
(B.3)
The Majorana condition for the (1 + 3)-dimensional spinor and the pseudo-Majorana
condition for the (0 + 2)-dimensional spinor are given by
= B ,
= B (2) ,
= B (2) ,
(B.4)
258
which implies 1 = i2 for components of the two component spinor = (1 , 2 )T , and

similarly for .
The Dirac equation for four component fermions in the general WessZumino model
reads
2

2 W 1 i 5
W 1 + i 5
i
+
(B.5)
j = 0.
i
Ai Aj
2
Ai Aj
2
j
The mode equation in x 1 , x 2 space is defined in terms of two component spinors n and
j
n

2W 1 + 3 j
2W 1 3
i
n = m(1)
ij 1 1 + 2 2
n n ,
Ai Aj 2
Ai Aj 2

2W 1 3 j
2W 1 + 3
i
n = m(2)
ij 1 1 + 2 2
n n .
Ai Aj 2
Ai Aj
2
(B.6)
The two component spinors ni , ni satisfy the pseudo-Majorana condition (B.4). Then we
find that mass eigenvalues are real

(1)
(2)
,
m(2)
.
(B.7)
m(1)
n = mn
n = mn
We can make a separation of variables for the Dirac equation (B.5) by means of real
fermionic fields cn and bn
X cn (x 0 , x 3 )n (x 1 , x 2 )
.
(B.8)
i =
ibn (x 0 , x 3 )n (x 1 , x 2 )
n
Using these mode functions we find that the fermion fields (cn , ibn )T satisfy the Dirac
equation in 1 + 1 dimensions in Eq. (3.46).
This representation can be related to the usual Weyl representation in Ref. [33] by the
following unitary matrix U
1 3 3 1 + 3 2
+
,
2
2
Weyl = U U,
U=
BWeyl = U BU = 2 2 .
(B.9)
The two component spinor in the Weyl representation is related to the components of the
four component spinor (B.2) in the representation (B.1) in this appendix as
1

2
.
(B.10)
=
Weyl i1
i2
Thus we obtain the mode equation (3.41)(3.44) in the Weyl representation.
259
References
[1] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315.
[2] I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G. Dvali, Phys. Lett. B 436 (1998) 257,
hep-ph/9804398.
[3] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370, hep-ph/9905221.
[4] L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
[5] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, N. Kaloper, Phys. Rev. Lett. 84 (2000) 586, hepth/9907209.
[6] C. Csaki, Y. Shirman, Phys. Rev. D 61 (2000) 024008, hep-th/9908186.
[7] A.E. Nelson, A new angle on intersecting branes in infinite extra dimensions, hep-th/9909001.
[8] K. Behrndt, M. Cvetic, Supersymmetric domain wall world from D = 5 simple gauged
supergravity, hep-th/9909058.
[9] K. Skenderis, P. Townsend, Phys. Lett. B 468 (1999) 46, hep-th/9909070.
[10] A. Chamblin, G.W. Gibbons, Supergravity on the Brane , hep-th/9909130.
[11] O. DeWolfe, D.Z. Freedman, S.S. Gubser, A. Karch, Modeling the fifth dimension with scalars
and gravity, hep-th/9909134.
[12] E. Bogomolnyi, Sov. J. Nucl. Phys. 24 (1976) 449.
[13] M.K. Prasad, C.H. Sommerfield, Phys. Rev. Lett. 35 (1975) 760.
[14] E. Witten, D. Olive, Phys. Lett. B 78 (1978) 97.
[15] E.R.C. Abraham, P.K. Townsend, Nucl. Phys. B 351 (1991) 313.
[16] G. Dvali, M. Shifman, Phys. Lett. B 396 (1997) 64, hep-th/9612128.
[17] G. Dvali, M. Shifman, Nucl. Phys. B 504 (1997) 127, hep-th/9611213.
[18] A. Kovner, M. Shifman, A. Smilga, Phys. Rev. D 56 (1997) 7978, hep-th/9706089.
[19] A. Smilga, A. Veselov, Phys. Rev. Lett. 79 (1997) 4529, hep-th/9706217.
[20] B. de Carlos, J.M. Moreno, Phys. Rev. Lett. 83 (1999) 2120, hep-th/9905165.
[21] G. Dvali, G. Gabadadze, Z. Kakushadze, Nucl. Phys. B 562 (1999) 158, hep-th/990103.
[22] V. Kaplunovsky, J. Sonnenschein, S. Yankielowicz, Nucl. Phys. B 552 (1999) 209, hepth/9811195.
[23] V.A. Rubakov, M.E. Shaposhnikov, Phys. Lett. B 125 (1983) 136.
[24] B. Chibisov, M. Shifman, Phys. Rev. D 56 (1997) 7990, hep-th/9706141.
[25] G. Gibbons, P.K. Townsend, Phys. Rev. Lett. 83 (1999) 1727, hep-th/9905196.
[26] S.M. Carroll, S. Hellerman, M. Trodden, Phys. Rev. D 61 (2000) 065001, hep-th/9905217.
[27] A. Gorsky, M. Shifman, Phys. Rev. D 61 (2000) 085001, hep-th/9909015.
[28] P.M. Saffin, Phys. Rev. Lett. 83 (1999) 4249, hep-th/9907066.
[29] D. Bazeia, F.A. Brito, Phys. Rev. Lett. 84 (2000) 1094, hep-th/9908090.
[30] H. Oda, K. Ito, M. Naganuma, N. Sakai, Phys. Lett. B 471 (1999) 148, hep-th/9910095.
[31] P. Townsend, in: PASCOS/Hopkins, 1995, p. 0271, hep-th/9507048.
[32] P. Townsend, Class. Quant. Grav. 17 (2000) 1267, hep-th/9911154.
[33] J. Wess, J. Bagger, Supersymmetry and Supergravity, Princeton University Press, 1991.
[34] D. Binosi, T. tel Veldhuis, Phys. Lett. B 476 (2000) 124, hep-th/9912081.
[35] D. Bazeia, F.A. Brito, Bags, junctions, and networks of BPS and non-BPS defects, hepth/9912015.
[36] M. Shifman, T. tel Veldhuis, Calculating the tension of domain wall junctions and vortices in
generalized WessZumino models, hep-th/9912162.
[37] S.M. Carroll, S. Hellerman, M. Trodden, BPS domain wall junctions in large extra dimensions,
hep-th/9911083.
[38] N. Arkani-Hamed, S. Dimopoulos, G. Dvali, N. Kaloper, Manyfold universe, hep-ph/9911386.
[39] S. Nam, JHEP 0003 (2000) 005, hep-th/9911104.
[40] I. Bakas, A. Brandhuber, K. Sfetsos, Domain walls of gauged supergravity, M-branes and
algebraic curves, hep-th/9912132.
260
[41] M. Cvetic, H. L, C.N. Pope, Domain walls and massive gauged supergravity potentials, hepth/0001002.
[42] C. Csaki, J. Erlich, T.J. Hollowood, Y. Shirman, Universal aspects of gravity localized on thick
branes, hep-th/0001033.
[43] R. Kallosh, A. Linde, JHEP 0002 (2000) 005, hep-th/0001071.
[44] S. Nam, K. Olsen, Domain wall junctions in supersymmetric field theories in D = 4, hepth/0002176.
[45] R. Altendorfer, J. Bagger, D. Nemeschansky, Supersymmetric RandallSundrum scenario,
hep-th/0003117.
[46] N. Alonso-Alberca, P. Meessen, T. Ortin, Supersymmetric brane-worlds, hep-th/0003248.
[47] J.P. Gauntlett, G.W. Gibbons, P.K. Townsend, Intersecting domain walls in MQCD, hepth/0004136.

Models of dynamical supersymmetry breaking with

gauged U (1)R symmetry
Noriaki Kitazawa a,1 , Nobuhito Maru b, , Nobuchika Okada c,2
a Department of Physics, Tokyo Metropolitan University, Hachioji, Tokyo 192-0397, Japan
b Department of Physics, Tokyo Institute of Technology, Oh-Okayama, Meguro, Tokyo 152-8551, Japan
c Theory Group, KEK, Tsukuba, Ibaraki 305-0801, Japan
Received 5 April 2000; accepted 26 June 2000
Abstract
We present simple models of dynamical supersymmetry breaking with gauged U (1)R symmetry.
The minimal supersymmetric standard model and supersymmetric SU(5) GUT are considered as
the visible sector. The anomaly cancellation conditions for U (1)R are investigated in detail and
simple solutions of the R-charge assignments are found. We show that this scenario of dynamical
supersymmetry breaking is phenomenologically viable with the gravitino mass of order 1 TeV or 10
TeV. 2000 Elsevier Science B.V. All rights reserved.
PACS: 12.60.Jv; 11.30.Na; 12.60.Cn
Keywords: Supergravity; Gauged U (1)R ; Dynamical supersymmetry breaking
1. Introduction
Supersymmety is motivated to solve the gauge hierarchy problem. However, since
the superparticles have not been observed yet, supersymmetry should be broken at low
energies. Spontaneous supersymmetry breaking at the tree-level [1,2] does not explain
why the discrepancy between the supersymmetry breaking scale and the Planck scale
is so large. On the other hand, in the models of dynamical supersymmetry breaking
the supersymmetry breaking scale is related to the Planck scale via the dimensional
transmutation [3]. In the light of this fact, many people have constructed the models of
the dynamical supersymmetry breaking [4,5] and discussed the phenomenology so far [6,
7]. As for the mediation mechanism of supersymmetry breaking to the visible sector, there
Corresponding author. E-mail: maru@th.phys.titech.ac.jp; JSPS Research Fellow; Department of Physics,
University of Tokyo, Hongo, Bunkyo-ku, Tokyo 113-0033, Japan.

1 kitazawa@phys.metro-u.ac.jp
2 okadan@camry.kek.jp; JSPS Research Fellow.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 1 5 - 6
262
N. Kitazawa et al. / Nuclear Physics B 586 (2000) 261274
are mainly two ways. One is the gravity mediation [6] (including the anomaly mediation [8,
9]), the other is the gauge mediation [7] (including the anomalous U (1) mediation [10,11]).
Many authors have extensively studied both scenarios and proposed interesting models.
In our previous letter [12], we proposed a simple mechanism of the dynamical
supersymmetry breaking with gauged U (1)R symmetry. Although the attempts to make
use of the gauged U (1)R symmetry in supergravity theories can be found in Refs. [13,
14], the mechanism of supersymmetry breaking was not the main topics in these works.
Supersymmetry can be dynamically broken by the interplay between the FayetIliopoulos
term of U (1)R and the dynamically generated superpotential due to the non-perturbative
dynamics of the gauge theory with vanishing cosmological constant. However, we did not
discuss in detail the anomaly cancellation for U (1)R taking into account the full particle
contents, since we focused on the dynamics of supersymmetry breaking in the hidden
sector. The main purpose of this paper is to pursuit this point. We discuss two models as
the visible sector: one is the minimal supersymmetric standard model (MSSM), the other
is the supersymmetric SU(5) grand unified theory (GUT). In the model of MSSM, we use
the GreenSchwarz mechanism for anomaly cancellation. We find quite simple solutions
of R-charge assignments in both cases. The spectrum of the supersymmetry breaking mass
is also discussed. The scalar fields receive the soft supersymmetry breaking masses, which
is the same order of the gravitino mass from the tree-level interactions of supergravity.
Moreover, the scalar fields with non-zero R-charges obtain additional soft supersymmetry
breaking masses through U (1)R D-term, which is also the same order of the gravitino mass.
The masses of gauginos in the MSSM depend on the form of gauge kinetic functions. If
the gauge kinetic function includes the higher-dimensional term, the gaugino masses are
generated in the same order of the gravitino mass or less. If the gauge kinetic function is
trivial, the gaugino masses are generated through the anomaly mediation in a few orders
smaller than the gravitino mass. We see that our scenario is phenomenologically viable
with the gravitino mass of order 1 TeV or 10 TeV. The comments on the generation of the
Higgs potential from higher-dimensional interactions are given.
This paper is organized as follows. In Section 2, we review our mechanism of dynamical
supersymmetry breaking with gauged U (1)R symmetry. Then this mechanism is applied to
MSSM of the visible sector in Section 3 and applied to the supersymmetric SU(5) GUT of
the visible sector in Section 4. The last section is devoted to the summary and discussion.
2. Dynamical supersymmetry breaking with gauged U (1)R symmetry

In this section, we review our previous letter, in which we proposed a simple mechanism
of dynamical supersymmetry breaking with gauged U (1)R symmetry in the context of the
minimal supergravity. The model is based on the gauge group SU (2)H U (1)R with the
following matter contents 3 :
3 In our previous letter [12], we did not discuss the cancellation of the gauge anomaly of [U (1) ]3 and the
R
mixed gravitational anomaly of U (1)R . We simply assumed there that these anomalies are cancelled out, if all
Q1
Q2
S
SU(2)H
2
2
1
263
U (1)R
1
1
+4
The general renormalizable superpotential at the tree-level is

W = S[Q1 Q2 ],
(1)
where the square brackets denote the contraction of SU(2) indices by the -tensor, is a
dimensionless coupling constant. We assume that is real and positive.
It is known that the superpotential is generated dynamically by non-perturbative
(instanton) effect of the SU(2) gauge dynamics [15]. The total effective superpotential
is found to be
Weff = S [Q1 Q2 ] +
5
,
[Q1 Q2 ]
(2)
where the second term is the dynamically generated superpotential and is the dynamical
scale of the SU(2)H gauge interaction. Note that the supersymmetric vacuum lies at hSi
and hQ1 i, hQ2 i 0, if only the F-term potential is considered.
Next, let us consider the D-term potential. The gauged U (1)R symmetry is impossible
in the globally supersymmetric theory, since the generators of the U (1)R symmetry and
supersymmetry do not commute with each other. On the other hand, in the supergravity
theory the U (1)R symmetry can be gauged as if it were a usual global symmetry [13,14,
16,17]. However, it should be noticed that the FayetIliopoulos term of the gauged U (1)R
symmetry appears due to the symmetry of supergravity. This fact is easily understood by
the standard formula for supergravity theories [18]. Using the generalized Khler potential
P
G = K + ln |W |2 , we have D = i qi (G/zi )zi , where qi is the U (1)R charge of the
field zi . Note that the contribution from the superpotential leads to the constant term, since
the superpotential is holomorphic and has U (1)R charge 2.
With the above particle contents, the D-term potential is found to be
2
gR2
4S S Q1 Q1 Q2 Q2 + 2MP2 ,
(3)
2
where MP = Mpl / 8 is the reduced Planck mass, gR is the U (1)R gauge coupling, and
the minimal Khler potential, K = S S + Q1 Q1 + Q2 Q2 , is assumed. 4 Note that the
supersymmetric vacuum conditions required by the D-term potential and by the effective
superpotential of Eq. (2) are incompatible. Therefore, supersymmetry is broken. This
consequence remains correct, if there is no other superfields which have negative U (1)R
charges, or if the other negatively charged superfields (if they exist) have no vacuum
expectation value.
VD =
particle contents are considered with an appropriate U (1)R charge assignment. In this paper, we will discuss this
issue in detail.
4 This assumption is justified by our result with M which means that the SU(2) gauge interaction is
P
H
weak at the Planck scale.
264
Let us analyze the total potential in our model. Here, note that the cosmological constant
should vanish. This requirement comes not only from the observations of the present
universe but also from the consistency of our discussion. Since it is not clear whether
the superpotential discussed above can be dynamically generated even in the curved space,
the spacetime should be flat for our discussion to be correct. Note that we cannot take the
usual strategy, namely, adding a constant term to the superpotential, since such a term is
forbidden by the U (1)R gauge symmetry. Therefore, it is a non-trivial problem whether we
can obtain the vanishing cosmological constant in our model.
Assuming that the potential minimum lies on the D-flat direction of the SU(2)H gauge
interaction, we take the vacuum expectation values such that hSi = s and hQi i = vi ,
where i and denote the flavor and SU(2)H indices, respectively. We can always make s
and v real and positive by symmetry transformations. The total potential is given by
2

5
K
2
2
2
2
V (v, s) = e (v + sW ) + 2v s 4 + W 3W
v
gR2
(4s 2 2v 2 + 2)2 ,
(4)
2
where K and W are the Khler potential and superpotential, respectively, which are given
by
+
(5)
K = s 2 + 2v 2 ,
5
(6)
W = sv 2 + 2 .
v
Here, all dimensionful parameters are taken to be dimensionless with the normalization
MP = 1. The first line in Eq. (4) comes from the F-term (except for W 2 term) and the
remainder is the D-term potential.
Since the potential is very complicated, it is convenient to make some assumptions
for the values of parameters. First, assume that gR , . Since the D-term potential
is proportional to gR2 and positive definite, the potential minimum is expected for VD to be
small as possible. If we assume s 1 and v 1, the potential can be rewritten as

(7)
V e2 2 310 .
5
It is found that 3 is required in order to obtain the vanishing cosmological
constant.
Let us consider the stationary conditions of the potential. Using the assumptions s 1
and v = 1 + y (|y| 1), the stationary conditions can be expanded with respect to s and
y. Considering the relations gR 5 , we can expand the condition V /y = 0 and
obtain
y s2
e2 2
.
2gR2
(8)
Using this result, we can also expand the expansion of the condition V /s = 0 and obtain
s
5
.
10
82
(9)
265
By the numerical analysis, the above rough estimation is found to be a good approximation.
The result of numerical calculations is the following:
y 4.7 103 ,
s 6.8 10
(10)
(11)
Here, we used the values of = 103 , 1.85 and gR = 1011 . For these values
of the parameters, we can obtain the vanishing cosmological constant. Note that the
numerical values of Eqs. (10) and (11) are almost independent of the actual value of ,
if the condition gR 5 is satisfied and the ratio /5 is fixed. This can be seen in the
approximate formulae of Eqs. (8) and (9). We can choose the value of in order to obtain
a phenomenologically acceptable mass spectrum.
As a result of the above analysis, we can estimate the gravitino mass as

5
m3/2 = eK/2 W 3.0 4 .
MP
(12)
This non-zero gravitino mass means that supersymmetry is really broken in the framework
of the supergravity. The gravitino mass contributes to the masses of scalar partners via
the tree-level interactions of supergravity. Note that there is another contribution, if scalar
partners have non-zero U (1)R charges. In this case, they also acquire the mass from the
vacuum expectation value of the D-term, and it is estimated as
m2D-term
= gR2 hDiq

5 2
7.3 4 q,
MP
(13)
where q is the U (1)R charge. This mass squared is always positive for the scalar fields
with positive U (1)R charges. The mass is the same order of the magnitude of the gravitino
mass. This is because gR is canceled out in the above estimation (see Eq. (8)). Note that
m3/2 and m2D-term is controlled by the strong coupling scale of SU(2)H gauge theory .
We give some comments. Our model has the same structure of the supersymmetry
breaking model with the anomalous U (1) gauge symmetry [10]. In the model, the Fayet
Iliopoulos term is originated from the anomaly of the U (1) gauge symmetry [1921]. On
the other hand, in our model the origin of the term is the symmetry of supergravity with
the gauged U (1)R symmetry. The FayetIliopoulos term appears even if the U (1)R gauge
interaction is anomaly free.
The mediation of supersymmetry breaking to the visible sector is discussed in
succeeding sections. We address the highly non-trivial problem whether the anomaly
cancellation of U (1)R can be done when we include the visible sector superfields with
positive semi-definite U (1)R charges [13,14]. We discuss two explicit models. One is the
model that the U (1)R anomalies are cancelled out by the GreenSchwarz mechanism. We
refer this model to the anomalous U (1)R model. The other is the model that the U (1)R
symmetry is anomaly free. We refer this model to the anomaly free U (1)R model.
266
3. Anomalous U (1)R model

Before discussing the model in detail, it is instructive to review the GreenSchwarz
mechanism in four dimensions [22]. Suppose that we have a gauge symmetry U (1)X
with mixed gauge anomalies of U (1)X G2i , where Gi denote other gauge groups. The
Lagrangian is not invariant under the U (1)X gauge transformation A A + (x):
Lgauge = (x)

Ci
ei ,
tr Fi F
2
8
(14)
ei ) are the field strength tensors (its dual) of Gi .

where Ci are anomaly coefficients and Fi (F
On the other hand, there could be another anomaly at the Planck scale through the dilaton
superfield S with the U (1)X transformation S S + 2i GS (x):

GS 1
ei ,
ki tr Fi F
(15)
2
2
is the GreenSchwarz coefficient which is calculated in the string theory [19]
LGS = (x)
where GS
as
1
tr qi .
(16)
192 2
Here, qi are R-charges of the fermionic components of the superfields. ki are the Kac
Moody levels of Gi . Note that the KacMoody levels are integers for non-Abelian gauge
groups but are not necessary integers for Abelian groups. Therefore, we can assume the
value of the KacMoody levels for Abelian groups appropriately. All the KacMoody
levels must have the same sign, since all the gauge couplings are generated by
GS =
1
= ki hSi,
gi2
(17)
where hSi is the vacuum expectation value of the dilaton superfield. From Eqs. (14) and
(15), the anomaly cancellation conditions lead to the GreenSchwarz relations:
Ci
= 2 2 GS .
ki
(18)
This relation should be satisfied for any i. It follows from this relation that all of Ci has the
same sign, since all of ki has the same sign.
Now, we discuss the model in detail. We consider the MSSM with all the Yukawa
couplings but the -term. The relevant anomaly coefficients are listed below:
1
CH = (q1 + q2 ) + 2,
2
3
C3 = (2q + u + d) + 3,
2

1
3
C2 = (3q + l) + h + h + 2,
2
2

4
1
1
1
1
CY = 3 q + u + d + l + e + h + h ,
6
3
3
2
2
(19)
(20)
(21)
(22)

CY Y = 3 q 2 2u2 + d 2 l 2 + e2 + h2 h 2 = 0,

Cg = 2(q1 + q2 ) + s + 3(6q + 3u + 3d + 2l + n + e) + 2 h + h 6,

CR = 2(q13 + q23 ) + s 3 + 3 6q 3 + 3u3 + 3d 3 + 2l 3 + n3 + e3

+ 2 h3 + h 3 + 18,
267
(23)
(24)
(25)
where q1 , q2 and s are R-charges of the fermionic components for Q1 , Q2 and S in the
hidden sector, and q, u, d, l, n, e, h, and h are those for the chiral superfields Q(3, 2, 16 ),
SS
S(1, 2, 1 ),
S(S
3, 1, 13 ), L(1, 2, 12 ), N(1, 1, 0) E(1, 1, 1), H (1, 2, 12 ), and H
U
3, 1, 23 ), D(
2
respectively. The representations and charges in the parenthesis are those under SU(3)C
SU(2)L U (1)Y . The coefficients CH , C3 , C2 , CY , CY Y , Cg and CR represent the anomaly
coefficients for U (1)R (SU(2)H )2 , U (1)R (SU(3)C )2 , U (1)R (SU(2)L )2 , U (1)R (U (1)Y )2 ,
(U (1)R )2 U (1)Y , U (1)R and (U (1)R )3 , respectively. Here we simply assume a family
independent charge assignment. Note that the gravitino contribution for the mixed
gravitational anomaly is 21 times that of a gaugino, while the gravitino contribution
for the U (1)R gauge anomaly is three times that of a gaugino [2326], and the dilatino
contributes (1) to both Cg and CR . Note also that the anomaly coefficient CY Y has to
vanish identically, since it cannot be cancelled by the GreenSchwarz mechanism.
As mentioned earlier, all the Yukawa coupling are included:
SH + yd QD
SH
S + ye LE H
S + yn LNH.
W = yu QU
(26)
The resulting conditions for R-charges are

q + u + h = 1,
q + d + h = 1,
(27)
l + e + h = 1,
(29)
l + n + h = 1.
(30)
(28)
Furthermore, we need the Yukawa coupling in the hidden sector

W = S[Q1 Q2 ],
(31)
which leads to the condition

s + q1 + q2 = 1.
(32)
One can immediately see that C3 = C2 = CY = 0 and Eqs. (27), (28) and (29) are
incompatible, since CY + C2 2C3 = 6. Therefore we have to use the GreenSchwarz
mechanism to cancel the anomalies.
Taking into account the simple assumption k2 = k3 , we have only two solutions. One
solution consists of all integer charges:
S = 6,
Q1 = Q2 = 2,
S
S
Q = U = D = L = N = E = 0,
(33)
S = 2,
H =H
(34)
where these are the charges of superfields. The corresponding anomaly coefficients are
268
C3 = C2 = 3,
CH = 1,
CY = 9,
CR = 9,
CY Y = 0,
Cg = 57.
(35)
(36)
Note that all the non-trivial anomaly coefficients have negative signs. Therefore, we assume
that all the KacMoody levels are positive with negative vacuum expectation value of the
dilaton superfield. We can freely choose the positive values of kR and kg to satisfy the
GreenSchwarz relation [22]. On the other KacMoody levels the following discussions
are required.
The GreenSchwarz relation
2 2 Cg
Cg
C2 C3 CH
CY
=
=
=
= 2 2 GS =
=
kY
k2
k3
kH
96
192 2
(37)
is satisfied, if we introduce 39 gauge singlets of vanishing R-charge with kY = 9, k2 = k3 =

3 and kH = 1. These values of KacMoody levels tell us the gauge coupling relations at
the Planck scale, namely,
3 = 2 = 3Y ,
H = 33 .
(38)
This coupling unification can be easily accomplished by introducing extra massive particles
which change the running of the gauge coupling constants and does not affect the anomaly
cancellation conditions. In fact, the gauge coupling unification is realized, if we introduce
extra massive particles so that the SU(3)C gauge coupling almost does not run and the
SU(2)L gauge coupling is asymptotically non-free.
Let us perform the potential analysis. We assume the SU(2)H D-flat condition. The
scalar potential in the present case is the same one in our previous letter except for the
D-term part 5
2

5
K
2
2
2
2
V (v, s) = e (v + sW ) + 2v s 4 + W 3W
v
gR2
(6s 2 4v 2 + 2)2 ,
(39)
2
where K and W are the Khler potential and the superpotential, respectively, which are
given by Eqs. (5) and (6). We assume , gR , v = 1 + y with y 1 and s 1, and
2
expand the potential with respect to y and s. The approximate formulae of the stationary
conditions V /y = 0 and V /s = 0 are
+
e(32 1610)
3
,
y s2
2 2
32 2gR2
(40)
105
.
(41)
92 3210
These formulae are correct within the order of magnitude, since the convergence of the
expansion is not so good. The numerical calculation gives
s
5 Although the FayetIliopoulos term due to the anomaly of U (1) should be included, but its magnitude is
R
suppressed compared to that of the gauged U (1)R symmetry. Hence, we simply neglect it.
1
v + 0.019,
2
s 0.13,
269
(42)
(43)
where we used the parameters, = 103 , = 7.35 and gR2 = 1011 . We have checked
that the cosmological constant can be fine-tuned to zero by adjusting for fixed .
Next, we estimate the spectrum of supersymmetry breaking masses. The gravitino mass
is given by

5
m3/2 = eK/2 W 4.1 4 .
MP
(44)
This gravitino mass contributes to the masses of scalar partners via the tree-level interaction
S, D,
S L, N, E vanish,
of supergravity. Since R-charges for Q, U

S, D,
S L, N, E .
(45)
m2i m23/2 i = Q, U
S have R-charge 2,
On the other hand, since H and H
m2H,HS m23/2 + 2gR2 hDi,

5 2
2
O(m23/2 ).
m3/2 + 2 8.2 4
MP
(46)
The contribution from hDi 6= 0 is the same order of the gravitino mass.
The gaugino mass can be generated in two different ways. In the case that the gauge
kinetic function is trivial, the anomaly mediation contribution dominates as
m i =
(gi2 )
2gi2
m3/2 ,
(47)
where (gi2 ) is the beta function for the standard model gauge groups. 6 In the case that the
gauge kinetic function is non-trivial and includes the term S 2 [Q1 Q2 ]3 /MP8 , for example,
the gaugino masses are
1
m3/2.
(48)
23
In both cases, there could be further contribution from the vacuum expectation value of Fcomponent of the dilaton [28]. The experimental bound on gaugino masses in the MSSM
[29] determines the order of the gravitino mass as 10 TeV in both cases.
Here, we comment on the -term. We can have the higher-dimensional interaction
m i
W =
S[Q1 Q2 ]2 S
HH
MP4
(49)
which respects all symmetries, where is a dimensionless constant. Plugging the vacuum
expectation value of S, Q1 and Q2 into Eq. (49), we obtain hSi. By adjusting 1,
the electroweak scale is generated.
6 More precisely, there is a correction of the same order of magnitude, see Ref. [27].
270
4. Anomaly free U (1)R model

In the last section, we discussed the scenario that supersymmetry breaking effect is
mediated to the MSSM as the visible sector. The GreenSchwarz mechanism is used to
cancel gauge anomalies of U (1)R and the simple solution was found. However, the Green
Schwarz mechanism requires the dilaton field, and this leads to new difficult problems
such as the dilaton stabilization and so on [28,30]. Therefore, it is desirable to consider the
case that U (1)R is anomaly free. As mentioned earlier, it is impossible to cancel all the
gauge anomaly for U (1)R in the MSSM. In this section, we consider the supersymmetric
SU(5) GUT instead of the MSSM as the visible sector. We have found that all the gauge
anomalies can be cancelled by introducing SU(5) SU(2)H gauge singlets with non-trivial
R-charges.
Suppose that we have N of SU(5) SU(2)H gauge singlets with R-charge 2. The
anomaly cancellation conditions are

3
1
1
(50)
U (1)R [SU(5)]2 : 3 f + a + h + h + 5 + 5 = 0,
2
2
2

[U (1)R ]: 2(q1 + q2 ) + s + 3 5f + 10a + n

+ 5 h + h + 24 + N + 7 = 0,
(51)

3
3
3
3
3
3
3
3
3
[U (1)R ] : 2 q1 + q2 + s + 3 5f + 10a + n + 5 h + h
+ 24 3 + N + 31 = 0,
(52)
where f, a, n, , h and h are the R-charges of the fermionic component for the superfields
S(S
S(S
F
5), A(10), N(1), (24), H (5) and H
5), respectively. The representations in the
parenthesis are those of SU(5). The superpotential of Eq. (31), and Yukawa couplings
SH
S + NF
SH
W AAH + AF
(53)
give the conditions of Eq. (32) and

2a + h = 1,
f + a + h = 1
n + f + h = 1.
(54)
(55)
(56)
We have a simple solution, in the case of N = 36 as

S = A = N = 0,
F
Q1 = 0,
S = 2,
H =H
Q2 = 2,
S = 4,
= 1,
(57)
where these are the charges of superfields. Note that Q1 and Q2 have different charges now.
This modification makes possible to have the special charge assignment with all positive
charges except for the fields which couple to SU(2)H gauge bosons.
The potential analysis is the same in Ref. [12] as long as we assume that the potential is
along the SU(2)H D-flat direction. The gravitino mass is estimated as
5
m3/2 = eK/2 W 4 .
MP
(58)
271
The gravitino mass contributes to the masses of scalar partners via the tree-level
interactions of supergravity. Note that there is another contribution, if scalar partners
have non-zero U (1)R charges. In this case, they also acquire the mass from the vacuum
expectation value of the D-term, and it is estimated as
5 2
q,
(59)
m2D-term = qgR2 hDi
MP4
where q is the U (1)R charge. This mass squared is always positive for the scalar partners
with positive U (1)R charges. The mass is the same order of the magnitude of the gravitino
mass.
The gaugino mass can be generated in two different ways. In the case that the gauge
kinetic function is trivial, the anomaly mediation contribution dominates as
m i =
(gi2 )
2gi2
m3/2 .
(60)
On the other hand, in the case that the gauge kinetic function is non-trivial and includes the
higher-dimensional term 7 S([Q1 Q2 ])2 /MP5 , for example, the gaugino masses are
mi m3/2 .
(61)
Considering the experimental bound on gaugino masses in the MSSM [29], the gravitino
mass is taken to be of the order of 10 TeV or 1 TeV in the former case or the latter case,
respectively. This phenomenological constraint requires the dynamical scale of the SU(2)H
gauge interaction to be of the order of 1015 GeV for both cases. This also means that is
extremely small, 1015 . Note that the requirement of this fine-tuned value of is the
result from the fine-tuning for vanishing cosmological constant. Furthermore, this small
Yukawa coupling is consistent with the above discussion in the following sense. Since
S has the vacuum expectation value, the mass for Qi is generated through the Yukawa
coupling in Eq. (2). The relation hSi is needed not to change our result from the
SU(2)H gauge dynamics.
Here, we comment on the Higgs potential. We would like to discuss whether it is possible
to construct the higher-dimensional operators that realize the gauge symmetry breaking as
SU(5) SU(3)C SU(2)L U (1)Y SU(3)C U (1)em and also realize the doublet
triplet splitting. We can explicitly write down the following interactions respecting all
symmetries. 8
W
1
SH + 2 [Q1 Q2 ]2 H
S 2 H
[Q1 Q2 ]H
MP
MP5
4
3
+ 4 S[Q1 Q2 ]2 tr( 2 ) + 3 [Q1 Q2 ] tr( 4 ),
MP
MP
(62)
7 The higher-dimensional term S([Q Q ])2 /M 5 in the gauge kinetic function can be forbidden to all orders
1 2
P
by the discrete symmetry.
8 More precisely, there are other terms of contracting the gauge indices in different ways, but we simply write
the representative.
272
where i (i = 1 4) are constants. By choosing i appropriately, the first two terms can
realize the doublettriplet splitting, while the last two terms can realize the gauge symmetry
breaking.
Unfortunately, there exists unwanted operators, for examples,
1
MP3
S
[Q1 Q2 ] H F
1
MP5
1
MP10
2

1
S 2,
HF
MP
[Q1 Q2 ]3 S 2 MP S 2 ,
[Q1 Q2 ]5 S 3 S 3 .
(63)
Although the first term seems to be harmless, the other terms are dangerous since
supersymmetry is restored if these operators are present. However, this problem is not
specific to our model, but inevitable and generic in supergravity models.
5. Summary and discussion

In this paper, we have shown that it is possible to construct simple and phenomenologically viable models of dynamical supersymmetry breaking with gauged U (1)R symmetry. Supersymmetry breaking occurs by the interplay between the dynamically generated
superpotential and the FayetIliopoulos term which appears due to the symmetry of supergravity. The cosmological constant can be fine-tuned to vanish at the minimum of the
potential. What is the most non-trivial in this class of models is the anomaly cancellations for U (1)R . We have presented two explicit models with anomaly cancellations as the
visible sector.
One is the MSSM with all the Yukawa couplings and without the -term. Anomalies
are cancelled by the GreenSchwarz mechanism. We have found a quite simple solution
of R-charge assignments for matter superfields in our model compared to the models
in Refs. [13,14]. Our solution consists of all integer charges. We have discussed the
gauge coupling unification which follows from the GreenSchwarz relations. The gauge
coupling unification is easily accomplished by introducing extra massive particles. The
spectrum of supersymmetry breaking masses has been estimated and turned out to be
phenomenologically viable as follows. The gravitino mass is of order 10 TeV. The scalar
masses are the same order of the gravitino mass. The gaugino masses are a few orders
smaller than the gravitino mass. We have also shown that it is possible to have the term. Unfortunately, we have to introduce the dilaton superfield to cancel anomalies in this
model. This leads to new difficult problems such as the dilaton stabilization and so on [28,
30].
The other is the supersymmetric SU(5) GUT. In this model, U (1)R anomalies are
cancelled by introducing the SU(5) gauge singlet superfields. We have also found a quite
simple solution of R-charge assignment. Unlike the anomalous U (1)R model in this paper
and the anomalous U (1) model in Refs. [10,11], the dilaton is not necessary since the
273
gauged U (1)R is anomaly free. Therefore, there is no dilaton stabilization problem. We

have estimated the mass spectrum of supersymmetry breaking and found that these models
are phenomenologically viable. The gravitino mass is of the order 1 TeV or 10 TeV
depending on the form of the gauge kinetic function. The scalar masses are the same order
of the gravitino mass. The gaugino masses are a few orders smaller than the gravitino mass
in the case that the gauge kinetic function is trivial, while the same order of it in the case
that the gauge kinetic function have higher-dimensional terms. We have shown that it is
possible to have the Higgs potential for the doublettriplet splitting and the GUT gauge
symmetry breaking.
It is interesting to apply our mechanism to the phenomenology of the anomaly mediation
scenarios. In this scenarios, if the visible sector consists of the MSSM, the slepton mass
squared becomes negative because the beta function coefficients of SU(2)L U (1)Y are
negative. This problem is easily avoided 9 by using our mechanism since the contribution
to the mass squared from the vacuum expectation value of U (1)R D-term is always positive
for the positively R-charged field. Unfortunately, this situation is not realized in our models
since R-charges of the matter fields are zero. However, it would be possible to construct
the models in which lepton superfields have positive charges.
Finally, note that our model is effectively reduced to the well known Polonyi model
which is the simplest supersymmetry breaking model at the tree-level. One can understand
this fact by freezing the Qi with its vacuum expectation values in the superpotential of Eq.
(2). Here, we would like to emphasize that the Polonyi model is derived as the effective
theory of our model. The Polonyi model does not specify the dynamics of supersymmetry
breaking. Also, unlike the Polonyi model, there is no dimensionful parameters other than
the Planck scale in our model. They are induced dynamically from the Planck scale.
Acknowledgements
This work was supported in part by the Grant-in-aid for Science and Culture Research
form the Ministry of Education, Science and Culture of Japan (#11740156, #3400, #2997).
N.M. and N.O. are supported by the Japan Society for the Promotion of Science for Young
Scientists.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
P. Fayet, J. Iliopoulos, Phys. Lett. B 51 (1974) 461.

L. ORaifeartaig, Nucl. Phys. B 96 (1975) 331.
E. Witten, Nucl. Phys. B 188 (1981) 513.
E. Poppitz, S. Trivedi, Ann. Rev. Nucl. Part. Sci. 48 (1998) 307.
Y. Shadmi, Y. Shirman, Rev. Mod. Phys. 72 (2000) 25, and references therein.
H.P. Nilles, Phys. Rep. 110 (1984) 1.
G.F. Giudice, R. Rattazzi, Phys. Rep. 322 (1999) 419, and references therein.
9 For related works, see [3133].
274
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
L. Randall, S. Sundrum, Nucl. Phys. B 557 (1999) 79.

G.F. Giudice, M. Luty, H. Murayama, R. Rattazzi, JHEP 9812 (1998) 027.
G. Dvali, A. Pomarol, Phys. Rev. Lett. 77 (1996) 3728.
P. Bintruy, E. Dudas, Phys. Lett. B 389 (1996) 503.
N. Kitazawa, N. Maru, N. Okada, hep-ph/9911251.
A.H. Chamseddine, H. Dreiner, Nucl. Phys. B 458 (1996) 65.
D.J. Castano, D.Z. Freedman, C. Manuel, Nucl. Phys. B 461 (1996) 50.
I. Affleck, M. Dine, N. Seiberg, Nucl. Phys. B 241 (1984) 493; Phys. Lett. B 137 (1984) 187.
S. Ferrara, L. Girardello, T. Kugo, A. van Proeyen, Nucl. Phys. B 223 (1983) 191.
T. Kugo, S. Uehara, Nucl. Phys. B 226 (1983) 49.
E. Cremmer, S. Ferrara, L. Girardello, A. van Proeyen, Nucl. Phys. B 212 (1983) 413.
M. Dine, N. Seiberg, E. Witten, Nucl. Phys. B 289 (1987) 589.
J. Atick, L. Dixon, A. Sen, Nucl. Phys. B 292 (1987) 109.
M. Dine, I. Ichinose, N. Seiberg, Nucl. Phys. B 293 (1988) 253.
M. Green, J. Schwarz, Phys. Lett. B 149 (1984) 117.
N.K. Nielsen, M.T. Grisaru, H. Rmer, P. van Nieuwenheuizen, Nucl. Phys. B 140 (1978) 477.
N.K. Nielsen, H. Rmer, Phys. Lett. B 154 (1985) 141.
L. Alvarez-Gaume, E. Witten, Nucl. Phys. B 234 (1983) 269.
J.P. Derendinger, Phys. Lett. B 151 (1985) 203.
J. Bagger, T. Moroi, E. Poppitz, hep-th/9911029.
N. Arkani-Hamed, M. Dine, S.P. Martin, Phys. Lett. B 431 (1998) 3219.
Particle Data Group, Eur. Phys. J. C 3 (1998) 1.
T. Barreiro, B. de Carlos, J.A. Casas, J.M. Moreno, Phys. Lett. B 445 (1998) 82.
A. Pomarol, R. Rattazzi, JHEP 9905 (1999) 013.
Z. Chacko, M.A. Luty, I. Maksymyk, E. Ponton, hep-ph/9905390.
E. Katz, Y. Shadmi, Y. Shirman, JHEP 9908 (1999) 015.

Consistent SO(6) reduction of type IIB supergravity

on S 5
M. Cvetic a,1 , H. L a,1, , C.N. Pope b,2 , A. Sadrzadeh b , T.A. Tran b
a Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA
b Center for Theoretical Physics, Texas A&M University, College Station, TX 77843, USA
Abstract
Type IIB supergravity can be consistently truncated to the metric and the self-dual 5-form. We
obtain the complete non-linear KaluzaKlein S 5 reduction ansatz for this theory, giving rise to
gravity coupled to the fifteen YangMills gauge fields of SO(6) and the twenty scalars of the coset
SL(6, R)/SO(6). This provides a consistent embedding of this subsector of N = 8, D = 5 gauged
supergravity in type IIB in D = 10. We demonstrate that the self-duality of the 5-form plays a
crucial role in the consistency of the reduction. We also discuss certain necessary conditions for
a theory of gravity and an antisymmetric tensor in an arbitrary dimension D to admit a consistent
sphere reduction, keeping all the massless fields. We find that it is only possible for D = 11, with
a 4-form field, and D = 10, with a 5-form. Furthermore, in D = 11 the full bosonic structure of
eleven-dimensional supergravity is required, while in D = 10 the 5-form must be self-dual. It is
remarkable that just from the consistency requirement alone one would discover D = 11 and type
IIB supergravities, and that D = 11 is an upper bound on the dimension. 2000 Elsevier Science
B.V. All rights reserved.
1. Introduction
Non-trivial KaluzaKlein sphere reductions of supergravity theories have been studied
in a number of contexts. Long ago it was demonstrated [1] that the linearised analysis
of the zero-mode fluctuations of the S 7 reduction of D = 11 supergravity [2,3] can be
extended to a fully non-linear and consistent embedding of maximal D = 4 gauged SO(8)
supergravity in D = 11. The fact that the truncation to the zero-mode sector is consistent,
despite the non-linearities of the theory, is rather non-trivial since there appears to be no
group-theoretic understanding of why it should work.
Corresponding author. E-mail: honglu@dept.physics.upenn.edu
1 Research supported in part by DOE grant DOE-FG02-95ER40893.
2 Research supported in part by DOE grant DOE-FG03-95ER40917.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 7 2 - 2
276
M. Cvetic et al. / Nuclear Physics B 586 (2000) 275286
More recently, a similar, although more explicit, demonstration of the consistency of the
S 4 reduction from D = 11 to the maximal D = 7 gauged SO(5) supergravity was given
[4,5]. The consistent reduction of massive type IIA supergravity on a locally S 4 space,
to give the N = 2 gauged SU(2) supergravity in D = 6, has also been obtained [6]. In
addition, certain reductions to truncations of the maximal gauged supergravities in various
dimensions have also been constructed. These have the advantage of being considerably
simpler than the maximal theories, allowing the reduction ansatz to be presented in a
more explicit form. Cases that have been constructed include the N = 2 gauged SU(2)
supergravity in D = 7 [7]; the N = 4 gauged SU(2) U (1) supergravity in D = 5 [8]; the
N = 4 gauged SO(4) supergravity in D = 4 [9]; and maximal abelian truncations in D =
4, 5 and 7 [10]. One can also consider non-supersymmetric truncations of the maximal
gauged supergravities. In [11,12] the consistent truncations of the D = 7, D = 5 and
D = 4 supergravities to the graviton plus scalar subsectors comprising only the diagonal
scalars in the SL(N, R)/SO(N) scalar submanifolds were considered (with N = 5, 6, and
8, respectively), and the consistent embeddings in D = 11 and D = 10 were constructed.
Although the full consistency of the S 7 and S 4 reductions of D = 11 supergravity
has essentially been demonstrated, no similar complete result exists for the S 5 reduction
of type IIB supergravity. The field content of the full N = 8 gauged supergravity [13]
consists of gravity; fifteen SO(6) gauge fields; twelve 2-form gauge potentials in the 6
and 6 representations of SO(6); 42 scalars in the 1 + 1 + 200 + 10 + 10 representations
of SO(6), and the fermionic superpartners. It is believed that this can arise from an S 5
reduction of type IIB supergravity; at the linearised level, the reduction ansatz was given in
[14]. However, at the full non-linear level, the only complete demonstrations so far are for
the consistent embedding of the maximal abelian U (1)3 truncation [10], the N = 4 gauged
SU(2) U (1) truncation [8], and the scalar truncation in [11,12]. The full metric ansatz
was conjectured in [15].
In this paper, we obtain the consistent reduction ansatz for a different truncation of the
maximal gauged D = 5 supergravity. One can consistently set the 1 + 1 + 10 + 10 of scalars
to zero, at the same time as setting the 6 + 6 of 2-form potentials to zero. The bosons that
remain, namely gravity, the 15 YangMills fields, and the 200 of scalars, come just from
the metric plus the self-dual 5-form sector of the original type IIB theory. We shall obtain
complete results for the consistent embedding of this subsector of the gauged N = 8 theory,
ij
with all fifteen of the SO(6) gauge fields A(1) , and the twenty scalars that parameterise the
full SL(6, R)/SO(6) submanifold of the complete scalar coset. These can be parameterised
by a unimodular symmetric tensor Tij .
Another way of expressing the truncation that we shall consider in this paper is as
follows. The type IIB theory itself can be consistently truncated in D = 10 so that just
gravity and the self-dual 5-form remain. The fifteen YangMills gauge fields and twenty
scalars that we retain in our ansatz are the full set of massless fields associated with Kaluza
Klein reduction from this ten-dimensional starting point. (The counting of massless fields
is the same as the one that arises from a toroidal reduction from the same ten-dimensional
starting point.) We shall see below that the self-duality condition on the 5-form plays an
essential rle in the consistency of the S 5 reduction.
277
This subsector of type IIB supergravity is particularly relevant for the AdS/CFT
correspondence [1618], because it is the metric and the self-dual 5-form that couple to
the D3-brane.
We also address the more general question of the circumstances under which a theory
admits a consistent sphere reduction. We show that in a D-dimensional theory of gravity
coupled to an n-form field strength, a consistent S n reduction that retains the full set of
SO(n + 1) YangMills fields together with coupled massless scalars is possible only when
D = 11 and n = 4 or 7, or in D = 10 with n = 5. Furthermore, in D = 11 the theory
has to be the bosonic sector of eleven-dimensional supergravity, with the FFA term, while
in D = 10 the 5-form must be self-dual. In all three cases the full set of massless scalars
includes a subset Tij described by the coset SL(n+1, R)/SO(n+1). (Such a coset structure
is absent for any other values of (D, n), and so it would not be appropriate to look for a
consistent reduction ansatz with scalars Tij of SL(n + 1, R)/SO(n + 1) for generic (D, n).)
Of the three cases where such a consistent sphere reduction is possible, the ten-dimensional
one is singled out as the only case where the consistent reduction includes only gravity plus
ij
the SO(n + 1) gauge fields A(1) and the scalars Tij . By contrast, for D = 11 reduced on S 4
one must additionally retain a set of five 2-form potentials, while for D = 11 reduced on
S 7 one must instead additionally retain 35 pseudoscalars as well as the 35 scalars Tij , in
order to achieve consistent reductions. 3
2. The SO(6) reduction ansatz on S 5
We parameterise the fields for this truncated theory as follows. The twenty scalars, which
are in the 200 representation of SO(6), are represented by the symmetric unimodular tensor
Tij , where i is a 6 of SO(6). The fifteen SO(6) YangMills gauge fields will be represented
ij
by the 1-form potentials A(1) , antisymmetric in i and j . The inverse of the scalar matrix Tij
is denoted by Tij1 . In terms of these quantities, we find that the KaluzaKlein reduction
ansatz is given by
2
= 1/2 ds52 + g 2 1/2 Tij1 Di Dj ,
d s10
b(5) + G
b(5) ,
b(5) = G
H

b(5) = gU (5) + g 1 T 1 DTj k (k Di )
G
ij
1
ij
k
`
g 2 Tik1 Tj1
` F(2) D D ,
2

b(5) = 1 i1 i6 g 4 U 2 Di1 Di5 i6
G
5!
5g 4 2 Di1 Di4 DTi5 j Ti6 k j k

i1 i2
10g 31 F(2)
Di3 Di4 Di5 Ti6 j j ,
(1)
(2)
(3)
(4)
3 This is related to the fact that one can consistently truncate five-dimensional maximal gauged supergravity to
the SO(6) gauge fields plus scalars of the coset SL(6, R)/SO(6). By contrast, the analogous truncations cannot
be performed in the maximal gauged supergravities in D = 7 and D = 4.
278
where
U 2Tij Tj k i k Tii ,
ij
ij
kj
Tij i j ,
jk
F(2) = dA(1) + gAik

(1) A(1) ,
DTij dTij + gAik

(1) Tkj + gA(1) Tik ,
i i = 1,
ij
Di di + gA(1) j ,
(5)
b(5) is derivable
and (5) is the volume form on the five-dimensional spacetime. Note that G
from the given expressions (1) and (3); we have presented it here because it is quite an
involved computation. The coordinates i , subject to the constraint i i = 1, parameterise
points in the internal 5-sphere. In obtaining the above ansatz we have been guided by
previous results in the literature, including the S 4 reduction ansatz from D = 11 that was
constructed in [4,5].
It is consistent to truncate the fields of type IIB supergravity to the metric and self-dual
b(5) . The ten-dimensional equations for motion for these fields are then given by
5-form H
bMP QRS H
bN P QRS ,
bMN = 1 H
R
96
b(5) = 0.
dH
(6)
The ansatz presented above satisfies these equations of motion if and only if the fivedimensional fields satisfy the equations

1
k1 k2
k3 k4
1
k`
D Tik1 Tj1
` F(2) = 2gTk[i DTj ]k 8 ij k1 k4 F(2) F(2) ,

mj
1
`k
F(2)
F(2)
D Tik1 DTkj = 2g 2 (2Tik Tj k Tij Tkk )(5) + Tik1 T`m

1
mp
1 1
`k
ij 2g 2 2Tk` Tk` (Tkk )2 (5) + Tpk
T`m F(2)
F(2) , (7)
6
together with the five-dimensional Einstein equation. These equations of motion can all be
derived from the five-dimensional Lagrangian
1
1
ij
1
k`
DTì Tik1 Tj1
L5 = R1 Tij1 DTj k Tk`
` F(2) F(2) V 1
4
4
1
i1 i2 i3 i4 i5 i6
i1 i2 i3 i4 i5 j j i6
F(2) A(1) gF(2)
A(1) A(1) A(1)
i1 i6 F(2)
48

2 2 i1 i2 i3 j j i4 i5 k ki6
+ g A(1) A(1) A(1) A(1) A(1) ,
5
(8)
where the potential V is given by

1
(9)
V = g 2 2Tij Tij (Tii )2 .
2
In (8) we have omitted the wedge symbols in the final topological term, to economise on
space. The Lagrangian is in agreement with the one for five-dimensional gauged SO(6)
supergravity in [13].
b(5) = 0, with the ansatz
To see this, we look first at the ten-dimensional equation d H
ij
given above. The terms involving structures of the form X(4) Di Dj , where
279
ij
X(4) represents a 4-form in the five-dimensional spacetime, give rise to the YangMills
b(5) and also from the final term in
equations above. Contributions of this kind come from d G
ij
ij
b(5) . The terms involving structures of the form X(5)
(i Dj ), where X(5) represents
d G
a 5-form in the five-dimensional spacetime, give rise to the scalar equations of motion
in (7). Since i Di = i di = 12 d(i i ) = 0, there is a trace subtraction in the scalar
ij
kk = 0 as the five-dimensional equation of motion.
equation, and we read off X(5) 16 ij X(5)
b(5) . Finally, all other structures arising
Contributions of this structure come only from d G
b5 = 0 vanish identically, without the use of any five-dimensional
from calculating d H
equations of motion. In deriving these results one needs to make extensive use of the
Schoutens over-antisymmetrisation identity, [i1 i6 Vi7 ] = 0.
It is worth remarking that it is essential for the consistency of the reduction ansatz that
b(5) should be self-dual. One cannot simply consider a reduction ansatz for a
the 5-form H
ten-dimensional theory consisting of gravity plus a non-self-dual 5-form, whose ansatz
b(5) = 0 would give perfectly
b(5) in (3). Although the Bianchi identity d G
is given by G
ij
(5) = 0 would
acceptable equations of motion for F(2) and Tij , the field equation d G
produce the (unacceptable) constraint
k1 k2
k3 k4
F(2)
= 0.
ij k1 k4 F(2)
(10)
b(5) together into the self-dual field H

b(5) that a consistent
b(5) and G
It is only by combining G
b(5) to
five-dimensional result is obtained, with (10) now combining with terms from d G
4
form part of the five-dimensional YangMills equations given in (7). It is interesting,
therefore, that self-duality of the 5-form is apparently forced on us by the requirements
of KaluzaKlein consistency, if we try to invent a ten-dimensional theory that can be
reduced on S 5 . Thus once again we see that supersymmetry and KaluzaKlein consistency
for sphere reductions seem to go hand in hand. 5
The KaluzaKlein S 5 reduction that we have obtained here retains the full set of massless
fields that can result from the reduction of gravity plus a self-dual 5-form in D = 10.
In other words, after the initial truncation of the type IIB theory in D = 10, no further
truncation of massless fields has been performed.
It should also be emphasised that it would be inconsistent to omit the fifteen SO(6)
gauge fields when considering the embedding of the twenty scalars Tij . This can be seen
1
DTj ]k appearing
from the YangMills equations in (7), which have a source term gTk[i
on the right-hand side. This is a quite different situation from a toroidal reduction, where
it is always consistent to truncate to the scalar sector, setting the gauge fields to zero. The
new feature here in the sphere reduction is that the scalar fields are charged under the gauge
4 If one were to consider the reduction of a ten-dimensional theory with a non-self-dual 5-form there would be
additional fields present in a complete massless truncation. These would comprise 10 = 4 + 6 vector potentials
and 5 = 1 + 4 scalars. However the inclusion of these fields would still not achieve a consistent reduction ansatz,
since the current (10), which is in the 15 of SO(6), could still not acquire an interpretation as a source term for
the additional fields. Thus the additional requirement of self-duality seems to be essential for consistency. (Of
course, anti-self-duality would be equally good.)
5 Of course, the initial truncation of the type IIB theory to its gravity plus self-dual 5-form sector is itself a nonsupersymmetric one, but the crucial point is that the consistency of the KaluzaKlein S 5 reduction is singling out
a starting point that is itself a subsector of a supersymmetric theory.
280
group. This is a general feature of all the sphere reductions, to D = 4, D = 5 and D = 7,

and thus in all cases it is inconsistent to include the full set of scalar fields without including
the gauge fields as well (see also [4,5,19]). (One can consistently truncate to the diagonal
1
DTj ]k = 0.)
scalars in Tij , setting all the gauge fields to zero, as in [11,12], since then Tk[i
Our testing of the consistency of the reduction ansatz (1)(4) has so far been restricted
b(5) . A full testing of the consistency of the tento checking the equations of motion for H
dimensional Einstein equations would be quite involved, and will be addressed in future
work. 6 In the next section we shall show that the reduction ansatz that we have obtained
in this paper reduces, with appropriate additional truncations, to results that have been
obtained previously. Since the complete consistency was proven in these earlier results,
including the ten-dimensional Einstein equations, this provides further supporting evidence
for the complete consistency of the ansatz that we have constructed here.
Finally in this section, we remark that the S 5 reduction that we have constructed here is
also consistent if we include the dilaton and axion of the type IIB theory. These simply
reduce according to the ansatz
= ,
= ,
(11)
where the unhatted quantities denote the fields in five dimensions. In this reduction they
do not appear in the previous ansatz for the metric and self-dual 5-form, and in D = 5 they
just give rise to the additional SL(2, R)-invariant Lagrangian
1
1
(12)
L(,) = d d e2 d d,
2
2
which is added to (8). Note in particular that and do not appear in the five-dimensional
scalar potential V .
3. Truncations to previous results

We can consider three different truncations of the reduction scheme of the previous
section, in order to make contact with previous results in the literature. The first of the
three involves truncating the twenty scalars Tij of the SL(6, R)/SO(6) coset to the diagonal
subset
(13)
Tij = diag(X1 , X2 , X3 , X4 , X5 , X6 ),
Q
ij
where i Xi = 1, and setting all the fifteen gauge fields F(2) to zero. This reduces to the
embedding that was obtained in [11], for which the complete proof of consistency was
constructed in [12].
The second possible truncation involves reducing the scalar sector still further, to

e1 , X
e1 , X
e2 , X
e2 , X
e3 , X
e3 ,
(14)
Tij = diag X
6 The experience in all previous work on consistent reductions is that although the actual checking of the higherdimensional Einstein equations is the most difficult part from a computational point of view, the consistency seems
to be assured once it has been achieved for the equations of motion for the antisymmetric tensor fields, which
itself is an extremely stringent requirement.
281
Q
12 , F 34 and F 56
ea = 1, but now retaining the three U (1) gauge fields F(2)
where a X
(2)
(2)
3
of the maximal abelian U (1) subgroup of SO(6). (It is easy to see, by looking at the
five-dimensional equations of motion in (7), that this is a consistent truncation.) The
truncated theory is supersymmetric, and describes five-dimensional U (1)-gauged simple
supergravity coupled to two U (1) vector multiplets. The consistent embedding of this
theory in type IIB supergravity was obtained in [10].
The third possible truncation involves retaining just a single scalar field X, by taking

(15)
Tij = diag X, X, X, X, X2 , X2 ,
at the same time retaining only the gauge fields of SU(2) U (1). It is convenient now to
ij
take the SO(6) indices i to range over 0 6 i 6 5. We take all the gauge potentials A(1) to
be zero except for the following:
1 1
23
A01
(1) = A(1) = A(1) ,
2
1 3
03
12
A(1) = A(1) = A(1) ,
2
1 2
31
A02
(1) = A(1) = A(1) ,
2
A45
(1) = B(1) .
(16)
We also parameterise the coordinates i of the internal 5-sphere as follows:

0 + i3 = cos cos 12 ei(+)/2,
1 + i2 = cos sin 12 ei()/2,
4 + i5 = sin ei .
(17)
Substituting (15), (16) and (17) into the metric ansatz (1), we obtain
2
2
= 1/2 ds52 + g 2 X1/2 d 2 + g 2 1/2 X2 s 2 d gB(1)
d s10
X
2
1
+ g 2 1/2X1 c2
(18)
i 2 gAi(1) ,
4
i
where c cos , s sin , = X2 s 2 + Xc2 , hi i 2 gAi(1) , and the i denote

the three left-invariant 1-forms of SU(2), given by 1 + i2 = ei (d + i sin d),
3 = d + cos d. This is the metric ansatz obtained in [8] for the embedding of fivedimensional N = 4 gauged SU(2) U (1) supergravity in D = 10. The internal 5-sphere
now has a geometrical interpretation as a foliation by S 3 S 1 , with parameterising the
foliating surfaces.
b(5) given in (3), leads to
Substituting (15), (16) and (17) into the ansatz for G
2
i
b(5) = gU 5 3sc X1 dX d + c
X2 F(2)
hj hk ij k
G
g
8 2 g2

sc
sc
i

X2 F(2)
hi d 2 X4 G(2) d d gB(1) ,
2
g
2 2g
(19)
where U = 2(X2 c2 + X1 s 2 + X1 ), again in agreement with the ansatz obtained in [8].

As a consistency check, we can also verify that substituting (15), (16) and (17) into the
b(5) in [8]. Thus
b(5) gives the same expression as the one obtained for G
ansatz (4) for G
we have verified that the gauged SO(6) embedding that we have obtained in this paper can
be truncated to the SU(2) U (1) embedding of the N = 4 theory whose consistency was
282
proven in [8]. This again provides further supporting evidence for the consistency of the
our gauged SO(6) reduction ansatz.
4. Consistency conditions for sphere reductions

One of the interesting outcomes from our analysis is that it is essential for the consistency
of the 5-form reduction ansatz that it be a self-dual 5-form, rather than an unconstrained
one. It seems, therefore, that the requirement of consistency of the sphere reduction has
singled out a ten-dimensional starting point that is itself embeddable in a supersymmetric
theory.
This raises the more general question of what possible higher-dimensional theories
might allow consistent KaluzaKlein sphere reductions. All the known examples are
associated with supersymmetric higher-dimensional theories, but one might wonder
whether this was just a reflection of the fact that these are the cases that have received the
most attention in the literature. However, the following argument seems to suggest that the
supersymmetric cases may be the only ones that can allow consistent S n sphere reductions,
in which all the massless fields (including the SO(n + 1) YangMills fields) are retained.
Consider a D-dimensional theory of gravity plus an n-form field strength, with the
Lagrangian
e1 LD = R
1 2
F ,
2n! n
(20)
where e = g. If this were to give a (D n)-dimensional theory with an SO(n + 1)

gauge group, as a consistent reduction on S n , it would be necessary that the ungauged
(D n)-dimensional theory obtained by reducing instead on the torus T n should have
a global symmetry group G that contains SO(n + 1) as a compact subgroup, since an
SO(n + 1) factor in the denominator group would be gauged in the spherical reduction.
A reduction on T n always produces a theory with a GL(n, R) global symmetry, which has
SO(n) as its maximal compact subgroup, and so this would be insufficient for allowing an
SO(n + 1) gauging. In special cases the GL(n, R) global symmetry can be enhanced [20,
21] to SL(n + 1, R), but this happens only if there is a conspiracy between axionic scalars
coming from the metric and axions coming from the form-field Fn . For this conspiracy to
occur, the strengths of the dilaton couplings to axions from these two sources must be
the same. 7 Specifically, if E denotes the set of n canonically-normalised dilatons that
result from the T n reduction then all the dilaton/axion couplings should be of the form
E
ai )2 = 4 for each i. (In fact the full set
eaEi (i )2 , where the constant vectors aEi satisfy (E
of aEi vectors would constitute the positive-root vectors of SL(n + 1, R) [20,21].) It was
shown in [22] that the strengths of dilaton couplings can be conveniently characterised in
terms of the quantity , related to the dilaton vector aE by
7 It is always the case that the counting of dilatons and axions would be consistent with the numerology required
for an SL(n + 1, R)/SO(n + 1) coset structure, but this does not in general imply that the enhanced coset actually
occurs.
aE 2 =
2(n 1)(D n 1)
,
D2
283
(21)
for an n-form field strength in D dimensions, since the quantity is preserved under
toroidal KaluzaKlein reduction. Furthermore, it was shown that the KaluzaKlein vectors
and axions coming from a toroidal reduction of the metric always have = 4. It therefore
follows that for the symmetry enhancement to SL(n + 1, R) to take place, the dilatoncoupling for the original n-form field 8 in (20) must also take the value = 4. An
enumeration of all possible cases shows that for Lagrangians of the form (20) one gets
= 4 only for
(D, n) = (11, 4),
(11, 7),
(10, 5).
(22)
The above considerations seem to single out the three cases in (22) as the only ones
where an S n reduction of a Lagrangian of the form (20) could consistently yield the
gauge fields of SO(n + 1) and the scalars Tij of the coset SL(n + 1, R)/SO(n + 1). 9
As we have seen in Section 2 of this paper, it can turn out that additional structure is also
required in order to achieve consistency, namely the self-duality of the 5-form in the case
of (D, n) = (10, 5). By the same token one can expect that a consistent reduction would
only be possible in the cases (D, n) = (11, 4) and (11, 7) if additional structure is also
present in the eleven-dimensional Lagrangian.
For example, if we consider the S 7 reduction from D = 11, then the total set of 70
spin-0 fields decompose as a 35v of scalars and a 35c of pseudoscalars. The 35 scalars
in Tij correspond to the 35v . Turning on these forces all the 28 gauge fields of SO(8)
to be excited, and in turn these excite the remaining 35c of pseudoscalars. In order for
an S 7 reduction that retains all these fields to be consistent, it is necessary to include an
F(4) F(4) A(3) term in the original Lagrangian (20), with precisely the coefficient dictated
by D = 11 supersymmetry. This point also emphasises that in the S 7 reduction one cannot
consider just the 35 scalars Tij in isolation; the full set of bosonic fields of N = 8 gauged
supergravity (and not merely the 28 gauge fields) must be included if the full set of Tij
scalars are present.
A similar situation arises with the S 4 reduction from D = 11. Including the full set of 14
scalars Tij in a consistent reduction will force the complete set of massless fields to be nonvanishing, including not only the ten YangMills fields of SO(5) but also the five 3-form
field strengths coming from the antisymmetric tensor. The consistency of the reduction is
then only possible if the FFA term of D = 11 supergravity is included.
Another possibility for obtaining further examples of consistent sphere reductions is to
include a dilaton in the higher-dimensional Lagrangian, whose coupling to the n-form field
strength is arranged to have = 4:
8 If there is no dilaton in the original higher-dimensional theory then the dilaton coupling is given by
setting aE 2 = 0 in (21), and it is this value that is preserved under toroidal reduction.
9 Consistent sphere reductions with scalars T for more general (D, n) values for the Lagrangian (20) have
ij
been suggested in [19]. We expect that a more complete analysis of the consistency of the reduction would
exclude such possibilities. In particular, scalars Tij parameterising the coset SL(n + 1, R)/SO(n + 1) only occur
in the three cases listed in (22).
284
1
1 a 2
e1 LD = R ()2
(23)
e Fn ,
2
2n!
with a 2 = 4 2(n 1)(D n 1)/(D 2). (Lagrangians of this type without the
restriction on the value of the constant a have also been discussed in [19] in the context of
sphere reductions.) In this case there will no longer be an AdSDn S n vacuum solution,
but rather a warped product of a domain wall and S n . The possibilities for achieving = 4
couplings are in fact rather limited, given that a should be a real number. First of all, for an
n-form with 4 6 n 6 D 4 it can be seen that we must have D 6 11. For example, if n = 4,
then we must have D 6 11, while for n = 5 we must have D 6 10. (It is only necessary
to consider forms with n 6 D/2 since Hodge duality maps those with n > D/2 into this
range.) Interestingly, for n = 3 one can achieve = 4 in an arbitrary dimension D; the
Lagrangian (23) then corresponds to the low-energy effective theory of the D-dimensional
bosonic string. We can expect that the full 3-sphere reduction of this Lagrangian (keeping
the complete SO(4) gauge fields of its isometry group, not merely the SU(2) subset of leftinvariant fields) will be consistent. Similarly, we can expect that it should be consistent to
reduce the theory with n = 3 on a (D 3)-sphere. Before the gauging, the scalar coset in
D = 3 is SO(D 2, D 2)/(SO(D 2) SO(D 2)). One of the SO(D 2) denominator
group factors can be gauged, and we obtain the scalar coset SL(D 2, R)/SO(D 2)
together with the additional gauge fields of SO(D 2), and a singlet scalar (which is the
original dilaton of the D-dimensional theory).
A consistent sphere reduction, albeit of a slightly different kind, has in fact been obtained
in an example where there is a dilaton in the higher-dimensional theory that couples to the
form-field. In [6] it was shown that a reduction of the massive type IIA supergravity on
an internal 4-dimensional space that is locally S 4 gives rise to the N = 2 SU(2) gauged
supergravity in D = 6.
5. Conclusions and further comments
In this paper we have constructed a consistent KaluzaKlein reduction ansatz for
embedding the subset of the fields of five-dimensional N = 8 gauged supergravity,
comprising gravity, the fifteen SO(6) gauge fields, and the twenty scalars of the
SL(6, R)/SO(6) submanifold of the full scalar manifold, into type IIB supergravity in
D = 10. This embedding can equivalently be viewed as a complete reduction ansatz (with
no truncation of massless fields) for the ten-dimensional theory comprising just gravity
plus a self-dual 5-form, which itself is a consistent truncation of type IIB supergravity.
A crucial point in the analysis is that in the gauged five-dimensional supergravity one
cannot consistently set the YangMills fields to zero, while retaining the full set of scalar
fields, unlike the situation in ungauged supergravity. In the context of the truncation that we
consider in this paper, where we retain the twenty scalars Tij , we cannot ignore the fifteen
SO(6) gauge fields, since the scalars act as sources for them. 10 We saw that the self-duality
10 This also implies that solutions built using these scalars will in general require non-vanishing YangMills
fields.
285
of the 5-form field of the type IIB theory plays an essential rle in the consistency of the
reduction.
More generally, we showed that if one starts from a theory comprising gravity and an nform field strength, then a consistent reduction on S n that retains the scalars Tij of the coset
SL(n + 1, R)/SO(n + 1) will also have to include at least the gauge fields of SO(n + 1),
and furthermore will only be possible for the S 4 and S 7 reductions of D = 11, and the
S 5 reduction of D = 10. In fact the consistency will in addition require that the theories
in D = 11 and D = 10 have the additional structures associated with D = 11 and type
IIB supergravity, namely the FFA term in D = 11, and the self-duality of the 5-form in
D = 10. In D = 7 the scalars and YangMills fields must be supplemented by the five
2-form potentials, while in D = 4 they must be supplemented by the 35 pseudo-scalars, in
order to achieve consistency. On the other hand in D = 5 no additional fields beyond the
ij
scalars Tij and the gauge fields A(1) are required for consistency. In fact in all three cases
this is related to the presence of the terms of the form (10), bilinear in YangMills fields.
In D = 7 this term acts as a source for the five 3-form fields; in D = 4 it acts as a source
for the 35 pseudo-scalars; but in D = 5 it acts as a source for the YangMills fields
themselves. This special feature of the D = 5 gauged supergravity may have implications
in the dual four-dimensional N = 4 super YangMills theory.
We also discussed the more general possibilities that might arise if one includes a dilaton
in the higher-dimensional theory. The possibilities for further examples of consistent
sphere reductions seem to be rather limited, as discussed in Section 4.
In a full S 5 reduction of type IIB supergravity there will be additional fields coming
from the reduction of the NSNS and RR 2-form potentials, and from the dilaton
and axion. A complete analysis of the S 5 reduction can therefore be expected to be
extremely complicated. In particular, for example, the 10 and 10 of pseudo-scalars lead
to a considerably more complicated metric reduction ansatz. The ansatz for a subset of
the fields that included one scalar and one pseudoscalar was derived in [23], and in [24].
However even in that case, the construction of the ansatz for the antisymmetric tensor fields
is rather involved, and has not yet been pushed to completion.
Acknowledgement
C.N.P. is grateful to the University of Pennsylvania for hospitality during the course of
this work.
References
[1] B. de Wit, H. Nicolai, The consistency of the S 7 truncation in D = 11 supergravity, Nucl. Phys.
B 281 (1987) 211.
[2] M.J. Duff, C.N. Pope, KaluzaKlein supergravity and the seven sphere, in: S. Ferrara,
J.G. Taylor, P. van Nieuwenhuizen (Eds.), Supersymmetry and Supergravity, Vol. 82, World
Scientific, Singapore, 1983.
[3] M.J. Duff, B.E.W. Nilsson, C.N. Pope, KaluzaKlein supergravity, Phys. Rep. 130 (1986) 1.
286
[4] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistent nonlinear KK reduction of 11d

supergravity on AdS7 S4 and self-duality in odd dimensions, hep-th/9905075.
[5] H. Nastase, D. Vaman, P. van Nieuwenhuizen, Consistency of the AdS7 S4 reduction and the
origin of self-duality in odd dimensions, hep-th/9911238.
[6] M. Cvetic, H. L, C.N. Pope, Gauged six-dimensional supergravity from massive type IIA,
Phys. Rev. Lett. 83 (1999) 5226, hep-th/9906221.
[7] H. L, C.N. Pope, Exact embedding of N = 1, D = 7 gauged supergravity in D = 11, Phys.
Lett. B 467 (1999) 67, hep-th/9906168.
[8] H. L, C.N. Pope, T.A. Tran, Five-dimensional N = 4 SU(2) U (1) gauged supergravity from
type IIB, Phys. Lett. B 475 (2000) 261, hep-th/9909203.
[9] M. Cvetic, H. L, C.N. Pope, Four-dimensional N = 4, SO(4) gauged supergravity from D =
11, hep-th/9910252, to appear in Nucl. Phys. B.
[10] M. Cvetic, M.J. Duff, P. Hoxha, J.T. Liu, H. L, J.X. Lu, R. Martinez-Acosta, C.N. Pope, H. Sati,
T.A. Tran, Embedding AdS black holes in ten and eleven dimensions, Nucl. Phys. B 558 (1999)
96, hep-th/9903214.
[11] M. Cvetic, S.S. Gubser, H. L, C.N. Pope, Symmetric potentials of gauged supergravities in
diverse dimensions and Coulomb branch of gauge theories, hep-th/9909121, to appear in Phys.
Rev. D.
[12] M. Cvetic, H. L, C.N. Pope, A. Sadrzadeh, Consistency of KaluzaKlein sphere reductions of
symmetric potentials, hep-th/0002056.
[13] M. Gnaydin, L.J. Romans, N.P. Warner, Compact and non-compact gauged supergravity
theories in five dimensions, Nucl. Phys. B 272 (1986) 598.
[14] H.J. Kim, L.J. Romans, P. van Nieuwenhuizen, Mass spectrum of ten-dimensional N = 2
supergravity on S 5 , Phys. Rev. D 32 (1985) 389.
[15] A. Khavaev, K. Pilch, N.P. Warner, New vacua of gauged N = 8 supergravity in five dimensions,
hep-th/9812035.
[16] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231, hep-th/9711200.
[17] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[18] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9802150.
[19] H. Nastase, D. Vaman, On the non-linear KK reductions on spheres of supergravity theories,
hep-th/0002028.
[20] E. Cremmer, B. Julia, H. L, C.N. Pope, Dualisation of dualities, Nucl. Phys. B 523 (1998) 73,
hep-th/9710119.
[21] E. Cremmer, B. Julia, H. L, C.N. Pope, Higher-dimensional origin of D = 3 coset symmetries,
hep-th/9909099.
[22] H. L, C.N. Pope, E. Sezgin, K.S. Stelle, Stainless super p-branes, Nucl. Phys. B 456 (1995)
669, hep-th/9508042.
[23] M. Cvetic, H. L, C.N. Pope, Geometry of the embedding of supergravity scalar manifolds in
D = 11 and D = 10, hep-th/0002099.
[24] K. Pilch, N.P. Warner, A new supersymmetric compactification of chiral type IIB supergravity,
hep-th/0002192.

D = 4 N = 1 type IIB orientifolds with continuous

Wilson lines, moving branes, and their field theory
realization
Mirjam Cvetic, Paul Langacker
Department of Physics and Astronomy, University of Pennsylvania, Philadelphia PA 19104-6396, USA
Received 13 June 2000; accepted 28 June 2000
Abstract
We investigate four-dimensional N = 1 type IIB orientifolds with continuous Wilson lines, and
their T-dual realizations as orientifolds with moving branes. When continuous Wilson lines become
discrete the gauge symmetry is enhanced and the T-dual orientifold corresponds to branes sitting at
the orbifold fixed points. There is a field theoretic analog describing these phenomena as D- and
F-flat deformations of the T-dual model, where the branes sit at the origin (original model without
Wilson lines) as well as a deformation of the T-dual model where sets of branes sit at the fixed
points (the model with discrete Wilson lines). We demonstrate these phenomena for the prototype
Z3 orientifold: we present an explicit construction of the general set of continuous Wilson lines as
well as their explicit field theoretic realization. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.Mj; 11.15.Ex
Keywords: Orientifolds; Wilson lines; Flat directions
1. Introduction
Four-dimensional N = 1 supersymmetric type IIB orientifolds (see [113] and references therein) provide a domain of perturbative string vacua with novel properties (as opposed to the perturbative heterotic solutions) with potentially interesting phenomenological
implications. One goal, that is far from being achieved, is the development of techniques
that would yield a larger class of solutions than those of based on symmetric orientifold
constructions. However, even within the current fairly limited class based on symmetric orbifolds, these models possess a rich structure of possible deformations which may
E-mail address: plg@langacker.hep.upenn.edu (P. Langacker).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 1 4 - 4
288
M. Cvetic, P. Langacker / Nuclear Physics B 586 (2000) 287302
provide a fruitful ground for further investigation of their phenomenological implications

(with the ultimate goal to identify classes of models with quasi-realistic features).
The deformations in the space of supersymmetric four-dimensional solutions always
have a field-theoretic realization, i.e., one identifies specific D- and F-flat directions of
the original (undeformed) model. The new (deformed) supersymmetric ground states
correspond to the exact string solutions, described as a power-series in the magnitude of the
vacuum expectation values of the fields responsible for the deformations. In particular, one
interesting phenomenon to explore is the blowing-up of the orientifold singularities [14
16], which are different in nature from that of perturbative heterotic orbifolds [17,18]. This
phenomenon is notoriously difficult to describe within the full string theory context, since
the metric of the blown-up space is not explicitly known. On the other hand the explicit
field-theoretic realization in terms of (non-Abelian) flat directions allows the determination
of the surviving gauge groups and massless spectrum.
Another set of deformations corresponds to the introduction of Wilson lines, both
continuous and discrete, and here one expects to have, parallel with the field theory
treatment, also the full string theory construction. The purpose of this paper is to address
the study of such continuous Wilson lines of four-dimensional N = 1 orientifolds, from
both the full string theory description, i.e., by constructing explicitly these Wilson lines,
and to find their T-dual interpretation, as well as from the field theory side, i.e., by
identifying the moduli space of D- and F-flat directions of the effective theory; these
deformations correspond on the string side (in the T-dual picture) to the motion of a
set of branes away from the fixed points. However, the construction of explicit continuous
Wilson lines allows for an explicit string theory realization where sets of branes are located
at an arbitrary distance away from the fixed points. On the other hand the field theoretical
approach is only perturbative in the vacuum expectation values (VEVs), and thus in the
string picture corresponds only to a deformation infinitesimally away from the undeformed
model, i.e., where branes are located at the orbifold fixed points.
The purpose of this paper is to set the stage for constructions of four-dimensional
N = 1 type IIB orientifolds with continuous Wilson lines, and their T-dual realizations as
orientifolds with moving branes. In particular, we concentrate on explicit constructions of
the continuous Wilson line solutions and the corresponding field theory realizations within
the prototype, Z3 orientifold model [1]. The explicit realization of these complementary
pictures provide a beautiful correspondence between the two approaches, and sets the stage
for further investigations of more involved orientifold models with continuous Wilson lines
[19].
Explicit examples of discrete Wilson lines have been constructed for a number of a
different orientifold models (see [8,11,12] and references therein). Continuous Wilson lines
were first addressed in [8]; however, the explicit unitary representation has not been given.
The connection of models with continuous Wilson lines to the T-dual models, where branes
are located away from the orbifold fixed points, while anticipated in general ([8,20,21] and
references therein), was exhibited for a number of examples [8]. It is also believed that in
general there should be a field theoretical realization of the same phenomena.
289
This paper advances these topics in several ways. In particular, we provide the first
explicit unitary representation of the continuous Wilson line, specifically constructed for
the Z3 orientifold, and construct for this model the most general set of continuous Wilson
lines that in the T-dual picture correspond to the moving branes. We show that these
solutions allow a continuous interpolation between the original model without Wilson lines
and the models with discrete Wilson lines. In addition, we provide a systematic analysis of
their realization on the field theory side.
The paper is organized in the following way. In Section 2.1 we summarize the salient
features of the Z3 orientifold construction. In Section 2.2 we proceed with the construction
of, first discrete and then continuous Wilson lines. When continuous Wilson lines become
discrete the gauge symmetry is enhanced and the T-dual orientifold corresponds to branes
sitting at the orbifold fixed points. In Section 3 we turn to the field theoretical analysis,
by first recapitulating the techniques for a classification of D- and F-flat directions
(Section 3.1). We then (Section 3.2) provide explicit constructions of such D- and F-flat
directions and demonstrate their one-to-one correspondence with the continuous Wilson
line string constructions.
2. Z3 orientifold with continuous Wilson lines

2.1. D = 4, N = 1 Z3 orientifold
We briefly summarize the construction [1] of four-dimensional Z3 type IIB orientifold
models. One starts with type IIB string theory compactified on a T 6 /Z3 (T 6 -sixtorus, G1 Z3 the discrete orbifold group) and mod out by the world-sheet parity
operation , which is chosen to be accompanied by the same discrete symmetry G2 Z3 ,
i.e., the orientifold group is G = G1 + G2 = Z3 + Z3 . (Closure requires gg 0
G1 = Z3 for g, g 0 G2 = Z3 .)
The compactified tori are described by complex coordinates Xi , i = 1, 2, 3. The action
of an orbifold group ZN on the compactified dimensions can be summarized via a twist
P
vector v = (v1 , v2 , v3 ) (subject to constraint 3i=1 vi = 1):
g : Xi e2ivi Xi .
(1)
For the Z3 orientifold v1 = v1 = v3 =

The tadpole cancellation, associated with the open-string modes, requires the inclusion
of an even number of D9 branes. (The case of additional discrete symmetries in the
orientifold group may require the presence of multiple sets of D5 branes as well.)
An open string state is denoted as |, ij i where denotes the world-sheet state and
i, j the ChanPaton indices associated with the end points on a D9 brane. The elements
g G1 = Z3 act on open string states as follows:

(2)
g : |, ij i (g )ii 0 g , i 0 j 0 g1 j 0 j .
1
3.
Similarly, the elements of G1 = Z3 act as

1
,
g : |, ij i (g )ii 0 g , j 0 i 0 g
j 0j
(3)
290
where we have defined g = g , up to a phase, in accordance with the usual rules for
multiplication of group elements. Note, that exchanges the ChanPaton indices.
Since the g form a projective representation of the orientifold group, consistency with
group multiplication implies some conditions on the g . Consider the G1 = G2 = Z3 case;
g 3 = 1, 2 = 1 and gg 0 Z3 for g, g 0 Z3 respectively imply:

= T ,
gk = gk .
(4)
g3 = 1,
It turns out that the tadpoles cancel, if we choose the plus sign for D9 branes (and the
minus sign for D5 branes) [8]. The explicit representation for the D9 brane sector is
symmetric and can be chosen real:

0 116
,
(5)
,9 =
116 0
where the subscript 9 denotes the D9 brane sector in which these matrices are acting.
Further, finiteness of string loop diagrams yields tadpole cancellation conditions which
constrain the traces of g matrices [1]
Tr(Z3 ) = 4.
(6)
The Z3 twist action on the tori is given by the twist vector v = ( 13 , 13 , 13 ) and its action
on ChanPaton matrices is generated by:

(7)
Z3 = diag 112 , 14 , 2 112 , 14 , where = e2i/3 .
This choice satisfies Eqs. (4) and (6). Open string states, whose ChanPaton matrices
will be denoted by (i) , i = 0, . . . , 3, in the following, give rise to spacetime gauge bosons
(i = 0) and matter states (i = 1, 2, 3).
Gauge bosons in the D9 brane sector arise from open strings beginning and ending on
D9 branes. Invariance of these states under the action of the orientifold group requires
T
1
(0) = ,9 (0) ,9
1
and (0) = g,9(0) g,9
.
(8)
With Eq. (5) the first constraint implies that the (0) are SO(32) generators, while the
constraints from the g,9 will further reduce the group.
The result is the gauge group:
U (12) SO(8).
(9)
The ChanPaton matrices of the matter states have to be invariant under the action of the
orientifold group as well. However, since the string vertices for the chiral matter superfields
involve the oscillator modes of the target space toroidal coordinates Xi , the ChanPaton
matrices now transform under the orbifold action (in order to render the physical states
invariant under the orbifold action), thus implying:
T
1
(i) = ,9 (i) ,9
1
and (i) = e2ivi g,9 (i) g,9
.
(10)
For Z3 orientifold this yields a matter content of three copies of

= (12, 8)1 ,
= (66, 1)+2 ,
= 1, 2, 3,
(11)
291
where the subscript refers to the U (1) charge of U (12). The closed string sector yields the
gravity supermultiplet and the 36 (chiral) supermultiplets corresponding to the 9 untwisted
(toroidal) and 27 twisted (blowing-up) sector moduli. The moduli are gauge singlets,
whose real and imaginary components arise from the NSNS and RR sector, respectively.
The renormalizable superpotential is of the form
b
W ia i [a,b] ,
(12)
where , , are family indices, {a, b}-U (12) indices, and i-SO(8) indices.
2.2. Wilson lines
Discrete Wilson line
When the action of the Wilson line on the ChanPaton matrices, which is represented
by a matrix W , is such that it commutes with Z3 , it depends only on discrete values of
parameters, i.e., it describes a discrete Wilson line. Let us focus on a Wilson line, acting
along the two-torus coordinate Xi , say i = 3. It satisfies the following algebraic consistency
conditions:
(Z3 W )3 = +1,
[Z3 , W ] = 0.
(13)
Further, tadpole cancellations require

Tr(Z3 ) = Tr(Z3 W ) = Tr Z3 W2 = 4.
(14)
To simplify the notation, let us rearrange the entries in the Z3 matrix (7) as follows:

Z3 = diag 2 16n , 16n , 14n , Z 1n ; 16n , 2 16n , 14n , Z 1n , (15)

(16)
Z = diag 2 , , 1 , = e2i/3 , n = {0, . . . , 4}.
The above consistency conditions reduce to the unique solution of the discrete Wilson line:

(17)
W = diag 1163n , 13n ; 1163n, 2 13n .
The surviving gauge symmetry is determined by a projection:
(0) = W (0) W1 ,
(18)
which further breaks the gauge group down to:

U (12 2n) SO(8 2n) U (n)3 .
(19)
The matter representation is determined by the condition:

(i) = W (i) W1 .
(20)
The matter comes in three copies and has the following representation (the subscripts
correspond to the self-evident U (1) charges):
= ((6 n)(11 2n), 1, 1, 1)2,
(21)
= (12 2n, 8 2n, 1, 1, 1)1,
(22)
1,1 ,
S = (1, 1, n, 1, n)
(23)
292
P = (1, 1, n,
n, 1)1,1 ,
n)1,1 ,
Q = (1, 1, 1, n,
(24)
a = 1, 2, 3.
(25)
The superpotential is of the form:

b
W ia i [a,b] + Si1 3 Pi2 1 Qi3 2
with 6= 6= ,
(26)
where , , are again family indices, {a, b}-U (12 2n) indices, i-SO(8 2n) indices,
and {i1,2,3}-respective indices for the three U (n) factors.
There is a T-dual interpretation of this solution. Namely, T-dualizing the original model
along, say, the third complex direction X3 , corresponds to the model with 32 D7 branes
sitting at the origin of the third complex plane. The action of discrete Wilson lines implies
that one can take n sets (n = 1, . . . , 4) of branes (each set containing six D7 branes) to be
at one of the two Z3 orbifold fixed points, located away from the origin. (This motion can
be accomplished in sets of six D7 branes; of six since each such set is moded out by six
elements of the combined Z3 and group action.) Note that the string states associated
with branes located at one Z3 orbifold fixed point, away from the origin (in a particular
complex plane), are related, by orientifold projection, to complex conjugate states of branes
located at the other fixed point away from the origin.
Continuous Wilson lines
As the next step we generalize the construction to the case of continuous Wilson lines.
We are still after a unitary representation W that satisfies conditions (13)(14), except that
now it need not commute with Z3 .
For convenience, we again rearrange the entries in the Z3 matrix,

Z3 = diag 2 16n , 16n , 14n , 1n Z; 16n , 2 16n , 14n , 1n Z . (27)
The ansatz for the Wilson line is taken to be of the form:

,
W = diag 1163n , W0 ; 1163n, W0
(28)
where W0 is a (3n 3n)-dimensional unitary matrix. Such a Wilson line can be uniquely
determined up to unitary transformations that commute with Z3 . The part of such unitary
transformations that affects W0 is of the form: Un U30 , where Un is a general (n n)dimensional unitary matrix and U30 is a (3 3)-dimensional diagonal unitary matrix.
Employing such transformations in turn enables one to cast W in the following most
general form:

(29)
W = diag 1163n , W1 , . . . , Wn ; 1163n, W1 , . . . , Wn ,
where Wi are (3 3)-dimensional unitary matrices subject to the following consistency
conditions:

(30)
(ZWi )3 = 13 .
Tr(ZWi ) = Tr ZWi2 = 0,
We have reduced the problem to finding an explicit representation of the matrices Wi .
Starting with a general ansatz:
293
w1 a b
W = a 0 w2 c ,
b 0 c0 w3
(31)
the conditions (30) yield the following general form:
w
a
b
a0
w+x
c
,
W0 =
aa 0 + wx aa 0 2 x(x + w)
2
w x
b
c
(32)
where a, b, c, w, x are complex numbers subject to the constraint det(W ) = 1. However,

unitarity imposes an immediate constraint x = 0; this is due to the fact that the co-factors
of the diagonal entries in the above matrix are all equal to w(w + x) aa 0 , and due to
unitarity they should be proportional to the complex conjugate diagonal entries in (32).
The unitarity conditions (equating W entries with the corresponding co-factors of W )
allow one to further constrain the parameters:
q
|w| = 1 |a|2 |a 0 |2 .
(33)
|c| = |a|,
|b| = |a 0 |,
0
At this point, for the sake of simplicity, we
shall change the notation (|a|, |a |, |w|)
0
2
02
(a, a , w) > 0, subject to the constraint w = 1 a a .
To further determine the phases a , b , c , a 0 , w , we introduce
A = 2a 0 + a + b c ,
(34)
B = 2a + a 0 b + c ,
(35)
= w + a + a 0 ,
(36)
and reduce the remaining unitarity conditions to the following set of equations:
aa 0wei + a 03 eiA = a 02 ,
0
aa we
0
aa we
+a e
3 iB
+w e
3 3iw
(37)
=a ,
(38)
=w .
(39)
((ZW )3 = 13 is automatically satisfied, provided (37)(39) hold; namely, (37) + (38) +

(39) is precisely the condition det(W ) = 1.) The solution of (37)(39) gives:
(a 4 + a 2 a 02 + a 04 + a 02 a 2 )
,
2a 03
(a 4 + a 2 a 02 + a 04 a 02 + a 2 )
,
cos(B ) =
2a 3
(w4 + w2 a 2 a 02 )
,
cos(3w ) =
2w3
cos(A ) =
(40)
(41)
(42)
as well as
cos() =
(a 4 + a 2 a 02 + a 04 a 02 a 2 )
.
2aa 0w
(43)
294
Two additional constraints of Eqs. (37)(39) turn out to be automatically satisfied. In

addition, the phases A , B , w and are not independent, i.e.,
3 = 3w + A + B .
(44)
One can show that the expressions (40)(43) indeed ensure cos(3) = cos(3w + A +
B ). (In proving this identity, the relationships: a 03 sin(A ) = a 3 sin(B ) = w3 sin(3w ) =
aa 0 w sin() that follow from (37)(39) are useful.)
Thus we have arrived at the following form of the matrix W :
aeia
a 0 eib
weiw
a 0 eia0
weiw
aeic .
(45)
W =
i(a +a 0 b ) 0 i(a +a 0 c )
ae
ae
weiw
It is determined up to diagonal unitary transformations U30 = diag(ei1 , ei2 , ei3 ), which
is the most general unitary matrix that commutes with Z, and thus provides an equivalency
class for the Wilson line representations. Thus, two phase parameters in W can be gauged
away, and the only remaining phases are the two gauge invariant phase parameters
A,B and the phase of the diagonal element w , all specified, by Eqs. (40)(42), in terms
of two real parameters, up to signs and multiples of 2 . The latter are further constrained
by (37)(39) and (44).
To summarize, the final form of (45) is specified in terms of three real, positive
parameters w, a, a 0 , subject to the constraint w2 + a 2 + a 02 = 1. (Equivalently the Wilson
line can be specified in terms of the two Euler angles and of the three-sphere,
introduced as: w = cos , a = sin cos , a 0 = sin sin .)
To study the monodromy properties of this Wilson line, it is convenient to solve (42) for
cos w , which has three roots:

1 i0
e + ei0 ,
2

1
cos(w )2 = ei0 + 2 ei0 ,
2

1
cos(w )3 = 2 ei0 + ei0 ,
2
where

1/3
p
(w4 + w2 a 2 a 02 )
, A
.
ei0 A + i 1 A2
2w3
cos(w )1 =
(46)
(47)
(48)
(49)
For the limit A = 1, i.e., w = 1, a = a 0 = 0, these solutions reduce to w = 0, , 2 ,

which are respectively the case of no Wilson line, or discrete Wilson lines, corresponding
(in the T-dual picture) to the set of D7 branes sitting at the origin or at one of the two fixed
points. The continuous Wilson line interpolates continuously between these limits. For that
purpose one has to find a path in the space of w, a, a 0 (or equivalently the Euler angles
and ) in which 30 varies from 0 to 2 . It is straightforward to find such paths. In
particular, 30 = 0 and 30 = correspond to A = +1 and 1, respectively, which occur
on the boundaries of the allowed region at the points a = a 0 = 0 and a = a 0 = 23 . There is
295
a continuous path between these, which passes 30 = 2 (A = 0) at, e.g., a = a 0 = 1 1 .

3
From A = 1 one can always move along either branch of the square root, allowing a
continuous interpolation, e.g., from 30 = 0 to 30 = and then to 30 = 2 and further.
For example, 3w can start from 0 and increase first to 2 and then to 4 as one moves
back and forth between A = +1 and A = 1. An examination of (37)(44) reveals that
A and B each decrease by as 3w increases by 2 . As 3w passes through 2 , A,B
and must increase discontinuously by 3 and 2 , respectively, in order to preserve the
signs of the angles. This occurs at a = a 0 = 0, where the phases are indeterminant, so the
changes in the Wilson line parameters are continuous. It is convenient to use the freedom
4
in U30 to require b = c , in which case an increase of w by 2
3 (or 3 ) is accompanied
by an increase in a,a 0 by the same amount. From (30) it is clear that if W is a continuous
Wilson line solution, then so are W and 2 W . We thus see that these solutions, related
by the discrete Z3 symmetry for fixed a and a 0 , can actually be related by a continuous
interpolation as 3w varies from 0 to 2 to 4 as a function of a and a 0 .
The surviving gauge group is generically:
U (12 2n) SO(8 2n) U (1)n ,
(50)
as long as W1 6= W2 6= Wn and (ai , ai0 ) 6= 0 (i = 1, . . . , n). However for special values

of the Wi parameters additional gauge enhancement can take place. In particular, W1 =
W2 = Wk (k 6 n) yields the gauge group enhancement of U (1)n to U (k) U (1)nk ,
with an obvious generalization to two sets W1 = W2 and W3 = W4 .
This general continuous Wilson line reduces to a hybrid (continuous/discrete) Wilson
line if k W matrices become diagonal with elements (discrete Wilson lines) (or
equivalently, 2 ) and the remaining (n k) W -matrices remain off-diagonal (continuous).
The gauge group U (1)n now becomes enhanced to U (k)3 U (1)nk .
The Wilson line (29) has a T-dual interpretation in terms of n sets of six D7 branes
each moving in, say, the third complex plane away from the orbifold fixed at the origin.
In particular, when a subset of k Wilson line elements become discrete, the T-dual picture
corresponds to k-sets of six D7 branes sitting at one (of the two) orbifold fixed point away
from the origin.
The above construction of the Wilson lines for the Z3 orientifold provides a generalization of the discrete Wilson lines (with n = 4) discussed in [8]. The continuous Wilson
line considered there corresponds to the case with W1 = W2 = W3 = W4 . Here, we have
generalized and given an explicit unitary representation of Wi in terms of two parameters.
(When one imposes the unitarity constraint on the symmetric matrix in [8], there is a relation between the magnitude and phase of the complex parameters, and the matrix becomes
a special one parameter case (with a = a 0 ) of the continuous Wilson described above.) In
the T-dual picture the position of the branes (in the third complex plane) should be parameterized by two real parameters, and thus for the full one-to-one correspondence with the
continuous Wilson line picture the Wilson line should depend on two real parameters (a
and a 0 ) as well.
296
Multiple continuous Wilson lines

As the last step we proceed with the construction of multiple Wilson lines. On general
grounds such Wilson lines are of the form (29) with the Wi s being of the form (45). The
elements in (45) are uniquely specified in terms of two real parameters, except for two
phases. As discussed above, for a single Wilson line these two phases can be removed
(gauged away) by a diagonal unitary transformation U30 that commutes with Z.
For the additional Wilson lines WJ one cannot gauge away the two phases. However,
the Wilson lines have to commute:
[W , WJ ] = 0.
(51)
This condition should in principle fix the undetermined phases. We checked that this is
indeed the case. We chose the first Wilson line W by fixing the gauge, i.e., choosing a
specific U30 , so that the phases for the matrices W are fixed as:
a = a 0 = b .
(52)
Then, if the phases for the corresponding matrices WII in the second Wilson line WII are
chosen as
cII = aII + c a ,
aII0 = bII ,
(53)
the two Wilson lines commute.

One can introduce the third Wilson line WIII , with the phases subject to the analogous
constraint:
cIII = aIII + c a ,
0 = b .
aIII
III
(54)
Such a Wilson line turns out to commute both with W and WII . (Of course, one still has
the freedom to make an overall unitary transformation U30 on all three Wilson lines.)
The additional Wilson line WII has again a T-dual interpretation. (The introduction
of additional Wilson lines does not break the generic gauge group (50).) T-dualizing the
original model along two, say, the second X2 and third X3 , complex directions corresponds
to the model with 32 D5 branes sitting at the origin of the second and third complex plane.
The action of continuous Wilson lines WII and W then corresponds to the independent
motion of n sets of six D5 branes in the second and the third complex planes, respectively.
Since each W and WII are fully specified by two real parameters, these parameters are in
one-to-one correspondence with the motion of the (n) sets of branes in the two complex
planes. Again, when any Wilson line element becomes diagonal (discrete) the solution
corresponds to a particular set of branes reaching the orbifold fixed point in the particular
plane.
The T-dual model, in which one has dualized all three complex directions X1,2,3 ,
corresponds to the C 3 /Z3 model of 32 D3 branes sitting at the origin. The introduction
of the third Wilson line WIII , which is also uniquely specified by two real parameters
for each WIII matrix, parameterizes the independent motion of n sets of (six) D3 branes
along the first plane (along with the independent motions in the second and third planes,
parameterized by WII and W , respectively).
297
3. Field theory realization

3.1. Classification of F- and D-flat directions
In this section we classify the F- and D-flat directions of the new supersymmetric
ground states that correspond to the deformation of the original model (and whose string
theoretical construction we provided in the previous section). We also show how to deform
from the discrete Wilson line solutions corresponding to n sets of branes located at the
orbifold fixed points away from the origin. For that purpose we utilize the one-to-one
correspondence [23] of D-flat directions with holomorphic gauge-invariant polynomials
(HIPs) built out of the chiral fields in the model. The constraints of F-flatness further
require that hW/p i = 0 and hW i = 0 for all of the massless superfields p in the
model. (The detailed analysis of the blown-up Z3 orientifold was given in [14], using this
technique. 1 )
We first construct a gauge invariant polynomial from the non-Abelian fields, which
is a sum of monomials involving the components of the fields. Then one monomial
term defines a D-flat direction. Each field in the monomial will typically have the same
vacuum expectation value (VEV). The D-flat constraints for both diagonal and off diagonal
generators of the non-Abelian gauge group are automatically satisfied. Other flat directions,
e.g., those with different phases for the VEVs of the fields in the monomial, are gauge
rotations of the original monomial.
One can also consider D-flat directions with more than one independent VEV, formed
as products of other HIPs. The flat directions correspond to products of monomials from
each of the HIPs, each with its own VEV. (See [14] for more detailed discussion of such
issues as overlapping HIPs, involving products of HIPs which have common multiplets.
Here, it suffices to check D-flatness for each case.)
3.2. Flat directions corresponding to Wilson line solutions
Unlike the blowing-up procedure [1416] in which the blow-up introduced Fayet
Iliopoulos terms [22] for the anomalous U (1) terms, and thus the supersymmetric
ground state solutions were achieved by HIPs which have non-zero U (1) charges, the
deformations corresponding to the continuous Wilson lines correspond to the polynomials
that are gauge invariant under the anomalous U (1) as well.
The D-flatness condition for SO(8) is
X X a

i AIij ja = 0,
(55)
DI =
,a i,j
1 These techniques were previously developed (see [24]) to construct the moduli space of the flat directions for
models based on perturbative heterotic string models. For simplicity, the flat direction analysis in [24] considered
only the non-Abelian singlet fields in the model, in which case the flat directions correspond to gauge invariant
holomorphic monomials. In the present model, the D-flat directions necessarily involve non-Abelian fields due to
the matter content.
298
where AI are generators of the vector representation of SO(8) and I = 1, . . . , 28. For
U (12),
X X a
XX
J
J b
i Tbab
i +
Tab [b,c] ,
(56)
[a,c]
DJ =
a,b,c
,i a,b
where T J
bJ
(T
T J T ) are the generator matrices for the fundamental (anti-fundamental)
representation of U (12) and J = 1, . . . , 144. (In the blown-up case one must add a constant
FayetIliopoulos term FI to the D-term for the anomalous U (1) [14].)
The F-flatness conditions of the original Z3 orientifold model are
b
ia i = 0,
i [a,b] = 0.
(57)
For the case of the original Z3 orientifold, one can construct D- and F-flat directions
from HIPs of the form ()(), in which each factor is separately U (12) invariant,
and only the product is SO(8), invariant, i.e.,
c d
i j [c,d] .
(58)
ia jb [a,b]
F-flatness is ensured by taking all fields from the same family, e.g., = 3 for definiteness.
It is necessary to consider a 6th order polynomial because the cubic () vanishes for a
single family index due to the symmetry (antisymmetry) in SO(8) [U (12)] indices. A flat
direction corresponds to a specific monomial in (58), e.g., to
31 32 3
3
1 2 [1,2] .
(59)
131 232 [1,2]
3
, with each of the three fields having the common
This is just the square 2 of 131 232 [1,2]
3
)2 .
VEV v, and will henceforth be denoted by (131 232 [1,2]
The direction (59) appears to break SO(8) U (12) to SO(6) U (10). In fact, there is
an additional surviving U (1) generated by a combination of the broken SO(8) and U (12)
generators. To see this, consider a concrete representation of non-Hermitian SO(8) and
U (12) generators labeled by I = (kl) and J = (cd), i.e., Akl and Tdc , with matrix elements
(Akl )ij = j k il j l ik ,
and
Tdc

ab
= ad bc ,
(60)
(61)
respectively. It is then straightforward to see that the Hermitian generator F12 , defined by

j
(62)
Fij = i Tji Ti + Aij ,
is conserved.
Several generalizations of (59) are possible. One can consider a product of flat directions
of the following schematic form:
n
Y
()2i ,
n = {1, . . . , 4},
(63)
i=1
2 It is easy to check that (58) is only D-flat for [a, b] = [c, d]. For example, ( 31 32 3 )( 33 34 3 ) is
1
2 [1,2]
1
2 [3,4]
not D-flat under U (12) because, in the terminology of [14], it involves overlapping U (12) polynomials.
299
where each factor is a sixth order polynomial analogous to (58), we have suppressed
the third family index, and each factor has a non-zero VEV vi for each field in
2i1 2i
2i1
2i 2i1,2i . (These directions are always F-flat.) For example, one can consider
3
131 232 [1,2]
2
3
333 434 [3,4]
2
(64)
for the case n = 2, where the two sets of fields have VEVs v1 and v2 , respectively. The
direction in (63) has the generic surviving gauge group
U (12 2n) SO(8 2n) U (1)n .
(65)
This flat direction provides a field theory realization of a motion of n sets of (six) D7
branes in, say, the third complex plane, away from the fixed point at the origin, and with
each set at a different location. The T-dual string theory construction in terms of a generic
continuous Wilson line, given by (29), was derived in the previous section. However, note
again that the field theory allows for a realization of such a motion of branes only in the
neighborhood of the original fixed point, i.e., the result is valid only in the power series
expansion in terms of the VEVs of the fields.
When p of the VEVs in (63) are equal 3 , the gauge factor U (1)p is enhanced to U (p).
For example, if v1 = v2 in (64), it is straightforward to show that the generators F12 , F34 ,
F13 + F24 , and F14 F23 are conserved, and form the surviving group U (2). (The U (1)
generator is F12 + F34 .) Most generally, the direction in which sets of pi of the n VEVs
P
are equal, with i pi = n = {1, . . . , 4}, leads to
Y
U (pi ),
(66)
U (12 2n) SO(8 2n)
i
in one to one correspondence with the continuous Wilson line solutions (29) with pi
(3 3)-dimensional matrices Wi equal, which in the T-dual picture realizes the the motion
of n sets of (six) D7 branes with groups of pi of them at the same position, thus providing
for an enhancement of gauge symmetries.
One can generalize the construction of flat directions to include fields with all three
family indices. For example, each factor in (63) can be generalized to a product of three
factors, one for each family with its own VEV vi but the same gauge group structure.
These directions are still F-flat provided one does not mix families within the same factor.
These solutions provide a field theoretical realization of the motion of branes in multiple
complex planes, whose T-dual string theory construction in terms of multiple continuous
Wilson line solutions was given at the end of the previous section.
On the other hand, the discrete Wilson lines (17) provided a T-dual realization in terms
of n sets of (six) D7 branes sitting at a fixed point away from the origin, in, say, the third
complex plane; the starting gauge group there is (19) with the massless particle content
3 One can show that the enhanced symmetry also holds if each pair of VEVs differs by a factor of or 2 .
e13 + F
e24 , and F
e14 F
e23 , where F
e13
For example, in (64) the conserved U (2) generators become F12 , F34 , F
v
v
v
e24 . We have not found an analog of this freedom in
i( v1 T31 v2 T13 + A13 ), v1 = or 2 , and similarly for F
2
1
2
the continuous Wilson line construction.
300
(21)(25). The flat direction corresponding to the moving of one set of branes away from
the fixed point (located away from the origin) is associated with the HIP
i
Si1 3 Pi2 1 Qi3 2 S131 P131 Q31

1 ,
(67)
where the restriction to a single family, say, = 3, ensures F-flatness. 4 This breaks the
U (n)3 symmetry down to U (1) U (n 1)3 , where the U (1) generator is the diagonal
sum t11 of three broken U (1) generators, where
tji tj(1)i + tj(2)i + tj(3)i .
(68)
In (68) t (l) , l = 1, . . . , 3, is the generator of the lth U (n) factor. Similarly, a product of
(SP Q)q with distinct gauge indices, such as, e.g.,
32 32 32
S2 P2 Q2 ,
(69)
S131 P131 Q31
1
generically breaks U (n)3 to U (1)q U (n q)3 . However, for special points with equal
VEVs for each of the factors 5 there is an enhanced symmetry U (q) U (n q)3 .
One can choose a hybrid flat direction composed of:
X
X
Y
Y
()2pi
(SP Q)qj , where
pi 6 4 n,
qj 6 n.
(70)
i
The surviving gauge group is

Y
X Y
X
pi SO 8 2n 2
pi
U (pi )
U (qj )
U 12 2n 2
i

X 3
qj .
U n
(71)
This field theory picture of course has an analogous Wilson line realization encoded
in a special choices of Wi matrices, including the choice of the specific branches for the
interpretation of the phases w .
4. Conclusions
In this paper we focused on the study of continuous Wilson lines within fourdimensional N = 1 type IIB orientifold models. We enforced unitarity and constructed
the most general set of continuous Wilson lines within the original Z3 orientifold [1] and
demonstrated that these models are in one-to-one correspondence with the T-dual models,
4 Each factor SP Q can again be replaced by a product of three factors, one with its own VEV for each family,
provided the family indices are not mixed within a factor. This corresponds to the multiple continuous Wilson
lines and in the T-dual picture to the motion of sets of branes in three complex planes.
5 There are also enhanced symmetries for the products (SP Q)q in the case in which the VEVs of each factor
have relative phases of or 2 . The generators of the enhanced symmetry are generalizations of (68) in which
v
there are corresponding phases in the coefficients of the t (l) . For example, if the two factors in (69) have v1 =
2
v (1)1
(2)1
v (3)1
or 2 , then t21 v1 t2 + t2 + v2 t2 , t12 t21 , t11 , and t22 form a conserved U (2). Again, we have not
2
1
found an analog on the string theory side.
301
where each Wilson line has an interpretation of n sets (n = 1, . . . , 4) of (six) branes moving
in one (of the three) complex planes. The number of parameters of such a continuous
Wilson line is in one-to-one correspondence with the parameters that specify the location
in the complex plane of each set of branes. When a sub-block of the continuous Wilson
line becomes discrete, i.e., the sub-block that commutes with the corresponding sub-block
of the Z3 element and depends on a discrete parameter, this corresponds in the T-dual
orientifold to branes sitting at the Z3 orbifold fixed points. The generic Wilson lines break
the original gauge group U (12) SO(8) down to U (12 2n) SO(8 2n) U (1)n .
A gauge enhancement takes place for special values of the Wilson line parameters, e.g.,
when k sub-blocks are equal then U (1)k is promoted to U (k). Similarly, when the full
Wilson line becomes discrete the gauge group is enhanced to U (12 2n) SO(8 2n)
U (n)3 . The Wilson line solution continuously interpolates between the limit of no Wilson
line and the discrete solutions.
We also analysed the field theoretic analog, describing the above string constructions
as D- and F-flat deformations of the effective field theory of the original model as well
as deformations of the models with discrete Wilson lines. The field theory describes these
string solutions only in the proximity of the original models, i.e., it allows only for a powerseries expansion in the vacuum expectation values of the chiral superfields, specified by
the holomorphic gauge invariant polynomials that parameterize the moduli space of the
supersymmetric deformations of the original models. We find the explicit form of the
holomorphic polynomials, that are in one-to-one correspondence with the parameters of
the string constructions with the continuous Wilson lines (and their T-dual interpretation),
thus quantifying the correspondence between the two complementary approaches.
The work sets the stage for further investigations of models with continuous Wilson
lines. In particular, the explicit construction of the continuous Wilson lines would allow
one to construct not only the massless, but also the massive spectrum of the string models.
The dependence of the mass spectra and the couplings on the continuous Wilson line
parameters, both from the explicit string construction as well as from the (perturbative)
field theory perspective, deserves further study.
Another more general direction involves a study of a general class of four-dimensional
N = 1 type IIB orientifold models, in order to establish the general (and precise) correspondence between the models with continuous Wilson lines, their T-dual interpretation,
as well as their field theory realization [19]. In general these models contain not only D9
branes but also, e.g., D5 branes. The latter can be located at different points on a particular two-torus T 2 , where they are point-like, thus allowing for even more involved models,
implementing simultaneously moving branes and the actions of continuous Wilson lines.
Investigation of these general classes of string solutions (by determining the gauge group,
the mass spectra and the couplings) would shed light on the properties of a broad class of
open string models within symmetric type IIB orientifold constructions, and may in turn
lead to a discovery of potentially realistic open-string solutions.
302
Acknowledgements
We would like to thank Angel Uranga for many communications and suggestions
regarding the work presented in the paper as well as for collaboration on related topics.
We also benefitted from discussions with L. Faccioli, S. Katz, M. Plmacher, and J. Wang.
The work was supported in part by U.S. Department of Energy Grant No. DOE-EY-76-023071 (M.C. and P.L.) and in part by the University of Pennsylvania Research Foundation
award (M.C.).
References
[1] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 385 (1996) 96,
hep-th/9606169.
[2] M. Berkooz, R.G. Leigh, Nucl. Phys. B 483 (1997) 187, hep-th/9605049.
[3] G. Zwart, Nucl. Phys. B 526 (1998) 378, hep-th/9708040.
[4] Z. Kakushadze, Nucl. Phys. B 512 (1998) 221, hep-th/9704059.
[5] Z. Kakushadze, G. Shiu, Phys. Rev. D 56 (1997) 3686, hep-th/9705163; Nucl. Phys. B 520
(1998) 75hep-th/9706051.
[6] L.E. Ibez, JHEP 9807 (1998) 002, hep-th/9802103.
[7] D. ODriscoll, hep-th/9801114.
[8] G. Aldazabal, A. Font, L.E. Ibez, G. Violero, Nucl. Phys. B 536 (1999) 29, hep-th/9804026.
[9] Z. Kakushadze, S.H.H. Tye, Phys. Rev. D 58 (1998) 126001, hep-th/9806143.
[10] G. Shiu, S.H.H. Tye, Phys. Rev. D 58 (1998) 106007, hep-th/9805157.
[11] Z. Kakushadze, Phys. Lett. B 434 (1998) 269, hep-th/9804110; Phys. Rev. D 58 (1998) 101901,
hep-th/9806044; Nucl. Phys. B 535 (1998) 311, hep-th/9806008.
[12] M. Cvetic, M. Plmacher, J. Wang, JHEP 0004 (2000) 004, hep-th/9911021.
[13] G. Aldazabal, L.E. Ibanez, F. Quevedo, A.M. Uranga, hep-th/0005067.
[14] M. Cvetic, L. Everett, P. Langacker, J. Wang, JHEP 9904 (1999) 020, hep-th/9903051.
[15] C. Beasley, B.R. Greene, C.I. Lazaroiu, M.R. Plesser, Nucl. Phys. B 566 (2000) 599, hepth/9907186.
[16] M.R. Douglas, B. Fiol, C. Romelsberger, hep-th/0003263.
[17] M. Cvetic, L. Dixon, unpublished.
[18] M. Cvetic, in: S.J. Gates, R. Mohapatra (Eds.), Proceedings of Superstrings, Cosmology and
Composite Structures, College Park, Maryland, March 1987, World Scientific, Singapore, 1987;
Phys. Rev. Lett. 59 (1987) 1795; Phys. Rev. Lett. 59 (1987) 2829.
[19] M. Cvetic, P. Langacker, M. Plmacher, A. Uranga, J. Wang, in preparation.
[20] J. Lykken, E. Poppitz, S.P. Trivedi, Nucl. Phys. B 543 (1999) 105, hep-th/9806080.
[21] L. Ibez, JHEP 9807 (1998) 002, hep-th/9806080.
[22] M.R. Douglas, G. Moore, hep-th/9603167.
[23] A. Luty, W.I. Taylor, Phys. Rev. D 53 (1996) 3399, hep-th/9506098.
[24] G. Cleaver, M. Cvetic, J.R. Espinosa, L. Everett, P. Langacker, Nucl. Phys. B 525 (1998) 3,
hep-ph/9711178; Nucl. Phys. B 545 (1999) 47, hep-ph/9711178.

UV/IR mixing in noncommutative field theory

via open string loops
Youngjai Kiem, Sangmin Lee
School of Physics, Korea Institute for Advanced Study, Seoul 130-012, South Korea
Abstract
We explicitly evaluate one-loop (annulus) planar and nonplanar open string amplitudes in the
presence of the background NSNS two-form field. In the decoupling limit of Seiberg and Witten,
we find that the nonplanar string amplitudes reproduce the UV/IR mixing of noncommutative field
theories. In particular, the investigation of the UV regime of the open string amplitudes shows that
certain IR closed string degrees of freedom survive the decoupling limit as previously predicted
from the noncommutative field theory analysis. These degrees of freedom are responsible for
the quadratic, linear and logarithmic IR singularities when the D-branes embedded in spacetime
have the codimension zero, one and two, respectively. The analysis is given for both bosonic and
supersymmetric open strings. 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
Certain noncommutative field theories [13] can be systematically derived from open
string theories in the presence of constant background NSNS two-form field (B field)
[510]. The upshot of these developments is that noncommutative field theories are more
stringy than what one might naively expect. For example, unlike the generic commutative
field theories arising as decoupling limits of string theories, noncommutative field theories
are T-duality invariant signaling its stringy nature [4]. Further considerations of loop effects
in noncommutative field theories [1113] add an intriguing new element in the analogy
between open string theories and noncommutative field theories, namely, the UV/IR
mixing. From the open string perspective, the simplest one-loop annulus diagram reveals
a prominent stringy character, the open/closed string channel duality. The UV regime of
the open string annulus amplitudes can naturally be interpreted as the IR closed string
degrees of freedom. This behavior closely parallels the UV/IR mixing of Refs. [12,13]
E-mail addresses: ykiem@kias.re.kr (Y. Kiem), sangmin@kias.re.kr (S. Lee).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 3 0 - 2
304
Y. Kiem, S. Lee / Nuclear Physics B 586 (2000) 303314
in noncommutative field theories, where one can interpret certain UV divergences coming
from nonplanar loops of high energy virtual particles as IR divergences.
In this paper, we embark upon the detailed study of one-loop annulus open string
amplitudes in the presence of constant background B field 1 and recover many features
found in Refs. [12,13]. The one-loop open string amplitudes turn out to be of the same
form as Ref. [12], except including the contributions from the massive string excitations.
Upon taking the decoupling limit of [10], massive excitations decouple, while some UV
degrees of freedom do not in nonplanar diagrams. Since our set-up is the string theory
framework, via the standard open/closed string duality, we can unambiguously identify
these extra degrees of freedom as IR closed string contributions. Their Wilsonian effective
action decoded from the annulus amplitudes also turns out to be the same as the one
proposed in Ref. [12] for the extra degrees of freedom responsible for the IR singularities of
the one-loop noncommutative field theory amplitudes. In particular, for (D 1) D-branes
(original critical open string theory), (D 2)-branes and (D 3)-branes, open string theory
calculations reproduce the quadratic divergences [12], linear divergences and logarithmic
divergences [13] caused by the extra degrees of freedom, respectively, where D is the
dimension of spacetime. In short, noncommutative quantum field theories arising as
limits of open string theories include closed string degrees of freedom, which survive
the decoupling limit, couple linearly to the D-brane world-volume open string degrees
of freedom and live in the bulk spacetime.
This paper is organized as follows. In Section 2, we compute the world-volume
propagators on an annulus in the presence of the constant background B field. In Section 3,
we evaluate the planar and nonplanar annulus diagrams in the bosonic open string theory.
Via open/closed string duality, we identify the IR closed string degrees of freedom, which
survive the decoupling limit, and study their properties. In Section 4, we extend our
analysis to open superstring amplitudes. We discuss further directions and implications
suggested by our analysis in Section 5.
2. World-sheet propagator on an annulus

One important ingredient in computing the one-loop string amplitude is the world-sheet
propagator on an annulus. It was first obtained in [15,16] using a world sheet coordinate
in which the two boundaries of the annulus are concentric circles. Here we present an
equivalent but more concise form of the propagator following the notations of [17].
First, in the absence of the B field, consider a rectangular torus whose modulus
parameter = iT is purely imaginary. The world-sheet propagator is
0
X (z)X (w) = G(z w),
2
(2.1)
where
1 We note that our calculations have overlaps with the earlier literature on open string amplitudes, such as
Refs. [14] and [15,16].
305
Fig. 1. The world sheet coordinate for annulus.

1 (|iT ) 2 2
+

[Im()]2 .
G() = log 0
1 (0|iT )
T
(2.2)
Here, 1 is the theta function defined as

1 (| ) = i
1 2
(1)m q 2 (m+ 2 ) m+ 2 ,
m=
q = exp(2i ),
= exp(2i).
The propagators are periodic under the two lattice transformations

z z + 1,
z z + iT ,
and they satisfy the flux conversation; the integral of 2 G(z) over the torus vanishes.
To turn torus propagators to annulus propagators, we place a mirror charge at w (and
at all their lattice translation points) for a source charge at w. This operation imposes
Neumann boundary conditions along the two boundaries at Re(z) = 0 and Re(z) = 1/2,
while maintaining the periodicity in z z + iT , thereby turning the original torus to an
annulus. Written explicitly, the propagators look like
0

.
X (z)X (w) = G(z w) + G(z + w)
2
(2.3)
The propagators with Dirichlet boundary condition can also be straightforwardly written
down. However, we will not need them for we will consider only open string vertex
insertions.
Our next task is to find the explicit form of analogous expressions that are valid when
B 6= 0. As noted in [18], one can bring the B field into a block-diagonal form and
consider each 2 2 block separately. Suppose for now that we turn on the B12 = B along
306
X1 = X and X2 = Y directions parallel to the D-branes under consideration. The boundary

conditions at Re(z) = 0 and Re(z) = 1/2 should be modified into 2

n X + iBt Y = 0Re(z)=0,1/2.
(2.4)
The answer is:

B 2 4
1 B2
2
[Re(z + w)]
2,
G(z
+
w)
+
X(z)X(w)
=
G(z
w)
+
0
1 + B2
1 + B2 T

4i
1 (z + w)
2
2B
+
Re(z
+
w)Im(z
+
w)
.
log
Y
(z)X(w)
=
0
1 (z + w)
T
1 + B2
(2.5)
(2.6)
When |z|, |w| 1, T , the quadratic terms are negligible and 1 (z|iT ) reduces to z, so that
we recover the propagators on a disk, which was obtained, for example, in Refs. [8,10,18].
The coefficients of the -function terms are uniquely determined by comparison with the
propagator on the disk. The quadratic terms are required by the periodicity in z z + iT
and flux conservation. 3
For the computation of the open string amplitudes, we insert open string vertex operators
along the boundaries at Re(z) = 0 and Re(z) = 1/2. For later convenience, we note that
for the planar insertions, the propagators become

X(0)X(iy) = 0
1
G(iy),
(2.7)
1 + B2

B
(y),
(2.8)
X(0)Y (iy) = i( 0 )
1 + B2
where we introduced the Heaviside step function (y) = y/|y|. For nonplanar insertions,
we have
0 B 2
1
G(1/2 + iy) +
,
2
1+B
2T 1 + B 2

B y
.
X(0)Y (1/2 + iy) = i(2 0 )
1 + B2 T
The quadratic term in (2.5),

X(0)X(1/2 + iy) = 0
(2.9)
(2.10)
2B 2 2
[Re(z + w)]
2,
(2.11)
1 + B2 T
deserves further comments. We note that (2.11) distinguishes between the planar and
nonplanar vertex insertions. In particular, it vanishes for planar insertions along Re(z) = 0
but gives a non-vanishing contribution for nonplanar insertions where one vertex is
separated from another. We will find that it will give finite contributions to the amplitudes
under the decoupling limit of Ref. [10] thus surviving in the effective noncommutative field
theory.
2 As in [15,16], it is possible to trade the quadratic terms of (2.5) with modified boundary conditions involving
a constant term on the right hand sides of (2.4). Our choice in this paper is to keep the boundary conditions (2.4)
intact.
3 The periodicity in this context means the periodicity of the physical objects, such as hXY i. We note that
hXY i itself is not periodic, but this does not give an ambiguity when computing physical amplitudes.
307
3. One-loop bosonic open string amplitudes

Utilizing the world-sheet propagators of Section 2, we explicitly evaluate the annulus
amplitudes, inserting two open string vertex operators along the world-sheet boundaries in
the presence of D-branes. In the following, we will ignore all numerical constants in the
overall normalization of the amplitudes, but the dimensions and dependence on coupling
constant will be unambiguous.
3.1. Planar and nonplanar bosonic open string amplitudes on an annulus
The one-loop amplitude in the presence of an (n 1)-brane is given by
Z
A=
dT
Z(T )
T
ZT
ZT
dy1
0
2 0
=g
dy2 V1 (p, y1 )V2 (p, y2 ) T
(3.1)
dT
(2 0 T )n/2 f1 (q)24T
T
ZT
dy I (p; y, T ).
(3.2)
The partition function part is computed in the same way as in [19];

Z
dn k X 2 0 T (k 2 +M 2 )
I ,
e
Z(T ) =
(2)n
(3.3)
f1 (q) = q 1/24
(1 q m ),
q = e2T .
(3.4)
m=1
The vertex operators for a tachyon and a gauge boson are, respectively,
VA = g XeikX ,
VT = g 0 eikX ,
(3.5)
where the coupling constant g is the one that appears in the low energy effective action of
open strings. Schematically,
Z

(3.6)
S = dn x ()2 + m2 2 + g 2 4 + .
A simple dimensional analysis shows that g is related to string coupling by g 2 =
( 0 )n/22 gst .
We first give the answers for the planar amplitudes. When B = 0, for the tachyon
tachyon insertion, we have

1 (iy|iT ) 2
exp 2 y 2 ,

(3.7)
I = 0
1 (0|iT )
T
while for gauge bosongauge boson insertion, we have
0 2

I = 1 2 J 00 + 0 (1 p)(2 p)(J 0 )2 e p J

0 2
= 0 (1 2 )p2 (1 p)(2 p) (J 0 )2 e p J + (total derivative in y),
(3.8)
308
where we define J (y) = G(iy). When we turn on the B field (B 6= 0), the answer is exactly
the same as the one for B = 0, except that the external momentum squared is evaluated with
respect to the open string metric G defined in Ref. [10] as

(3.9)
G B1 B .
We now consider the nonplanar insertions where B 6= 0 effect is conspicuous. For the
tachyontachyon insertions, we get

2 (iy|iT ) 2
exp 2 y 2 exp p p ,
(3.10)
I = 0
1 (0|iT )
T
2 0 T
and the gauge bosongauge boson insertions yield

0 2 (1 p)(2 p)

0
2
I = (1 2 )p (1 p)(2 p) (K ) +
0 T 2

pp
,
exp 0 p2 K
2 0 T
(3.11)
where we define K(y) = G(1/2 + iy). We introduced -product and -product following
Ref. [12] as
1
p = p ,
p p p (G) p ,
4
where is the noncommutativity parameter defined in [10],

2 0 ( + B)1 B( B)1 .
(3.12)
(3.13)
The sign in the definition of p p is introduced to make it nonnegative.

The B dependence in the amplitudes come from two combinations p p and p. Due
to the prefactor 1/T in front of p p, the effect of noncommutativity becomes stronger
as we approach the UV corner of the moduli integral. For the higher spin world-volume
fields, there are polarization dependences, as exemplified in p for the gauge boson
amplitudes.
3.2. Open/closed string duality and the decoupling limit
For the rest of this section, we only consider the tachyon amplitudes in detail for
simplicity. We first review the well-known world-sheet duality for B = 0 to contrast it
with the B 6= 0 situation. The relation between the nonplanar tachyontachyon amplitude
Z
A=
0
dT
(2 0 T )n/2 f1 (q)24 T
T
ZT
0

2 (iy|iT ) 2
2 2

y
dy 0
exp
1 (0|iT )
T
(3.14)
and the one-loop amplitudes in the field theory of the type (3.6) is manifest in the region
T 1, where we can expand (3.14) in e2T . For example, when y T , the open string
diagram looks very much like the field theory diagram in Fig. 2(a). This intuitive picture is
confirmed by an explicit calculation which shows that
A = g2 0
X
I
Z
aI
X
dT
0
2
(2 0 T )n/2 T e2 T MI =
T
dn k
309
aI g 2
,
k 2 + MI2
(3.15)
where aI are some numerical coefficients.

In the opposite end T 1, the usual channel duality allows us to rewrite (3.14) from the
point of view of the closed strings. In particular, using the modular transformation of the
theta functions, we find
Z
A=
0
dS n/212
S
f1 (q)
18
S
Z1

2
dx 4 (x|iS) ,
(3.16)
where S = 1/T and q = exp(2S). The picture now is a closed string connecting the
open string states as in Fig. 2(b). For the case of the spacetime filling 25-brane, (3.16) can
be expanded to give
X bJ 2
X Z
0 2
2
bJ dS e2S (p +MJ )/4
(3.17)
A=
p2 + MJ2
J
J
for some numerical constants bJ . 4 The coupling constant appears in the low energy
effective action of the form
Z
(3.18)
S = dn x ,
where is a closed string field.
The noncommutative field theory arises in the decoupling limit 0 0 while keeping
G and fixed [10]. We note that in the bosonic string theory, the mass spectrum is
known to be 0 MI2 = NI 1 (open) and 0 MJ2 = 4(NJ 1) (closed) for nonnegative
Fig. 2. Open/closed string channel duality.

4 When computing perturbative string amplitudes, the external momenta are always put on-shell. Here we are
assuming that the final expression holds for off-shell amplitudes as well.
310
integers NI and NJ . In this limit, therefore, if we ignore the tachyons, all but the
contribution from massless intermediate states disappear, as can be seen from (3.17). When
B = 0, the massless intermediate degrees of freedom give a trivial IR divergence that
should be cancelled with other divergences via FischlerSusskind type mechanism. When
B 6= 0, their contribution is non-trivial as we will see shortly; we need to take the Wilsonian
point of view regarding the cutoff and the effective degrees of freedom.
3.3. What survives the decoupling limit
The key issue is to identify the contributions from the T 1 UV regime to the open
string moduli integral when B 6= 0. For this purpose, in the spirit of string field theory
[2022], we explicitly introduce a short distance UV regulator 1/2 in the open string
description; the regulated open string contribution comes from the region of the moduli
space where 2 0 T > 1/2 . Then, as depicted in Fig. 3, the possible extra UV degrees of
freedom originating from the extreme UV open string loops should come from the corner of
moduli space where 0 < 2 0 T < 1/2 . When 2 0 T goes below the UV cutoff, we have
a factorization channel where the original nonplanar annulus diagram becomes two string
states connected by a long closed string tube. 5 Via open/closed string channel duality, it is
natural to investigate 0 < 2 0 T < 1/2 corner of the open string moduli space in terms of
the closed string picture. We thus resort to the nonplanar amplitude expression in the closed
string channel, Eq. (3.16). In this channel, the open string UV cutoff 1/2 transforms to the
closed string IR cutoff 2 . As shown in Fig. 3, the open string UV regime gets mapped to
the closed string IR regime S/(2 0 ) > 2 . The contribution to the nonplanar amplitude
from these IR closed string degrees of freedom can be computed as
Z
AIR = A() A() '
dS n/212 ppS/(2 0)
2
0
S
e(pp+1/ )S/(2 ) , (3.19)
e
S
where the IR regulated closed string amplitude A() is defined as
Fig. 3. Domain of moduli integral.

5 In [23], the same amplitude was computed using the openclosed string field theory for B = 0. Division of
the moduli space into two connected parts is inherent in their formalism.
Z
A() '
dS n/212 (pp+1/2)S/(2 0)
S
e
,
S
311
(3.20)
explicitly introducing the cut off 2 . The A() defined in (3.20) is the IR regulated
amplitude where we restrict the closed string moduli integral to the distance up to the IR
cutoff scale 2 . In (3.19), we wrote down only the part of the amplitude that survives the
decoupling limit and we deleted the tachyonic intermediate contribution. Reinstating the
tachyonic contribution would produce the negative eigenvalue for the quadratic effective
2
in (3.19), indicating
action for the small value of p p by shifting p p into p p +Mtachyon
the tachyonic instability [12]. For the spacetime filling 25-brane (n = 26), we have

1
1
2 ( 0 )2
(3.21)
=
AIR = 2 ( 0 )2
2
p p p p + 1/
p p + 2 (p p)2
from (3.19). From the long tube IR closed string picture of Fig. 2, we find that (3.21)
is nothing but the propagator of the extra degree of freedom multiplied by the coupling
constant. From the low energy effective description point of view, the extra degree of
freedom (denoted as field) then has the effective Lagrangian of the form
Z
Z

26
2
2
(3.22)
dx + ( ) + dn x ,
where is a generic world-volume open string scalar field. From our derivation, it is clear
that field with the effective action (3.22) gives the effective description of the long tube
IR closed strings at low energies. The effective action (3.22) is identical to the one found
in the noncommutative field theory one-loop analysis [12].
For the codimension one 24-brane, (3.19) yields
2 0

2 0

p
,
(3.23)
AIR =
pp
p p + 1/2
while for the codimension two 23-brane, we get

AIR = 2 log(p p) log(p p + 1/2 ) .
(3.24)
We note that (3.23) and (3.24) are the same as the 1PI amplitudes found in Ref. [13] for
the extra low energy degrees of freedom. From our derivation, it is clear that they also
represent the IR closed string degrees of freedom; they live in the bulk spacetime while
the open string degrees of freedom are confined on a codimension one and two D-brane,
respectively. The extra dimensions found in Ref. [13] are indeed spacetime dimensions
transversal to the brane, at least when the noncommutative field theory under consideration
derives from the decoupling limit of open string theory.
4. One-loop open superstring amplitudes

In this section, we repeat the calculations of the previous section for the case of open
superstrings. The main finding that the spacetime supersymmetry makes the two point
312
amplitudes vanish regardless of the value of B is consistent with Ref. [24]. In Ref. [24],
it was argued that for the supersymmetric gauge theories with sixteen supercharges, the
noncommutative IR singularities do not show up. In our present context with parallel
D-branes, we clearly have sixteen supercharges.
The answer for the amplitude is given by
8 ZT

Z
4
X

dT
a
0
n/2 fa (q)
(2 T )
(1)
T dy V1 (p, y)V2 (p, 0) ,
A=
T
f1 (q)
a=2
(4.1)
where the index a labels spin structures. For the definitions of fa (q), see Ref. [19]. The
vertex operators for massless gauge bosons are
V (p, y) = (X + ip )eipX .
(4.2)
In addition to the terms in Eqs. (3.8) and (3.11), we have the contraction of four
world-sheet fermions. After the summation over spin structures, however, the two point
function completely vanishes due to the Jacobis fundamental formulae [25]. Naturally,
one attributes this property to the spacetime supersymmetry. In fact, only the terms with
eight or more world-sheet fermions give nonzero contribution.
5. Discussions
The term (2.11) should be present in the world-sheet propagators as a consequence
of the boundary conditions (2.4). Its contribution to nonplanar amplitudes is the crucial
exponential factor exp((p p)/(2 0 T )) necessary for the emergence of IR closed string
degrees of freedom. Its existence, however, might seem rather puzzling from the open
string theory point of view; the direction Re(z) is the spatial direction along which the
open string lies. Therefore, it implies that there is a term in the mode expansion of X,
which is linearly proportional to Re(z). For closed strings, this would usually signal the
presence of a non-trivial winding state, while we are apparently considering open strings.
This behavior, however, is consistent with Ref. [26] where the thermodynamic evidence
for the winding states in noncommutative field theories is given. As was shown from the
calculations in Section 3, its contribution to amplitudes is from the corner of the open string
moduli space where the dual closed string interpretation is appropriate. From this point of
view, the winding states of Ref. [26] represent nothing but closed string states. We note
that the authors of Ref. [26] do not find any wrapping states, which would correspond to
higher extended objects than strings.
For the commutative field theories, the amplitude AIR becomes a simple divergence
involving the cutoff 2 without the momentum dependence. In realistic models, via the
FischlerSusskind mechanism, the cutoff dependence in open string channel gets cancelled
by closed string sigma model divergence, ultimately resulting the vanishing beta function.
In fact, the dual supergravity background geometries of commutative field theories have
the asymptotic isometry group isomorphic to the conformal group, a familiar AdS/CFT
correspondence [2729]; in this context, we consider the perturbations around a conformal
313
fixed point. On the other hand, the conjectured dual supergravity backgrounds of the
noncommutative field theories do not have the asymptotic isometry group isomorphic
to the conformal group [30,31]. The emergence of the non-trivial IR closed strings in
noncommutative field theories appears to be related to the nonconformality of the worldvolume theory. Naturally, the detailed investigation of the FischlerSusskind mechanism in
the noncommutative context along the line of Ref. [32] should reveal interesting physics.
The noncommutative field theory calculations mimic the open string calculations to a
remarkable degree, as shown from the analysis in this paper. By turning the viewpoint
around, one might consider using the noncommutative field theory as a useful guide that
provides us with a systematic organization tool for the open/closed string loop diagrams.
In this spirit, the disentangling of higher loop diagrams in noncommutative field theories
via open string perturbation theory should be an exciting venue, especially in relation to
string field theory [10,20].
Note added
After the completion of the first version of this paper, Ref. [34] appeared. After reading
[34], we found some calculational errors in Section 3.1 for the gauge boson amplitudes in
the original version of our paper. The corrected calculation revealed the (1 p)(2 p)
part in Eq. (3.11). As noted in [34], some on-shell string calculations can be extended to offshell in field theory limits. Reinstalling the part that vanishes in on-shell, the [(1 2 )p2
(1 p)(2 p)] part, in Eqs. (3.8) and (3.11), we find that our gauge boson amplitudes
are identical to the ones given in Ref. [34]. We emphasize that our boundary propagator
expressions, Eqs. (2.7)(2.10), are identical to Eqs. (2.42)(2.45) in [34] obtained by using
the boundary state formalism. To explicitly see this, we need variable changes 2y =
log |/ 0 | and 2T = log k.
Acknowledgements
We are grateful to Seungjoon Hyun, Kimyeong Lee, Hyeonjoon Shin and Piljin Yi for
useful discussions. Y.K. would like to thank Chang-Yeong Lee for useful correspondence,
especially for bringing Ref. [26] to our attention. While this paper was being written,
Ref. [33] came up, which has some overlap with the material presented in Section 2 and
Section 3.1.
References
[1] A. Connes, Noncommutative Geometry, Academic Press, 1994.
[2] A. Connes, M. Rieffel, in: Operator Algebras and Mathematical Physics, Iowa City, Iowa, 1985,
Contemp. Math. Oper. Alg. Math. Phys. 62 (1987) 237.
[3] A. Connes, M.R. Douglas, A. Schwarz, J. High Energy Phys. 9802 (1998) 003, hep-th/9711162.
[4] M.R. Douglas, C. Hull, J. High Energy Phys. 9802 (1998) 008, hep-th/9711165.
314
[5] M.M. Sheikh-Jabbari, Phys. Lett. B 425 (1998) 48, hep-th/9712199; Phys. Lett. B 450 (1999)
119, hep-th/9810179.
[6] Y.-K.E. Cheung, M. Krogh, Nucl. Phys. B 528 (1998) 185, hep-th/9803031.
[7] C.-S. Chu, P.-M. Ho, Nucl. Phys. B 550 (1999) 151, hep-th/9812219; hep-th/9906192.
[8] V. Schomerus, J. High Energy Phys. 9906 (1999) 030, hep-th/9903205.
[9] F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, hep-th/9803067; J. High Energy Phys. 02 (1999)
016, hep-th/9810072; hep-th/9906161.
[10] N. Seiberg, E. Witten, J. High Energy Phys. 9909 (1999) 032, hep-th/9908142.
[11] T. Filk, Phys. Lett. B 376 (1996) 53.
[12] S. Minwalla, M. Van Raamsdonk, N. Seiberg, hep-th/9912072.
[13] M. Van Raamsdonk, N. Seiberg, hep-th/0002186.
[14] E.S. Fradkin, A.A. Tseytlin, Phys. Lett. B 163 (1985) 123.
[15] C.G. Callan, C. Lovelace, C.R. Nappi, S.A. Yost, Nucl. Phys. B 288 (1987) 525.
[16] A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 (1987) 599.
[17] S. Lee, S.-J. Rey, Nucl. Phys. B 508 (1997) 107, hep-th/9706115.
[18] S. Hyun, Y. Kiem, S. Lee, C.-Y. Lee, Nucl. Phys. B 569 (2000) 262, hep-th/9909059.
[19] J. Polchinski, hep-th/9611050.
[20] E. Witten, Nucl. Phys. B 268 (1986) 253.
[21] B. Zwiebach, Nucl. Phys. B 390 (1993) 33.
[22] J. Khoury, H. Verlinde, hep-th/0001056.
[23] T. Asakawa, T. Kugo, T. Takahashi, Prog. Theor. Phys. 102 (1999) 427, hep-th/9905043.
[24] A. Matusis, L. Susskind, N. Toumbas, hep-th/0002075.
[25] E.T. Whittaker, G.N. Watson, A Course of Modern Analysis, Cambridge Univ. Press, 1927,
p. 467.
[26] W. Fischler, E. Gorbatov, A. Kashani-Poor, S. Paban, P. Pouliot, J. Gomis, hep-th/0002067.
[27] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200.
[28] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[30] A. Hashimoto, N. Itzhaki, Phys. Lett. B 465 (1999) 142, hep-th/9907166.
[31] J.M. Maldacena, J.G. Russo, J. High Energy Phys. 9909 (1999) 025, hep-th/9908134.
[32] D. Berenstein, R.G. Leigh, Phys. Rev. D 60 (1999) 106002, hep-th/9904104.
[33] O. Andreev, H. Dorn, hep-th/0003113.
[34] A. Bilal, C.-S. Chu, R. Russo, hep-th/0003180.

The type IIA NS5-brane

Igor Bandos a,1 , Alexei Nurmagambetov a, , Dmitri Sorokin b,2
a Institute for Theoretical Physics, NSC Kharkov Institute of Physics and Technology, Akademicheskaya 1,
UA-61108, Kharkov, Ukraine

b INFN, Sezione di Padova, Via F. Marzolo, 8, 35131 Padova, Italia
Received 6 April 2000; revised 6 June 2000; accepted 19 June 2000
Abstract
The kappa-invariant worldvolume action for the NS5-brane in a D = 10 type IIA supergravity
background is obtained by carrying out the dimensional reduction of the M5-brane action. 2000
Elsevier Science B.V. All rights reserved.
PACS: 11.15.-q; 11.17.+y
Keywords: Superbranes; Self-dual gauge fields; Dimensional reduction
1. Introduction
The M5-brane plays an important role in studying properties of M-theory [1,2], the
theory of strings and associated field theories. For instance, many physically important
multibrane configurations, realized to be relevant to a brane description of non-Abelian
gauge theories [3,4] and a brane-world scenario [5], can be considered as a specific
compactification of a single D = 11 M5-brane down to lower dimensions (with or without
its subsequent T-dualization).
A direct dimensional reduction of D = 11 spacetime with an M5-brane down to tendimensional spacetime produces a so-called NS5-brane of type IIA supergravity which
has been intensively studied in relation to six-dimensional gauge theories [6] and little
string theories [7]. 3
Corresponding author. E-mail: ajn@kipt.kharkov.ua
1 bandos@kipt.kharkov.ua
2 dmitri.sorokin@pd.infn.it; On leave from Institute for Theoretical Physics, NSC Kharkov Institute of Physics
and Technology, Kharkov, 61108, Ukraine.
3 The double dimensional reduction of the M5-brane action [810] is well known to result in a type IIA D = 10
Dirichlet 4-brane. It has also been shown [11,12] how by reducing the M5-brane action one may arrive at a
duality-symmetric D3-brane action.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 8 - 9
316
I. Bandos et al. / Nuclear Physics B 586 (2000) 315330
The verification of the quantum consistency of M-theory requires, in particular, finding

a mechanism of anomaly cancellation in the presence of M5-branes. It has been shown
that the anomaly problem has a natural solution in the case of D = 10 NS5-branes [13,14],
while in the case of the D = 11 M5-brane the situation is much more subtle and requires
additional study [1318]. Mechanisms for the M5-brane anomaly cancellation proposed
recently in [15,16] include as an important feature the reduction of the structure group
SO(5) of the normal bundle of the M5-brane down to its SO(4) subgroup. Such a reduction
implies an existence of a covariantly constant vector field and, therefore, looks very much
as a dimensional reduction (to be more precise, the dimensional reduction is a particular
case of such an M5-brane framing [15,16]). 4 These facts provide us with a motivation
to study in more detail the dynamical and symmetry properties of the NSIIA five-brane
by constructing a full worldvolume action describing its dynamics in a type IIA D = 10
supergravity background.
By now the action for the NS5-brane has been constructed up to a second order in the
field strength of a two-rank self-dual worldvolume gauge field of the five-brane and only
in a background of the bosonic sector of IIA D = 10 supergravity [19,20].
The aim of this paper is to get a full, nonlinear and -symmetric, NS5-brane action in a
curved IIA D = 10 target superspace by carrying out the direct dimensional reduction of
the D = 11 M5-brane action [810], and thus filling in a gap in the list of worldvolume
actions for supersymmetric extended objects found in string theory.
The fact that the NS5-brane can be regarded as an M5-brane propagating in a
dimensionally reduced D = 11 supergravity background substantially simplifies the
analysis of the NS5-brane model, in particular, allowing one to derive its symmetries and
dynamical properties directly from those of the M5-brane.
For instance, the physical field content of the IIA D = 10 NS5-brane is the same as of
the M5-brane. The bosonic sector consists of three degrees of freedom corresponding to the
two-rank self-dual worldvolume field and five worldvolume scalars. In the case of the M5brane the five scalar fields describe its oscillations in a D = 11 background in the directions
transversal to the M5-brane worldvolume, while in the case of the NS5-brane four scalar
fields correspond to transversal oscillations in a D = 10 background, and the fifth scalar
field (corresponding to the compactified dimension of the D = 11 space) decouples and
becomes a purely worldvolume field. This results in the abovementioned reduction of
the M5-brane normal bundle structure group SO(5) down to SO(4). For both five-branes
eight fermionic fields can be associated with brane oscillations in Grassmann directions
of corresponding target superspaces.
To get the action describing the dynamics of the physical modes of the NS5-brane as a
dimensionally reduced M5-brane action we first briefly remind the structure and properties
of the latter.
In Sections 25 we consider bosonic M5- and NS5-branes and in Section 6 we describe
the full target-superspace covariant and -invariant NS5-brane action.
4 In contrast to [15,16] the analysis of Ref. [17] is based on the assumption that a full understanding of anomaly
cancellation requires keeping the full SO(5). We are thankful to Jeff Harvey for clarifying this difference in the
approaches.
317
2. The M5-brane action

In the absence of interactions with antisymmetric tensor fields of D = 11 supergravity
the action for the bosonic sector of the M5-brane has the following form [8]
"q
#
p
Z

g
6
mn
r
H
det gmn + iHmn + q
Hmnr a ,
(1)
S = d
[
4 aa
where
m, n, . . . = 0, . . . , 5
are vector indices of d = 6 worldvolume coordinates m ;
. . . = 0, . . . , 10
m,
n,
are vector indices of D = 11 target space coordinates X m ;
(11)
gmn = m X m gm n n X n
(2)
is the worldvolume metric induced by embedding the five-brane into a D = 11 gravity

(11)
background with a metric gm n (X) (we use the almost minus Minkowski signature
(+ )), Hmnl ( ) = 3[m bnl] is the field strength of the worldvolume antisymmetric
tensor field bmn ( ),
1
H mnr
q
r a,
H mn
[
aa
H mnl =
1
p mnlrsq Hrsq ,
3! g
(3)
a( ) is an auxiliary scalar field ensuring the covariance of the model, and

[ m a g mn n a
aa
(4)
denotes the scalar product of the d = 6 vector m a with respect to the metric (2). In what
follows the hat over quantities indicates that they correspond to or induced by the elevendimensional theory.
In addition to the usual gauge symmetry of the b2 field
a( ) = 0,
bmn = 2[m n] ( ),
(5)
the action (1) is invariant under the following transformations [810]

a( ) = 0,
a = ( ),
bmn = 2[m ( )n] a( ),

"
#
a
s a
ps
b
Hmn Hmnp g q
,
bmn = q
[
[
aa
aa
where
bmn = p2 LDBI ,
H
g H mn
LDBI

.
det gmn + iH mn
.
bmn defined in (8) reduces to H mn
Note that at the linearized level, H
(6)
(7)
(8)
318
The symmetries (6) and (7) are characteristic of the covariant approach [21] to the
Lagrangian description of duality-symmetric fields. They ensure the b2 field equation of
motion to reduce to a self-duality condition, as well as the connection with non-covariant
formulations [2231]
Let us briefly describe how one derives the symmetries (6) and (7) and gets the selfduality condition [8,21].
To this end note that the second term in the action (1) can be written in terms of
differential forms
Z
Z
p
1
H mn Hmnr r a
d6 L1 d6 g q
[
4 aa
Z
1
v H3 iv H3 ,
(9)
=
2
M6
where 5
v = d m vm ,
vk q
k a
(10)
[
aa
1 m
1
d d n d l Hlnm ,
iv H3 d m d n vk g kl Hlnm .
(11)
3!
2
The variation of the first term in (1) with respect to the gauge field and the scalar a( ) can
be written in terms of differential forms as
Z
Z
Z
q

6
6
b H ,
H
(12)
d LDBI d det gmn + iHmn =
2
2
H3
M6
b and H are constructed respectively from the tensors (8) and (3)
where 2-forms H
2
2
bnm ,
b 1 d m d n H
H
2
2
H 2 = iv H3 d m d n H nm
2
(13)
and is the Hodge operation in d = 6 dimensions. 6

Using the identities
iv v = 0,
6 iv 6 v,
iv H3 = H3 v,
(14)
v H3 iv H3 = v iv H3 H3 + H3 H3 ,
(15)
v H3 i v H3 = v H3 iv H3 = v v iv H3 iv H3
(16)
5 In our notation d m1 d m6 = d6 m1 m6 .
6 To have = I we define
2 =
1
d m4 d m1 gm1 m4 n1 n2 n1 n2 ,
2!4!
4 = +
1
1
1
d m2 d m1 g m1 m2 n1 n4 n1 n4
dm2 dm1 m1 m2 n1 n4 n1 n4 .
2!4!
2!4!
g
319
and
b = v H H
b H
v H
2
2
2
2
abcdef
b H
b
abcdef H
Hbc Hde vf ,
bc de v f =
(17)
one can rewrite the variation of the Lagrangian (1) in the form

Z
Z
1
1
6
H3 H3 da F2 H3 da da F2 F2 , (18)
d L
2
2
M6
where
F2 q

b iv H3 = 1 d m d n Fnm ,
H
2
[
aa
1
or
Fmn q
1
[
aa
bmn Hmnl g lk q
H
(19)
k a
(20)
[
aa
Since H3 = db2 , the variation (18) can be written (up to a total derivative) in the
following form 7
Z
Z
d(da F2 ) (b2 aF2 ),
(21)
d 6 L
M6
from which the invariance of the action under (6) and (7) becomes evident.
From (21) it also follows that the equation of motion of b2 field is
d(da F2 ) = 0,
(22)
and the equation of motion of a(x) is a consequence of Eq. (22). It can be shown [21]
that, using the symmetry (6), the second-order equation (22) reduces to the first-order selfduality condition

1
b iv H3 = 0,
H
(23)
F2 q
[
aa
or in components
bmn = Hmnl g lk q k a .
H
[
aa
(24)
To prove this note that H 2 is invariant under the transformations (6)

q
[ v 1
b2 = da 1 aa
q

[ iv (v d1 ) 0.
H 2 iv H 3 = aa
7 We use conventions where external derivative acts from the right:
dq =
1 mq
d m1 d n n m1 mq ,
d
q!
d(p q ) = p dq + ()q dp q .
320
Hence, the transformations of the two-form (19) reduce to

1
iv H3 = iv (v d1 ).
F2 = q
[
aa
Eq. (25) is simplified when one takes into account that iv da =
Then
(25)
q
q
[ iv v = aa.
[
aa
F2 = d1 + iv d1 v,
and
(da F2 ) = da d1 .
(26)
We now observe that Eq. (26) is similar to the general solution of Eq. (22) for da F2 .
This means that the general solution of Eq. (22) can be gauged to zero with the use of the
symmetry (6), and Eq. (23) appears just as a result of such gauge fixing.
= H g lk k a
bmn is defined in (8) and reduces to H mn
at the
Remember that H
mnl
[
aa
=H
linearized level, Eq. (24) becoming the conventional self-duality condition H lmn
lmn .
Further details on the classical dynamics of the M5-brane the reader may find in [810,
3239].
3. Dimensional reduction of D = 11 gravity and the NS5-brane action

The procedure of the direct dimensional reduction assumes a compactification of some
of target-space spatial dimensions (one in our case), the worldvolume of the p-brane being
not compactified. A standard (string frame) ansatz for the target-space vielbein under the
KaluzaKlein reduction of one spatial dimension has the following form

y = X 10 ,
X m = Xm , y ,
E a = E a , E 10 dX m Em a X ,

1
2
2
a
E 10 = e 3 dy dXm Am e 3 F ,
E a = e 3 dXm em (X),
!
1
2
e 3 em a e 3 Am
a
,
(27)
em =
2
0
e3
where y is the coordinate compactified into a torus, and the reduction means that the
background fields, such as components of (27), do not depend on y which is now
considered as an intrinsic scalar field in the 5-brane worldvolume. (X) is the dilaton field
and Am (X) is the Abelian vector gauge field of D = 10 IIA supergravity. The U (1)-gauge
transformations of Am (X) and y are
Am (X) = m (0)(X),
y = (0) (X).
(28)
This ansatz leads to the following expression for the D = 11 target space metric in terms
(10)
(X) = em a ema , Am (X) and (X)
of the D = 10 metric gmn
!

2
4
(10)
e 3 gmn
e2 Am An
e 3 Am
(11)
gm n =
(29)
4
4
e 3 An
e 3
and, consequently, to the following form of the six-dimensional induced metric (2)

2
gmn = e 3 gmn e2 Fm Fn .
321
(30)
In (30)
(10)
(X)n Xn ,
gmn = m Xm gmn
m = 0, . . . , 9
(31)
is the six-dimensional metric induced by embedding the 5-brane worldvolume into the
ten-dimensional curved spacetime and
Fm = m y Am ,
(32)
where Am = m Xm Am (X) is the worldvolume pullback of Am (X) and Fm is the pullback

of the one-form F introduced in (27).
Fm defined in (32) can be considered as a field strength of the worldvolume scalar field
y( ). It is invariant under the U (1) gauge transformations (28).
In what follows we will also use an expression for the inverse worldvolume metric

2
e2 F m F n
.
(33)
g mn = e 3 g mn +
1 e2 F 2
The NS5-brane action follows from the M5-action (1) with the background metric
having a particular form (29) and the coordinate X 10 = y( ) being considered as an
intrinsic worldvolume scalar field. To present the explicit form of the NS5-brane action we
should rewrite all its constituents in terms of D = 10 fields, and to rescale worldvolume
fields and their scalar products with respect to the worldvolume induced metric (31).
For instance, the Hodge duality (3) is now redefined with respect to the metric (31)
r
1
g mnp
H
mnlrsd Hrsd ,
,
H mnl =
(34)
H mnp =
g
3! g
and the M5-brane field strength H mn (3) is related to its NS5-brane counterpart H mn as
r s
1
g (a)2 mn
k a
kmnpqr Hpqr p
,
H mn =
,
(35)
H
H mn =
g aa
3! g
[
(a)2
where the scalar product (4) has also been correspondingly redefined as
[ k a g ks s a = e 3 N 2 (a)2,
aa
2
(a)2 l ag lm m a,
with N standing for

s
N
1
e2 (F a)2
= e 3
1+
(a)2 (1 e2 F 2 )
[
aa
.
(a)2
(36)
(37)
In view of Eqs. (30), (35), (36) and (37) the antisymmetric tensor entering the DBI-like
part of the M5-brane action is reexpressed in terms of H lm as follows
p
1
= gml g nk H lk = gml g nk g/g e 3 N 1 H lk ,

H mn

2
(38)
gml = e 3 gml e2 Fm Fl .
322
As a result, substituting (30)(38) into the action (1), we get the action for a bosonic 5brane coupled to the metric, the dilaton and the gauge vector field of type IIA D = 10
supergravity
Z
q
S = d6 e2 det(gmn e2 Fm Fn )
s

e (gmp e2 Fm Fp )
H np
det m n + i p
N det(m n e2 Fm F n )

Z
p a
1
e2 F k F p
1
p
.
(39)
d6 g 2 H mn Hmnk g kp +
2
2
4
N
1e F
(a)2
Since the action (39) is nothing but the M5-brane action for a special choice of the
D = 11 metric (29), its variation with respect to the gauge field b2 ( ) and the auxiliary
scalar a( ) has the form of Eq. (21), and hence (39) is also invariant under the symmetries
(6) and (7) which, as we have seen, produce the self-duality condition (24).
To rewrite the transformations and the self-duality condition (24)
in the form adapted to the NS5-brane propagating in the D = 10 background, let us
bmn (8)
introduce the NS5 counterpart of the tensor H
2 Lkin.NS5
=
,
(40)
Hmn
g H mn
where Lkin.NS5 denotes the first (DBI-like) term in the action (39), which is just the DBIlike term of the M5-action (1) written in the D = 10 adapted worldvolume frame. Using
bmn and Hmn

(35), it is easy to find the relation between H
s
(a)2
bmn
Hmn
=H
.
(41)
[
aa
Taking into account Eqs. (24), (33), (36) and (41) we obtain the following form of the local
worldvolume symmetries
a = 0,
bmn = 2[m a n] ( ),
a = ( ),
"
#

2 F p F s
a
1
e
a
p
2 Hmnp g ps +
Hmn
bmn = p
N
1 e2 F 2
(a)2
(a)2
(42)
(43)
and the self-duality equation for the NS5-brane gauge field b2

1
e2 F p F s
s a
ps
p
Hmn = 2 Hmnp g +
,
(44)
N
1 e2 F 2
(a)2
defined in (37) and (40).
with N and Hmn
In addition to the worldvolume diffeomorphisms and the symmetries (42) and (43), the
action (39) (by construction) has gauge symmetries (5) and (28).
Thus, we have obtained the action describing the worldvolume dynamics of the bosonic
5-brane propagating in the KaluzaKlein part (29) of the IIA D = 10 supergravity
background. In the next section we extend this action to describe coupling of the NS5brane to antisymmetric gauge fields of IIA D = 10 supergravity.
323
4. Coupling to the background gauge fields

When the M5-brane couples to the 3-form background field C (3) of D = 11 supergravity
the field strength H3 gets extended by the worldvolume pullback of C (3)
H (3) H (3) = db(2) C (3) .
(45)
As a result, up to a total derivative, the variation (18) of the action (1) with respect to
bmn ( ) and a( ) acquires an additional term in comparison with Eq. (21)

Z
Z
1
d(da F2 ) (b2 aF2 ) + dC 3 b2 .
(46)
d 6 L =
2
The symmetries (6) and (7) spoiled by the last term of (46) are restored if to the action (1)
one adds the WessZumino term [40]

Z
1 (2) (3)
(6)
.
(47)
C + db C
SWZ =
2
M6
As it was shown in [8], the symmetries (6) and (7) uniquely fix the relative factor between
SWZ and the action (1). In (47) C (6) is the pullback of a six-form gauge potential whose
field strength is D = 11 Hodge-dual to the field strength of C (3)
1
dC (6) + C (3) dC (3) = dC (3).
2
(48)
In addition to the symmetries (6) and (7) with H (3) generalized as in (45), the M5brane action (1) extended by the WessZumino term (47) is invariant under the following
transformations of the antisymmetric gauge fields
1
C (6) = d (5) C (3) C (3) , C (3) = d (2),
2

(2)
(2)
b = X( ) .
(49)
(50)
To get the form of the coupling of the NS5-brane to the antisymmetric gauge fields of
type IIA D = 10 supergravity we should dimensionally reduce C (3) , C (6) and the Wess
Zumino term (47) of the M5-brane. The dimensional reduction of C (3) produces a tendimensional RR three-form C (3) and an NSNS two-form B (2)

1
C (3) = dX l dX n dX m C m n l X
3!

1
1
= dXl dXn dXm Cmnl (X) + dXn dXm Bmn (X) dy dXl Al
3!
2
(51)
C (3) + B (2) F ,
and the dimensional reduction of C (6) produces a ten-dimensional five-form C (5) and a
six-form B (6) which are dual to C (3) and B (2) , respectively,
C (6) = B (6) + C (5) F ,
the duality relations can be easily derived by the dimensional reduction of Eq. (48).
(52)
324
Thus, the field strength of the self-dual gauge field of the NS5-brane coupled to the
D = 10 background gauge fields is extended as follows
H (3) = db(2) C (3) B (2) F ,
(53)
and the NS5-brane action (39) is enlarged with the following WessZumino term

Z
1
1
B (6) + C (5) F + db(2) C (3) + db(2) B (2) F ,
SWZ =
2
2
(54)
M6
where F = d m (m y Am ) is now the worldvolume pullback of the two-form (51).

We have now obtained the action for the NS5-brane propagating in a background of the
bosonic sector of type IIA D = 10 supergravity
Z
q
s

e (gmp e2 Fm Fp )
n
np
H
det m + i p
N det(m n e2 Fm F n )

Z
p a
1
e2 F k F p
1
6
mn
kp
Hmnk g +
p
d g 2 H
4
N
1 e2 F 2
(a)2

Z
1
1
+
(55)
B (6) + C (5) F + db(2) C (3) + db(2) B (2) F .
2
2
M6
This action is invariant under the worldvolume gauge transformations (5), (42), (43) and
(28), with H (3) now having the form (53), and under target-space gauge transformations
C (3) = d (2) + d (1) A,
b (2) = (2) (1) dy,
B (2) = d (1) ,
A = d (0),
y = (0),
(56)
under which H (3) = 0, and

1 (2)
1
d C (3) d (1) A C (3) ,
2
2
1
1
1
(57)
C (5) = d (4) d (2) B (2) + d (1) C (3) d (1) A B (2) .
2
2
2
Before proceeding with the consideration of the full super-NS5-brane action let us
demonstrate how the action of Refs. [19,20] is obtained from Eq. (55).
B (6) = d (5) + d (4) A
5. NS5-brane action in the second order approximation

The action of [19,20] is a second-order approximation in powers of Hmnk of the NS5brane action, with the self-duality condition being regarded as an extra (actually on-shell)
constraint. To get the second-order action we should expand (55) in series of H (3) and
truncate it down to the second order in H (3) assuming the worldvolume gauge field to be
325
weak. Since the WessZumino term is already linear and quadratic in H , we shall write
down only the kinetic part of the action. 8
To carry out such a truncation the simplest way is to first truncate the M5-brane action
(1) and then perform its dimensional reduction. Up to the second order in H (3) the M5brane action has the form

Z
p
1 mn 1 mn
6
p
+ H
Hmnp v +
(58)
S = d g 1 Hmn H
4
4
with the self-duality condition (24) reducing to
= 0.
Hmnl H mnl
(59)
Taking into account the expressions

1
= p
mnkqrs H qrs
H mnk
3! g
and
mnlpqr mnlst v = (3!)2 p[s qt rv] ,
after some algebra one can rewrite (58) in the following form [21]
Z
p
1
S = d6 g 1 Hmnl H mnl
24

p
1
m a H mnl H mnl Hnlp H nlp
a + . (60)
[
8aa
Discarding in (60) the term containing the auxiliary field a( ) and the anti-selfdual
tensor H H (which is zero on the mass shell (59)), and carrying out the direct
dimensional reduction of (60) we recover the NS5-brane action of [19,20]
Z
q

e4
1 2
mnk
mnk
p
+3
Fm H
Hnkp F + .
e Hmnk H
1
(61)
24
1 e2 F 2
Alternatively, the action (61) can be obtained directly by truncating the NS5-brane action
(55), and discarding terms containing the auxiliary field and the linearized NS5-brane selfduality condition (44)

e2 F p Fl
p
Hmnp l +
= 0.
Hmnl
1 e2 F 2
8 We should note that our choice of the dimensionally reduced C and C differs from that in [19,20], so the
3
5
WessZumino term in Eq. (55) is related to the one of Refs. [19,20] by the following field redefinition:
y c(0) ,
B (2) B (2) ,
b(2) a (2) ,
C (3) B (2) A C (3) ,
1
C (5) C (5) C (3) B (2) ,
B (6) C (5) A B (6) .
2
The WZ term of [19,20] also contains the curl of an auxiliary worldvolume 5-form field which ensures the exact
gauge invariance of the WZ term.
326
6. The -symmetric super-NS5-brane action

To generalize the results of previous sections to describe the propagation of an NS5brane in a curved IIA D = 10 target superspace parametrized by ten bosonic coordinates
Xm and 32-component Majorana-spinor fermionic coordinates forming a IIA, D = 10
superspace coordinate system

(62)
Z M = Xm , ,
we again start with an M5-brane propagating in a generic D = 11 supergravity background
parametrized by eleven bosonic coordinates X m and 32-component Majorana-spinor
fermionic coordinates forming a D = 11 superspace coordinate system

(63)
Z M = X m , = Z M , y ,
where we have separated the eleventh coordinate y = X10 keeping in mind the dimensional
reduction of D = 11 superspace down to type IIA D = 10 superspace.
D = 11 superspace geometry is described by a supervielbein

A
(64)
E A = dZ M E Z = E a , E ,
M
= (a,
) are locally flat tangent superspace indices, by a superconnection
where A
B
B
w = dZ M w Z ,
A
MA
(65)
and by a three-superform generalization of the bosonic gauge field (51)

1
(66)
C (3) = dZ N dZ M dZ L CL M N (Z).
3!
The supervielbein, the superconnection and the gauge superfield are subject to
supergravity constraints which put the superfield formulation of eleven-dimensional
supergravity on the mass shell. An explicit form of the D = 11 supergravity constraints
relevant to the description of M5-brane dynamics the reader may find in [9,3739,45].
The super-M5-brane action has the similar form as the bosonic action (1) enlarged with
the WZ term (47), where the worldvolume induced metric is now

a
(67)
gmn = m Z M n Z N E Z E M a Z ,
N
C (3)
C (6)
and
are worldvolume pullbacks of the three-superform (66) and its sixand
superform dual [45,46].
In addition to all symmetries discussed above and target-space superdiffeomorphisms
the super-M5-brane action is invariant under the following fermionic -symmetry
transformations [9,10]

,
i E Z M E Z = I S
i E a Z M E Z = 0,
M

1
(68)
a = 0,
b2 = i C 3 dZ M 3 dZ M 2 Z M 1 C M M M Z ,
1 2 3
2
where the spinor matrix has the following expansion in products of D = 11 Dirac
matrices
p
S = q
det(g + iH )
327

i
(6)
mnl
mn
+ Hmn vl
+ tm vn
,
2

1 m1 m6
m m Z M E Z a ,
m1 m6 ,
M
6!
1
1

b H
b vq .
(69)
Hpl vq mnkplq H
t m = mnkplq H nk
nk
pl
8
8
As is characteristic of all superbranes, for the M5-brane action to be -symmetric the
superbackground must satisfy the supergravity constraints [9,39,45]. When they are taken
into account, from (68) we get
H 3 = i F4 = E a E b E a b I S ,
(6) =
gmn = 4iE
(m
n) I S
(70)
We now turn to the consideration of the super-NS5-brane action. It can be obtained from
the super-M5-brane action by the direct dimensional reduction of the D = 11 supergravity
superfields. A consistent ansatz for the dimensionally reduced supervielbein (64) was
proposed in [48]. This is the following superfield generalization of Eq. (27)

1
2
2
E 10 = e 3 (Z) dy dZ M AM (Z) e 3 (Z) F ,
(71)
E a = e 3 (Z) E a ,
E = e 6 (Z) E (Z) + F (Z),
1
(72)
where E A (Z) = dZ M EM = (E a , E ) are supervielbeins of type IIA D = 10 supergravity,

(Z) is the dilaton superfield, AM (Z) are components of the one-form gauge superfield
A = dZ M AM (Z), and (Z) is a Grassmann-odd Majorana spinor superfield, which is
actually the Grassmann derivative of the dilaton superfield (Z).
The superfields which describe IIA D = 10 supergravity are subject to the constraints
which are obtained from the D = 11 supergravity constraints using the ansatz (71),
(72) and solving for Bianchi identities. Different forms of these constraints have been
considered in [4851].
We do not write the super-NS5-brane action explicitly since it has exactly the same form
as Eq. (55) where now the worldvolume induced metric is
a
gmn = m Z M n Z N EN (Z)EMa (Z),
(73)
and all the bosonic background fields are replaced with corresponding superfields, in
particular, B (6) , C (5) , C (3) and B (2) are the worldvolume pullbacks of the type IIA D = 10
superforms
1 An
E E A1 CA1 An (Z).
(74)
n!
Note that the spinor superfield does not appear in the action (55).
The super-NS5-brane action is invariant under -symmetry transformations obtained
from Eqs. (68) by substituting into the latter the ansatz (71) and (72)
,
(75)
i E Z M EM (Z) = I
C (n) (Z) =
328
i E a = 0,
i F = 0
y = i E A (Z),
b2 = i C3 + F i B2 ,
a = 0,
(76)
where A (Z) is a fermionic component of the KaluzaKlein connection form

A dZ M AM = E A + E a Aa .
7. Conclusion and discussion

To summarize, we have obtained the covariant -symmetric action for the super-NS5brane in a IIA D = 10 supergravity background by the direct dimensional reduction of
the M-theory super-five-brane action. In addition to worldvolume diffeomorphisms, gauge
symmetry, -symmetry and background supergravity symmetries the super-NS5-brane
action possesses special local symmetries ensuring the covariance of actions with self-dual
gauge fields and serving for deriving the self-duality condition directly from the action as
a consequence of the equation of motion of the gauge field.
An interesting problem for future study is to construct the Lagrangian description of the
consistent coupling of a type IIA supergravity action to an NS5-brane source. The latter
requires the construction of a duality-symmetric version of type IIA supergravity by the
dimensional reduction of the duality-symmetric D = 11 supergravity [46]. The truncation
of such a IIA supergravity action shall produce the duality-symmetric version of the N =
1, D = 10 supergravity, which should naturally couple to a heterotic five-brane [52,53].
Note that recent investigations of interacting brane actions [54] may provide one with a
possibility of making this coupling supersymmetric.
Another problem for further studying is to perform the T-duality transformation of the
complete NS5-brane action and to arrive at a non-linear and supersymmetric action for
a type IIB D = 10 KaluzaKlein (KK) monopole. A quadratic approximation for the
bosonic part of this action has been constructed in [20]. One of possible ways of deriving
appropriate T-duality transformation rules for the antisymmetric gauge fields is to T-dualize
the duality-symmetric version of type IIA supergravity to the duality-symmetric version of
type IIB supergravity [55,56].
As it was noted in [47] and proved in the second order approximation in [20], the type IIB
D = 10 KK monopole is expected to be a self-dual object under the S-duality symmetry
of type IIB supergravity. The construction of the complete action for the type IIB KK
monopole should allow one to explicitly verify this statement.
Acknowledgements
The authors are grateful to Kurt Lechner, Paolo Pasti and Mario Tonin for interest to
this work and valuable discussions and to Christopher Hull, Jeffrey Harvey, Bernard Julia
and Kellog Stelle for useful comments. I.B. and A.N. also acknowledge kind hospitality
extended to them at the Abdus Salam International Centre for Theoretical Physics where
329
part of this work was done. This work has been partially supported by the Ukrainian GKNT
Grant 2.5.1/52 and INTAS Grant No 96-0308.
References
[1] P.K. Townsend, Nucl. Phys. (Proc. Suppl.) 58 (1997) 163, hep-th/9609217.
[2] P.K. Townsend, Four lectures on M-theory, in: Proc. of the ICTP Summer School on HEP and
Cosmology, Trieste, 1997; hep-th/9612121.
[3] A. Hanany, E. Witten, Nucl. Phys. B 492 (1997) 152.
[5] K. Bachas, Desert energy or transverse space, hep-th/9907023; Class. Quant. Grav. 17 (2000)
1, hep-th/0001093, and references therein.
[6] E. Witten, JHEP 01 (1998) 001.
[7] O. Aharony, A brief review of little string theories, Class. Quant. Grav. 17 (2000) 929.
[8] P. Pasti, D. Sorokin, M. Tonin, Phys. Lett. B 398 (1997) 41, hep-th/9701037.
[9] I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, Phys. Rev. Lett. 78
(1997) 4332, hep-th/9701149.
[10] M. Aganagic, J. Park, C. Popescu, J.H. Schwarz, Nucl. Phys. B 496 (1997) 191, hep-th/9701166.
[11] D. Berman, Phys. Lett. B 409 (1997) 153159, hep-th/9706208; Nucl. Phys. B 533 (1998) 317
332, hep-th/9804115.
[12] A. Nurmagambetov, Phys. Lett. B 436 (1998) 289.
[13] E. Witten, J. Geom. Phys. 22 (1997) 103, hep-th/9610234.
[14] K. Becker, M. Becker, Fivebrane gravitational anomalies, hep-th/9911138.
[15] L. Bonora, C.S. Chu, M. Rinaldi, JHEP 12 (1997) 007, hep-th/9710063.
[16] L. Bonora, M. Rinaldi, Normal bundles, pfaffians and anomalies, hep-th/9912214.
[17] D. Freed, J.A. Harvey, R. Minasian, G. Moore, Adv. Theor. Math. Phys. 2 (1998) 601, hepth/9803205.
[18] E. Witten, Duality relations among topological effects in string theory, hep-th/9912086.
[19] E. Bergshoeff, Y. Lozano, T. Ortin, Nucl. Phys. B 518 (1998) 363.
[20] E. Eyras, B. Janssen, Y. Lozano, Nucl. Phys. B 531 (1998) 275, hep-th/9806169.
[21] P. Pasti, D. Sorokin, M. Tonin, Phys. Lett. B 352 (1995) 59; Phys. Rev. D 52 (1995) R4277;
Phys. Rev. D 55 (1997) 6292.
[22] D. Zwanziger, Phys. Rev. D 3 (1971) 880.
[23] S. Deser, C. Teitelboim, Phys. Rev. D 13 (1976) 1592.
[24] R. Floreanini, R. Jackiw, Phys. Rev. Lett. 59 (1987) 1873.
[25] M. Henneaux, C. Teitelboim, in: Proc. Quantum Mechanics of Fundamental Systems 2,
Santiago, 1987, p. 79; Phys. Lett. B 20 (1988) 650.
[26] J.H. Schwarz, A. Sen, Nucl. Phys. B 411 (1994) 3.
[27] M. Perry, J.H. Schwarz, Nucl. Phys. B 489 (1997) 47.
[28] J.H. Schwarz, Phys. Lett. B 395 (1997) 191.
[29] A. Maznytsia, C.R. Preitschopf, D. Sorokin, Nucl. Phys. B 539 (1999) 438.
[30] Y.-G. Miao, H.J.W. Mueller-Kirsten, Self-duality of various chiral boson actions, hepth/9912066.
[31] Y.-G. Miao, R. Manvelyan, H.J.W. Mueller-Kirsten, Self-duality beyond chiral p-form actions,
hep-th/0002060.
[32] P.S. Howe, E. Sezgin, Phys. Lett. B 394 (1997) 62.
[33] P.S. Howe, E. Sezgin, P.C. West, Phys. Lett. B 399 (1997) 49.
[34] I. Bandos, K. Lechner, A. Nurmagambetov, P. Pasti, D. Sorokin, M. Tonin, Phys. Lett. B 408
(1997) 135.
330
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
D. Sorokin, P.K. Townsend, Phys. Lett. B 412 (1997) 265.

E. Bergshoeff, D. Sorokin, P.K. Townsend, Nucl. Phys. B 533 (1998) 303.
P. Claus, R. Kallosh, A. Van Proeyen, Nucl. Phys. B 518 (1998) 117.
P. Claus, Phys. Rev. D 59 (1999) 066003.
D. Sorokin, Superbranes and superembeddings, hep-th/9906142; Physics Reports (in press).
O. Aharony, M theory and string dualities, in: Gauge Theories, Applied Supersymmetry and
Quantum Gravity, Imperial Colledge Press, 1997.
E. Bergshoeff, M. de Roo, T. Ortin, Phys. Lett. B 386 (1996) 85, hep-th/9606118.
M.J. Duff, J.X. Lu, Nucl. Phys. B 354 (1991) 129.
M.J. Duff, R.R. Khuri, J.X. Lu, Phys. Rep. 259 (1995) 213.
E. Cremmer, B. Julia, H. L, C.N. Pope, Nucl. Phys. B 535 (1998) 242, hep-th/9806106.
A. Candielo, K. Lechner, Nucl. Phys. B 412 (1994) 479.
I. Bandos, N. Berkovits, D. Sorokin, Nucl. Phys. B 522 (1998) 214.
C. Hull, Nucl. Phys. B 509 (1998) 216, hep-th/9705162.
M.J. Duff, P.S. Howe, T. Inami, K.S. Stelle, Phys. Lett. B 191 (1987) 70.
J.L. Carr, S.J. Gates Jr., R.N. Oerter, Phys. Lett. B 189 (1987) 68.
M. Cederwall, A. von Gussich, B.E.W. Nilsson, P. Sundell, A. Westerberg, Nucl. Phys. B 490
(1997) 179, hep-th/9610148.
E. Bergshoeff, E. Cowdall, P.K. Townsend, Phys. Lett. B 410 (1997) 13, hep-th/9706094.
E. Witten, Nucl. Phys. B 460 (1996) 541, hep-th/9511030.
J. Mourad, Nucl. Phys. B 512 (1998) 199, hep-th/9709012.
I. Bandos, W. Kummer, Phys. Lett. B 462 (1999) 254, hep-th/9905144; Nucl. Phys. B 565
(2000) 291, hep-th/9906041.
G. DallAgata, K. Lechner, D. Sorokin, Class. Quant. Grav. 14 (1997) L195.
G. DallAgata, K. Lechner, M. Tonin, JHEP 9807 (1998) 017.

Five-dimensional gauge theories and local

mirror symmetry
Tohru Eguchi a , Hiroaki Kanno b,
a Department of Physics, Faculty of Science, University of Tokyo, Tokyo 113, Japan
b Department of Mathematics, Graduate School of Science, Hiroshima University,
Higashi-Hiroshima 739, Japan

Abstract
We study the dynamics of 5-dimensional gauge theory on M4 S 1 by compactifying type II/M
theory on degenerate CalabiYau manifolds. We use the local mirror symmetry and shall show that
the prepotential of the 5-dimensional SU (2) gauge theory without matter is given exactly by that of
the type II string theory compactified on the local F2 , i.e., Hirzebruch surface F2 lying inside a noncompact CalabiYau manifold. It is shown that our result reproduces the SeibergWitten theory at
the 4-dimensional limit R 0 (R denotes the radius of S 1 ) and also the result of the uncompactified
5-dimensional theory at R .
We also discuss SU (2) gauge theory with 1 6 Nf 6 4 matter in vector representations and show
that they are described by the geometry of the local F2 blown up at Nf points. 2000 Elsevier
Science B.V. All rights reserved.
PACS: 11.15.-g; 11.30.Pb; 02.10.Rn
Keywords: Mirror symmetry; SeibergWitten theory; Instanton expansion; Supersymmetric gauge theory
1. Introduction
Recent developments in non-perturbative string theory and M theory have led to new
insights into the relation between low energy field theory and string theory: it has been
argued in particular that non-perturbative dynamics takes place in low energy field theory
and higher gauge symmetries emerge when compactifying CalabiYau and K3 manifolds
degenerate and some of their homology cycles vanish. For instance, when K3 surface
develops an A-D-E singularity, there appears an enhanced A-D-E gauge symmetry in
6-dimensions. Similarly, when a family of P1 s shrinks to zero size along a rational curve
Corresponding author. E-mail: kanno@math.sci.hiroshima-u.ac.jp
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 7 5 - 8
332
T. Equchi, H. Kanno / Nuclear Physics B 586 (2000) 331345
in CalabiYau threefold, we obtain a 4-dimensional SU (2) gauge theory with N = 2

supersymmetry.
In this article we would like to study the dynamics of SUSY gauge theories in 5dimensions by compactifying the M theory on degenerate CalabiYau manifolds. We
construct an effective action of an SU (2) gauge theory on a 5-dimensional space M4 S 1
where M4 is the Minkowski space and S 1 is a circle of radius R. In the limit of R 0 this
theory reproduces the standard N = 2 SUSY gauge theory in 4-dimensions, i.e., Seiberg
Witten theory [1,2]. Thus our 5-dimensional model gives an M-theoretic generalization of
SeibergWitten theory by incorporating KaluzaKlein excitations. In the opposite limit
R our model reduces to the gauge theory in uncompactified 5-dimensions M5 .
Characteristic features of this theory have been studied using the brane-probe picture [3]
and also from the point of view of classical geometry of collapsing del Pezzo surfaces
[46] and the behavior of the low-energy effective gauge coupling has been determined
exactly.
In this paper we would like to propose an exact solution of the 5-dimensional theory on
M4 S 1 which reproduces the known results at both limits R 0 and R . As it turns
out, our prepotential follows directly from that of the type II string theory compactified on
singular CalabiYau manifolds using the method of local mirror symmetry [710]. In the
case of pure SU (2) gauge theory without matter our result is obtained from the type II
theory compactified on the local F2 , i.e., Hirzebruch surface F2 lying inside a CalabiYau
threefold which is the canonical bundle over F2 . Similarly, SU (2) gauge theories with Nf
matter in vector representations are also obtained from the type II theory compactified on
the local F2 blown up at Nf points (0 6 Nf 6 4). We will find that our model at R =
has an infinite bare coupling constant and yields a non-trivial interacting field theory in the
infra-red limit.
The local mirror symmetry is a method of mirror symmetry adapted in the case of noncompact CalabiYau manifolds. Suppose, for instance, we are given a compact CalabiYau
threefold which is an elliptic fibration over Fn . One considers the limit of the size of the
fiber tE going to . Then the resulting non-compact manifold is modeled by the local Fn ,
i.e., Fn inside a CalabiYau with the normal bundle being given by the canonical bundle of
Fn . The limit of tE may also be considered as the limit of shrinking Fn with the size
of the fiber kept fixed. Fn (n = 0, 1, 2) and its various blow ups are the del Pezzo surfaces
and these are in fact the type of manifolds which featured in the geometrical interpretation
of 5-dimensional gauge theory [46].
2. SU (2) gauge theory without matter

Let us start from the case of 5-dimensional gauge theory without matter. We consider
the local F2 model which is described by the toric data given in the Appendix. Following
the standard procedure [710] one obtains a curve in the B-model given by
P = a0 x + a1 x 2 + a2 + a3 + a4
1
= 0.
(2.1)
333
Introducing a new variable y = a2 a4 / the curve is rewritten as

y 2 = (a1 x 2 + a0 x + a3 )2 4a2a4 .
(2.2)
Complex moduli of the B-model are defined by

zF =
a1 a3
,
a02
zB =
a2 a4
.
a32
(2.3)
If we choose a1 = a3 = 1, a0 = s, 4a2 a4 = K 4 , we find

y 2 = (x 2 + sx + 1)2 K 4 ,
zF =
1
,
s2
zB =
K4
.
4
(2.4)
(2.4) is in fact the curve proposed by Nekrasov [11] for the description of 5-dimensional
gauge theory and its properties have been studied in [12] in detail.
If we introduce a variable U which is the analogue of u of the SeibergWitten solution,
the parameter s is written as
s = 2R 2 U,
where R is the radius of S 1 . In terms of U and R the curve reads as

2
1
K 4.
y 2 = x 2 R4 U 2 4
R
(2.5)
(2.6)
By comparing (2.6) with the SeibergWitten curve y 2 = (x 2 u)2 4 , we find the

correspondence between the parameters U and u

1
2
2
(2.7)
u R U 4 .
R
As we see from (2.7), U variable describes two copies of the u-plane. Strong coupling
region u 0 maps to U 1/R 2 and thus the two copies are separated by a distance of
order 1/R 2 .
In the brane-probe interpretation of SeibergWitten solution [13,14], u-plane is
identified as a local region around one of the four fixed points (O-7 planes) in type I0
theory compactification to 8-dimensions on T 2 (u = 0 is identified as the location of
the O-7 plane). Then the curve (2.6) describes a theory which contains two of these O7 planes. Since the fixed plane acts like a reflecting mirror, D3-brane probe will possess an
infinite number of mirror images when inserted into a background of two orientifold planes.
These mirror images are separated by distances n/R 2 , n Z and open strings connecting
them generate KaluzaKlein modes of supersymmetric gauge fields. Thus the curve (2.6)
effectively describes a theory on a 5-dimensional manifold M4 S 1 with R being the
radius of S 1 . In the limit of R 0 one of the u-planes moves off to and the model (2.6)
is expected to reduce to the SeibergWitten theory.
Periods of the B-model of local F2 (2.4) is determined by solving differential
equations (GelfandKapranovZelevinskij (GKZ) system) associated with the toric data
(see Appendix). Differential operators are given by
334

L1 = zF 4z2F + 2zF + zF 2zB zF ,

L2 = zB 2zB zF + 1 2zB zF z2B ,
(2.8)
(2.9)
where
zF zF
,
zF
zB zB
.
zB
(2.10)
These operators have a regular singular point at zF = zB = 0: they possess singlelog solutions F , B behaving as F = log zF + and B = log zB + at
zF , zB 0. There also exists a double-log solution behaving as = (log zF )2 +
(log zF )(log zB ) + . We identify the two single-log solutions as the Khler parameters
tF , tB of the A-model: tF represents the size of the P1 fiber of F2 and tB the size of its base
P1 . tF is given by
"
#
X (2m 1)!
X
2(2n 1)!
n m
m
z z
zB ,
tF F = log zF +
(n 2m)!n!m!2 F B
m!2
n>1,m>0
m>1
s
!
1
1
+
1
= 2 log
4zF
4zF
#
"
X (2m 1)!
X
2(2n 1)!
n m
m
z z
zB .
(2.11)
+
(n 2m)!n!m!2 F B
m!2
n>1,m>1
Similarly, tB is given by
m>1
1
+
tB B = 2 log
4zB
!
1
1 .
4zB
(2.12)
In our interpretation as the 5-dimensional gauge theory, the size of the fiber tF is
identified as the vacuum expectation value A of the scalar field in the vector-multiplet
tF = 4RA.
(2.13)
On the other hand, the size of the base tB is related to the dynamical mass parameter as
etB = 4R 4 4 .
(2.14)
Then the mirror transformation (2.12) becomes

K=
2R
1 + 4R 4 4
(2.15)
Note that (2.13) and (2.14) are in fact the identification of variables suggested in [7,15].
One can invert the relations (2.11), (2.12) perturbatively and express the B-model
parameters zF ,zB in terms of tF , tB . In the case of local mirror symmetry the holomorphic
solution of GKZ system is a constant and the mirror transformation is simpler than in the
compact CalabiYau case. One may then represent the double-log solution in terms of the
Khler parameters
X

X 1
n
2 n
q
+
4q
n
q
B
F
n2 F
n=1
n=1

qF2 + 36qF3 + 260qF4 + 1100qF5 + + O qB3 ,
335
= tF2 + tF tB + 4
+ qB2
(2.16)
where
qF = etF ,
qB = etB .
(2.17)
We identify the double-log solution as AD , the dual of the variable A

= 8iRAD .
(2.18)
First two terms of (2.16) represent the classical intersection numbers of the CalabiYau
manifold and the remaining terms represent the contribution of world-sheet instantons.
According to Ref. [10] the double-log solution has a generic form

2
X X X
X
q kn q km
i j
i
ti tj hJ J i +
x ti dn,m 1 32 ,
(2.19)
=
k
i,j =1
k=1 n,m>0
for a local model of a surface S (with two Khler parameters). Ji denotes the Khler classes
of S and hJi Jj i their intersection numbers. Numerical coefficients x i are defined by
X
x i Ji ,
(2.20)
c1 (S) =
i
where c1 (S) is the first Chern class of S. dn,m gives the number of rational holomorphic
curves in the homology class nJ1 + mJ2 . Sum over k in (2.19) represents the multiplecover factor.
By comparing (2.16) and (2.19) (set q1 = qF , q2 = qB ) we find [7,10]
d1,0 = 2,
d1,2 = d2,2 = 0,
dn,0 = 0,
n > 1,
d3,2 = 6,
dn,1 = 2n,
d4,2 = 32,
...
By integrating AD over A we have the prepotential F for the local F2 model

"
X

X 1
tF2 tB
tF3
1
n
n
+4
q + 4qB
nqF
F=

3
2
32iR 2
n3 F
n=1
n=1
#

2 1 2
3
4
5
3
q + 12qF + 65qF + 220qF + + O qB .
+ qB
2 F
(2.21)
(2.22)
3. Small and large radius limits

Let us next examine the small and large radius limits R 0, of (2.22). We first
consider the 4-dimensional limit R 0. Due to the relations (2.13), (2.14) small R
corresponds to the base of F2 becoming large (tB ) and the fiber becoming small
(tF 0) while the ratio etB /tF4 is kept fixed. At small R, we have K 2R and U
cosh(2RA). As explained by Katz, Klemm and Vafa [7], quantum parts of F are suppressed
336
because of the powers of qB R 4 and the surviving contributions come from the divergent
P
parts of the series dn,m qFn over qF as qF = e4RA 1. At small R the gauge coupling
= 2 F /A2 behaves as
X
i
8i
2i X qFn
4i
log(4R 4 4 )
R 4 4
n3 qFn +
= RA
n=1
n=1

4
A
3i
2i
+ .
(3.1)
log 2 2
16 A4
(3.1) reproduces the one-loop beta function and the one-instanton contribution to the
SeibergWitten solution [1]. We can check the agreement with SeibergWitten theory for
higher instanton terms.
In (3.1) the world-sheet instanton expansion of type II theory is converted into the
space-time instanton expansion of gauge theory in the R 0 limit. Coefficients of the
m-instanton amplitudes of gauge theory are determined by the asymptotic behavior of the
number of holomorphic curves dn,m as n with fixed m.
Let us next examine the R limit of the uncompactified 5-dimensional gauge
theory. It is known [16,17] that the gauge theory on M5 has no quantum corrections and its
gauge coupling is simply expressed in terms of classical intersection numbers of Calabi
Yau manifold. In fact by taking R world-sheet instanton terms disappear and we
have
2
5 = 2 A
(3.2)
lim
R 2iR
(we have rescaled so that 5 corresponds to the gauge coupling of 5-dimensional theory,
5 = 1/g52 ). In the next section we will discuss local F2 model blown up at Nf points. We
then find that the above formula is generalized as
!
Nf
Nf
X
X
2
1
1
|A Mi |
|A + Mi | .
(3.3)
5 = 2 A
16
16
i=1
i=1
(3.3) is exactly the behavior of gauge coupling of SU (2) theory with Nf matter in vector
representations (at infinite bare coupling) [35]. Thus we reproduce correct results also in
the uncompactified limit R .
We should note that the local models of F0 and F1 also reproduce SeibergWitten
solution in the R 0 limit [7] since the asymptotic behavior of the number of holomorphic
curves dn,m are the same for all models Fi , i = 0, 1, 2. F0 , F1 , however, have a different
classical topology from F2 and do not reproduce (3.2) at R = . Thus the F2 model is
singled out as the unique candidate for the description of 5-dimensional gauge theory on
M4 S 1 .
4. SU (2) theory with matter
Let us next consider the case of SU (2) gauge theory coupled to Nf matter (1 6 Nf 6 4)
in vector representations. We first discuss the Nf = 1 case. Relevant geometry is given by
337
the local F2 with one-point blown up. Corresponding curve is given by (see Appendix)

1
2
2
2
3
.
(4.1)
y = (x + sx + 1) K xw +
w
Comparison with the Nf = 1 SeibergWitten curve suggests the identification
w = eRM ,
(4.2)
where M denotes the bare mass of the matter multiplet. Complex moduli of the B-model
are given by (we denote zF E as zF and zF +E as zF 0 for notational simplicity)
w2
1
K3
,
zF 0 =
.
(4.3)
,
zB =
2
s
sw
4w
In this case the GKZ system is given by five partial differential equations in three variables
(see Appendix) and is of considerable complexity. Here we content ourselves with the
analysis of prepotential at the tree and one-loop level ignoring the spacetime instantons.
This suffices for our purpose of extracting the R behavior of the theory. The small
R behavior has already been studied in [8] and argued to reproduce SeibergWitten theory
(we have also verified that the 1-instanton term is correctly reproduced).
When we ignore instantons, single log solutions are given by
X (2n 1)!
(zF zF 0 )n ,
(4.4)
tF F = log(zF ) +
n!2
n=1
X (2n 1)!
(zF zF 0 )n ,
(4.5)
tF 0 F 0 = log(zF 0 ) +
n!2
zF =
n=1
tB B = log(zB ).
(4.6)
Identification with the variables of the gauge theory is given by

tF = 2R(A M),
tF 0 = 2R(A + M),
etB =
2R 3 3
.
w
(4.7)
Then by inverting relations (4.4), (4.5) we find

e2R(AM)
,
1 + e4RA
The double log solution
zF =
zF 0 =
e2R(A+M)
.
1 + e4RA
(4.8)
n
X 4(2n 1)! X
1
1
(zF zF 0 )n
= tF2 + tF2 0 + 2tF tF 0 + tB (tF + tF 0 ) +
2
2
n!
j
n=1
j =1
!
X X (n + ` 1)! 1
zn z` 0 ,
+
+
n!`!
n` F F
n>`
(4.9)
n<`
can then be re-expressed as

1
= tF2 + tF2 0 + 2tF tF 0 + tB (tF + tF 0 )
2
X 1
X 1
X 1
4nRA
2nR(A+M)
e
e2nR(AM) ,
+4
n2
n2
n2
n=1
n=1
n=1
(4.10)
338
using (4.8).
When we take the derivative in A, (4.10) gives

1
i
2 log sinh(2RA) log sinh R(A + M) sinh R(A M) .
=
4
In the 4-dimensional limit (4.11) becomes

1
i
2
log A
(4.11)
(4.12)
which gives the 1-loop beta function of SeibergWitten theory. On the other hand in the
5-dimensional limit we find

2
1
1
= 2 A |A M| |A + M|
(4.13)
5 = lim
R 2iR
16
16
as we have claimed before. We note that in formulas (3.2) and (4.13) 5 does not contain an
2 . Thus the
additive constant which is identified as the bare coupling constant 5,B = 1/g5,B
2 = which yields non-trivial
theory sits at the infinite bare coupling constant limit g5,B
5-dimensional theory in the infra-red regime [3].
We can similarly study the system with more matter up to Nf = 4 using the local mirror
symmetry. Curves of the B-model are given by (see Appendix)
y = (x + sx + 1) K
2
4Nf
Nf

Y
1
wi x +
,
wi
(4.14)
i=1
where parameters wi correspond to the bare masses of the matter multiplets

wi = eRMi ,
i = 1, . . . , 4.
(4.15)
(4.14) agrees with the curve suggested by [11] and [18] (at Nf = 4 the factor K 4Nf
should be replaced by a dimensionless parameter q).
In the cases Nf > 2 the analysis of GKZ system becomes further involved: we will
instead use a simpler method based on the PicardFuchs (PF) equation derived for the
elliptic curves (4.14). It turns out that the PF equation of the elliptic curve is an ordinary
differential equation of third order in the variable A and can be studied relatively easily.
This is the method used in Ref. [12]. This system, however, is not complete unlike that of
GKZ case. The periods are determined only up to integration constants and one can not
precisely fix the mirror transformation. This ambiguity, however, affects only the quantum
part of the computation and one still obtains precise results for the prepotential at the tree
and one-loop level.
First we note that the quadratic curve y 2 = ax 4 + 4bx 3 + 6cx 2 + 4dx + e is transformed
into the Weierstrass form
y 2 = 4x 3 g2 x g3 ,
(4.16)
by the relation
g2 = ae 4bd + 3c2,
g3 = ace + 2bcd ad 2 b2 e c3 .
(4.17)
339
We regard g2 , g3 as functions of the parameter s. Periods of the elliptic curve (4.16) obey
the PF equation [19,20]
d
d2
+ c0 (s) = 0,
+ c1 (s)
2
ds
ds
(4.18)

3
dg2
d
dg3
2g2
3
g3 ,
c1 = log
ds
2
ds
ds
(4.19)
c0 =
(4.20)
where

1 d2
1
1
dg3 2
d
dg2 2
c1 log +
log
12
g
2
12 ds
12 ds 2
16
ds
ds
and = (g2 )3 27(g3 )2 denotes the discriminant of the curve. Since dA/ds is one of the
periods of the curve, it satisfies the PF equation. Regarding s as a function of A, we obtain
2 2
2 2
4
d s
d s
ds
ds
ds d3 s
3
+ c1
c0
= 0.
(4.21)
3
2
2
dA dA
dA dA
dA
dA
This determines s in terms of A.
Similarly dAD /ds = dA/ds d2 F /dA2 satisfies the PF equation and we obtain
2 2 3
1 4
ds
d F
d s d F
d3 F
ds
3
+
c
= 0.
1
dA
dA
dA4
dA2 dA3
dA3
This equation can be integrated once and we find

3
dg2
const
dg3
ds
d3 F
2g2
3
g3
=
.
3
ds
ds
dA
dA
(4.22)
(4.23)
Solving (4.23) determines the prepotential (const in the right-hand-side is fixed by a

suitable normalization of F ).
In the case Nf = 2, we find the gauge coupling at the classical and one-loop level as

2

1Y
i
log sinh R(A + Mi ) sinh R(A Mi ) . (4.24)
2 log sinh(2RA)
=
4
i=1
In the 4-dimensional limit (4.24) reads as

1
2i
2 2 log A,
(4.25)
which gives the beta function of Nf = 2 SeibergWitten theory. On the other hand, in the
five-dimensional limit we have
!
2
2
X
X
2
1
1
|A Mi |
|A + Mi |
(4.26)
5 = 2 A
16
16
i=1
i=1
in agreement with (3.3).

We have checked that the local model of F2 blown up at 3 and 4 points (see Appendix)
also reproduce (3.3). Thus we have obtained a model which appears to describe correctly
the physics of 5-dimensional theory up to the number of flavors Nf = 4 making use of the
mirror symmetry.
340
5. Discussions
Since F2 is obtained from F1 which is a one-point blow up of P2 , we effectively have up
to 5-point blow ups of P2 describing the gauge theory. Unfortunately, beyond the 5-point
blow up mirror symmetry of del Pezzo surfaces can not be described by toric geometry
and we can not apply our analysis for these cases. In fact in the range Nf > 5 something
drastic must happen, since in this asymptotically non-free region SeibergWitten solution
does not exist and we should not have a smooth 4-dimensional limit. On the other hand, in
this range the 5-dimensional gauge theories are expected to possess En , n = 6, 7, 8 global
symmetries and are of particular interests. It is a challenging problem to clarify the physics
of gauge theory with En symmetry. It may shed some light on the nature of asymptotically
non-free field theories in 4-dimensions.
Our result shows that from the point of view of the type II/M theory compactified
on CalabiYau manifold, low energy 5-dimensional theory on M4 S 1 emerges most
naturally with its prepotential being exactly the same as that of the string theory. On the
other hand, 4-dimensional SeibergWitten theory appears only in the fine-tuned limit of
the Khler parameters. It seems possible that our 5-dimensional model is an example of
an M-theoretic lift of various 4-dimensional quantum field theories. We may imagine
most of the 4-dimensional SUSY field theories in fact have a lift to 5-dimensions where
the quantum effects of the loops and instantons are replaced by purely geometrical effects
of world-sheet instantons. It will be also quite interesting to see if there is a further lift of
quantum field theory to 6-dimensions as suggested by the duality between F- and M-theory.
Acknowledgements
We would like to thank M. Jinzenji, M. Naka and Y. Ohta for discussions. Researches
of T.E. and H.K. are supported by the fund for the Special Priority Area No.707, Japan
Ministry of Education. We also acknowledge the stimulating atmosphere at Summer
Institute 99 where this research was started.
Appendix A
A.1. Nf = 0
For pure YangMills case we take the Hirzebruch surface F2 lying inside a CalabiYau
manifold. Its toric diagram has five vertices;
0 = (0, 0),
3 = (1, 0),
1 = (1, 0),
2 = (0, 1),
4 = (2, 1).
(A.1)
This is a two-dimensional reflexive polyhedra No. 4 in Fig. 1 of [10]. The charge vectors
satisfying the linear relation
`(k)
i i = 0
341
(A.2)
are given by
`(F ) = (2; 1, 0, 1, 0),
`(B) = (0; 0, 1, 2, 1).
The constraint among B-model variables Yi , i = 0, 1, . . . , 4

Y `(k)
Y `(k)
Yi i =
Yi i
(k)
ì >0
(A.3)
(A.4)
(k)
ì <0
gives
Y02 = Y1 Y3 ,
Y2 Y4 = Y32 ,
(A.5)
and we have a solution

s4
.
Y = sx, x 2 , , s 2 ,
(A.6)
Set s = 1. Then we obtain the following curve in the B-model side

X
ai Yi ,
PNf =0 =
= a0 x + a1 x 2 + a2 + a3 + a4
1
= 0.
(A.7)
Introducing a new variable y = a2 (a4 / ), we can rewrite the curve as

y 2 = (a1 x 2 + a0 x + a3 )2 4a2a4 .
(A.8)
For each charge vector `(k) , a corresponding complex structure modulus is given by
Y `(k)
ai i .
(A.9)
zk =
i
In the present case we have

a1 a3
a2 a4
zB = 2 .
zF = 2 ,
a0
a3
(A.10)
By setting
a1 = a3 = 1,
a0 = s,
4a2 a4 = K 4 ,
(A.11)
we have
1
K4
.
,
z
=
B
4
s2
GKZ system is defined by a system of differential operators
(k)
(k)
Y ì
Y ì
=
.
ai
ai
(k)
(k)
zF =
ì >0
ì <0
In the F2 case it is given by
(A.12)
(A.13)
342
L1 = zF (4z2F + 2zF ) + zF (2zB zF ),
(A.14)
L2 = zB (2zB zF + 1)(2zB zF ) z2B ,
(A.15)
zi zi
,
zi
i = F, B.
A.2. Nf = 1
It is known that matters in the vector representation are generated by blowing up the
manifold (see, for instance, [21]). We expect that a blow up of the local F2 provides the
description of gauge theory with Nf = 1 matter. The corresponding reflexive polyhedra is
No. 6 of [10] and given by the vertices
0 = (0, 0),
1 = (1, 0),
3 = (1, 1),
2 = (0, 1),
4 = (1, 0),
5 = (1, 1).
(A.16)
We see that the charge vectors are

`(B) = (0; 0, 0, 1, 2, 1),
`(F E) = (1; 0, 1, 1, 1, 0),
`(E) = (1; 1, 1, 1, 0, 0).
(A.17)
The constraint
Y3 Y5 = Y42 ,
Y0 Y3 = Y2 Y4 ,
Y0 Y2 = Y1 Y3 ,
(A.18)
is solved by

tx s
Y = sx, sx 2 , , , s, s .

Setting s = 1 gives the curve
x
1
PNf =1 = a0 x + a1 x 2 + a2 + a3 + a4 + a5 ,
or
(A.19)
(A.20)
y 2 = (a1 x 2 + a0 x + a4 )2 4a5(a2 x + a3 ).
(A.21)
If we set
a1 = a4 = 1,
a2 = w,
a0 = s,
4a5 = K 3 ,
a3 = w1 ,
(A.22)
we obtain the curve [11]

1
.
y = (x + sx + 1) K wx +
w
Complex moduli are given by
2
zB =
a3 a5 K 3
,
=
4w
a42
zF E =
a2 a4 w 2
,
=
a0 a3
s
Complete set of differential equations is given by [10]
(A.23)
zF +E =
a1 a3
1
=
.
a0 a2 sw2
(A.24)
L1 = B (B F E + F +E ) zB (2B F E )(2B F E + 1),
343
(A.25)
L2 = (F E F +E )(F E 2B )
zF E (F E + F +E )(F E B F +E ),
(A.26)
L3 = F +E (F +E + B F E ) zF +E (F +E + F E )(F +E F E ),
(A.27)
L4 = (F E F +E )B zB zF E (F E + F +E )(2B F E ),
(A.28)
L5 = F +E (F E 2B )
zF E zF +E (F E + F +E )(F E + F +E + 1).
(A.29)
We note that the last two operators are necessary to obtain a unique double log solution .
A.3. Nf = 2
We choose the following vertices
0 = (0, 0),
1 = (1, 0),
4 = (1, 0),
2 = (0, 1),
5 = (1, 1),
3 = (1, 1),
6 = (1, 1).
(A.30)
This is a reflexive polyhedra No. 8 of [10]. The curve is given by

PNf =2 = a0 x + a1 x 2 + a2
x
1
x2
+ a3 + a4 + a5 + a6 = 0,
(A.31)
or
y 2 = (a1 x 2 + a0 x + a4 )2 4a5(a6 x 2 + a2 x + a3 ).
(A.32)
Substituting the relation

a1 = a4 = 1,
a6 = w1 w2 ,
a0 = s,
4a5 = K 2 ,

w2 w1
a2 =
+
,
a3 = (w1 w2 )1 ,
w1 w2
(A.33)
we obtain the complex structure moduli

zB =
a3 a5
K2
=
,
4w1 w2
a42
zF2 =
a1 a3
1
,
=
a0 a2 s(w12 + w22 )
zF1 =
a2 a4 w12 + w22
,
=
a0 a3
s
zF3 =
a1 a2 w12 + w22
=
.
a0 a6
sw12 w22
(A.34)
A.4. Nf = 3
Polyhedron for the 3-point blow up is obtained by adding a vertex 7 = (0, 1) to (A.30)
(No. 12 of [10]). Elliptic curve is given by
PNf =3 = a0 x + a1 x 2 + a2
x
1
x2
+ a3 + a4 + a5 + a6 + a7 x = 0
(A.35)
or
y 2 = (a1 x 2 + a0 x + a4 )2 4(a6x 2 + a2 x + a3 )(a7 x + a5 ).
(A.36)
344
By choosing the variables as

a0 = s,
a1 = 1,
a4 = 1,
a5 =

K w1 w2
K
+
,
,
a3 =
4 w2 w1
4w1 w2
Kw1 w2
,
a7 = w3 ,
a6 =
4
a2 =
1
,
w3
(A.37)
we find the complex moduli

a3 a5
K
a2 a4 1 2
=
,
zF1 =
= w1 + w22 ,
2
4w1 w2 w3
a0 a3
s
a4
1
a1 a3 1
=
=
,
a0 a2
s w12 + w22

a1 a2 1 1
1
a4 a7 w32
.
=
=
+
=
=
,
z
F
4
a0 a6
s w12 w22
a0 a5
s
zB =
zF2
zF3
(A.38)
A.5. Nf = 4
Polyhedron for the 4-point blow up is obtained by further adding a vertex 8 = (1, 1)
to the polyhedron of Nf = 3 (No. 15 of [10]). Elliptic curve is given by
PNf =4 = a0 x + a1 x 2 + a2
+ a5 + a6
x
1
+ a3 + a4
x2
+ a7 x + a8 x 2 = 0.
(A.39)
By eliminating it becomes
y 2 = (a1 x 2 + a0 x + a4 )2 4(a6x 2 + a2 x + a3 )(a8 x 2 + a7 x + a5 ).
By choosing the variables as
(A.40)

w1 w2
a1 = 1,
a2 =
+
q,
a0 = s,
w2 w1
1
1
q,
a4 = 1,
a5 =
,
a3 =
w1 w2
w3 w4

w3 w4
a7 =
+
,
a8 = w3 w4 ,
a6 = w1 w2 q,
w4 w3
(A.41)
we find the complex moduli

a3 a5
q
=
,
2
w1 w2 w3 w4
a4
1
a1 a3 1
=
=
,
a0 a2
s w12 + w22
zB =
zF2
zF4 =
a4 a7 w32 + w42
,
=
a0 a5
s
a2 a4 1 2
= (w + w22 ),
a0 a3 s 1

a1 a2 1 1
1
zF3 =
=
+
,
a0 a6 s w12 w22

a1 a7 1 1
1
zF5 =
=
+
.
a0 a8
s w32 w42
zF1 =
(A.42)
345
References
[1] N. Seiberg, E. Witten, Electro-magnetic duality monopole condensation and confinement in
N = 2 supersymmetric YangMills theory, Nucl. Phys. B 426 (1994) 19, hep-th/9407087.
[2] N. Seiberg, E. Witten, Monopoles, duality and chiral symmetry breaking in N = 2 supersymmetric QCD, Nucl. Phys. B 431 (1994) 484, hep-th/9408099.
[3] N. Seiberg, Five-dimensional SUSY field theories, non-trivial fixed points and string dynamics,
Phys. Lett. B 388 (1996) 753, hep-th/9608111.
[4] D.R. Morrison, N. Seiberg, Extremal transitions and five-dimensional supersymmetric theories,
Nucl. Phys. B 483 (1997) 229, hep-th/9609070.
[5] M.R. Douglas, S. Katz, C. Vafa, Small instantons, del Pezzo surfaces and type I0 theory, Nucl.
Phys. B 497 (1997) 155, hep-th/9609071.
[6] K. Intriligator, D.R. Morrison, N. Seiberg, Five-dimensional supersymmetric gauge theories and
degenerations of CalabiYau spaces, Nucl. Phys. B 497 (1997) 56, hep-th/9702198.
[7] S. Katz, A. Klemm, C. Vafa, Geometric engineering of quantum field theory, Nucl. Phys. B 497
(1997) 173, hep-th/9609239.
[8] A. Klemm, On the geometry behind N = 2 supersymmetric effective actions in four dimensions,
hep-th/9705131.
[9] S. Katz, P. Mayr, C. Vafa, Mirror symmetry and exact solution of 4D N = 2 gauge theories I,
Adv. Theor. Math. Phys. 1 (1998) 53, hep-th/9706110.
[10] T.-M. Chiang, A. Klemm, S.-T. Yau, E. Zaslow, Local mirror symmetry: calculations and
interpretation, hep-th/9903053.
[11] N. Nekrasov, Five-dimensional gauge theories and relativistic integrable systems, Nucl. Phys.
B 531 (1998) 323, hep-th/9609219.
[12] H. Kanno, Y. Ohta, PicardFuchs equation and prepotential of five-dimensional SUSY gauge
theory compactified on a circle, Nucl. Phys. B 530 (1998) 73, hep-th/9801036.
[13] A. Sen, F-theory and orientifolds, Nucl. Phys. B 475 (1996) 526, hep-th/9605150.
[14] T. Banks, M.R. Douglas, N. Seiberg, Probing F-theory with branes, Phys. Lett. B 387 (1996)
278, hep-th/9605199.
[15] A. Lawrence, N. Nekrasov, Instanton sums and five-dimensional gauge theories, Nucl. Phys.
B 513 (1998) 239, hep-th/9706025.
[16] A.C. Cadavid, A. Ceresole, R. DAuria, S. Ferrara, 11-dimensional supergravity compactified
on CalabiYau threefolds, Phys. Lett. B 357 (1995) 76, hep-th/9506144.
[17] E. Witten, Phase transitions in M-Theory and F-Theory, Nucl. Phys. B 471 (1996) 195, hepth/9603150.
[18] A. Brandhuber, N. Itzhaki, J. Sonnenschein, S. Theisen, S. Yankielowizc, On the M-Theory
approach to (compactified) 5D field theories, Phys. Lett. B 415 (1997) 127, hep-th/9709010.
[19] A. Klemm, S. Theisen, M.G. Schmidt, Correlation functions for topological LandauGinzburg
models with c 6 3, Int. J. Mod. Phys. A 7 (1992) 6215.
[20] A. Klemm, B.H. Lian, S.S. Roan, S.T. Yau, A note on ODEs from mirror symmetry, hepth/9407192.
[21] S. Katz, C. Vafa, Matter from geometry, Nucl. Phys. B 497 (1997) 146, hep-th/9606086.

On the DrellLevyYan relation to O(s2)

J. Blmlein a, , V. Ravindran a,b , W.L. van Neerven a,c
a DESY Zeuthen, Platanenallee 6, D-15735 Zeuthen, Germany
b Mehta Research Institute of Mathematics and Mathematical Physics (MRI), Chhatnag Road, Jhusi, Allahabad,
211019, India
c Instituut-Lorentz, Universiteit Leiden, P.O. Box 9506, 2300 HA Leiden,The Netherlands
Abstract
We study the validity of a relation by Drell, Levy and Yan (DLY) connecting the deep inelastic
structure (DIS) functions and the single-particle fragmentation functions in e+ e annihilation which
are defined in the space-like (q 2 < 0) and time-like (q 2 > 0) regions respectively. Here q denotes
the momentum of the virtual photon exchanged in the deep inelastic scattering process or the
annihilation process. An extension of the DLY-relation, which originally was only derived in the
scaling parton model, to all orders in QCD leads to a connection between the two evolution kernels
determining the q 2 -dependence of the DIS structure functions and the fragmentation functions,
respectively. In relation to this we derive the transformation relations between the space- and timelike splitting functions up to next-to-leading order (NLO) and the coefficient functions up to NNLO
both for unpolarized and polarized scattering. It is shown that the evolution kernels describing the
combined singlet evolution for the structure functions F2 (x, Q2 ), FL (x, Q2 ) where Q2 = |q 2 | or
F2 (x, Q2 ), F2 (x, Q2 )/ ln(Q2 ) and the corresponding fragmentation functions satisfy the DLY
relation up to next-to-leading order. We also comment on a relation proposed by Gribov and Lipatov.
1. Introduction
Neutral current deep inelastic leptonnucleon scattering and single nucleon inclusive
production in e+ e pair-annihilation are formally related by crossing the kinematic
channels. Already before the advent of Quantum Chromodynamics (QCD) Drell, Levy, and
Yan mentioned the possibility [1,2] that the deep inelastic scattering structure functions at
the one side and the nucleon fragmentation functions in e+ e pair-annihilation on the other
side may be related by an analytic continuation from the t- to the s-channel. The hadronic
Corresponding author. E-mail: blumlein@ifh.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 2 2 - 3
350
J. Blmlein et al. / Nuclear Physics B 586 (2000) 349381
tensors for the space-like process of deep inelastic scattering (DIS) and the time-like single
nucleon inclusive reaction would therefore be related by
(S)
(T )
(q, p) = W
(q, p).
W
(1)
Here p denotes the nucleon momentum and q is the 4-momentum transfer to the hadronic
system, with q 2 < 0 for deep inelastic scattering and q 2 > 0 for e+ e -annihilation.
At that time the physical nucleons were considered as bound states built up of bare
nucleons and pions in the context of the Yukawa theory. The interactions were described
by BetheSalpeter [35] or Faddeev-type [6,7] equations and their generalization [811] 1
aiming at a perturbative description of the structure and fragmentation functions. In these
theories neither infrared nor collinear singularities are occurring. One may think off a
general representation of the structure and fragmentation functions in terms of current
current expectation values. However, it was already shown [1] that for e+ e -annihilation
also diagrams of distinct connectedness appear (cf. also [1416]) which are absent in DIS
so that a proof of Eq. (1) at the non-perturbative level becomes very difficult. The relation
could be established for the aforementioned ladder-models [1,1421]. 2 Within QCD the
picture changes. Here it turns out that a thorough perturbative description of the structure
functions and fragmentation functions is not possible. However, perturbation theory may
be used to describe the scaling violations of these functions at large values of |q 2 | where the
running coupling constant is small. The QCD-improved parton model, moreover, exhibits
a similarity with the approach by Drell, Levy, and Yan, since in the range where single
parton states are dominating, i.e., for the contributions of lowest twist, the non-perturbative
contributions factorize. One therefore may calculate the respective one-particle evolution
kernels and study their behavior under the crossing from the t- to the s-channel. A further
complication in the case of a vector-theory as QCD is the emergence of infrared and
also collinear singularities which have an essential impact on the crossing because of the
behavior of the kernels at x = 1. 3 Here x denotes the Bjorken scaling variable which
will be defined differently for the time-like and space-like region except for x = 1 where
both definitions lead to the same value for x. As a consequence of the BlochNordsiek
theorem [27,28] the kernels become distribution-valued for x = 1 [29].
In leading order for unpolarized scattering the crossing relations mentioned above were
given in [30]. Similar relations hold in the polarized case. In this order these kernels
are nothing but the lowest order splitting functions Pkl(0) (x), which are obtained from the
N(0)
inverse Mellin transforms of the anomalous dimensions kl [3133].
It is the aim of the present paper to investigate the validity of the DrellLevyYan (DLY)
relation, if applied to perturbatively calculable partonic structure functions and quantities
related to them up to the level of two-loop order. To establish this crossing relation
1 Later on massive vector meson ladder-models were studied in [12,13], where the crossing relation Eq. (1) was
verified for the respective kernels.
2 For a review on the early developments see [22].
3 Eq. (1) was originally postulated assuming that those terms are absent [1]. See also the subsequent discussion
in [23] pointing out that the exponent of the structure functions (1 x)p near x = 1 [2426] needs not to be
integer.
351
between space- and time-like processes one has to study scheme-invariant quantities which
are the physical evolution kernels for specific choices of observables as the unpolarized
and polarized structure and fragmentation functions or derivatives of them w.r.t. q 2 .
Furthermore, conditions are derived for the transformation of the splitting and coefficient
functions from the space- to the time-like case. For the coefficient functions we extend
the discussion to the NNLO level. Other relations between the splitting functions such
as supersymmetric relations and relations due to conformal symmetry were discussed
elsewhere, cf., e.g., [34,35].
The paper is organized as follows. Basic relations for the deep inelastic structure and
fragmentation functions are summarized in Section 2. In Section 3 scheme-invariant
combinations of coefficient and splitting functions are constructed for the space- and
time-like processes both for unpolarized and polarized deep inelastic reactions where
we consider two principal examples. The DrellLevyYan relation is studied in detail in
Section 4. We also comment on a relation by Gribov and Lipatov [36] which emerged
in the same context. Section 5 contains the conclusions. In the appendix we present the
differences between to the space- and time-like coefficient functions at O(s2 ) as well as
the convolution relations which are needed for the investigation of the DLY-relation.
2. Structure functions and fragmentation functions

Deep inelastic scattering (DIS) of a lepton (l) off a hadron target (P ) is described by the
process
l(k1 ) + P (p) l(k2 ) + X0 ,
q = k1 k2 ,
q 2 = Q2 < 0.
(2)
where X0 represents an inclusive final state. When a single gauge boson is exchanged
between the incoming lepton and the hadron the above process factorizes into the leptonic
part and the remaining hadronic part. In the case of forward scattering the scattering matrix
element can be written in terms of the leptonic tensor L and the hadronic tensor W by
|M|2 = L W .
(3)
The hadronic tensor [3740] contains the unpolarized and polarized deep inelastic structure
functions Fi and gi . If the process is mediated by photon only we have i = 1, 2 in both the
polarized and unpolarized case. Notice that instead of F1 we can also take the longitudinal
structure function FL . At asymptotic values of the kinematic variables structure functions
only depend on Q2 and the Bjorken scaling variable
xB =
Q2
,
2p.q
0 6 xB 6 1.
(4)
In QCD the Q2 dependence of the structure functions is only logarithmic and it accounts
for the violation of scaling. In the context of the parton model the structure functions
can be expressed in terms of quark and gluon densities and the corresponding space-like
(S)
coefficient functions Ci,k (k = q, g) 4
4 Similar relations hold for the polarized structure functions g (x, Q2 ) and g (x, Q2 ) on the level of twist 2.
1
2
352
(S)
Fi (xB , Q2 ) = xB
Nf
X
j =1

+ fg
Z1
ej2
xB
"

dz 1 S xB 2
Q2
(S)S
, f Ci,q z, 2
f
z Nf q z
f

#
xB 2
Q2
Q2
(S)
(S)NS
NS xB
2
, f Ci,g z, 2 + fqj
, f Ci,q
z, 2
,
z
z
f
f
i = 2, L,
(5)
where ej denotes the charge of the j th quark flavor and Nf represents the number of light
flavors. The scale f , appearing in the above equation, denotes the factorization scale
which is introduced while removing the collinear singularities from the partonic structure
functions. In addition one encounters a dependence on the renormalization scale r which
arises in the renormalization procedure. For convenience this scale is put equal to the
factorization scale in the following. Notice that the structure functions Fi and gi do not
depend on these scales. However the parton densities and the coefficient functions, which
do depend on these scales, satisfy renormalization group equations which will be shown
below.
In Eq. (5) the index (S) in the structure functions indicates the space-like nature of
the process (q 2 < 0). Furthermore, in Eq. (5) appear the singlet S and non-singlet NS
combinations of parton densities which are defined by
fqS
z, 2f
Nf
X

fqi z, 2f + fqi z, 2f ,
(6)
i=1
and

1 S
fq z, 2f ,
z, 2f = fqi z, 2f + fqi z, 2f
fqNS
i
Nf
(7)
respectively. Corresponding formulae hold for polarized scattering. In this case the
polarized parton densities and polarized coefficient functions are denoted by 1fk (z, 2f )
(k = q, g) and 1Ci,k (z, Q2 /2f ) (i = 1, 2).
Whereas in deep inelastic scattering the constituent structure of the nucleons is studied,
hadroproduction at e+ e colliders provides us with information about the fragmentation
process of these constituents into the hadrons. This information is contained in the
fragmentation functions observed in the reaction [1]
2 ) P (p) + X0 ,
l(k1 ) + l(k
q = k1 + k2 ,
q 2 Q2 > 0,
(8)
where the symbols have he same meaning as in Eq. (2). These fragmentation functions
are the analogues of the DIS structure functions. Therefore, in the QCD improved parton
model these functions can be expressed in a similar way in terms of parton fragmentation
densities Dk (k = q, g) multiplied by time-like coefficient functions, i.e.,
Fi(T ) (xE , Q2 ) = xE
nf
X
j =1
Z1
ej2
xE
"

dz 1
Q2
(T )S
S xE
2
, f Ci,q z, 2
D
z Nf q z
f

+ Dg

#
xE 2
Q2
Q2
(T )
(T )NS
NS xE
2
, f Ci,g z, 2 + Dqj
, f Ci,q
z, 2
,
z
z
f
f
353
i = 2, L,
(9)
where the corresponding scaling variable for the process in Eq. (9) is defined by
xE =
2p.q
,
Q2
0 6 xE 6 1.
(10)
The symbol T appearing within parentheses in Eq. (9) denotes that the fragmentation
functions are measured in time-like processes. The scales f and r are defined in the
same way as in Eq. (5) where like in DIS we set the renormalization scale equal to the
factorization scale. Furthermore, the definitions for the singlet and non-singlet parton
fragmentation functions are the same as those for the parton densities given in Eqs. (6),
(7). Similarly as in DIS one can also study the annihilation processes in Eq. (8) where the
hadron P is polarized. This entails the definition of the polarized fragmentation functions
(T )
(T )
denoted by g1 and g2 for which one can present a similar formula as in Eq. (9). Very
often one also encounters the transverse structure function which in the time-like and
space-like case is given by

i
1 h (R)
F2 x, Q2 FL(R) x, Q2 ,
F1(R) x, Q2 =
2x
with
R = S (x = xB ),
R = T (x = xE ).
(11)
3. Scheme-invariant combinations
In this section we give a short outline of the origin of the factorization scheme dependence of the anomalous dimensions (splitting functions) and the coefficient functions. We
also show how this dependence disappears in the evolution of the structure functions w.r.t.
the kinematic variable Q2 . The discussion below deals with the DIS structure functions but
the conclusions also hold for the fragmentation functions. The partonic structure functions
bi,k (i = 1, 2, L; k = q, g), representing the QCD radiative corrections, contain
denoted by F
various divergences. First these divergences have to be regularized for which the most convenient way is to choose the method of n-dimensional regularization. Using this method the
singularities reveal themselves in the form of pole terms of the type 1/ j , with n = 4 + ,
bi,k . The infrared divergences cancel between virtual and bremsstrahlung
in the quantity F
contributions by virtue of the BlochNordsieck theorem [27,28]. Due to the Kinoshita
LeeNauenberg theorem [41,42] all the final state mass singularities are canceled too since
the DIS structure function is an inclusive quantity. Then one is left with only two types
of singularities. The first one originates from the ultraviolet region. This type of singularities is removed via a redefinition of the parameters appearing in the QCD Lagrangian. An
example is the coupling constant which becomes equal to s (2r ) where r is the renormalization scale. After coupling constant renormalization the hadronic structure function
can be written as follows
354
Fi (x, Q ) =
2
k=q,g
!

2 2
2 Q
Fik s r , 2 , 2 , fk (x),
r
(12)
where the symbol denotes the Mellin-convolution defined by

Z1
(f g)(z) =
Z1
dz2 f (z1 )g(z2 )(z z1 z2 ).
dz1
0
(13)
Furthermore, fk is defined as the bare parton density which is scale independent and is an
unphysical object because of the singular behavior of Fik . Notice that the latter depends
on the scale r and therefore on the renormalization scheme w.r.t. the coupling constant.
The parameter originates from n-dimensional regularization because in this method the
coupling constant gets a dimension. The second type of singularity originates from the
collinear region which can be attributed to the vanishing mass of the initial state parton
represented by either the (anti-)quark or the gluon. Hence the in Eq. (12) represents the
collinear singularities which are removed from the partonic structure function via mass
factorization and transferred to a transition function lk as follows

2 2
bik z, s 2r , Q , ,
F
2 2r
!

X
2 2f
2f 2f
2 Q
2
(14)
Ci,l s r , 2 , 2 lk s r , 2 , 2 , (z).
=
r
f r
l=q,g
This procedure provides us with the coefficient function denoted by Ci,l . Substitution of
Eq. (14) into Eq. (12) leads to the result

!
X
2 2f
2f 2f
2
2 Q
2
Ci,l s r , 2 , 2 fl s r , 2 , 2
(x),
(15)
Fi (x, Q ) =
r
f r
l=q,g
where the renormalized parton density is defined as

!

X
2f 2f
2f 2f
2
2
lk s r , 2 , 2 , fk (z).
fl z, s r , 2 , 2 =
r
r
(16)
k=q,g
Since the mass factorization can be carried out in various ways one is left with an additional
scheme dependence which comes on top of the renormalization scheme dependence
entering the coupling constant in Eq. (12). The former only shows up in the parton densities
and the coefficient functions and it only disappears in specific combinations representing
physical quantities. Hence physical quantities are invariant under scheme transformation.
Like in the case of renormalization, mass factorization leads to the introduction of a scale
f called mass factorization scale which is related to the factorization scheme dependence.
Like in the latter case f drops out in physical quantities as the DIS structure functions or
fragmentation functions. The change of the parton densities and the coefficient functions
with respect to a variation in the scales r and f is determined by the renormalization
group equation (RGE) [4345]. The renormalization group equation of the parton densities
355
follow from the one presented for the transition functions lk . The latter takes the following
form

1
2
2
+
a

1
,

2f
s
lm
lm
s
f
f
2
2f
as 2f
!

2f
2
mk as f , 2 , 1, (z) = 0,

2
s f
, 1 = (1 z),
(17)
as 2f
4
where we have set r = f for simplicity. The functions Pij (as , , z) appearing in the
above equation are the splitting functions. Furthermore, the beta-function is defined by

2

2
2 d a s r
= 0 as2 2r 1 as3 2r
(18)
a s r r
2
dr
The same equation as in Eq. (17) also applies to the parton density because of the definition
in Eq. (16). The scale dependence of the coefficient function in Eq. (15) is given by

1
2
2
2
1lm + Plm as f ,
+ a s f
f
2
2f
as 2f
!

Q2
(19)
Ci,m as 2f , 2 , 1 (z) = 0.
f
As has been mentioned above scheme transformations such as
X
X
1
Zlm mk ,
Ci,l
,
C i,m Zml
lk
m=q,g
(20)
m=q,g
will not alter the physical observable like, e.g., the structure functions and fragmentation
functions. The relation between the splitting and coefficient functions computed in two
different schemes is found to be
X
X

d
Zlm Pmn Z 1 nk 2(as )
Zlm
(21)
Z 1 mk ,
Plk =
da
s
m=q,g
{m,n}=q,g
X

C i,m Z 1 ml .
(22)
Ci,l =
m=q,g
Below we present the relation between the coefficient functions computed in two different
schemes up to order as2 (Q2 ). Notice that we have chosen here 2f = Q2 in order to get rid
off the logarithms ln(Q2 /2f ) which usually appear. Up to O(as2) one obtains

(1)
(2)
(1)
(2)
(1) 2
(1)
(1)
+ Zqq
+ Zqg
Zgq
+ as2 C i,q + Zqq
Ci,q = (1 z) + as C i,q + Zqq

(1)
(1)
(1)
(1)
+ C i,q Zqq
+ C i,g Zgq
+ ,
(23)

(1)
(2)
(1)
(2)
(1)
(1)
(1)
+ Zqg
+ Zqg
+ Zqg
Zgg
+ Zqq
+ as2 C i,g
Ci,g = as C i,g

(1)
(1)
(1)
(1)
+ C i,g Zgg
+ .
(24)
+ C i,q Zqg
356
In the subsequent part of the paper it is much more convenient to derive the expressions in
the Mellin transform representation so that one can avoid the convolution symbol . The
Mellin transform of a function f (z) is given by
Z1
f
(N)
dz zN1 f (z).
(25)
In this way Eq. (13) can be written as

Z1
(f g) =
dz zN1 (f g)(z) = f N g N .
(26)
Since the structure functions are scheme independent they become renormalization group
invariants. Hence they satisfy the RG equation

2

+
a
(27)
FiN (x, Q2 ) = 0.
2f
s
f
2f
as 2f
The equation above follows from combining Eqs. (17), (19) and (15). However, the
independence of the structure function on the scales f and r is not manifest when
multiplying parton densities with coefficient functions. In particular when the perturbation
series of a physical quantity is computed up to finite order there is a residual dependence on
these unphysical scales (see, e.g., [4649]). Their influence is expected to become smaller
when higher order terms in the perturbation series are taken into account [50]. To avoid
the problem of the factorization scheme dependence of the structure function when the
perturbation series is truncated up to fixed order it is better to study evolution equations for
the structure functions with respect to a physical scale which is represented by a kinematic
variable like Q2 . In these type of evolution equations the kernels are factorization scheme
independent order by order in perturbation theory. However, the dependence on the choice
of renormalization scheme and therefore the dependence on r remains so that one is able
to obtain a better estimate of the theoretical error on s . For such an equation one needs
two different structure functions called FA (x, Q2 ) and FB (x, Q2 ). Examples are A = 2
and B = L or FA and Q2 dFA /dQ2 . Limiting ourselves to the singlet case, the evolution
equation for the non-singlet structure functions is even more simple, one can write

2f
Q2
N
2
N
2
N
2
FI (Q ) = fq as f , 2 CI,q as f , 2
Q0
f

2
f
Q2
N
as 2f , 2 , I = A, B,
(28)
+ fgN as 2f , 2 CI,g
Q0
f
N (I = A, B; l = q, g) as matrix elements so that the equation
Here one can view the CI,l
above has the form
N N
N N
CAq CAg
fq
FA
=
.
(29)
N
N
N
FB
fgN
CBq CBg
357
The coefficient functions satisfy the RG-equation in Eq. (19) and the solution is given by
the T-ordered exponential [51]

)#!
" ( asZ Q2

Q2

N (x)
N
2
N
2
, (30)
dx
CI,l as f , 2 = CI,m as Q , 1 Tas exp
2(x)
f

ml
as 2f
where N is the anomalous dimension matrix defined by

Z1
lkN
dz zN1 Plk (z).
(31)
We will now differentiate the coefficient functions w.r.t. Q2

N
N
CI,k
as 2f , Q2 /2f
as 2f , Q2 /2f
CI,k
2
2

Q
= as Q
Q2
as Q2

N
as Q2 , 1
CI,m

2
N 1
2

a
Q
,1
C
= (as Q
s
m,J
as Q2

N

1
N
Q2
1 N
as Q2 , 1 mn
as Q2 C N n,J as Q2 , 1 CJ,k
as 2f , 2 . (32)
CI,m
2
f
One can show that the expression above is invariant under scheme transformations. The
latter are given by
X
X
1
1
N N
N
Zlm
mn Z N nk + 2(as )
Zlm
(33)
Z N mk ,
lkN =
a
s
m=q,g
{m,n}=q,g
X
X
1
1

N
N
N N 1
C I,m
=
C N l,I =
Zlm
(34)
Z N ml ,
C m,I .
CI,l
m=q,g
m=q,g
Q2 -dependence
only resides in the coefficient function the same evolution

Since the
equation as in Eq. (32) also applies to FIN in Eq. (28). For a short-hand notation we
introduce the evolution variable t

as Q2
2
,
(35)
t = ln
0
as (Q20 )
so that we obtain

FAN
1
=
4
N
KAA
N
KAB

FAN
,
N
N
FBN
KBA
KBB
FBN
where the physical (scheme invariant) kernel is given by

N (t)
CI,m

0 as Q2
N
N 1
N
N
N 1
CI,m
(t)
(t)
C
(t)
.
C m,J (t)
KI J = 4
mn
n,J
t
(as Q2 )
t
(36)
(37)
The kernels KINJ depend both on the anomalous dimensions lkN and the coefficient
N (Q2 ) but the latter two quantities are combined in a factorization scheme
functions CI,l
358
independent way. This factorization scheme independence of KINJ holds order by order
in perturbation theory. Using the series expansions for the anomalous dimensions and
coefficient functions in terms of the strong coupling constant
lkN =

(n)
asn+1 Q2 N lk ,
X

(n)
N
CI,l
asn Q2 C N I,l ,
Q2 =
n=0
n=0
l, k = q, g,
I = A, B,
(38)
one can compute order by order the coefficients in the perturbation series of the kernel
KINJ =

(n)
asn Q2 K N I J .
(39)
n=0
(n)
Notice that the coefficients (K N )I J are not invariant with respect to a finite renormalization of the coupling constant. This dependence is removed when the perturbation series in
Eq. (39) is resummed in all orders.
3.1. F2 (x, Q2 ) and FL (x, Q2 )
Let us consider now two specific examples, choosing the structure functions F2 (x, Q2 )
and FL (x, Q2 ) or the structure function F2 (x, Q2 ) and its slope F2 (x, Q2 )/t as the
observables FA,B (x, Q2 ). In this combination of observables it is convenient to normalize
the structure function FL (x, Q2 ) to its gluonic contribution in lowest order. This is because
FL (x, Q2 ) vanishes in zeroth order of s due to the CallanGross relation, cf. Eq. (11),
contrary to the structure function F2 (x, Q2 ). Therefore this normalization accounts for
keeping the same order in the coupling constant for the two quantities

FLN Q2
N(S)
(40)
Q2 ,
FBN Q2 =
FAN Q2 = F2
N(1) .
as Q2 CL,g
(1)
(1)
Since both the coefficient functions CLq and CLg are scheme invariants such a
normalization is possible. We expand now the kernels KINJ for this choice of observables
into a series in as . The lowest order contribution is well-known, cf., e.g., [59],
N(0)
N(0)
= qq
K22
N(0)
KL2
N(0)
= gq
N(0)
N(0)
= gg
+
KLL
N(1)
CL,q
N(0) ,
N(1) qg
CL,g
N(1) !2
CL,q
N(0)
qg
N(1)
CL,g
N(1)
CL,q N(0)
.
N(1) qg
CL,g
N(0)
N(0)
K2L
= qg
,
N(1)
CL,q
N(1)
CL,g
To next-to-leading order in as (Q2 ), one finds

N(0)
N(0)
gg
qq
,
(41)

N(1)
N(1)
CL,q
1 N(0) CL,q
1 N(0)
N(1) N(0)
N(1)
qq N(1) qg
qg
+ N(1) C2,g qq
0
0
CL,g
CL,g
" N(2)
N(1) !2
N(1) N(2) #
CL,q
CL,q
CL,q CL,g
N(1)
N(1) N(0)
N(0)
+
C
+ C2,g
gq
qg
2,g
N(1)
N(1)
N(1) N(1)
CL,g
CL,g
CL,g CL,g
!
N(1)
N(1)
CL,q
CL,q
N(1) N(0)
N(1)
N(1)
N(1) C2,g gg + 20 C2,q N(1) C2,g ,
CL,g
CL,g

1 N(0)
N(1)
N(1)
N(1)
N(1)
N(0)
N(0)
qg
C2,g qq
gg
+ 20 C2,g
K2L = qg
0
N(1)
N(2) !
CL,q
CL,g
N(1)
N(1)
N(0)
+ C2,q + N(1) C2,g N(1) qg
,
CL,g
CL,g

N(1)
1 N(0) CL,q
1 N(0)
N(1)
N(1)
N(1)
= gq
gq
+ N(1) qq
qq
KL2
0
0
C
N(1)
K22
359
N(1)
= qq
(42)
(43)
L,g

N(1) 2
N(1)
CL,q
CL,q
1 N(0)
1 N(0)
N(1)
N(1)
qg
gg
N(1)
N(1)
0 qg
0 gg
CL,g
CL,g
#
" N(2)
N(1)
N(1) !2
CL,q N(1)
CL,q
CL,q
N(1)
N(0)
N(1) C2,q +
C2,g qq
+
N(1)
N(1)
CL,g
CL,g
CL,g
"
N(1) !3
N(1) N(2)
N(1) !2 N(2)
CL,q
CL,q CL,q
CL,q
CL,g
N(1)
C2,g + 2 N(1) N(1)

N(1)
N(1)
N(1)
CL,g
CL,g CL,g
CL,g
CL,g
#
N(1) !2
N(1)
N(2) !
CL,q
C
C
L,q
L,g
N(1)
N(1)
N(1)
N(0)
N(0)
C2,q qg
+
C
C2,q + N(1) gq
N(1)
N(1) 2,g
CL,g
CL,g
CL,g
" N(2)
#
N(1) !2
N(1)
CL,q
CL,q
CL,q
N(1)
N(1)
N(0)
+
C2,g N(1) C2,q gg
N(1)
N(1)
CL,g
CL,g
CL,g
N(2)
N(1) N(2) !
CL,q CL,g
CL,q
+ 20
N(1) N(1) ,
(44)
N(1)
CL,g
CL,g CL,g

N(1)
1 N(0) CL,q
1 N(0)
N(1)
N(1)
N(1)
KLL = gg gg + N(1) qg qg
0
0
CL,g
" N(2)
#
N(1)
N(1) N(2)
N(1) !2
CL,q
CL,q CL,g
CL,q
CL,q
N(1) N(0)
N(1)
N(0)
N(1) N(1) +
C2,g qg
N(1) C2,g qq +
N(1)
N(1)
CL,g
CL,g
CL,g CL,g
CL,g
N(1)
N(2)
CL,q
CL,g
N(1) N(0)
N(1) N(0)
C2,g gq
+ N(1) C2,g gg
+ 20 N(1) .
(45)
CL,g
CL,g
360
It is evident that the representation in terms of Mellin-moments is of advantage when

compared to the corresponding x-space expressions. In the latter case one has to find the
N and N appear in the denominators of
inverse Mellin transforms of quantities where Ci,k
kl
the expressions above which in general is not possible.
3.2. F2 (x, Q2 ) and F2 (x, Q2 )/t
A second example concerns the structure function F2 (x, Q2 ) and its slope. Both
quantities are well measurable in the present-day deep inelastic scattering experiments.
The observables FA,B (x, Q2 ) are here

FAN Q2 = F2(S)N Q2 ,

FBN Q2 = F2(S)N Q2 .
t
This example has been considered before in [54]. In leading order one obtains
N(0)
N(0)
= 0,
K2d = 4,

1
N(0)
N(0) N(0)
N(0) N(0)
gg qg
gq
,
Kd2 = qq
4
The next-to-leading order kernels read:
K22
N(1)
K22
(46)
= 0,
N(0)
Kdd
N(0)
N(0)
= qq
+ gg
.
(47)
(48)
N(1)
K2d
= 0,
(49)
h
i
1
N(1)
N(0) N(1)
N(1) N(0)
N(1) N(0)
N(0) N(1)
qq + gg
qq qg
gq qg
gq
Kd2 = gg
4

0 N(1) N(0)
1
N(0) N(0)
N(0) N(0)
N(0)
gg gq
qg
+ gg
20
qq
+ C2,q qq
20
2
N(1) h
i

0 C2,g
N(0) 2
N(0) N(0)
N(0) N(0)
N(0)
+
2
0
qq
qq
gg
qg
gq
qq
N(0)
2 qg
!
N(0) N(1)
qq qg
0
N(1)
qq
,
(50)
N(0)
2
qg

1 N(0)
N(1)
N(1)
N(1)
N(0)
+ gg
qq + gg
Kdd = qq
0

i
20 h N(1) N(0)
N(1)
N(0)
N(1)
20 qg
21 . (51)
+ 40 C2,q
N(0) C2,g qq gg
qg
For this combination in next-to-leading order the evolution depends on two evolution
kernels only. In the case of polarized deep inelastic scattering similar relations apply
considering the structure function g1 (x, Q2 ) and its slope. The anomalous dimensions and
coefficient functions of the unpolarized case have to be substituted by those for polarized
scattering.
Although enforced by Eq. (36) one has still to show that the kernels (41)(45), (47)(51)
are scheme-independent by an explicit calculation, which we have done using Eqs. (33),
(34) for the next-to-leading order contributions. In leading order the scheme-invariance
361
is visible explicitly, since the leading order anomalous dimensions and the lowest order
coefficient functions for FL (x, Q2 ) are scheme invariants.
With the help of the evolution Eq. (36) we are now prepared to ask for the validity of
crossing relations between different space- and time-like quantities in perturbative QCD.
Relations of this kind are henceforth called DrellLevyYan (DLY) relations, although the
original reasoning of these and other authors was quite different. One condition to ask
such a question at all is that the behavior of all contributing parts under crossing from
space- to time-like momentum transfer are controlled. At large momentum transfers |q 2 |
the single parton picture applies and the non-perturbative parton densities factorize. This
makes it possible to study the respective evolutions kernels without reference to the nonperturbative input densities. Even if a crossing relation for these quantities does not exist,
one still may investigate whether it exists for the perturbative evolution kernels. A further
condition for this investigation is that the latter quantities are scheme-invariant, as in
Eq. (36).
4. DrellLevyYan relations
In the following we study in detail an interesting relation between deep inelastic lepton
hadron scattering and e+ e annihilation into a hadron and anything else, proposed by
Drell, Levy and Yan [1]. Here we first briefly illustrate the idea behind the work of DLY for
completeness. In field theory, the deep inelastic e+ e annihilation can be related to matrix
elements of hadronic electromagnetic current operators similar to that of deep inelastic
leptonhadron scattering. The crucial difference, apart from the ones which originate from
the kinematics, is that the annihilation process is not related to the forward Compton
amplitude contrary to deep inelastic scattering because in the former process the hadron is
observed in the final state. Nevertheless, both processes are related by crossing symmetry
which any field theory enjoys. This motivated DLY to study the process in detail and
then relate it to the deep inelastic scattering process. From the structure of the hadronic
S (q, p) (space-like) and W T (q, p) (time-like) and using the standard reduction
tensors W
formalism one can infer that

T
S
(q, p) = W
(q, p),
W
(52)
where the momenta within the respective parentheses of the above quantities are the same
as those defined in the beginning of the paper. In the Bjorken limit for both deep inelastic
scattering and deep inelastic annihilation for q 2 = Q2 , p.q and q 2 = Q2 , p.q
, respectively, the scaling structure and fragmentation functions satisfy the following
relation 5 :

1
(S)
(T )
, i = 1, 2, L.
(53)
Fi (xB ) = (1)2(s1+s2 ) xE Fi
xE
5 Here we indicate the overall signs in case of the scattering of particles of different spin, cf. [60]. In the original
work of DLY [1] the Yukawa theory was discussed which does not contain gauge bosons.
362
Here it has been assumed that non-perturbative input parton densities can be decoupled
(T )
trivially and are the same. In other words, the functions Fi (xE ) are the analytic
(S)
continuations of the corresponding functions Fi (xB ) from 0 < xB 6 1 to 1 6 xE < .
This is true only when the continuation is smooth, i.e., if there are no singularities for
example at x = 1, etc. This relation is called DLY-relation in the literature.
In this section, we study this property in more detail extending earlier work [34]. It
is particularly interesting to study the above transformation at the level of the splitting
functions and coefficient functions which constitute the physical quantities such as the
structure and fragmentation functions. Then we show how these relations are preserved for
the physical quantities by looking at the kernels discussed in the previous section. Apart
from scaling violation one also encounters distributions of the type

i
ln (1 z)
, i = 0, 1, 2, . . . ,
(54)
(1 z),
1z
+
which destroy the continuation through z = 1. Here the distribution (lni (1 z)/(1 z))+
is represented by

i
lni (1 z)
lni+1
ln (1 z)
+ (1 z)
,
(55)
= (1 z)
1z
(i + 1)
(1 z)
+
where 1. It turns out that the DLY-relation is violated for the coefficient functions
and splitting functions separately because both are scheme dependent. This in particular
happens when we adopt the MS-scheme. Here the relation is already violated up to oneloop order for the coefficient functions. Although one can choose other schemes in which
Eq. (53) is preserved (see [34]) up to one-loop order we do not know whether this will hold
up to any arbitrary order in perturbation theory.
Let us start with the simplest examples and consider the scheme-invariant evolution
kernels Eqs. (41), (47).
4.1. The DrellLevyYan relations at leading order
In the case of the scheme-independent evolution kernels describing the evolution of
F2 (x, Q2 ) and F2 (x, Q2 )/t, respectively, or its polarized counterpart for the structure
function g1 (x, Q2 ), only the transformation of two combinations of the leading order
anomalous dimensions has to be considered, cf. (47). These are the determinant and the
trace of the singlet anomalous dimension matrix at leading order. In both quantities the
color factors of the off-diagonal elements enter only as a product. The unpolarized and
polarized leading order splitting functions read

3
1 + z2
(0)
(0)
(56)
(z) = 1Pqq
(z) = 4CF
+ (1 z) ,
Pqq
(1 z)+ 2

(0)
(z) = 8TR Nf z2 + (1 z)2 ,
(57)
Pqg
2

(0)
2
(58)
1Pqg (z) = 8TR Nf z (1 z) ,
(0)
(z) = 4CF
Pgq
1 + (1 z)2
,
z
(59)
1 (1 z)2
,
z

1z
z
(0)
+ z(1 z) + 20 (1 z),
+
Pgg (z) = 8CA
(1 z)+
z

1
(0)
+ 1 2z + 20 (1 z).
1Pgg (z) = 8CA
(1 z)+
(0)
1Pgq
(z) = 4CF
The crossing relations of the leading order splitting functions are

CF
(0)
(0) 1
(0)
(0) 1
,
Pqg
,
= zPqq
=
zPqg
Pqq
z
2Nf Tf
z

2Nf Tf
(0)
(0) 1
(0)
(0) 1
Pgq =
zPgq
,
Pgg = zPgg
,
CF
z
z
363
(60)
(61)
(62)
(63)
where one demands

(1 z) (1 z).
(64)
Eq. (63) is easily verified and implies the validity of the crossing relation from spaceto time-like evolution kernels Eq. (47), i.e., the validity of the DLY-relation for this
case. For the second set of physical evolution kernels the DLY-relation follows at leading
order referring to the transformation relations for the leading order longitudinal coefficient
functions, Eqs. (78), (79) in an analogous way.
4.2. NLO splitting function
As we know, the splitting functions and coefficient functions are not physical quantities
due to their factorization scheme dependence. Hence, the naive continuation rule for these
quantities may be violated, which is indeed the case in most of the schemes, e.g., in the
MS scheme characteristic of n-dimensional regularization. It was demonstrated by Curci,
Furmanski and Petronzio [64,65] that by an appropriate modification of the continuation
rule in the MS scheme one can show that the time-like splitting functions are related to
their space-like counter parts. Since the modification of the continuation rule has to do
with the scheme one adopts, it simply amounts to finding finite renormalization factors. It
was shown that the finite renormalization factors can be constructed from the -dependent
part of the splitting function when computed in dimensional regularization [66]. In addition
to this, care should be taken when dealing with quark and gluon states which was not the
case in the work by DLY, where a color and flavor neutral field theory was discussed. The
transformation rules are:
The diagonal elements of the space-like flavor singlet splitting functions Pqq , Pgg
have to be multiplied by (1).
The off-diagonal elements of the singlet splitting functions matrix have to be
multiplied by CF /(2Nf Tf ) for Pqg and 2Nf Tf /CF for Pgq , respectively, accounting
for the interchange of the initial and final state particles under crossing.
364
Note that these transformations are automatically accounted for in the case of the leading
order physical evolution kernels discussed in the previous paragraph. Keeping this in mind
and using the known splitting functions [61,64,65,67] and the continuation rules
ln(1 z) ln(1 z) ln(z) + i,
(65)
ln() ln() + i,
(66)
one finds that

(1)(S)
(1)(T )
(T )(1)
(T )(1)
(0)
(T )(1)
(0)
Pqq
= 20 Zqq
+ Zqg
Pgq
Zgq
Pqg
,
Pqq

(1)(S)
(1)(T
)
(T
)(1)
(T
)(1)
(0)
(0)
(0)
Pgq
= 20 Zqg
+ Zqg
Pgg Pqq + Pqg
Pqg

(T )(1)
(T )(1)
Zqq
Zgg
,

(1)(S)
(1)(T )
(T )(1)
(T )(1)
(0)
(0)
(0)
Pqg
= 20 Zgq
+ Zgq
Pqq
Pgg
+ Pgq
Pgq

(T )(1)
(T )(1)
Zgg
Zqq
,
(1)(S)
(1)(T )
(T )(1)
(T )(1)
(0)
(T )(1)
(0)
Pgg
= 20 Zgg
+ Zgq
Pqg
Zqg
Pgq
,
Pgg
(67)
(68)
(69)
(70)
where the quantities with a bar denote that they are continued from z 1/z with the
appropriate factors in front. These quantities read in explicit form:

CF
(n)
(n) 1
(n)
(n) 1
,
Pqg (z) =
,
zP
Pqq (z) = zPqq
z
2Nf Tf qg z
(71)

2Nf Tf
(n)
(n) 1
(n)
(n) 1
zPgq
,
Pgg (z) = zPgg
.
Pgq (z) =
CF
z
z
The relations given in Eqs. (67)(71) remain true for the polarized splitting functions [62,
63,66] as well. The renormalization factors appearing in the Eqs. (67)(70) are given by

(72)
ZijT (1) = Pj(0)
i ln(z) + aj i .
The constants aij are different in the unpolarized and polarized case. For unpolarized
scattering they read
1
aqg = ,
2
whereas in the polarized case
aqq = agg = 0,
aij = 0.
1
agq = ,
2
(73)
(74)
The logarithms in the renormalization factors originate from the kinematics. In dimensional
bi,k in Eq. (12) from
regularization, when one continues the partonic structure function F
the space-like to the time-like region one obtains an additional factor z which when
multiplied with the pole in yields ln(z). Since the pole is always associated with the
splitting functions, one has the function Pij(0) along with ln(z). The z-independent constant
aij , which is also multiplied by the splitting function, results from the polarization average.
For deep inelastic scattering one averages the processes with one gluon in the initial state
by a factor 1/( + 2). Such an average is not needed for the annihilation process since here
365
the gluon appears in the final state. Notice that the average over the polarization sum does
not show up in the polarized structure functions. Therefore in this case the constants aij
are zero.
The transformation behavior of the non-singlet splitting functions in NLO have been
worked out in [64,65] where also the relations for the NLO non-singlet coefficient
functions were presented.
4.3. NLO coefficient functions
Now, let us study how space-like and time-like coefficient functions are related.
The coefficient functions are expected to violate the DLY-relation due to their scheme
dependence. Here we first present the relations between the space-like and time-like
coefficient functions Ci,k (z) (i = 1, L; k = q, g). The leading order transverse coefficient
functions are identical. At next-to-leading order, in the MS scheme [6870], the coefficient
functions are related by the Z-factors in Eq. (72) as follows:

(T )(1)
(S)(1) 1
(T )(1)
= Zqq
,
(75)
C1,q (z) + z C1,q
z

CF
1 (T )(1)
(S)(1) 1
(T )(1)
(z)
.
(76)
C
= Zqg
2zC1,g
2 1,g
2Nf Tf
z
Since the coefficient functions depend on the hard scale of the process, one has to replace
the space-like q 2 by the time-like q 2 in addition to Eqs. (65)(66). This leads to the
following continuation rule
2
2
Q
Q
ln 2
i.
(77)
ln 2
f space-like
f time-like
The Z-factors get contributions from two sources. The first one is z-dependent and comes
from the phase space integrals. The time-like phase space acquires an extra factor z
which gives a finite contribution when being multiplied with the pole terms 1/. The pole
term originates from the collinear divergence in n-dimensional regularization. The second
term originates from the polarization average which is again absent in the time-like case.
The continuation rules given in Eqs. (65)(64), (77) are essential to get the constant (2)
right when one goes from the space-like to the time-like region. Notice that the space-like
coefficient function contains 4 (2)(1 z) and the time-like one contains 8 (2)(1 z).
The difference which is 12 (2) can be understood to originate from the one-loop vertex
correction when one continues from the space-like to time-like region in Q2 . The same
also holds when other regularization methods for the collinear divergences are chosen. It
is worth noticing that if one would replace ln(1 z) ln(1 z) ln(z) contrary to the
prescription in Eq. (65) one would obtain an additional term 12 (2) on the right-hand side
of Eq. (75).
The zeroth order longitudinal coefficient functions are identically zero so that the first
order contributions are scheme independent. This implies that there are no pole terms in
bL,k . Hence, there is no left-over finite piece
the corresponding partonic structure function F

which could arise from the z - or n-dimensional polarization average. We find
366

z (S)(1) 1
(T )(1)
CL,q (z) CL,q
= 0,
2
z

CF
1 (T )(1)
(S)(1) 1
CL,g (z) +
= 0.
z CL,g
2
2Nf Tf
z
(78)
(79)
4.4. NNLO coefficient functions

4.4.1. Longitudinal coefficient functions
We consider the NNLO correction to the longitudinal coefficient function. We follow
the results given in [71,7377] for the space-like and [79,80] for the time-like case. It turns
out that the coefficient functions are related by the Z-factors through the matrix-valued
convolutions

z (S)(2) 1
(T )(2)
CL,q (z) + CL,q
2
z

z (1)(S) 1
CF
z (S)(1) 1
(T )(1)
(T )(1)
CLq
+ Zgq
,
(80)
= Zqq
CL,g
2
z
2Nf Tf
2
z

CF
1 (T )(2)
(S)(2) 1
CL,g (z) +
zCL,g
2
2Nf Tf
z

z
CF
1
z (1)S 1
(S)(1)
(T )(1)
(T )(1)
+ Zgg
.
(81)
CL,q
= Zqg
CL,g
2
z
2Nf Tf
2
z
The right-hand side of the above equation contains the convolutions of Z-factors with the
continued NLO longitudinal space-like coefficient functions. We have found this pattern
by comparing the scheme transformation which we derived in the last section. The reason
for this structure relies on the fact that CL,i is obtained as the difference between C2,i and
C1,i . Since the NLO coefficient functions involve various Nielsen integrals, we used the
following identities to simplify the expressions:

1
1
(82)
= Li2 (z) ln2 (z) (2),
Li2
z
2

1
1
= ln2 (z) Li2 (1 z),
(83)
Li2 1
z
2

1
1
(84)
= ln3 (z) + S1,2 (1 z),
S1,2 1
z
6

1
1
= ln3 (z) + S1,2 (1 z) Li3 (1 z) + ln(z)Li2 (1 z),
(85)
Li3 1
z
6

1
1
(86)
= Li3 (z) + ln3 (z) + (2) ln(z),
Li3
z
6

1
1
(87)
= S1,2 (z) + Li3 (z) ln(z)Li2 (z) ln3 (z) + (3).
S1,2
z
6
If we do not continue ln(1 z) and replace ln(1 z) ln(1 z) ln(z), then terms
proportional to (2) are not compensated between space-like and time-like coefficient
functions and hence the relations given in Eqs. (80), (81) are no longer true. Although
367
(2)S,T
formally of NNLO, the coefficient functions CLq(G)
(z) may be combined to physical
evolution kernels together with the NLO splitting functions as shown in Section 3.1. In
Section 4.5 we will show that because of the transformation in Eqs. (80), (81) the physical
evolution kernels in Sections 3.1 and 3.2 remain DLY-invariant.
4.4.2. Transverse coefficient functions

In NNLO physical evolution kernels for the transverse structure and fragmentation
function can only be constructed when the space-like and time-like three-loop splitting
functions are known. If they become available one can extend Eqs. (47)(51) up to second
order. Here we consider the relation between the space- and time-like coefficient functions
using the transformation relations (64), (65), (66) and (77) for unpolarized and polarized
scattering.
The space-like coefficient functions for unpolarized scattering are computed in [7377]
whereas the time-like ones can be found in [79,80]. The transverse coefficient functions
are related by (see Appendix A.1):

(T )(2)
(S)(2) 1
(z) + z C1,q
C1,q
z
"
#

1
(1)S 1
(T )(1)
(T )(1)
(0)
(T )(1)
(0)
2(z)Pqq
+ 20 Zqq
+ Zgq
Pqg Zqg
Pgq ln(z)
=
4
z

1 (T )(1)
(S)(1) 1
(T )(1)
(T )(1)
Zqq
+ Zqq
zC1,q
+ Zqq
2
z

C
1
1 (T )(1)
F
(S)(1)
(T )(1)
(T )(1)
Zqg
+ Zgq
zC
+ Zgq
2
2Nf Tf 1,g
z
2
2
Q
1 (0)
(0)
Pgq
+ 12CF2 (2) 2 ln 2 3 (1 z).
(88)
+ Pqg
8
f
For the polarized NNLO coefficient functions which were derived in [7377] and [79,80],
we find that the form of Eqs. (88) is the same but the term
1 (0)
(0)
P Pqg
8 gq
does not occur.
Similarly for the gluonic coefficient functions we find
"

#
CF
1 (T )(2)
(S)(2) 1
C
(z)
2z C1,g
2 1,g
2Nf Tf
z
"

CF
1
(S)(1) 1
(T )(1)
(T )(1)
(0)
+ 20 Zqg
2zPqg
+ Zqg
Pqq
=
4 2Nf Tf
z
#

1
(T )(1)
(0)
(T
)(1)
(0)
(T
)(1)
(0)
Zqq
Pqg + Zgg
Pqg Zqg
Pgg
ln(z) +
2

1 (T )(1)
(S)(1) 1
(T )(1)
(T )(1)
Zqq
+ Zqg
zC1,q
+ Zqg
2
z
368

CF
1 (T )(1)
(S)(1) 1
(T )(1)
(T )(1)
+ Zqg
Zgg
+ Zgg
zC1,g
2
2Nf Tf
z

1 (0)
1 (0)
(0)
(0)
Pqq
.
0 Pqg + Pqg Pgg
8
16
(89)
For the polarized case, the terms

1 (0)
1
(0)
(0)
(0)
+ Pqg
Pgg
Pqq
0 Pqg
8
16
in Eq. (89) are absent. Since we do not have to average over the initial state gluon
polarization in the case of polarized scattering the term ln(z) + 1/2 multiplying the first
bracket in Eq. (89) is replaced by ln(z), cf. also Eqs. (72), (74).
4.5. NLO physical evolution kernels
After having found the relations between space-like and time-like splitting and
coefficient functions, we investigate the DLY-transformation for the physical evolution
kernels presented in Sections 3.1 and 3.2. In order to do this, we define the difference
between the time-like quantities KITJ and the continued space-like quantities K ijS by
KI J = KITJ K ISJ ,
(90)
where K ISJ is obtained by transforming KISJ to the time-like region using the continuation
rules (65), (66), (64), (77). Application of the DLY-transformations provides us with the
following results
N(1)
N(1)
= qq
K22
"
N(1)
C L,q
N(1)
C L,q
C L,g
N(1)
C L,g
N(2)
CL,q
N(1)
C L,g
N(1)
gq
+
N(1)
N(1) 2
CL,q
N(1)
C L,g
N(1) N(0)
C2,g
qq
N(1)
C2,g
N(1)
N(2)
C L,q
CL,g
N(1)
N(1)
C L,g C L,g
#
N(0)
qg
!
N(1)
C L,q
N(1)
N(1)
N(1)
N(0)
N(1) gg C2,g + 20 C2,q N(1) C2,g
C L,g
C L,g
N(1)
(T )N(1)
N(0) (T )N(1)
N(0) (T )N(1)
20 Zqq
gq
Zqg
+ qg
Zgq
= qq
N(1)
C L,q
N(1)
(T )N(1)
(T )N(1) N(0)
(T )N(1) N(0)
+ N(1) gq
+ 20 Zqg
Zqg
qq + Zqq
qg
C L,g
N(1)
N(0)
+ gq
C2,g
N(1)
C L,q

(T )N(1) N(0)
(T )N(1) N(0)
Zgg
qg + Zqg
gg
.
N(1)
Substituting the expressions for qq
N(1)
and gq
N(1)
= 0.
K22
For the remaining evolution kernels one obtains
(91)
using Eqs. (67), (69), we get

(92)

N(1)
K2L
369

N(1)
(T )N(1)
(T )N(1)
N(0)
N(0)
= gq
20 Zqg
+ Zqg
gg
qq

N(0)
(T )N(1)
(T )N(1)
Zgg
Zqq
,
qg
(93)
N(1)
N(1)
N(0)
(T )N(1) N(0)
(T )N(1) N(0)
= gg
20 gg
+ Zqg
gq Zgq
qg
KLL
N(1)
C L,q h
N(1)
C L,g
N(1)
(T )N(1)
(T )N(1) N(0)
(T )N(1) N(0)
20 Zqg
+ Zqg
qq Zqg
gg
gq
i
(T )N(1) N(0)
(T )N(1) N(0)
Zqq
qg + Zgg
qg ,
!
N(1)
N(1)
N(1) 2
C
C
C L,q
L,q
L,q
N(1)
N(1)
N(1)
N(1)
N(1)
gq N(1) gg
KL2 = qg + N(1) qq
N(1)
C L,g
C L,g
C L,g
(94)
(T )N(1)
(T )N(1) N(0)
(T )N(1) N(0) N(0)
20 Zgq
Zgq
qq + Zqq
gq gq
(T )N(1)
(T )N(1) N(0)
Zgg
+ Zgq
gg
+
+
N(1) h
C L,q
N(1)
C L,g
(T )N(1)
(T )N(1) N(0)
(T )N(1) N(0)
Zqg
gq + Zgq
qg
20 Zqq
N(1)
C L,q
N(1)
C L,g
!2
(T )N(1)
(T )N(1) N(0)
(T )N(1) N(0)
Zqg
qq Zgg
qg
20 Zqg
(T )N(1) N(0)
(T )N(1) N(0)
+ Zqq
qg + Zqg
gg
N(1) h
C L,q
N(1)
C L,g
i
(T )N(1)
(T )N(1) N(0)
(T )N(1) N(0)
Zqg
gq + Zgq
qg .
20 Zgg
(95)
The explicit expressions for the differences in the coefficient function are given in
Appendix A.1 as well as a series of involved Mellin convolutions leading to Nielsen
integrals (see Appendix A.2), which are necessary in the explicit calculation.
Using Eqs. (67)(70), leads to
KL2
N(1)
= 0,
(96)
N(1)
K2L
N(1)
KLL
= 0,
(97)
= 0.
(98)
The physical evolution kernels KI,J for the evolution of the structure functions F2 and
FL are thus DLY-invariant to next-to-leading order if continued from the space-like to the
time-like region.
We turn now to the physical evolution kernels in next-to-leading order where we
choose the physical quantities F2 , F2 /t as a basis. Here only two evolution kernels
are contributing, which change under the DLY-transformation as follows:
Kd2 =

0 N(1)
(T )N(1)
N(0)
N(0)
C2q Zqq
+ gg
20
qq
2
370
Kdd

N(1)
(T )N(1)
N(0) 2
N(0) N(0)
Z
qq
gg
C
qq
qg
2g
N(0)
0
2gq

N(0) N(0)
N(0)
+ 2qg
gq 20 qq
,

0
N(1)
(T )N(1)
N(0)
N(0)
= 2 N(0) C2g Zqg
gg
20
qq
gq

N(1)
(T )N(1)
Zqq
.
+ 40 C1q
(99)
(100)
From Eqs. (75), (76), (78), (79) we can derive that

Kd2 = 0,
(101)
Kdd = 0.
(102)
From these results it is clear that the time-like physical evolution kernels KijT can be
directly derived from the space-like physical evolution kernels using the continuations
in Eqs. (65), (66), (64), (77) where one has to account for the corresponding changes
in the overall color factors. The Z T -factors which are needed for the transformation of
the splitting and coefficient functions cancel in the expression above. In the future one
can extend the investigation performed in this section to physical evolution kernels at the
NNLO-level, provided the three-loop anomalous dimensions are calculated. For the choice
of observables (F2 , FL ) one also needs the three-loop coefficient functions.
We finally would like to comment on a relation derived by Gribov and Lipatov in [36]
for the leading order kernels for a pseudoscalar and a vector field theory. 6 One may write
it in the form
K(xE , Q2 ) = K(xB , Q2 ),
(103)
where K and K denote the time- and space-like evolution kernels, respectively, and xB =
1/xE . One verifies, that this relation holds in leading order for the space- and time-like
splitting functions of QCD, Eqs. (56)(62), without changing the -function, Eq. (64).
Starting with next-to-leading order, this relation is not preserved. For the physical nonsinglet evolution kernels this was shown in [64,65,67] and for some singlet combinations
in [67]. We find, that also for the physical singlet combinations, Eqs. (42)(45) , (48)(51),
this relation is violated as well.
5. Conclusions
The old question, whether the scattering cross sections of deep inelastic scattering
e + P e + X0 are related to the annihilation cross section e+ + e P + X0
by a crossing relation changing from t- to s-channel was newly discussed. Since in both
reactions non-perturbative quantities such as the structure and fragmentation functions
contribute the above question cannot be answered by means of perturbation theory for
6 See also [81,82] for related work.
371
the process as a whole. However, since both the parton densities involved in the space- and
time-like process factorize if the virtuality Q2 = |q 2 | of the four-momentum transfer is
large a related question can be asked for the crossing behavior of the respective evolution
kernels, which are computable within perturbation theory. In the calculation of both
inclusive processes only two types of singularities occur, the collinear singularity and the
ultraviolet singularity. These divergences are absorbed into the bare parton densities and the
coupling constant, respectively. Two distinct renormalization group equations are implied.
They quantify the impact of the factorization and the renormalization scale on the DIS
structure functions and fragmentation functions when the perturbation series is truncated
up to a given order. However one can construct factorization-scale independent evolution
kernels which describe the scheme-invariant evolution of these physical quantities in terms
of a kinematic variable given by Q2 . This scheme invariant evolution is guaranteed up
to any finite order in perturbation theory. Notice that in finite order this method does not
remove the dependence of the physical quantities on the renormalization scheme of the
strong coupling constant or its scale r .
The first example of the application of the physical evolution kernels is the coupled
structure functions F2 (x, Q2 ) and FL (x, Q2 ) associated with the corresponding fragmentation functions in e+ e -annihilation. A second example is given by F2 (x, Q2 ) and
F2 (x, Q2 )/ ln(Q2 ). Contrary to the splitting functions (anomalous dimensions) and coefficient functions the evolution kernels of the examples above are factorization scheme
independent. For that purpose transformation relations have been derived for the splitting
functions up to NLO and the coefficient functions up to NNLO. We have also shown that
these kernels are invariant under the DrellLevyYan-transformation up to next-to-leading
order. On the other hand the GribovLipatov relation, which is valid in leading order, is
already violated at next-to-leading order. It remains to be seen how the physical evolution
kernels behave under the DLY crossing relation at NNLO, which presumes the knowledge of the yet unknown three-loop splitting functions (space- and time-like) as well the
three-loop longitudinal coefficient functions in the first example above.
Acknowledgement
We would like to thank P. Menotti for providing us a reprint of [22]. Discussions with S.
Kurth in an early phase of this work are acknowledged. This work was supported in part
by EU contract FMRX-CT98-0194 (DG 12-MIHT).
Appendix A
A.1. Coefficient functions
In this appendix we list the difference of the space- and time-like coefficient functions
in the MS scheme, which are used in Section 4 to study the validity of the DLY-relation.
Here the expressions also contain the logarithms
372
Lf = ln Q2 /2f
(A.1)
which arise when the factorization scale 2f is chosen to be different from Q2 .

The difference between the longitudinal non-singlet coefficient functions corresponding
to the processes + q q + g + g and q 0 + q + g + g, respectively, are given by

(2)NS
(A.2)
CL,q = CF2 4 2z ln(z) ln(z) 16 Li2 (1 z) + 8 8z ,
where q 0 denotes the quark in the final state which undergoes fragmentation into a hadron
P (see Eq. (8)). The difference between the longitudinal purely singlet coefficient functions
respectively,
corresponding to the processes + q q + q + q and q 0 + q + q + q,
are given by

4 2
(2)PS
CL,q = Nf Tf CF 8 6 + 4z z ln(z) + 16 ln2 (z) 16
3

128 2
304
+ 64z
z .
(A.3)
9z
9
The same is done for the longitudinal gluonic coefficient functions corresponding to the
respectively. The difference in
processes + g g + q + q and g 0 + g + q + q,
the coefficient function is given by

24
2
(2)
+ 4z
CL,g = CF2 8 1 + z ln(z) + 8 ln2 (z) 28 +
z
z

1 2
2
1
ln2 (z)
+ CA CF 16 4 + z z ln(z) 16 1 +
z
3
z

40
248
1
24z + z2 .
Li2 (1 z) 8 +
(A.4)
+ 32 1
z
9z
9
Notice that for the computation of the coefficients functions above and the ones following
hereafter one also needs the virtual contributions to the zeroth and first order partonic
processes. The differences between the transverse coefficient functions emerge from the
same processes as mentioned above Eqs. (A.2), (A.3), (A.4). In the non-singlet case we
have
(2)NS
C1,q

1
ln(z)
2
= CF CA CF 8
2 (2) 4 ln(z) ln(1 + z) + ln2 (z)
2
1+z
+ 4 2(1 z)(2) + 4(1 z) + 4(1 z) ln(z) ln(1 + z)

2
2
1+z
+ 2(1 + z) ln(z) (1 z) ln (z) ln(z) 16 Li2 (z) ln(z)
1+z

10
8
1 + 11z ln(z)
+ Nf Tf CF
9
1z

ln(z) 67
2
+ ln (z) 2 (2)
+ CA CF 4
1z 9

53 187
2
z + 2(1 + z) ln(z) (1 + z) ln (z) ln(z) + 4 (2)(1 + z) ln(z)

+2
9
9
"
+ CF2
ln(z)
1z
373

8Lf 6 + 4 ln(1 z) ln(1 z) + 6Lf 18
!

16
ln(z) ln(z)
+ 4Lf + 6
3

+ 4 4(1 + z)Lf + 1 + 5z 2(1 + z) ln(1 z) ln(z) ln(1 z)
+ 2 2(5 + z)Lf + 14 + 40z + 2(1 + z) ln(z) ln(1 z)

+ 6(1 + z)Lf ln(z) 8(2 + z) ln(z) + 7(1 + z) ln2 (z) ln(z)
!

12
+ 7 + 7z ln(z) 10 + 22z
+ 2Li2 (1 z) 4
1z

2
24
+ 13 + 13z S1,2 (1 z) + 36(1 z) + 32 (2)
1 z ln(z)
+8
1z
1z
#

(A.5)
+ 12 (2) 9 12Lf + 4L2f (1 z) .
For the purely singlet difference we obtain
(2)PS
C1,q
"

8
8 4 + 6z + z2 Lf + ln(1 z)
3

160
368 2
2
+ 2z ln(z) 160
112z
z
16 1
3z
9z
9
!

10
ln(z) ln(z) ln(z)
+ 4(1 + z) 4 ln(1 z) + 4Lf +
3

38 2
38
z z
Lf + ln(1 z)
8 1+
9z
9

4 2
+ 16 2(1 + z) ln(z) 2 3z z Li2 (1 z)
3
= Nf Tf CF
#
1808 2
1168 224 640
+
z+
z .
+ 32(1 + z)S1,2 (1 z)
9
27z
9
27
The difference between the gluonic coefficient functions equals
(2)
C1,g
"

184 2
32
24
704
+ 66z +
z + 48
+ 16z + z2 Lf
9z
9
z
3

16
32
100
+ 10z ln(z) + 40
+ 12z + z2 ln(1 z)
+ 4
3z
z
3

1
2
+ 4 2 + + z ln(1 + z) 16 1 + + z ln(z)Lf
z
z
= 2CA CF
188 +
(A.6)
374

1
2
+ 8 2 + + z ln(z) ln(1 + z) 16 1 + + z ln(z) ln(1 z)
z
z
!

40 64 52
2
2
2
+
+ z ln (z) + 4 2 z ln (1 z) ln(z)
3
3z
3
z

104 2
356
+ 4z
z ln(1 z)
+ 32 +
9z
9

104 2
284
2
+ 4z
z Lf
+ 2 2 z ln2 (1 z) + 24 +
z
9z
9

3
+ 32 1 2z ln(z)
z
!

32 2
8
2
+ 16 2 z ln(1 z) + Lf + 16 + + 16z + z Li2 (1 z)
z
z
3

2
2
+ 4 2 + + z 1 + 2 ln(z) Li2 (z) + 16 2 + + z Li3 (1 z)
z
z

856 2
418 4168 668
16
10z S1,2 (1 z)
+
z
z
+8 8
z
9
27z
9
27
#

4
+ 8 (2) 3 + + 2z 1 + 2 ln(z)
z
"

3
8
2
+ 2CF 42 + 29z (14 11z) ln(z) + 16 3 z ln(1 z)
z
z

4
+ 4(2 z) ln(z)Lf 4 2 z ln(z) ln(1 z)
z

10
2
16 2 z ln(1 z)Lf + (2 z) ln2 (z)
z
3
!

2
12 2 z ln2 (1 z) ln(z)
z
!

2
2
52
18z 8 2 z Lf 6 2 z ln(1 z) ln(1 z)
+ 64
z
z
z

10
3z Lf
+ 2 16
z
!

2
6
+ 8 (2 z) ln(z) 2 2 z ln(1 z) + Lf + 6 3z Li2 (1 z)
z
z
#

8
106
2
+ 50z .
+ 16 2 z Li3 (1 z) + 8 10 5z S1,2 (1 z) 169 +
z
z
z
(A.7)
375
We have computed the same differences between the coefficient functions corresponding
to the structure function g1 (x, Q2 ) which describes polarized scattering. The analogues of
Eqs. (A.5), (A.6), (A.7) are given by
(2)NS
1C1,q

"

1
ln(z)
2
2 (2) 4 ln(z) ln(1 + z) + ln2 (z)
= CF CA CF 8
2
1+z
+ 4 2(1 z)(2) + 4(1 z) + 4(1 z) ln(z) ln(1 + z)
#
2
1+z
+ 2(1 + z) ln(z) (1 z) ln (z) ln(z) 16Li2 (z) ln(z)
1+z
#
"

10
8
1 + 11z ln(z)
+ Nf Tf CF
9
1z
"

ln(z) 67
+ ln2 (z) 2 (2)
+ CA CF 4
1z 9
#

53 187
z + 2(1 + z) ln(z) (1 + z) ln2 (z) ln(z) + 4 (2)(1 + z) ln(z)

+2
9
9
"

ln(z)
2
8Lf 6 + 4 ln(1 z) ln(1 z) + 6Lf 18
+ CF 4
1z

16
lg(z) ln(z)
+ 4Lf + 6
3

+ 4 4(1 + z)Lf + 1 + 5z 2(1 + z) ln(1 z) ln(z) ln(1 z)
2
+ 2 2(5 + z)Lf + 18 + 36z + 2(1 + z) ln(z) ln(1 z)

+ 6(1 + z)Lf ln(z) 2(7 + 5z) ln(z) + 7(1 + z) ln2 (z) ln(z)

12
+ 7 + 7z ln(z) 2 + 14z
+ 2Li2 (1 z) 4
1z

24
2
+ 13 + 13z S1,2 (1 z) + 36(1 z) + 32 (2)
1 z ln(z)
+8
1z
1z
#

(A.8)
+ 12 (2) 9 12Lf + 4L2f (1 z) ,
(2)PS
1C1,q
= Nf Tf CF
"

16(1 4z) Lf + ln(1 z) 184 + 24z 16(2 + 3z)
!

10
ln(z) ln(z) ln(z)
ln(z) + 4(1 + z) 4Lf + 4 ln(1 z) +
3

48(1 z) Lf + ln(1 z)
376
#

+ 16 2(1 + z) ln(z) + 1 4z Li2 (1 z) + 32(1 + z)S1,2 (1 z) 112 + 112z ,
(A.9)
(2)
1C1,g
= 2CA CF

212 + 48z 32(1 2z)Lf + 4(8 + 5z) ln(z) 16(1 3z) ln(1 z)
8(4 + z) ln(z)Lf + 8(2 + z) ln(z) ln(1 + z) 8(4 + z) ln(z) ln(1 z)

4
(26 + 5z) ln2 (z) 4(2 z) ln2 (1 z) ln(z)
3

+ 32(1 z) ln(1 z) + Lf

+ 16 2 + z (2 z) Lf + ln(1 z) (8 z) ln(z) Li2 (1 z)
+ 8(2 + z) ln(z)Li2 (z) + 16(2 z)Li3 (1 z) 32(5 z)S1,2 (1 z)

+ 224 224z + 8 (2)(8 3z) ln(z)

2
+ 2CF 22 + 40z + 4(2 z)Lf 4(14 9z) ln(1 z) + 4(8 5z) ln(z)
+ 16(2 z) ln(1 z)Lf 4(2 z) ln(z)Lf + 4(2 z) ln(z) ln(1 z)

10
(2 z) ln2 (z) + 12(2 z) ln2 (1 z) ln(z)
3

8(1 z) ln(1 z) + Lf

+ 4 10 + 5z + 4(2 z) ln(1 z) + Lf 2(2 z) ln(z)

Li2 (1 z) 16(2 z)Li3 (1 z) 40(2 z)S1,2 (1 z) + 96 96z .
(A.10)
A.2. Convolutions
Here we list the convolutions of a series of functions, which are needed for the
investigation of the DLY-relation in Section 4. Using the definition in Eq. (13) we obtain

ln(z)
1
1 3
ln(z)
=
4S1,2(1 z) 2 ln(z)Li2 (1 z) ln (z) ,
(1 z) (1 z) (1 z)
6
(A.11)

ln(z)
1
ln(z)
= 2 S1,2(1 z) + ln(z)Li2 (1 z) ,
(1 z)
z
z
(A.12)
1
ln(z)
z2 ln(z) =
3 24z + 27z2 24S1,2(1 z)z2
(1 z)
12

3(1 + 4z + 5z2 ) ln(z) 2z2 ln3 (z) 12z2 ln(z)Li2 (1 z) ,
(A.13)
377
ln(z)
z ln(z) = 2 + 2z 2z S1,2 (1 z) ln(z) z ln(z)
(1 z)
z
(A.14)
ln3 (z) z ln(z)Li2 (1 z),
6
1
ln(z)
ln(z) = 2S1,2(1 z) ln3 (z) ln(z)Li2 (1 z),
(A.15)
(1 z)
6

ln(1 z) 1
ln(z)
= S1,2 (1 z) ln(1 z)Li2 (1 z) + Li3 (1 z) ,(A.16)

(1 z)
z
z
1n
ln(z)
z2 ln(1 z) = 4 5z + 9z2 8 S1,2 (1 z)z2 + ln(1 z)
(1 z)
4
+ 4z ln(1 z) 5z2 ln(1 z) 3 ln(z) 4z ln(z)
+ 2 ln(1 z) ln(z) + 4z ln(1 z) ln(z)

2z2 ln(1 z) ln2 (z) + 2 + 4z 4z2 ln(1 z)
o

4z2 ln(z) Li2 (1 z) + 4z2 Li3 (1 z) ,
(A.17)
ln(z)
z ln(1 z) = 2 + 2z 2z S1,2 (1 z) + ln(1 z) z ln(1 z)
(1 z)
z
ln(z) + ln(1 z) ln(z) ln(1 z) ln2 (z)
2

1 + z ln(1 z) + z ln(z) Li2 (1 z) + zLi3 (1 z),
(A.18)
1
ln(z)
ln(1 z) = 2 S1,2(1 z) ln(1 z) ln2 (z)
(1 z)
2

ln(1 z) + ln(z) Li2 (1 z) + Li3 (1 z), (A.19)

ln(1 z)
1
1
ln(z)
ln(z) ln2 (1 z) 2S1,2 (1 z)

=
(1 z)
1z
(1
z)
2
+

1 2
(A.20)
ln(z)Li2 (1 z) ln (z) ln(1 z) ,
2

1
1
1
ln(z)
ln(z) ln(1 z) ln2 (z) ,

=
(A.21)
(1 z) (1 z)+ (1 z)
2

1
ln(z) ln(1 z)
=
2 S1,2(1 z) ln(1 z) ln2 (z) 2 ln(z)Li2 (1 z) ,
z
z
2z
(A.22)

ln(1 z)
1 1 2
ln(z)
ln (1 z) ln(z)
=
z
1z
z 2
+

(A.23)
+ ln(1 z)Li2 (1 z) Li3 (1 z) ,
378

1
1
ln(z)
= Li2 (z) + (2) ,

z
(1 z)+
z

ln(1 z)
2
z ln(z)
1z
+

5
3
3
+ ln(1 z)
= z2 S1,2 (1 z) +
4
4z 4
(A.24)
3
1
1
ln(1 z) ln(1 z) ln(z)
4z2
z
4
1
3
ln(1 z) ln(z) + ln2 (1 z) ln(z)
2
2
1
3
3
+ ln2 (z) ln(1 z) ln2 (z) Li2 (1 z)
4
2
2

+ ln(1 z)Li2 (1 z) ln(z)Li2 (1 z) Li3 (1 z) ,
z2 ln(z) z2 ln(1 z) =
z2 ln(z)
(A.25)

z2
2 S1,2(1 z) + ln(1 z) ln2 (z) + 2 ln(z)Li2 (1 z) ,
2
(A.26)
5
1
1
3
= z + z2 z2 ln(z)
(1 z)+
4
4
2
z2 2
ln (z) + z2 ln(z) ln(1 z) + z2 Li2 (1 z),
(A.27)
2

1
ln(1 z)
= z S1,2 (1 z) + ln(1 z) ln(1 z) ln(z)
z ln(z)
1z
z
+
1 2
1
ln (1 z) ln(z) ln(z) ln(1 z) + ln2 (z)
2
2
1
ln(1 z) ln2 (z) Li2 (1 z) + ln(1 z)
2

Li2 (1 z) ln(z)Li2 (1 z) Li3 (1 z) , (A.28)
+
z ln(z)
z
1
= z ln(z) 1 + z ln2 (z) + zLi2 (1 z) + z ln(z) ln(1 z),
(1 z)+
2
(A.29)
z ln(z) z ln(1 z) = z S1,2 (1 z)

ln(1 z)
ln(z)
1z
z 2
ln (z) ln(1 z) z ln(z)Li2 (1 z),
2
(A.30)

+
= S1,2 (1 z) +
1 2
ln (1 z) ln(z)
2
1
ln(1 z) ln2 (z) + ln(1 z)Li2 (1 z)
2
ln(z)Li2 (1 z) Li3 (1 z),
379
ln(z) ln(1 z) = S1,2 (1 z)

ln(z)
1 2
ln (z) ln(1 z) ln(z)Li2 (1 z),
2
1
1
= ln2 (z) + Li2 (1 z) + ln(z) ln(1 z),
(1 z)+
2
ln(z)
1
ln(1 z)
1
(1 z) (2),
=2
(1 z)+ (1 z)+
(1 z)
(1 z)

ln(1 z)
1 2
1
ln (1 z),
=
z
1z
2z
+
1
1
1
= ln(1 z),
z (1 z)+
z

(A.31)
(A.32)
(A.33)
(A.34)
(A.35)
(A.36)

3
3
1
1
+ z z2 ln(1 z) z(1 z) + z2 ln2 (z)
2
2
2
2
+
1 2 2
+ z ln (1 z) z2 ln(z) ln(1 z) z2 Li2 (1 z), (A.37)
2
3
1
1
= + z z2 z2 ln(z) + z2 ln(1 z),
(A.38)
z2
(1 z)+ 2
2
z2
ln(1 z)
1z
ln(1 z)
z
1z

+
= (1 z) ln(1 z) + z ln(z) z ln(z) ln(1 z)

1
zLi2 (1 z) + z ln2 (1 z),
2
1
= 1 z + z ln(1 z) z ln(z),
(1 z)+

1
ln(1 z)
= Li2 (1 z) ln(z) ln(1 z) + ln2 (1 z),
1
1z
2
+
1
= ln(1 z) ln(z).
(1 z)+
(A.39)
(A.40)
(A.41)
(A.42)
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
S.D. Drell, D.J. Levy, T.M. Yan, Phys. Rev. 187 (1969) 2159; Phys. Rev. D 1 (1970) 1617.
J. Pestieau, P. Roy, Phys. Lett. B 30 (1969) 483.
E.E. Salpeter, H.A. Bethe, Phys. Rev. 84 (1951) 1232.
J. Schwinger, Proc. Nat. Acad. Sci. 37 (1951) 452; Proc. Nat. Acad. Sci. 37 (1951) 455.
J.C.Y. Chen, A.C. Chen, Adv. Atom. Molecul. Phys. 8 (1972) 71.
L.D. Faddeev, Sov. Phys. JETP 12 (1961) 1014.
S. Weinberg, Phys. Rev. 133 (1964) B232.
380
[8] S.D. Drell, D.J. Levy, T.M. Yan, Phys. Rev. Lett. 22 (1969) 744.
[9] S.J. Chang, P.M. Fishbane, Phys. Rev. Lett. 24 (1970) 847; Phys. Rev. D 2 (1970) 1084; Phys.
Rev. D 2 (1970) 1173.
[10] S.D. Drell, T.D. Lee, Phys. Rev. D 5 (1972) 1738.
[11] S.D. Drell, D.J. Levy, T.M. Yan, Phys. Rev. D 1 (1970) 1035.
[12] V.N. Gribov, L.N. Lipatov, Phys. Lett. B 37 (1971) 78.
[13] P.M. Fishbane, J.D. Sullivan, Phys. Lett. B 37 (1971) 68.
[14] A. Suri, Phys. Rev. D 4 (1971) 510.
[15] R. Gatto, G. Preparata, Nucl. Phys. B 47 (1972) 313.
[16] H.D. Dahmen, F. Steiner, Phys. Lett. B 43 (1973) 217.
[17] R. Gatto, P. Menotti, Nuovo Cimento A 2 (1971) 881; Nuovo Cimento A 7 (1972) 118.
[18] R. Gatto, P. Menotti, I. Vendramin, Lett. Nuovo Cimento 4 (1972) 79; Ann. Phys. (New York) 79
(1973) 1.
[19] P.V. Landshoff, J.C. Polkinghorne, R.D. Short, Nucl. Phys. B 28 (1971) 225.
[20] P.V. Landshoff, J.C. Polkinghorne, Phys. Rev. D 6 (1972) 3708; Nucl. Phys. B 53 (1973) 473.
[21] G. Altarelli, L. Maiani, Phys. Lett. B 41 (1972) 480; Nucl. Phys. B 51 (1973) 509.
[22] P. Menotti, in: Proceedings of the Informal Meeting on Electromagnetic Interactions, Frascati,
May 23, 1972, Pisa Preprint SNS 3/72.
[23] P.V. Landshoff, Phys. Lett. B 32 (1970) 57.
[24] S.D. Drell, T.M. Yan, Phys. Rev. Lett. 24 (1970) 181.
[25] G.B. West, Phys. Rev. Lett. 24 (1970) 1206.
[26] E.D. Bloom, F. Gilman, Phys. Rev. Lett. 25 (1970) 1140.
[27] F. Bloch, A. Nordsieck, Phys. Rev. 52 (1937) 54.
[28] D.R. Yennie, S.C. Frautschi, H. Suura, Ann. Phys. (New York) 13 (1961) 379.
[29] I.M. Gelfand, G.E. Shilov, Generalized Functions, Vol. 1, Academic Press, New York, 1964.
[30] L.N. Lipatov, Sov. J. Nucl. Phys. 20 (1975) 181.
[31] D.J. Gross, F. Wilczek, Phys. Rev. D 8 (1973) 3633; Phys. Rev. D 9 (1974) 980.
[32] H. Georgi, D. Politzer, Phys. Rev. D 9 (1974) 416.
[33] G. Altarelli, G. Parisi, Nucl. Phys. B 126 (1977) 298.
[34] J. Blmlein, V. Ravindran, W.L. van Neerven, Acta Phys. Polon. B 29 (1998) 2581, and
references therein.
[35] A.V. Belitsky, D. Mller, A. Schfer, Phys. Lett. B 450 (1999) 126.
[36] V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 (1972) 675.
[37] S.D. Drell, J.D. Walecka, Ann. Phys. (New York) 28 (1964) 18.
[38] E. Derman, Phys. Rev. D 19 (1979) 133.
[39] J. Blmlein, N. Kochelev, Nucl. Phys. B 498 (1997) 285.
[40] J. Blmlein, A. Tkabladze, Nucl. Phys. B 553 (1999) 427.
[41] T. Kinoshita, J. Math. Phys. 3 (1962) 650.
[42] T.D. Lee, M. Nauenberg, Phys. Rev. 133 (1964) B 1549.
[43] C.G. Callan Jr., Phys. Rev. D 2 (1970) 1541.
[44] K. Symanzik, Comm. Math. Phys. 18 (1970) 227; Comm. Math. Phys. B 39 (1971) 49.
[45] G. Parisi, Phys. Lett. B 39 (1972) 643.
[46] A.D. Martin, R.G. Roberts, W.J. Stirling, Phys. Lett. B 266 (1991) 173.
[47] M. Virchaux, A. Milsztain, Phys. Lett. B 274 (1992) 221.
[48] J. Blmlein, S. Riemersma, W.L. van Neerven, A. Vogt, Nucl. Phys. B (Proc. Suppl.) 51C (1996)
97.
[49] J. Blmlein, A. Vogt, DESY 99-072.
[50] W.L. van Neerven, A. Vogt, Nucl. Phys. B 568 (2000) 263.
[51] A.J. Buras, Rev. Mod. Phys. 52 (1980) 199.
[52] G. tHooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.
[53] P. Breitenlohner, D. Maison, Comm. Math. Phys. 52 (1977) 11.
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
[71]
[72]
[73]
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
381
W. Furmanski, R. Petronzio, Z. Phys. C 11 (1982) 293.

M. Glck, E. Reya, Phys. Rev. D 25 (1982) 1211.
G. Grunberg, Phys. Rev. D 29 (1984) 2315.
M. Glck, E. Reya, A. Vogt, Phys. Rev. D 45 (1992) 3986; Phys. Rev. D 48 (1993) 116.
J. Blmlein, V. Ravindran, W.L. van Neerven, J. Phys. G 25 (1999) 1551.
S. Catani, Z. Phys. C 75 (1997) 665.
A.P. Bukhvostov, L.N. Lipatov, N.P. Popov, Sov. J. Nucl. Phys. 20 (1975) 287.
W. Furmanski, R. Petronzio, Phys. Lett. B 97 (1980) 437.
R. Mertig, W.L. van Neerven, Z. Phys. C 70 (1996) 637.
W. Vogelsang, Phys. Rev. D 54 (1996) 2023; Nucl. Phys. B 475 (1996) 47.
G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
E.G. Floratos, R. Lacaze, C. Kounnas, Phys. Lett. B 98 (1981) 89; Phys. Lett. B 98 (1981) 285.
M. Stratmann, W. Vogelsang, Nucl. Phys. B 496 (1997) 41.
E.G. Floratos, R. Lacaze, C. Kounnas, Nucl. Phys. B 192 (1981) 417.
W.A. Bardeen, A.J. Buras, D.W. Duke, T. Muta, Phys. Rev. D 18 (1978) 3998.
G. Altarelli, R.K. Ellis, G. Martinelli, Nucl. Phys. B 143 (1978) 521; Nucl. Phys. B 146 (1978)
544 (Erratum); Nucl. Phys. B 157 (1979) 461.
B. Humpert, W.L. Van Neerven, Nucl. Phys. B 184 (1981) 225.
J. Sanchez Guillen et al., Nucl. Phys. B 353 (1991) 337.
W.L. van Neerven, E.B. Zijlstra, Nucl. Phys. B 417 (1994) 61; Nucl. Phys. B 426 (1994) 245,
Erratum.
W.L. van Neerven, E.B. Zijlstra, Phys. Lett. B 272 (1991) 127; Phys. Lett. B 273 (1991) 476;
Nucl. Phys. B 383 (1992) 525; Phys. Lett. B 297 (1992) 377.
J. Blmlein, S. Kurth, DESY-97-160; hep-ph/9708388; Phys. Rev. D 60 (1999) 014018.
J. Blmlein, hep-ph/0003100.
S. Moch, J. Vermaseren, hep-ph/9912355.
E.B. Zijlstra, Ph.D. Thesis, Leiden University, 1993.
P.J. Rijken, W.L. van Neerven, Nucl. Phys. B 487 (1997) 233; Phys. Lett. B 392 (1997) 207;
Phys. Lett. B 386 (1996) 422.
P.J. Rijken, W.L. van Neerven, Nucl. Phys. B 523 (1998) 245.
P.J. Rijken, Ph.D. Thesis, Leiden University, 1997.
N. Christ, B. Hasslacher, A.H. Mueller, Phys. Rev. D 6 (1972) 3543.
S. Ferrara, R. Gatto, G. Parisi, Phys. Lett. B 44 (1973) 381.

Single spin asymmetry in heavy flavor

photoproduction as a test of pQCD
N.Ya. Ivanov a, , A. Capella b,1 , A.B. Kaidalov c,2
a Yerevan Physics Institute, Alikhanian Br.2, 375036 Yerevan, Armenia
b Laboratoire de Physique Theorique, Universite de Paris XI, Batiment 210, F-91405 Orsay Cedex, France
c Institute of Theoretical and Experimental Physics, B.Cheremushkinskaya 25, 117259 Moscow, Russia
Received 22 November 1999; revised 29 May 2000; accepted 20 June 2000
Abstract
We analyze in the framework of pQCD the properties of the single spin asymmetry in heavy flavor
production by linearly polarized photons. At leading order, the parallelperpendicular asymmetry
in azimuthal distributions of both charm and bottom quark is predicted to be about 20% in a wide
region of initial energy. Our analysis shows that the next-to-leading order corrections practically
do not affect the Born predictions for the azimuthal asymmetry at energies of the fixed target
experiments. Both leading and next-to-leading order predictions for the asymmetry are insensitive
to within few percent to theoretical uncertainties in the QCD input parameters: mQ , R , F , QCD
and in the gluon distribution function. We estimate also nonperturbative contributions to azimuthal
distributions due to the gluon transverse motion in the target and the final quark fragmentation. Our
calculations show that nonperturbative corrections to a B-meson azimuthal asymmetry are negligible.
We conclude that measurements of the single spin asymmetry would provide a good test of pQCD
applicability to heavy flavor production at fixed target energies. 2000 Elsevier Science B.V. All
rights reserved.
PACS: 12.38.Bx; 13.88.+e; 13.85.Ni
Keywords: Perturbative QCD; Heavy flavor photoproduction; Single spin asymmetry
1. Introduction
In the framework of perturbative QCD, the basic spin-averaged characteristics of heavy
flavor hadro-, photo- and electroproduction are calculated up to the next-to-leading order
(NLO) [18]. During the last ten years, these NLO results have been widely used for a phenomenological description of available data (for a review see [9]). At the same time, the key
Corresponding author. E-mail: nikiv@uniphi.yerphi.am
1 capella@qcd.th.u-psud.fr
2 kaidalov@vxitep.itep.ru
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 9 - 0
N.Ya. Ivanov et al. / Nuclear Physics B 586 (2000) 382396
383
question still remains open: How to test the applicability of QCD at fixed order to the heavy
quark production? On the one hand, the NLO corrections are large; they increase the leading order (LO) predictions for both charm and bottom production cross sections approximately by a factor of 2. For this reason, one could expect that the higher order corrections
as well as the nonperturbative contributions can be essential in these processes, especially
for the c-quark case. On the other hand, it is very difficult to compare directly, without
additional assumptions, the pQCD predictions for spin-averaged cross sections with experimental data because of a high sensitivity of the theoretical calculations to standard uncertainties in the input QCD parameters. The total uncertainties associated with the unknown
values of the heavy quark mass, mQ , the factorization and renormalization scales, F and
R , QCD and the parton distribution functions are so large that one can only estimate the
order of magnitude of the pQCD predictions for total cross sections [7,8]. In particular, at
the energies of the fixed target experiments the theoretical calculations are especially sensitive to the value of the heavy quark mass which plays the role of a cutoff parameter for
infrared singularities of the theory. For the shapes of the one- and two-particle differential
distributions the above uncertainties are moderate in comparison with the ones for total
cross sections; however they are also significant. In fact these uncertainties are of the same
order as the contributions of nonperturbative effects (such as the primordial transverse motion of incoming partons and the hadronization phenomena) which are usually used for a
phenomenological description of the charm and bottom differential spectra [9,10].
Since the spin-averaged characteristics of heavy flavor production are not well defined
quantitatively in pQCD it is of special interest to study those spin-dependent observables
which are stable under variations of input parameters of the theory. In this paper we analyze
the charm and bottom production by linearly polarized photons, namely, the reactions

(1.1)
+N Q Q + X.
In particular, we calculate the single spin asymmetry parameter, A(s), which measures the
parallelperpendicular asymmetry in the quark azimuthal distribution:
A(s) =
1 d (s, = 0) d (s, = /2)

.
P d (s, = 0) + d (s, = /2)
(1.2)
Here d (s, ) d
d (s, ), P is the degree of linear polarization of the incident photon
beam, s is the centre of mass energy of the process (1.1) and is the angle between
the beam polarization direction and the observed quark transverse momentum. The main
results of our analysis can be formulated as follows:
At fixed target energies, the LO predictions for azimuthal asymmetry (1.2) are not
small and can be tested experimentally. For instance,
LO
LO
(1.3)
A s = 400 GeV2 Charm A s = 400 GeV2 Bottom 0.18.
The NLO corrections, both real and virtual, practically do not affect the Born
predictions for A(s) at fixed target energies.
At energies sufficiently above the production threshold, both leading and nextto-leading order predictions for A(s) are insensitive (to within few percent) to
384
uncertainties in the QCD parameters: mQ , R , F , QCD and in the gluon

distribution function. This implies that theoretical uncertainties in the spin-dependent
and spin-averaged cross sections (the numerator and denominator of the fraction
(1.2), respectively) cancel each other with a good accuracy.
Our analysis shows that the pT - and xF -spectra of the single spin asymmetry
also depend weekly on the variations of QCD parameters. To be specific: Theoretical
uncertainties in the pQCD predictions for the azimuthal asymmetry differential distribution
are always significantly smaller than the ones in calculations of the shape of the
corresponding spectrum of unpolarized cross section.
We conclude that the single spin asymmetry is an observable quantitatively well defined
and rapidly convergent in pQCD. Measurements of asymmetry parameters would provide
a good test of the fixed order QCD applicability to charm and bottom production.
Another important question is how the fixed order predictions for asymmetry parameters
are affected by nonperturbative contributions usually used for the description of unpolarized spectra. We estimate the nonperturbative corrections to A(s) due to the transverse
motion of partons in the target. For this purpose we use a parametrization of the intrinsic
transverse momentum distribution proposed in Ref. [10] (so-called kT -kick). It is shown
that the kT -kick corrections to the b-quark azimuthal asymmetry A(s) are negligible. Because of the smallness of the c-quark mass, the kT -kick corrections to A(s) in the charm
case are larger; they are of order of 20%.
Modeling nonperturbative effects with the help of the kT -kick and the Peterson
fragmentation function we analyze also the pT - and xF -behavior of azimuthal asymmetry
in D- and B-meson photoproduction. It is shown that for the B-meson both the kT -kick
and the fragmentation contributions are small almost in the whole region of kinematical
variables pT and xF . For the D-meson, the kT -kick corrections are essential at low values
of pT , pT . mc , and rapidly vanish at pT > mc .
The paper is organized as follows. In Section 2 we analyze the properties of heavy quark
azimuthal distributions at leading order. We also give the physical explanation of the fact
that the pQCD predictions for A(s) are approximately the same at LO and at NLO. The
details of our calculations of radiative corrections are too long to be presented in this paper; they will be reported separately in a forthcoming publication [11]. In Section 3 we
discuss the nonperturbative contributions caused by both kT -kick and Peterson fragmentation mechanisms. A comparison of the QCD predictions with the Regge model ones is also
given.
2. Single spin asymmetry in pQCD

2.1. Leading order predictions
At leading order, O(em S ), the only partonic subprocess which is responsible for heavy
quark photoproduction is the two-body photongluon fusion:
(k ) + g(kg ) Q(pQ ) + Q(pQ ).
(2.1)
385
Fig. 1. Photongluon fusion diagrams at leading order.
The cross section corresponding to the Born diagrams (see Fig. 1) is:
2 (2 )
eQ

em S
d2
R
s , t, , 2R = C
dx d
s

1 + x 2 2(1 2 )( 2 x 2 )
+
(1 + P cos 2) ,
1 x 2
(1 x 2 )2
(2.2)
where P is the degree of the photon beam polarization; is the angle between the
observed quark transverse momentum, pEQ,T , and the beam polarization direction. In (2.2)
C is the color factor, C = TF = Tr(T a T a )/(Nc2 1) = 1/2, and eQ is the quark charge
in units of electromagnetic coupling constant. We use the following definition of partonic
kinematical variables:
t = (k pQ )2 ;
t m2Q
;
x = 1 + 2
s
s = (k + kg )2 ;
u = (kg pQ )2 ;
s
4m2Q
;
= 1
s
2
pEQ,T
=

s 2
x 2 .
4
(2.3)
In the Born approximation, the invariant cross section of the single inclusive hadronic
process,
(k ) + N(kN ) Q(pQ ) + X(pX ),
can be written in the form
EQ d 3
(s, t, u, ) =
d3 pQ
(2.4)

2s d2

dz s + t + u 2m2Q g z, 2F
s , t, , 2R . (2.5)
dt d
Here g(z, 2F ) describes gluon density in a nucleon N evaluated at a factorization scale F .

The hadronic variables are related to the partonic ones as follows:
s = (k + kN )2 = s /z;
t = (k pQ )2 = t.
u = (kN pQ )2 = u/z
m2Q (1/z 1);
(2.6)
In this paper we will discuss photoproduction processes only at fixed target energies. For
this reason we do not take into account the contribution of so-called hadronic or resolved
component of the photon. This contribution is small at the energies under consideration [7].
386
To analyze the azimuthal distributions in photoproduction, it is convenient to use the

parallelperpendicular asymmetry parameters. Apart from the quantity A(s) defined by
(1.2), we will consider also the parameters Ap (pT2 ) and Ax (xF ),

1 d2 (s, pT2 , = 0) d2 (s, pT2 , = /2)
,
Ap pT2 =
P d2 (s, pT2 , = 0) + d2 (s, pT2 , = /2)
Ax (xF ) =
1 d2 (s, xF , = 0) d2 (s, xF , = /2)

,
P d2 (s, xF , = 0) + d2 (s, xF , = /2)
(2.7)
(2.8)
which describe the dependence of azimuthal asymmetry on the transverse momentum,

2 , and on the Feynman variable, x = p /p
pT2 pEQ,T
F
l
l,max , of observed particle,
2
2
2
respectively. The quantities d (s, pT , ) and d (s, xF , ) are the cross section (2.5)
integrated over xF and over pT2 , respectively. The quantity d (s, ) in (1.2) corresponds
to the cross section (2.5) integrated over both xF and pT2 .
Fig. 2. Single spin asymmetry, A(E ), in the c-quark production as a function of beam energy
E = (s m2N )/2mN ; the QCD LO predictions with and without the inclusion of kT -kick effect.
Fig. 3. The same as in Fig. 2, but for the b-quark case.
387
Unless otherwise stated, the CTEQ3M [12] parametrization of the gluon distribution
function is used. The default values of the charm and bottom mass are mc = 1.5 GeV
and mb = 4.75 GeV; 4 = 300 MeV and 5 = 200 MeV. The default values of the
factorization scale F chosen for the A(s) asymmetry calculation are F |Charm = 2mc for
the case of charm production and F |Bottom = mb for the bottom case [9,10]. Calculating
p,x
the pT - and xF -dependent asymmetries Ap (pT2 ) and Ax (xF ), we use F |Charm =
q
q
p,x
2 m2c + pT2 for charm and F |Bottom = m2b + pT2 for bottom. For the renormalization
scale, R , we use R = F .
Let us discuss the pQCD predictions for the asymmetry parameters defined by (1.2),
(2.7) and (2.8). Our calculations of A(s) at LO for the c- and b-quark are presented by
solid lines in Fig. 2 and Fig. 3, respectively. One can see that at energies sufficiently above
the production threshold the single spin asymmetry A(E ) depends weekly on E , E =
(s m2N )/2mN .
The most interesting feature of LO predictions for A(E ) is that they are practically
insensitive to uncertainties in QCD parameters. In particular, changes of the charm quark
mass in the interval 1.2 < mc < 1.8 GeV affect the quantity A(E ) by less than 6% at
energies 40 < E < 1000 GeV. Remember that analogous changes of mc lead to variations
of total cross sections from a factor of 10 at E = 40 GeV to a factor of 3 at E = 1 TeV.
The extreme choices mb = 4.5 and mb = 5 GeV lead to 3% variations of the parameter
A(E ) in the case of bottom production at energies 250 < E < 1000 GeV. The total
cross sections in this case vary from a factor of 3 at E = 250 GeV to a factor of 1.5
at E = 1 TeV. The changes of A(E ) are less than 3% for choices of F in the range
1
2 mb < F < 2mb . For the total cross sections, such changes of F lead to a factor of 2.7 at
E = 250 GeV and of 1.7 at E = 1 TeV. We have verified also that all the CTEQ3 versions
of the gluon distribution function [12] as well as the CMKT parametrization [13] lead to
asymmetry predictions which coincide with each other with accuracy better than 1.5%. 3
Fig. 4. Scaling behavior of the asymmetry parameter A() as a function of = 4m2Q /s at QCD LO.
3 Note, however, that the fixed order predictions for azimuthal asymmetry are insensitive to uncertainties in
388
In the Born approximation scaling holds: with a good accuracy the quantity A(s) is a
function of the variable , = 4m2Q /s, so that
LO

A
s
Bottom
2
m
LO
A(s)
m2c
b
(2.9)
Charm
The scaling behavior of A() (i.e., its independence of QCD /mQ ) is demonstrated in
Fig. 4. 4
Fig. 5. pT2 -distribution of azimuthal asymmetry, Ap (pT2 ), in a D-meson production; the QCD LO
predictions with and without the inclusion of the kT -kick and Peterson fragmentation effects.
Fig. 6. The same as in Fig. 5, but for the pT -distribution of azimuthal asymmetry, Ap (pT ), in a
B-meson production.
QCD parameters only at energies sufficiently above the production threshold. For instance, at E < 10 GeV,
changes of the charm quark mass in the interval 1.2 < mc < 1.8 GeV lead to 100% variations of the QCD
predictions for A(s).
4 Eq. (1.3) does not contradict to Eq. (2.9). As one can see from Fig. 4, the asymmetry parameter A() is not a
monotonic function of variable . So, Eq. (1.3) reflects only the fact that A( = 0.25) A( = 0.025).
389
In Fig. 5/Fig. 6 we show by a solid line the LO predictions for the pT2 -/pT -dependent
parameter Ap (pT2 )/Ap (pT ) describing the azimuthal asymmetry in charm/bottom production. pT -distributions grow rapidly with pT up to pT mQ and fall slowly as pT becomes
larger than the heavy quark mass. The dependence of pT -spectra on variations of mQ is
shown in Fig. 7 and Fig. 8. One can see that, contrary to A(s), the form of the pT2 -spectrum
is sensitive to the value of charm quark mass. Note however that variations of the Ap (pT2 )
shape due to changes of mc are significantly smaller than the corresponding ones for the
unpolarized distribution d2 (s, pT2 ) [9]. We have also verified that a change in the other indpT
put parameters has practically no effect on the computed values of Ap (pT2 ) for both charm
and bottom.
In the Born approximation the pT -distribution of azimuthal asymmetry is practically a
function of two variables: = 4m2Q /s and = pT2 /m2Q . The function Ap ( ) at different
values of is shown in Fig. 9.
Fig. 7. Dependence of the pT2 -spectrum, Ap (pT2 ), on the c-quark mass, mc , at QCD LO.
Fig. 8. Dependence of the pT -spectrum, Ap (pT ), on the b-quark mass, mb , at QCD LO.
390
Fig. 9. Scaling behavior of the asymmetry parameter Ap ( ) as a function of , = pT2 /m2Q , at

different values of at QCD LO.
Fig. 10. xF -distribution of the single spin asymmetry, Ax (xF ), in a D-meson production; the QCD
LO predictions with and without the inclusion of the kT -kick and Peterson fragmentation effects.
Fig. 11. The same as in Fig. 10, but for the case of a B-meson production.
391
We have calculated also the dependence of single spin asymmetry on xF . The LO

predictions for Ax (xF ) defined by (2.8) in the case of c- and b-quark production are
presented by solid lines in Fig. 10 and Fig. 11, respectively. xF -distributions take their
maximal values approximately at xF 0.5 and fall to zero as xF 1.
2.2. Radiative corrections
Let us briefly discuss the NLO corrections to the single spin asymmetry A(s). It is well
known that, at order O(em S2 ), the main heavy quark photoproduction mechanism is the
real gluon emission in the photongluon fusion:
(k ) + g(kg ) Q(pQ ) + Q(pQ ) + g(pg ).
(2.10)
According to the analysis [3,4], practically whole contribution to the heavy quark
production in photonhadron reactions at fixed target energies, E . 1 TeV, originates
from the so-called initial-state gluon bremsstrahlung (ISGB) mechanism which is due to
the Feynman diagrams with the t-channel gluon exchange (see Fig. 12). Since the gluon
distribution function increases very steeply at small z, g(z) 1/z, the order-S2 photon
hadron cross section is determined at E . 1 TeV by the threshold, s 4m2Q , behavior of
the g cross section. Near the threshold, a large logarithmic enhancement of the t-channel
gluon exchange diagrams contribution takes place in the collinear, pEg,T 0, and soft,
pEg 0, limits [3,4]. Effectively, the ISGB contribution is proportional to the Born g
differential cross section:
S

d LO
d ISGB
K(s )
s, t
s, t ,
dt
dt
s 4m2Q ,
(2.11)
where K(s ) is an enhancement factor.

Since the azimuthal angle is the same for both g and QQ centre of mass systems in
the collinear and soft limits, it seems natural to expect that the Eq. (2.11) can be generalized
to the spin-dependent case substituting the spin-averaged cross sections in (2.11) by the dependent ones. Indeed, the threshold enhancement of the ISGB mechanism is due to the
gluon propagator pole in the diagrams in Fig. 12 which is a common factor for both spindependent and spin-independent amplitudes.
It is well known that the shapes of differential cross sections of heavy quark production
in photonhadron reactions are not sensitive to radiative corrections [3,4]. For instance, the
normalized quantity
Fig. 12. t -channel gluon exchange diagrams in the photongluon fusion.
392

f s, pT2 =
where int (s) =

d
1
s, pT2 ,
2
int (s) dpT
dpT2
d
dpT2
(2.12)
2 , is practically the same at LO and at

(s, pT2 ) and pT2 pEQ,T
NLO. This fact means that non-small contributions of the enhancement factor K(s ) (NLO
corrections) to theR pT -dependent and to the pT -integrated photonhadron cross sections,
d NLO(s, pT2 ) dz g(z) d ISGB (zs, pT2 ), cancel each other in the ratio (2.12 ) with a
good accuracy. One can assume that the same situation takes place also for the single spin
asymmetry which is a ratio of the -dependent cross section to the -averaged one.
Our calculations show that it is really the case. Radiative corrections, both real
and virtual, to the photongluon fusion mechanism practically do not affect the Born
predictions for the single spin asymmetry in heavy flavor photoproduction at fixed target
energies. Moreover, the azimuthal asymmetry is independent (to within few percent) of the
theoretical uncertainties in the QCD input parameters (mQ , R , F and QCD ) at NLO
too. The details of our NLO analysis are too long to be presented here and will be given in
a separate publication [11].
As to the photonquark fusion, q QQq, the dominant production mechanism in
this reaction is the so-called flavor excitation (FE), also arising from the diagrams with
the t-channel gluon exchange in Fig. 12. However the contribution of the FE mechanism
may be essential at superhigh energies only [4]. For instance, the contribution of the
q reactions to the unpolarized bottom photoproduction makes only 510% from the
contribution of the g processes at E . 1 TeV. It is evident that an account of the
photonquark reactions cannot affect significantly the predictions of the photongluon
fusion mechanism at fixed target energies in the polarized case too.
3. Nonperturbative contributions
Let us discuss how the pQCD predictions for single spin asymmetry are affected by
nonperturbative contributions due to the intrinsic transverse motion of the gluon and
the fragmentation of produced heavy quark. Because of the c-quark low mass, these
contributions are especially important in the case of charmed particle production. In our
analysis, we use the MNR model [10] parametrization of the gluon transverse momentum
distribution,
kEg = zkEN + kEg,T .
(3.1)
According to [10], the primordial transverse momentum, kEg,T , has a random Gaussian
distribution:

kT2
1
1 d2 N
exp 2 ,
=
(3.2)
N d2 kT
hkT2 i
hkT i
2 . It is evident that the inclusion of this effect results in a dilution of
where kT2 kEg,T
azimuthal asymmetry. In [9,10], the parametrization (3.2) (so-called kT -kick) together with
the Peterson fragmentation function [14],
393
a
(3.3)
y[1 1/y /(1 y)]2
R1
(where a is fixed by the condition 0 dyD(y) = 1), have been used to describe the single
inclusive spectra and the QQ correlations in photo- and hadroproduction at NLO. It was
shown that the available data on charm photoproduction allow to choose for the averaged
intrinsic transverse momemtum, hkT2 i, any value between 0.5 and 2 GeV2 .
Our calculations of the parameter A(s) at LO with the kT -kick contributions are
presented in Fig. 2 and Fig. 3 by dashed (hkT2 i = 0.5 GeV2 ) and dotted (hkT2 i = 1 GeV2 )
curves. One can see that in the c-quark case the kT -kick reduces the value of A(s)
approximately by 15% at hkT2 i = 0.5 GeV2 and by 20% at hkT2 i = 1 GeV2 . The kT -kick
corrections to the bottom asymmetry are systematically smaller; they do not exceed 5% for
both hkT2 i = 0.5 GeV2 and hkT2 i = 1 GeV2 .
Nonperturbative contributions to the pT -dependent asymmetry parameters Ap (pT2 ) and
p
A (pT ) are shown in Fig. 5 and Fig. 6. It is seen that the kT -kick corrections decrease
rapidly with the increase of the heavy quark transverse mass. For the B-meson, they are
negligibly small in the whole region of pT . In the case of a D-meson production, the kT kick corrections are essential only in the region of low pT . mc .
Calculating the hadronization effect contributions we use for the parameter that
characterizes the Peterson fragmentation function the values D = 0.06 for a D-meson
and B = 0.006 for a B-meson [15]. Strictly speaking, according to the factorization
theorems, the application of the fragmentation function formalism can only be justified
in the region of high pT . It is seen from Fig. 5 and Fig. 6 that for pT > mQ the account
of the fragmentation function (3.3) leads to a reduction of the pT -spectra. Fragmentation
corrections to Ap (pT2 ) are of order of 15% in the case of a charmed meson production;
they are less than 5% for a B-meson case.
For completeness, in Fig. 10 and Fig. 11 we presented nonperturbative contributions to
the xF -distributions of single spin asymmetry. One can see that the kT -kick corrections
to Ax (xF ) in the bottom case are small in the whole region of xF . As expected, the
xF -distribution of the c-quark azimuthal asymmetry is more sensitive to the kT -kick
corrections.
So, we can conclude that nonperturbative corrections to the b-quark asymmetry
parameters (1.2) and (2.7) due to the kT -kick and Peterson fragmentation effects practically
do not affect predictions of the underlying perturbative mechanism: photongluon fusion.
To illustrate how strongly the azimuthal distributions depend on the basic subprocess
dynamics let us consider the mechanisms of the photon fusion with nonvector primordials.
2 0). In the
The latter can be nonperturbative, color or white objects (say, reggeons at pO
Born approximation, the partonic cross sections corresponding to the reactions
D(y) =
+OQ+Q
(3.4)
have the following form:

2
eQ
em QSQ
4(1 2 )( 2 x 2 )
2
d2 O=S
=C
(1 + P cos 2) , (3.5)
dx d
s
1 x 2
(1 x 2 )2
394
eQ em QP Q 2
d2 O=P
=C
,
dx d
s
1 x 2
2
(3.6)
where the cross sections O=S and O=P describe the photon fusion with a scalar (O = S)
and a pseudoscalar (O = P ) object, respectively. One can see from (2.2) and (3.5), (3.6)
that only the gluon contribution leads to a positive azimuthal asymmetry. In the case of
a scalar primordial, the asymmetry parameters are predicted to be negative. The photon
fusion with a pseudoscalar object (say, 0 -meson) leads to an -independent cross section.
Note that the expressions (3.5), (3.6) do not correspond to any concrete models; they
are presented only as illustrative exercises demonstrating how the azimuthal distribution
reflects the nature of the underlying subprocess.
Completing the section let us compare the QCD predictions for a D-meson single spin
asymmetry with the Regge pole model ones. In principle, at arbitrary values of variables
xF and pT , the spin dependent part of a D-meson photoproduction cross section can
be of two types. The first, |[Ee, nE 0 ]pET |2 = pET2 sin2 , corresponds to the contribution of
the t-channel exchanges with natural spin-parity while the second, |(EepET )|2 = pET2 cos2 ,
describes the contribution of the t-channel singularities of unnatural spin-parity. In the
above expressions, eE is the polarization vector of the photon, nE0 is a unit vector in
the direction of the photon momentum and is the azimuthal angle of the D-meson
momentum about nE0 : tan = ([Ee, nE 0 ]pET )/(EepET ). Really, the Regge pole model is only
applicable in the narrow region xF 1 and pT2 s (so-called triple reggeon exchange
limit). In this region, the cross section of the inclusive reaction is determined by the
contribution of the leading (rightmost in the j -plane) Regge trajectory. The diagram, which
describes the pseudoscalar meson photoproduction in the triple reggeon exchange limit, is
sketched in Fig. 13. Since the relevant leading Regge trajectory, D , is of natural spinparity, the spin structure of the corresponding reggeon-particle vertex, D D, is written
as:

D D
2
2 1
(pT ) [Ee, nE0 ]pET = pET2 (1 cos 2).
2
(3.7)
It is seen from (3.7) that, in contrast to QCD, the Regge model predicts in the region xF 1
a large negative azimuthal asymmetry such that Ax (xF )|Regge 1 as xF 1.
Fig. 13. Triple reggeon diagram for the process N DX.
395
4. Conclusion
In the investigation of mechanisms which are responsible for heavy flavor production
at fixed target energies, the reactions (1.1) with a linearly polarized photon beam are of
special interest. As it is shown in the present paper, the QCD LO predictions for the
single spin asymmetry in both charm and bottom production are non-small and can be
tested experimentally; in a wide region of initial energy the parameter A(s) defined by
(1.2) is of order of 20%. Our analysis shows that the NLO corrections practically do
not affect the Born predictions for A(s) at fixed target energies. Unlike the unpolarized
cross section, the single spin asymmetry is an observable quantitatively well defined in
QCD. In particular, at both LO and NLO, A(s) is independent to within few percent of
the theoretical uncertainties in mQ , R , F , QCD and in the gluon distribution function.
Our calculations show that the pT - and xF -distributions of the azimuthal asymmetry in
bottom production are practically insensitive to nonperturbative contributions due to the
primordial transverse motion and Peterson fragmentation effects.
So, the single spin asymmetry in heavy flavor production by linearly polarized
photons is an observable quantitatively well defined, rapidly convergent and insensitive
to nonperturbative contributions. Measurements of the azimuthal asymmetry in bottom
production would be a good test of the conventional parton model based on pQCD.
Due to the c-quark low mass, nonperturbative contributions to the charm production can
be essential. As it is shown in our paper, the pQCD and Regge approaches lead to strongly
different predictions for the single spin asymmetry in the region of low pT and xF 1.
Data on the pT - and xF -distributions of the azimuthal asymmetry in D-meson production
could make it possible to discriminate between these mechanisms.
Concerning the experimental aspects, the linearly polarized high energy photon beams
can be generated using the Compton back-scattering of the laser light off the lepton beams
(see, for instance, [16,17] and references therein). According to the above references, this
method promises to provide the beams of real photons with a definite polarization and high
monochromaticity.
Acknowledgements
This work was supported in part by the grant NATO OUTR.LG971390.
References
[1]
[2]
[3]
[4]
[5]
P. Nason, S. Dawson, R.K. Ellis, Nucl. Phys. B 303 (1988) 607.

W. Beenakker, H. Kuijf, W.L. van Neerven, J. Smith, Phys. Rev. D 40 (1989) 54.
R.K. Ellis, P. Nason, Nucl. Phys. B 312 (1989) 551.
J. Smith, W.L. van Neerven, Nucl. Phys. B 374 (1992) 36.
P. Nason, S. Dawson, R.K. Ellis, Nucl. Phys. B 327 (1989) 49; Nucl. Phys. B 335 (1990) 260
(Erratum).
396
[6] W. Beenakker, W.L. van Neerven, R. Meng, G.A. Schuler, J. Smith, Nucl. Phys. B 351 (1991)
507.
[7] M.L. Mangano, P. Nason, G. Ridolfi, Nucl. Phys. B 373 (1992) 295.
[8] S. Frixione, M.L. Mangano, P. Nason, G. Ridolfi, Nucl. Phys. B 412 (1994) 225.
[9] S. Frixione, M.L. Mangano, P. Nason, G. Ridolfi, hep-ph/9702287; in: A.J. Buras, M. Lindner
(Eds.), Heavy Flavours II, Advanced Series on Directions in High Energy Physics, World
Scientific, Singapore, 1998.
[10] M.L. Mangano, P. Nason, G. Ridolfi, Nucl. Phys. B 405 (1993) 507.
[11] N.Ya. Ivanov, A. Capella, A.B. Kaidalov, in preparation.
[12] H.L. Lai, J. Botts, J. Huston et al., Phys. Rev. D 51 (1995) 4763.
[13] A. Capella, A.B. Kaidalov, C. Merino, J. Tran-Thanh Van, Phys. Lett. B 337 (1994) 358.
[14] C. Peterson, D. Schlatter, I. Schmitt, P. Zerwas, Phys. Rev. D 27 (1983) 105.
[15] J. Chrin, Z. Phys. C 36 (1987) 163.
[16] I.F. Ginzburg, G.L. Kotkin, S.L. Panfil, V.G. Serbo, V.I. Telnov, Nucl. Instr. Meth. 219
(1984) 5.
[17] V. Telnov, hep-ph/9805002; in: P. Chen (Ed.), Quantum aspects of Beam Physics: Proceedings,
World Scientific, Singapore, 1998, p. 173.

Two-loop QCD anomalous dimensions

of flavour-changing four-quark operators
within and beyond the Standard Model
Andrzej J. Buras a , Mikoaj Misiak b,c, , Jrg Urban a
a Physik Department, Technische Universitt Mnchen, D-85748 Garching, Germany
b Theory Division, CERN, CH-1211 Geneva 23, Switzerland
c Institute of Theoretical Physics, Warsaw University, Hoza 69, PL-00-681 Warsaw, Poland
Received 25 May 2000; accepted 6 July 2000
Abstract
We calculate the two-loop QCD anomalous dimension matrix (ADM) ( (1) )NDR in the NDRMS
scheme for all the flavour-changing four-quark dimension-six operators that are relevant in both the
Standard Model and its extensions. Both currentcurrent and penguin diagrams are included. Some
of our NDRMS results for 1F = 1 operators overlap with the previous calculations, but several
others have never been published before. In the case of 1F = 2 operators, our results are compatible
with the ones obtained by Ciuchini et al. in the regularization-independent renormalization scheme,
but differ from their NDRMS results. In order to explain the difference, we calculate the ADM of
1F = 2 operators again, extracting it from the ADM of 1F = 1 operators. 2000 Elsevier Science
1. Introduction
Renormalization group short-distance QCD effects play an important role in the
phenomenology of non-leptonic weak transitions of K-, D- and B-mesons. An essential
ingredient in any renormalization group analysis is the anomalous dimension matrix
(ADM), which describes the mixing of the relevant local four-quark operators under
renormalization [1,2].
The operators considered in the present paper have the form
1 Ak 2 3 Bk 4 ,

1 Ak 2 3 Bk 4 ,
E-mail address: mikolaj.misiak@cern.ch (M. Misiak).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 3 7 - 5
(1.1)
398
A.J. Buras et al. / Nuclear Physics B 586 (2000) 397426
k
where , are colour indices and A,B
are generic Dirac matrices given explicitly below.
The subscripts i in i are flavour indices. In the case of FCNC transitions with 1F = 2,
such as neutral meson mixing, one has
1 = 3 ,
2 = 4 .
(1.2)
V A (bd)
V A relevant in the
Known examples are the operators (s d)V A (s d)V A and (bd)
0
0
0
0
Standard Model (SM) to K K and Bd B d mixing, respectively.
Four-quark operators that occur in the SM calculations of flavour-changing processes
do not form a complete set of all the dimension-six four-quark operators. Other operators
need to be considered in many extensions of the SM, e.g., in the Supersymmetric Standard
Model (SSM) (see, e.g., Ref. [3]). For instance, the SSM and SM predictions for K 0 K 0
and Bd0 B d0 mixing can have similar precision only if the two-loop ADM for all the 1F = 2
operators is known.
The main purpose of the present paper is a calculation of the two-loop ADM for all
the dimension-six flavour-changing four-quark operators in the NDRMS scheme ( MS
scheme with fully anticommuting 5 ). Our main findings are the NDRMS anomalous
dimensions of the operators with Dirac structures (cf. Eq. (1.1)):
Ak Bk = (1 5 ) (1 5 ),

Ak Bk = (1 5 ) (1 5 ) .
(1.3)
For these operators, our two-loop results differ from the NDRMS ones of Ciuchini et al.
[4], but are compatible with their RI-scheme ADM. For all the other operators, no new
calculation is actually necessary all the two-loop results can be extracted from the
existing Standard Model ones.
Our paper is organized as follows. In Section 2, we perform a direct calculation of
the NDRMS-scheme ADM of 1F = 2 operators. This is a relatively straightforward
computation, since all the methods are already known from similar SM calculations (see,
e.g., Refs. [58]). The only novelty here is the introduction of evanescent operators that
vanish by the Fierz identities.
In Section 3, we compute the NDRMS ADM for such 1F = 1 operators, to which
only the currentcurrent diagrams are relevant. Some of the 1F = 1 results have never
been published before. The ones that are not new agree with the old SM calculations. The
subject of Section 4 are 1F = 1 operators containing one quarkantiquark pair of the same
flavour. We identify the operators to which the so-called penguin diagrams are relevant, and
give the corresponding anomalous dimensions.
In Section 5, we derive the matrix 1r necessary for transforming the Wilson coefficients
from the NDRMS to the RI scheme (originally called the MOM scheme) that is more
useful for non-perturbative calculations of hadronic matrix elements [9].
Section 6 is devoted to performing a consistency check of our 1F = 1 and 1F = 2
results. The currentcurrent ADM of 1F = 1 operators is transformed there to such an
operator basis, in which the 1F = 2 results can be easily read off. This calculation serves
also as a preparation for the comparison with Ciuchini et al. [4]. Comparison with this
article and other existing literature is the subject of Section 7. We conclude in Section 8.
399
In Appendix A, we list the evanescent operators relevant to the 1F = 2 calculation.

In Appendix B, an analogous list for the 1F = 1 case is presented. Appendix C contains
additional evanescent operators that become important only when one wants to derive the
1F = 2 results from the 1F = 1 ones, as in Section 6. Appendix D is devoted to recalling
and generalizing the notion of Greek projections. Finally, Appendix E contains a list of
separate contributions from different diagrams to the one- and two-loop ADMs for 1F = 1
operators with Dirac structures (1.3).
2. Direct calculation of the ADM in the 1F = 2 case
For definiteness, we shall consider here operators responsible for the K 0 K 0 mixing.
There are 8 such operators of dimension 6. They can be split into 5 separate sectors,
according to the chirality of the quark fields they contain. The operators belonging to the
first three sectors (VLL, LR and SLL) read

= s PL d s PL d ,
QVLL
1

= s PL d s PL d ,
s PR d ,
QSLL
QLR
1 = s PL d
1

= s PL d s PL d , (2.1)
s PR d ,
QSLL
QLR
2 = s PL d
2
where = 12 [ , ] and PL,R = 12 (1 5 ). The operators belonging to the two
and QSLL
by interchanging
remaining sectors (VRR and SRR) are obtained from QVLL
i
1
PL and PR . Since QCD preserves chirality, there is no mixing between different sectors.
Moreover, the ADMs in the VRR and SRR sectors are the same as in the VLL and SLL
sectors, respectively. In the following, we shall consider only the VLL, LR and SLL sectors.
In dimensional regularization, the four-quark operators from Eq. (2.1) mix at one
loop into the evanescent operators listed in Appendix A. Specifying these evanescent
operators is necessary to make precise the definition of the NDRMS scheme in the
effective theory [5,8,10,11]. An important novelty in the present case (when compared
to 1F = 1 calculations) is the necessity of introducing evanescent operators that vanish in
4 dimensions by the Fierz identities. The Fierz identities cannot be analytically continued
to D dimensions. Therefore, they have to be treated in dimensional regularization in the
same manner as the identity
= g + g g + i 5 ,
(2.2)
i.e., appropriate evanescent operators have to be introduced.

and QSLL
As an example, consider the operators QSLL
1
2 . When these operators are
inserted into one- and two-loop diagrams, the operators

= s PL d s PL d ,
(2.3)
Q SLL
1

SLL
(2.4)
Q2 = s PL d s PL d
and QSLL
are generated. In 4 dimensions these operators can be expressed through QSLL
1
2
by using the Fierz identities
400
1
1
(PL )ij (PL )kl = (PL )il (PL )kj ( PL )il ( PL )kj ,
2
8
1
( PL )ij ( PL )kl = 6(PL )il (PL )kj ( PL )il ( PL )kj ,

2
which give
(2.5)
1
1
= QSLL
+ QSLL
,
(2.6)
Q SLL
1
1
D=4 2
8 2
1
= 6 QSLL
+ QSLL
.
(2.7)
Q SLL
2
1
D=4
2 2
These relations can be used in the calculation of one-loop ADM. In the case of two-loop
calculations, in the NDRMS scheme, where Dirac algebra has to be performed in D 6= 4
dimensions, these relations have to be generalized to
1
1
= QSLL
+ QSLL
+ E1SLL ,
(2.8)
Q SLL
1
2 1
8 2
1
= 6 QSLL
+ QSLL
+ E2SLL .
(2.9)
Q SLL
2
1
2 2
Here, E1SLL and E2SLL are the evanescent operators that vanish in 4 dimensions by Fierz
identities. They are simply defined by (2.8) and (2.9) and are given in Appendix A.
The effective Lagrangian can be written separately for each sector. It takes the form
Leff =
2

2 2 X

G2F MW
V
Zq
Ci () Qi + (counterterms)i ,
V
t
d
t
s
2
4
(2.10)
where Zq is the quark wave-function renormalization constant.

The coefficients Ci () satisfy the Renormalization Group Equation (RGE)
d E
E
C() = ()T C()
d
(2.11)
governed by the ADM () that has the following perturbative expansion:

() =

s () (0) s2 () (1)
+
+ O s3 .
2
4
(4)
(2.12)
The ADM in the MS or MS scheme is found from one- and two-loop counterterms in
the effective theory, according to the following relations (equivalent to Eqs. (4.26)(4.37)
of Ref. [5]):
(0) = 2a 11,
(1)
= 4a
12
(2.13)
2b c.
(2.14)
The matrices a 11 , a 12 and b in the above equations parametrize the MS-scheme

counterterms in Eq. (2.10) (for D = 4 2)

X
s X 11
s2 X 1 22 1 12
aik Qk +
bik Ek +
a + aik Qk
(counterterms)i =
4
(4)2
2 ik

k

+ (two-loop evanescent counterterms) + O s3 .
(2.15)
401
Fig. 1. One-loop diagrams.
The matrix c is recovered from one-loop matrix elements of the evanescent operators.
Let us denote by hEk i1loop the one-loop K 0 K 0 amplitude with an insertion of some
evanescent operator Ek . The pole part of such an amplitude is proportional to some linear
combination of tree-level matrix elements of evanescent operators. The remaining part in
the limit D 4 can be expressed by tree-level matrix elements of the physical operators
Qi . The finite coefficients of these matrix elements define the matrix c as follows:

X
X
1 X
dkj hEj itree +
ekj hFj itree
cki hQi itree + O(). (2.16)
hEk i1loop =

j
Here, Fj stand for such evanescent operators that are not necessary as counterterms for
the one-loop Green functions with insertions of the physical operators Qi . The matrices c
and a 12 depend on the structure of Fj , but (1) does not.
The matrices (0) = 2a 11 , b and c in each sector are found from the one-loop d s s d
diagrams presented in Fig. 1 with insertions of the physical operators Qi , as well as the
evanescent operators Ek . We calculate only the annihilation-type diagrams, i.e., we
drop all the diagrams where fermion lines connect the incoming and outgoing particles.
Dropping such diagrams consistently at the tree level, at one loop and (later) at two loops
does not alter the final results for the renormalization constants.
All the one- and two-loop diagrams considered in the present article are calculated using
two different methods. In both of them, a covariant gauge-fixing term
Lgf =
1 a a
G G
2
(2.17)
is used, and the physical masses are set to zero. In the first method, the external quarks
are assumed to have momentum p. In the second method, the external momenta are set
to zero, but a common mass parameter is introduced in all the propagator denominators
as IR regulator [12]. The two methods give the same results for the MS renormalization
constants. The ADMs calculated from these renormalization constants with the help of
Eqs. (2.13) and (2.14) are independent of the gauge-fixing parameter .
We begin with presenting the ADM in the SLL sector, because in this very sector our
results are going to differ (at two loops) from those of Ref. [4]. The matrices (0)SLL and
b SLL are found to be the following:
!
1
1
6N + 6 + N6
2 N
(0)SLL
=
,
(2.18)
24 48
2N + 6 N2
N
402
b SLL =
1
2
1
8 8 2N
1
2
(2.19)
where N stands for the number of colours.

In order to find the matrix a 12 , we need to calculate two-loop diagrams obtained from
the ones in Fig. 1 by including one-loop corrections on the gluon lines or adding another
gluon that couples to the open quark lines. Of course, one-loop diagrams with counterterm
insertions need to be included, too. All the two-loop diagrams and the corresponding colour
factors are the same as in Fig. 2 and Table 2 of Ref. [5]. However, in the present article, we
also consider additional Dirac structures (1.3) in the four-quark vertices.
Inserting the calculated matrix a 12 into Eq. (2.14), we obtain the two-loop ADM. Its
entries are found to be the following:
136 12
107
2
10
203 2 107
10
N +
N+
f,
+ Nf f
2
6
3
3
N
3
3
3N
2N
31
9
4
1
1
1
(1)SLL
+ 2 f+
f,
= N
12
36
9
N
N
18
9N
704 208 320 136
176
364
(1)SLL
N
2 +
f+
f,
=
21
3
3
N
N
3
3N
21
2
343 2
188 44
26
(1)SLL
N + 21N
+
+
f,
(2.20)
22
=
Nf 6f +
18
9
N
9
9N
2N 2
where f stands for the number of active flavours. The above equation is one of the main
results of the present paper.
Proceeding analogously in the VLL sector, we reproduce the well-known results for the
[13]:
one- and two-loop anomalous dimensions of the operator QVLL
1
(1)SLL
11
6
,
N
22 39
57
2
19
2
+
f.
+ f
(1)VLL = N
6
3
N
3N
2N 2 3
(0)VLL = 6
(2.21)
The matrix b in the VLL sector reads

1
1
b VLL = 5 2N
2 .
(2.22)
Finally, our results for the LR sector read

!
6
12
N
(0)LR
=
,
0 6N + N6
(1)LR
137
15
22
6 + 2N 2 3N f
71
9
4 N + N 2f
b LR =
200
6
44
3 N N 3 f
479
15
10
2
203
6 N + 6 + 2N 2 + 3 Nf
1 1
0 5 2N
0 0
2
1 1
0 0
0 0 2N
2
(2.23)
!
22
3N f
, (2.24)

.
(2.25)
403
As mentioned in the introduction, all the comparisons with existing literature are relegated
to Section 7.
3. Currentcurrent contributions to the ADM of 1F = 1 operators
In the present section, we evaluate contributions from the currentcurrent diagrams to
the ADM of 1F = 1 operators. For this purpose, we choose the operators in such a manner
that all the four flavours they contain are different: s , d, u,
c. In such a case, the only
possible diagrams are the currentcurrent ones.
Twenty linearly independent operators can be built out of four different quark fields.
They can be split into 8 separate sectors, between which there is no mixing. The operators
belonging to the first four sectors (VLL, VLR, SLR and SLL) read

= s PL d u PL c = Q VL VL ,
QVLL
1

= s PL d u PL c = QVL VL ,
QVLL
2

= s PL d u PR c = Q VL VR ,
QVLR
1

= s PL d u PR c = QVL VR ,
QVLR
2

= s PL d u PR c = Q LR ,
QSLR
1

= s PL d u PR c = QLR ,
QSLR
2

= s PL d u PL c = Q LL ,
QSLL
1

= s PL d u PL c = QLL ,
QSLL
2

= s PL d u PL c = Q TL TL ,
QSLL
3

= s PL d u PL c = QTL TL ,
(3.1)
QSLL
4
where on the r.h.s. we have shown the notation of Ref. [4].
The operators belonging to the four remaining sectors (VRR, VRL, SRL and SRR) are
obtained from the above by interchanging PL and PR . Obviously, it is sufficient to calculate the ADMs only for the VLL, VLR, SLR and SLL sectors. The mirror operators in
the VRR, VRL, SRL and SRR sectors will have exactly the same properties under QCD
renormalization.
The evanescent operators for the VLL, VLR, SLR and SLL sectors are listed in
Appendix B. Calculation of the renormalization constants and the ADMs proceeds along
the same lines as in the previous section. The relevant divergences in one- and two-loop
diagrams in the cases of VLL, VLR and SLR sectors are given in Refs. [5] and [6]. For
completeness we give in Appendix E the corresponding results for the SLL sector. These
have not been published so far in the NDRMS scheme.
Our final results for the 1F = 1 ADMs are as follows:
!
6
N6
(0)VLL
=
,
(3.2)
6 N6
404
(1)VLL
(0)VLR
22
3
19
6 N
6N +
6
N
479
15
6 + 2N 2
3
100
3 N+ N
0 6N +
6
N
6
N
+
+
10
3 Nf
22
3 f
(3.3)
71
2 N
22
3N f
137
6
15
2N 2
18
N
+ 4f
!
, (3.5)
22
3N f
(3.6)
100
3 N+
2
203
6 N +
48
N
N
2
6
N
479
6
1
N
3
N
15
2N 2
N1
N2 4N
2N
+
+
22
3 f
10
3 Nf
22
3N f
, (3.7)
1
2
107
N 2 148
10
+
f,
2Nf
2
2
3
2N
3N
64 16
178
(1)SLL
N+
+ f,
=
12
3
N
3
4
f
107 2 71
1
(1)SLL
N
,
=
Nf +
13
36
18 N 2 18
9N
8
f
109
(1)SLL
N+ ,
=
14
36
N
18
104
(1)SLL
,
= 26N +
21
N
107
10
203 2 28
10
(1)SLL
N +
f,
=
+ Nf
22
6
3
3
3N
2N 2
2
1
89
(1)SLL
= N + f,
23
18
N
9
4
53
1
(1)SLL
f,
= 2 +
24
18 N
9N
176
676 2 1880 320 88
(1)SLL
N
2 Nf +
f,
=
31
3
3
N
3
3N
448 88
820
(1)SLL
N+
f,
=
32
3
N
3
21
2
257 2 116
22
(1)SLL
N
+
f,
=
+ Nf +
33
18
9
9
9N
2N 2
(1)SLL
=
11
(3.4)
0
6N +
(0)SLL = 48
N + 24N
24
48
39
2
N + 3f
57
2
3N
f
2N 2
137
15
22
6 + 2N 2 3N f
71
18
2 N N + 4f
(1)SLR =
22
3
6
N
19
6 N+
6
N
2
203
6 N +
(1)VLR =
(0)SLR
57
2
3N
f
2N 2
39
2
+ N + 3f
2
N
(3.8)
405
50
8
N f,
3
3
416 176
488
(1)SLL
N+
f,
=
41
3
N
3
776 320 176
(1)SLL
2 +
f,
=
42
3
N
3N
40 8
22
(1)SLL
+ f,
= N
43
3
N
3
21
2
343
28
26
(1)SLL
N2 +
+
f.
(3.9)
=
Nf +
44
2
18
9
2N
9
9N
Equation (3.9) is one of the main results of this work.
The careful reader has already noticed that the following equalities hold up to O(s2 ):
(1)SLL
34
VLL
VLL
= 22
,
11
VLL
VLL
12
= 21
,
VLR
SLR
11
= 22
,
VLR
SLR
= 11
,
22
VLR
SLR
12
= 21
,
VLR
SLR
21
= 12
.
At one loop, these equalities are a consequence of the Fierz identities

( PL )ij PL kl = ( PL )il PL kj ,

( PL )ij PR kl = 2(PR )il (PL )kj ,
(3.10)
(3.11)
(3.12)
as well as the flavour- and chirality-blind character of QCD interactions. Since the Fierz
identities are satisfied in four spacetime dimensions only, the relations (3.10) could be
potentially broken at two loops in the NDRMS scheme. Surprisingly, they are not. 1
On the contrary, analogous relations are broken at two loops in the SLL sector. Because
of the Fierz relations (2.5), the one-loop matrix (0)SLL must satisfy the following identity
(cf. Eqs. (9) and (10) of Ref. [4]):
b
b (0)SLLF
(0)SLL = F
with
0 12 0
1
8
1 0 1 0
8
b=
F
.
2
0 6 0 12
6
1
2
(3.13)
(3.14)
No similar relation holds for (1)SLL in the NDRMS scheme. As it has already been said,
this is not surprising, because the Fierz identities are not true in D 6= 4 dimensions.
It is unclear to us whether the symmetries (3.10) for the VLL, VLR and SLR sectors are
preserved at two loops in the NDRMS scheme only by coincidence, or if there is some
reason beyond this. As we shall see in Section 6, this question is related to the properties
of one-loop matrix elements of certain evanescent operators.
1 In Section 4, where the penguin diagrams are considered, no invariance under Fierz rearrangement is observed
at two loops for the operators with VLL Dirac structure. A detailed discussion of this fact can be found in Ref. [6].
406
4. Penguin contributions to the ADM of 1F = 1 operators

In the present section, we shall describe additional contributions to the ADM of 1F = 1
operators that are due to penguin diagrams. Such contributions may arise only when the
operators contain one quarkantiquark pair of the same flavour.
For definiteness, let us consider 1S = 1 operators. In the SM analysis of Ref. [6], 10
such operators were considered 2

Q1 = s PL u u PL d ,
Q2 = s PL u u PL d ,
X

X

q PL q ,
Q4 = s PL d
q PL q ,
Q3 = s PL d
q
Q5 = s PL d
X
q PR q

Q6 = s PL d
X

q PR q ,
X

3
eq q PR q ,
Q7 = s PL d
2
q
X

3
eq q PR q ,
Q8 = s PL d
2
q
Q9 =
Q10 =
X

3
eq q PL q ,
s PL d
2
q
X

3
eq q PL q .
s PL d
2
q
(4.1)
Their one- and two-loop ADMs, including currentcurrent and penguin diagrams, can be
found in Appendices A and B of Ref. [6]. They were also obtained in Ref. [7]. The same
results hold for the mirror copies of the SM operators, i.e., for the operators obtained from
the ones in Eq. (4.1) by PL PR interchange.
Beyond SM, new linearly independent operators appear. Their Dirac structures are as
in Eq. (3.1). Our aim is to find a minimal set of linearly independent new operators. In
the process of identifying these operators, we shall use four-dimensional Dirac algebra,
including the Fierz relations (2.5), (3.11) and (3.12). It turns out that only 3 additional
operators (and their mirror copies) undergo mixing via penguin diagrams into other fourquark operators in Eq. (4.1). These are

Q11 = s PL d d PL d + s PL s ,

Q12 = s PL d d PR d + s PR s ,

(4.2)
Q13 = s PL d d PR d + s PR s .
The remaining elements of the operator basis can be chosen in such a manner that massless
penguin diagrams with their insertions vanish. The first three of the remaining operators
have the structure of Q11 , . . . , Q13 , but with a relative minus sign between the two terms.
The next two have the structure of Q5 and Q6 , but the sum over flavour-conserving currents
2 Our operators here differ from the ones in Ref. [6] by a global normalization factor of 4. Of course, it does not
affect their ADM. The factor of 4 can be absorbed into the global normalization factor of the effective Lagrangian,
as the first ratio on the r.h.s. of Eq. (2.10). In this case, the Wilson coefficients of our operators are exactly the
same as those in Ref. [6].
407
is replaced by a difference between the analogous u-quark and c-quark currents. Their
mirror copies have to be included, as well. Further operators have the SLL and SRR Dirac
structures as in Eq. (1.3), or they have the form

(4.3)
s PL,R d q PR,L q ,

(4.4)
s PL,R d q PR,L q ,
where q has flavour different from s and d. It is straightforward to convince oneself that
we have not missed any linearly independent 1S = 1 operator in the above considerations.
Massless penguin diagrams with insertions of the operators (1.3), (4.3) and (4.4) vanish,
because
Tr(Sodd PL,R ) = 0 and PL,R Sodd PL,R = 0,
(4.5)
where Sodd is a product of an odd number of Dirac -matrices. For dimensional reasons,
only massless penguin diagrams can cause mixing into other four-quark operators. This
means that all the 1S = 1 operators, except for Q1 , . . . , Q13 and their mirror copies, mix
only due to currentcurrent diagrams, i.e., their ADMs are identical to the ones we have
already calculated in Sections 2 and 3.
At the two-loop level, a complication arises because generally the Fierz relations
could be broken in D 6= 4 dimensions. Consequently, our use of these relations in the
identification of linearly independent operators could be put in question. However, as we
have already discussed in Section 2 and will elaborate in Section 5, this complication can be
avoided by introducing appropriate evanescent operators that vanish in four dimensions by
Fierz identities. This allows us to restrict the basis of new physical operators (undergoing
penguin mixing) to the one in Eq. (4.2), even at the two-loop level.
The introduction of evanescent operators that vanish in four dimensions by Fierz
identities turns out to have no effect on the two-loop ADM in the case of the operators with
VLR and SLR structures, because the Fierz identity (3.12) remains valid at two loops in
the NDRMS scheme, even if the penguin insertions are considered [6]. On the other hand,
as pointed out in Ref. [6], the Fierz identity (3.11) is broken at two loops in the NDRMS
scheme through penguin diagrams, although it remains valid for currentcurrent diagrams.
As a result, the mixing of the operator

(4.6)
Q011 = s PL d d PL d + s PL s
with the operators in Eq. (4.1), through penguin diagrams, differs from the one of Q11 at the
two-loop level. This can be easily verified by using the results of Ref. [6]. As Q011 = Q11 in
D = 4 dimensions due to the Fierz identity (3.11), Q011 was not included in the basis (4.2).
By working with Q11 and the evanescent operator Q011 Q11 , the explicit appearance of
Q011 can be avoided at any number of loops, so that the basis (4.2) remains unchanged.
The above discussion implies that the only additional ADMs we need to find in the
present section are:
The 33 matrix cc describing the mixing of Q11 , . . . , Q13 among themselves.
408
The 34 matrix p describing the mixing of Q11 , . . . , Q13 into Q3 , . . . , Q6 via

penguin diagrams. (Only Q3 , . . . , Q6 are generated by massless QCD penguin
diagrams with four-quark operator insertions.)
The matrix cc is given by currentcurrent diagrams only. It takes the form
!
VLL
1F
0
=2
(4.7)
cc =
VLR
0
1F
=1
VLL and VLR taken from Eqs. (2.21), (3.2) and (3.3).
with 1F
=2
1F =1
(0)
s (1)
p + that originates from penguin diagrams can be
The matrix p = p + 4
easily extracted from Sections 3.2 and 5.3 of Ref. [6]. We find
T

(4.8)
p(0) = 43 , 43 , 0 N1 , 1, N1 , 1 ,
16
172
356
40
112
6N 64
27 3N + 27N 2
27 27N 2 6N + 3N
2
460
500
352
32
22
27 N 3 27N
27 N + 27N
3
.
p(1)T =
(4.9)
244
20
188
140
148
4
6N + 3N
6N 27 + 3N 27N 2
27 + 27N 2
172
27 N
2
3
260
27N
220
27 N
508
27N
22
3
The above discussion changes very little in the case of 1F = 1 operators, in which F
is the up-type flavour. Similarly to the 1S = 1 case, all the contributions from penguin
diagrams can be easily extracted from Ref. [6].
5. Transformation of the Wilson coefficients to the RI scheme

The ADMs calculated in the present work are given in the NDRMS scheme that
is most convenient for perturbative calculations. However, after the Wilson coefficients
are evolved with the help of RGE (2.11) down to a low energy scale, it might be
necessary to transform them to another scheme that is more appropriate for nonperturbative calculations of hadronic matrix elements [9]. One such scheme is the socalled Regularization-Independent (RI) scheme (originally called the MOM scheme) used
in Ref. [4]. Below, we shall give relations between the NDRMS-renormalized and RIrenormalized Wilson coefficients of all the operators considered in Sections 2 and 3.
For completeness, we begin with the definition of the RI scheme. For the massless quark
propagator, the renormalization condition can be written as

i
1
S(p)
= 1,
(5.1)
R
4
p
p 2 =2
where is the subtraction scale. A simple one-loop calculation is necessary to verify that
the renormalized inverse propagator in the RI scheme reads

p2
s
1
1
CF
ln 2
p 1
(5.2)
+ O s2 ,
S(p)R = i/
4
2
409
where CF = (N 2 1)/(2N) and is the gauge-fixing parameter (cf. Eq. (2.17)).

In dimensional regularization, the corresponding quark wave-function renormalization
constant reads

1
s
1
CF
+ ln(4) +
,
(5.3)
ZqRI = 1
4

2
provided the subtraction scale is identified with the standard MS renormalization scale.
Conditions similar to Eq. (5.1) are imposed on renormalized matrix elements of the
operators (2.1) and (3.1) among four external quarks with the same momentum p. The
quarks are assumed to be massless here. For the 1F = 2 operators, such matrix elements
have the following form
VLL

VLL
2
VLL 2
+ B11
(p ) p p s PL d s PL d tree ,
Q1 R = AVLL
11 (p ) Q1
tree
QLR
1
QLR
2
LR

2
LR 2
= ALR
s PR d tree
11 (p ) Q1 tree + B11 (p ) p p s PL d
LR

2
LR 2
+ ALR
s PR d tree ,
12 (p ) Q2 tree + B12 (p ) p p s PL d
LR

2
LR 2
= ALR
s PR d tree
21 (p ) Q1 tree + B21 (p ) p p s PL d
LR

2
LR 2
+ ALR
s PR d tree ,
22 (p ) Q2 tree + B22 (p ) p p s PL d
SLL
SLL
SLL
2
2
+ ASLL
,
Q1 R = ASLL
11 (p ) Q1
12 (p ) Q2
tree
tree
SLL
SLL
SLL
2
2
+ ASLL
.
Q2 R = ASLL
21 (p ) Q1
22 (p ) Q2
tree
tree
(5.4)
The formfactors Bij (p2 ) originate from UV-finite parts of Feynman diagrams and are
scheme-independent. Note that in all the matrix elements multiplied by Bij (p2 ), only
colour-singlet quark currents occur. Colour-octet currents are removed from these terms
with the help of the following Fierz identities (which are independent from the ones in
Eqs. (2.5), (3.11) and (3.12)):

1
p2 PL il PL kj ,
(5.5)
2

1
1
(5.6)
/ PR kl = p p PR il PL kj + p2 PR il PL kj ,
p
/ PL ij p
2
2

1
/ PR il p
(5.7)
/ PL kj p2 PR il PL kj .
p p PL ij PR kl = 2 p
2
No Bij formfactors occur in the SLL sector thanks to the four-dimensional identity
p
/ PL
ij
p
/ PL
kl
il
p
/ PL
kj

1
= p2 PL ij PL kl .
4
The RI renormalization condition reads
p p PL
= p
/ PL
ij
PL
kl
Aij (2 ) 2 Bij (2 ) = ij ,
with
(5.8)
(5.9)
410
1
4
VLL
LR , and B LR ,
for B11
, B11
21
(5.10)
0 otherwise.
The renormalization condition (5.9) can be equivalently written as
(p2 = 2 ) = ij ,
Aeffective
ij
(5.11)
obtained from Eqs. (5.4) by making the following ad hoc replacements

with Aeffective
ij
1
pPL,R ) p2 ( PL ) ( PL,R ),
(/
pPL ) (/
4
p p ( PL,R ) ( PR,L ) 0.
(5.12)
In the case of 1F = 1 operators, the general structure of one-loop matrix elements is

similar to that in Eq. (5.4), but the number of formfactors is larger, because operators with
colour-octet currents are now linearly independent. The matrix elements can be written as

(p2 ) Qj tree + Ni ,
(5.13)
Qi R = Aeffective
ij
where Ni vanish under the replacements (5.12). The RI renormalization condition then has
the same form as in Eq. (5.11).
In each of the sectors, the RI-renormalized Wilson coefficients can be obtained from the
NDRMS-renormalized ones with the help of the following relation

s () T
1rMSRI () CE MS () + O s2 ,
(5.14)
CE RI () = 1
4
where
1rMSRI ()

ij

4 RI 2
2
Aij (p ) AMS
ij (p ) .
s ()
(5.15)
The above relations can be easily derived from the fact that the renormalized matrix
element of the whole effective Hamiltonian is scheme-independent, i.e.,
RI
MS
E
E
p2 ) = CE MS () Q(,
p2 ) .
CE RI () Q(,
(5.16)
Again, the RI subtraction scale and the standard MS renormalization scale have been tacitly
identified. The external states must be the same in Eq. (5.16). Consequently, the RI-scheme
MS (p2 ) that enters
renormalization constant (5.3) must be used for external quark lines in Aij
into Eq. (5.15).
The dependence on p2 and the explicit dependence on cancels out in 1r MSRI (5.15).
However, one should not forget that this matrix depends on the gauge-fixing parameter
that is, in turn, -dependent.
Once the RI renormalization conditions have been specified, finding the explicit form of
1r MSRI is only a matter of a straightforward one-loop computation. Our results for the
1F = 2 operators are as follows:

12 ln 2
3
3
4 ln 2
7
VLL
+
4 ln 2 +
,
1rMSRI = 7 12 ln 2 +
N
N
2 2N
N
1r LR
=
MSRI
2
2 ln 2
1
2 ln 2
N + N + 2N + N

1 + ln 2 12 ln 2
411
4 + 4 ln 2 + (1 + 4 ln 2)
,

ln 2 3N 1 2 ln 2
4N + N2 + 2 N
N
2
2N

2 ln 2
1
1
3N
2 ln 2
5
,
+
+
+
= 4N + 7 + 4 ln 2 +
N
N
2 2N
2
N

i
h
13
2 ln 2 5 ln 2
5
1
ln 2 ln 2
5
SLL
+
+
+
=
1rMSRI
,
12
12 12N
3
6N
24 6N
3
6N

i
h
40 ln 2
8
8 ln 2
12
SLL
32 ln 2 +
2 + + 16 ln 2
,
=4
1rMSRI
21
N
N
N
N

i
h
5
28 ln 2 26 ln 2
N
7
5
8 ln 2 10 ln 2
7
SLL
+
+
+
+
.
=
1rMSRI
22
3 3N
3
3N
2
6 6N
3
3N
SLL
1rMSRI
11
In the 1F = 1 case, we find

3 4 ln 2
3 4 ln 2
7
12
ln
2
+
N7 + 12Nln 2 2N
N
,

2
=
1r VLL
MSRI
3 4 ln 2
7 12 ln 2 + 32 4 ln 2
N7 + 12Nln 2 2N
N

ln 2 3N 1 2 ln 2
2 2 ln 2 + 1 2 ln 2
4N + N2 + 2 N
N
2
2N

,

=
1r VLR
MSRI
2 + 2 ln 2 + 1 + 2 ln 2
2 2 ln 2 12 + 2 ln 2
N
N
2N
N
=
1r SLR
MSRI
2
2 ln 2
1
2 ln 2
N + N + 2N + N

2 2 ln 2 + 1 2 ln 2

2 2 ln 2 12 + 2 ln 2
,

ln 2 3N 1 2 ln 2
4N + N2 + 2 N
2
2N
N

5
2 ln 2
1
2 ln 2
3N
+ +
+
+
=
,
2
N
N
2N
N

i
h
1
7
SLL
+ 2 ln 2 ,
= 2 ln 2
1rMSRI
12
2
2

i
h
13
5 ln 2
N
1
ln 2
N
SLL
+
+
+
,
=
1rMSRI
13
2
12N
6N
8
6N
6N

i
h
5 ln 2
1
ln 2
7
SLL
=
1rMSRI
,
14
12
6
24
6
i
h
SLL
= 1 2 ln 2 + (1 2 ln 2),
1rMSRI
SLL
1rMSRI
11
21

2 ln 2
3N
1
2 ln 2
5
+
,
22
N
N
2
2N
N

i
h
1 ln 2
13 5 ln 2
SLL
=
,
1rMSRI
23
12
6
6
6
h
SLL
1rMSRI
= 4N +
412

1
5 ln 2
ln 2
13
+
,
24
12N
6N
6N
6N

i
h
8
8 ln 2
12 40 ln 2
SLL
+
+ 6N +
= 4N
,
1rMSRI
31
N
N
N
N
i
h
SLL
= 8 40 ln 2 + (2 8 ln 2),
1rMSRI
h
SLL
1rMSRI
32

5
26 ln 2
5
10 ln 2
5N
+
N +
,
33
2
3N
3N
6N
3N

i
h
11 10 ln 2
25 26 ln 2
SLL
=
,
1rMSRI
34
6
3
6
3
i
h
SLL
= 12 40 ln 2 + (8 8 ln 2),
1rMSRI
h
SLL
1rMSRI
41

8
8 ln 2
12 40 ln 2
+
,
42
N
N
N
N

i
h
1 10 ln 2
5 26 ln 2
SLL
+
,
=
1rMSRI
43
3
3
3
3

i
h
26 ln 2
N
5
10 ln 2
5
SLL
+
+
+
.
=
1rMSRI
44
3N
3N
2
6N
3N
h
SLL
1rMSRI
In Section 7, the above results will be used in performing the comparison with Ref. [4].
6. Recovering the ADM of 1F = 2 operators from 1F = 1 results
Let us now use our 1F = 1 anomalous dimensions from Section 3 to find again the
ADM of 1F = 2 operators. This will serve as a cross-check of our findings and as a
preparation for the comparison with Ref. [4] in Section 7.
Starting from Eq. (3.1), we shall pass to another operator basis where the operators are
either symmetric or antisymmetric under d c interchange. Next, the flavours of both
quarks and both antiquarks will be set equal. For definiteness, we shall do it first in the
SLL sector. The superscript SLL will be understood for all the relevant quantities below,
and we shall not write it explicitly.
In four spacetime dimensions, passing to the new operator basis would be equivalent to
performing a simple linear transformation of the operators. In the framework of dimensional regularization, introducing additional evanescent operators becomes necessary. In
the SLL sector, only two evanescent operators were needed in the 1F = 1 calculation (see
Appendix B). Now, we need to introduce six additional evanescent operators in this sector.
They are defined in Appendix C.
We begin with a redefinition of the physical operators Qi (i = 1, . . . , 4) that amounts to
adding to them appropriate linear combinations of the evanescent operators Ei :
Qi Qi +
8
X

Wik Ek Qi new ,
413
(6.1)
k=1
where
0 0 12 0
1
8
00 0
0 0 0 0 0 0 0 0
W =
.
0 0 6 0 12 0 0 0
(6.2)
0 0 0 0 0 00 0
The new operators read

1
1
Q1 F + Q3 F ,
Q2 new = Q2 ,
2
8

1
Q4 new = Q4 ,
Q3 new = 6 Q1 F + Q3 F ,
2
where

Q3 F = s PL c u PL d .
Q1 F = s PL c u PL d ,
Q1
new
(6.3)
(6.4)
(6.5)
In 4 spacetime dimensions, the transformation (6.1) would be equivalent to performing

the Fierz rearrangement of Q1 and Q3 , as Ek would not contribute. Since the Fierz
identities cannot be analytically continued to D dimensions, the Fierz rearrangement
must be understood in terms of the transformation (6.1), so long as the MS scheme is
used. The MS-renormalized one-loop matrix elements of Q1 and Q3 are affected by this
transformation. This means that the renormalization scheme is changed. We pass from one
version of the NDRMS scheme to another, even though the evanescent operators remain
unchanged.
After the redefinition (6.1), we perform a simple linear transformation of the operators
with
Qi

new
14
R = 1
4
4
X
j =1
1
2

Rij Qj new
1
16
0 14
1
1
2 16
0 14
(6.6)
.
0
1
2
(6.7)
1
2
As one can easily check, our final operator basis is {Q+

1 , Q2 , Q1 , Q2 }, where
Q
1 =

1
s PL d u PL c s PL c u PL d ,
2

1
s PL d u PL c s PL c u PL d .
2
The ADM transforms as follows:
Q
2 =
(6.8)
414
(0) R (0) R 1 ,

(1) R (1) + 1r , (0) + 20 1r R 1 ,
(6.9)
(6.10)
2
r reflects in the usual manner [14] change of the
where 0 = 11
3 N 3 f . The matrix 1
renormalization scheme that follows from Eq. (6.1). The explicit form of 1r is [8]
1r = W c,
(6.11)
provided W e = 0. The matrices c and e are found from one-loop matrix elements of
evanescent operators, as in Eq. (2.16). The product W e indeed vanishes in our case, and
?
?
?
?
?
?
?
?
3
17
1
3
5
1
4N N
N
4
16
4N
16
?
?
?
?
.
(6.12)
c =
7
5
13
7N 28
21
N
+
N
4
N
4
?
?
?
?
?
?
?
?
?
Here, stars denote non-vanishing elements of c that are irrelevant for us, since they do not
affect the matrix
1
1
3
12 N + N1
12
4 N 4N
2
0
0
0
0
1r = W c =
.
(6.13)
1
1
1
8N + 44
36
N
N
2
N
2
0
After the transformation (6.9), (6.10), the ADM in the basis {Q+
1 , Q2 , Q1 , Q2 } is found
to have the form
!
+
22
022
,
(6.14)
44 =
022 22
where = (0) +
(0) =
s (1)
+ ,
4
6N 6 +
24
48
N
6
N
12
1
N
2N 6
2
N
and
136 12
107
2
10
203 2 107
10
N
N+
f,
+ Nf f
2
6
3
3
N
2N
3
3
3N
31
9
4
1
1
1
(1)
2 f+
f,
12 = N
36
9
N
N
18
9N
704 208 320 136
176
364
(1)
N
2
f+
f,
21
=
3
3
N
3
3N
N
(1)
=
11
(6.15)

(1)
22
343 2
188 44
26
21
2
Nf 6f +
N 21N
+
f.
2
18
9
N
2N
9
9N
415
(6.16)
One can easily verify that the matrix + is equal to the one we have already found in
Eqs. (2.18) and (2.20). It must be so, because the operators Q+
i from Eq. (6.8) reduce to
from
Eq.
(2.1)
when
the
flavour
replacements
c
d
and
u
s are made. Moreover,
QSLL
i
the evanescent operators listed in Appendices B and C can be linearly combined to the ones
that are either symmetric or antisymmetric under d c interchange. When the flavour
replacements c d and u s are made, the antisymmetric operators vanish, while the
symmetric ones become equal to those in Appendix A. Thus, we have shown how to extract
the 1F = 2 results from the 1F = 1 ones.
Let us now briefly describe the analogous transformations in the VLL and
LR VLR SLR sectors. All the necessary evanescent operators are given in Appendices B and C. The relevant matrices W and R are the following:
!
!
1 1
0 01 0 0 0
1
VLL
VLL
,
=
,
R
=
W
2 1 1
0 00 0 0 0
LR
W 412
VLR 0
W 26
26
026 W SLR
!
=
26
0 1 2 0
026
026
1 VLL
W
!
,
1
1
0 0 1
R LR = 2
.
2 0 1 2 0
1
2
2W VLL
(6.17)
0 0 1
, QLR+
, QLR
,
Consequently, the final operator bases are {QV1 LL+ , QV1 LL } and {QLR+
1
2
1
},
where
QLR
2

1
s PL d u PL c s PL c u PL d ,
2

1
s PL d u PR c s PR c u PL d ,
=
QLR
1
2

1
s PL d u PR c s PR c u PL d .
=
(6.18)
QLR
2
2
An important simplification in the present case is that the one-loop matrix elements of
the evanescent operators E3VLL , E3VLR and E3SLR from Appendix C vanish in the limit
D 4, after subtraction of the MS-counterterms proportional to evanescent operators
only. This means that the third rows of cVLL , cVLR and cSLR vanish (cf. Eq. (2.16)). Consequently, 1r VLL = W VLL cVLL = 0 and 1r LR = W LR cLR = 0. This is why the two-loop
1F = 1 matrices of the VLL, VLR and SLR sectors exhibited Fierz symmetry in
Eq. (3.10). The transformations of the two-loop ADMs in the VLL and LR sectors thus
look as if we worked in 4 dimensions, i.e., they reduce to simple multiplications by the
corresponding R-matrices
and their inversions. The final results are
QV1 LL =
416
VLL
22
V LL+
V LL
!
,
LR
44
LR+
22
022
LR
022 22
!
,
(6.19)
where
(0)V LL = 6
(0)LR
(1)LR =
6
,
N
6
N
12
6N + N6
(1)V LL =
!
22 39
57
2
19
2
N
f,
f
2
6
3
N
2N
3
3N
137 + 15 22 f
6
3N
2N 2
71
9
4 N N 2f
6
44
200
3 NN 3 f
479
15
10
22
2
203
6 N + 6 + 2N 2 + 3 Nf 3N f
!
.
(6.20)
One can see that V LL+ and LR+ are identical to our 1F = 2 results in Eqs. (2.21),
(2.23) and (2.24).
7. Comparison with previous ADM calculations

In the present section, we compare our findings from Sections 2, 3 and 4 with the
previously published results for anomalous-dimension matrices.
7.1. One-loop results
As far as the one-loop QCD ADMs of four-quark operators are concerned, the historical
order of their evaluation was as follows:
Currentcurrent contributions to the one-loop ADM of 1F = 1 operators belonging
to the VLL and VLR sectors were originally calculated in Refs. [15,16]. These results
were also immediately applicable to the SLR sector, because the Fierz rearrangement
has a trivial effect at one loop. For the same reason, one-loop anomalous dimensions
of the 1F = 2 operators belonging to the VLL and LR sectors could have been
immediately read off from these articles. Thus, after 1974, the only unpublished oneloop currentcurrent anomalous dimensions were those of the SLL sector, both in the
1F = 1 and 1F = 2 cases.
One-loop penguin contributions to the ADM of the Standard Model operators were
originally evaluated in Refs. [1719]. As we have shown in Section 4, penguin
contributions to the ADM of other (beyond-SM) flavour-changing dimension-six
operators can be easily extracted from the SM calculations, both at one and at two
loops.
To our knowledge, the first published results for (0)SLL occur in Refs. [20] and [4],
for the 1F = 2 and 1F = 1 cases, respectively.
The one-loop ADMs given in the present article agree with all the papers quoted
above. However, in order to perform comparisons, one often needs to make simple
linear transformations, because different operator bases are used by different authors. For
417
SLL
instance, the results for (0)SLL in Ref. [20] are given in the basis {QSLL
1 , Q1 }. In order
to compare them with our Eq. (2.18), one should use the relation (2.6). Similarly, Eqs. (6.7)
and (6.9) need to be used for comparing our (0)SLL in Eq. (3.8) with the corresponding
results in Ref. [4].
7.2. Two-loop results
The history of previous two-loop computations is as follows:
The currentcurrent anomalous dimensions of the 1F = 1 operators belonging to
the VLL sector were originally calculated in Ref. [21] (in the DREDMS scheme),
and confirmed in Ref. [5] (where the NDRMS and HVMS results were also
given).
The remaining elements of the two-loop QCD ADM for 1F = 1 operators relevant in
the SM were calculated in Refs. [6,7]. New results in these papers were the current
current contributions in the VLR sector, as well as all the penguin contributions.
The SLR sector results in the 1F = 1 case, as well as the 1F = 2 results
for the VLL and LR sectors could be easily derived from them with the help
of Fierz identities, because the NDRMS-renormalized one-loop matrix elements
remain invariant under Fierz transformations, except for the currentcurrent ones
in the SLL sector, and the penguin ones in the VLL sector. Therefore, in the early
1990s, the only unknown two-loop anomalous dimensions were those of the SLL
sector.
The first calculation of the two-loop ADM in the SLL sector was performed by Ciuchini et al. [4], in both the 1F = 1 and 1F = 2 cases. The ADM was calculated
there in the so-called FRI renormalization scheme. The transformation rules were
given to the LRI scheme (Landau-gauge RI-scheme) and to the NDRMS scheme.
Currentcurrent anomalous dimensions for the remaining sectors were recalculated
as well.
Penguin contributions to the ADM of non-SM operators are considered for the first
time in the present article.
All the two-loop results presented here agree with the previous calculations mentioned
above, except for the NDRMS ones for the SLL sector found in Ref. [4]. Below, we
explain the reason for this disagreement.
7.3. Comparison with Ref. [4]
In Ref. [4], the two-loop ADM for 1F = 1 operators of the SLL sector was given in the
basis defined in Eq. (13) of that paper, which is equivalent to our Eq. (6.8). It was presented
in the so-called FRI renormalization scheme, and the transformation rules to the NDR
MS scheme were appended. Applying these transformation rules to their FRI-scheme
ADM, one obtains results that differ from our Eq. (6.16). In particular, a mixing between
+
Q
i and Qi occurs, which is absent in our result (6.16). We could obtain their result if
we ignored the transformation (6.1) and, consequently, used 1r = 0 in our Eq. (6.10).
However, the final results would then correspond to the basis
418

1

s PL d u PL c s PL d u PL c
2

1
s PL d u PL c ,
8

1
s PL d u PL c 6 s PL d u PL c
Q0 2 =
2

1
s PL d u PL c ,
2
Q0 1 =
1
2
(7.1)
rather than the one in Eq. (6.8). In 4 spacetime dimensions, the operators (6.8) and (7.1)
are identical, thanks to the Fierz identities (2.5). However, in D dimensions they are not.
Consequently, their NDRMS-renormalized matrix elements differ at one loop, and it is
not surprising that the two-loop ADM depends on which of the two bases is used.
We informed the authors of Ref. [4] about our findings prior to publication of the
present article. They responded that although their NDRMS results had been claimed to
correspond to the basis (6.8), the NDRMS renormalization conditions had been actually
imposed in the basis (7.1). However, they had forgotten to mention this in their article.
Unfortunately, such a mistake in the presentation has the same effect on the final result as
a mistake in the calculation that amounts to missing 1r 6= 0 in Eq. (6.10).
As far as the two-loop ADM for 1F = 2 operators of the SLL sector is concerned, the
situation is as follows. If we made the flavour replacements c d and u s in the basis
(7.1), but did not change anything in the ADM, we could interpret this ADM as the one for
1F = 2 operators, as the authors of Ref. [4] did. However, it would correspond to quite
non-standard conventions for the treatment of the evanescent operators obtained from Q0
1
and Q0
2 after the flavour replacements. One would need to assume that the finite one-loop
matrix elements of these evanescent operators are not renormalized away, contrary to the
usual procedure for any evanescent operator [5,8,10,11]. Such non-standard conventions
make the RGE evolution more complicated, because one has to deal with a 4 4 instead
of a 2 2 ADM in the NDRMS RGE for the SLL sector, in the 1F = 2 case. The
calculation of the one-loop matrix elements becomes more involved, as well.
In the 1F = 2 case, no calculation is necessary to convince oneself that the results of
Ref. [4] cannot correspond to the NDRMS renormalization conditions imposed in the
basis (6.8) (their Eq. (13)). Once the c d and u s replacements have been made, the
operators Q
i in Eq. (6.8) vanish identically in D dimensions. Therefore, they cannot mix
into the Q+
i operators, independently of what the treatment of evanescent operators is. On
+
the other hand, mixing of Q
i into Qi was claimed to be found in the NDRMS scheme
in Ref. [4]. Therefore, an inconsistency is clearly seen.
In the remainder of this section, we shall verify that our NDRMS results are compatible
with the LRI ones of Ref. [4]. By differentiating Eq. (5.14) with respect to , one obtains

0 s ()
0 s2 ()
T
RI
T
T
E
1r
1rMSRI () +
()
()
RI ()C () =
MSRI
8 2
8 2

s ()
T
T
1rMSRI
() MS
() CE MS () + O s3 ,
(7.2)
+ 1
4
419
where we have used the RGE (2.11),

0 s ()2
d
s () =
+ O s3 , and
d
2
0

s ()
d
() + O s2 .
() =
d
2
(7.3)
We have also used the fact that the dependence of 1r MSRI on originates solely from
its dependence on the gauge-fixing parameter ().
Next, we use Eq. (5.14) again to express CE MS () by CE RI () in Eq. (7.2). Then, the first
two terms of the perturbative expansion (2.12) of RI can be easily read off
(0)
(0)
,
MS
RI =
(1)
= (1) + 1r MSRI , (0)
RI
MS
MS

0
1r MSRI .
+ 2 0 +
(7.4)
(7.5)
Armed with our explicit expressions for 1r MSRI given in Section 5 and with the values
of
2
11
0 = N f
3
3

and

13
2
N + f,
2
6
3
(7.6)
we can easily calculate the RI-scheme ADM from our MS results, for arbitrary . Setting
then 0, we recover all the LRI-scheme anomalous dimensions given in Ref. [4].
As far as the FRI-scheme ADMs of Ref. [4] are concerned, we can confirm them as
well. However, it should be emphasized that the FRI scheme is not equivalent to the RI
scheme considered in Section 5 for any choice of . The FRI scheme cannot be defined
beyond perturbation theory, because different external momenta are chosen in different
diagrams when the renormalization conditions are specified. Therefore, in our opinion, the
main advantage of the RI scheme is lost.
8. Conclusions
In the present paper, we have calculated the two-loop QCD anomalous dimensions
matrix (ADM) ( (1) )NDR in the NDRMS scheme for all the four-fermion dimension-six
flavour-changing operators that are relevant to both the Standard Model and its extensions.
The 1F = 2 two-loop results can be found in Eqs. (2.20), (2.21) and (2.24). While the
matrices in Eqs. (2.21) and (2.24) could be extracted from the already published results,
the two-loop NDRMS ADM (2.20) for the SLL operators defined in Eq. (2.1) is correctly
calculated for the first time here.
The 1F = 1 two-loop results for operators containing four different quark flavours can
be found in Eqs. (3.3), (3.5), (3.7) and (3.9). While the matrices in Eqs. (3.3), (3.5) and
(3.7) could be extracted from the already published results, the two-loop NDRMS ADM
(3.9) for the SLL operators defined in Eq. (3.1) is correctly calculated for the first time
here.
420
Penguin contributions to the ADM of non-SM operators have been considered for
the first time here. These contributions can be easily extracted from the existing SM
calculations. We have identified the relevant non-SM operators in the 1S = 1 case, and
presented the corresponding ADM explicitly in Eqs. (4.8) and (4.9).
We have demonstrated that the main findings of our paper, given in Eqs. (2.20) and (3.9),
are compatible with each other, i.e., we have shown how to properly transform the ADMs
from the 1F = 1 to the 1F = 2 case. In this context, we have pointed out that in the
process of this transformation it is necessary to introduce additional evanescent operators
that vanish in four spacetime dimensions because of the Fierz identities.
We have also given the rules that allow transforming our NDRMS ADMs to the
corresponding results in the RI scheme, for arbitrary gauge-fixing parameter . They can
be found in the end of Section 5.
The 1F = 1 two-loop ADMs for all the operators defined in Eq. (3.1) were previously
presented in Ref. [4], in the Q
i basis. In the case of VLL, VLR and SLR operators, there is
full agreement between their and our results. The case of SLL operators is more subtle. We
can confirm their LRI-scheme results (RI scheme with = 0). However, their NDRMS
ADM is compatible with ours only after correcting their Eq. (13), i.e., after changing the
definitions of their SLL operators to the ones given in Eq. (7.1).
After such a correction in Eq. (13) of Ref. [4], also their 1F = 2 NDRMS results
are compatible with ours, provided they are understood in terms of quite non-standard
conventions for the treatment of evanescent operators. In their conventions, the two-loop
1F = 2 NDRMS ADM is a 44 rather than 22 matrix, which makes the RGE evolution
and calculating low-energy matrix elements unnecessarily complicated. Consequently, the
results presented here should be more useful for phenomenological applications.
Acknowledgements
We thank the authors of Ref. [4] for extensive discussions concerning their paper.
Furthermore, we would like to thank Christoph Bobeth and Gerhard Buchalla for
carefully reading the manuscript. A.B. and J.U. acknowledge support from the German
Bundesministerium fr Bildung und Forschung under the contract O5HT9WOA0. M.M.
has been supported in part by the Polish Committee for Scientific Research under grant
2 P03B 014 14, 1998-2000.
Appendix A
Here, we specify the evanescent operators that are necessary as counterterms for oneloop diagrams with insertions of the 1F = 2 operators (2.1).

,
E1VLL = s PL d s PL d QVLL
1

,
E2VLL = s PL d s PL d + (16 + 4)QVLL
1
421

E3VLL = s PL d s PL d + (16 + 4)QVLL
,
1

1
,
E1LR = s PL d s PR d + QLR
2 1

E2LR = s PL d s PR d + 2QLR
2 ,

E3LR = s PL d s PR d + (4 4)QLR
1 ,

E4LR = s PL d s PR d + (8 + 8)QLR
2 ,

E5LR = s PL d s PR d 6QLR
2 ,

E6LR = s PL d s PR d + 3QLR
1 ,

1
1
QSLL
,
E1SLL = s PL d s PL d + QSLL
1
2
8 2

1
QSLL
,
E2SLL = s PL d s PL d 6QSLL
1
2 2

E3SLL = s PL d s PL d + (64 + 96)QSLL
1
+ (16 + 8)QSLL
2 ,

E4SLL = s PL d s PL d 64QSLL
1
+ (16 + 16)QSLL
2 .
The evanescent operators for the VRR and SRR sectors, i.e., EkVRR and EkSRR are
obtained by replacing L by R in the definitions of EkVLL and EkSLL .
The operators E1VLL , E1LR , E2LR , E1SLL and E2SLL vanish in four spacetime dimensions
because of the Fierz identities (3.11), (3.12) and (2.5). The operators E2VLL , E3VLL , E3LR ,
E4LR , E3SLL and E4SLL vanish by the four-dimensional identity (2.2). Finally, E5LR and E6LR
vanish in four dimensions, because they become full contractions of self-dual and selfantidual antisymmetric tensors.
The evanescent operators listed here would look somewhat simpler if we removed from
them all the terms proportional to . It would be equivalent to changing one version of the
MS scheme to another. Keeping the terms proportional to in the above equations makes
our NDRMS scheme equivalent to the one where the so-called Greek projections are
used (see Appendix D).
Appendix B
Here, we specify the evanescent operators that are necessary as counterterms for oneloop diagrams with insertions of the 1F = 1 operators (3.1).

,
E1VLL = s PL d u PL c + (16 + 4)QVLL
1

,
E2VLL = s PL d u PL c + (16 + 4)QVLL
2
422

E1VLR = s PL d u PR c + (4 4)QVLR
,
1

,
E2VLR = s PL d u PR c + (4 4)QVLR
2

E1SLR = s PL d u PR c 6QSLR
1 ,

E2SLR = s PL d u PR c 6QSLR
2 ,

E1SLL = s PL d u PL c + (64 + 96)QSLL
1
+ (16 + 8)QSLL
3 ,

E2SLL = s PL d u PL c + (64 + 96)QSLL
2
+ (16 + 8)QSLL
4 .
The remaining evanescent operators (for the VRR, VRL, SRL and SRR sectors) are
obtained by interchanging L and R above.
Appendix C
This appendix contains definitions of the additional evanescent operators that are not
necessary as one-loop counterterms in the 1F = 1 effective Lagrangian in Section 3.
However, they have to be included before performing transformation to the plusminus
basis in Section 6.

,
E3VLL = s PL c u PL d QVLL
1

,
E4VLL = s PL c u PL d QVLL
2

,
E5VLL = s PL c u PL d + (16 + 4)QVLL
1

,
E6VLL = s PL c u PL d + (16 + 4)QVLL
2

1
,
E3VLR = s PR c u PL d + QVLR
2 1

1
,
E4VLR = s PR c u PL d + QVLR
2 2

,
E5VLR = s PR c u PL d + 3QVLR
1

,
E6VLR = s PR c u PL d + 3QVLR
2

E3SLR = s PR c u PL d + 2QSLR
1 ,

E4SLR = s PR c u PL d + 2QSLR
2 ,

E5SLR = s PR c u PL d + (8 + 8)QSLR
1 ,

E6SLR = s PR c u PL d + (8 + 8)QSLR
2 ,
423

1
1
E3SLL = s PL c u PL d + QSLL
QSLL
,
1
2
8 3

1
1
QSLL
,
E4SLL = s PL c u PL d + QSLL
2
2
8 4

1
QSLL
,
E5SLL = s PL c u PL d 6QSLL
1
2 3

1
QSLL
,
2
2 4

1
+ (16 + 16)QSLL
3 ,

2
+ (16 + 16)QSLL
4 .
The remaining evanescent operators (for the VRR, VRL, SRL and SRR sectors) are
obtained by interchanging L and R above.
Appendix D
In the present appendix, the notion of Greek projections [2,5,22] is recalled and
generalized to the case of SLL-sector operators. Let us denote the Dirac structure of the
operator in Eq. (1.1) by A B . The insertion of this operator in one- and two-loop
diagrams results in new Dirac structures like
n A n B ,
(D.1)
where n = 1 2 n . Several examples of such structures occur in Appendices AC.

It has been suggested in Ref. [22] to project them onto physical operators as follows. One
defines the projection G so that the following equality is satisfied

(D.2)
G n A n B = G A B .
In the case of A = B = PL , performing the projection G amounts to replacing by
on both sides of the above equation and contracting the indices using D-dimensional
Dirac algebra. In this manner, the coefficient is determined. One finds for instance:

+ O 2
(D.3)
G s PL d u PL c = (16 4) G QVLL
1
as defined in Eq. (3.1).
with QVLL
1
It has been pointed out in Ref. [5] that for a proper treatment of counterterms in twoloop calculations, one has to use Eq. (D.3) only as a prescription for defining an evanescent
operator. In the case at hand, this is the operator E1VLL of Appendix B. As discussed in
Ref. [2], in the case of VLR and SLR operators, the analogous projections are performed by
replacing by 1 and , respectively. Examples of the corresponding evanescent operators
can be found in Appendices AC.
424
The projections in the SLL sector are slightly more involved. In the case of the insertion
or QSLL
of QSLL
1
3 , the r.h.s. of Eq. (D.2) has to be generalized to a linear combination of
SLL
these two operators. The same applies to the pair (QSLL
2 , Q4 ). The projection G is now
performed by replacing by . After the projection, one finds linear combinations
of g and on both sides of the equation. This allows extracting the coefficients in
question. One finds, for instance,

G s PL d u PL c

+ (16 8) G QSLL
+ O 2 .
(D.4)
= (64 96) G QSLL
1
3
The corresponding evanescent operator is E1SLL in Appendix B. An alternative approach to
projections can be found in Ref. [11].
Appendix E
In this appendix, the 1/ and 1/ 2 poles in the one- and two-loop diagrams are given
for the 1F = 1 calculation in the SLL sector. Analogous results for the remaining sectors
can be found in Refs. [5] and [6]. The gauge-fixing parameter is set to unity here, i.e.,
the Feynmant Hooft gauge is used.
SLL
Each insertion results in a linear combination of QSLL
1 , . . . , Q4 , after subtracting the
evanescent counterterms (see Appendix B) or, alternatively, after performing the Greek
projections (see Appendix D). Table 1 gives the singularities (without colour factors) in
the coefficients of the resulting operators, for each diagram separately. The numbering of
the diagrams and values of the colour factors are exactly as in Figs. 1, 2 and Tables 1, 2 of
Ref. [5]. The multiplicity factors of the diagrams are included.
In the two-loop case, the singularities include one-loop diagrams with counterterm
insertions. The counterterms proportional to evanescent operators are multiplied by an
additional factor 1/2, and, at the same time, the term 2bc in Eq. (2.14) is ignored.
Correctness of such a trick has been justified in Refs. [5,10].
SLL
The singularities from Table 1 apply for the pair (QSLL
2 , Q4 ), too. After including
colour factors and summing the diagrams, the 1/ singularities build a 4 4 matrix in the
SLL
SLL
SLL
basis {QSLL
1 , Q2 , Q3 , Q4 }

s 2
s
B1 +
B2 + O s3 ,
B=
4
4
from which the anomalous-dimension matrix can be obtained by means of

(0)SLL
= 2 2 a1 ij + B 1 ij ,
ij

ij(1)SLL = 4 2 a2 ij + B 2 ij .
(E.5)
(E.6)
(E.7)
Here, a1 and a2 originate from 1/ singularities in the quark field renormalization
constants. They read

a1 = CF ,
a2 = CF
425

1
3
17
CF
N+ f .
4
4
2
(E.8)
Remembering the trick applied to evanescent operators here, it is easy to verify that
Eqs. (E.6) and (E.7) are equivalent to Eqs. (2.13) and (2.14) from Section 2.
Table 1
Pole parts of the one- and two-loop diagrams with insertions of QSLL
and QSLL
1
3 . The colour factors
are omitted, whereas the multiplicity (M) is taken into account. The numbering is according to Fig. 2
of Ref. [5]. While the singularities in front of the resulting QSLL
and QSLL
are the same in this table,
1
2
they become different after the inclusion of colour factors. The same comment applies to QSLL
and
3
SLL and QSLL insertions are equal to
QSLL
.
When
the
colour
factors
are
omitted,
the
results
for
Q
4
2
4
those for QSLL
and QSLL
insertions, respectively
1
3
QSLL
1
D M
QSLL
and QSLL
1
2
QSLL
3
QSLL
and QSLL
3
4
QSLL
and QSLL
1
2
QSLL
and QSLL
3
4
1/ 2
1/
1/ 2
1/
1/ 2
1/
1/ 2
1/
8
2
2
0
1/2
1/2
0
24
24
0
6
6
2
16
16
0
0
2
4
9
1
5/4
2
4
9
1
7/4
2
0
4
0
0
2
0
2
0
1/2
2
0
2
0
1/2
4
8
8
0
0
4
2
0
1/2
5/4
4
2
0
1/2
5/4
4
8
4
0
0
4
2
0
1/2
1/4
4
2
0
1/2
1/4
4
8
4
0
0
4
8
4
2
2
4
8
4
0
0
4
8
4
2
2
4
4
10
1
1
4
4
10
1
2
1
16
0
0
0
1
4
5
1
1/4
1
4
5
1
3/4
4
24
20
0
0
4
6
2
3/2
1/4
4
6
2
3/2
1/4
4
0
0
0
3
8f
0
0
2 5N + 2f 26N
3
3
5N f 17N +
30 2
0
0
12
6
72
5N f 17N +
31 2
0
0
12
6
72
0
48
48
0
0
0
0
24
24
0
24
24
96
0
96
0
48
48
0
48
48
0
72
72
0
0
0
76
52
0
24
24
0
20
20
0
28
28
64
0
64
0
64
112
0
28
4
0
108
108
144
0
0
12
12
0
0
0
0
6
6
0
6
6
0
0
0
0
12
12
0
12
12
0
18
18
0
5N 2f
3
3
0
7
7
4
2
2
4
8
4
0
0
4
0
0
0
0
2
22
0
5
5
0
6
18
0
4f
16N
9 + 9
1 2
2 2
3 2
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
f
44f
10N 4f 32N + 8f
134N
36 20N 8f 3 + 3
3
3
9
9
f
44f
4f 62N
20f
134N
10N
36 20N 8f 3 + 3 3 + 3
9 9
426
References
[1] G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125.
[2] A.J. Buras, hep-ph/9806471; in: R. Gupta, A. Morel, E. de Rafael, F. David (Eds.), Probing the
Standard Model of Particle Interactions, Elsevier Science, Amsterdam, 1999, p. 281.
[3] F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321.
[4] M. Ciuchini, E. Franco, V. Lubicz, G. Martinelli, I. Scimemi, L. Silvestrini, Nucl. Phys. B 523
(1998) 501.
[5] A.J. Buras, P.H. Weisz, Nucl. Phys. B 333 (1990) 66.
[6] A.J. Buras, M. Jamin, M.E. Lautenbacher, P.H. Weisz, Nucl. Phys. B 400 (1993) 37.
[7] M. Ciuchini, E. Franco, G. Martinelli, L. Reina, Nucl. Phys. B 415 (1994) 403.
[8] K. Chetyrkin, M. Misiak, M. Mnz, Nucl. Phys. B 520 (1998) 279.
[9] D. Becirevic et al., hep-lat/0002025, and references therein.
[10] M.J. Dugan, B. Grinstein, Phys. Lett. B 256 (1991) 239.
[11] S. Herrlich, U. Nierste, Nucl. Phys. B 455 (1995) 39.
[12] K. Chetyrkin, M. Misiak, M. Mnz, Nucl. Phys. B 518 (1998) 473.
[13] A.J. Buras, M. Jamin, P.H. Weisz, Nucl. Phys. B 347 (1990) 491.
[14] A.J. Buras, M. Jamin, M.E. Lautenbacher, P.H. Weisz, Nucl. Phys. B 370 (1992) 69; Nucl. Phys.
B 375 (1992) 501 (Addendum).
[15] G. Altarelli, L. Maiani, Phys. Lett. B 52 (1974) 351.
[16] M.K. Gaillard, B.W. Lee, Phys. Rev. Lett. 33 (1974) 108.
[17] A.I. Vainshtein, V.I. Zakharov, M.A. Shifman, Sov. Phys. JETP 45 (1977) 670.
[18] F.J. Gilman, M.B. Wise, Phys. Rev. D 20 (1979) 2392.
[19] B. Guberina, R.D. Peccei, Nucl. Phys. B 163 (1980) 289.
[20] J.A. Bagger, K.T. Matchev, R.J. Zhang, Phys. Lett. B 412 (1997) 77.
[21] G. Altarelli, G. Curci, G. Martinelli, S. Petrarca, Nucl. Phys. B 187 (1981) 461.
[22] N. Tracas, N. Vlachos, Phys. Lett. B 115 (1982) 419.

Next-to-leading order calculation of

the color-octet 3S1 gluon fragmentation function
for heavy quarkonium
Eric Braaten a , Jungil Lee b,
a Physics Department, Ohio State University, Columbus, OH 43210, USA
b II. Institut fr Theoretische Physik, Universitt Hamburg, 22761 Hamburg, Germany
Abstract
The short-distance coefficients for the color-octet 3S1 term in the fragmentation function for a
gluon to split into heavy quarkonium states is calculated to order s2 . The gauge-invariant definition
of the fragmentation function by Collins and Soper is employed. Ultraviolet divergences are removed
using the MS renormalization procedure. The longitudinal term in the fragmentation function agrees
with a previous calculation by Beneke and Rothstein. The next-to-leading order correction to the
transverse term disagrees with a previous calculation. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 13.87.Fh; 13.60.Le; 13.88.+e
Keywords: Quarkonium; Fragmentation; Fragmentation function; Nonrelativistic QCD (NRQCD); Color-octet;
Next-to-leading order
1. Introduction
The cross sections for heavy quarkonium states probe the production of heavy-quark
antiquark pairs with small relative momenta. Many of the theoretical uncertainties in
quarkonium production decrease at large transverse momentum. Factorization theorems
for inclusive single-hadron production [1] guarantee that the dominant mechanism for
producing heavy quarkonia with high pT is fragmentation [2], the production of a parton
which subsequently decays into the quarkonium state and other partons. This process is
described by a fragmentation function D(z, ), where z is the longitudinal momentum
fraction of the quarkonium state and is a factorization scale.
Corresponding author. E-mail: jungil@mail.desy.de
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 9 6 - 5
428
E. Braaten, J. Lee / Nuclear Physics B 586 (2000) 427439
The NRQCD factorization formalism [3] can be used to factor the fragmentation
functions D(z, ) for quarkonium into NRQCD matrix elements, which can be regarded
as phenomenological parameters, and short-distance factors, which depend on z and are
calculable in perturbation theory. Most of the phenomenologically relevant short-distance
factors begin at order s2 and have been calculated to leading order. However, there is one
short-distance factor that begins at order s . It is the color-octet 3S1 term in the gluon
fragmentation function, whose NRQCD matrix element is denoted hO8 (3S1 )i. This term
is of particular phenomenological importance. Braaten and Yuan showed that it must
be included in the gluon fragmentation function for triplet P-wave states in order to
avoid an infrared divergence in the short-distance coefficient of the color-singlet matrix
element hO1 (3PJ )i [4]. Braaten and Fleming argued that the hO8 (3S1 )i term is also
phenomenologically necessary in the gluon fragmentation function for spin-triplet S-wave
states in order to explain the production rate of direct J / and 0 at large pT at the
Tevatron [5]. This led to the remarkable prediction by Cho and Wise that J / and 0 at
large pT should be transversely polarized [6].
In the earliest calculations of fragmentation functions for heavy quarkonium [2,7,8],
the short-distance factors were deduced by comparing the cross sections for quarkonium
production with the form predicted by the factorization theorems for inclusive singlehadron production. However, the fragmentation functions can also be defined formally
as matrix elements of bilocal operators in a light-cone gauge [9] or, more generally, as
matrix elements of non-local gauge-invariant operators [1]. The gauge-invariant definition
of Collins and Soper was first applied to calculations of the fragmentation functions for
heavy quarkonium by Ma [10]. The definition is particularly convenient for carrying out
calculations beyond leading order in s . It was used by Ma to calculate the short-distance
factor of the color-octet 3S1 term in the gluon fragmentation function to next-to-leading
order in s [11].
In this paper, we calculate the short-distance coefficients for the color-octet 3S1 term in
the fragmentation function for a gluon to split into heavy quarkonium states to order s2 .
We use the gauge-invariant definition of the fragmentation function given by Collins
and Soper [1], and we remove ultraviolet divergences using the MS renormalization
procedure. Our result for the longitudinal term in the fragmentation function agrees with a
previous calculation by Beneke and Rothstein [12]. Our result for the next-to-leading order
correction in the transverse term disagrees with a previous calculation by Ma [11].
2. Gauge-invariant definition
The fragmentation function DgH (z, ) gives the probability that a gluon produced in
a hard-scattering process involving momentum transfer of order decays into a hadron
H carrying a fraction z of the gluons longitudinal momentum. This function can be
defined in terms of the matrix element of a bilocal operator involving two gluon field
strengths in a light-cone gauge [9]. In Ref. [1], Collins and Soper introduced a gaugeinvariant definition of the gluon fragmentation function that involves the matrix element
429
of a nonlocal operator consisting of two gluon field strengths and eikonal operators. One
advantage of this definition is that it avoids subtleties associated with products of singular
distributions. The gauge-invariant definition is also advantageous for explicit perturbative
calculations, because it allows the calculation of radiative corrections to be simplified by
using Feynman gauge.
The gauge-invariant definition of Collins and Soper is
(g )zN2
DgH (z, ) =
16(N 1)k +
Z+
+
dx eik x
+

+
h0|Gc (0)E (0 )cb PH (zk + ,0 ) E(x )ba G+
a (0 , x , 0 )|0i.
(1)
The operator E(x ) in (1) is an eikonal operator that involves a path-ordered exponential
of gluon field operators along a light-like path:
"
#
Z
E(x )ba = P exp +ig
dz A+ (0+ , z , 0 )
(2)
ba
where A (x) is the matrix-valued gluon field in the adjoint representation: [A (x)]ac =
if abc Ab (x). The operator PH (p+ ,p ) in (1) is a projection onto states that in the asymptotic
2 )/p+ , p ), where
future contain a hadron H with momentum p = (p+ , p = (m2H + p
mH is the mass of the hadron. In the definition (1), the hard-scattering scale can
be identified with the renormalization scale of the nonlocal operator. In perturbative
calculations, it is convenient to use dimensional regularization to regularize ultraviolet
divergences. The prefactor in the definition (1) has therefore been expressed as a function
of the number of spatial dimensions N = 3 2.
For any state H that can be defined in perturbation theory, the definition (1) can be used
to calculate the fragmentation function DgH (z, ) as a power series in s . A convenient
set of Feynman rules for this perturbative expansion is given in Ref. [1]. If the state consists
of a QQ pair with invariant mass p2 , the lowest order diagram is shown in Fig. 1. The
circles connected by the double pair of lines represent the nonlocal operator consisting of
the gluon field strengths and the eikonal operators. The momentum k = (k + , k , k ) flows
into the circle on the left and out the circle on the right. The cutting line represents the
projection onto states that in the asymptotic future include a QQ pair with total momentum
Fig. 1. Leading order Feynman diagram for g QQ.
430
p = (zk + , p2 /(zk + ), 0 ). If the heavy quark has relative momentum q in the QQ rest
frame, the invariant mass is p2 = 4(m2Q + q 2 ).
The fragmentation of a gluon into heavy quarkonium state H involves many momentum
scales, ranging from the hard-scattering scale , which we will assume to be larger than
mQ , to momenta much smaller than mQ where nonperturbative effects are large. The
NRQCD factorization formalism allows the systematic separation of momentum scales
of order mQ and larger from scales of order mQ v or smaller, where v is the typical relative
velocity of the heavy quark in the hadron. The factorization formula has the form
X
H
dmn (z, ) Onm
.
(3)
DgH (z, ) =
mn
The short-distance coefficients dmn (z, ) are independent of the quarkonium state H and
can be calculated as a perturbation series in s (mQ ). All long-distance effects are factored
H i, which can be expressed as matrix elements in an
into the NRQCD matrix elements hOnm
effective field theory. They have the general form
H
(4)
Onm = h0| Kn PH Km |0i,
where Km and Kn are constructed out of color matrices, spin matrices, and covariant
derivatives and the operator PH projects onto states that in the asymptotic future contain
a quarkonium state H . The NRQCD matrix elements are nonperturbative but they are
universal, with the same matrix elements describing inclusive production in other high
energy processes.
The threshold expansion method [13,14] is a general prescription for determining
the short-distance coefficients dnm (z, ) in the NRQCD factorization formula (3). The
diagrams for the fragmentation function of a QQ pair are computed in perturbation theory,
except that the QQ pair is allowed to be in a different state on the two sides of the finalstate cut. To the left of the cutting line, the Q and Q have relative momentum q in the QQ
rest frame and color and spin states specified by Pauli spinors and . To the right of the
cutting line, the Q and Q have relative momentum q 0 and color and spin states specified
by 0 and 0 . After expanding around the threshold q = q 0 = 0, the resulting expression for
the diagrams has the form
X
dmn (z, ) 0 n 0 m ,
(5)
mn
where the color and spin matrices m and n are polynomials in the relative momenta q
and q 0 . The short-distance coefficients dnm (z, ) in the factorization formula (3) can then
be read off from this expression. For example, if the sum of the diagrams includes the
color-octet 3S1 terms

(6)
NmQ dT (z)( ij z i z j ) + dL (z)zi z j 0 j T a 0 i T a ,
then the fragmentation function includes the terms
DgH (z) =

N
dT (z)( ij z i z j ) + dL (z)zi z j h0| j T a PH i T a |0i. (7)
4mQ
431
If we sum over the spin states of H , the matrix element in (7) is proportional to ij and (7)
reduces to

(8)
DgH (z) = (N 1)dT (z) + dL (z) O8H 3S1 ,
where hO8H (3S1 )i is the color-octet 3S1 matrix element defined in Ref. [3]:

O8H 3S1 =
1
h0| i T a PH i T a |0i.
4mQ
(9)
The factor of 1/(4mQ ) accounts for the relativistic normalization of the projection operator
used in Refs. [13,14].
3. Leading order
The only Feynman diagram of order s for the fragmentation process g QQ is shown
in Fig. 1. Using the Feynman rules of Ref. [1], the expression for the diagram can be
easily written down in terms of spinors u and v that describe the color, spin, and relative
momentum states of the Q and Q. We can allow for the Q and Q on the right side of the
final-state cut to have different color and spin states and different relative momentum q 0 by
replacing their spinors by u0 and v 0 . The resulting expression is

p2
s 2

(1
z)
g
n
n
T a v.
(10)
v 0 T a u0 u
2(N 1)(p2 )2
(k n)2
Setting q = q 0 = 0, we can replace the Dirac spinors with Pauli spinors by substituting
u
T a v = 2mQ Li i T a ,
(11)
where Li is a boost matrix that satisfies the identities

g Li Lj = ij ,
pn
n Li = p z i .
p2
(12)
(13)
The expression (10) then reduces to

s 2
(1 z) ij z i z j 0 j T a 0 i T a .
2
8(N 1)mQ
(14)
Comparing with (6), we can read off the order-s terms in the short-distance functions
dT (z) and dL (z) defined in (7):
dT(LO) (z) =
(LO)
dL
s 2
8N(N 1)m3Q
(1 z),
(z) = 0.
The dependence on the number of spatial dimensions N agrees with that in Ref. [14].
(15)
(16)
432
4. Virtual corrections
The Feynman diagrams for the fragmentation function for g QQ at order s2 consist
of virtual corrections, for which the final state is QQ, and real-gluon corrections, for which
the final state is QQg. The diagrams with virtual-gluon corrections to the left of the cutting
line are shown in Fig. 2. The black blob in Fig. 2(a) includes the vertex corrections and
propagator corrections shown in Fig. 3.
We calculate the diagrams using Feynman gauge. In this case, the diagram Fig. 2(b)
vanishes, because the gluon attaching to the eikonal line gives a factor of n . When
contracted with the factor k ng p n from the circle on the left side of the cut,
it gives a factor (k p) n = 0. In the limit q = q 0 = 0, all the other diagrams can
be reduced to the leading order diagram in Fig. 1 times a multiplicative factor. For the
diagrams in Fig. 2(a), this follows from the Dirac equation for the heavy-quark spinors.
For the remaining diagrams, we must also use the fact that the contraction of n with the
right side of the diagram vanishes.
The virtual corrections contribute only to the transverse short-distance function dT (z)
defined in (7). We will express the various contributions in the form of the leading-order
result (15) times a multiplicative factor. The sum of the diagrams in Fig. 2(a), together with
their complex conjugates, is
Fig. 2. The Feynman diagrams of order s2 for g QQ with QQ final states. There are additional
contributions from the complex-conjugate diagrams.
433
Fig. 3. One loop correction diagrams for g QQ.

dT(2a)(z) = dT(LO) (z) 2 Re + + (ZQ 1) ,
(17)
where is the vertex correction factor from the diagrams in Fig. 3(b), is the propagator
correction factor for a gluon with invariant mass 4m2Q from the diagrams in Fig. 3(d), and
1/2
ZQ 1 comes from the wavefunction renormalization factors ZQ for the heavy quark
from the diagrams in Fig. 3(c). These correction factors are given by

1 (1 + ) 13
s 2 13 (1 + )
+ 3 ln 2 i ,
+
(18)
=
m2Q
12 UV
12 IR
6
2

s 2
1 (1 + ) 2 (1 + ) 4
2
ln
2
,
(19)
ZQ 1 =
m2Q
3 UV
3 IR
3

s 2 i 15 2nf (1 + ) 93 10nf
e
+
,
(20)
=
m2Q
12
UV
36
where nf is the number of light quark flavors. The subscripts on the poles in indicate
whether the divergences are of ultraviolet or infrared origin. The sum of the virtual
corrections given in (17) is

s 2
(2a)
(LO)
(1 + )
dT (z) = dT (z)
m2Q

123 10nf
12 nf
3
+ 2 ln 2 .
(21)
3UV
2IR
18
434
Mas result for these diagrams [11] differs by an additive constant 5.487 inside the square
brackets and by setting N(N 1) 3(N 1) in dT(LO) (z).
The contribution from the diagrams in Fig. 2(c) with its complex conjugate is

(2c)
(LO)
(22)
dT (z) = dT (z) 24s 2 Re i(2k nIABC IAB ) ,
where the scalar integrals IAB are given in Appendix A. The contribution from the sum
of the diagrams in Figs. 2(d) and 2(e) with their complex conjugates is

(2d,e)
(LO)
(z) = dT (z) 96s 2 m2Q Re i(k nIABCD IABD ) .
(23)
dT
The sum of the virtual corrections given in (22) and (23) is

s 2
(2c,d,e)
(LO)
(z) = dT (z)
(1 + )
dT
m2Q

3
2
3(1 )
+ 6 ln 2 + 6 ln2 2 .
+
+3
2UV IR 2UV

2
(24)
Mas result for these diagrams [11] differs by an additive constant 3 + 32 2 6 ln2 2
inside the square brackets and by setting N(N 1) 3(N 1) in dT(LO) (z).
The total virtual correction is the sum of (21) and (24):

s 2
(virtual)
(LO)
(z) = dT (z)
dT
m2Q

2
3(1 ) (1 + )
(1 + ) 177 10nf
+ 8 ln 2 + 6 ln2 2 , (25)
+ 0
+
2
UV IR
UV
18
2
where 0 = (33 2nf )/6.
5. Real gluon corrections

The Feynman diagrams for the real-gluon corrections to the fragmentation function for
g QQ pair are shown in Fig. 4. We draw the 5 left-half diagrams only, but they must
be multiplied by their complex conjugates to give a total of 25 diagrams. We calculate the
diagrams using Feynman gauge. After a considerable amount of algebra, they reduce to
(real)
dT (z) =
s 2
8N(N 1)m3Q

2 Z

3s
t 1
2
1
dx 2 ,
1
2
(1 ) mQ
z(1 z)
x
(1z)/z
(26)
dL(real) (z) =
2
8Nm3Q
3s
(1 )

2 1 z 2
m2Q
dx
t
x2
(27)
(1z)/z
where t = (1 z)(zx + z 1), x = 2q p/p2 , q is the final-state gluon momentum, and p

is the QQ momentum. Integration over x can be done using the following identities:
435
Fig. 4. The Feynman diagrams of order s2 for g QQ with QQg final states. There are a total of
25 diagrams, but only the left halves of the diagrams are shown.
Z
dx
t 1
1
= z(1 z)12
B(1 + , 1 ),
UV
x2
(28)
dx
t
= z(1 z)12 B(1 + , 1 ).
x2
(29)
(1z)/z
Z
(1z)/z
The subscript on in (28) indicates that the pole has an ultraviolet origin. In (26), there
is also an infrared divergence associated with the limit z 1. It can be made explicit by
using the expansion:

1
1
ln(1 z)
1
=
(1 z) +
2
+ O 2 .
(30)
2IR
(1 z)+
1z
(1 z)1+2
+
The final results for the real-gluon corrections are

s 2
s 2
(real)
(1 + )
dT (z) =
m2Q
8N(N 1)m3Q

3(1 )
1z
z
3(1 )
+ z(1 z)
(1 z) +
+

2UV IR
UV
(1 z)+
z

6 ln(1 z)
+ 6 2 z + z2 ln(1 z) ,
(31)
z
1z
+
s
3s 1 z
(real)
.
(32)
dL (z) =
3
z
8NmQ
Note that the double-pole term proportional to 1/(UV IR ) in (31) exactly cancels its
counter part in the virtual correction (25). The sum of (25) and (31) is free of infrared
436
divergences. Mas result for the real-gluon corrections [11] is completely different from the
sum of (31) and (32). In particular, he found that the [ln(1 z)/(1 z)]+ terms cancelled.
In the way we organized the calculation, there is no possibility of such a cancellation.
6. Renormalization
The sum of the order-s2 corrections to dT (z) in (25) and (31) still contains ultraviolet
divergences in the form of single poles in . These divergences are cancelled by the
renormalization of the coupling constant s in the leading-order expression (15) and by
the renormalization of the nonlocal operator in (1). The renormalization of s in the MS
scheme can be carried out by making the following substitution in (15):

s
1
0 4e
,
(33)
s s 1
2
UV
where 0 = (33 2nf )/6. The operator renormalization in the MS scheme can be carried
out by making the following substitution
(LO)
(LO)
dT (z) dT (z)
1
s
4e
2
UV
Z1
dy
(LO)
Pgg (y) dT (z/y),
y
(34)
where Pgg (y) is the gluon splitting function:

0
z
1z
+ z(1 z) +
(1 z) .
+
Pgg (z) = 6
(1 z)+
z
6
(35)
The sum of the two contributions of order s2 from renormalization is

dT(ren) (z) =
s 2
1)m3Q
(3)
s
4e
UV
8N(N

0
z
1z
+ z(1 z) +
(1 z) .
(36)
+
(1 z)+
z
3
This cancels the single ultraviolet poles in the sum of (25) and (31). Our final result for the
transverse fragmentation function is obtained by adding the order-s2 corrections to dT (z)
from (25), (31), and (36) and taking 0:

s ()
1
s ()
A()(1 z) + ln
Pgg (z)
(1 z) +
dT (z, ) =
2mQ 2
48m3Q

6 ln(1 z)
,
(37)
+ 6 2 z + z2 ln(1 z)
z
1z
+
where the coefficient A() is

13
2 2
+ 8 ln 2 + 6 ln2 2.
+
+
A() = 0 ln
2mQ
6
3
2
(38)
The result (37) disagrees with the final result obtained by Ma [11]. Our final result for the
longitudinal fragmentation function is obtained by setting 0 in (32):
dL (z, ) =
s2 () 1 z
.
8m3Q z
437
(39)
This agrees with the result of Beneke and Rothstein [12]. The fragmentation probabilities
obtained by integrating dT (z) and dL (z) diverge, because these functions behave like 1/z
as z 0. The higher moments of the fragmentation functions however are well-defined.
7. Discussion
The color-octet 3S1 term in the fragmentation function is important for calculating the
production at large pT of spin-singlet S-wave states, like the J /, and spin-triplet P-wave
states, like the cJ . We can deduce the fragmentation functions for each of their spin states
by using the approximate spin symmetry of NRQCD to simplify the expression (7). The
functions dT (z) in (37) and dL (z) (39) give the fragmentation functions for the transverse
and longitudinal spin states of the J /, respectively:
J / 3
S1 ,
DgJ /(1) (z) = dT (z) O8
J / 3
S1 .
DgJ /(0) (z) = dL (z) O8
(40)
(41)
The sum over spin states is

J / 3
S1 .
DgJ / (z) = 2dT (z) + dL (z) O8
The fragmentation functions for each of the spin states of the cJ are [16]

Dgc0 (z) = 2dT (z) + dL (z) O8 c0 3S1 ,

Dgc1 (0)(z) = 3dT (z) O8 c0 3S1 ,

3
Dgc1 (1) (z) = dT (z) + dL (z) O8 c0 3S1 ,
2

Dgc2 (0)(z) = dT (z) + 2dL(z) O8 c0 3S1 ,

3
Dgc2 (1) (z) = dT (z) + dL (z) O8 c0 3S1 ,
2

Dgc2 (2) (z) = 3dT (z) O8 c0 3S1 .
The sums over spin states are

DgcJ (z) = (2J + 1) 2dT (z) + dL (z) O8 c0 3S1 .
(42)
(43)
(44)
(45)
(46)
(47)
(48)
(49)
In order to give accurate predictions for the production of quarkonium at large pT ,

it is important to know the next-to-leading order correction to the color-octet 3S1 term
in the gluon fragmentation function. Unfortunately our calculation of the short-distance
coefficient disagrees with the previous calculation by Ma. An independent calculation of
this important function is therefore essential.
438
Acknowledgements
This work was supported in part by the US Department of Energy Division of High
Energy Physics under grant DE-FG02-91-ER40690, by the Alexander von Humboldt
Foundation, and by the Korea Institute for Advanced Study. J.L. would like to thank the
OSU theory group for its hospitality during his stay in Columbus.
Appendix A. Integral table

In this appendix, we present the explicit values of the integrals encountered in evaluating
the virtual-gluon corrections. These integrals have the form
Z
dN+1 l
1
,
(A.1)
IAB =
(2)N+1 AB
where the denominator AB can be a product of 1, 2, 3, or 4 of the following factors:
A = l 2 + i,
(A.2)
B = (l p) + i = l 2l
2
p + 4m2Q
+ i,
C = (p l) n + i,
D = (l p/2)
m2Q
(A.3)
(A.4)
+ i = l l p + i.
2
(A.5)
The momentum p is that of a QQ pair with zero relative momentum (p2 = 4m2Q ) and n
is light-like (n2 = 0). The integrals IA and IB vanish in dimensional regularization. By
symmetry under p l p, we have IAD = IBD . Some of the integrals can be reduced to
ones with fewer denominators by using the identity A + B 2D = 4m2Q :
4m2Q IABD = 2(IAD IAB ),
4m2Q IABCD
= IACD + IBCD 2IABC .
The independent integrals that need to be evaluated are therefore

i
i
(1 + ) 2 (1 )
e
,
IAB =
(4)2 m2Q
UV (2 2)

i
4 (1 + )
,
IAD =
(4)2 m2Q UV (1 2)
(A.6)
(A.7)
(A.8)
(A.9)
i
i
(1 + ) 2 (1 )
e
,
(A.10)
2
2
(4) p n mQ
UV IR (1 2)

2

+i
4 (1 + )
2
2
6
ln
2
+
O

2
ln
2
+

IACD =
, (A.11)
(4)2 p n m2Q
UV
3

i
4 (1 + )
.
(A.12)
IBCD =
(4)2 p n m2Q
UV IR
IABC =
439
The subscripts on the poles in indicate whether the divergences are of ultraviolet or
infrared origin.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
J.C. Collins, D.E. Soper, Nucl. Phys. B 193 (1981) 381; Nucl. Phys. B 194 (1982) 445.
E. Braaten, T.C. Yuan, Phys. Rev. Lett. 71 (1993) 1673; Phys. Rev. D 52 (1995) 6627.
G.T. Bodwin, E. Braaten, G.P. Lepage, Phys. Rev. D 51 (1995) 1125.
E. Braaten, T.C. Yuan, Phys. Rev. D 50 (1994) 3176.
E. Braaten, S. Fleming, Phys. Rev. Lett. 74 (1995) 3327.
P. Cho, M.B. Wise, Phys. Lett. B 346 (1995) 129.
E. Braaten, K. Cheung, T.C. Yuan, Phys. Rev. D 48 (1993) 4230.
E. Braaten, K. Cheung, T.C. Yuan, Phys. Rev. D 48 (1993) R5049.
G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
J.P. Ma, Phys. Lett. B 332 (1994) 398.
J.P. Ma, Nucl. Phys. B 447 (1995) 405.
M. Beneke, I.Z. Rothstein, Phys. Lett. B 372 (1996) 157; Phys. Lett. B 389 (1996) 769
(Erratum).
E. Braaten, Y.-Q. Chen, Phys. Rev. D 54 (1996) 3216.
P. Cho, M.B. Wise, S.P. Trivedi, Phys. Rev. D 51 (1995) 2039.

Static correlation lengths in QCD at high

temperatures and finite densities
A. Hart a , M. Laine b,c , O. Philipsen b
a Dept. of Physics and Astronomy, Univ. of Edinburgh, Edinburgh EH9 3JZ, Scotland, UK
b Theory Division, CERN, CH-1211 Geneva 23, Switzerland
c Department of Physics, P.O. Box 9, FIN-00014 Univ. of Helsinki, Finland
Abstract
We use a perturbatively derived effective field theory and three-dimensional lattice simulations
to determine the longest static correlation lengths in the deconfined QCD plasma phase at high
temperatures (T & 2Tc ) and finite densities ( . 4T ). For vanishing chemical potential, we refine
a previous determination of the Debye screening length, and determine the dependence of different
correlation lengths on the number of massless flavours as well as on the number of colours. For
non-vanishing but small chemical potential, the existence of Debye screening allows us to carry out
simulations corresponding to the full QCD with two (or three) massless dynamical flavours, in spite
of a complex action. We investigate how the correlation lengths in the different quantum number
channels change as the chemical potential is switched on. 2000 Elsevier Science B.V. All rights
reserved.
PACS: 12.38.Mh; 11.10.Wx; 12.38.Gc; 11.10.Kk
Keywords: Finite-temperature field theory; Quarkgluon plasma; Lattice QCD simulations; Field theory in three
dimensions
1. Introduction
With the advent of RHIC and ALICE, there is a growing need for a precise understanding of various properties of QCD at temperatures up to a GeV. At the moment, we are
still far from a satisfactory level in this respect, even for the equilibrium properties of the
plasma. Indeed, even though the system can in principle be described as a gas of quarks
and gluons, a fully perturbative computation with these degrees of freedom does in practice
not work well at any reasonable temperatures below 1010 GeV, since the perturbative series is badly convergent due to infrared sensitive contributions [15]. On the other hand,
E-mail addresses: hart@ph.ed.ac.uk (A. Hart), mikko.laine@cern.ch (M. Laine), owe.philipsen@cern.ch
(O. Philipsen).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 1 8 - 1
444
A. Hart et al. / Nuclear Physics B 586 (2000) 443474
the only systematic fully non-perturbative method available, four-dimensional (4d) lattice
simulations, is severely restricted in the presence of light dynamical fermions, and even
more so in the presence of a finite baryon density; the current state of the art is summarised
in [6]. A first principles solution of any problem related to non-equilibrium phenomena
remains even further in the future.
The physics problem we address in this paper is the determination of various static
correlation lengths. To tackle this problem, we will employ a method which can overcome
some of the difficulties mentioned above. The method combines analytic and numerical
techniques, in a way that both are used in a regime where they are well manageable.
First, perturbation theory is employed in deriving, by what is called dimensional
reduction [714,4,5], an effective theory for the long range degrees of freedom of the
system. At high temperatures all such degrees of freedom are bosonic, since fermions
are screened by non-zero Matsubara frequencies T n, with n odd. The reduction step
can thus be carried out with dynamical, massless fermions and, as we shall see, with a
finite chemical potential. Second, non-perturbative lattice simulations are used to study the
infrared sensitive dynamics of the remaining degrees of freedom. The theory to be studied
with simulations is the SU(3) + adjoint Higgs model in three dimensions (3d), and many
of its properties have already been determined [15].
The method we employ suffers clearly from a number of restrictions. First of all, in
QCD it is limited to temperatures above Tc , roughly T & 2Tc . 1 This can be inferred from
a comparison of the dynamical scales described by the effective theory to those integrated
out [5] as well as, more concretely, from a direct quantitative comparison of the correlation
lengths measured within 3d for SU(2) [22], with those determined using 4d SU(2) lattice
simulations at finite temperature (without fermions) [23]. 2 Second, as we shall see, it is
also limited by the largest chemical potential that can be reached. We will be able to go up
to . 4T .
If there are restrictions, there are also strengths. What can be done within the effective
theory can be done quite precisely. The simulations are comparatively easy technically,
they yield high accuracy continuum limits for the quantities under investigation, and hence
there is little ambiguity in the conclusions. For instance, studies in 3d have produced
detailed and accurate insight into the relative sizes of perturbative and non-perturbative
contributions to the Debye screening length [22,26], a quantity which could be directly
relevant for such signals of the quarkgluon plasma as J / suppression.
In the present work, we extend the study of [22] first to SU(3) YangMills theory, and
then to two- and three-flavour QCD in the massless limit. In comparing with earlier results
in SU(3) [5,26], we improve on the accuracy of Debye screening length determination and
measure the screening lengths also in the other quantum number channels. We also discuss
1 For the electroweak theory, in contrast, similar methods allow for a precise determination of the properties of
the phase transition at T = Tc for Higgs masses mH 30 . . . 250 GeV [1621].
2 For early work with similar conclusions both in SU(2) and SU(3), see [912]. Recently, analogous results have
been reached also by considering gauge fixed correlators [24], as well as by considering dimensional reduction
of pure SU(3) from (2 + 1)d to 2d [25].
445
explicitly the Nc and Nf dependence of our results here Nc is the number of colours
and Nf is the number of (massless) quark flavours.
We then proceed to extend our calculations to finite baryon density. Our approach is
straightforward: dimensional reduction leads to an additional complex term in the effective action, which has to be absorbed in the observables for practical simulations. This
reweighting displays in principle similar problems as chemical potential simulations in
four dimensions (for reviews, see [6,27]). We find that for the effective theory there exists a
range of volumes and ratios /T , however, for which the problems are in practice manageable. We demonstrate this by investigating a range of imaginary chemical potentials [28],
for which we find complete agreement between reweighted calculations and those using
the exact action. We then employ the reweighting technique to the case of real chemical
potentials.
Let us note that our approach relies on /(T ) being small, but no restriction is imposed
on fermionic masses mi , i = 1, . . . , Nf , which we take to be zero. In 4d simulations, on
the other hand, progress has been possible when the mi are taken to be very large [3034],
whereas the temperature can be small.
2. Continuum formulation for = 0
We start by reviewing the result of the dimensional reduction step for = 0. We follow
closely Ref. [5].
The effective theory emerging from hot QCD by dimensional reduction [7,8] is the
SU(3) + adjoint Higgs model with the action

Z
2
1
Tr Fij2 + Tr[Di , A0 ]2 + m23 Tr A20 + 3 Tr A20 ,
(2.1)
S = d3 x
2
where Fij = i Aj j Ai + ig[Ai , Aj ], Di = i + ig3 Ai , Fij , Ai , and A0 are all traceless
3 3 Hermitian matrices (A0 = Aa0 Ta , etc.), and g32 and 3 are the gauge and scalar
coupling constants with mass dimension one, respectively. The physical properties of the
effective theory are determined by the two dimensionless ratios
x=
,
2
g3
y=
m23 ( 3 = g32 )
g34
(2.2)
where 3 is the MS dimensional regularization scale in 3d. The parameters x, y, as well

as the scale g32 , are via dimensional reduction functions of the temperature T and the QCD
scale MS . For Nc = 3, the result is [5]
24 2
1
T,
33 2Nf ln( g /MS )
9 Nf
1
,
x=
33 2Nf ln( x /MS )
g32 =
3
2
(9 Nf )(6 + Nf ) 486 33Nf 11Nf 2Nf
+
+ O(x).
y(x) =
144 2x
96(9 Nf ) 2
(2.3)
(2.4)
(2.5)
446
We have assumed here Nf massless flavours, but the inclusion of quark masses is also
possible in principle.
The scales entering in Eqs. (2.3), (2.4) are
i = T i ,

3ci + 4Nf ln 4
E
T 7.0555T ,
i = exp
,
T = 4e
66 4Nf
54 22Nf
4
cx =
+ Nf .
cg = 1,
9 Nf
3
(2.6)
(2.7)
(2.8)
To be explicit, for Nf = 2 we obtain

24 2 T
,
29 ln(8.11 T /MS )
7
,
x=
29 ln(6.91 T /MS )
g32 =
y=
7
15
+
.
2
18 x 28 2
(2.9)
Corrections to these expressions are of relative magnitude O(s /)2 O(x 2 ).

An important question is how small one can in practice take T /MS and still have
only small corrections in Eqs. (2.3)(2.5). As was observed in [5], for y(x) the expansion
appears to converge surprisingly fast, with an error on a few percent level even at T MS .
The expansions need a priori not be as good for every parameter, however.
Consider for instance g32 . One may expect higher loop graphs to amount to an increase in
the effective scale factor g , since there are many massive ( 2T ) particles inside the
loops instead of one. Thus an upper bound on the error in g32 can be obtained by evaluating
the 2-loop QCD running coupling at the 1-loop scale g 7T (for Nf = 0). This gives
a correction of up to 25%, which is quite large. An explicit computation of g32 including
effects of order O(g 6 ) would thus be welcome in order to clarify whether the expansion
for g32 in reality converges as rapidly as for y(x) or not. Such computations are beyond our
scope here and we will simply use Eqs. (2.3)(2.5); a comparison with direct 4d simulations at Nf = 0 turns out to be quite satisfactory within this procedure, and thus we will
work under the assumption that the error in g32 , x is similarly small as in y.
Finally, we may sometimes want to express the temperature to which our simulations
correspond in terms of Tc , rather than MS . In view of Eq. (2.9), this requires knowledge
of the deconfinement temperature Tc /MS , which can only be fixed with 4d simulations.
For the case Nf = 0, we take the value Tc /MS = 1.03(19) from [35]. 3 For Nf 6= 0, the
determination of Tc is not accurate, and we will take Tc /MS = 1.0 as a reference. Our
results however only depend on T /MS , so they can be reinterpreted as corresponding to
some other temperature relative to Tc if need be.
2.1. Operators and quantum numbers
The physical observables on which we shall focus in this paper are spatial correlation
lengths of QCD at finite temperatures. For the practical calculations we reinterpret the 3d
3 A recent computation [36] favours a slightly larger central value, but well within the error bars.
447
theory in Eq. (2.1) which we use to compute them, as a (2 + 1)d one. We take operators
to live on the (x1 , x2 )-plane, and the correlations are taken in the x3 direction. Thus we
compute the spectrum of the 2d Hamiltonian of the effective theory, whose eigenvalues
then correspond to screening masses or inverse spatial correlation lengths of the 4d theory
at finite temperature. Thus, unless otherwise stated, bound states, glueballs, etc. refer
to the eigenstates of the effective Hamiltonian and to the screening masses of the 4d theory
in the above sense.
We use the quantum number notation
R:
Di Di ,
Fij Fij ,
A0 A0 ,
P:
D1 D1 ,
D2 D2 ,
F12 F12 ,
C:
Di Di ,
Fij Fij ,
A0 A0 .
(2.10)
A0 A0 ,
(2.11)
(2.12)
The action in Eq. (2.1) is invariant under these operations. In the finite temperature context,
the operation R in the 3d theory is a remnant of the 4d time reversal operation [37].
The parity P means a parity on the 2d plane, i.e., a reflection across the x1 -axis. The
charge conjugation C is a non-trivial quantum number only for SU(3), since for SU(2) it is
just a global gauge transformation i 2 , so that there are no gauge invariant operators odd
in C [37]. In addition to these discrete transformations, we define a rotation in the (x1, x2 )plane, with the corresponding angular momentum J . Thus the full symmetry group is
SO(2) Z2 (R) Z2 (P ) Z2 (C), and accordingly we classify our operators and states
by JRP C .
We note from Eqs. (2.10)(2.12) that apart from D2 , the action of RP C on the operators
in the (x1, x2 )-plane corresponds to complex conjugation (and x2 x2 ). Thus any real
operator which does not contain D2 is even in RP C. This means, in particular, that for the
scalar J = 0 operators in which no single D2 can appear, there are effectively only two
quantum numbers left out of the original three, thus four different channels. The lowest
dimensional gauge invariant operators of these types in the J = 0 channels are:
JRP C = 0++
+ :
JRP C = 0
+ :
2
Tr A20 , Tr F12
,...
3
Tr F12
, Tr A20 F12 , . . .
JRP C = 0+
:
Tr A0 F12 , . . .
JRP C
2
Tr A30 , Tr A0 F12
,... .
= 0+
:
(2.13)
All operators here with C = 1 vanish identically for SU(2). Going out of the plane (i.e.,
allowing also for F13 , F23 ), other channels would in principle become possible [37], but
we do not expect them to change our conclusions.
3. Lattice formulation for = 0
To simulate the effective theory in Eq. (2.1) on the lattice, we use the discretised version

X
X

1
Tr (x)(x)
1 Re Tr Uij (x) + 2
S=
3
x
x,i>j
448
X

2
Tr (x)Ui (x)(x + )Ui (x) +
2 Tr (x)(x) 1 ,
(3.1)
x,i
where the fields in the lattice action have been rescaled relative to the continuum, A0 (x) =
(/a)1/2 (x) where a is the lattice spacing, and Uij (x) denotes the elementary 1 1
plaquette in the i, j -plane located at x. The parameters of the continuum and lattice theory
are up to two loops related by a set of equations which become exact in the continuum
limit [38,39],
3x 2
6
,
=
,
2
2
ag3

3x
3.1759115
5
2 1
3
+
1+ x
y=
18
4
3

1
+
60x 20x 2 (ln + 0.08849) + 34.768x + 36.130 .
(3.2)
2
16
For a given pair of continuum parameters x, y these equations determine the lattice
parameters , as a function of the lattice spacing, and hence govern the approach to the
continuum limit, . We have not implemented O(a) improvement [40] here, since
the discretisation effects in the correlation lengths are already quite small at the values of
we are using.
=
3.1. Observables
Operators for all the quantum number channels considered can be constructed from
a number of basic operator types. We consider operators involving only scalar fields or
products of scalar and gauge field variables,

R3 (x) = Tr 3 (x) ,
R2 (x) = Tr 2 (x) ,

Li (x) = Tr (x)Ui (x)(x + )Ui (x) ,

B2 (x) = Tr 2 (x)Uij (x) .
(3.3)
B1 (x) = Tr (x)Uij (x) ,
Furthermore, we have loop operators constructed from link variables only,

Cij11 (x) = Tr Ui (x)Uj (x + )Ui (x + )Uj (x) , i, j = 1, 2, i 6= j,
(3.4)
and in addition to the elementary plaquette C 11 , we also consider squares of size 2 2

as well as rectangles of size 1 2, 1 3, 2 3. Another useful pure gauge operator is the
Polyakov loop along a spatial direction j ,
Pj(L) (x) = Re Tr
L1
Y
Uj (x + m),
j = 1, 2,
(3.5)
m=0
which can be used to extract the 3d string tension. It also gives useful information about
finite volume effects via so called torelon states [41,42].
Operators with definite quantum number assignments are constructed from the above
types by taking linear combinations with appropriate transformation properties. Clearly
449
operators containing an even number of scalar fields, such as B2 , will couple to R = +1

states, and those with odd, such as B1 , to R = 1. The different P , C channels can be
chosen by utilising the projection operators 12 (1P ), 12 (1C), where, e.g., CB1 (x1 , x2 ) =
B1 (x1 , x2 ), P B1 (x1 , x2 ) = B1 (x1 , x2 ). Different spin states are obtained by employing
the operation R(n ) which rotates the operators by n(/2) around x. Note also that even
though lattice rotations are restricted to multiples of /2, we are in practice close enough
to the continuum limit that continuum symmetries are reproduced within the statistical
errors; thus we keep the continuum notation in terms of J even on the lattice.
In the following we list the operators used corresponding to Eq. (2.13):
0++
+ channel:
R: R2 (x),
L: Re(L1 (x) + L2 (x)),
C: real part of symmetric combinations of C 11 , C 22 , C 12 , C 13 , C 23 ,
(L)
(L)
P : P1 (x) + P2 (x),
2
2
(L)
T : P1 (x) + P2(L) (x) ,
0
+ channel:
C: imaginary part of symmetric combinations of C 11 , C 22 , C 12 , C 13 , C 23 ,
P
B: Im 4n=1 R(n )B2 (x),
0+
channel:
P
B: Im 4n=1 R(n )B1 (x),
0+
channel:
R: R3 (x),
P
B: Re 4n=1 R(n )B1 (x).
We have also measured higher spin states. For 2++
+ we have a fairly large basis, identical to
the one described in [22]. The J = 1 states prove to be quite heavy, so that their relevance
for the 4d finite temperature system is not immediately obvious at modest temperatures,
and thus for simplicity we consider here only the ground states in the channels with
R = 1, without specifying the other quantum numbers.
3.2. Blocking and matrix correlators
For later reference, let us recall the general principles of how the operators discussed
above can be used for a reliable extraction of correlation lengths [4346].
The eigenstates of the 2d Hamiltonian in the region of parameter space where we work
are bound states. In order to increase the overlap of our operators onto such extended
states, we construct smeared link and scalar field variables i as described in [22]. For
every quantum number channel, we then measure the correlation matrix

(3.6)
Cij (t) = i (t)j (0) ,
450
where we have denoted t = x3 . This matrix can be diagonalised by solving a generalised

eigenvalue problem,
C 1 (0)C(t)v n = n (t)v n ,
(3.7)
where n (t) exp(aMn t). We carry out this procedure at t = a, 2a, 3a, which gives
somewhat differing eigenvectors v n , and check that the final outcome remains the same
within error bars. We normalise the eigenvectors according to v n C(0)v n = 1, so that n =
P (i)
i v n i satisfy n n = 1, and are thus the normalized eigenstates of the 2d Hamiltonian.

The final results are extracted from the correlation functions

(3.8)
Gn (t) = n (t)n (0) = v n C(t)v n ,
by computing effective masses

Gn (t + 1)
,
aMeff,n (t) = ln
Gn (t)
(3.9)
and ensuring that these have attained a stable plateau value. Information about the
composition of n in terms of the operators used in the simulation, which we shall quote
P
(j )
in some of our tables, is obtained from the overlaps hi n i = j Cij (0)v n .
3.3. Simulation and analysis
In our Monte Carlo simulation of the lattice action, Eq. (3.1), link variables are
updated by a combination of heat bath and over-relaxation steps with algorithms described
in [47]. The scalar fields are generated by a combination of heat bath and reflection
steps [48]. One compound sweep consists of several over-relaxation and reflection
updates following each heat bath update of gauge and scalar fields. Measurements are
taken after every compound sweep. Typically, we gathered between 5 000 and 20 000
measurements depending on the lattice sizes. Statistical errors are estimated using a
jackknife procedure with bin sizes of 100250 measurements.
4. Numerical results for = 0

In this section we discuss simulations of the effective theory for parameter values
corresponding to hot SU(3) gauge theory with Nf = 0, 2, 3, 4 flavours of fermions at zero
baryon density. Detailed investigations of finite volume effects and scaling of correlation
lengths have been performed for the 3d pure SU(2) and SU(2) + adjoint Higgs theories
in [22,47], as well as for the 3d pure SU(3) theory in [26,47]. In these works explicit
continuum extrapolations were performed. For large enough the scaling behaviour was
found to be quite good, with continuum results differing from those on a fine lattice by only
a few percent. Our gauge couplings here were chosen to produce a similar lattice spacing,
and checks also show good scaling. The parameter combinations and the lattices used for
them are collected in Table 1.
451
Table 1
The lattice parameters and sizes used for calculations at = 0. Let us stress that for Nf > 0, T /Tc
means really T /MS , see Section 2
= 21
= 28
0.39188
L2 T = 283 , 383
L2 T = 403
0.009636
4.0
L2 T = 403
0.10304
0.43668
L2 T = 303
L2 T = 303
L2 T = 303
L2 T = 403
L2 T = 303
L2 T = 303
Nf
T /Tc
2.0
0.11346
1011
1.5
2
3
4
2.0
0.09191
0.4830
1.5
0.08680
0.47890
2.0
0.07814
0.52741
2.0
0.062921
0.569680
4.1. The mass spectrum for Nf = 0

The main features of the spectrum, displayed in detail in Tables 5, 7 in Appendix A,
are the same as those found in previous studies of the confinement region for SU(2)
with fundamental [42] or adjoint [22] scalar fields. There is a dense spectrum of bound
states, consisting of a replication of the glueball spectrum found in the d = 2 + 1 pure
SU(3) theory [47], and additional bound states of scalars, with little mixing between
the two. This may be concluded from the fact that, as indicated in Table 5, some
eigenstates are composed predominantly of purely gluonic operators C, with practically
no contribution from operators carrying the same quantum numbers but containing scalars.
Comparing with [47], we find that our gluonic states have quantitatively the same masses
as the corresponding glueballs in the pure YangMills theory; the comparison is shown
in Table 2.
As can be observed from Table 2, however, in the physically interesting region of
couplings the lightest states in given quantum number channels are not glueballs but
always states including the scalar field . This can be understood in the sense that in
Table 2
Mass estimates M/g32 at T = 2Tc and Nf = 0. Our data for
the glueballs are compared with the results obtained in the pure
gauge theory [47]
gauge-Higgs
JRP C
0++
+
0
+
2++
+
pure gauge
scalar
glueball
glueball
0.994(19)
2.95(16)
2.511(65)
3.61(32)
2.575(18)
3.841(28)
2.46(10)
3.50(25)
3.795(27)
3.355(94)
4.14(28)
4.257(33)
452
the finite temperature context the phase transition is (for Nf = 0) assumed to be driven
by the Z(N) symmetry related to A0 (or on the lattice), and thus is the dominant
infrared degree of freedom. Let us now discuss the corresponding correlation lengths in
more detail.
4.1.1. The magnetic sector, R = +1
Consider first operators with R = +1. We will call this the magnetic sector, even
though the operators can include an even number of A0 -fields. This sector determines the
correlation length related to the lowest lying glueballs and all other 0++
+ observables, as
well as that felt by the real part of the 4d temporal Polyakov loop [37,49,50] (not to be
confused with the spatial Polyakov loop in the 3d theory). In order to express the masses in
units of temperature, we use the perturbative expression for the gauge coupling in Eq. (2.3)
to arrive at g32 (T = 2Tc ) 2.7T and g32 (T = 1011Tc ) 0.25T . The spectrum in these units
is shown in Fig. 1.
Let us first discuss the channel 0++
+ . From Table 5 in Appendix A, we note that the
lightest state (open circle in Fig. 1 (left)) is dominantly Tr A20 , while the next state (filled
2 . Thus, the nave ordering of states consisting
circle in Fig. 1 (left)) is dominantly Tr F12
2
of the fields A0 gT and Ai g T is reversed non-perturbatively: at low temperatures it
is the A0 which is responsible for the lightest gauge invariant state in the system.
As the temperature is increased, however, the ordering gets changed. The lowest state
becomes heavier with increasing temperature, as the scalar constituent mass parameter
y 1/2 is then growing (for a discussion of the constituent picture in a similar context,
2 does not contain much A admixture.
see [24,51]). On the other hand, the glueball Tr F12
0
Correspondingly its mass, in units of g32 , is quite insensitive to y, so that for large enough
temperatures indeed the glueball corresponds to the lightest excitation. This situation is,
however, only realised at high temperatures T > 102 Tc ; an extreme example is shown
Fig. 1. The spectrum of screening masses in various quantum number channels at Nf = 0, T = 2Tc
(left), Nf = 0, T 1011 Tc (right). Filled symbols denote 3d glueball states, which have become the
lightest excitations at T 1011 Tc . The states 1+ are much heavier than 1 at high temperatures,
and thus not visible on the right.
453
in Fig. 1 (right). In this limit the A0 may be integrated out, leaving the 3d pure SU(3)
gauge theory as an effective theory [7,8].
For the channel 0
+ , the behaviour is similar, but the masses are much larger. Let
us mention here already that in the 4d theory with fermions at modest temperatures
T 2Tc , all the higher-lying bosonic states in general are probably heavier than some
gauge invariant states consisting of fermionic fields ,

which are not addressed by
our theory, and whose mass we may expect to be . 2T . (At lower temperatures such
states can be even lighter, since in the chiral limit they are expected to represent the critical
degrees of freedom.)
To get an impression of the quality of dimensional reduction for Nf = 0, we now
state Tr A20 with that found directly from a 4d simulation at
compare the 0++
+
Nf = 0 [23]. Our value is M[0++
+ ] = 2.71(6)T , to be compared with M = 2.60(4)T
reported in [23]. In complete analogy with the SU(2) case [22], it is thus found that
dimensional reduction quantitatively reproduces the lowest screening mass of the 4d Yang
Mills theory at temperatures as low as T = 2Tc .
We can also carry out another comparison. The real part of the 4d temporal Polyakov
loop carries quantum numbers 0++
+ , and non-perturbatively mixes with other operators
in that channel. Correspondingly, its decay should also be determined by the bound state
Tr A20 . Indeed, the exponential decay of the 4d temporal Polyakov loop correlator is in 4d
measured to be 2.5T [52], which is quite compatible with the value 2.7T we find here.
On the other hand, in contrast to SU(2) where even several excitations agree between
the full and the effective theory, we find here deviations of about 20% in comparing
with [23]. This may signal that the effective theory is not as reliable for states whose
mass is approaching 2T . However, we also believe that there could be some room for
improvement upon this apparent disagreement. The 4d simulations are quite difficult, and
it is not easy to guarantee at this stage that the infinite volume and continuum limits are
reached for the excited states.
4.1.2. The electric Debye sector, R = 1
Of particular phenomenological relevance for the QCD plasma is the Debye mass,
whose inverse gives the length scale over which the colour-electric field is screened. The
expansion in powers of coupling constants for this quantity reads
mLO
Nc mLO
O(g 3 T )
mD
D
D
ln
=
+
+
c
+
.
N
c
4
g32
g32
g32
g32
The perturbative leading order result is [5355]

Nc Nf 1/2
LO
+
gT = g32 y + O g 3 T ,
mD =
3
6
(4.1)
(4.2)
but at next-to-leading order g32 , only a logarithm can be extracted [56] (the 2nd term on
the right hand side of Eq. (4.1)), whereas the coefficient cNc is entirely non-perturbative.
To determine it, we use the gauge invariant definition of the Debye mass based on the
Euclidean time reflection symmetry as given in [37]. According to this definition, the
454
Table 3
The different contributions to the Debye mass, Eq. (4.1). The perturbative part means the 1st and 2nd
terms on the right-hand side of Eq. (4.1)
mD /g32
pert. part
c3
T = 2Tc
1.70(5)
0.514
1.65(6)
T 1011 Tc
3.82(12)
2.165
1.65(6)
O(g 3 T )/g32
0.46(6)
0.00(12)
mD /T
4.6(2)
0.96(3)
Debye mass corresponds to the mass of the lightest state odd under this transformation.
The remnant of Euclidean time reflection symmetry in our reduced model is the scalar
reflection symmetry R, and the Debye mass thus corresponds to the mass of the lightest
R = 1 state of our spectrum. As shown in Fig. 1, this is the 0+
ground state Tr A0 F12 ,
+
3
while 0 Tr A0 is slightly heavier.
With this definition, the coefficient cNc can be measured separately from the exponential
decay of a Wilson line in a 3d pure gauge theory [37]. The measurements for Nc = 2, 3
have been performed in [26] with the results c2 = 1.14(4), c3 = 1.65(6). On the other hand,
the mass we have measured here also includes the O(g 3 T ) correction in Eq. (4.1). We
can thus now estimate the magnitude of the remainder; our results are shown in Table 3.
We find that the O(g 3 T ) corrections are less than 30% even at temperatures as low as
T = 2Tc , and they disappear entirely for asymptotically large temperatures. The sign is
negative, so that the complete result is somewhat smaller than estimated in [26] based on
the perturbative contributions and c3 alone, 5T instead of 6T . The result is also about
20% smaller than what can be extracted from [5], a difference which we presume to be due
to the relatively small lattice sizes used there.
We can conclude that for temperatures of physical interest, the Debye mass according to
this definition is entirely non-perturbative, with its next-to-leading order correction being
larger than the leading term. This remains to be the case up to T 107Tc [26]. But even
at T 1011Tc , as the table shows, there are sizeable corrections to the leading behaviour.
This leads to the conclusion that the scale dominating the Debye mass is g 2 T with nonperturbative physics for all temperatures of interest, in contrast to the nave expectation
gT . The reason for this behaviour is the large non-perturbative coefficient cNc that
appears in front of the correction term in the series in g.
4.1.3. Nc scaling between SU(2) and SU(3)
Let us finally discuss the scaling with Nc . In [47] it was found for SU(Nc ) pure gauge
theories with Nc = 2, . . . , 5 that the differences in the mass spectra can be accounted for
by the leading order 1/Nc2 corrections in a large Nc expansion, and that the coefficients
of these are remarkably small. In theories with various scalar fields, the glueball content
has been found to be practically identical to that of pure gauge theories for Nc = 2, 3
[22,42,57], and thus the scaling behaviour with Nc is preserved. It is then natural to ask
if the same scaling behaviour holds for the scalar bound states, which are not present in
the YangMills theory. In Fig. 2 we plot some of our low lying states from the current
455
+ ++
Fig. 2. The scaling with Nc of some of the low lying 0++
+ (left) and 0 , 2+ (right) states. Filled
symbols denote glueballs, and the lines are to guide the eye only.
analysis as well as the SU(2) case [22] (we have to focus on states with C = +1 in order
to have an SU(2) counterpart). Indeed we observe a similar scaling for the scalar states as
for the glueballs. This suggests that the screening masses of hot SU(3) gauge theory are
close to the Nc limit, and hence large Nc methods may be useful approximations to
analytically deal with some of the non-perturbative aspects discussed here.
4.2. The mass spectrum for Nf 6= 0
Having convinced ourselves that dimensional reduction of pure YangMills theory
works at the very least for the lowest screening masses extracted from the static
bosonic correlation functions of the theory, we may now move ahead to exploit
the advantage of this formalism, the easy inclusion of fermions. For the case of
massless fermions, all that is required is to evaluate Eqs. (2.3)(2.5) for the desired
Nf and simulate the effective theory with the corresponding parameters. Each of these
simulations gives results completely analogous to those of the Nf = 0 case, except
for a slight shift in the values. For the reader interested in the detailed numbers,
we collect our data in Tables 6, 8 in Appendix A, at temperatures T = 2MS ,
and 1.5MS , which probes the lower limit of the range of validity of dimensional
reduction.
With the tables at hand, the spectrum shown in Fig. 1 for Nf = 0 is now also known
for Nf = 2, 3, 4. In Fig. 3 we display the dependence on Nf of the low lying J = 0
states. The general behaviour of slightly rising mass values with Nf can be understood
from Eqs. (4.1), (4.2). Increasing Nf increases the value of mLO
D corresponding to the
bare scalar mass, and hence the masses of the bound states increase as well. In units of
the temperature, the increase from Nf = 0 to the phenomenologically interesting case
Nf = 2 . . . 3 is about 30% for 0++
+ , . 20% for the other channels. The correlation lengths
decrease accordingly.
456
Fig. 3. The Nf dependence of the J = 0 spectrum at T = 2Tc (to be more precise, for Nf > 0 Tc
means really MS , see Section 2). Filled symbols denote glueball states. We should stress that only
the part M . 2T of the spectrum can be expected to directly represent the lowest states in 4d finite
temperature QCD.
4.3. The spatial string tension

Finally, let us discuss one observable of the 3d theory which does not have an
interpretation as a physical 4d correlation length. A Polyakov loop in a spatial direction
couples to a flux loop state (or torelon) that winds around the periodic boundaries of the
finite volume. The exponential fall-off of its correlator is related to the mass of the flux
loop, according to
X
(L)

(L) (x) ' eaMP (L)t , aMP (L) = a 2 L L.
(4.3)
Pj (x + t 3)P
j
x,j =1,2
Such a flux loop state can be easily identified through its energy scaling linearly with
the size of the lattice as seen in, e.g., Table 9. Since the scalar fields are in the adjoint
representation, they cannot screen the colour flux represented by the Polyakov loop in
fundamental representation, in contrast to the case with fundamental scalars [58]. Thus, a
string tension can be defined by the slope of the static potential at infinite separation, just
as in pure gauge theory. Accordingly, the coefficient L corresponds to the string tension
at separation L. For large enough L, an estimate for the string tension in infinite volume is
then provided by the relation [59]
.
(4.4)
a 2 = a 2 L +
6L2
We have extracted the string tension by diagonalising a basis of various smeared Polyakov
loops, and the results are given in Tables 9, 10.
We note that the string tension in continuum units is rather insensitive to the precise
parameter values x, y, as it fluctuates by less than 4% between the cases considered.
Moreover, all values are similarly close to the ones obtained in pure gauge theory [47].
457
This is a further indication of the insensitivity of the pure gauge sector to the presence of
the adjoint scalar fields. In the context of YangMills theory at finite temperature, the string
tension of the effective theory corresponds to the spatial string tension as measured in 4d
simulations at finite temperature. Indeed, our value /g32 = 0.569(10) on the finest
lattice for T = 2Tc , Nf = 0 is in good agreement with the spatial 4d result s /g 2 T =

0.586(5) given in [60].
5. Extension of the method to finite density

As we have mentioned in the Introduction, there is no satisfactory algorithm to simulate
lattice QCD in four dimensions at finite baryon density, with small quark masses. On the
other hand, in the experimental situation of heavy ion collisions there always is a net baryon
density, which in the collisions at and above AGS and SPS energies can be estimated
to correspond to /T . 4.0 [61]. As long as the temperature is sufficiently above Tc ,
dimensional reduction is applicable for the lowest lying screening masses extracted from
static correlation functions in QCD, as we have seen in the previous sections. We now
proceed to dimensionally reduce QCD at finite temperature and density. As we shall see,
it is possible to perform simulations in the /T -range of interest in this framework.
5.1. The effective theory with 6= 0
At leading order, the introduction of 6= 0 leads to a very simple modification of the
effective 3d theory in Eq. (2.1). New effects come only from fermions, where we change
pf pf i in the loop momenta; here pf denotes the fermionic Matsubara frequencies.
The 1-loop effective potential computed in [62] tells then that the effective action for the
Matsubara zero mode of A0 has for 6= 0 the terms (Nc = 2, 3)

Nf
Nf
Nc Nf
+
+ 2 2 Tr A20 + ig 3 2 Tr A30
VA0 = g 2 T 2
3
6
2
3

6
+
N
N
c
f
2
+ g4
(5.1)
Tr A20 .
24 2
There is of course no term linear in A0 , unlike in the Abelian case [63]. Note furthermore
that for SU(2), Tr A30 = 0.
In general, the inclusion of 6= 0 also leads to other new operators than Tr A30 . However,
as can be seen from the effective potential computed in [62], there are no higher order
operators involving only A0 , at least at 1-loop and 2-loop levels. On the contrary, there
could be operators such as Tr A0 Fij2 . As is usual in effective field theories, we expect
such non-renormalizable operators to give contributions suppressed with respect to the
dynamical effects arising within the effective theory in Eq. (5.1) by a power of the scale
hierarchy, and therefore we ignore them here. It is perhaps also worthwhile to mention
that since no 5 appears in perturbation theory in QCD and since the construction of the
effective theory in Eq. (5.1) is only sensitive to the ultraviolet and thus purely perturbative,
no ChernSimons type operators are expected to be generated.
458
Fig. 4. The function D() from Eq. (5.4), compared with the small- limit.
1/2 A3d ,
Going to 3d units (A4d
0 T
0
Z
Sz = d3 x g33 Tr A30 ,
R
d d3 x T 1 d3 x) and denoting
the dominant changes due to 6= 0 in the action in Eq. (2.1) are then

2 3N
Nf
f
;
y y 1+
.
S S + izSz , z =
T 3 2
T 2Nc + Nf
(5.2)
(5.3)
Thus, one new operator is generated in the effective action, and one of the parameters
which already existed, gets modified. We note that the new operator is quite special and
it, for instance, does not generate new ultraviolet divergences for the parameters in the
original action in Eq. (2.1).
A more precise computation of the effective action requires the inclusion of 2-loop
effects in Eq. (5.1), as well as a 1-loop computation of the wave function corrections of A0 ,
Ai and the gauge coupling g32 . We proceed as explained in [5], changing pf pf i in
the fermion propagators, and including the 2-loop contributions into the effective potential
from [62,64]. Let

Z
1
1
2P
D() = (4)
((pf + iT )2 + p2 )2 (pf2 + p2 )2
pf
Z
=

1
1
dp 1 ep
+
.
p 1 + ep ep+ + 1 ep+ + 1
At small , D() (7 (3)/2)2 . We then obtain

Nf
2
2
2
D
,
x +O x
g3 = g3 =0 1
6 + Nc Nf
T
(5.4)
(5.5)

Nf
D
x + O x2 ,
6 + Nc Nf
T

(8 Nf )(4 + Nf )
3Nf
2
1+
yNc =2 =
144 2 x
4 + Nf T

x = x =0 1
(5.6)
192 2Nf 7Nf2 2Nf3
96 2 (8 Nf )

3Nf (11 + 2Nf )(4 Nf ) 2
+ O(x),
1+
192 2Nf 7Nf2 2Nf3 T

3Nf
(9 Nf )(6 + Nf )
2
1+
yNc =3 =
6 + Nf T
144 2 x
+
459
(5.7)
486 33Nf 11Nf2 2Nf3

1+
96 2 (9 Nf )
3Nf (7 + Nf )(9 2Nf )
486 33Nf 11Nf2 2Nf3

Nf
3 7 + Nf
2
x
+
O
x
1
+
zNc =3 =
.
T 3 2
2 9 Nf
2
+ O(x),
(5.8)
(5.9)
In the case we actually study, Nc = 3, Nf = 2, these reduce to

2
x ,
1 D
7
T

2
x ,
x = x =0 1 D
7
T

15
3 2
7
+
1
+
,
y=
4 T
18 2 x 28 2

27
2
1+ x ,
z=
T 3
14
g32
= g32 =0
(5.10)
(5.11)
(5.12)
(5.13)
where g32 , x|=0 are from Eq. (2.9), and corrections are of relative order O(x 2 ). Let us
recall that x 0.1 at T 2Tc .
Due to the small value of x, we will in the analysis which follows for simplicity ignore
even the O(x) corrections in g32 , x, z in Eqs. (5.10), (5.11), (5.13), and use the leading
order expressions displayed already in Eq. (5.3). All of the sub-dominant effects neglected
work in the same direction and would strengthen the changes we see in the mass spectrum.
For z this is obvious because, according to Eq. (5.13), for a given /T the physical value
would be larger than what we have used. For x, Eq. (5.11) and the negative value of D
imply that the fact that we have kept x unchanged means that we have effectively moved
slightly up in T /MS . In g32 a major part of this effect cancels since the correction is the
same in Eqs. (5.10), (5.11). Therefore, M/T remains essentially the same for given M/g32 ,
even though the temperature is slightly higher.
460
5.2. A complex action and reweighting

The generic problem of SU(3) simulations at non-vanishing chemical potential now
appears as follows. For 6= 0, there is the imaginary term izSz in the action, and the
conventional importance sampling will not work. 4 A way to circumvent this is to do
the importance sampling with the action at z = 0, while exp(izSz ) is included in the
operators. We will refer to this procedure as reweighting. Reweighting is also applied
in the Glasgow method suggested for 6= 0 simulations in 4d (for a review, see [65]);
however, in our approach we do not need to carry out any expansion related to z. Denoting
by brackets the expectation value with respect to the action in Eq. (3.1), operators even O+
and odd O in R therefore appear as

O+ exp(izSz ) = O+ cos(zSz ) ,

(5.14)
O exp(izSz ) = i O sin(zSz ) .
Since Eq. (5.14) involves trigonometric functions, the problem manifests itself if
configurations with Sz 1 occur frequently. In this case there are cancellations in the
functional integral, and the Monte Carlo method will soon lose its accuracy.
Configurations with Sz 1 need not always be typical, however. Indeed, let us consider
the distribution of Sz in Eq. (5.2) for z = 0. The centre of the distribution is at zero. The
fact that there is Debye screening, y > 0, guarantees that the distribution is approximately
Gaussian. Its width is easily computed, and is given by the 2-loop scalar sunset graph on
the lattice, which has been evaluated in [38,39]:

1/2
z 6L 3/2
.
(5.15)
5 ln 1/2 + 0.58849
1(zSz )
4
3y
We note that the width grows with the lattice size L (or, more precisely, with the physical
lattice size L/) 5 . Nevertheless, if z is small enough, the infinite volume and continuum
scaling limits of the correlation functions may be extracted before Eq. (5.15) grows to
values 1. In these cases we can in practice carry out all the measurements before
the oscillations set in. The feasibility of this procedure can be checked by monitoring
the numerical distribution of zSz , the argument of the reweighting factors. Accurate
calculations should be possible whenever the bulk of the distribution is contained in the
interval (, ). Once the distribution gets significantly broader, the breakdown should
show up in the correlation functions simply as a noisy signal.
Even assuming that the oscillations are under control, reweighting according to
Eq. (5.14) of course requires some care. The importance sampling of the Monte Carlo
is dominated by the minimum of the action at z = 0, which is unmodified by including
Sz in the operators. If the z = 0 minimum is widely separated from the true z 6= 0
minimum in configuration space, there is a danger of biasing the Monte Carlo towards
the wrong configurations. In order to check this, we have tested the reweighting method by
4 It can be seen directly in 4d that the sign problem is strongly correlated with the imaginary part of the temporal
Polyakov loop [33,34]. This corresponds precisely to zSz in the 3d language.
5 For a fixed physical lattice size, the width grows with , but only logarithmically.
461
considering z imaginary, z i . In this case the action is real and the scalar update can
be modified by a Metropolis step to include Sz , and hence update with the exact action.
The results can then be compared with those obtained from Eq. (5.14), now with
cos(zSz ) cosh( Sz ),
i sin(zSz ) sinh( Sz ).
(5.16)
Simultaneously, we change the sign of 2 in y in Eq. (5.3).

Finally, as a small technical detail let us note that in practice we implement Sz
in Eq. (5.2) on the lattice by including an O(a) improved [40] version of Tr A30 , as
in Eq. (2.10) of [15].
5.3. Determining masses with a complex action
Assuming that our reweighting procedure works, we then have to extract masses in the
changed symmetry situation. After the inclusion of 6= 0, the action of the effective theory
is no longer real, and no longer invariant in R, C. It is however still invariant under RC and
+
P . In the scalar channel this means that the operators 0++
+ and 0 in Eq. (2.13) can couple
+
to each other, and similarly 0+ and 0 . These two channels are still distinguished,
however, by the parity P .
We measure the correlation matrix between the two channels which used to be decoupled
for = 0. From Eq. (5.14) we know that the result is of the block form

A(t) izD(t)
C(t) =
,
(5.17)
izD(t) B(t)
where A(t), B(t) are symmetric matrices (possibly of different sizes), representing the
correlations within the two channels as in Eq. (3.6). We have factored out z in the offdiagonal blocks, to make it clear that the result is odd in z and vanishes for z 0. In order
to facilitate the solution of the eigenvalue problem in Eq. (3.7), we may note that we can
equivalently consider the eigenvalue problem for the real correlation matrix

A(t) zD(t)
e
C(t) =
,
(5.18)
zD(t) B(t)
obtained from C(t) by a similarity transformation Z = diag (1, i). Finally, for imaginary
chemical potentials z = i we have a correlation matrix of the usual symmetric form as
in Eq. (3.6),

A (t) D (t)
.
(5.19)
C (t) =
D (t) B (t)
In addition, let us note that physical observables such as masses (i.e., inverse correlation
lengths) must be even in z, since a change of its sign can be compensated for by the field
redefinition A0 A0 in the action in Eq. (5.3). Since the mass spectrum is well defined
and there are no massless modes at z = 0, we moreover expect that the system is analytic
in z for small |z|, the expressions of the masses starting with z2 . In particular, masses must
remain real for small enough z (even though for the eigenvalues of a fully general matrix
of the form in Eqs. (5.17), (5.18) this is not always the case).
462
We can use these observations to discuss the form of the mass eigenstates. For the cases
in Eqs. (5.17), (5.18), we do not in general have orthonormal eigenvectors in the usual
sense. In order for the procedure to make physical sense, we however expect (and observe)
that the eigenvectors are independent of t within statistical accuracy, with eigenvalues of
the form exp(aMt), as in Eq. (3.7).
Consider now first Eq. (5.19). Since the eigenvalues are even in , the eigenvectors
should be of the forms

vA
,
v 2A + 2 v 2B = 1,
v (1) =
vB

v A
(2)
(5.20)
,
2 v 2A + v 2B = 1,
v =
vB
where normalisation conditions have also been shown. With analytic continuation, we can
then directly apply this to the case of Eq. (5.17),

vA
(1)
,
v 2A z2 v 2B = 1,
v =
izv B

izv A
(2)
(5.21)
,
z2 v 2A + v 2B = 1,
v =
vB
as well as to the case of Eq. (5.18), using the similarity transformation:

vA
(1)
,
v 2A z2 v 2B = 1,
v =
zv B

zv A
(2)
,
z2 v 2A + v 2B = 1.
v =
vB
(5.22)
We observe that the metric has changed, but otherwise the situation is quite analogous to
the usual one in Eq. (5.20).
Fortunately, in practice even less needs to be changed in the usual numerical analysis
based on a matrix of the form in Eqs. (3.6), (5.19), if we use Eq. (5.18). There are algorithms finding the eigenvalues and eigenvectors for a general non-symmetric real matrix.
The only change is in the normalisation according to Eq. (5.22), which amounts to fixing
an overall coefficient for each eigenvector. However, as we only consider the correlations
between the eigenvectors, cf. Eq. (3.8), and do not need to verify orthogonality explicitly,
we can equally well employ a wrong scalar product and normalisation based on the old
type of metric in Eq. (5.20) (with z). Then everything goes precisely as before.
Finally, let us illustrate the general expectations for the mass pattern with a simple 2 2
matrix as an analogue of Eq. (5.17). Consider two real scalar fields , , with a mass term
1
1
V = m2 2 + M 2 2 + i.
2
2
(5.23)
For M m and < (M 2 m2 )/2, the mass eigenvalues are real, m2 + (/M)2 ,
M 2 (/M)2 . Thus the lightest mass goes up, which is what we expect for a real chemical
potential. In fact there is another effect contributing in the same direction, since according
463
to Eq. (5.3) the parameter y grows with /T . In the case of a complex = i, on

the contrary, the lightest mass becomes even lighter, the heaviest mass even heavier:
m2 (/M)2 , M 2 + (/M)2 . Again, the change of y contributes in the same direction
for the lightest mass.
6. Numerical results for 6= 0
We now proceed to present the results of our first numerical explorations of 6= 0.
We restrict our attention to the case T = 2Tc , Nf = 2, corresponding to x = 0.0919. As we
have explicitly checked for good scaling behaviour at = 0, we work exclusively with
= 21 in this section. The parameters and lattices considered are summarised in Table 4,
and the detailed results are displayed in Tables 11, 12 in Appendix A.
6.1. Consistency checks and the general pattern
We begin by assessing the distribution of the argument of the reweighting factors in
Eq. (5.14). We find that any typical histogram can be easily fitted to a Gaussian, from
which we determine its width. The plots in Fig. 5 show the corresponding distributions
as a function of /T and the lattice volume, with behaviour as expected from Eq. (5.15).
The volume chosen for the /T series is large enough to be free from finite size effects in
the = 0 simulations. As the histograms show, for these volumes the distribution of the
argument of the reweighting factor is entirely contained within one period for /T 6 2.0.
On the other hand, at /T = 4.0 the tails of the distribution are significantly spreading.
Nevertheless, the bulk of the distribution is still within the region without sign flips. For the
same number of measurements at /T = 2.0, 4.0, however, the errors on the correlation
functions are indeed by a factor of three larger in the latter case, confirming the onset of the
Table 4
The lattice parameters and sizes used for calculations with 6= 0.
All are for T = 2MS , Nf = 2, x = 0.0919, = 21. The
parameter z is from Eq. (5.3)
/T
L2 T
0.5 i
0.47382
0.0338 i
203
1.0 i
0.44630
0.0675 i
203
1.5 i
0.40042
0.1013 i
303
0.5
0.49218
0.0338
303
1.0
0.51970
0.0675
303
1.5
0.56558
0.1013
303
2.0
0.62981
0.1351
303 , 403
4.0
1.07026
0.2702
103 , 143 , 183 , 303
464
Fig. 5. Distribution of zSz [A0 ] for the reweighting procedure (Eq. (5.14)).
sign problem. Going to still larger /T will thus lead to increasingly noisy signals, until a
mass determination becomes impossible.
In the next step, we test the reliability of the reweighting procedure by comparing with
the correct importance sampling in the case of an imaginary chemical potential. This
comparison can only be carried out for a limited range of /T . In the full action, the
imaginary chemical potential term weakens the metastability [5] of the symmetric phase,
and tunnelling during typical simulations to the unphysical phase becomes possible. This
can be countered by increasing the lattice volume, but in so doing we increase the width
of the distribution of zSz [A0 ]. We find however that there exists a window of opportunity
between these two limits where we can usefully compare the two algorithms. In the left
panels of Fig. 6, the ground states of the 0+ and 0 channels are shown as a function of
imaginary /T . First, we note the correct qualitative behaviour of a mass decrease with
|/T |, as described after Eq. (5.23). Second, we observe complete agreement within the
small statistical errors between the reweighted observables and those measured with the
full action.
With these tests passed, we can thus display the same states for the case of real chemical
potential in the right panels of Fig. 6. Again in agreement with the qualitative behaviour
expected from the discussion after Eq. (5.23), the lowest mass values are growing with
increasing /T , since the constituent mass parameter y 1/2 gets larger, see Eq. (5.3).
Thus the corresponding correlation lengths decrease. We note again, however, that purely
2 ) are rather insensitive
gluonic states (or, in the 4d language, purely magnetic states Tr F12
to /T , in analogy with their behaviour under variations of y discussed earlier. This means
that at large enough /T > 4.0, they may become the lightest states in the system, as
suggested by the top right panel of Fig. 6.
At /T = 2.0 we have performed a finite volume check, comparing lattices of L =
30 and 40 at = 21. We find the low lying masses to be consistent within statistical
errors, see Table 12. This is encouraging, implying that the reweighting does not magnify
finite volume effects unduly, and that it is indeed possible to extrapolate screening
masses to the infinite volume in a finite chemical potential context, before the onset of
oscillations.
465
Fig. 6. The dependence of the lowest thermal masses on /T for imaginary (left) and real (right)
chemical potential. For clarity of presentation, we show the lightest state only for each quantum
number assignment. The quantum number assignments refer to the eigenstates at /T = 0. Filled
symbols correspond to glueballs. Exact and reweighted simulations (left) are explained in
Section 5.2.
6.2. The Debye screening length at finite baryon density

Let us now consider what happens to the Debye screening length as defined in
Section 4.1.2. This definition is based on the R-symmetry of our effective theory [37].
For 6= 0 the extra term in the action spoils the symmetry, resulting in a merging of the
various RC-channels, while only P remains as a good quantum number. The question then
arises, how can the Debye mass be defined in such a situation [37]?
We now note the following. For = 0, the lightest R = 1 state was 0+
Tr A0 F12 .
This is, however, also the lightest state in the channel P = 1. Thus, even after 6= 0,
the same operator continues to determine the long distance decay of a number of R = 1
operators. The values can be seen in Fig. 6, bottom right. If we define this largest P = 1
correlation length to correspond to Debye screening, then 6= 0 does not change the
situation in an essential way, and the corresponding correlation length decreases with .
An example of a 4d gauge invariant operator with such a behaviour is Tr F03 F12 [37].
466
++
3
2
On the other hand, the 0+
operator Tr A0 does couple to 0+ Tr A0 after the
inclusion of 6= 0. Therefore, its correlation length strictly speaking increases by a factor
of almost two as soon as 6= 0, as can be seen from the top right panel in Fig. 6. There
we have labeled the states with the quantum numbers of the /T = 0 limit; however, for
3
/T > 0, any operator which originally had just the quantum numbers 0+
such as Tr A0 ,
++
couples now also to 0+ . The eigenstate shown in Fig. 6 which does have a larger mass, is
a particular linear combination of the two types of original states. Nevertheless, the overlap
between the two operators Tr A20 , Tr A30 is very small for small /T , so that even for
Tr A30 there is an intermediate distance range where correlations should decay as shown
with 0+
in Fig. 6, as discussed in [37]. Thus no abrupt change is observed. An example
of a gauge invariant 4d operator with such a behaviour is the imaginary part of the 4d
temporal Polyakov loop.
7. Conclusions
In this paper, we have carried out mass measurements in the different quantum number
channels within the 3d SU(3) + adjoint Higgs theory. We have stayed on the dimensional
reduction curve in the symmetry restored phase of the 3d theory, and the results are
thus expected to correspond to spatial correlation lengths in the deconfined quarkgluon
plasma phase of QCD. This interpretation appears to be relatively accurate at least down to
T 2Tc . We first considered a vanishing chemical potential = 0, and then also extended
the measurements to a finite /T . 4.0.
For = 0, Nf = 0, we believe that the asymptotic correlation lengths related to the
real and imaginary parts of the 4d temporal Polyakov loop, as well as to other gauge
invariant bosonic operators, are now relatively well understood. For instance, at T 2Tc
2MS the real part of the Polyakov loop (with quantum numbers 0++
+ according to
our conventions discussed in Section 2.1) decays exponentially at (3T )1 , while the
1
imaginary part (0+
) decays at (5T ) , as can be observed from Fig. 1. The Debye
screening length according to the definition of [37] turns out to be determined by operators
of the type 0+
Tr F03 F12 , and is slightly longer than the screening length of the
imaginary part of the Polyakov loop, (4.6T )1 .
We have also explicitly studied the effects of Nf dynamical fermions on the longest
correlation lengths. For fixed T /MS , the correlation lengths decrease as Nf is increased.
For instance, in the phenomenologically interesting case of Nf = 2, 3 at T 2MS , the
real part of the Polyakov loop decays exponentially at (4T )1 , while the imaginary part
at (6T )1 . As a theoretical point, it is interesting to note that in units of g32 , the screening
lengths scale quite well with Nc .
We have then extended the measurements to 6= 0. We have demonstrated that
the use of dimensional reduction allows one to carry out simulations corresponding to
phenomenologically interesting values of quark masses, temperatures, and /T (although
not at the point of the phase transition). Simulations are possible because the only term in
467
the super-renormalizable 3d action suffering from the sign problem, i Tr A30 , comes with
a very small numerical coefficient.
In general, we observe that as /T is switched on at a fixed temperature, the screening
lengths decrease. This can be interpreted so that the lightest excitations, involving A0 ,
are less critical and correspondingly heavier. Thus, staying at a fixed temperature but
increasing , we are apparently moving further away from the critical line, which hence
has to bend down. This is in complete accordance with what we qualitatively know about
the phase diagram in the (, T )-plane.
2 . Thus
However, there is one state whose mass does not increase, the 3d glueball Tr F12
at large enough /T , similarly to the case of large enough T /MS at = 0, it becomes
the lightest excitation in the system. We find that the crossover corresponds roughly to
y 1.1 in terms of the dimensionless parameter defined in Eqs. (2.2), (5.3). For Nf = 2,
this corresponds roughly to T /MS 102 at /T 0.0, and /T 4.0 at T /MS 2.
Finally we have addressed the behaviour of the imaginary part of the 4d temporal
Polyakov loop. As the chemical potential is switched on, its asymptotic correlation length
suddenly increases contrary to the other observables discussed, and agrees for all 6= 0
with that of the real part of the 4d temporal Polyakov loop.
The task remains to establish a direct connection between the screening masses
discussed here and observables relevant for the phenomenology of heavy ion collision
experiments.
Note added
After the submission of our paper, a letter appeared by Gavai and Gupta [66], who
measure the spatial meson (pion) correlation length at T (1.5 . . . 2.9)Tc for Nf = 4
light dynamical fermions. They observe that at T 2Tc the meson mass measured
has become lighter than the lightest bosonic mass (represented by Tr A2
from
0
in the 3d language), so that the bosonic gauge degrees of freedom are no longer the
lightest dynamical degrees of freedom, and thus the standard dimensionally reduced
effective theory should not be reliable any more. This observation is in accordance with
our comments in Section 4.1.1. However, Gavai and Gupta make the further suggestion
that even in the temperature range T (2 . . . 10)Tc where the meson mass is no longer
the lightest one, but is still below its asymptotic value ( 2T in the case of the chiral
continuum limit), dimensional reduction may not be reliable, in contrast to the situation in
the purely bosonic case Nf = 0. Let us comment that there is no hard evidence for the latter
suggestion. In fact, it seems quite possible to us that the decrease from 2T could be
partly accounted for precisely by the confining bosonic 3d gauge field dynamics related to
A0 , Ai . Furthermore, we see no evidence for why the bosonic correlation lengths we have
measured here would not be satisfactorily reproduced by the 3d theory in this temperature
range. Thus we believe that our bosonic mass measurement for Nf > 0 (both at /T = 0
and /T > 0) do reproduce at least the qualitative features of the physical 4d theory at all
temperatures T & 2Tc .
468
As already discussed by Gavai and Gupta, one might also expect that chiral effects
responsible for the behaviour seen by them are at Nf = 4 somewhat larger than in the
realistic case Nf = 2 . . . 3. For instance, Tr A20 is lighter for smaller Nf as we have
still reaches the same asymptotic value, so that the level
discussed in this paper, while
crossing invalidating dimensional reduction should take place at a smaller temperature.
Acknowledgements
We acknowledge useful discussions with R.D. Pisarski and K. Rajagopal. This work was
partly supported by the TMR network Finite Temperature Phase Transitions in Particle
Physics, EU contract no. FMRX-CT97-0122. The work of A.H. was supported in part by
UK PPARC grant PPA/G/0/1998/00621.
Appendix A. Tables
For completeness, we collect in this appendix the numerical results for the correlation
length measurements we have carried out.
Table 5
++
Mass estimates and dominant operator contributions in the 0++
+ and 2+ channels for Nf = 0. The
dominant operator types contributing are denoted with (without) parentheses if hi k i < (>)0.5
T = 2Tc
T 1011 Tc
= 21
L = 38
0++
+
L = 28
M/g32
Ops.
M/g32
Ops.
1.022(15)
R, L
1.022(11)
2.44(14)
2.51(11)
= 28
= 28
L = 40
L = 40
M/g32
Ops.
M/g32
R, L
0.994(19)
R, L
2.55(13)
2.511(66)
2.86(14)
Ops.
2.92(14)
(R, L)
2.62(11)
2.65(8)
3.75(21)
3.70(18)
2.95(14)
(L)
2.95(16)
(R, L)
4.28(28)
(P )
3.79(18)
3.66(21)
3.61(32)
4.89(27)
R, L
4.62(28)
(R, L)
4.14(25)
(T )
3.66(36)
(T )
M/g32
Ops.
M/g32
Ops.
M/g32
Ops.
M/g32
Ops.
3.60(18)
2.63(7)
2.82(13)
2.94(17)
3.55(19)
R, L
3.44(16)
R, L
3.355(94)
R, L
4.26(19)
4.23(25)
3.92(21)
4.14(28)
4.35(28)
4.94(53)
4.06(28)
3.94(33)
5.05(44)
2++
+
469
Table 6
++
Mass estimates and dominant operator contributions in the 0++
+ and 2+ channels for various
Nf > 0 on L = 30, 40 for = 21, 28, respectively. The dominant operator types contributing are
shown in the superscripts
T = 1.5MS
Nf = 2
T = 2.0MS
Nf = 3
Nf = 2
= 21
0++
+
(R,L)
(C)
(T )
M/g32
Nf = 3
Nf = 4
Nf = 2
= 21
M/g32
M/g32
M/g32
= 28
M/g32
M/g32
1.183(22)
1.316(18)
1.351(18)
1.477(18)
1.557(22)
1.335(19)
2.51(5)
2.16(18)
2.576(50)
2.51(13)
2.59(11)
2.55(7)
2.64(9)
2.86(19)
2.60(13)
2.90(18)
2.83(18)
2.74(9)
3.14(13)
3.26(18)
3.29(11)
3.35(21)
3.44(18)
3.25(12)
(C)
5
(T )
6
3.68(21)
3.83(24)
3.51(33)
3.86(25)
3.76(28)
4.01(35)
3.84(32)
4.11(25)
4.25(32)
4.05(33)
2++
+
M/g32
M/g32
M/g32
M/g32
M/g32
M/g32
(R,L)
(T )
2.69(12)
2.88(11)
2.84(18)
2.98(12)
2.90(14)
2.74(7)
3.56(14)
3.73(15)
3.69(21)
3.57(28)
3.77(19)
3.59(12)
(C)
3
(T )
4
4.27(18)
4.08(28)
3.76(23)
4.27(33)
4.16(32)
4.14(24)
3.83(28)
4.40(33)
4.06(28)
3.87(24)
(R,L)
Table 7
++
Masses M/g32 for channels other than 0++
+ , 2+ at Nf = 0. The stars in the
superscripts denote excited states
T = 2Tc
= 21
L = 38
0
+
L = 28
T 1011 Tc
= 28
= 28
L = 40
L = 40
2.57 (11)
2.53 (9)
2.46 (10)
3.83 (19)
0
+
3.71 (28)
3.91 (21)
3.50 (25)
4.48 (28)
0+
1.799 (35)
1.771 (32)
1.699 (43)
3.82 (12)
0+
3.28 (18)
3.25 (18)
3.27 (18)
5.07 (28)
0+
4.96 (28)
1.85 (7)
1.90 (5)
1.85 (5)
0+
3.06 (33)
3.06 (18)
3.07 (12)
1+
1
3.47 (28)
2.76 (28)
3.49 (30)
2.99 (21)
3.33 (28)
2.98 (14)
4.42 (28)
470
Table 8
++
Masses M/g32 for channels other than 0++
+ , 2+ , for various Nf > 0, on L = 30, 40 with
= 21, 28, respectively. The stars in the superscripts denote excited states
T = 1.5MS
Nf = 2
T = 2.0MS
Nf = 3
Nf = 2
= 21
0
+
0
+
0+
0+
0+
0+
1+
1
Nf = 3
Nf = 4
Nf = 2
= 21
= 28
2.83 (7)
2.96 (7)
2.74 (14)
2.84 (21)
3.16 (25)
2.83 (7)
3.71 (21)
3.94 (19)
3.70 (18)
3.58 (28)
3.84 (25)
3.64 (14)
1.932 (32)
1.977 (43)
1.960 (43)
2.079 (35)
2.103 (67)
1.932 (43)
3.46 (11)
3.52 (21)
3.50 (21)
3.56 (11)
3.66 (16)
3.56 (14)
2.19 (7)
2.27 (11)
2.391 (35)
2.604 (70)
2.677 (70)
3.31 (24)
3.29 (19)
3.31 (15)
3.53 (21)
3.60 (11)
3.29 (10)
3.59 (25)
3.10 (18)
3.92 (20)
3.17 (25)
4.06 (16)
2.91 (16)
4.11 (32)
3.28 (28)
4.08 (25)
3.21 (18)
3.73 (15)
3.08 (12)
Table 9
Spatial Polyakov loop masses and string tensions for Nf = 0
T = 2Tc
= 21
aMP (L)
/g32
T 1011 Tc
= 28
= 28
L = 40
L = 38
L = 28
L = 40
1.042 (40)
0.749 (20)
0.585 (17)
0.625 (20)
0.167 (4)
0.166 (3)
0.122 (2)
0.126 (2)
0.585 (15)
0.580 (11)
0.569 (10)
0.588 (10)
Table 10
Spatial Polyakov loop masses and string tensions corresponding to 3d parameters with Nf > 0, on
L = 30, 40 for = 21, 28, respectively
T = 1.5MS
Nf = 2
Nf = 3
T = 2.0MS
Nf = 2
= 21
aMP (L)
/g32
Nf = 3
Nf = 4
= 21
Nf = 2
= 28
0.760(21)
0.818(15)
0.798(19)
0.847(35)
0.820(19)
0.588(11)
0.161(3)
0.167(2)
0.165(2)
0.170(4)
0.167(2)
0.123(2)
0.563(11)
0.584(8)
0.577(8)
0.595(15)
0.584(8)
0.574(10)
471
Table 11
Mass estimates and dominant operator contributions in the 0+ and 0 channels for imaginary 6= 0
from reweighted and exact actions. The dominant operator types contributing are shown in the
superscripts. Some states on the lattice size L = 30 (used for /T = 1.5 i) have been reordered
for clarity of presentation
/T = 0.5 i
/T = 1.5 i
L = 20
rewtd.
L = 20
rewtd.
L = 20
exact
L = 30
rewtd.
L = 30
exact
M/g32
M/g32
M/g32
M/g32
M/g32
0.85 (3)
0+
(R2 ,L)
/T = 1.0 i
1.34 (3)
1.20 (2)
1.18 (2)
0.89 (3)
(P )
2
(R )
3 3
(C)
4
1.94 (6)
1.89 (6)
1.73 (11)
2.54 (14)
2.96 (21)
2.21 (9)
2.22 (7)
2.21 (11)
1.81 (6)
1.81 (6)
2.54 (18)
2.53 (9)
2.50 (13)
2.45 (11)
2.45 (18)
M/g32
M/g32
M/g32
M/g32
M/g32
2.04 (3)
1.87 (4)
1.86 (3)
1.76 (3)
1.77 (7)
2.91 (15)
2.88 (9)
2.81 (13)
2.46 (7)
2.42 (9)
3.15 (25)
3.32 (18)
3.45 (20)
3.29 (21)
3.23 (25)
3.80 (35)
3.75 (25)
3.63 (28)
3.50 (27)
3.70 (28)
0
(B1 )
(B )
2 2
(B )
3 1
(C)
4
Table 12
Mass estimates and dominant operator contributions in the 0+ and 0 channels for real 6= 0. The
dominant operator types contributing are denoted with (without) parentheses if |hi k i| < (>)0.5
/T = 0.5
/T = 1.0
/T = 1.5
L = 30
L = 30
L = 30
0+
M/g32
Ops.
M/g32
Ops.
M/g32
Ops.
1.37(4)
R2 , L
1.48(2)
R2 , L
1.59(2)
R2 , L
2.29(8)
C, (R3 )
2.61(3)
C, (R3 )
2.50(5)
2.51(9)
R3 , (C)
2.61(4)
R3 , (C)
2.78(5)
R3
2.64(17)
2.84(8)
2.82(7)
M/g32
Ops.
M/g32
Ops.
M/g32
Ops.
2.01(2)
B1
2.02(2)
B1
2.14(3)
B1
2.84(12)
B2
3.12(6)
B2
3.18(7)
B2
3.49(18)
B1
3.56(7)
B1
3.67(11)
B1
3.74(21)
3.68(11)
3.73(12)
472
Table 12 (continued)
/T = 2.0
/T = 4.0
L = 30
L = 40
0+
M/g32
Ops.
1.73(3)
2.45(8)
L = 30
M/g32
Ops.
M/g32
Ops.
R2 , L
1.80(11)
R2 , L
2.58(16)
R2 , L
2.43(21)
2.69(18)
3.01(14)
R3
2.71(21)
R3
3.02(14)
M/g32
Ops.
M/g32
Ops.
M/g32
2.25(5)
B1
2.19(7)
B1
2.43(20)
B1
3.45(8)
B2
3.37(25)
B2
4.17(22)
B2 , C
3.57(11)
3.78(35)
3.70(16)
B1
3.85(36)
B1
Ops.
References
[1] A.D. Linde, Phys. Lett. B 96 (1980) 289.
[2] D.J. Gross, R.D. Pisarski, L.G. Yaffe, Rev. Mod. Phys. 53 (1981) 43.
[3] P. Arnold, C. Zhai, Phys. Rev. D 50 (1994) 7603, hep-ph/9408276; Phys. Rev. D 51 (1995)
1906, hep-ph/9410360.
[4] E. Braaten, A. Nieto, Phys. Rev. Lett. 76 (1996) 1417, hep-ph/9508406; Phys. Rev. D 53 (1996)
3421, hep-ph/9510408.
[5] K. Kajantie et al., Nucl. Phys. B 503 (1997) 357, hep-ph/9704416; Phys. Rev. Lett. 79 (1997)
3130, hep-ph/9708207.
[6] F. Karsch, Plenary talk at Lattice 99, hep-lat/9909006, to appear in the Proceedings.
[7] P. Ginsparg, Nucl. Phys. B 170 (1980) 388.
[8] T. Appelquist, R.D. Pisarski, Phys. Rev. D 23 (1981) 2305.
[9] S. Nadkarni, Phys. Rev. Lett. 60 (1988) 491.
[10] T. Reisz, Z. Phys. C 53 (1992) 169.
[11] L. Krkkinen, P. Lacock, D.E. Miller, B. Petersson, T. Reisz, Phys. Lett. B 282 (1992) 121;
Nucl. Phys. B 418 (1994) 3, hep-lat/9310014.
[12] L. Krkkinen, P. Lacock, B. Petersson, T. Reisz, Nucl. Phys. B 395 (1993) 733.
[13] S. Huang, M. Lissia, Nucl. Phys. B 438 (1995) 54, hep-ph/9411293; Nucl. Phys. B 480 (1996)
623, hep-ph/9511383.
[14] K. Kajantie, M. Laine, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 458 (1996) 90, hepph/9508379; Phys. Lett. B 423 (1998) 137, hep-ph/9710538.
[15] K. Kajantie, M. Laine, A. Rajantie, K. Rummukainen, M. Tsypin, JHEP 9811 (1998) 011, heplat/9811004.
[16] K. Kajantie et al., Nucl. Phys. B 466 (1996) 189, hep-lat/9510020; Nucl. Phys. B 493 (1997)
413, hep-lat/9612006.
[17] M. Grtler et al., Nucl. Phys. B 483 (1997) 383, hep-lat/9605042; Phys. Rev. D 56 (1997) 3888,
hep-lat/9704013.
[18] O. Philipsen et al., Nucl. Phys. B 528 (1998) 379, hep-lat/9709145.
[19] K. Rummukainen et al., Nucl. Phys. B 532 (1998) 283, hep-lat/9805013.
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
473
F. Csikor, Z. Fodor, J. Heitger, Phys. Rev. Lett. 82 (1999) 21, hep-ph/9809291.

M. Laine, JHEP 06 (1999) 020, hep-ph/9903513.
A. Hart, O. Philipsen, Nucl. Phys. B 572 (2000) 243, hep-lat/9908041.
S. Datta, S. Gupta, Nucl. Phys. B 534 (1998) 392, hep-lat/9806034; Phys. Lett. B 471 (2000)
382, hep-lat/9906023.
F. Karsch, M. Oevers, P. Petreczky, Phys. Lett. B 442 (1998) 291, hep-lat/9807035.
P. Bialas, A. Morel, B. Petersson, K. Petrov, T. Reisz, hep-lat/0003004.
M. Laine, O. Philipsen, Nucl. Phys. B 523 (1998) 267, hep-lat/9711022; Phys. Lett. B 459
(1999) 259, hep-lat/9905004.
M. Alford, Nucl. Phys. B (Proc. Suppl.) 73 (1999) 161, hep-lat/9809166.
M. Alford, A. Kapustin, F. Wilczek, Phys. Rev. D 59 (1999) 054502, hep-lat/9807039.
M. Lombardo, hep-lat/9908006.
T.C. Blum, J.E. Hetrick, D. Toussaint, Phys. Rev. Lett. 76 (1996) 1019, hep-lat/9509002.
R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, hep-lat/9903004.
J. Engels, O. Kaczmarek, F. Karsch, E. Laermann, Nucl. Phys. B 558 (1999) 307, heplat/9903030.
P. de Forcrand, V. Laliena, Phys. Rev. D 61 (2000) 034502, hep-lat/9907004.
K. Langfeld, G. Shin, hep-lat/9907006.
J. Fingberg, U. Heller, F. Karsch, Nucl. Phys. B 392 (1993) 493.
B. Beinlich, F. Karsch, E. Laermann, A. Peikert, Eur. Phys. J. C 6 (1999) 133, hep-lat/9707023.
P. Arnold, L.G. Yaffe, Phys. Rev. D 52 (1995) 7208, hep-ph/9508280.
K. Farakos, K. Kajantie, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 442 (1995) 317,
hep-lat/9412091.
M. Laine, A. Rajantie, Nucl. Phys. B 513 (1998) 471, hep-lat/9705003.
G.D. Moore, Nucl. Phys. B 493 (1997) 439, hep-lat/9610013; Nucl. Phys. B 523 (1998) 569,
hep-lat/9709053.
C. Michael, J. Phys. G 13 (1987) 1001.
O. Philipsen, M. Teper, H. Wittig, Nucl. Phys. B 469 (1996) 445, hep-lat/9602006; Nucl. Phys.
B 528 (1998) 379, hep-lat/9709145.
G.C. Fox, R. Gupta, O. Martin, S. Otto, Nucl. Phys. B 205 (1982) 188.
B. Berg, A. Billoire, Nucl. Phys. B 221 (1983) 109.
M. Lscher, U. Wolff, Nucl. Phys. B 339 (1990) 222.
A.S. Kronfeld, Nucl. Phys. B (Proc. Suppl.) 17 (1990) 313.
M. Teper, Phys. Rev. D 59 (1999) 014512, hep-lat/9804008.
B. Bunk, Nucl. Phys. B (Proc. Suppl.) 42 (1995) 566.
S. Nadkarni, Phys. Rev. D 33 (1986) 3738.
E. Braaten, A. Nieto, Phys. Rev. Lett. 74 (1995) 3530, hep-ph/9410218.
W. Buchmller, O. Philipsen, Phys. Lett. B 397 (1997) 112, hep-ph/9612286.
O. Kaczmarek, F. Karsch, E. Laermann, M. Ltgemeier, hep-lat/9908010.
E.V. Shuryak, Zh. Eksp. Teor. Fiz. 74 (1978) 408, Sov. Phys. JETP 47 (1978) 212.
J.I. Kapusta, Nucl. Phys. B 148 (1979) 461.
D.J. Gross, R.D. Pisarski, L.G. Yaffe, Rev. Mod. Phys. 53 (1981) 43.
A.K. Rebhan, Phys. Rev. D 48 (1993) 3967, hep-ph/9308232; Nucl. Phys. B 430 (1994) 319,
hep-ph/9408262.
E.M. Ilgenfritz, A. Schiller, C. Strecha, Eur. Phys. J. C 8 (1999) 135, hep-lat/9807023.
O. Philipsen, H. Wittig, Phys. Rev. Lett. 81 (1998) 4056, hep-lat/9807020; Phys. Rev. Lett. 83
(1999) 2684 (Erratum).
P. de Forcrand, G. Schierholz, H. Schneider, M. Teper, Phys. Lett. B 160 (1985) 137.
F. Karsch, E. Laermann, M. Ltgemeier, Phys. Lett. B 346 (1995) 94, hep-lat/9411020.
J. Cleymans, K. Redlich, Phys. Rev. C 60 (1999) 054908, nucl-th/9903063, and references
therein.
474
[62] C.P. Korthals Altes, R.D. Pisarski, A. Sinkovics, Phys. Rev. D 61 (2000) 056007, hepph/9904305.
[63] S.Yu. Khlebnikov, M.E. Shaposhnikov, Phys. Lett. B 387 (1996) 817, hep-ph/9607386.
[64] C.P. Korthals Altes, Nucl. Phys. B 420 (1994) 637, hep-th/9310195.
[65] I.M. Barbour, S.E. Morrison, E.G. Klepfish, J.B. Kogut, M. Lombardo, Nucl. Phys. A (Proc.
Suppl.) 60 (1998) 220, hep-lat/9705042.
[66] R.V. Gavai, S. Gupta, hep-lat/0004011.

Suppressing monopoles and vortices:

a possibly smoother approach to scaling?
Rajiv V. Gavai
Department of Theoretical Physics, Tata Institute of Fundamental Research, Homi Bhabha Road,
Mumbai 400005, India
Abstract
Suppressing monopoles and vortices by introducing large chemical potentials for them in the
Wilson action for the SU(2) lattice gauge theory, we study the nature of the deconfinement phase
transition on N3 N lattices for N = 4, 5, 6 and 8 and N = 8 16. Using finite size scaling
theory, we obtain / = 1.93 0.03 for N = 4, in excellent agreement with universality.
Corresponding determinations for the N = 5 and 6 lattices are also found to be in very good
agreement with this estimate. The critical couplings for N = 4, 5, 6 and 8 lattices exhibit large
shifts towards the strong coupling region when compared with the usual Wilson action, and suggest
a lot smoother approach to scaling. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.15.Ha; 11.10.Wx; 12.38.Gc; 12.38.Mh
Keywords: Smoother scaling; Deconfinement transition; SU(2) lattice gauge theory; Monopole and vortex
suppression; Finite size scaling theory
1. Introduction
Quantum field theories need regularization schemes to control divergences. The
regularization schemes, many different types of which have been used in performing
calculations, do not themselves affect physics in any manner, as they are eliminated at
the end of all calculations. Long distance physics, such as confinement of quarks in
quantum chromodynamics or determination of the hadronic spectrum, is conveniently
studied using the lattice regularization. There is a lot of freedom in defining a lattice
field theory. In particular, a variety of different choices of the lattice action correspond
to the same quantum field theory in the continuum. While many numerical simulations
have been performed for the Wilson action [1] for the gauge theories, other choices, some
motivated by the desire to find a smoother continuum limit, have also been used. These
E-mail address: gavai@tifr.res.in (R.V. Gavai).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 3 8 - 7
476
R.V. Gavai / Nuclear Physics B 586 (2000) 475488
actions differ merely by irrelevant terms in the parlance of the renormalization group: in
the naive classical continuum limit of a 0, they all reduce to the same YangMills action
and the differing terms are of higher order in a.
Investigations of the deconfinement phase transition for mixed actions, obtained by
extending the Wilson action by addition of an adjoint coupling term, showed [25]
surprising challenges to the above notion of universality. These actions [6,7] are

X
1
1
1 TrF UP + A 1 TrA UP
,
(1)
SBC =
2
3
P
and

X 1
+ V P TrF UP .
SV =
2
(2)
Here UP denotes the directed product of the basic link variables which describe the
gauge fields, U (x), around an elementary plaquette P . F and A denotes that the traces
are evaluated in the fundamental and adjoint representations respectively and the formula
TrA U = |TrF U |2 1 is used. P are Z2 plaquette fields associated with the plaquette P
and the partition function in the Villain case of Eq. (2) has a sum over all possible values
for each of them. The first term in both the equations describes the standard SU(2) Wilson
action whereas the second term adds an adjoint SO(3) action. For zero adjoint coupling,
i.e., for the Wilson action, several finite temperature investigations have shown the presence
of a second order deconfinement phase transition. Its critical exponents have been shown
[8,9] to be in very good agreement with those of the three dimensional Ising model, as
conjectured by Svetitsky and Yaffe [10]. The verification of the universality conjecture
strengthened our analytical understanding of the deconfinement phase transition which,
however, came under a shadow of doubt by the results for the mixed actions. Following
the deconfinement phase transition into the extended coupling plane by simulating these
actions at finite temperature, it was found on a range of temporal lattice sizes for both
actions (1) and (2) that:
(a) The deconfinement transition was of second order, and in agreement with the
conjectured universality exponents, for small values of the adjoint coupling. It
became definitely of first order for large values A or V .
(b) The deconfinement order parameter acquired a nonzero value discontinuously at
the transition point for large adjoint couplings.
(c) There was no evidence of any other separate bulk transition at those large adjoint
couplings, as expected from the results of Refs. [6,7]. In fact, simulations on
larger symmetric lattices even suggested [11] a lack of a bulk phase transition at
that adjoint coupling where a first order deconfinement transition for a lattice of
temporal size four was clearly seen.
Recently it was shown [12,13] that suppression of some lattice artifacts such as Z2
monopoles and vortices do restore the universality for the action (2): no first order
deconfinement transition was found in the entire coupling plane in that case. In this paper,
we address this question for the action (1) in the same manner and find that unlike the
477
Villain action, one gains an additional bonus. The approach to scaling seems to become
smoother than that for the original Wilson action. The organization of the paper is as
follows: In Section 2 we define the action we investigate and briefly recapitulate the
definitions of various observables used and their scaling laws. We present the detailed
results of our simulations in the next section and the last section contains a brief summary
of our results and their discussion.
2. The model and the observables

Bhanot and Creutz [6] found that the lattice theory defined by the extended action of
Eq. (1) has a rich phase structure, shown in Fig. 1. Similar results were obtained for the
Villain action (2) by Halliday and Schwimmer [7]. In either case, the = 0 axis describes
an SO(3) model which has a first order phase transition, denoted by point A in Fig. 1. At
A (or V ) = , the theory reduces
to a Z2 lattice gauge theory with again a first order
1
crit
phase transition at = 2 ln(1 + 2 ) 0.44 [14]. For both the mixed actions, these first
order transitions extend into the coupling plane, as shown in Fig. 1 by continuous lines.
These lines meet at a triple point C and continue as a single line of first order transitions
which ends at an apparently critical point D. The proximity of D to the A = 0 line, which
defines the Wilson action, has commonly been held responsible for the abrupt change from
the strong coupling region to the scaling region for the Wilson action. It has also been
attributed as a possible reason for the dip in the non-perturbative -function obtained by
Monte Carlo Renormalization Group methods. Indeed, its relative closeness to the A = 0
line for the SU(2) theory compared to the SU(3) theory has been thought [15,16] of as a
possible reason for the shallower dip in the former case.
Fig. 1. The phase diagram for the action (1), showing the first order bulk phase transition lines. Taken
from Ref. [6] but with the endpoint D as obtained in Ref. [11].
478
The bulk transition in the Villain form of the SO(3) gauge theory is known to be caused
by a condensation of Z2 monopoles in the strong coupling phase [17]. Defining c =
Q
pc P for an elementary cube c, one finds that c = 1 characterizes a monopole
located in c. These monopoles are absent in the weak coupling region, and can be
P
suppressed at stronger couplings by adding a term, M c (1 c ) to the action (2) and
setting M large. Using this extra term, Ref. [13] demonstrated a clear merging of the
second order deconfinement line with a first order bulk transition line for M = 1 for
the Villain action (2). Moreover, two separate transitions were shown to exist on the same
lattice near the merging point, thereby shedding some light on the paradoxes (a)(c) above.
While pointing to the peculiar role the bulk transitions play in affecting the deconfinement
transitions, these results also suggested that Z2 electric current loops or vortices have to
be suppressed in restoring the universality for the mixed action fully. Defining l for a link
l (x, )
at a point x on the lattice in the th direction as a product of P of all those
Q
= 1 signals the
plaquettes which share the link l, i.e., l = pl
P , one finds that l P
link to be a part of an Z2 electric current loop. Adding further a term E l (1 l ) to the
action (2) in addition to the monopole suppression term above, Refs. [12,13] showed that
universality is restored in the entire coupling plane for large E .
While a similar mechanism is expected to work for the BhanotCreutz mixed action
(1) as well, it is clear that both the monopole and vortex suppression terms added above
do not depend on the gauge variables U (x) directly and have to be generalized suitably
for addition to SBC . One possible way is to define c and l by replacing P in them by
sign(TrF UP ). Thus the mixed action with chemical potentials for these monopoles and
vortices in that case is given by

X
1
1
1 TrF UP + A 1 TrA UP
SBC,S =
2
3
P
X
X
(1 c ) + E
(1 l ),
(3)
+ M
Q
Q
where c = pc sign(TrF UP ) and l = pl
sign(TrF UP ). Comparing the naive
classical continuum limit of Eq. (3) with the standard SU(2) YangMills action, one
obtains
2A
1
.
(4)
= +
2
4
3
gu
Here gu is the bare coupling constant of the continuum theory. Since the asymptotic scaling
equation for the above mixed action with additional (irrelevant) couplings M and E can
be easily written down in terms of gu , it is clear that the introduction of a non-zero A ,
M , or E does not affect the continuum limit: the theory for each A , including the usual
Wilson theory for A = 0.0, flows to the same critical fixed point, guc = 0, in the continuum
limit for all M and E and has the same scaling behavior near the critical point.
One sees that even for A = 0 Eq. (3) corresponds to a modified Wilson action with
different densities of the monopoles and vortices depending on the values of M and E ,
respectively. By analogy with the works of Refs. [12,13] for the Villain action, one expects
479
to eliminate all bulk transition lines in Fig. 1 by setting M and E large. In particular,
the critical point D is expected to be absent in that case, leading perhaps to a lot smoother
transition from the strong coupling region to the scaling region for those values. It has to
be stressed though that universality has to be tested afresh for Eq. (3) even for A = 0 to be
sure that the above naive argument about the M and E terms being irrelevant is correct.
This is what we do in the following by determining a critical index of the deconfinement
phase transition. We then check whether the passage to scaling is affected by studying the
deconfinement transition as a function of the temporal lattice size.
It is interesting to note that no similar change occurs on the fundamental axis (V =
0) in the case of the Villain action since the variables P are decoupled from the gauge
variables in that case and can be integrated out exactly for any observable depending solely
on U (x). However, even in that case the c and l could have been defined in terms of
sign(TrF UP ) and similar results as we obtain below would be obtained.
We studied the deconfinement phase transition on N3 N lattices by monitoring its
order parameter and the corresponding susceptibility for N = 4, 5, 6 and 8 and N = 8,
10, 12, 14, 15, and 16. We used the simple Metropolis algorithm and tuned it to have an
acceptance rate 40%. The expectation values of the observables were recorded every
20 iterations to reduce the autocorrelations. Errors were determined by correcting for the
autocorrelations and also by the jack knife method. The observables monitored were the
average plaquette, P , defined as the average of TrF UP /2 over all independent plaquettes,
and the absolute value of the average of the deconfinement order parameter [18], L(E
n),
over the three-dimensional lattice spanned by nE , where L is defined by
N
Y
1
U0 (E
n, ) .
L(E
n) = TrF
2
(5)
=1
n, ) is the timelike link at the lattice site (E

n, ). In order to monitor the nature of
Here U0 (E
deconfinement and bulk phase transitions, we also define the susceptibilities for both |L|
and P :

(6)
|L| = N3 hL2 i h|L|i2 ,

3
2
2
(7)
P = 6N N hP i hP i .
According to the finite size scaling theory [19], the peak of the |L| (or plaquette)
susceptibility at the location of the deconfinement (or bulk) transition should grow on
N3 N lattices like
max
N ,
|L|orP
(8)
for fixed N . For a second order transition, /, where and characterize the growth
of the |L| (plaquette)-susceptibility and the correlation length near the critical temperature
(coupling) on an infinite spatial lattice. If the phase transition were to be of first order
instead, then one expects the exponent = 3, corresponding to the dimensionality of the
space [20]. In addition, of course, the average |L| or plaquette is expected to exhibit a sharp,
or even discontinuous, jump and the corresponding probability distribution should show a
480
double peak structure in case of a first order phase transition. Such an analysis of the |L|susceptibility for the Wilson action, where only is nonzero in Eq. (3), yielded [8,9] an
exponent = 1.930.03, in good agreement with the corresponding value (1.9650.005)
for the three-dimensional Ising model, and the universality conjecture [10]. Universality
of the continuum limit of lattice gauge theories predicts a similar deconfinement transition
belonging to the same universality class as the three-dimensional Ising model for all values
of A , M and E .
3. Results of the simulations

In view of the fact that even for A = 0, the action (3) differs from the Wilson action
in a nontrivial manner for nonzero M and E , we concentrate here on A = 0 and large
M and E . Our aim is to check the impact of the suppression of monopoles and vortices,
defined by the sign(TrF UP ) as above, on the critical exponent on lattices with fixed N .
We then wish to study the scaling behaviour of the deconfinement transition by varying
N . We chose M = 1 and E = 5 throughout this work. Variations with respect to these
as well as A should in principle be investigated although one would expect the results to
display universality for sufficiently large M and E if the A = 0 results do so.
3.1. N = 4
The deconfinement phase transition on N3 4 lattices for N = 8, 10, 12 and 14 lattices
was studied by first making short hysteresis runs on the smallest lattice to look for abrupt or
sharp changes in both the average plaquette hP i and the order parameter h|L|i. In the region
of strong variation of h|L|i, longer runs were made to check whether the |L|-susceptibility
exhibits a peak. Histogramming technique [21] was used to extrapolate to nearby couplings
for doing this. A fresh run was made at the |L| peak position and the process repeated
until the input coupling for the run was fairly close to the output peak position of the
susceptibility. The same procedure was used for the bigger lattices also but by starting
from the c of the smaller lattice. No peak was found in the vanishingly small plaquette
susceptibility throughout, suggesting a lack of any nearby bulk transition. This should be
contrasted [22,23] with the results for the M = E = 0, which is known to exhibit a peak.
Typically 100200 thousand measurements (24 million Monte Carlo iterations) were used
to estimate the magnitude of the peak height and the peak location for each N . Table 1
lists our final results for all the N used. The errors on c were estimated by varying the
bin size while those for max were taken to be the maximum of the errors for all bin sizes.
Fitting the peak heights in Table 1 to Eq. (8), we obtained
= 1.93 0.03.
(9)
Fig. 2 displays the very good quality of the fit. The critical exponent is in excellent
agreement with the values for both the standard Wilson action [8,9] and the 3-dimensional
Ising model quoted in previous section. Fitting the peak locations c,N by the usual finite
size scaling expression,
481
Table 1
The values of at which long simulations were performed on N3 4
max
lattices, c and the height of the |L|-susceptibility peak, |L|
N
c,N
max
|L|
8
10
12
14
1.37
1.344
1.331
1.34
1.366(7)
1.360(5)
1.345(2)
1.343(2)
9.34 0.07
14.34 0.11
20.44 0.29
27.48 0.64
Fig. 2. Variation of the peak height of |L|-susceptibility with spatial size N for fixed N = 4. The
curve is a fit to the Eq. (8).
c,N = c, + B/N1/ ,
(10)
where = 0.63 is the correlation length exponent for the 3-dimensional Ising model and B
is a constant, we obtained c, = 1.326 0.006, which is shifted by about one from the
corresponding value for the M = E = 0 case which is 2.2986 0.0006 [8,9]. Inspired
by the agreement of above, we assumed universality to be true for as well in Eq. (10).
However, any reasonable variation of between 0.33 and 1 changes the infinite volume
extrapolation for c by a few ( 23) per cent only. Thus the shift = cWilson c =
0.97 appears to be dominantly due to nonzero M and E , i.e., suppression of monopoles
and vortices.
3.2. N = 5, 6 and 8
In order to minimize finite spatial volume effects, we chose to work with N > 2N
always, as seen in Section 3.1. Consequently, the full 4-volume N3 N increased rapidly
482
Fig. 3. The deconfinement order parameter h|L|i as a function of for various temporal lattices N .
For N = 6, the circles (triangles) correspond to a 123 (153 ) spatial lattice. The inset displays average
plaquette hP i on all N as a function of . The lines are smooth extrapolations of the data to guide
the eye.
as we increased N . This resulted in progressive shrinking of the coupling interval in

which the histogramming technique was reliable. We therefore used many longer runs
in the region of strong variation of h|L|i to obtain the susceptibility directly and used the
histogramming only for the finer determination of the critical coupling. Fig. 3 exhibits our
results for h|L|i and hP i (shown in the inset) as a function of for N = 5, 6 and 8. A
deconfinement phase transition is clearly visible for all of them. The behaviour of the order
parameter for two spatial volumes 123 and 153 for N = 6 also supports the existence of
a transition. This is more clearly seen in the corresponding |L| determinations, shown
in Fig. 4. The plaquette, on the other hand, describes a smooth and unique curve for all
N and N values. As the inset in Fig. 3 shows, plaquette values for all these lattices fall
on the same curve, indicating an absence of any bulk transition. Tables 2 and 3 list the
estimated maxima of |L| for N = 5 and 6 for two different spatial volumes along with
the corresponding peak locations. As seen in Fig. 4, they are rather close to the input
at which the long runs were made. Using our value for from Eq. (9), determined for
N = 4, and the peak height for the smaller spatial volume, the max on the bigger lattice
can be predicted. These predictions are listed in the respective tables in the last column
and can be seen to be in very good agreement with the direct Monte Carlo determinations.
Alternatively, one can fit Eq. (8) to the peak heights in Tables 2 and 3 and determine
again:
= 1.99 0.17 for N = 5
and = 1.80 0.14 for N = 6.
(11)
483
Fig. 4. The |L|-susceptibility as a function of on lattices with N = 6. The continuous lines are
extrapolations using the histogramming technique.
Table 2
Same as Table 1 but for on N3 5 lattices
N
c,N
max
|L|
max
predicted
10
15
1.545
1.545
1.570(5)
1.558(2)
13.17 0.17
29.57 0.77
28.78 0.41
Table 3
Same as Table 1 but for on N3 6 lattices
N
c,N
max
|L|
max
predicted
12
15
1.75
1.70
1.735(5)
1.702(2)
16.34 0.45
24.39 0.41
25.13 0.70
Both these determinations agree with the canonical values as well as our own value in
Eq. (9). Not only is the universality of the deconfinement phase transition thus verified
on three different temporal lattice sizes, but it also confirms that the same physical phase
transition is being simulated on them, thus approaching the continuum limit of a 0 in a
progressive manner by keeping the transition temperature Tc constant in physical units.
484
3.3. Scaling of Tc
Table 4 lists the c (estimated by extrapolating to infinite volume whenever possible)
for all the N values we used. The corresponding values for the usual Wilson action, i.e.,
M = 0 and E = 0 case, are also given in Table 4 along with the shifts caused by switching
on these two couplings. The shifts decrease with increasing N but nevertheless remains
sizeable even for the largest lattice we used. Their decrease smoothens the approach to
the scaling limit, as we shall see below. Fig. 5 shows aTc = N1 as a function of the
corresponding critical for both our simulations with suppression of monopoles and
vortices and the standard Wilson action (without any such suppression). The latter are taken
Table 4
The values of c at which the deconfinement phase transition takes place on a lattice with temporal
extension N for Eq. (3) for M = 1 and E = 5 (column 2) and the usual Wilson action (column 3),
taken from Ref. [24]. The last column lists the shift = cWilson c
N
cWilson
4
5
6
8
1.327 0.007
1.56 0.01
1.70 0.01
1.933 0.02
2.2986 0.0006
2.3726 0.0045
2.4265 0.0030
2.5115 0.0040
0.9720.007
0.8130.011
0.7270.010
0.5790.020
Fig. 5. 1/N as a function of c . The squares are from this work while circles are from Ref. [24].
The full lines depict the 2-loop asymptotic scaling relation of Eq. (12), normalized at N = 8 in both
cases. The dashed line denotes Eq. (14), normalized the same way.
485
from the compilation of Ref. [24]. The full lines in the figure show the 2-loop asymptotic
scaling relation
2

1
4b0 b1 /b0
exp
,
(12)
aTc =
N
8b0
where
11
17
, and b1 =
,
(13)
2
24
96 4
are the first two coefficients of the perturbative -function for the SU(2) YangMills
theory. The curve in each case was normalized to pass through the N = 8 data point.
The dashed line describes a phenomenological scaling equation which is similar to the
Eq. (12) but with the exponent increased by a factor of two:
2

1
4b0 b1 /b0
exp
.
(14)
aTc =
N
4b0
b0 =
One sees deviations from asymptotic scaling for both the Wilson action and our action
with suppression of monopoles and vortices. The deviations for the same range of N
seem larger for our action but then one is also considerably deeper in the strong coupling
region of the Wilson action where one a priori would not have even expected any scaling
behaviour. As the agreement of our results with the dashed line of Eq. (14) in Fig. 5 shows,
scaling may hold in this region of for the suppressed action, since the relation between
a and in this region (or g 2 ) is similar to the asymptotic scaling relation, differing only
in the exponent which will cancel in dimensionless ratios of physical quantities. It is clear
that as , the difference between the two actions must vanish. The shifts in Table
4 do show such a trend although the limiting point is not reached by N = 8 definitely.
It seems likely though that the trend of evenly spaced transition points for our action will
continue and the dashed line traced by its transition points will merge with the Wilson
action by N 25 or so, as suggested by its approach to the data for the Wilson action. If
this were to be so, a much smoother approach to continuum limit is to be expected after the
suppression of monopoles and vortices. In particular, one expects that dimensionless ratios
of physical quantities at the deconfinement phase transition couplings should be constant,
already from 1.33, which is the transition point for the N = 4.
One possible interpretation of the results in Fig. 5 is that the proximity of the point D in
Fig. 1 for the usual Wilson action causes the nontrivial curvature visible in the data for the
Wilson action, and consequently its approach to the continuum limit is not so smooth. A
strong suppression of monopoles and vortices, as performed here, eliminates D, resulting
in a smoother approach to scaling. Of course, for very small lattice spacings (or large ),
no significant difference between the two will be seen but for sizeable values of the cut-off
one may expect the action with suppression to exhibit a better and smoother approach to the
continuum limit. We intend to check this by measuring the glueball spectrum at the critical
couplings for N = 48. In the meantime, one can try to check this hypothesis by using Eq.
(14) to convert our () results to (T /Tc ), i.e., as a function of a dimensionless ratio for
various N . Fig. 6 depicts the susceptibility as a function of T /Tc on lattices with N =
486
Fig. 6. Same as Fig. 4 but as a function of T /Tc and on additional lattices, as indicated.
5, 6 and 8. Ideally one would have expected all susceptibility data for the same physical
volume to fall on the same curve for different N . These are the three lowest curves with
physical volume 8 T 3 but with N = 5, 6 and 8. Unfortunately, the order parameter h|L|i
is not ultra-violet safe; it contains divergent contributions in the continuum, making it N
(or a)-dependent even as a function of T /Tc . Consequently, the corresponding s are
close but not on any universal curve. On the other hand, an increase in physical volume
to 15.6 T 3 (the 153 6 data) and 27 T 3 (the 153 5 data) does seem to sharpen the
susceptibility peak progressively, as expected. Although no quantitative analysis can be
done meaningfully due to the cut-off dependence of the order parameter itself, the results
do show the right trend and thus support a possible scaling in the coupling region of these
data points.
4. Summary and discussion

The phase diagram of the mixed action of Eq. (1) in the fundamental and adjoint
couplings, and A , has been a crucial input in understanding many properties of the
SU(2) and SU(3) lattice theories and their continuum limits. The cross-over to the scaling
region from the strong coupling region, as well as the dip in the non-perturbative -function
have been attributed to the location of the end point D of the line of bulk first order phase
transition. In fact, even the relative shallowness of the dip for the SU(2) case compared
to the SU(3) case is thought to be due to the closeness of the corresponding end point
to the A = 0 Wilson axis. Adding extra irrelevant terms to the action one obtains the
487
modified action of Eq. (3) in which monopoles and vortices can be suppressed by setting
the additional couplings to large values. Based on the works [12,13] for Villain action,
one expects the phase diagram to change completely in that case. In particular, no phase
transition lines or their critical end point D will be there, causing a smoother transition
from the strong coupling region to the scaling region.
In this paper we studied the deconfinement phase transition on the fundamental axis
in the (, A ) coupling plane but with M = 1 and E = 5, i.e., with strong suppression
of monopoles and vortices. Our finite size scaling analysis yielded 1.93 0.03 for the
critical exponent / for lattices with N = 4. This value is in excellent agreement
with that [8,9] for the Wilson action and the three-dimensional Ising model, thus verifying
the naive universality of the modified action. However, as a result of the suppression, the
critical coupling is shifted by about unity compared to the Wilson case. Our results on
N = 5 and 6 also yielded similar values for albeit with larger errors, confirming that the
same physical transition was being studied this way as a function of the lattice cut-off, a.
While the aTc = N1 was found to vary slower than expected from the asymptotic scaling
relation (12) for N = 48, the data did obey a similar relation with a factor of two larger
exponent. A straightforward extrapolation suggests the results from the modified action
will merge with those of Wilson action for large N (of about 25), as expected in the limit
of vanishing lattice spacing a. This suggests that the suppression makes the approach from
the strong coupling side to the scaling side much smoother than that for the unsuppressed
Wilson action, allowing us to simulate the theory at smaller . It will be interesting to see
if dimensionless ratios of physical quantities such as glueball masses or string tension with
Tc are constant in the range of critical couplings explored here. Since the phase diagrams
for SU(3), and indeed SU(N) lattice gauge theories, is similar and the same mechanism
is expected to work for them, it will also be interesting to study such suppression in those
theories as well. However, additional possibilities for topological objects may add further
complications and may make it necessary to suppress them as well.
Acknowledgements
It is a pleasure to acknowledge interesting discussions with Sourendu Gupta.
References
[1] K. Wilson, Phys. Rev. D 10 (1974) 2445.
[2] R.V. Gavai, M. Grady, M. Mathur, Nucl. Phys. B 423 (1994) 123.
[3] M. Mathur, R.V. Gavai, Nucl. Phys. B 448 (1995) 399; Nucl. Phys. B (Proc. Suppl.) 42 (1995)
490.
[4] R.V. Gavai, M. Mathur, Phys. Rev. D 56 (1997) 32.
[5] P.W. Stephenson, hep-lat/9604008.
[6] G. Bhanot, M. Creutz, Phys. Rev. D 24 (1981) 3212.
[7] L. Caneschi, I.G. Halliday, A. Schwimmer, Nucl. Phys. B 200 (1982) 409.
[8] J. Engels, J. Fingberg, M. Weber, Nucl. Phys. B 332 (1990) 737.
488
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
J. Engels, J. Fingberg, D.E. Miller, Nucl. Phys. B 387 (1992) 501.

B. Svetitsky, L.G. Yaffe, Nucl. Phys. B [FS6] 210 (1982) 423.
R.V. Gavai, Nucl. Phys. B 474 (1996) 446.
R.V. Gavai, M. Mathur, Phys. Lett. B 458 (1999) 331.
Saumen Datta, R.V. Gavai, Nucl. Phys. B (Proc. Suppl.) 8384 (2000) 366; Phys. Rev. D 62
(2000) 054512.
F.J. Wegner, J. Math. Phys. 12 (1971) 2259.
A. Patel, S. Otto, R. Gupta, Phys. Lett. B 159 (1985) 143.
U. Heller, F. Karsch, Phys. Rev. Lett. 54 (1985) 1765.
I.G. Halliday, A. Schwimmer, Phys. Lett. B 102 (1981) 337.
L. McLerran, B. Svetitsky, Phys. Rev. D 24 (1981) 450.
M.N. Barber, in: C. Domb, J.L. Lebowitz (Eds.), Phase Transitions and Critical Phenomena,
Vol. 8, Academic Press, New York, 1983, p. 146.
M.S. Challa, D.P. Landau, K. Binder, Phys. Rev. B 34 (1986) 1841.
A.M. Ferrenberg, R.H. Swendsen, Phys. Rev. Lett. 61 (1988) 2635.
J. Engels, T. Scheideler, Phys. Lett. B 394 (1999) 147.
M. Nauenberg, T. Schalk, R. Brower, Phys. Rev. D 24 (1981) 548.
J. Fingberg, U. Heller, F. Karsch, Nucl. Phys. B 392 (1993) 493.

Non-linear realization of 0-extended

supersymmetry
Hitoshi Nishino
Department of Physics, University of Maryland at College Park, College Park, MD 20742-4111, USA
Received 24 February 2000; accepted 26 June 2000
Abstract
As generalizations of the original VolkovAkulov action in four-dimensions, actions are found
for all space-time dimensions D invariant under N non-linear realized global supersymmetries. We
also give other such actions invariant under the global non-linear supersymmetry. As an interesting
consequence, we find a non-linear supersymmetric BornInfeld action for a non-Abelian gauge group
for arbitrary D and N, which coincides with the linearly supersymmetric BornInfeld action in
D = 10 at the lowest order. For the gauge group U (N ) for M(atrix)-theory, this model has N 2 extended non-linear supersymmetries, so that its large N limit corresponds to the infinitely many
(0 ) supersymmetries. We also perform a duality transformation from F into its Hodge dual
N1 D2 . We next point out that any ChernSimons action for any (super)groups has the nonlinear supersymmetry as a hidden symmetry. Subsequently, we present a superspace formulation for
the component results. We further find that as long as superspace supergravity is consistent, this
generalized VolkovAkulov action can further accommodate such curved superspace backgrounds
with local supersymmetry, as a super p-brane action with fermionic kappa-symmetry. We further
elaborate these results to what we call simplified (Supersymmetry)2 -models, with both linear and
non-linear representations of supersymmetries in superspace at the same time. Our result gives a
proof that there is no restriction on D or N for global non-linear supersymmetry. We also see that
the non-linear realization of supersymmetry in curved spacetime can be interpreted as nonperturbative effect starting with the flat spacetime. 2000 Elsevier Science B.V. All rights
reserved.
1. Introduction
The study of non-linear supersymmetry realizations has a very long history dating to
the first effort by Volkov and Akulov [1], who gave the initial example of supersymmetry
in four-dimensions (D = 4). This initial work was further elaborated by its linearization
This work is supported in part by NSF grant # PHY-93-41926.
nishino@nscpmail.physics.umd.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 2 0 - X
492
H. Nishino / Nuclear Physics B 586 (2000) 491517
[2], and generalized to extended local supersymmetries in D = 4 [3,4]. The importance

of non-linear realizations of supersymmetry has been revealed in many different contexts.
For example, during the first revolution of string theory [5], there was strong motivation
for supersymmetrization [6] of the BornInfeld action [7,8], and initial results were
summarized in [9] where the lower-order terms were explicitly and systematically
presented.
The recent discovery of gauge-fixed supersymmetric BornInfeld action in D 6 10 to
all orders [10] has drawn much attention in the context of D-brane dynamics [1119]
related to superstrings [5] and M-theory [2023]. The non-Abelian generalization of these
formulations was also considered in [24]. In such a formulation, there is already nonlinear realization of supersymmetry, interpreted as the signal of broken supersymmetry.
In D = 4 formulations, non-linear supersymmetry has been interpreted in terms of
partially broken N = 2 supersymmetry [2527], with NambuGoldstone (NG) multiplets.
The relationship between the non-linear and linear supersymmetry in superspace has been
also studied in [28,29]. These global supersymmetric systems have been further applied to
the question of spontaneous breaking of local supersymmetry (supergravity) [3032].
Despite of these developments over the decades since 1970s, however, including the
recent D-brane models, there still has been no universal guiding principle that can
systematically yield new invariant actions for non-linear realization of supersymmetry
in arbitrary space-time dimensions. This seems true also for the interpretation of nonlinear realization as broken supersymmetry [2527]. On top of this, even though the
original spirit of VolkovAkulov formulation of non-linear supersymmetry [1] was offshell without auxiliary fields present and no physical superpartner fields in D = 4, this
aspect has not been maintained in many formulations recently presented. Perhaps one of
the reasons is that D-brane theory [10,11,18] necessitates the superpartner vector field for
the gaugino.
In the case of linear realizations of supersymmetry, it is well-known that there
are restrictions on the spacetime dimensions D and the number N of extended
supersymmetries, such as N 6 8 in D = 4. However, very little has been stated about such
restrictions on D or N for extended supersymmetries in the case of non-linear realizations.
In this paper, we will take a step toward a universal understanding of non-linear
realization of supersymmetry in arbitrary dimensions. We will present simple actions
that are invariant under non-linearly realized supersymmetry. We also give some matter
Lagrangians that are invariant under such global but non-linear supersymmetry. Such
matter Lagrangians coupled to NG-fermion within D = 4 have been studied in many
contexts of phenomenological models [33,34], or non-linear realizations for extended
supersymmetries [3,4]. We stress that our result is valid in any spacetime dimensions D as
well as any number N of extended supersymmetries. We find results applicable to arbitrary
non-linearly realized N -extended supersymmetries, as long as the spacetime allows the
definition of spinor fields. As a by-product, we also give a supersymmetric BornInfeld
action for an arbitrary non-Abelian gauge group D and N . We also establish a superspace
formulation for these component results, whose geometrical significance is clearer than in
a component formulation. We further give the super p-brane [35] interpretation of some
493
of these actions, on supergravity backgrounds. This will be achieved by introducing the

maximal-rank superpotential CA1 AD in an arbitrary even dimensions D = 2k, which
was first formulated in the context of super eight-brane [36]. Motivated by this, we also
present a possible (Supersymmetry)2-formulation [37], in which the non-linear and linear
realizations of supersymmetries coexist simultaneously within the same superspace.
2. Non-linear realization of supersymmetry

2.1. Generalized VolkovAkulov action
We first setup our supersymmetry transformation which is the starting point in this
paper. As mentioned above, there is no restriction on the spacetime dimensions D in
our formulation, except that we have the usual signature (D 1, 1) with the flat metric
(mn ) = diag(, , , , +), with only one time coordinate. Our system initially has
only one NG-fermion, which may be Majorana, Weyl, MajoranaWeyl, or Dirac spinor.
As will be seen, there is no restriction on such spinorial property for our formulation.
The rule for a supersymmetry transformation of the NG-fermion is essentially the same
as those in [1] or [10], but we have to keep in mind that the spacetime dimension D is
arbitrary 1 :

(2.1)
Q = + i + ,
where the parameter of the supersymmetry generator Q does not depend on space
time coordinates x , since we are dealing only with global supersymmetry. The quantity
defined by 2 , 3

i = i ,
(2.2)
behaves like the parameter for general coordinate transformation. It is easy to see the
closure of supersymmetry on , as a routine check for consistency of the system:

m
,
(2.3)
Q (1 ), Q (2 ) = 2i 2 1 = P 2i 2 m 1 P 12
m 2i( m ), reflecting the underlying
where P is the usual translation, and 12
2
1
4
supersymmetry algebra

{Q , Q } = 2 m Pm .
(2.4)
A fundamental observation is that the quantity defined by

1 We do not introduce any dimensionful constants in this paper. We just put dimensionful constants like the
Planck constant to unity.
2 Here the gamma-matrix is the same as m . The usage of the index instead of m becomes clearer
m
shortly.
3 Here the underlines for the indices , , . . . are for any additional indices for spinors, such as dotted-ness
for chiralities, or USp(N )-indices needed in some spacetime dimensions D [38]. Accordingly, the -matrix may
also contain any relevant metric for such indices.
4 The index for the N -extended supersymmetry can be implicit in the underlined indices. We will return to this
point later.
494

E m m + i m m + m ,
(2.5)
transforms under Q as

Q E m = i E m + i E m

= E m + E m = G E m .
(2.6)
m
This is a significant result, because it means that the field E transforms under
supersymmetry, as if it were a vielbein transforming under the general coordinate
transformation G with the parameter not only in D = 4 [1] but also in any space
time dimensions! Now the advantage of using the two distinct sets of indices m, n, . . . and
, , . . . on E m is also clear: They respectively behave like the curved frame and local
Lorentz frame indices.
Once this is understood, it is straightforward to confirm the invariance of the action
Z
Z
Z

(2.7)
IE dD x det E m dD x E dD x LE , E det E m ,
under the variation in (2.1), as a total divergence 5 :

Q LE = EEm Q E m = EEm E m + E m = E .
(2.8)
We can reconfirm this formal proof by the direct variation with respect to the NG-fermion
whose details are to be skipped here.
As the -field equation or the original Lagrangian (2.7) reveals, the action IE contains
the standard kinetic term for the NG-fermion, upon the expansion det(E m ) 1 +
i( ) + O(4 ). Therefore, the NG-fermion is regarded as a physical field, also
with self-interaction terms that make our system highly non-trivial.
2.2. More generalizations
As the reader may have already noticed, once the transformation rule of the vielbein is
as appears in (2.5), we can write other Lagrangian densities constructed from the vielbein.
For example, we can write Lagrangians in terms of the scalar curvature made up of the
metric 6 G mn E m E n with its inverse metric G : G G = , such as the
Hilbert action:
Z
IR dD x GR(G),
R(G) G R (G), R (G) R (G),
n o
n
o n
on o n on
o
+
,
R (G)
n
o 1
G ( G + G G ),

2
because G transforms as
(2.9)
5 The existence of the inverse vielbein E is taken for granted otherwise the action is ill-defined.
m
6 We use G
instead of g as a reminder that this is a composite metric but not a fundamental one.
Relevantly, G should not be confused with the Einstein tensor R (1/2)G R.

Q G = G + G + G = G G ,
495
(2.10)
as if it were a metric tensor. Accordingly, the curvature tensors above transform as

Q R (G) = G R (G),

Q R(G) = G R(G).
(2.11)
Q R (G) = G R (G),
In the case we need more canonical kinetic term for the NG-fermion , we can simply
add the action IE (the analog of a cosmological term) to IR , so that a kinetic term for the
-field is present, while higher-order terms are regarded as interaction terms. In fact, if the
total action is IE + IR , the lowest-order term in IR is already at the fourth order in as an
interaction term.
Some readers may wonder if the Riemann or Ricci tensor built out of the metric G
mn E m E n vanishes by the use of the -field equation. We show this is not the case, e.g.,
for the Ricci tensor

R (G) = 2i , , , (| , |) , (| , |) + , ( ),

(2.12)
+ O 4 .
Here , , , , etc. 7 Obviously, even though i O(3 ) by the
NG-fermion equation at the lowest order, there still remains non-vanishing component
even for the scalar curvature R(G). We can further elaborate our result, introducing more
higher-derivative Lagrangians, such as (R (G))2 , (R (G))2 , or R(G)2 , etc.
We stress here that the invariance of IE or IR relied only on the particular transformation
rule (2.1) of the vielbein, but did not use any property related to the spacetime dimensions,
such as Fierz arrangements. This implies that the non-linear realizations of supersymmetry
should be valid in D, where we can define a spinor field with the usual kinetic term. In
this context, the specific nature of the spinors (Weyl, Majorana, or Dirac) does not matte.
The generalization to extended supersymmetry is also straightforward, assigning an
explicit N -supersymmetry index i on i . Accordingly, (2.1) and (2.2) are now

i i i .
(2.13)
Q i = i + i ,
If we suppress the explicit contraction index i as understood, then formally the same
equations as (2.4) through (2.9) follow, and there is no restriction on the value of N . We
have thus a non-linear realization of arbitrarily extended global supersymmetries.
Note that we are dealing with two different spacetimes, the a flat spacetime with
no gravity and with no curvature, while the other spacetime has non-vanishing curvature
(2.12), in which we can define general coordinate transformations. Interestingly, the action
IE is invariant in both of these spacetime simultaneously.
Our result applies even to D > 12 for the invariant actions under non-linear supersymmetry, including of course includes D = 12 of F-theory [39] or D = 13 of S-theory [40],
and so forth, as well as the M(atrix)-theory in D = 11 [4144]. As a matter of fact, we will
see a more explicit link with the M(atrix)-theory [4144] in Section 3.3.
7 Since only lowest-order terms are under question here, the raising/lowering indices by the metric G
or
does not matter here.
496
3. Matter Lagrangians
3.1. Bosonic Lagrangians
We have so far offered a possible meaning of our vielbein, only with the NG-fermion.
The next natural question is whether or not we can introduce other matter fields. In the case
of D = 4, we already know that such Lagrangians are easily constructed [3,4,6,28,29].
However, because of the latter applications to super p-branes, etc., we generalize these
into D and N , even though such a generalization looks straightforward. We show that the
answer to this question is in the affirmative, i.e., we can introduce other matter fields, as
long as our general covariance G is maintained.
First, we can show that the algebra of supersymmetry closes on the following scalars
a , a non-Abelian gauge field A I , an n-th rank covariant tensor field T1 n with no
restriction on symmetry or antisymmetry:

(3.1a)
Q a = i a = a = G a ,
I
I
I
I
(3.1b)
Q A = A + A = G A ,

Q T1 n = T1 n + 1 T2 n + + n T1 n1

= G T1 n .
(3.1c)
Here and throughout this paper we use exactly the same as (2.1). We use the indices
a, b, . . . for an appropriate representation of a certain compact non-Abelian gauge group
G for the scalars, while I, J, . . . for the adjoint representation of G. Needless to say, if
some of the indices on the tensor are contravariant ones, we can change the signs of terms
corresponding to those contravariant indices in (3.1c), following the usual textbook result.
Interestingly, the closure on the vector or tensors generates no gauge transformations as in
(2.3), in addition to the standard translation:

m
,
(3.2)
Q (1 ), Q (2 ) = P 12
where is an arbitrary field in (3.1). This is another aspect of the non-linear supersymmetry
different from the linear supersymmetry, in which such gauge transformations are induced
in the commutator algebras [45].
The construction of some invariant matter actions is now an easy task, as long as our
general covariance is maintained via G or E m . A typical bosonic action is like

Z

1
1
D
a
a
I
I
(3.3)
IB = d x G + G D D G G F F ,
2
4
where D a is the usual gauge covariant derivative D a a + A I (T I )ab b for
an appropriate representation of the generators T I of the gauge group G, and F I
A I A I + f I J K A J A K is the usual field strength. These gauge covariant
combinations transform under Q as

Q D a = D a + D a = G D a ,
(3.4a)

I
I
I
I
I
(3.4b)
Q F = F + F + F = G F .
497
It is not a problem at all to generalize the action (3.3) to more sophisticated interactions. If
the above result is only for D = 4, it is nothing new [3,4,6,28,29,33,34], but our point here
is that these actions are valid for N -extended supersymmetry in D.
3.2. Non-Abelian supersymmetric BornInfeld Lagrangians
With these relations at hand, it is now straightforward even to supersymmetrize a Born
Infeld Lagrangian. This can be easily done, because the combination H G + F
for an Abelian field strength F transforms covariantly: Q H = Q (G + F ) =
G ( )H , therefore the action 8
Z
Z

1/2

1/2
D
ISBI d x det(G + F )
= dD x det(H )
(3.5)
is invariant under supersymmetry: Q ISBI = 0. This sort of supersymmetric BornInfeld
Lagrangian in particular dimensions, such as in D = 10 is nothing new [9,10], but our
Lagrangian is valid for D and N .
It is not too difficult to generalize this result to the non-Abelian gauge group. Following
the non-supersymmetric non-Abelian BornInfeld action in [24], we can easily postulate a
non-Abelian analog of the supersymmetric BornInfeld action (3.5), as
Z
Z

1/2
1/2
dD x STr det(H )
,
INASBI = dD x STr det G I + F I T I
(3.6)
where the generators T I are in an appropriate matrix representation, and I in G I is
the unit matrix for such a representation. The determinant operation here is only for the
indices , but not for the group indices. The STr is the symmetrized trace defined by
STr(A1 An ) (1/n!) Tr(A1 An + all permutations). Due to the Q -transformation
property

(3.7)
Q H = Q G I + F I T I = G H ,
of the combination H G I + F I T I , together with the property of STr, it is clear
that the action INASBI is invariant under Q . As before, since INASBI is valid for D and
N , our result is a generalization of [9,10], up to discrepancy in higher-order terms. In fact,
it is easy to see that the lowest-order terms in our (3.6) agree with those in [9,10].
Needless to say, we can also add the antisymmetric potential b to (3.5):
H = G + F + b ,
while the standard kinetic term for its field strength G 3[ b] is

Z
1
D

GG G G G G ,
Ib = d x
12
(3.8)
(3.9)
as in super D-branes [18]. In other words, there can be a general antisymmetric component
b for the metric G , maintaining the Q -invariance as well as the b-field gauge
8 We need the negative signs in the square roots, because det( ) < 0.
mn
498
invariance: b = 2[ ] . Adding further an appropriate minimal couplings of A to

scalars like IB (3.3), there is no worry that the field strength F is completely gauged
away in (3.8) by the field redefinition of b .
For the N -extended supersymmetry for N > 2, we can introduce an SO(N) gauge field
A ij (i, j, . . . = 1, 2, . . . , N) minimally coupled to i , when all the components of are
Majorana spinors or (Majorana-)Weyl spinors in the same chirality. Such a vector field is
also subject to the covariant transformation rule (3.1b). Relevantly, the derivative in (2.5)
is replaced by the covariant derivative D i i + A ij j :

(3.10)
E m = m + i i m D i .
One important case is when i, j, . . . indices are adjoint ones I, J, . . . = 1, . . . , dim G for
D I I + f I J K A J K :

E m = m + i I m D I .
(3.11)
Then the lowest-order terms of our action (3.6) in D = 10 coincide with those in linearly
supersymmetric BornInfeld action [10]. This seems to imply that our action (3.6) can be
of equal importance for the M(atrix)-theory [4144] as the D-brane actions [1118]. 9
Interestingly, the number N of extended supersymmetries in our formulation will be
N = dim G, e.g., N = N 2 for G = U (N ) used for M(atrix)-theory [4144]. Therefore, the
usual large N limit for U (N ) [4144] corresponds to N = N 2 , namely infinitely
many (0 ) extended supersymmetries in our formulation!
3.3. Fermionic Lagrangians
We can incorporate other fermionic fields rather easily, because we need only global
Lorentz invariance for our action, much like the case of general covariance, which is not
really present. In fact, we can show that the action 10
Z

i
IF = dD x EEm a m D a ,
(3.12)
2
for a fermion a is actually invariant under supersymmetry:

(3.13)
Q a = i a = a = G a .
We use the indices a, b, . . . for some appropriate representation of the group G that
belongs to, and D is the gauge covariant derivative D a a + A I (T I )ab b . The
invariance Q IF = 0 is due to the transformations

(3.14)
Q D a = D a + D a = G D a ,
like a general covariant vector, with respect to the index . As in the case of , there is no
essential restriction on the property of the spinor , such as Weyl, Majorana, Majorana
Weyl, or Dirac spinor, with any additional group indices in arbitrary D.
9 See, however, the recent development about subtlety at higher orders [46]. We do not get into these details in
this paper.
10 In order to avoid any misinterpretation, we always use E m or E explicitly in this paper, instead of
m
m Em , etc.
499
Once we have a fermionic field , then we can modify (2.5) as

E m = m + i m + i a m D a .
(3.15)
( a m D a )
transforms as a covariant vector with respect to the

This is because
index . Accordingly we have the alternative fermionic action
Z

(3.16)
IE0 = dD x det m + i m + i a m D a .
We have thus two actions in good contrast, IE0 for fermions, and INASBI for bosons.
3.4. ChernSimons Lagrangians
Our results above have a wide range of applications. Since INASBI needs no explicit
metric, but the GN-field forms a composite metric, the first interesting application is the
ChernSimons action for an arbitrary (super)group in D = 2n + 1 [4852]:
Z
ICS =
z
}|
{

STr F F F A + ChernSimons completion
MD
=
M D+1
F ,
STr F
}
| F{z
(3.17a)
(3.17b)
n+1
to be added to the above actions: I ICS + INASBI + IE0 + IB . Here STr acts on fermionic
generators as an antisymmetrized trace, as a supergroup generalization of the symmetrized
trace for ordinary groups. In (3.17b), the hatted field strength F is defined in the extended
spacetime D +1 = 2n+2 with one additional Vainberg coordinate 0 6 y 6 1 [53,54] with
the indices (x ) (x 0 , x 1 , . . . , x 2n , x 2n+1 y) (x , y) by
(

F A A + A , A ,
F A A + A , A
Fy y A = Fy ,

(3.18)
A (x , 1) A x , A x , 0 = 0, A y = 0,
R1
in such a way that the leading term (3.17a) is generated by the y-integration 0 dy [53].
The gauge group here can be even a supergroup, such as OSp(N|m) for ChernSimons
supergravity in D = 2n + 1 [4852]. Accordingly, the generators T I in (3.6) should be
those of the gauge (super)group under consideration. Due to our transformation rule (3.1b),
the invariance Q ICS = 0 is clear.
The important point is that in the conventional ChernSimons formulations including
those with supergroups [4852], the introduction of additional matter fields has been
problematic, because the kinetic terms of such matter fields need some metrics or
vielbeins. However, since INASBI has no fundamental metric, it is compatible with such
formulations without a metric by construction. In this sense, the non-linear realization of
supersymmetry is more natural for ChernSimons theories. Due to the minimal couplings
in IF or IB , there are still non-trivial interactions of the gauge fields to other matter
bosonic or fermionic fields.
500
To put this differently, the important conclusion is that any ChernSimons action (3.17)
formulated in D = 2n + 1 for a (super)group is invariant under hidden global non-linear
supersymmetry dictated by (3.1b). Moreover, once the action IE (2.7) is added to ICS ,
then the GN-field becomes physical, and the total action is no longer invariant under
arbitrary general coordinate transformations, so that the GN-fermion is no longer gauged
away. Therefore the global supersymmetry is not a superficial symmetry any longer by
the presence of IE . This has a very far-reaching conclusion, because any ChernSimons
theory for a supergroup [4852] except in D = 3 [55] has been supposed to have no direct
link with the Poincar supersymmetry with diffeomorphism, in particular in the context of
M-theory [4951,56,57]. 11 This missing link has been a major drawback with the Chern
Simons formulation [4852] in the context of spacetime supersymmetry. In other words,
the non-linear representation of supersymmetry may well provide a missing link between
ChernSimons theory for supergroups [4852] and spacetime supergravity theories [59]
with diffeomorphism.
3.5. Generalized -model Lagrangians
We can try even a more sophisticated example. Suppose we have a coset G/H , whose
coordinates are parametrized by the scalars a . Now instead of H in (3.5), we consider

(3.19)
H G + F + gab () a b ,
where gab () is the metric on this coset. Due to the transformation property of a under
Q , it is clear that the action
Z

1/2
,
(3.20)
I = dD x det G + F + gab () a b
is invariant under Q . As we easily see, the lowest order-term in the determinant contains
the same kinetic term for the scalars a as the conventional -model. We can further add
some WessZuminoNovikovWitten (WZNW) term to I , as an analog of WZNW-terms
used in D-brane theories [10,1214,18].
3.6. Duality transformation of F into N1 D2
As another interesting by-product, we mention the duality transformation [60] for the
field strength F into N1 D2 in D. We can show this is not too difficult, even with
the general metric tensor like that in (3.20). Such a duality transformation has been done
in D = 4 [6163], but our formulation here is complimentary, applicable to D 6= 4 and
N .
We start with the generalized supersymmetric BornInfeld action
Z

1/2
0
,
(3.21)
ISBI dD x det(g + F )
11 See, however, Ref. [58] in which extended objects are introduced to ChernSimons supergravity to study such
a direct link with Poincar supergravity.
501
for an Abelian field strength F . Here g can be any generalized metric containing any
fields other than A or F , e.g.,

(3.22)
g G + gab () a b ,
in the case of (3.20). The crucial point here is that g should have no coupling to A for
the duality transformation [60] to work.
The Lagrangian (3.21) is not very convenient for duality transformations, because the
lowest-order term in F is not manifestly quadratic in F , is not manifest. In order to make
the F 2 -term more manifest, we first introduce the generalized vielbein
g = e m mn e n .
(3.23)
Note that this e m is more generalized than (2.5), and it contains at least the latter as a part
of it. Using such a vielbein, we see that

det(g + F ) = det e m m n + Fm n en = det e m (det en ) det m n + Fm n

(3.24)
= det(g ) det m n + Fm n ,
where as usual Fmn em en F , Fm n nr Fmr , etc. Note also that the antisymmetry
F t = F implies that

det m n + Fm n = det(I + F ) = det (I + F )t = det(I F )
1/2

1/2
= det (I + F )(I F )
= det(I + F ) det(I F )

1/2 n
1/2
= det I F 2
= det m F 2 m n
,
(3.25)
with (F 2 )m n Fm r Fr n . Hence (3.21) is equivalent to
Z
1/4

0
= dD x g det m n F 2 m n
,
ISBI
(3.26)
where g det(g ) and the F 2 -term is now manifest.

We next linearize the determinant operation, in such a way that the F 2 -term is involved
only as a quadratic term, by introducing a symmetric matrix auxiliary field Pm n :
Z

0
= dD x a g cP 1/4 + P 1/4 P 1 m n n m F 2 n m ,
(3.27)
ISBI
where P det(Pm n ), and the constants a and c are
1
c = 4 D,
a= ,
4
in order to get the normalized algebraic field equation for Pm n :

Pm n = m n F 2 m n .
(3.28)
(3.29)
Here the matrix multiplication for F 2 is done with the local Lorentz indices m, n, . . ..
By these preliminaries, it is now easy to perform the duality transformation [60] from
F into N1 mD2 . As usual, we introduce the Lagrange multiplier potential M1 D3
into a constraint action, so that the total action is now
502

dD x ac g P 1/4 + a g P 1/4 P 1 m n n m F 2 n m

2
1 D3 M1 D3 F
+
(D 1)!

Z

= dD x ac g P 1/4 + a g P 1/4 P 1 m n n m F 2 n m

2
1 D2

N1 D2 F .
(D 2)!
00
ISBI
=
(3.30a)
(3.30b)
Here F is no longer a field strength of A , but is regarded as an independent field variable

[60], while N1 D2 (D 2)[1 M2 D2 ] . The constraint term at the end of (3.30a)
forces the Bianchi identity for F for consistency. Our next step is to obtain the algebraic
field equation of F , as
00
e = 0,
ISBI = Q F + F Q 2N
(3.31)
where
Q P 1/4 P 1
P 1/4 e m en P 1

m
(3.32)
and
1
e 1
1 D2 N1 D2 .
N
g (D 2)!
(3.33)
The field equation (3.31) is conveniently expressed as a D D matrix equation

Q, F
em n .
= 2N
(3.34)
e, we split Q into Q = I 2q, and get

In order to solve (3.34) for F in terms of P and N
the solution as an infinite series in terms of q:
F =

r
X
r
X
r s e rs X X 1 r
e Q)rs .
=
q Nq
(I Q)s N(I
2r s
s
r=0 s=0
(3.35)
r=0 s=0
Even though this is an infinite series, it is a closed form, starting with the lowest-order term
e + O(q), as desired. Our final step is to substitute (3.35) into (3.30b), in order
F = N
to eliminate F . After this, all the F -terms are replaced by its dual N1 D2 , and thus we
have achieved the duality transformation on a generalized metric (3.23).
Eq. (3.28) indicates that the case of D = 4 is exceptional or singular. This is because the
original field strength and its Hodge-dual have the same rank. This seems also related to the
possible SL(2, R) duality invariance between these two field strengths [6163]. Other than
this exceptional case of D = 4, our action is still superinvariant, because of the general
coordinate invariance valid in D 6= 4 and N .
503
4. Superspace formulations
4.1. Second and third-rank superfield strengths
Once we have understood the component formulation of the non-linear realization of
supersymmetry, our next question is how to reformulate the same results in superspace
[64]. This is a non-trivial question, because even though we know that such a formulation
is definitely possible in D = 4 with auxiliary fields [2527], or without auxiliary fields
in D = 10 [10], D = 6 or D = 7 [1517], it is no longer trivial that we can repeat it in
arbitrary spacetime dimensions D. In particular, it is no longer automatic that superspace
Bianchi identities [64] are satisfied, when supersymmetry is realized non-linearly. This is
due to possible extra symmetries present in the system, that prevent us from formulating
the component results in superspace [55].
We answer this question, by the actual investigation of conventional superspace Bianchi
identities [64]. As the first non-trivial example, we study the Bianchi identity 12
[A FBC) I T[AB| D FD|C) I 0,
(4.1)
for the superfield strength FAB I of a non-Abelian gauge vector superpotential AA I ,

corresponding to (3.1b).
Our superspace constraints are summarized as

(4.2a)
T c = 2i c ,

(4.2b)
= + i c c i c c ,

(4.2c)
Fb I = +i c Fcb I i c Fcb I = P a Fab I ,
F I =

1 a
( a ) Fab I = +P a P b Fab I ,
2

P b i b .
(4.2d)
There is no a priori restriction on D here. 13 The most important constraint is (4.2d), which
is similar to the postulates given in [9,65] in D = 10 as a special case of D. This is because
F I can be rewritten as

1
1
(c ) abc Fab I +
(c1 c5 ) a c1 c5 b Fab I , (4.3)
16
1920
in terms of undotted chiral indices in D = 10 after Fierzing, where the second term has
been suggested in [9,65]. For an arbitrary spacetime dimension D, we keep the original
form (4.2d), because forms like (4.3) needs Fierzing depending on D.
F I =
12 In this section of superspace, we use the indices A = (a, ), B = (b, ), . . . for superspace coordinates. The
underlined spinorial indices , , . . . include any possible implicit indices such as dottedness for chiralities, or
some USp(N ) indices needed for extended supersymmetry, depending on D [38]. Our symmetrizations in this
section are not normalized, i.e., M[A NB) MA NB .
13 Depending on D, we have to switch between the antisymmetry and symmetry of the metric tensor for spinorial
indices [38]. For such cases, we need to switch some signs for contracted indices in the constraints here, but the
basic structure is still universal.
504
The confirmation of Bianchi identities (4.1) goes in a way similar to the conventional
case, once we know the structure of F I . However, there are also important differences.
For example, it turns out to be easier to proceed the conformation of Bianchi identities
from the highest mass dimension two (d = 2) to the lowest d = 1/2 backward, as opposed
to the conventional case, where the lowest-dimensional one is the simplest. The reason for
this is that the lowest dimensional constraint (4.2c) is more involved than the higher ones.
Another important aspect of this system is that the Bianchi identities with these constraints
do not yield any field equation for the NG-fermion . This is natural, because as we saw
in component results, the transformation rule for is independent of the choice of the total
action, e.g., IE versus IE + IR , etc., namely the system of non-linear supersymmetry is
essentially off-shell.
Another instructive example is the third-rank superfield strength GABC , satisfying the
Bianchi identity
1
1
[A GBCD) T[AB| E GE|CD) 0.
6
4
Our superspace constraints are
Gbc = P a Gabc ,
Gc = i(c ) + P a P b Gabc ,
G = P a P b P c Gabc ,
(4.4)
(4.5a)
(4.5b)
in addition to (4.2a) and (4.2b).

Note the interesting patterns of involvement of the factor P b in (4.5) as well as (4.2),
namely all the spinorial index , , . . . in the superfields are converted into the bosonic
ones a, b, . . . by the factor P a . By reading these patterns, it is easy to generalize these
results to more higher-rank superfield strengths. The role of such a factor has been noted,
e.g., in D = 6 [1517], but our point is that our constraints are valid for D and N .
Even though we skip the details for these higher-rank superfield strengths in this paper,
the geometrical structure of this superspace formulation is much clearer than component
formulation. In the next subsection, we address ourselves to the question of coupling to
curved superspace backgrounds with supergravity, based on this particular role for the
factor P b .
4.2. Super (2k 1)-brane interpretations 14
As the pattern of the factor P a suggests, all of the effects of the GN-fermions can be
reinterpreted as part of the vielbein superfields. For example, the first form (2.5) of E m
suggests that the GN-fermion can be interpreted as nothing else than the -coordinates in
flat superspace, as has been also pointed out by several authors [10,1518,28,29] for special
cases of D = 10 [10], D = 6, D = 7 [1517], or D = 4 [28,29]. In our paper, we generalize
this feature to more general dimensions, because this process seems imperative, if we also
want to introduce curved backgrounds with supergravity, and we need to consider more
geometrical superspace formulation.
14 We avoid such terminology as (D 1)-brane, because it is confusing with the D-branes.
505
If this scenario really works in more general dimensions, we must have the geometric
structure of a base manifold and a target superspace at the same time, as special cases
[10,1517] indicate. Accordingly, we must have an analog of GreenSchwarz type super
p-brane action [35] in which the base manifold (world-supervolume) and the bosonic
coordinates in the target superspace have the same dimensions, with appropriate socalled fermionic -symmetry [35,66,67]. Such a formulation can accommodate curved
superspace supergravity backgrounds, whose classical superfield equations can be satisfied,
only when the action has the -invariance. The subtlety here is that there are two bosonic
coordinates involved both for the world-supervolume and the target superspace with the
same dimensionality, so that we need to avoid the confusion between them. This spirit of
having the same dimensionality both for the world-supervolume and the target spacetime
was also presented in the context of doubly-supersymmetric p-branes or L-branes [68,69],
or superembeddings [19,70].
There has already been a general formulation for such a GreenSchwarz -model action
in general even dimensions D = 2k in [36] in which the D = 2k-th rank antisymmetric
potential superfield CA1 AD is introduced, as one more generalization of super p-branes
in [35], and its -invariance has been confirmed. As a matter of fact, the generalized
VolkovAkulov action IE (2.7) with the determinant form of E m already suggests the
involvement of such maximally-ranked CA1 AD -superfield. In Ref. [36], we have seen
that a similar putative GreenSchwarz action in odd dimensions D = 2k 1 instead of
D = 2k leads to a trivial -symmetry, due to the absence of chirality projection -matrix
like 5 .
Following the results in [36], our (2k 1)-brane action in general D = 2k-dimensions
should be

Z
1
2k
gg ij ab i a j b (k 1) g
SGS = d +
2

1 i1 i2k
A1
A2k

i1 i2k CA2k A1 ,
(4.6)
+
(2k)!
where i, j, . . . = 0, 1, 2, . . . , D 1 = 2k 1 are the curved indices for the worldsupervolume with the coordinates i and the metric gij , while i A (i Z M )EM A
are the ordinary pull-backs from the target superspace with the coordinates (Z M )
(Xm , ) into the world-supervolume. The superpotential CA1 A2k has its superfield
strength HA1 A2k+1 , satisfying the Bianchi identities
1
1
[A1 HA2 A2k+2 )
T[A A | B HB|A3 A2k+2 ) 0,
(2k + 1)!
2[(2k)!] 1 2
while the supertorsion satisfies the T -Bianchi identities [64]
(4.7)

1
(4.8)
[A TBC) D T[AB| E TE|C) D R[AB|e f Mf e |C) D 0.
2
The H -Bianchi identities (4.7) at mass dimensions d = 1/2 and 1 [36] require the
conditions:
T(| H| )d1 dD1 = 0,
T[c1 |(| H|)|c2 cD ] = 0.
(4.9)
506
There are of course other Bianchi identities satisfied by other superfield strength,
depending on the supergravity multiplet for a given D and N , e.g., the fourth-rank
superfield strength FABC for D = 10, N = 2A, that we do not present here explicitly.
The superspace constraints at mass dimension d = 0 relevant to our -transformation
below are

(4.10a)
T c = 2i c ,

(4.10b)
Hc1 c2k1 = +i c1 c2k1 2k+1 ,
where 2k+1 is an analog of 5 in D = 4, and all other independent components in
HA1 A2k+1 are zero [36].
Even though our superspace constraints (4.10) look so simple, we can accommodate all
the curved supergravity effects in the system, such as the gravitino Ea a , as long as
the H - and T -Bianchi identities above, as well as those for other superfield strengths are
satisfied, with curved supertorsions, supercurvatures and superfield strengths.
As an important note to avoid confusion, we no longer use the flat case constraints
(4.2b)(4.2d), (4.5), once we have introduced the GreenSchwarz action. They will be
recovered, though, when we consider the special flat superspace limit we mention shortly.
It is easy to see that SGS has the -invariance under
1
E = (I + ) , E a = 0,
2
(1)(k1)(2k+1)/2 i1 i2k

i1 a1 i2k a2k (a1 a2k ) = (2k+1 ) ,

(2k)! g
(4.11a)
(4.11b)
following Ref. [36], whose details we do not repeat here. As in [36], the field equation of
gij
gij = ab i a j b ,
(4.12)
is algebraic, so we do not have to worry about its variation under .

In order to go back to flat superspace, we perform the identifications ,
Xm i , 15 so that SGS coincides with the generalized VolkovAkulov action (2.7), after
eliminating gij by its algebraic field equation (4.12). In fact, the s coincides with the
Es in (2.5):

S a i i a + i a i , i i i .
(4.13)
i a i a + i
In other words, in the flat target superspace, the fermionic coordinates are nothing else than
the GN-fermions [10,1517]. On the other hand, the -transformation stays formally the
same as (4.11), except that we no longer have, e.g., the gravitino in the target superspace:
Ea | a 0. Accordingly, our previous flat space constraints (4.2b)(4.2d) or (4.5)
will be recovered, due to the identification in the superspace vielbeins Em a [64].
Finally, the WZNW-term in (4.6) is shown to be equivalent to g, as follows: First,

we can identify Ca1 a2k a1 a2k C(Z) with a scalar superfield C(Z). Second, due to
15 This identification makes sense, because the dimension of the i s and that of Xm s coincide.
507
Hb1 b2k = 0, we can set C(Z) = const. Hence the WZNW-term is proportional to

[(2k)!]1 i1 i2k a1 a2k i1 a1 i2k a2k = det i a = g.
The introduction of the U (1) field strength Fij i Aj j Ai with the fundamental
vector Ai on the world-supervolume is not difficult. For this purpose, we modify SGS as
p
Z
p

1
0
= d2k
g g ij ab i a j b + Fij (k 1) g
SGS
2

1 i1 i2k
i1 A1 i2k A2k CA2k A1 ,

(4.14)
+
(2k)!
where g ij is an auxiliary field both with symmetric and antisymmetric components, playing
a role of a generalized metric, and g ij is its inverse with g det( g ij ). Since the field
equation of gij is algebraic:
gij = ab i a j b + Fij ,
we can eliminate gij to get the D-p-brane type action [10,1519]

Z

1/2
0
2k
SGS d det ab i a j b + Fij

1 i1 i2k
A1
A2k

+
i1 i2k CA2k A1 ,
(2k)!
(4.15)
(4.16)
0 has the -invariance under (4.11) and A = 0, as

also with the last WZNW-term. The SGS
i
is easily confirmed. We repeat here that these actions are valid not only in D = 10 like type
IIA or IIB superstrings [10,1217,19], but also in any arbitrary even dimensions D = 2k
where we can build supergravity backgrounds consistently in superspace [36]. 16
In this section, we have shown how to accommodate curved supergravity backgrounds
into our system, and in particular, we have seen how the generalized VolkovAkulov action
(2.7) can be interpreted as a special case of GreenSchwarz super-(2k 1)-brane action
in general D = 2k dimensions. At the point of introducing supergravity backgrounds, we
lose the universality on D, due to the satisfaction of T -Bianchi identities, which needs
consistent supergravity constraints. To put this differently, we have seen in this section
that, when the background superspace is flat, supersymmetry is global and non-linear, with
no restriction on D or N , and we can always define GreenSchwarz super p-brane actions.
However, when superspace is curved with supergravity with Lorentz covariance, there is a
restriction, such as D 6 11 by the ordinary Bianchi identities for the curved backgrounds.
4.3. Simplified (Supersymmetry)2-models

The degeneracy of the dimensionalities between the target spacetime and the worldsupervolume suggests some formulation similar to what is called (Supersymmetry)2models [37]. Namely, we expect two kinds of supersymmetries realized simultaneously in
16 This allows us to even beyond D = 2k > 12 [71,72], if we can give up Lorentz covariance.
508
the target spacetime and the world-supervolume. 17 In this sub-section, we show that such
a formulation indeed makes sense in general superspace, with the simplification that these
two supersymmetries are realized within the same spacetime dimensions. Such a sharing
of the common spacetime is possible, because of the degeneracy of the dimensionalities
between the base and target spacetime. Accordingly, we have only one set of superspace
coordinates (Z M ) (x m , ), which we temporarily call simplified (Supersymmetry)2model. 18
Our starting point is to postulate the superspace analog of our component vielbein (2.5)
by

(0) a
S a M E (0) a + (1)M i ( a ) M , (4.17a)
+ (1)M i
EM a = EM
M
(0)
= M .
E M = EM
(4.17b)
(0) A
is the ordinary flat superspace vielbein [64]:
Here EM

O
m a
a m
(0) A
(0) M
,
E
EM
a
A
i
+i m

.
(4.18)
The form for EM a in (4.17a) is a simple generalization of (2.5) into superspace, while
there might be an additional term like M in (4.17b), which is excluded here just for
simplicity. The Grassmann parity (1)M is needed in (4.17a), because of the contracted
spinorial indices on . The in (4.17) is a spinor superfield, whose non-linear
supersymmetry transformation rule is
Q = + M M ,
(4.19)
where

m = i m ,
= 0.
(4.20)
Here as in component case, m a m a , while we avoid the usage of m Ea m a .

We can also obtain the inverse matrix components perturbatively for EA M , as

(0) m
S m DA + O 4 ,
(1)A i
(4.21a)
E A m = EA
EA = A ,
O(4 )-terms
(4.21b)
O(0 )-terms.
ignored compared with the

Note the absence of up to
dependent terms in (4.21b), which is confirmed to all orders by the general method to
get an inverse of a supermatrix [59].
We can easily show that (4.19) induces the supersymmetry transformation of the
vielbeins

Q EM A = N N EM A + M N EN A ,

(4.22)
Q EA M = N N EA M EA N N M ,
17 Similar formulation was also presented for L-brane models [68,69], in which various off-shell field strengths
dual to scalars or vectors on the world-supervolume are introduced.
18 We can in principle elaborate this formulation by introducing the second set of coordinates (x 0m , 0 ), in
order to distinguish the base and target spacetimes, but we skip such a formulation in this paper.
509
due to the restriction (4.20). Here we can confirm (4.22b) only up to O(3 )-terms because
of the ignored terms in (4.21). Eq. (4.22) implies nothing else than the general coordinate
transformation of the vielbein in superspace [64].
When we consider some other superfields, such as a real scalar superfield , its
supersymmetry transformation rule is

(4.23)
Q = N N = G M ,
to comply with (4.21). It is not difficult to show that the commutator of two such non-linear
supersymmetries on or induces the desirable translations:

a
,
(4.24)
Q (1 ), Q (2 ) = +P 12
a +2i( a ) like the component case (2.3).
with 12
2
1
Note that since we are dealing with superspace, we have an additional manifest linear
supersymmetry Q0 dictated by [64]
Q0 = 0 D ,
(0)
(0)
Q0 = 0 D ,
(0)
(4.25)
(0)
with DA EA M M . The commutator of two such linear supersymmetries induces the

familiar translation operator:

(4.26)
Q0 01 , Q0 02 = +P 0a12 ,
where 0a12 +2i( 02 a 01 ) similarly to our non-linear supersymmetry (2.3). Interestingly,
by simple commutator computation on and , we see that these two sorts of
supersymmetries are commuting each other:

(4.27)
Q (), Q0 0 = 0,
as desired, with no interference.
Having seen the existence of two sorts of supersymmetries, a natural question is
whether or not these two supersymmetries are equivalent to each other after all, and if
not, how the usual theorem about maximal number of supersymmetries [73,74] in each
spacetime dimension can be avoided? Our answer to the first question is that these two
supersymmetries are not equivalent to each other connected by super Weyl rescaling (field
redefinitions), because the vielbeins (4.17) for the supersymmetry Q gives geometry in
superspace different from the flat one Q0 . As for the second question, we understand
that the usual argument on maximal number of supersymmetries is based on the linear
representation that mixes physical fields with helicity differences of 1/2. Since the nonlinear representation does not shuffles such helicities, the usual restriction on the maximal
number of extended supersymmetries seems to be avoided here. This is the reason why we
can set up two different supersymmetries at the same time, as a (Supersymmetry)2-model.
We can also postulate a possible action as a superspace generalization of IE in (2.7): 19
19 To comply with the superspace notation in [64], we use E sdet(E M ) with the inverse power compared
A
with our component convention.
510
Z
IE
e
N
d x d sdet EM
dD x dN E 1 ,
(4.28)
e 2[D/2] N is the number of the -coordinates for N supersymmetry in D,

where N
counting each component of a Majorana spinor as unity. However, we immediately notice
that such an action will produce higher-derivative terms, as the mass dimension of the integral indicates, as usual in superspace [64]. In fact, the superdeterminant is expanded
as

S a Da + O 4 ,
(4.29)
E 1 = 1 + i
with Da Ea M M . As a simple dimensional analysis reveals, some of the component
fields in have higher-order derivatives in their kinetic terms.
In order to overcome this problem, we propose two models. The first one is to introduce
some spurion scalar superfields [75] that is to be inserted into the above Lagrangian:
Z
Z

e
N
D
1
(4.30)
IE = d x d E = dD x id0 a Da + O 4 ,
with
e
= N d0 ,
(4.31)
where d0 6= 0 is an analog of the D-component in D = 4 for the highest sector component

field of , and we identify | . The kinetic term in (4.30) is physical with no negative
energy ghosts or higher-derivatives involved. Even though this first model has the drawback
due to the explicit breaking of both supersymmetries, it may still be of some mathematical
interest, because of the coexistence of two sorts of supersymmetries. We expect more
improved models to be developed in order to avoid such explicit breakings.
Our second model is much more advanced. We introduce a constraint action with a
Lagrange multiplier superfield LMN for a bilinear combination of [76]:
Z

e 1
(4.32)
IL2 dD x dN E 1 Lab Db Da ,
2
where L has no (anti)symmetry other than Lab = Lba . Accordingly, the L
transforms in the standard fashion:

Q Lab = M M Lab = G M Lab ,
Q0 Lab = 0 D(0) Lab ,
(4.33)
so that the total action I IE + IL2 is invariant under both Q and Q0 . Notice here that
the derivative in IL2 deletes the effect of the inhomogeneous term in the transformation of
(4.19). Clearly, the superfield equation for L from the total action I implies the vanishing
of Da :
I = Db

Da = 0
(4.34)
H Da = 0,
L
because all the indices a, , b, are free. This superfield equation preserves both the nonlinear Q and linear Q0 supersymmetries. In particular, the effect of the inhomogeneous
ab
511
term in (4.19) disappears under the derivative. On the other hand, the -superfield equation
is

1
EM
I = E 1 (1)M
E M + E M L cd Dd Dc
2

1 ab
Db + O 5 .
(4.35)
Da E L
Since Da = 0 on-shell, each term in (4.35) vanishes, and there arises no problem of
higher-derivative kinetic term for . As usual in this sort of formulation, Lab does not
enter any superfield equations effectively, once (4.34) is satisfied. Even if we add other
matter actions to the above total action I , the NG-fermion is frozen with no dynamics.
The effective disappearance of the multiplier Lab from the set of physical field
equations suggests the existence of appropriate gauge transformation that can eliminate
Lab [76]. In fact, it is easy to show that the total action I = IE + IL2 is invariant up to
O(4 )-terms under the local -symmetry:
= b Db ,

bc 1

ab
a b
a
c ab
= 2i
Dc L
+ Dc L
L
2

2
( a b) + O ,
(4.36)
where a is a local vectorial parameter.

This second model (4.32) with the constraint action IL2 is more complicated than the
first one (4.30), due to the supersymmetric invariance maintained at the action as well as
the superfield equation level. Similarly to the component case in Section 2 for ordinary
spacetime, the coexistence of two sorts of superspaces is common to these two models,
one with the usual flat superspace metric, and the other with non-vanishing curvature with
non-trivial vielbeins. To our knowledge, these models here are the first examples of this
kind with the realization of two different supersymmetries at the same time. In particular,
in the second model, both of them are exact at the action level.
We draw readers attention to the point that our formulation with the unbroken non-linear
supersymmetry is distinguished from other formulations with the non-linear realization as
a manifestation of spontaneously broken linear supersymmetry, such as in Refs. [2529].
The interesting aspect of our formulation is that both non-linear and linear supersymmetries
coexist in superspace with no mutual inconsistency, that was not emphasized before.
5. Concluding remarks
In this paper we have first studied non-linear realization of N -extended global
supersymmetry in arbitrary dimensions D, and we have given actions invariant under
non-linear supersymmetry transformations. These actions and transformations are valid
in arbitrary dimensions D and for arbitrary N , including of course D > 12, or N > 9
in D = 4. This is because the vielbein field (2.5) transforms like the general coordinate
transformation with the parameter i( ) under the non-linear supersymmetry,
512
even though supersymmetry is global. The action IE has non-trivial features, because of
the kinetic term of the NG-fermion, with infinitely many interaction terms, despite of its
superficially simple form. Even though it seems almost straightforward to generalize the
original VA action [1] to D and N , these generalizations will be of importance for later
sections.
These formulations have no need for the bosonic superpartner fields that are indispensable in other formulations, such as in [2527], independent of D or N . This feature is very
similar to the original model by VolkovAkulov [1] with the single NG-fermion. Reviewing also the transformation rule given in [10], we notice that the vector field does not enter
the transformation rule of the NG-fermion, and this signals the decoupling of the vector
field from the NG-fermion system. From this viewpoint, it is natural that such a vector
field can be completely decoupled from the NG-fermion system, in some choice of frame
without spoiling global supersymmetry.
For linear representation of supersymmetry, interacting models are possible only
D 6 11, unless Lorentz symmetry is broken, such as in those formulations in [71,72].
Therefore it has been believed that any putative interacting supersymmetric model in
D > 12 with Lorentz covariance is doomed to fail. The result in this paper have
broken this taboo, namely we have seen that it is certainly possible to formulate global
non-linear supersymmetry in D > 12, as long as spinor fields can be defined. In other
words, global supersymmetry and Lorentz symmetry can coexist for any given D and N ,
only at the expense of the linear realization of local supersymmetry. We now know that
interacting models with non-linear realization of global supersymmetries are possible even
for N > 9 in D = 4, or N > 2 in D = 11, or in D > 12. Since the non-linear realization of
supersymmetry is possible even in D > 12 with global Lorentz symmetry, we do not take
the standpoint that non-linear supersymmetry is a reflection of broken supersymmetries
[2527].
Although it emerged just as an interesting by-product, the supersymmetric BornInfeld
action for a non-Abelian gauge group for D and N we gave is a new result in this
paper. Like other actions, this action also has non-trivial but consistent interactions with
the NG-fermion, and it agrees in the special case of D = 10 with the result in [10] for
the lowest-order terms. When the NG-fermion carries the adjoint index of the non-Abelian
group, the lowest-order terms of our BornInfeld action (3.6) coincide with those in the
linearly supersymmetric action [10] in D = 10. This seems to imply that our BornInfeld
action (3.6) provides an equally important foundation for the M(atrix)-theory [4144]
as the linearly supersymmetric BornInfeld action [10] for D-branes [1117,19], up to
some recently-discovered subtlety at higher orders [46]. Interestingly enough, since we
have dim G-extended supersymmetry, the usual large N limit for G = U (N ) [4144]
corresponds to N = N 2 , i.e., infinitely many (0 ) extended supersymmetry limit
for our non-linear realization.
Additionally, we have shown how to perform duality transformation from an Abelian
field strength F in a BornInfeld (type) action into its Hodge-dual field strength
N1 D2 in D 6= 4 and N , excluding the case of D = 4, in which certain subtlety arises
related to the SL(2, R) duality [6163].
513
As another important application, we have pointed out that our supersymmetric Born
Infeld action INASBI is a suitable matter action compatible with ChernSimons actions,
in which there is no fundamental metric from the outset. The gauge groups can be even a
supergroup such as OSp(N|m), etc. Another important aspect is the fact that any Chern
Simons action (3.16) for any supergroup formulated in D = 2n + 1 [4852] has hidden
global supersymmetry under (3.1b) realized non-linearly. This may well provide the longstanding missing link between the ChernSimons theories of supergroup and Poincar
supergravity with diffeomorphism [59], especially in the context of the superalgebra in
M-theory [4952,58].
In our formulation, we found that the minimal field content for supersymmetric Born
Infeld action is the NG-fermion and a vector field, both for Abelian and non-Abelian
cases. If we perform simple dimensional reduction from the starting dimensions D into
one dimension lower: D 1, then one component of the vector field becomes a scalar
field, so that more and more scalars are generated, if we go further down [10]. However,
this is not the only scenario we can think of, because in our formulation, at each dimension
D we can formulate a supersymmetric BornInfeld action with the minimal set of fields,
i.e., one NG-fermion and a vector field, but no scalars or any other fields. This is also one
of the distinctive features of our formulation compared with other D-brane formulations
[10,1214].
Our result of extended supersymmetries on N has other analogs. For example, similar
situations have been already presented for infinitely many local supersymmetries in D = 3
[77] or arbitrarily many global supersymmetries in D = 1 [78]. These so-called 0 aspects
of supersymmetries have already shown up in many different contexts in the past, and now
we have explicit models at hand that elucidate such universality more explicitly in D.
We have also reformulated the component results in superspace, and in particular,
confirmed the Bianchi identities for a second-rank superfield strength for a non-Abelian
group and a third-rank superfield strength, as explicit examples. We have seen that the
components F I or G are needed, and the special factor P b i( b ) is involved
in a peculiar way, as a generalization of those given in [1517]. Our constraints have
essentially off-shell structure, namely the NG-fermion equations are not implied by the
Bianchi identities, as opposed to the usual on-shell superspace formulations in higherdimensions, or constrained superfield formulations in D = 4 [2,2527]. Remarkably, these
constraints are valid for D and N , which, to our knowledge, have not been emphasized
in the literature.
As another important result, we have presented an universal GreenSchwarz -model
super (2k 1)-brane action [35,36] in D = 2k-dimensions that can accommodate even
curved superspace backgrounds, when we need to include supergravity, by introducing
the maximally antisymmetric potential superfield Ca1 a2k [36]. The requirement of
satisfaction of T -Bianchi identities for supergravity with Lorentz covariance restricts the
dimensionality D to be D = 2k (1 6 k 6 5).
This (2k 1)-brane formulation suggests the significance of the target spacetime whose
dimensionality coincides with that of the world-supervolume. Motivated by this, we have
given a simplified (Supersymmetry)2-models as another by-product of our formulations,
514
in which linear and non-linear representations of supersymmetries coexist at the same time.
This is a simplified version of more general (Supersymmetry)2-models, initiated in 2D
[37]. We have seen that the original ansatz of forming a vielbein in terms of NG-fermion is
directly generalized into superfields, and we have also given some explicit models with the
simultaneous realization of both linear and non-linear supersymmetries. In particular, the
second model has both of these supersymmetries explicit at the action level. In this sense,
our formulation is distinguished from others, such as Refs. [2529], in which non-linear
supersymmetry is regarded as spontaneous breaking of linear supersymmetry.
We saw that neither the metric nor the vielbein field in our formulation is an elementary
field, but they are effectively composite in terms of the NG-fermion. From this viewpoint,
our theory is similar to the Bars and MacDowell theory [79] in which the metric tensor
is not an elementary field, but is a composite bilinear form of gravitino which is the
fundamental field. The main difference, though, is that the formulation in [79] contains a
gravitino as a gauge field for the local supersymmetry, while in our theory supersymmetry
is global but is realized non-linearly. Most of our actions in general spacetime dimensions
are only globally supersymmetric in flat D dimensions, except the super (2k 1)-brane
actions which can accommodate curved superspace backgrounds.
One might say that a possible drawback in our formulation is the loss of uniqueness
of actions for particle physics. This is because there are essentially no restrictions on
possible invariant Lagrangian by supersymmetry, as long as the general covariance G
is maintained in terms of our metric or vielbein for D and N . However, we instead
stress the importance of universality of supersymmetry more than uniqueness of a
particle theory. Our result here opened a completely new avenue, i.e., supersymmetry
does not necessarily restrict the spacetime dimensions for possible interacting models for
physics. To our knowledge, the significance of such universality has have been overlooked,
or at least not emphasized during the decades since the first work in [1].
One final point to be stressed is the coexistence of two sorts of spacetimes or even
superspaces simultaneously, which are not equivalent to each other: One with flat metric,
while the other with highly non-trivial vielbeins with general coordinate transformations.
We have seen this for component as well as superspace formulations. However, this is in
a sense puzzling, because it is well-known in mathematics to be impossible to embed a
curved spacetime into a flat spacetime of the same dimensionality. Our solution to
this puzzle is that the curved spacetime is non-perturbatively realized from the flat
spacetime, by adding up infinite series in terms of NG-fermions.
Moreover, the coexistence of two different superspaces is also related to the coexistence
of two sorts of supersymmetries, one in linear and the other in non-linear representations.
These aspects of non-linear supersymmetry may well be different manifestations of the
intricate feature of M-theory [2023] for curved supergravity in D = 11 realized first as a
global supersymmetry in flat spacetime, via a simple matrix theory [4144].
We believe that the results given in this paper can provide a first step into the studies
of more general features of non-linear supersymmetries in D and N , motivated also
by the recent development of D-branes [11]. We have even seen that the non-linear
supersymmetry can coexist with linear supersymmetry as simple (Supersymmetry)2-model
515
in superspace. We have also seen that the non-linear realization as compared with the linear
realization of supersymmetry has more freedom for accommodating a much wider class of
models.
Acknowledgements
We are grateful to J. Bagger, S.J. Gates, Jr., F. Gonzalez-Rey, M. Luty, M. Rocek,
J.H. Schwarz and W. Siegel for helpful discussions. Special acknowledgement is for
S.J. Gates, Jr., for many suggestions to improve the paper.
Note added
After completing this paper, we have encountered a recent work by P.C. West [80],
in which superembedding [19] for non-linear realizations is performed by enlarging the
automorphism group of supersymmetry. Also, after the acceptance of this paper for
publication, we received a correspondence from the author of [81] that his paper also
dealt with a similar subject of non-linear supersymmetry for the supergroup OSp(N; 4)
in D = 4.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
D.V. Volkov, V.P. Akulov, JETP Lett. 16 (1972) 438; Phys. Lett. B 46 (1973) 109.
M. Rocek, Phys. Rev. Lett. 41 (1978) 451.
S. Ferrara, L. Maiani, P.C. West, Zeit. Phys. C 19 (1983) 267.
S. Ferrara, Problems in unification and supergravity, in: Proceedings of La Jolla Unification
Workshop, 1983.
M. Green, J.H. Schwarz, E. Witten, Superstring Theory, Vol. I and II, Cambridge University
Press, 1987.
S. Cecotti, S. Ferrara, Phys. Lett. B 187 (1987) 335.
M. Born, L. Infeld, Proc. Roy. Soc. (London) A 144 (1934) 425.
A.A. Tseytlin, BornInfeld action, supersymmetry and string theory, hep-th/9908105; in:
M. Shifman (Ed.), Y. Golfand Memorial, World Scientific, 2000.
E. Bergshoeff, M. Rakowski, E. Sezgin, Phys. Lett. B 185 (1987) 371.
M. Aganagic, C. Popescu, J.H. Schwarz, Phys. Lett. B 393 (1997) 311; Nucl. Phys. B 495
(1997) 99.
J. Polchinsky, TASI lectures on D-branes, hep-th/9611050.
I.A. Bandos, D.P. Sorokin, M. Tonin, Nucl. Phys. B 497 (1997) 275.
R. Kallosh, VolkovAkulov theory and D-branes, hep-th/9705118.
V.A. Akulov, I. Bandos, W. Kummer, V. Zima, Nucl. Phys. B 527 (1998) 61.
M. Cederwall, A. von Gussich, B.E.W. Nilsson, A. Westerberg, Nucl. Phys. B 490 (1997) 163.
M. Cederwall, A. von Gussich, B.E.W. Nilsson, P. Sundell, A. Westerberg, Nucl. Phys. B 490
(1997) 179.
T. Adawi, M. Cederwall, U. Gran, M. Holm, B.E.W. Nilsson, Int. J. Mod. Phys. A 13 (1998)
4691.
E. Bergshoeff, P. Townsend, Nucl. Phys. B 490 (1997) 145.
516
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
D. Sorokin, Phys. Rep. 329 (2000) 1.

C. Hull, P.K. Townsend, Nucl. Phys. B 348 (1995) 109.
E. Witten, Nucl. Phys. B 443 (1995) 85.
P.K. Townsend, Four lectures on M-theory, in: Proceedings of ICTP Summer School on High
Energy Physics and Cosmology, Trieste, June, 1996; hep-th/9612121.
P.K. Townsend, M-theory from its superalgebra, hep-th/9712004, NATO Advanced Study
Institute, Cargese, May 1997.
A.A. Tseytlin, Nucl. Phys. B 501 (1997) 41.
J. Bagger, J. Wess, Phys. Lett. B 138 (1984) 105.
J. Hughes, J. Polchinski, Nucl. Phys. B 278 (1986) 147.
J. Bagger, A. Galperin, Phys. Rev. D 55 (1997) 1091.
E.A. Ivanov, A.A. Kapustnikov, Nucl. Phys. B 333 (1990) 439; Phys. Lett. B 143 (1984) 379;
J. Phys. G 8 (1982) 167; J. Phys. A 11 (1978) 2375.
T. Uematsu, C. Zachos, Nucl. Phys. B 201 (1982) 250.
U. Lindstrom, M. Rocek, Phys. Rev. D 19 (1979) 2300.
A.A. Kapustnikov, Theor. Math. Fiz. 47 (1981) 198.
S. Samuel, J. Wess, Nucl. Phys. B 221 (1983) 153; Nucl. Phys. B 233 (1984) 488.
T.E. Clark, S.T. Love, Phys. Rev. D 39 (1989) 2391; Phys. Rev. D 54 (1996) 5723.
T.E. Clark, T. Lee, S.T. Love, G.-H. Wu, Phys. Rev. D 57 (1998) 5912.
A. Achucarro, J.M. Evans, P.K. Townsend, D.L. Wiltshire, Phys. Lett. B 198 (1987) 441.
H. Nishino, Phys. Lett. B 457 (1999) 51.
S.J. Gates Jr., H. Nishino, Class. Quant. Grav. 3 (1986) 391.
T. Kugo, P.K. Townsend, Nucl. Phys. B 211 (1983) 157.
C. Vafa, Nucl. Phys. B 469 (1996) 403.
I. Bars, Phys. Rev. D 55 (1997) 2373.
T. Banks, W. Fischler, S.H. Shenker, L. Susskind, Phys. Rev. D 55 (1997) 5112.
A. Bilal, Fortschr. Phys. 47 (1999) 5.
T. Banks, TASI lecture note on matrix theory, hep-th/9911068.
W. Taylor IV, The M(atrix) model of M-theory, hep-th/0002016, and references therein, NATO
Advanced Study Institute, Akureyri, Iceland, Aug. 1999.
J. Wess, B. Zumino, Nucl. Phys. B 70 (1974) 39.
I.L. Buchbinder, S.M. Kuzenko, A.A. Tseytlin, On low-energy effective actions in N = 2, 4
superconformal theories in four dimensions, hep-th/9911221.
Z. Lalak, S. Lavignac, H.P. Nilles, Target-space duality in heterotic and type I effective
Lagrangians, hep-th/9912206.
A.H. Chamseddine, Phys. Lett. B 233 (1989) 291; Nucl. Phys. B 346 (1990) 213.
M. Baados, R. Troncoso, J. Zanelli, Phys. Rev. D 54 (1996) 2605.
R. Troncoso, J. Zanelli, Phys. Rev. D 58 (1998) 101703.
R. Troncoso, J. Zanelli, Higher dimensional gravity and local anti-de Sitter symmetry, hepth/9907109.
C.R. Preitschopf, T. Hurth, P. van Nieuwenhuizen, A. Waldron, Nucl. Phys. B Proc. Suppl. 56
(1997) 310.
M.M. Vainberg, Variational Methods for the Study of Non-Linear Operators, Holden Day, San
Fransisco, 1964.
S.J. Gates Jr., H. Nishino, Phys. Lett. B 173 (1986) 46.
H. Nishino, S.J. Gates Jr., Int. J. Mod. Phys. A 8 (1993) 3371.
P. Horava, Phys. Rev. D 59 (1999) 046004.
M. Baados, Nucl. Phys. B Proc. Suppl. 88 (2000) 17.
P. Mora, H. Nishino, Phys. Lett. B 482 (2000) 222.
P. van Nieuwenhuizen, Phys. Rep. C 68 (1981) 189.
H. Nicolai, P.K. Townsend, Phys. Lett. B 98 (1981) 257.
517
[61] G.W. Gibbons, D.A. Rasheed, Nucl. Phys. B 454 (1995) 185; Phys. Lett. B 365 (1996) 46.
[62] M.K. Gaillard, B. Zumino, Non-linear electromagnetic self-duality and Legendre transformations, in: P.C. West, D. Olive (Eds.), Duality and Supersymmetric Theories, Cambridge University Press, hep-th/9712103.
[63] D. Brace, B. Morariu, B. Zumino, Duality invariant BornInfeld theory, hep-th/9905218.
[64] S.J. Gates Jr., M.T. Grisaru, M. Rocek, W. Siegel, Superspace, Benjamin/Cummings, Reading,
MA, 1983.
[65] S.J. Gates Jr., S. Vashakidze, Nucl. Phys. B 291 (1987) 172.
[66] L. Brink, J.H. Schwarz, Phys. Lett. B 100 (1981) 310.
[67] W. Siegel, Class. Quant. Grav. 2 (1985) L95.
[68] P.S. Howe, E. Sezgin, Phys. Lett. B 390 (1997) 133.
[69] P.S. Howe, O. Raetzel, I. Rudychev, E. Sezgin, Class. Quant. Grav. 16 (1999) 705.
[70] P.S. Howe, E. Sezgin, P.C. West, hep-th/9705093; Contributed to Kharkov 1997, Supersymmetry and Quantum Field Theory, p. 64.
[71] H. Nishino, E. Sezgin, Phys. Lett. B 388 (1996) 569.
[72] H. Nishino, Phys. Lett. B 428 (1998) 85; Phys. Lett. B 437 (1998) 303; Phys. Lett. B 452 (1999)
265; Nucl. Phys. B 523 (1998) 450; Nucl. Phys. B 542 (1999) 217.
[73] W. Nahm, Nucl. Phys. B 135 (1978) 149.
[74] R. Haag, J. opuszanski, M.F. Sohnius, Nucl. Phys. B 88 (1975) 61.
[75] L. Girardello, M.T. Grisaru, Nucl. Phys. B 194 (1982) 65.
[76] W. Siegel, Nucl. Phys. B 238 (1984) 307.
[77] H. Nishino, S.J. Gates Jr., Nucl. Phys. B 480 (1996) 573.
[78] S.J. Gates Jr., L. Rana, Phys. Lett. B 352 (1995) 50; Phys. Lett. B 369 (1996) 269.
[79] I. Bars, W.S. MacDowell, Phys. Lett. B 71 (1977) 111.
[80] P. West, Automorphisms, nonlinear representations, and branes, hep-th/0001216.
[81] J. Lukievski, Phys. Lett. B 121 (1983) 135.

Peeling and multi-critical matter

coupled to quantum gravity
Martin G. Harris 1 , John F. Wheater
Department of Physics, University of Oxford, Theoretical Physics, 1 Keble Road, Oxford OX1 3NP, UK
Received 9 December 1999; revised 16 May 2000; accepted 3 July 2000
Abstract
We show how to determine the unknown functions arising when the peeling decomposition is
applied to multi-critical matter coupled to two-dimensional quantum gravity and compute the loop
loop correlation functions. The results that = 2 + 2/(2K 3) and = 1 3/2K agree with the
slicing decomposition, and satisfy Fisher scaling. 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.60.-m; 04.60.Kz; 04.60.Nc
Keywords: Multi-critical matter; Quantum gravity; Hausdorff dimension
1. Introduction
One of the outstanding problems in the theory of two-dimensional quantum gravity
is the effect of matter fields on the Hausdorff dimension. In models of discretized twodimensional quantum gravity we define the grand canonical partition function for an
ensemble of graphs (which for the moment we assume are triangulations) G by
X
e|G| wG ,
(1)
Z() =
GG
where |G| denotes the number of triangles in G, and wG the partition function of any matter
fields in the theory on the graph G (for an introduction to this material see for example
[1]). To define the Hausdorff dimension [2,3] we first define the geodesic distance dG (i, j )
between two links i and j as the minimum number of triangles which must be traversed to
E-mail addresses: martin.harris@mathengine.com.uk (M.G. Harris), j.wheater1@physics.ox.ac.uk

(J.F. Wheater).
1 Present address: MathEngine PLC, The Oxford Centre for Innovation, Mill Street, Oxford OX2 0JX.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 3 2 - 6
M.G. Harris, J.F. Wheater / Nuclear Physics B 586 (2000) 518546
519
get from the centre of one link to the centre of the other. Then we introduce the two-point
function
X
X

e|G| wG
dG (i, j ) r .
(2)
H(r, ) =
GG
i,j G
We expect that H has the asymptotic behaviour [3,4]

H(r, ) em()r ,
r 1g ,
m()r 1,
m()1 r 1,
(3)
where, as c , the mass gap vanishes as

m() ( c )g .
(4)
In general it is also convenient to consider a more general correlation function between

boundary loops of length l1 and l2 ; (2) is essentially the correlator for minimum length
loops. Note that it follows from (2) that
X
X
H(r, ) =
e|G| |G|2 wG ( c )str ,
(5)
r
GG
where, in unitary theories, str is the string susceptibility exponent, and inserting the
form (3) we conclude that
g (2 g ) = str
(6)
which is the Fisher scaling relation. At least in unitary theories the Hausdorff dimension
dH is given by dH g = 1 and has the geometrical meaning that in the continuum limit the
average volume is related to the geodesic size by hV i R dH [1].
Analytic calculations of the scaling behaviour of the correlation functions (2) were
first done by means of the slicing decomposition introduced by Kawai et al. [2] and then
somewhat later Watabiki [5] introduced the peeling decomposition. For pure gravity (i.e.,
wG = 1, str = 1/2) both peeling and slicing decompositions keep track of the geodesic
distance and give the same results which tell us directly that the Hausdorff dimension of
the ensemble is 4.
When matter fields are introduced the situation becomes more complicated. The time
scale, usually called the string time t, introduced in the decompositions that have been
formulated is no longer by construction the geodesic distance, nor indeed are the time
scales for different decompositions necessarily equivalent. However the above discussion
of correlation functions can be repeated in terms of the string time t instead of the geodesic
distance r leading to another pair of exponents, and , which are also expected to satisfy
the Fisher scaling relation. For example in the c = 2 model the scaling with string time
has been calculated completely by the peeling decomposition [6] with the result that = 12
which would imply that dH = 2 if the string time and geodesic distance are proportional.
In fact high precision numerical calculations [7] find dH = 3.58 0.04 in agreement with
the formula [8,9]
25 c + 49 c
(7)
dH = 2
25 c + 1 c
520
derived using scaling arguments for diffusion in Liouville theory. For unitary matter
complete calculations have not been made but it seems that = | |str /2 [10]; the implied
value of dH is in contradiction with the results of numerical simulations which suggest that
dH is close to 4 [11] but are not in particularly good agreement with (7) either. It seems
certain that when matter is present the string time and the geodesic distance have different
scaling dimensions but the relation between them is unknown.
In this paper we will be concerned with the matrix models with a critical point
corresponding to the (p, q) = (2, 2K 1), K = 2, 3, 4, . . . , multi-critical models coupled
to quantum gravity [1217]. For K > 2 the ensemble of graphs G now allows polygons
with {4, . . . , 2K} sides and there are K independent coupling constants, with polygons of
order {6, 10, 14, . . .} having negative weights which makes the models non-unitary. The
partition function is again defined by
X
e|G| wG ,
(8)
Z() =
GG
where now G denotes the number of polygons, is the coupling constant conjugate to the
number of polygons, and wG depends on the remaining K 1 coupling constants of the
theory. K = 2 corresponds to pure gravity; the second coupling constant which appears
in this matrix model is conjugate to the length of the boundary of the graph. Since there
are now in general many couplings, there are many correlation functions which are second
derivatives with respect to the couplings so there are many susceptibilities. We define the
susceptibility

2 Z() X |G| 2
=
e
|G| wG ( c ) ,
2
(9)
GG
where the exponent is known to take the value K 1 at the multi-critical point. The
string time, tG (i, j ), separating two links is now defined as the minimum number of
polygons which must be traversed to get from the centre of one link to the centre of the
other and the two-point function is defined as
X
X
e|G| wG
(tG (i, j ) t).
(10)
H(t, ) =
GG
i,j G
We expect (and shall confirm) that, if all the couplings except are set to their values at
the multi-critical point, H has the asymptotic behaviour
H(t, ) em()t ,
t 1 ,
m()t 1,
m()1 t 1,
(11)
where, as c , the mass gap vanishes as

m() ( c ) ,
(12)
and that the Fisher scaling relation (which may be obtained by similar manipulations as
before)
(2 ) =
(13)
521
is satisfied. One could hold that t is still the geodesic distance but this is slightly
problematic for large K; it implies that all sides of a given polygon, no matter how many
sides it has, are separated from one another by geodesic distance 1 and we shall argue
in Section 6 that the continuum limit of t is not a sensible continuum geodesic distance.
These models have been analyzed using the slicing decomposition in the same way as
pure gravity [18] and they have also been considered using the peeling decomposition
in the scaling limit but the discretized equations have not been solved completely [5,19].
In this paper we will examine their peeling decomposition in detail and explain how to
solve completely the non-trivial differential equations that arise. We have two motivations
for this; to check whether the results are indeed the same as for slicing, and the intrinsic
interest of the method of solution.
This paper is structured as follows. In Section 2 we briefly describe the standard peeling
calculation for pure gravity and then derive the evolution equation for the multi-critical
models. In Section 3 we consider the K = 2 case and show that it always gives the standard
pure gravity results. Then in Section 4 we show in detail how to calculate the exponent
for K = 4 and describe how the calculation extends to all higher even K. In Section 5 we
explain how to calculate for all even K and in Section 6 we give our conclusions.
2. The peeling decomposition and evolution equations

We start by reviewing the calculation in [5] for the simplest pure gravity model which
has matrix model potential
1
1
U () = 2 g 3 .
2
3
(14)
The matrix model generates the dual graphs to the triangulations G (see Eq. (1)) with
g = e and wG = 1. The SchwingerDyson equation for connected Greens functions
is obtained by marking one external line and pulling it out to expose the vertex to
which it is attached [20], see Fig. 1. In the peeling decomposition we assign a time
variable to this process; a single iteration advances t by an amount 1/n so that we
obtain
An (t + 1/n) = n,2 + gAn+1 (t) + g
n
X
Am (t)Anm+1 (t).
m=1
Fig. 1. The SchwingerDyson equation for the potential (14).
(15)
522
We are interested in the loop-loop correlation function; suppose for the moment that at
t = 0 the entry loop is a one-loop and form the quantity
Gn (t) =
An (t)
A1 (0)
(16)
which is the amplitude for an exit n-loop at time t given an entry 1-loop at time t = 0.
Differentiating (15) we obtain
Gn (t + 1/n) = gGn+1 (t) + 2g
n
X
Am (t)Gnm+1 (t).
(17)
m=1
If we restrict to spherical topology then Am (t) may be replaced by the disk amplitude
for m legs, Am , because the branch can never rejoin the main tube (see Fig. 2). The
next step is to approximate the time by a continuous variable to obtain the evolution
equation
X
1 dGn (t)
= gGn+1 (t) + 2g
Am Gnm+1 (t)
Gn (t) +
n dt
n
(18)
m=1
with the initial condition that

An (0)
= n,1 .
Gn (0) =
A1 (0)
(19)
Note that by differentiating (18) we can show iteratively that all the derivatives of Gn (t) at
t = 0 are finite. Defining the generating function
G(t, x) =
x n Gn (t)
(20)
n=1
Eq. (18) becomes

F (x)G
G
=x
,
t
x
x
(21)
where
F (x) = 2gA(x) + g x,
Fig. 2. The two-loop function in spherical topology.
(22)
523
and the form of the disk amplitude

A(x) =
An x n
(23)
n=1
is known [10]. Note that the function F (x) contains just the universal scaling part of the
amplitude A(x); it is given by
F (x) = (f x)(1 8fg 4gx)1/2 ,
(24)
where f is the root of the cubic

(1 8fg)f 2 g 2 = 0
(25)
which is positive and vanishes as g 0. It is then straightforward to solve the evolution

equation which gives

x
U t + J (x) ,
(26)
G(t, x) =
F (x)
where
1
dJ
=
dx
F (x)
(27)
and the function U (y) is fixed by the initial conditions to satisfy

1=

1
U J (x) .
F (x)
(28)
This leads to the scaling behaviour at large y

U (y) ' ( c )3/4
cosh( c )1/4 y
sinh3 ( c )1/4y
(29)
where we have suppressed various constant factors; so we deduce = 1/4 and = 4 in

agreement with the Fisher scaling relation [3].
Now we turn to the multi-critical models for which we will use the notation of Ref. [21].
The potential for the Kth multi-critical model is given by
U () = 2 +
K
X
gpK 2p ,
(30)
p=1
where the couplings at the multi-critical point are

gpK = (1)p1
1
K!(p 1)!
p,1 .
(K p)!2p! 2
(31)
This time we will deal with disconnected graphs. The evolution equation follows from
the SchwingerDyson equation, shown in Fig. 3, just as in the pure gravity case; we obtain
the evolution equation
An (t + 1/n) =
n2
X
j =0
Anj 2 (t)Aj (t) 2
K
X
k=1
kgkK An+2k2 (t).
(32)
524
Fig. 3. The SchwingerDyson equation for disconnected graphs of the multi-critical potential (30).
Proceeding as before, but assuming that the entry loop is an m-loop and defining
An (t)
,
Gn,m (t) =
Am (0)
we find
n2
K
X
X
1 dGn,m (t)
=2
Gj,m (t)Anj 2
2kgkK Gn+2k2,m (t)
Gn,m (t) +
n
dt
j =0
(33)
(34)
k=1
with the initial condition that

Gn,m (0) = n,m .
(35)
Again, note that by differentiating (34) we can show iteratively that all the derivatives of
Gn,m (t) at t = 0 are finite.
Defining the generating function
Gm (x, t) =
x n Gn,m (t)
(36)
n=1
we obtain the partial differential equation

(
!)
K
X
2kgkK
Gm
2
=x
Gm 2x A(x) 1
t
x
x 2k2
k=1
K
X
k=1
2kgkK
2k3
X
Gj,m (t)(2k 2 j )x j 2k+2 .
(37)
j =1
At this stage we should make several remarks. Firstly that one reason for dealing with
the disconnected graphs is technical convenience; the required disk amplitudes are known
and take a simple form [21], and the structure of the evolution equations is similar to the
3 case. However it is also very clear that because the graph ensemble is disconnected
there is no direct correspondence between the string time t and a geodesic distance the
latter can only be sensibly defined on connected graphs. These equations are more difficult
to solve than the pure gravity example reviewed above because of the presence of the a
priori unknown functions Gj,m (t) which have to be determined by the self-consistency
and analyticity properties of the solutions. We will work our way through the problem in a
number of steps. First we will study the solution in the K = 2 case and show that it gives
the standard pure gravity results.
525
3. Universality in the K = 2 case

It is simpler to work at the multi-critical points initially and to compute the exponent
directly at the critical point; we leave the exponent for Section 5. For the K = 2 multicritical model we have

Gm
=x
Gm 2x 2 A(x) 1 2g12 4g22 x 2 4g22 x 1 G1,m (t) ,
(38)
t
x
where g12 = 1/2, g22 = 1/12 and A(x) is given by

3/2
1
1
1 .
(39)
A(x) = 2 + 4 1 4x 2
x
6x
Thus Gm satisfies

Gm (1 4x 2 )3/2
x 1 Gm
1
=
(40)
+ 2 G1,m (t)
2
t
x
3x
3x
with the initial condition
Gm (0, x) = x m .
(41)
The potential (30) is an even function of the fields and therefore Gn,m (t), which is the
amplitude for an entrance m-loop and exit n-loop, can only be non-zero if n and m are
both odd or both even. Thus if m is odd there is an unknown function on the r.h.s. of (40)
whereas if m is even there is no such problem; we will consider the even and odd cases
separately.
3.1. Even m
When m is even G1,m (t) = 0 and we can solve (40) immediately to obtain

x
U t + J (x) ,
Gm (t, x) =
F (x)
where
dJ
1
(1 4x 2 )3/2
,
=
,
3x
dx
F (x)
1/2
3
1 4x 2
1 ,
J (x) =
4
and the function U (y) is determined by the initial condition (41)

x
U J (x) .
xm =
F (x)
Thus we obtain
1
U (y) 3
y
for large y and hence
1
Gn,m (t) 3
t
for even n and m so that = 4 as expected for all these amplitudes.
(42)
F (x) =
(43)
(44)
(45)
(46)
526
3.2. Odd m
When m is odd G1,m (t) 6= 0 and the presence of the unknown function on the r.h.s. of
(40) complicates matters; however this is more typical of the general multi-critical models
than the even m case and so we shall study it in some detail. Although the differential
equation can of course still be solved in the t domain it is more convenient to work with
the Laplace transformed correlation functions
Sn,m (s) =
G
Gn,m (t)est dt,
Sm (s, x) = 1
G
x
Gm (t, x)est dt.
(47)
Taking the Laplace transform of (40) we obtain the equation

Sm (s, x)F (x) + 1 G
S1,m (s).
Sm (s, x) x m1 = G
sG
x
3x 2
Integrating this differential equation we find
(48)
!

Zx
S1,m (s)
S1,m (s)
1
s
G
G
sJ
(x)
sJ
(y)
m1
Sm (s, x) =
e
dy e
y
. (49)
G
F (x)
3x
(1 4y 2)3/2
0
Sm (s, x) has a power series expansion in x

The function G
X
Sn,m (s)
Sm (s, x) =
x n1 G
G
(50)
which we expect from (49) and (43) to have finite radius of convergence 1/2; within
the radius of convergence the coefficients are the Laplace transforms of the correlation
functions. Unless the Gn,m (t) grow faster than exponentially at large t, something
Sn,m (s) will have an asymptotic series
we do not expect, their Laplace transforms G
representation at large s. This series can be obtained by successive integration by parts of
the definition (47); as we observed in Section 2, Gn,m (t) and all its (finite order) derivatives
are finite at t = 0 so we obtain the formal series
X k
Sn,m (s) = 1
, 0 = n,m .
(51)
G
s
sk
k=0
(Of course the k depend on n and m but we will always suppress such dependence
for clarity.) We will now show that imposing the condition that the coefficients in the x
Sm (s, x) behave like (51) at large s is sufficient to fix the unknown function
expansion of G
S
G1,m (s).
Sm (s, x),
It is more convenient to impose the condition on the integral of G
Zx
0
m
Sm (s, y) dy = x 1 esJ (x)
G
sm s
Zx
dy e
0
sJ (y)

S1,m (s)
sG
m1
y
.
(1 4y 2 )3/2
(52)
527
Using (50), (51) and (52) the consistency condition is that

Zx
gm = e
sJ (x)
dy e
0
sJ (y)

S1,m (s) 1
sG
m1
y
.
(1 4y 2)3/2
s
(53)
That is to say the small x, large s, expansion of gm contains no terms O(1) or higher in s.
Substituting (51) and expanding the exponential in its Taylor series (which is of course
allowed for all values of the argument) we get
!
Zx
p
X
X k
p m1
1
s
y
.
(54)
J (x) J (y)
gm = dy
p!
sk
(1 4y 2 )3/2
0
p=0
k=0
For x < 1/2 (54) clearly has a power series expansion in x so the requirement that there
are no O(1) terms in s becomes
!
Zx
X
p p
1
m1
= 0.
(55)
J (x) J (y)
C0 = dy y
(1 4y 2)3/2
p!
p=0
First we show that this constraint alone implies that all higher powers of s vanish as well;
for O(s) we get
!
Zx
X
m1
p p
1
J (x) J (y)
C1 = dy J (x) J (y) y
(1 4y 2 )3/2
p + 1!
p=0
= 0,
(56)
and observe that

dC1 dJ
=
C0 = 0
dx
dx
(57)
provided x < 1/2. It follows that C1 (x) is a constant; but C1 (0) = 0 therefore
C1 (x) = 0.
(58)
Any pair Ck (x) and Ck+1 (x), with k > 0 are related in the same way and so it is
straightforward to proceed inductively to show that all Ck>0 (x) are zero.
Taking the Laplace transform of (55) with respect to J we find that
Z
0
x(J )m sJ
S1,m (s)
e
dJ = G
m
esJ
dJ,
3x(J )
(59)
S1,m (s) is determined; making the change

where x(J ) is obtained by inverting (43). Thus G
1
of variables x = 2 tanh and integrating by parts we get
R
3
m1
e 4 s cosh sech2 d
0 tanh
S
.
(60)
G1,m (s) =
R
3
2m1 s 0 cosh e 4 s cosh d
528
S1,m (s) is positive for all real, positive s and straightforward to check that
It is clear that G
its second derivative with respect to s diverges logarithmically as s 0; it follows that
1
(61)
t3
at large t. The remaining integrals in (49) yield functions which are analytic in s in the
neighbourhood of the origin provided x < xc . Thus we can conclude that
G1,m (t)
1
(62)
t3
for all odd n and m. So the result = 4 is universal for all amplitudes Gn,m (t) in agreement
with every other calculation for pure gravity.
S1,m (s) can be fixed more simply by examining (49). As x xc , J (x) diverges
In fact G
Sm (s, x) grows faster than exponentially in s, which is impossible, unless
and hence G
Gn,m (t)

Z1/2
S1,m (s)
sG
sJ (x)
m1
dx e
x
=0
(1 4x 2)3/2
(63)
which is the same condition as (60). We have explored the more indirect route because this
will help in the multi-critical case.
4. The multi-critical models

The multi-critical evolution equation (37) may be written

1
x 1 Gm (x, t)
=
x Gm (x, t)F (x)
t
x
2k3
K
X
X
2kgkK
Gj,m (t)(2k 2 j )x j 2k+1 ,
(64)
j =1
k=1
where, taking A(x) from [21], we have

K
X
2kgk
F (x) x 2x A(x) 1
x 2k2
2
!
K 32K
x
1 4x 2
= 2KgK
K 1
2
(65)
k=1
The solution to (64) takes the form [22]

1
U0 t + J (x)
x 1 Gm =
F (x)
K
X
2kgkK
2k3
X
j =1
k=1
(2k 2 j )
F (x)
Zt

d Gj,m ( )Uj k t + J (x) ,
where, as before,
Zx
K2
X
3

dy
2 rK+ 2
=
hK
hK
J (x) =
r 1 4x
K1
F (y)
0
r=0
(66)
(67)
529
and the coefficients hK

r are easy to compute. The function U0 (y) is fixed by the initial
conditions

1
U0 J (x)
(68)
x m1 =
F (x)
and the functions Uj k (y) by requiring that (66) is a solution of (64),

1
Uj k J (x) .
(69)
F (x)
Because of the unknown functions Gj,m (t), (66) is not of course a complete solution. Note
that J (x) is always an even function which diverges as |x| 12 and which is positive for
even K and negative for odd K. To determine the critical behaviour from the properties at
large t we need to know the behaviour of the functions U0 ( ) and Uj k ( ) for large positive
argument. For even K we see by considering (68) and (69) as |x| 12 that
x j 2k+1 =
2K1
2K1
U0 ( ) 2K3 ,
Uj k ( ) 2K3 .
(70)
However for odd K only the large negative argument behaviour is determined. This
phenomenon always occurs in calculations for the multi-critical models. Extrapolating the
solution to positive time leads to a singularity at finite time; it seems to us quite likely that
this is an artefact of the truncation of the original finite difference equation (32) into a first
order differential equation (34) and that the solution may be stabilised by higher derivative
terms. From now on we will concentrate on the even K models and start by studying the
K = 4 case in detail.
4.1. K = 4, even m
As for the K = 2 model we will consider the cases of even and odd m separately. For
even m we have

1
4
4
x 1 Gm (x, t)
=
x Gm (x, t)F (x) + G2,m (t)

t
x
35x 5 5x 3
2
G4,m (t),
(71)
+
35x 3
with
(1 4x 2)7/2
.
35x 5
Taking the Laplace transform gives
F (x) =
(72)

S
4
4
m1
S
S
=
Gm (s, x)F (x) + G2,m (s)

s Gm (s, x) x
x
35x 5 5x 3
2 S
G4,m (s),
(73)
+
35x 3
where the transformed correlation functions are defined as in (47); integrating (73) we
obtain
530
Sm (s, x) =
G
(
1 S
1
1 SC
1 SC
G2,m (s) +
G4,m (s) + G
(s)
4
2
F (x) 35x
35x
35 6,m
Zx
+e
sJ (x)

dy esJ (y) y m1 +
s
35y 5F (y)
SC (s)
y 5G
6,m
SC (s) + y G
S2,m (s)
+ y3G
4,m

)
,
(74)
where we have introduced the combinations

S4,m (s) 14G
S2,m (s),
SC (s) = G
G
4,m
S2,m (s) 14G
S4,m (s) + G
S6,m (s).
SC (s) = 70G
G
6,m
(75)
S6,m (s) in (74); this happens because F (x) is singular at x = 0

Note the appearance of G
which makes the evaluation of the limits of integration slightly non-trivial. As before it is
convenient to deal with
Zx
m
Sm (s, y) dy = x + gm ,
(76)
G
sm
s
0
where now
Zx
gm (s, x) = e
sJ (x)
0

dy esJ (y) y m1 +
s
SC (s) + y 3 G
SC (s)
y 5G
6,m
4,m
35y 5F (y)

S2,m (s)
(77)
+ yG
and we deduce that

1
gm
s
(78)
Sm (s, x) has the correct small x, large s, expansion.

in order to fulfill the constraint that G
S6,m (s) as explained at the end of Section 2 by
In fact we can immediately determine G
S
requiring that Gm (s, x) does not grow faster than any exponential of s as x 12 which
implies that

(79)
gm s, 12 = 0
S2,m (s) and G
S4,m (s) which still have to be determined.
S6,m (s) is related to G
and so G
Defining
X k
SC (s) = 1
,
G
6,m
s
sk
k=0
X k
SC (s) = 1
,
G
4,m
s
sk
k=0
X k
S2,m (s) = 1
G
s
sk
531
(80)
k=0
(of course the , and coefficients depend on m but we have suppressed this to avoid
clutter) and substituting into (77) we obtain the O(s 0 ) constraint
Zx
C0 =
k
X

1
(J
(x)
J
(y))
dy y m1
y 5 k + y 3 k + yk = 0.
k!
35y 5F (y)
(
k=0
(81)
Identical manipulations to the K = 2 case show that C0 = 0 ensures Ck>0 = 0 also. There
are in fact three separate constraints hidden in (81) which are sufficient to determine the
k , k and k coefficients. To see this we proceed by writing everything in terms of J .
J (x) is an even function with a small x expansion
J (x) =
35 6 245 8
x +
x +
6
4
(82)
so it follows that, by reverting the series,
m/2 X
k
x(J )m
= J 1/3
ak J 1/3 ,
m
(83)
k=0
where we take the real positive cube root of J (remember that m is even and again we
suppress the m dependence on ak ). Similarly we have
Zx
y dy
1
35y 5F (y)
J (x) J (y)
k
ZJ
=
dJ 0
(J J 0 )k
35x(J 0)4
ZJ
=
dJ 0 (J J 0 )k J 02/3
X
l=0
bl J 01/3
l
l=0
bl J k+(l+1)/3
(k + 1) ( l+1
3 )
(k +
l+1
3
+ 1)
(84)
and
Zx
l+2
k X
1
k+(l+2)/3 (k + 1) ( 3 )
J
(x)
J
(y)
.
y dy
=
c
J
l
35y 5F (y)
(k + l+2
3 + 1)
3
(85)
l=0
We will not need explicit expressions for the coefficients al , bl and cl . Substituting (83),
(84) and (85) into C0 we obtain
532
C0 (J ) = J 1/3
m/2 X
ak J 1/3
k
k=0
+
+
1 X k J k+1
35
(k + 1)!
k=0
( l+1
3 )
k bl J k+(l+1)/3
l+1
(k + 3 + 1)
k=0 l=0
XX
( l+2
3 )
.
k cl J k+(l+2)/3
l+2
(k + 3 + 1)
k=0 l=0
X
X
(86)
Letting N be an integer, (86) yields three conditions corresponding to terms O(J N ),

1
2
O(J N+ 3 ) and O(J N+ 3 ), respectively. The corresponding coefficients are easily extracted
from (86) and we find
X
X
N1
a3Nm/2 (N + 1) +
NL b3L1 (L) +
NL c3L2 (L) = 0,
35
N
L=1
L=1
(87)
a3N+1m/2 N +
4
3
N
X
NL b3L L +
1
3
L=0
N
X
NL c3L1 L +
1
3
= 0,
L=1
(88)
a3N+2m/2 N +
5
3
N
X
NL b3L+1 L +
L=0
2
3
N
X
NL c3L L +
2
3
= 0.
L=0
(89)
Now suppose that 0 , . . . , N1 and 0 , . . . , N1 are known; from (87) we can obtain
0 , . . . , N1 . Then (88) contains only these known coefficients together with N which
is thus determined. Now (89) determines N . Knowing 0 , . . . , N1 and 0 , . . . , N1
we now determine N from (87). Proceeding in this way all the coefficients, and thus the
right-hand sides of (80), can be obtained iteratively. To analyse the asymptotic behaviour
S2,m (s) etc. at small s.
of the correlation functions we need to know the behaviour of G
Resumming the relations (87), (88), and (89), and using the integral representation
Z
(n)
= d n1 e s ,
(90)
sn
0
we get
0=
SC (s)
G
6,m
35s
SC (s)
+G
4,m
0
Z
d e
S2,m (s)
a3Nm/2 + G
N=1
d e s
X
N=1
c3N2 N1 ,
d e s
b3N1 N1
N=1
(91)
Z
0=
d e
a3N+1m/2
S2,m (s)
+G
N=0
SC (s)
+G
4,m
Z
d e
d e s
b3N N 3
N=0
c3N1 N 3 ,
a3N+2m/2
N+ 23
S2,m (s)
+G
N=0
SC (s)
+G
4,m
d e
(92)
N=1
Z
0
0=
N+ 13
533
Z
0
d e s
d e s
b3N+1 N 3
N=0
c3N N 3 .
(93)
N=0
These three equations determine the unknown functions that we need.

Now consider the equation
J (x) = ,
(94)
where is real and positive; it can be rewritten as the quintic equation for u x 2

2
24 2
1 + (1 4u)5 30u2 10u + 1 = 0.
7
When = 0 it reduces to

4u3 10 95u + 256u2 = 0
(95)
(96)
which has a complex conjugate pair of roots
95 i9 15
,
(97)
u=
512
and three roots vanishing when = 0; any integer power of these roots, which we denote
by u1,2,3 , has a series expansion
u1 ( )
M

1/3 M
n
DMn 1/3 ,
n=0
u2 ( )
u3 ( )
M
M
X
1/3 M
DMn 1/3 )n ,
n=0
X
2 1/3 M
n
DMn 2 1/3 ,
(98)
n=0
where is a complex cube root of unity and 1/3 is the real cube root of .
Now take linear combinations of (91)(93) with the coefficients being different powers
of to obtain

S

Z
SC (s)
G
1
1 G
2,m (s)
4,m
s
m
C
S
+ G6,m (s)
+
(99)
ui ( ) 2
0 = d e
35 ui ( )2
ui ( )
0
534
with i = 1, 2, 3.
As all the roots of (95) converge on u = 1/4 like

c
1
1 2/5 + ,
u=
4
(100)
where is a fifth root of unity and c is a constant. The flow of the roots in the complex
plane as varies is shown in Fig. 4. We see that u1 flows into the = point as (100)
4
with = 1; and that the complex conjugate pair u2,3 flow into the pair with = ei 5 .
Note that there are no degenerate roots of (94) in the interval = (0, ) (this follows from
the fact that F is regular in this interval).
Now define the function J by

(101)
J u = x 2 = J (x)
and make the change of variable
= J (ui )
in (99). The constraint equations then take their final form

S

Z
SC (s)
G
dJ s J (u) 1 m
1 G
2,m (s)
4,m
C
S
2
e
+ G6,m (s) ,
+
u
0 = du
du
35
u2
u
(102)
(103)
Pi
where the contours Pi in the complex u-plane are shown in Fig. 4. Note that the i = 1
constraint is the same as (79) after an integration by parts and that these equations are
guaranteed to have a unique solution by the argument immediately following (87)(89).
To find the asymptotic large t dependence of the correlation functions it suffices to find
the leading non-trivial s-dependence of their Laplace transforms at small s; for this purpose
we need the integrals in (103) up to and including O(s 1/5 ). Replacing J by its explicit
form we need
Fig. 4. The flow of the roots of (95) in the complex u plane; the arrows denote the direction of
increasing .
IiF =
35
2
Z
Pi
535
du u2(1 4u)7/2F (u)

exp s A(1 4u)5/2 + B(1 4u)3/2 + C(1 4u)1/2 , (104)
1
where F (u) = {u2 , u1 , u 2 m }, A = 7/64, B = 35/96 and C = 35/64 and we have

dropped a factor exp(7s/24) which cancels in (103). Now make the change of variables
z = s(1 4u)5/2 .
Then all the integrals in (102) are linear combinations of
Z

Hk = s (2k7)/5 dz z2(k1)/5 exp(Az) exp Bs 2/5 z3/5 + Cs 4/5 z1/5 ,
(105)
(106)
Pi0
where k = 1, 2, . . . . The contours Pi0 in the complex z-plane are shown in Fig. 5. The
0 contours
second exponential in (106) is then expanded in its Taylor series, and the P2,3
collapsed onto the branch cut as shown in Fig. 6; all the integrals then become elementary.
Eqs. (103) can then be solved to give
0 for (106) in the complex z plane. All contours start at z = s and

Fig. 5. Integration contours P1,2
go out to infinity; the cut is for the fractional powers of z. The contour P30 is simply the complex
conjugate of P20 .
Fig. 6. (a) Distortion of the integration contour P20 . If the integrand has an integrable singularity at
the origin then the circular part of the new contour can also be collapsed onto the branch cut as shown
in (b).
536
S2,m (s) = 35 (m + 2)
G
27+m
2/5

7s
4
175 ( 45 ) 2
m + 6m + 8 1 + 2 cos
+ O s 4/5 ,
+
7+m
1
6.2
5
64
(5)
S4,m (s) = 35 (3m 2)
G
26+m
2/5

7s
4
175 ( 45 ) 2
m
+
6m
+
8
1
+
2
cos
+ O s 4/5 ,
+ 7+m
1
2
5
64
(5)
S6,m (s) =
G
35
(15m2 50m + 64)
26+m m
2/5

7s
4
875 ( 45 ) 2
m + 6m + 8 1 + 2 cos
+ O s 4/5 .(107)
+ 7+m
1
2
5
64
(5)
These expressions have the expected properties; they are positive and decreasing functions
of s for small positive s. It follows from (107) that G2,m (t), G4,m (t) and G6,m (t) all have
the same asymptotic behaviour at large t namely
Gj,m (t) t 7/5
(108)
for j = 2, 4, 6 and hence the exponent = 12/5. To determine the behaviour of the higher
correlation functions it is simplest to return to (74). All the integrals yield functions of s
which are analytic in some neighbourhood of the origin provided x < 1/2 and so we can
conclude that every coefficient of the x expansion of Gm (x, t) behaves the same way and
that (108) is valid for all (even) j and m.
4.2. K = 4, odd m
The method is very similar to the even m case. After Laplace transforming the evolution
Sm (s, x) we obtain
equation and solving for G
(
1 SC
1 SC
1 S
Sm (s, x) = 1
G (s)
G1,m (s) +
G (s) +
G
F (x) 35x 5
35x 3 3,m
35x 5,m
Zx
+e
sJ (x)
dy e
sJ (y)

y m1 +
S1,m (s)
+G
s
35y 5F (y)
)
SC (s) + y 2 G
SC (s)
y 4G
5,m
3,m
The consistency conditions can be cast in the form

S

Z
SC (s) G
SC (s)
G
dJ s J (u) 1 m
1 G
1,m (s)
3,m
5,m
2
e
+
+
u
.
0 = du
du
35
u5/2
u3/2
u1/2
Pi
This time we need the integrals
(109)
(110)
IiF =
35
2
Z
du

1
u2
(1 4u)7/2 u2 , u1 , 1, u 2 (m+1) es J (u) .
1/2
u
537
(111)
Pi
To the required order in s these can be calculated by expanding the u1/2 factor about
u = 1/4; then term by term the integrals are just the Hk . The resulting expressions for
the singular terms in s are simple but the constants appear as infinite sums; fortunately
the constants are only needed for the sub-leading s dependence so this method suffices
(the reason for this is explained in the next sub-section). The final expressions for the
correlation functions are very similar to those for even m

35
m2 + 4m + 3
28+m m

2/5

4
7s
175 ( 45 ) 3
2
m + 9m + 23m + 15 1 + 2 cos
+
6.28+m m ( 15 )
5
64

+ O s 4/5 ,

S3,m (s) = 35 3m2 + 4m + 1
G
7+m
2
m
2/5

7s
4
175 ( 45 ) 3
2
m
+
9m
+
23m
+
15
1
+
2
cos
+ 8+m
5
64
2
m ( 15 )

+ O s 4/5 ,

S5,m (s) = 35 15m2 20m + 29
G
7+m
2
m
2/5

7s
875 ( 45 ) 3
4
2
m + 9m + 23m + 15 1 + 2 cos
+ 8+m
1
2
5
64
(5)

4/5
+O s
.
(112)
S1,m (s) =
G
Thus we can conclude that the exponent always takes the value 12/5 in the K = 4 model.
4.3. General even K
The method is very similar to the K = 4 case. After Laplace transforming the evolution
Sm (s, x) we obtain
equation and solving for G
(
2K2
X
K
SC
Sm (s, x) = 1
x 1p G
2KgK
G
2Kp1,m (s)
F (x)
p=2
Zx
dy esJ (y) y m1
+ esJ (x)
0
2K2
2Kg K s X p1 C
Sp,m (s)
2K3 K
y
G
y
F (y)
p=1
where
!)
,
(113)
538
SC
G
2Kp1,m (s) =
K
X
kgkK
k=2
K
KgK
S2kp1,m (s).
G
(114)
SC
Note that for given m (either odd or even) half of the G
p,m (s) in (113) are automatically
zero so there are K 1 undetermined functions. Proceeding as before we next examine the
roots of the equation
J (x) =
(115)
which can be rewritten as a degree 2K 3 polynomial equation for u x 2 taking the

generic form
2
(116)
(1 + a )2 (1 4u)2K3 1 + b1 u + + bK2 uK2 = 0,
where a, b1 , . . . , bK2 are constants. There are K 1 roots which vanish at = 0 and
K 2 roots which do not; the K 1 vanishing roots correspond to the K 1 functions
which have to be determined. As all the roots converge on u = 1/4 like

c
1
1
+ ,
(117)
u=
2
4
2K3
where c is a constant and is a (2K 3)rd root of unity. The roots which vanish at = 0
flow into

K
K
4n
(118)
, n = + 1, . . . , 0, . . . , 1.
= exp i
2K 3
2
2
From now on we will use the integer n to label the roots. The consistency conditions
become
(
)
Z
2K2
SC
X G
dJ s J (u) 1 m
p,m (s)
K
e
u 2 + 2KgK
,
(119)
0 = du
du
u(2K2p)/2
p=1
Pn
where the K 1 paths Pn follow in the complex plane the roots which vanish at the origin.
Note that in the case of K = 4 the conditions for even m, (103), and for odd m, (110),
can be merged into the form (119); for any fixed m we have K 1 equations for K 1
unknowns.
Restricting ourselves to even m all the integrals in (119) can be written as linear
combinations of
!
Z
K2
X
pK+ 21
K
K
rK+ 23
exp shK1 s
hr (1 4u)
.
(120)
Kp = du (1 4u)
r=0
Pn
Note that the factor exp(shK

K1 ) cancels out in (119) so from now on we drop it. After the
substitution
z = s(1 4u)K+ 2
we obtain
(121)
539
Fig. 7. (a) A contour P 0 with n = 3 and (b) the deformed version of it. The branch cut on the positive
real axis is suppressed for clarity.
2(2K 3)Kp = s
2p
1+ 2K3
Z
dz z
2p
2K3
exp
hK
0 z
K2
X
2r
2r
2K3 z1 2K3
hK
r s
r=1
Pn0
(122)
where the contours Pn0 encircle the origin n times before heading off to real positive
infinity, see Fig. 7a. As we will demonstrate shortly all these integrals are needed up
to and including the first positive power of s and so they can be separated into three
classes:
1. p < K 1; the integrals are divergent as s 0 but the singularities of the integrand
are integrable at z = 0 so the contour can be deformed as shown in Fig. 7b. The integral
from the starting point to the first blob is given by
2K 3
+ O(s).
(123)
2(2K 3)Kpfin =
2p 2K + 3
The portion of the contour between the two blobs gives no contribution and the leg
stretching out to infinity gives
2(2K 3)Kpdiv
1+ 2p
2K3
= sei2n
i2n 2r !
Z
K2
X
2K3
2p
se
2K3
K
K
dR R
exp h0 R
hr R
.
R
(124)
r=1
To obtain the integral up to the desired order the second exponential factor can be Taylor
expanded and then integrated term by term.
2. p = K 1; after an integration by parts the above construction can be used and we
get

1

i2n 2K3
2K4
se
+ .
(125)
2(2K 3)KK1 = (2K 3) 1 hK
0
2K3
3. p > K 1; the leading non-analytic term is of higher order than we need to consider
so
2(2K 3)Kp =
2K 3
.
2p 2K + 3
(126)
540
Now observe that in all these integrals s always appears with a factor ei2n and that
there are no other phase factors. Thus all of the constraint equations (119) can be obtained
from the n = 0 case by making the replacement s sei2n . They can thus be written in
the form
X
K1
C
S
Fp sei2n G
F0 sei2n +
2p,m (s) = 0,
(127)
p=1
where
Fp (w) = w
K1
X
!
fp,q w
2q
2K3
+ fp,K w + .
(128)
q=0
We have truncated the expansion of Fp (w) in anticipation of the following. Eq. (127) can
now be written in the matrix form
Df + s 2K3 f0,K1 + D f G + s 2K3 fG = 0,
1
(129)
where

pq = exp i2(K/2 + p)(1 + 2q/(2K 3)) ,
D pq = pq s 2(p1)/(2K3)1,
f p = f0,p1 + f0,K sp,1 ,

p = exp i2(K/2 + p)/(2K 3) ,
pq
pq
= fp,q1 + fq,K sp,1 ,

= exp i2(K/2 + p)/(2K 3) fq,K2 ,
SC .
Gp = G
2p,m
(130)
Now D is clearly non-singular and it is straightforward to check that is non-singular.

There does not seem to be any simple way of writing the elements of f for general K
but we expect it too is non-singular. Then the leading order solution for G is a constant
vector; furthermore the next term in the solution is O(s 2/(2K3)). We have already shown
by explicit solution that this is indeed what happens for K = 4; using Maple we have also
checked it for K = 6, 8, 10. It follows that
=
4(K 1)
2K 3
(131)
SC
for the correlation functions G
p,m , p = 2, 4, . . . , 2(K 1). The relationship (114)
C
Sp,m is non-singular so this conclusion applies also to G
Sp,m , p =
Sp,m and G
between G
2, 4, . . . , 2(K 1). Finally we note as usual that the integrals in (113) are analytic functions
of s in the neighbourhood of the origin provided x < 1/2 so the conclusion extends to all
the correlation functions.
541
5. The exponent
To find we need to study the scaling behaviour as the multi-critical point is approached
and to do this consider the modified couplings
g1K = g1K ,
gpK = (1 )gpK ,
p > 2.
(132)
In the graphical expansion the power of 1 is the number of vertices in the graph so
is related to in (8) by
1 = exp( c ).
(133)
By definition the multi-critical point is attained as 0. The disk amplitude A(x)

can still be calculated by exploiting the connection with topological gravity [21]; the
topological potential is
V (z) = Kz + z + (1 )(1 z)K
(134)
and the disk amplitude

I

1/2
z V (z)
dz
,
1 4zx 2
1 V 0 (z) log
A(, x) =
2i
z
(135)
where the contour encircles the branch cut of the logarithm. The branch points are at z = 0
and
zc
' 1 K 1/K ,
(136)
z=
4
where K is a constant. Collapsing the contour onto the cut gives
zc /4
Z
1/2

dy 1 4yx 2
+ (1 )(1 y)K1 .
A(, x) = K
(137)
Integrating we find that

F (x) x 2x A(, x) 1
2
K
X
2k g K
k=1
= xK 1 zc x 2
1/2
k
x 2k2
x(1 )
K
X
Kk
k 1
2kgkK
2
K 1/K
1 zc x 2
2k2
x
k=1
(138)
which reproduces (65) when = 0.
At small x, even for finite we still get the leading behaviour
F (x) x 32K
which implies that
Zx
dy
x 2K2.
J (x) =
F (y)
0
(139)
(140)
542
Fig. 8. The contours for K = 4 at finite .
Thus the considerations described in detail for K = 4 in Section 4 up to Eq. (93) go through
as before the only difference is that all the various coefficients are now functions of .
However the contours Pi are modified; when we make the change of variable (102)
= J (ui )
(141)
the = endpoint of the contour occurs at the value of ui where J diverges. The points
where J diverges are of course determined by the zeros of F . Defining w through
x 2 = zc + K 1/K w
(142)
we get using (138)

J (w) J u = (zc + K 1/K w)1
Zw
w1/2 (zc + K 1/K w)1/2
(K 1/K )1/2
dw.
=
P
K k1
2
K + 2(1 )(K 1/K )K1 K
k=1 kgk w
(143)
The term K in the denominator is sub-leading; its only effect is to shift the zeros by an
amount O(1/K ) and we discard it to obtain
3
Zw 1/2
w
(zc + K 1/K w)1/2
(K 1/K ) 2 K
dw.
(144)
J (w) =
PK
K k1
4(1 )
k=1 kgk w
Note that the denominator has precisely K 1 simple zeroes; one is real and positive and
the others come in complex conjugate pairs. Each zero is the end-point of one of the K 1
contours, see Fig. 8. Since we will only need the leading scaling behaviour we can also
approximate (zc + K 1/K w) by zc in (144) so that
(K 1/K ) 2 K
J (w) =
K)
4(1 )(KgK
3
Z
w
w1/2
dw
QK1
Sk )
k=1 (w w

3
Z
K1
(K 1/K ) 2 K X
w1/2
=
R
dw,
k
K)
ww
Sk
4(1 )(KgK
k=1
543
(145)
where w
Sk is a zero of the denominator of (144) and Rk the corresponding residue.
Sl ; when w is close to
Now suppose that w lies on the contour Pl which terminates at w
the end-point we get

3
Rl
w+ w
Sl
(K 1/K ) 2 K
log
+ Ql ,
(146)
J (w) =
K)
w
Sl
w w
Sl
4(1 )(KgK
where Ql is the accumulated finite contribution from all the other poles in (145). Note that
Sl > 0
the branch cut for the square roots is taken down the negative real axis so that Re w
for all l. Inverting (146) we find that

Tl exp 13/2K Sl J + 1 2

,
(147)
w=w
Sl
Tl exp 13/2K Sl J 1
Sl is
where Sl and Tl are (complex) constants of O(1). It turns out that the quantity Rl / w
never pure imaginary and thus Re Sl 6= 0; except for the single real root, Sl is not purely
real either and hence the correlation functions have some oscillatory behaviour which is
typical of a non-unitary theory. The formula (147) is valid for J ; expanding we get
!
X

(l)
13/2K
n exp n
(148)
Sl J ,
w=w
Sl 1 +
n=1
where
Sl = Sl ,
= Sl ,
if Re Sl > 0,
if Re Sl < 0.
(149)
Using (142) and (148) we see that the conditions (119) can be written in the form
0=
m/2 X
Al
1
zc + K 1/K w
Sl
+
s
s + n13/2K Sl
n=1
!
(n)
M Gl
(0) X
K 1
M G l+
+ 2KgK
,
s
s + n13/2K Sl
(n)
(150)
n=1
(n)
where Al
and M (n) are collections of coefficients and
= zc + K 1/K w
Sp
M (0)
pq
K1q
(151)
The apparent singularity in (150) at s = 0 is of course not present provided M (0) is

invertible. M (0) is of the Vandermonde form and therefore
(K1)(K2)/2 Y
(S
wp w
Sq ).
(152)
det M (0) = K 1/K
p>q
544
Since none of the roots are degenerate M (0) is invertible. Therefore the first non-analyticity
occurs at s 13/2K and hence the mass gap exponent is
=1
3
.
2K
(153)
6. Discussion
We have found that = 1 3/2K and = 2 + 2/(2K 3) and therefore, since
= 1/K, the Fisher scaling relation (13) is indeed satisfied. The results show that the
functions which are initially undetermined in the peeling calculation do not in fact change
the conclusions one would draw simply by ignoring their contribution in (66). The value
of agrees with that obtained by slicing [18].
To compare it is necessary to examine the continuum limit for the perturbation (132)
away from the multi-critical point. The multi-critical models have K independent coupling
constants so there are K independent directions along which the multi-critical point can
be approached; if chosen appropriately these directions correspond in the continuum limit
to operators of (length) dimensions {1, 2, 3, . . . , K} [17]. The procedure for taking the
continuum limit is explained in [1]. By definition boundary loops with l legs in the dual
graph (i.e., which are l links long) have continuum length L = la where a is the length of
one link. This implies that the generating function variable x conjugate to l must be related
to a continuum quantity X by x = xc eaX where xc is the radius of convergence of the
X)
disk amplitude (138). We can construct a non-trivial continuum disk amplitude Ac (,
provided that
K,
a
(154)
where is some continuum coupling; we obtain

X) + ,
A(, x) = A0 (, x) a K1/2Ac (,
(155)
where A0 is the analytic (non-universal) part of the disk amplitude. Note that is not
the cosmological constant; in the continuum theory the volume V must have dimension
(length)2 and therefore the cosmological constant must correspond to whichever
direction in the space of lattice couplings leads in the continuum limit to an operator
of dimension 2. It is straightforward to check that this is accomplished by the modified
couplings
p1 (K 2)!(p 1)!
,
gpK = gpK (1)
(K p 2)!2p!
gpK = gpK ,
p = K, K 1.
p 6 K 2,
(156)
The weight for polygons of 2p sides is modified by a p-dependent factor and so the lattice
quantity whose continuum limit is the volume is a complicated object with polygons of
different number of sides weighted in different ways; the volume is not the number of
polygons.
545
Using the scaling (154) for in the large t behaviour of the two-point function we find
that

(157)
exp t = exp ta K = exp T
and hence we will get a consistent non-trivial scaling limit provided the continuum string
time scales as T ta K . This also agrees with [18] (where what we call the string time is
called the geodesic distance).
The structure of these multi-critical surfaces seems slightly bizarre. Recall that the
coupling is conjugate to the number of polygons which therefore behaves roughly like
1 for typical surfaces in the ensemble. On the other hand from (157) we have that the
typical t must be roughly and therefore
h#polygonsi hti1/ .
(158)
When K gets large 1 and so these surfaces have an almost linear structure when
viewed in terms of polygons. However, as we discussed in Section 1, the polygons can
be very large so that, for example, the number of links traversed in a typical cycle can
also be very large and it does not follow that the surfaces in the continuum limit are at all
one-dimensional. It does however cast doubt on the idea that it makes sense to identify the
string time T with a continuum geodesic distance.
Another manifestation of this can be seen by considering a perturbation in terms of the
cosmological constant . Then we expect that the two point function behaves as

K
(159)
exp 2 T
(it is a straightforward calculation along the lines of Section 5 but using the couplings (156)
to check this). The average volume of the system will be hV i 1 ; on the other hand
2
from (159) the string time extent must be of order Ts 2 and hence hV i TsK . 2
Interpreting T as a geodesic distance leads to the conclusion that the Hausdorff dimension
2
which vanishes as K and is less than two for all K > 2. However we can
is dH = K
look at this another way; in [18] the probability distribution for the length L of exit loops
at time T given a point-like entrance loop at T = 0 was calculated. At large but finite K
1
the distribution shows that there is typically one macroscopic exit loop of length L T K
but that as K the distribution function becomes a delta-function. Using this result
we can relate the volume to the typical boundary length as K and find
K
V L2
(160)
which is typical of a smooth flat surface of Hausdorff dimension 2. This is the behaviour
we expect because K corresponds to central charge c where semi-classical
properties are recovered; the volume and loop length have highly anomalous behaviour
relative to the string time but exactly what we expect relative to each other. Note that our
discussion above implies that the macroscopically large boundary loops are often made
from a finite number of polygons with diverging number of sides.
2 One might worry that the negative weights appearing in the multi-critical models cause some cancellation
which alters this conclusion. We have checked explicitly that for K = 4 at least this does not happen.
546
The multi-critical models are non-unitary and have negative central charge, and so it is
plausible that in the continuum theory the Hausdorff dimension is given by (7) which works
well for the c = 2 model. As c (7) gives dH = 2 and so it is clear that the geodesic
distance implicit in the derivation of this formula (which is done directly in the continuum
using scaling arguments and Liouville theory) cannot be equivalent to the continuum limit
of the string time, T . Since we do not know what the relationship between these distance
measures really is, it is not clear what calculation in the discretized formulation would be
needed to check (7) for general K; in any case it must be highly non-trivial because (7)
gives irrational values for dH when K > 2 whereas discretized calculations are almost sure
to give rational values for exponents.
Acknowledgement
We acknowledge useful conversations with Thordur Jonsson. This work was supported
in part by PPARC Grant GR/L56565.
References
[1] J. Ambjrn, B. Durhuus, T. Jonsson, Quantum Geometry, Cambridge Monographs on
Mathematical Physics, Cambridge, 1997.
[2] H. Kawai, N. Kawamoto, T. Mogami, Y. Watabiki, Phys. Lett. B 306 (1993) 19.
[3] J. Ambjrn, Y. Watabiki, Nucl. Phys. B 445 (1995) 129.
[4] J. Ambjrn, J. Jurkiewicz, Y. Watabiki, Nucl. Phys. B 454 (1995) 313.
[5] Y. Watabiki, Nucl. Phys. B 441 (1995) 119.
[6] J. Ambjrn, C. Kristjansen, Y. Watabiki, Nucl. Phys. B 504 (1997) 555.
[7] J. Ambjrn et al., Nucl. Phys. B 511 (1998) 673.
[8] Y. Watabiki, Prog. Theor. Phys. Suppl. 114 (1993) 1.
[9] N. Kawamoto, Fractal structure of quantum gravity in two dimensions, in: (Eds.) K. Kikkawa,
M. Ninomiya, Proceedings of the 7th Nishinomiya-Yukawa Memorial Symposium, November
1992, Quantum Gravity, World Scientific.
[10] M. Ikehara, N. Ishibashi, H. Kawai, T. Mogami, R. Nakayama, N. Sasakura, Phys. Rev. D 50
(1994) 7467.
[11] J. Ambjrn, K.N. Anagnostopoulos, Nucl. Phys. B 497 (1997) 445.
[12] V. Kazakov, Mod. Phys. Lett. A 4 (1989) 2125.
[13] M. Staudacher, Nucl. Phys. B 336 (1990) 349.
[14] D.J. Gross, A.A. Migdal, Phys. Rev. Lett. 64 (1990) 127.
[15] M. Douglas, S. Shenker, Nucl. Phys. B 335 (1990) 635.
[16] E. Brzin, V. Kazakov, Phys. Lett. B 236 (1990) 144.
[17] G. Moore, N. Seiberg, M. Staudacher, Nucl. Phys. B 362 (1991) 665.
[18] S.S. Gubser, I.R. Klebanov, Nucl. Phys. B 416 (1994) 827.
[19] J. Ambjrn, Y. Watabiki, Int. J. Mod. Phys. A 12 (1997) 4257.
[20] I.K. Kostov, Nucl. Phys. B 326 (1989) 583.
[21] J. Ambjrn, M.G. Harris, M. Weis, Nucl. Phys. B 504 (1997) 482.
[22] R. Courant, D. Hilbert, Methods of Mathematical Physics, Vol. 2, Wiley.

Operator product expansion of the lowest weight

CPOs in N = 4 SYM4 at strong coupling
Gleb Arutyunov a , Sergey Frolov b,1 , Anastasios C. Petkou c,
a Sektion Physik, Universitt Mnchen, Theresienstr. 37, D-80333 Mnchen, Germany
b Department of Physics and Astronomy, University of Alabama, Box 870324, Tuscaloosa, AL 35487-0324, USA
c Department of Physics, Theoretical Physics, University of Kaiserslautern, Postfach 3049,
67653 Kaiserslautern, Germany

Abstract
We present a detailed analysis of the 4-point functions of the lowest weight chiral primary
operators O I tr( (i j ) ) in N = 4 SYM4 at strong coupling and show that their structure is
compatible with the predictions of AdS/CFT correspondence. In particular, all power-singular terms
in the 4-point functions exactly coincide with the contributions coming from the conformal blocks
of the CPOs, the R-symmetry current and the stress tensor. Operators dual to string modes decouple
at strong coupling. We compute the anomalous dimensions and the leading 1/N 2 corrections to the
normalization constants of the 2- and 3-point functions of scalar and vector double-trace operators
with approximate dimensions 4 and 5, respectively. We also find that the conformal dimensions
of certain towers of double-trace operators in the 105, 84 and 175 irreps are non-renormalized.
We show that, despite the absence of a non-renormalization theorem for the double-trace operator
in the 20 irrep, its anomalous dimension vanishes. As by-products of our investigation, we derive
explicit expressions for the conformal block of the stress tensor, and for the conformal partial wave
amplitudes of a conserved current and of a stress tensor in d dimensions. 2000 Elsevier Science
PACS: 11.25.Hf; 11.25.Sq
Keywords: AdS/CFT correspondence; Conformal field theory; Non-perturbative methods
1. Introduction
The AdS/CFT correspondence [13] is arguably the best currently available way of
getting nontrivial dynamical information for the strong coupling behavior of certain
E-mail addresses: arut@theorie.physik.uni-muenchen.de (G. Arutyunov), frolov@bama.ua.edu (S. Frolov),

petkou@physik.uni-kl.de (A.C. Petkou).
1 On leave of absence from Steklov Mathematical Institute, Gubkin str. 8, 117966 Moscow, Russia.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 3 9 - 9
548
G. Arutyunov et al. / Nuclear Physics B 586 (2000) 547588
conformal field theories. In particular, the N = 4 supersymmetric SU(N) YangMills

2 N is
theory in four dimensions (SYM4 ) at large N and at strong t Hooft coupling = gYM
dual to type IIB supergravity on the AdS5 S 5 background. The supergravity fields are dual
to certain quasi-primary operators in SYM4 . According to [2,3], the generating functional
for the connected Green functions of these operators coincides with the on-shell value of
type IIB supergravity action which has to be further modified by the addition of definite
boundary terms [4]. Thus, computing n-point correlation functions in the supergravity
approximation is generally divided into two independent problems finding first the
supergravity action up to the nth order and then evaluating its on-shell value. Although a
covariant action for type IIB supergravity is unknown, one can use the covariant equations
of motion [57] and the quadratic action [8] to find cubic actions [911] for its physical
fields and to compute the corresponding 3-point functions using the technique developed
in [12].
Computing 4-point functions [1326] in the supergravity approximation in general
requires the derivation of the supergravity action up to fourth order. The part of the
action relevant to the massless modes, corresponding to the dilaton and axion fields, was
already known as pointed out in [13] where the calculation of the corresponding 4-point
functions was initiated. The complete expression for the 4-point functions was obtained
in [20] and was further analyzed in [23]. Unfortunately, these modes are dual to the
rather complicated operators trF 2 + and trF F and the analysis performed in [23] was
unavoidably incomplete.
It is known that all operators dual to the type IIB supergravity fields belong to
short representations of the conformal superalgebra SU(2, 2|4) and are supersymmetric
descendents of Chiral Primary Operators (CPOs) of the form OkI = tr( (i1 ik ) ). CPOs
are dual to scalar fields s I that are mixtures of the five form field strength on S 5 and the
trace of the graviton on S 5 . The relevant part of the quartic action of type IIB supergravity
for the scalars s I was found in [27] and was then used in [28] to compute the 4-point
functions of the simplest CPOs O I = tr( (i j ) ). In the present paper we use these 4-point
functions to analyze in detail the Operator Product Expansion (OPE) of the lowest weight
CPOs at strong coupling.
It is widely believed that the structure of a Conformal Field Theory (CFT) is encoded
in the OPE since knowledge of the latter allows, in principle, the calculation of all n-point
functions. Thus, in the context of AdS/CFT correspondence one would eventually like to
prove that the 4-point functions (and in general n-point functions) of CPOs in the boundary
CFT computed in the supergravity approximation admit an OPE interpretation. This is a
rather complicated problem because an infinite number of quasi-primary operators may in
principle appear in the OPE of two CPOs. Therefore, the best one can presently do is to
show that the leading terms in a double OPE expansion of the 4-point functions exactly
match the contributions of the conformal blocks of the first few quasi-primary operators
with the lowest conformal dimensions. This is the main line of investigation which we
follow in the present work in our analysis of the 4-point function of the lowest weight
CPOs in N = 4 SYM4 .
549
Our study shows that there are four singular terms in the OPE of two lowest weight
CPOs corresponding to the identity operator, the lowest weight CPO itself, the R-symmetry
vector current and the stress tensor. These three nontrivial operators are dual to the scalars
s I , the vector fields A and the graviton h that appear in the exchange Feynman
diagrams of type IIB supergravity. The most singular terms in the 4-point functions
computed in the supergravity approximation exactly coincide with the contributions
coming from the conformal blocks of the above three operators.
We compare the strong coupling OPE with the free field theory OPE, and explicitly
observe, at weak coupling, the splitting of the R-symmetry current and of the stress tensor
into 2 and 3 terms respectively which belong to different supermultiplets. Only one term
in each splitting is dual to a supergravity field and survives at strong coupling while the
others acquire large dimensions and decouple. A similar type of splitting also occurs in the
case of the double-trace operators transforming in the 84 and 175 irreps.
We also analyze the leading nonsingular terms in the OPE which are due to doubletrace operators of the schematic form : m O I n O J : with free field conformal dimensions
4 + m + n. A generic property of any correlation function computed in the supergravity
approximation is the appearance of logarithmic terms. In an unitary CFT logarithmic terms
have a natural interpretation in terms of anomalous dimensions of operators [29] and such
an interpretation was used in the past in studies of the O(N) vector model [3032]. Since
the operators dual to the supergravity fields have protected conformal dimensions, the
logarithmic terms in the correlation functions of supergravity can only be attributed to
anomalous dimensions of double-trace operators.
We show that among the scalar double-trace operators with free field conformal
dimension 4, the only one acquiring an anomalous dimension is the operator : O I O I :,
which transforms in the trivial representation of the R-symmetry group SO(6) SU(4).
The anomalous dimension of this operator is found to be 16/N 2 and coincides with the
anomalous dimension of the operator B which was calculated in [23]. This is consistent
with the fact that B is a supersymmetric descendent of : O I O I : . It is worth noting that
among the non-renormalized operators we find a double-trace scalar operator in the 20
irrep of SO(6) whose non-renormalization property does not follow from the shortening
condition discussed in [33,34].
Finally, we compute the anomalous dimensions of the double-trace vector operators with
free field conformal dimension 5 transforming in the 15 and 175 irreps, respectively. We
show that there are several towers of traceless symmetric tensor operators in the 105, 84 and
175 irreps, whose anomalous conformal dimensions vanish. Some of these tensor operators
are not subject to any known non-renormalization theorem.
The 4-point functions of CPOs also allow us to find the leading 1/N 2 corrections to
the normalization constants of the 3-point functions involving two CPOs and one doubletrace operator with low conformal dimension. In the case when a double trace operator
has protected dimension we interpret these corrections as manifestation of the splitting
of the free field theory operator in two orthogonal parts carrying different representations
of supersymmetry. The first one has protected both the dimension and the normalization
constant, the other one acquires infinite anomalous dimension and disappears at strong
550
coupling. To make this interpretation precise one should further show that the linear
splitting arising due to the difference between normalization constants in free theory and
at strong coupling is consistent with the fact that the split fields transform in different
representations of supersymmetry. It would be quite interesting to investigate such a
property in more detail.
The plan of the paper is as follows. In Section 2 we recall how logarithmic terms are
related to anomalous conformal dimensions in an unitary CFT and in the framework of
the AdS/CFT correspondence. In Section 3 we discuss the structure of the OPE of the
lowest weight CPOs in free field theory and at strong coupling. In Section 4 we compute
anomalous dimensions and first corrections to the 2- and 3-point normalization constants
of double-trace operators of approximate dimensions 4 and 5. A discussion of the results
obtained and our conclusions are presented in Section 5. Several technical issues are
considered in five appendices. In the Appendix A we discuss a decomposition of a bilocal operator which is a normal-ordered product of two quasi-primary scalar operators
into a sum of conformal blocks of local tensor primary operators. In the Appendix B
explicit formulae for conformal partial amplitudes of scalar, conserved vector current and
stress tensor are derived. A convenient series representation used throughout the paper is
obtained in the Appendix C. In the Appendix D we discuss the projectors which single
out the contributions of irreps occurring in the decomposition 20 20 of SO(6) from the
4-point function of CPOs. In the Appendix E an explicit formula for the conformal block
of the stress tensor is derived.
2. Anomalous dimensions and logarithmic terms in CFT

An arbitrary unitary CFT is completely characterized by a set of quasi-primary operators
Oi of conformal dimensions i and by their OPE
X
1
C k (x y, y )Ok (y).
(2.1)
Oi (x)Oj (y) =
i +j k ij
|x
y|
k
Here the sum runs over the set of all the quasi-primary operators and i, j, k are multiindices which in general include the indices of the R-symmetry and of the Lorentz groups.
The operator algebra structure constants Cijk (x y, y ) can be decomposed in a power
series in x y and y . Without loss of generality one can assume that the operators Oi are
orthogonal

ij
Oi (x)Oj (0) = Ci 2 ,
x i
where Ci is a normalization constant of the 2-point function. Then the operator algebra
structure constants are fixed by the conformal dimensions i , j , k , and by the ratio
Cij k /Ck , where the structure constants Cij k appear in the 3-point functions

Oi (x)Oj (y)Ok (z) =
|x
y|i +j k |x
Cij k
.
z|i +k j |y z|j +k i
(2.2)
551
The conformal dimensions and the structure constants depend on the coupling constants of
the CFT. In principle, the OPE (2.1) allows one to compute any correlation function in the
CFT. In particular, 4-point functions are given by the following (schematic) double OPE
expansion
X
1
Oi (x)Oj (y)Ok (z)Ol (w) =
m
i
j
|z w|k +l m
m |x y|
m
Cijm (x y, y )Ckl
(z w, w )
Cm
.
|y w|2m
(2.3)
Thus we see that the short distance expansion of exact CFT correlation functions does
not contain logarithmic terms. Suppose, however, that one can only calculate correlation
functions up to some order in the coupling constant or another small parameter of
the CFT. Then it is clear from (2.3) that logarithmic terms would appear due to the
nontrivial dependence of conformal dimensions on the coupling or on the small parameter.
These terms can be easily found representing the conformal dimensions as = (0) +
(1) , where (0) is the canonical part and (1) is the anomalous coupling constant
dependent part. Such a representation leads then to an expansion for the two-point
(1)
functions of the form |x| = 1 + (1) log |x| + , connecting the logarithmic terms
to the anomalous dimensions, that may be used to compute the latter. It is worthwhile to
note that at the nth order of perturbation theory one encounters terms of the form (log |x|)n .
The N = 4 SYM4 theory provides an example of such a logarithmic behavior of
correlation functions, both in the weak coupling standard perturbation expansion [3538]
and also in the supergravity approximation [20,28]. Due to superconformal invariance all
quasi-primary operators of SYM4 belong either to short or long representations of the
conformal superalgebra SU(2, 2|4) and in the framework of the AdS/CFT correspondence
fall into three classes:
(i) Chiral operators dual to the type IIB supergravity fields which belong to short
representations and have protected conformal dimensions. The simplest operators
in this class are the lowest weight CPOs O I = tr( (i j ) ).
(ii) Operators dual to multi-particle supergravity states which are obtained as normalordered products of the chiral operators, e.g., the double-trace operators : O I O J :.
They may belong either to short or long representations and have conformal
dimensions restricted from above.
(iii) Operators dual to string states (single- or multi-particle) which belong to long
representations and whose conformal dimensions grow as 1/4 in the strong
coupling limit. The simplest example of such an operator is the Konishi operator
tr( i i ).
In the supergravity approximation to the AdS/CFT correspondence the operators dual
to string states decouple from the spectrum and one can calculate the connected n-point
functions of chiral operators dual to the supergravity fields to leading order which is
1/N n2 . Since the expansion parameter is 1/N 2 , an n-point function contains logarithmic
terms of the form (log |x|)[(n2)/2] . In particular, a 4-point function can have only log |x|dependent terms, and cannot have, say, terms of the form (log |x|)2 . Moreover, since chiral
552
operators have protected conformal dimensions only the operators dual to multi-particle
supergravity states contribute to log-dependent terms.
The AdS/CFT correspondence predicts a simple form of the OPE of chiral operators
in the strong coupling limit. Let O1 and O2 be operators dual to the supergravity fields
1 and 2 respectively and let the supergravity action contain the non-vanishing cubic
couplings N1 12k 1 2 k with some fields k . Then, the OPE of O1 and O2 takes the form
(suppressing the indices of the operators and structure constants)
O1 (x)O2(y) =
1 X
1
C k (x y, y )Ok (y)
N
|x y| 1 +2 k 12
k

+ : O1 (x)O2 (y) : ,
(2.4)
where Ok is an operator dual to k . Here we denote by [: O1 (x)O2 (y) :] an infinite

sum of tensor quasi-primary operators and their descendents, which are dual to multiparticle supergravity states. In general these operators acquire anomalous dimensions and
are responsible for the appearance of logarithms in correlation functions. An important
property of the operators dual to multi-particle supergravity states is that their structure
constants are of order 1, while the structure constants of the operators dual to supergravity
fields are of order 1/N . Due to such a property, the sum of these operators coincides
in the limit N with the corresponding free field theory normal-ordered operator
fr
fr
: O1 (x)O2 (y) :. This can be seen as follows. A 4-point function of chiral operators
is given by a sum of a disconnected contribution which is of order 1 and a connected
Green function which is of order 1/N 2 . Since the structure constants of the operators
dual to supergravity fields are of order 1/N , they do not contribute to the disconnected
part of the 4-point function. Thus only the normal-ordered operators contribute. The
disconnected part is given by a sum of products of 2-point functions of chiral operators,
hence it does not depend on the coupling constant and N (we assume that all the chiral
operators are orthonormal) and coincides with the free field disconnected part. Therefore,
in the limit N the sum [: O1 (x)O2 (y) :] has to coincide with the free field normalordered product : O1fr (x)O2fr (y) :, 2 that is, decomposed into a sum of local tensor quasiprimary operators. However, at finite N an infinite number of the tensor operators acquire
anomalous dimensions and their structure constants get 1/N 2 corrections to their free field
values. For this reason it seems hardly possible to prove that a 4-point function computed
in the supergravity approximation admits an OPE interpretation. This would require the
knowledge of the conformal partial wave amplitude of an arbitrary tensor operator. Another
reason that complicates the analysis of 4-point functions is that in general one should split
the free field theory double-trace operators into a sum of operators with the same free field
theory dimensions, each one transforming irreducibly under the superconformal group.
In the context of the present work we are able to successfully deal with both the above
problems.
2 One can easily see that the normal-ordered product : O fr (x)O fr (y) : is the only term of order 1 in the free
1
2
field OPE of chiral operators.
553
3. OPE of the lowest weight CPOs

In this section we study the OPE of the lowest weight CPOs in free field theory and
at strong coupling. Recall that the normalized lowest weight CPOs in N = 4 SYM4 are
operators of the form
O I (x) =

23/2 2 I
Cij tr : i j : ,
where the symmetric traceless tensors CijI , i, j = 1, 2, . . . , 6, form a basis of the 20 of

SO(6) and satisfy the orthonormality condition
CijI CijJ = I J .
Using for the Wick contractions the following propagator
g 2 ab ij
j
ai b = Y M 2 ,
(2)2 x12
(3.1)
where a, b are color indices and xij = xi xj , one finds the following expressions for the
free field theory 2-, 3- [9] and 4-point functions of O I :

I1 I2
O I1 (x1 )O I2 (x2 ) fr = 2 ,
x12

1 23/2C I1 I2 I3
,
O I1 (x1 )O I2 (x2 )O I3 (x3 ) fr =
2 x2 x2
N x12
13 23

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) fr
"
#
I1 I2 I3 I4
I1 I3 I2 I4
I1 I4 I2 I3
=
+ 4 4 + 4 4
4 x4
x12
x13 x24
x14 x23
34
"
#
C I1 I3 I2 I4
C I1 I3 I4 I2
C I1 I2 I3 I4
4
+ 2 2 2 2 2 + 2 2 2 2 + 2 2 2 2 ,
N x12 x23 x34 x41 x13 x32 x24 x41 x13 x34 x42 x21
(3.2)
where the first term in the 4-point function represents the contribution of disconnected
diagrams. We have also introduced the shorthand notations C I1 I2 I3 = CiI11i2 CiI22i3 CiI33i1 and
I
C I1 I2 I3 I4 = CiI11i2 CiI22i3 Ci33i4 CiI44i1 for the trace products of matrices C I .

3.1. Free field theory OPE
The simplest way to derive the OPE in free field theory is to apply Wicks theorem.
Using the propagator (3.1) we find the following formula for the product of two CPOs
O I1 (x1 )O I2 (x2 ) =

I1 I2
23 2 I1 I2
+
Cik Ckj : tr i (x1 ) j (x2 ) :
4
2
x12
Nx12
+ : O I1 (x1 )O I2 (x2 ) :
(3.3)
On the r.h.s. of (3.3) we have bi-local operators of the form : O (x1 )O (x2 ) :, where O is
either i or O I1 and O is either j or O I2 . To find the operator content of the r.h.s. of (3.3)
554
one should perform the Taylor expansion of the operator O and rearrange the resulting
series as a sum of conformal blocks of local quasi-primary operators. It is clear that in
free field theory any bilocal operator : O (x1 )O (x2 ) : may be represented as an infinite
sum of conformal blocks of symmetric traceless rank l tensor operators with dimensions
+ + l + 2k,
: O (x)O (0) :=
X
l,k=0

1

x 2k x1 xl O(k)
(0) ,
1 l
(l + 2k)!
(3.4)
where the square brackets [ ] are used to denote the whole conformal block of a quasiprimary operator. In an interacting theory the tensor quasi-primary operators may acquire
anomalous dimensions. Explicit expressions of the tensor operators through O , O are
unknown and the best we can do is to find the first few terms in the series. In particular, as
shown in Appendix A, the terms up to two derivatives are given by the following formula
: O (x)O (0) : = : O (0)O (0) : +x : O (0)O (0) :
1
+ x x : O (0)O (0) :
2
1

1 2

(0) + x T (0) .
= O (0) + x O (0) x x T
2
2
(3.5)
Here the quasi-primary operators are given by
O = : O O :,

1
O = : O O O O :,
2

1
: O O :
T = : O O + O O :
2
2(2 + 1)

+1 2

2
2
:O O : +: O O :+:O O : ,
+
8
2 + 1

1 2
1

2
2
:O O : +: O O :+:O O : ,
T =
8
2 1
where is the conformal dimension of the operators O , O which takes the values 1 and
2 in the cases under consideration.
Obviously the conformal dimensions of the scalar operators O and T are equal
to 2 and 2 + 2, respectively, the dimension of the vector operator is 2 + 1 and the
dimension of the traceless symmetric tensor operator is 2 + 2. Consider first the case
when = 1. The scalar operator tr( i j ) is decomposed into a sum of the traceless part
in the 20 which is a lowest weight CPO O I and the trace part. The trace part is the
2
2
normalized Konishi scalar field K = 32
1/2 tr( ). If = 1 the vector and tensor operators
ij
are conserved and the operator T vanishes because of the on-shell equation 2 i = 0.
In fact the conserved current transforms in the 15 irrep of SO(6) and is the R-symmetry
ij
current of the free field theory of 6 scalars i . Decomposing the tensor operator T into
irreducible representations of the R-symmetry group SO(6), i.e., into the traceless and
555
ii coincides with
trace parts with respect to the indices i, j , one sees that the trace part T
ij
the stress tensor of the free field theory. The Konishi scalar and the traceless part of T
are dual to string modes and are expected to decouple in the strong coupling limit.
To complete the consideration of the free field theory OPE we have to decompose
the remaining operators into irreducible representations of SO(6) SU(4). One has the
general decomposition of the 20 20 of SU(4) as
20 20 = [0, 0, 0] + [0, 2, 0] + [0, 4, 0] + [2, 0, 2]
+ [1, 0, 1] + [1, 2, 1].
(3.6)
The representations in the first and the second lines of (3.6) are symmetric and
antisymmetric in the indices of the 20s I1 , I2 , respectively. The dimensions of the
representations are
D([0, 0, 0]) = 1,
D([2, 0, 2]) = 84,
D([0, 2, 0]) = 20,
D([0, 4, 0]) = 105,
D([1, 0, 1]) = 15,
D([1, 2, 1]) = 175.
(3.7)
I I
Introducing the orthonormal ClebshGordon coefficients CJ1D2

I1 I2 I1 I2
CJ
CJ 0 = JD JD0 ,
D
D
where JD is the index of an irrep of dimension D, as well as the operators

O JD = CJ1D2 : O I1 O I2 :,
I I
OJD = CJ1D2 OI1 I2 ,

I I
(3.8)
we can write
I1 I2 J20
I1 I2 J105
I1 I2 J84
O
+ CJ
O
+ CJ
O ,
: O I1 O I2 : = I1 I2 O1 + CJ
20
105
84

1
I1 I2 J15
I1 I2 J175
O + CJ
O
,
OI1 I2 = : O I1 O I2 : : O I1 O I2 : = CJ
15
175
2
I1 I2
and T I1 I2 . Note that the operators have the following
and a similar decomposition for T
3
free field theory 2-point functions

J1 J2
,
O J1 (x1 )O J2 (x2 ) = 2 + O(1/N 2 )
8
x12

I (x12 ) J J
J
1 2,
O 1 (x1 )OJ2 (x2 ) = 4 + O(1/N 2 )
10
x12
where I (x) = 2x x /x 2 . The precise values of the normalization constants will

be determined in the next section. Due to the definition of the double-trace operators, the
3-point normalization constants which appear in the following 3-point functions

O I1 (x1 )O I2 (x2 )O JD (x3 ) = COOOD
I1 I2
CJ
D
|x12|4D |x13|D |x23 |D
3 The only exception is the operator O = (1/20) : O I1 O I2 : in the singlet representation, whose normalization
1
constant is 1/10 + O(1/N 2 ).
556
I1
O (x1 )O
I2
I1 I2 2
2 x
CJ
x23 x31 x31
23
D

(x2 )OJD (x3 ) = COOOD
|x12|5D |x13|D +1 |x23 |D +1

are equal to the 2-point normalization constants COD .
Combining all pieces together we obtain the first few terms in the free field OPE of the
CPOs as
O I1 (x1 )O I2 (x2 )
23/2 I1 I2 I 1 I
2
1
I1 I2
C
O + 1/2 I1 I2 2 K
= 4 +
2
N
3 N
x12
x12
x12

fr 4 2 x12
x12 I1 I2 I I
27/2 2 x12 I1 I2 J15 2 2 I1 I2 x12 x12
C
C
J
T
+
T
J
2
2
2
15
N x12
3N
N x12
x12

I1 I2 J20
I1 I2 J105
I1 I2 J84
O
+
C
O
+
C
O
+ I1 I2 O1 + CJ
J
J
20
105
84

I I
I I
(3.9)
+ CJ1152 x12 OJ15 + CJ11752 x12 OJ175 + .
fr is the stress tensor of the free field theory of six scalar fields, while the normalized
Here T
R-symmetry current JJ15 is defined as follows

J 1
JJ15 = Cij 15 tr : i j : : i j : ,
2
J
where the antisymmetric tensors Cij 15 form a basis of the 15 of SO(6) and satisfy the
J
J0
orthogonality condition Cij 15 Cij 15 = J15 J15 . The R-symmetry current has the following
2-point function
0
J15
JJ15 (x1 )J

2 J15 J 0 I (x12 )
15
(x2 ) =
.
6
8 4
x12
We would like to stress that in addition to the above fields the OPE contains infinite towers
of both single-trace as well as double-trace operators.
3.2. Strong coupling OPE
As was explained in the previous section, the strong coupling OPE of CPOs is easily
determined from the cubic terms in the scalars s I dual to the lowest weight CPOs in
the type IIB supergravity action. There are three different cubic vertices in the action
describing the cubic couplings among the three scalars s I , the interaction of the scalars
with the graviton and the interaction with the SO(6) vector fields. Thus, according to the
discussion in the previous section the strong coupling OPE has the form
O I1 (x1 )O I2 (x2 ) =
I1 I2
4
x12
23/2 I1 I2 I 1 I 27/2 2 x12 I1 I2 J15

C
CJ15 R
O +
2
2
N
3N x12
x12

(1)

2 2 I1 I2 x12 x12
I1 I2 1
x
T
+
O1
12
2
15N
x12
(1)
(1)
(1)
I1 I2
I1 I2 105 J105
I1 I2 84 J84
J20
20
x
x
x
O
+
C
O
+
C
+ CJ
J105 12
J84 12 O
20 12

(1)
(1)
I1 I2 15 J15
I1 I2 175 J175
+ CJ
x
x
x
O
+
C
+ .
12
J175 12 x12 O
15 12
557
(3.10)
Here R 15 is the R-symmetry current and T is the stress tensor of N = 4 SYM4 .

J
The structure constants of the operators O I , R 15 , T are found by requiring that the
above OPE reproduces the known 3-point functions of two CPOs with another CPO, the
R-symmetry current and the stress tensor respectively, as the latter were computed in
the supergravity approximation in [9,10]. The operator algebra structure constants of the
double-trace operators in (3.10) are chosen to be 1, which means that their 2- and 3-point
normalization constants are kept equal. The anomalous dimensions 1 , 20 , . . . , 175 of
the double-trace operators will be determined in the next section by studying the 4-point
functions of the CPOs.
Comparing (3.10) with (3.9), we see that the structure of the strong coupling OPE is
simpler than the corresponding free field theory one. Instead of having an infinite number of
single-trace operators as in (3.9), we find in (3.10) only three single-trace operators giving
rise to the most singular terms. The coefficients in front of the R-symmetry current and the
stress tensor are, however, different from the ones in (3.9). The reason is that the free field
J
fr receiving contribution only from bosons may be represented as
operators J 15 and T
1
2 J15
;
JJ15 = RJ15 + K
3
3
1
10
18
fr
T
= T + K + ,
5
35
35
(3.11)
J15
and K are vector and tensor operators from the Konishi supermultiplet
where K
which has as leading component that scalar K, while is the leading component of
a new supersymmetry multiplet. The splitting (3.11) is explained by the fact that T ,
K and have pairwise vanishing two-point functions [39,40] and belong to different
supersymmetry multiplets. The operators in the Konishi supermultiplet as well as are
dual to string modes and therefore decouple in the strong coupling limit.
A splitting analogous to (3.11) may also occur for the free field theory double-trace
operators. However, there is an important difference. If we assume that all operators have
free field theory 2-point normalization constants of order 1, then the splitting has the
following schematic form
O fr = O gr +
1 str
O ,
N
where a free field theory double-trace operator O fr is split into a sum of operators O gr dual
to supergravity multi-particle states, and operators O str dual to string states. As follows
from the discussion in the previous section the coefficient in front of O str has to be of
order 1/N , because otherwise one would not reproduce the disconnected part of the 4point function. Such a splitting manifests itself in the 1/N 2 corrections to 2- and 3-point
normalization constants of double-trace operators. In what follows we will be mostly
interested in double-trace operators with free-field dimensions 4 and 5. We will see that
such a splitting does occur for all the operators except the operators in the 20 and 105
irreps.
558
4. Anomalous dimensions of double-trace operators

In this section we determine the anomalous dimensions of double-trace operators and the
leading 1/N 2 corrections to their 2- and 3-point function normalization constants CD (N).
To this end, we study the asymptotic behavior of the 4-point functions of CPOs in the direct
2 , x 2 0. Since we know all the 4-point functions, we do not need to consider
channel x12
34
the crossed channels. It is well-known that a conformally-invariant 4-point function is
given as a general analytic function of two variables, which are here conveniently chosen
to be the biharmonic ratios
2
2 x2
x12
x 2 x34
34
,
v
=
.
u = 12
2 x2
2 x2
x13
x
24
14 23
We also use in the following the variable Y = 1 v/u. The biharmonic ratios above and
2 , x 2 0.
the variable Y have the property that u, v, Y 0 as x12
34
To perform the computation we need to know the contributions of various quasi-primary
operators and their descendents in the 4-point functions of CPOs, i.e., the conformal
partial wave amplitudes of quasi-primary operators. We restrict ourselves mainly to the
contributions of scalar, vector and second rank symmetric traceless tensor operators. Let
the OPE of CPOs be of the form

J
1 J COOT x12 x12
I1 I2 COOS
S
+
T
O I1 (x1 )O I2 (x2 ) = CJ
4
6
S
T
CS x
CT x
12
12

COOV x12 J
V
+
,
(4.1)
+
CV x 5V
12
where J denotes an index of an irreducible representation of the R-symmetry group SO(6),
I1 I2
are the ClebshGordon coefficients and S , T , V are the conformal dimensions
CJ
of the scalar, tensor and vector operators respectively. For any of the operators, CO and
COO O denote the normalization constant in the 2-point function hO(x1 )O(x2 )i and the
coupling constant in the three-point function hO I (x1 )O J (x2 )O(x3 )i, respectively. Then,
one can show that the short-distance expansion of the conformal partial amplitudes of the
scalar S, tensor T and vector V operators can be written as [32]
I I
I I
C 1 2C 3 4
O I1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) = J 4 4J
x12x34

2

S
3S
C
S + 2
S
Y+
v 1+
Y +
OOS v 2 1 +
CS
4
16(S 1)(S + 1)
4

2
T
C
1
1
1
v
vY
+ OOT v 2 1 Y 2
CT
4
T
T

V 1
C2
1
Y + .
(4.2)
+ OOV v 2
CV
2
The formulas for the leading contributions of a rank-2 traceless symmetric tensor and a
vector can be generalized to the case of a rank-l traceless symmetric tensor of dimension
l and one gets a leading term of the form
l l
2
559
Y l.
For this reason a term of the form v /2 F (Y ) in a 4-point function contains, in principle,
the contributions not only from a scalar operator, but also from any symmetric tensor
operator of rank l and conformal dimension + l. Moreover, (4.2) shows that the
(0)
anomalous dimensions are related to terms of the type v S

((0)
V 1)/2
(0)
(T 2)/2
/2
log v for scalar operators,
Y 2 log v
Y log v for vector operators and v

for rank-2 tensor operators.
The 4-point functions of CPOs were computed in the supergravity approximation in [28]
and can be written as follows

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )
I1 I2 I3 I4
4 x4
x12
34
I1 I3 I2 I4
I1 I4 I2 I3
4 x4
4 x4
x13
x14
24
23

C

8
I I I I
2 2
2 2
+ 2 2 21 2 23 4 2 x13
x24 x14
x23 D2222
N
x12 x34
2
2
2
2
x24
D1212 x13
D2121 + x14
D2112 + x23
D1221

2 x2 + x2 x2 x2 x2 )
(x13
3
1
24
14 23
12 34
D3322 + D2222
2 D2211 +
+
2
2
2x34
x34

1
2
D2233 3D2222 + t + u ,
(4.3)
+ 2CI+1 I2 I3 I4 2 D2211 + 4x34
x34

where CI1 I2 I3 I4 = 12 CI1 I2 I3 I4 CI2 I1 I3 I4 and t and u stand for the contributions of the
t- and u-channels obtained by the interchange 1 4 and 1 3, respectively. The Dfunctions are defined as
I1 I2 I3 I4
D1 2 3 4 (x1 , x2 , x3 , x4 )
Z
d1+1 +2 +3 +4
x0
d+1
.
= d x 2
[x0 + (x x1 )2 ]1 [x02 + (x x2 )2 ]2 [x02 + (x x3 )2 ]3 [x02 + (x x4 )2 ]4
(4.4)
It is convenient to represent D-functions in the form
D1 2 3 4 (x1 , x2 , x3 , x4 )
=
D 1 2 3 4 (v, Y )
2 )
(x12
1 +2 3 4
2
2 )
(x13
1 +3 2 4
2
2 )
(x23
2 +3 +4 1
2
.
2 )4
(x14
As shown in Appendix C, a D-function

is given by a convergent series in v and Y . In terms
of the biharmonic ratios u and v the 4-point function acquires the form

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )

1
= 4 4 I1 I2 I3 I4 + u2 I1 I3 I2 I4 + v 2 I1 I4 I2 I3
x12 x34
560
(

1
8
v3
v2
2
3
+v
+ 2 2 4 4 CI1 I2 I3 I4 D2222 2v 2 + vu v u
N x12x34
u
u

2

v3
2v
2
v +
+ D 2112 2v vu + v 2
+ D1212
u
u

4v 3
+ 4v 2 D 3223
+ D 2211 vu v 2 + D 2323
u

v3
+ v2 u + v3
+ CI+1 I2 I3 I4 D 2222 12v 2 vu
u

3

v
+ D 2112 vu + v 2
+ D 1212 v 2 +
u

4v 3
D2323 + 4v 2 D 3223
+ D 2211 2v vu v 2 + 8v 2 D 3322 +
u

3
v
v2 u v3
+ CI1 I3 I2 I4 D 2222 6v 2 + vu +
u

3

v
+ D 2112 vu + v 2
+ D 1212 v 2
u

4v 3
2
2
D2323 + 4v D3223
+ D2211 vu + v +
u

1
3
v2
v2
+ I1 I2 I3 I4 v 2 D 2222 v D 2211 + D 3322 v +
2
2
u

1
v3
3
+ v3
+ I1 I3 I2 I4 v 2 D 2222 v 2 D 1212 + D 2323 v 2
2
2
u

)
3
1 2
v
I1 I4 I2 I3 3 2
2
3
v D2222 v D2112 + D 3223 v +
+v
.
+
2
2
u
(4.5)
This 4-point function is given as a sum of contributions from quasi-primary operators

transforming in the six irreducible representations (3.6) of SO(6). It is clear that to obtain
a contribution of operators belonging to a D-dimensional irrep one should multiply the
I1 I2 I3 I4
CJD which is a projector onto the irrep.
4-point function by a SO(6) tensor CJ
D
In what follows it will be sometimes useful to compare the short-distance expansion of
the 4-point function (4.5) with the one of the free field 4-point function (3.2), which in
terms of the biharmonic ratios takes the form

O I1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) fr
1 h
= 4 4 I1 I2 I3 I4 + u2 I1 I3 I2 I4 + v 2 I1 I4 I2 I3
x12 x34
i
4
+ 2 (u + v)CI+1 I2 I3 I4 + (v u)CI1 I2 I3 I4 + uvCI1 I3 I2 I4 .
N
(4.6)
561
4.1. Projection on the singlet

First we project the 4-point function on the singlet part that amounts to applying to it
From the strong coupling OPE (3.10) we expect to find the stress tensor
contribution and a contribution of the double-trace scalar operator O1 of approximate
dimension 4.
The result for the connected part is

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 1
"

8
3v 3
I1 I2 I3 I4
2
2
3
3vu
+
3v
=
u
+
3v
9v
D
2222
4 x4
u
20 2 N 2 x12
34

v3
19 2
19
v +3
+ D 2112 3vu + v 2
+ D 1212
6
u
6

20 2
10
v2
2
+ D2211 v 3vu 3v + D3322 20 + 20v + v

3
u
3

3
3
41v
41 2 v
+ v 3 + D 3223
v +
+ v3 .
+ D 2323 v 2 +
3u
3
u
1 I1 I2 I3 I4
.
400
Using the formulas for the D-functions

from the Appendix C, we can find that the most
singular terms of the v-expansion are

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )
1
I1 I2 I3 I4
4 x4
N 2 x12
34

vF1 (Y ) + v 2 F2 (Y ) + v 2 log vG2 (Y ) ,
(4.7)
where
4Y 2 8Y
4(6 + 6Y Y 2 ) log(1 Y )
+
,
Y3
3Y 3
1680 + 3360Y 2108Y 2 + 428Y 3 21Y 4
F2 (Y ) =
15(1 Y )Y 4

4 1140 1890Y + 962Y 2 151Y 3 + 5Y 4
log(1 Y )
15Y 5
16(Y 2)(6 6Y + Y 2 )
Li(Y ),
+
Y5

4(6 6Y + Y 2 ) 12 12Y + Y 2 6(Y 2) log(1 Y )
+
.
G2 (Y ) =
Y 1
Y
3Y 4
F1 (Y ) =
Expanding the functions in powers of Y we then obtain

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 1
#
"

4
43 2
1 I1 I2 I3 I4 2
2
2 47
vY + v
log v
v Y .
= 2 4 4
N x12 x34 45
225 5
225
(4.8)
562
Comparing this asymptotics with (4.2), we see that the contribution from a scalar field of
dimension 2 is absent, as it should be, since the Konishi field acquires large anomalous
dimension and decouples in the strong coupling limit. We also get the relation:
2
COOT
2
=
.
4CT
45N 2
Since for COOT one has COOT =
4 4
3 2 N
one finds at strong coupling
102
,
4
which represents the normalization of the complete stress tensor of the N = 4 SYM4 [44].
As it was discussed above, a term of the form vF (Y ) contains, in general, contributions
from all traceless symmetric tensor operators of rank l and dimension 2 + l. However,
comparing F1 (Y ) in (4.7) with the corresponding term in the conformal partial wave
amplitude of the stress tensor (B.15) we see that they coincide. Thus, the strong coupling
OPE does not contain single-trace rank-l traceless symmetric tensors with dimension 2 + l
in its singlet part. Nevertheless, it may in principle contain tensors of dimension 4 + l or
higher. However, as it was shown in Section 3 a possible single-trace scalar operator of
dimension 4 vanishes. Thus the only scalar operator of approximate dimension 4 is the
double-trace operator O1 . 5
The formula (4.8) also allows us to determine the anomalous dimension of O1 .
Assuming the existence at strong coupling of a scalar field with dimension = (0) +
(1) , where (0) = 4 and (1) is the anomalous dimension, we find that
CT =
1
v /2 = v 2 + (1) v 2 log v + .
2
Since there is only one operator of approximate dimension 4, we do not face the problem
of operator mixing and from (4.2) we get
2
4
1 COOO1 (1)
= 2.
2 CO1
5N
2
/CO1 the O(1) result which is 1/10. In this
Since (1) is of order 1/N 2 we use for COOO
1
way we obtain
16
,
(4.9)
N2
for the anomalous dimension of O1 . This coincides with the anomalous dimension of the
operator B considered in [23], as it should be, since B is a descendent operator of O1 .
(1) =
4 This value of the coupling constant is fixed by a conformal Ward identity [41], the same value was also
obtained in the supergravity approximation in [10].
5 The free field theory operator O fr probably splits into a linear combination of O and an operator O str dual
1
1
1
to a string mode. However, the coefficient in front of O1str is of order 1/N , and even if the latter operator does
not decouple in the strong coupling limit it cannot contribute to log-dependent terms in 4-point functions. In the
following, when discussing double-trace operators in other irreps we simply assume that operators such as O1str
above do decouple, making at the same time a consistency check to confirm our assumption.
563
We can also find the leading 1/N 2 correction to the 2- and 3-point normalization
constant CO1 . Writing as

1 (1)
1
1 + 2 CO1 ,
CO1 =
10
N
and taking into account that CO1 = COOO1 , we find from the term of order v 2
38
.
15
Finally, we can make a consistency check of our computation. Namely, since we know
corrections to the conformal dimension, (1) = 16/N 2 and to the structure constant
we can compute the term of order v 2 Y by using (4.2), in order to compare it with
the corresponding value obtained from our 4-point function. Taking into account the
contribution of the stress tensor we get from (4.2) and from the expansion of our 4-point
function the same number 43/225. This also confirms that there is only one operator of
approximate dimension 4 in the strong coupling OPE, and that the operator O1str decouples
in the strong coupling limit.
We can also compute the 2-point normalization constant in free field theory by using
(3.1) and the definition of the operator. A simple calculation gives

1
2
1+
.
CO1 =
10
3N 2
Thus, not only the conformal dimension but also the 2- and 3-point normalization constants
get 1/N 2 corrections in the strong coupling limit.
(1)
CO
=
1
4.2. Projection on 20
According to (4.2), to obtain the contribution of the operators transforming in a Ddimensional irrep, we should multiply the 4-point function by the projector onto the
representation

1 I1 I2 I3 I4
C C ,
(4.10)
PD I I I I =
1 2 3 4
D JD JD
where
X II II II II
CJ1D2 CJ3D4 CJ1 0 2 CJ3 0 4 ,
D =
Ii
is the dimension of the irrep so that PD2 = 1/D .

The projector on the 20 can be easily found by taking into account that the Clebsh
I1 I2
is proportional to the SO(6) tensor C I1 I2 I3 . Then, one can show
Gordon coefficient CJ
20
that

3
1
+
I I I I .
C
(4.11)
P20 I I I I =
1 2 3 4
100 I1 I2 I3 I4 6 1 2 3 4
Using the Table 1 from Appendix D for the contractions of the projector with the SO(6)
tensors appearing in the 4-point function, we find the contribution of the operators in the
20 to the connected part of the 4-point function
564

O I1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 20
I1 I2 I3 I4 "

3v 3 3 2
8 CJ20 CJ20
3
3 3
2
+ v u+ v
= 2 2
D2222 18v vu
4 x4
N
2
2u
2
2
x12
34

4 2
3
3 2
4 2 3v 3
3
10
+ D2112 vu + v + D2211
v vu v
+D1212 v +
3
2u
2
3
3
2
2

#
3
3
19v
19 2 v
40
+ v 3 + D 3223
v +
+ v3 .
(4.12)
+ v 2 D 3322 + D 2323 v 2 +
3
3u
3
u
Expanding the D-functions

in powers of v, we obtain

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )
20
I1 I2
I3 I4

1 CJ20 CJ20
vF1 (Y ) + v 2 F2 (Y ) + v 2 log vG2 (Y ) ,
2
4
4
N
x12x34
(4.13)
where
40 log(1 Y )
,
3Y
8(65 65Y + 6Y 2 ) 20(74 49Y + 2Y 2 )
log(1 Y )
F2 (Y ) =
3(1 Y )Y 2
3Y 3
160(Y 2)
Li(Y ),
+
Y3

40 12 12Y + Y 2 6(Y 2) log(1 Y )
+
.
G2 (Y ) =
3Y 2
Y 1
Y
F1 (Y ) =
Expanding the above functions in powers of Y we finally obtain

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 )
I1 I2
I3 I4
1 CJ20 CJ20
4 x4
N 2 x12
34
20

26
4
40
v + v 2 (1 + Y ) v 2 Y 2 log v .
3
9
3
(4.14)
The analysis of the results obtained follows the one in the previous subsection. Firstly,
comparing F1 (Y ) in (4.13) with the corresponding term of the conformal partial amplitude
of a scalar operator of dimension 2 (B.2), we see that they coincide. 6 Therefore, all singletrace rank-l traceless tensors of dimension 2 + l transforming in the 20 are absent in
the OPE. Then, the only scalar operator of approximate dimension 4 is the double-trace
operator O20 . Moreover, we see that log v-dependent terms appear starting from the term
v 2 Y 2 log v. Thus we conclude from (4.2) that the double-trace operator O20 has protected
conformal dimension. It is worth noting that the non-renormalization of the conformal
dimension of this operator is not related to the shortening condition discussed in [33] and is
a prediction of the AdS/CFT correspondence. The first operators which acquire anomalous
dimensions are scalar and tensor operators of approximate dimension 6.
2
6 Recall that for the lowest weight CPOs one has COOO = 40 .
CO
3
565
The first 1/N 2 correction to the 2- and 3-point normalization constant CO20 can also be
easily found. Writing the constant as

1 (1)
CO20 = 2 1 + 2 CO20 ,
N
and taking into account the contribution of the single-trace operator O I and that CO20 =
COOO20 , we find from the term of order v 2
1
(1)
CO20 = .
3
The 2-point normalization constant can be also computed in free field theory by using (3.1)
and the definition of the operator (3.8) and appears to coincide with the value obtained in
the strong coupling limit

1
.
CO20 = 2 1 +
3N 2
Thus, both the conformal dimension and the 2-point function normalization constant of the
double-trace operator in the 20 are non-renormalized in the strong coupling limit. This also
shows that in this case there is no splitting, and the free field theory double-trace operator
coincides with O20 .
The free field theory OPE (3.9) and the strong coupling OPE (3.10) do not contain
single-trace operators transforming in the 105 irrep. Thus, only double-trace operators
contribute to this part of the 4-point function. The corresponding connected contribution
can be easily found using the Table 1 from Appendix D and is given by

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 105
I1 I2 I3 I4 "

8 CJ105 CJ105
v3
2
2
3
v
= 2 2
+
vu
+
u
v
D
3v
2222
4 x4
N
u
x12
34

v3
1
1
+ D 2112 vu + v 2 + D 2211 vu + v 2
+ D 1212 v 2
2
u
2
#
3

v3
3v
3
2
2
3
+ v + v + D3223 3v + v +
.
(4.15)
+ D2323
u
3


I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 105
I I
=
where
I1 I2
3 4

1 CJ105 CJ105 2
v F2 (Y ) + v 3 F3 (Y ) + v 4 F4 (Y ) + v 4 log vG4 (Y ) ,
2
4
4
N
x12 x34
(4.16)
566
4
,
1Y
4(Y 2)
8
log(1 Y ),
F3 (Y ) =
(1 Y )Y 2 Y 3
4(28 28Y + 3Y 2 ) 8(38 25Y + Y 2 )
log(1 Y )
F4 (Y ) =
(1 Y )Y 4
Y5
96(Y 2)
Li(Y ),
+
Y5

8 12 12Y + Y 2 6(Y 2) log(1 Y )
+
.
G4 (Y ) = 4
Y 1
Y
Y
F2 (Y ) =
Since only double-trace operators contribute, it is useful to compare (4.16) with the
corresponding part of the free field theory 4-point function (4.6)

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) fr 105
#
I1 I2 I3 I4 "

1 CJ105 CJ105 2
1
v2 4
= 2
.
(4.17)
v 1+
+ 2
4 x4
N
(1 Y )2
N 1Y
x12
34
The first term on the r.h.s. of this equation shows the disconnected part of the free field
theory 4-point function. Comparing the term of order 1/N 2 in (4.17) with the term v 2 F2 (Y )
in (4.16), we see that they coincide. This means that the conformal dimensions and the
leading corrections in 1/N 2 to 2- and 3-point functions normalization constants of any
symmetric traceless rank-2k tensor operator of dimension 4 + 2k transforming in the
105 coincide with the ones computed in free field theory. Thus, all these operators are
non-renormalized in the strong coupling limit. The first correction to the 2- and 3-point
functions normalization constant of the double-trace operator O105 can be easily found
from (4.17) and is given by

2
CO105 = 2 1 + 2 .
N
The non-renormalization of the double-trace operator O105 follows from the shortening
conditions derived in [33,34], and was also checked in perturbation theory at small YM
coupling in [33,36,38,45].
The expansion (4.16) also shows that the first log v-term appears at order v 4 . Therefore,
all symmetric traceless rank-2k tensor operators of dimension 6 + 2k transforming in the
105 have protected conformal dimensions. Note, however, that the normalization constants
of their 2- and 3-point functions certainly receive corrections at strong coupling, which
are encoded in the function F3 (Y ). The vanishing of anomalous dimensions of these
tensor operators does not seem to follow from any known non-renormalization theorem.
These results also demonstrate that the free field theory symmetric traceless rank-2k tensor
operators of dimension 4 + 2k do not split, while the ones with dimension 6 + 2k do.
1
+ the first double-trace operator in the 105 which acquires
Since G4 (Y ) = 10
anomalous dimension is the scalar operator with approximate dimension 8.
567
Just as it was the case for the operators in the 105, only double-trace operators
transforming in the 84 irrep can contribute to this part of the 4-point function. The
corresponding connected contribution is again found by using the Table 1 from Appendix
D:

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 84

I1 I2 I3 I4 "

v3
8 CJ84 CJ84
uv uv 2 v 3
2
+
+
= 2 2
D2222 6v
4 x4
N
2
2
2
2u
x12
34

3
v
uv
uv v 2
2
2
v D2211
+
+ D2112
+D1212 v +
2u
2
2
2
#

3
3
3v
v
+ D 3223 3v 2 + v 3 +
.
(4.18)
+D 2323 v 2 + v 3
u
u


I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 84
I I
I1 I2
3 4

1 CJ84 CJ84 2
= 2
v F2 (Y ) + v 3 F3 (Y ) + v 3 log vG3 (Y ) ,
4
4
N
x12x34
(4.19)
where
8(3 3Y + Y 2 ) 12(Y 2)
+
log(1 Y ),
(1 Y )Y 2
Y3
8(Y 2)(21 21Y + 2Y 2 )
F3 (Y ) =
(1 Y )Y 4
4(228 + 264Y 80Y 2 + 3Y 3 )
144(Y 2)2
+
log(1
Y
)
Li(Y ),
Y5
Y5

12(Y 2) 12 12Y + Y 2 6(Y 2) log(1 Y )
+
.
G3 (Y ) =
Y4
Y 1
Y
F2 (Y ) =
Since the first log v-term appears at order v 3 , all symmetric traceless rank-2k tensor
operators of dimension 4+2k transforming in the 84 have protected conformal dimensions.
The first double-trace operator in the 84 which acquires an anomalous dimension is the
scalar operator with approximate dimension 6. However, contrary to the case of the 105
irrep, the leading 1/N 2 corrections to the normalization constants of the 2- and 3-point
functions of these operators differ from their free field theory values. To see this we
compare (4.19) with the corresponding part of the free field theory 4-point function (4.6)

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) fr 84

I1 I2 I3 I4

1 CJ84 CJ84 2
1
v2 2
.
(4.20)
= 2
v
1
+
+
4 x4
N
(1 Y )2
N2 1 Y
x12
34
568
Expanding (4.19) and (4.20) in powers of Y , we obtain the normalization constants of

2- and 3-point functions of the operator O84 at strong coupling and in free field theory
correspondingly as

3
1
str
fr
CO84 = 2 1 2 .
CO84 = 2 1 2 ,
N
N
The vanishing of the anomalous dimensions of the double-trace operator O84 follows from
the shortening conditions discussed in [33,34] and was also shown in perturbation theory
str and C fr again may find a
at small YM coupling in [38,45]. The difference between CO
O84
84
natural explanation in the fact that the corresponding free field theory operator undergoes
a linear splitting on O84 and K84 , where O84 has protected both its dimension and the
normalization constants of the 2- and 3-point functions, while the operator K84 belongs to
the Konishi multiplet [38] and, therefore, decouples at strong coupling.
By using the projector (P15 )I1 I2 I3 I4 constructed in the Appendix D and the results of
Table 1 we find the following contribution of the operators in 15 to the connected part of
the 4-point function

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 15

I1 I2 I3 I4 "

8 CJ15 CJ15
2v 3
4v 2
2
3
2uv
= 2 2
+
2v
D
4v
+
2uv
2222
4 x4
N
u
u
x12
34

3v 2 4v 2 2v 3
3v 2
+
+
+ D 2112 4v 2uv +
+D 1212
2
u
u
2
#

3

7v
v3
2
2
3
2
3
D2323 + 7v + v +
D3223 .
+D2211 2vu 2v + v v
u
2u
Expansion of the D-functions in powers of v produces now the following expression for
leading terms

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 15
I I
I1 I2
3 4

1 CJ15 CJ15
= 2
vF1 (Y ) + v 2 F2 (Y ) + v 2 log vG2 (Y ) .
4
4
N
x12x34
(4.21)
Here the functions F1 , F2 and G2 are given by

16
2Y + (Y 2) log(1 Y ) ,
2
Y
4(Y 2)(56 56Y + 5Y 2 )
F2 (Y ) =
(Y 1)Y 3
8(152 + 176Y 53Y 2 + 2Y 3 )
192(Y 2)2
+
log(1
Y
)
Li(Y ),
Y4
Y4

16(Y 2) 12 12Y + Y 2 6(Y 2) log(1 Y )
+
.
G2 (Y ) =
Y3
Y 1
Y
F1 (Y ) =
569
Expansion in powers of Y produces the following leading terms

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 15

I1 I2 I3 I4
CJ
CJ
12 2
16 2
8
15
15
vY
+
v
v
=
Y
Y
log
v
.
4 x4
3
25
5
N 2 x12
34
(4.22)
The absence of the terms vY log v shows that the vector operator of the dimension 3,
J
which is the R-symmetry current R 15 , has protected conformal dimension. According
to the discussion above, the function F1 (Y ) may receive contributions from single-trace
rank 2k + 1 traceless tensors of dimension 2k + 3 transforming in 15, which is what
indeed happens in the free field theory limit. However, comparing the function F1 (Y ) with
the relevant part of the conformal partial amplitude of the conserved vector current of
dimension 3 (B.10) one concludes that they coincide, therefore, the corresponding tensors
are absent in the strong-coupling OPE. Next, comparing (4.22) with Eq. (4.2) we read off
the value of the ratio
2
COOR
8
=
.
2CR
3N 2
1/2
Since the value of COOR is fixed by the conformal Ward identity to be COOR = 2 2 N one
finds
32
CR =
8 4
which corresponds to the normalization of the two-point function of the complete Rsymmetry current of the N = 4 SYM4 [41,42].
The function F2 (Y ) receives contributions both from the R-symmetry current and from
traceless symmetric rank 2k + 1 tensors with approximate dimension 2k + 5. Since RJ15
is non-renormalized, the presence of the function G2 shows that operators from the above
tensor tower acquire anomalous dimensions. We can now find the anomalous dimension
of the lowest current O15 in this tower whose free field theory counterpart OJ15 with
conformal dimension (0) = 5 was discussed in Section 3. Comparing the coefficient in
front of v 2 Y log v in (4.22) with the asymptotic (4.2) one finds
2
16
1 COOO15 (1)
= 2.
4 CO15
5N
The free field result gives
(1) =
2
COOO
15
CO15
= 4, therefore we find
16
,
5N 2
(4.23)
J
for the anomalous dimension of O 15 . Since the conformal partial amplitude of the
conserved vector current is known one can also find the correction to the normalization
constant CO15 in the same manner as was done in the previous cases. Indeed, the v2 Y term
in (4.22) is split as
8
4
12 2
v Y = v 2 Y v 2 Y.
25
15
75
570
Here the first term is a contribution of the conformal partial amplitude of the R-symmetry
current, while the second one is related to the correction to the normalization constant
CO15 . One can easily see that in the free field theory COOO15 = CO15 = 4 is an exact result,
i.e., it does not receive 1/N 2 corrections. Thus, if we write

1 (1)
CO15 = 4 1 + 2 CO15 ,
N
then from (4.2) it follows that
(1)
=
CO
15
2
.
75

Only double-trace operators transforming in the 175 appear in the free field theory
OPE (3.9) and in the strong coupling OPE (3.10). Applying the projector (P175 )I1 I2 I3 I4
constructed in Appendix D to the 4-point function we find the following expression for the
contribution of the operators in the 175:

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 175
I1 I2 I3 I4 "
8 CJ175 CJ175
v2
v2
= 2 2
+
D
D2112
1212
4 x4
2
2
N
x12
34
#

v3
v3
2
3
2
3
+ v +v
D2323 + v v
D3223 .
u
u
Expanding D functions in v we keep the leading terms v 2 and v 3

I
O 1 (x1 )O I2 (x2 )O I3 (x3 )O I4 (x4 ) 175
I I
I1 I2
3 4

1 CJ175 CJ175 2
v F2 (Y ) + v 3 F3 (Y ) + v 3 log vG3 (Y )
2
4
4
N
x12 x34
with
4(Y (Y 2) + 2(Y 1) log(1 Y ))
,
Y 2 (Y 1)
4(28 28Y + 3Y 2 ) 8(38 25Y + Y 2 ) log(1 Y )
F3 (Y ) =
Y 3 (Y 1)
Y4
96(Y 2)
+
Li(Y ),
Y4

8 12 12Y + Y 2 6(Y 2) log(1 Y )
+
.
G3 (Y ) = 3
Y
Y 1
Y
F2 (Y ) =
The function F2 receives contributions from tensor operators of rank 2k + 1 with

approximate dimensions 2k + 5. Since the term proportional to v 2 log v is absent, we
conclude that these tensor operators have protected conformal dimensions. The lowest
J
current O 175 among them, with dimension 5, was discussed in Section 3. Note that these
571
operators also contribute to F3 together with operators of rank 2k + 1 and approximate

dimensions 2k + 7. For the two terms of the Y -expansion one finds

1 2
3 4

1 CJ175 CJ175
4 2
2 2

v Y 2v Y . (4.24)
O (x1 )O (x2 )O (x3 )O (x4 ) 175 = 2
4 x4
N
3
x12
34
I I
I1
I2
I3
I I
I4
This allows us to determine the 1/N 2 correction to the 2- and 3-point normalization
constant CO175 of the operator OJ175 . Taking into account that in free field theory
fr
fr
= CO
= 4 as can be easily seen from the free field theory 4-point function
COOO
175
175
(4.6), we write as

1 (1)
CO175 = 4 1 + 2 CO175 .
N
Then from the first term of order v 2 in (4.24) one finds
2
(1)
CO175 = .
3
Apparently, the splitting mechanism is again at work, i.e., the corresponding free field
theory operator is split in two orthogonal parts carrying different representation of the
supersymmetry; one has protected both its dimension and the normalization constants,
while the other one is dual to a string mode and decouples at strong coupling.
5. Conclusions
We studied in detail the 4-point functions of the lowest weight CPOs and we showed
that they have a structure compatible with the OPE of CPOs predicted by the AdS/CFT
correspondence. We demonstrated that all power-singular terms in the 4-point functions
exactly match the corresponding terms in the conformal partial wave amplitudes of the
CPOs, of the R-symmetry current and of the stress tensor. As these operators are dual to
type IIB supergravity fields, we concluded that the operators dual to string modes, which
appear in the free field theory OPE, decouple in the strong coupling limit.
We also computed the anomalous dimensions and the leading 1/N 2 corrections to the
normalization constants of the 2- and 3-point functions of the scalar double-trace operators
with approximate dimension 4 and of vector operators with approximate dimension 5.
The only scalar double-trace operator that acquires an anomalous dimension appears to be
the operator in the singlet of the R-symmetry group SO(6). The double-trace operator
in the 20 seems to be protected, however as this does not follow from the shortening
condition discussed in [33,34] we do not have a satisfactory explanation for such a nonrenormalization property.
The anomalous dimension of the singlet operator is negative, hence this operator is
relevant and can be used to study non-conformal deformations of the N = 4 SYM4 . All
other scalar double-trace operators have protected dimension 4 and are marginal. They can
be added to the Lagrangian in order to study conformal deformations. Nevertheless, it is
572
unclear at present how dual deformations of type IIB supergravity (or string theory) can be
described.
We have also found several towers of traceless symmetric double-trace operators in
the 105, 84 and 175 irreps, whose anomalous conformal dimensions vanish. The rank-2k
tensor operators of dimension 6 + 2k satisfy the shortening condition A0 ) of [34]. However,
even if they contain the highest weight states of the SU(2, 2|4) superalgebra the shortening
condition A0 ) does not imply non-renormalization of the corresponding multiplets. On the
other hand operators from other towers are certainly not the highest weight states, and at
present we are not aware if the lowest weight states of their supermultiplets satisfy the
shortening condition responsible for non-renormalization.
There are two interesting facts related to the structure of the leading log-dependent terms
in the 4-point functions. Namely, all the functions G(Y ) which appear in (4.7), (4.13) and
so on, differ from each other by some simple rational factors. We expect that this is an
indication that the anomalous dimensions of all double-trace operators may be related by
some relatively simple formula. Then, the leading log v-dependent terms appear in the 4point functions exactly at the same order of v where the dilogarithm Li appears for the first
time.
Acknowledgements
We would like to thank A. Tseytlin for valuable comments. G.A. is grateful to S. Theisen
and, especially, to S. Kuzenko for discussions of the structure of the Konishi multiplet. S.F.
is grateful to S. Mathur and, especially, to A. Tseytlin for valuable discussions. A.C.P.
wishes to thank W. Rhl for sharing with him his insight on CFT. The work of G.A.
was supported by the Alexander von Humboldt Foundation and in part by the RFBI grant
N99-01-00166. The work of S.F. was supported by the US Department of Energy under
grant No. DE-FG02-96ER40967 and in part by RFBI grant N99-01-00190. The work of
A.C.P. was supported by the Alexander von Humboldt Foundation. G.A. and A.C.P. wish
to acknowledge the warm hospitality and financial support of the E.S.I. in Vienna where
part of the work was done.
Appendix A. Free field OPE and conformal blocks

A quasi-primary field of the CFT appearing in the OPE together with all its derivative
descendents is known as a conformal block. If two fields O and O transforming in some
representation of an R-symmetry group have the one and the same conformal dimension
then their OPE has the following structure
O (x)O (0) =
1
(x 2 )
+
1
2 (2O )
C(x, )O (0) +
1
(x 2 )
1
2 (2T +2)
1
(x 2 )
1
2 (2J +1)
C (x, )T
(0) + .
C (x, )J (0)
(A.1)
573
Here we identify the leading quasi-primary fields with conformal dimensions O , J and
T as a scalar O , a vector current J and a symmetric traceless second rank tensor T

respectively. The OPE coefficient C(x, ) denotes a power series in derivatives generating
the conformal block [O ] of the scalar O . Similarly we denote the OPE coefficients for
for the other fields.
The structure of the conformal blocks is uniquely fixed by the conformal symmetry and
it may be found by requiring consistency of the OPE with 2- and 3-point functions of the
fields involved. In particularly, the conformal block of a scalar field with dimension is
given by the following differential operator [32,4951]:
C(x, y ) =
1
1
1
2 O , 2 O
Z1
X
m=0
1
m!(O + 1)m

1 1
1
dt t (1 t) 2 O
t (1 t)x 2 1y et xy ,
4
(A.2)
where the Euclidean spacetime dimension d enters as d = 2, xy = x y, , 1y = y2

and we use the Pochhammer symbol (a)n = (a + n)/ (a). In what follows we need to
specify explicitly the first three terms of C(x, y ) in the derivative expansion:
1
O + 2
(xy )2
C(x, y ) = 1 + (xy ) +
2
8(O + 1)
O
x 2 1y + .
16(O + 1)(O + 1 )
(A.3)
The conformal blocks of a conserved vector current and a conserved second rank tensor
with canonical dimensions 2 1 and , respectively, are also available. For a vector
current one has [30,31]
C (x, y ) =
x X
1
B(, )
m!()m
m=0

m
1
2
dt t (1 t)
t (1 t)x 1y et xy
4
0

1
x
1 2
2
= x + x (xy ) +
( + 1)(xy ) x 1y +
2
4(2 + 1)
2
Z1
1
(A.4)
and for a conserved symmetric traceless tensor one finds (see Appendix E)
C (x, y ) =
X
x x
1
B( + 1, + 1)
m!( + 1)m
m=0
Z1

m

1
dt t (1 t)
t (1 t)x 2 1y et xy
4
= x x + .
(A.5)
574
Using the above formulae, one can now consider the operator product : O (x)O (0) :
in a free field theory and find explicit expressions for J and T . Indeed, from
the Taylor expansion one sees that the leading component is a quasi-primary field
O =: O (x)O (0) : with conformal dimension O = 2, therefore it should appear
in the OPE with its whole conformal block. Subtracting from the Taylor expansion the
first three terms of the conformal block of the scalar with dimension 2 we find at
the next level another quasi-primary operator O that turns out to be a vector current
J = 12 : ( O O O O ) : with dimension J = 2 + 1. Now subtracting from

what we get the first two terms of the conformal block of the vector current 7 and
decomposing the resulting second rank tensor on the traceless and trace parts we are left
with two new fields, one is a tensor and another one is a new scalar, which are given by

1
: O O + O O :
: O O :
2
2(2 + 1)

+1 2

2
2
:O O : +: O O :+:O O : ,
+
4
2 + 1

1
+1 2
: O O : + : 2O O : + : O 2O : .
=
4
2 + 1
=
T
The transformation properties of these fields under the conformal group show that the
are both quasi-primary. Thus, for = 2 we get the desired result (3.5). Note that T is
conserved while T vanishes on-shell as soon as = + 1. Clearly with the knowledge
of the conformal blocks of the higher rank tensor operators the procedure of identifying
the quasi-primary operators on the r.h.s of (3.4) may be extended to any desired order.
Appendix B. Conformal partial wave amplitudes of a scalar, a conserved vector

current and the stress tensor
The full contribution of the conformal block of an operator carrying and irreducible
representation of the conformal group into the 4-point function is known as the conformal
partial wave amplitude (CPWA). The scalar CPWA was computed in [4951] by evaluating
the corresponding scalar exchange diagram. If we consider operators with the same
conformal dimension, then the CPWA of a scalar operator with dimension S contributes
to its 4-point function as [32]:
HS (v, Y ) = v
S /2
n
X
v
n=0
4
1
2 S n
n! (S )2n (S + 1 )n
2 F1
1
2 S

+ n, 12 S + n; S + 2n; Y ,
(B.1)
7 We do not assume here that J is conserved, however the first two terms in the conformal blocks of the
conserved and non-conserved vector currents are the same.
575
where we have represented the result as the convergent series in conformal variables v and
Y . The first few terms of the v, Y expansion of HS (v, Y ) are given in (4.2). In particular,
for S = 2 the first term of v-expansion reads as
3
vF1 (Y ) + ,
(B.2)
40
where F1 (Y ) is defined in Section 4.2.
The CPWA of traceless symmetric tensors of dimension and rank l, corresponding to
irreducible representations of dimension and spin l of SO(d, 2), can be also calculated
in CFT as the relevant graphs reduce to sums of scalar exchanges. Using the following
normalization prescriptions [4648] for the 2- and 3-point functions of the exchanged
tensor fields

N (, l)
(x
)
I
(x
)
traces
,
I
M1 ,..,l (x1 )M1 ,..,l (x2 ) = C,l
12
12
l
l
1
1
sym
2
x12

, l)
g ,l
N (;

l
1
2
trace terms ,
O(x1)O(x3 )M1 ,2 ,..,l (x5 ) =
1
1 2 2 1
2 )
2 (x x ) 2
(x13
( 2 ) 2 l
15 35
HS (v, Y ) =
where the normalization constants are taken to be

2 ( + l) (d 1)
,

1
(2) 2 d 12 d (d + l 1)
N (, l) =
2+ 2 + 2 l
, l) =
N (;
(2) 2 d
!1/2

+ 12 + 12 l 12 d 12 + 12 l 2 12 + 12 l

d 12 + 12 l 12 d + 12 + 12 l 2 12 d 12 + 12 l
and
(1, 2; 3) =
(x13 )
2
x13
(x23 )
2
x23
2 (1, 2; 3) =
2
x12
2 x2
x13
23
the contribution of the tensor field to the 4-point function of a scalar operator with
dimension takes the form
(x1 , x3 ; x2, x4 ; , l)
= ;,l
1
1 2
1 d+ 1
2 )
2 (x )
2
2
(x13
24
dd x5
{e1 el traces}{e0 1 e0 l traces}

1
15 35
is then given by
The constant ;,l
=
;,l
g 2
,l
C,l
(x 2 x 2 ) 2 (x 2 x 2 ) 2 d 2
25 45

1
1
22+ 2 d+ 2 l 12 + 12 l + 12 + 12 l 12 d

,
(2)d/2 12 d + 12 + 12 l d 12 + 12 l
(B.3)
576
where we have introduced the concise notation
(1, 3; 5)
(2, 4; 5)
,
e0 =
,
e =
2
1/2
2
1/2
| (1, 3; 5)|
| (2, 4; 5)|
e e = e0 e0 = 1.
One can show that for the general tensor exchange (B.3) is reduced to a finite sum of
four-star integrals S(a1 , a2 ; a3, a4 ):
Z
1
(B.4)
S(a1 , a2 ; a3, a4 ) = d4 x5 2a 2a 2a 2a ,
1
2
x x x 3x 4
15
25
35
45
which can be directly evaluated. The final result is obtained after dropping the shadow
series of the four-star integral, as the latter corresponds to the exchange of the shadow
tensor field with dimension d .
Here, we apply the general formula (B.3) to the two cases we are interested in the paper;
the case of the conserved vector current with = d 1 and l = 1 and the stress tensor
with = d and l = 2. Choosing to work directly in d = 4, the contribution of a conserved
vector field in the scalar four-point function is given by
Z
1
e e0
.
(B.5)
d4 x5
2 (x1 , x2 ; x3, x4 ; 3, 1) = 2;3,1
3
1
2 )2 2 (x 2 )2 2
2 x 2 ) 32 (x 2 x 2 ) 12
(x12
(x
34

15 25
The inner product

1
e e0 =
2
e e0
35 45
can be written as
!1/2"
2
x 2 x 2 x 2 x 2
x24
15 25 35 45
2 x2
x12
x 2 x 2
34
25 45
2
x14
2
x13
2
x23
#
.
(B.6)

2
2
S(2 + , 1 + ; 1 , ) x23
S(1 + , 2 + ; 1 , ) .
+ x13
(B.7)
x 2 x 2
15 45
x 2 x 2
15 35
x 2 x 2
25 35
Substituting (B.6) into (B.5) we obtain four 4-star functions as

1
1
2 (x1 , x2 ; x3, x4 ; 3, 1) = 2;3,1 2
2 )2
21
2
(x12) (x34
2
2
x24 S(1 + , 2 + ; , 1 ) x14
S(2 + , 1 + ; , 1 )
Note that we have also regularized the dimension of the vector field as = 3 + 2 to
deal with the singularities contained in the four-star functions involved into (B.7). The
singularities are avoided by keeping the regulating parameter non-zero in the intermediate
stages of the calculation. The analyticity of the exchange graph then ensures that taking
the limit 0 at the end of the calculation one recovers the correct result. Using the
expression for the four-star function derived in [26] we then obtain (here we present the
formula for general d and to ensure a wider applicability of our result)

2 12 2 + 12
1
2
(x1 , x2 ; x3, x4 ; , 1) = ;,1

1
2
2 )
3
2 2 2 + 2 () (x12x34

HV (v, Y ) + shadow part ,
where the function HV (v, Y ) represents the CPWA of the vector current
577
3 1 X v n Y m
1
HV (v, Y ) = v 2
4
n!m! (1 + )n ()2n+m
n,m=0
"
2

+1
1 2 +1 2
1 2
+
2
2
2
2
n
n+m
n
n+m

1
+1
+1
1
2
2
2
2
2
n
n
n+m
n+m

2
2 #
1
+1
Y
.
2
2
n
n+m
(B.8)
For = d 1 = 3 the CPWA of the vector current simplifies to give
1
3 X vn Y m
HV (v, Y ) = v
4
n!m! (2)n (3)2n+m
n,m=0

(2)2n (1)2n+m + (1 Y )(1)2n (2)2n+m 2(1)n (2)n (2)n+m (1)n+m , (B.9)
and it is normalized to start as HV (v, Y ) = 12 Y + (cf. (4.2)). To make a comparison

with the supergravity results in Section 4.5 we need to single out in Eq. (B.9) the leading-v
contribution. Putting in the previous formula n = 0 and performing the summation in m
we obtain
3
(B.10)
HV (v, Y ) = vF1 (Y ) + ,
16
where F1 (Y ) is defined in Section 4.5.
Analogously, the contribution of the stress tensor is given by

Z
e e 14 e0 e0 14
1
4
d x5
2 (x1 , x2 ; x3, x4 ; 4, 1) = 2;3,1 2
3/2 2 2 1/2 . (B.11)
(x34 )2
x 2 x 2
x x
15 25
35 45
Using then (B.6) and regularizing the tensor dimension as = 4 + 2 we obtain

1
1 2
x24 S(1 + , 3 + ; 1 , 1 )
2 (x1 , x2 ; x3, x4 ; 4, 1) = 2;3,1 2
2 x2
(x34 )2 x12
34
4
2
S(3 + , 3 + ; 1 , 1 ) + x13
S(3 + , 1 + ; 1 , 1 )
+ x14
2
2 2
S(1 + , 3 + ; 1 , 1 ) 2x24
x14 S(2 + , 2 + ; 1 , 1 )
+ x23
2 2
2 2
x23S(2 + , 2 + ; 1 , 1 ) + 2x24
x13 S(2 + , 2 + ; , )
2x13
2 2
2 2
x23S(1 + , 3 + ; 1 , 1 ) 2x14
x13 S(3 + , 1 + ; , )
2x24

1
2 2
(B.12)
x23S(2 + , 2 + ; , ) S(2 + , 2 + ; , ) .
+ 2x14
4
One then observes that (B.12) contains a number of four-star functions which are O() and
therefore vanish in the 0 limit. These are all the four-star functions with in the last
two positions. Then, by virtue of
(2 2) 1
= + O(),
(1 )
2
(B.13)
578
the remaining four-star functions give a finite result which reads

2 (x1 , x2 ; x3, x4 ; 4, 1) =

2
1
2,4,2 4 4 HT (v, Y ) + shadow part ,
12
x12 x34
where HT (v, Y ) represents the CPWA of the stress tensor:
5 X vn Y m
1
HT (v, Y ) = v
4
n!m! (3)n (4)2n+m
n,m=0
2 2
(3)n (1)n+m + 2(3)n (1)n (3)n+m (1)n+m + (1 Y )2 (1)2n (3)2n+m

2(3)n (2)n (1)n+m (2)n+m 2(1 Y )(1)n (2)n (3)n+m (2)n+m . (B.14)
The normalization of HT (v, Y ) is fixed such that its v, Y expansion reproduces the
corresponding terms in (4.2). Again to establish a link with supergravity results in Section
4.1 we single out the v term in Eq. (B.14) and, performing the summation in m, get
45
vF1 (Y ) + ,
(B.15)
8
where F1 (Y ) is the function defined in Section 4.1. This completes the construction of the
CPWA for conserved vector and tensor currents.
HT (v, Y ) =
Appendix C. Series representation for D-functions

Here we derive a representation for the D 1 2 3 4 -functions in a form of a convergent
series in v and Y variables by using a technique similar to [18].
We start with the definition (4.4). Standard Feynman parameter manipulations based on
the formula
Z
1
1
=
dt t 1 et z ,
z ()
0
and two integrals

Z
P
P
P
d i i
1
2 d1+ i
dx0 = (St ) 2
e i t i x 0 x 0
2
Z
d/2
1 P
t t x2
dd xE eti |Ex Exi | = d/2 e St i<j i j ij ,
St
i
2
lead to
D1 2 3 4 (x1 , x2 , x3 , x4 )
Z
1 +2 +3 +4
2
= K dt1 dt4 t11 1 t44 1 (St )
0

1
2
2
exp
t1 t2 x12 + + t3 t4 x34 ,
St
d
2

,
where the short-hand notations

3 +4
d/2 1 +2 +
2
K=
St = t1 + t2 + t3 + t4 ,
2 (1 ) (4 )
were introduced. Performing the change of variables
X

ti
1/2
tj0 ti0 uti0 ,
det
= 2u4 ,
ti = St ti0 =
tj0
d
2
579

,
one obtains
Z
D1 2 3 4 (x1 , x2 , x3 , x4 ) = 2K
"
dt1 dt4 t11 1 t44 1 exp
#
ti tj xij2
i<j
Now we rescale the variables ti : ti i ti , where the constant parameters i are chosen to
induce the following scale transformations
1
1
1
1
t1 t3 2 t1 t3 ,
t1 t4 2 t1 t4 ,
t2 t3 2 t2 t3 ,
t1 t2 2 t1 t2 ,
x12
x13
x14
x23
and as the consequence
t2 t4 =
x2
t2 t3 t1 t4
2 132 t2 t4 ,
t1 t3
x14 x23
t3 t4 =
x2
t2 t3 t1 t4
2 122 t3 t4 .
t1 t2
x14 x23
Under this rescaling the integral transforms into

D1 2 3 4 (x1 , x2 , x3 , x4 )
=
D 1 2 3 4 (v, Y )
2 )
(x12
1 +2 3 4
2
2 )
(x13
where
1 +3 2 4
2
2 )
(x23
2 +3 +4 1
2
(C.1)
2 )4
(x14
dt1 dt4 t11 1 t22 1 t3 3 t44 1

v
exp t1 t2 t1 t3 t1 t4 t2 t3 t2 t4 vt3 t4 ,
u
and the integral is understood as a function of the conformal variables v and Y .
Next, using the MellinBarnes integral representation
D 1 2 3 4 (v, Y ) = 2K
1
exp[z] =
2i
r+i
Z
ds (s)zs ,
r < 0,
1
|arg z| < ,
2
ri
for the two exponentials in the last formula which involve v/u and v the integral reduces
to
D 1 2 3 4 (v, Y )
s
Z
d ds
v
(s)
()v
= 2K
u
(2i)2
Z

dt1 dt4 t11 1 t22 +s1 t33 +1 t44 +s+1 exp t1 t2 t1 t3 t1 t4 t2 t3 .
580
The following change of variables:

t1 t2 = u1 ,
t1 t3 = u2 ,
t1 t4 = u3 , t2 t3 = u4 ,

ti
1
det
,
=
uj
2u1 u2
allows one to perform the t-integration with the result

D 1 2 3 4 (v, Y )

Z
dds
(s) ()
=K
(2i)2
1 +2 3 4
2
2 +3 +4 1
2
1 +3 2 4
2
s
s

v
+ s + (4 + s + )v
.
u
The s-integration is then performed by using the integral and series representations for the
hypergeometric function F (a, b, c; 1 z):
F (a, b, c; 1 z) =
(c)
(a) (b) (c a) (c b)
Zi
1
ds zs (s) (c a b s) (a + s) (b + s),
2i
i
and
F (a, b, c; 1 z) =
(c) X (a + m) (b + m)
(1 z)m ,
(a) (b)
(c + m)m!
m=0
where one needs to substitute

2 + 3 + 4 1
+ ,
b = 4 + ,
c = 3 + 4 + 2.
a=
2
Thus one arrives at the convergent hypergeometric series in the variable Y :
D 1 2 3 4 (v, Y )
(Z

d
Ym
()
=K
m!
2i
1 +2 3 4
2
m=0
2 +3 +4 1
2
3 +4 +1 2
2

+ (3 + )
(3 + 4 + 2 + m

+ + m (4 + + m)v
)
.
(C.2)
Since for any D-function

occurring in the 4-point function of CPOs the quantity 1 +
2 3 4 is an integer, the final MellinBarnes integral receives a contribution from
double poles and, therefore, the integration can be done by using the general formula

Z
X

ds 2
vn
d
s
(s)g(s)v =
g( ) =n , (C.3)
2(n + 1)g(n) g(n) ln v
2i
d
(n!)2
C
n=0
valid for any function g(s) regular at s = 0. In this way we arrive at the representation for
D-functions
in terms of double convergent series in v and Y variables.
581
Below we list explicitly the series representations for D-functions

we used in the paper
D 2222 (v, Y ) = 2
X
Y m v n (n + 2)2 (2 + n + m)2
m! (n!)2
(4 + 2n + m)
m,n=0

1
1
+ (4 + 2n + m) (n + m + 2) ln v ,

n+1
2
2 X Y m v n (n + 2) (n + 1) (n + m + 1) (n + m + 2)
D 2112 (v, Y ) =
2
m! (n!)2
(3 + 2n + m)
m,n=0

1
+ 2(3 + 2n + m)

n+1

(n + m + 1) (n + m + 2) ln v ,
D 1212 (v, Y ) = 2
X
Y m v n (n + 1)2 (n + m + 2)2
m! (n!)2
(3 + 2n + m)
m,n=0

1
(3 + 2n + m) (n + m + 2) ln v ,
2
2
D 2211 (v, Y ) =
2
D 3322 (v, Y ) =
4
X
Y m v n n (n + 1)2 (n + m + 1)2
m! (n!)2
(2 + 2n + m)
m,n=0

1
2(n + m + 1) + 2(2 + 2n + m) ln v ,
n
X
Y m v n n (n + 2)2 (2 + n + m)2
m! (n!)2
(4 + 2n + m)
m,n=0

3n + 1
+ 2(4 + 2n + m) 2(2 + n + m) ln v ,

n(n + 1)
2 X Y m v n (n + 2)2 (3 + n + m)2
D 2323 (v, Y ) =
2
m! (n!)2
(5 + 2n + m)
m,n=0

1
1
+ (5 + 2n + m) (3 + n + m) ln v ,

n+1
2
2 X Y m v n (n + 2) (n + 3) (2 + n + m) (3 + n + m)
D3223 (v, Y ) =
4
m! (n!)2
(5 + 2n + m)
m,n=0

3n + 5
+ 2(5 + 2n + m)

(n + 1)(n + 2)

(2 + n + m) (3 + n + m) ln v .
(C.4)
582
Appendix D. Projectors
Here we give an explicit construction of the projectors that single out the contributions
of irreps occurring in the decomposition 20 20 of SO(6) from the 4-point function of the
lowest weight CPOs.
J
Matrices CijI and Cij 15 introduced in Section 3 obey to the following summation
formulae [28]:
X J J
X

1
1
1
1
I
CijI Ckl
= ik j l + il j k ij kl ,
Cij 15 Ckl 15 = ik j l il j k .
2
2
6
2
J15
I1 I2
I1 I2
and CJ
It is then easy to check that the orthonormal ClebshGordon coefficients CJ
20
15
are given by
31/2 I1 I2 I
1
J
I1 I2
C
,
CJ
= 1/2 CijI1 CjI2k Cik15 .
(D.1)
15
51/2
2
The other coefficients are constructed in a similar manner. Irreps 84, 105 and 175 are
D
with the normalization condition
described by traceless rank 4 tensors CijJkl
I1 I2
=
CJ
20
J0
D
D
Cij kl
= JD JD .
CijJkl
84
is antisymmetric in i, k and in j, l and symmetric under permutation of the
Tensor Cij kl
84
105
= 0. Then Cij kl
is
pairs ij and kl. It is also required to obey the condition ij klmn Cklmn
175
is symmetric in i, k and in j, l and antisymmetric
a totally symmetric and, finally, Cij kl
under permutation of the pairs ij and kl.
A projector on the contribution of irrep D into the 4-point function is defined by
I1 I2 I1 I2
CJ20 and
(4.10) with D being the dimension of the representation. The sums CJ
20
I1 I2 I1 I2
CJ15 are computed straightforwardly by using Eqs. (D.1). To find the other projectors
CJ
15
we introduced the following three tensors QID1 I2 being elements of the corresponding
representations:

I2
I1 I2
I1 I2
I2
Ckj
Cil + Ckl
Cij CilI1 Ckj
QI841 I2 ij kl = CijI1 Ckl

1 I1 I2
I1 I2
I1 I2
I1 I2
Cmj il + Ckm
Cml ij Cim
Cml kj
Cim Cmj kl Ckm
4

1 I1 I2
I2
I1 I2
I1 I2
il + Clm
Cmk ij Clm
Cmi kj
+ Cj m Cmi kl CjI1m Cmk
4

1
I1 I2 ij kl il kj ,
10
+
1 I2
QI105

ij kl
I2
I1 I2
I1 I2
I1 I2
I2
= CijI1 Ckl
+ Cik
Cj l + CilI1 CjI2k + Ckl
Cij + Ckj
Cil + CjI1l Cik
1
1
I1 I2
I1 I2
I1 I2
I2
ij Ckm
Cml + Clm
Cmk kl Cim
Cmj + CjI1m Cmi
5
5
1
1
I1 I2
I1 I2
I2
I1 I2
+ Ckm
Cmj
ik Cj m Cml + Clm Cmj il CjI1m Cmk
5
5
583
1
1
I1 I2
I1 I2
I1 I2
I1 I2
j k Cim
Cml + Clm
Cmi j l Cim
Cmk + Ckm
Cmi
5
5

1
ij kl + ik j l + il j k I1 I2 ,
+
20
1
I1 I2
I2
I1 I2
I1 I2
= Cik
Cj l CjI1l Cik
ij Ckm
Cml Clm
Cmk
8
1
1
I1 I2
I1 I2
I1 I2
I2
Cml Clm
Cmi il Ckm
Cmj CjI1m Cmk
kj Cim
8
8
1
I1 I2
I1 I2
kl Cim Cmj Cj m Cmi .
8
Clearly one may write

I1 I2 JD
Cij kl = D QID1 I2 ij kl ,
CJ
D
1 I2
QI175
ij kl
(D.2)
where D is a normalization constant. Then one finds

0

I I
I1 I2 I3 I4
I1 I2 JD I3 I4 JD
CJD = CJ
Cij kl CJ 0 Cij kl = D2 QID1 I2 ij kl QD3 4 ij kl
CJ
D
D
D
with the normalization constant D following from

D = D2 QID1 I2 ij kl QID1 I2 ij kl
and, therefore,
PD

I1 I2 I3 I4

QID1 I2 ij kl QID3 I4 ij kl

=
.
QIDJ mnsp QIDJ mnsp
In this way one obtains the following explicit expressions for projectors singling out the
contributions of the irreps:

1
3
1
+
C
P20 I I I I =
I I I I ,
P15 I I I I = CI1 I2 I3 I4 ,
1 2 3 4
1 2 3 4
30
100 I1 I2 I3 I4 6 1 2 3 4

1
1
2I1 I3 I2 I4 + 2I1 I4 I2 I3 + I1 I2 I3 I4 4CI1 I3 I2 I4 2CI+1 I2 I3 I4 ,
P84 I I I I =
1 2 3 4
504
5

1
1
16
2I1 I3 I2 I4 + 2I1 I4 I2 I3 + I1 I2 I3 I4 + 8CI1 I3 I2 I4 CI+1 I2 I3 I4 ,
P105 I I I I =
1 2 3 4
1260
5
5

1
I I I I I1 I4 I2 I3 + CI1 I2 I3 I4 .
P175 I I I I =
1 2 3 4
350 1 3 2 4
1 I1 I2 I3 I4

these projectors provide the orthogonal
One may check that together with 400
I
I
I
I
1
2
3
4
decomposition of the unity .
The following formulae
380
20
,
CI1 I2 I3 I4 CI2 I1 I3 I4 = ,
CI1 I2 I3 I4 CI1 I3 I2 I4 = 0,
3
3
200
20
,
CI+1 I2 I3 I4 CI1 I2 I3 I4 = 0,
CI+1 I2 I3 I4 CI1 I3 I2 I4 = .
CI+1 I2 I3 I4 CI+1 I2 I3 I4 =
3
3
are helpful to find the contractions of the projectors with tensors describing the 4-point
function. The results for contractions are summarized in the Table 1.
CI1 I2 I3 I4 CI1 I2 I3 I4 =
584
Table 1
The values of contractions of the projectors with tensors describing the structure of the 4-point
function of the lowest weight CPOs
Tensor
CI+I I I
1 2 3 4
CII I I
1 2 3 4
CI1 I3 I2 I4
I1 I3 I2 I4
I1 I4 I2 I3
(1/400)I1 I2 I3 I4
(P15 )I1 I2 I3 I4
(P20 )I1 I2 I3 I4
(P84 )I1 I2 I3 I4
(P105 )I1 I2 I3 I4
(P175 )I1 I2 I3 I4
1/6
0
5/3
0
0
0
0
2
0
0
0
0
1/60
0
1/6
1/2
1
0
1/20
1
1
1
1
1
1/20
1
1
1
1
1
Appendix E. Conformal block of the conserved 2nd rank tensor

Here we sketch the derivation of the conformal block of the conserved second rank
tensor. We do not use this result in the paper, however, we feel that it might be useful for
subsequent studies of the OPE.
We start with (A.1) and suppress the inessential indices and . The 3-point function is
given by the following expression

O(x)O(0)T (y) =
1
(x 2 ) 2 (2T +2) (y 2 ) 2 (T +2) ((y
1
x)2 ) 2 (T +2)
1

2 2
x y (y x)2 y 4 (y x) (y x)

(x y)4 y y + y 2 (y x)2 ((y x) y + y (y x) ) ,
where for simplicity we choose the constant COOT to be equal to unity. Compatibility
of the 3-point function with the conservation law requires the dimension of the tensor to
be canonical, i.e., T = d, where d = 2 is a spacetime dimension. However, in what
follows we meet certain divergences and that is why we keep in some places = d T
as a regularization parameter. Substituting Eq. (A.1) into the 3-point function, we get an
equation defining the conformal block
1
1
1
exy
1
T (T 2) (y 2 ) 12 (T 2)
2
(y ) 2 (T 2)
+
1
(y 2 ) 2 (T 2)
1
xy
(y 2 ) 2 (T 2)
1
1
1
1
exy
1
1
(
2)
(T 2)2
2
2
T
(y ) 2
(y ) 2 (T 2)
+
1
1
(y 2 ) 2 (T 2)
xy
1
1
(y 2 ) 2 (T 2)
1
1
1
d T
1
exy
+
exy
1
1
1
1
(
2)
dT
(y 2 ) 2 T
(y 2 ) 2 T
(y 2 ) 2 T
(y 2 ) 2 (T 2)
1
1
2
exy
+
1
d(T 2)2 (y 2 ) 12 (T 2)
2
(y ) 2 (T 2)
=
585

1 k
(x, y ) T (0)T (y) ,
k!
k=2
where we have introduced the following representation

C (x, y ) =
X
1 k
(x, y ).
k!
k=2
Using the series representation for the exponentials on the l.h.s. of the equation defining
the conformal block, one then obtains an equation for k (x, y ).
The 2-point function of the second rank tensor can be written in the form [41,43]

1
1
E; 2 2 ,
T (0)T (y) =
(T 3)(T 2)T (T + 1)
(y ) T
where again for simplicity we choose the constant CT to be the unity. Here E; is a
tensor with the following structure
T 3
4(T 2)

1 2
2 ( + ) + ( ) + longitudinal .
2
If we suppose that the conformal block acting on the 2-point function is symmetric traceless
and transversal, then only the first term here is of importance and one gets
4(T 1)(T 1 )(T ) k

1
(x, y ) 2 .
k (x, y ) T (0)T (y) =
(T 2)T (T + 1)
(y ) T
E; =
Now we substitute every function 1/(y 2 )a appearing in the equation defining the conformal
block for its Fourier transform
Z
1
eipy
1
2(a) ( a)
=
2
dp 2 a
2
a
2
(y )
(a) (2)
(p )
and find 8
Z
24 (2)(2 + 1)
(p q )(p q )
2
(p
)
(ixq)k .
dq
2
( 1) (2 )
((p q)2 q 2 )1+/2
Since the conformal block is applied to the traceless transversal operator (2-point function)
in the last expression we have omitted all trace and longitudinal terms proportional to
and to p , respectively. The equation can be then brought to the form
k (x, ip) =
k (x, ip) =
4
(p2 )
(2)(2 + 1)
Ik
2 ( 1) (2 ) 2 2 1
8 We keep only there where it is actually needed to compute the limit 0.

1; 2 + 1 ,
(E.1)
586
again modulo inessential trace and longitudinal terms. Here we introduced the following
integral
Z
(ixq)k
Ik (1 ; 2 ) = dq
((p q)2 )1 (q 2 )2
that is explicitly evaluated to give

[k/2]
X k (2m)!
1 2 2 m 2 1 2
k2m
(ixp)
(p )
x p
Ik (1 ; 2 ) =
(1 ) (2 )
4
2m m!
m=0
(k m + 2 ) (m + 1 )
(1 + 2 m ).
(k + 2 1 2 )
Again neglecting the trace and longitudinal contributions, we evaluate the limit 0
and normalize the resulting expression such that the first nontrivial term 2 starts as
2 (x, ip) = 2x x + . In this way we find the following expression
k (x, y ) = x x
[k/2]1
X
m=0
(2 + 2)
( + 1)

m
(k m + 1)
k!
1
(xy )k2m2 x 2 1y .
(k 2m 2)!m!
(k + 2)
4
Finally, performing the summation in k we recover the expression C (x, y ) for the
conformal block of the conserved second rank tensor given in Appendix A.
References
[1] J. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor.
Math. Phys. 2 (1998) 231.
[2] G.G. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string
theory, Phys. Lett. B 428 (1998) 105, hep-th/9802109.
[3] E. Witten, Anti-de Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9805028.
[4] G. Arutyunov, S. Frolov, On the origin of the supergravity boundary terms in the AdS/CFT
correspondence, Nucl. Phys. B 544 (1999) 576, hep-th/9806216.
[5] J.H. Schwarz, Covariant field equations of chiral N = 2 D = 10 supergravity, Nucl. Phys. B 226
(1983) 269.
[6] J.H. Schwarz, P.C. West, Symmetries and transformations of chiral N = 2 D = 10 supergravity,
Phys. Lett. B 126 (1983) 301.
[7] P.S. Howe, P.C. West, The complete N = 2 D = 10 supergravity, Nucl. Phys. B 238 (1984) 181.
[8] G. Arutyunov, S. Frolov, Quadratic action for type IIB supergravity on AdS5 S 5 , JHEP 08
(1999) 024, hep-th/9811106.
[9] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three-point functions of chiral operators in
D = 400 , N = 4 SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
[10] G. Arutyunov, S. Frolov, Some cubic couplings in type IIB supergravity on AdS5 S 5 and
three-point functions in SYM4 at large N , Phys. Rev. D 61 (2000) 064009, hep-th/9907085.
[11] S. Lee, AdS(5)/CFT(4) four-point functions of chiral primary operators: Cubic vertices, Nucl.
Phys. B 563 (1999) 349, hep-th/9907108.
587
[12] D. Freedman, S. Mathur, A. Matusis, L. Rastelli, Correlation functions in the CFT(d)/AdS(d +

1) correspondence, Nucl. Phys. B 546 (1999) 96, hep-th/9804058.
[13] H. Liu, A.A. Tseytlin, On four-point functions in the CFT/AdS correspondence, Phys. Rev. D 59
(1999) 086002, hep-th/9807097.
[14] D. Freedman, S.D. Mathur, A. Matusis, L. Rastelli, Comments on 4-point functions in the
CFT/AdS correspondence, Phys. Lett. B 452 (1999) 61, hep-th/9808006.
[15] G. Chalmers, K. Schalm, The large Nc limit of four-point functions in N = 4 super YangMills
theory from anti-de Sitter supergravity, hep-th/9810051.
[16] E. DHoker, D. Freedman, Gauge boson exchange in AdSd+1 , Nucl. Phys. B 544 (1999) 612,
hep-th/9809179.
[17] J.H. Brodie, M. Gutperle, String corrections to 4-point functions in the AdS/CFT correspondence, Phys. Lett. B 445 (1999) 296, hep-th/9809067.
[18] H. Liu, Scattering in anti-de Sitter space and operator product expansion, hep-th/9811152.
[19] E. DHoker, D. Freedman, General scalar exchange in AdSd+1 , Nucl. Phys. B 550 (1999) 612,
hep-th/9811257.
[20] E. DHoker, D. Freedman, S. Mathur, A. Matusis, L. Rastelli, Graviton exchange and complete
4-point functions in the AdS/CFT correspondence, hep-th/9903196.
[21] E. DHoker, D. Freedman, L. Rastelli, AdS/CFT 4-point functions: How to succeed at zintegrals without really trying, hep-th/9905049.
[22] Sanjay, On direct and crossed channel asymptotics of four-point functions in AdS/CFT
correspondence, Mod. Phys. Lett. A 14 (1999) 1413, hep-th/9906099.
[23] E. DHoker, S.D. Mathur, A. Matusis, L. Rastelli, The operator product expansion of N = 4
SYM and the 4-point functions of supergravity, hep-th/9911222.
[24] L. Hoffmann, A.C. Petkou, W. Rhl, A note on the analyticity of AdS scalar exchange graphs
in the crossed channel, hep-th/0002025, to appear in Phys. Lett. B.
[25] C.P. Herzog, OPEs and 4-point functions in AdS/CFT correspondence, hep-th/0002039.
[26] L. Hoffmann, A.C. Petkou, W. Rhl, Aspects of the conformal operator product expansion in
AdS/CFT correspondence, hep-th/0002154.
[27] G. Arutyunov, S. Frolov, Scalar quartic couplings in type IIB supergravity on AdS5 S 5 , hepth/9912210, to appear in Nucl. Phys. B.
[28] G. Arutyunov, S. Frolov, Four-point functions of lowest weight CPOs in N = 4 SYM4 in
supergravity approximation, hep-th/0002170, to appear in Phys. Rev. D.
[29] K. Symanzik, Small-distance-behaviour analysis and Wilson expansions, Commun. Math.
Phys. 23 (1971) 49.
[30] K. Lang, W. Rhl, The critical O(N) -model at dimension 2 < d < 4: a list of quasi-primary
fields, Nucl. Phys. B 402 (1993) 573.
[31] K. Lang, W. Rhl, The critical O(N) -model at dimension 2 < d < 4 order 1/N 2 : Operator
product expansions and renormalization, Nucl. Phys. B 377 (1992) 371.
[32] A.C. Petkou, Conserved currents, consistency relations and operator product expansions in the
conformally invariant O(N) vector model, Ann. Phys. 249 (1996) 180, hep-th/9410093.
[33] L. Andrianopoli, S. Ferrara, On short and long SU(2, 2/4) in the AdS/CFT correspondence,
hep-th/9812067.
[34] S. Ferrara, A. Zaffaroni, Superconformal field theories, multiplet shortening and the
AdS5 /SCFT4 correspondence, hep-th/9908163.
[35] B. Eden, P.S. Howe, C. Schubert, E. Sokatchev, P.C. West, Simplifications of four-point
functions in N = 4 supersymmetric YangMills theory at two loops, hep-th/9906051.
[36] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, On logarithmic behaviour in N = 4 SYM theory,
hep-th/9906188.
[37] B. Eden, C. Schubert, E. Sokatchev, Three-loop four-point correlator in N = 4 SYM, hepth/0003096.
588
[38] M. Bianchi, S. Kovacs, G. Rossi, Y.S. Stanev, Anomalous dimensions in N = 4 SYM at order
g 4 , hep-th/0003203.
[39] D. Anselmi, The N = 4 quantum conformal algebra, Nucl. Phys. B 541 (1999) 369, hepth/9809195.
[40] D. Anselmi, Quantum conformal algebras and closed conformal field theory, Nucl. Phys. B 554
(1999) 415, hep-th/9811149.
[41] H. Osborn, A. Petkou, Implications of conformal invariance in field theories for general
dimensions, Ann. Phys. (N.Y.) 231 (1994) 311, hep-th/9307010.
[42] R. Manvelyan, A.C. Petkou, R-current anomalies in the (2, 0) tensor multiplet in d = 6 and
AdS/CFT correspondence, hep-th/000310, to appear in Phys. Lett. B.
[43] J. Erdmenger, H. Osborn, Conserved currents and the energy momentum tensor in conformally
invariant theories for general dimensions, Nucl. Phys. B 483 (1997) 431, hep-th/9605009.
[44] H. Liu, A.A. Tseytlin, D = 4 Super-YangMills, D = 5 gauge supergravity and D = 4
conformal supergravity, Nucl. Phys. B 553 (1998) 88, hep-th/9804083.
[45] W. Skiba, Correlators of short multi-trace operators in N = 4 supersymmetric YangMills,
Phys. Rev. D 60 (1999) 105038, hep-th/9907088.
[46] G. Mack, Group theoretical approach to conformal-invariant quantum field theory, in: Caianiello
(Ed.), Renormalization and Invariance in Quantum Field Theory, Plenum Press, New York,
1974.
[47] V.K. Dobrev, V.B. Petkova, S.G. Petrova, I.T. Todorov, Dynamical derivation of vacuum
operator product expansion in Euclidean conformal quantum field theory, Phys. Rev. D 13
(1976) 887.
[48] E.S. Fradkin, M.Ya. Palchik, New developments in d-dimensional conformal quantum field
theory, Phys. Rep. 300 (1998) 1.
[49] S. Ferrara, R. Gatto, A.F. Grillo, Conformal invariance on the light-cone and canonical
dimensions, Nucl. Phys. B 34 (1971) 349.
[50] S. Ferrara, R. Gatto, A.F. Grillo, Tensor representations of conformal covariant operator product
expansions, Ann. Phys. 76 (1973) 161.
[51] S. Ferrara, R. Gatto, A.F. Grillo, G. Parisi, Covariant expression of the conformal four-point
function, Nucl. Phys. B 49 (1972) 77.

Instantons and multi-instantons in curvilinear

coordinates
A.A. Abrikosov Jr.
ITEP, 117 259, Moscow, Russia
Abstract
Instantons in regular and singular gauges and explicit multi-instanton solutions are generalized to
curvilinear coordinates. Expressions can be simplified by fitting gauge orientation of the solutions to
the vierbein. The gauge transformation replaces -symbols used in Cartesian formalism by constant
-symbols and generates a compensating addition to the gauge potential of pseudoparticles. Typical
examples (4-spherical, (2 + 2)- and (3 + 1)-cylindrical coordinates) are studied explicitly. Effects of
singularities of the compensating field on gauge dependent quantities are discussed. 2000 Elsevier
Science B.V. All rights reserved.
PACS: 12.38.Aw; 12.38.Lg; 11.15.-q; 11.15.Kc
Keywords: Instantons; Multi-instanton solutions; Curvilinear coordinates; Confinement models
Introduction
Instantons form an essential nonperturbative element of nonabelian gauge theories
applied in physics of strong and weak interactions. But although a significant knowledge
has been acquired since the pioneering paper [1] the topic is not exhausted yet. Among
others the question of the role of instantons in quark confinement remains unsettled.
Probably the problem will stand unless new methods and ideas appear. Reviews of the
current state of affairs can be found in [2,3].
Usually instantons are studied in 4-dimensional Euclidean space parametrized by
Cartesian coordinates. The demand for non-Cartesian coordinates first came from
investigation of anomalies and index theorems on manifolds with boundaries. Adaptation
of the results (see [4,5]) to practical problems of chiral symmetry breaking requires explicit
realizations of general formulae.
E-mail address: persik@vxitep.itep.ru (A.A. Abrikosov).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 5 0 - 8
590
A.A. Abrikosov Jr. / Nuclear Physics B 586 (2000) 589608
From physical point of view it would be promising to examine pseudoparticles in

phenomenological environment provided by the confinement models such as the QCDstring or MIT-bag. However Cartesian coordinates are not the best choice for such
objects. For example strings looks more natural in (2 + 2)-cylindrical coordinates while
a (3 + 1)-frame may be convenient for the bag. Studies of pseudoparticles against
nonuniform background fields may require even more imagination.
A notable progress in the instanton physics was associated with the exact multi-instanton
solutions. The general configuration [6] proved too complicated and up to now the most
universal solution known explicitly is that found by Jackiw, Nohl, and Rebbi (JNR) who
have generalized the famous t Hoofts ansatz [7]. Most of instanton-based models start
from pseudoparticles in singular gauge that are a specific case of the latter solution.
Another advantage of the ansatz is that it makes possible to calculate exact Green functions
in multipseudoparticle field [8]. Besides it provides a way to study QCD at nonzero
temperature [9,10] and quark density [11].
The purpose of the present paper is to approach JNR and t Hoofts multipseudoparticle
solutions and single instantons in curvilinear coordinates. This offers a chance to benefit
from the spatial symmetry of other physical objects. This must simplify calculations and,
consequently, could widen the research field.
The basic component of the Cartesian JNR and t Hoofts solutions are the so-called
-symbols (-tensors) that project (anti-)selfdual tensors onto the SU(2) gauge group, [12].
However, in non-Cartesian frames -symbols are no longer constant tensors but depend
on coordinates. Our idea is to substitute them by constant -symbols in the vierbein
formalism. This can be done by means of a gauge transformation of the original (anti-)
selfdual field. The resulting potential is a sum of the instanton part and compensating field
which allows for geometry. It appears due to the gauge transformation and is uniquely
defined by the vierbein. Calculation of the compensating connection for given vierbein is
straightforward. In the present paper we shall discuss general aspects of the approach and
illustrate it by explicit examples.
The paper has the following structure. We start from reminding the basics: Section 1
introduces instanton and multi-instanton solutions while Section 2 deals with curvilinear
coordinates. Section 3 is dedicated to the compensating gauge connection. First we
calculate the compensating gauge potential and then discuss the relation between - and
-symbols. After that we turn to pseudoparticles in non-Cartesian frame in Section 4. We
calculate vector potentials and gauge field strengths for a general N -instanton configuration
and for an instanton in singular and regular gauges. 1
Section 5 presents a detailed study of instanton in 4-spherical coordinates. We calculate
explicitly the (gauged) pseudoparticle fields starting from singular and regular gauges.
If the instanton center lies at the origin the both results coincide. The calculation of the
ChernSimons number proves that being gauge dependent it feels the singularity of the
compensating field. In Section 6 we compute the compensating connection for (2 + 2)and (3 + 1)-cylindrical coordinates. The results are summarized in the conclusion.
1 More precisely our solutions coincided with those in Cartesian coordinates.
591
1. Instantons and multi-instantons

This section reviews basic facts on instantons in 4-dimensional Euclidean space. For the
time being we shall focus on the Cartesian coordinates and make no distinction between the
upper and lower indices. We shall consider the YangMills theory with the SU(2) gauge
group. The action of the gauge field is:
Z
Z
a )2
(F

1
4
a
a
abc b c 2
=
x
A
+

A
A
.
(1)
d
S = d 4x
4g 2
4g 2
Vector potential is denoted by Aa with = 1, . . . , 4 and a = 1, 2, 3 being the Lorentz and
group indices, respectively. Throughout the paper we shall use for gauge fields the matrix
notation. The covariant derivative in fundamental representation is:
i
(2)
D = i A = a Aa ,
2
where a are Pauli matrices and hats indicate matrices. Later we shall insert into the
covariant derivative the Levi-Civita and spin connections that are absent in Cartesian
coordinates. The commutator of covariant derivatives gives the field strength:

a
b = 1 a F
= i D , D .
(3)
F
2
The classical equations for Aa have instanton, or pseudoparticle, solutions. Instantons
are characterized by the topological properties: their field is selfdual, (a), and has the unit
topological charge, (b):
1
I
I
b
b
= F
,
F
2
Z
1
I bI
b
F = 1.
d 4 x tr F
32 2
(4a)
(4b)
A is
b
In addition to instantons there exist anti-instantons. The anti-instanton field F
antiselfdual and has the topological charge 1, i.e., the signs of the right-hand sides of
the equations (4) must be reversed. From here on we shall speak mostly about instantons
indicating generalization to anti-instantons if necessary.
Pseudoparticle field depends on gauge. The most popular for solitary instantons are
regular and singular gauges. The fields of instantons of radius centered at x 0 in these
gauges are (up to uniform gauge rotations):
+
+

1

ln reg x, x 0 =
ln
,
A +
reg =
0
2
2
(x x )2 + 2

2
0

ln
ln
1
+
.
x,
x
=
A +
sing
sing
2
2
(x x 0 )2
(5a)
(5b)
+ = a a and = a a are the matrix versions of the t Hoofts -symbols:

Here
a
,
for , = 1, 2, 3,
(6)
a =
()
a = ()
() a , for = 4.
592
These are antisymmetric in Lorentz indices and selfdual or antiselfdual, respectively:

1
1
+

=
,
=
.
(7)
2
2
Their properties can be found in [12] or obtained directly from the matrix representation
(9) below.
In practical calculations it is convenient to make use of the two hermitean conjugated
sets of 2 2 matrices (here the Latin indices stand for three spatial dimensions):
= ( a , i),
= ( a , i).
The following equations relate

+
,
= + i
(8)
to -symbols:
= + i
.
(9)
Separating the symmetric and antisymmetric parts of these expressions we obtain:
+
[
] = i
,
= {
} = ,
{ }
[ ]
= i
.
(10)
The singular gauge (5b) can be obtained from the regular one (5a) by means of the gauge
transformation:

b1 A I (x) N
b1 N
b+ + i N
b+
=N
(11)
A I (x)
sing
reg
b+1 = r /r,
b+ =
and N
with the matrices N
b+1 ).
b = N
be swapped for anti-instantons, N
r /r
where r = x x0 ( and must
Another way to derive (5b) is to carry out the inversion, r r 2 /r 2 . However

this changes the orientation of coordinate system and converts the instanton to an antiinstanton. This is corrected by the replacement.
The practical advantage of singular gauge is that the vector potential falls rapidly away
from the pseudoparticle. Hence a direct superposition of instantons and anti-instantons (the
so-called dilute instanton gas) is an approximate solution of the classical field equations.
The dilute gas approximation is an important component of instanton physics.
The solution (5) can be easily generalized to higher topological charges. One has simply
to include into (x) pieces describing additional pseudoparticles:
sing (x) Nt Hooft(x) = 1 +
N
X
i=1
i2
,
(x xi )2
(12)
i and xi being their radii and positions. This solution of duality equations bears the name
of the t Hoofts ansatz [7]. It has the topological charge N , depends on 5N parameters
{xi , i } and does not allow for independent gauge rotations of instantons.
The further generalization belongs to Jackiw, Rebbi and Nohl who have proposed the
following form of the -function
N
(x) =
JNR
N+1
X
i=1
i2
.
(x xi )2
(13)
Their solution depends on 5N + 4 parameters (the uniform scaling of s does not matter)
and is the most general multi-instanton configuration known explicitly. (The solution [6]
which takes into account all degrees of freedom proved to be too complicated.)
593
Throughout the paper we shall discuss topological solutions of the type (5) with different
functions .
2. Curvilinear coordinates
In order to distinguish curvilinear coordinates from Cartesian ones we shall denote them
by q , q etc. The metric tensor is now g (q) and
ds 2 = dx2 = g (q) dq dq .
(14)
The metric g (q) and its inverse g (q) = [g (q)]1 are used for raising and lowering
indices: A = g A ; A = g A .
Let us start from fields that are singlets with respect to the gauge group. Covariant
:
derivatives of vectors are taken with the help of the Levi-Civita connection
D A = A + A
and D A = A
A .
(15)
The latter is unambiguously defined by the condition that the metric tensor is covariantly
constant, D g = 0:

g
g
g
1
= g
+
.
(16)
2
q
q
q
The metric g may be decomposed into vierbeins ea (from here on we reserve for those
the Latin indices). Let ab be the flat Euclidean metric, ab = ab = diag(1, 1, 1, 1). Then
g (q) = ab ea (q) eb (q),
while ab = g (q) ea (q) eb (q).
(17)
One may convert spatial indices into vierbein ones, Aa = ea A . Raising and lowering of
the latter is performed by means of the tensors : Aa = ab Ab and Aa = ab Ab . The
inverse of the vierbein is ea :
ea eb = ba
and ea ea = .
(18)
Covariant derivatives of quantities with vierbein indices are defined in terms of the spin
connection Ra b
D Aa = Aa + Ra b Ab
and D Aa = Aa Ab Rb a .
(19)
Vierbeins are covariantly constant and this fixes the spin connection. Solving the equation
D ea = ea + ea eb Rb a = 0 we obtain:

eb = ea D e b .
(20)
Ra b = ea eb + ea
The matrices R are antisymmetric with respect to the exchange a b. (This follows from
ea eb = ba = const.) Sometimes it is convenient to expand Ra b in terms of generators of

the O(4) group, (Lmn )ab :
i
a
= Bmn (Lmn )ab ,
Rb
2

a
where (Lmn )ab = i m
bn na bm .
(21)
594
The coefficients Bmn are antisymmetric, Bmn = Bnm ,

i
(22)
Bmn = (R )ab (Lmn )ba = Rmn .
2
Decomposition of metric into vierbeins is not unique and is defined up to orthogonal
rotations with respect either to Greek or to Latin indices. The transformations ea
oba (q)eb and ea O (q)ea with matrices oba (q) and O (q), that satisfy the conditions
g = g O O
and ab = cd oac obd ,
(23)
leave relations (17) invariant. Therefore they define new vierbeins that are equivalent to
the initial one. Once the spin connection is defined in terms of vierbein it suffers the same
ambiguity. In flat space there always exists the vierbein (ea = (x /q )ea ) associated
with the original Cartesian frame such that the spin connection R is identically zero.
It is convenient to use the freedom in order to make some components of vierbein zero.
In general ea is a (4 4)-matrix. However if the coordinate frame is orthogonal 2 and the
metric tensor may be diagonalized,
g (q) = G(q) ,
then one can use the diagonal vierbein
p
ea = G(q) a .
(24)
(25)
We shall call it the natural vierbein. Only four of its components are not zero.
The last thing is to extend the covariant derivative to spin- 21 fields. We would like
matrices a with Latin indices to be covariantly constant. Let us define -matrices so that
{a , b } = 2ab .
(26)
Then their commutators

i
1
ab = [a , b ],
(27)
2
4
are the rotation generators for spin- 21 fields. The covariant derivative acting on bispinors is
i ab
B ab .
4
It is easy to check directly that the covariant derivatives of -matrices are zero:

i
D a = a + Bmn b (Lmn )ba 12 [mn , a ] = 0.
2
D =
(28)
(29)
3. The compensating gauge connection

3.1. Definition and properties
Now let us turn to gauge fields in curvilinear coordinates. One may define a and a
matrices with vierbein indices by analogy with (8). We shall show that it is possible to
2 This class incorporates many coordinate systems including spherical, cylindric, parabolic, elliptical and
others.
595
introduce compensating gauge fields A

such that either a b or a b become covariantly
constant. The compensating field is a pure gauge provided that the space is flat.
It is convenient to turn back to the spin connection Bmn and use the following
representation of -matrices:
!
!

0 a
ab
0
0
a b
, a b =
, ab =
,
(30)
a =
a 0
0 +
0
a b
ab
where
+
= i [a
b]
ab
and ab
= i [a b]
(31)
+ and . One may define -symbols with

are the vierbein analogs of the Cartesian
coordinate indices as follows:

+
+
= ea eb ab
and
= ea eb ab
.
(32)
In order to escape confusion we shall use only ab that are just constant numerical matrices.
From the relation (29) applied to the block-diagonal matrix a b it is easy to deduce for
the separate blocks that:

d
c d
+ Ra
b i A +
(33a)
D a b = a b c d ac Rb
, a b = 0,

d
c d
+ Ra
b i A
(33b)
D a b = a b c d ac Rb
, a b = 0.
The vector-potentials A +
and A are:
1 mn +
i ab
(34a)
A +
= B mn = a R b ,
4
4
1 mn
i
ab
(34b)
A
= B mn = a R b .
4
4
Thus the gauge connections A +

and A are projections of the selfdual and antiselfdual
ab
parts of the spin connection R onto the SU(2) gauge group. In general only one of
the bilinears a b , a b can be made covariantly constant by the appropriate choice of
compensating field. The equations (34) are the most straightforward way to calculate A .
the corresponding
(32) becomes covariantly constant too.)
(Note that along with ab
In flat Euclidean space either of the potentials A +

and A is a pure gauge, i.e., the field
b
strengths F = A A i[A , A ] are zero. In order to show this let us again
resort to the help of the spin connection. Note that matrices = ea a are covariantly
constant, that is

i
(35)
D = Bmn mn , = 0,
4
where D is the isotopic singlet part of the covariant derivative. It follows from (35) that
b , ] R = 0.
[D , D ] = i[G
(36)
b is the commutator of spinor

Here R is the Riemann curvature tensor and G
components of covariant derivatives:
596

i mn
i kl
b
G = i B mn , B kl =
4
4
b
F
0
b+
F
!
.
(37)
The last term in (36) can be written as R = 4i R [ , ].

b and one obtains the relation
Separating the diagonal blocks of the matrices G
between the compensating field strength and the Riemann curvature of the space:
b+ = 1 R + ,
(38a)
F
4
b = 1 R .
(38b)
F
4
b = 0 provided that R = 0. Inasmuch as simple changes of variables
b+ = F
Thus F
x q can not produce curvature both of the compensating fields must be pure gauges
in flat space.
3.2. Relation between ab and symbols
The pure gauge vector-potentials A +

and A may be represented as
1
1
A +
(q) = i+ (q) + (q) or A (q) = i (q) (q).
(39)
This defines the (2 2)-matrices + and up to the left multiplication by a constant

unitary matrix U : U . However the freedom may be eliminated by requiring that
gauge rotated into .
Let us rewrite the condition of ab being covariantly constant (33) in Cartesian
x-coordinates where the spin connection R is absent. It will read (presently we may drop
the -superscripts):

q q a b
e e ab = ea eb ab + ea eb 1 , ab = 0, (40)
x x
where ea = (q /x )ea is our vierbein expressed in the x-frame. This is equivalent to

1 ea eb ab 1 = 0 or ea eb ab 1 = const.
(41)
D (x) = D
1
are constant tensors that project selfdual and antiselfdual
We conclude that ea eb ab
antisymmetric tensors onto the SU(2) group. Hence it must be equal (up to a gauge
depends on orientation of the

rotation) to the corresponding . (Duality of ea eb ab
a
vierbein e , see appendix.) Uniform gauge rotations do not affect the compensating fields
(39) and we may fix up to a phase factor ei by demanding that:
ea eb ab = 1
or ab = ea eb .
(42)
The question is whether there are solutions to these equations. One may figure out the
two conditions. First, the both sides of (42) must be normalized in the same way. In order
to prove this we take the square of the first equation:
ea eb ab ec ed cd = 1 .
(43)
This results into the identity ab ab = = 12. Thus the first condition is fulfilled.
597
However this does not guarantee existence of solutions. It is necessary that both sides of
the equations were of same duality. This is the case if the two sets a and ea differ by a
rotation and/or an even permutation. 3 (Another way is to say that the change of variables
must respect parity.) Gauge transformations do not interfere with duality and it suffice to
check the equivalence of (42) to their duals,
1 a b

e e ab = 1
or ab
= abcd ec ed
.
2
(44)
A well-known example of parity violating procedure is inversion q = x /x 2 that may

be implemented to derive the singular gauge, (5b) from the regular one, (5a). In order to
restore duality altered by the inversion the replacement is necessary.
4. Instanton with the compensating field

A straightforward way to transform the instanton to curvilinear coordinates would be
simply to change variables: x q and A A = A (x /q ). However if for
some reasons the use of a particular vierbein is preferable then it is convenient to match the
selfdual field with the vierbein. This may be done by means of one of the gauge rotations
discussed in the preceeding section. Obviously if the q-system is orthogonal it makes sense
to work with the natural vierbein and the corresponding -symbols. In most of the cases
one can recognize in the new configuration the original solution. Still in the next section
we shall meet the example of different starting gauges leading to the same final result.
.
Let us begin with the singular gauge (5b) that contains the antiselfdual symbol
This dictates the choice of the matrix for the transformation. The combined change of
variables and gauge transform give:

x I
(q)
1
1

(q) (q) + i
(q)
.
(45)
A
A
sing (q) = (q)
q sing
q
With the help of the relations (39) and (42) this may be immediately reduced to

1 a b

(46)
A
sing (q) = e ab e ln sing (q) + A .
2
+ and the matrix
On the other hand the instanton field in regular gauge (5b) depends on
+ must be employed:

1 a + b

+
(47)
A
reg (q) = e ab e ln reg (q) + A .
2
An interesting feature of the expressions (46), (47) is that they do not contain
explicitly. One needs only A
which may be found directly from (34). That is much easier
than solving (42) for . From here on we shall omit the superscript in A
(q). In fact
if the sign of the topological charge is not crucial one may apply the formulae (46), (47)
and (34) without checking relative duality of ab and ea eb .

3 Although gauge transformation rotate the traceless Pauli -matrices into each other, they do not affect the
a
4 -matrix. Hence permutations can not be reduced to gauge transforms.
598
The duality equation (4a) in the non-Cartesian frame takes the form
g
bab = 1 abcd F
b or F
bcd ,
b =
F
F
2
2
where g = det kg k. The topological charge is (compare to (4b)),
Z
Z
1
1
b 4
b
bab F
bcd g d 4 q.
tr
F
d
q
=
F

abcd tr F
q=

32 2
32 2
(48)
(49)
We shall calculate the field strength in two ways. First we shall derive a general
expression that works in the multi-instanton case as well. Then we shall demonstrate that
for one pseudoparticle the formulae do simplify both in regular and singular gauges.
Let us denote for brevity the first addends in the right-hand sides of (46), (47) by A I .
Then the full covariant derivative D is:

b
b
b = X
b i A I + A
(50)
D X
, X X ,
and the field strength equals

I

I
b (q) = D A I + A D A I + A
F
+ i A + A , A + A .
(51)
Note that the Levi-Civita connection has dropped out of the result.
b . Thus its only role is
The pure gauge compensating field A does not contribute to F
b
e is covariantly constant. Hence the covariant derivatives
to ensure that the factor ea ab
act only onto the logarithms:

1
1
b
b
I
b
= ea ab
e D ln + ea ab
e D ln i A I , A I ,
F
2
2
and do not contain gauge terms:
ln (q).
D ln (q) = ln (q)
(52)
(53)
The commutator can be calculated with the help of the identity

[ab , cd ] = 2i(bc ad + ad bc ac bd bd ac ),
(54)
that follows from the properties of SU(2) generators but may be proven directly. Denoting
for brevity ()2 = g we obtain:

2 a b
2
()2
I
b
g
e ab e g
F sing =
2 2
()2
()2
D a b
b D
e ab e + ea ab
.
(55)
e
+
2
2
Mind that so long we referred neither to the specific form of (q) nor to the instanton
number. Thus the formula works for all the solutions listed in Section 1. However
expression may be significantly simplified for solitary instantons in regular or singular
gauges. Let us start from the regular gauge in Cartesian coordinates,

I
b
=
F
reg
+
2
(r 2 + 2 )2
(56)
599
with r 2 = (x x0 )2 and transform it to q-coordinates. Obviously the transformation

+ . According to the version of (42) the result is:
affects only
+
+ b
2ea ab
e

I
b
F reg = 2
(r + 2 )2

I
bab
or F
=
reg
+
2ab
.
(r 2 + 2 )2
(57)
The last version is explicitly selfdual.

The expression (55) in the one instanton case can be simplified too. Once more we
shall start from (56). Remember that the Cartesian pseudoparticle in singular gauge can be
obtained from that in regular gauge by means of the transformation (11). This results into
I |
I | N
b
b+1 F
b
F
sing = N
reg b+ . Now we should carry out the simultaneous change of variables
and gauge transform:
+

x x 1 b1 2
I
b+ .
b
N
N
=
F
sing
q q + (r 2 + 2 )2
(58)
b+1 = r (note that the

b-matrices as: N
b+ = r and N
It is convenient to rewrite N
Cartesian g = ). After a little algebra with the use of (42) one obtains:

I
b
= N 1
F
+
sing
+
2
(r 2 + 2 )2
N+ ,
N+ = a ea r,
N+1 = a ea r.
(59)
We note that the Levi-Civita and gauge connections have dropped out giving the compact
results (57), (59). Unfortunately for multipseudoparticle solutions one still has to apply the
clumsy formula (55). Later in Section 5.2 we shall present the case when matrices N+
become degenerate (i.e., N+ = i) so that (57) and (59) coincide.
It is obvious that all the reasoning may be literally repeated for anti-instantons. The only
+
ab
and a a . The matrices N+ , N+1
difference is that one has to interchange ab
1
1
must be substituted by N = N+ and N = N+ . Certainly the signs of the right-hand
sides of the equations (48) must be reversed as well.
5. Instanton in O(4)-spherical coordinates

In order to show how the general theory works we shall apply it to the familiar case. Let
us derive explicit formulae for one instanton placed at the origin of 4-dimensional spherical
coordinates. The example happens to be instructive and reveals two unexpected features.
First, it turns out that (in this particular setting) our prescription converts pseudoparticles
both in singular and regular gauges to the same form. In a sense this is an evidence in
favor of the approach. Second, the compensating vector potential exhibits a singularity.
At first sight the singularity of the field that is a pure gauge must be of no importance.
However in presence of a pseudoparticle the singularity comes to life and contributes to
the ChernSimons number and other gauge dependent quantities.
We begin from general description of 4-dimensional spherical coordinates and then
focus on the pseudoparticles. Finally we present the calculation of the ChernSimons
number.
600
5.1. 4-dimensional spherical coordinates

Spherical coordinates make a natural choice for problems involving single Euclidean
pseudoparticle. The set of spherical coordinates includes radius and three angles: q =
(r, , , ). The polar axis is aligned with x 1 and
x 1 = r cos ,
(60a)
x = r sin sin cos ,
(60b)
x = r sin sin sin ,
(60c)
x = r sin cos .
(60d)
2
3
4
The metric tensor and the natural vierbein are:

g = diag 1, r 2 sin2 , r 2 sin2 sin2 , r 2 ,
ea
= diag(1, r sin , r sin sin , r).
(61a)
(61b)
It is convenient to introduce the matrix notation for the Levi-Civita symbols, b = k k.

The standard calculation gives:
0 0 0 0
1
0
0 0
(62a)
br =
0 0 1 0 ,
1
0 0 0
r
0
0
0 r sin2
0
0
cot
b = r
(62b)
,
0
0
cot
0
1
0
0
0 sin 2
2
0
0
r sin2 sin2
0
1
0
0
0
sin 2
2
,
(62c)
b =
1 cot
0
cot
r
1
0
0
0
sin 2 sin2
2
0
0
0
r
0 cot
0
0
b
(62d)
= 0
0
cot 0 .
1
0
0
0
r
By means of the relation (20) we may find the spin connection. Sticking once again to
b = kR a k, one obtains:
the matrix notation, R
b
br = 0,
R
0
sin
b =
R
0
0
0
b =
R
0
1
0
0
0
0
0
sin
0
0
cos
0
0
0
0
0 1
0 0
,
0 0
0 0
0
b =
R
sin sin
0
0
0
cos
0
601
(63a)
0
cos
,
0
0
(63b)
(63c)
sin sin
cos
0
cos sin
0
.
cos sin
0
(63d)
The compensating gauge potentials A + and A depend on the choice of -matrices. Let
+
and ab
must be of correct duality in
us remind that their antisymmetrized products ab
order to make the equations (42) and (44) equivalent. For example, if x , y , z stand for
the standard Pauli matrices we can take:
a = (i, z , y , x ),
a = (i, z , y , x ).
(64)
Convolutions of the -matrices (64) with the spin connection, (34), result into the
compensating gauge fields:
y
z
x
cos sin ,
A
,
=
2
2
2
y
z
x
cos sin sin + cos sin ,
A
=
2
2
2
A
=
A
r = 0.
(65)
1
=
)
Either of these fields is a pure gauge generated by the corresponding unitary (
matrix:
+ i cos cos
sin sin
2
2
2
2
+ i sin cos
cos sin
2
2
2
2
sin
+ i sin cos
2
2
2
2 .

sin sin
i cos cos
2
2
2
2
(66)
cos
The matrices + and for selfdual and antiselfdual cases differ by the sign of the polar
angle .
The compensating connection is singular since neither A

nor A go to zero at =
0, and = 0, . As long as the field is a pure gauge this singularity is not observable.
Nevertheless in presence of physical fields it may tell on gauge variant quantities.
602
5.2. Instantons in 4-spherical coordinates

We shall consider instantons in singular and regular gauges with their centers at
the origin. The analysis reveals an amusing coincidence. Remember that the instanton
+
and ab
.
gauge potentials in these two cases involve different t Hooft symbols ab
+
Hence according to our prescription the compensating vector potentials are A and A ,
respectively. It turns out that if the -matrices are taken in the form (64) then the equation
(46) for the singular gauge and that (47) for the regular one give the same result. Let us
take reg and sing (5) in the form:
reg (r) = (r 2 + 2 )1
and sing(r) = 1 + 2 /r 2 .
(67)
Substituting those into the equations (46), (47) with the compensating potentials given by
(65) we obtain for instantons (A+ ) and anti-instantons (A ):
A
r = 0,
2 2

,
2
y
z
cos sin 1 2
2
2
r +

y
2 2
x
z
+ cos sin ,
A = cos sin sin 1 2
2
2
2
r +
2

2
2
x
A
1 2
.
=
2
r + 2
A
=
(68a)
(68b)
(68c)
(68d)
Note that the potentials A interpolate between A at the origin and A at infinity:
lim A
(q) = A (q)
r0
and
lim A
(q) = A (q).
(69)
The field strength is given by the expression (57). Comparison of the latter with (59)
proves that as a rule regular and singular gauges are different. However in the present case
the matrices N = i and the transformation converting the gauges into each other is
trivial. The degeneracy is specific to our choice of coordinates. Note that we had taken
q 1 = r and associated with it 1 = i. According to (59) these are the necessary and
sufficient conditions of the coincidence of the two gauges in curvilinear coordinates. Hence
this coincidence is stable with respect to reparametrizations of the angles , , and
-matrices as long as r and 1 stay intact.
5.3. The singularity and the ChernSimons number
It is well known that the topological charge density (49) may be represented as a total
divergence:

b F
b
tr F
2i

b
= K =
tr A F + A A A .
(70)
32 2
16 2
3
According to the Gauss theorem the space integral of the lhs may be reduced to the surface
integral of K over the boundary. However the current K is not gauge invariant. Thus the
distribution of K over the boundary depends on gauge even though the topological charge
603
H
q = K dS does not. A text book example is the instanton in the A 4 = 0 gauge. Here
the two nonzero contributions to q come from the hyperplanes x4 = (i, j, k = 1, 2, 3):

Z
d 3 xE ij k
2i x4 =+
b
tr Ai Fj k + Ai Aj Ak
.
(71)
q(A4 = 0) =
3
16 2
x4 =
R
The integral Q0 (x4 ) = d 3 xE K 4 (x4 ) over a 3-dimensional manifold is called the Chern
Simons number. In the example above Q0 () = 12 leading to q = 1. Thus the
topological charge may be interpreted as the increment of the ChernSimons number.
Let us try to carry out the same procedure for the gauge field (68) in spherical
coordinates. At first sight there is a striking similarity with the A 4 = 0 case. According
to (69) now Ar = 0 and the instanton interpolates between A + at r = 0 and A at r = .
It would be nice to take the radius (or
H rather ln r/) for the 4th coordinate and associate
the ChernSimons number Q(r) = S 3 K r (r) d d d with a sphere Sr3 of radius r. The
r
analogy would be complete provided that one had the contributions Q() = 1/2 from the
infinite sphere and Q(0) = 1/2 from the infinitesimal sphere at the origin. 4 However
the explicit calculation proves that this is not the case because the singular lines = 0,
carry the ChernSimons number as well.
The topological charge in 4-spherical coordinates is given by the integral over a half
infinite parallelepiped {0 6 r 6 , 0 6 6 , 0 6 6 , 0 6 6 2}:
1
q=
32 2
Z
dr
Z
d
Z2
d
b F
b .
d tr F
(72)
When applying the Gauss theorem one must take into account the entire boundary
including the lateral faces. The components of the ChernSimons current are:

sin
2 2
1 2 r 2 sin2
+
1
,
(73a)
Kr =
2 2
r 2 + 2
8
(r 2 + 2 )2
2 r sin 2 sin
,
4 2 (r 2 + 2 )2
2 r cos
,
K = 2 2
4 (r + 2 )2
K =
(73b)
K = 0.
(73c)
Converting (72) to the surface integral and keeping only nonzero pieces we write:
Z
q=
Z
d
Z2
d
r= Z Z
=
Z2

dK
+ dr d dK
.
r
r=0
(74)
=0
The first addend in the rhs is Q() Q(0) and the second comes from the integration
along the hyperstrips = 0, which correspond to the Cartesian x 1 x 4 -plane. Each of
the four addends contributes 1/4 to the final value q = 1. We see that the singularity of
4 In regular gauge, (5a), the ChernSimons number is concentrated at infinity q = Q () = 1, in contrast to
r
the singular one, (5b), where q = Qs (0) = 1.
604
compensating gauge field is not absolutely harmless. The reason for its coming to being
was that K is not a gauge invariant.
br /8 2 . Multiplication of the singular
It is easy to see that K = r tr A F
br results into the singularity of K .
gauge potential A by the instanton field F
Thus the singularity of the seemingly unobservable compensating field A leads to
nonzero consequences in presence of the physical field A I . In general singular gauge
transformations may generate singularities of gauge variant quantities. Obviously this is
specific neither to the instanton nor to the current choice of the O(4)-spherical coordinates.
6. Cylindrical coordinates
Let us describe two other coordinate systems that are relevant to physics. We shall start
from (2 + 2)-cylindrical coordinates, i.e., the geometry of thick euclidean QCD-string or
vortex. Then we shall consider (3 + 1)-cylindrical coordinates which are common to bags
or glueballs.
6.1. (2 + 2)-cylindrical coordinates
These coordinates may be appropriate for objects with axial symmetry, such as strings,
vortices or quarkantiquark pairs. We parametrize the x 1 x 2 -plane by polar coordinates
q 1 = r and q 2 = and leave q 3,4 = x 3,4 :
x 1 = r cos ,
x 2 = r sin ,
x 3 = z,
x 4 = t.
(75)
The metric tensor and the natural vierbein are, respectively:

g = diag(1, r 2 , 1, 1) and ea = diag(1, r, 1, 1).
(76)
Only three of the Levi-Civita symbols (16) are not zero:

r
= r
and r = r = r 1 .
Direct calculation (see

component. Namely:
0 1
1 0
b =
R
0 0
0 0
(77)
Eq. (20)) proves that the spin connection has only one nonzero
0
0
0
0
0
0
,
0
bz = R
bt = 0.
br = R
R
(78)
Convolutions of the -matrices (64) with the spin connection R give the compensating
fields that are singular at r = 0:
z
A
(79)
A
r = Az = At = 0.
= ,
2
Finally, for completeness we present explicitly the unitary gauge matrices + and .
i
i
exp
exp
1
2
2 , 1 = .
(80)
=
i
2 exp i
exp
2
2
605
The difference between the matrices + and is in the sign of the polar angle
(corresponding to left and right rotations).
6.2. (3 + 1)-cylindrical coordinates
This geometry is typical for objects that are spherically symmetric in 3 dimensions. In
addition to the famous MIT-bag and glueballs one may list the monopole and the caloron
that is a periodic instanton chain along the 4th axis. It takes the place of instanton at nonzero
temperature [9]. Now we parametrize the spatial sector, i.e., x 1 , x 2 , x 3 by spherical
coordinates q 1 = r, q 2 = and q 3 = leaving q 4 = x 4 .
x 1 = r sin cos ,
x 2 = r sin sin ,
x 3 = r cos ,
x 4 = t.
(81)
The metric tensor and the natural vierbein are well known:
g = diag(1, r 2 , r 2 sin2 , 1),
ea = diag(1, r, r sin , 1).
(82)
Since the transformation (81) affects only the spatial sector bt = 0 and s are
essentially 3-dimensional:
0 0 0
(83a)
br = 0 r 0 ,
1
0 0
r
0 r
0
1
b =
(83b)
0
0
,
r
0 0 cot
0
0
r sin2
0
sin 2
(83c)
b = 0
.
2
1
cot
0
r
The nonzero components of the spin connection are:
0 1 0 0
0
0
sin 0
0
cos 0
,
b = 0
b = 1 0 0 0 ,
R
R
0 0 0 0
sin cos
0
0
0 0 0 0
0
0
0
0
b
b
(84)
Rr = Rt = 0.
We shall use the same set of -matrices (64) as before. This leads to the following
compensating connections:
y
z
x
cos sin , A
(85)
A
r = At = 0.
= , A =
2
2
2
This vector potential is singular at = 0, . The gauge matrices generating the above
connections are:
606

i
i
exp
2
2
1
1
2
2
=
, = . (86)

i
i
2 exp + +
exp + + 2
2
2
2
We note again that + and differ by the sign of the polar angle .
exp
7. Conclusion
The purpose of the paper was to analyze selfdual classical fields in non-Cartesian
coordinates. We showed that the explicit multi-instanton solutions by t Hoofts and by
Jackiw, Nohl and Rebbi as well as instantons in regular and singular gauges can be
economically generalized to curvilinear coordinates. The gauge orientation of the field
can be matched with the vierbein by means of a gauge transformation. As a result the
varying -symbols with coordinate indices are replaced by the constant ab -symbols
with vierbein ones. The price for ab being constant is the appearance of the compensating
field. The origin of the latter is the coordinate-dependent gauge transformation associated
with the vierbein. Thus in most of the cases the proposed solution is a gauge rotated version
of the original one.
The compensating gauge potential may be obtained in a straightforward manner. The
calculation proceeds in three steps.
(16).
(1) Starting from the metrics g one can find the Levi-Civita connections
a
(2) Covariant differentiation of the vierbein ea leads to the spin connection R (20).
(3) Convolution of the spin connection with -matrices gives the compensating gauge
potential (34).
The first two points are the standard calculation of spin connection. The last one is
nothing but projecting the selfdual (or antiselfdual) component of the antisymmetric tensor
Rab onto the SU(2) gauge group.
The gauge potential of pseudoparticle is the sum of the instanton part that is similar
to the Cartesian formula and the compensating field (46), (47). Having constant ab (31)
and covariantly constant -symbols (32) notably simplifies calculations and is worth
the appearance of additive compensating background. Singularities of the compensating
connection do not spoil physical quantities.
I would like to thank A.A. Rosly and E. Gozzi for encouraging discussions. It is a
pleasure to acknowledge the hospitality of the Department of Theoretical Physics of Trieste
University that I enjoyed several times and where this research was completed as well as
the financial support of INFN and COFIN-97-MURST of Italy. The work was done with
partial support of the RFBR grants 97-02-16131 and 00-02-17808.
Appendix A. Duality of -symbols
+ 1
= (q /x )(q /x ) ea eb
It was mentioned in Section 3.2 that ea eb ab
+ 1 is a constant selfdual or antiselfdual tensor, see (41). Now we shall prove this
ab
607
statement. First let us remind how coordinate changes x q affect the Levi-Civita
antisymmetric pseudotensors. The change of variables leads to the substitution:

q
q q q q

E
=

=
det

x
x x x x

= () ,
g
(A.1)
where g = | det kg k| and is the ordinary -symbol ( 1234 = 1). The sign of the last
expression depends on the relative orientation of the q and x coordinate systems. To be
definite we shall assume that the sign is plus.
The second formula relates -symbol with coordinate (greek) and vierbein (latin)
indices:

(A.2a)
ea eb ec ed = det ea abcd = () g abcd ,
1

abcd ea eb ec ed = () .
g
(A.2b)
Now the ()-option is associated with the sign of det kea k defined by the orientation of
the vierbein. For time being we shall stick to the plus sign again.
+ 1
. The gauge rotation respects duality and we
Let us turn to properties of ea eb ab
+
1
:
may omit the , matrices. Let us calculate the dual of ea eb ab
1 a b + 1 q c q d +

e e ab =
e
e .
2
2
x x cd
With the help of the equation (A.1) one can show that:
1 a b + 1 x x c d +

e e ab =
e e ,
2
2
g q q cd
(A.3)
(A.4)
and then using (A.2b) and the identity ea ea = one arrives at:
1 x x c d + 1 x x abcd +
+
cd = ea eb ab
e e =
e
e
.
2
g q q cd 2 q a q b
(A.5)
is selfdual (and ea eb ab
is antiselfdual) provided that the vierbein
This proves that ea eb ab
and the Cartesian coordinate frame are oriented in the same way.
+
depends on the sign of
Generally speaking we may say that the duality of ea eb ab
+
a
a
itself the
det kq /x k det ke k = det ke k. In order that it had the same duality as ab
a
a
Cartesian expression e of the vierbein e must be oriented in the right way.
References
[1]
[2]
[3]
[4]
A.A. Belavin, A.M. Polyakov, A.S. Schwartz, Yu.S. Tyupkin, Phys. Lett. B 59 (1975) 85.
M.A. Shifman (Ed.), Instantons in Gauge Theories, World Scientific, Singapore, 1994.
T. Schfer, E.V. Shuryak, Rev. Mod. Phys. 70 (1998) 323.
P. Forgacs, L. ORaifeartaigh, A. Wipf, Nucl. Phys. B 293 (1987) 559.
608
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
A.J. Niemi, G.W. Semenoff, Nucl. Phys. B 269 (1986) 131.

M.F. Atiah, N.J. Hitchin, V.G. Drinfeld, Yu.I. Manin, Phys. Lett. A 65 (1977) 185.
R. Jackiw, C. Nohl, C. Rebbi, Phys. Rev. D 15 (1977) 1642.
L.S. Brown, R.D. Carlitz, D.B. Creamer, Ch. Lee, Phys. Rev. D 17 (1978) 1583.
B.J. Harrington, H.K. Sheppard, Phys. Rev. D 17 (1978) 1583.
D.J. Gross, R.D. Pisarski, L.G. Yaffe, Rev. Mod. Phys. 53 (1981) 43.
A.A. Abrikosov Jr., Nucl. Phys. B 218 (1983) 459.
G. t Hooft, Phys. Rev. D 14 (1976) 3432.
Nuclear Physics B 586 [FS] (2000) 611640

Exact solution of the supersymmetric sinh-Gordon

model with boundary
Changrim Ahn a , Rafael I. Nepomechie b,
a Department of Physics, Ewha Womans University, Seoul 120-750, South Korea
b Physics Department, P.O. Box 248046, University of Miami, Coral Gables, FL 33124 USA
Received 19 May 2000; accepted 6 July 2000
Abstract
The boundary supersymmetric sinh-Gordon model is an integrable quantum field theory in 1 + 1
dimensions with bulk N = 1 supersymmetry, whose bulk and boundary S matrices are not diagonal.
We present an exact solution of this model. In particular, we derive an exact inversion identity and
the corresponding thermodynamic Bethe ansatz equations. We also compute the boundary entropy,
and find a rich pattern of boundary roaming trajectories corresponding to c < 3/2 superconformal
models. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.55.Ds; 11.30.Pb; 11.10.Kk; 11.25.Hf; 05.50.+q
Keywords: Sinh-Gordon model; Supersymmetry; Boundary S-matrix; Bethe ansatz; Duality; Integrable spin
chain
1. Introduction
The supersymmetric sinh-Gordon (SShG) model [16] is one of the simplest examples
of a (1 + 1)-dimensional integrable quantum field theory with N = 1 supersymmetry.
Indeed, the particle spectrum consists of one boson and one fermion which have equal
mass and which enjoy factorized scattering [7,8]. As such, SShG is a valuable toy model.
In this article, we consider the boundary SShG model, with boundary conditions that
preserve the bulk integrability, but not necessarily the bulk supersymmetry [913]. In
addition to its usefulness as a simple prototype, we expect that this model may also have
applications to quantum impurity problems [14].
An interesting feature of the boundary SShG model is that the S matrices which
have been conjectured for both bulk and boundary scattering are not diagonal. Our main
objective is to perform a thermodynamic Bethe ansatz (TBA) analysis [6,1519] for this
E-mail address: nepomechie@physics.miami.edu (R.I. Nepomechie).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 4 0 - 5
612
C. Ahn, R.I. Nepomechie / Nuclear Physics B 586 [FS] (2000) 611640
model, using these conjectured S matrices as inputs. Such analysis can provide checks on
the input S matrix data, as well as information about the underlying boundary conformal
field theory [20,21].
Conventional wisdom suggests that the problem of determining the necessary Bethe
ansatz equations is intractable, due to the fact that both the bulk and boundary S matrices
are not diagonal. Nevertheless, we succeed to determine the Bethe ansatz equations and
carry out the TBA analysis for the boundary SShG model. This is the first example of
a model defined on an open interval whose both bulk and boundary S matrices are nondiagonal for which Bethe ansatz equations are obtained. 1
We also obtain an expression for the boundary entropy [19,26] for the boundary SShG
model. Moreover, we find a rich pattern of boundary roaming trajectories corresponding
to c < 3/2 superconformal models [27,28], thereby generalizing previous work on bulk
[2931] and boundary [32,33] roaming.
The outline of this article is as follows. In Section 2 we review the scattering theory
of the boundary SShG model, which serves as our input. Here we also show that the
strongweak duality symmetry of the bulk model (see, e.g., [34]) extends also to the model
with boundary. Moreover, we introduce the notations and conventions which are used
throughout the paper. In Section 3 we formulate the so-called Yang matrix [35] and relate it
to a commuting transfer matrix, which is the true starting point of any TBA analysis. For the
problem at hand, we require a boundary version of the Yang matrix [36,37], which presents
an interesting complication with respect to the more familiar case of periodic boundary
conditions. In Section 4 we use the open-chain fusion formula [38] to derive an exact
inversion identity, using which we obtain the eigenvalues of the transfer matrix in terms of
roots of certain Bethe ansatz equations. That such an inversion identity exists is presumably
due to the fact that the bulk S matrix satisfies the so-called free fermion condition [3941].
In Section 5 we use these results to derive the TBA equations. Certain remarkable identities
lead to very simple formulas, in particular for the boundary entropy. In Section 6 we use
our result for the boundary entropy to obtain boundary roaming trajectories. Finally, in
Section 7 we discuss our results and describe some possible generalizations.
2. Review of boundary SShG scattering theory

In this section, we review the bulk and boundary S matrices which have been proposed
for the boundary supersymmetric sinh-Gordon model. As mentioned in the introduction,
these S matrices will be used as inputs in the calculations that follow. We also show that the
strongweak duality symmetry of the bulk model extends also to the model with boundary.
1 For the multichannel Kondo model [2224], the corresponding transfer matrix is that of a closed spin chain
with an impurity, rather than an open spin chain with boundaries. For the boundary sine-Gordon model with
non-diagonal boundary S matrix [9], certain exact results have been obtained [25] by analytic continuation from
a regime with diagonal bulk scattering; however, the corresponding Bethe ansatz equations have not yet been
determined.
613
2.1. Bulk
In order to understand the SShG scattering theory, it is essential to first consider a related
model, namely, the supersymmetric sine-Gordon (SSG) model, whose Euclidean-space
Lagrangian density is given by
LSSG
bulk =

1
M2
z + z + iM
cos +
z z +
sin2 ,
2
2 2
(2.1)
z ) are the components of a Majorana

where (z, z ) is a real scalar field, (z, z ) and (z,
spinor field, and is the dimensionless coupling constant. The Lagrangian density for
the supersymmetric sinh-Gordon (SShG) model is obtained by analytic continuation to
imaginary coupling, i.e., setting = i with real:
LSShG
bulk =

1
M2
z + z + iM
cosh
+
sinh2 .
z z +
2
2
2
(2.2)
Both of these models have N = 1 supersymmetry (without topological charge) [1,2] and
are integrable [3,4]. 2
Observe that SSG has a periodic potential, which admits classical soliton solutions that
interpolate between neighboring minima. Correspondingly, it has been proposed [4244]
that the SSG quantum spectrum consists of supersymmetric multiplets of kinks of mass
m and breathers (bound states of kinks) of mass mn = 2m sin(n), n = 1, 2, . . . , [1/2],
where
=
2 /4
,
1 ( 2 /4)
(2.3)
and [x] denotes integer part of x. Hence, breathers can be present only if 0 < < 1/2.
The lightest (n = 1) breathers are identified as the elementary particles (boson, fermion)
corresponding to the fields in the Lagrangian density (2.1).
Upon making the analytic continuation to SShG (which is the model of primary interest),
we see that the potential is no longer periodic, and hence, there are no longer any classical
soliton solutions. Thus, the SShG quantum spectrum does not contain kinks; it consists
only of the elementary particles of some mass m corresponding to the fields in the
Lagrangian density (2.2), i.e., corresponding to the n = 1 SSG breather. Setting = i
in Eq. (2.3), we obtain
=
2 /4
B,
1 + ( 2 /4)
(2.4)
where we have introduced the SShG parameter B.

2 Of course, a similar relation exists between the usual (non-supersymmetric) sine-Gordon (SG) and sinhGordon (ShG) models. It is more straightforward to infer the scattering theory for the trigonometric (SG, SSG)
models than for the hyperbolic (ShG, SShG) models, because the former have kinks with topological charge,
whose non-diagonal S matrices must satisfy highly restrictive constraints [7,8,42,43]. The S matrices for the
hyperbolic models are inferred by analytic continuation of the corresponding breather S matrices, as is explained
in more detail below.
614
Since the SShG spectrum corresponds to the n = 1 SSG breather, we infer that the
SShG S matrix S( ) for two particles of rapidities 1 , 2 (and corresponding energy Ei =
m cosh i and momentum Pi = m sinh i , i = 1, 2) is given by the analytic continuation of
the n = 1 SSG breather S matrix,
S( ) = SShG ( )SSUSY ( ),
(2.5)
where = 1 2 . The scalar factor SShG ( ) is given by

sinh i sin(2B)
.
(2.6)
sinh + i sin(2B)
This is the S matrix of the usual (non-supersymmetric) sinh-Gordon model [45,46], which
is the analytic continuation of the n = 1 SG breather S matrix [47], but with a different
dependence on the coupling constant. It satisfies
SShG ( ) =
SShG ( )SShG ( ) = 1,
SShG ( ) = SShG (i ).
(2.7)
The factor SSUSY ( ) is given by 3

SSUSY ( ) = Y ( )R( ),
(2.8)
where R( ) is a 4 4 matrix acting on the tensor product space V V , where V is the

2-dimensional vector space of 1-particle states. We choose {|b( )i, |f ( )i} to be the basis
of V (corresponding to a boson, fermion with rapidity , respectively); and hence, the basis
of V V is given by {|b1, b2 i, |b1 , f2 i, |f1 , b2 i, |f1 , f2 i}, where |b1 , b2 i |b(1), b(2 )i,
etc. In this basis, R( ) is given by
0 d( )
a+ ( ) 0
0
b c( ) 0
(2.9)
R( ) =
,
0 c( ) b
0
d( )
a ( )
with
2i sin B
,
b = 1,
sinh
sin B
i sin B
,
d =
.
c=
sinh 2
cosh 2
a ( ) = 1
(2.10)
It is important to note that the matrix elements of R( ) satisfy the free fermion condition
[6,39]
a+ a + b 2 = c 2 + d 2 .
The matrix R( ) is a solution of the
(2.11)
YangBaxter equation 4
3 The matrix S
SUSY ( ) for SSG was first obtained [5] in terms of an unknown parameter by solving the
constraints coming from supersymmetry and factorization. The identification of in terms of was made in
[44].
4 We use the very useful convention (which is standard in the spin-chain literature [4850], but unfortunately
not in the field theory literature), whereby Rij ( ) acts nontrivially on the ith and j th vector spaces. For instance,
in the YangBaxter equation, the R matrices act on V 3 , and therefore R12 ( ) = R( ) I, R23 ( ) = I R( ),
etc., where I is the unit matrix.
615
R12 (1 2 )R13 (1 3 )R23 (2 3 )

= R23 (2 3 )R13 (1 3 )R12 (1 2 ).
(2.12)
This matrix is both P and T invariant,

P12 R12 ( )P12 = R12 ( ),
R12 ( )t1 t2 = R12 ( ),
where ti denotes transposition in the ith space, and P is the permutation matrix
1 00 0
0 0 1 0
P =
0 1 0 0 .
0 00 1
(2.13)
(2.14)
Moreover, Y ( ) is a scalar factor given by

Y ( ) =
sinh 2
sinh 2 i sin B
!
Z
dt sinh(it/) sinh(t (1 + B)) sinh(tB)
,
exp
t
cosh t cosh2 2t
(2.15)
which is a solution of the unitarity and crossing constraints

Y ( )Y ( ) =
sinh2
sinh2
+ sin2 B
Y ( ) = Y (i ).
(2.16)
Let us denote the total scalar factor by Z( )

Z( ) = SShG ( )Y ( ).
(2.17)
One can show that Z( ) has the integral representation [6]

Z( ) =
sinh 2
sinh 2 + i sin B
!
Z
dt sinh(it/) sinh(t (1 B)) sinh(tB)
exp
,
t
cosh t cosh2 2t
(2.18)
which is the same as the expression in Eq. (2.15), except with B B. It has no poles 5
in the physical strip (0 < Im < ), provided B lies in the range
0 < B < 1,
which corresponds to 0 < 2 < .
In short, the proposed SShG bulk S matrix S( ) is given by
5 Although Y ( ) has a pole at = i2B , it is canceled by a corresponding zero of S
ShG ( ).
(2.19)
616
S( ) = Z( )R( ),
(2.20)
where the scalar factor Z( ) is given by Eq. (2.18) and the matrix R( ) is given by
Eqs. (2.9), (2.10).
It is known (see, e.g., [34]) that the SShG bulk S matrix is invariant under the strong
which implies
weak duality transformation 4/,
B 1 B.
(2.21)
Indeed, this invariance can be checked by inspection of the matrix elements (2.10) of
R( ) and the expression (2.18) for Z( ). (The factors SShG ( ) and Y ( ) are not separately
invariant.) Note that this transformation maps the range (2.19) into itself.
2.2. Boundary
We turn now to boundary conditions and boundary scattering, following the framework
developed by Ghoshal and Zamolodchikov [9]. An investigation of which boundary terms
can be added to the bulk SShG model (2.2) without spoiling (classical) integrability has
led to the following results [10]: the boundary Lagrangian
LSShG
boundary = cosh ( 0 ) + M + + ,
M 6= 1,
(2.22)
breaks supersymmetry but preserves integrability; and

LSShG
boundary =
cosh
2
(2.23)
preserves both supersymmetry and integrability. Notice that the boundary terms (2.22)
involve a total of 5 boundary parameters , 0 , M, , . If , are nonzero, then fermion
number is not conserved.
The proposed boundary S matrix S( ) for a particle of rapidity is given by (compare
with Eq. (2.5)) 6 , 7
()
S( ) = SShG ( ; , ) SSUSY ( ; ).
(2.24)
The scalar factor SShG ( ; , ), which depends on two boundary parameters , , is given
by

4B
4iB
SShG ( ; , ) = X0 ( ) X1 ;
X1 ;
,
(2.25)
where
X0 ( ) = (1)(1 + 2B)(2 2B),
X1 ( ; F ) =
1
,
(1 F )(1 + F )
(2.26)
with
6 We make an effort to distinguish boundary quantities from the corresponding bulk quantities by using Sans
Serif letters to denote the former, and Roman letters to denote the latter.
7 The boundary S matrix is obtained in terms of a set of boundary parameters (, , , ) by solving the
boundary YangBaxter equation. The relation of these parameters to those in LSShG
boundary (2.22) is not known.
(x)
sinh
sinh
617
ix
4
.
ix
4
(2.27)
This is the boundary S matrix of the usual (non-supersymmetric) boundary sinh-Gordon

model, which is the analytic continuation of the n = 1 boundary sine-Gordon breather S
matrix [51]. It satisfies
SShG ( ; , )SShG ( ; , ) = 1,
SShG

i
i
+ ; , SShG (2 ) = SShG
; , .
2
2
(2.28)
()
The factor SSUSY ( ; ) is given by [11,12]

()
SSUSY ( ; ) = Y() ( ; ) R()( ; ),
(2.29)
where is a discrete parameter which can be either +1 or 1, and is a continuous

boundary parameter. R() ( ; ) is a 2 2 matrix acting on the vector space V of 1-particle
states, which is given by
!
()
cosh 2 G()
i sinh
+ + i sinh 2 G
()
R ( ; ) =
,
(2.30)
()
()
i sinh
cosh 2 G+ i sinh 2 G
where
r cosh + e

sinh2 2
, if 0 = ,
1
+
sin
B
()
G 0 =

sinh2 2
r sinh + e
, if 0 =
1 + sin B
and

r=
2( + sin B)
sin B
(2.31)
1/2
.
(2.32)
The matrix R() ( ; ) is a solution of the boundary YangBaxter equation [52]

()
()
R12 (1 2 ) R1 (1 ; ) R12(1 + 2 )R2 (2 ; )

()
()
= R2 (2 ; ) R12(1 + 2 ) R1 (1 ; )R12 (1 2 ).
(2.33)
We remark that for , the matrix R() ( ; ) becomes diagonal and commutes with
S of the supersymmetry charges [1113].
linear combinations Q Q
In order to determine the scalar factor Y()( ; ), we recall that the full boundary S
matrix must satisfy boundary unitarity S( )S( ) = I and boundary cross-unitarity [9],
which can be written in matrix form as
t0

i
i
+ P01 S01 (2 )t1 = S1
.
(2.34)
tr0 S0
2
2
We observe that the matrix R()( ; ) satisfies
618
R() ( ; )R()( ; ) = h( )I,
(2.35)
where

cosh ,
h( ) = c0 + c1 sinh2 + c2 sinh4
2
2
and

c0 =
c1 =
r 2 cosh2 ,
r 2 sinh2 ,
(2.36)
if = +1,
if = 1,
r 2 e2
+ 2,
1 + sin B
c2 =
r 2 e2
.
(1 + sin B)2
(2.37)
Also,

()
tr0 R0
i
+ ;
2
t0

P01 R01 (2 )
t1
()
= g( ) R1

i
; ,
2
(2.38)
where
g( ) =
sinh i sin B
.
sinh
(2.39)
Setting
()
()
Y() ( ; ) = Y0 ( )Y1 ( ; ),
()
(2.40)
()
it follows that Y0 ( ) and Y1 ( ; ) must satisfy

()
()
Y0 ( )Y0 ( ) cosh = 1,
()
Y0
and

i
() i
+ Y (2 )g( ) = Y0
,
2
2

2
4
+ c2 sinh
c0 + c1 sinh
= 1,
2
2

() i
() i
+ ; = Y1
; ,
Y1
2
2
(2.41)
()
()
Y1 ( ; )Y1 ( ; )
(2.42)
respectively.
For simplicity, we shall henceforth restrict our attention to the case = +1, and so we
shall drop the superscript (). We propose the following integral representations for Y0 ( )
and Y1 ( ; ):
!
Z
1 dt sinh(2it/) sinh(t (1 + B)) sinh(tB)
i
Y0 ( ) =
,
exp
2
t
cosh2 t cosh2 2t
2 sinh 2 + i
4
0
Y1 ( ; ) =
619
1
r cosh
Z
exp 4
0

t
dt
cosh(t /) cosh (1 2B)
t
2

!

i
it
t 1
t
1+
sinh
sinh t cosh
sinh
,
2
2
2
(2.43)
where is a function of the boundary parameter defined by

= cos1 1 + e2 (1 + sin B) .
(2.44)
In order to streamline the notation, let us denote the set of boundary parameters {, , }
by , and denote the total scalar factor by Z( ; )
Z( ; ) = SShG ( ; , )Y( ; )
= Z0 ( )Z1 ( ; ).
(2.45)
The proposed SShG boundary S matrix is then given by

S( ; ) = Z( ; )R( ; ).
(2.46)
We now observe that the boundary S matrix is also invariant under the strongweak
duality transformation (2.21). Indeed, it is evident that the matrix R( ; ) (2.30) has this
invariance, if we assume that the parameter remains invariant under this transformation.
Let us now consider the scalar factor. The part of the scalar factor that does not depend on
boundary parameters can be written in the form
Z0 ( ) = X0 ( )Y0 ( )
=
2 sinh
(2.47)

i
4
!
Z

1 dt sinh(2it/)
cosh(t (1 2B))(1 + 2 cosh t) + cosh t ,
exp
4
t cosh2 t cosh2 2t
(2.48)
in which the duality invariance is manifest. (The factors X0 ( ) and Y0 ( ) are not separately
invariant.) Finally, the part of the scalar factor which does depend on boundary parameters,

4B
4iB
Z1 ( ; ) = X1 ;
X1 ;
Y1 ( ; ),
(2.49)
is also invariant under duality, since each of its factors are separately invariant (provided
the boundary parameters B and B are assumed to remain invariant).
620
3. Yang equations
Having specified the bulk and boundary S matrices, we are ready to start the TBA
program. The first step is to formulate the Yang matrix and relate it to a commuting transfer
matrix. Since this is not obvious for the case of boundaries, we begin by reviewing the more
familiar case of periodic boundary conditions.
3.1. Closed
We consider N particles of mass m with real rapidities 1 , . . . , N and two-particle S
matrix S( ) (2.20) in a periodic box of length L 1/m. The Yang equation [18,35] for
particle 1 (which has momentum P1 = m sinh 1 ) is given by

(3.1)
eiLm sinh 1 Y(1) I |1 , . . . , N i = 0,
where Y(1) is the Yang matrix 8
Y(1) = S1N (1 N )S1,N1 (1 N1 ) S12 (1 2 ),
(3.2)
which acts on V N . There are similar equations, and corresponding matrices Y(i) , for the
other particles i = 2, 3, . . . , N .
The objective is to diagonalize Y(i) . The key to this problem is to relate Y(i) to
an inhomogeneous closed-chain transfer matrix, for which there are well-developed
diagonalization techniques. (For reviews, see, e.g., [4850].) Indeed, consider the transfer
matrix (see Fig. 1)

(3.3)
closed( |1 , . . . , N ) = tr0 S0N ( N ) S02 ( 2 )S01 ( 1 ) ,
with inhomogeneities 1 , . . . , N . Notice that we have introduced an additional (auxiliary) 2-dimensional vector space denoted by 0. The product of S matrices inside the trace
(the so-called monodromy matrix) acts on V (N+1) ; but after performing the trace over the
auxiliary space, one is left with an operator which acts on the (quantum) space V N . Because S( ) satisfies the YangBaxter equation, the transfer matrix commutes for different
values of
Fig. 1. Closed-chain transfer matrix.

8 We remind the reader that we are using the convention explained in footnote 4.

closed( |1 , . . . , N ), closed ( 0 |1 , . . . , N ) = 0.
621
(3.4)
Let us now evaluate this transfer matrix at = 1 . Using the fact that S(0) = P (the
permutation matrix (2.14)) and P 2 = I, we see that
closed(1 |1 , . . . , N )

= tr0 (P01P01 )S0N (1 N ) (P01 P01 )S02 (1 2 )P01 .
(3.5)
Finally, using P01 S0i P01 = S1i and tr0 P01 = I1 , we conclude that closed(1 |1 , . . . , N ) =
Y(1) . In general, we have
Y(i) = closed(i |1 , . . . , N ),
i = 1, . . . , N.
(3.6)
This is the sought-after relation. In order to diagonalize the Yang matrices Y(i) , it suffices
to diagonalize the commuting closed-chain transfer matrix closed( |1 , . . . , N ). That
calculation, as well as the corresponding bulk TBA analysis, is described in [6].
3.2. Open
We now turn to the case with boundaries, which is our primary interest in this paper. We
therefore consider N particles of mass m with real rapidities 1 , . . . , N in an interval of
length L 1/m, with bulk S matrix S( ) (2.20) and boundary S matrix S( ; ) (2.46).
The Yang equation for particle 1 is given by [36,37]

(3.7)
e2iLm sinh1 Y(1) I |1 , . . . , N i = 0,
where the Yang matrix Y(1) is now given by
Y(1) = S1 (1 ; )S21 (1 + 2 ) SN1 (1 + N )
S1 (1 ; + )S1N (1 N ) S12 (1 2 ),
(3.8)
where the subscripts denote the left and right boundaries. (There are similar matrices Y(i)
for the other particles.) In analogy with the case of periodic boundary conditions, the key
to diagonalizing the Yang matrix is to relate it to an inhomogeneous open-chain transfer
matrix [53] (see Fig. 2)

( |1 , . . . , N ) = tr0 S0 ( + i; + )t0 S0N ( N ) S01 ( 1 )

(3.9)
S0 ( ; )S01 ( + 1 ) S0N ( + N ) ,
which commutes for different values of
Fig. 2. Open-chain transfer matrix.
622

( |1 , . . . , N ), ( 0 |1 , . . . , N ) = 0.
(3.10)
Using the boundary cross-unitarity relation (2.34) as well as the YangBaxter equation
(2.12), (2.33), one can show that
Y(i) = (i |1 , . . . , N ),
i = 1, . . . , N.
(3.11)
A proof for the case N = 2 is presented in Appendix A. 9 Hence, in order to diagonalize

the Yang matrices Y(i) , it suffices to diagonalize the commuting open-chain transfer matrix
( |1 , . . . , N ). It is to this task that we devote the following section.
4. Inversion identity and transfer-matrix eigenvalues
In this section, we consider the problem of determining the eigenvalues of the
inhomogeneous open-chain transfer matrix (3.9). Our approach will be to first derive an
exact so-called inversion identity. This approach has been used in the past to diagonalize
simple (e.g., Ising) closed-chain transfer matrices [6,17,54].
4.1. Inversion identity
Instead of working with the dressed transfer matrix (3.9), it is more convenient (see
footnote 10) to strip away the scalar factors from the bulk and boundary S matrices, and to
work instead with the bare transfer matrix

t( |1 , . . . , N ) = tr0 R0 ( + i; + )t0 R0N ( N ) R01 ( 1 )

(4.1)
R0 ( ; )R01 ( + 1 ) R0N ( + N ) ,
where R( ) is given by (2.9) and R( ; ) is given by (2.30) with = +1.
There are two key points involved in obtaining the inversion identity. The first key point
is to observe that the bulk S matrix degenerates into a one-dimensional projector for a
certain value of (= i):
1 0 0 1
1 0 0 0 0
.
(4.2)
S(i)
2 0 0 0 0
1 0 0
Hence, it is possible to fuse [48,55,56] in the auxiliary space, and thereby obtain a fusion
formula of the form [38]
t( |1 , . . . , N )t( + i|1, . . . , N ) t( |1 , . . . , N ) + ,
(4.3)
where t( |1, . . . , N ) is a fused open-chain transfer matrix (see Fig. 3), and here
represents a product of certain quantum determinants [57,58]. The fused transfer matrix is
e ) and the fused boundary S matrix e
R( ; ),
constructed from the fused bulk S matrix R(
using the fused 3-dimensional (instead of 2-dimensional) auxiliary space.
9 We therefore fill a gap left open in [36], where it was first observed that the open-chain Yang matrix is related
to the Sklyanin transfer matrix; but neither the precise form of the relation nor its proof was given.
623
Fig. 3. Fused open-chain transfer matrix.
e ) and e
The second key point is that both R(
R( ; ) can be brought to upper-triangular
form by a -independent similarity transformation. This remarkable fact is presumably due
to the fact that R( ) satisfies the free fermion condition (2.11) (cf., [40,41]). As a result,
the fused transfer matrix is proportional to the identity matrix
t( |1 , . . . , N ) I.
(4.4)
It follows from the fusion formula that the transfer matrix obeys an exact inversion
identity
t( |1 , . . . , N )t( + i|1, . . . , N ) = f ( )I,
(4.5)
where f ( ) is a calculable scalar function. We find (see Appendix B for more

details)
f ( ) =
16 sinh2
sinh( iB) sinh( + iB)(1 + sin B)2 sin2 B
(

N
Y
cosh 12 ( j ) iB cosh 12 ( + j ) iB
a

cosh 12 ( j )
cosh 12 ( + j )
j =1
b

!
j ) iB sinh 12 ( + j ) iB

sinh 12 ( j )
sinh 12 ( + j )
N
Y
sinh
j =1

j ) + iB cosh 12 ( + j ) + iB

cosh 12 ( j )
cosh 12 ( + j )
N
Y
cosh
j =1
1
2 (

!)
j ) + iB sinh 12 ( + j ) + iB
,

sinh 12 ( j )
sinh 12 ( + j )
N
Y
sinh
j =1
where
1
2 (
1
2 (

i
1
iB +
a = sinh
2
2

i
i
1
1
iB +
+ e sinh2
iB +
+
e sinh2
2
2
2
2
2
[ + ],
(4.6)
624

1
i
+
b = sinh2
iB +
2
2

i
1
1
iB +
+ e sinh2
iB +
e sinh2
2
2
2
[ + ],

i
1
+
iB +
c = sinh2
2
2

i
1
1
iB +
+ e sinh2
iB
e sinh2
2
2
2
[ + ],

i
1
iB
+
d = sinh2
2
2

i
1
1
iB +
+ e sinh2
iB +
e sinh2
2
2
2
[ + ].
i
+
2
i
+
2

(4.7)
Notice that the function f ( ) is invariant under the duality transformation B 1 B. This
inversion identity is one of the main results of this paper. We have checked it numerically
up to N = 3.
4.2. Eigenvalues
We now proceed to determine the eigenvalues of the transfer matrix. First, observe that
by virtue of the commutativity property (3.10), the bare transfer matrix t( |1, . . . , N ) has
eigenstates |1 , . . . , N i which are independent of ,
t( |1 , . . . , N )|1 , . . . , N i = L( |1 , . . . , N )|1 , . . . , N i,
(4.8)
where L( |1 , . . . , N ) are the corresponding eigenvalues. Acting on |1 , . . . N i with the

inversion identity, we obtain the corresponding identity for the eigenvalues
L( |1 , . . . , N )L( + i|1, . . . , N ) = f ( ).
(4.9)
Moreover, one can show that the bare transfer matrix t( |1, . . . , N ) is a periodic
function of with period 2i 10
t( + 2i|1, . . . , N ) = t( |1, . . . , N ),
whose asymptotic behavior for large is given by
c
t( |1 , . . . , N ) e3 I, for ,
32
where
4ie ++
.
c=
(1 + sin B) sin B
Correspondingly, the eigenvalues obey
(4.10)
(4.11)
(4.12)
10 This is not the case for the dressed transfer matrix ( | , . . . , ), due to the presence of the scalar factors.
N
1
625
L( + 2i|1, . . . , N ) = L( |1 , . . . , N ),
c
(4.13)
L( |1 , . . . , N ) e3 , for .
32
The eigenvalues L( |1 , . . . , N ) are uniquely determined by the zeros and poles of
f ( ), together with periodicity and asymptotic behavior. Indeed, observe that f ( ) is a
product of two factors. Let = zk+ , zk be zeros of the first, second factors, respectively.
Then zk+ obeys

1 +
2 (zk j ) iB

tanh 12 (zk+ j )
N
Y
tanh
j =1

1 +
2 (zk + j ) iB

tanh 12 (zk+ + j )
tanh
+
i
2 zk
=
+
sinh2 12 iB + i
2 + zk

+
sinh2 12 iB + i
+ e sinh2 12 iB + i
e
2
2 + zk

+
e sinh2 12 iB + i
+ e sinh2 12 iB + i
2
2 zk
sinh2
1
2
iB +
[ + ],
(4.14)
and zk obeys

1
2 (zk j ) + iB

tanh 12 (zk j )
N
Y
tanh
j =1
1
2
sinh2 12

e
sinh2
iB +
iB
1
2
2 1
e
sinh 2
sinh2
[ + ].
i
2
i
2

1
2 (zk + j ) + iB

tanh 12 (zk + j )
tanh
+ zk
+ zk

iB +
iB +
i
+ e
2

i
+ e
2
1
2
2 1
sinh 2
sinh2
iB
iB +
i
2
i
2
+ zk
+ zk

(4.15)
These are our magnonic Bethe ansatz equations.

It follows 11 that f ( ) can be represented as
!
N
Y

c2
sinh zk+ sinh + zk+ sinh zk sinh + zk
f ( ) = sinh2
16
k=0
!1
N
Y
2
2
sinh ( k ) sinh ( + k )
.
(4.16)
k=1
It now follows by similar arguments that

11 Indeed, let us denote the right-hand side of Eq. (4.16) by F ( ). We observe that both f ( ) and F ( ) have the
same periodicity (namely, i , which is half the period of L( |1 , . . . , N )), the same zeros and poles in the strip
i /2 < < i /2, and the same asymptotic behavior. (The apparent poles of f ( ) at = iB are canceled
by corresponding zeros.) Hence, the function g( ) = F ( )/f ( ) is regular everywhere in the complex plane,
and thus must be constant by Liouvilles theorem. By considering the limit , we see that this constant
must be 1.
626
L( |1 , . . . , N )

1
1
zk+ sinh
+ zk+
2
2
k=0

!

1
1

sinh
zk
+ zk
sinh
2
2

N
Y

1
1
k sinh
+ k
sinh
2
2
k=1

!1

1
1
k cosh
+ k
cosh
2
2
= c sinh
N
Y
sinh
(4.17)
is the unique solution to the inversion identity (4.9) with the properties (4.13). Note that
there are N + 1 pairs of roots zk , whereas in the case of periodic boundary conditions [6]
there are only N . The appearance of the additional pair of roots z0 is due to the fact that
the boundary S matrix R( ; ) is not diagonal. The existence of these roots is essential for
obtaining the correct asymptotic behavior; and it can be easily checked for the case N = 0.
In summary, the eigenvalues of the bare transfer matrix (4.1) are given by (4.17), where
zk satisfy Eqs. (4.14), (4.15).
4.3. Structure of Bethe ansatz roots
Before performing the thermodynamic (N ) limit (which is the subject of the next
section), it is necessary to first understand the structure of the Bethe ansatz roots. Following
[6], we observe that the Bethe ansatz Eqs. (4.14), (4.15) have roots of the form

xk + iB,
x iB,
+
zk = k
(4.18)
zk =
xk + iB + i,
xk iB i,
where xk are real and satisfy

N
Y
tanh 12 (xk j iB) tanh 12 (xk + j iB) sinh2 12 i
2 + xk

tanh 12 (xk j + iB) tanh 12 (xk + j + iB) sinh2 12 i
2 xk
j =1

sinh2 12 iB + i
+ e sinh2 12 i
e
2
2 xk

[ + ] = 1,
e sinh2 12 iB + i
+ e sinh2 12 i
2
2 + xk
k = 0, 1, . . . , N.
(4.19)
Evidently, for each xk , there are 4 possible combinations of roots (zk+ , zk ). However, by
considering the limit B 0, one can argue that only 2 of these combinations are allowed,
which we denote by k = +1 and k = 1, respectively:

zk+ = xk + iB, zk = xk iB i ,
k = +1:

zk+ = xk + iB + i, zk = xk iB .
(4.20)
k = 1:
Hence, the eigenvalues are specified by {xk , k }, k = 0, . . . , N :
627
QN
( xk ) k ( + xk )
,
L( |1 , . . . , N )0 N = c sinh QNk=01 k
k=1 4 sinh( k ) sinh( + k )
(4.21)
where

1
1
( ) = sinh ( iB) cosh ( + iB) ,
2
2
(4.22)
k = 1, and xk satisfy (4.19).

To close this section, we observe that the dressed transfer matrix (3.9) is simply related
to the bare transfer matrix (4.1) by
( |1 , . . . , N )
= Z( + i; + )Z( ; )
N
Y

Z( k )Z( + k ) t( |1, . . . , N ),
(4.23)
k=1
where the scalar factors Z( ) and Z( ; ) are introduced in Eqs. (2.17), (2.45). Hence, the
eigenvalues ( |1 , . . . , N ) of ( |1, . . . , N ) are given by
( |1 , . . . , N )0 N =
Z( ; + )Z( ; )Z(2 )
N
Y
i
2
Z( k )Z( + k )L( |1 , . . . , N )0 N ,
(4.24)
k=1
where L( |1 , . . . , N )0 N is given by (4.21). Here we have used the fact

Z( + i; ) =
Z( ; )Z(2 )
i
2
(4.25)
which follows from the bulk unitarity (2.7), (2.16) and boundary cross-unitarity (2.28),
(2.41), (2.42) relations.
5. Thermodynamic Bethe ansatz analysis

Having obtained the eigenvalues of the transfer matrix and the Bethe ansatz equations,
we can proceed to the derivation of the TBA equations and boundary entropy. We begin
by briefly reviewing the general framework. Following [9,19] we consider the partition
function Z+ of the system on a cylinder of length L and circumference R with left/right
boundary conditions denoted by (see Fig. 4)
Z+ = tr eRH+ = eRF
= hB+ |eLHP |B i
hB+ |0ih0|B ieLE0 ,
for L .
(5.1)
In the first line, Euclidean time evolves along the circumference of the cylinder, and H+
is the Hamiltonian for the system with spatial boundary conditions . In passing to the
second line, we rotate the picture, so that time evolves parallel to the axis of the cylinder;
628
Fig. 4. Cylinder on which the partition function Z+ is defined.
HP is the Hamiltonian for the system with periodic boundary conditions, and |B i are
boundary states which encode initial/final (temporal) conditions. In the third line, we consider the limit L ; the state |0i is the ground state of HP , and E0 is the corresponding
eigenvalue. The quantity lnhB+ |0ih0|B i is the sought-after boundary entropy [19,26]. 12
Taking the logarithm of the above expressions for the partition function, one obtains
RF LE0 + lnhB+ |0ih0|B i.
(5.2)
Whereas the free energy F has a leading contribution which is of order L, here we seek
the subleading correction which is of order 1.
5.1. Thermodynamic limit
We proceed to compute F using the TBA approach [6,1519]. To this end, we introduce
the densities P ( ) of magnons, i.e., of real Bethe ansatz roots {xk } with k = 1,
) of particles {k } and holes, respectively.
respectively; and also the densities 1 ( ) and (
Computing the logarithmic derivative of the magnonic Bethe ansatz equations (4.19), we
obtain 13
Z

1
P+ ( ) + P ( ) =
d 0 1 ( 0 ) ( 0 ) + ( + 0 )
2
0
+
where

tanh
1
ln
i
tanh

sinh
1
ln
( ) =
i
sinh
( ) =

1
( ) + 2 ( ) + + ( ) + ( ) ,
2L
(5.3)

1
4 cosh sin B
2 ( iB)
,
=
1
cosh
2 cos 2B
2 ( + iB)

1 i
1
2 2 +
,
=
1 i
cosh
2 2
12 More precisely, we shall compute the dependence of the boundary entropy on the boundary parameters. The
term in the boundary entropy which is constant (independent of boundary parameters) seems to be difficult to
compute even for simpler models [19,59].
13 The term 1 ( ) originates from the exclusion [36,37] of the Bethe ansatz root x = 0.
k
2 L

e sinh2
1
( ) =
ln
i
e sinh2
4 cosh cos
,
=
cosh 2 + cos 2
1
2
1
2
iB +
iB +
i
+ e sinh2
2

i
+ e sinh2
2
1 i
2 2
1 i
2 2
629

(5.4)
where is defined in (2.44). Defining 1 ( ) for negative values of to be equal to 1 (| |),

we obtain the final form
P+ ( ) + P ( ) =
1
1
(1 )( ) +
2
2L

( ) + 2 ( ) + + ( ) + ( ) ,
(5.5)
where denotes convolution

Z
(f g)( ) =
d 0 f ( 0 )g( 0 ).
(5.6)
We next consider the Yang equations, which imply (see Eqs. (3.7), (3.11))
e2iLm sinh k (k |1 , . . . , N ) = 1,
k = 1, . . . , N,
(5.7)
where ( |1 , . . . , N ) is the eigenvalue of the dressed transfer matrix ( |1 , . . . , N ),

which is given by (4.24). Computing the logarithmic derivative, we obtain
)
1 ( ) + (
(
Z

1
2m cosh + d 0 1 ( 0 ) Z ( 0 ) + Z ( + 0 )
=
2
0
Z
+

d 0 P+ ( 0 )+ ( 0 ) + P ( 0 ) ( 0 )
+ P ( 0 )+ ( + 0 ) + P+ ( 0 ) ( + 0 )

)
1
Z ( ) 2Z (2 ) +
Im ln Z( ; + ) +
Im ln Z( ; ) , (5.8)
+
L
where
Z ( ) =
Im ln Z( ),
( ) =
Im ln ( ).
(5.9)
Using the fact ( ) = 12 ( ), and defining P ( ) for negative values of to be equal

to P (| |), we obtain
)
1 ( ) + (

1
1
m
(1 Z )( ) +
(P+ P ) ( )
= cosh +
4

2
1
Im ln Z( ; + ) +
Im ln Z( ; ) . (5.10)
Z ( ) 2Z (2 ) +
+
2L
630
We now use (5.5) to eliminate P , and use the expressions (2.45), (2.49) to separate the
various factors in Z( ; ) to obtain
)
1 ( ) + (

1
1
1
m
cosh +
P+ +
1 Z

=
2
2
4

1
1
1
Z
+2
Im ln Z0 ( ) Z (2 )

+
2L
4
4

1
Im ln Y1 ( ; + )

+
4 +

1
Im ln Y1 ( ; )

+
4

4+ B
4i+ B
1
ln X1 ;
X1 ;
+
i

1
4 B
4i B
ln X1 ;
X1 ;
.
(5.11)
+
i
Noting the bulk identity [33]

Z ( )
1
( )( ) = 0,
4
(5.12)
and its boundary counterparts

1
Im ln Y1 ( ; )
( )( ) = 0,
4
1
1
Im ln Z0 ( ) Z (2 )
( )( ) = ( ) + ( ),
4
4
(5.13)
we remain with the rather simple result

)=
1 ( ) + (
1
m
cosh +
(P+ )( )
2

1
4+ B
4i+ B
1
( ) + 2 ( ) + ;
+ ;
+
2L
2

4i B
4 B
+ ;
,
(5.14)
+ ;
where
( ; F ) =
4 cosh cos(F /2)

1
ln X1 ( ; F ) =
.
i
cosh 2 + cos F
(5.15)
The thermodynamic limit of the magnonic Bethe ansatz equations and the Yang
equations, given by (5.5) and (5.14), respectively, are the main results of this subsection.
Notice that the former depends on the boundary parameters , while the latter depends
on the (boundary sinh-Gordon) boundary parameters , .
631
5.2. TBA equations and boundary entropy

The free energy F is given by
F = E T S,
(5.16)
where the temperature is T = 1/R, the energy E is

E=
N
X
k=1
L
m cosh k =
2
Z
d 1 ( )m cosh ,
(5.17)
and the entropy S is [15,17]

L
S=
2

d (1 + )
ln(1 + )
1 ln 1 ln

+ (P+ + P ) ln(P+ + P ) P+ ln P+ P ln P .
(5.18)
Extremizing the free energy (F = 0) subject to the constraints

1
1 ,
2
1
P+ ,
(5.19)
= 1 +
2
(which follow from Eqs. (5.5), (5.14), respectively) we obtain a set of TBA equations which
is the same as for the case of periodic boundary conditions [6,33]
P = P+ +
1
( L2 )( ),
2
1
( L1 )( ),
0 = 2 ( ) +
2
r cosh = 1 ( ) +
(5.20)
where

r = mR,
Li ( ) = ln 1 + ei () ,

P
,
2 = ln
.
1 = ln
1
P+
(5.21)
We next evaluate F using also the constraints (5.5), (5.14) and the TBA equations. From
the boundary (order 1) contribution, we obtain (see Eq. (5.2)) the boundary entropy
lnhB+ |0ih0|B i

Z (
4+ B
1
4i+ B
1
d ( ) + 2 ( ) + ;
+ ;
=
4
2

4i B
4 B
+ ;
L1 ( )
+ ;
)

+ ( ) + 2 ( ) + + ( ) + ( ) L2 ( ) .
(5.22)
632
In particular, the dependence of the boundary entropy of a single boundary on the boundary
parameters is given by 14
sB (, , )
)

Z (
4B
4iB
1
d ;
+ ;
L1 ( ) + ( )L2 ( ) ,
=
4
(5.23)
where the kernels ( ; F ) and ( ) are defined in Eqs. (5.15) and (5.4), respectively. The
term involving L1 , which had previously been conjectured [33], depends on the boundary
sinh-Gordon parameters , . The term involving L2 , which had not been anticipated,
depends on the boundary parameter (which appears in R( ; ), i.e., the non-diagonal
part of the boundary S matrix). This expression for the boundary entropy is another of the
main results of this paper.
6. Boundary roaming trajectories

One application of our result (5.23) for the boundary entropy is to obtain boundary
roaming trajectories corresponding to c < 3/2 superconformal models. In order to best
explain this result, it is helpful to first recall earlier work on bulk and boundary roaming.
Zamolodchikov [29] first considered the TBA equations for the bulk ShG (nonsupersymmetric) model with the coupling constant analytically continued to complex
values,
(6.1)
= i0, 0 1.
2
The corresponding effective central charge ceff (r) interpolates (roams) between the
values
6
, p = 3, 4, 5, . . .
(6.2)
cp = 1
p(p + 1)
corresponding to the unitary c < 1 minimal models [20]. Indeed, a plot of ceff (r) vs.
log(r/2) reveals a staircase with plateaus at values of ceff (r) equal to cp .
This result was later generalized [32] to the boundary ShG model: choosing the value of
r so that ceff (r) lies on some plateau, the boundary entropy sB (F ) (where F is a boundary
parameter) interpolates between values corresponding to various conformal boundary
conditions [21].
The original work [29] was also generalized [33] to the bulk SShG (supersymmetric)
model. The TBA equations with a similar analytic continuation of the coupling constant
(6.3)
B = i0 , 0 1
2
cause the effective central charge ceff (r) to interpolate between the values
14 For the case = 1, we obtain a similar result, except the parameter appearing in the kernel ( ) is now
given by = cos1 (1 + e2 (1 sin B )) instead of by Eq. (2.44).
cp =

3
8
1
,
2
p(p + 2)
p = 4, 6, 8, . . .
633
(6.4)
corresponding to the even unitary c < 3/2 minimal models [27,28]. Precisely this set of
TBA equations had been conjectured earlier in [30], and then further generalized in [31].
Finally, let us consider the model of primary interest here, namely, boundary SShG.
For simplicity, we fix 0 = 0 in the boundary Lagrangian (2.22), which corresponds to
= 0. 15 Due to the roaming limit (6.3), we should rescale the remaining two parameters
, so that the boundary entropy can be a function of well-defined (finite) boundary
parameters. For this purpose we set 0 and while keeping 0 and 0 2
finite. Let us introduce new boundary parameters f1 and f2 defined by (see Eq. (2.44))
20
f1 ,
1
1 + e0 2 cosh f2 .
2
(6.5)
We can reexpress the boundary entropy (5.23) in terms of these parameters as

sB = sB(1) + sB(2) , where
Z
1
(i)
d ( ; fi )Li ( ),
sB =
4
i = 1, 2,
(6.6)
with
( ; f ) =
4 cosh cosh f
.
cosh 2 + cosh 2f
(6.7)
To compute the roaming boundary entropy, we fix a value of r where ceff (r) lies on a
plateau (6.4). Then, as we change the boundary roaming parameters f1 and f2 , we check
if the boundary entropy interpolates between the values [33] 16
sin r sin s
p
p+2

sB (r, s) = ln
sin p
sin p+2
= sB (r, 1) + sB (1, s)
(6.8)
corresponding to conformal boundary states (r, s) (which, in turn, correspond 17 to primary

fields (r,s) ).
15 Consider the boundary SSG model first. When = 0 the total Lagrangian respects C symmetry due to the
0
Z2 symmetry . Therefore, the boundary S matrix should respect C symmetry, namely the soliton and
antisoliton should scatter equally on the boundary. Since the topological sector of the SSG S matrix is encoded
in the SG part, the boundary parameter should vanish as it does in the SG model [9]. This holds also for the
boundary SShG S matrix because the two models are related by the fusion procedure.
16 Note that this expression satisfies s (1, 1) = 0. The correct expression for the conformal boundary entropies
B
has an additional constant term (i.e., independent of both r and s); we neglect this term here, since we are
mostly interested in differences sB (r, s) sB (r 0 , s 0 ), for which the constant term cancels.
17 We recall [21] that for each bulk primary field
(r,s) , there corresponds a conformal boundary state |h (r,s) i
(which, for brevity, we denote here by (r, s) ) such that the partition function Z(1,1)(r,s) for the CFT on a cylinder
with conformal boundary states (1, 1) and (r, s) is given by Z(1,1)(r,s) = (r,s) (q), i.e., the character of (r,s) .
In particular, Z(1,1)(1,1) = (1,1) (q) is the character of the unit operator.
634
(1)
Fig. 5. Boundary roaming trajectories: sB vs. f1 .

(1)
Indeed, we can see clearly from Fig. 5 [33] that sB interpolates between boundary
entropies of the conformal boundary states

(a 2, 1),
(1, a)
(6.9)
(a, 1),
a odd.
Similarly sB(2) generates the new flow (see Fig. 6) 18

(a 2, 1),
(1, a)
(a, 1),
a even.
(6.10)
While these flows are generated by changing one parameter while fixing the other, we
can generate more general flows by changing f1 and f2 simultaneously. In view of the
additivity property (6.8), these two sets of flows can be combined to generate additional
flows for the total boundary entropy sB
(s, r),
(s 2, r),
(r, s)
(6.11)
(s, r + 2),
(s 2, r + 2), r s = odd.
Note that r s = even/odd corresponds to the NeveuSchwarz/Ramond sectors, respectively.
18 For p > 4, we cannot associate any conformal boundary state to the final plateau (i.e., for asymptotically large
values of the boundary parameter f2 ), since there is no state (0, 1).
635
(2)
Fig. 6. Boundary roaming trajectories: sB vs. f2 .
7. Discussion
We have presented the exact solution of the boundary SShG model an integrable QFT
whose bulk and boundary S matrices are not diagonal. In particular, we have derived an
exact inversion identity (4.5)(4.7), as well as the TBA equations and boundary entropy
(5.23). Moreover, we have uncovered a rich pattern of boundary roaming trajectories,
which remain to be understood in detail.
Although the boundary SShG model has a special feature which allows it to be solved by
an inversion identity (namely, the bulk S matrix satisfies the free-fermion condition (2.11)),
it is by no means the only such model. Indeed, there are infinite families of integrable QFTs
with N = 1 or N = 2 supersymmetry [6063] that have this property. These models have
bulk and boundary S matrices which are similar to those of SShG, and therefore, we expect
similar inversion identities to hold. We hope to report on these models in the near future
[64].
Finally, we recall [53] that one can readily obtain the Hamiltonian of an integrable open
quantum spin chain with N spins from any homogeneous open-chain transfer matrix t( |0)
(4.1). Indeed, the Hamiltonian H is given by

t( |0) ,
(7.1)
H
=0
which commutes with t( |0). For the R matrices which we have considered here (2.9),
(2.30), the corresponding Hamiltonian is that of a certain anisotropic XY chain with both
636
bulk and boundary magnetic fields. By determining the eigenvalues (4.17) of the transfer
matrix, we have evidently also solved the corresponding open quantum spin chain. It
would be interesting to exploit this solution to determine properties of this model in the
thermodynamic limit.
Acknowledgements
We thank O. Alvarez, D. Bernard, E. Corrigan, G. Delius, P. Dorey, P. Fendley,
M. Martins and H. Saleur for helpful comments and/or correspondence. One of us (R.N.)
is grateful for the hospitality at the APCTP in Seoul (where this work was initiated) and at
the CRM in Montreal (where the results were first reported). This work was supported in
part by KOSEF 1999-2-112-001-5 (C.A.) and by the National Science Foundation under
Grant PHY-9870101 (R.N.).
Appendix A. Relation of Yang matrix to Sklyanin transfer matrix

In Section 3.2, we stated that the Yang matrix (3.8) is related to the Sklyanin open-chain
transfer matrix (3.9) in the following way (3.11):
Y(i) = (i |1 , . . . , N ),
i = 1, . . . , N.
(A.1)
We present here a proof for the case N = 2. Evaluating the transfer matrix at = 1 , we
have

(1 |1 , 2 ) = tr0 S0 (1 + i; + )t0 S02 (1 2 )S01 (0)

S0 (1 ; )S01 (21 )S02 (1 + 2 )

= tr0 S02 (1 2 )P01 S0 (1 ; )(P01 P01 )S01 (21)(P01 P01 )

(A.2)
S02 (1 + 2 )(P01 P01 )S0 (1 + i; + )t0 = .
In passing to the second line, we have used the cyclic property of the trace, as well as
S(0) = P and P 2 = I, where P is the permutation matrix (2.14).

= S1 (1 ; ) tr0 S02 (1 2 )S01 (21 )S12 (1 + 2 )

(A.3)
P01 S0 (1 + i; + )t0 = .
Here we have used P01 X0 P01 = X1 , and the P symmetry of the R matrix (2.13).

= S1 (1 ; ) tr0 S12 (1 + 2 )S01 (21 )S02 (1 2 )

P01 S0 (1 + i; + )t0 = .
Here we have used the YangBaxter equation (2.12).
(A.4)
637
= S1 (1 ; )S12 (1 + 2 )

tr0 S01 (21 )(P01 P01 )S02 (1 2 )P01 S0 (1 + i; + )t0
= S1 (1 ; )S12 (1 + 2 )

tr0 S01 (21 )P01 S0 (1 + i; + )t0 S12 (1 2 )
= S1 (1 ; )S12 (1 + 2 )S1 (1 ; + )S12 (1 2 ).
(A.5)
In passing to the last line, we have used the boundary cross-unitarity relation (2.34) with
t1
= i
2 1 , and the crossing relation S01 (i ) = S01 ( ). Comparing the last line to
the expression (3.8) for the Yang matrix, we conclude that
(1 |1 , 2 ) = Y(1) .
(A.6)
For higher values of N , the proof is similar.
Appendix B. Derivation of inversion identity

In Section 4.1, we give the important inversion identity (4.5)(4.7). Here we explain in
more detail how we derived it. As already mentioned in text, the main idea is to formulate
the fusion formula, following Ref. [38], to which we shall refer as I. 19
Although the dressed bulk S matrix S( ) (2.20) is regular at = 0, the bare bulk S
matrix R( ) (2.9) has a pole there. In order to avoid complications from this spurious pole,
in this appendix we rescale R( ) by the factor sinh ; i.e., we take R( ) to be given still by
(2.9), but now with matrix elements
a ( ) = sinh 2i sin B, b( ) = sinh ,
c( ) = 2i sin B cosh , d( ) = 2 sin B sinh .

2
2
(B.1)
Keeping in mind the symmetries (2.13) of the R matrix, the unitarity relation (I 2.3) is

sinh2 + sin2 B ,
( ) = 4 cosh2
(B.2)
R12 ( )R12 ( ) = ( )I,
2
2
and the crossing relation (I 2.4) is
R12 ( ) = V1 R12 ( )t2 V1 ,
with 20
= i,

V=
1 0
0 1
(B.3)

.
e
The matrix R12 ( ) at = is proportional to the one-dimensional projector P
12
19 In order to facilitate comparison with [38], we use here similar notations.
20 Alternatively, choosing = i , one has V = I.
(B.4)
638
1
0
1
Pe12 =
2 0
1
0
0
0
0
0 1
0 0
,
0 0
0 1
Pe12
2
= Pe12
.
(B.5)
As explained in I, from the corresponding degeneration of the (boundary) YangBaxter

equation, one can derive identities which allow one to prove that fused (boundary) S
matrices satisfy generalized (boundary) YangBaxter equations.
The fused R matrix is given by (I 2.13)
e+ R13 ( )R23 ( + )Pe+ ,
Rh12i3 ( ) = P
12
12
(B.6)
e+ = I Pe . An important observation (which one can verify by direct calculation)

where P
12
12
is that the fused R matrix can be brought to upper triangular form by a similarity
transformation 21
X12 Rh12i3 ( )X12 = upper triangular,
where the 4 4 matrix X is independent of , and is given by
1
0
0
2
2
0 sin B cos B 0
2
2
X2 = I.
X=
,
B
0 cos B
sin
0
2
2
1
0
0
1
2
(B.7)
(B.8)
It follows that the fused monodromy matrices 22 (I 4.7), (I 5.4), (I 5.5)

Th12i ( ) = Rh12iN ( ) Rh12i1 ( ),
b
Th12i ( + ) = Rh12i1 ( ) Rh12iN ( ),
(B.9)
also become triangular by the same transformation.

Denoting (as in I) our bare boundary S matrices R( ; ), R( + i; + ) by K ( ),
+
K ( ), respectively, the corresponding fused matrices are given by (I 3.5), (I 3.9)
e+ K ( )R12 (2 + )K ( + )Pe+ ,
( ) = P
Kh12i
12 1
2
12
+ + t

+
+
1
e+ t12 ,
Kh12i ( ) = Pe12 K1 ( ) R12 (2 3)K2 ( + )t2 P
12
(B.10)
since M = V t V = I.
Remarkably, the fused K matrices are also brought to upper triangular form by the same
similarity transformation
21 This observation is similar to, but not the same as, the one made by Felderhof [40,41]. Indeed, in our language,
+
)
he shows that R13 ( )R23 ( + ) (i.e., the expression for the fused transfer matrix without the projectors Pe12
can be brought to triangular form by a (somewhat more complicated) -independent similarity transformation.
Although for the case of periodic boundary conditions both approaches lead to the inversion identity, this appears
to be no longer true for the case of boundaries.
22 For simplicity, we consider here the homogeneous case ( = 0, i = 1, . . . , N ).
i
X12 Kh12i
( )X12 = upper triangular.
639
(B.11)
It follows that the fused transfer matrix t( ), which is given by (I 4.5), (I 4.6)
+
t( ) = tr12 Kh12i
( )Th12i ( )Kh12i
( )Tbh12i ( + ),
(B.12)
is proportional to the identity matrix,

t( ) I,
(B.13)
where the proportionality factor is determined from the diagonal elements of the various
triangular matrices.
The fusion formula is given by (I 4.17), (I 5.1)
t( )t( + )

1
t( ) + {K + ( )}{K ( )}{T ( )}{Tb( )} ,
=
(2 + 2)
(B.14)
where the transfer matrix t( ) is given by (4.1) (see also (I 4.1), (I 4.2)), and the quantum
determinants [57,58] are given by (I 4.15), (I 5.3), (I 5.7)
{T ( )} = {Tb( )} = ( + )N ,

e K ( )R12 (2 + )K ( + )V1 V2 ,
{K ( )} = tr12 P
12 1
2

e V1 V2 K + ( + )R12 (2 3)K + ( ) .
{K + ( )} = tr12 P
12
2
1
(B.15)
Reverting to the original normalization of the R matrix by rescaling each of the transfer
matrices t( ) in (B.14) by (sinh )2N , introducing the inhomogeneities i in the obvious
way, and factoring the result into a product of two factors, we arrive at the results (4.5)
(4.7). 23
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
P. Di Vecchia, S. Ferrara, Nucl. Phys. B 130 (1977) 93.

J. Hruby, Nucl. Phys. B 131 (1977) 275.
S. Ferrara, L. Girardello, S. Sciuto, Phys. Lett. B 76 (1978) 303.
L. Girardello, S. Sciuto, Phys. Lett. B 77 (1978) 267.
R. Shankar, E. Witten, Phys. Rev. D 17 (1978) 2134.
C. Ahn, Nucl. Phys. B 422 (1994) 449.
A.B. Zamolodchikov, Al.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
A.B. Zamolodchikov, Sov. Sci. Rev. A 2 (1980) 1.
S. Ghoshal, A.B. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 3841.
T. Inami, S. Odake, Y.-Z. Zhang, Phys. Lett. B 359 (1995) 118.
C. Ahn, W.M. Koo, J. Phys. A 29 (1996) 5845.
C. Ahn, W.M. Koo, Nucl. Phys. B 482 (1996) 675.
M. Moriconi, K. Schoutens, Nucl. Phys. B 487 (1997) 756.
H. Saleur, 1998 Les Houches lectures, cond-mat/9812110.
23 We have refrained from giving explicit results for the intermediate steps, which are rather unwieldy and not
very illuminating. We have done these computations with the help of M ATHEMATICA.
640
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
C.N. Yang, C.P. Yang, J. Math. Phys. 10 (1969) 1115.

Al.B. Zamolodchikov, Nucl. Phys. B 342 (1990) 695.
Al.B. Zamolodchikov, Nucl. Phys. B 358 (1991) 497.
A.B. Zamolodchikov, Al.B. Zamolodchikov, Nucl. Phys. B 379 (1992) 602.
A. LeClair, G. Mussardo, H. Saleur, S. Skorik, Nucl. Phys. B 453 (1995) 581.
A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333.
J. Cardy, Nucl. Phys. B 324 (1989) 581.
A.M. Tsvelick, P.B. Wiegmann, Adv. Phys. 32 (1983) 453.
N. Andrei, K. Furuya, J.H. Lowenstein, Rev. Mod. Phys. 55 (1983) 331.
P. Fendley, Phys. Rev. Lett. 71 (1993) 2485.
P. Fendley, F. Lesage, H. Saleur, J. Stat. Phys. 85 (1996) 211.
I. Affleck, A.W.W. Ludwig, Phys. Rev. Lett. 67 (1991) 161.
M. Bershadsky, V. Knizhnik, M. Teilman, Phys. Lett. B 151 (1985) 31.
D. Friedan, Z. Qiu, S.H. Shenker, Phys. Lett. B 151 (1985) 37.
Al.B. Zamolodchikov, Resonance factorized scattering and roaming trajectories, unpublished
preprint, 1991.
M.J. Martins, Phys. Lett. B 304 (1993) 111.
P. Dorey, F. Ravanini, Nucl. Phys. B 406 (1993) 708.
F. Lesage, H. Saleur, P. Simonetti, Phys. Lett. B 427 (1998) 85.
C. Ahn, C. Rim, J. Phys. A 32 (1999) 2509.
G. Mussardo, Nucl. Phys. B 532 (1998) 529.
C.N. Yang, Phys. Rev. Lett. 19 (1967) 1312.
P. Fendley, H. Saleur, Nucl. Phys. B 428 (1994) 681.
M. Grisaru, L. Mezincescu, R.I. Nepomechie, J. Phys. A 28 (1995) 1027.
L. Mezincescu, R.I. Nepomechie, J. Phys. A 25 (1992) 2533.
C. Fan, F.Y. Wu, Phys. Rev. B 2 (1970) 723.
B.U. Felderhof, Physica 65 (1973) 421.
B.U. Felderhof, Physica 66 (1973) 279, 509.
C. Ahn, D. Bernard, A. LeClair, Nucl. Phys. B 346 (1990) 409.
D. Bernard, A. LeClair, Commun. Math. Phys. 142 (1991) 99.
C. Ahn, Nucl. Phys. B 354 (1991) 57.
S.N. Vergeles, V.M. Gryanik, Yad. Fiz. 23 (1976) 1324.
B. Schroer, T.T. Truong, P.H. Weisz, Phys. Lett. B 63 (1976) 422.
I.Ya. Arefeva, V.E. Korepin, JETP Lett. 20 (1974) 312.
P.P. Kulish, E.K. Sklyanin, in: Lecture Notes in Physics, Vol. 151, Springer, 1982, p. 61.
V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method, Correlation
Functions and Algebraic Bethe Ansatz, Cambridge Univ. Press, 1993.
R.I. Nepomechie, Int. J. Mod. Phys. B 13 (1999) 2973.
S. Ghoshal, Int. J. Mod. Phys. A 9 (1994) 4801.
I.V. Cherednik, Theor. Math. Phys. 61 (1984) 977.
E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, 1982.
M. Karowski, Nucl. Phys. B 153 (1979) 244.
P.P. Kulish, N.Yu. Reshetikhin, E.K. Sklyanin, Lett. Math. Phys. 5 (1981) 393.
A.G. Izergin, V.E. Korepin, Sov. Phys. Doklady 26 (1981) 653.
A.G. Izergin, V.E. Korepin, Nucl. Phys. B 205 (1982) 401.
P. Dorey, I. Runkel, R. Tateo, G. Watts, Nucl. Phys. B 578 (2000) 85.
K. Schoutens, Nucl. Phys. B 344 (1990) 665.
C. Ahn, Prog. Theor. Phys. 118 (1995) 165.
M. Moriconi, K. Schoutens, Nucl. Phys. B 464 (1996) 472.
P. Fendley, K. Intriligator, Nucl. Phys. B 380 (1992) 265.
C. Ahn, R.I. Nepomechie, in preparation.

Strong coupling approach to the supersymmetric

Kondo model
P. Coleman a, , C. Ppin b , A.M. Tsvelik b
a Materials Theory Group, Department of Physics and Astronomy, Rutgers University, 136 Frelinghausen Road,
Piscataway, NJ 08854, USA

b Department of Physics, Oxford University, 1 Keble Road, Oxford OX1 3NP, UK
Received 24 February 2000; accepted 26 June 2000
Abstract
We carry out the strong coupling expansion for the SU(N) Kondo model where the impurity
spin is represented by a L-shaped Young tableau. Using second order perturbation theory around
the strong coupling fixed point it is shown that when the antisymmetric component of the Youngtableau contains more than N/2 entries, the strong-coupling fixed point becomes unstable to a twostage Kondo effect. By comparing the strong coupling results obtained here with the result using a
supersymmetric large N expansion, we are also able to confirm the validity of the the supersymmetric
formalism for mixed symmetry Kondo models. 2000 Elsevier Science B.V. All rights reserved.
PACS: 78.20.Ls; 47.25.Gz; 76.50+b; 72.15.Gd
1. Introduction
In this paper we present a strong-coupling treatment of a single-impurity Kondo
model where the spin is a higher representation of the group SU(N). We consider spin
representations that can be tuned continuously from being antisymmetric to being fully
symmetric. This work is motivated in part by a desire to understand how the presence of
strong Hunds interactions between electrons modify the spin quenching process. These
issues become particularly important in heavy electron systems, where the localized
electrons can be subject to Hunds interactions which far exceed their kinetic energy [1,2].
The basic model of interest is the single-site Kondo model, described by the Hamiltonian
X
J
(1)
H=
k ck ck + c (0)0 c (0) S.
N
E-mail address: coleman@physics.rutgers.edu (P. Coleman).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 1 9 - 3
642
P. Coleman et al. / Nuclear Physics B 586 [FS] (2000) 641667
P
1
k ck creates an
ns
1
2
= ( , , . . .) are the form
Here, the spin sums run over N > 1 possible values, c (0) =
electron at the origin, ns is the number of sites. The matrices 0

a basis of N 2 1 traceless SU(N) matrices, where Tr[ a b ] = ab and S = (S 1 , S 2 , . . .) is
a spin describing a particular representation of SU(N). The above model was first derived
by Coqblin and Schrieffer [5], who showed that a rare earth ion containing a single, spin
orbit coupled f -electron corresponds to the above model, with N = 2j + 1, and the spin
in the fundamental representation of SU(N).
When we come to consider more complex local moment systems, we need to consider
the atomic spins formed by combining more than one elementary spin. Previous treatments
of this model have considered local moments described by symmetric or antisymmetric
representations of SU(N), denoted by the Young tableaux [68]
(a) Antisymmetric
(b) Symmetric
6
nf
.
.
.
.

nb 2S
?
The first representation describes nf elementary spins that have been combined into
a purely antisymmetric spin wavefunction; the second describes nb 2S elementary
spins that have been combined into a purely symmetric spin wavefunction. These
two representations are of particular interest because the former can be described by
a combination of nf spin, or Abrikosov pseudo fermions, whereas the latter can
be described by a combination of nb 2S Schwinger bosons", and is the natural
generalization of spin-S to SU(N). For a given number of spins, the antisymmetric and
symmetric spin combinations represent two extremes where the Casimir S2 attains its
extremal values. Loosely speaking, the antisymmetric and symmetric spin representations
are the combination of spins with the smallest and largest total spin, respectively. The
symmetric spin configuration can thus be thought of as a state where a large Hunds
coupling has maximized the total spin.
In this paper, we are interested in a class of L-shaped spin representations composed
of spins which interpolate between these two extremes, as shown in Fig. 1 below. This
family of representations enables us to examine the effect of progressively turning on
the Hunds interaction the effect of progressively increasing the strength of the Hunds
interaction between the constituent spins inside an atom. In are a real multi-electron local
moment, such as a U 3+ ion, containing three localized f -electrons, the Hunds interaction
also imposes the crystalline symmetry, leading to a model with a far lower symmetry. Our
toy representation enables us to separate the leading order effect of the Hunds interaction
from the additional complications of lowering the symmetry.
In a previous paper [3], we showed that the above mixed symmetry representations of
SU(N) spins can be described using a super-symmetric spin representation. In particular,
if b and f ( = (1, N)) are Bose and Fermi creation operators, respectively, then a mixed
643
Fig. 1. A sequence of L-shaped Young tableaux which interpolate between and antisymmetric and
symmetric representation. Each tableau has six boxes, corresponding to six elementary spins.
symmetry representation of is obtained by writing the spin S as a sum of a bosonic and a

fermionic spin
S = b 0 b + f 0 f ,
(2)
= (N 2 1)/2 independent SU(N)
( 1 , . . . , M )
represents the M
generators.
where 0
In this way, the spin representation combines aspects of the bosonic Schwinger boson
representation of spins and the fermionic Abrikosov pseudo-fermion representation of
spins. This spin operator commutes with the the operators
X
X
f b ,
=
b f .
(3)
=
=1,N
=1,N
These operators interconvert the Bose and fermion fields and form the generators of the
supergroup SU(1|1). An irreducible L-shaped representation of SU(N) is obtained by
imposing two constraints
n f + n b = Q,
1
n f n b + [ , ] = Y,
Q
(4)
where Q and Y are determined from the Young tableau via the relations Q = nf + nb and
Y = nf nb . These constraints also commute with the generators and , so the entire
spin representation is supersymmetric.
This mixed representation of the spin-operators allows a consideration of the properties
of Kondo models with L-shaped Young-tableaux by developing a supersymmetric field
theory and then carrying out a large-N expansion. In the conventional one-channel Kondo
model, the spin is screened from S to S 12 by a process that is characterized by a
single temperature scale, the Kondo temperature. In the classic picture, this leads to a
screening cloud of dimension l = vF /TK , where vF is the Fermi velocity of electrons in the
conduction sea and TK is the single Kondo temperature. One of the unexpected results of
the new analysis, was that under some circumstances a local moment under the influence
of a Hunds interaction can undergo a two-stage Kondo effect associated with two separate
644
Fig. 2. Illustrating a local moment that is screened by a two-stage screening process in a single
scattering channel. The screening cloud contains a shell structure, with an inner and outer
conduction electron cloud which reduce the total moment from S to S = S 1.
Kondo temperatures TK1 and TK2 , leading to a screening cloud with internal structure,
characterized by two screening length scales
l1 =
vF
,
TK1
l2 =
vF
,
TK2
(5)
where vF is the Fermi velocity of the electrons, as illustrated in Fig. 2. In the language
of the renormalization group, a new fixed point occurs where the impurity is screened in
two stages reducing the effective spin from S to to S = S 1. In this paper we provide
a complimentary strong-coupling treatment of the same model. Our results confirm the
key results obtained in the large N expansion, providing an important check on the validity
of the supersymmetric field theory.
It is well known that under the renormalization group, the antiferromagnetic Kondo
model renormalizes towards strong coupling [9,10]. The weak-coupling beta function for
the Kondo model is independent of representation, and for a single channel model, takes
the form
(g) =

g3
dg()
= g 2 +
+ O g4 ,
d
N
(6)
where g = J and is the conduction electron density of states. For the simplest
one-channel Kondo models, the coupling constant g flows smoothly from the unstable
weak-coupling fixed point, to a stable strong coupling fixed point. However, in certain
cases the strong coupling fixed point is itself unstable. The most famous example of
such behaviour is the two-channel Kondo model, where the flow to strong coupling is
intercepted by an intermediate coupling fixed point that is characterized by a non-Fermi
liquid properties [4,1113]. The two stage Kondo effect discussed here is another example
of an instability at strong coupling. As we shall see, the strong coupling fixed point
becomes unstable to a second Kondo effect at an exponentially smaller temperature.
The classic analysis [4] of the stability of the strong coupling fixed point of a Kondo
model follows Wilsons method [14] of formulating the Kondo model on a lattice:
H = t
X
n>1,
645
J
c (n + 1)c (n) + H.c. + c (0)0 c (0) S.
N
(7)
The strong-coupling fixed point is obtained by first setting t = 0, and solving for the
ground-state of the one-site problem
J
c (0)0 c (0) S
(8)
N
which leads to a partially screened local moment with spin S . When a finite t/J is
restored, virtual charge fluctuations of electrons onto, and off site 0 induce an effective
interaction between the spin density at site 1 and the residual moment, given by
HK =
H (1) = J S c (1)0 c (1),
(9)
where J = O(t 2 /J ) determines the strength of the coupling between the residual local
moment and the conduction electron at site 1. The stability of the strong coupling
fixed point is determined by the sign of J . If this coupling is ferromagnetic (J < 0),
then a residual ferromagnetic Kondo effect with the low energy electrons causes J
to scale logarithmically to zero, decoupling the residual spin from the conduction sea
and stabilizing the strong-coupling fixed point. Conversely if the effective coupling is
antiferromagnetic (J > 0), then the effective model flows to strong coupling and the
strong coupling fixed point of the initial Kondo model becomes unstable.
To get a better idea of how antiferromagnetic coupling can arise in our single-channel
model, consider a local moment, described by an L-shaped representation of SU(N),
denoted by the Young tableau
2S S=
6
nf
(10)
?
where we have replaced nb = 2S. In the ground-state of the strong-coupling Hamiltonian
HK , electrons form a singlet with the fermionic part of the spin creating a partially screened
moment, denoted by a Young tableau with a completely filled first row:
2S 2S 1

-
6
S = (0e (0) + S) =
(11)
c0
c0
? c0
where in this example we have taken N = 8. Since the first column of the tableau is a singlet
(with N boxes), it can be removed from the tableau, leaving behind a partially screened
646
spin S 1/2, described by a row with 2S 1 boxes. If we now couple the electron at the
origin with electrons at site 1 via a small hopping matrix element t J , then the virtual
charge fluctuations of electrons in and out of the singlet at the origin will lead to a residual
coupling between the partially screened moment and the electrons at the neighboring site
1
H (1) = J S c (1)0 c (1),
(12)
where J t 2 /J . In the SU(2) Kondo model, only electrons parallel to the residual
moment S can hop onto the origin, which gives rise to a ferromagnetic coupling J < 0. In
the SU(N) case, electrons can hop provided they are not in the same spin state as electrons
at the origin. The sign of the coupling J depends on the number of conduction electrons
nc = n nf , bound at the origin. For example, if nf = 1, so that nc = N 1 in the ground
state, electrons hopping onto the origin will have to be parallel to the residual spin, so in
this case the coupling is ferromagnetic, J < 0. By contrast, if nc N , there are many
ways for the electron to hop onto the origin with a spin component that is different to the
residual moment, so the residual interaction will be antiferromagnetic, J > 0. In this case,
the strong-coupling fixed point becomes unstable, and a second-stage Kondo effect occurs,
binding a further N 1 electrons at site 1 to form a state represented by the tableau

2S
S = (0e0 + 0e1 + S) =
c1
c1
c1
N
c1
c0 c1
c0 c1
? c0 c1
2S 2

(13)
which corresponds to a residual spin S = S 1. This final configuration is stable, because

an electron at site 2 can only hop onto site 1 if it is parallel to the unquenched moment,
so the residual interaction between site 2 and site 1 will be ferromagnetic.
To examine the stability of the strong coupling fixed point in detail, this paper follows
the method of Nozires and Blandin [4], using second order perturbation theory about the
strong-coupling fixed point to determine the sign of the residual interaction between the
unscreened spin and the bulk of conduction electrons. If J < 0 the residual coupling is
ferromagnetic and the strong coupling fixed point is stable: the impurity residual spin is
S = S 1/2 after screening. If J > 0 the residual interaction is antiferromagnetic and
the strong coupling fixed point becomes unstable, leading to the two-stage Kondo effect
S = S 1. In parallel with this approach, we carry out a large N treatment of the strongcoupling limit, using the technique developed in our previous paper. By comparing the two
techniques we are able to confirm the validity of the field theoretic approach developed in
our earlier work. Both methods are able to confirm that for N > 2, J changes sign when
647
Table 1
Strong coupling
Large N
Ground state energy Eg
NJ [(1 n f )(n f + q/N]
NJ (1 n f )n f
1E = E (ne + 1) Eg
J (1 n f q/N)
J (1 n f )
1E = E (ne + 1) Eg
J (1 n f + n b q/N)
J (1 n f + n b )
nf N
(N1)(Nnf Q/N)
QN
(Q+Nn )(1+N+Q2n
Q/N)
f
f

(Nnf )(Qnf )N
(N1)(Q+Nn )(n +Q/N)
f
f
t2
(2S)J
Effective coupling
constant
t 2 (1/2n )
J (1n )n (1fn +n )
f f
f
b
the number of bound-conduction electrons is less than N/2, and in the large N limit is
given by

1
f
t2
2 n
,
(14)
J =
J (1 n f )n f (1 n f + n b )
where n f = nf /N and n b = 2S/N .
Table 1 gives the main results, comparing the strong coupling and large N expressions
for the ground-state energy Eg , the excitation energies 1E and 1E to add an electron
to the ground state in a spin configuration that is parallel or anti-parallel to the residual
spin. The last row compares the effective coupling constant J calculated by both methods.
In these expressions, we denote n f = nf /N , n b = nb /N , q = Q/N .
2. The one site impurity problem
To begin, we start with the one-dimensional rendition of the SU(N) Kondo model,
X
J
(15)
c (n + 1)c (n) + H.c. + c (0)0 c (0) S,
H = t
N
n>0,
where c (j ) creates an electron at the site of the local moment, S is the spin of the
local moment and 0 is a generator of the SU(N) group. For simplicity we have put
the impurity at the beginning of the chain. The spin of the impurity is represented by an
L-shaped Young tableau while the conduction electron is represented with a single box.
In the strong coupling limit, only the local part of the Hamiltonian is relevant and the
problem becomes single sited. In order to determine what is the ground state of the single
site problem, we consider the impurity described by a Young tableau with Q boxes. In
order to simplify further calculations, we denote by nb the number of boxes in the row and
by nf the number of boxes in the column minus one. (This will correspond in the next
section to put some bosons in the row as well as in the corner and to fill the remaining
648
Fig. 3. Screening of the impurity. n1 and n2 denote the number of conduction electrons screening the
impurity in the first and second column of the Young tableau.
of the column with fermions.) In the strong coupling limit, the conduction electrons will
be trapped at the impurity site, screening the impurity. We note n1 and n2 the number of
conduction electrons screening the impurity respectively in the first and second column of
the Young tableau (see Fig. 3).
The energy of the ground state can be expressed in terms of second order Casimirs
E=

J 2
J
Sc Si =
Stot S2i S2c ,
N
2N
(16)
where S2tot is the second order Casimir of the impurity screened by the n1 + n2 conduction
electrons; S2i is the Casimir of the free impurity and S2c is the Casimir of the conduction
electrons which screen the impurity.
If we normalize the generators of the fundamental representation of SU(N) according
to Tr[ ] = , then the expression for the Casimir of an arbitrary irreducible
representation is [15]
S2 =
X
Q(N 2 Q)
+
mj (mj + 1 2j ),
N
(17)
j =1,N
where m j is the number of boxes in the j th row from the top. The energy of the ground
state is then given by

J
2n2 (N n2 ) + (nf + n1 n2 )(N nf n1 n2 1)
E=
2N
1
1
(Q + n1 + n2 )2 nf (N nf 1) + Q2
N
N
1
(n1 + n2 )(N + 1 n1 n2 ) + (n1 + n2 )2 .
(18)
N
The energy in the [n1 , n2 ] plane has the form represented in Fig. 4. In the range of
parameters we are interested in (n1 [0, N nf 1], n2 [0, N 1]), only one minimum
remains and the ground state of the problem is found to be
n1 = N nf 1,
n2 = 0
(19)
which corresponds to the usual one-stage Kondo model where the impurity spin is screened
by 1/2.
649
Fig. 4. Single impurity energy as a function of the parameters n1 and n2 . We see that the minimum
is given when n1 is maximum (n1 = N nf 1) and n2 is minimum (n2 = 0).
Fig. 5. In the special case where Q/N = k is an integer and n1 = k, it costs no energy to add
additional electrons to the ground-state and the fixed point behavior of this particular Kondo model
will then involve valence fluctuations between the different degenerate configurations. The figure
illustrates the situation where Q = N = 6 and n1 = 1.
A special exception to this case occurs when Q/N = k is an integer, when, if n1 =

Q/N = k, E/n2 = 0 for all n2 . The point n1 = Q/N = k, n2 [1, N 1] corresponds
to a line of degenerate ground states. An example is illustrated in Fig. 5. In this case, the
strong coupling fixed point is also unstable, but the fixed point physics will be governed
by valence fluctuations between the degenerate states of different n2 . We shall exclude this
special case from the discussion here, leaving it for future work.
3. Second order perturbation theory around the strong coupling fixed point
In the leading order in 1/J two processes of excitation appear. Suppose the screened
impurity is at site zero. Then one electron from site one can hop briefly to site zero,
then return (process 1, Fig. 6). Alternatively one electron from site zero can make a
virtual hop to site one and return (process 2). In the first process the intermediate state has
one more electron at site one (nc + 1), where nc is the number of conduction electrons
which screen the impurity in the ground state). We note this state |GS + 1, 0i. In the
second process of excitations the intermediate state has nc 1 electron at site zero and
650
Fig. 6. Process I: an electron makes a virtual incursion from site 1 onto site 0.
Fig. 7. Process II: an electron at site zero makes a virtual incursion to site 1.
Fig. 8. (a) Parallel configuration: the spin of the electron at site 1 is symmetrized with the impurity
spin. (b) Anti-parallel configuration: the spin of the electron at site 1 is anti-symmetrized with the
impurity spin.
two electrons at site one. We note it |GS 1, 2i. If we call |GS, 1i the initial state (the
site 0 has the impurity in the ground state and the site 1 has one electron), then using
second order perturbation theory, the energy shift due to the perturbation is given by
1E = t 2
|hGS + 1, 0|c1c2 |GS, 1i|2

|hGS 1, 2|c2c1 |GS, 1i|2
+ t2
,
E(ne ) E(ne + 1)
E(ne ) E(ne 1)
(20)
where E(ne ) is the energy of the initial state; E(ne + 1) is the energy of the intermediate
state in process 1 and E(ne 1) is the energy of the intermediate state in process 2.
Now for each process of excitation the spin of the electron at site 1 can is either
symmetrically (Fig. 8(a)), or antisymmetrically (Fig. 8(b)) correlated with the spin at the
impurity. For SU(2) spins, this corresponds to a spin that is either parallel or anti-
651
parallel to the impurity spin, and we shall adopt the same convention for the SU(N) case.
There are two corresponding possibilities for the energy shifts
M(2)
M(1)
+
,
1E =
E(ne ) E (ne + 1) E(ne ) E(ne 1)
(21)
M(1)
M(2)
1E =
+
,
E(ne ) E (ne + 1) E(ne ) E(ne 1)
where, for example, M(1) and M(1) are the matrix elements for process one in the parallel
and antiparallel configurations, respectively.
Supposing now that the effective Hamiltonian around the strong coupling fixed point
takes the form
H (1) = J S c (1)0 c (1)
(22)
by computing the energy difference between the parallel and anti-parallel configurations,
the effective Kondo coupling constant is then
1E 1E = Jeff (2S).
(23)
Thus by evaluating the energy shifts in the second order perturbation theory we are able to
determine the sign and hence the stability of the fixed point.
3.1. Evaluation of the energies
First, consider the energy of the ground state at site one. It is given by
J
[(N 1 nf )(nf + (N + Q)/N)].
(24)
N
Suppose now one electron jumps from site 1 to site 0 in the parallel state, as in Fig. 8(a),
the the energy of the intermediate state is

J 2
S2imp S2el ,
S
E (ne ) =
2N N+1
E(ne ) =
where S2N+1 is the Casimir of the intermediate state; S2imp is the Casimir of the impurity
before screening and S2el is the Casimir of all (nc + 1) conduction electrons involved into
the intermediate state. The resulting energy is given by

J
E (ne + 1) = 2 nf N 2 nb nf Nnf .
N
Similarly if the starting spin configuration is the anti-parallel one, as in Fig. 8(b), the
energy of the intermediate state is

J
E (ne + 1) = nb (1 nf /N) + nf (N nf nf /N) .
N
For process 2, the energy of the intermediate state does not depend on whether the spin
configuration is parallel or anti-parallel, and is given by
E(ne 1) E(ne ) =
J
(nf + 1 + Q/N).
N
652
Fig. 9. Ground state after the screening of the impurity spin by conduction electrons. Note that we
have changed our notations here and have given a spin S + 1/2 to the initial impurity spin.
In conclusion the two energy differences associated with process 1 are given by
E (ne + 1) E(ne ) = (N nf Q/N),

N
(25)
E (ne + 1) E(ne ) = J (N nf + nb Q/N),

N
where we have replaced nf + 1 nf and nb nb . The excitation energy associated with
process 2 is

J
nf + Q/N
(26)
E(ne 1) E(ne ) =
N
for both spin configurations.
3.2. Matrix elements
The calculation of the matrix elements is much more complex and requires a detailed
expression for each state involved in terms of operators. We will begin by introducing the
notation, using the ground state as an example. Then we will review the matrix elements
for each process of excitation.
3.2.1. Notation
The screened impurity in its ground state is given by the following Young tableau
represented in Fig. 9: Each box is filled with a field to which an index is attached. In order
to describe the state we have to first symmetrize the indices of the fields in the row and then
antisymmetrize the indices in the column. We choose to fill the row with nb = 2S bosons
in a given spin state (say, all the bosons have index A), and the rest of the column will
be filled with nf = nf 1 f -fermions and nc conduction electrons so that nf + nc = N .
With these conventions the ground state can be expressed as
"
X
1
2S
(P)f1 fn cn +1 cN1 |0i
bA
|GSi =
f
f
NGS
P { }{1...N}{A}
#
X
X 2S1
b i
(P)f1 fn cn +1 cN1 |0i , (27)
bA
+
i 6=A
P ({ })({1...N}{i })
653
Fig. 10. Schematic tensor product between the screened impurity in the ground state and a conduction
electron. Note that the number of boxes in the column of the screened impurity is N (the impurity is
screened).
where (1 . . . n ) = P(1 . . . n ) are permutations of the set of indices {1 n } strictly

ordered. (P) is the sign of the permutation. NGS is the normalization factor

N 1 1/2
(2S 1)! nf ! nc !,
NGS = 2S + N 1
nf

where ab is the number of ways of choosing a elements out of a group of b possible
choices and nc = (N 1 nf ). In all that follows we will lighten the notation by keeping
track of the degenerescences in any expression of states. For example the state |GSi can be
written
"
X
1
2S1
nf ! nc ! bA
0 f1 fn cn +1 cN1 |0i
|GSi =
f
f
NGS
S nf {1...N}{A}
X
i 6=A
2S1
b i
bA
X
Snf ({1...N}{i })
0 f1
#
fn cn +1
f
f
cN1 |0i
, (28)
where we suppose that the two sets of indices (1 . . . nf ) and (nf +1 . . . N1 ) are ordered
in a strictly increasing order. 0 is the sign of the residual permutation of indices due to the
fact that the two sets of indices are not ordered with respect to each other. Snf {2 . . . N}
denotes the partition of the two sets of indices (1 . . . nf ) and (nf +1 . . . N1 ) chosen
among {2 . . . N} and strictly ordered inside each set. The important point here is that the
above sum runs over linearly independent states only. Thus when we have to evaluate the
norm of a state, the prefactors in front of the sum have to be squared. We give in the
appendix the detailed evaluation of the norm of the ground state as an example.
3.2.2. Decomposition rule into parallel and anti-parallel states
When we add an additional electron to the ground-state, the state that forms is a linear
combination of two states, one in which the spin is symmetrically correlated with the
impurity, and another in which or antisymmetrically correlated with the partially screened
impurity, as illustrated by the Young tableau (10). In the analogous SU(2) problem, these
two states would correspond to adding an electron into a state in which the spin that is
654
either parallel or anti-parallel to the impurity spin. Written out in operator language,
then if is the spin index of the added electron,
parallel
z
}|
{

1
c |GSi + (2S 1)cA |int()i

c |GSi =
2S
anti-parallel
}|
{
z

1
(2S 1)c |GSi (2S 1)cA |int()i .
2S
(29)
|int()i is a state where the index of the new box has been exchanged with the indices
of the bosons in the ground state (here, remember, all the bosons have index A). In each
case the spin index of the added electron is symmetrized, or anti-symmetrized with the the
boson spin indices, leaving the indices of the first column untouched because this column
is a filled singlet in the ground state. The state |int()i can be written
|int()i
1
=
NGS
+
"
N1
2S1
b
nf nf !nc !(bA )
X
i =
6 {A,}
2S2
bA
2
b
S nf {1...N}{A}
N1
2S2
b bi
nf nf !nc !(bA )
N1
nf nf !nc !
0 f1 fn cn
X
f +1
0 f1 fn cn
S nf {1...N}{A}
X
Snf {1...N}{A}
0 f1
cN1 |0i
fn cn +1
f
f
f +1
cN1 |0i
#
cN1 |0i
, (30)
where the first line corresponds to one index in the bosonic row of the Young tableau; the
second line corresponds to one index in the bosonic row and one index i ({1 . . . N}
{A, }) and the third line to two indices in the bosonic row. We note that the state
|int()i is not normalized: we have hint ()|int ()i = 1/(2S).
It is now convenient to use the above decomposition rule (29) to define two initial states
|(GS, 1A ) i and |(GS, 1 ) i, where the electron is added in a parallel and antiparallel
spin state, respectively. |(GS, 1A ) i is the state where one electron of index A at site 1
is in the parallel configuration with the ground state at site 0. Its thus the projection of
|GSi onto its parallel part. (Here c1,A

is the creation operator for an electron at site 1
c1,A
of index A.) We choose to give an index A to the electron at site 1 because it is already in
the parallel configuration with the ground state at site 0:

(GS, 1A ) c |GSi.
(31)
A
Note that this state is normalized. |(GS, 1 ) i is the state where one electron of index at
site 1 is in the anti-parallel configuration with the ground state at site 0. We require 6= A
so that this configuration exists. We want this state to be normalized (by convention) so
that

(GS, 1 ) =

2S 1
|int()i .
c1, |GSi c1,A
2S
655
(32)
3.2.3. Process 1, parallel case

In this process, one electron in the initial state |(GS, 1A ) i makes a virtual incursion
onto site 0 from site 1. We get the following expression for the matrix elements:
X
GS + 1 , 0c c1, GS , 1A 2 ,
(33)
M(1) = t 2
0,
denotes the creation operator for an electron at site 0 of index and c1,
where c0,
denotes the creation operator for an electron at site 1 of index . hGS + 1 , 0| is the
intermediate state, with one more electron at site 0 and no electron at site 1. By convention
we suppose that both the intermediate and initial states are normalized. In the numerator
c1, |GS , 1A i = |GS + 1 i, where is a multiplicative

in (33), we notice that c0,
constant. Indeed while the electron is transferred from site two to site one it stays in the
same spin configuration with respect to the impurity spin. Thus (33) can be rewritten
X

(GS, 1A ) c1,
.
(34)
M(1) = t 2
0 c0, 0 c0, c1, (GS, 1A )
, 0
As electron at site 1 has index A, we get
|GSi.
M(1) = t 2 hGS|c0,A c0,A
c0,A |GSi and finally

Then M1 = hGS|1 c0,A

nc
,
M(1) = t 2 1
2S 1 + N
(35)
where nc /(2S 1 + N) is the number of conduction electrons of index 1 in the ground

state. The explicit evaluation can be found in the appendix.
3.2.4. Process 1, anti-parallel case
The initial state is |GS , 1 i where one electron at site 1 is in the anti-parallel
configuration with the ground state at site 0. With the same reasoning as before, we have
X

(GS, 1 ) c1,
.
(36)
M(1) = t 2
0 c0, 0 c0, c1, (GS, 1 )
, 0
Remembering the electron at site 1 has index ,
|GSi.
M(1) = t 2 hGS|c0, c0,
Finally, as derived in the appendix,

nc
2
.
M(1) = t 1
N 1
(37)
(38)
656
3.3. Process 2, parallel case

The initial state is |(GS, 1 ) i where one electron at site 1 with index A is in the parallel
configuration with the ground state at site 0. One electron at site 0 jumps onto site 1 and
comes back:
X
GS 1 , 2c c0, (GS, 1A ) 2 .
(39)
M(2) = t 2
1,
As above we consider that the initial state and the intermediate state |GS 1 , 2i (there
is one less electron in the ground state and one more at site 1) are normalized and with the
same proportionality argument,
X

GS , 1A c0,
M(2) = t 2
0 c1, 0 c1, c0, (GS, 1A )
, 0
=t
2
, 0
We thus get
M(2) = t 2

(GS, 1A ) c0,
.
0 , 0 c1, c1, 0 c0, (GS, 1A )
c0, (GS, 1A ) t 2 hGS|c0,A

c0,A |GSi.
(GS, 1A ) c0,
(40)
(41)
We define
c0, (GS, 1A ) t 2 hGS|c0,A

c0,A |GSi,A
(42)
M(2), t 2 hGS, 1A |c0,
P
so that M(2) = M(2), . The second term in (42) is the number of conduction
electrons in the ground state that we have already calculated in (35) and which equals
nc /(2S 1 + N). The first term in this equation requires some exact evaluation with the
expression of the state in terms of operators. After some algebra done in the appendix, we
get
(
nc (2S + N 2)
2
(43)
M(2), = t (N 1)(2S + N 1) , 6= A,
0,
= A.
An then
M(2) = t 2 nc

nc
.
2S + N 1
(44)
3.3.1. Process two, anti-parallel case

The initial state is |(GS, 1 ) i where one electron at site 1 with index is in the antiparallel configuration with the ground state at site 0. Note that 6= A. On electron at site 0
jumps onto site 1 and comes back:
X

(GS, 1 ) c0,
M(2) = t 2
0 c1, 0 c1, c0, (GS, 1 )
, 0
=t
, 0

(GS, 1 ) c0,
.
0 , 0 c1, c1, 0 c0, (GS, 1 )
(45)
Remembering that the electron at site 1 has index we get

X
c0, GS , 1 t 2 hGS|c0,
c0, |GSi .
(GS, 1 ) c0,
M(2) = t 2
657
(46)
We define
c0, (GS, 1 ) t 2 hGS|c0,

c0, |GSi,
(47)
M(2), t 2 (GS, 1 ) c0,
P
so that M(2) = M(2), . After decomposing each state with creation operators, we get
(cf. appendix) three possibilities depending on the indices:
nc
t2
,
= A,
2S + N 1
= ,
(48)
M(2), = 0,
nc (2S + N 2)
t
, otherwise
(N 1)(2S + N 1)
and after summation upon the indices we find

nc
nc
+
.
(49)
M(2) = t 2 nc
N 1 (N 1)(2S + N 1)
3.3.2. Energy shifts
We evaluate the energy shifts with the expression
1E 1E =
E(ne ) E (ne + 1)
+
and we get
1E
1E
M(1)
M(1)
E(ne ) E (ne + 1)
M(2) M(2)
(50)
E(ne ) E(ne 1)

nf N
t2
=
J (N 1)(N nf Q/N)
QN
(Q + N nf )(N + Q + 1 2nf Q/N)

(N nf )(Q nf )N
(n 1)(Q + N nf )(nf + Q/N)
(51)
where we have put nf = nf + 1, in keeping with our initial definition of L-shaped Young
tableaux. This expression is valid for any (Q, nf , N).
In the large N limit when Q/N = q, nf /N = n f and (2S + 1)/N = n b , we get

1 n f
1 n f
n f
t2
1
q
. (52)
1E 1E =
J 1 n f
n f
n b + 1 n b + 1 n f
n f
Reducing to the same denominator and dividing by n b in order to get the effective
coupling (23), we finally have

1
f
t2
2 n
.
(53)
J =
J (1 n f )n f (1 n f + n b )
658
In the next section, we will check this result in the framework of the large N approach,
using the formalism of paper [3].
4. Strong coupling in the large N approach

In order to test the formalism developed in [3], we need to rederive the effective coupling
constant J in the large N limit with the path integral formulation.
4.1. Mean-field theory
The impurity is described within a SU(N) representation in form of a L-shaped Young
tableau. The impurity Kondo model is written
HK
HY
Ho
HQ
}|
{
z
}|
{
z
}|
{ z
z
}|
{
X
bb
J
k ck ck + c 0 c S + (Q Q0 ) + (Y Yo )
QY,
H=
N
Qo
(54)
k,
where Ho describes the conduction electron sea, HK is the interaction between the
P
creates
conduction electron spin density, and the local moment, where c = ns 1/2 k ck
an electron at the site of the local moment (ns = number of sites). HQ and HY impose
b = nf nb + 1 [b f = Yo , f b ].
b = nf + nb = Qo and Y
the constraints given by Q
Q
nf and nb are respectively the number of f -electrons and bosons which parametrize the
representation.
In the strong coupling limit, this Hamiltonian becomes
J X
J X
f c c f
b c c b + (Q Q0 ) + (Y Yo ).
(55)
H =
N
N
After factorizing the Kondo interaction, we obtain

X
X

c
c V f + f V c +
c b + b
H=
N
+ (Q Q0 ) + (Y Yo ).
+ (V V )
J
At the large N saddle point, the fluctuating variable V acquires a static value, which we
take to be hV i = hV i = V . The average value of the Grassman field is zero, so the Bose
field is unhybridized in the large N limit. In order that nb 0(N), the Bose field must
condense so that the chemical potential of the Bose field, b = must vanish, i.e.,
= . The mean-field Hamiltonian is then
X
N
V c f + f c + V 2 + 2 nf 2 Qo Yo ,
(56)
HMF = H =
J
where the interaction term entering into Y is ignored at this level of approximation.
After diagonalization of the fermionic Hamiltonian, we find two eigenvalues E =
p
2 + V 2 and the ground state hamiltonian can be written
HGS =

V2
2 N n f ,
E+ a +
a+ + E a a + 2N
J
659
(57)
where a = 1 (c f ) and n f = nf /N . The ground-state corresponds to complete

2
occupation of all N states in the lowest level, so that the ground-state energy is

q
V2
2
2
2 n f .
(58)
EGS = N + V +
J
Minimizing EGS with respect to and V 2 gives the two saddle-point equations
= 2n f
1 p
2 + V2
1
1
p
= ,
2
2
J
2 +V
(59)
so that
q
J = 2 2 + V 2,
= J (1 2n f ).
(60)
These are the mean-field equations of the strong coupling fixed point. Substituting back
into Eq. (58), we obtain
EGS = NJ n c n f ,
(61)
where n c = nc /N = 1 n f is the number of conduction electrons in the ground-state.

4.2. Excitation energies
Suppose now that we want to describe the excitation energies of the large N limit, to
compare with those obtained in a direct strong-coupling expansion (25). In the mean-field,
the energy level E is completely filled. If we add electrons to the impurity, they must go
into the upper level. We can add a large number of electrons into this level, so we identify
this excitation with the anti-parallel excitation states, giving an excitation energy
q
(62)
1E = E+ = + 2 + V 2 = J (1 n f ).
This matches the large N limit of Eq. (25). The leading large N calculation is able to
capture the energy to add an electron into an anti-parallel configuration because one can
add many electrons into these states, giving an energy change of order O(N). By contrast,
one can no more than one electron in the parallel spin configuration, since it is not possible
to symmetrize more than one electron at a given site. In this case, the change in energy is
of order O(1), and we must consider the Gaussian corrections to the mean-field theory to
extract this excitation energy.
4.3. Gaussian fluctuations
The most important fluctuations about the mean-field theory, are the fluctuations of the
-field. The relevant interaction part of the Lagrangian is given by
660
L =
N
c .
c b + b
J
Fluctuations of the alpha field thus mediate an interaction between the partially screened
moment and the conduction electrons, given by
where the full line denotes a conduction electron propagator, G1

c () = (
V2
2 ),
a wavy line indicates a boson propagator G1

b () = ( b ), where b = is the
chemical potential of the Bose field and a curly line indicates the propagator. Poles in
this propagator describe excitations associated with the action of the operator c b on the
ground-state. But since hb i = nb A , this operator describes the addition of an electron

into a parallel spin configuration with the unscreened local moment. Thus, to find out the
energy to add an electron to the spin in the parallel spin configuration, we must look for
poles in the propagator. The action for the fluctuations is given by
X
1
()D
()(),
(63)
Sfluc = N
where
D 1 () =
X
1
+T
Gc ( + )Gb ().
J
(64)
Now stationarity of the action with respect to the quantity V gives

X
V
V
F
= 0.
= +T
Gc ()
2
V
Using this to replace 1/J in the inverse propagator, we obtain

X
1
1
Gc ( + )
.
D 1 () = T
b + 2
(65)
Carrying out this sum using the contour-integral method, we obtain

I
dz nb (z)
1
D 1 ()
=
( 2 )
2i (z b )(z + in E+ )(z + in E )

X
f (E )
nb (b )
+
=
( E + )( E ) = (E )(E E )
=
n b f (E )
.
( E )(E E )
But at T = 0, f (E ) = 1 and f (E + ) = 0 which leads to

n b
1 + n b
( 2 )
p
+
D 1 () =
E+ E 2 V 2 + 2
(66)
661
so that finally
D 1 () =
( 2 )

p
,
( E+ )( E ) 2 V 2 + 2
(67)
where = E+ + J n b . Notice, that D() has poles at both = and = 2 . The

appearance of a second pole at = 2 is a ghost which factors out of the entire
partition function when we fix the gauge properly [3], and it does not contribute to physical
excitations. We can identify the excitation energy
1E = = J (1 n f + n b )
as the physical energy to add an electron in the perpendicular spin configuration. This result
agrees with the large N limit of result (22). Result defines the pole in the propagator.
4.4. Renormalized spin interaction between conduction electrons and residual spin
The residual interaction between the partially quenched moment and the electrons at site
1 is given in strong-coupling by the diagram
where the t denotes the hopping matrix element for an electron moving between site 1
and site 0. In this diagram, the external legs do not contribute to the interaction amplitude.
The total interaction strength at low energies is consequently
J
= t 2 [Gc ()]2 D()|=0 ,
N
(68)
where
Gc () =
V2
2
(69)
is the propagator for a conduction electron at site 0. At zero frequency, we have

2
.
V2
From Eq. (67), we have
Gc () =
(70)
2
JV2
(71)
J t 2 (2 )
.
2 V 2
(72)
D(0) =
so that
J =
662
Now using the mean-field equations, we get

= J (1 n f + n b ),
1 2n f
V2
= 2
.
2
J n f (1 n f )
so that finally
J =
t 2 (1/2 n f )
.
J (1 n f )n f (1 n f + n b )
(73)
This result corresponds to the result obtained using a direct strong coupling expansion (53).
5. Conclusion
In this paper we have investigated the effect of the representation in the description of the
SU(N) Kondo model. We used a special class of Young tableaux which are L-shaped and
we obtain a path integral expression of our model with an additional symmetry between
fermions and bosons: the supersymmetry. This formalism enables us to tune the Hunds
interactions inside the representation, which can be used to describe complex atoms like
U 3+ .
We have found that changing the representation gives rise to new fixed points in the
physics of the Kondo model. When the number of fermions (antisymmetric components
in the L-shaped Young tableau) is larger than N/2 (in the large N limit), then a two-stage
Kondo appears, where the impurity spin is screened twice by two clouds of conduction
electrons, leading to a resulting spin of S = S 1. In addition to the two-stage Kondo
effect, a class of representations (where for example nb = 2 and nf = N 1 ) lead to a
degenerate ground state and thus to a non-Fermi liquid. We didnt investigate so far the
properties of this new fixed point.
Alternatively, this paper is a test for the formalism developed in our previous work
about a large N field theory for the SU(N) Kondo model. We have given here the full
renormalization group argument to prove the instability of the fixed point leading to the
two-stage Kondo effect. We have matched the instability criterions in the limit of large N .
The excitation energies involved in adding an extra electron to the screened impurity both
in the parallel and anti-parallel states have been reproduced.
Acknowledgement
This research was supported in part by NSF grant DMR 9983156 and research funds
from the EPSRC, UK.
663
Appendix A. Some detailed evaluations

A.1. The norm of the ground state
We illustrate the use of the notation in formula (28) by checking that hGS|GSi = 1.
2S 2S
bA |0i = (2S)! we have
Using the relation h0|bA
hGS|GSi =
1
(2S)!
2
NGS
N1
2
2
nf (nf !) (nc !)
+ (2S 1)!(N 1)

N1
2
2
nf (nf !) (nc !)
(A1)
so that
hGS|GSi =
1
(2S + N 1)(2S 1)!
2
NGS
N1
2
2
nf (nf !) (nc !)
= 1.
(A2)
A.2. Number of conduction electrons of different indices in the ground state

A.2.1. Number of electrons of index 1
In Eq. (28) there are no c-electrons of index 1 in the first line. We thus get
X
1 X 2S1
(bA )
b i
0 f1 fn cn +1 cN1 |0i.
c1 |GSi =
f
f
NGS
i 6=A
S nf ({1...N}{A,i })
(A3)
The choice of indices for the f -electrons cannot be 1 any more because it is attributed to
cA |GSi we get
the c-electron in that sum. Noting ncA = hGS|cA

nf
(N 1) N2
nc
,
(A4)
=
ncA =
N1
(2S
+
N 1)
(2S + N 1)
nf
where nc is the total number of c-electrons in the ground state.

A.2.2. Number of electrons of index 6= A in the ground state
"
X
1
2S1
)
nf !nc !
0 f1 fn cn +1 cN1 |0i
(bA
c |GSi =
f
f
NGS
S nf ({1...N}{A, }
X
i 6=
2S1
(bA
)
b i
0 f1
Snf ({1...N}{,i })
#
fn cn +1
f
f
cN1 |0i
(A5)
Thus we get
2
1
nf
(2S + N 2) N2
(2S 1)! nf !nc !
2
NGS
nc (2S + N 2)
.
=
(2S + N 1)(N 1)
nc =
(A6)
664
We note that summing over the indices, we can check that the total number of conduction
electrons in the ground state is equal to nc = N nf 1.
A.3. Process 1, anti-parallel case: evaluation of the matrix element
The electron at site 1 has index 6= A.
r

2S 1
c1, |GSi c1,A

|int()i ,
|GS , 1 i =
2S
(A7)
the creation operator for an electron of index at site 1. Due to

where we have noted c1,
the presence of the electron at site 1 the two state in (A7) are anti-parallel. We get

2S 1
hGS|c1, c1,
|GSi + hint ()|c1,A c1,A
|int()i
2S
2S 1
[hGS|GSi + hint ()|int ()i]
=
2S
= 1.
hGS , 1|GS , 1i =
(A8)
Now starting from Eq. (37) the full matrix element for this case is given by

2S 1
hGS|1 c c |GSi + hint |1 cA

cA |inti + 2hint|c cA |GSi
2S
nc (2S)
2S
1
1
= t2
2S
(2S + N 1)(N 1)

nc N
nc
1
1
+2
+
2S 1
(2S + N 1)(N 1)
(2S + N 1)(N 1)

n
c
.
(A9)
= t2 1
N 1
M(1) = t 2
A.4. Process 2, parallel case: evaluation of the matrix element

We suppose that the electron at site 1 has index A. From Eq. (41) it is clear that if
the hopping electron had index = A the corresponding matrix element vanishes. For a
hopping electron with index 6= A, we can study the state |(GS, 1A ) i in the first term of
Eq. (41).
We have
|(GS, 1A ) i
"
1
2S1
nf !nc !
bA
=
NGS
+
X
i 6=A
2S1
b i
bA
Thus for 6= A we get
0 f1 fn c2,A
cn
S nf {1...N}{A}
X
Snf ({1...N}{i })
0 f1
f +1
cN1 |0i
#
fn c2,A
cn +1
f
f
cN1 |0i
. (A10)
c |(GS, 1A ) i
"
1
2S1
nf !nc !
bA
=
NGS
X
f +1
S nf ({1...N}{A, }
2S1
b i
bA
i 6=A
0 f1 fn c2,A
cn
665
cN1 |0i
#
0 f1
Snf ({1...N}{,i })
fn c2,A
cn +1
f
f
cN1 |0i
(A11)
and when taking the norm we get
M(2), = t 2
nc (2S + N 2)
(N 1)(2S + N 1)
if 6= A.
(A12)
A.5. Process 2, anti-parallel case: evaluation of the matrix element

In this process we choose on site 1 an electron with an index 6= A. The second term
in Eq. (46) corresponds to the number of c-electrons of index 6= 1 in the ground state
which has been evaluated in Section A.2.1. We study the first term in Eq. (46). We have
three different cases.
A.5.1. 6= and 6= A
r

2S 1
c |int()i .
c1, c |GSi c1,A
2S
We then have the two states:
c |GS , 1 i =
c |GSi

1
2S1
nf !nc !
bA
=
NGS
+
bA
2S1
bi
i 6=A
and
1
c |int()i =
NGS
0 f1 fn cn
Snf ({1...N}{A, }
0 f1 fn cn
N1
2S1
nf nf !nc !(bA )
X
X
i ({1...N}{A, })
c
|0i
; (A14)
N1
+1
0 f1 fn cn
f
f +1
cN1 |0i
N1
2S2
b b i
nf nf !nc !(bA )
S nf ({1...N}{A, })
cN1 |0i
S nf ({1...N}{A, })
f +1
S nf ({1...N}{,i })
"
(A13)
0 f1 fn cn
2S2 2
N1
(b )
nf nf !nc !(bA )
f +1
cN1 |0i
666
0 f1
S nf ({1...N}{A, })
fn cn +1
f
f
cN1 |0i
. (A15)
For 6= these two states are anti-parallel and we get

nc (2S + N 2)
,
(N 1)(2S + N 1)
nc (2S + N 2)
c1, c1,
c0, |int()i =
hint ()|c0,
2S(N 1)(2S + N 1)
c1, c1,
c0, |GSi =
hGS|c0,
(A16)
so that we get
M(2), = t 2
nc (2S + N 2)
(N 1)(2S + N 1)
if 6= and 6= A.
(A17)
A.5.2. = A
The two states in Eq. (A13) are still anti-parallel in that case. We have
X
1 X 2S1
(bA )
b i
0 f1 fn cn +1 cN1 |0i;
cA |GSi =
f
f
NGS
i 6=A
Snf ({1...N}{A,i })
(A18)
1
cA |int()i =
NGS
"
X
i {1...N}{A}
N1
2S2
b b i
nf nf !nc !(bA )
0 f1 fn cn
S nf {1...N}{A}
N1
nf nf !nc !
bA
S nf {1...N}{A}
f +1
cN1 |0i
fn cn +1
f
f
cN1 |0i
2S2
0 f1
2
#
.
(A19)
As a result we get
M(2),A = t 2
nc
.
2S + N 1
A.5.3. =
It is easy from Eq. (46) to check that in this case M,(2) = 0.

References
[1]
[2]
[3]
[4]
D.L. Cox, M. Makivic, Physica B 199200 (1994) 391.

M. Norman, Phys. Rev. Lett. 72 (1994) 2077.
P. Coleman, C. Ppin, A.M. Tsvelik, cond-mat 9909073.
Ph. Nozires, A. Blandin, J. Phys. (Paris) 41 (1980) 193.
(A20)
667
[5] B. Coqblin, J.R. Schrieffer, Phys. Rev. 185 (1969) 847.

[6] M. Hammermesh, Group Theory and Its Application to Physical problems, Addison Wesley,
1962, pp. 198.
[7] Lichtenberg, Unitary Symmetry and Elementary Particle, Academic Press, New York.
[8] H.F. Jones, Groups Representations and Physics, Institute of Physics, 1990.
[9] P.W. Anderson, Phys. Rev. 164 (1967) 352.
[10] P. Nozires, J. Phys. (Paris) Colloq. 37 (1976) C1-271.
[11] A.M. Tsvelik, P.B. Wiegmann, J. Stat. Phys. 38 (1985) 125.
[12] A.M. Tsvelik, J. Phys. C 18 (1985) 159.
[13] C. Destri, N. Andrei, Phys. Rev. Lett. 52 (1984) 364.
[14] K.G. Wilson, Rev. Mod. Phys. 47 (1975) 773.
[15] A. Jerez, N. Andrei, G. Zarand, Phys. Rev. B 58 (1998) 3814.

Chiral random matrix model for critical statistics

A.M. Garcia-Garcia, J.J.M. Verbaarschot
Department of Physics and Astronomy, SUNY, Stony Brook, NY 11794, USA
Received 18 April 2000; accepted 30 May 2000
Abstract
We propose a random matrix model that interpolates between the chiral random matrix ensembles
and the chiral Poisson ensemble. By mapping this model on a non-interacting Fermi-gas we show
that for energy differences less than a critical energy Ec the spectral correlations are given by chiral
Random Matrix theory whereas for energy differences larger than Ec the number variance shows
a linear dependence on the energy difference with a slope that depends on the parameters of the
model. If the parameters are scaled such that the slope remains fixed in the thermodynamic limit, this
model provides a description of QCD Dirac spectra in the universality class of critical statistics. In
this way a good description of QCD Dirac spectra for gauge field configurations given by a liquid of
instantons is obtained. 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.30.Rd; 12.39.Fe; 12.38.Lg; 71.30.+h
Keywords: QCD Dirac spectra; Spectral correlations; Critical statistics; Fermi-gas method
1. Introduction
By now it has been well-established that the smallest eigenvalues of the QCD Dirac
operator are correlated according to a Random Matrix theory with the global symmetries
of the QCD partition function [1] (see [2,3] for recent reviews and a complete list of
references). In particular, this has been confirmed by the analysis of the low-energy
effective theory [47], universality studies [812], lattice QCD simulations [4,1323] and
the study two-sublattice theories with disorder [2428]. This means that the dynamical
details of QCD are not important on energy scales of the order of the average level spacing.
The natural question that can be asked is at what energy scale the dynamics of QCD
becomes relevant and how does this manifest itself in the Dirac spectrum.
The answer to this question has been understood within the context of effective theories
[47,29,30]. The effective theory for the QCD Dirac spectrum is known as the partially
quenched effective partition function and was originally introduced to study the quenched
Corresponding author. E-mail: verbaarschot@tonic.physics.sunysb.edu
PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 3 6 2 - X
A.M. Garcia-Garcia, J.J.M. Verbaarschot / Nuclear Physics B 586 [FS] (2000) 668685
669
approximation in QCD [3135]. The central observation is that in the domain where the
mass dependence of the effective partition function is given by the contribution of the
constant fields (the zero momentum modes) the Dirac eigenvalues are correlated according
to chiral Random Matrix theory. The relevant mass scale can thus be identified as the scale
for which the Compton wavelength of the lightest particle becomes equal to the size of
the box, i.e., M 1/Ls . In QCD, in the phase of spontaneously broken chiral symmetry,
the lightest particles are the Goldstone modes with a mass given by M 2 Km (with m the
quark mass and K = /F 2 in terms of the pion decay constant F and the chiral condensate
). The critical scale is thus given by [4]
Ec =
1
.
KL2s
(1)
In the context of disordered condensed matter systems [3639] this energy scale is known
as the Thouless energy and also in this article we will adopt this name.
A more intuitive interpretation of the Thouless energy has been given in the theory
of mesoscopic systems [36,37]. The time scale h /Ec is the time for which an initially
localized wave packet diffuses all over space. For this reason the eigenvalues are correlated
according to Random Matrix theory for energy differences below Ec (known as the
ergodic regime). At shorter time scales, different wave functions do not necessarily overlap
resulting in a weakening of correlations of the corresponding eigenvalues. For energy
differences beyond the inverse elastic collision time e the corresponding eigenvalues
are completely uncorrelated (the Poisson ensemble). The domain inbetween Ec and h /e
is known as the diffusive or AltshulerShklovskii domain. A third energy scale is the
average level spacing 1. The ratio Ec /1 is identified in mesoscopic physics as the
dimensionless conductance. It is equal to the number of subsequent levels correlated
according to Random Matrix theory. The existence of these domains has been confirmed
by numerical simulations of the Anderson model [40].
In QCD the average level spacing is related to the order parameter of the chiral phase
transition, the chiral condensate, by the BanksCasher formula [41] according to 1 =
/V (with V the volume of Euclidean spacetime). The prediction from
(1) is that the
2
number of eigenvalues correlated according to chRMT is of the order F V .
For increasing disorder the number of subsequent eigenvalues described by Random
Matrix theory decreases. For strong disorder we expect that all states become localized with
uncorrelated eigenvalues. We thus expect a critical value of the disorder for which the three
scales, 1, Ec and h /e coincide. In particular, the dimensionless conductance becomes
volume independent [42]. It has been conjectured [43] that at this point the eigenvalue
correlations are described by a new universality class known as critical statistics (see [44
46] for a review). In this class, only the short range correlations of the eigenvalues are
described by the usual random matrix ensembles whereas the number variance, 2 (n),
shows a linear n-dependence beyond this domain. What is relevant for QCD is that such
behavior has been observed in numerical simulations of the 4-dimensional Anderson model
[47].
670
The volume dependence of the Thouless energy has been investigated by means of lattice
QCD simulations [17,19,23] and instanton-liquid simulations [5]. In essence, results from
lattice QCD simulations are in complete agreement with theoretical results from partial
quenched chiral perturbation theory. However, the results from instanton liquid simulations
seem to deviate from the prediction (1) with a scale independent constant K; in that case
the Thouless energy only shows a weak volume dependence. This raises the question
whether the Dirac eigenvalues might be described by critical statistics. To address this issue
we generalize a random matrix model for critical statistics [48,49] to include the chiral
symmetry of the QCD partition function (Section 2). In Section 3 we map our model on
a partition function of noninteracting fermions. This model is solved in the semi-classical
limit in Section 4, where we obtain analytical expressions for the microscopic spectral
density and the two-point correlation function. Comparisons with instanton simulations
are shown in Section 5 and concluding remarks are made in Section 6.
2. Definition of the model

The random matrix model of Moshe et al. [48] is defined by the partition function
Z
Z
dU eb Tr([U,H ][U,H ] ) ,
Z = dH e TrH H
(2)
where H is a Hermitian and U a Unitary N N matrix. The integration measures dH and

dU are given by the Haar measure. This model can be interpreted as the zero-dimensional
limit of the KazakovMigdal model [50]. It interpolates between the Gaussian Unitary
ensemble (b = 0) and the Poisson ensemble (b ). Using the invariance of the measure,
the integral over U can be replaced by an integral over the eigenvalues of U . For b
this partition function is dominated by matrices H that commute with arbitrary diagonal
unitary matrices. This set of matrices is the ensemble of diagonal Hermitian matrices which
is known as the Poisson ensemble. What is nice about this model is that it preserves the
unitary invariance which enables us to take full advantage of the existing random matrix
theory methods. In order to obtain a nontrivial b-dependence in the thermodynamic limit,
the parameter b has to be scaled as
b = h2 N 2 .
(3)
It has been shown that in this limit the model (2) is related [49] to both a banded random
matrix model [51] with a power-like cutoff [52] and to random matrix models with an
asymptotic Tr log2 H probability potential [53]. The correlation functions of the latter
model have been derived by means of q-orthogonal polynomials [53,54] and Painlev
equations [55].
In this paper we are interested in chiral random matrix ensembles defined as ensembles
of N N random matrices with the structure

0 C
,
(4)
D=
C 0
671
where C is an arbitrary complex matrix n (n + ) matrix (N = 2n + ). Since the matrix

D has exactly zero eigenvalues is interpreted as the topological quantum number.
The chiral Gaussian Unitary ensemble (chGUE) with Nf flavors is defined as the ensemble
of matrices D with matrix elements C distributed according to the Gaussian probability
distribution
P (C) detNf (D + m)eN Tr CC .
(5)
Here, for simplicity we have taken all quark masses equal to m. The probability distribution
P (C) has the unitary invariance
C U CV 1 ,
(6)
where U and V are unitary matrices. Since an arbitrary complex matrix can always be
brought to diagonal form by this transformation this invariance allows us to factorize the
probability distribution in a product over the eigenvalues of C and the unitary matrices that
diagonalize C.
The generalization of the model of Moshe et al. to the chiral ensembles is immediate.
The interpolating model is defined by the partition function
Z
Z
2
2
Nf
( 2 /4h) Tr D D
(7)
dU e( hN /4) Tr[D,U ][D,U ] ,
Z = dC det (D + m)e
where U has the chiral block structure

U 0
U=
,
0 V
(8)
with U an n n unitary matrix and V an (n + ) (n + ) unitary matrix. If and h

are constants as in (7) we will refer to this model as the critical chiral unitary ensemble.
As we will see below, in order to make contact with the Thouless energy in the partially
quenched effective partition function we have to scale h with an additional factor 1/ N .

The unitary invariance of this partition function follows from the invariance of the Haar
measure. In comparison to [48], an additional factor 2 / h has been included in the
probability distribution of the matrix elements. As we will see below, this will guarantee
that in the thermodynamic limit the spectral density is h-independent to leading order in h.
We will also find that the partition function is normalized such that the represents the
chiral condensate by means of the BanksCasher relation = (0)/N (with (0) the
spectral density around = 0).
Decomposing into the blocks of D and U the partition function can be written as
Z
2 2
2
Z = mNf dC detNf [C C + m2 ]e(1+2h N )( /2h) Tr CC

Z
2
2
1
dU dV e hN Re Tr U CV C .
(9)
The arbitrary complex matrix C can be decomposed according to
C = U1 U2 ,
(10)
672
and the integral over C can be expressed as an integral over the eigenvalues k and the
unitary matrices U1 and U2 . Up to a irrelevant constant, the Jacobian of this transformation
is given by
Y 2+1
k ,
(11)
J () = 2 2i
k
where

Y 2
k 2l
2i =
(12)
k<l
is the Vandermonde determinant. Because of the unitary invariance, the U1 and U2

dependence in the second exponent can be absorbed in a redefinition of U and V , and the
integrations over U1 and U2 just result in an overall constant. Remarkably, the integral over
U and V in (9) is an ItzyksonZuber type integral which is known analytically [5658]
Z
detk,l |I (zk l )|
1
,
(13)
dU dV ez Re Tr U V = C1
Q
2 ({2i }) nk=1 2
k
where C1 is a normalization constant. The Vandermonde determinants cancel resulting in
the partition function
Z Y
n
n
Y
N
dk
2k + m2 f
Z = m
k=1
k=1

1
1
2 2
2
2
2
det(k l ) 2 e( 2 +h N )( /2h)(k +l ) I (h 2 N 2 k l ).
k,l
(14)
The instanton-liquid Dirac spectra that will be described by this model were obtained in
the quenched approximation and for zero total topological charge. Therefore we will not
attempt to solve this model for arbitrary Nf but instead focus on the technically simpler
case of Nf = 0. Since the topological charge does not give rise to additional complications
we will consider the case of arbitrary . Below we will thus analyze the joint probability
distribution

1
1
2 2
2
2
2
(15)
(1 , . . . , n ) det(k l ) 2 e( 2 +h N )( /2h)(k +l ) I (h 2 N 2 k l ).
k,l
3. Fermi-gas representation of the interpolating chiral Random Matrix model

In this section we rewrite the joint probability distribution (15) in terms of the fermionic
n-particle matrix element
(x1 , . . . , xn ) Chx1 xn |eH |x1 xn i,
where H is the separable Hamiltonian

X
4 2 1
2 2
+
x
k2 +
H=
k ,
4xk2
k
(16)
(17)
and C is an irrelevant constant. The eigenfunctions of the single particle Hamiltonian are
known in terms of Laguerre polynomials. Specifically,

4 2 1
2 2
+
x
x2 +
n = n n ,
4x 2
673
(18)
has the solutions

(19)
n = (4n + 2 + 2),
1/4
1
2
2
ex /2 (x )+ 2 Ln (x 2 ).
(20)
n =
hn
The normalization factor hn = (n + )!/n! ensures that the n are normalized to unity
Z
n (x)n (x) dx = 1.
(21)
Using completeness, the many-particle matrix element can be written as

X
n (xi )en n (xj ).
hx1 xn |eH |x1 xn i = det
ij
(22)
The sum over n can be performed analytically using the identity

X
2 xyz
x+y
Ln (x)Ln (y)zn (xyz)/2
exp z
=
I
.
hn
1z
1z
1z
(23)
n=0
With z = exp(4) this results in

hx1 xn |eH |x1 xn i

xi xj

cosh 2 2
2

exp
xi xj .
= det
xi + xj I
ij sinh 2
2 sinh 2
sinh 2
(24)
Comparing this expression with (15) we find that

=
4h2 N 2 + 1
cosh 2 = 1 +
1
2h2 N 2
2
,
2h
(25)
(26)
The joint probability distribution of the eigenvalues is thus given by an n-particle

diagonal matrix element of the density operator. The average spectral density is equal to
the average particle density. It is obtained by integrating over the positions of all particles
except one. The integral can be performed by rewriting the matrix elements (22) in terms
of a sum over permutations and 0 ,
1 X X
() ( 0 )(1) (x1 ) (n) (xn )
n! <<
0
1
k k
0 (1) (x1 ) 0 (n) (xn ),
(27)
where () is the sign of the permutation. Performing the integrations over x2 , . . . , xn , by

orthogonality we find that the only nonzero contribution is for = 0 . Then the remaining
sum over (1 ) is just a sum over 1 , . . . , n . For the canonical ensemble we thus find the
one-particle density,
674
h(x)i
X
k

P
X X
1
(x xk ) =
|i (x)|2e k k .
Zn <<
1
(28)
In an occupation number representation this can be rewritten as

P
1 X X
ni |i (x)|2 e k ni k ,
h(x)i =
Zn P
(29)
i ni =n i
where the sum is over all ni {0, 1} subject to the condition given below the summation
sign and the canonical partition function is given by
P
X
e i ni i .
(30)
Zn =
P
i ni =n
For completeness we give the following exact expressions for the canonical partition
function [59,60]
e2n 2n
,
(1 e4 )(1 e8 ) (1 e4n )
and the one-particle density
2
Zn =
h(x)i =
|k (x)|2
(31)
N
X
Y
1
(1)r e4(krnr+ 2 r(r+1))
r=0
They have been obtained by writing the constraint

Z
P
X
1
ni =
d ei(N i ni ) ,
N
2

1 e4p .
(32)
p=Nr
P
i
ni = n as
(33)
and summing the geometric series after using the explicit expression (20) for the n .
Instead of working with the canonical ensemble we eliminate the constraint on the sum
over the ni by working with the grand canonical ensemble. The grand canonical partition
function is defined by
Y
X

z n Zn =
1 + ek + ,
(34)
Z=
n
z = e
where
given by
is the fugacity. The one particle density in the grand canonical ensemble is
1X n
z Zn (x)
Z n
P
1 X X
ni |i (x)|2 e k nk (k ) ,
=
Z
h(x)i =
(35)
ni {0,1} i
which can be evaluated to be

X
1
|i (x)|2
.
h(x)i =
(
1 + e i )
(36)
The fugacity z = e is determined by the condition that the total number of particles is
equal to n, i.e.,
1 + e(i )
= n.
675
(37)
The connected two-particle correlation function is defined by

X
h(x xk )(y yl )i h(x)ih(y)i.
R2 (x, y) =
(38)
k6=l
It can be obtained from the n-particle matrix element (16) by integration over all
coordinates (eigenvalues) except x1 and x2 . The remaining sum over (1) and (2) can
be rewritten as
P
X X

i (x1 )j (x2 ) i (x1 )j (x2 ) j (x1 )i (x2 ) e k k .
(39)
1 <<N i6=j
The diagonal term, i = j , is zero allowing us to include it in the summation. The first term
can then be identified as the square of the average particle density. It is the disconnected
contribution to the two-particle correlation function. The connected part of the two-particle
distribution function can thus be written as
P
X X
1
i (x1 )j (x2 )j (x1 )i (x2 )e i i .
(40)
R2 (x1 , x2 ) =
Zn <<
n
i,j
This correlation function does not include the contributions from the self-correlations of the
eigenvalues. After all, our starting point was the joint distribution of different eigenvalues.
In an occupation number representation R2 (x1 , x2 ) simplifies to
#2
"
P
X
X
1
ni i (x1 )i (x2 ) e k nk l .
(41)
R2 (x1 , x2 ) =
Zn n +n +=n
1
Using this representation, the two-particle density in the grand canonical ensemble is found
to be

2
X

1

i (x)i (y)
(42)
R2 (x, y) =
.

1 + e(i )
i
Using similar manipulations as for the canonical partition function and the corresponding one-particle density it is possible to simplify the exact analytical expression for the
connected two-particle correlation function in the canonical ensemble,
R2 (x, y) =
|k (x)l (y)|2
k,l
(1)r e4(kr+lsn(r+s)+ 2 (r+s)(r+s+1))
r,s=0
N
Y
(1 e4p ).
(43)
p=Nrs
This result can be used to compare the two ensembles but we will not address this question
in this article.
676
4. Semiclassical calculation
In this section we calculate the microscopic spectral density and the two-point
correlation function using semiclassical methods starting from the expressions for the
grand canonical ensemble derived in previous section. We are thus interested in the
region around x = 0. Because of the hard edge at x = 0 we cannot simply do a WKB
approximation by replacing the wave functions by plane waves but instead have to use
Bessel functions. This follows immediately from the wave equation for x 0,

4 2 1
x J (kx) = k 2 x J (kx).
(44)
x2 +
2
4x
Alternatively, one can exploit the asymptotic relation between Laguerre polynomials and
Bessel functions.
4.1. The microscopic spectral density
R
Taking into account the normalization of the Bessel functions, 0 k dk xx 0 J (kx)
J (kx 0 ) = (x x 0 ), to fix the constants in the integration measure, we arrive at the
following expression for the single particle density
Z
h(x)i =
kx dk
0
J2 (kx)
1 + ek
2 +2 x 2
(45)
R
The chemical potential is determined by the condition dx (x) = n. Since this integral
is over all x, the use of the semiclassical expressions for the wave functions is not justified
but instead we have to rely on the exact wave functions. In the limit, , the sum over
i in (37) can be replaced be an integral which can be performed analytically resulting in
1
.
(46)
e4n 1
In the limit n 1 the semiclassical expression for the spectral density is thus given by
e(2+2) =
Z
h(x)i =
kx dk
0
J2 (kx)
1 + ek
2 +2 x 2 (4n22)
(47)
In this limit the semiclassical expression (45) leads to the correct value of the chemical
potential. Below we will show for n 1 and finite x (in units of the average level
spacing) the term 2 x 2 can be neglected relative to 4n.
An estimate for the average spectral density near zero but many level spacings away
from x = 0 is obtained by using the leading order asymptotic expansion of the Bessel
functions and calculating the integral in (47) in the limit of a degenerate Fermi-gas. This
results in
k
(48)
= ,
where k 4n is the radius of the Fermi-sphere. At fixed h in the limit n 1 we have
677
1
.
(49)
2hn
Using these results and invoking the BanksCasher formula the parameter can be
identified as the chiral condensate,
2 n and

= .
(50)
2n
We will now show that our approximations are self-consistent. The condition that we are
close to the degenerate Fermi-gas can be written as k/k 1. We thus have to impose the
requirement that
lim
1
h
k
= 1.
2
4
k
2 k
(51)
The conditions kx 1 and 2 x 2 4n can be combined into

1
x 2 4n,
(52)
4n
or, in units of the average level spacing, x = u/,
the range of validity of the above
asymptotic results is given by
u2 2
1

4n.
4n
4n
(53)
Because of the second inequality it is justified to neglect the term 2 x 2 which we will do
in the remainder of this section.
By partial integration the expression (47) for the spectral density can be rewritten as
Z
h(x)i =
dk
xk 3 [J2 (kx) J+1 (kx)J1 (kx)]

4 cosh2 2 (k 2 4n)
(54)
Using the BanksCasher formula one finds that the chiral condensate depends on h. The
leading order correction is given by
2 2
(h)
=1
h + .
(55)
96
The corresponding spectral density will be denoted by (h)
2n(h)/ . The microscopic spectral density is then given by

u
1
s (u) = lim
n 2n(h)
2n(h)
Z
k 2 s0 (ku/(h))
dk
,
(56)
=
2
(h)
h cosh2 (k h1)
0
where s0 is the microscopic spectral density for the chGUE [61,62]

u
s0 (u) = J2 (u) J+1 (u)J1 (u) .
2
(57)
678
The interpretation is clear. The oscillations in the microscopic spectral density due to
the Bessel functions are smeared out over a distance hu by the integration over k. The
oscillations are thus visible up to a distance of u 1/ h.
In the limit of small h the main contribution to the integral comes from the region around
the Fermi-surface. In this limit we can derive an approximate formula correct to order h2
at fixed uh. To this end we change integration variables in (56) according to k = 1 + ht
neglecting terms that are subleading in h. This results in
Z
s (u) =
dt
s0 (u + hut)
cosh2 (2t)
(58)
Next we Taylor expand the microscopic spectral density as follows

s0 (u + hut) = s0 (u)

1
+ hut J2 + J+1 J1
2

f1
1 (hut)2
4J J1 +
+
2 2!
u

3
g1 g2
1 (hut)
2
2
4 J1
+ 2
J+1
+
+
2 3!
u
u

4
h1 h2 h3
1 (hut)
2
4 J J1 +
+ 2+ 3 ,
+
2 4!
u
u
u
(59)
where the terms that will be neglected are denoted by fi , gi , hi , etc. At small values of u
these terms are of order h2 u whereas for large u they are suppressed by order 1/u2 (notice
the factor u in (57)). By inspection one easily finds that the neglected terms are at most of
order h2 independent of the value of u. The leading order terms can be easily resummed to
1
s0 (u + hut) = s0 (u) J (u)J1 (u)[cos(2hut) 1] + O(h2 )
2
+ terms odd in t.
The integral over t in (58) can be performed analytically resulting in

1
hu
1 + O(h2 ).
s (u) = s0 (u) J (u)J1 (u)
2
2 sinh(hu/2)
(60)
(61)
In Fig. 1 we compare the exact expression for the microscopic spectral density (56) to this
approximate formula. We observe that even for a value of h as large as h = 0.3 the two
results are very close.
4.2. Two-point function
In this subsection we derive a semiclassical expression for the two-point correlation
function. To this end the wave functions in the expression (42) for the connected two-point
correlation function are replaced by Bessel functions,
679
Fig. 1. The microscopic spectral density s (u) versus u for a value of h = 0.3. The exact result
(full curve) and the approximate result (dashed curve) nearly coincide.
Z
2

xy J (kx)J (ky)

R2 (x, y) = k dk
.
2

1 + ek 4n
(62)
By partial integration with respect to k the correlation function can be expressed as

2

Z
K(x k,
y k)

2
k dk
R2 (x, y) =
,
h
cosh2 ((k 2 1)/ h)
(63)
where K(x, y) is the two-point kernel for the chGUE given by [61,62]
K(x, y) =
xJ+1 (x)J (y) yJ (x)J+1 (y)

xy
.
x2 y2
To order h2 the correlation function can be simplified to

2

Z

K(x (1
+ ht), y (1
+ ht))

dt
R2 (x, y) =
+ O(h2 ).

cosh2 2t
(64)
(65)
In the same way as for the one-point function we now will derive an approximate formula
for the two-point function correct to order h2 at fixed value of uh. This can be done
conveniently by using the following summation formula
J+1 (x + xh)J (y + yh) J (x + xh)J+1 (y + yh)
[J+1 (x)J (y) J (x)J+1 (y)] cos[(x y)h]
+ [J (x)J (y) J+1 (x)J+1 (y)] sin[(x y)h].
(66)
This formula has been derived by means of a Taylor expansion and a subsequent
resummation employing the following approximate derivative formulas
680

k2n [J+1 (kx)J (ky) J (kx)J+1 (ky)]k=1
(1)n (x y)2n [J+1 (x)J (y) J (x)J+1 (y)],
(67)
and

k2n1 [J+1 (kx)J (ky) J+1 (kx)J (ky)]k=1
(1)n+1 (x y)2n1 [J (x)J (y) J+1 (x)J+1 (y)].
(68)
They have been obtained by means of recursion relations for Bessel functions neglecting
terms that are suppressed by order 1/x or 1/y. One can easily show that the combined
powers of x and y in the prefactor is always larger than the combined powers of x and y
that have been neglected. Since only even terms in t contribute to the integral in (65) our
final result for the two-point function, correct to order h2 , is given by
3

(y )
+ J (x )J
+1 (y )
2 h xy J+1 (x )J
R2 (x, y) =
8
sinh((x + y) 2 h/4)

3 h xy J+1 (x )J
(y )
J (x )J
+1 (y )
2
(69)
+
.
8
sinh((x y) 2 h/4)
In the limit x, y |x y| the analytical result for the two-point function of the model of
Moshe et al. [48] is recovered from the leading order asymptotic expansion of the Bessel
functions. For unitary invariant ensembles, it can be shown that the result of [48] for critical
statistics and WignerDyson statistics are the only two possibilities [63]. At this moment it
is not clear whether this argument can be extended to the chiral unitary ensembles as well.
The number variance of the eigenvalues near x = 0 is obtained by integrating the twopoint correlation function including the self-correlations
L/
Z(h)
(L) =
2
L/
Z(h)

dy (x y)h(x)i + R2 (x, y) .
dx
0
(70)
We study its asymptotic behavior in the limit L . Starting from the expressions
(54) and (62), 2 (L) can be simplified by means of an orthogonality relation for Bessel
functions. To leading order in h we find
L/
Z(h)
(L) =
2
x dx
0
Z
k dk
0
h
L
= L.
k 4
J2 (kx)
4 cosh2 2 (k 2 4n)
(71)
The same asymptotic result can be derived from the analytical result (69). (In this case the
average spectral density does not depend on h (see (58)).) Such linear term, first proposed
in [36,37], is believed to be characteristic for universal critical statistics [43] valid at the
mobility edge and has been related to the multifractality of the wave functions [44,45,64
67].
681
Fig. 2. The number variance 2 (L) versus L for a value of h = 0.3. The exact result (full curve) and
the approximate result (dashed curve) nearly coincide.
In Fig. 2 we compare the number variance derived from the approximate analytical result
(69) and from the exact result (62). Clearly, even for a value of h as large as 0.3 the two
curve are barely distinguishable.
The asymptotic linear behavior of the number variance seems to be contradicted by the
sum rule
Z

dy (x y)h(x)i + R2 (x, y) = 0.
(72)
The resolution of this paradox [49,68] is probably best illustrated by considering the
Poisson ensemble for n uncorrelated eigenvalues with average spacing .
To satisfy
instead of zero for uncorrelated eigenvalues
the sum-rule we have that R2 (x, y) = /n
resulting in the number variance 2 (L) = L L2 /n. We conclude that an asymptotic
linear behavior is possible if the thermodynamic limit is taken before the limit L .
To make contact with the partially quenched effective partition function, for which the
number of subsequent eigenvalues around zero that are correlated according to the chGUE
scales as n [47], we have to scale h as h h/ n. In this limit microscopic universality

[1,812] is recovered for the interpolating chiral unitary ensemble.
5. Comparison with instanton simulations

Spectral correlations have been studied in great detail for both lattice QCD simulations
[4,15,16,18,2022] and instanton-liquid simulations [5] (see [2] for a recent review). In
lattice QCD they were studied by means of the disconnected scalar susceptibility [23],
682
Fig. 3. The number variance 2 (L) versus L for Dirac eigenvalues of instanton gauge field
configurations (points) and our interpolating random matrix model. The total number of instantons
and the parameter h of our model are given in the label of the figure.
and complete agreement with partially quenched chiral perturbation theory was found. In
particular, it was shown that
the number of subsequent eigenvalues around zero described
by chRMT scales as F 2 V . In instanton simulations a weaker volume dependence of
the number of such eigenvalues was observed suggesting an approach to a critical point
similar to a localization transition. Indeed the multifractality index of the fermionic wave
functions was found to be nonzero. We thus compare the instanton data with the model
in previous section at fixed value of the parameter h. Results for the number variance,
2 (L) versus L are shown in Fig. 3. The closed and open circles represent results [5]
for the eigenvalues of the Dirac operator with field configurations given by an ensemble
of instantons and an equal number of anti-instantons with a total density of 1 fm4 . The
total number of (anti)-instantons is given in the label of the figure. In the same figure we
show the result for the chGUE (dotted curve) and results for the model (62) for h = 0.18
(full curve) and h = 0.23 (dashed curve). We observe that both the slope and the range
of agreement with the chGUE curve only shows a week volume dependence. Outside
this domain the data show a linear L-dependence. Both features are nicely reproduced
by the critical random matrix model. The slightly positive curvature of the instanton data
might be a remnant of the L2 dependence predicted for the AltshulerShklovsky domain
[36,37].
6. Conclusions
We have analyzed a chiral random matrix model that interpolates between the chGUE
and the chiral Poisson ensemble. This model is a generalization of a model originally
683
proposed by Moshe et al. [48]. It has been mapped onto a gas of non-interacting fermions
and was solved by means of statistical mechanics methods. To leading order in the
deviation from the chGUE we have obtained compact analytical expressions for the
microscopic spectral density and the two-point level correlation function. We have shown
that this critical chiral random matrix model provides a good description of the level
correlations of Dirac eigenvalues for gauge field configurations given by a liquid of
instantons.
The number variance of the critical chiral random matrix model shows a linear
L-dependence for a large L (in units of the average level spacing) whereas it coincides
with the chGUE result for small values of L. The characteristic feature is that the transition
point between these two domains is stable in the thermodynamic limit. This situation
is very different for a non-linear -model description of disordered systems where this
transition point or the Thouless energy is determined by the competition between the mass
term and the kinetic term. In that case one finds the scaling behavior Ec D/L2s with
D the diffusion constant and Ls the linear size of the sample. The theoretical reason
for a scale independent dimensionless conductance (i.e., the Thouless energy in units
of the average level spacing) is that the localization length diverges at a critical value
of the disorder. In the approach to this limit the diffusion constant has to become scale
dependent. If Ec , in units of the average level spacing, becomes scale independent the
diffusion constant has to be scale dependent leading to a multi-fractal scaling of the wave
functions.
The weak volume dependence and the linear number variance observed in correlations of eigenvalues of the QCD Dirac operator with instanton liquid gauge field
configurations suggests that we are dealing with a critical system close to a localization transition. Indeed the same numerical simulations suggest a small nonzero
multifractality index of the wave-functions. On the other hand, the dimensionless
conductance found in lattice QCD simulations scales according to our expectations
from chiral perturbation theory. At this moment we do not have a good explanation for this discrepancy. It could simply be that the expected scaling behavior is
only recovered for much larger volumes in instanton simulations. Indeed, a very slow
approach to the thermodynamic limit has been found for other quantities such a
quenched chiral logarithms. Clearly, more work has to be done to resolve this issue.
Acknowledgements
This work was partially supported by the US DOE grant DE-FG-88ER40388. One of us
(A.M.G.) was supported by laCaixa Fellowship Program. J.J.M.V. thanks the Institute
for Nuclear Theory at the University of Washington for its hospitality and partial support
during the completion of this work. Dominique Toublan is thanked for a critical reading of
the manuscript.
684
References
[1] E.V. Shuryak, J.J.M. Verbaarschot, Nucl. Phys. A 560 (1993) 306.
[2] J.J.M. Verbaarschot, T. Wettig, Ann. Rev. Nucl. Part. Sci. (2000), hep-ph/0003017.
[3] J.J.M. Verbaarschot, Lectures given at APCTPRCNP Joint International School on Physics of
Hadrons and QCD, Osaka, Japan, 1998; 1998 YITP Workshop on QCD and Hadron Physics,
Kyoto, Japan, 1998, in: H. Yabu et al. (Eds.), Physics of Hadrons and QCD, World Scientific,
Singapore, 1998.
[4] J.J.M. Verbaarschot, Phys. Lett. B 368 (1996) 137.
[5] J.C. Osborn, J.J.M. Verbaarschot, Phys. Rev. Lett. 81 (1998) 268; Nucl. Phys. B 525 (1998)
738.
[6] J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 540 (1999) 317.
[7] P.H. Damgaard, J.C. Osborn, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 547 (1999) 305.
[8] E. Brzin, S. Hikami, A. Zee, Nucl. Phys. B 464 (1996) 411.
[9] G. Akemann, P.H. Damgaard, U. Magnea, S. Nishigaki, Nucl. Phys. B 487 [FS] (1997) 721.
[10] A.D. Jackson, M.K. Sener, J.J.M. Verbaarschot, Nucl. Phys. B 479 (1996) 707.
[11] T. Guhr, T. Wettig, Nucl. Phys. B 506 (1997) 589.
[12] A.D. Jackson, M.K. Sener, J.J.M. Verbaarschot, Nucl. Phys. B 506 (1997) 612.
[13] M.A. Halasz, J.J.M. Verbaarschot, Phys. Rev. Lett. 74 (1995) 3920.
[14] M.A. Halasz, T. Kalkreuter, J.J.M. Verbaarschot, Nucl. Phys. B Proc. Suppl. 53 (1997) 266.
[15] M.E. Berbenni-Bitsch, S. Meyer, A. Schfer, J.J.M. Verbaarschot, T. Wettig, Phys. Rev. Lett. 80
(1998) 1146.
[16] M.E. Berbenni-Bitsch, M. Gockeler, 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.
[17] 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.
[18] P.H. Damgaard, U.M. Heller, A. Krasnitz, Phys. Lett. B 445 (1999) 366.
[19] M. Gockeler, H. Hehl, P.E.L. Rakow, A. Schafer, T. Wettig, Phys. Rev. D 59 (1999) 094503.
[20] R.G. Edwards, U.M. Heller, J. Kiskis, R. Narayanan, Phys. Rev. Lett. 82 (1999) 4188.
[21] B.A. Berg, H. Markum, R. Pullirsch, Phys. Rev. D 59 (1999) 097504.
[22] F. Farchioni, I. Hip, C.B. Lang, Phys. Lett. B 471 (1999) 58.
[23] M.E. Berbenni-Bitsch, M. Gckeler, H. Hehl, S. Meyer, P.E.L. Rakow, A. Schfer, T. Wettig,
Phys. Lett. B 466 (1999) 293; Nucl. Phys. B Proc. Suppl. 83 (2000) 974.
[24] R. Gade, F. Wegner, Nucl. Phys. B 360 (1991) 213.
[25] R. Gade, Nucl. Phys. B 398 (1993) 499.
[26] A. Altland, B.D. Simons, Nucl. Phys. B 562 (1999) 445.
[27] K. Takahashi, S. Iida, Nucl. Phys. B 573 (2000) 685.
[28] T. Guhr, T. Wilke, H.A. Weidenmller, hep-th/9910107.
[29] J. Gasser, H. Leutwyler, Phys. Lett. B 188 (1987) 477; Nucl. Phys. B 307 (1988) 763.
[30] H. Leutwyler, A. Smilga, Phys. Rev. D 46 (1992) 5607.
[31] C. Bernard, M. Golterman, Phys. Rev. D 46 (1992) 853.
[32] S. Sharpe, Phys. Rev. D 46 (1992) 3146.
[33] C. Bernard, M. Golterman, Phys. Rev. D 49 (1994) 486, hep-lat/9311070.
[34] M.F.L. Golterman, K.C. Leung, Phys. Rev. D 57 (1998) 5703.
[35] M.F.L. Golterman, Acta Phys. Pol. B 25 (1994).
[36] B.L. Altshuler, B.I. Shklovskii, Zh. Eksp. Teor. Fiz. 91 (1986) 343.
[37] B.L. Altshuler, I.Kh. Zharekeshev, S.A. Kotochigova, B.I. Shklovskii, Zh. Eksp. Teor. Fiz. 94
(1988) 343.
[38] T. Guhr, A. Mller-Groeling, H.A. Weidenmller, Phys. Rep. 299 (1998) 189.
[39] G. Montambaux, in: E. Giacobino, S. Reynaud, J. Zinn-Justin (Eds.), Quantum Fluctuations,
Les Houches, Session LXIII, Elsevier Science, 1995, cond-mat/9602071.
[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]
685
D. Braun, G. Montambaux, Phys. Rev. B 52 (1995) 13903.

T. Banks, A. Casher, Nucl. Phys. B 169 (1980) 103.
A. Aronov, V. Kravtsov, I. Lerner, Phys. Rev. Lett. 74 (1995) 1174.
B.I. Shklovskii, B. Shapiro, B.R. Sears, P. Lambrianides, H.B. Shore, Phys. Rev. B 47 (1993)
11487.
A. Mirlin, Phys. Rep. 326 (2000) 259.
A.D. Mirlin, F. Evers, Phys. Rev. Lett. 84 (2000) 3690.
M. Janssen, Phys. Rep. 295 (1998) 1.
I.Kh. Zharekshev, B. Kramer, Ann. Phys. (Leipzig) 7 (1998) 442.
M. Moshe, H. Neuberger, B. Shapiro, Phys. Rev. Lett. 73 (1994) 1497.
V. Kravtsov, K. Muttalib, Phys. Rev. Lett. 79 (1997) 1913.
V.A. Kazakov, A.A. Migdal, Nucl. Phys. B 397 (1993) 214.
T.H. Seligman, J.J.M. Verbaarschot, M.R. Zirnbauer, J. Phys. A 18 (1985) 2751.
A.D. Mirlin, Y.V. Fyodorov, F.M. Dittes, J. Quezada, T.H. Seligman, Phys. Rev. E 54 (1996)
3221.
K.A. Muttalib, Y. Chen, M.E.H. Ismail, V.N. Nicopoulos, Phys. Rev. Lett. 71 (1993) 471.
Y. Chen, M.E.H. Ismail, K.A. Muttalib, J. Phys. Cond. Mat. 4 (1992) L417; J. Phys. Cond.
Mat. 5 (1993) 177.
S.M. Nishigaki, Phys. Rev. E 58 (1998) R6915; Phys. Rev. E 59 (1999) 2853.
F.A. Berezin, F.I. Karpelevich, Doklady Akad. Nauk SSSR 118 (1958) 9.
T. Guhr, T. Wettig, J. Math. Phys. 37 (1996) 6395.
A.D. Jackson, M.K. Sener,
J.J.M. Verbaarschot, Phys. Lett. B 387 (1996) 355.

M. Caselle, A. DAdda, S. Panzeri, Phys. Lett. B 293 (1992) 161.
D. Boulatov, V.A. Kazakov, Int. J. Mod. Phys. A 8 (1993) 809.
J. Verbaarschot, I. Zahed, Phys. Rev. Lett. 70 (1993) 3852.
J. Verbaarschot, Phys. Rev. Lett. 72 (1994) 2531; Phys. Lett. B 329 (1994) 351.
E. Kanzieper, Private communication.
B. Huckenstein, L. Schweitzer, Phys. Rev. Lett. 72 (1994) 713.
J. Chalker, V. Kravtsov, I. Lerner, JETP Lett. 64 (1996) 836.
J. Chalker, I. Lerner, R. Smith, Phys. Rev. Lett. 77 (1996) 554.
V.E. Kravtsov, A.M. Tsvelik, cond-mat/0002120.
V. Kravtsov, in: Proc. of Correlated Fermions and Transport in Mesoscopic Systems, Moriond
Conference, Les Arcs, 1996, cond-mat/9603166.

Magnetic impurities in a correlated electron system

with spin-triplet pairing
A.A. Zvyagin a,b , P. Schlottmann c,
a Max Planck Institut fr Physik komplexer Systeme, Nthnitzer Str. 38, D-001187 Dresden, Germany
b B. I. Verkin Institute for Low Temperature Physics and Engineering, Ukrainian National Academy of Sciences,
47 Lenin Avenue, Kharkov, 61164, Ukraine

c Department of Physics, Florida State University, Tallahassee, FL 32306, USA
Received 4 April 2000; revised 28 June 2000; accepted 12 July 2000
Abstract
We consider an exactly solvable two-band model of electrons moving in one dimension and
interacting with a -function spin-exchange potential. The interaction leads to the formation of spintriplet bound states of the Cooper type (hard-core bosons that do not condensate and exist at all
temperatures). Without destroying the integrability we introduce a finite concentration of mixedvalent impurities with two magnetic configurations of spin S and S 12 , respectively, which hybridize
with the conduction states of the host. We derive the Bethe ansatz equations diagonalizing the
correlated host with impurities and discuss the ground state properties as a function of magnetic
field, the crystalline field splitting of the bands, the concentration of the impurities and the Kondo
exchange coupling. In the zero-field ground state the spin-triplet states are effectively free bosons and
a threshold band splitting is required to break up a triplet pair. The gap of the unpaired electron
excitations is gradually reduced by the band splitting and closes at . For a splitting larger than
a fraction of the electrons is spin-polarized. The impurities are antiferromagnetically correlated with
each other and the effect of a magnetic field is to suppress these correlations as well as to gradually
align the spins of the spin-triplets with the field. Spin- 12 impurities always enhance the gap , while
with impurities of higher spin the gap decreases if the coupling between the triplets and impurities is
weak (small Kondo temperature), but increases otherwise. For small TK the gap is closed at a critical
concentration xcr and the two electron bands have different populations. For x > xcr we obtain
a zero-magnetization phase with three coexisting components, namely, itinerant ferromagnetism
of unpaired electrons, antiferromagnetism of the impurities and superconducting fluctuations of
ferromagnetically correlated spin triplets. The critical exponents of correlation functions are also
briefly discussed in the conformal field theory limit. 2000 Elsevier Science B.V. All rights reserved.
PACS: 75.20.Hr; 71.27.+a; 75.30.Mb; 71.10.Li; 71.10.Pm; 74.20.Mn
Keywords: Magnetic impurities; Correlated electron systems; Triplet superconductivity; Bethe ansatz
E-mail address: schlottm@martech.fsu.edu (P. Schlottmann).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 5 2 - 1
A.A. Zvyagin, P. Schlottmann / Nuclear Physics B 586 [FS] (2000) 686710
687
1. Introduction
There has been increased interest in magnetic impurities embedded into correlated
electron hosts, in particular in view of experimental realizations in, e.g. standard BCSlike superconductors [1], high-Tc cuprates [2,3], heavy fermion alloys and compounds
(for a recent review see, e.g. [4]), and quantum wires and carbon nanotubes [5]. For
instance, the non-Fermi-liquid behavior in transport and thermodynamic properties of
cuprates and some heavy fermion compounds observed experimentally [24], has been
interpreted in terms of quantum critical points and the multichannel Kondo problem (for
a review see, e.g. [6] and references therein). Non-Fermi-liquid behavior in underdoped
high-Tc compounds appears to be related to the gap for low-lying excitations and chargeordered stripes, and has been a challenge for theorists for more than a decade.
On the other hand, unconventional superconductivity has been discovered in several
classes of compounds, such as quasi-one-dimensional organic conductors, heavy fermions,
high-Tc cuprates and ruthenates. Common to these systems is the formation of anomalous
(non-s-wave) Cooper pairs, probably due to mechanisms other than the usual phonon
mediated attraction. While in U-based heavy fermion compounds the order parameter is
complicated by the spinorbit coupling and appears to be spin-dependent [7], in Sr2 RuO4
and in (TMTSF)2 PF6 it is believed to be an odd-parity (p-wave) spin-triplet pairing [815]
(for some high-Tc compounds the gap does not seem to depend on the magnetic field,
see [16]). Spin-triplet pairing is otherwise realized in the superfluid phases of He3 .
In this paper we consider an integrable one-dimensional (1D) two-band correlated
electron model with triplet pairing as the host to mixed-valent magnetic impurities
and study it by means of Bethes ansatz. Exact solutions in 1D can be obtained by
nonperturbative methods, which are not applicable to higher-dimensional models. Thus
exact results in 1D may serve as a testing ground for approximative methods for
models representing more realistic situations. Interactions also manifest themselves most
dramatically in 1D, where they transform the electron gas into a Luttinger liquid.
The Bethe ansatz method has recently been used to investigate isolated magnetic
impurities in single band 1D correlated hosts [1720]. Suitable correlated lattice hosts
are the Hubbard chain, the supersymmetric tJ model and the SU(3)-invariant tJ
model embedding Anderson, Kondo and CoqblinSchrieffer impurities. The effects of the
interactions in the host on the screening of the impurity have also been investigated with
several other methods [2124], e.g. abelian and non-abelian bosonization, renormalization
groups, poor-mans scaling, boundary conformal field theory and numerical methods.
An impurity can be located on a link or at the open boundary of the chain. In the former
case the impurity is a pure elastic scatterer, i.e. no reflection wave is generated, while in
the latter case it is a backward scatterer, i.e. only a reflecting wave (but no transmission)
is generated. It has been shown that the low-energy magnetic properties of an impurity
located at the edge are very similar to those of an impurity situated on a link with periodic
boundary conditions [25]. Furthermore, models with impurities on a link can easily be
generalized to finite concentrations of impurities [2629].
688
Several kind of integrable 1D models of interacting electrons with a finite concentration

of impurities have been studied.
(i) In Refs. [26,27] magnetic impurities embedded into the Hubbard model with
attractive U were shown to gradually close the spin-gap of the Cooper-pair-like
bound states as a function of the concentration of impurities. The gap closes at
a critical concentration xcr and for x > xcr a phase with itinerant ferromagnetism
coexist with the Cooper pairs. The impurities in Refs. [26,27] are noninteracting,
i.e. they have a truly magnetic character, even in the ground state.
(ii) A finite concentration of mixed-valent impurities embedded into the supersymmetric and SU(3) invariant tJ models yield an impurity band, which arises from the
interactions among the impurities [28,29]. Hence, in addition to the Kondo effect
the spins compensate each other via antiferromagnetic correlations giving rise to
a singlet ground state.
(iii) In Ref. [30] we considered a two-band model with -function interaction giving
rise to spin-singlet Cooper-pair bound states with magnetic impurities. Again,
the impurities form a band and compensate each other antiferromagnetically.
The interference between superconducting and antiferromagnetic correlations can
be constructive or destructive. In the destructive interference regime a critical
concentration exists at which the gap closes. At a higher concentration itinerant
ferromagnetism is induced which coexists with ferromagnetism of the impurity
spins and the superconducting fluctuations.
In the present contribution we investigate the effect of impurities on electrons paired
in spin-triplets. The properties of the two-band host with -function exchange potential
giving rise to triplet pairing has been studied in Ref. [31]. Interacting magnetic impurities
are now embedded into this host without destroying the integrability. Depending on the
coupling of the impurities to the host the impurities may suppress the superconducting
fluctuations or enhance them. For S > 1, a large off-resonance shift of the impurities, and
weak coupling among the itinerant electrons, a critical concentration xcr is obtained at
which the gap (binding energy of the Cooper-pairs) is closed. For a strong coupling of
the impurities to the electrons the gap increases, i.e. the impurities do not break the pairs,
but the antiferromagnetic correlations among them rather enhance the superconducting
fluctuations.
The rest of the paper is organized as follows. In Section 2 the model is defined through
the scattering matrices between electrons and of the electrons off the impurity. The form
of the impurity Hamiltonian is given for a linearized dispersion of the conduction band.
The Bethe ansatz equations (BAE) diagonalizing the transfer matrix for the model are
introduced in Section 3. The ground state properties of an isolated impurity are discussed in
Section 4. In Section 5 we analyze the ground state, the low-temperature thermodynamics
and the asymptotic behavior of correlation functions for a finite concentration of impurities.
Concluding remarks follow in Section 6.
689
2. The model
As the host we consider an integrable model in 1D, consisting of electrons with spin
in two parabolic electron bands of equal mass labelled by m = 1, 2. The two bands
are separated by a crystalline field splitting D that lifts the degeneracy of the bands.
The electrons interact with each other via a contact (-function) exchange interaction of
strength c. The Hamiltonian for the two-band host is [3133]:
X
Z
(x) 2 /x 2 m (x)
m
Hhost = dx
m,
Z
c
+D
dx 0
X
m0 , 0
0
0
(x x 0 )m
(x)m
0 0 (x )m0 (x )m 0 (x)
(x)1 (x) 2
(x)2 (x)
1
(1)
(x) creates an electron with spin at the site x in the band m. The equal
where m
effective mass of the bands is a necessary condition for the integrability. The interaction
is attractive for two electrons forming a spin triplet and leads to paired electron bound
states. These spin-triplet pair bound states exist at all temperatures, while long-range order
is naturally absent in 1D. For small enough band (crystalline-field) splitting the unbound
electron states exhibit a gapped behavior [3133].
The two-electron wave function can be written as a product of (i) a coordinate wave
function referring to the positions and momenta of the electrons, (ii) a spin wave function,
and (iii) an orbital wave function involving the band labels. The global symmetry of the
wave function has to be antisymmetric under the exchange of the two particles. Hence,
if the spin and orbital parts have the same symmetry, the coordinate wave function is
antisymmetric and vanishes if x1 = x2 , so that the particles cannot interact. Interacting
electrons then necessarily form a spin triplet and orbital singlet or a spin singlet and orbital
triplet. The former situation corresponds to an attractive exchange interaction, while for
the latter the coupling is repulsive. The two-particle scattering matrix then factorizes into
a spin and a band part [3234]
b = k I + icP k Im icPm ,
R(k)
k + ic
k ic
(2)
where k is the momentum transfer, and Im, and Pbm, denote the identity matrices and
permutation operators in the band and spin channels, respectively. When applied to a triplet
wave function (either in the spin or the orbital sector) the corresponding scattering matrix
yields one, while if it acts on a singlet it gives rise to a phase shift. For the case of spin
and orbital singlets the two phase factors cancel and there is no effective interaction. Since
the scattering matrix for each channel satisfies the YangBaxter relation, their product also
does and model (1) is integrable [3234].
To introduce the integrable impurity we proceed along the lines pioneered in Ref. [35]
(see also [36,37]) for spin chains and extended to correlated electron models in
690
Refs. [1720]. To preserve the integrability the impurity-host two-particle scattering matrix
has to satisfy the YangBaxter relation with the scattering matrix of the host, Eq. (2).
Here we consider the two-parametric scattering matrix, proposed in Refs. [38,39], which
is diagonal in the band sector, and in spin space it has the form [40]

0
0 0
0
0
b
SMM
0 (k) = 0 MM 0 + (M |M + ) M |M +
0

(3)
ic(2S + 1)/ k ic S + 12 PbMM
0,
b 0 0 =
where is the off-resonance shift and S > 12 is the spin of the impurity. Here P
MM
0 MM 0 + 0 M 0 M+2 , |M| 6 S and the ClebschGordan coefficient

(4)
(M |M + ) = SM; 12 S 12 S 12 (M + )
preserves the spin rotational invariance. The impurity can temporarily absorb the spin of
one conduction electron to form an effective spin S 0 = S 12 , i.e. the impurity is mixedvalent and its wave function is a linear superposition of two different spin configurations.
For a linearized dispersion of the itinerant electrons about the Fermi level, the impurity
Hamiltonian for a diluted alloy of Ni well-separated impurities situated at the positions xj
has the form [1719]
Himp =
Ni
X
X 0 0
0 0
S M S M
|j | /vF
j
j
j =1
Mj0
p
+ c(2S + 1)/vF

(xj )|SMj i S 0 Mj0 + H.c. , (5)
Mj |Mj0 m
m,,Mj ,Mj0
where the bras and kets denote the impurity states, vF is the Fermi velocity of the
conduction electrons, and is the chemical potential. Each impurity has to satisfy the
completeness condition
X
X

S 0 M 0 S 0 M 0 +
|SMihSM| = 1.
(6)
M0
In principle, it is possible to incorporate different impurities into the host, e.g. of different
spins S or a distribution of j . However, in this paper we limit ourselves to two situations,
namely an isolated impurity and a finite concentration of identical impurities. In Eq. (5)
j represents the energy of the charged impurity state of spin S 0 , i.e. the energy difference
between the two configurations. On the other hand, j also measures the off-resonance
shift of the impurity with respect to the center of the host bands. The latter interpretation
becomes evident in the context of the Bethe ansatz equations.
The monodromy matrix is the product of the host two-particle electronelectron
scattering matrices and electronimpurity scattering matrices. Because of the YangBaxter
relations this product does not depend on the relative positions of the impurities, and
gives rise to a large degeneracy of space configurations of the impurities. This locality
structure is an artifact of the integrability. The transfer matrix is the trace over the auxiliary
subspace of the monodromy. Since the monodromy matrices also satisfy the Yang
Baxter relation, transfer matrices of different spectral parameters commute and can all be
691
diagonalized with a common basis of eigenstates. This establishes the exact integrability
of the problem [1719].
For integrable models the Hamiltonian can be obtained as the logarithmic derivative
of the transfer matrix. Unfortunately, this procedure is very cumbersome and its outcome
depends on the relative positions of the impurities. A simple situation corresponds to the
absence of mobile electrons, so that the system only consists of impurities of spin S
and the interaction Hamiltonian takes the form of the BabujianTakhtajan model (the
integrable generalization of the Heisenberg chain with SU(2)-symmetric coupling between
spins S) [41,42].
3. Bethe ansatz equations

Bethes ansatz parameterizes the eigenfunctions and eigenvalues in terms of three sets
of rapidities, namely:
e
(i) charge rapidities {kj }N
j =1 , where Ne is the number of electrons,
M is the number of electrons with down spins, and
(ii) spin rapidities { }M
=1 , where
n
(iii) band (orbital) rapidities { }=1 , where Ne n and n are respectively the number
of electrons in the majority and minority bands.
Note that for nonzero crystalline field D the bands are populated differently. As the vacuum
state we chose the state with all the electrons having spin up and populating the band
m = 2. The eigenvalues and eigenfunctions are obtained following the standard Bethe
ansatz procedure, i.e. the cancellation of the so-called unwanted terms (see, e.g. [17
19,25]) yield the Bethe ansatz equations (BAE) for periodic boundary conditions:
e
ikj L
M
Y
e11 (kj
=1
N
e2Si (
e1 (kj ),
=1
Ne
Y
e1 ( kj ) =
j =1
Ne
Y
n
Y
M
Y
e2 ( ),
=1
e1 ( kj ) =
j =1
n
Y
e2 ( ),
(7)
=1
where em () = (2 imc)/(2 + imc), j = 1, . . . , Ne , = 1, . . . , M, = 1, . . . , n , and

L is the length of the box. The magnetization is given by M z = 12 Ne + SNi M, the
orbital moment by L = Ne 2n , and the energy of the system is
E=
Ne
X
j =1
kj2 2x
M
X
a2S ( ),
(8)
=1
where x = Ni /L. The first set of terms corresponds to the kinetic energy of the host
electrons and the second one arises from the impurities [28,29]. Here an () is the Fourier
transform of exp(n|c|/2). An isolated impurity corresponds to Ni = 1. For Ne = 0 we
recover the BabujianTakhtajan Heisenberg chain [41,42].
692
In a model with open boundary conditions, also the reflected waves have to be
considered. The procedure to obtain the open boundary BAE is described in Refs. [4345]
and the BAE are similar to those for periodic boundary conditions (independently if
impurities are present or not). Also interesting is the limit x 1 in which the system
consists of a lattice of interacting impurities.
In the thermodynamic limit (L, Ne , M, n with Ne /L, M/L and n /L kept fixed)
the solutions of the BAE can be classified as:
(i) real charge rapidities corresponding to unbound itinerant electrons;
(ii) pairs of complex conjugated charge rapidities representing triplet spin (orbitalsinglet) pairs;
(iii) spin bound states (-strings); and
(iv) inter-band bound states ( -strings).
This classification of states is identical to the one of the host without impurities [31]. The
impurities do not add new types of states, but do modify the distribution densities of the
rapidities. The integral equations satisfied by the density distributions are linear, so that the
contributions of the host and the impurities can always be separated.
In thermal equilibrium the dressed energies of the excitations ((k) for unbound
electrons, 9( ) for spin-triplet pairs, n () for spin strings of length n, and n ( ) for
the interband strings of length n) satisfy

(k) = k 2 12 H D + T a1 ? ln 1 + e9/T
T

an ? ln 1 + en /T 1 + en /T ,
n=1
14 c2

12 H + T a2 ? ln 1 + e9/T

+ T a1 ? ln 1 + e/T
9( ) = 2
T
T ln 1 + e
n ( )/T
X

[n6=1 an1 + an+1 ] ? ln 1 + en /T ,
n=1
= 2Dn + T an ? ln 1 + e/T
+T

An,m ? ln 1 + em /T ,
m=1
min(2S,n)
X

a2S+n+12l ( )
T ln 1 + en ()/T = H n 2x
l=1
X

An,m ? ln 1 + em /T
T an ? ln 1 + e/T + T
m=1

T [n6=1 an1 + an+1 ] ? ln 1 + e9/T ,
(9)
where H is the magnetic field, T the temperature and ? denotes convolution. The kernel
An,m () is the Fourier transform of coth(|c|/2)[exp(|n m||c|/2) exp((n +
m)|c|/2)]. The effect of the impurities enters through the driving term for the spin
693
rapidities (-strings), which is proportional to x. This driving term is analogous to the

one of the BabujianTakhtajan Heisenberg chain of spin S.
The integral equations for the density functions ((k) for unbound electrons, 0 ( ) for
spin-triplet pairs, n () for spin strings of length n, and n ( ) for the interband strings
of length n) are obtained from Eqs. (9) by differentiating with respect to the chemical
potential, e.g. 2[1 + exp((k)/T )](k) = (k)/. The energy, the number of
electrons, the spin magnetization and the orbital moment (population difference of the
two bands) are given by
Z
Z

E
2
= dk k (k) + 2 d 2 14 c2 0 ( )
L
min(2S,n)
X
X Z
d n () a2S+n+12l ( ),
2x
Ne
=
L
n=1
l=1
dk (k) + 2
d 0 ( ),
Z
Z
Z
X
Mz
= xS + 12 dk (k) + d 0 ( )
n d n (),
L
n=1
Z
Z
X
L
= dk (k) 2
n d n ( ).
L
(10)
n=1
4. Properties of a single impurity

4.1. Summary of properties of the host
To understand the behavior of an isolated impurity we first briefly summarize the
properties of the host without impurities [31]. The most important physics is contained in
the ground state and at low temperature. Taking the T 0 limit in Eqs. (9) we have that for
x 0 only unbound electrons, spin-triplet pairs, spinons (spin strings of length 1) and twospin bound states (spin strings of length 2) can have a negative dressed energy, i.e. can be
populated in the ground state. Hence, in general the groundstate consists of four Dirac seas.
The populated states correspond to rapidities in the intervals | | 6 Q with 9(Q) = 0,
|k| 6 B with (B) = 0, || 6 A1 with 1 (A1 ) = 0, and || 6 A2 with 2 (A2 ) = 0.
The integration limits Q, B, A1 and A2 determine the Fermi points of the Dirac seas.
From the statistical point of view all the states correspond to hard-core particles, while the
symmetry of the wavefunctions depends on the strength of the interaction and is in general
different from fermions and bosons, being of anion nature [46]. The energy of the particles
in the four Dirac seas is given by the dressed energies, i.e. 9( ) for the pairs, (k) for
the unpaired electrons, 1 () for the spinons, and 2 () for the two-spin strings, and carry
a physical momentum given by
694
Z
ppair ( ) = 2

d 0 h0 ( 0 ) + 0 ( 0 ) ,
Zk
punp (k) = 2

dk 0 h (k 0 ) + (k 0 ) ,
Z
p1 () = 2

d0 1h (0 ) + 1 (0 ) ,
Z
p2 () = 2

d0 2h (0 ) + 2 (0 ) ,
(11)
respectively. Here the subindex h denotes the corresponding distribution of rapidity holes.
In this way, the rapidities parameterize the energy and the momentum.
The solution of the ground state equations in general requires a numerical approach.
The equations, however, simplify considerably in three limiting cases. Consider first D =
H = 0, which corresponds to the degenerate band limit with Limp = 0, which is sometimes
referred to as the orbital singlet. From Eqs. (9) we see that the band of unbound electrons
is empty in this limit, i.e. B = 0. For H = 0 the Dirac sea of two-spin bound states
is completely filled (A2 = ), while the band of spinons has zero spectral weight and
is irrelevant. The full band is eliminated via Fourier transformation and we obtain for
the dressed energy (density) of the spin-triplet orbital-singlet bound states (which carry
a charge 2e):

9( ) = 2 2 c2 /2 2,
0 ( ) + h0 ( ) =
1
.
(12)
This means that for D = H = 0 the ground state of the itinerant electrons is only
determined by the spin-triplet orbital-singlet pairs. These pairs are effectively free, i.e.
they are noninteracting and their dispersion law, Eq. (12), corresponds to free (hard-core)
bosons with symmetric wavefunctions [31]. From 9(Q) = 0 we have that for D =
H = 0 the chemical potential (Fermi energy) is equal to = Q2 (c/2)2 . The ground
state energy, the number of itinerant electrons and the Fermi momentum are given by
E = (4QL/)[(Q2 /3) (c/2)2 ], Ne = 4QL/ and pF = (/2)(Ne /L), respectively.
Electronhole excitations within the 9 band yield a gapless continuum. On the other hand,
the spectrum of unbound charges has a gap given by
1
c2
Q2
=
2
c
ZQ
d
Q
( 2 Q2 )
.
cosh(/c)
(13)
The three relevant energy bands (triplet pairs, unpaired electrons and two-spin strings) as
a function of physical momentum are displayed in Fig. 1(a) for Q = c = 1. The chemical
potential corresponds to zero-energy. Note that the maximum momentum of the two-spin
695
Fig. 1. Energy bands as a function of momentum for the Dirac seas forming the ground state of the
host system: (a) for D = H = 0, Q = c = 1, and (b) for D = 0, H = 1, Q = c = 1. Here the Fermi
level has zero energy. Both cases correspond to the degenerate band limit and the unpaired electron
states are gapped. In case (a) only three bands, namely, the triplet pairs, Epair , unpaired electrons,
Eunp , and two-spin strings, E2 , are shown. The spinon band, E1 , has zero spectral weight in this
case and has collapsed to a point in the origin. In case (b) H > Hs , so that all the spins are polarized
and the bands of the spinons and two-spin strings are empty (positive energy).
strings is pF /2 and that their energy is negative. The fourth band (spinons) has zero spectral
weight and reduces to one point at the origin.
Let us now consider the other two important limiting cases. If the crystalline field D is
very strong, the band with index m = 1 will be empty, and the system is reduced to the
gas of electrons interacting via a repulsive -function potential (for a review of exactly
solved models of strongly correlated electrons see [47]). On the other hand, if a very large
magnetic field is applied, all electrons are spin-polarized (the spin-index plays no role)
and the model reduces to the many-body problem of spinless fermions in two bands (the
index m is a pseudo-spin) with an attractive interaction [47]. In this latter case for D = 0
the unpaired electrons are again gapped, so that only the band of triplet pairs is populated.
The energy bands for this situation is depicted in Fig. 1(b) for Q = c = 1 and H = 1. Again
pF = (/2)(Ne /L) and in this case all four bands have a finite momentum range.
As a function of field (assuming D = 0 for simplicity) the two-spin string band, 2 (),
is no longer completely filled and has a Fermi surface at 2 (A2 ) = 0. The Fermi point
A2 (H ) is a decreasing function of the field, from A2 = for H = 0 to A2 = 0 at Hs ,
the saturation magnetic field, at which all spins point upward (ferromagnetic state). The
critical field Hs can be expressed as
ZQ
Hs = (1/2)
Q

d a1 ( ) + a3 ( ) 9( ),
(14)
696
and the saturation magnetization is M z = S + 12 Ne for H > Hs . The phase transition at Hs

is of second order. In Eq. (14) the dressed energy 9( ) is the solution of
ZQ
9( ) = 2 c /2 2 H
2
d 0 a2 ( 0 )9( 0 ),
(15)
where the field H can actually be absorbed into the chemical potential. Eq. (15) means
that the spin-triplet pairs are no longer free hard-core bosons, but rather anion-like
quasiparticles with nonsymmetric wavefunctions [46]. Note that superconducting and
antiferromagnetic correlations coexist in the Dirac sea of spin-triplet pairs for H < Hs
(no off-diagonal long-range order).
To illustrate the possible phases of the host, we show in Fig. 2 a schematic phase diagram
in the H D plane. Of course only the first quadrant needs to be considered. The phase
diagram consists of four curves arising from Q(H, D) = 0, B(H, D) = 0, A1 (H, D) = 0
and A2 (H, D) = 0, which correspond to the van Hove singularities of the empty bands.
Eight different phases are then possible: in phase A all four bands are populated, while
in phases B and C only the two-spin string and spinon bands are empty, respectively,
but all other bands partially occupied. In D only the triplet pairs and unpaired electron
states are occupied (all spin-polarized). In phase E there are only unpaired electrons and
spinons, with the spinons being gradually depopulated as a function of field. In F there
are only unpaired (spin-polarized) electrons. Phases G and H are gapped for the unpaired
electrons. Hence, only triplet pairs (in G and H) and two-spin strings (in G) are present.
Phase H is again spin-polarized. The three simple cases referred to above correspond
to the origin and large D or H along the corresponding axis. Note that although we
Fig. 2. Schematic phase diagram in the H D plane for the host. There are four phase boundary
lines that arise when one of the four bands gets empty (van Hove singularities). They are labeled
accordingly with Q = 0, B = 0, A1 = 0, and A2 = 0. This gives rise to eight different phases denoted
AH. In phase A all four bands populated, in phases B and C three bands are populated, with the
two-spin string and spinon bands empty, respectively, and in D only the triplet pairs and unpaired
electron states are occupied (all spin-polarized). In phase E there are only unpaired electrons and
spinons, with the spinons being gradually depopulated as a function of field. In F there are only
unpaired (spin-polarized) electrons. Phases G and H are gapped for the unpaired electrons. Hence,
only triplet pairs (in G and H) and two-spin strings (in G) are present. Phase H is again spin-polarized.
Dc (H = 0) is the same as in Eq. (13). The details of the diagram may depend on Ne /L and c.
697
have four bands in the ground state there are only three external variables determining
the occupations, namely, the chemical potential (total number of electrons), the external
magnetic field (magnetization) and the band splitting D (relative populations of the two
electron bands) [31]. The critical band-splitting Dc is only weakly dependent on H , as it
can be verified analytically in the weak and strong coupling limits, for which the gap is
field independent.
4.2. Equations for the impurity
To study an isolated impurity we set x = 0 in Eqs. (9) but keep x = 1/L in the
equations for the density functions [1719]. The integral equations obeyed by the impurity
distribution densities in the ground state are
ZQ
h (k) + (k) =
d a1( k) 0 ( )
ZA1
ZA2
d a1 ( k)1 () +
+
A1
ZQ
h0 ( ) + 0 ( ) =
d a2 ( k)2 (),
A2
d 0 a2 ( 0 ) 0 ( 0 )
ZB
ZA1
dk a1 ( k)(k) +
d a2 ( )1 ()
A1
ZA2
+

d a1 ( ) + a3 ( ) 2 (),
A2
ZA2
2h () + 2 () +
d0 [2a2( 0 ) + a4 ( 0 )]2 (0 )
A2
ZQ
ZB
d a1 ( ) + a3 ( ) ( ) +
=
Q
dk a2 ( k)(k)
ZA1
A1
min(2,2S)
X

d0 a1 ( 0 ) + a3 ( 0 ) 1 (0 ) +
a2S+32l ( ).
(16)
l=1
The limits of integration are the same as for the host. The equation for the density of
spinons (spin strings of length 1), 1 , coincides with the second of Eqs. (16) if its left-hand
side is replaced by [1h ( ) + 1 ( )] + a2S ( ).
698
4.3. Zero-field properties

The ground state properties of the impurity are obtained by solving the system of integral
equations (16). As mentioned above this, in general, requires a numerical approach, except
for limiting cases, e.g. (a) the degenerate band limit (D = 0) and (b) the very large magnetic
field limit (all the electrons are spin polarized). In the latter case the spin (and thus the
magnetic impurity) plays no role. Below we limit ourselves to the degenerate band limit.
We distinguish the cases S = 1/2 and S > 1. For S > 1 the impurity contributions to the
densities of pairs and spin strings of length two are
1 () + 1h () = 0,
0

1
() + ei(S1)|c| ,
2 () =
2 cosh(c/2)
0 () + h0 () = eiS|c| .
(17)
Here the hat denotes Fourier transform. The impurity magnetization is then given by
z
= S 1, independently of the electron population in the triplet pair band. Clearly,
Mimp
the spin of the impurity is underscreened for S > 1 by the low-lying excitations of the host,
which carry spin 1, and is totally screened if S = 1. The valence of the impurity,
ZQ
hNiimp = 2
Q

2
1 Q +
1 Q
tan
+ tan
,
d a2S ( ) =
cS
cS
(18)
depends explicitly on the total number of electrons in the host via Q. The impurity valence
is zero for an empty band (Q = 0) and increases monotonically with Q until a triplet pair is
localized at the impurity site for Q . Here measures the coupling to the host, i.e. for
| | Q the impurity is on resonance with the host (strong coupling), while if | | Q it is
off-resonance (weak coupling). The valence of the impurity also decreases with increasing
spin S (classical spin limit).
For S = 12 we obtain, if D = H = 0:
ei
,
2 cosh(c/2)

1
ei
0
() 1 () +
,
2 () =
2 cosh(c/2)
2 cosh(c/2)
1 () + 1h () =
0 () + h0 () =
ei|c|
.
2 cosh(c/2)
(19)
While for S > 1 the spectral weight of the spinons is zero in the ground state, this is not
the case for an impurity of spin 12 . Consequently, the host equations do not define the
integration limit A1 , since it is an irrelevant parameter. With an impurity driving term,
which has a mesoscopic weight of the order 1/L, the limits T 0 and L are an
z
= 12 2h (0). It
delicate issue. The impurity magnetization for all fields is given by Mimp
z
= 0, independent of the
follows that the zero-field impurity magnetization is zero, Mimp
value of A1 . The valence of the impurity is equal to
699
ZQ
hNiimp = 2
d G2 ( ),
(20)
where Gn (x) is defined as

Z
1
exp(iux n|cu|/2)
.
du
Gn (x) =
2
2 cosh(cu/2)
(21)
Hence, the impurity valence increases as a function of the band filling (Q) and decreases
with the off-resonance shift .
4.4. Magnetic field dependence
The ground state properties of the impurity in a magnetic field are given by 2h , which
satisfies a Fredholm equation (third of Eqs. (16)) with two driving terms, namely a Kondo
driving term and a term involving the density function of charged spin-triplet pairs 0 . The
magnetization of the impurity is then the sum of the Kondo magnetization and the one
arising from the spin-fluctuations due to the valence admixture. Assuming that the Kondo
temperature and the Zeeman splitting are much smaller than the band width for the charged
spin-triplet pairs, we can neglect the contribution of the valence fluctuations to the impurity
magnetization.
For S > 1 the ground state of the impurity is magnetic and the susceptibility diverges
as H 0. For small fields the Fredholm equation for spin strings of length two can be
reduced to a hierarchical sequence of Wiener-Hopf integral equations. Keeping only the
leading equation of this hierarchy the solution only depends on the parameter | | A2 ,
which is usually parameterized by 1 ln(H /TK ), where TK exp(| |) plays the role
of a Kondo temperature. The next order contribution in the hierarchical sequence is smaller
by factor A1
2 and also depends on | | + A2 . Such a dependence indicates an interference
of the two Fermi points of the band of spin strings of length two, i.e. backscattering or
more than one partial wave playing a role. However, in the Kondo limit of very large | |
(TK H Q/c) we can neglect the nonleading contributions, so that

z
= M0 1 (LH)1 ln(LH)/(LH)2 + ,
(22)
Mimp
where LH = | ln H /ATK| (A is a S-dependent universal constant). Here, for H TK the
upper sign and M0 = S 1 are to be selected, while TK H Q/c corresponds to the
lower sign and M0 = S.
For S = 1 the ground state is a singlet and the zero-field magnetic susceptibility is finite,
z
H TK1 (plus corrections due to the mixed-valence), which is
i.e. for small fields is Mimp
reminiscent of the Fermi-liquid like behavior.
The most interesting situation is S = 12 , which corresponds to the overscreened impurity
(here we consider A1 = 0). Instead of a single pole in the WienerHopf solution as for
S = 1, or a branch cut for S > 1, one has the second-order pole. Then the leading field
z
(4H / 3)LH, and the magnetic
dependence of the impurity magnetization is Mimp
susceptibility has a logarithmic divergence as for the two-channel Kondo problem.
700
If D = 0 all conduction electrons are bound in spin-triplet pairs, independently of

the magnetic field. This drastically differs from the situation of spin-singlet pairs (see
Refs. [32,33]), where a magnetic field exceeding a critical field breaks up singlet-pairs,
in analogy to the BCS theory [48,49].
4.5. Effect of band splitting
To populate the -band the crystalline field splitting has to be larger than a critical value
Dc = , which is the energy required to overcome the binding energy and break up a pair.
If the band splitting D is changed, e.g. by applying pressure, the electron gas shows no
response unless D > Dc . Hence, the properties of the impurity for D < Dc are qualitatively
the same as for D = 0. At D = Dc the system undergoes a second-order phase transition.
For D > Dc the unbound electrons in the ground state break up the orbital singlet (i.e., the
two electron bands have different populations) and the spinon band acquires a nonvanishing
occupation at the expense of the spin-two-strings, but the total zero-field magnetization of
the itinerant electrons remains zero. For S > 1 the magnetic impurity is only marginally
affected by the population of the -band for D > Dc and H = 0; its valence becomes
D-dependent (e.g. for d = (D Dc )/Dc 1 its absolute value increases with d), but
the Kondo magnetization remains the same as for D < Dc . On the other hand, due to the
presence of the unbound electrons and spinons (which carry spin 12 ) the overscreened fixed
point for S = 12 is quenched and a singlet groundstate is expected [50].
4.6. Low-temperature behavior
The Sommerfeld expansion of the free energy and the dressed energies yields the
low-temperature specific heat of the host, which is proportional to T and given by C =
1
1
+ vpair
+ v11 + (3/2)v21 ]. Here vunb , vpair , v1 and v2 are the Fermi velocities
(T /3)[vunb
of the corresponding bands. If a band is empty its group velocity does not contribute to C.
At the critical fields, e.g. Dc or Hs (at second-order phase transitions one Fermi velocity
is
zero), the van Hove singularity of the empty band yields a low-T term proportional
to T .
For S = 1 the magnetic susceptibility of the impurity is Curie-like for T TK with the
usual logarithmic corrections, but for T TK it is finite (reminiscence of the Fermi-liquid
like behavior) [1719]. For the underscreened case, S > 1, the susceptibility is Curie-like
with a Curie constant corresponding to an effective spin of S 1 at low-T and to a free
spin S at high-T . In the overscreened case, at low T the impurity susceptibility diverges as
(T ) ln(T /TK ), and is again Curie-like at high T . The specific heat of the impurity for
S = 1 is Fermi-liquid like at low T , Cimp = T /3TK . For S > 1 at T H the degeneracy
of the underscreened impurity spin gives rise to a Schottky anomaly. For the overscreened
heat of the impurity is Cimp T ln(T /TK ). The impurity
case S = 12 the low-T specific
properties also manifest T singularities at the critical points of the host.
701
5. Finite concentration of impurities

Magnetic impurities in BCS-superconductors with spin-singlet pairing break the timereversal symmetry and act as Cooper pair breakers [48,49], reducing this way the critical
temperature and the gap. The gap closes before the superconductivity is destroyed, an
effect that is known as gapless superconductivity [48,49]. In superconductors with spintriplet pairing, usually both magnetic and nonmagnetic impurities affect the superconducting state. The effects of the impurities are quite different in the present 1D correlated
electron model.
For a finite concentration x of magnetic impurities the impurity driving terms affect
in first place the spin space, giving rise to an impurity band in the ground state. The
Bethe ansatz solution of our model holds for all Ne and Ni . For Ni = 0 we have the
pure host, for Ni = 1 an isolated impurity embedded into the host, and for Ne = 0 and
finite x we obtain the BabujianTakhtajan Heisenberg chain of spin S. Hence, for finite x
the impurities interact among each other and are antiferromagnetically correlated. This is a
consequence of the second term in the expression of the energy (8). When both x and Ne /L
are finite, the magnetic interactions among the impurities and those between the impurities
and the itinerant electrons give rise to a large frustration of magnetic bonds [28,29]. The
situation x = 1, for instance, is also of interest since it corresponds to a lattice of impurities
(Anderson lattice-like model).
As already mentioned for finite x the BAE and the energy eigenvalues do not depend
on the relative positions of the impurities. This locality structure is a consequence of the
integrability condition (YangBaxter triangular relation) and manifests itself in an impurity
band with a relatively weak dispersion. The Hamiltonian itself, however, depends on the
spatial arrangement of the impurities. This gives rise to a large degeneracy of Hamiltonians
corresponding to the same BAE [28,29].
We introduce an equal number of impurities with + and to avoid a net chirality.
A finite concentration of impurities drastically affects the properties at low temperatures
and moderate magnetic fields. In contrast to the isolated impurity, where the ground state
in general consists of four Dirac seas (unbound electrons, spin-triplet orbital-singlet pairs,
and spin strings of lengths 1 and 2), a fifth Dirac sea (the impurity band) of spin strings of
length 2S appears if x 6= 0. The impurity band yields low energy spin excitations. If the
impurities have spin S = 1 or S = 12 they cannot generate a new Fermi sea and there are
only four Dirac seas in the ground state as in the pure host. In this case the impurities only
modify the existing bands.
The ground state integral equations satisfied by the dressed energies for the five classes
of states are obtained from Eqs. (9) in the limit T 0:
= k 2 12 H D a1 ? 9 + a1 ? 1 + a2 ? 2 + a2S ? 2S ,

9 = 2 2 14 c2 12 H a2 ? 9 a1 ?
+ a2 ? 1 + (a1 + a3 ) ? 2 + (a2S1 + a2S+1 ) ? 2S ,
1 = H + aX
2 ? 9 + a1 ? a2 ? 1 (a1 + a3 ) ? 2
x
a2S ( ) (a2S1 + a2S+1 ) ? 2S ,
702
2 = 2H (2a2 + a4 ) ? 2 + a2 ? + (a1 + a3 ) ? 9 (a1 + a3 ) ? 1

X min(2,2S)
X
KS ? 2S x
a2S+32l ( ),
"
2S = 2SH a4S + 2
2S1
X
l=1
a2l ? 2S + a2S ? + (a2S1 + a2S+1 ) ? 9
l=1
(a2S1 + a2S+1) ? 1 KS ? 2 x
2S
XX
a4S+12l ( ).
(23)
l=1
Here ? denotes convolution over the intervals for which the dressed energies are negative
Pmin(2,2S)1
a2S+22l (x).
and KS (x) = a|2S2| (x) + a2S+2 (x) + 2 l=1
5.1. Zero-field solution
We now study the zero-field properties of the impurity band for S > 1, the case for which
the impurities add new physics to the model. The analysis is simplified because the filling
of the Dirac seas for all spin-strings is governed by the field H . Each of the three Dirac seas,
1 , 2 and 2S is either completely filled or empty if H = 0. By Fourier transformation
we eliminate the spinons and spin two-strings and obtain an equation involving only the
2S-strings [31]. Here 2S decouples from the and 9 potentials, because as x 0 the
2S-string band is not populated. The only driving term is then proportional to x. The
dressed energy 2S satisfies an integral equation identical to that of the BabujianTakhtajan
Heisenberg chain [41,42] for an effective spin S 1 (the spin two-strings screen the
impurities) and a driving term scaled by the concentration of impurities x. The integral
equation is easily solved yielding
2S () =
x/2c
x/2c
.
cosh[( )/c] cosh[( + )/c]
(24)
In zero field the band of 2S strings is completely filled and yields zero magnetization, i.e.
the impurity spins are all spin-compensated due to antiferromagnetic correlations among
themselves, in analogy to a Heisenberg chain. The antiferromagnetic correlations and the
singlet ground state arise from the expression of the energy of our model, which leads to
the occupation of the 2S spin-string band. Similarly, we obtain for H = 0
1 () = G0 ? ,
2 () = G0 ? 9,
(25)
i.e. the band of spin two-strings is completely filled, and the one of spinons is empty in
the gapped case and filled if the chain has no gap. The dressed energies for the spinons
and two-strings do not depend explicitly on the concentration of impurities, so that their
properties are qualitatively the same as for the chain without impurities.
To clarify the above we compare the dilute limit x 0 with the behavior at finite
concentrations. In the dilute limit, x 0, the 2S band is empty and hence there is no
spin compensation among the impurities (only an underscreened Kondo effect). The case
x > 0 is completely different because 2S () is negative and occupied for all . The point
x 0 is then singular, in analogy to models for magnetic impurities embedded into the
703
supersymmetric tJ chain [28,29]. In a finite magnetic field the long wavelength spin
states are not populated, i.e. the large || tails of 2S () are positive and not occupied,
yielding a finite magnetization. The zero-field susceptibility is constant at T = 0, as for the
BabujianTakhtajan model. In Refs. [26,27] a different model was considered, in which
the impurity spins embedded into the Hubbard chain with attractive U do not interact with
each other. The impurities are then not spin compensated and act similarly to a magnetic
field. In that case the limit x 0 is not singular.
5.2. Magnetic field dependence
We now consider the gapped system in a high magnetic field and explicitly distinguish
the cases: (a) S > 1, (b) S = 1 and (c) S = 12 . In case (a) 2 and 2S are occupied in zero
field and gradually depopulated as a function of field. Two critical fields are then expected
at the van Hove singularities of the empty bands. For small x first the impurity band 2S is
f
emptied and at higher fields the band of two-strings is depopulated. For H > Hs1 , where

f
(26)
Hs1 Hs + (2x) a2S+1( ) + a2S1 ( )
and Hs is given by Eq. (14), the system is totally spin-saturated, i.e. ferromagnetic, and all
f
the bands for spin excitations are empty. Note that Hs1 > Hs . The crossovers at the two
critical fields correspond to second-order phase transitions with mean-field exponents.
For case (c) the band of spinons is filled in zero field, driven by the S = 12 impurities.
For finite but small x the spinon band is emptied first as a function of field, giving rise
f
to a second order phase transition with mean-field exponents. At fields larger than Hs2
Hs +2xa2( ) again the system is totally spin polarized (ferromagnetic) and all spin bands
are empty.
Finally, in case (b) only the spin two-strings plays a role. Depending on x the dressed
energy of the spin bound states can reveal up to three-minima as a function of . One
is centered at = 0 and is due to the itinerant electrons (spin-triplet pairs), while the
other two arise from the impurities and are centered at = . This three-valley structure
is manifested only for large and x. The bottom of each of the three minima can be
approximated by a parabola, so that the band has up to two 1D van Hove singularities (the
ones at = occur at the same energy), each defining a second order phase transition
as a function of H . The minimum centered at = 0 determines the critical field of the
transition to the spin-saturated phase.
The impurity spins and the triplet pairs of the conduction electrons are antiferromagnetically coupled with each other, yielding a singlet zero-field ground state with frustrated
bonds. The frustration is gradually broken up as a function of field and is totally lifted at the
second order phase transitions. Such a spin frustration is often observed in heavy fermion
compounds. For D < Dc the antiferromagnetic fluctuations of the impurity spins and the
spin-triplet charged pairs of the itinerant electrons coexist with the superconducting correlations of the latter. While a magnetic field gradually breaks up the antiferromagnetism, the
superconducting correlations are not suppressed, i.e. the effective spin one of the triplets is
gradually aligned to the field, but the bound states are not destroyed.
704
5.3. Gap
The hybridization impurities also affect the gap of the unbound electron excitations.
As discussed above, for D < Dc and H = 0 Eqs. (23) simplify considerably, because the
unbound electron excitations are gapped B = 0 and the Dirac seas of the spin strings of all
lengths are either empty or completely filled, and their dressed energies can be eliminated
via Fourier transformation (see Eqs. (24) and (25)). The spin fluctuations renormalize
the dressed energies for both, pairs and unpaired electrons, decreasing their respective
energies. Hence, the chemical potential and the Fermi points of the pairs Q (related by
the zeroes of 9( )) are also renormalized. If each impurity contributes with hNiimp [see
Eqs. (18) and (20)] conduction electrons to the system, then Q remains unchanged. The
gap for the unbound electron excitations, = (k = 0), is one-half of the smallest energy
required to depair a spin-triplet orbital-singlet pair. The renormalization of the gap due to
the impurities is proportional to x.
We first consider impurities of spin S = 12 in zero-field and D = 0. For B = 0 and
keeping Q unchanged by the impurities, the dressed energies of the paired and unpaired
electrons and the gap renormalization are
X

G2 ( ),
9( ) = 2 2 14 c2 x
(k) = k 2 G0 ? 9 x
G1 (k ),
#
" ZQ
G2 ( )
(x) (0) X 1
=
G1 ( )
d
x
c
2 cosh(/c)
1 1
G2 (Q )
2 c
ZQ
Q
!
1
.
d
2 cosh(/c)
(27)
In the last expression both terms are positive, so that the gap increases linearly with x with
the addition of spin 12 impurities.
Similarly, for S > 1 and H = D = 0 we obtain (B = 0 and Q unchanged):
X

a2S ( )],
9( ) = 2 2 14 c2 x
(k) = k G0 ? 9 x
2
G2S (k ),
#
" ZQ
a2S ( )
(x) (0) X 1
=
G2S ( )
d
x
c
2 cosh(/c)
1 1
a2S (Q )
2 c
ZQ
Q
!
1
.
d
2 cosh(/c)
(28)
In the last expression, the second term is always positive and enhances the gap, while the
first term is negative and reduces the gap, both linearly with x. The effect of the impurities
705
Fig. 3. (a) Gap of unpaired electrons as a function of impurity concentration x for Q = 0.1,
interaction strength c = 0.1, impurity spin S = 1 and several values of the impurity off-resonance
shift . The increase or decrease of the gap with x is linear. (b) Rate of change of the gap with x for
S = 1 as a function of for various band-fillings Q. Note that c3 d((x)/(0))/dx is only a function
of Q/c and /c.
on the gap depends on the off-resonance shift . The linearity in x and the -dependence
are shown in Fig. 3(a). Note that c3 d[(x)/(0)]/dx is a function only of Q/c and /c,
which is displayed in Fig. 3(b) as a function of /c for several Q/c. For Q the rate
of change is zero for all and for it also vanishes for all Q. For a given Q/c
the general trend of the gap is to decrease for large /c and to increase for small values
of /c. The gap decreases with x when the impurities are off-resonance with the conduction
electrons, i.e. when they are weakly coupled to the itinerant electrons. In addition the
impurity spins compensate each other as they do in a Heisenberg antiferromagnet, so that
they do not actually break the time-reversal symmetry. The rate of the reduction of the gap
is larger for weak interactions c and can give rise to a critical concentration xcr at which
the gap is closed. Since there are no occupied unbound electron states in the gapped phase
(x < xcr ), the magnetic impurities do not break the spin-triplet pairs, but rather weaken
them. This contrasts the suppression of superconductivity by magnetic impurities in BCSlike superconductors [48,49].
The increase of with x can either be considered a constructive interference of the
superconducting and antiferromagnetic fluctuations or the consequence of an enhanced
density of states of the electrons due to the presence of the impurities. For small the
impurity rapidity is on-resonance with the conduction electrons and the impurities and the
host are strongly coupled. This stabilizes the pairs and increases the gap with x.
For fixed Q the decrease of the gap is linear in x. However, if we assume that the total
number of electrons is fixed (i.e. each impurity contributes with a fraction of electron), then
Q is slightly x-dependent, giving rise to a weak curvature in (x).
There is co-existence of superconducting and magnetic fluctuations. For x < xcr the band
of the spin 2S-strings is filled for H = 0 (as for the Heisenberg chain), so that the impurity
spins compensate each other via antiferromagnetic correlations (yielding a singlet ground
state). Also the spins of the spin-triplet Cooper pairs are antiferromagnetically correlated,
so that superconducting and antiferromagnetic fluctuations coexist. For x > xcr the gap
is closed and the band of unpaired electrons is gradually populated. Simultaneously the
706
populations of the spinon and spin two-string bands change, since the number of Cooper
pairs is reduced. The total magnetization of the system is 12 times the number of holes
in the 2S-string band. According to Eqs. (23), the 2S band remains completely filled for
x > xcr , so that the zero-field magnetization is zero. In other words, the ferromagnetism
of the unpaired electrons is compensated by a change in the spectral weight of the spinons
and spin two-strings. Note that the two electron bands are now populated differently, so
that the orbital singlet state is broken up. The itinerant ferromagnetic component for x
slightly larger than xcr , yields a small but nonzero B x xcr (B are the Fermi points
for the unbound electrons). The corresponding 1D van Hove singularity indicates a second
order phase transition. The effect of an external magnetic field is to gradually break up the
antiferromagnetic frustration, align the spins of the triplet pairs, introduce a hole population
in the band of 2S-strings, and hence a magnetization even for small fields.
5.4. Comparison with other integrable models
We now compare the present results with those for other integrable models containing
a finite concentration of impurities. For instance, in the supersymmetric tJ model with
impurities [28,29], for H = 0 the electrons are paired into singlet states, but there is no
binding energy associated with them. The spectrum is gapless and again the spin 2Sstrings form an impurity band that compensates the impurity spins via antiferromagnetic
correlations. The key difference with the present model is the lack of gap. In our gapped
system the electrons do not respond to a crystalline field smaller than the critical one.
On the other hand, the Hubbard model with attractive U has a spin-gap in its excitation
spectrum, analogously to the present model [30]. Magnetic impurities reduce the spin-gap,
and eventually close it at a critical concentration [26,27]. For x > xcr the unpaired electron
band is populated and gives rise to itinerant ferromagnetism. For the model considered
in Refs. [26,27] the spin 2S-strings are not occupied, and hence the impurity spins are all
aligned rather than antiferromagnetically correlated. In this case the system has no response
to a magnetic field smaller than Hc (the field closing the gap).
Recently we considered hybridization impurities in the same two-band correlated
electron model but with opposite sign of c, which leads to the formation of spin-singlet
orbital-triplet pairs [30]. For S > 1 the gap of the unbound electron excitations decreases
for small off-resonance shifts, while it increases for larger | |. This behavior is opposite
to what we obtain for the present model. A second difference is that for the spin-singlet
pairs there exists a phase for x > xF > xcr with a spontaneous magnetization at H = 0,
while there is no such phase for the present spin-triplet model. Also the increase of the
gap for any parameters for S = 12 impurities is a special feature of the spin-triplet pairs.
The key difference of the present study and Ref. [30] is the mixed valence character of the
impurities, mandated by the YangBaxter relations, which involves configurations with
spins S and S + 12 in Ref. [30], while configurations of spins S and S 12 for the spintriplet case.
707
5.5. Conformal invariance

The low-T specific heat has a contribution from each rapidity band with Fermi surface
1
1
+ vpair
+ v11 +
proportional to the inverse of its group velocity, i.e. C = (T /3)[vunb
1
]. These terms correspond to the Dirac seas of unbound
(3/2)v21 + (3S/(S + 1))v2S
electrons, paired spin-triplets, and spin strings of lengths 1, 2 and 2S (assuming S > 1). The
Sommerfeld linear-temperature coefficient is in general large because of the impurity
band, which is flat and has a small group velocity. This effect of the impurities is
reminiscent of heavy-fermion-like behavior. As a function of H , D or one or more Fermi
seas may be emptied (or populated) giving
rise to a 1D van Hove singularity contribution
to the specific heat, proportional to T . At the tricritical point the leading term is of
order T 1/4 .
The asymptotes of correlation functions in the conformal limit correspond to the
semidirect product of in general four Gaussian models and the level-2S WessZumino
NovikovWitten (WZNW) model of conformal anomaly c = 3S/(S + 1) (not to be
confused with the coupling constant). The WZNW model can be mapped onto a gapless
Gaussian (c = 1) model and a sector of (2S 1)/(S + 1) parafermions. The latter are
massive and affect the finite-size corrections to the ground state energy only as a trivial
constant shift, whilst they are present in the low temperature corrections (e.g. in the specific
heat). Note that the spin two-strings are actually a level-2 WZNW model of central charge
c = 3/2 (see the specific heat). The conformal dimensions for the right- and left-moving
low-lying excitations are given by [51,52]

Nk 2
2rS r 2
t
,
(29)
n
+
j = Zkj Dk
j
2Zkj
8S(S + 1)
where Nj represents the deviation of the total number of particles in band j from its
average value, and Dj is the backward scattering quantum number. Here n
j are the
numbers of particlehole excitations about the two Fermi points of the Dirac sea j , Zj k
is the matrix of dressed charges and 1/Zkj denotes its inverse. The last term with r = 1,
2, . . . , 2S corresponds to the parafermionic sector. The conformal dimensions pertain to
the minimal choice of possible numbers Dk , Nk , n
k and r.
The matrix of dressed charges Zj k is given by the functions j k () taken at the
Fermi points, where j k () satisfy the integral equations obtained by differentiating the
dressed energies for each rapidity band, Eqs. (23), with respect to the effective chemical
potentials j , e.g. for k we have j k = k /j . Here j, k label the five bands populated
in the ground state and take the values unb,pair, 1, 2, 2S. Similarly, j represents , 9,
1,2,2S , and the effective chemical potentials j for these bands are [ + D + (H /2)],
2( + D), H , 2H , and 2SH , respectively. The solution of this set of, in general, 25 linear
integral equations can be obtained numerically.
The situation simplifies in some limiting cases for which some of the Dirac seas are
either completely filled or empty. For example, for D = H = 0 and x < xc there are only
pairs and spin excitations in the system. For low energy excitations the charge and spin
sectors are effectively decoupled, with scaling dimension = 1 for the spin-triplet pairs
708
(gas of free hard-core bosons), and =
r(r+2)
4(S+1) ,
r = 1, 2, . . . , 2S for the spin strings (if the
impurities are of spin S = then = for the spinons). This shows again the coexistence
of superconducting and antiferromagnetic fluctuations. The critical behavior due to the van
Hove singularity of an empty band of rapidities appears also in the functional dependence
of the components of the dressed charge matrix.
1
2
1
2
6. Conclusions
We studied the properties of magnetic impurities embedded into a correlated electron
host involving two bands and a contact exchange interaction between the electrons. The
exchange interaction is repulsive in the spin channel and attractive in the orbital sector,
and binds the electrons into spin-triplet (orbital-singlet) Cooper-pair-like bound states.
A splitting between bands larger than a critical one, , is required to overcome this
binding energy. In this case a fraction of the electrons is unpaired and spin polarized.
We consider magnetic hybridization impurities embedded into the host without destroying
the integrability. The impurities are only elastic scatterers, i.e. the scattering of electrons
off the impurities does not produce a reflection amplitude. As a consequence the energy
spectrum of the system does not depend on the relative positions of the impurities, giving
rise to a locality structure with a flat dispersion (heavy fermions). The impurities are
in a mixed valent state, which is a superposition of two ionic configurations with a
magnetic ground state. The properties of the system strongly depend on the concentration
of impurities. Three situations have to be distinguished, namely, (i) isolated impurities, (ii)
a concentration x of impurities of spin S = 12 , and (iii) a concentration x of impurities of
spin S > 1.
Isolated impurities are partially screened via the Kondo effect at low T by the spintriplet pairs, so that the ground state magnetization is S 1 (S > 1). Hence, an isolated
impurity is magnetic for S > 1, but its ground state is a singlet for S = 1. For S = 12
the spin-triplet bound states of the conduction electrons cannot exactly compensate the
impurity spin and overcompensated Kondo behavior (a quantum critical point) is obtained.
For large magnetic fields in all cases an asymptotically free spin S is recovered.
The ground state of the host without impurities is in general described in terms of four
Fermi seas of rapidities (spin-triplet Cooper-pairs, unpaired electrons, spinons and twostrings of spin rapidities). For S > 1 the impurities introduce a fifth Dirac sea, namely,
a band of spin 2S-strings. The spectral weight of the 2S-strings is proportional to x, i.e.
it tends to zero as x 0. Hence, x = 0 is a singular point. For finite x and in the gapped
phase, the zero-field ground state of the system is nonmagnetic, i.e. the impurity spins
compensate each other via antiferromagnetic correlations similarly to a Heisenberg chain.
The spin-triplet pairs are also antiferromagnetically correlated. All the antiferromagnetic
correlations are gradually broken up as a function of a magnetic field.
A finite concentration of S = 12 impurities increases the binding energy of the spintriplets. This constructive interference between antiferromagnetic and superconducting
fluctuations can be attributed to an effective increase of the electron density of states.
709
For impurities with S > 1, on the other hand, the effect of the impurities on the gap
depends on the value of the off-resonance shift . For small the impurities and the triplet
pairs are strongly coupled and the interference of antiferromagnetic and superconducting
fluctuations is again constructive. However, if is large the coupling between impurities
and charges is weak (in the sense of the Kondo effect) and the gap can decrease. For small
c there exists a critical concentration xcr at which the gap is closed. For x > xcr the band
of spin-polarized unpaired electrons acquires a finite population (with a quantum phase
transition at xcr ), but the overall magnetization of the system is still zero.
Acknowledgements
P.S. acknowledges the support by the US Department of Energy under grant DE-FG0298ER45707 and by the US National Science Foundation under grant DMR98-01751.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
A.V. Balatsky, S.A. Trugman, Phys. Rev. Lett. 79 (1997) 3767.

N. Nagaosa, P.A. Lee, Phys. Rev. Lett. 79 (1997) 3755.
C. Ppin, P.A. Lee, Phys. Rev. Lett. 81 (1998) 2779.
H.v. Lhneysen, J. Magn. Magn. Mater. 200 (1999) 532.
A. Yacoby, M. Heiblum, D. Mahalu, D. Shtrikman, Phys. Rev. Lett. 74 (1995) 4047.
P. Schlottmann, P.D. Sacramento, Adv. Phys. 42 (1993) 641.
J.A. Sauls, Adv. Phys. 43 (1994) 113.
Y. Maeno, H. Hashimoto, K. Yoshida, S. Nishizaki, T. Fujita, J.G. Bednorz, F. Lichtenberg,
Nature (London) 372 (1994) 532.
T.M. Rice, M. Sigrist, J. Phys. Condens. Matter 7 (1995) L643.
K. Ishida, H. Mukuda, Y. Kitaoka, K. Asayama, Z.Q. Mao, Y. Mori, Y. Maeno, Nature
(London) 396 (1998) 658.
T. Ishigiro, K. Yamaji, G. Saito, Organic Superconductors, Springer-Verlag, Berlin, 1998.
C. Bergermann, S.R. Julian, A.P. Mackenzie, S. Nishizaki, Y. Maeno, Phys. Rev. Lett. 84 (2000)
2662.
S. Belin, K. Behnia, Phys. Rev. Lett. 79 (1997) 2125.
I.I. Lee, M.J. Naughton, G.M. Danner, P.M. Chaikin, Phys. Rev. Lett. 78 (1997) 3555.
I.J. Lee, D.S. Chow, W.G. Clark, J. Strouse, M.J. Naughton, P.M. Chaikin, S.E. Brown, Preprint
cond-mat/0001332.
K. Gorny, O.M. Vyasilev, J.A. Marindale, V.A. Nandor, C.H. Pennington, P.C. Hammel,
W.L. Hults, J.L. Smith, P.L. Kuhns, A.P. Reyes, W.G. Moulton, Phys. Rev. Lett. 82 (1999)
177.
A.A. Zvyagin, P. Schlottmann, J. Phys. Condens. Matter 9 (1997) 3543.
P. Schlottmann, A.A. Zvyagin, Phys. Rev. B 55 (1997) 5027.
P. Schlottmann, A.A. Zvyagin, Nucl. Phys. B 501 (1997) 728.
A.A. Zvyagin, P. Schlottmann, Phys. Rev. B 56 (1997) 300.
D.-H. Lee, J. Toner, Phys. Rev. Lett. 69 (1992) 3378.
A. Furasaki, N. Nagaosa, Phys. Rev. Lett. 72 (1994) 892.
P. Frjdh, H. Johannesson, Phys. Rev. Lett. 75 (1995) 300.
P. Frjdh, H. Johannesson, Phys. Rev. B 53 (1996) 3211.
A.A. Zvyagin, Phys. Rev. Lett. 79 (1997) 4641.
710
[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]
P. Schlottmann, Phys. Rev. B 57 (1998) 10638.

P. Schlottmann, A.A. Zvyagin, preprint.
P. Schlottmann, Nucl. Phys. B 525 [FS] (1998) 697.
A.A. Zvyagin, P. Schlottmann, Nucl. Phys. B 565 [FS] (2000) 555.
A.A. Zvyagin, P. Schlottmann, Phys. Rev. B 60 (1999) 6292.
P. Schlottmann, Phys. Rev. Lett. 68 (1992) 1916.
P. Schlottmann, Phys. Rev. B 49 (1994) 6132.
A.M. Tsvelik, Zh. Eksp. Teor. Fiz. 93 (1987) 1329; Sov. Phys. JETP 66 (1987) 754.
N. Andrei, H. Johannesson, Phys. Lett. A 100 (1984) 108.
K. Lee, P. Schlottmann, Phys. Rev. B 37 (1988) 379.
P. Schlottmann, J. Phys. Condens. Matter 3 (1991) 6617.
A.A. Zvyagin, P. Schlottmann, J. Phys. A 31 (1998) 1981.
A.A. Zvyagin, H. Johannesson, M. Granath, Europhys. Lett. 41 (1998) 213.
P. Schlottmann, Z. Phys. B 59 (1985) 391.
L.A. Takhtajan, Phys. Lett. A 87 (1982) 479.
H.M. Babujian, Nucl. Phys. B 215 [FS] (1983) 317.
F. Woynarovich, Phys. Lett. A 108 (1985) 401.
F. Alcaraz, M.N. Barber, M.T. Batchelor, R.J. Baxter, G.R.W. Quispel, J. Phys. A 20 (1987)
6397.
A.A. Zvyagin, H. Johannesson, Phys. Rev. Lett. 81 (1998) 2751.
F.D.M. Haldane, Phys. Rev. Lett. 67 (1991) 937.
P. Schlottmann, Int. J. Mod. Phys. B 11 (1997) 355.
B.T. Mattias, H. Suhl, E. Corenzwit, Phys. Rev. Lett. 1 (1958) 92.
A.A. Abrikosov, L.P. Gorkov, Sov. Phys. JETP 12 (1961) 1243.
P. Schlottmann, K.J.-B. Lee, Phys. Rev. B 52 (1995) 6489.
V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and
Correlation Functions, Cambridge University Press, 1993, and references therein.
F.C. Alcaraz, M.J. Martins, J. Phys. A 21 (1988) 4397.

Nested Bethe ansatz for PerkSchultz model

with open boundary conditions
Guang-Liang Li a,b, , Rui-Hong Yue a,b , Bo-Yu Hou b
a CCAST (World Lab.), P.O. Box 8730, Beijing 100080, China
b Institute of Modern Physics, P.O. Box 105, Northwest University, Xian 710069, China
Received 21 December 1999; revised 14 June 2000; accepted 28 June 2000
Abstract
By using the algebraic Bethe ansatz, we find the exact solution of PerkSchultz model with general
open boundary conditions. The Bethe ansatz equation and the eigenvalue of the transfer matrix are
obtained, which depends on two free parameters dominating the different grade boundary. Under
the special condition m = 1, the results in present paper give the solution of multichannel t-J model
with general boundary condition. Our results also recover those obtained by A. Gonzalez-Ruiz, Nucl.
Phys. B 424 [FS] (1994) 468. 2000 Elsevier Science B.V. All rights reserved.
PACS: 05.50+q; 64.60.Cn; 75.10.Hk; 75.10.Jm
Keywords: PerkSchultz model; Reflecting matrix; Nested Bethe ansatz; Open boundary
1. Introduction
The quantum inverse scattering method (QISM) [17] presents a unified approach
scheme for the construction of integrable quantum systems and the simultaneous
eigenvectors of commuting integrals of motion of a quantum system. The Bethe ansatz
solutions of the integrable model will make it possible to study the thermodynamic
properties of the model, such as correlation functions [7], specific heat, magnetic
susceptibility [811], and finite size effects [1214]. So it is necessary to obtain the Bethe
ansatz solutions for the integrable models.
In the framework of QISM, many works have been done on the integrable models with
boundary conditions [5,1532], especially that with open boundary conditions [5,2232].
The solving of the model with open boundary conditions will enable us to investigate the
boundary effect of the system. It may be also helpful for us to study the impurities problems
which have attracted considerable interest recently [3344].
E-mail addresses: lgl@phy.nwu.edu.cn (Guang-Liang Li), yue@phy.nwu.edu.cn (Rui-Hong Yue).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 0 ) 0 0 4 1 6 - 8
712
G.-L. Li et al. / Nuclear Physics B 586 [FS] (2000) 711729
There are two ways to construct the integrable models with impurities. One is to embed
impurities into periodic chain without violating integrability by Andrei and Johannesson
for the Heisenberg model [33] and later into various models including the supersymmetric
t-J model of interacting electrons by various authors [3439]. Another is to put impurities
on the two ends of an open chain, such as Heisenberg model [40,41] and Kondo-type
impurities coupled to correlated electron system [4244]. Usually, the impurities embedded
in these systems are magnetic ones, which describes the spin interactions between
particles and impurities. The open boundary mentioned here means the boundary fields.
Such open boundary models can also be considered as a simple boundary non-magnetic
impurities models by putting two potential scatters instead of magnetic impurities on both
ends. Thus, the study of the models with open boundary fields will help us to analyse the
boundary effects since the boundary fields can change continuously.
In this paper, we will apply the nested Bethe ansatz method to find the eigenvalues
and the eigenvectors of the transfer matrix for PerkSchultz model [45] with general
boundary conditions. The model contains n fermionic and m bosonic grade bases. Based
on the reflection equation, we find the general diagonal solution, which determine the
general boundary interaction in the Hamiltonian. When m = 1, the model reduces into
the q-deformed version of the generalized supersymmetric t-J model with n components.
The Hamiltonian contains particle hopping term, the nearest-neighbour sln spinspin
interaction and the contribution of boundary fields. While m = 2, n = 1, the model recover
one discussed in Ref. [32]. When both the boundary matrices K + and K do not depend
on spectrum parameter, our result reduces to that of [27].
The paper is organized as following. In Section 2 we introduce the open PerkSchultz
model by finding the solution K of reflection equation which determines the nontrivial
boundary terms in the Hamiltonian. Section 3 contributes to the diagonalization of the
transfer matrix of the model with general boundary conditions in the frame work of nested
Bethe ansatz. In Section 4 the summary of our main results is presented and some tedious
calculation are given in the appendix.
2. The vertex model and integrable boundary conditions

The vertex weight R(u) of the PerkSchultz model are defined as [45]

aa
(u) = sin + (1)a u , a = 1, 2, . . . , m + n,
Raa
ab
(u) = sin(u)(1)a b ,
Rab
a 6= b = 1, 2, . . . , m + n,
ab
(u) = sin()eiusign(ab),
Rba
ab
(u) = 0, for other cases,
Rcd
a 6= b = 1, 2, . . . , m + n,
where is an anisotropy parameter and

0, a = 1, . . . , m,
a =
1, a = m + 1, . . . , m + n.
For convenience, we denote
(1)
(2)
a(u) = sin(u + ), b(u) = sin(u),

w(u) = sin( u), c (u) = sin()eiu .
713
(3)
The R-matrix defined by (1), (2) satisfies YangBaxter equation

R12 (u v)R13 (u)R23 (v) = R23 (v)R13 (u)R12 (u v),
(4)
where R12 (u), R13 (u) and R23 (u) act in C m+n C m+n C m+n with R12 (u) = R(u) 1,
R23 (u) = 1 R(u), etc. Additionally, the R-matrix also fulfills the following properties
t1 t2
(u),
P T invariance: P12 R12 (u)P12 = R12
(5)
unitarity: R12 (u)R21 (u) = sin(u + ) sin( u) id,

crossing symmetry:
t1
t2
(u)M1 R12
(u d)M11
R12
(6)
= sin(u) sin(u + d) id, (7)
where d = m n and M is a (m + n) (m + n) matrix

Mbc = bc Mb ,

1 6 b 6 m,
(1)b ei2(b1),
Mb =
(1)b ei2(2mb) , m < b 6 m + n.
(8)
One can also verify that M is a symmetry matrix of the R-matrix:

[M M, R12 (u)] = 0.
(9)
The local transfer matrix L(u) satisfying the following equation [24,45,46]:
R12 (u v)L1 (u)L2 (v) = L2 (v)L1 (u)R12 (u v),
(10)
where L1 (u) = L(u) 1, L2 (u) = 1 L(u). The standard row-to-row monodromy matrix
for an N N square lattice is defined by
T (u) = LN (u) L1 (u)
= R0N (u) R01 (u),
(11)
where L(u) is assumed to be in the fundamental representation. We can prove that T (u)
also satisfies the YangBaxter equation
R12 (u v)T1 (u)T2 (v) = T2 (v)T1 (u)R12 (u v),
(12)
where the operator T is an (m + n) (m + n) matrix of the operators acting in the quantum

N
.
space Vn+m
Now, we can use Mezincescue and Nepomechies generalized formalism to construct
integrable systems with open boundary conditions. In our case, the reflection equations
take the following form [47,48]:
R12 (u v)K1 (u)R21 (u + v)K2 (v)
= K2 (v)R12 (u + v)K1 (u)R21 (u v),
R12 (u + v)K1+ (u)t1 M11 R21 (u v d)M1 K2+ (v)t2

= K2+ (v)t2 M1 R12 (u v d)M11 K1+ (u)t1 R21 (u + v).
Obviously, there is an isomorphism between K + (u) and K (u).
(13)
(14)
714

d t
: K (u) K + (u) = K u
M.
2
(15)
Given a solution K (u) of Eq. (13), one can also find a solution K + (u) of Eq. (14). But
in a transfer matrix of an integrable lattice, K (u) and K + (u) need not satisfy Eq. (15). In
this paper, we will take Eq. (15) to define K + .
Solving the reflection Eq. (13), we have the following general diagonal solution
which [49]
K (u, ) =
n
X
PaA (u, )Eaa ,
(16)
a=1
where
PaA (u, ) =
eiu sin( + u), 1 6 a 6 A,

A < a 6 m + n.
eiu sin( u),
(17)
Correspondingly,
K + (u, ) =
n
X
PaB (u, )Eaa M
(18)
a=1
with
PaB (u, ) =
ei(u+d/2) sin( u d/2),

1 6 a 6 B,
i(u+d/2)
sin( + u + d/2), B < a 6 m + n.

e
(19)
In above Eqs. (17) and (19), and both are free parameters. The integer numbers A and
B are also free parameters which take value from 1 to m + n.
Taking Sklyanins formalism, the double-row monodromy matrix is defined as:
U (u) = T (u)K (u, )T 1 (u),
(20)
where T 1 (u) is the inverse of T (u) in the auxiliary and quantum spaces. The double-row
monodromy matrix U (u) satisfies the reflection equation
R12 (u v)U1 (u)R21 (u + v)U2 (v) = U2 (v)R12 (u + v)U1 (u)R21 (u v).
(21)
Then the transfer matrix is defined as:

t (v) = tr K + (v, )U (v).
(22)
Using the reflection equations (14), (21) and the properties of the R-matrix (5)(9), one
can prove
[t (u), t (v)] = 0.
(23)
So the transfer matrix constitutes a one-parameter commutative family which ensures the
integrability of the model. As indicated by Sklyanin, the transfer matrix is related to the
Hamiltonian of the quantum chain with nearest-neighbour interaction and boundary terms
through the following relation
715
t 0 (0) = 2 tr K + (0)H + tr K + (0)0 ,
(24)
where
H=
N1
X
Hj +1,j +
j =1
tr K0+ (0)H0,N
1
K1 (0)0 +
2 sin( )
tr K + (0)
(25)
and
Hj +1,j = Rj0 +1,j (0)Rj1
+1,j (0).
(26)
From Eq. (22), one can derive the explicit expression of the Hamiltonian, which is omitted
here because it is not used in the following discussion. In order to compare it with the
SUq (1|2) supersymmetric t-J model, we give the Hamiltonian under m = 1, which is
defined:
)
(N1
XX
cj,s cj +1,s + cj +1,s cj,s P

H =P
j =1 s
N1
X
cos()
j =1
!
nj,a nj +1,a + cos() (nj + nj +1 ) cos() (nj nj +1 )
a=1
N1
X
j =1
N1
X
n
X
Sj Sj
+1

+ Sj Sj+1 + i sin() (nj
X
i sin()
j =1
nj,a nj +1,b
a<b
!
nj +1 )

nj,a nj +1,b
a>b
+ Ha + Hb + constant id,
(27)
where cjs (cj s ) creates (annihilates) an electron with spin component s, s = 1, 2, . . . , n

located j th site. nj is density operator, Sj is the spin matrix at site j . + denotes the
set roots of the su(n) algebra. P is a operator projecting out doubly occupied states. The
constraint is that more than one electron on each site is strictly prohibited. Ha and Hb are,
respectively,
Ha =
n+1
ei X
n1,a1 ,
sin( )
(28)
a=A+1
Hb =
n+1
X
ei((2Bd/2))
nN,a1
sin( + (d/2 2 + B))
a=B+1
with d = 1 n.
(29)
716
3. Nested Bethe ansatz for open boundary conditions

Rewriting the Eq. (20)
R12 (u )ac11ca22 U (u)c1 d1 R21 (u+ )db11 cd22 U (v)d2 b2
= U (v)a2 c2 R12 (u+ )ac11dc22 U (u)c1 d1 R21 (u )db11 db22 ,
(30)
where the repeated indices sum over 1 to m + n, u = u v, u+ = u + v and introducing

a set of notations for convenience:
A(v) = U (v)11 ,
Ba (v) = U (v)1a ,
Ca (v) = U (v)a1 ,
Dab (v) = U (v)ab ,
2 6 a, b 6 m + n,
(31)
we have the commutation relations from Eq. (30), see Ref. [27]
a(u v)b(u + v)
Bb (u)A(v)
a(u + v)b(u v)
c+ (u v)b(2u)
Bb (v)A(u)
a(2u)b(u v)
c (u + v)
ecb (u),
Bc (v)D
(32)
a(u + v)
R12 (u + v + )ac11dc22
ec1 d1 (u)
ea1 b1 (u)Bb2 (v) =
R21 (u v)db11 db22 Bc2 (v)D
D
b(u v)b(u + v + )
c+ (u v)
ed2 b2 (v)
R12 (2u + )ad12 db11 Bd1 (u)D
b(2u + )b(u v)
c+ (u + v)b(2v)
1
R12 (2u + )ab12 db21 Bd2 (u)A(v) (33)
+
b(2u + ) a(u + v)a(2v)
for the cases m > 1 and
w(u v)b(u + v)
Bb (u)A(v)
A(v)Bb (u) =
w(u + v)b(u v)
c+ (u v)b(2u)
Bb (v)A(u)
w(2u)b(u v)
c (u + v)
ecb (u),
Bc (v)D
(34)
w(u + v)
R12 (u + v )ac11dc22
ec1 d1 (u)
ea1 b1 (u)Bb2 (v) =
R21 (u v)db11 db22 Bc2 (v)D
D
b(u v)b(u + v )
c+ (u v)
ed2 b2 (v)
R12 (2u )ad12 db11 Bd1 (u)D
b(2u )b(u v)
c+ (u + v)b(2v)
1
R12 (2u )ab12 db21 Bd2 (u)A(v) (35)
+
b(2u ) w(u + v)w(2v)
for the case m = 0, where all indices take values from 2 to m + n, and the repeated indices
e is defined by
sum over 2 to m + n. The new operators D
A(v)Bb (u) =
eab (v) = Dab (v) ab

D
R12 (2v)a1
1a
R12 (2v)11
11
A(v).
717
(36)
Considering the vertex model defined by Eqs. (1), (2), the explicit form for Eq. (36) are
c+ (2v)
A(v), m 6= 0,
Dab (v) ab
a(2v)
eab (v) =
(37)
D
Dab (v) ab c+ (2v) A(v), m = 0.

w(2v)
Introducing the vacuum state,
|vaci =
N
Y
(1, 0, . . . , 0)t ,
(38)
where t denotes the transposition. The action of the double-row monodromy matrix on the
vacuum state is (see Appendix A)
N 1
N

R (u)11
A(u)|vaci = R(u)11
11
11 K11 (u)|vaci = (u)|vaci,

c+ (2u)
e
K11 (u) bN (u)b N (u)|vaci
Dab (u)|vaci = ab Kaa (u)
R(2u)11
11
= ab a (u)|vaci,
Ca (u)|vaci = 0,
Ba (u)|vaci 6= 0,
(39)
where b(u)
= R 1 (u)ab
ab , a 6= b. Note that the action of Ba (u) on the vacuum state is not
proportional to the vacuum state. Now we construct the eigenvectors of transfer matrix
t (u). It take the form
(v1 , . . . , vL ) = Bb1 (v1 ) BbL (vL )|vaciF b1 bL .
(40)
Recalling the definition of the transfer matrix, we rewrite the transfer matrix as:
t (u) =
=
=
m+n
X
a=1
m+n
X
a=2
m+n
X
Ka+ (u)U (u)aa

eaa (u) +
Ka+ (u)D
(m+n
X
c+ (2u)
Ka+ (u)
a=2
a(2u)
)
+ K1+ (u)
A(u)
eaa (u) + S1 (u)A(u),

Ka+ (u)D
(41)
a=2
where
S1 (u) =
m+n
X
a=2
Ka+ (u)
c+ (2u)
+ K1+ (u).
a(2u)
Using the Eq. (41), we then obtain the action of t (u) on (see Appendix B)
(42)
718
t (u) = (u)S1 (u)
L
Y
a(vj u)b(vj + u)
b(vj u)a(vj + u)
j =1
F
+
b1 bL
L
Y
j =1
Bb1 (v1 ) BbL (vL )|vaci
a(vj u)a(u vj )
L
(1) (u) (2)(u,
{vi })db11 d
bL
b(u vj )b(vj + u + )
F b1 bL Bd1 (v1 ) BdL (vL )|vaci

L
X
c+ (vk + u)
b(2vk )
c+ (vk u)
(d1 )
S1 (u) +
T1 (u)
+
b(vk u)
a(vk + u)b(2u + )
a(2vk )
k=1
L
Y
j =1,j 6=k
a(vj vk )b(vj + vk )
L b1 bL
(vk )S(vk , {vi })db11 d
bL F
b(vj vk )a(vj + vk )
(43)
Bd1 (u)Bd2 (v1 ) Bdk (vk1 )Bdk+1 (vk+1 ) BdL (vL )|vaci

L
X
c+ (u vk )
sin() (1)(vk )
c (u + vk )
(d )
S1 (u) +
T1 1 (u)
a(u + vk )
b(u vk )b(2u + )
T1(d1 ) (vk )
k=1
L
Y
j =1,j 6=k
a(vk vj )a(vj vk )
b(vk vj )b(vj + vk + )
dL (2)
b1 bL
L
(vk , {vi })cb11c
S(vk , {vi })dc11c
bL F
L
Bd1 (u)Bd2 (v1 ) Bdk (vk1 )Bdk+1 (vk+1 ) BdL (vL )|vaci,
where
L
{vi })cb11c
(2) (u,
bL
m+n
X

Ka+ (u) L(1) (u,
v1 ) L(1) (u,
vL )K (u,
(1) )
a=2
vL )1 L(1) (u,
v1 )1
L(1) (u,
=
m+n
X
c1 cL
b1 bL aa

c c
Ka+ (u) T (1) (u,
{vi })K (u,
(1) )T (1) (u,
{vi })1 b1 bL aa ,
1
(44)
a=2
i
(1)
sin(2u)e 2 N
b (u)b N (u),
(u) =
sin(2u + )
L(1) (v1 , vi )aa12 bb12 = R12 (v1 + vi )aa12 bb12 ,

dL
and u = u + /2, (1) = /2, vi = vi + /2. T1(d1 ) (u), S(vk , {vi })dc11c
L are defined in
Appendix B. From the above Eq. (43), one can see that the function is not the eigenstate
of t (u) unless F s are the eigenstates of (2) and the sum of the third and the fourth term
in the above equation is zero, which will give a restriction on the L spectrum parameters
{vi }. So, we arrive at the following results:
719
If F is the eigenstate of (2) with the eigenvalue (2) satisfying Eq. (46), then is the
eigenstate of t (u) with the eigenvalue (1) ,
(1) (u) = (1) (u)S1 (u)
L
Y
a(vj u)b(vj + u)
b(vj u)a(vj + u)
j =1
+ (1)(u)
L
Y
j =1
a(vj u)a(u vj )
(2) (u, {vi }),
b(u vj )b(vj + u + )
(45)
where
(1) (u, {vi })F = (2) (u, {vi })F,
(2)(vk , {vi }) =
L
(1) (vk )b(2vk )T1(d1 ) (vk ) i Y b(vj + vk )
,
e
a(vk vj )
(1)(vk )a(2vk ) sin()
(d1 6 B),
j =1,6=k
(46)
= (u). Therefore, the diagonalization of t (u) is reduced to finding the
where
eigenvalue of (2) . The explicit expression of (2) (see Eq. (43)) implies that (2) can be
considered as the transfer matrix of an L-sites quantum chain, in which every spin takes
m + n 1 values. The related YangBaxter equation is the same as the one of t (u), except
R being an (m + n 1)2 (m + n 1)2 matrix. Hence, we can use the same method to find
the eigenvalue of (2) . Repeating the procedure m times, one can reduce to a subsystem
(m+1) which is an n n matrix in auxiliary space. The related YangBaxter equation is
also defined by Eq. (1), but one should notice that in this case all a = 1 due to m = 0.
e by the second Eq. (37). The
In order to diagonalize (m+1) , we need the definition of D
elements of T (m) satisfy Eqs. (34) and (35). Following the same procedure, one can further
reduce the (m+1) into the (m+2) subsystem. The late has the same structure as the former.
In this case one finally obtains the eigenvalue of (m+n1) . This is the well-known nested
Bethe ansatz. Because the wave-functions are not needed in this paper, we omit them here.
The eigenvalue and the constraint on the spectral parameters read as

(k) u, vi(k1) , vi(k)
(1) (u)
Pk
a(vj(k) u)b(u + vj(k) + (k 1))
(k1) Y
= Sk (u) (k) u, vi
(k)
(k)
j =1 b(vj u)a(u + vj + (k 1))
(k)
Pk
(k1) Y
u, vi
(k)
(k)
a(vj u)a(u vj )
b(u vj(k) )b(u + vj(k) + k)

(k) (k+1)
u, vi , vi
(1 6 k 6 m),
j =1
(k+1)

(k) u, vi(k1) , vi(k)
(k)
(k)
Pk
w(vj u)b(u + vj + (2m k + 1))
(k1) Y
= Sk (u) (k) u, vi
(k)
(k)
j =1 b(vj u)w(u + vj + (2m k + 1))
(47)
720

(k)
Pk
(k1) Y
+ (k) u, vi
(k)
(k)
b(u vj )b(u + vj + (2m k))

(k) (k1)
u, vi
vi
(m + 1 6 k 6 n)
j =1
(k+1)
(k)
a(vj u)a(u vj )
(48)
and

(k+1) vl(k) , vi(k) , vi(k+1)
=
(k) (vl(k) , {vi(k1) }) Tk (vl(k) ) i

e
(k) (v (k) , {v (k1) }) sin()
l
sin(2vl(k)
(k)
(k)
Pk
sin(vl + vj + (k 1))
+ (k 1)) Y
(k)
sin(2vl
+ k)
j =1,6=l
(k)
(k)
sin(vj vl
(1 6 k 6 m), (49)

(k+1) vl(k) , vi(k) , vi(k+1)
=
(k) (vl(k) , {vi(k1) }) sin(2vl(k) + (2m k + 1))Tk (vl(k) )
ei
(k)
(k1)
(k)
(k) (vl , {vi
}) sin() sin(2vl + (2m k))
Pk
Y
sin(vl(k) + vj(k) + (2m k + 1))
(m + 1 6 k
(k)
(k)
sin(v
v
+
)
j =1,6=l
l
j
6 m + n 1),
(50)
where

sin(2u + d)
d
i((k2+B+d/2)+u)
e
,
sin + B u
2
sin(2u + k)
1 6 k 6 B,

i((k2+d/2)u)
+ d + u sin(2u + d) ,
e
sin
Sk (u) =
2
sin(2u + k)
B < k 6 m,

sin(2u + d)
d
i((d/2+2mk)u)
,
sin + + u
e
2
sin(2u + (2m k))
m<k6m+n
(51)
for 1 6 B 6 m and when m < B 6 m + n

sin(2u + d)
d
i((k2+2mB+d/2)+u)
(2m
B)
u
,
sin
2
sin(2u + k)
1 6 k 6 m,

d
i((2mk+2mB+d/2)u)
sin + (2m B) + u
e
2
(52)
Sk (u) =
sin(2u + d)
, m < k 6 B,
sin(2u + (2m k))

sin(2u + d)
d
i((d/2+2mk)u)
+
u
,
sin
+
e
2
sin(2u + (2m k))
B <k6m+n
and

d
i((k1+B+d/2)+u)
+ B u sin(2u + d),
e
sin
1 6 k 6 B,

i((k1+d/2)u)
d
sin + + u sin(2u + d),

Tk (u) = e
2
B
<
k
6
m,

i((d/2+2mk1)u)
+ + u sin(2u + d),
sin
m<k6m+n
721
(53)
for 1 6 B 6 m and while m < B 6 m + n

d
i((k1+2mB+d/2)+u)
(2m
B)
u
sin(2u + d),
sin
1 6 k 6 m,

ei((2mk1+2mB+d/2)+u) sin + (2m B) u

2
(54)
Tk (u) =
sin(2u + d), m < k 6 B,

i((d/2+2mk1)u)
sin + + u sin(2u + d),

e
B < k 6 m + n,
(k) u,
(k)
P
(k1)
k1
Y
sin(u + vj
+ k)
k (u),
(k1)
u + )
j =1 sin(vj
(k1)
vi
= P
k1
1 6 k 6 m,
Y sin(u + vj(k1) + (2m k))
k (u), m + 1 6 k 6 n,
(k1)
sin(vj
u )
j =1
(55)
P
(k1)
(k1)
k1
Y
sin(u + vj
+ (k 1)) sin(u vj
)
(k1)
(k1)
u + ) sin(u vj
+ )
j =1 sin(vj
sin(2u
+
(k
1)))
k , 1 6 k 6 m,
(k1)
sin(2u + k)
(56)
u, vi
= P
(k1)
(k1)
k1
Y
sin(u + vj
+ (2m k + 1)) sin(u vj
)
sin(vj(k1) u + ) sin(u vj(k1) + )
j =1
sin(2u + (2m k + 1)))
k , m + 1 6 k 6 m + n 1.
sin(2u + (2m k))
Here k (u) and k are defined by

i(u+(k1)/2)
sin( + u),
1 6 k 6 A,
e
k (u) = ei(u+(k1)/2) sin( A u),
A < k 6 m,
i(u+(2mk+1)/2)
sin( A u), m < k 6 m + n
e
for 1 6 A 6 m and when m < A 6 m + n
(57)
722
i(u+(k1)/2)
sin( + u),
1 6 k 6 m,
e
k (u) = ei(u+(2mk+1)/2) sin( + u),
m < k 6 A,
i(u+(2mk+1)/2)
sin( (2m A + 2) u), A < k 6 m + n,
e
i/2
, 1 6 k 6 A,
e
k = ei3/2 , A < k 6 m,
i3/2
,
m<k6m+n
e
for 1 6 A 6 m and while m < A 6 m + n
i/2
e
, 1 6 k 6 m,
k = ei/2 ,
m < k 6 A,
i3/2
, A < k 6 m + n.
e
(58)
(59)
(60)
In the above representation, vj(1) = vj , vj(0) = 0, P0 = N, P1 = L and (m+n) = 1,

Pm+n = 0 are assumed. Notice that (k) (u, {vik1 }) vanishes at the special points vi(k1)
due to the factor sin(u vi(k1) ) appearing in (k) . Taking u = vi(k1) in formulae (47) and
(48), we can get another kind of constraints on (k)

(k) vl(k1) , vi(k1) , vi(k)

= (k) vl(k1) , vi(k1) Sk vlk
Pk
Y
sin(vj(k) vl(k1) + ) sin(vj(k) + vl(k1) + (k 1))
sin(vj(k) vl(k1) ) sin(vj(k) + vl(k1) + k)
j =1
(61)
1 6 k 6 m,

(k) vl(k1) , vi(k1) , vi(k)

= (k) vl(k1) , vi(k1) Sk vlk
Pk
Y
sin(vj(k) vl(k1) ) sin(vj(k) + vl(k1) + (2m k + 1))
j =1
sin(vj(k) vl(k1) ) sin(vj(k) + vl(k1) + (2m k))
(62)
m + 1 6 k 6 m + n.
Now, changing the index k into k + 1 in the above formulae, we can obtain constrains
on (k+1) . Comparing these with Eqs. (47)(50), one can derive out the following Bethe
ansatz equations
Y sin(vl(k) vj(k1) + ) sin(vl(k) + vj(k1) + k)
Pk1
j =1
sin(vl(k) vj(k1) ) sin(vl(k) + vj(k1) + (k 1))

Y sin(vl(k) vj(k+1) ) sin(vl(k) + vj(k+1) + (k + 1))
Pk+1
j =1
(k)
sin(vl
(k+1)
vj
(k)
) sin(vl
(k+1)
+ vj
+ k)
(k)
(k)
(k)
(k)
Pk
Y
sin(vl vj ) sin(vl + vj + (k 1))
j =1,6=l
sin(vl(k) vj(k) + ) sin(vl(k) + vj(k) + (k + 1))
sin(2vlk + (k + 1))Sk+1 (vlk )k+1 (vlk )k

Tk (vlk )k (vlk )
(1 6 k 6 m 1),
ei
723
(63)
Y sin(vl(m) vj(m1) + ) sin(vl(m) + vj(m1) + m)
Pm1
j =1
(m)
sin(vl
(m1)
vj
(m)
) sin(vl
(m1)
+ vj
+ (m 1))
Y sin(vl(m) vj(m+1) ) sin(vl(m) + vj(m+1) + (m 1))
Pm+1
j =1
(m)
sin(vl
(m+1)
vj
(m)
+ ) sin(vl
(m+1)
+ vj
+ m)
(64)
sin(2vlm + (m 1))Sm+1 (vlm )m+1 (vlm )m i

e ,
Tm (vlm )m (vlm )
Y sin(vl(k) vj(k1) ) sin(vl(k) + vj(k1) + (2m k))
Pk1
j =1
sin(vl(k) vj(k1) ) sin(vl(k) + vj(k1) + (2m k + 1))

Y sin(vl(k) vj(k+1) ) sin(vl(k) + vj(k+1) + (2m k 1))
Pk+1
j =1
sin(vl(k) vj(k+1) + ) sin(vl(k) + vj(k+1) + (2m k))
Pk
Y
sin(vl(k) vj(k) + ) sin(vl(k) + vj(k) + (2m k + 1))
j =1,6=l
sin(vl(k) vj(k) ) sin(vl(k) + vj(k) + (2m k 1))
sin(2vlk + (2m k 1))Sk+1 (vlk )k+1 (vlk )k

Tk (vlk )k (vlk )
(65)
ei
(m + 1 6 k 6 m + n 1).
The above Bethe ansatz equations can be simplified by introducing the following new
variables
(
(k)
wi k/2,
1 6 k 6 m,
(k)
(66)
vj =
(k)
wi (2m k)/2, m + 1 6 k 6 m + n.
Then the Bethe ansatz equations then take the form
Y sin(vl(k) vj(k1) /2) sin(vl(k) + vj(k1) /2)
Pk1
j =1
sin(vl(k) vj(k1) + /2) sin(vl(k) + vj(k1) + /2)

Y sin(vl(k) vj(k+1) /2) sin(vl(k) + vj(k+1) /2)
Pk+1
j =1
(k)
sin(vl
(k+1)
vj
(k)
+ /2) sin(vl
(k+1)
+ vj
(k)
(k)
(k)
(k)
Pk
Y
sin(vl vj + ) sin(vl + vj + )
j =1,6=l
sin(vl(k) vj(k) ) sin(vl(k) + vj(k) )
+ /2)
(67)
724
= k vlk
(1 6 k 6 m + n 1, k 6= m),
Y sin(vl(m) vj(m1) /2) sin(vl(m) + vj(m1) /2)
Pm1
sin(vl(m) vj(m1) + /2) sin(vl(m) + vj(m1) + /2)
j =1
Y sin(vl(m) vj(m+1) + /2) sin(vl(m) + vj(m+1) + /2)
Pm+1
(m)
sin(vl

= m vlm ,
j =1
(m+1)
vj
(m)
/2) sin(vl
(m+1)
+ vj
/2)
(68)
where
Tk (vlk k/2)k (vlk k/2)
ei , 1 6 k 6 m 1,
k + )S
k k/2)
k k/2)
sin(2v
(v
(v
k+1
k+1
k
l
l
l
Tm (vlm m/2)m (vlm m/2)
ei , k = m,
m
m
m
(69)
k (vlk ) = sin(2vl )Sm+1 (vl m/2)m+1 (vl m/2)m
sin(2v k )S (v k (2m k)/2) (v k (2m k)/2)
k+1 l
k+1 l
k i
e ,
k (2m k)/2) (v k (2m k)/2)
T
(v
k
k
l
l
m < k 6 m + n 1.
(k)
The function (1)(u, . . .) must not be singular at u = vj (1 6 j 6 pk , 1 6 k 6 m + n 1)

since the transfer matrix t (u) is an analytic function of u. In fact, the Eq. (46) comes from
the condition under which the unwanted term vanishes. One can understand this constraint
from another point of view: from Eq. (45), we know that u = vj = vj(1) is a pole of (1) (u).
In order to keep the analyticity of (1) (u), one should need the residue of (1)(u) at vj
vanishing, which also gives the constraint (46). So, (1) (u) is analytic at vj . Similarly,
Eqs. (67) ensure the analyticity of (1) (u) at all vj(k) . Therefore, the eigenvalues of the
transfer matrix are analytic functions if the previous Bethe ansatz equations are satisfied.
The energy spectrum of 1-dimensional quantum system defined by Eq. (24) can be
derived from (1) (u). It is
E =
PX
1 =L
k=1
sin[(d 1)]
2 sin()
+ N cot()
cos( + 2vk ) cos() sin() sin(d)
+
where
B0 =
ei(d/2+) sin[(B 0 d)]

ei
+
,
2 sin( ) sin(d) sin( + d/2 B 0 )
B,
2m B,
1 6 B 6 m,
m + 1 6 B 6 m + n.
(70)
(71)
4. Summary
In this paper, we solve PerkSchultz model under general open boundary conditions
(general diagonal solution of reflection equations) and obtain the Bethe ansatz equations.
725
The corresponding energy spectrum of the 1-dimensional quantum system defined by

Eq. (24) is derived. Using the Bethe ansatz equations and energy spectrum, we can
study the boundary contributions to the thermodynamic quantities, such as magnetic
susceptibility and specific heat. When m = n = 2, we can obtain the exact solution for
one strongly correlated electrons model (EsslerKorepinSchoutens model) [50]. While
m = 2 and n = 1, the obtained Bethe ansatz equations and the spectrum coincide with that
obtained by GonzalezRuiz [32]. When A = B = m + n, we can recover the result obtained
in Ref. [27].
Here we only consider the case that the boundary matrices are general diagonal cnumber ones. For the non-diagonal c-number boundary matrices, the model can not be
solved as we know. However in special case that the non-diagonal boundary matrices are
operator ones like that in Refs. [4244], the model can be solved.
Acknowledgements
We thank Prof. K.J. Shi for his useful discussions. G.L. Li is supported partly by the
Science Fund of Northwest University.
Appendix A
Acting the operators on the vacuum state, we can easily find
A(u)|vaci = (u)|vaci,
Ca (u)|vaci = 0,
Ba (u)|vaci 6= 0,
N 1

11 N
(u) = R(u)11
11 K11 (u) R (u)11 .
(A.1)
eab (u) on the vacuum state. From the definition of Dab (u),
Now, we consider the action of D
we have
(u)T 1 (u)1b |vaci

Dab (u)|vaci = T (u)a1 K11
(u)T 1 (u)cb |vaci.

+ T (u)ac Kcc
(A.2)
The contribution of the first term can not be calculated directly but it can be worked out by
using the following method. Taking v = u in the YangBaxter equation, we can get:
T21 (u)R12 (2u)T1 (u) = T1 (u)R12 (2u)T21 (u).
(A.3)
Taking special indices in this relation and applying both sides of this relation to the vacuum
state, we find:

c+ (2u)
ab (u) T (u)ac T 1 (u)cb |vaci.
(A.4)
T (u)a1 T 1 (u)1b |vaci =
11
R(2u)11
Substituting this relation to Eq. (A.2), we have the result:
726

Dab (u)|vaci = ab
c+ (2u)
K11
(u)(u)
R(2u)11
11

c+ (2u)
N
N (u) |vaci.
(u)
K
(u)
b
(u)
b
+ Kaa
11
R(2u)11
11
(A.5)
Then we have

c+ (2u)
eab (u)|vaci = ab Kaa

(u)
K
(u)
bN (u)b N (u)|vaci.
D
11
R(2u)11
11
(A.6)
Appendix B
By repeatedly using the commutation relations between Bs which take the form (see
Ref. [27])
e12 (u1 u2 )d2 d1 Bd2 (u2 )Bd1 (u1 ),
Bb1 (u1 )Bb2 (u2 ) = R
b2 b1
(B.1)
where

ab
11
e21 (u)ab
R
cd = R21 (u)cd R(u)11 .
(B.2)
we can commute B(vk ) with B(vk1 ), . . . , B(v1 ), respectively.

Bb1 (v1 ) BbL (vL )|vaci
L
= S(vk , {vi })db11 d
bL Bd1 (vk )Bd2 (v1 ) Bdk (vk1 )Bdk+1 (vk+1 ) BdL (vL )|vaci,
(B.3)
where
L
S(vk , {vi })db11 d
bL =
L
Y
j =1+k
2 3
e12 (v1 vk )d1 d2 R
e
b j d j R
c2 b1 12 (v2 vk )c3 b2
c d
e12 (vk1 vk )ck1 dk .

R
bk bk1
(B.4)
Now, we calculate the action of the transfer matrix on . Applying Eq. (32) to the Bethe
ansatz state defined by Eq. (40), we have
A(u)Bb1 (v1 ) BbL (vL )|vaciF b1 bL
=
L
Y
a(vj u)b(vj + u)
(u) Bb1 (v1 ) BbL (vL )|vaciF b1 bL
b(vj u)a(vj + u)
j =1
L
L
X
c+ (vk u)b(2vk ) Y a(vj vk )b(vj + vk )
(vk )F b1 bL
a(2vk )b(vk u)
a(vj + vk )b(vj vk )
j =1,j 6=k
k=1
L
Bd1 (u)Bd2 (v1 ) Bdk (vk1 )Bdk+1 (vk+1 ) BdL (vL )|vaciS(vk , {vi })db11 d
bL
L
L
X
c (vk + u) Y
k=1
a(vk + u)
j =1,j 6=k
a(vj vk )a(vk vj )
c c
S(vk , {vi })b11 bLL
b(vj + vk + )b(vk vj )
727

L(1) (vk , v1 ) L(1) (vk , vk1 )L(1) (vk , vk+1 ) L(1) (vk , vL )(vk )
L(1) (vk , vL )1 L(1) (vk , vk1 )1 L(1) (vk , vk+1 )1
d d
L(1) (vk , vL )1 c 1c L
1
Bd1 (u)Bd2 (v1 ) Bdk (vk1 )Bdk+1 (vk+1 ) BdL (vL )|vaciF b1 bL .
(B.5)
eaa (u) on . We have

Next, we evaluate the action of Ka+ (u)D
m+n
X
eaa (u)Bb1 (v1 ) BbL (vL )|vaciF b1 bL

Ka+ (u)D
a=2
m+n
X
Ka+ (u)F b1 bL
L
Y
j =1
a=2
1
b(u vj )b(u + vj + )
s1 q1
p1 d 2
s2 q2
1
R12 (u + v1 + )ad
p1 q1 R21 (u v1 )ab1 R12 (u + v2 + )p2 q2 R21 (u v2 )s1 b2
p
dL
sL qL
R12 (u + vL + )pL1
L qL R21 (u vL )sL1 bL
Bd1 (v1 )Bd2 (v2 ) BdL (vL )pL sL pL (u)|vaci + u.t.,
(B.6)
{vi }) (44), we rewrite

where u.t. stands for unwanted term. Using the definition of (2) (u,
relation (B.6) as:
=
m+n
X
Ka+ (u)F b1 bL
j =1
a=2
L
Y
(2)
a(u vj )a(vj u)
(1)(u)
b(u vj )b(u + vj + )
L
(u,
{vi })db11 d
bL Bd1 (v1 )Bd2 (v2 ) BdL (vL )|vaci + u.t.
The unwanted terms in Eq. (B.6) is

u.t. =
L
X
d d
S(uk , {vi })b11 bLL F b1 bL (vk )T1
(d1 )
(u)
k=1
L
Y
a(vj vk )b(vj + vk )
c+ (u + vk )b(2vk )
a(u + vk )a(2vk )b(2u + )
a(vj + vk )b(vj vk )
j =i,6=k
Bd1 (u)Bd2 (v1 ) Bdk (vk1 )Bdk+1 (vk+1 ) BdL (vL )|vaci
+
L
X
c c
S(uk , {vi })b11 bLL F b1 bL T1
(d1 )
(u)
k=1
L
Y
a(vj vk )a(vk vj )
c+ (u vk )
b(u vk )b(2u + )
b(vk vj )b(vj + vk + )
j =i,6=k
L (vk , v1 ) L (vk , vk1 )

(1)
(1)
L(1) (vk , vk+1 )

L(1) (vk , vL )(vk )L(1) (vk , vL )1 L(1) (vk , vk+1 )1
(B.7)
728
L(1) (vk , vk1 )1 L(1) (vk , v1 )1
d1 dL
c1 cL
Bd1 (u)Bd2 (v1 ) Bdk (vk1 )Bdk+1 (vk+1 ) BdL (vL )|vaci,
(B.8)
where
(d1 )
T1
(u) =
m+n
X
1
Ka+ (u)R(2u + )ad
d1 a .
(B.9)
a=2
Combining the Eqs. (B.5), (B.7) and (B.8) and using the following important relation
(1)
L (vk , v1 ) L(1) (vk , vk1 )L(1) (vk , vk+1 ) L(1) (vk , vL )(vk )
d d
L(1) (vk , vL )1 L(1) (vk , vk+1 )1 L(1) (vk , vk1 )1 L(1) (vk , v1 )1 c 1c L
1
c c
S(vk , {vi })b11 bLL

=
sin() (1) (vk )

T1(d1 ) (vk )
dL (2)
L
S(vk , {vi })dc11c
(vk , {vi })cb11c
bL
L
(B.10)
we arrive at Eq. (43).

References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
P.P. Kulish, E.K. Sklyanin, in: Lecture Notes in Physics, Vol. 151, Springer, Berlin, 1982, p. 61.
H.B. Thacker, Rev. Mod. Phys. 53 (1982) 253.
E.K. Sklyanin, J. Phys. A 21 (1988) 2375.
H.J. de Vega, Int. J. Mod. Phys. A 4 (1989) 2371.
V.E. Korepin, N.M. Bogoliubov, A.G. Izergin, Quantum Inverse Scattering Method and
Correlation Function, Cambridge Univ. Press, Cambridge, 1993.
C.N. Yang, C.P. Yang, Phys. Rev. 150 (1966) 321.
M. Takahashi, Prog. Theor. Phys. 46 (1971) 1388.
A. Klmper, R.Z. Bariev, Nucl. Phys. B 458 (1996) 623.
P.A. Pearce, A. Klmper, Phys. Rev. Lett. 66 (1991) 623.
C. Destri, H.J. de Vega, Nucl. Phys. B 438 (1995) 413.
E.H. Lieb, F.Y. Wu, Phys. Rev. Lett. 20 (1968) 1445.
B. Sutherland, Phys. Rev. B12 (1975) 3795.
P.B. Wiegmann, JETP Lett. 31 (1980) 364.
N. Andrei, Phys. Rev. Lett. 45 (1980) 379.
N. Andrei, C. Destri, Phys. Rev. Lett. 52 (1984) 364.
P.B. Wiegmann, A.M. Tsvelik, JETP Lett. 38 (1983) 596.
A.M Tsvelik, J. Phys. C 18 (1985) 159.
V. Pasquier, H. Saleur, Nucl. Phys. B 316 (1990) 523.
B.Y. Hou, K.J. Shi, Z.X. Yang, R.H. Yue, J. Phys. A 24 (1991) 3825.
H.J. de Vega, E. Lopes, Phys. Rev. Lett. 67 (1991) 489.
A. Foerster, M. Karowski, Nucl. Phys. B 396 (1993) 611.
A. Foerster, M. Karowski, Nucl. Phys. B 408 [FS] (1993) 512.
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
R.H. Yue, H. Fan, B.Y. Hou, Nucl. Phys. B 462 (1996) 167.
C. Destri, H.J. de Vega, Nucl. Phys. B 361 (1992).
C. Destri, H.J. de Vega, Nucl. Phys. B 374 (1992) 692.
S.M. Fei, R.H. Yue, J. Phys. A 27 (1994) 3715.
H.J. de Vega, A. Gonzalez-Ruiz, Nucl. Phys. B 417 (1994) 553.
A. Gonzalez-Ruiz, Nucl. Phys. B 424 [FS] (1994) 468.
N. Andrei, H. Johannesson, Phys. Lett. A 100 (1984) 108.
P.A. Bares, cond-mat/9412011.
G. Bedrfitg, F.H.L. Eler, H. Frahm, Phys. Rev. Lett. 77 (1996) 5098.
G. Bedrfitg, F.H.L. Eler, H. Frahm, Nucl. Phys. B 489 (1997) 697.
A. Foerster, J. Links, A. Tonel, Nucl. Phys. B 552 [FS] (1999) 707.
J. Links, A. Foerster, J. Phys. A 32 (1999) 147.
H. Frahm, A.A. Zvyagin, J. Phys. C 9 (1997) 9939.
Y.P. Wang, Phys. Rev. B 56 (1997) 14045.
Y.P. Wang, J.H. Dai, Z.N. Hu, F.C. Pu, Phys. Rev. Lett. 79 (1997) 1901.
Z.N. Hu, F.C. Pu, Y. Wang, J. Phys. A 31 (1998) 5241.
G. Bedrfitg, H. Frahm, J. Phys. A 32 (1999) 4585.
J.H. Perk, C.L. Schultz, Phys. Lett. A 84 (1981) 407.
E. Lopes, Nucl. Phys. B 370 (1992) 636.
L. Mezincescue, R.I. Nepomechie, J. Phys. A 24 (1991) L19.
L. Mezincescue, R.I. Nepomechie, Int. J. Mod. Phys. Lett. A 6 (1991) 2497.
R.H. Yue, H. Liang, High Energy Physics and Nuclear Physics 20 (6) (1996) 514.
F.H. Essler, V.E. Korepin, K. Schoutens, Phys. Rev. Lett. 68 (1992) 2960.
729

CUMULATIVE AUTHOR INDEX B581B586
Abada, A.
Abrikosov Jr., A.A.
Accomando, E.
Achiman, Y.
Ahn, C.
Akemann, G.
Aliev, T.M.
Alvarez, O.
Alvarez, O.
Alvarez, O.
Amors, G.
Anastasiou, C.
Andreev, O.
Antoniadis, I.
Arhrib, A.
Arnowitt, R.
Arutyunov, G.
B585 (2000) 45
B586 (2000) 589
B585 (2000) 124
B584 (2000) 46
B586 (2000) 611
B583 (2000) 739
B585 (2000) 275
B582 (2000) 139
B584 (2000) 659
B584 (2000) 682
B585 (2000) 293
B585 (2000) 763
B583 (2000) 145
B583 (2000) 35
B581 (2000) 34
B585 (2000) 124
B586 (2000) 547
Balog, J.
Bandos, I.
Br, O.
Barbieri, R.
Barcel, C.
Barr, S.M.
Barros, M.
Baulieu, L.
Blanger, G.
Benakli, K.
Benaoum, H.B.
Berera, A.
Bertola, M.
Bertolini, M.
Bialas, P.
Bialas, P.
Bianchi, M.
Bijnens, J.
Bilal, A.
Binoth, T.
Blanchard, P.
Blumenhagen, R.
Blmlein, J.
Blmlein, J.
Bohm, A.R.
B583 (2000) 614

B586 (2000) 315
B581 (2000) 499
B585 (2000) 28
B584 (2000) 415
B585 (2000) 79
B584 (2000) 719
B581 (2000) 604
B581 (2000) 3
B583 (2000) 35
B585 (2000) 554
B585 (2000) 666
B581 (2000) 575
B582 (2000) 393
B581 (2000) 477
B583 (2000) 368
B584 (2000) 216
B585 (2000) 293
B582 (2000) 65
B585 (2000) 741
B583 (2000) 368
B582 (2000) 44
B581 (2000) 449
B586 (2000) 349
B581 (2000) 91
Bonini, M.
Boudjema, F.
Braaten, E.
Brignole, A.
Brink, L.
Bronnikov, K.A.
Bros, J.
Bruckmann, F.
Buchbinder, I.L.
Buras, A.J.
B585 (2000) 253

B581 (2000) 3
B586 (2000) 427
B582 (2000) 759
B586 (2000) 183
B584 (2000) 436
B581 (2000) 575
B584 (2000) 589
B584 (2000) 615
B586 (2000) 397
Caffo, M.
Campbell, B.A.
Campos, I.
Capdequi Peyranre, M.
Capella, A.
Capitani, S.
Carena, M.
Carloni Calame, C.M.
Carter, G.W.
Casas, J.A.
Castellani, C.
Castro Alvaredo, O.A.
Ceresole, A.
Chakrabarti, S.K.
Chalmers, G.
Chan, Ch.S.
Chetyrkin, K.G.
Chetyrkin, K.G.
Chu, C.-S.
Chu, C.-S.
Cirillo, E.N.M.
Coleman, P.
Coste, A.
Cox, J.
Creminelli, P.
Cski, C.
Cski, C.
Cvetic, M.
Cvetic, M.
Cvetic, M.
Czyz, H.
B581 (2000) 274

B581 (2000) 240
B581 (2000) 499
B581 (2000) 34
B586 (2000) 382
B582 (2000) 762
B586 (2000) 92
B584 (2000) 459
B582 (2000) 571
B581 (2000) 61
B583 (2000) 542
B581 (2000) 643
B585 (2000) 143
B582 (2000) 627
B585 (2000) 517
B581 (2000) 156
B583 (2000) 3
B586 (2000) 56
B582 (2000) 65
B585 (2000) 193
B583 (2000) 584
B586 (2000) 641
B581 (2000) 679
B583 (2000) 331
B585 (2000) 28
B581 (2000) 309
B584 (2000) 359
B584 (2000) 149
B586 (2000) 275
B586 (2000) 287
B581 (2000) 274
732
DallAgata, G.
Damgaard, P.H.
de Azcrraga, J.A.
Delfino, G.
Denef, F.
Derendinger, J.-P.
Derkachov, S.
Diakonov, D.
Daz, M.A.
Di Clemente, V.
Donato, F.
Dorn, H.
Dorsner, I.
Dubin, A.Yu.
Dubin, A.Yu.
Dutta, B.
B585 (2000) 143

B583 (2000) 347
B581 (2000) 743
B583 (2000) 597
B581 (2000) 135
B582 (2000) 231
B583 (2000) 691
B582 (2000) 571
B583 (2000) 182
B581 (2000) 61
B581 (2000) 3
B583 (2000) 145
B585 (2000) 79
B582 (2000) 677
B584 (2000) 749
B585 (2000) 124
Eden, B.
Eguchi, T.
Einhorn, M.B.
Ellis, J.
Engelhardt, M.
Engelhardt, M.
Enqvist, K.
Epele, L.N.
Erdmenger, J.
Erickson, J.K.
Erler, J.
Erlich, J.
Erlich, J.
Espinosa, J.R.
Evans, N.
Eyras, E.
B581 (2000) 523

B586 (2000) 331
B582 (2000) 216
B586 (2000) 92
B585 (2000) 591
B585 (2000) 614
B582 (2000) 763
B583 (2000) 454
B585 (2000) 517
B582 (2000) 155
B586 (2000) 73
B581 (2000) 309
B584 (2000) 359
B586 (2000) 3
B581 (2000) 391
B584 (2000) 251
Fabrizio, M.
Fanchiotti, H.
Farhi, E.
Ferrndez, A.
Ferreira, C.N.
Feruglio, F.
Finster, F.
Fortunato, S.
Fosco, C.D.
Freund, M.
Fries, R.J.
Frhlich, J.
Frolov, S.
B583 (2000) 542

B583 (2000) 454
B585 (2000) 443
B584 (2000) 719
B581 (2000) 165
B582 (2000) 759
B584 (2000) 387
B583 (2000) 368
B582 (2000) 716
B585 (2000) 105
B582 (2000) 537
B583 (2000) 381
B586 (2000) 547
Gallot, L.
Gandolfo, D.
Gannon, T.
Garcia-Garcia, A.M.
Garca Canal, C.A.
Garousi, M.R.
Gates Jr., S.J.
Gavai, R.V.
B586 (2000) 206

B583 (2000) 368
B581 (2000) 679
B586 (2000) 668
B583 (2000) 454
B584 (2000) 284
B584 (2000) 109
B586 (2000) 475
Geyer, B.
Gherghetta, T.
Ghoshal, D.
Gitman, D.M.
Glover, E.W.N.
Godbole, R.
Gonnella, G.
Gonzlez-Sprinberg, G.A.
Gorini, V.
Grlich, L.
Gorsky, A.
Govindarajan, T.R.
Graham, N.
Grandjean, O.
Greene, B.R.
Grojean, C.
Grzadkowski, B.
Grzadkowski, B.
Guan, X.-W.
Guimares, M.E.X.
Gukov, S.
Gukov, S.
Gunion, J.F.
Guruswamy, S.
B581 (2000) 341

B586 (2000) 141
B584 (2000) 300
B584 (2000) 615
B585 (2000) 763
B581 (2000) 3
B583 (2000) 584
B582 (2000) 3
B581 (2000) 575
B582 (2000) 44
B584 (2000) 197
B583 (2000) 291
B585 (2000) 443
B583 (2000) 381
B584 (2000) 480
B584 (2000) 359
B583 (2000) 49
B585 (2000) 3
B583 (2000) 721
B581 (2000) 165
B584 (2000) 69
B584 (2000) 109
B583 (2000) 49
B583 (2000) 475
Harlander, R.V.
Harmark, T.
Harris, M.G.
Harshman, N.L.
Hart, A.
Hassan, S.F.
Heinrich, G.
Heinzl, T.
Helayl-Neto, J.A.
Heller, U.M.
Herdeiro, C.A.R.
Hikami, K.
Hioki, Z.
Hisano, J.
Holland, K.
Hollik, W.
Hollowood, T.J.
Hollowood, T.J.
Hong, D.K.
Hormuzdiar, J.
Hou, B.-Y.
Howe, P.S.
Hsu, S.D.H.
Huber, P.
Huiszoon, L.R.
Hull, C.M.
B586 (2000) 56
B585 (2000) 567
B586 (2000) 518
B581 (2000) 91
B586 (2000) 443
B583 (2000) 431
B585 (2000) 741
B584 (2000) 589
B581 (2000) 165
B583 (2000) 347
B582 (2000) 363
B581 (2000) 761
B585 (2000) 3
B584 (2000) 3
B583 (2000) 331
B581 (2000) 34
B581 (2000) 309
B584 (2000) 359
B582 (2000) 451
B581 (2000) 391
B586 (2000) 711
B581 (2000) 523
B581 (2000) 391
B585 (2000) 105
B584 (2000) 705
B583 (2000) 237
Inoue, R.
Intriligator, K.
Irvine, S.E.
Ishibashi, N.
B581 (2000) 761

B581 (2000) 257
B584 (2000) 795
B583 (2000) 159

Iso, S.
Ito, K.
Ivanov, N.Ya.
B583 (2000) 159

B586 (2000) 231
B586 (2000) 382
Jaffe, R.L.
Janik, R.A.
Jurco, B.
B585 (2000) 443

B586 (2000) 163
B584 (2000) 784
Kaidalov, A.B.
Kamimura, K.
Kaminsky, K.
Kanno, H.
Karakhanyan, D.
Kaste, P.
Kausch, H.G.
Kawai, H.
Kaya, A.
Kazama, Y.
Ketov, S.V.
Ketov, S.V.
Kiem, Y.
Kikukawa, Y.
Kim, J.E.
Kimura, Y.
Kinar, Y.
Kirschner, R.
Kiselev, V.V.
Kiselev, V.V.
Kitazawa, N.
Kitazawa, Y.
Kitazawa, Y.
Kniehl, B.A.
Kogan, I.I.
Kogut, J.B.
Krs, B.
Kotikov, A.V.
Kovacs, S.
Kovalsky, A.E.
Kramer, G.
Krauth, W.
Krykhtin, V.A.
Khn, J.H.
Kurosawa, K.
Kyae, B.
B586 (2000) 382

B585 (2000) 219
B581 (2000) 240
B586 (2000) 331
B583 (2000) 691
B582 (2000) 203
B583 (2000) 513
B583 (2000) 159
B583 (2000) 411
B584 (2000) 171
B582 (2000) 95
B582 (2000) 119
B586 (2000) 303
B584 (2000) 511
B582 (2000) 296
B581 (2000) 295
B583 (2000) 76
B583 (2000) 691
B581 (2000) 432
B585 (2000) 353
B586 (2000) 261
B581 (2000) 295
B583 (2000) 159
B582 (2000) 514
B584 (2000) 313
B582 (2000) 477
B582 (2000) 44
B582 (2000) 19
B584 (2000) 216
B585 (2000) 353
B582 (2000) 514
B584 (2000) 641
B584 (2000) 615
B586 (2000) 56
B584 (2000) 3
B582 (2000) 296
Laine, M.
Laine, M.
Langacker, P.
Lazar, M.
LeClair, A.
Lee, H.M.
Lee, J.
Lee, S.
Lerche, W.
Lerda, A.
Lesgourgues, J.
B582 (2000) 277

B586 (2000) 443
B586 (2000) 287
B581 (2000) 341
B583 (2000) 475
B582 (2000) 296
B586 (2000) 427
B586 (2000) 303
B582 (2000) 203
B586 (2000) 206
B582 (2000) 593
733
Li, G.-L.
Li, T.
Likhoded, A.K.
Lindner, M.
Lipatov, L.N.
Logan, H.E.
Lpez, A.
Losada, M.
Losada, M.
Low, I.
L, H.
L, H.
Lucas, P.
Ludwig, A.W.W.
Lunardini, C.
Lunardini, C.
Lscher, M.
Lst, D.
Ltken, C.A.
B586 (2000) 711

B582 (2000) 176
B585 (2000) 353
B585 (2000) 105
B582 (2000) 19
B586 (2000) 39
B582 (2000) 716
B582 (2000) 277
B585 (2000) 45
B585 (2000) 395
B584 (2000) 149
B586 (2000) 275
B584 (2000) 719
B583 (2000) 475
B583 (2000) 260
B584 (2000) 459
B582 (2000) 762
B582 (2000) 44
B582 (2000) 203
Macfarlane, A.J.
Mangano, M.L.
Martins, M.J.
Maru, N.
Marucho, M.
McDonald, J.
Melnikov, V.N.
Merten, C.
Metsaev, R.R.
Mikhailov, A.
Miramontes, J.L.
Misiak, M.
Moeller, N.
Montagna, G.
Montes, X.
Morel, A.
Moschella, U.
Moultaka, G.
Mouslopoulos, S.
Mukhopadhyay, B.
Mller, B.
Muramatsu, T.
B581 (2000) 743

B582 (2000) 759
B583 (2000) 721
B586 (2000) 261
B583 (2000) 454
B582 (2000) 763
B584 (2000) 436
B584 (2000) 46
B586 (2000) 183
B584 (2000) 545
B581 (2000) 643
B586 (2000) 397
B583 (2000) 105
B584 (2000) 459
B582 (2000) 259
B581 (2000) 477
B581 (2000) 575
B581 (2000) 34
B584 (2000) 313
B582 (2000) 627
B582 (2000) 537
B584 (2000) 171
Naganuma, M.
Nakamura, A.
Nastase, H.
Nastase, H.
Nepomechie, R.I.
Nersesyan, A.A.
Niclasen, R.
Nicrosini, O.
Niedermaier, M.
Niedermayer, F.
Nierste, U.
Nishino, H.
Nomura, Y.
B586 (2000) 231

B584 (2000) 528
B581 (2000) 179
B583 (2000) 211
B586 (2000) 611
B583 (2000) 671
B583 (2000) 347
B584 (2000) 459
B583 (2000) 614
B583 (2000) 614
B586 (2000) 39
B568 (2000) 491
B584 (2000) 3
734
Nurmagambetov, A.
B586 (2000) 315
Oda, H.
Oeckl, R.
Ohsawa, A.
Okada, N.
Okawa, Y.
Oleari, C.
Onishchenko, A.I.
zpineci, A.
B586 (2000) 231

B581 (2000) 559
B581 (2000) 73
B586 (2000) 261
B584 (2000) 329
B585 (2000) 763
B581 (2000) 432
B585 (2000) 275
Panda, S.
Papazoglou, A.
Patrascioiu, A.
Paul, P.L.
Pelizzola, A.
Ppin, C.
Pernici, M.
Pershin, V.D.
Peschanski, R.
Petersson, B.
Petkou, A.C.
Petrov, K.
Philipsen, O.
Piccinini, F.
Pickering, A.
Pilaftsis, A.
Pliszka, J.
Polyakov, A.
Pomarol, A.
Pope, C.N.
Pope, C.N.
Ptter, B.
B584 (2000) 251

B584 (2000) 313
B583 (2000) 614
B581 (2000) 156
B583 (2000) 584
B586 (2000) 641
B582 (2000) 733
B584 (2000) 615
B586 (2000) 163
B581 (2000) 477
B586 (2000) 547
B581 (2000) 477
B586 (2000) 443
B584 (2000) 459
B581 (2000) 523
B586 (2000) 92
B583 (2000) 49
B581 (2000) 116
B586 (2000) 141
B584 (2000) 149
B586 (2000) 275
B582 (2000) 514
Quirs, M.
Quirs, M.
B581 (2000) 61
B583 (2000) 35
Radu, E.
Rasmussen, J.
Ravindran, V.
Recknagel, A.
Reinhardt, H.
Reisz, T.
Remiddi, E.
Restrepo, D.A.
Rtey, A.
Robaschik, D.
Rosier-Lees, S.
Ross, G.G.
Rossi, G.
Ruelle, P.
Rummukainen, K.
Russo, R.
Russo, R.
Rychkov, V.
B585 (2000) 637

B582 (2000) 649
B586 (2000) 349
B583 (2000) 381
B585 (2000) 591
B581 (2000) 477
B581 (2000) 274
B583 (2000) 182
B583 (2000) 3
B581 (2000) 449
B581 (2000) 3
B584 (2000) 313
B584 (2000) 216
B581 (2000) 679
B583 (2000) 347
B582 (2000) 65
B585 (2000) 193
B581 (2000) 116
Sadrzadeh, A.
Saito, T.
Sakai, N.
Sakai, S.
Samuel, S.
Santamaria, A.
Santiago, J.
Santoso, Y.
Satz, H.
Sauser, R.
Savc, M.
Savvidy, K.G.
Schaeffer, R.
Schfer, A.
Schalm, K.
Schaposnik, F.A.
Scharnhorst, K.
Schellekens, A.N.
Schlottmann, P.
Schmidhuber, C.
Schomerus, V.
Schreiber, E.
Schubert, C.
Schubert, C.
Schupp, P.
Schwetz, M.
Sciuto, S.
Seiler, E.
Semenoff, G.W.
Sen, A.
Sevrin, A.
Shirman, Y.
Shiu, G.
Shore, G.M.
Simn, J.
Smirnov, A.Yu.
Smoller, J.
Sokatchev, E.
Sommer, R.
Sonnenschein, J.
Sonoda, H.
Sorokin, D.
Stanev, Y.S.
Staudacher, M.
Stein, E.
Stephanov, M.A.
Strigazzi, P.
Strumia, A.
Suneeta, V.
Suzuki, H.
B586 (2000) 275

B584 (2000) 528
B586 (2000) 231
B584 (2000) 528
B585 (2000) 715
B582 (2000) 3
B584 (2000) 313
B585 (2000) 124
B583 (2000) 368
B582 (2000) 231
B585 (2000) 275
B585 (2000) 567
B581 (2000) 575
B582 (2000) 537
B584 (2000) 480
B582 (2000) 716
B581 (2000) 718
B584 (2000) 705
B586 (2000) 686
B585 (2000) 385
B583 (2000) 381
B583 (2000) 76
B585 (2000) 407
B585 (2000) 429
B584 (2000) 784
B581 (2000) 391
B585 (2000) 193
B583 (2000) 614
B582 (2000) 155
B584 (2000) 300
B581 (2000) 135
B581 (2000) 309
B584 (2000) 480
B581 (2000) 409
B585 (2000) 219
B583 (2000) 260
B584 (2000) 387
B581 (2000) 523
B582 (2000) 762
B583 (2000) 76
B585 (2000) 725
B586 (2000) 315
B584 (2000) 216
B584 (2000) 641
B582 (2000) 537
B582 (2000) 477
B586 (2000) 206
B585 (2000) 28
B583 (2000) 291
B585 (2000) 471
Talavera, P.
Tamada, M.
Tanaka, T.
Taylor, W.
Taylor, W.
B585 (2000) 293

B581 (2000) 73
B582 (2000) 259
B583 (2000) 105
B585 (2000) 171

Terashima, S.
Tok, T.
Toublan, D.
Tran, T.A.
Trentadue, L.
Tricarico, E.
Trigiante, M.
Troost, J.
Tseytlin, A.A.
Tsvelik, A.M.
B584 (2000) 329

B584 (2000) 589
B582 (2000) 477
B586 (2000) 275
B583 (2000) 307
B585 (2000) 253
B582 (2000) 393
B581 (2000) 135
B584 (2000) 233
B586 (2000) 641
Urban, J.
B586 (2000) 397
Vafa, C.
Vaidya, S.
Vainshtein, A.
Valle, J.W.F.
Vaman, D.
Vaman, D.
van der Bij, J.J.
van Neerven, W.L.
van Nieuwenhuizen, P.
Vasiliev, M.A.
Verbaarschot, J.J.M.
Verbaarschot, J.J.M.
Verbeni, M.
Verlinde, H.
Vernizzi, G.
Vidal, J.
Visser, M.
Volkov, M.S.
B584 (2000) 69
B583 (2000) 291
B584 (2000) 197
B583 (2000) 182
B581 (2000) 179
B583 (2000) 211
B585 (2000) 637
B586 (2000) 349
B581 (2000) 179
B586 (2000) 183
B582 (2000) 477
B586 (2000) 668
B583 (2000) 307
B581 (2000) 156
B583 (2000) 739
B582 (2000) 3
B584 (2000) 415
B582 (2000) 313
735
Wagner, C.E.M.
Walcher, J.
Walton, M.A.
Wang, Y.-J.
Weigel, H.
Weiss, N.
Weisz, P.
Wess, J.
West, P.C.
Wheater, J.F.
White, B.E.
Wiedemann, U.A.
Wipf, A.
Wipf, A.
Witten, E.
Witten, E.
Wittig, H.
B586 (2000) 92
B582 (2000) 203
B584 (2000) 795
B583 (2000) 671
B585 (2000) 443
B583 (2000) 76
B583 (2000) 614
B584 (2000) 784
B581 (2000) 523
B586 (2000) 518
B581 (2000) 409
B582 (2000) 409
B582 (2000) 313
B584 (2000) 589
B584 (2000) 69
B584 (2000) 109
B582 (2000) 762
Yau, S.-T.
Yue, R.-H.
Yung, A.
B584 (2000) 387

B586 (2000) 711
B584 (2000) 197
Zarembo, K.
Zayas, L.A.P.
Zee, A.
Zhang, R.-J.
Zhitnitsky, A.
Zvyagin, A.A.
Zwanziger, D.
Zwirner, F.
B582 (2000) 155

B582 (2000) 216
B585 (2000) 395
B586 (2000) 3
B582 (2000) 477
B586 (2000) 686
B581 (2000) 604
B582 (2000) 759

Complete Two-Loop Dominant Corrections To The Mass of The Lightest in The Minimal Supersymmetric Standard Model

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Complete Two-Loop Dominant Corrections To The Mass of The Lightest in The Minimal Supersymmetric Standard Model

Uploaded by

Copyright:

Available Formats

Nuclear Physics B 586 (2000) 338

Complete two-loop dominant corrections

Received 29 March 2000; accepted 26 June 2000

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

2. CP-even Higgs boson masses to two-loop order

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

which is exact up to corrections of order O(m4Z /m2A0 ). 3

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

corrections to mh0 , we take equal soft masses MQ

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

+ 12 24K 18 ln 2 Xt 3 + 16K 3 ln 2 4X t Yt + Yt2

The notations used are X t = A t + cot , Yt = A t tan , with reduced parameters z

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

3. Renormalization group resummation

where we have used At = Xt s2 + Yt c2 , A2t + 2 = Xt2 s2 + Yt2 c2 and m2H2 + 2 = m2A0 c2 .

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

The one-loop threshold correction is

and the two-loop threshold correction reads

Q0t = (mt m2t )1/3 ,

Q2 = e1/3mt ' 1.4 mt .

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

Several approximations have been made to quantities in Eqs. (3)(5), in particular we

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

In this paper we calculate radiative corrections to the lightest MSSM CP-even

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

Appendix A. One- and two-loop scalar functions

(1 x)m21 + xm22 x(1 x)p2 i

The remaining functions can be related to A0 and B0 as follows

In all one-loop expressions of radiative corrections, we adopt a (modified) minimal

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

A.2. Two-loop scalar functions

The function J is symmetric in m1 , m2 and I symmetric in m1 , m2 and m3 .

I m21 , m22 , m23 = (d 3) m23 m21 m22 I m21 , m22 , m23

m23 + m21 m22

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

where Li2 (x) is the dilogarithm function

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

with m + n = 5. In this expansion the constant number K is given by

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

Appendix B. MSSM in the leading approximation

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

where ct = cos t, st = sin t, c2t = cos 2t and

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

B.2. Renormalization group equations

2 + M 2 + A2 . Then we need one-loop RGEs for those masses

where Dn = s2 , c2 , s2 , c2 and En = s2 , s2 , s2 , s2 for n = 1, 2, 3, 4. [Here we use

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

Appendix C. One-loop self-energies

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338

where H 0 tL t1 = ct H 0 t1 t1 st H 0 t2 t1 , H 0 tR t1 = st H 0 t1 t1 + ct H 0 t2 t1 , etc. The symbols

J.R. Espinosa, R.-J. Zhang / Nuclear Physics B 586 (2000) 338