Nucl - Phys.B v.568 PDF

Nuclear Physics B 568 Ž2000.
3–39
www.elsevier.nlrlocaternpe
SUSY Higgs at the LHC: Large stop mixing effects and

associated production
a
´
G. Belanger , F. Boudjema a , K. Sridhar b
a
´
Laboratoire de Physique Theorique, LAPTH, Chemin de BelleÕue, B.P. 110, F-74941 Annecy-le-Vieux,
Cedex, France 1
b
Department of Theoretical Physics, Tata Institute of Fundamental Research, Homi Bhabha Road, 400 005
Mumbai, India
Received 3 May 1999; accepted 5 August 1999
Abstract
We revisit the effect of the large stop mixing on the decay and production of the lightest SUSY
Higgs at the LHC. We stress that whenever the inclusive two-photon signature is substantially
reduced, associated production, Wh and tth, with the subsequent decay of the Higgs into photons
is enhanced and becomes an even more important discovery channel. We also point out that these
reductions in the inclusive channel do not occur for the smallest Higgs masses where the
significance is known to be lowest. We show that in such scenarios the Higgs can be produced in
the decay of the heaviest stop. For not too heavy masses of the pseudo-scalar Higgs where the
inclusive channel is even further reduced, we show that large stop mixing also allows the
production of the pseudo-scalar Higgs through stop decays. These large mixing scenarios therefore
offer much better prospects than previously thought. As a by-product we have recalculated t˜1 t˜1) h
production at the LHC and give a first evaluation of the t˜1 t˜1) Z. q 2000 Elsevier Science B.V. All
rights reserved.
PACS: 12.60.Jv; 14.80.Cp; 14.80.Ly
1. Introduction
The most popular alternative to the Standard Model, S M , is supersymmetry which

at the moment fits in very well with all the precision data. So well in fact that some see
in the latest precision data as preferring a low Higgs mass a very good evidence for
SUSY. In fact a low mass for one of the scalar Higgses is the most robust limit of any
supersymmetric model, contrary to all the other Žs.particles of the model which may
1
´ a` l’Universite´ de Savoie.
URA 14-36 du CNRS, associee
0550-3213r00r$ - see front matter q 2000 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 4 9 7 - 6
4 ´
G. Belanger et al.r Nuclear Physics B 568 (2000) 3–39
have rather high masses. In the minimal scenario of SUSY, the light Higgs mass can not
exceed ; 130 GeV. Considering the existing LEP2 2 direct searches which indicate a
mass greater than about 90 GeV means that the lightest SUSY Higgs is confined to a
small mass range. Yet this range of Higgs masses poses considerable problems for the
™™
LHC. For a review see Ref. w5x. The dominant decay into bb is not exploitable,
especially in the inclusiÕe production channel gg h bb, due to the huge QCD
background. One therefore has to rely on the much smaller two-photon signal w6,7x.
However, especially for the LHC, the two-photon decay of the light Higgs to which
dedicated detectors are being designed constitutes a challenge. Moreover many effects
either due to the direct w8–14x or indirect Žloop. w13,15–18x contributions of the rich
SUSY spectrum enter the predictions of the two-photon rate of the supersymmetric
Higgs. These can lead to a substantial reduction of the supersymmetric Higgs signal as
compared to the standard model Higgs. Take for instance the rather simple scenario
w13,19x where all sparticles, apart from the parameters of the Higgs sector, are very
heavy and where mixing effects are negligible. This is the scenario which has been
extensively investigated by the ATLAS w20,21xrCMS w22x collaborations which leads to
the much celebrated MA –tan b Higgs discovery potential of the LHC. For short, we will
refer to this model as Class-H scenario. In this scenario the two-photon Higgs signal can
be much reduced compared to the S M Higgs especially as one lowers the mass of the
pseudo-scalar boson, A0 . Nonetheless, even in this scenario, this channel covers a large
part of the MA –tan b discovery plane, while when MA gets small so that the two-photon
signal gets too small, one can extend the discovery potential by exploiting the signatures
of the then not too heavy additional Higgses w19–22x. It is therefore important to inquire
how much the important two-photon signal can get reduced and equally important to
investigate when this reduction occurs, whether new mechanisms for Higgs production
open up or are enhanced. Could the latter then make up for the loss in the former?
Considering that a general SUSY model furnishes an almost untractable number of
parameters to give an unambiguous answer, apart from the Class-H scenario only partial
investigations w15–18x within specific models have been conducted. To quantify how the
rate of the two-photon signal can be affected as the SUSY parameters are varied, it is
instructive to take as a reference point the signal for an S M Higgs with a mass that of
the lightest SUSY Higgs. In Ref. w17x this has been done within the mSUGRA
™™
hypothesis w23,24x but considering only the dominant inclusive Higgs production
channel: gg h gg . One does find indeed that this ratio can be much smaller than
unity even for relatively large MA Žwhich is generic in mSUGRA. and hence making it
more difficult to search for the SUSY Higgs than for the same mass S M Higgs.
However, it is known w25x that within mSUGRA other channels for Higgs production
may open up, like the cascading of the heavier neutralino to a lighter one and a Higgs,
thus offering the fantastic possibility of not only discovering supersymmetry but
allowing an easy detection of the Higgs w26x before its observation in the two-photon
channel. Recently, it has been argued w18x, that even in the large MA region, the
so-called decoupling limit w27x, if one introduces large mixing in the stop sector a very
substantial reduction can also ensue in the inclusiÕe two-photon Higgs signal. This
2
The limit on the mass of the SUSY Higgs depends on the SUSY parameters. For an update on the limits
on the Higgs mass see Refs. w1–4x.
´
G. Belanger et al.r Nuclear Physics B 568 (2000) 3–39 5
effect together with the issue of the mixing in the Higgsinorgaugino sector had been
studied previously by comparing the rates with and without mixing w15,16x. It was found
that there were small regions in parameter space where the rate for the two-photon
Higgs signal could be either very much reduced or very much enhanced by the
inclusion of mixing.
When large reductions in an important channel occur it is crucial to find out how
other channels are affected. What has not been stressed in the previous studies w15–18x,
especially in the case of large mixing, is the importance of the associated Higgs
production w28–33,35x Žthe K-factor for WH in the SM is calculated in Ref. w34x. and
even if no efficient b-tagging were possible, how in these scenarios these processes can
salvage the Higgs signal. Within the S M and in the no-mixing scenarios, both CMS
w22,36,37x and ATLAS w21,38x have now shown that associated Higgs production
ŽWh,Zh and tth., with the subsequent decay of the Higgs into photons, can provide an
invaluable Higgs signal, when enough luminosity has been accumulated. This is
because, although associated production has lower rates than the inclusive channel, the
corresponding signals are not plagued by as much background. The CMS analysis w22x
for the S M Higgs shows that already with an integrated luminosity of 30 fby1 the gg l
Ž l s e, m . leads to a significance higher than 5 Žthus an observable Higgs signal. for the
range of light Higgs masses we are interested in. For a high luminosity of 100 fby1 this
significance improves to more than 10 and is higher than the significance in the
inclusive channel for practically all Higgs masses in the range of interest. Although it is
known that the ATLAS analyses are less optimistic 3 when it comes to the two-photon
signal, either in the associated or inclusive channel w21x, it remains that at high
luminosity the associated production provides a better reach in the MA –tan b plane w21x.
One should therefore also inquire in the case of the SUSY Higgs if the rates for
associated production are reduced together with the inclusive rates or if they can rather
help the discovery potential. At the same time if the rates for the SUSY Higgs are very
much affected this generally means that some sort of non-decoupling of some of the
SUSY particles is taking place. These particles should then be observed directly.
Moreover since their coupling to the Higgs can not be negligible, these same particles
could trigger Higgs production, through their decays for instance or through new
associated productions. Another important aspect to address is the impact of stop mixing
on the Higgs mass and its conjunction with the reduction in the inclusive channel.
Indeed, the significance in the inclusive two-photon channel is very much dependent on
the Higgs mass, even in the narrow range allowed by SUSY w21,22x, contrary to the
associated two-photon channel where the significances are rather flat as a function of the
Higgs mass in the range of interest w21,22x.
The present paper revisits the case of the large mixing in the stop sector w15,16,18x,
how the WhrZh and tth associated production saves the day when the inclusive channel
drops to critical levels and how other new channels for Higgs production open up. To set
the stage, Section 2 is intended as a reminder of how much a reduction in the usual light
Higgs signals can occur and is tolerable in the Class-H scenario. This will serve as a
benchmark when we study whether the other scenarios could give reductions which are
3
The differences between ATLAS and CMS are quantified in w39x.
6 ´
much worse than those obtained with lowering MA , a situation of some concern
especially if no new production mechanism is exploitable. We will also present some
approximations for evaluating the reductions due to MA which will be useful even when
we study the stop mixing case.
Our analysis of the large mixing scenario in the stop sector is contained in Section 3.
We first consider the large MA limit. While we confirm that large reductions in the
inclusive channel can occur, we point out that in most cases these are no worse than
what is obtained with a low MA in the no-mixing case. Moreover we will show that if
tan b ) 3 an increase in the inclusiÕe channel is possible. This increase is not possible
for low values of tan b as studied in w18x because the effect is associated with a too low
Higgs mass already excluded by LEP2. We also carefully analyse for which Žlight.
Higgs masses these reductions occur. We will show that contrary to the no-mixing
scenario where the signals in the inclusive channels are lowest for the lowest Higgs
masses, in the case of large stop mixing the most drastic drops in the inclusive channels
do not occur for the lightest Higgs mass possible. Indeed the effect of mixing tends to
increase the mass of the Higgs compared to its value in the absence of mixing.
Considering that the significances for the S M Higgs in the inclusive channel are lowest
for the lowest possible Higgs masses in our range, ; 90–130 GeV, means that the
largest reductions do not necessarily correspond to the lowest Higgs signal. For instance
a reduction of 0.4 may well be tolerated for a Higgs mass of 110 GeV but a reduction of
0.8 may be ’’too much’’ when it occurs in conjunction with m h s 90 GeV. More
importantly we find that at the same time as the inclusive channel decreases, the
associated production increases and has much better significances than with an
S M Higgs or with a corresponding SUSY Higgs where the stop mixing have been
switched off. We will explain why this is so. It should also be pointed out that the large
reductions in the inclusive channel occur mostly when one of the stops becomes rather
light, below about 200 GeV. In many instances, as first suggested by Refs. w40,41x,
associated t˜1 t˜1 h production can provide a new channel to search for the light Higgs. We
will quantify how much one can benefit from this additional channel. Most studies
w15,16,18x have assumed equality of all soft squark masses which almost invariably
leads to a maximal mixing angle <sin2 u t˜ s 1 <, where u t˜ is the mixing angle in the stop
sector. Maximal mixing should be viewed as a very special singular point in the large
array of the SUSY parameters and even though justified for the first two families as
suggested by the mSUGRA w24x Žfor a review see Ref. w23x. scenario is quite unnatural
for the third family especially in view of the large Yukawa coupling. We show,
nonetheless, that maximal mixing is not always required for the reductions in the
two-photon rates to occur. However, moving away even slightly from this singular
mixing angle, while not changing much the previous conclusions, can open yet another
Higgs production channel. We point out that provided m t̃ 2 is not too large, say
m t̃ 2 ( 500 GeV, so that its production rate is large, t˜2 can provide a source of Higgs
through its decay into the lighter stop thanks to a sizeable Yukawa t˜2 t˜1 h coupling. This
coupling is controlled by the same parameters that make the t˜1 t˜1 h coupling large and
™ ™™
which lead to a reduction of the inclusive channel. We will compare the rate for this new
Higgs production mechanism s Ž pp t˜2 t˜2) t˜2 t˜1) h q t˜2) t˜1 h. with the associated light-
est stop pair production mechanism s Ž pp t˜1 t˜1) h. w40,41x and show that the cascade
decay of the t˜2 can be substantial. This is akin w25x to the mixing in the higgsino–gaugino
´
™
sector which has been shown w26x to allow a direct Higgs production through the
cascade decay x 20 x 10 h. We then move to the analysis of the combined effect of
allowing for smaller pseudo-scalar masses together with large stop mixing. For moderate
MA our conclusions are little changed, the associated productions offering always a good
channel. When MA gets rather small Ž MA ; 250 GeV., the usual reduction, as compared
to the S M , in both the inclusive and associated production occurs. This is irrespective
of mixing and can be explained along the lines of what happens in the Class-H scenario.
Including the large mixing effects from the stops could decrease even further the signal
from the inclusive channel, but the same effect again helps increase the associated
production channel. Therefore the reach in this channel alone is better than what
previously studied by ATLAS w20,21x and CMS w22x in the MA –tan b plane for the
no-mixing Class-H scenario. Luckily in these situations with both a low MA and large
™ ™ ™ ™ ™
mixings in the stop sector we find that beside the new channels for Higgs Ž h. production
s Ž pp t˜2 t˜2 t˜2 t˜1 h. and s Ž pp t˜1 t˜1) h., one can also have s Ž pp t˜2 t˜2 t˜2 t˜1 A..
™
There are even instances where the pseudo-scalar Higgs triggers h production through
A Zh. Independently of the extreme mixing scenario studied here we advocate to
exploit the potentially large Yukawa coupling of the stops to search for the HiggsŽes.
through the cascade decays of these third generation squarks. In all our discussion we do
not mention rescuing the Higgs signal through its decay into bb in the associated
production w42x which would be possible provided good b-tagging is available as
discussed by ATLAS w21,43x. This is a difficult issue w44x especially at high luminosity
and further simulation studies are needed. Note also, as we will discuss in the next
section, that there can be scenarios where the branching ratio of the Higgs into bb also
is reduced. We will not investigate this possibility. Section 4 gives our conclusions.
2. A warm up: Variation with M A , the no mixing case
In order to compare the various effects of lowering the masses of the SUSY particles,
we start by briefly reviewing the situation when the masses of all sparticles but those of
the Higgs sector are set to a high scale, M˜ S s 1 TeV. The mass of the pseudo-scalar
Higgs is let free. Moreover in the illustration we have also taken the Higgs mixing
parameter such that m s y180 GeV and the SUŽ2. gaugino mass M2 s 500 GeV with
the traditional GUT assumption on the gaugino masses which at the electroweak scale
translates as
5
M1 s tan2u W M2 . Ž 2.1 .
3
Therefore strictly speaking we have allowed rather light charginos and neutralinos. All
the tri-linear A-terms were set to zero. These kind of scenarios w13,19,20,22x, with high
masses of sfermions, have been assumed in the simulation searches for the HiggsŽes. by
the ATLASrCMS Collaboration leading to the much advertised MA –tan b plots.
Meanwhile it has been known for some time that as MA increases one reaches a
decoupling regime w27x whereby at low energy only the lightest neutral Higgs appears in
the spectrum with the important property that its couplings are essentially the same as
those of the standard model. This kind of S M-like Higgs should be easiest to discover
8 ´
Fig. 1. Variation of R g ggg with MA , for tan b s 2.5 Žfull. and tan b s10 Ždotted..
at the LHC. However, as the mass of the pseudoscalar decreases the production rates of
the lightest Higgs also decrease. The reduction in the inclusive two-photon rate of the
Higgs, as compared to the S M , is defined through the ratio
G SUSY Ž h ™ gg . = BR Ž h ™ gg . .
SUSY
Ž h ™ gg . = BR Ž h ™ gg .
R g ggg s Ž 2.2 .
G SM SM
This ratio is calculated by taking the same mass for the S M Higgs as the one that is
derived for the SUSY Higgs once all the SUSY parameters are set. Throughout this
paper we use HDECAY w45x to calculate all the couplings, widths and branching ratios of
the Higgs. This program incorporates the leading two-loop corrections for the Higgs
masses following Refs. w46,47x. We show in Fig. 1 how this ratio decreases with MA .
This ratio can drop to as little as ; 30% for MA s 200 GeV and tan b s 10. Though
trivial in this case, it is useful to point for later that as the pseudo-scalar Higgs mass
decreases so do the other Higgs masses,4 therefore the most drastic drops occur for the
lowest range of the lightest Higgs, see Fig. 2. This is particularly drastic for tan b s 2.5
GeV, where the drop occurs around m h ; 90 GeV. It is for these low masses that the
significance of the S M Higgs is also lowest w20,22x and therefore for this low tan b this
would constitute the worst scenario for the discovery of the lightest SUSY Higgs
through its two-photon decay.5
What is troublesome for a low MA is that the branching fraction into two photons is
the main reason behind the drop, as shown in Fig. 2. This ratio is defined as
BRSUSY Ž h ™ gg . .
Ž h ™ gg .
Rgg s SM Ž 2.3 .
BR
For instance for MA s 200 GeV and tan b s 2.5, the ratio of the branching fraction
into photons, Rgg , is reduced to about 0.5 with respect to what it would be in the S M .
This reduction accounts for much of the reduction in R g ggg , R g ggg s 0.4. Therefore one
4
In the analysis we have required Mh )90 GeV.
5
Of course, for MA ( 2 m t there is a chance of discovering the other Higgses.
´
Fig. 2. Ža. R g ggg versus Mh as MA varies from 200 GeV to 1000 GeV for different values of tan b .
m sy180 GeV. The lowest values of R g ggg and Rgg correspond to MA s 200 GeV. Žb. R g ggg versus Rgg .
expects also a considerable drop in the Higgs signal even in the associated channel Wh
and tth with the subsequent decay of the Higgs into two photons. These channels have
been shown to be invaluable w20–22,38x especially when a high luminosity has been
accumulated.
To get an understanding of these gross features and compare with what happens in
other scenarios, it is worth discussing how the various couplings, tth, bbh and WhrZh
that enter both the associated production and, at the loop level, the inclusive production
are influenced by a change in MA . This is best illustrated and most transparent in the
large MA , so-called decoupling, limit which has been shown to be already operative at
200 GeV w27x. Take the tth coupling which differs from the S M by the factor R,
g cos a
Vt t h s R mt with R s , Ž 2.4 .
2 MW sin b
where a is the usual angle that appears in the diagonalization of the CP-even neutral
Higgs mass matrix. As was shown elsewhere w48,49x, in this limit and up to radiatiÕe
corrections we may introduce the factor r,
tan a tan b s y Ž 1 q r . with r < 1, Ž 2.5 .

10 ´
where r collects all MA dependence and other radiative corrections which also occur
in the computation of the Higgs masses. Neglecting the latter we have
2 MZ2 tan2b y 1
r, 0 0. Ž 2.6 .
MA2 tan2b q 1
Then the reduction factor which appears in tth is
2
1 q tan2b r
R s 2 2
, R,1y . Ž 2.7 .
1 q tan b q r q 2 r 1 q tan2b
Likewise it is found that in hbb there is an enhancement factor which especially, for
larger values of tan b , is more substantial than the reduction in the top vertex
tan2b
Rbbh , 1 q r . Ž 2.8 .
1 q tan2b
On the other hand, the WWhrZZh vertex, controlled by sinŽ a y b ., is much less
affected: it only shows a quadratic dependence on r:
r2 tan2b
RV V h , 1 y 2
. Ž 2.9 .
2 Ž 1 q tan2b .
™
In the S MG Ž h gg . is dominated by the W loop which interferes destructively
with the top. Since, in this scenario the dominant W coupling is hardly affected at
™
moderate MA the little change in the top Žbottom. coupling has negligible effect on
G Ž h gg .. However, this is not the case for the branching fraction into photons. Here,
since the total width is dominated by the width into bb, which is larger than in the S M ,
™
the branching ratio into photons will be reduced, especially as tan b increases, see Eq.
Ž2.8.. On the other hand, we expect a slight decrease in the G Ž h gg .. This is because
™
vertex is reduced. Therefore the main effect in the production rate pp h gg is due
to the reduction in Br Ž h gg .. This very crude argument gives the correct order of
™™
it is dominated by the top loop in the S M , and therefore it is reduced roughly as the tth
magnitude in the different drops in R g ggg and Rgg shown in the figures. Writing for
example
G SM Ž h ™ bb . , 1 y 4 M 2
tan2b Ž tan2b y 1 .
Ž h ™ bb .
Z
Rgg , 1 y 2 2 Ž 2.10 .
G SUSY M A Ž 1 q tan2b .
we recover Rgg s 0.483 for MA s 200 GeV and tan b s 2.5 which compares very well
with the full calculation. Moreover in a first approximation, the change in the width into
gluons can be mostly accounted for by the change in the tth vertex, in which case we
may write
4 MZ2 tan2b y 1
R g ggg , 1 y . Ž 2.11 .
MA2 1 q tan2b
For larger tan b and especially for low values of MA the approximation is acceptable
but not as good. This is partly due to the effect of radiative corrections on the h bb ™
´
coupling through the diagonalisation of the neutral Higgs mass matrices.6 Especially for
large tan b these corrections are no longer so suppressed compared to the MA2 correc-
tions w17,49–51x. In our case the effect is rather marginal since the only mixing
parameter, m , is rather small compared to the SUSY scale. However, let us stress that in
all the analyses in this paper even when considering large values of the tri-linear
coupling Žsee next section. the branching ratio into bb is hardly affected. Because our
aim is to concentrate on the effect of the tri-linear coupling of the top sector we do not,
in the present paper, analyse the case with very large tan b as these would require to
analyse the sbottom sector and also for large m possible reductions in the h bb
branching ratio.
™
When considering the associated channels, beside the reduction in the two-photon
branching ratio, a further, even though slight, reduction factor affects tth production
while we expect Wh to be much less affected. This is borne out by the numerical
analysis shown in Figs. 1–3. Once again we define, for the associated productions, ratios
normalized to the S M rates for the same Higgs mass:
s SUSY Ž pp ™ Wh. = BR Ž h ™ gg .
SUSY
Ž pp ™ Wh . = BR Ž h ™ gg .
R Wgg s Ž 2.12 .
s SM SM
™
and similarly for the associated top production: R t tgg . At the level of the cross sections,
s Ž pp Wh,Zh,tth., the ratios are assumed to be given by the ratios of the squares of
the WWh and tth couplings. We clearly see, Figs. 1–3, that a lowering of MA in case of
no mixing not only results in a lowering of m h but also in a reduction of both the
inclusive and associated two photon channels, as compared to the S M signal for the
same m h . The worst hit channels are the direct production and the tth. The Wh channel
is slightly less affected. Note that ttgg versus R g ggg shows almost no tan b depen-
dence, Fig. 3. This is due to the dominance of the top loop in the gg h production,
controlled by the same vertex that enters the associated tth cross section. These
™
reductions occur for the lightest Higgs mass and are due essentially to the drop in the
branching ratio of the Higgs into two photons, see Figs. 1–3. Since as we pointed earlier
the significances in the associated channels are rather flat with respect to the Higgs mass
in the range we are interested in, this explains why the 5s discovery region based on the
associated channel in the MA –tan b plane w20,22x are almost independent of tan b
Ž2.5 ( tan b ( 10.. On the other hand, the discovery region based on the inclusive
channel shows a strong difference between low and high tan b values. This is due
essentially to the low significances for low Higgs masses which translates into low
significances for tan b in case of no mixing for a fixed value of MA . Therefore although
the reduction due to a low MA is slightly worse for tan b s 10 than for tan b s 2.5, Fig.
1, the significance in the direct channel is much better for tan b s 10 Ž m h ; 110 GeV.
than for tan b s 2.5 Ž m h ; 93 GeV.. This observation is to be kept in mind and shows
the importance of localising where in terms of m h any reduction, especially, in the
inclusive direct channels occurs. Take for instance the CMS analysis w22x. It is found
that already with a low luminosity of 30 fby1 the significance for the S M Higgs is
6
In this discussion this applies especially to the off-diagonal terms of the Higgs mass matrix.
12 ´
Fig. 3. Variation of R g ggg versus R Wgg , R t tgg with 200( MA (1000 GeV and tan b s 2.5 Žfull., 10 Ždotted..
larger in the associated channel than in the inclusive channel for m h - 105 GeV and is
above 5. Translated to Class-H this means that for MA 0 450 GeV associated production
allows observability of the light Higgs for all values tan b s 2.5–10 whereas direct
production extends the reach in MA for tan b s 10 Ž MA ; 400 GeV.. With a higher
luminosity of 100 fby1 , the CMS analysis shows that the reach in MA is better in the
associated channel for all values of tan b and especially for low tan b . In terms of the
ratio Rgg , this analysis translates into discovery for Rgg ) 0.4 corresponding to MA ) 220
GeV Ževen slightly better for tan b s 2.5, see Fig. 3.. Note that one can recover the
obsvervability region of the SUSY Higgs of the CMS analysis by combining our results
for the ratios R with their analysis for the S M . As stated earlier the ATLAS w21x
analysis requires higher luminosities and the above numbers correspond roughly to a
luminosity of 300 fby1 to take full advantage of the associated production.7
7
For lower luminosities the ATLAS significances in the associated channels are based on a Poisson
statistics. We thank Guillaume Eynard for providing us with his code and the ‘‘data’’ for the SM Higgs in the
separate channels WhrZh and tth, see also Ref. w38x.
´
3. Mixing in the stop sector
To discuss the stop sector and define our conventions, we turn to the weak eigenstate
basis where the mass matrix in the t˜L ,t˜R involves the SUSY soft-breaking masses: the
common SUŽ2. mass m ˜ Q˜ 3 and the UŽ1. mass m˜ U˜3 R , beside the mixing, m˜ 2t˜L R
4
˜ 2Q˜ 3 q m2t q 12 MZ2 1 y
m2t̃ L s m ž sin2u W cos Ž 2 b .
/ Ž 3.1 .
3
˜ U2˜3 R q m2t q 23 MZ2 sin2u W cos Ž 2 b . ,

m2t̃ R s m
m
m2t̃ L R s ym t A t q
ž / ' ym t A˜t . Ž 3.2 .
tan b
One sees that apart from the soft SUSY-breaking parameters, m ˜ Q˜ 3 , m˜ U˜3 R and the
tri-linear top term Ž A t ., there appears also the ubiquitous tan b and the higgsino mass
term m.
The stop mass eigenstates are defined through the mixing angle u t̃ , with the lightest
stop, t˜1 ,
t˜1 s cos u t˜ t˜L q sin u t˜ t˜R . Ž 3.3 .

It is quite useful to express the mixing angle as w52,53x Žfor a compendium on the
phenomenology of third generation sfermions see, Ref. w54x.
y2 m t A˜t 2 m2t̃ L R
tan Ž 2 u t˜ . s or sin Ž 2 u t˜ . s .
MZ2 cos2 b 8 sW2 m2t˜1 y m2t˜2
m2Q˜ 3 y mU2˜3 R q
˜ ˜
2 ž 1y
3 /
Ž 3.4 .
For further reference note, in the case of equal soft SUSY breaking masses for the
left and right sector of the stop Ž m˜ 2Q˜ 3 s m˜ U2˜3 R ., that apart from the case of extremely
small mixing A˜t s O Ž MZr10., one has maximal mixing: sinŽ2 u t˜. , 1. In this case we
have
m t 12 A˜t tan b 2 q 1
tan Ž 2 u t̃ . , . Ž 3.5 .
MZ MZ tan b 2 y 1
3.1. The t˜1 t˜1 h Õertex
Mixing in the stop sector not only allows one of the stops to be rather light, but this
light stop can have rather large Yukawa couplings. Let us therefore discuss this
coupling. The stop–stop Higgs couplings, like the stop mass matrix, emerge essentially
from the F-terms in the scalar potential Žthere is a residual D-term component A MZ2 ..
14 ´
With the angle a in the Higgs mixing matrix, the t˜1 t˜1 h coupling is Žwe write the
potential.
m t cos a
Vt˜1 t˜1 h s yg
MW sin b ž Ž A t y mtan a . sin u t˜cos u t˜ y m t
MZ2 sin b 1 2 2
q
m t cos a
sin Ž a q b .
žž 2
y
3
sin2u W cos 2u t˜ q
/ 3
sin2u W sin2u t˜
// .
Ž 3.6 .
The vertex does involve some important parameters which stem from the Higgs
sector, notably the angle a . In the decoupling limit w27x which we are most interested in
and up to radiative corrections Ž3.6. reads
1 mt m r
Vt˜1 t˜1 h s qgR
MW ½ m2t q sin u t˜cos u t˜ sin u t˜cos u t˜ m 2t˜1 y m2t˜2 y
ž ž / tan b /
1 2 2
qMZ2 Ž Ž 2 q r . cos 2b y 1 .
žž 2
y
3
sin2u W cos 2u t˜ q
/ 3
sin2u W sin2u t˜
/5 .
Ž 3.7 .
We see that in the limit r < 1 where r is neglected, the t˜1 t˜1 h very much simplifies.
Note that neglecting the correction due to r, the coupling no longer depends on m.
Notice also that Eq. Ž3.7. shows that this correction is reduced as tan b gets larger.
Discarding the r correction altogether, we end up with the compact formula
2 2
g ž m˜ y m˜ / q m q M
t1 t2 1 2
Vt˜1 t˜1 h ,
MW ž sin2 Ž 2 u t˜ .
4
2
t
2
Z cos Ž2 b . žž 2
y
3
sin2u W /
2
=cos 2u t˜ q
3
sin2u W sin2u t˜
// . Ž 3.8 .
We also confirm that the tan b dependence in the vertex is also hardly noticeable. Eq.
Ž3.8. makes it clear that even for maximal mixing, sin2 2 u t̃ ; 1 the contribution of the
stops and that of the top cancel each other thus leading to a very small vertex. The dip
occurs for values of the mixing angle such that
2
4 m 2t
sin 2 u t̃ , . Ž 3.9 .
m 2t˜2 y m2t˜1
On the other hand, when the mixing is negligible, the vertex is accounted for almost
entirely by the top mass and therefore has the same strength as the tth vertex.
The t˜2 t˜2 h vertex can be obtained from t˜1 t˜1 h by sin u t˜ cos u t˜ and m t˜1 m t˜2 .
Therefore if the t˜2 t˜2 h and t˜1 t˜1 h vertices were to be added, the mixing terms do not
l l
survive, as expected since the latter mix the left and right states. This is to be kept in
™ ™
mind. In situations where the stop masses are of the order of the top mass so that they
both contribute to h gg or h gg , the effect of mixing will, to a large extent, be
washed away.
´
Already at this point we can attempt to predict the general features in R g ggg and Rgg
that will be introduced by large mixing in the stop sector. Consider the large MA limit
where the t˜1 t˜1 h vertex is most transparent, see Eq. Ž3.8.. Naturally the stop will
™ ™
contribute if its mass is not too large and if its coupling to the Higgs is also large. When
there is no mixing, only the diagonal m 2t term in Eq. Ž3.8. will, in both G Ž h gg . and
™
G Ž h gg ., interfere constructiÕely with the top quark contribution. We therefore
expect an enhancement of G Ž h gg ., that is of the inclusive production. On the other
™
hand, the fact that the toprstop loops and W interfere destructively, means that
G Ž h gg . will get smaller. Nonetheless since the W loop is much larger than the top
loop, the reduction in the two-photon decay width will be modest compared to the
enhancement in the two gluon width. Considering that at large MA the width into bb
™™
Žthus the total width. is hardly affected by mixing and hence sensibly the same as in the
S M , direct production s Ž pp h gg . is enhanced. At the same time associated
WhrZh and tth with the subsequent two-photon decay of the Higgs will be reduced
somehow. For moderate mixing the t˜1 t˜1 h vertex gets vanishingly small: here no effect is
to be expected, either in any of the associated productions nor in the direct production.
When the mixing gets very large so that now, it is the term in m 2t̃ 2 in Eq. Ž3.8. which
dominates, the sign of the interferences between the stop and the top quark loop gets
reversed. In this situation direct production can get extremely small, the stop loop
cancelling the top loop. In the two photon decay, on the other hand when this
cancellation takes place it still leaves the large W contribution. Nonetheless, the increase
in Rgg will be modest compared to the dramatic decrease in R g ggg . Since the total
width is hardly affected by these mixing effects the direct inclusive production will be
much reduced. However, associated WhrZh and tth gets enhanced in these situations.
3.2. pp ™ t˜ t˜ h at the LHC

)
1 1
Because t˜1 is relatively light and its coupling to the Higgs Ž h. large, associated stop
cross sections can, exceptionally, be of the order of that of the associated top cross
section w40,41x or even larger. At the LHC this cross section is essentially induced by
gluon gluon fusion and is therefore directly proportional to the square of the t˜1 t˜1 h
vertex. We have recalculated this cross section with the help of a modified version of
CompHep w55–58x to properly take into account the radiative corrections to the Higgs
mass and couplings. For our analysis we have found it useful to calculate the cross
section at the LHC by taking, as a reference point, the m2t term only in the t˜1 t˜1 h vertex,
Eq. Ž3.8.. The cross section can then be easily evaluated by specifying as independent
input parameters m t̃ 1 and m h only. The corresponding cross sections are shown in Figs.
4 and 5. We have made a polynomial fit, in the variables m h y m t˜1 to these cross
sections that reproduces the full results with a precision better than 2%, which is well
within the uncertainty due to the choice of scale and structure function. Once a set of
SUSY parameters is given, apart from the stop masses and tan b it will also furnish the
corresponding Higgs mass, m h , and the proper t˜1 t˜1 h vertex can be evaluated. One can
™
then properly normalise our cross sections. Considering the relative complexity of the
pp t˜1 t˜1 h cross section this method is much more efficient when we are scanning over
™
many SUSY parameters as done in the present analysis since we do not have to
recalculate the pp t˜1 t˜1 h for each scan. Our results agree with those shown in Refs.
16 ´
Fig. 4. t˜1 t˜1 h at the LHC as a function of m t˜1 and for a range of SUSY Higgs masses. The t˜1 t˜1 h vertex is set in
the limit of large MA with no mixing and no D-term, see text for how to normalise it when the SUSY
parameters are fixed. Also shown is t˜1 t˜1 Z. For the latter the vertex has been computed with cos 2u t˜ s1r2, i.e.
maximal mixing. For other values of the mixing, rescale by using the vertex Žcos 2u t̃ r2y 23 sW 2 .
. We have taken
the CTEQ4 structure function with a scale set at the invariant mass of the subprocess.
w40,41x as well as in Ref. w59x, however the largest cross sections shown in w40,41x do
not pass our constraint on the Higgs mass m h ) 90 GeV andror Dr Žsee below..
3.3. The t˜2 t˜1 h,A,H Õertex and t˜2 ™ t˜ h,H,A

1
The t˜1 t˜2 h vertex may be cast into

1 cos2 u t̃ 2 mt m r
Vt˜1 t˜2 h s qgR
MW ½ 4 ž sin2 u t˜ m2t˜1 y m2t˜2 y
ž / tan b /
´
2 1
qMZ2 sin2 u t˜ Ž cos2 b q rcos 2b . ž 3
sin2u W y
4 /5
™ qgR 4 M1 W
sin4u t˜ m2t˜1 y m2t˜2 .
ž / Ž 3.10 .
It is crucial to note that within the approximation of neglecting the r terms and the
D-terms, this coupling does not survive in the maximal mixing scenario, it is propor-
™
tional to sinŽ4u t̃ .. Nonetheless because of it its Yukawa nature this can be a rather large
coupling and therefore phase-space allowing Br Ž t˜2 t˜1 h. can be large. Considering
Fig. 5. As in Fig. 4 but as function of the Higgs mass. The tth is also shown for comparison for the same set
of structure functions and by only taking into account the gluon gluon processes. For the latter including the
small quark initiated process, our results agree with Ref. w60x. Also shown is ttZ.
18 ´
Fig. 6. Next-to-leading order t˜1 pair production at the LHC, for three representative values of the gluino mass.
We used the code given to us by Michael Spira w61,62x.
that t˜2 pair production exceeds 1pb for m t˜2 ( 500 GeV Žsee Fig. 6., t˜2 can trigger Higgs
™
Ž h. production.8
Contrary to x 20 x 10 h whose branching ratio can reach 100% for some of the SUSY
parametersw25x and thus very efficiently triggers Higgs production, Br Ž t˜2 t˜1 h. w26x ™
™ ™
can never reach 100%. This is because, independently of other decay modes into the b,b˜
™
sector, there is always the competing larger decay rate t˜2 t˜1 Z. Indeed when the
splitting is large t˜2 t˜1 Z can be approximated by t˜2 t˜1 f 0 , f 0 being the neutral
Goldstone boson, with an effective coupling g 4 M1 W sin2 u t˜Ž m2t˜2 y m2t˜1 . s gr2 MW m t Ž A t q
mrtan b ..
™
When MA is small, t˜2 can also provide a welcome source of pseudo-scalar Žand
heavy Higgses. through t˜2 t˜1 A, H. What’s more, the strength of the t˜2 t˜1 A ™
8
t˜1 t˜2 is completely negligible at the LHC w61x. ŽFor review on QCD corrections to SUSY particles
production at hadron colliders, see Ref. w62x..
´
coupling does not depend on the stop mixing angle:

mt At
Vt˜1 t˜2 A s ig
2 MW ž tan b
ym .
/ Ž 3.11 .
™
The decay t˜2 t˜1 H is generally smaller and vanishes when the mixing is maximal.
In the decoupling limit this becomes
mt At
Vt˜1 t˜2 H ; igcos2 u t˜
2 MW ž tan b /
ym . Ž 3.12 .
Of course, to calculate the branching ratios of t˜2 into Higgses we have evaluated all
possible widths of t˜2 , without QCD corrections though. We have checked our numbers
against those of w63–67x as well as the output of GRACE w69x. ŽFor a description of the
program see, for example, Ref. w68x..
For a general recent review of stop decays see Refs. w54,70x. ŽRef. w70x contains a
nice review of QCD corrections to stop decays..
™
For further reference note that whenever stop mixing is not excessively small, we can
reach Br Ž t˜2 t˜1 h. ; 10%. Associated t˜2 t˜1 A in mSUGRA has also been entertained
recently w59x. However, in the mSUGRA scenario the mixing is generally not large and
the stops are usually heavy leading to small cross section for Higgs production through
stops. But then in this same scenario large drops in the inclusive production due to stop
mixing hardly occur either.
3.4. Constraints from low Higgs masses, Dr and CCB
Large values of the t˜1 t˜1 h vertex which lead to the largest pp t˜1 t˜1 h and the ™
sharpest drop in R g ggg occur when the mixing is large with a large splitting between the
two stop physical masses. It is, however, for this configuration that one has some strong
constraints which preclude the highest values of the cross section. For instance, one has
to be wary that imposing a lower bound on the Higgs mass, from its non-observation at
LEP2 say, can restrict drastically the sin2 u t˜–m t˜2 parameter space. This constraint is very
much dependent on tan b . Much less dependent on tan b but a quite powerful one, for
the values of m t̃ 1 that we have entertained, is the constraint coming from Dr w71–
75,77–79x. ŽFor earlier analyses, see Ref. w76x.. Taking the present limit Dr - 0.0013
applicable to New Physics with a light Higgs w80x, which here means essentially the
contribution from stops and sbottoms Žand marginally the Higgs sector 9 . generally
excludes region of the parameter space where the t˜1 t˜1 h is largest. In our Dr constraint
we have relied on the two-loop calculation of w81,82x, which can enhance Dr by as
much as 10% even with a heavy gluino.
One more constraint one needs to mention. In the stop sector and in the presence of
large mixing as is the case here, one often has to check whether the parameters do not
induce colour and charge breaking global minima ŽCCB. w83–88x. ŽFor a recent
summary see Ref. w89x.. It has been argued that the constraints based on the global
minima may be too restrictive w90–92x. It was shown that for a wide range of
9
For light stops in the decoupling limit the sbottom–stop contribution when substantial gives a positive
contribution, whereas the Higgs sector contributes a negligible negative contribution.
20 ´
parameters, the global CCB minimum becomes irrelevant on the ground that the time
required to reach the lowest energy state exceeds the present age of the universe. Taking
the tunneling rate into account results in a milder constraint which may be approximated
w90–92x by
A2t q 3 m2 - 7.5 MQ2˜ 3 q Mt˜2R .

ž / Ž 3.13 .
When presenting our results we will, unless otherwise stated, impose the limits
m h ) 90 GeV, Dr - 0.0013 together with the mild CCB constraint Eq. Ž3.13.. Consider-
ing that the CCB constraint is rather uncertain, it is worth pointing out that our CCB
constraint hardly precludes points which are not already rejected by Dr and m h .
Apart from the indirect constraints we also imposed, the model independent limit,
m t̃ 1,m b˜ 1 ) 80 GeV from present direct searches w93x. Our limit on the stop, is however
superseded by our constraint that the lightest neutralino is the LSP and that t˜1 c x 10
w94x is always open. When taking m s yM2 s 250 GeV with the unification condition,
™
x 10 , 120 GeV and thus m t˜1 ) 120 GeV.
3.5. tan b s 2.5
We start our analysis by considering the case with tan b s 2.5. Although this value is
not far from being excluded by the direct LEP2 searches w1–4x, depending on the exact
SUSY parameters, we study it here in order to compare our results with those in w18x and
to show a feature which is not present for higher values of tan b .
3.5.1. The case of a common mass in the third generation squark sector
We first revisit the case w15,16,18x of allowing, at the electroweak scale, a common
mass for all the supersymmetric masses of the third generation squarks: m ˜ Q˜ 3 s m˜ U˜3 R s
m˜ D˜ 3 R ˜ 3˜
s m . Taking a common value for the&SUŽ 2. and UŽ1 .
<
masses
<
shows that unless
the effective tri-linear term is negligible, A t; 0, this leads to sin2 u t˜ s 1, see Eq. Ž3.4..
We note that contrary to what is claimed in w18x this situation, although common for the
first two generation of squarks, occurs only in exceptional situations in a model such as
mSUGRA. Moreover in mSUGRA A t is controlled almost entirely by m1r2 , the
common gaugino mass, and thus would not be excessively large w95–100x. Leaving this
aside, this assumption helps keep the number of parameters to a minimum while
concentrating on the impact of mixing. To that effect we have set, apart from the
common third family scalar quark m ˜ 3̃ which was allowed to vary in the range 100–1000
GeV and MA s 1 TeV, all other sfermion masses to 500 GeV. Moreover we have
assumed the unification condition for the gaugino masses and set the Higgsino mass
M2 s ym s 250 GeV. We then scanned over A t , y1000( A trŽGeV. ( 1000 and m ˜ 3˜ .
Note that since we are scanning over both positive and negative values of A t , some
important mixing effects sensitive to the sign of A t = m are covered even though we
have fixed the sign of m. Among the 2 = 10 4 generated point for each tan b half passed
all the constraints.
First, as shown in Fig. 7, we do confirm that the reduction in the two-photon signal in
the direct channel is most dramatic for the lowest values of the stop mass, although a
low stop mass does not always mean that a reduction has to occur. As a matter of fact
´
Fig. 7. R g ggg versus m t˜1 for tan b s 2.5, m sy250 GeV and MA s1 TeV.
there are more points that generates a low m t̃ 1 and give R g ggg 0 0.6, say, than those that
give R g ggg ( 0.6. Note that most points clustering around values corresponding to little
mixing or large stop masses. Therefore the very rare situations corresponding to very
sharp drops could be interpreted as at best unnatural. It is also worth pointing out that
values such that R g ggg 0 1.0 are not obtained for tan b s 2.5. We have verified that
while, in principle, this was possible for tan b s 2.5 this possibility was ruled out by the
requirement of having m h 0 90 GeV.
As stressed numerous times, for the intermediate mass Higgs in the direct channel
decaying into two photons, the significance depends crucially on the Higgs mass. It is
therefore important to localise for which values of the Higgs mass, the reductions are
most drastic. For tan b s 2.5 we see, Fig. 8, that this reduction gets worse, R g ggg , 0.2
for Higgs masses clustered around ; 103 GeV.
It is important to note, on the other hand, that for Higgs masses around 90 GeV
where the Ž S M . Higgs signal is most difficult to extract, the effect of the stop is rather
negligible Žhere there is no mixing hence the low mass of the Higgs which does not get
further radiative corrections.. Therefore this is a welcome point. As compared to the
case of CLASS-H with MA s 180 GeV and tan b s 2.5, for which Mh s 90 GeV, R g ggg
reaches 0.3 whereas for the same Higgs mass Žand tan b . our points cluster around one.
An even more important remark concerns the behaviour of the branching ratio into two
photons. We find, see Fig. 8, that the branching ratio into photons in this SUSY scenario
increases at the same time as the direct production decreases, in sharp contrast to what
happens in CLASS-H when MA decreases. This confirms our expectations. The R g ggg
versus Rgg can be considered as a signature of this scenario. In the corresponding
scatter plot of Fig. 8, the points fall almost along a line and shows that when
R g ggg ( 0.8, Rgg 0 1. Considering that in this large MA scenario and even in the
presence of large mixing the tth and Wh are sensibly the same as in the standard model,
the associated Higgs production with the Higgs decaying into two-photon should pose
no problem with the high luminosity LHC. We do not show the ratios for ttgg and Wgg
as these are given essentially by the ratio Rgg , see Fig. 8. To conclude, for this value of
tan b s 2.5, when 90 ( m h ( 100 GeV observability of the lightest SUSY Higgs Žh. is
quite similar to that of the S M . Above these values, if the direct production is not
possible, the branching into photons is larger than the S M and thus associated produc-
tion provides more chance of detecting the Higgs. For instance, taking the S M Higgs
22 ´
Fig. 8. As in Fig. 7 but for Ža. R g ggg versus Mh , Žb. R g ggg versus Rgg and Žc. Rgg versus Mh .
CMS analysis w22x with a luminosity of 100 fby1 as a guide, shows that it is only in the
range 100–105 GeV where values below 0.6 are possible for R g ggg that the Higgs may
not be observed in the direct channel. The same analysis shows, however, that with the
values that we obtain in the associated channels that there is no problem of cornering the
Higgs. Note that for the most critical drop in the direct channel we have obtained a
enhancement factor of up to 1.35 in the associated production. For such values even the
ATLAS simulation w21,38x with a luminosity of 100 fby1 indicates observation in the
associated channels.
Finally, another note of optimism in the case where the drop in the inclusive
production is severe is that production of h in association with stops could help also. As
´
Fig. 9. As in Fig. 7 but for R g ggg versus s Ž t˜1 t˜1 h. Žfb..
shown in Fig. 9, whenever R g ggg ( 0.6, s Ž t˜1 t˜1 h. is in excess of 100 fb and can reach as
much as ; 740 fb. As a comparison, for these extreme cases for which m h ; 100–105
GeV, one has s Ž tth. , 500 fb. Considering that, see Fig. 7, these helpful t˜1 t˜1) h cross
sections are for values of m t̃ 1 ( 250 GeV for which R g ggg ( 0.6, t˜1 with our choice of
parameters will decay exclusively into c x 10 . It remains to be seen whether this
constitutes a viable signal and whether we could use the Higgs decays into bb, which by
the way is not much affected at these low values tan b by these mixing effects. The
signal would be bb q jets q pu T . Note that the continuum t˜1 t˜1 Z is quite small. For
m t̃ 1 s 120 GeV and maximal stop mixing angle, after folding with Br Ž Z bb . the
continuum leads to a dismal raw cross section of about 1 fb.
™
3.5.2. Lifting the degeneracy in the third family scalar masses
We have already argued that the scenario with exactly equal squark masses for the
third generation is very special and even unnatural. Taking a more general framework,
™
we move away from the case of maximal mixing. As we have discussed this can open
up new possibilities, notably t˜2 t˜1 h decays. For illustration, we have taken m
˜ t˜3 R s 200
GeV, m ˜ b˜ 3 R s 500 GeV and allowed 50 ( m˜ Q˜ 3 ( 500 GeV. In order to compensate for
the deviation from maximal mixing, the trilinear coupling was allowed to vary in the
range y2000( A t ( 2000 GeV. However, very few points with < A t < 0 1200 GeV pass
our constraints, essentially from Dr . As expected the general features found in the case
of maximal mixing are still present here, even though with our parameters the drops are
not as dramatic as in the maximal mixing case. Another observation is that m h ) 105
GeV is not generated. This is because contrary to the previous case the stop masses do
not extend to 1 TeV and hence the radiative corrections to the Higgs mass are not
optimal. Nonetheless as seen in Fig. 10 a ratio R g ggg as low as 0.4 is possible and
occurs for low t˜1 masses. Again this drop occurs for a small range of Higgs masses
sensibly the same as in the maximal mixing case, m h ; 103–104 GeV, Fig. 10.
However, when this occurs one is saved by the fact that the branching ratio into photons
is larger than in the S M , Fig. 11. Moreover we still find that when R g ggg gets too
24 ´
Fig. 10. Ža. R g ggg versus m t˜1 for tan b s 2.5, m sy250 GeV and MA s1 TeV, when we allow different
scalar masses for the third generation as given, see text. Žb. As in Ža. but for R g ggg versus Mh .
™ ™ ™
small pp t˜1 t˜1) h is of the order 100 fb reaching a maximum of 200 fb when R g ggg is
lowest, Fig. 12. The main novelty here is pp t˜2 t˜2) t˜2 t˜1) h, with Br Ž t˜2 t˜1 h. s ™
O Ž10%.. Because this stems from a two-body cross section, it can lead to quite large
s Ž t˜2 t˜1) h. reaching as much 600 fb, and therefore in many instances larger than the
continuum t˜1 t˜1) h, Fig. 12. What is also worth noting is that these large cross sections do
not necessarily occur when one has large drops in the inclusive two-photon channel.
Moreover the signature in this channel should be cleaner, taking advantage of the
cascade decays of the other t˜2 starting with t˜1 Z,b˜ 1W,bx˜q 1,2 , . . . Of course there are
points where neither t˜2 t˜1 h10 nor t˜1 t˜1 h exceeds 10 fb, Fig. 12. However, in this case the
reduction in R g ggg is quite modest.
10
With our set of parameters one would expect that some points with maximal mixing are generated.
However, we have checked that these do not pass all the constraints. This explains why we never get a
vanishingly small t˜2 t˜1 h cross section.
´
Fig. 11. As in Fig. 10 but for R g ggg versus Rgg and Rgg versus Mh .
3.5.3. Stop mixing with a low MA

We have seen in Section 2, in the case of no-mixing, that as MA decreases both the
inclusive and associated two-photon channels decrease, mainly because of an increase in
the width into bb which dominates the total width and hence reduces the two-photon
branching ratio. Since hbb is hardly affected by the mixing effect, this decrease due to
MA will also be present in the case of mixing and hence reduces the significance of the
two-photon channel. This overall reduction, independent of mixing, can be evaluated by
using Eq. Ž2.10.. Likewise since the t˜1 t˜1 h vertex carries the same reduction R as the tth
vertex, see Eq. Ž3.7., an overall MA- reduction in R g ggg , which can be approximated by
Eq. Ž2.11., will take effect beside the pure large stop mixing effects that we have
discussed in the MA s 1 TeV limit. We first consider the case of a moderate MA s 350
GeV with all other masses as in Subsection 3.5.1. Fig. 13 show that those points for
which at large MA the effect of mixing were most drastic on R g ggg are not much further
reduced. They occur for masses which are sensibly the same as with the much larger
MA . Note however that the optimal values of R g ggg are reduced from about 1 to 0.8 and
occur also for Mh ; 90 GeV. This reduction is essentially what we would have obtained
by applying the factor R g ggg calculated using Eq. Ž2.11.. For this value of Mh now
detectability may be a problem if the luminosity is low, especially that the corresponding
26 ´
™ ™ t˜ t˜ h. and Žc.
™ ™ ™ ™
Fig. 12. As in Fig. 10 but for Ža. R g ggg versus s Ž pp t˜1 t˜1 h., Žb. R g ggg versus s Ž pp 2 1
s Ž pp t˜1 t˜1 h. versus s Ž pp t˜2 t˜1 h.. s Ž pp t˜2 t˜1 h. ' s Ž pp t˜2 t˜1) hq t˜2) t˜1 h..
Rgg is about 0.75, which may also preclude detection in the associated Higgs produc-
tion, Fig. 13. With these values occurring at such low values of m h , even CMS w22x with
30 fby1 will miss the Higgs, but again there should be no problem in the associated
production after collecting ; 100 fby1 . Still, whenever mixing becomes important and
reduces R g ggg significantly, associated production should be no problem. For instance
™
when R g ggg is below 0.4, Rgg 0 0.95. In these configurations Br Ž h gg . benefits ™
™
from the increase in G Ž h gg . which is not completely offset by the increase in
G Ž h bb .. In these configurations with small t˜1 , t˜1 t˜1 h could help with s Ž t˜1 t˜1 h. s
100–780 fb, Fig. 14.
´
Fig. 13. Ža. R g ggg versus Mh for tan b s 2.5, m sy250 GeV and MA s 350 GeV, with equal squark masses.
Žb. As in Ža. but for Rgg versus Mh . Žc. As in Ža. but for R g ggg versus Rgg .
We may argue that had we taken a much lower value of MA we would have
introduced a larger reduction in Rgg which may affect dangerously the associated
production. We would then be in a situation where the inclusive cross section is down
because of large mixing in the stops and the associated production small mainly because
the branching into photons is down as a result of MA being low. Note however that in
these situations we would be far from the decoupling regime, with all Higgses being
relatively light and a very light stop having large couplings to the Higgses. One
™™
consequence of this light spectrum is that, even in the case of maximal stop mixing
where t˜2 t˜1 h is inhibited, the t˜1 t˜2 A coupling is large, Eq. Ž3.11. and can be such that
it triggers t˜2 t˜1 A. This is because large mixing and large splitting between the stop,
28 ´
Fig. 14. As in Fig. 13 but for R g ggg versus s Ž pp ™ t˜ t˜ h..

1 1
allows enough phase space for a relatively light pseudo-scalar. To illustrate this fact, we
have lowered MA to 250 GeV. The gross features found for MA s 350 GeV are still
Fig. 15. Ža. Rgg versus Mh for tan b s 2.5, m sy250 and MA s 250 TeV, with equal squark masses. Žb. As
in Ža. but for R g ggg versus Rgg
´
™
present as concerns the inclusive production of h Fig. 15, with an overall reduction
factor due to Br Ž h gg . which is slightly larger. At the same time the location of the
drops are shifted to slightly lower values of m h , which is a direct consequence of a low
MA . However, as shown in Fig. 16, s Ž t˜2 t˜1 A. can be quite large and often exceeds t˜1 t˜1 h.
Note that s Ž t˜2 t˜1)A. may be large even for points where the inclusive two-photon cross
section is lowest, whereas t˜2 t˜1) h is largest for regions where the inclusive cross section
is most affected. Therefore we see that combining different channels in this scenario
offers much better prospects than in the no-mixing case with the same low value of MA .
To start with, when the direct production is very much reduced, associated production
has a better significance in the case of very large stop mixing compared to the no mixing
™ ™
case for the same MA . Another interesting point is that although the main decay of A
™ ™
will be into bb, we also find that A Zh can be substantial. For instance, the chain
s Ž pp t˜2 t˜2 t˜2 t˜1 A t˜2 t˜1 Zh. can reach as much as 350 fb Žfor this point m t˜1 s 129
™
GeV, m t˜2 s 396 GeV.. For larger values of the stop masses Ž m t˜1 s 235 GeV, m t˜2 s 525
GeV., the same chain corresponds to 43 fb. The decay t˜2 t˜1 H is also possible, but the
corresponding cross section, s Ž t˜2 t˜1) H . is below 10 fb. This is because the branching
ratio into H is about a factor cos 2 2 u t̃ down compared to the branching ratio into A,
while H and A are almost degenerate in mass, Eqs. Ž3.11. – Ž3.12.. To end this section
Fig. 16. As in Fig. 16 but for Rgg versus s Ž t˜1 t˜1 h. and s Ž t˜2 t˜1 A..
30 ´
™
let us mention that when the mass of the pseudo-scalar gets small, below 2 m t , one
should also investigate direct gg A, H production. A low mass t˜1 has no effect either
on the production or decay Žwe are in a scenario where m t̃ 2 ) MA . of A, the usual
™
channels should not be much affected. For H, one needs to critically review how the
production is affected and whether H t˜1 t˜1 can be exploited. The phenomenology is
certainly richer here and the HiggsŽes. should not be missed.
3.6. tan b s 5
We now move to a larger tan b . We go through basically the same steps as those in
Subsection 3.5. For the same scenarios we will scan over the same mass ranges. One
general new feature will have to do with the fact that for larger tan b we obviously have
larger Higgs masses. In most cases this will help. However, on the whole similar
conclusions will be reached.
3.6.1. The case of a common mass in the third generation squark sector with large MA
Again R g ggg is most affected when the t˜1 mass is smallest, Fig. 17. In the maximal
mixing case, one new feature compared to tan b s 2.5 is that the ratio R g ggg can be
larger than one, for small m t̃ 1, reaching almost ; 1.3. This is even more welcome that it
Fig. 17. Ža. R g ggg versus m t˜1 for tan b s 5, m sy250 GeV and MA s1 TeV. Žb. As in Ža. but for R g ggg
versus Mh .
´
occurs for Higgs masses in the range 92–98 GeV, Fig. 17. As a matter of fact, this is
consistent with the argument we gave earlier: in this case t˜2 is not too heavy so that the
top and t˜1 loop interfere and since the scale in the stop sector is not too high, the
radiative corrections to the lightest Higgs mass are far from maximal. Considering that,
especially in the lower end of this range, the significance in the direct channel are
usually Ž S M or no-mixing. smallest, such scenarios can make it easier to discover h
even in the direct channel. Of course, light t˜1 Žwith much heavier t˜2 . can also lead to a
much reduced R g ggg . When this happens it occurs for higher Higgs masses, clustered
around Mh s 115 GeV. Though for this range of m h significances in the direct
production are much better, for certain values of the parameters the drop is too severe:
R g ggg - 0.4. But again this occurs simultaneously with an enhanced Rgg : Rgg ) 1.2,
™
Fig. 18. Again the smaller R g ggg the larger Rgg . As with the lower tan b when the
direct production drops, s Ž pp t˜1 t˜1 h. increases. When R g ggg - 0.6 this cross section
is in excess of 100 fb up to , 650 fb, for the smallest value of R g ggg , Fig. 19. Note
also that when R g ggg ) 1 this additional cross section is below 100 fb.
3.6.2. The case of a common mass in the third generation squark sector with MA s 350
GeV
The discussion is essentially the same as the one we presented for tan b s 2.5 with
MA s 350 GeV. The overall reduction factor from the lowering of MA which affects
Fig. 18. Ža. As in Fig. 17 but for R g ggg versus Rgg . Žb. As in Ža. but forRgg versus Mh .
32 ´
Fig. 19. As in Fig. 17 but for Rgg versus s Ž t˜1 t˜1 h..
Rgg is sightly smaller Žabout 0.76. but then the reductions in R g ggg are for m h ; 115
GeV, Fig. 20. Note also that s Ž t˜1 t˜1 h. production, Fig. 21, is only slightly smaller than
with tan b s 2.5 Žthis is due to a higher Higgs mass. and therefore is a useful addition
when the direct channel drops too much. For R g ggg - 0.2 one gets as much as 400 fb.
For larger tan b and small MA , de-excitation of t˜2 into t˜1 is not as efficient as for the
Fig. 20. Ža. As in Fig. 18 but for R g ggg versus Rgg . Žb. As in Ža. but forRgg versus Mh .
´
Fig. 21. As in Fig. 17 but for Rgg versus s Ž t˜1 t˜1 h..
lower tan b with the rather moderate values of m that we have considered in this study.
This is evident from Eq. Ž3.11., but as we see t˜1 t˜1 h still plays its role.
Fig. 22. Ža. R g ggg versus Rgg for tan b s 5, m sy250 GeV and MA s1 TeV, when we allow different scalar
masses for the third generation as given, see text. Žb. As in Ža. but for R g ggg versus t˜2 t˜1 h.
34 ´
3.6.3. Lifting the degeneracy in the third family scalar masses

Taking unequal masses as in Section 3.5.3, the reductions in the direct production are
less pronounced. We do not get below R g ggg , 0.55, while values up to 1.25 are still
possible for R g ggg . There is also little change in where these reductions or enhance-
ments occur as a function of the Higgs mass. Again when R g ggg - 0.8, Rgg ) 1, Fig. 22.
As with tan b s 2.5 when the mass degeneracy is lifted, the channel t˜2 t˜1 h opens
up.This leads to typical cross sections of the order of 100 fb, especially for regions
™
where the drop in the direct inclusive two-photon channel is the largest, see Fig. 22. This
cross section can be larger than 1pb in situation where R g ggg is little reduced. Of course
our scans do show some regions where this cross section is unusable, typically when the
mixing and stop splitting is small, but then as Fig. 22 shows the inclusive two-photon
channel is unaffected. Of course continuum s Ž t˜1 t˜1) h. is still useful when large drops
occurs Žit is then around 100 fb. but note that for our choice of parameters s Ž t˜2 t˜1) h. is
practically always larger, Fig. 23.
3.7. tan b s 10
Apart from the location, in terms of the Higgs mass, of where the largest drop in
R g ggg occurs, that is around 118 GeV, all the general features we found in the case with
Fig. 23. As in Fig. 22 but for R g ggg versus t˜1 t˜1 h and t˜2 t˜1 h versus t˜2 t˜1 h.
´
Fig. 24. Ža. R g ggg versus m h for tan b s10, m sy250 GeV and MA s1 TeV, Žb. As in Ža. but for R h ™ gg
versus Mh , Žc. As in Ža. but for R g ggg versus Rgg .
tan b s 5 are recovered again, Fig. 24. Note that we do not get more noticeable
reduction either in R g ggg or Rgg due to the larger tan b , and also that R g ggg ) 1 are
possible. Similar observations to those made for tan b s 5 can be made here even when
we consider different splitting and lowering of masses, especially as concerns the
importance of t˜2 t˜1) h. Some of these results are summarised in Fig. 24.
4. Conclusions
We have in this paper reinvestigated the fate of the photon signal of the lightest
SUSY Higgs at the LHC when large tri-linear mixing terms in the stop sector are
36 ´
present. Previous investigations w15,16,18x had drawn a very pessimistic picture of these
scenarios. Our analysis shows that if we exploit all the consequences of these scenarios
and not pick out only the Higgs signal in the inclusive channel these models have an
excellent discovery potential. First, the large reductions in the inclusive two-photon
signal not only require large mixing but also that one of the stops be relatively light.
Although this has not been stressed in the text, a first signal of these scenarios will be
t˜1 t˜1) production with a cross section of order ; 100pb. Even though it may be argued
that regions with the largest drops in the inclusive two-photon channel correspond to a
very light stop and are likely to lead to a signature, jets q pu T , which is difficult. In any
case it should be stressed that a hallmark of these scenarios is that whenever the signal
™
and makes it up for the drop in the former channel. Moreover when s Ž pp h gg . ™™
in the inclusive channel drops that in the associated WhrZh and tth channels increases
gets too small the continuum pp t˜1 t˜1 h w40,41x reaches values of order few 100 fb.
™
More importantly we find that since these situations imply a large mass splitting
between the two stops, t˜2 t˜1 h can be substantial leading to another source of Higgs
™
production with a yield larger than in the continuum and with a better signature than the
t˜1 t˜1) h continuum. We have shown that t˜2 t˜1) h occurs whenever the stop mixing angle
does not take its maximal value, <sin2 u t̃ < s 1, which is often unnaturally assumed on the
basis of equal soft SUSY breaking masses for the SUŽ2. and UŽ1. sfermions of the third
generation, at the electroweak scale. We have also shown that although when MA gets
small the two-photon signals Žboth direct and associated. get further reduced Žthis
for low values of tan b , one can trigger A production through the cascade t˜2 t˜1 A,
beside the usual channels for A productions. Moreover one should not forget that
™
happens even in the absence of mixing., with large tri-linear mixing terms and especially
especially with not too small tan b , tan b ) 3, scenarios with light stops Žbut small
mixing. do give an increase in the direct channel, but then a decrease in the associated
two-photon channels. The overall conclusion we can draw almost resembles that of a
no-lose scenario: whenever an effect reduces a particular signal it opens up new
channels or enhances other channels. We have not discussed the use of h bb in the™
associated tth channel which in these scenarios should allow detection. This requires
rather good b-tagging facilities, as shown in w43x. This should certainly add to the
discovery potential. The new associated stop Higgs signatures deserve a full simulation
to critically quantify how beneficial these additions can be. In general there is a lack of
detailed study of stop phenomenology at the LHC despite some important theoretical
issues related to the third generation sfermions. As has been pointed out by several
authors w101–106x, the idea of an inverted hierarchy of the SUSY spectrum whereby the
third generation sfermions are, at the electroweak scale, much lighter than the first two
is compelling and quite plausible. This helps solve the flavour problem in SUSY since
very large masses for the superpartners of the first two generations can suppress FCNC,
contributions to electric dipole moments and lepton flavour violations. This would still
not go against naturalness since these particles couple weakly to the Higgs, at the heart
of the fine-tuning problem. Naturalness does on the other hand require the stops and
sbottoms Žand the electroweak gauginos higgsinos. to be rather light, like in the
scenarios we have studied and could also with a light stop make electroweak baryogene-
sis w107–112x work.
´
Acknowledgements
We would like to thank Guillaume Eynard for providing with his programme for
calculating the significance of the two-photon associated Higgs signal using the ATLAS
simulation and for very useful discussions. We also thank Michael Spira for promptly
providing us with the code for the NLO stop pair production at the hadron colliders and
Andrei Semenov for advice on the use of CompHep. We are also grateful to Elzbieta
Richter-Waş for helpful discussions and communication,as well as providing us with an
advance copy of the ATLAS TDR on Higgs Physics. K.S. acknowledges the hospitality
of LAPTH where part of this work was done. This work is done under partial financial
support of the Indo-French Collaboration IFCPAR-1701-1 Collider Physics.
References
w1x F. Gianotti, Opal talk for the LEP Committee, March 1999, http:rralephwww.cern.chrALPUBrsemi-
narrlepc_mar99.pdf.
w2x N.J. Kjaer, Delphi talk for the LEP Committee, March 1999, http:rrdelphiwww.cern.chrdelfigsrfig-
uresrniels990324.ps.gz.
w3x G. Bobbink, L3 talk for the LEP Committee, March 1999, http:rrl3www.cern.chrconferencesrpsr
Bobbink_LEPC9903.ps.
w4x D.A. Glenzinski, Opal talk for the LEP Committee, March 1999, http:rrwww1.cern.chrOpalrplotsr
glenzinskirmain.ps.gz.
w5x J.F. Gunion, H.E. Haber, G.L. Kane, S. Dawson, The Higgs Hunter’s Guide ŽAddison-Wesley, Reading,
1990..
w6x R.K. Ellis, I. Hinchliffe, M. Soldate, J.J. van der Bij, Nucl. Phys. B 297 Ž1988. 221.
w7x J. Gunion, G. Kane, J. Wudka, Nucl. Phys. B 299 Ž1988. 231.
w8x J. Gunion, H. Haber, Nucl. Phys. B 278 Ž1986. 449.
w9x H. Baer, D. Dicus, M. Drees, X. Tata, Phys. Rev. D 36 Ž1987. 1363.
w10x K. Griest, H. Haber, Phys. Rev. D 37 Ž1988. 37.
w11x H. Baer, M. Bisset, D. Dicus, C. Kao, X. Tata, Phys. Rev. D 47 Ž1992. 1062.
w12x A. Djouadi, J. Kalinowski, P.M. Zerwas, Z. Phys. C 57 Ž1993. 569.
w13x H. Baer, M. Bisset, C. Kao, X. Tata, Phys. Rev. D 46 Ž1992. 1067.
w14x H. Baer, M. Bisset, D. Dicus, C. Kao, X. Tata, Phys. Rev. D 47 Ž1992. 1062.
w15x B. Kileng, Z. Phys. C 63 Ž1993. 87.
w16x B. Kileng, P. Osland, P.N. Pandita, Z. Phys. C 71 Ž1996. 87, hep-phr9506455.
w17x G.L. Kane, G.D. Kribs, S.P. Martin, J.D. Wells, Phys. Rev. D 50 Ž1996. 213.
w18x A. Djouadi, Phys. Lett. B 435 Ž1998. 101, hep-phr9806315.
w19x Z. Kunszt, F. Zwirner, Nucl. Phys. B 385 Ž1992. 3.
w20x E. Richter-Wa̧s et al., ATLAS Internal note, PHYS-No-074 Ž1996..
w21x For an update on the Higgs analysis in ATLAS see, the ATLAS Technical Design Report, the ATLS
Collaboration, 1999, to appear. http:rratlasinfo.cern.chrAtlasrGROUPSrPHYSICSrHIGGSrhiggs-
wwwr Analyses-notes.htmlrhiggs-sugra.ps.gz.
w22x For an update on Higss searches in CMS see, R. Kinnunen, D. Denegri, CMS Note 1997r057, April
1997.
w23x H.P. Nilles, Phys. Rep. 110 Ž1984. 1.
w24x R. Arnowitt, A. Chamseddine, P. Nath, Applied N s1 Supergravity ŽWorld Scientific, Singapore,
1984..
w25x H. Baer, M. Bisset, C. Kao, X. Tata, Phys. Rev. D 50 Ž1993. 316.
w26x G. Polesello, L. Poggioli, E. Richter-Wa̧s, J. Sderqvist, ATLAS Internal Note, PHYS-No-111, Oct.
1997.
w27x H. Haber, in: Perspectives in Higgs Physics II, ed. G.L. Kane ŽWorld Scientific, Singapore, 1998..
w28x S.L. Glashow, D.V. Nanopoulos, A. Yildiz, Phys. Rev. D 18 Ž1978. 1724.
38 ´
w29x Z. Kunszt, Nucl. Phys. B 247 Ž1984. 339.

w30x J.F. Gunion et al., Nucl. Phys. B 294 Ž1987. 621.
w31x D. Dicus, S.S.D. Willenbrock, Phys. Rev. D 39 Ž1989. 751.
w32x W. Marciano, F. Paige, Phys. Rev. Lett. 66 Ž1991. 2433.
w33x J. Gunion, Phys. Lett. B 261 Ž1991. 510.
w34x T. Han, S.S.D. Willenbrock, Phys. Lett. B 273 Ž1990. 167.
w35x J. Gunion, L. Orr, Phys. Rev. D 46 Ž1992. 2052.
w36x S. Abdullin, A. Starodumov, N. Stepanov, CMS TNr93-86 Ž1993..
w37x M. Dzelayin, Z. Antunovic, D. Denegri, R. Kinnunen, CMS TNr96-091 Ž1996..
w38x For a detailed analysis of the two-photon signal in the associated production in ATLAS see, G. Eynard,
PhD thesis, LAPP-T-98r02, LAPP-Annecy, May 1998.
w39x D. Froidevaux, F. Gianotti, E. Richter-Wa̧s, ATLAS Internal Note, PHYS-No-64, Feb. 1995.
w40x For a study of stop stop Higgs production at the LHC see, A. Djouadi, J.L. Kneur, G. Moultaka, Phys.
Rev. Lett. 80 Ž1998. 1830. hep-phr9711244;
w41x A. Djouadi, J.L. Kneur, G. Moultaka, hep-phr9903218.
w42x J. Dai, J.F. Gunion, R. Vega, Phys. Rev. Lett. 71 Ž1993. 2699.
w43x D. Froidevaux, E. Richter-Wa̧s, ATLAS Internal Note, PHYS-No-043, Sep. 1994.
w44x ¨
V. Drollinger, T. Muhler, R. Kinnunen, CMS Note 1998r003 and 1999r01.
w45x A. Djouadi, J. Kalinowski, M. Spira, Comp. Phys. Comm. 108 Ž1998. 56, hep-phr9704448.
w46x M. Carena, J.R. Espinosa, M. Quiros, C.E.M. Wagner, Phys. Lett. B 355 Ž1995. 209.
w47x M. Carena, M. Quiros, C.E.M. Wagner, Nucl. Phys. B 461 Ž1996. 407.
w48x ´
G. Belanger, F. Boudjema, T. Kon, V. Lafage, hep-phr9711334, Eur. J. Phys., in press.
w49x W. Loinaz, J.D. Wells, Phys. Lett. B 445 Ž1998. 178, hep-phr9808287.
w50x H. Baer, J.D. Wells, Phys. Rev. D 57 Ž1998. 4446, hep-phr9710368.
w51x M. Carena, S. Mrenna, C.E.M. Wagner, Preprint ANL-HEP-PR-98-54, CERN-THr98-262, FERMI-
LAB-PUB-98r250-T, Sep. 1998. hep-phr9808312.
w52x M.M. Nojiri, Phys. Rev. D 51 Ž1995. 6281.
w53x M.M. Nojiri, K. Fujii, T. Tsukamoto, Phys. Rev. D 54 Ž1996. 6756.
w54x W. Porod, PhD Thesis, Wien Univ. 1998. hp-phr9804208.
w55x E. Boos et al., Proc. XXVIth Rencontres de Moriond, ed. J. Tran Thanh Van ŽEditions Frontieres, `
Gif-sur-Yvette, 1991. p. 501.
w56x E. Boos et al., Proc. of the Int. Conf. on Computing in High Energy Physics, ed. Y. Watase, F. Abe,
Universal Academy Press, Tokyo, 1991, p. 391.
w57x E. Boos et al., Int. J. Mod. Phys. C 5 Ž1994. 615.
w58x E. Boos et al., SNU CTP preprint 94-116, Seoul, 1994, hep-phr9503280.
w59x A. Dedes, S. Moretti, preprint RAL-TR-1998-081, Dec. 1998. hep-phr9812328.
w60x Z. Kunszt, S. Moretti, W.J. Stirling, Z. Phys. C 74 Ž1997. 479.
w61x W. Beenakker et al., Nucl. Phys. B 515 Ž1998. 3, hep-phr9710451.
w62x T. Plehn, Doctoral thesis, DESY-THESIS-1998-024, July 1998. hep-phr9809319.
w63x A. Bartl, W. Majerotto, W. Porod, Z. Phys. C 64 Ž1994. 499.
w64x A. Bartl, W. Majerotto, W. Porod, Z. Phys. C 68 Ž1995. 518, Erratum.
w65x A. Bartl et al., Z. Phys. C 73 Ž1997. 549.
w66x A. Bartl et al., Z. Phys. C 76 Ž1997. 469.
w67x A. Bartl et al., Phys. Lett. B 435 Ž1998. 118.
w68x M. Jimbo, H. Tanaka, T. Kaneko, T. Kon, hep-phr9503363.
w69x J. Fujimoto et al., Comp. Phys. Commun. 111 Ž1998. 185, hep-phr9711283.
w70x S. Kraml, Stop and Sbottom Phenomenology in the MSSM, PhD Thesis, Univ. of Vienna, hep-
phr9903257.
w71x M. Drees, K. Hagiwara, Phys. Rev. D 42 Ž1990. 1709.
w72x R. Barbieri, M. Frigeni, F. Giuliani, H.E. Haber, Nucl. Phys. B 341 Ž1990. 309.
w73x M. Drees, K. Hagiwara, A. Yamada, Phys. Rev. D 45 Ž1992. 1725.
w74x ` Mod. Phys. Lett. A 9 Ž1994. 211.
D. Garcia, J. Sola,
w75x P.H. Chankowski et al., Nucl. Phys. B 417 Ž1994. 101.
w76x R. Barbieri, L. Maiani, Nucl. Phys. B 224 Ž1983. 32.
w77x C.S. Lim, T. Inami, N. Sakai, Phys. Rev. D 29 Ž1984. 1488.
´
w78x Z. Hioki, Prog. Theor. Phys. 73 Ž1985. 1283.

w79x ` Nucl. Phys. B 253 Ž1985. 47.
J.A. Grifols, J. Sola,
w80x J. Erler, P. Langacker, hep-phr9809352.
w81x A. Djouadi et al., Phys. Rev. Lett. 78 Ž1997. 3626.
w82x A. Djouadi et al., Phys. Rev. D 57 Ž1998. 4179.
w83x ` D.R. T Jones, S. Raby, Nucl. Phys. B 222 Ž1983. 11.
J.M. Frere,
w84x M. Claudson, L. Hall, I. Hinchcliffe, Nucl. Phys. B 228 Ž1983. 501.
w85x C. Kounnas, A.B. Lahanas, D.V. Nanopoulos, M. Quiros, ´ Nucl. Phys. B 236 Ž1984. 438.
w86x J.F. Gunion, H.E. Haber, M. Sher, Nucl. Phys. B 306 Ž1988. 1.
w87x P. Langacker, N. Polonsky, Phys. Rev. D 50 Ž1994. 5824.
w88x A. Strumia, Nucl. Phys. B 482 Ž1996. 24.
w89x J.A. Casas, Perspectives on Supersymmetry, ed. G.L. Kane ŽWorld Scientific, Singapore., to be
published, hep-phr9707475.
w90x A. Kusenko, P. Langacker, G. Segre, Phys. Rev. D 54 Ž1996. 5824, hep-phr9602414.
w91x A. Kusenko, P. Langacker, Phys. Lett. B 391 Ž1997. 29, hep-phr9608340.
w92x A. Kusenko, Nucl. Phys. B ŽProc. Suppl.. 52A Ž1997. 67, hep-phr9607287.
w93x C.M. Holck, CDF Collaboration, Abstract 652, ICHEP 98, Vancouver, BC, July 98.
w94x K. Hikasa, M. Kobayashi, Phys. Rev. D 36 Ž1987. 724.
w95x M. Drees, M.M. Nojiri, Nucl. Phys. B 369 Ž1992. 54.
w96x M. Drees, S.P. Martin, hep-phr9504324.
w97x V. Barger, M.S. Berger, P. Ohmann, Phys. Rev. D 47 Ž1993. 1093.
w98x D.J. Castano, E.J. Piard, P. Ramond, Phys. Rev. D 49 Ž1994. 4882.
w99x M. Carena, M. Olechowski, S. Pokorski, C.E.M. Wagner, Nucl. Phys. B 419 Ž1994. 213.
w100x M. Carena, C.E. M Wagner, Nucl. Phys. B 452 Ž1995. 45.
w101x S. Dimopoulos, G.F. Giudice, Phys. Lett. 357 Ž1995. 573.
w102x A. Pomarol, D. Tommasini, Nucl. Phys. B 466 Ž1996. 3.
w103x G. Dvali, A. Pomarol, Phys. Rev. Lett. 77 Ž1996. 3728.
w104x G. Dvali, A. Pomarol, Nucl. Phys. B 522 Ž1998. 3.
w105x A.G. Cohen, D.B. Kaplan, A.E. Nelson, Phys. Lett. B 388 Ž1996. 588.
w106x J.L. Feng, C. Kolda, N. Polonsky, hep-phr9810500.
w107x J.R. Espinosa, Nucl. Phys. B 475 Ž1996. 273.
w108x J.M. Cline, M. Joyce, K. Kainulainen, Phys. Lett. B 417 Ž1998. 79, hep-phr9708393.
w109x M. Carena, M. Quiros,´ C.E.M. Wagner, Nucl. Phys. B 524 Ž1998. 3, hep-phr9710401.
w110x J.M. Moreno, M. Quiros,´ M. Seco, Nucl. Phys. B 526 Ž1998. 489, hep-phr9801272.
w111x M. Laine, K. Rummukainen, Phys. Rev. Lett. 80 Ž1998. 5259, hep-phr9804255.
w112x J.M. Cline, preprint McGill-98r27, hep-phr9810267.
Nuclear Physics B 568 Ž2000. 40–59
Radiative large-angle Bhabha scattering in collinear kinematics

V. Antonelli a , E.A. Kuraev b, B.G. Shaikhatdenov b,1
a
Dipartimento di Fisica, UniÕersita` di Milano-Bicocca and INFN Sezione di Milano, Milano, Italy
b
Joint Institute for Nuclear Research, Dubna, 141980, Russia
Received 30 June 1999; received in revised form 14 September 1999; accepted 20 September 1999
Abstract
The process of large-angle high energy electron–positron scattering with emission of one hard
photon almost collinear to one of the charged particles’ momenta is considered. The differential
cross section with radiative corrections due to emission of virtual and soft real photons calculated
to a power accuracy is presented. Emission of two hard photons and total expressions for radiative
correction are given in leading logarithmical approximation. The latter are illustrated by numeric
estimates. A relation of the results with structure function formalism is discussed. q 2000 Elsevier
Science B.V. All rights reserved.
PACS: 12.20.ym; 12.20.Ds
1. Introduction
The process of electron–positron scattering is commonly used for luminosity mea-

surements at eqey colliders. It has almost pure electrodynamical nature and could
therefore be described to any desired precision within a framework of perturbative QED.
Nevertheless the accuracy of modern experiments is ahead of that provided by theory. A
lot of work has recently been done to uplift the theoretical uncertainty to about one per
mille under conditions of small-angle Bhabha scattering at LEP1 w1x and afterwards up
to 0.05–0.06% w2,3x.
The large-angle kinematics of the Bhabha scattering process is extensively used for
calibration purposes at eqey colliders of moderately high energies, such as f , Jrc , B,
and crt factories and LEP2. At the Born and one-loop levels the process was
investigated in detail in Refs. w4–8x, taking into account both QED and electroweak
effects.
1
On leave of absence from the Institute of Physics and Technology, Almaty.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 2 5 - 2
V. Antonelli et al.r Nuclear Physics B 568 (2000) 40–59 41
In paper w9x we considered Bhabha scattering to O Ž a . order exactly improved by the

structure function method. The latter, based on the renormalization group approach,
allows to evaluate the leading radiative corrections to higher orders, including all the
terms ; Ž a L s . n, n s 2,3, . . . , where L s s lnŽ srm2 . is a large logarithm, s is the total
center-of-mass Žcms. energy of incoming particles squared and m is the mass of
fermion.
To reach the one per mille accuracy it is required to take into account radiative
corrections ŽRC. up to third order within the leading logarithmic approximation ŽLLA.
and up to second order in the next-to-leading approximation ŽNLA.. In a series of papers
several sources of these corrections were considered in detail w10–14x.
In a recent publication w14x the contribution due to virtual and soft photon corrections
to large-angle radiative Bhabha scattering was calculated for the general case of hard
photon emission at large angle with respect to all charged particles’ momenta. In the
present work we are going to consider the complementary specific kinematics, in which
the photon moves within a narrow cone of small opening angle u 0 < 1 together with
one of the incoming or outgoing charged particles. Thus, the result obtained here may be
used in experiments with the tagging of the scattered electron Žpositron. in detectors of
small aperture u 0 < 1.
Our paper is organized as follows. In Section 2 the Born level cross section of
radiative Bhabha scattering is revised in the collinear kinematics of photon emission
along the initial Žscattered. electron. We introduce here the physical gauge of real
photon that is extensively used in the next sections. In Section 3 a set of crossing
transformations which enables us to consider in some detail only the scattering type
amplitudes of loop corrections to the process is described. Besides, we restrict ourselves
to the kinematics of hard photon emission along the initial electron. In Section 4 the
corrections due to virtual and soft real photon emission in the case k 1 I p 1 are
considered. The general expression for correction in the case of hard photon emission
along the scattered electron is given in Section 5. In Section 6 we consider a
contribution Žin LLA. coming from two hard photon emission and derive a general
expression for radiative correction. In conclusion we discuss the relation with structure
function approach and the accuracy of the results obtained. Some useful expressions for
loop integrals are given in Appendix A and the results of numeric estimates are given in
graphs.
2. Born expressions in collinear kinematics
Let us begin revising the radiative Bhabha scattering process
ey Ž p 1 . q eq Ž p 2 . ™e y
Ž pX1 . q eq Ž pX2 . q g Ž k 1 . Ž 1.
at the tree level. We define the collinear kinematical domains as those in which the hard
$
photon is emitted close Žwithin a narrow cone with opening angle u 0 < 1. to the
X
$
incident Ž u 1Ž2. s p 1Ž2. k 1 - u 0 . or the outgoing electron Žpositron. Ž u 1Ž2. s pX1Ž2. k 1 - u 0 .
direction of motion. Because of the symmetry between electron and positron we may
restrict ourselves to a consideration of only two collinear regions, which correspond to
42 V. Antonelli et al.r Nuclear Physics B 568 (2000) 40–59
the emission of the photon along the electron momenta. The two remaining contributions
to the differential cross section of the process Ž1. can be obtained by the substitution Q ,
d scoll s 1qQ
Q
ž ll /
p1
pX1
p2
pX2
d s g Ž k 1 I p 1 . qd s g Ž k 1 I pX1 . 4 . Ž 2.
To begin with, let us recall the known expression w15,16x in Born approximation for the
general kinematics, i.e. assuming all the squares of the momenta transferred among
fermions to be large compared to the electron mass squared:
a3 d 3 pX1d 3 pX2 d 3 k 1
d s 0g s T dG , d Gs d 4 Ž p 1 q p 2 y pX1 y pX2 y k 1 . , Ž 3.
8p 2 s ´ 1X ´ X2 v 1
2
S 16 m2 s t1
Ts
tt 1 ss1
ss1 Ž s 2
q s12 . q tt1 Ž t 2
q t 12 . q uu1 Ž u 2
q u12 . y
x 2X 2 ž t1
q
s
q1
/
2 2 2
16 m 2 s t 16 m2 s1 t1 16 m2 s1 t
y
x 1X 2 ž t
q q1 y
s x 22 / ž t1
q
s1
q1
/ y
x 12 ž t
q
s1
q1 ,
/
s s1 t1 t u1 u
Ss4 q y y q q ,
x1 x2 x 1X x 2X x 1 x 1X x 2 x 2X x 2 x 1X x 1 x 2X
2 2 2
s s Ž p1 q p 2 . , s1 s Ž pX1 q pX2 . , t s Ž p 2 y pX2 . ,
2 2 2
t 1 s Ž p 1 y pX1 . , u s Ž p 1 y pX2 . , u1 s Ž p 2 y pX1 . ,
X
x i s 2 pi k 1 , x 1,2 s 2 pX1,2 k 1 .
In the collinear kinematical domain in which k 1 I p 1 the above formula takes the form
a 3 d3 k1 1 d 3 pX1d 3 pX2
d s 0g Ž k 1 I p 1 . s FF d 4 Ž Ž 1 y x . p 1 q p 2 y pX1 y pX2 .
p 2s v1 x1 ´ 1X ´ X2
Ž 4.
2
1q Ž1yx . 2 m2
s dWp 1d s 0 Ž Ž 1 y x . p 1 , p 2 . , Fs y ,
x Ž1yx . x1
2
s1 t
Fs
ž t
q
s1
q1 ,
/
where
´ 1X 1yx
s1 s s Ž 1 y x . , y1 s s2 ,
´ a
´ X2 2 y 2 x q x 2 q cx Ž 2 y x .
y2 s s , a s 2 y x q cx , v1 s ´ x ,
´ a
m2
s s 4´ 2 , x1 s
s
2
x Ž 1 y c1 b . , bs ( 1y
´2
,
Ž1 yx . Ž1 yc.
2
$ $
t s t 1 Ž 1 y x . s ys , c s cos Ž p 1 pX1 . , c1 s cos Ž p 1 k 1 . ,
a
3
a 1yx d k1
dWp 1 s 2
F . Ž 5.
2p x1 v1
Here yi are the energy fractions of the scattered leptons and d s 0 Ž p 1Ž1 y x ., p 2 . is the
cross section of the elastic Bhabha scattering process.
Throughout the paper we use the following relations among invariants:
s1 q t q u1 s 4 m2 y x 1 f 0, s q t 1 q u s 4 m2 q x 1 f 0.
In the case k 1 I pX1 we have
a 1 d3 k1
d s 0g Ž k 1 I pX1 . s X F˜ Ž 1 y x . d s 0 Ž p1 , p 2 . , Ž 6.
2p 2 x 1 v1
2
1q Ž1yx . 2 m2
F̃ s y .
x x 1X
These expressions could also be inferred by using the method of quasi-real electrons w17x
and starting from the non-radiative Bhabha cross section.
After integration over a hard collinear Ž k 1 I p 1 . photon angular phase space, the cross
section of radiative Bhabha scattering in the Born approximation is found to be
2
d s 0g 4a 3 1 q Ž 1 y x . 1yx
s L0 y 2
d xd c k 1 Ip 1 s x x
2
3 y 3 x q x 2 q 2 cx Ž 2 y x . q c 2 Ž 1 y x Ž 1 y x . .
=
ž Ž 1 y x . Ž 1 y c . a2 /
= Ž 1 q O Ž u 02 . . , Ž 7.
2
where L0 s ln Ž ú 0rm . . And in the case k 1 I pX1 it reads
2 2
d s 0g a 3 1q Ž1yx . 1yx 3 q c2
LX0 y 2
d xd c X
k 1 Ip 1
s
4s x x ž 1yc /Ž 1 q O Ž u 02 . . , Ž 8.
´ 1X u 0 2
LX0 s ln ž / , ´ 1X s ´ Ž 1 y x . .
m
The simplest way to reproduce these results is to use the physical gauge for the real
photon which in the beam cms sets the photon polarization vector to be a space-like
3-vector el having density matrix
0, if m or n s 0, k1
Ý emlenl ) s
l
½ dmn y nm nn , m s n s 1,2,3, ns
v1
,
with the properties
t 1 u1
Ý < el < 2 s y2, Ý < p1 el < 2 s ´ 2 Ž 1 y c12 . , Ý < pX1 el < 2 s s
,
l l l
Ý Ž p1 el . Ž pX1 el . ; u ™ 0u .
)
Ž 9.
l
These properties enable us to omit mass terms in the calculations of traces and, besides,
to restrict ourselves to the consideration of singular terms Žsee Eq. Ž10.. only, both at
the Born and one-loop level. As shown in Ref. w18x, this gauge is proved useful for a
description of jet production in quantum chromodynamics; it is also very well suited to
our case because it allows to simplify a lot the calculation with respect, for instance, to
the Feynman gauge. What is more, it possesses another very attractive feature related
with the structure of the correction to be mentioned below Žsee Appendix A..
™
With these tools at our disposal let us turn now to the main point. The contributions,
which survive the limit u 0 0, arise from the terms containing
2 2
Ž p1 e . e2 Ž pX1 e .
, , . Ž 10 .
x 12 x1 x1
Other omitted terms Žin particular those which do not contain a factor xy1
1
. can be safely
neglected since they give a contribution of the order of u 02 which determines the
accuracy of our calculations
a 2
m
1qO žpu / 0 Ls ,
´
< u 0 < 1. Ž 11 .
In the realistic case this corresponds to an accuracy of the order of per mille.
3. Crossing relations
In this and the next section we shall consider the case k 1 I p 1. In the case of photon
emission along pX1 one can get the desired expression by using the left-to-right
permutation
l yp < M <
l yp /
X
p1
< M < 2k 1 I pX1 s Q 1 2
ž p2 X
2
k 1 I p1 . Ž 12 .
From now on we deal with scattering type amplitudes ŽFD. with the emission of a
hard photon by the initial electron. This is possible due to the properties of the physical
gauge. The contribution of annihilation type amplitudes may be derived by applying the
momenta replacement operation as follows:
2
D < M < annihilation s Q Ž pX1 l yp . 4D < M <
2
2
scattering '
2
Q1 4 D < M < scattering . Ž 13 .
In considering FD with two photons in the scattering channel Žbox FD. one may
l
examine only those with uncrossed photons because a contribution of the others may be
obtained by the permutation p 2 ypX2 . Thus the general answer becomes
1
½
< M < 2k 1 I p 1 s Re Ž 1 q Q1 . w G q L x q
s1 t
Ž 1 q Q1 . Ž 1 q Q 2 . s1 t Ž B q P . 5 ,
Ž 14 .
with the permutation operators acting as
Q1 F Ž s1 ,t 1 , s,t . s F Ž t , s,t 1 , s1 . , Q 2 F Ž s,u, s1 ,u1 . s F Ž u, s,u1 , s1 . .
4. Virtual and soft photon emission in k 1 I p 1 kinematics
One-loop QED RC Žwhich are described by seventy two Feynman diagrams. can be
classified out into the two gauge invariant subsets Žsee Fig. 1.:
Ø single photon exchange between electron and positron lines ŽG,L-type.;

Ø double photon exchange between electron and positron lines ŽB,P-type..
Fig. 1. Some representatives of FD for radiative Bhabha scattering up to second order: Ž1. is the vertex
insertion; Ž2. is the vacuum polarization insertion; graphs denoted by Ž3., Ž4. are of the L-type, Ž5. is of
G1-type, Ž6. is of G 2-type, Ž7. is of B-type and Ž8. is of P-type.
For L-type FD Žsee Fig. 1Ž3,4.. the initial spinor uŽ p 1 . is replaced by

Ž ar2p . A 2 kˆ 1 eu
ˆ Ž p1 ., with
1 r 2 r 2 y3rq2 1 p2
A2 s
x1 ½ y
2 Ž r y 1.
x1
q
2 Ž r y 1.
2
Lr q
r 6
yLi 2 Ž 1 y r . q
,
5
Lr s ln r , rs 2 .
m
The relevant contribution to the matrix element squared and summed over spin states
reads
A 2 s13 y u13 2Ž 2 y x .
D < M < 2L s 2 9p 2a 4 2
Yy W , Ž 15 .
x 1 s1 t 1yx
2
x 2
Y s 4 Ž p1 e . y e 2x 1 , W s Ž p1 e . .
1yx
The contribution of vertex insertion, vacuum polarization2 and G 1-type Žsee Fig.
1Ž1,2,5.. has the following form:
s13 y u13
D < M < 2P , G , G a s 2 10p 2a 4 P t q Gt q 14 Ga 2 Y, Ž 16 .
t s1 x 12
p2
P t s 13 L t y 59 , Gt s Ž Ll y 1 . Ž 1 y L t . y 14 L t y 14 L2t q ,
12
p2
Ga s y3L2t q 4 L t Lr q 3 L t q 4 Ll y 2ln Ž 1 y r . y q 2Li 2 Ž 1 y r . y 4,
3
m yt zdx
Ll s ln , L t s ln 2 , Li 2 Ž z . s y ln Ž 1 y x . .
H
l m 0 x
Here l is as usual the IR cut-off parameter to be cancelled at the end of calculus against
a soft photon contribution.
For the contribution of G 2-type FD Žsee Fig. 1Ž6.. with four denominators we obtain
s13 y u13
D < M < G2 s 2 9a 4p 2
ts1 x 1 Ž 1 y x .
2Ž 2 y x .
= Ž J y J1 . Y q W Ž J11 y J1 q xJ1 k y xJk . . Ž 17 .
1yx
It turns out that only the scalar integral and the coefficients before p 1 ,k 1 in the vector
and tensor integrals give a non-vanishing contribution in the limit u 0 0 ™
d 4 k Ž 1,k m ,k m k n . mn
H s Ž J , J1 p 1m q Jk k 1m , J11 p 1m p 1n q Jk k k 1m k 1n q J1 k Ž p 1 k 1 . . ,
ip 2 Ž 0 . Ž 1 . Ž 2 . Ž q .
Ž 0 . s k 2 y l2 , Ž 1. s k 2 y 2 p1 k , Ž 2 . s k 2 y 2 pX1 k ,
mn
Ž q . s k 2 y 2 k Ž p1 y k 1 . y x 1 , Ž ab . s a m b n q an b m ,
and the terms having no p 1 or k 1 momentum in the decomposition have been omitted
for their unimportance.
2
For realistic applications one should also add to P the contributions due to m and t leptons and hadrons.
The B-type FD shown in Fig. 1Ž7. with uncrossed legs gives

1
D < M < 2B s 2 9p 2a 4 Y Ž u13 y s13 . s1Ž B q a y b . y u13 s1
s1 t x 12
2
ž
= c q a1X 2X q a1X 2 q
s1 /
a g q s13 Ž c w t y u1 x q 2 J0 . , Ž 18 .
where the coefficients are associated with scalar, vector and tensor integrals over loop
momentum
d 4 k Ž 1,k m ,k m k n . d4 k 1
H ip 2
s Ž B, B m , B mn . , J0 s H ip 2
,
b 0 b1 b 2 b 3 b1 b 2 b 3
b 0 s k 2 y l2 , b 1 s k 2 q 2 pX1 k , b 2 s k 2 y 2 pX2 k , b 3 s k 2 y 2 qk q t ,
m
q s pX2 y p 2 , B m s Ž apX1 q bpX2 q cp 2 . ,
B mn s a g g mn q a1X 1X pX1m pX1n q a 22 p 2m p 2n q a 2X 2X pX2m pX2n
mn mn mn
q a1X 2 Ž pX1 p 2 . q a1X 2X Ž pX1 pX2 . q a 22X Ž p 2 pX2 . .
For P-type FD Žsee Fig. 1Ž8.. with uncrossed photon legs we have
s13 y u13 2Ž 2 y x .
D < M < 2P s 2 9p 2a 4 Y Ž E y E1 . q
t x 1Ž 1 y x . 1yx
=W Ž E11 y E1 q xE1 k y xEk . . Ž 19 .
™
Here we are using the definition Žwith tensor structures giving no contributions in the
limit u 0 0 dropped.
d 4 k Ž 1,k m ,k m k n .
H ip 2
s Ž E, E1 p 1m q Ek k 1m , E11 p 1m p 1n q Ek k k 1m k 1n
a0 a1 a2 a3 a4
qE1 k Ž p 1m k 1n q p 1n k 1m . . ,
a 0 s k 2 y l2 , a1 s k 2 y 2 p 1 k , a2 s k 2 y 2 k Ž p1 y k 1 . y x 1 ,
a3 s k 2 q 2 p 2 k , a4 s k 2 y 2 qk q t.
Note that in the evaluating of P-type FD we are allowed to put k 1 s xp1 , thus keeping
only p 1 momentum containing terms in the decomposition.
Collecting all the contributions Žfor the explicit expressions of all the coefficients see
Appendix A. given above we arrive at the general expression for the virtual corrections
with r s x w1 q Ž úrm. 2 x < srm2 ,
2Re Ý Ž M0) M . k Ip 1 1
11 4 2
2 a p 2yx w p2
s
x1
FF
½ 1yx F
F q 2 Ll Ž 2 y L t y L t 1 y L s y L s1 q L u q L u1 . q
3
r
qLi 2 Ž x . y 101
18 q ln q L2u1 y L2t y L2s1 q Lr ln Ž 1 y x . q 113 L t
1yr
s1 1 t 3 y u13 s1 2 u1 Ž u12 q s12 . y ts12 u1
yq q ln2 q Pq3 ln q ln2
t F s12 t t 2
4 t s1 t
2 u1 Ž u12 q t 2 . y t 2 s1 u1 s1 u1 t u1 s1 t
q
4 ts12
ln2
s1
q
2t
ln
t
q
2 s1
ln
s1
y 34 p 2
ž t
q
s1 /5 ,
Ž 20 .
where we have used the following definitions:
x p2
qs Li 2 Ž 1 y r . y q Li 2 Ž x . q Lr ln Ž 1 y x . ,
ryx 6
s13 y u13 p 1 t 3 y u13

Ps
s1 t 2 a ž 1 y Pt /
y 1 y 13 L t q 59 q
s12 t
p 1
=
a
Re
ž 1 y P s1 /
y 1 y 13 L s1 q 59 ,
P s1 s 13 Ž L s1 y ip . y 59 , F s x 1 A 2 q t 1 x 1 Ž J11 y J1 q xJ1 k y xJk . ,

1 1 s1 yu yu1
ws y , L s1 s ln , L u s ln , L u1 s ln ,
x r m2 m2 m2
yt yt1
L t s ln 2
, L t 1 s ln .
m m2
After integration over x 1 one gets additional large logs of the form L0 s L s q
lnŽ u 02r4.. Terms containing the last factor have to be cancelled against a contribution
coming from the emission of the hard photon outside a narrow cone u - u 0 < 1 Žand
supplied by the same set of virtual and soft corrections., which was considered in Ref.
w14x. In the case of two hard photon emission it is necessary to consider four kinematical
regions, namely when both are emitted insideroutside a cone and one insideranother
outside.
Fortunately enough, the w-structure, which obviously violates the factorization
feature, does not contribute in LLA due to a cancellation of large logs in F . What for a
correction to the above structure coming from P-type graph it vanishes in the sum of FD
with crossed and uncrossed photon legs Žfor a more comprehensive account see
Appendix A..
The total expression can be obtained by summing virtual photon emission corrections
and those arising from the emission of an additional soft photon with energy exceeding
no D ´ < ´ .
The emission of a soft photon is seen as a process factored out of a hard subprocess
Žin our case the latter is exactly a hard collinear photon emission. so this is seemingly
come into an evident conflict with a hard collinear emission. Nevertheless, arguments
similar to those given in the paper devoted to the problem of DIS with a tagged photon
w19,20x may be applied in the present paper: the factorization of the two in the
differential cross section is present and we are, hence, allowed to consider a soft photon
emission restricted as usual by
D´
< 1. Ž 21 .
´
Thus the soft correction can be written as
Ý < M < 2hardqsoft s Ý < M < 2B wsoft Ž k 1 I p1 . , Ž 22 .

a d3 k p1 pX1 p2 pX2 2
wsoft Ž k 1 I p 1 . s y
4p 2
Hv-D ´ 'k 2
q l2 ž y
p1 k
q
pX1 k
q
p2 k
y
pX2 k / ,
where M B denotes the matrix element of the hard photon emission at the Born level and
in the kinematics k 1 I p 1 it reads
2 11a 3p 3
Ý < M < B2 s x1
F F. Ž 23 .
Now let us check the cancellation of the terms containing Ll. Indeed it takes place in the
sum of contributions arising from emission of virtual and soft real photons. To show that
we bring the soft correction into the form
a D´
wsoft Ž k 1 I p 1 . s
p ½ž
2 ln
´
q Ll /Ž y2 q L s q L s1 q L t q L t 1 y L u y L u1 .
q 12 Ž L2s q L2s1 q L2t q L2t 1 y L2u y L2u1 . q ln y 1 Ž L u1 y L s1 y L t 1 .
2p 2 y1
qln y 2 Ž L u y L t y L s1 . q ln Ž y 1 y 2 . y y 12 ln2
3 y2
1 q c1X 2X 1 q c1X 1 y c 2X
qLi 2 ž 2 / q Li 2 ž 2 / q Li 2 ž 2 /
1 y c1X 1 q c 2X
yLi 2 ž 2 / y Li 2 ž 2 /5 , Ž 24 .
where c i are the cosines of emission angles of ith particle with respect to the beam
direction Ž p 1 in cms., c1X 2X is the cosine of the angle between scattered fermions in cms
of the colliding particles and yi are their energy fractions and in the case k 1 I p 1 we
have
1 q c1X 2X 1yx 1 y cX2 y1Ž 1 q c .
cX1 s c, s1y , s . Ž 25 .
2 y1 y 2 2 2 y2 Ž 1 y x .
Then the cancellation of infrared singularities in the sum is evident from comparison
of Eqs. Ž20., Ž24.. The terms with lnŽ D ´r´ . should be cancelled when adding a
contribution of a second hard photon having energy above the registration threshold D ´ .
The complete expression for the correction in the case k 1 I p 1 reads
2
R s 2Re Ý Ž M0) M . q < M < soft
2 11a 4p 2 2yx w D´
s
x1
FF
½ 1yx F
F q 4ln ž / ´
y1 q L t 1 q 12 yln Ž 1 y x . ž
s s1
q2ln / q 113 L t q Ž Lr y L t . ln Ž 1 y x . y L t ln Ž y 1 y 2 . q ln2
yu yt
yu p2
qln y 1 ln Ž 1 y x . q ln Ž y 1 y 2 . 1 q ln ž / y q Li 2 Ž x . y 101
18 y q
s 3
r y1 yu 1 q c1X 2X
qln y 12 ln2 q ln Ž 1 y x . ln q Li 2 ž /
1yr y2 s 2
1 q c1X 1 y c 2X 1 y c1X 1 q c 2X
qLi 2 ž 2
3
/ q Li 2 ž 2 / y Li 2 ž 2 / ž /
y Li 2
2
1 t y u13 s1 2 u1 u12 q s12
Ž . y ts12 u
2 1
q Pq3 ln q ln
F s12 t yt 4 t 2 s1 t
2 u1 Ž u12 q t 2 . y t 2 s1 yu s1 u1 t yu
q ln2 q ln q ln
4 ts12 s 2t t 2 s1 s
s1 t
y 34 p 2
ž t
q
s1 /5 , Ž 26 .
1
d s Ž k 1 I p1 . s R dG .
2 p 5s
11
5. Kinematics k 1 I pX1
We put here a set of replacements one can use in order to obtain the modulus of
matrix element squared and summed over spin states for the case k 1 I pX1 , starting from
l l
the analogous expression for k 1 I p 1 ŽEq. Ž14.. and using the replacement of momenta
p 1 ypX1 , p 2 ypX2 . The last operation results in the following substitutions:
x y ™ x
1yx
, x 1 yx 1X , s s1 , ™
u u1 , t t, t1 t1 .l l ™ ™
Ž 27 .
Then under these permutations the expression for virtual corrections given in Eq. Ž20.
gets transformed yielding the following result for the collinear domain k 1 I pX1:
2 11a 4p 2 2 y x w̃
2Re Ý Ž M0) M . k Ip s 1
X
1
x 1X
F˜F˜
½ 1 y x F˜
F˜ q 2 Ll Ž 2 y L t y L t 1 y L s y L s1
p2 yx j
qL u q L u1 . q
3
q Li 2 ž 1yx / y 101
18 q ln ž /
jq1
s
qL2u y L2t y L2s y Lj ln Ž 1 y x . q 113 L t q ln2
yt
3 3
1 t yu s 2 u Ž u q s . y ts 2
2 2
q Pq3 ln q
F˜ s2 t yt 4t2s
u 2 uŽ u2 q t 2 . y t 2 s yu s u
=ln2 q 2
ln2 q ln y q˜
t 4 ts s 2t t
t yu s t
q
2s
ln
s
y 34 p 2 ž t
q
s /5 , Ž 28 .
with
s 3 y u3 p 1 t 3 y u3 p 1
P̃ s
st 2
a ž 1 y Pt /
y 1 y 13 L t q 59 q 2
s t a
Re
ž 1 y Ps
y1
/
s t 2
1yx 1 x 1X
1
y Ls q
3
5
9 , F˜ s ž q q1 , / w˜ s y q , js
t s x j m2
and F˜ ,q˜ derived upon applying a set of replacements from Eq. Ž27. on the quantities
F ,q .
The contribution from the soft photon emission is described by
a D´ 1yc
wsoft Ž k 1 I pX1 . s ž
4 ln q Ll /ž y1 q L s q ln q 12 ln Ž 1 y x . /
p ´ 1qc
1yc 1yc
qL2s q 2 L s ln y 12 ln2 Ž 1 y x . q ln Ž 1 y x . q ln2
1qc 2
2
1qc 2p 1qc 1yc
yln2 y q 2Li 2 ž / y 2Li 2 ž / . Ž 29 .
2 3 2 2
The total correction for the case k 1 I pX1 has the following form:
R˜ s 2Re 2
Ý Ž M0) M . q < M < soft
2 11a 4p 2 2 y x w˜ D´
s
x 1X
F˜F˜
½ 1 y x F˜
F˜ q 4ln ž / ´
1yc p2 yx
= y1 q L s q 12 ln Ž 1 y x . q ln
ž / q q Li 2 ž /
1qc 3 1yx
j s
y 101
18 q ln ž /
jq1
y 2ln2 Ž 1 y x . q 113 L t y Lj ln Ž 1 y x . q ln2
yt
1qc 1yc
y 23 p 2 q ln Ž 1 y x . y q˜ q 2Li 2 ž / y 2Li 2 ž /
2 2
1 t 3 y u3 s 1 u 1
q P˜ q 3 ln q ln2 ž / Ž 2 u Ž u 2 q s 2 . y ts 2 . q 4 ts 2
F˜
2 2
s t yt 4t s t
yu s u t yu s t
=ln2 ž /Ž
s
2 uŽ u2 q t 2 . y t 2 s . q
2t
ln
t
q
2s
ln
s
y 34 p 2 ž t
q
s /5 ,
Ž 30 .
1
d s Ž k 1 I pX1 . s R˜ d G .
2 p 5s 11
Performing the integration over a hard photon angular phase space Žinside a narrow
cone. we put the RC to the cross section coming from virtual and soft real additional
X
Fig. 2. The ratio J r L t JL Žsee Eq. Ž31.. versus x s v 1 r ´ for the case: Ža. k 1 I p1 ; Žb. k 1 I p1 .
photons valid to a logarithmic accuracy in the form

d s g ŽVqS . d s 0g a D´
s C qL t JLqJ . Ž 31 .
d xd c d xd c p ´
In Fig. 2a,b the ratio of JrL t JL versus x for the collinear kinematics considered
above is given.
6. Two hard photon emission and results in LLA
Turning to the structure of the result obtained, it should be noted that all the terms
quadratic in large logarithms L t 1 ; L s1 ; L u 4 Lr are mutually cancelled out as it
should be.
From the formula Ž26. it immediately follows that Župon doing an integration over a
hard photon angular Žwithin a narrow cone. phase space. the w-term that is not
proportional to F , which is in fact the kernel of the non-singlet electron structure
function, is not dangerous in the sense of a feasible violation of the expected Drell–Yan
form of the cross section, because it does contribute only at next-to-leading order.
Performing the above mentioned integration and confining ourselves to LLA we get
for the sum of virtual and soft photons
d s g Ž SqV . d s 0g a D´
s L 4ln q 113 y 12 ln Ž 1yx . yln Ž y 1 y 2 . . Ž 32 .
d xd c d xd c p ´
The LLA contribution coming from the emission of a second hard photon with total
energy exceeding D ´ consists of a part corresponding to the case in which both hard
photons Žwith total energy ´ x . are emitted by the initial electron w10x
d s 2g d s 0g a x PQŽ 2 . Ž 1 y x . D´
s L 2
q 12 ln Ž 1 y x . y ln y 34 , Ž 33 .
d xd c d xd c p 4 Ž1 q Ž 1 y x . . ´
1qz2 1qz
PQŽ2. Ž z. s2
1yz
Ž 2ln Ž 1 y z . y ln z q 32 . q 2
ln z y 1 q z ,
and the remaining part which describes the emission of a second hard photon along
scattered electron and positrons. The latter, upon combining with the part of contribu-
tions of soft and virtual photons to our process,
d s 0g 3 a D´
L ln q 34 ,
d xd c p ´
may be represented via electron structure function in the spirit of the Drell–Yan
approach
2
d s 0g a 1q Ž1yx .
¦ ;d xd c k 1 Ip 1
s
2p x
L0 d z 2 d z 3 d z 4 D Ž z 2 . D Ž z 3 . D Ž z 4 .
H
d s 0 Ž p 1 Ž 1 y x . , z 2 p 2 ;q1 ,q2 .
= , Ž 34 .
dc
with the non-singlet structure function D Ž z . w21x

a a 2 1
DŽ z . sd Ž1yz . q
2p
L P Ž1. Ž z . q žp/
2
L
2!
P Ž2. Ž z . q . . . , Ž 35 .
™0
P Ž1 ,2. Ž z . s lim d Ž 1 y z . PŽD1,2 . q Q Ž 1 y D y z . PQŽ1 ,2. Ž z . 4 ,
D
1qz2
PDŽ1. s 2ln D q 32 , PQŽ1. Ž z . s ,
1yz
2 2p 2
PDŽ2. s Ž 2ln D q 32 . y , ...
3
These functions describe the emission of Žreal and virtual. photons both by final
electron and by positrons. The multiplier before the integral stands for the emission of a
hard photon by the initial electron. Thus Eq. Ž34. actually represents the partially
integrated Drell–Yan form of the cross section. Quite the same arguments are applicable
to the second case in which a hard photon is emitted by the final electron.
The cross section of the hard sub-process eŽ p 1 z 1 . q eŽ p 2 z 2 . eŽ q1 . q eŽ q2 . enter-
ing Eq. Ž34. has the form
™
d s 0 Ž z 1 p 1 , z 2 p 2 ;q1 ,q2 .
dc
2
8pa 2 z 12 q z 22 q z 1 z 2 q 2 c Ž z 22 y z 12 . q c 2 Ž z 12 q z 22 y z 1 z 2 .
s 2
. Ž 36 .
s z1Ž 1 y c . Ž z1 q z 2 q c Ž z 2 y z1 . .
The momenta of scattered electron q1 and positron q2 are completely determined by the
energy–momentum conservation law
2 z1 z 2 $
q10 s ´ , q10 q q20 s ´ Ž z 1 q z 2 . , c s cosq1 , p1 ,
z1 q z 2 q c Ž z 2 y z1 .
$ $
z 1 sinq1 , p1 s z 2 sinq2 , p1 .
In general their energies differ from those detected in experiment ´ 1X , ´ X2 , namely
´ 1X s q10 z 3 , ´ X2 s q20 z 4 ,
whereas the emission angles are the same in LLA.
Collecting the two expressions presented in Eqs. Ž32., Ž33. one can rewrite the result
in LLA as
ds g d s 0g
d xd c k 1 Ip 1
s ž d xd c / k 1 Ip 1
1 q d14 ,
²d s 0grd xd c : a x PQŽ 2 . Ž 1 y x .
d1 s
ž d s 0grd xd c / k 1 Ip 1
y1q
p
L 2
3 y ln Ž y 1 y 2 . q
4 Ž1 q Ž 1 y x .
2
.
.
Ž 37 .
For the case k 1 I pX1 the correction is found to be

ds g d s 0g
d xd c X
k 1 Ip 1
s ž d xd c / k 1 Ip 1
X
1 q d 1X 4 ,
²d s 0grd xd c : a x PQŽ 2 . Ž 1 y x .
d 1X s
ž d s 0grd xd c / X
k 1 Ip 1
y1q
p
L 2
3 q
4 Ž1 q Ž 1 y x .
2
.
,
2
d s 0g a 1q Ž1yx .
¦ ;
d xd c X
k 1 Ip 1
s
2p x
LX0 d z 1d z 2 d z 4 D Ž z 1 . D Ž z 2 .
H
d s 0 Ž z 1 p 1 , z 2 p 2 ;q1 ,q2 .
= D Ž z4 . , Ž 38 .
dc
with LX0 s L0 q 2lnŽ1 y x ..
For the case when the energies of scattered fermions are not detected, the expressions
Ž34., Ž38. may be simplified due to Hd z D Ž z . s 1 and z 3 , z 4-independence of the
integrand in k 1 I p 1 kinematics Ž z 4-independence in the k 1 I pX1 case..
The x-dependence of d 1 is shown in Fig. 3 for different values of the cosine of
scattering angle c. For a hard photon emission by final particles the correction d 1X
strongly depends on the experimental conditions of particle detection: the energy
thresholds of detection of scattered fermions. This dependence for d 1 is much weaker,
namely about 1%.
In conclusion let us recapitulate the results given in Eqs. Ž37., Ž38.. They both respect
the Drell–Yan form for a cross section in LLA. Nevertheless a certain deviation away
from RG structure function representation at a second order of PT in k 1 I p 1 kinematics
is observed. The term destroying expectations based on RG approach comes from
Fig. 3. The x-dependence of d 1 Žsee Eq. Ž37.. for different values of the cosine of scattering angle c.
Other parameters chosen are: u 0 s 0.1, ´ s1 GeV.
definite contribution of a soft photon emission, the term with lnŽ y 1 y 2 . in Eq. Ž37. which
mere consequence of a complicate kinematics of 2 3 type hard subprocess Žsee w14x.; ™

cannot be included into the structure function approach. Its appearance is presumably a
for such a kind of process the validity of the Drell–Yan form for a cross section was not
proved so far. Another possible way out is a careful analysis of a conflict between a soft
and hard collinear photon emission. We have used the factorized form of a soft photon
emission Ž22. under the condition Ž21.. But, to the moment, this representation in the
peculiar case at hand is not rigorously proved as well.
The accuracy of our calculations of virtual and soft photon corrections is determined
by the omitted terms of the order of
a m2 a
ž
1 q O u 02
p
Ls ,
s p
Ls ,
/ Ž 39 .
which corresponds to a per mille level. The accuracy of the correction coming from two
hard photon emission is determined by O ŽŽ arp .lnŽ4ru 02 .. and at 1% level.
Acknowledgements
We thank A.B. Arbuzov for a critical reading of the manuscript, many valuable
comments and participating at the early stage of the investigation. We are also indebted
to L. Lipatov for many discussions elucidating factorization issues. One of us ŽEAK. is
grateful to the physical department of Insubria University ŽComo. for a warm hospitality
during accomplishment of the final part of this work and to the Landau Network-Centro
Volta grant for financial support. The support of EAK by INTAS grant 93-1867 ext. and
of EAK and BGS by Russian Foundation for Basic Research grant 99-02-17730 is
acknowledged.
Appendix A
Here we give the expressions for the quantities associated with G-type integrals:
1 p2
Jsy y2 Ll L t 1 q 2 L t 1 Lr y L2t y 2Li 2 Ž x. y ,
x 1 t1 6
1 r dz ln z A x Aqq
J1 s
t1 x 1
H0 1yz 1ylz
s
t1 x 1 ž 1q
ryx / s
t1 x 1
,
1 r dz z ln z
Jk s y H ,
t1 x 1 r 0 1yz 1ylz
1 r dz z ln z
J11 s y
t1 x 1
H0 Ž1yz . Ž1yl z . ž 1q
1yz / ,
1 r zd z z ln z
. ž /
J1 k s H 1q ,
t1 x 1 r 0 Ž1yz . Ž1yl z 1yz
p2 x x1
A s Li 2 Ž 1 y r . y q Li 2 Ž x . q Lr ln Ž 1 y x . , ls , rs .
6 r m2
Ž A.1 .
In the limit r 4 1 we have
F s x 1 A 2 q t 1 x 1 Ž J11 y J1 q xJ1 k y xJk . s y 12 q O Ž ry1 .
and that is the reason why w-structure does contribute only to next-to-leading terms.
In general the expression for 5-denominator one-loop scalar, vector and tensor
integrals are some complicated functions of five independent kinematical invariants Žin
the derivation we extensively use the technique developed in Ref. w22x.. In the limit
m 2 < x 1 < s ; yt they may be considerably simplified because of singular 1rx 1
terms only kept:
1 1
Es D 0124 q D 0123 ,
s1 t
1
E1 s yxEk s Ž D 0134 y Ž 1 y x . D 0234 y xD1234 q x 1 E . ,
2 x1
1 x x 2p 2
D 0124 s L2r q 2 Lr ln y ln 2
y ,
xt1 x 1 1yx 1yx 3
1 p2
Re D 0123 s L2s1 y 2 L s1 Lr y 2 L s Ll q q 2Li 2 Ž x . ,
sx1 6
1 5p 2
Re D 0234 s L2s1 q 2 L s1 Ll y 2 Lr L s1 q 2 L s1 L t y ,
s1 t 6
1 7p 2
Re D 0134 s L2s q 2 L s Ll y 2 Ž L t 1 q ln Ž x . . L s q 2 L s L t q ,
st 6
1 7p 2
Re D 1234 s y yL2s q 2 L s Ž L t 1 q ln Ž x . . q 2 L s1 Ll y . Ž A.2 .
s1 xt 1 6
The structure E11 q xE1 k has the form 1rŽ s x 1 . f Ž x, x 1 . and will vanish after performing
the operation Ž1 q Q 2 . s1 tP given in Ž14. which yields a contribution of P-type graphs
with crossed and uncrossed photon legs.
The following coefficient for the scalar integral is obtained in the calculation of
B-type FD:
1 p2
Bs L2s1 q 2 L s1 Ll y 2 L s1 Lr q 2 L s1 L t q . Ž A.3 .
s1 t 6
For the vector integral coefficients we get

1
asy yp 2 s1 q 2 u1 Li 2 Ž 1 y r . y s1 L2t q tL2s1 y 2 tL s1 L t ,
2 s1 u1 t
1 2p 2
bsy q 2Li 2 Ž 1 y r . y 2 L2s1 q 4 L s1 Lr y 2 L s1 L t ,
2 s1 t 3
1 p2
cs 2 u1 Li 2 Ž 1 y r . q Ž 4 u1 q 6 t . q Ž t y 2 u1 . L2s1
2 s1 u1 t 6
ys1 L2t q 4 u1 L s1 Lr q 2 s1 L s1 L t . Ž A.4 .
The relevant quantities for tensor B-type integrals are:

1 r 1 2
a1X 2X s ž Lr y L t , / ag s y Ž L s y Lt . qp 2 ,
s1 t ry1 4 u1 1
1 2 1 1 r
a1X 2 s y Ž Lt y L s . qp 2 q Ž Ls y Lt . y s ž /
Lr y L s1 ,
2 u12 1
tu1 1
1t ry1
1 3
4p 2
J0 s 2 L2s1 y 2 L s1 Lr y Li 2 Ž1yr . y . Ž A.5 .
s1 3
As has been mentioned in the text, the physical gauge exploited provides a direct
extraction of the kernel of the structure function out of the traces both in the tree- and
loop-level amplitudes. The pattern emerging
Ž pˆ 1 y kˆ 1 q m . eˆ Ž pˆ 1 q m . eˆ Ž pˆ 1 y kˆ 1 q m .
2
s 4 Ž p 1 e . Ž pˆ 1 y kˆ 1 . y e 2x 1 kˆ 1 f Ž 1 y x . Ypˆ 1 ,
2yx
kˆ 1 eˆ Ž pˆ 1 q m . eˆ Ž pˆ 1 y kˆ 1 q m . f Ž 1 y x . 2 ž /
W y Y pˆ 1 , Ž A.6 .
1yx
shows this clearly.
References
w1x A.B. Arbuzov et al., Phys. Lett. B 383 Ž1996. 238.

w2x B.F.L. Ward et al., Phys. Lett. B 450 Ž1999. 262.
w3x G. Montagna et al., Nucl. Phys. B 547 Ž1999. 39.
w4x W. Beenakker et al., Nucl. Phys. B 349 Ž1991. 323.
w5x F.A. Berends et al., Nucl. Phys. B 206 Ž1982. 61.
w6x G. Montagna et al., Nucl. Phys. B 401 Ž1993. 3.
w7x W. Placzek et al., preprint CERN-THr99-07, Geneva Ž1997., hep-phr9903381.
w8x M. Caffo, R. Gatto, E. Remiddi, Nucl. Phys. B 252 Ž1985. 378.
w9x A.B. Arbuzov et al., J.H.E.P. 10 Ž1997. 001.
w10x A.B. Arbuzov et al., Nucl. Phys. B 483 Ž1997. 83.
w11x A.B. Arbuzov et al., Nucl. Phys. B 474 Ž1996. 271.
w12x A.B. Arbuzov et al., Phys. Atom. Nucl. 60 Ž1997. 591.

w13x A.B. Arbuzov et al., Mod. Phys. Lett. A 13 Ž1998. 2305.
w14x A.B. Arbuzov et al., J.E.T.P. 88 Ž1999. 213.
w17x V.N. Baier et al., Nucl. Phys. B 65 Ž1973. 381.
w18x A. Bassetto, M. Ciafaloni, G. Marchesini, Phys. Rep. 100 Ž1983. 201.
w19x H. Anlauf et al., J.H.E.P. 10 Ž1998. 013.
w20x H. Anlauf et al., Phys. Rev. D 59 Ž1999. 14003.
w21x E.A. Kuraev, V.S. Fadin, Sov. J. Nucl. Phys. 41 Ž1985. 466.
w22x W.L. van Neerven, J.A.M. Vermaseren, Phys. Lett. B 137 Ž1984. 241.
Isolated-photon production in polarized pp collisions

Stefano Frixione a , Werner Vogelsang b
a
CERN, TH DiÕision, GeneÕa, Switzerland
b
C.N. Yang Institute for Theoretical Physics, State UniÕersity of New York at Stony Brook, Stony Brook, NY
11794-3840, USA
Received 19 August 1999; accepted 9 September 1999
Abstract
We perform a detailed study of the production of isolated prompt photons in polarized

hadronic collisions, in the centre-of-mass energy range relevant to RHIC. We compare the results
obtained for a traditional cone-isolation prescription, with those obtained by imposing an isolation
condition that eliminates any contribution to the cross section from the fragmentation mechanism.
The latter prescription will allow us to present the first fully consistent next-to-leading order
calculation in polarized prompt-photon production. We will discuss the theoretical uncertainties
affecting the cross section, addressing the issue of the reliability of the perturbative expansion, for
both inclusive isolated-photon and photon-plus-jet observables. Finally, we will study the depen-
dence of our predictions upon the polarized parton densities, and the implications for the
measurability of the gluon density. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 13.88.qe; 13.85.Qk; 12.38.Bx
1. Introduction
A direct measurement of the gluon distribution in the proton is both very interesting
and very difficult. This is true, in particular, for the spin-dependent gluon density D g of
a longitudinally polarized proton. So far, the only information on the nucleon polarized
parton distributions comes from polarized deep-inelastic lepton–nucleon scattering
ŽDIS.. Here, in principle, D g could be determined from scaling violations; however, in
practice this procedure is afflicted by very large uncertainties due to the limited accuracy
of the data and, in particular, to the fact that so far only fixed-target DIS experiments
have been carried out in the polarized case, which, consequently, have a rather limited
lever arm in Q 2 . Thus, the proton spin-dependent gluon density D g is experimentally
constrained only very little so far.
In order to perform a direct determination of the gluon distribution, one needs to
consider physical processes, which are predominantly initiated by gluons at the parton
level; the contamination from quark-initiated subprocesses must be under good theoreti-
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 5 7 5 - 1
S. Frixione, W. Vogelsangr Nuclear Physics B 568 (2000) 60–92 61
cal control; and finally, the process must take place at a measurable rate, taking
experimental efficiencies into proper account. Only a few processes are known that meet
all these requirements. One example is the production of heavy flavours in photon–
hadron collisions; the cross section for this process in the case of polarized scattering
has recently been computed w1,2x in QCD at next-to-leading order ŽNLO.. The COM-
PASS collaboration at CERN w3x, and possibly even HERA in the polarized configura-
tion w4–6x, will be able to exploit charm production in order to constrain the polarized
gluon density. Such measurements are on the other hand severely limited by the low
experimental efficiency of charmed-meson tagging.
A second possibility is given by jet hadro- and photoproduction; also in this case,
QCD cross sections for polarized collisions are now known to NLO w7,8x. It will indeed
be attempted to determine D g, at the forthcoming polarized RHIC pp collider w9,10x, by
a measurement of the spin asymmetry in jet production, and the recent study of Ref. w7x
has demonstrated that this approach has very promising prospects.
™
In the unpolarized case, the classical tool for determining the gluon density at
intermediate and large x has been prompt-photon production, pp g X and pN g X,
in fixed-target experiments w11x. Indeed, data on prompt photons have been the
™
backbone of the gluon determination in many analyses of parton densities. The main
™ ™
reason for this is that, at leading order, a photon in the final state is produced in the
reactions qg g q and qq g g, with the contribution of the former subprocess being
obviously sensitive to the gluon and usually dominant over that of the latter. It is the
‘point-like’ coupling of the photon to the quark in these subprocesses that is responsible
for a much cleaner signal than, say, for the inclusive production of a p 0 , which proceeds
necessarily through a fragmentation process.
The cleanliness of the signal is of course an advantage that also counts at colliders.
The aim of this paper is to provide a detailed study, performed in perturbative QCD at
NLO accuracy, of the production mechanism for prompt photons at polarized hadronic
colliders, such as RHIC. In the rest of this introduction, we will briefly review the
present theoretical knowledge of prompt-photon production. In particular, we will
explain why, in our opinion, a theoretical reappraisal of this matter is needed, before the
data-taking will start at RHIC.
Prompt-photon data obtained at hadronic colliders have been used as a constraint for
the unpolarized gluon density w11x. Thinking of the polarized case again, it is clear that
the production of photons with polarized beams at RHIC is likely to be a very promising
source of information on D g. We note that having pp reactions Žas opposed to pp ones
™
as hitherto in the unpolarized case. is also an advantage, since the competing LO
subprocess qq g g does not receive any contributions from valence–valence scatter-
ing here. Compared to jets, the prompt-photon reaction shows a larger spin asymmetry,
even though of course the jet rate is much higher at a given p T , resulting in smaller
statistical errors. Eventually, both reactions, plus also charm production and p 0 produc-
tion in polarized pp scattering, can be utilized to constrain D g at RHIC, and it will be
interesting to see how far the various channels will provide compatible pieces of
information, andror whether they will complement one another.
Unfortunately, the cleanliness of the prompt-photon signal alleged above is limited.
As is well known, photons can also be produced through a fragmentation process, in
which a parton, scattered or produced in a QCD reaction, fragments into a photon plus a
number of hadrons. The problem with the fragmentation component in the prompt-pho-
62 S. Frixione, W. Vogelsangr Nuclear Physics B 568 (2000) 60–92
ton reaction is twofold: first, it brings us back to the dependence on non-perturbative

fragmentation functions, similar to the case of pion production addressed above, even
though for prompt-photon production of course only a certain part of the total signal
depends on the fragmentation functions. So far, the photon fragmentation functions are
only insufficiently known; first information is emerging from the LEP experiments
™
w12,13x. Secondly, all QCD partonic reactions contribute to the fragmentation compo-
™
nent; thus the advantage of having a priori only one partonic reaction Ž qq g g .
competing with the signal Ž qg g q . is lost, even though some of the subprocesses
relevant to the fragmentation part at the same time result from a gluon initial state.
Numerical studies w14–17x for photon production in unpolarized collisions, based on
predictions w14,18–20x for the photon fragmentation functions that turned out to be
compatible with the sparse LEP data, demonstrate that the fragmentation component is
not messing up things too much, even though it cannot be neglected in a careful study.
In the fixed-target regime, it amounts to an effect of about 20%. At collider energies, it
would easily make up for about half of the observed photons; however, here the
situation is saved by the so-called ‘isolation’ cut, which is imposed on the photon signal
in experiment. Isolation is an experimental necessity: in a hadronic environment the
study of photons in the final state is complicated by the abundance of p 0 ’s, eventually
decaying into pairs of g ’s. The isolation cut simply serves to improve the signal-to-noise
ratio: if a given neighbourhood of the photon is free of energetic hadron tracks, the event
is kept; it is rejected otherwise. In principle, there is a large freedom in the choice of
™
specific isolation cuts, the only requirement being that they must strongly suppress the
background p 0 gg , while keeping the signal at a measurable rate. Traditionally,
isolation is realized by drawing a cone of fixed aperture in azimuthal angle-pseudorapid-
ity space around the photon, and by restricting the hadronic transverse energy allowed in
this cone to a certain fraction Žof the order of less than 10%. of the photon transverse
energy. In this way, it is clear that the fragmentation contribution, resulting from an
essentially collinear process, will be diminished w21x. In actual numbers, it is not
expected w15,16x that it will remain responsible for more than 15–20% of the photon
signal after isolation.
Studies of the backgrounds to prompt-photon production expected for RHIC have
been reported in Refs. w22–26x, based on parton-shower Monte Carlo event generators. It
is anticipated w22–24x that isolation cuts will also have to be applied in prompt-photon
measurements at polarized RHIC. When working out theory predictions for the
Žun.polarized cross sections and the resulting spin asymmetry, it is crucial that the
calculation properly treats all effects mentioned so far, in particular those related to
fragmentation and isolation.
It is the objective of this paper to do just this. Of course, it must be pointed out that
basic idea of determining D g in ™™ ™

several studies with a similar focus have been presented before. In particular, after the
p p g X had been formulated and developed in Ref.
w27–34x, the QCD corrections to the ‘direct’ Ži.e. non-fragmentation. component of
polarized prompt-photon production were first calculated in Refs. w35–37x and applied
for phenomenological predictions in Refs. w38,39x. There, the fragmentation component
was neglected altogether, and no isolation cut was imposed. The calculations presented
in Refs. w35–37x are fully analytical and can be used only for calculating the single-in-
clusive photon cross section, and not for looking at, say, photon-plus-jet final states.
Still, it was demonstrated w40x how the effects of the isolation cut can be implemented
into these calculations in the approximation that the isolation cone be rather narrow.
photons in Ref. w41,42x and for ™ ™

™ g q jet q X in Ref. w43x. In these papers, a Monte
Much more complete phenomenological studies were presented for the case of inclusive
pp
Carlo code for the NLO corrections to the direct part of the cross section was developed
and employed, which readily allows the isolation constraints to be taken into account.
The results presented in Refs. w41–43x suffer from the fact that the fragmentation
component to the polarized cross section could be treated at the LO level only, since the
corresponding NLO corrections had not yet been calculated. This is potentially haz-
ardous, since beyond LO the direct as well as the fragmentation part of the cross section
depends on the factorization convention chosen in subtracting the final-state collinear
singularities. Only their sum is free of scheme-dependence and is physical, which
implies that a fully consistent NLO calculation affords knowledge of both production
mechanisms at NLO, even though using a LO fragmentation contribution instead of the
NLO one presumably introduces only a minor error from a numerical point of view. We
emphasize that this situation has not improved since then, and it is not the purpose of
this paper to do this. Still, this paper will present the first fully consistent NLO
calculation of polarized prompt-photon production.
We believe that we have several good reasons for presenting a further phenomenolog-
ical study on prompt photons at RHIC. First of all, our predictions will be based on the
use of a new type of photon isolation constraint, introduced recently by one of us w44x,
which has the virtue of entirely eliminating the unwanted fragmentation component to
the cross section. The basic idea of Ref. w44x is not only to restrict the total hadronic
energy falling into the isolation cone, but to allow less and less hadronic energy, the
closer to the photon it is deposited, until eventually no energy at all is allowed exactly
collinear to the photon. In this way, no collinear configuration is possible, and
fragmentation does not contribute to the cross section. This feature allows us to present
complete and theoretically consistent NLO predictions for polarized prompt-photon
production. This will be done in terms of a dedicated Monte Carlo program, which
implements the new type of isolation in both polarized and unpolarized hadronic
collisions. The isolation constraint we promote should be straightforwardly imple-
mentable in experiment. Here, we have in mind in particular the PHENIX detector at
RHIC, with the very fine granularity of its electromagnetic calorimeter w22,23x; also the
STAR detector w24x, with its much larger angular coverage and its ability to see jets,
appears to offer promising prospects.
Another motivation for our study is to look in more detail at the main theoretical
uncertainties in the calculation, resulting from the scale dependence of the results.
Furthermore, previous theoretical studies have not really sufficiently addressed the
question of the experimentally achievable statistical accuracy in the various conceivable
measurements of the spin asymmetries in prompt-photon production. This will also be
done in this paper. Here, we somewhat disagree with some previous conclusions
concerning the usefulness of inclusive measurements compared with photon-plus-jet
ones.
Presenting a study on prompt photons, we cannot ignore a disagreeable development
found in the unpolarized case in the last few years concerning the comparison between
theory predictions and data. While things worked out very well in the first decade or so
of prompt-photon experiments, the agreement with the more recent, and the most
precise, data sets w45–49x is rather poor and sometimes so bad that the situation cannot
possibly be saved even by ‘fine’-tuning the unpolarized gluon density! Clearly, if this
situation persists, we will have to worry about whether one can really interpret, in the
polarized case, future RHIC data straightforwardly in terms of D g. A possible remedy
for this trend has been brought forward in terms of a smearing of the transverse
momenta of the initial partons participating in the hard scattering w50–53x, required to be
substantially larger than what is already introduced by the NLO calculation. This
approach still remains to be set on a more solid foundation – eventually it should be
accounted for to some extent by a k T-resummation calculation w54x with perturbative as
well as non-perturbative components. Furthermore, threshold resummations w55,56x,
aiming at the high-pT end, have been shown w57x to lead to a certain improvement in the
fixed-target regime. We also note that possible inconsistencies between the various data
sets themselves have been pointed out w17x. It remains to be seen whether or not the
perform the first measurements on ™™ ™

agreement between data and theory will be in better shape by the time RHIC will
p p g X, as a result of the present experimental
and, in particular, theoretical efforts in this field. Whatever the solution will eventually
be, NLO theory, as employed in this work, will certainly be an indispensable part of it; it
should then be straightforward to implement the lessons from the unpolarized case to the
polarized one. It should not be forgotten either that RHIC itself should be able to
provide new and complementary information also in the unpolarized case – never before
have prompt-photon data been taken in pp collisions at energies as high as 'S s 200–
500 GeV.
The remainder of the paper is organized as follows. In Section 2 we will provide the
framework for our calculations. The new isolation definition will be presented in more
detail, as well as the main ingredients for our Monte Carlo code. The remaining sections
are devoted to numerical studies for RHIC. Section 3 deals with the single-inclusive
cross section. In Subsection 3.1 we focus on issues related to perturbative stability,
theoretical uncertainties and the effects of isolation. Subsection 3.2 is devoted to studies
of spin asymmetries for the prompt-photon process at RHIC and of the sensitivity to
D g. In Section 4 we consider more differential variables, such as photon-plus-jet ones.
Here, Subsection 4.1 discusses some general features of photon-plus-jet observables and
also addresses their perturbative stability, while Subsection 4.2 presents phenomenologi-
cal results. Finally, we summarize our work in Section 5.
2. Isolated photons in perturbative QCD

2.1. Isolation prescriptions
The production of isolated photons in hadronic collisions can be written in perturba-
tive QCD as follows:
d sA B Ž K A , K B ; Kg . s dx 1 dx 2 f aŽ A. Ž x 1 , m F . f bŽ B . Ž x 2 , m F .
H
=d sâisol
b ,g Ž x 1 K A , x 2 K B ; K g ; m R , m F , mg .
q dx 1 dx 2 dzf aŽ A. Ž x 1 , mXF . f bŽ B . Ž x 2 , mXF .

H
=d sâisol X X Ž c.
b ,c Ž x 1 K A , x 2 K B ; K g rz ; m R , m F , mg . Dg Ž z , mg . , Ž 1.
where A and B are the incoming hadrons, with momenta K A and K B respectively, and
a sum over the parton indices a, b and c is understood. In the first term on the r.h.s. of
Eq. Ž1., denoted as the direct component, the subtracted Žfactorized. partonic cross
sections d sâisol
b,g get contributions from all the diagrams with a photon leg. On the other
hand, the subtracted partonic cross sections d sâisol
b,c appearing in the second term on the
r.h.s. of Eq. Ž1. Ždenoted as the fragmentation component., get contribution from the
pure QCD diagrams, with one of the partons eventually fragmenting in a photon, in a
way described by the Žperturbatively uncalculable but universal. parton-to-photon
fragmentation function DgŽ c.. Eq. Ž1. is to be regarded as a generic expression for the
cross section: it will apply to unpolarized as well as polarized cross sections; in the latter
case one simply has to substitute the parton densities f iŽ h. and the partonic cross sections
d sâisol
b, r with their spin-dependent counterparts, D f i
Ž h.
and d Dsâisol
b, r respectively. Note,
however, that the parton-to-photon fragmentation functions DgŽ c. are always the unpolar-
ized ones since we are not measuring the polarization of the produced photon.
As the notation in Eq. Ž1. indicates, the isolation condition is embedded into the
partonic cross sections. As mentioned in the introduction, for all the isolation conditions
known at present, except that of Ref. w44x, as well as for the case of totally inclusive
Žnon-isolated. photon production, neither the direct nor the fragmentation components
are separately well defined at any fixed order in perturbation theory: only their sum is
physically meaningful. In fact, the direct component is affected by quark-to-photon
collinear divergences, which are subtracted by the bare fragmentation function that
appears in the unsubtracted fragmentation component. Of course, this subtraction is
arbitrary as far as finite terms are concerned. This is formally expressed in Eq. Ž1. by the
presence of the same scale mg in both the direct and fragmentation components: a finite
piece may be either included in the former or in the latter, without affecting the physical
predictions. The need for introducing a fragmentation contribution is physically better
motivated from the fact that a QCD hard scattering process may produce, again through
a fragmentation process, a r meson that has the same quantum numbers as the photon
and can thus convert into a photon, leading to the same signal.
Owing to the presence of the fragmentation remnants, which surround the emitted
photon, the effect of the isolation cuts will be a stronger suppression of the fragmenta-
tion component relative to the direct component, with respect to the case of totally
inclusive photon production. Since the parton-to-photon fragmentation functions are
extremely poorly known, one may adopt two opposite points of view.
1. Define the isolation cuts in order to suppress as much as possible the fragmentation
component. The resulting cross section will be useful to measure the incoming gluon
density or to test the predictions of the underlying theory. In this context, the
unknown fragmentation functions are regarded as uncertainties affecting the theoreti-
cal predictions.
2. Define the isolation cuts in order to keep a non-negligible contribution from the
fragmentation component. The comparison between data and the resulting cross
section will eventually be used to extract the parton-to-photon fragmentation func-
tions. This strategy makes most sense if the initial state is as clean as possible, which
is the case for eqey collisions.
The former criterion leads to the so-called cone approach w21,40,58–61x. After
tagging the photon, one draws a cone of half-angle R 0 around it. The word ‘cone’ can
be misleading, being motivated by eqey physics. Here, the cone is drawn in the
pseudorapidity-azimuthal angle plane, and corresponds to the set of points
½ ( 2
CR 0 s Ž h , f . N Ž h y hg . q Ž f y fg . ( R 0 ,
2
5 Ž 2.
where hg and fg are the pseudorapidity and azimuthal angle of the photon, respectively.
The quantity in Eq. Ž2. is boost-invariant, and is therefore suited to be used in collider
physics. For the photon to be defined as isolated, the total amount of hadronic transverse
energy ET ,had Ž R 0 . found in this cone must fulfill the following condition:
ET ,had Ž R 0 . ( e c p Tg , Ž 3.
where e c is a small number, and p Tg is the transverse momentum of the photon. This
isolation prescription was proven to be infrared-safe at all orders of perturbation theory
in Ref. w62x. The smaller e c , the tighter the isolation. Loosely speaking, for vanishing e c
™
the direct component behaves like log e c , while the fragmentation component behaves
™
like e c log e c . Thus, for e c 0 Eq. Ž1. diverges. This is obvious since the limit e c 0
corresponds to a fully isolated cross section, which cannot be a meaningful quantity,
whether experimentally Žbecause of limited energy resolution. or theoretically Žbecause
there is no possibility for soft particles to be emitted into the cone..
On the other hand, if one actually wants to measure the fragmentation functions, then
the so-called democratic approach should be adopted w63,64x. The basic idea here is to
treat the photon as a QCD parton in a jet-clustering algorithm, and then to impose a cut
on the hadronic energy contained in the ‘jet’, which also contains the photon. This
approach has so far been used only in eqey physics, and we will not discuss it any
further in this paper.
In the spirit of the cone approach, an alternative definition of the isolated photon has
been proposed w44x. After drawing a cone of half-angle R 0 around the photon axis, all
the cones of half-angle R ( R 0 are considered; their definition is identical to the one
given in Eq. Ž2., with R 0 replaced by R. Denoting by ET ,had Ž R . the total amount of
hadronic transverse energy found in each of these cones, the photon is isolated if the
following inequality is satisfied:
ET ,had Ž R . ( eg p Tg Y Ž R . , Ž 4.
for all R ( R 0 . A sensible choice for the function Y is the following:
n
1 y cos R
Y Ž R. s
ž 1 y cos R 0 / , n s 1. Ž 5.
It has been proved in Ref. w44x that such a choice allows the definition of an
isolated-photon-plus-jet cross section, which is infrared-safe to all orders in QCD
perturbation theory and still does not receive any contribution from the fragmentation
mechanism. In this case, therefore, only the first term on the r.h.s. of Eq. Ž1. is different
from zero, and it does not contain any mg dependence. The reader can find all the
technical details concerning the isolation prescription based on Eq. Ž4. of Ref. w44x.
Here, we will just recall the main ideas. The fundamental property of the function Y is
R ™0
lim Y Ž R . s 0, Ž 6.
the function being different from zero everywhere except for R s 0. This implies that
the energy of a parton falling into the isolation cone CR 0 is correlated to its distance Žin
the h – f plane. from the photon. In particular, a parton becoming collinear to the photon
is also becoming soft. When a quark is collinear to the photon, there is a collinear
divergence; however, if the quark is also soft, this divergence is damped by the quark
vanishing energy Žprovided that the energy vanishes fast enough; this condition is not
very restrictive, and the form in Eq. Ž5. easily fulfills it.. When a gluon is collinear to
the photon, then either it is emitted from a quark, which is itself collinear to the photon
– in which case, what was said previously applies – or the matrix element is finite.
Finally, it is clear that the isolation condition given above does not destroy the
cancellation of soft singularities, since a gluon with small enough energy can be emitted
anywhere inside the isolation cone. The fact that this prescription is free of final-state
QED collinear singularities implies that the direct part of the cross section is finite. As
far as the fragmentation contribution is concerned, in QCD the fragmentation mecha-
nism is purely collinear. Therefore, by imposing Eq. Ž4., one forces the hadronic
remnants collinear to the photon to have zero energy. This is equivalent to saying that
the fragmentation variable z is restricted to the range z s 1. Since the parton-to-photon
fragmentation functions do not contain any d Ž1 y z ., this means that the fragmentation
contribution to the cross section is zero, because an integration over a zero-measure set
is carried out.
We stress that the function given in Eq. Ž5. is to a very large extent arbitrary. Any
sufficiently well-behaved function, fulfilling Eq. Ž6., could do the job, the key point
being the correlation between the distance of a parton from the photon and the parton
energy, which must be strong enough to cancel the quark-to-photon collinear singularity.
We also remark that the traditional cone-isolation prescription, Eq. Ž3., can be recovered
from Eq. Ž4. by setting Y s 1 and eg s e c . In the rest of this paper, as a short-hand
notation, we will indicate the ‘traditional’ isolation obtained by imposing Eq. Ž3. as
definition A, and that obtained by imposing Eq. Ž4. as definition B.
At first sight, the new isolation approach appears to be stricter than the traditional
one. On the other hand, the fact that for the new constraint one also considers the angle
between the photon and hadrons in the cone, is a real virtue here: for the traditional
criterion, one would reject a hadron of, say, 2 GeV wherever it is located in the cone,
just because its energy exceeds the limit. Of course, if the cone size is 0.7, and the
hadron has a distance of 0.6 with respect to the photon – why should one want to reject
such an event? This situation is improved with the new constraint: hadrons in the cone
that are still quite far away from the photon are allowed to have more energy than those
close to the photon. In this way, one can well allow a hadron to have 2 GeV, or even
more, at a distance of 0.6. This little example implies that a detailed comparison
between the traditional and the new isolation methods is certainly of some interest, and
this will also be performed in this paper. One can then eventually decide which isolation
to use in actual experiment.
2.2. NLO computer codes
In order to give phenomenological predictions, we will use two different computer

codes in the cases of definitions A and B. As far as definition A is concerned, we use
the NLO program of Refs. w37,40x for the direct part of the cross section. This program
is based on an inclusive calculation of the contributing partonic subprocesses ab g X

to NLO, where X contains a sum over the appropriate partonic final states, fully
™
integrated over their phase spaces. The advantage of this approach is that all contribu-
tions to a given partonic channel, such as virtual and real-emission ones, and collinear
counterterms, can be added before any numerical implementation. In this way, not only
all singularities that appear in the calculation cancel, but there is no need to introduce
any soft or collinear cut-off at intermediate stages of the numerical calculation either.
One therefore ends up with a rather fast and accurate code. However, the drawback of
this is the inclusiveness of the program: there is no handle on, say, an extra jet since the
partons in the final state have been integrated over. As it stands, it would even seem
impossible to implement isolation in such a code since this clearly affords to have
control over partons falling into the isolation cone. However, as was shown in Ref. w40x,
one can fairly straightforwardly make up for this latter deficit, provided the opening of
the isolation cone is not too big. The idea is as follows: at NLO, the isolated cross
section can also be viewed as the non-isolated one minus the cross section for a parton
to be in the cone, having more energy than that allowed by the isolation cut. This latter
‘subtraction’ cross section can be approximated in a fairly simple calculation in the limit
of a narrow isolation cone, since it is dominated by almost collinear quark–photon
configurations. As a result, the ‘subtraction’ piece turns out to behave like A ln R 0 q B
q O Ž R 20 .; A, B are presented in w40x for both the unpolarized and the polarized cases.
Note that the ‘subtraction’ piece itself inevitably depends on the final-state factorization
scale mg introduced in Eq. Ž1..
As we discussed in the previous subsection, for definition A the fragmentation
mechanism contributes Žcf. Eq. Ž1... We also stated in the introduction that in the
polarized case we presently cannot calculate this part at NLO and thus have to stick to a
LO calculation for it. In contrast to this, in the unpolarized case the NLO corrections to
the relevant partonic scatterings are known w65,66x. The calculation and computer code
presented in Refs. w65,66x were also fully inclusive in the sense that the unobserved
partons had been integrated over their full phase spaces. However, as was shown in Ref.
w40x, the above ‘narrow-cone’ approximation can also be used to implement isolation in
the NLO fragmentation component as calculated with the program of Refs. w65,66x.
When calculating the unpolarized prompt-photon cross section for definition A in this
paper Žfor the purpose of computing asymmetries., we will always include the fragmen-
tation part at NLO, making use of the program of Refs. w65,66x, along with the
modifications for isolation developed in Ref. w40x.
The code relevant to definition B works in a completely different way, being fully
exclusive in the variables of the photon and of the Žone or two. final-state QCD partons.
It is based on the formalism presented in Refs. w67,68x, which allows the computation of
any infrared-safe observables for any kind of scattering particles, without requiring
algebraic manipulations on the matrix elements. The formalism adopts the subtraction
method in order to deal with soft and collinear singularities, and therefore both the
matrix elements and the phase space are treated without any approximation. The code
used in this paper is an extension of that presented in Ref. w69x, which deals with the
production of isolated photons in unpolarized hadronic collisions. Notice that the
formalism of Refs. w67,68x, although originally designed for the case of unpolarized
collisions, extends – basically without any modifications – to the case of polarized
collisions. A detailed discussion on this topic can be found in Ref. w7x. We finally
mention the fact that the code outputs the kinematical variables of the photon and of the
final-state partons, plus a suitable weight. Therefore, the isolation condition, the
jet-reconstruction algorithm, and any cuts matching the experimental setup can be
implemented as the final step of the computation algorithm. This allows us to plot as
many observables as we want in one single computer run.
It is easy to see that one can extend the ideas behind the ‘narrow-cone’ approxima-
tion, used for the definition-A code, also to the isolation given by definition B. In this
way, we have been able to compare extensively the results of the two codes. We found
excellent agreement of the two programs over a wide range of kinematical variables, and
also for cone openings of even R 0 s 0.7, if only central values of rapidity are
considered. This suggests the correctness of the two – entirely independent – codes. It
also implies that the ‘narrow-cone’ approximation has a rather large region of validity
and can be well used for practical applications. We recall, however, that the correspond-
ing code is only suitable for fully inclusive photon observables, and not for photon-plus-
jet ones.
3. Inclusive isolated-photon observables
In this section, we study the inclusive properties of isolated photons. More exclusive
observables, such as correlations between the photon and the accompanying jets, will be
discussed in Section 4. We will consider centre-of-mass energies spanning the range
'S s 200–500 GeV. We will carefully investigate the differences induced by the
different isolation prescriptions we deal with in this paper. We will address the issue of
the perturbative stability of our results, and study the dependence of the cross sections
upon the polarized parton densities.
3.1. Effects of isolation and discussion of theoretical uncertainties
Unless otherwise specified, we will use the following parameters, as a default for our
calculations:
1 GeV
R 0 s 0.4, ec s , definition A; Ž 7.
p Tg
R 0 s 0.4, eg s 1, n s 1, definition B. Ž 8.
It is worth emphasizing at this point that we have chosen eg 4 e c : for traditional
isolation A, e c has to be small – otherwise, isolation is totally ineffective. For isolation
B, on the other hand, eg may be chosen large, as we discussed previously. A large eg
only means that one still allows considerable amounts of hadronic energy in the cone,
provided it is deposited far away from the photon. We add that it is actually desirable
theoretically in any isolation to have a ‘large’ value of e Žs eg or e c .: soft-gluon
emission into the cone generates logarithms w21,40,44x of e , with an extra power at each
further order of perturbation theory, which for very small e eventually threaten to spoil
the perturbative expansion. A study on the structure of the logarithms appearing in the
isolated-photon cross section in eq ey collisions has been given in Ref. w62x. Pending a
more thorough investigation of these points in the case of isolation B, it seems likely
that being able to choose eg s O Ž1. is clearly a virtue of this isolation method.
The default value for the factorization and renormalization scales will be indicated by
m 0 , which will be taken equal to the transverse momentum of the photon in the case A,
and equal to half of the total transverse energy of the event in the case B. These two
choices, which slightly differ beyond LO, are due to the different structure of the codes
computing the isolated-photon cross section in cases A and B, as described in Section 2.
It would be possible to set m 0 s p Tg in case B Žhowever, this choice, although formally
correct, is less appropriate than the one adopted here: since the code is fully exclusive in
the variables of the photon and of the final-state partons, the reference scale, which is
directly related to the hardness of the process, should also depend upon the transverse
momenta of the partons.; this would result in differences with our default choice that are
completely negligible when compared to the other sources of theoretical uncertainty
affecting the cross section. We will adopt throughout the two-loop expression for a S , the
LQC D value being that associated with the parton densities used. Our default parton
density sets will be the NLO ‘standard’ set of Ref. w70x ŽGRSV STD. and MRST w53x
for the polarized and unpolarized scattering respectively. In the case of the definition A,
we will use the NLO GRV w18,19x set of parton-to-photon fragmentation functions.
We have to note here that, while the value of LQCD associated with the MRST set
Ž L5 MS s 220 MeV. is close to the central value of the latest PDG world average Žat two
loops, L5 MS s 237q26 w x.
y24 MeV 71 , all the available polarized density sets have a value
which is much lower, consistent with that extracted from DIS data some years ago.
Thus, by adopting the value of LQC D associated with a given set, we have the
unpleasant situation in which, in the computation of asymmetries, the numerator and the
denominator have different L’s. Still, we preferred not to violate the correlation
between the parton densities and LQC D . This correlation is expected to be particularly
strong in the case of the gluon density, which is of great importance here. Since a
smaller LQC D entails a smaller strong coupling, our predictions for asymmetries would
have become somewhat larger than the ones we present below, had we decided to adopt
the same value of LQC D in the polarized and unpolarized cross sections. This situation
has already been encountered in Ref. w7x, for jet physics. There, it has been shown that
using the same LQC D in the polarized and unpolarized cross sections would increase the
asymmetry by 15% Žrelatively. at the most. In the case of photon production, the
difference is even smaller. As we will see, the effect is therefore completely negligible,
with respect to the differences in the predictions of the asymmetries induced by the
choice of different parton densities.
In what follows, in order to assess the importance of the radiative QCD corrections,
we will often compare the NLO and Born results. Throughout the paper, by ‘Born
result’ we will mean the prediction obtained by convoluting the lowest-order partonic
cross sections Ž O Ž a em a S . and O Ž a S2 . for the direct and the fragmentation contributions
respectively. with the NLO-evolved parton densities and, if needed, fragmentation
functions. Also, the two-loop expression of a S will be used. There is of course a lot of
freedom in the definition of a Born-level result. However, we believe that with this
definition one has a better understanding of some issues related to the stability of the
perturbative series. This is especially true in polarized physics, where the data are not
sufficient to determine the parton densities with a good accuracy, and where large
Žartificial. differences can arise between sets fitted at LO or NLO to the available DIS
data. For a detailed discussion on this point, see for example Ref. w7x.
We start by considering the transverse momentum spectrum of isolated photons. In
the lower part of Fig. 1 we plot the Born Žhistogram with symbols. and NLO Žsolid
histogram. results for the polarized cross section, obtained at 'S s 500 GeV with the
isolation definition B. A cut hg - 0.35 has been imposed, which is suitable for the
PHENIX experiment. As can be seen from the figure, the inclusion of the radiative
corrections gives a sizeable effect as far as the normalization is concerned Žin the first
bin, the ratio of NLO over Born result is about 1.8., while the shape is almost unaffected
Žthe Born being only slightly harder than the NLO result.. Since the radiative corrections
are large, one may wonder whether the NLO result is a sensible quantity to compare
with data. A rigorous answer to this question can only come from a complete NNLO
calculation. Lacking that, we study the scale dependence of our results, as customary in
perturbative QCD, to see whether the inclusion of radiative corrections leads to a milder
dependence upon the scales, as compared to the one of the Born result. Here, it is
especially important to study the separate dependence upon the renormalization and
factorization scales, because cancellation effects between the two may hide some
problems. We present the scale dependence of the p Tg spectrum in the upper and central
parts of Fig. 1. There, we show the ratio of the cross section obtained by setting the
Fig. 1. Transverse-momentum spectrum of the isolated photon, in the case of definition B, for polarized pp
collisions at 'S s 500 GeV. The polarized parton densities used are GRSV STD. The scale dependences of
the Born and NLO results are also shown; see the text for details.
scales equal to m 0r2 and 2 m 0 , over the cross section for the default values of the scales.
We stress that only one scale is varied at a time. The renormalization Žfactorization.
scale variation corresponds to the dotted Ždashed. curves; the curves decreasing for
increasing p Tg correspond to m R s m 0r2 and m F s 2 m 0 , respectively. From the figure,
it is apparent that the inclusion of the radiative corrections reduces the scale dependence
in the whole p Tg range considered, with a possible exception in the case of the m F
dependence, for p Tg equal to 16–20 GeV, where there is basically no m F dependence.
The reduction is stronger in the case of m F dependence than in the case of m R
dependence. The fact that there is a point in the p Tg spectrum where there appears to be
no factorization scale dependence is purely accidental; it can be traced back to the
behaviour of the parton densities with respect to the hard scale. In fact, the gluon density
increases with the scale in the x range corresponding to the low-pTg region, while it is
decreasing when the scale is increasing for larger x values, probed when a harder
photon is produced. We can conclude from Fig. 1 that the perturbative expansion seems
to be reliable in this case; in all cases, the radiative corrections reduce the size of the
dependence of the p Tg spectrum upon the scales. We must comment on the fact that this
conclusion is not specific to the kinematical configuration considered in Fig. 1: we
verified that the same kind of behaviour can be seen in a larger hg range Žwe studied
the case y1 - hg - 2., and also at lower centre-of-mass energies Ž'S s 200 GeV..
Furthermore, almost the same results are obtained in the case of unpolarized collisions.
We now turn to the case when the photon is isolated according to definition A. The
results are presented in Fig. 2. In the lower part, we display the ratio of the cross section
over that obtained with definition B. In this case, the scales are fixed to their default
values. The result at the Born level is again displayed as a histogram with symbols. The
Born result in the case of definition A is always higher than that relevant to definition B.
This is easy to understand, since at this order the result for the direct part is independent
of the isolation condition, and the photon isolated with definition A gets a contribution
from the fragmentation part, which is not present in the case of definition B. Things of
course change at NLO: having an additional parton around, the isolation condition is
effective also in the direct part. We must also remark that, in the case of isolation A, the
fragmentation contribution is only included at LO. A consistent computation at NLO
would presumably produce a slightly larger cross section Žfor example, in the case of
unpolarized collisions, the inclusion of radiative corrections in the fragmentation compo-
nent enhances the full cross section at high p T by about 3%.. The effect is much larger
in the case in which there is no isolation condition, and the photon is fully inclusive. We
will further comment on this fact below. As in the previous case, we also studied the m R
and m F dependence of the spectrum; in doing so, the factorization and renormalization
scales of the direct and of the fragmentation components have been set to the same
value: mXF s m F and mXR s m R Žsee Eq. Ž1... The results are displayed in the upper parts
of Fig. 2. Note that for definition A we have an additional pair of lines Ždot-dashed.,
corresponding to the results obtained by varying the final-state factorization scale mg ,
which enters the fragmentation functions. The m F and m R dependence is very similar to
the one relevant to definition B, displayed in Fig. 1, and the same comments made
previously apply here. On the other hand, the mg dependence is extremely small, and
gives a negligible contribution to the theoretical error affecting the cross section. The
almost identical scale dependence in the case of definitions A and B also implies that the
Fig. 2. Transverse-momentum spectrum of the isolated photon, in the case of definition A. The ratio of the
cross section over that obtained with definition B is shown in the lower part. The rest of the figure displays the
scale dependence, at the Born and NLO levels.
ratio of cross sections plotted in the lower part of Fig. 1 is, to a good extent, independent
of the scale choice.
For completeness, we present in Fig. 3 the corresponding predictions for the fully
inclusive non-isolated prompt-photon cross section. We first discuss the scale depen-
dence of the results, displayed again in the upper two parts of the figure. At low p Tg , the
m R dependence turns out to be larger than that of the isolated-photon cross sections,
while at large p Tg the two appear to be pretty similar Žthe isolation condition is less and
less restrictive as the transverse momentum of the photon is increased, since it is more
and more difficult to have a hard parton, in the surroundings of the photon, that does not
pass the isolation cuts.. The mg dependence of the fully inclusive cross section is much
larger than that of the isolated-photon cross section obtained with definition A. How-
ever, its effect is still smaller than that due to m R and m F .
The comparison of the non-isolated cross section with the one isolated according to
definition B is shown in the lower part of Fig. 3. One immediately notices a striking
feature: at NLO, the isolated cross section becomes larger than the unpolarized one at
large p Tg . Clearly, this finding is at odds with the physical expectation that any
meaningful isolation cut should reduce the number of events with respect to the number
obtained for no isolation at all. The origin of the problem we encounter resides in the
fact that the fragmentation contribution to the non-isolated cross section Žand, obviously,
also for the results for isolation A presented in Fig. 2. has only been calculated at the
Fig. 3. Same as in Fig. 2, for non-isolated photons.
LO level since, as we pointed out earlier, the NLO corrections to the fragmentation
contribution have not been calculated so far in the polarized case. We expect that once
the proper NLO fragmentation component is included in the calculation of the polarized
cross section, the disagreeable feature of Fig. 3 will disappear. This view is corroborated
by the observation that we find exactly the same pattern in the unpolarized case: there,
everything can be calculated consistently at NLO, and the non-isolated cross section
turns out to be larger than the one obtained for both types of isolation we consider.
However, we checked that, if we compute the fragmentation contribution only at LO in
the unpolarized case, we indeed obtain a non-isolated cross section that is smaller than
the isolated one for definition B, much as happens in Fig. 3. The figure clearly points
out the importance of consistency in the NLO calculation. A NLO calculation of the
fragmentation component of the polarized prompt-photon cross section is highly desir-
able for the future. On the basis of Fig. 3 we would predict non-negligible positive
corrections to the LO result. For the time being, our present results for isolation A and
the non-isolated case have only limited reliability. Fortunately, fragmentation is really
important only in the non-isolated case, which is not the one relevant to experiment. For
a Žtraditionally. isolated cross section, it contributes a relatively small fraction of the full
result, and this fraction decreases rapidly towards larger p Tg . We are therefore fairly
confident that our predictions for definition A are numerically not too far off the true
NLO answer.
For these reasons, we refrain from performing a detailed study of the uncertainty in
the cross section for isolation A. We only state that we have also calculated the
Fig. 4. Average transverse energy in the cone annuli around the photon.
unpolarized and the polarized cross sections, using set I of the photon fragmentation
functions of Ref. w20x. We find that the fragmentation component to the cross section
decreases by about 23% at p Tg f 10 GeV, and by about 50% at the high-pTg end. For
the full Ži.e. direct plus fragmentation. cross section, the effect is obviously much
smaller, generally below 2%.
There is another striking property of the curve in the lower part of Fig. 2: it is Õery
close to unity over the whole range of p Tg Žin fact, from what we just discussed, we
would expect it to be even closer to unity at large p Tg , had we been able to include the
fragmentation component at NLO in case of definition A.. In other words, the two types
of isolation, albeit so different from a physics point of view, lead to almost identical
cross sections. To some extent, this is certainly due to the choices we made for e c , eg in
Eqs. Ž7., Ž8.: had we chosen, say, eg s e c there1 , then isolation B would have become
stricter than isolation A, and the corresponding curve in Fig. 2 would have been above
unity everywhere. Our choices in Eqs. Ž7., Ž8. presumably created a certain ‘balance’
between the two isolations. However, we found that there is more to the similarity of the
two isolated cross sections. When performing runs at larger values of R 0 , we found that
the cone-size dependence of the cross section is extremely mild for both types of
isolation. This indicates that isolation is most effective close to the photon and does not
affect the cross section too much at larger distances from the photon. To illustrate this
point, we plot in Fig. 4, as a function of the distance R from the photon, the amount of
hadronic transverse energy deposited on average in a cone annulus between R y D R and
R q D R, where D R s 0.025. We do this for isolation of type B, considering two
realistic values of the isolation-cone size, R 0 s 0.4 and R 0 s 0.7, and one extreme value
R 0 s 0.005. We have chosen 'S s 500 GeV, and the photon variables have been
1
This is simply an example; as discussed at the beginning of this subsection, this is actually not a desirable
choice.
integrated over p Tg ) 10 GeV, y1 - hg - 2 Žwe used an extended rapidity coverage in

order to reduce as much as possible the statistical errors on our results; however, rapidity
is not an issue here, and our conclusions apply to any rapidity range.. The reason for
having a plot obtained with R 0 s 0.005 is the following: with such a narrow isolation
cone, the distribution in hadronic transverse energy around the photon is basically
identical Žfor non-zero R . to what would be obtained in the case of non-isolated
photons. Thus, we can have a clear idea of the effect of imposing an isolation condition
with a realistic value for R 0 . By inspection of the solid histogram in Fig. 4, we see that a
sizeable amount of energy is deposited only at small distances from the photon, and – at
least in the framework of a NLO calculation – not much energy is delivered for
R ) 0.2–0.3. This is the reason why there is no real difference between the cross
sections obtained with different, but physically sensible, values of R 0 . Notice that this
conclusion holds regardless of whether we use isolation A or B. To avoid misunder-
standings concerning Fig. 4: of course at NLO we have only one extra particle at our
disposal Žthe third one balancing the other two in transverse momentum., so isolation
only becomes effective if this particle falls into the isolation cone. We are not saying in
™
Fig. 4 that this extra particle is usually soft: the point is rather that, thanks to the
collinear singularity of the corresponding 2 3 matrix elements at R s 0, the extra
particle simply happens to be close to the photon more often than far away from it, so
that on aÕerage more energy is deposited close to the photon. Incidentally, one can
convince oneself that, for small R and D R < R, the quantity shown in Fig. 4 has to be
proportional to D Rrs Ž R 0 . = Ž d s Ž u .rdu .usR , where u is the angle between the
momenta of the photon and the other particle in the cone, and s Ž R 0 . is the total cross
section for a given isolation-cone size R 0 and a given kinematical range for the photon
variables Žhere, p Tg ) 10 GeV and y1 - hg - 2.. For a quark parallel to the photon
Žwhich is the only configuration producing a collinear singularity at u s 0., one thus
finds immediately that the curves in Fig. 4 should fall A 1rR if R ) R 0 , i.e. outside the
isolation cone. This is exactly the pattern observed in the figure. Inside the cone the
™
isolation is effective, and for the isolation of definition B employed here, one expects
the curve to fall like some power of R as R 0. Finally, we also note that the
normalization factor 1rs Ž R 0 . is the reason why the histograms for the three different
R 0 in Fig. 4 do not exactly coincide even if R ) 0.7. A comparison of the histograms
relevant to R 0 s 0.4 and R 0 s 0.7 at R s 1 nicely demonstrates how weak the depen-
dence of the isolated-photon cross section on the cone size is.
We finally study the distribution of the photon in pseudorapidity hg . We impose here
the transverse momentum cut p Tg ) 10 GeV. Our results are summarized in Fig. 5; they
have been obtained in the case of definition B. The upper histograms correspond to
NLO, while the lower ones represent the Born level results. The solid histograms have
been obtained with default scales. They are pretty similar in shape; the Born is only
slightly broader, with a deeper dip at hg s 0. The size of radiative corrections is as
expected from what we previously found for the p Tg spectrum: the NLO result is about
a factor 1.6 higher than the Born one. The pairs of dotted histograms are obtained by
setting the renormalization scale equal to m 0r2 and 2 m 0 . It is clear that also in this case
the radiative corrections have the effect of reducing the scale dependence; by varying
m R , one obtains NLO cross sections that differ from the default one by 10% at most. The
effect of varying the factorization scale is much smaller than that associated with
Fig. 5. Pseudorapidity spectrum of isolated photons Ždefinition B.. The cross sections obtained by varying the
renormalization and factorization scales are also shown.
changes in m R . This can be understood by looking at Fig. 1: the effects at low p Tg

Ž m F s m 0r2 returns a cross section smaller than the default one. and at large p Tg
Ž m F s m 0r2 returns a cross section larger than the default one. tend to compensate
when integrating over p Tg , as done for Fig. 5. The scale dependence of the densities is
again responsible for this behaviour. When plotting the hg distribution for an integration
over, say, just the region 10 - p Tg - 12 GeV, one would see a much larger m F
dependence, of the order of that shown in the low-pTg region in Fig. 1. We also
considered the hg spectrum in the case when the photon is isolated according to
definition A. The scale dependence, as already in the case of the p Tg spectrum, is
identical to that presented in Fig. 5. The shape is almost the same as the one relevant to
definition B, for all scale choices.
3.2. Spin asymmetries and sensitiÕity to D g
In spin physics, experiments usually focus on spin asymmetries, since many system-
atic uncertainties cancel out in this ratio of polarized and unpolarized cross sections. In
what follows, we will therefore study the quantity
d DsrdpTg
Ap Tg s , Ž 9.
d srdpTg
as a function of p Tg . We will also consider a similar asymmetry, with p Tg replaced by
hg . More studies on asymmetries will be presented in Section 4. In Eq. Ž9., it is
understood that the same kinematical cuts are applied to both the numerator and the
denominator. The measurability of a spin asymmetry for a given process, as far as

statistics are concerned, is of course determined by the counting rate. The quantity
1 1
Ž Ap Tg . min s P 2 '2 s Le Ž 10 .
can be regarded as the minimal asymmetry that can be detected experimentally or,
equivalently, as the expected statistical error of the measurement, for a given integrated
luminosity relevant to parallel or antiparallel spins of the incoming particles, L , beam
polarizations P and a detection efficiency e ( 1; s is the unpolarized cross section
integrated over a certain range in p Tg Žthis range is denoted as a p Tg bin.. We assumed
protons are equal, L is L °' L . If this were not the case, the quantity 2 s L in Eq.
here that the luminosities relevant to parallel and antiparallel spins of the incoming
Ž10. would have to be substituted with 4s L i L ° rŽ L i qL L ° ..

At present, the largest source of uncertainty on the theoretical predictions for
asymmetries is clearly the choice of the polarized parton densities. We established in the
previous subsection that the perturbative expansion of single-inclusive cross sections is
under control. In the remainder of this section, we will therefore concentrate on the
dependence of the asymmetries upon the densities. There are many NLO-evolved sets
available, which mainly differ in the gluon sector, which is rather poorly constrained by
DIS data Žapart, perhaps, from the first moment.. In a previous study relevant to jet
physics w7x, we saw that most of the sets give almost identical results as far as the shape
Fig. 6. Asymmetry as a function of pTg , for various polarized densities, at different centre-of-mass energies.
The minimally observable asymmetry Ždot-dashed histogram. is also shown.
of the asymmetry is concerned, the main difference being in the absolute normalization.
For this reason, we will limit ourselves in this paper to three sets: GRSV STD, our
default set; GRSV MAXg w70x, which has a much larger gluon density and is thus
expected to return the highest cross sections; and set C of Ref. w72x ŽGS-C., which has a
rather small gluon density whose shape is dramatically different from that of all the
other sets, turning negative at high x for Q 2 not too large.
In Fig. 6 we present our results for the asymmetry as a function of p Tg . A cut
hg - 0.35 has been applied. In the left part of the figure we plot the asymmetries
obtained at 'S s 200 GeV, while in the right part we present the results for 'S s 500
GeV. The solid, dashed and dotted histograms correspond to the NLO predictions
obtained with GRSV STD, GRSV MAXg and GS-C respectively. The corresponding
symbols Žsee the labels on the figure. are the results obtained at the Born level. Finally,
the dot-dashed histogram is the minimally observable asymmetry, as defined in Eq. Ž10..
We have chosen L s 100 pby1 , P s 1 and e s 1. Of course, the latter two choices are
not realistic; however, in adopting this ‘ideal-world’ situation, we can estimate the
optimally achievable accuracy for a given integrated luminosity. Note that the assumed
value for L is conservative; one expects in the best case to eventually obtain L s 160
pby1 rpolarization at 'S s 200 GeV and L s 400 pby1rpolarization at 'S s 500
GeV. In any case, it is straightforward to rescale our predicted minimally observable
asymmetry 2 if one prefers other values for L , P or e . We also emphasize that we have
chosen rather small bins in p Tg , D p Tg s 2 GeV. It would certainly seem advantageous
in the actual data analysis to increase the bin size when going to larger p Tg , as is indeed
a commonly adopted procedure in the unpolarized prompt-photon experiments.
From Fig. 6, we see that the shapes of the asymmetries obtained using the GRSV
STD and GRSV MAXg sets are quite similar, but the difference in normalization is
sizeable; this is consistent with what we observed in our study of jet physics w7x. On the
other hand, the result obtained for GS-C looks completely different. At NLO, this
asymmetry becomes negative in the region 10 - p Tg - 30 GeV at 'S s 200 GeV, and in
the region 30 - p Tg - 45 GeV at 'S s 500 GeV. This is due in the first place to large
cancellations between the contributions of various partonic channels: while the GRSV
STD and GRSV MAXg results are dominated by the contribution of the qg-initiated
subprocess, this is not true in the case of GS-C, where the gluon is so small that
quark–quark scatterings Žof opposite sign. are in absolute value of the same order or
larger, in particular in the central pseudorapidity region, which is of interest here. For
the same reason, the asymmetries for set GS-C obtained at the Born level turn out to be
always substantially larger than those obtained at NLO. The issue of Žnon-.dominance of
the qg subprocess, which is obviously of key interest for the extraction of the polarized
gluon density from isolated-photon data, will be examined in more detail in the
following.
2
Our cuts used in Fig. 6, in particular hg - 0.35, actually correspond to those of the PHENIX experiment.
Note, however, that all our results have been integrated over the full 2p of azimuthal angle, whereas the
PHENIX electromagnetic calorimeter only covers half the azimuth. This implies that for a correct comparison
our minimally observable asymmetry has to be multiplied by a factor of '2 , in addition to introducing the
appropriate values for L , P, e . Our results for Ž Ap Tg . min are then found to be consistent with those reported
in w22,23x.
It is instructive to compare the asymmetries at the two different centre-of-mass

energies considered in Fig. 6. As is well known, at smaller centre-of-mass energies, the
asymmetries are generally larger; for example, when going from 'S s 200 GeV to
'S s 500 GeV the asymmetry obtained with GRSV STD decreases by a factor of about
3.6 Ž1.8. at p Tg s 10 GeV Ž p Tg s 50 GeV.. This feature is readily explained by the fact
that, for fixed p Tg , at larger 'S one probes smaller values of x in the parton
distributions. Since the unpolarized parton densities are steeper than the polarized ones
towards small x’s, one therefore gives more weight to the unpolarized cross section in
the denominator of the asymmetry. However, when comparing the predicted asymme-
tries with the minimum observable asymmetry, it is clear that, at a fixed value of p Tg ,
and except for the first few p Tg bins, the situation at 'S s 500 GeV is more favourable
than that at 'S s 200 GeV. On the other hand, as far as a measurement of D g at a given
x is concerned, one should rather look at the asymmetries at fixed x Tg s 2 p Tgr 'S , since
this corresponds to the value at which the parton densities are probed predominantly. As
can be seen from Fig. 6, the asymmetries corresponding to the GRSV sets approximately
scale with x Tg . Then, the quantity deciding about which energy is more favourable, is the
minimally observable asymmetry at a given x Tg . This quantity does not scale with x Tg ,
as can be inferred from the figure. For 'S s 500 GeV, one finds a value of Ž Ap Tg . min
larger than for the lower energy, making the higher-energy option appear less favourable.
However, two points should be kept in mind here: firstly, in both plots in Fig. 6 the
same value for the integrated luminosity has been used, whereas in reality one
anticipates a higher Žby a factor of 2 to 3. luminosity for 'S s 500 GeV. Secondly, the
lower cut-off for p Tg will certainly be the same for both energies, which means that at
'S s 500 GeV one can explore a region of x Tg that is inaccessible at 'S s 200 GeV.
Also, even if one considers the same x Tg value for the two energies, the parton densities
are still being probed at rather different scales, of the order of x Tg'S r2, the correspond-
ing p Tg values. Thus, it will be interesting to see whether measurements performed at
different centre-of-mass energies will yield information that is consistent, and compati-
ble with QCD evolution.
The same asymmetries as presented in Fig. 6 were also computed in a larger
pseudorapidity range, y1 - hg - 2, relevant to the STAR experiment. In the case of the
GRSV sets, only very minor differences were found, and the same conclusions as drawn
above apply Žwe have also to take into account the fact that the minimally observable
asymmetry decreases by a factor that can be as large as 2; therefore, in this larger
pseudorapidity range the situation is even more favourable.. In the case of GS-C, the
asymmetry increases significantly, becoming larger than the minimally observable
asymmetry for p Tg - 22 GeV at 'S s 500 GeV, and for p Tg - 14 GeV at 'S s 200
GeV. Also, in this region of p Tg the Born and NLO results are close to each other,
displaying a behaviour similar to that of the GRSV sets. The reason is again linked to
the dominance of the qg-initiated partonic subprocess: in fact, at large hg ’s, this
subprocess accounts for most of the full NLO cross section also in the case of GS-C, as
will be shown below. It thus follows that, if the true polarized gluon density is similar to
the one of the GS-C set, an extended pseudorapidity coverage is mandatory in order to
be able to see it experimentally. Finally, we mention that we computed the asymmetries
also by varying the isolation parameters R 0 and n. When setting R 0 s 0.7 andror
n s 2, we did not find any noticeable difference with the results presented above.
In Fig. 7 we present the asymmetry as a function of hg . A cut p Tg ) 10 GeV has

been applied. We again show the results obtained at two different centre-of-mass
energies, at both the Born and the NLO level. As in the case of the p Tg distribution, the
shapes of the asymmetries obtained with the two GRSV sets are almost identical, while
the one obtained with GS-C behaves quite differently. The NLO result for GS-C
becomes negative around hg s 0 at 'S s 200 GeV Žthe Born asymmetry remains
positive., consistently with what we observed before in the low-pTg region at this
energy. As remarked before, away from the central-hg region the GS-C asymmetry is
larger than the minimally observable one, and the Born and the NLO results become
similar. On the other hand, in this region the Born and NLO results for the GRSV sets
differ more than around hg s 0. This is simply related to the fact that here the polarized
cross sections fall more rapidly at the NLO level than at the Born level Žcf. Fig. 5.,
contrary to the case of the unpolarized cross section.
We now return to the issue of the dominance of the qg-initiated subprocess in
prompt-photon production at RHIC. In Fig. 8 we show the p Tg Žleft. and hg Žright.
dependences of the polarized and unpolarized cross sections, using the three sets of
polarized parton densities employed in the previous figures. As before, the isolation of
definition B and the default choice of scales have been adopted. We have used 'S s 200
GeV here Žwe checked that at 'S s 500 GeV we obtain results that lead us to the very
same conclusions.; cuts are as before. The symbols show in each case the cross section
Fig. 7. As in Fig. 6, but as a function of hg .

Fig. 8. pTg and hg spectra of isolated photons in polarized and unpolarized collisions. The results obtained by
retaining only the contribution of the qg-initiated partonic subprocess are also shown Žsymbols..
that is obtained by keeping only the qg-initiated subprocess. One can clearly see that in
the polarized case, for the two GRSV density sets, the qg subprocess alone produces a
result that is almost identical to the full answer. As we have checked, this comes about
to some extent because the other subprocesses all give small contributions, but also
because they tend to cancel one another to a good approximation. This explains why for
set GS-C a different pattern is found: slightly different quark densities and a vastly
different gluon distribution make the cancellation of the non-qg channels imperfect, and
the gluon density is not large enough to render the signal from qg scattering dominant,
except for large <hg <. However, in view of Fig. 6, this finding does not really create a
problem: if the gluon is indeed as small as embodied in the GS-C set, the measurement
at RHIC will anyway only give an asymmetry compatible with zero, and we will not be
in a position to actually unfold D g from the data. If, on the other hand, D g is sizeable,
it is an encouraging result that the polarized cross section provides a very direct measure
of it. Note that in the unpolarized case the qg channel is generally responsible for only
O Ž80%. of the cross section.
We finally mention that we have also computed the asymmetries by isolating the
photon according to definition A. In the case of the GRSV sets, the results are almost
identical to those shown here. In the case of GS-C, some difference can be seen in the
central pseudorapidity region, where the asymmetry tends to be smaller in the case of
definition A. Part of this effect results from the different scale choices adopted for the
two definitions, as discussed at the beginning of this section.
4. Isolated-photon-plus-jet observables
In the production process, the transverse momentum of the prompt photon is balanced
by that of the high-pT outgoing hadrons. It may be decided to neglect the properties of
these hadrons, and to study the inclusive production of the photon, as we did in Section
3. On the other hand, the study of the correlations between the photon and the associated
hadrons gives a more thorough information on the underlying dynamics. Also, from the
experimental point of view, photon-plus-hadron events can be used as a means of
calibrating the hadronic calorimeter. In this section, we will consider photon-plus-jet
observables. Our predictions are relevant to the STAR experiment w24x at RHIC, where
one of the main goals is indeed to determine D g from prompt-photon-plus-jet events.
As is customary in any fixed-order computation in perturbative QCD, our predictions are
given at the parton level Ži.e. our jet-finding algorithm deals with partons and not with
hadrons.. We will adopt here a k T-algorithm, namely that proposed in Ref. w73x, with
D s 1. We will only discuss the case when the photon is isolated following the
prescription B; unless otherwise specified, we will adopt the isolation parameters given
in Eq. Ž8.: R 0 s 0.4, eg s 1, n s 1. Since our computation is based on five-leg
amplitudes, we are able to predict the photon-plus-one-jet observables at NLO, and the
photon-plus-two-jet observables at LO.
4.1. General features and perturbatiÕe stability
The number of jets that accompany the photon does depend not only upon the
dynamics, but also upon the jet-finding algorithm and the kinematical cuts imposed on
Table 1
Total rates Žin pb; the entries relevant to unpolarized scattering have been multiplied by 10y3 . for
isolated-photon-plus-jet events, at 'S s 500 GeV
pT j 010 GeV pT j 012 GeV pT j 014 GeV
0yjet 1yjet 2yjet 0yjet 1yjet 2yjet 0yjet 1yjet 2yjet
GRSV STD 115.1 168.0 15.87 136.7 151.4 10.92 178.0 113.3 7.692
GRSV MAXg 206.7 294.7 29.28 244.6 265.8 20.30 317.2 199.1 14.41
GS-C 27.73 41.57 1.121 37.30 32.52 0.604 49.61 20.50 0.312
MRST=10y3 11.61 13.27 0.730 14.37 10.77 0.471 18.45 6.854 0.310
the jets. This is documented in Table 1, where the total rates are presented for events
satisfying
p Tg 0 10 GeV, y1 ( hg ( 2, p T j 0 10,12,14 GeV, y1 ( hj ( 2,
Ž 11 .
in the case of polarized and unpolarized pp collisions at 'S s 500 GeV Ž p T j and hj are
the jet transverse momentum and pseudorapidity, respectively.. The cuts on hg ,hj
considered here are relevant to the STAR detector. As one might expect, there is only a
small fraction of events where the photon is accompanied by two jets Žin the case of
two-jet events, the transverse-momentum cut is applied to both jets; it is clear that the
inclusion of radiative corrections for photon-plus-two-jet observables will not change
what was said before.. On the other hand 3 , a sizeable number of events falls in the class
denoted by ‘0-jets’, which is constituted by those events where the jetŽs. do not pass the
imposed transverse-momentum or pseudorapidity cuts; at the lowest transverse-momen-
tum cut, this is mainly due to the fact that the pseudorapidity cut is not symmetric
around h s 0. Since the number of jets is directly related to the hardness of the event,
large differences can be seen in the ratio of two-jet over one-jet rates, when different
™
parton densities are considered. In particular, this ratio is the larger the slower the gluon
density approaches zero for x 1; while GRSV STD and GRSV MAXg return almost
the same ratio Žthe shape of their gluon being basically the same., the result for GS-C is
much smaller, since the gluon density in this case has a dip at intermediate x values. It
follows that a first rough piece of information on the behaviour of the gluon density at
large x can be obtained by simply looking at the total photon-plus-jet rates. We also
notice that the result for the ratio of one-jet over two-jet rates in the case of unpolarized
scattering Žfourth row in Table 1. lies in between that of the GRSV sets and that of the
GS-C set, consistently with the fact that the shape of the MRST gluon density is softer
than that of GRSV STD and harder than that of GS-C. This also implies that the ratio of
rates corresponding to different numbers of jets is not very sensitive to the polarization
of the beams.
We must stress that, in the case of p T j 0 10 GeV and p Tg 0 10 GeV, the zero-jet and
one-jet rates are rather pathological in perturbative QCD Žon the other hand, their sum is
well-behaved.. Indeed, when equal transverse-momentum cuts are imposed on the
3
By definition, for a given row in Table 1, the sum ‘0-jet’q‘1-jet’q‘2-jet’ is the same for each of the
three pT j cuts, and corresponds to the inclusive isolated-photon rate.
photon and the hardest jet, large logarithms appear in the cross section, which in
principle should be resummed. Roughly speaking, at any fixed order in perturbation
theory, for ‘symmetric cuts’ the radiation of real gluons cannot compensate the large and
negative contribution of the virtual diagrams. The mechanism is identical to the one that
can be observed in two-jet correlations, in the case when the two jets have the same
minimum transverse momentum cut. This matter was discussed at length in Ref. w74x, to
which we refer the reader for more details. For illustration, we consider here the total
rate Žno h cuts have been applied; these would just change the absolute normalization
which is of no interest in what follows.:
sg j Ž D . s s Ž p Tg 0 10 GeV, p T j 0 10 GeV q D . Ž 12 .
as a function of D, for both polarized and unpolarized collisions, at different centre-of-
mass energies. By definition, the jet is the hardest of the jets of the event. The results are
displayed in Fig. 9. The plots in this figure are completely analogous to the ones in Fig.
4 of Ref. w74x. The main point is that a negative slope is here visible at D s 0, implying
that the cross section decreases here with the decreasing cut on p Tg , clearly signalling
a failure of the perturbative expansion. We remark, however, that, at variance with the
case of jet–jet correlations, in the case of isolated-photon-plus-jet production, a value of
D s 1 GeV already appears to be perfectly safe. We also remind the reader that, even in
the case of equal transverse-momentum cuts, the perturbative expansion is reliable
everywhere except in some corners of the phase space Žexamples will be given below..
Inspecting Fig. 9, we finally note that, when going from 'S s 200 GeV to 'S s 500
GeV, the cross section increases much more in the unpolarized than in the polarized
Fig. 9. Total photon-plus-jet rates, in polarized and unpolarized collisions, at different centre-of-mass energies.
case. This implies that, as in the case of inclusive observables discussed in the previous
section, at fixed final-state kinematical variables the asymmetries for photon-plus-jet
observables are larger at the smaller centre-of-mass energies.
We now turn to the issue of the perturbative stability of our results for correlations
between photon and jets. Since we are interested in NLO predictions, we will only
consider photon-plus-one-jet quantities; in the case when two jets are present in the
event, only the hardest jet is retained. As in subsection 3.1, we will assume that we
obtain a Žrelatively sound. indication of the stability of the cross sections if the
variations induced by changing the renormalization and factorization scales with respect
to their default values are small. In Fig. 10 we present the result for the invariant mass
distribution of the photon-jet system, in the case of polarized collisions at 'S s 500
GeV. The pseudorapidities of both the photon and the jet are required to be in the range
y1 ( h ( 2, and we impose p Tg 0 10 GeV and p T j 0 10 GeV. The results of both the
NLO computation Župper curves. and Born computation Žlower curves, which have been
rescaled in order to make them clearly distinguishable from the NLO ones. are
displayed. Similarly to the case of the inclusive transverse momentum distribution of
isolated photons Žcf. Fig. 1., in most of the range in Mg j the change of cross section
induced by a variation of the renormalization scale is of the order of 10% at NLO, and
larger at the Born level. However, a quite dramatic effect is seen at threshold, when
radiative corrections are included: the cross section in the first bin becomes negative,
and the scale dependence displays a pathological behaviour in this range, the cross
Fig. 10. Renormalization-scale dependence of the invariant mass spectrum of the photon–jet system, at the
Born and NLO levels. The result for unequal lower transverse-momentum cuts is also shown Žinset..
section becoming smaller for decreasing renormalization scales. This effect is exactly a
consequence of the fact that the minimum transverse-momentum cuts on the photon and
the jet are equal. Indeed, in the inset of Fig. 10 we show Žin the threshold region. the
invariant mass of the photon–jet system in the case when p T j 0 12 GeV. It is obvious
that here the scale dependence is as expected, and the cross section in the first bin
Žalthough not visible in the figure. remains positive. However, the first two bins show a
scale sensitivity comparable to the one of the Born result. In fact, close to the threshold
the NLO result is effectively a LO one, since the threshold at the Born level is in this
case at Mg j s 24 GeV. In the invariant mass range not close to threshold, the scale
dependence in the case of unequal transverse-momentum cuts is practically identical to
the one displayed in the main body of Fig. 10. We also studied the factorization scale
dependence of the invariant mass distribution. In the region not close to the threshold,
there is a clear improvement when going from LO to NLO; again, the results for equal
and unequal transverse-momentum cuts are very similar. At threshold, the same consid-
erations as given above apply. As in the case of the single-inclusive photon transverse-
momentum spectrum displayed in Fig. 1, the cross section for larger Žsmaller. factoriza-
tion scales is larger Žsmaller. than the default one at small invariant masses, while it is
smaller Žlarger. than the default for large invariant masses. This behaviour is almost
entirely due to scale dependence of the parton densities, as already discussed in the
single-inclusive case.
We performed a thorough study of the renormalization and factorization scale
dependence of many photon-jet observables. In particular, we considered the photon
Žjet. transverse momentum and pseudorapidity distributions, when cuts on the recoiling
jet Žphoton. are imposed, as suggested in Ref. w43x. Among the photon-jet correlations,
we considered the transverse momentum of the pair p TŽ g j., the azimuthal distance in the
transverse plane Dfg j , the distance in the h – f plane D Rg j , and the variables
p Tg ehg q p T j eh j p Tg eyhg q p T j eyh j
x1 s , x2 s , Ž 13 .
'S 'S
which, at the Born level, coincide with the Bjørken-x values of the incoming partons. In
all these cases, a reduction in the relative size of the scale dependence is seen at the
whenever the Born contribution is already present at the level of a 2 2 scattering ™

NLO with respect to the Born level, whenever such a comparison is meaningful Žthat is
process.: a change of the scales within the limits as above induces a variation of the
™
results of about 10% or less. In the regions of the phase space where the partonic
contributions start at the 2 3 level Žfor example, p TŽ g j. ) 0, Dfg j - p , D Rg j - p ., our
NLO results have a scale dependence larger than elsewhere, of the order of 15% to 20%,
since they are effectively LO. Finally, as in the case of the invariant-mass distribution,
there are corners of the phase space where the perturbative results are not reliable in the
case of equal transverse-momentum cuts. Among those, the case of D Rg j s p is
particularly interesting, since here the Born threshold falls inside the range available at
NLO. This case has been described, on general grounds, in Ref. w75x.
4.2. Spin asymmetries
We now turn to the study of asymmetries for photon-plus-jet cross sections. We

follow the procedure of Subsection 3.2, namely we study the dependence of the
asymmetries on the choice of polarized parton densities, at the Born and NLO levels.
Here, we restrict ourselves to 'S s 200 GeV. We verified that the pattern when going to
'S s 500 GeV is similar to the one already described in the preceding section; namely,
at fixed final-state kinematics we get smaller asymmetries, with however also smaller
minimally observable asymmetries. In Fig. 11 we show the asymmetries as functions of
the invariant mass Žleft. and x 1 Žright.. The photon-plus-jet events have been selected by
imposing equal transverse-momentum cuts on the photon and on the hardest jet
Ž p T ) 10 GeV; pseudorapidities are restricted to the range y1 - h - 2.. As discussed
previously, this choice only affects the threshold region of the invariant mass, where our
predictions should not be considered as reliable. As in the case of single-inclusive
quantities, the results for GRSV STD and GRSV MAXg are pretty similar in shape,
although sizeably different in normalization. On the other hand, GS-C has a clearly
distinguishable signature, showing a dip at intermediate values of the invariant mass and
in the region around x 1 s 0.1. It is very easy to trace the origin of this behaviour back to
the shape of the GS-C gluon. The NLO results are smaller than those at LO, as they
already were in the case of inclusive observables. The difference between Born and
NLO results is not big at small invariant masses and in the whole x 1 range, while it
grows larger in the tail of the invariant-mass distribution, since the K-factor of the
unpolarized cross section is larger in this region than that of the polarized cross section.
Figure 11 also presents the minimally observable asymmetry Ždot-dashed histograms.,
Fig. 11. Asymmetries at 'S s 200 GeV, as functions of invariant mass Žleft. and x 1 Žright.. The NLO
Žhistograms. and Born Žsymbols. results are both shown. The minimum observable asymmetry is displayed by
the dot-dashed histogram.
calculated according to Eq. Ž10.. A bin size of 2 GeV has been chosen for Mg j , and of
0.05 for log x 1; as before, L s 100 pby1 and e s P s 1. It is apparent that, if the
polarized densities are as predicted by the GS-C set, the measurement of the asymmetry
at RHIC will produce a result compatible with zero, even if quite large integrated
luminosities are attained. This result is consistent with what we already found in Section
3.2; however, in the case of single-inclusive variables the situation appeared to be
slightly better, when the enlarged pseudorapidity range y1 - h - 2 was considered.
In this context, we would like to comment on the findings of Ref. w43x, where it was
observed that, by looking at photon-plus-jet events, instead of considering only the
inclusive variables of the photon, one gets larger asymmetries. Also, photon-plus-jet
observables enhance the sensitivity to the shape of the parton densities and can be used
for a more straightforward deconvolution of D g from data w24x. Although we agree with
these observations, we doubt that photon-plus-jet correlations will give us a better
chance of measuring the gluon density than inclusive observables. In fact, there is in
practice the problem that the Žtheoretical. minimally observable asymmetry is larger in
the case of photon-plus-jet quantities than in the case of inclusive-photon quantities. The
situation is summarized in Fig. 12, for the case of the hg spectrum with p Tg 0 10 GeV.
In the case of the photon-plus-jet observable, the following cuts have been imposed on
the jet: p T j 0 10 GeV, hj ( 0.5. From the figure, it is apparent that, in spite of the fact
that the asymmetry is increased when cutting on the jet variables, the measurement
would be more difficult, since the minimally observable asymmetry is enhanced by a
Fig. 12. Asymmetries at 'S s 200 GeV, as functions of photon pseudorapidity, with Ždotted. and without
Žsolid. a transverse-momentum cut on the recoiling jet. The corresponding minimally observable asymmetries
Ždashed and dot-dashed, respectively. have been rescaled in order to coincide with the respective asymmetries
at hg s 0.
larger factor with respect to the asymmetry. Of course, this is a purely theoretical
estimate, which assumes that the experimental efficiency is the same in the case of
inclusive-photon or photon-plus-jet events Žhere taken to be equal to 1.. This is probably
unrealistic, but it seems unlikely that the photon-plus-jet efficiency will be higher than
that for photon-tagging only. Therefore, without any detailed study at the detector level,
it seems improbable that the photon-plus-jet asymmetries will be the preferred tool for
pinning down the polarized gluon density. We finally also have to add that in Ref. w43x
the photon transverse momentum was constrained in a bin around p T j s 10 GeV of
width 1 GeV, instead of having p Tg ) 10 GeV as in Fig. 12. In view of the discussion
relative to Fig. 9, the kinematical constraints imposed in Ref. w43x appear to be more
problematic from the perturbative point of view than those adopted in this paper for
producing Fig. 12.
5. Conclusions
We have performed a phenomenological study of prompt-photon production by

polarized colliding proton beams at RHIC. Our main motivation has been to put in
practice a recently proposed alternative way of isolating the photon from the hadronic
background, which eliminates the ill-understood fragmentation contribution from the
cross section, so that only the direct component is left. In this way, an optimally clean
photon signal results. This has also enabled us to perform the first fully consistent
next-to-leading order calculation for polarized prompt-photon production.
We have compared the cross sections and asymmetries for the new isolation with
those obtained for a traditional cone-type isolation as used hitherto in unpolarized
collider prompt-photon experiments. We find only small differences, implying that using
the new ‘clean’ isolation is not accompanied by a reduction in the number of events.
Performing a consistent next-to-leading order study, we were in a position to examine
the perturbative stability of the cross sections by studying their dependence on the
factorization and renormalization scales. We find that the scale dependence is signifi-
cantly reduced when going from the Born level to the next-to-leading order, and that the
scale dependence is moderate at next-to-leading order. This generally holds true for both
inclusive isolated-photon production and for isolated-photon-plus-jet production.
We have presented phenomenological results for the spin asymmetries and their
expected statistical errors in the RHIC experiments PHENIX and STAR. Consistently
with previous studies, the results are highly sensitive to the size of the spin-dependent
gluon density D g Ž x,Q 2 ., and it turns out that, unless D g is very small, it should clearly
be possible to rather accurately determine it in the region 0.04 - x - 0.25, at scales of
the order of the transverse momentum of the photon, 10 GeV - p Tg - 30 GeV. In this
context, we find that as far as statistics is concerned, inclusive-photon measurements
seem to be somewhat more favourable than photon-plus-jet ones.
Acknowledgements
We are happy to thank G. Ridolfi for his collaboration at an early stage of this work.
We are also grateful to N. Saito for useful information. The work of S.F. is supported in
part by the EU Fourth Framework Programme ‘Training and Mobility of Researchers’,

Network ‘Quantum Chromodynamics and the Deep Structure of Elementary Particles’,
contract FMRX-CT98-0194 ŽDG 12 - MIHT..
References
w1x I. Bojak, M. Stratmann, Nucl. Phys. B 540 Ž1999. 345.

w2x I. Bojak, M. Stratmann, Phys. Lett. B 433 Ž1998. 411.
w3x G. Baum et al., COMPASS Collaboration, CERNrSPSLC 96-14.
w4x S. Frixione, G. Ridolfi, Phys. Lett. B 383 Ž1996. 227.
w5x M. Stratmann, W. Vogelsang, Z. Phys. C 74 Ž1997. 641.
w6x M. Stratmann, W. Vogelsang, Proc. of the 1995r96 workshop on Future Physics at HERA, Hamburg,
Germany, ed. G. Ingelman, A. De Roeck, R. Klanner, p. 815.
w7x D. de Florian, S. Frixione, A. Signer, W. Vogelsang, Nucl. Phys. B 539 Ž1999. 455.
w8x D. de Florian, S. Frixione, Phys. Lett. B 457 Ž1999. 236.
w9x RHIC Spin Collaboration, D. Hill et al., letter of intent RHIC-SPIN-LOI-1991, updated 1993.
w10x G. Bunce et al., Particle World 3 Ž1992. 1.
w11x For a compilation of the prompt-photon data in the unpolarized case, see: W. Vogelsang, M. Whalley, J.
Phys. G 23 Ž1997. A1.
w12x ALEPH Collaboration, D. Buskulic et al., Z. Phys.C 69 Ž1996. 365.
w13x OPAL Collaboration, K. Ackerstaff et al., Eur. Phys. J. C 2 Ž1998. 39.
w14x P. Aurenche, P. Chiappetta, M. Fontannaz, J.Ph. Guillet, E. Pilon, Nucl. Phys. B 399 Ž1993. 34.
w15x ¨ L.E. Gordon, E. Reya, W. Vogelsang, Phys. Rev. Lett. 73 Ž1994. 388.
M. Gluck,
w16x W. Vogelsang, A. Vogt, Nucl. Phys. B 453 Ž1995. 334.
w17x P. Aurenche, M. Fontannaz, J.Ph. Guillet, B. Kniehl, E. Pilon, M. Werlen, Eur. Phys. J. C 9 Ž1999. 107.
w18x ¨ E. Reya, A. Vogt, Phys. Rev. D 48 Ž1993. 116.
M. Gluck,
w19x ¨ E. Reya, A. Vogt, Phys. Rev. D 51 Ž1995. 1427, ŽE..
M. Gluck,
w20x L. Bourhis, M. Fontannaz, J.Ph. Guillet, Eur. Phys. J. C 2 Ž1998. 529.
w21x E.L. Berger, J. Qiu, Phys. Rev. D 44 Ž1991. 2002.
w22x Y. Goto, talk given at the 7th Int. Workshop on Deep Inelastic Scattering and QCD ŽDIS 99., Zeuthen,
Germany, April 1999.
w23x Y. Goto, N. Hayashi, N. Saito, Background Study for Prompt Photon Production at PHENIX Žinternal
RIKEN report..
w24x L.C. Bland, hep-exr9907058, to be published in the Proc. of Workshop on Physics with Electron
Polarized Ion Collider – EPIC’99, Bloomington, Indiana, April 1999.
w25x ¨
S. Gullenstern, ¨ Phys. Rev. D 51 Ž1995. 3305.
P. Gornicki, L. Mankiewicz, A. Schafer,
w26x G.P. Skoro, M. Zupan, M.V. Tokarev, hep-phr9905511.
w27x N. Mobed, C. Papavasiliou, M. Svec, Phys. Rev. D 26 Ž1982. 3284.
w28x E.L. Berger, J. Qiu, Phys. Rev. D 40 Ž1989. 778.
w29x S. Gupta, D. Indumathi, M. Murthy, Z. Phys. C 42 Ž1989. 493.
w30x S. Gupta, D. Indumathi, M. Murthy, Z. Phys. C 44 Ž1989. 356, ŽE..
w31x H.-Y. Cheng, S.-N. Lai, Phys. Rev. D 41 Ž1990. 91.
w32x C. Bourrely, J.Ph. Guillet, J. Soffer, Nucl. Phys. B 361 Ž1991. 72.
w33x P. Mathews, R. Ramachandran, Z. Phys. C 53 Ž1992. 305.
w34x P. Chiappetta, P. Colangelo, J.Ph. Guillet, G. Nardulli, Z. Phys. C 59 Ž1993. 629.
w35x A.P. Contogouris, B. Kamal, Z. Merebashvili, F.V. Tkachov, Phys. Lett. B 304 Ž1993. 329.
w36x A.P. Contogouris, B. Kamal, Z. Merebashvili, F.V. Tkachov, Phys. Rev. D 48 Ž1993. 4092.
w37x L.E. Gordon, W. Vogelsang, Phys. Rev. D 48 Ž1993. 3136.
w38x A.P. Contogouris, Z. Merebashvili, Phys. Rev. D 55 Ž1997. 2718.
w41x L.E. Gordon, Nucl. Phys. B 501 Ž1997. 197.
w42x L.E. Gordon, G.P. Ramsey, Phys. Rev. D 59 Ž1999. 074018.
w43x S. Chang, C. Coriano, L.E. Gordon, Phys. Rev. D 58 Ž1998. 074002.

w44x S. Frixione, Phys. Lett. B 429 Ž1998. 369.
w45x CDF Collaboration, F. Abe et al., Phys. Rev. Lett. 73 Ž1994. 2662.
w46x CDF Collaboration, F. Abe et al., Phys. Rev. Lett. 74 Ž1995. 1891, ŽE..
w47x S. Kuhlmann, CDF Collaboration, FERMILAB-CONF-99-165-E.
w48x E706 Collaboration, L. Apanasevich et al., Phys. Rev. Lett. 81 Ž1998. 2642.
w49x UA6 Collaboration, G. Ballocchi et al., Phys. Lett. B 436 Ž1998. 222.
w50x J. Huston et al., Phys. Rev. D 51 Ž1995. 6139.
w51x H. Baer, M. Hall Reno, Phys. Rev. D 54 Ž1996. 2017.
w52x L. Apanasevich et al., Phys. Rev. D 59 Ž1999. 074007.
w53x A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 4 Ž1998. 463.
w54x H.-L. Lai, H. Li, Phys. Rev. D 58 Ž1998. 114020.
w55x E. Laenen, G. Oderda, G. Sterman, Phys. Lett. B 438 Ž1998. 173.
w56x S. Catani, M.L. Mangano, P. Nason, J. High Energy Phys. 9807 Ž1998. 024.
w57x S. Catani, M.L. Mangano, P. Nason, C. Oleari, W. Vogelsang, J. High Energy Phys. 9903 Ž1999. 025.
w58x H. Baer, J. Ohnemus, J.F. Owens, Phys. Rev. D 42 Ž1990. 61.
w59x P. Aurenche, R. Baier, M. Fontannaz, Phys. Rev. D 42 Ž1990. 1440.
w60x E.W.N. Glover, W.J. Stirling, Phys. Lett. B 295 Ž1992. 128.
w61x Z. Kunszt, Z. Trocsanyi, Nucl. Phys. B 394 Ž1993. 139.
w62x S. Catani, M. Fontannaz, E. Pilon, Phys. Rev. D 58 Ž1998. 094025.
w63x E.W.N. Glover, A.G. Morgan, Z. Phys. C 62 Ž1994. 311.
w64x A. Gehrmann-De Ridder, E.W.N. Glover, Nucl. Phys. B 517 Ž1998. 269.
w65x F. Aversa, P. Chiappetta, M. Greco, J.Ph. Guillet, Phys. Lett. B 211 Ž1988. 465.
w66x F. Aversa, P. Chiappetta, M. Greco, J.Ph. Guillet, Nucl. Phys. B 327 Ž1989. 105.
w67x S. Frixione, Z. Kunszt, A. Signer, Nucl. Phys. B 467 Ž1996. 399.
w68x S. Frixione, Nucl. Phys. B 507 Ž1997. 295.
w69x S. Frixione, hep-phr9809397, Proc. of the 29th Int. Conf. on High-Energy Physics ŽICHEP 98.,
Vancouver, Canada, July 1998.
w70x ¨ E. Reya, M. Stratmann, W. Vogelsang, Phys. Rev. D 53 Ž1996. 4775.
M. Gluck,
w71x Particle Data Group, C. Caso et al., Eur. Phys. J. C 3 Ž1998. 1.
w72x T. Gehrmann, W.J. Stirling, Phys. Rev. D 53 Ž1996. 6100.
w73x S.D. Ellis, D. Soper, Phys. Rev. D 48 Ž1993. 3160.
w74x S. Frixione, G. Ridolfi, Nucl. Phys. B 507 Ž1997. 315.
w75x S. Catani, B. R Webber, J. High Energy Phys. 9710 Ž1997. 005.
Self-gravitating fundamental strings and black holes

Thibault Damour a , Gabriele Veneziano b
a
Institut des Hautes Etudes Scientifiques, 91440 Bures-sur-YÕette, France
b
Theory DiÕision, CERN, CH-1211 GeneÕa 23, Switzerland
Received 8 July 1999; accepted 15 September 1999
Abstract
The configuration of typical highly excited Ž M 4 M s ; Ž a X .y1r2 . string states is considered

as the string coupling g is adiabatically increased. The size distribution of very massive single
string states is studied and the mass shift, due to long-range gravitational, dilatonic and axionic
attraction, is estimated. By combining the two effects, in any number of spatial dimensions d, the
most probable size of a string state becomes of order l s s '2 a X when g 2 MrM s ; 1. Depending
on the dimension d, the transition between a random-walk-size string state Žfor low g . and a
compact Ž; l s . string state Žwhen g 2 MrMs ; 1. can be very gradual Ž d s 3., fast but continuous
Ž d s 4., or discontinuous Ž d 0 5.. Those compact string states look like nuggets of an ultradense
state of string matter, with energy density r ; gy2 M sdq1. Our results extend and clarify previous
work by Susskind, and by Horowitz and Polchinski, on the correspondence between self-gravitat-
ing string states and black holes. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w; 04.70.Dy
1. Introduction
Almost exactly thirty years ago the study of the spectrum of string theory Žknown at
the time as the dual resonance model. revealed w1,2x a huge degeneracy of states
growing as an exponential of the mass. This led to the concept of a limiting ŽHagedorn.
temperature THag in string theory. Only slightly more recently Bekenstein w3x proposed
that the entropy of a black hole should be proportional to the area of its horizon in
Planck units, and Hawking w4x fixed the constant of proportionality after discovering that
black holes, after all, do emit thermal radiation at a temperature THaw ; Ry1BH .
When string and black hole entropies are compared one immediately notices a
striking difference: string entropy 1 is proportional to the first power of mass in any
1
The self-interaction of a string lifts the huge degeneracy of free string states. One then defines the entropy
of a narrow band of string states, defined with some energy resolution M s Q D E < M, as the logarithm of the
number of states within the band D E.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 5 9 6 - 9
94 T. Damour, G. Venezianor Nuclear Physics B 568 (2000) 93–119
number of spatial dimensions d, while black hole entropy is proportional to a d-depen-

dent power of the mass, always larger than 1. In formulae
Ž dy1 .r Ž dy2 .
aXM Area dy1
R BH Ž g 2 MrMs .
Ss ; ; MrMs , S BH ; ; ; ,
ls GN GN g2
Ž 1.1 .
where, as usual, a X is the inverse of the classical string tension, l s ; 'a X " is the
quantum length associated with it 2 , M s ; "ra X is the corresponding string mass scale,
'
R BH is the Schwarzschild radius associated with M,
1r Ž dy2 .
R BH ; Ž GN M . , Ž 1.2 .
and we have used that, at least at sufficiently small coupling, the Newton constant and
a X are related via the string coupling by GN ; g 2 Ž a X .Ž dy1.r2 Žmore geometrically,
l Pdy 1 ; g 2 l sdy1 ..
Given their different mass dependence, it is obvious that, for a given set of the
fundamental constants GN , a X , g 2 , S s ) S BH at sufficiently small M, while the opposite
is true at sufficiently large M. Obviously, there has to be a critical value of M, Mc , at
which S s s SBH . This observation led Bowick et al. w5x to conjecture that large black
holes end up their Hawking-evaporation process when M s Mc , and then transform into
a higher-entropy string state without ever reaching the singular zero-mass limit. This
reasoning is confirmed w6x by the observation that, in string theory, the fundamental
string length l s should set a minimal value for the Schwarzschild radius of any black
hole Žand thus a maximal value for its Hawking temperature.. It was also noticed w5,7,8x
that, precisely at M s Mc , R BH s l s and the Hawking temperature equals the Hagedorn
temperature of string theory. For any d, the value of Mc is given by
Mc ; M s gy2 . Ž 1.3 .
Susskind and collaborators w7,9,10x went a step further and proposed that the
spectrum of black holes and the spectrum of single string states be ‘‘identical’’, in the
sense that there be a one to one correspondence between Žuncharged. fundamental string
states and Žuncharged. black hole states. Such a ‘‘correspondence principle’’ has been
generalized by Horowitz and Polchinski w11x to a wide range of charged black hole
states Žin any dimension.. Instead of keeping fixed the fundamental constants and letting
M evolve by evaporation, as considered above, one can Žequivalently. describe the
both sides of and through, the string |

physics of this conjectured correspondence by following a narrow band of states, on
black hole transition, by keeping fixed the
entropy 3 S s S s s S BH , while adiabatically 4 varying the string coupling g, i.e. the ratio
between l P and l s . The correspondence principle then means that if one increases g
2 X
Below, we shall use the precise definition l s ''2 a " , but, in this section, we neglect factors of order
unity.
3
One uses here the fact that, during an adiabatic variation of g, the entropy of the black hole
dy 1
SBH ; ŽArea.r GN ; R BH r GN stays constant. This result Žknown to hold in the Einstein conformal frame.
applies also in string units because S BH is dimensionless.
4
The variation of g can be seen, depending on one’s taste, either as a real, adiabatic change of g due to a
varying dilaton background, or as a mathematical way of following energy states.
T. Damour, G. Venezianor Nuclear Physics B 568 (2000) 93–119 95
each Žquantum. string state should turn into a Žquantum. black hole state at sufficiently
strong coupling, while, conversely, if g is decreased, each black hole state should
‘‘decollapse’’ and transform into a string state at sufficiently weak coupling. For all the
reasons mentioned above, it is very natural to expect that, when starting from a black
hole state, the critical value of g at which a black hole should turn into a string is given,
in clear relation to Ž1.3., by
g c2 M ; M s , Ž 1.4 .
and is related to the common value of string and black-hole entropy via
1 1
g c2 ; s . Ž 1.5 .
SBH Ss
Note that g c2 < 1 for the very massive states Ž M 4 M s . that we consider. This justifies
our use of the perturbative relation between GN and a X .
In the case of extremal BPS, and nearly extremal, black holes the conjectured
correspondence was dramatically confirmed through the work of Strominger and Vafa
w12x and others w13–15x leading to a statistical mechanics interpretation of black-hole
entropy in terms of the number of microscopic states sharing the same macroscopic
quantum numbers. However, little is known about whether and how the correspondence
works for non-extremal, non-BPS black holes, such as the simplest Schwarzschild black
hole 5. By contrast to BPS states whose mass is protected by supersymmetry, we shall
consider here the effect of varying g on the mass and size of non-BPS string states.
Although it is remarkable that black-hole and string entropy coincide when R BH s l s ,
this is still not quite sufficient to claim that, when starting from a string state, a string
becomes a black hole at g s g c . In fact, the process in which one starts from a string
state in flat space and increases g poses a serious puzzle w7x. Indeed, the radius of a
typical excited string state of mass M is generally thought of being of order
1r2
R rw
s ; l s Ž MrM s . , Ž 1.6 .
X
as if a highly excited string state were a random walk made of MrMs s a Mrl s
segments of length l s w16–19x. ŽThe number of steps in this random walk is, as is
natural, the string entropy Ž1.1... The ‘‘random walk’’ radius Ž1.6. is much larger than
the Schwarzschild radius for all couplings g ( g c , or, equivalently, the ratio of
self-gravitational binding energy to mass Žin d spatial dimensions.
dy2 1
GN M R BH Ž M . M Ž4yd .
2
Ž R rw
s .
dy2
;
ž R rw
s
/ ;g 2
ž /
Ms
Ž 1.7 .
remains much smaller than one Žwhen d ) 2, to which we restrict ourselves. up to the
transition point. In view of Ž1.7. it does not seem natural to expect that a string state will
‘‘collapse’’ to a black hole when g reaches the value Ž1.4.. One would expect a string
state of mass M to turn into a black hole only when its typical size is of order of
5
For simplicity, we shall consider in this work only Schwarzschild black holes, in any number d ' Dy1
of non-compact spatial dimensions.
R BH Ž M . Žwhich is of order l s at the expected transition point Ž1.4... According to Eq.

Ž1.7., this seems to happen for a value of g much larger than g c .
Horowitz and Polchinski w20x have addressed this puzzle by means of a ‘‘thermal
scalar’’ formalism w21–23x. Their results suggest a resolution of the puzzle when d s 3
Žfour-dimensional space-time., but lead to a rather complicated behaviour when d 0 4.
More specifically, they consider the effective field theory of a complex scalar field x in
d Žspace-time. dimensions Žwith period b in Euclidean time t ., with mass squared
m2 Ž b . s Ž4p 2 a X 2 .y1 w Gtt b 2 y b H2 x, where Gmn is the string metric and by1
H the
Hagedorn temperature. They took into account the effect of gravitational Žand dilatonic.
self-interactions in a mean field approximation. This leads to an approximate Hartree-like
equation for x Ž x ., which admits a stable bound state, in some range g 0 - g - g c , when
d s 3. They interpret the size of the bound state ‘‘wave function’’ x Ž x . as the ‘‘size of
the string’’, and find that Žin d s 3. this size is of order
1
lx ; 2 . Ž 1.8 .
g M
They describe their result by saying that ‘‘the string contracts from its initial Žlarge.
size’’, when g ; g 0 ; Ž MrMs .y3 r4 , down to the string scale when g ; g c ;
Ž MrMs .y1 r2 . This interpretation of the length scale lx , characterizing the thermal
scalar bound state, as ‘‘the size of the string’’ is unclear to us, because of the formal
nature of the auxiliary field x which has no direct physical meaning in Minkowski
space-time. Moreover, the analysis of Ref. w20x in higher dimensions is somewhat
g 0 ; Ž MrM s .Ž dy6.r4 ) g c for the direct process Žstring

|™
inconclusive, and suggests that a phenomenon of hysteresis takes place Žwhen d 0 5.:
the critical value of g corresponding to the string black hole transition would be
black hole., and g c for the
reverse one. Finally, they suggest that, in the reverse process, a black hole becomes an
excited string in an atypical state.
The aim of the present work is to clarify the string |
black hole transition by a
direct study, in real space-time, of the size and mass of a typical excited string, within
the microcanonical ensemble of self-graÕitating strings. Our results lead to a rather
simple picture of the transition, in any dimension. We find no hysteresis phenomenon in
entropy S, for both directions of the string |

higher dimensions. The critical value for the transition is Ž1.4., or Ž1.5. in terms of the
black hole transition. In three spatial
dimensions, we find that the size Žcomputed in real space-time. of a typical self-graÕitat-
ing string is given by the random walk value Ž1.6. when g 2 ( g 02 , with g 02 ;
Ž MrMs .y3 r2 ; Sy3r2 , and by
1
R typ ; 2 , Ž 1.9 .
g M
when g 02 ( g 2 ( g c2 . Note that R typ smoothly interpolates between R rw
s and l s . This
result confirms the picture proposed by Ref. w20x when d s 3, but with the bonus that
Eq. Ž1.9. Žwhich agrees with Eq. Ž1.8.. refers to a radius which is estimated directly in
physical space Žsee below., and which is the size of a typical member of the
microcanonical ensemble of self-gravitating strings. In all higher dimensions6 , we find
6
With the proviso that the consistency of our analysis is open to doubt when d08.
that the size of a typical self-gravitating string remains fixed at the random walk value
Ž1.6. when g ( g c . However, when g gets close to a value of order g c , the ensemble of
self-gravitating strings becomes Žsmoothly in d s 4, but suddenly in d 0 5. dominated
by very compact strings of size ; l s Žwhich are then expected to collapse with a slight
further increase of g because the dominant size is only slightly larger than the
Schwarzschild radius at g c ..
Our results confirm and clarify the main idea of a correspondence between string
states and black hole states w7,9–11,20x, and suggest that the transition between these
states is rather smooth, with no apparent hysteresis, and with continuity in entropy, mass
and typical size. It is, however, beyond the technical grasp of our analysis to compute
any precise number at the transition Žsuch as the famous factor 1r4 in the Bekenstein–
Hawking entropy formula..
2. Size distribution of free string states
The aim of this section is to estimate the distribution function in size of the ensemble
of free string states of mass M, i.e. to count how many massive string states have a
given size R. This estimate will be done while neglecting the gravitational self-interac-
tion. The effect of the latter will be taken into account in a later section.
Let us first estimate the distribution in size by a rough, heuristic argument based on
the random walk model w16–19x of a generic excited string state. In string units
Ž l s ; My1s ; 1 , the geometrical shape in d-dimensional space of a generic massive
. Ž .
string state can be roughly identified with a random walk of M steps of unit length. We
can constrain this random walk to stay of size Q R by considering a diffusion process,
starting from a point source at the origin, in presence of an absorbing sphere SR of
radius R, centered on the origin. In the continuous approximation, the kernel K t Ž x,0.
giving the conditional probability density of ending, at time t, at position x, after having
started Žat time 0. at the origin, without having ever gone farther from the origin than the
distance R, satisfies: Ži. the diffusion equation E t K t s D K t , Žii. the initial condition
K 0 Ž x,0. s d Ž x ., and Žiii. the ‘‘absorbing’’ boundary condition K t s 0 on the sphere SR .
The kernel K t can be decomposed in eigenmodes,
K t Ž x ,0 . s Ý cn Ž x . cn Ž 0 . eyE n t , Ž 2.1 .
n
where cnŽ x . is a normalized Žreal. L2 basis Ž Hd d x cnŽ x . cmŽ x . s dn m ; Ý n cnŽ x . cnŽ y .

s d Ž x y y .. satisfying
D cn s yEn c , Ž 2.2 .
in the interior, and vanishing on SR . The total conditional probability of having stayed
within SR after the time t is the integral HB R d d x K t Ž x,0. within the ball BR : < x < ( R. For
large values of t, K t is dominated by the lowest eigenvalue E0 , and the conditional
probability goes like c 0 eyE 0 t , where c 0 is a numerical constant of order unity. The
eigenvalue problem Ž2.2. is easily solved Žin any dimension d ., and the s-wave ground
state can be expressed in terms of a Bessel function: c 0 Ž r . s N Jn Ž k 0 r .rŽ k 0 r . n with
(
n s Ž d y 2.r2. Here, k 0 s E0 is given by the first zero jn of Jn Ž z .:k 0 s jnrR. The
important information for us is that the ground state energy E0 scales with R like
E0 s c1rR 2 , where c1 s O Ž1. is a numerical constant. This scaling is evident for the
Dirichlet problem Ž2.2., whatever be the shape of the boundary. Finally, remembering
that the number of time steps is given by the mass, t s M, we expect the looked for
conditional probability, i.e. the fraction of all string states at mass level M which are of
size Q R, to be asymptotically of order
2
f Ž R . ; eyc 1 M r R . Ž 2.3 .
This estimate is expected to be valid when MrR 2 4 1, i.e. for string states which are
much smaller Žin size. than a typical random walk R 2rw ; M Žbut still larger than the
string length, R R 1.. In the opposite limit, R 2 4 M, the kernel K t Ž x,0. can be
approximated by the free-space value K tŽ0. Ž x,0. s Ž4 p t .yd r2 expŽyx 2rŽ4 t .., with
t s M, so that the fraction of string states of size R R 4 R rw will be of order
2
; eyc 2 R r M , with c 2 s O Ž1..
As the result Ž2.3. will be central to the considerations of this paper, we shall now go
beyond the previous heuristic, random walk argument and derive the fraction of small
string states by a direct counting of quantum string states. For simplicity, we shall deal
with open bosonic strings Ž l s ' '2 a X , 0 ( s ( p .
X m Ž t , s . s Xcm
m
Ž t , s . q X˜ m Ž t , s . , Ž 2.4 .
m
Xcm Žt ,s . s x m q 2 aX p m t , Ž 2.5 .
a nm
X̃ m Ž t , s . s i l s Ý eyi nt cos n s . Ž 2.6 .
n/0 n
m Ž
Here, we have explicitly separated the center of mass motion X cm with w x m, p n x s i h mn .
from the oscillatory one X˜ Žw a m , a n x s m dmqn h .. The free spectrum is given by
m m n 0 mn
a X M 2 s N y 1, where Ž a P b ' hmn a m b n ' ya 0 b 0 q a i b i .

` `
Ns Ý ayn P a n s Ý n Nn . Ž 2.7 .
ns1 ns1
Here Nn ' a†n P a n is the occupation number of the nth oscillator Ž a nm s 'n a nm ,w a nm ,anm† x
s h mn dn m , with n,m positive..
The decomposition Ž2.4. – Ž2.6. holds in any conformal gauge ŽŽ Et X m " Es X m . 2 s 0..
One can further specify the choice of world-sheet coordinates by imposing
nm X m Ž t , s . s 2 a X Ž nm p m . t , Ž 2.8 .
where n m is an arbitrary timelike or null vector Ž n P n ( 0. w24x. Eq. Ž2.8. means that the
n-projected oscillators nm a mm are set equal to zero. The usual ‘‘light-cone’’ gauge is
obtained by choosing a fixed, null vector nm . The light-cone gauge introduces a
preferred Ž‘‘longitudinal’’. direction in space, which is an inconvenience for defining the
Žrms. size of a massive string state. As we shall be interested in quasi-classical, very
massive string states Ž N 4 1. it should be possible to work in the ‘‘center of mass’’
gauge, where the vector n m used in Eq. Ž2.8. to define the t-slices of the world-sheet is
taken to be the total momentum p m of the string. This gauge is the most intrinsic way to
describe a string in the classical limit. Using this intrinsic gauge, one can covariantly 7
define the proper rms size of a massive string state as
1 2
R 2 ' ² Ž X˜H m
Ž t , s . . :s ,t , Ž 2.9 .
d
where X˜H m
' X˜ m y p m Ž p P X˜ .rŽ p P p . denotes the projection of X˜ m ' X m y X cm
m Ž .
t
m
orthogonally to p , and where the angular brackets denote the Žsimple. average with
respect to s and t . The factor 1rd in Eq. Ž2.9. is introduced to simplify later formulas.
So defined R is the rms value of the projected size of the string along an arbitrary, but
fixed spatial direction. ŽWe shall find that this projected size is always larger than
' 3 a Xr2 ; i.e. string states cannot be ‘‘squeezed’’, along any axis, more than this..
In the center of mass gauge, pm X˜ m vanishes by definition, and Eq. Ž2.9. yields
simply
1
R 2 s l s2 R , Ž 2.10 .
d
with
` ayn P a n q a n P ayn
R' Ý . Ž 2.11 .
ns1 2 n2
The squared-size operator R, Eq. Ž2.11., contains the logarithmically infinite contribu-
tion Ý 1rŽ2 n.. Without arguing with the suggestion that this contribution may have a
physical meaning Žsee, e.g., Ref. w25x., we note here that this contribution is state-inde-
pendent. We are interested in this work in the relative sizes of various highly excited,
quasi-classical states. A concept which should reduce to the well-defined, finite rms size
of a classical Nambu string in the classical limit. We shall therefore discard this
state-independent contribution, i.e. work with the normal-ordered operator
` a†n P a n ` Nn
: R :s Ý s Ý . Ž 2.12 .
ns1 n ns1 n
™
We shall assume that we can work both in the center-of-mass Žworld-sheet. gauge
Ž pm a mm 0. and in the center-of-mass ŽLorentz. frame ŽŽ p m . s Ž M,0... This means
that the scalar product in the level occupation number Nn runs over the d spatial
dimensions: Nn s a†n P a n s Ý dis1 Ž a ni . † a ni . The ‘‘wrong sign’’ time oscillators a n0 are set
equal to zero. The Virasoro constraints then imply, besides the mass formula a X M 2 s N
y 1, the usual sequence of constraints on physical states, L n < f : s 0, with L n
s 12 Ý m Ý dis1 a nym
i
a mi . These constraints mean that the d oscillators a ni at level n are
not physically independent.
The problem we would like to solve is to count the number of physical states, in the
Fock space of the center-of-mass oscillators a ni , having some fixed values of N and R
Žwe henceforth work only with the normal-ordered operator Ž2.12. without adorning it
with the : : notation.. The Virasoro constraints make this problem technically quite
7
In an arbitrary conformal gauge, the definition Ž2.9. is gauge-dependent Žin spite of the use of the
mŽ .
orthogonal projection. because both the definition of Xcm t , and that of the Ž s ,t .-averaging depend on the
choice of world-sheet gauge. Even if we were using the Žmore intrinsic but more complicated. average with
weight ydet ga b d s dt s Ž Es X m . 2 d s dt , the dependence upon X cm
' m Ž .
t would remain.
difficult. However, we know from the exact counting of physical states Žwithout size
restriction. in the light-cone gauge that the essential physical effect of the Virasoro
constraints is simply to reduce the number of independent oscillators at any level n from
d s D y 1 Žin the center-of-mass gauge. to d y 1 s D y 2. If we Žformally. consider d
as a large parameter 8 , this change in the number of effective free oscillators should have
only a small fractional effect on any other coarse-grained, counting problem. We shall
assume that this is the case, and solve the much simpler counting problem where the d
oscillators a ni are considered as independent 9. To solve this problem we pass from a
microcanonical problem Žfixed values of N and R . to a grand canonical one Žfixed
values of some thermodynamical conjugates of N and R .. Let us introduce the formal
‘‘partition function’’
Z d Ž b ,g . ' Ý exp Ž yb N Nni y g R Nni ., Ž 2.13 .
Nni 4
where the sum runs over all sequences Žlabeled by n 0 1 and i s 1, . . . ,d . of indepen-
dent occupation numbers Nni s Ž a in . † a ni s 0,1,2, . . . , and where N w Nni x and R w Nni x are
defined by Eqs. Ž2.7., and Ž2.12., with Nn ' Ý dis1 Nni. Note that Ž2.13. is not the usual
thermodynamical partition function, and that b is not the usual inverse temperature.
Indeed, b is a formal conjugate to N , a X M 2 and not to the energy M. In particular,
because the degeneracy grows exponentially with M Žand not M 2 . its Laplace trans-
form Ž2.13. is defined for arbitrary values of b . We associate with the definition Ž2.13.
that of a formal grand canonical ensemble of configurations, with the probability
p Nni 4 s Zy1 i
d Ž b ,g . exp Ž yb N Nn y g R Nn
i
. Ž 2.14 .
of realization of the particular sequence Nni of occupation numbers. The mean values of
N w Nni x and R w Nni x in this ensemble are
E cd Ž b ,g . E cd Ž b ,g .
Nsy , Rsy , Ž 2.15 .
Eb Eg
where we denote
cd Ž b ,g . ' ln Z d Ž b ,g . . Ž 2.16 .
The second derivatives of the thermodynamical potential cd Ž b ,g . give the fluctuations
of N and R in this grand canonical ensemble,
2
E 2 cd Ž b ,g . 2
E 2 cd Ž b ,g .
Ž DN . s , Ž D R. s . Ž 2.17 .
Eb2 Eg2
Let us define as usual the entropy SŽ b ,g . as the logarithm of the number of string
configurations having values of Nand R equal to N and R, Eqs. Ž2.15., within the
8
It would be interesting to see if one can technically implement a large d approach to our counting
problem.
9
We tried to work in the light-cone gauge, with dy1 independent oscillators. However, the necessary
inclusion of the longitudinal term My2 Ž pP X˜ . 2 in Ž2.9., which is quadratic in the longitudinal oscillators
ayn s p
Ž q .y1 Ltransverse
n , leads to a complicated, interacting theory of the dy1 transverse oscillators.
precision of the rms fluctuations Ž2.17. w26x. This definition means that, in the
saddle-point approximation, Z d Ž b ,g . , exp w S y b N y g R x, i.e.
cd Ž b ,g . , S Ž b ,g . y b N y g R , Ž 2.18 .
or
E cd Ž b ,g . E cd Ž b ,g .
S , cd Ž b ,g . y b yg . Ž 2.19 .
Eb Eg
In other words, the entropy SŽ N, R . is the Legendre transform of cd Ž b ,g ..
Because of the Žassumed. independence of the d oscillators in Ž2.13., one has
`
yd
Z d Ž b ,g . s Ł w1 y eyŽ b nq g r n. x , Ž 2.20 .
ns1
i.e.
cd Ž b ,g . s d c 1 Ž b ,g . , Ž 2.21 .
with
` g
c 1 Ž b ,g . s y Ý
ns1
ž
ln 1 y exp yb n y
n / . Ž 2.22 .
We shall check a posteriori that we are interested in values of b and g such that
b < 'b g < 1 . Ž 2.23 .

For such values, one can approximate the discrete sum Ž2.22. by a continuous integral
over x s b n. This yields
IŽ d .
c 1 Ž b ,g . s , where d ' 'b g , Ž 2.24 .
b
and
` 2
IŽ d . 'y H0 dx ln 1 y eyŽ x q d r x.
. Ž 2.25 .
As d s b g < 1, we can try to expand I Ž d . in powers of d : I Ž d . s I Ž0. q d I X Ž0. q

'
oŽ d .. ŽThough the integral Ž2.25. is expressed in terms of d 2 , its formal expansion in
powers of d 2 leads to divergent integrals.. The zeroth-order term is I Ž0. s yH0`dx ln Ž1
y eyx . s p 2r6, while
` du d ` du
I X Ž 0 . s lim y2
d ™0 H0 u e d Ž uq1ru .
y1
s y2 H0 2
u q1
s yp .
Hence, using Ž2.21.,

1
cd Ž b ,g . s
b
'
CyD b g qoŽ b g . ' Ž 2.26 .
with
p2
Cs d, Dsp d . Ž 2.27 .
6
ŽThe notation D in Ž2.27. should not be confused with the space-time dimension d q 1..
The thermodynamic potential Ž2.26. corresponds to the mean values
1
Cy
2
D bg ' D
N, , R, , Ž 2.28 .
b2 2d
and to the entropy
2 CyD d D2
S, '
,2 C N 1y , Ž 2.29 .
b 8C R
i.e.
d 3d 1 d aX 1r2
l s2
S,2 p ( 6
N 1y
4 R
,2 p ž / 6
M 1 y 34
R2
. Ž 2.30 .
(
The lowest-order term 2 p d Nr6 is the usual ŽHardy–Ramanujan. result for d
independent oscillators, without size restriction. The factor in bracket, 1 y
Ž3r4.Ž l s2rR 2 ., with l s2 s 2 a X , gives the fractional reduction in the entropy brought by
imposing the size constraint R 2 , dy1 l s2 R. Under the conditions Ž2.23. the fluctua-
tions Ž2.17. are fractionally small. More precisely, Eqs. Ž2.17. yield
2 2
Ž DN . Ms Ž D R. b Ž R 2rl s2 .
;b; , ; ; . Ž 2.31 .
N2 M R2 d Ž MrMs .
As said above, though we worked under the Žphysically incorrect. assumption of d
independent oscillators at each level n, we expect the result Ž2.29. to be correct when
d 4 1. ŽWe recall that the exact result for S in absence of size restriction is
(
2 p Ž d y 1 . Nr6 .. Note the rough physical meanings of the auxiliary quantities b , g
and d : b ; Ž N .y1r2 ; Ž MrM s .y1, d ; Ry1 ; l s2rR 2 , g ; Ž N .1r2rR
R 2 ; M l s5rR 4 .
Summarizing, the main result of the present section is that the number Ž‘‘degeneracy’’.
of free string states of mass M and size R Žwithin the narrow bands defined by the
fluctuations Ž2.17.. is of the form
D Ž M , R . ; exp c Ž R . a 0 M , Ž 2.32 .
X
where a 0 s 2 p ŽŽ d y 1. a r6.1r2 and
c1 R2
c Ž R. s 1 y ž R2 /ž 1 y c2
M2 / , Ž 2.33 .
with the coefficients c1 and c 2 being of order unity in string units. ŽWe have added, for
completeness, in Eq. Ž2.32. the factor 1 y c 2 R 2rM 2 which operates when one consid-
ers very ‘‘large’’ string states, R 2 4 R rw

2 Ž
as discussed below Eq. Ž2.3.... The coeffi-
cient c Ž R . gives the fractional reduction in entropy brought by imposing a size
constraint. Note that Žas expected. this reduction is minimized when c1 Ry2 ; c 2 R 2rM 2 ,
(
™
i.e. for R ; R rw ; l s MrM s . ŽThe absolute reduction in degeneracy is only a factor
O Ž1. when R ; R rw .. Note also that c Ž R . 0 both when R ; l s and when R ;
l s Ž MrMs .. ŽThe latter value corresponding to the vicinity of the leading Regge
trajectory J ; a X M 2 ..
3. Mass shift of string states due to self-gravity
In this section we shall estimate the mass shift of string states Žof mass M and size
R . due to the exchange of the various long-range fields which are universally coupled to
the string: graviton, dilaton and axion. As we are interested in very massive string states,
M 4 M s , in extended configurations, R 4 l s , we expect that massless exchange domi-
nates the Žstate-dependent contribution to the. mass shift.
The evaluation, in string theory, of Žone loop. mass shifts for massive states is
technically quite involved, and can only be tackled for the states which are near the
leading Regge trajectory w27–30x. ŽIndeed, the vertex operators creating these states are
the only ones to admit a manageable explicit oscillator representation.. As we consider
states which are very far from the leading Regge trajectory, there is no hope of
computing exactly Žat one loop. their mass shifts. We shall resort to a semi-classical
approximation, which seems appropriate because we consider highly excited configura-
tions. As a starting point to derive the mass-shift in this semi-classical approximation we
shall use the classical results of Ref. w31x which derived the effective action of
fundamental strings. The one-loop exchange of gmn , w and Bmn leads to the effective
action
I eff s I0 q I1 , Ž 3.1 .
where I0 is the free ŽNambu. string action
Ž with d 2 s1 ' d s1 dt 1 ,g 1 s ydet ga b Ž X m Ž s1 ,t 1 . . . :

I0 s yT d 2 s 1 g 1 ,
H ( Ž 3.2 .
and I1 the effect of the one-loop interaction Ž X 1m ' X m Ž s 1 ,t 1 ., . . . .
2
I1 s 2p HHd s 1 d 2s 2 GF Ž X 1 y X 2 . g 1 g 2 Ctot Ž X 1 , X 2 . ,
( ( Ž 3.3 .
where GF is Feynman’s scalar propagator ŽI GF Ž x . s yd D Ž x .., and Ctot Ž X 1 , X 2 . s

Jw Ž X 1 . P Jw Ž X 2 . q Jg Ž X 1 . P Jg Ž X 2 . q JB Ž X 1 . P JB Ž X 2 . comes from the couplings of w ,
gmn and Bmn to their corresponding world-sheet sources Žindices suppressed; spin-struc-
ture hidden in the dot product.. The exchange term Ctot takes, in null Žconformal.
coordinates s "s t " s , the simple left–right factorized form w31x
(g (g
1 2 Ctot Ž X 1 , X 2 . s 32 GN T 2 Ž Eq X 1m Eq X 2 m . Ž Ey X 1n Ey X 2 n . . Ž 3.4 .
Here, T s Ž2 p a X .y1 is the string tension, GN is Newton’s constant 10 and E "s ErEs "s
1 m
2 Et " Es . Let us define P " s P " s
Ž . m Ž ".
by Ž l s s '2 a X as above.
2 E " X m s l s P"
m
, Ž 3.5 .
so that, for an open Žbosonic. string Žwith a 0m ' l s p m .,
q`
"
m
P" s Ý a nm eyi n s . Ž 3.6 .
y`
Using the definition Ž3.5. and inserting the Fourier decomposition of GF yields
4 GN dD k 1 2
I1 s H Ž 2p . D 2 HHd s 1 d 2s 2 Ž Pq Ž X 1 . P Pq Ž X 2 . .
p k yi ´
= Ž Py Ž X 1 . P Py Ž X 2 . . e i kPŽ X 1yX 2 . , Ž 3.7 .

where one recognizes the insertion of two gravitational vertex operators V mn Ž k; X . s
mŽ . n Ž . i kP X
Pq X Py X e at two different locations on the world-sheet, and with two
opposite momenta for the exchanged graviton11 . ŽNote that the exchanged graviton is
m Ž .. 2
off-shell.. It is convenient to use the Virasoro constraints Ž P " X s 0 to replace in
1 m .2 m mŽ mŽ
Eq. 3.6 P " X 1 P P " X 2 s y 2 D P " where D P " ' P "
Ž . Ž . Ž . Ž X1 . y P " X 2 .. It is
m m Ž
important to note that the zero-mode contribution a 0 drops out of D P " i.e.
m ˜ m
D P " s D P " is purely oscillatory . .
Writing that the correction I1 to the effective action I eff Žwhich gives the vacuum
persistence amplitude; see, e.g., Eq. Ž7. of Ref. w31x. must correspond to a phase shift
0
yHd E dt s yHd M dXcm , in the center-of-mass frame of the string, yields Žwith the
normalization 2.5 the link I1 s yl s2Hdt M d M s y 12 l s2Hdt d M 2 . Let us also define
Ž ..
D X m ' X 1m y X 2m and decompose it in its zero-mode part D Xcm m
s l s2 p m Žt 1 y t 2 . and
its oscillatory part D X˜ m s X˜1m y X˜2m. Finally, the mass-shift can be read from
2 GN dD k 1 l s2 kP p Žt 1 y t 2 .
H dt d M 2 s y 2 H Ž 2p . D 2 HHd s 2
1 d 2s 2 e i W Ž k ,1,2 . ,
p l s
k yi´
Ž 3.8 .
where Ž1 and 2 being short-hands for Žt 1 , s 1 . and Žt 2 , s 2 ., respectively.
m 2 2 ˜
W Ž k ,1,2 . s Ž D Pq Ž 1,2 . . Ž D Pyn Ž 1,2 . . e i kPD X Ž1 ,2. . Ž 3.9 .
Interpreted at the quantum level, the classical result Ž3.8. gives Žmodulo some ordering
problems, which are, however, fractionally negligible when considering very massive
states. the mass-shift d MN2 of the string state < N : when replacing W Ž k,1,2., on the
right-hand side of Eq. Ž3.8., by the quantum average ² N < W Ž k,1,2. < N :. Here, we shall
mainly be interested in the real part of d M 2 , obtained by replacing Ž k 2 y i ´ .y1 by the
principal part of Ž k 2 .y1 Ždenoted simply 1rk 2 ., i.e. the Feynman Green’s function
GF Ž x . by the half-retarded-half-advanced one Gsym Ž x .. ŽThe imaginary part of d M 2
10
Normalized, in any dimension, by writing the Einstein action as Ž16 p GN .y1Hd D x g RŽ g .. '
11
For simplicity, we call ‘‘graviton’’ the exchanged particle, which is a superposition of the graviton, the
dilaton and the axion.
gives the decay rate, i.e. the total rate of emission of massless quanta.. As L0 y 1 is the
‘‘Hamiltonian’’ that governs the t-evolution of an open string, the vanishing of
Ž L0 y 1. < N : for any physical state ensures that ² N < W Ž k,1,2. < N : is t-translation
invariant, i.e. that it depends only on the difference t 12 ' t 1 y t 2 , and not on the
average t ' 12 Žt 1 q t 2 .. This means that the double world-sheet integration d 2s 1 d 2s 2
s dt 1 d s 1 dt 2 d s 2 s dt dt 12 d s 1 d s 2 on the right-hand side of Eq. Ž3.8. contains a
formally infinite infra-red ‘‘ volume’’ factor Hd t which precisely cancels the integral
Hd t on the left-hand side to leave a finite answer for d M 2 .
™ ™
It is also important to note the good ultraviolet behaviour of Eq. Ž3.8.. The ultraviolet
™
limit k ` corresponds to the coincidence limit Žt 2 , s 2 . Žt 1 , s 1 . on the world-sheet.
™
Let us define u ' sq q y y
1 y s 2 , Õ ' s1 y s 2 and consider the coincidence limit u 0,
Õ 0. In this limit the vertex insertion factors Ž D Pq . 2 Ž D Py . 2 tend to zero like u 2 Õ 2 ,
while the Green’s function blows up like wŽ D X . 2 xyŽ Dy2.r2 A Ž uÕ .yŽ Dy2.r2 . The result-
ing integral, Hdu dÕ Ž uÕ .yŽ Dy6.r2 , has its first ultraviolet pole when the space-time
dimension D s d q 1 s 8. This means probably that in dimensions D 0 8 the exchange
of massive modes Žof closed strings. becomes important. Our discussion, which is
limited to considering only the exchange of massless modes, is probably justified only
when D - 8.
Following the Žapproximate. approach of Section 2 we shall estimate the average
mass shift d M 2 Ž R . for string states of size R by using the grand canonical ensemble
with density matrix
y1
r ' Ž Z d Ž b ,g . . exp Ž yb N yg : R : . , Ž 3.10 .
where the operators N and : R:, defined by Eqs. Ž2.7. and Ž2.12., belong to the Fock
space built upon d sequences of string oscillators a ni Žformal ‘‘center-of-mass’’
oscillators.. For any quantity Q Žbuilt from string oscillators. we denote the grand
canonical average as ² Q :b ,g ' tr Ž Q r .. Using the t-shift invariance mentioned above,
Eq. Ž3.8. yields
2 GN dd k d v
d Mb2,g s y H Ž 2p .
p l s2 dq1
w k 2 yv 2 yi´ x
l s2 M v t 12
= dt 12 d s 1 d s 2 eyi
H ²W Ž k ,1,2 . :b ,g , Ž 3.11 .
where we have separated k m in its center-of-mass components k 0 s v , k i s k, and
where t 12 ' t 1 y t 2 as above.
We shall estimate the grand canonical average ²W :b ,g in a semi-classical approxima-
tion in which we neglect some of the contributions linked to the ordering of the operator
W, but take into account the quantum nature of the density matrix r , Eq. Ž3.10.. The
discreteness of the Fock states built from theŽ a in . †, and the Planckian nature of r will
play a crucial role in the calculation below. ŽBy contrast, a purely classical calculation
would be awkward and ill-defined because of the problem of defining a measure on
classical string configurations, and because of the Rayleigh–Jeans ultraviolet catastro-
phe.. To compute ²W :b ,g it is convenient to define it as a double contraction of the
coefficient of zm11 zm22 zm33 zm44 in the exponentiated version of W,
m1
Wz ' :exp zm11 D Pq m2
q zm22 D Pq m3
q zm33 D Py m4
q zm44 D Py q ik P D X˜ : Ž 3.12 .
We shall define our ordering of W by working with the normal ordered exponentiated
operator Ž3.12. Žand picking the term linear in zm11 zm22 zm33 zm44 .. The average ²Wz :b s
tr ŽWz r . Žwhere, to ease the notation, we drop the extra label g . can be computed by a
generalization of Bloch’s theorem. Namely, if A denotes any operator which is linear in
the oscillators a ni , we have the results
1 1
² e A :b s exp 2
² A2 :b ; ² :e A : :b s exp 2
² : A2 : :b , Ž 3.13 .
as well as their corollaries
1 1
² e A : 0 s exp 2
² A2 : 0 ; e A s :e A : exp 2
² A2 : 0 , Ž 3.14 .
™
where ²W : 0 denotes the vacuum average Žobtained in the zero temperature limit
by1 0.. The simplest way to prove these results is to use coherent-state methods w32x
Žsee also,Ref. w24x and Appendix 7.A of Ref. w33x.. For instance, to prove the second
equation Ž3.13. it is sufficient to consider a single oscillator and to check that Ždenoting
q s eye , with e s b n q grn Žlabel n suppressed., so that Z s Ž1 y q .y1 , and < b . '
exp Ž ba† . <0:.
† † †
² e c1 a e c 2 a :b s Zy1 tr Ž e c1 a e c 2 a q a a
.
2
d b )
b † †
s Zy1 H p
eyb Ž b< ec a 1 e c2 a q a a < b .
d2 b )
b )
s Ž1yq. H p
eyb Ž b< ec b 1 e c 2 a < qb .
d2 b ) )
s Ž1yq. H eyŽ1 yq . b b
e c1 b qc 2 q b
p
s e c1 c 2 q rŽ1yq . , Ž 3.15 .
and to recognize that qrŽ1 y q . s w e e y 1xy1 is the Planckian mean occupation number
² a† a:b .
If we apply the second equation Ž3.13. to an expression of the type
4
Wz s :exp
ž Ý
is1
z i A i q B :,
/
one gets a Wick-type expansion for the coefficient Žsay W1234 . of z 1 z 2 z 3 z4 :
1
w BBx
W1234 s e 2
Ž w A1 A 2 x w A 3 A 4 x q 2 terms q w A1 B x w A 2 B x w A 3 A 4 x
q5 terms q w A1 B x w A 2 B x w A 3 B x w A 4 B x . , Ž 3.16 .
where w AB x denotes the ‘‘thermal’’ contraction w AB x ' ² : AB: :b .
The looked-for grand canonical average of W Ž k,1,2., Eq. Ž3.9., is given by replacing
B s ik P D X˜ and A1 s A 2 s D Pq
m
, A 3 s A 4 s D Pyn
in Eq. Ž3.16.. This leads to
1
²W :b ,g s ey 2 ² :Ž kPD X˜ . 2 2
2
: :b ,g
½² :Ž D P q . : :b ,g ² : Ž D Py . : :b ,g q . . . 5 , Ž 3.17 .
where the ellipsis stand for other contractions Žwhich will be seen below to be
subleading..
The calculation of the various contractions w AB x ' ² : AB: :b in Eqs. Ž3.16., Ž3.17. is
easily performed by using the basic contractions among the oscillators a n ,a†m Ž n,m ) 0.
Žwhich are easily derived from the definition Ž3.10. of the density matrix.
† † d i j dn m
² :a ni Ž a mj . : :b ,g s ² : Ž a mj . a in : :b ,g s
, Ž 3.18 .
een y1
where e n s b n q grn. The other contractions w aax and w a†a† x vanish. In terms of the
a-oscillators, the basic contraction reads w a ni a mj x s d i j dnqm
0 < n <rŽexpŽ e < n < . y 1., where
now n and m can be negative but not zero . Using these basic contractions, and the
Ž .
oscillator expansion Ž2.6. of X˜ m Žand noting that, in the center-of-mass frame only the
spatial components of X˜ m survive. one gets
` x n Ž 1,2 .
2
² : Ž k P D X˜ . : :b ,g s 2 k 2 l s2 Ý e , Ž 3.19 .
ns1 n Ž e y 1.
n
with
x n Ž 1,2 . s cos 2 n s 1 q cos 2 n s 2 y 2 cos n s 1 cos n s 2 cos n t 12 . Ž 3.20 .
Similarly, the oscillator expansion Ž3.6. yields
` n pq
n Ž 1,2 .
² : D Pq . 2 : :b ,g s 4 d Ý , Ž 3.21 .
ns1 een y1
with
pq q q
n Ž 1,2 . s 1 y cos n Ž s 1 y s 2 . s 1 y cos n Ž t 12 q s 1 y s 2 . . Ž 3.22 .
.2
The result for Ž D Py is obtained by changing s
ys 1 q s 2 ..
q
™ y
s in Eq. Ž3.22. Ži.e. s 1 y s 2 ™
We can estimate the values of the right-hand sides of Eqs. Ž3.19. and Ž3.21. by using
the following ‘‘statistical’’ approximation. In the parameter range discussed in Section 2
the basic sums Ý n " 1 Ž e e n y 1.y1 , appearing in Eqs. Ž3.19., Ž3.21., see their values
dominated by a large interval, D n 4 1, around some n 0 4 1, of values of n, so that
one can, with a good approximation, replace the discrete sum over n by a formal
continuous integral over a real parameter. In such a continuous approximation one can
integrate by parts to show that any ‘‘oscillatory’’ integral of the type Hdn f Ž n. cos n s s
w ny1 f Ž n. sin n s x y Hdn ny1 f X Ž n. sin n s is, because of the factors ny1 , numerically
much smaller than the non-oscillatory one Hdn f Ž n.. ŽHere s denotes some combination
of s 1 and s 2 , like 2 s 1 , 2 s 2 , s 1 " s 2 .. Alternatively, we can say that, for generic
values of s 1 and s 2 , one can treat in Eqs. Ž3.20. or Ž3.22. cos n s 1 and cos n s 2 as
statistically independent random variables with zero average. Within such an approxima-
tion one can estimate Ž3.19. by replacing x nŽ1,2. by 1 Žbecause cos 2 n s 1 q cos 2 n s 2 s
1 q 12 Žcos 2 n s 1 q cos 2 n s 2 ... Similarly, one can estimate Ž3.21. by replacing pn" 1.
The resulting estimates of Eqs. Ž3.19. and Ž3.21. introduce exactly the grand canonical
™
averages of the quantities : R: and N:
2
1
2
² : Ž k P D X˜ . : :b ,g , k 2 ² R 2 :b ,g , Ž 3.23 .
2 2 X
² : Ž D Pq . : :b ,g , ² : Ž D Py . : :b ,g , 4 ² N :b ,g , 4 a M 2 . Ž 3.24 .
Furthermore, one can check that the other contractions Žlike w D Pq D Py x or w D Pq k P

D X˜ x. entering Eq. Ž3.17. are all of the ‘‘oscillatory’’ type which is expected to give
subleading contributions.
Inserting the results Ž3.23., Ž3.24. into Eqs. Ž3.17. and Ž3.11. leads to a trivial integral
over t 12 Ž Hd t 12 expŽyi l s2 M v t 12 . s 2p d Ž v .rŽ l s2 M .. and, hence, to the follow-
ing result for d M s d M 2rŽ2 M .
2
R2
2
dd k eyk
d Mb ,g , y4 p GN M H Ž 2p . d
. Ž 3.25 .
k2yi ´
The imaginary part of d M is easily seen to vanish in the present approximation.

Finally,
M2
d Mb ,g , yc d GN , Ž 3.26 .
R dy2
with the Žpositive12 . numerical constant
1 y1
dy2 2
Ž dy2 .
cd s Ž 4p . , Ž 3.27 .
2
equal to 1r 'p in d s 3.
The result Ž3.26. was expected in order of magnitude, but we found useful to show
how it approximately comes out of a detailed calculation of the mass shift which
incorporates both relativistic and quantum effects. It shows clearly that perturbation
theory breaks down, even at arbitrarily small coupling, for sufficiently heavy and
compact strings. Let us also point out that one can give a simple statistical interpretation
of the calculation Ž3.16. of the normal-ordered vertex operator W Ž k,1,2., with the basic
contractions Ž3.18.. The result of the calculation would have been the same if we had
simply assumed that the oscillators a in were classical, complex random variables with a
Gaussian probability distribution A exp y 12 Ž e e n y 1 . < a in < 2 . This equivalence under-
lies the success of the classical random walk model of a generic excited string state. The
fact that the random walk must be made of MrM s independent steps is linked to the
fact that the Planckian distribution of mean occupation numbers, Nn s ŽexpŽ b n q grn.
y 1.y1 is sharply cut off when n R by1, i.e. from Eq. Ž2.28., when n R MrM s . More
precisely, using the same ‘‘statistical’’ approximation as above, one finds that the slope
correlator ² : Es X˜ i Žt , s 1 . Es X˜ j Žt , s 2 .: :b ,g decays quite fast when < s 2 y s 1 < R M srM.
Finally, let us mention that, by using the same tools as above, one can compute the
imaginary part of the mass shift d M s d Mreal y i Gr2, i.e. the total decay rate G in
massless quanta. The quantity G is, in fact, easier to define rigorously in string theory
because, using Ž k 2 y i ´ .y1 s PP Ž k 2 .y1 q ip d Ž k 2 . in Ž3.8., it is given by an integral
where the massless quanta are all on-shell. When g s 0 Ža consistent approximation for
12
The sign d M - 0 was classically clear Ževen when taking into account relativistic effects., say in ds 3,
from the starting formulas Ž3.3., Ž3.4. where Gsym Ž x . s Ž4p .y1 d Ž x 2 . ) 0 and 4Ž Eq X1PEq X 2 . Ž Ey X1PEy X 2 .
s Ž Eq D X . 2 Ž Ey D X . 2 ) 0 because E " D X m is purely spacelike in the center-of-mass frame. The same
conclusion would hold in the light-cone gauge.
a result dominated by n ; by1 . one can use the covariant formalism with D s d q 1
oscillators to find,after replacing a discrete sum over n by an integral over v ,
GN n 2
G s cXd 2 Hd v v dy2
ž bn / , Ž 3.28 .
M l s
e y1
where cXd is a numerical constant, and where n s M l s2 vr2. The spectral decomposi-
tion of the total power radiated by the b-ensemble of strings is then simply deduced
from Ž3.28. by adding a factor " v in the integrand,
GN n 2
P s cXd 2 Hd v v dy1
ž bn / . Ž 3.29 .
M l s
e y1
The results Ž3.28., Ž3.29. agree with corresponding results Žfor closed strings. in Ref.
w10x and in w34x Žnote, however, that the factor M 2 in Eq. Ž3.2. of Ref. w34x should be M
and that the constant contains GN and powers of l s .. The integrals Ž3.28., Ž3.29. are
dominated by n ; by1 , i.e. v ; M s . This gives for the integrated quantities:
G;g2 M, P ; g 2 M Ms . Ž 3.30 .
The second equation Ž3.30. means that the mass of a highly excited string decays
exponentially, with half-evaporation time
string
ls
tevap ' MrP ; . Ž 3.31 .
g2
Let us anticipate on Section 3 and note that, at the transition l ' g 2 MrMs ; 1 between
string states and black hole states, not only the mass and the entropy are Žin order of
magnitude at least. continuous, but also the various radiative quantities: total luminosity
P, half-evaporation time tevap , and peak of emission spectrum. Indeed, for a black hole
decaying under Hawking radiation the temperature is TBH ; Ry1 BH and
y2 y2 rŽ dy2. BH
P BH ; Ry2
BH ; l s l , tevap ; R BH S BH ; l s gy2 l d rŽ dy2. . Ž 3.32 .
4. Entropy of self-gravitating strings
In the present section we shall combine the main results of the previous sections, Eqs.
Ž2.32. and Ž3.26., and heuristically extend them at the limit of their domain of validity.
We consider a narrow band of string states that we follow when increasing adiabatically
™
the string coupling g, starting from g s 0 13 . Let M0 , R 0 denote the ‘‘bare’’ values Ži.e.
for g 0. of the mass and size of this band of states. Under the adiabatic variation of
g, the mass and size, M, R, of this band of states will vary. However, the entropySŽ M, R .
remains constant under this adiabatic process: SŽ M, R . s SŽ M0 , R 0 .. We assume, as
usual, that the variation of g is sufficiently slow to be reversible, but sufficiently fast to
13
Alternatively, we can consider the coupling as an adjustable parameter Žit is so in perturbation theory. and
just follow how different physical quantities change as g is varied, without pretending that the change takes
place in physical time.
be able to neglect the decay of the states. We consider states with sizes l s < R 0 < M0
for which the correction factor,
2 2
c Ž R 0 . , Ž 1 y c1 Ry2
0 . Ž 1 y c 2 R 0rM 0 . , Ž 4.1 .
in the entropy
S Ž M0 , R 0 . s c Ž R 0 . a 0 M0 , Ž 4.2 .
™
is near unity. ŽWe use Eq. Ž2.32. in the limit g 0, for which it was derived.. Because
of this reduced sensitivity of c Ž R 0 . on a possible direct effect of g on R Ži.e.
RŽ g . s R 0 q d g R ., the main effect of self-gravity on the entropy Žconsidered as a
function of the actual values M, R when g / 0. will come from replacing M0 as a
function of M and R. The mass-shift result Ž3.26. gives d M s M y M0 to first order in
g 2 . To the same accuracy 14 , Eq. Ž3.26. gives M0 as a function of M and R,
M2 g2 M
M0 , M q c 3 g 2
R dy2
s M 1 q c3
ž R dy2 / , Ž 4.3 .
where c 3 is a positive numerical constant.

Finally, combining Eqs. Ž4.1. – Ž4.3. Žand neglecting, as just said, a small effect linked
to d g R / 0. leads to the following relation between the entropy, the mass and the size
Žall considered for self-gravitating states, with g / 0.
1 R2 g2 M
S Ž M , R . , a0 M 1 yž R2 /ž 1y
M2 /ž 1q
R dy2 / . Ž 4.4 .
For notational simplicity, we henceforth set to unity the coefficients c1 , c 2 and c 3 . There
is no loss of generality in doing so because we can redefine l s , R and g to that effect,
and use the corresponding Žnew. string units. The main point of the present paper is to
emphasize that, for a given value of the total energy M Žand for some fixed value of g .,
the entropy SŽ M, R . has a non-trivial dependence on the radius R of the considered
string state. Eq. Ž4.3. exhibits two effects varying in opposite directions: Ži. self-gravity
favors small values of R Žbecause they correspond to larger values of M0 , i.e. of the
‘‘bare’’ entropy., and Žii. the constraint of being of some fixed size R disfavors both
small Ž R < 'M . and large Ž R 4 'M . values of R. For given values of M and g, the
most numerous Žand therefore most probable. string states will have a size R ) Ž M; g .
which maximizes the entropy SŽ M, R .. Said differently, the total degeneracy of the
complete ensemble of self-gravitating string states with total energy M Žand no a priori
size restriction. will be given by an integral Žwhere D R is the rms fluctuation of R
given by Eq. Ž2.17..
dR SŽ M , R.
DŽ M . ; HDR e ; e SŽ M , R ) . , Ž 4.5 .
which will be dominated by the saddle point R ) which maximizes the exponent.
14
Actually, Eq. Ž4.3. is probably a more accurate version of the mass-shift formula because it exhibits the
real mass M Žrather than the bare mass M0 . as the source of self-gravity.
The value of the most probable size R ) is a function of M, g and the space
dimension d. To better see the dependence on d, let us first consider the case Žwhich we
generically assume. where the correction factors in Eq. Ž4.4. Žparentheses on the
right-hand-side. are very close to unity so that
S Ž M , R . , a0 M Ž 1 y V Ž R . . , Ž 4.6 .
where
1 R2 g2 M
V Ž R. s q y . Ž 4.7 .
R2 M2 R dy2
One can think of V Ž R . as an effective potential for R. The most probable size R )
must minimize V Ž R .. This effective potential can be thought of as the superposition of:
Ži. a centrifugal barrier near R s 0 Žcoming from the result Ž2.30.., Žii. an harmonic
potential Žforbidding the large values of R ., and Žiii. an attractive Žgravitational.
potential. When g 2 is small the minimum of V Ž R . will come from the competition
between the centrifugal barrier and the harmonic potential and will be located around the
2 2
value Ry2 '
) , R ) rM , i.e. R ) , M s R rw . This random walk value will remain
Žmodulo small corrections. a local minimum of V Ž R . Ži.e. a local maximum of SŽ M, R ..
as long as g 2 MrR )dy 2 < Ry2 2 2
) , i.e. for g < g 0 with
g 02 ' M Ž dy6.r2 . Ž 4.8 .

More precisely, working perturbatively in g 2 , the minimization of V Ž R . yields
dy2 g2
R ) , 'M 1 y
ž 8 g 02 / . Ž 4.9 .
Note that, when g 2 < g 02 , the value of V Ž R . at this local minimum is of order
Vmin , q2 Ry2 ) , q2 M
y1
, i.e. that it corresponds to a saddle-point entropy SŽ M, R ) .
, a0 M Ž1 y Vmin . , a 0 M y O Ž1. which differs essentially negligibly from the ‘‘bare’’
entropy a0 M Ž4 1.. To study what happens when g 2 further increases let us consider
separately the various dimensions d 0 3. We shall see that the special value g 02 , Eq.
Ž4.8., is significant Žas marking a pre-transition, before the transition to the black hole
state. only for d s 3. For d 0 4, the only special value of g 2 is the critical value
g c2 ; My1 , Ž 4.10 .
around which takes place a transition toward a state more compact than the usual
random walk one.
4.1. d s 3
Let us first consider the Žphysical. case d s 3, for which g 02 ; My3r2 < g c2 ; My1.
In that case, when g 2 becomes larger than g 02 , the Žunique. local minimum of V Ž R .
slowly shifts towards values of R lower than R rw and determined by the competition
between the centrifugal barrier 1rR 2 and the gravitational potentialy g 2 MrR.
In the approximation where we use the linearized form Ž4.6., Ž4.7., and where Žfor
g 2 4 g 02 . we neglect the term R 2rM 2 , the most probable size R ) is
2
R Ž)ds3. , 2 , when My3r2 < g 2 < My1 . Ž 4.11 .
g M
Note that as g 2 increases between My3 r2 and My1 , the most probable size R Ž)ds3.
smoothly interpolates between R rw and a value of order unity, i.e. of order the string
length. Note also that Vmin , yg 2 MrŽ2 R ) . , yg 4 M 2r4 remains smaller than one
when g 2 Q My1 so that the saddle-point entropy SŽ M, R ) . , a 0 M Ž1 y Vmin . never
differs much from the ‘‘bare’’ value a 0 M.
When g 2 , in its increase, becomes comparable to My1 , the radius becomes of order
one and it is important to take into account the Žsupposedly. more exact expression Ž4.4.
Žin which the factor Ž1 y Ry2 . plays the crucial role of cutting off any size R ( 1.. If we
neglect, as above, the term R 2rM 2 Žwhich is indeed even more negligible in the region
R ; 1. but maximize the factored expression Ž4.4., we find that the most probable size
R ) reads
1 q '1 q 3 l2
R Ž)ds3. , , when g 2 4 My3r2 , Ž 4.12 .
l
where we recall the definition
l'g2 M . Ž 4.13 .
When l < 1, the result Ž4.12. reproduces the simple linearized estimate Ž4.11..
When l R 1, Eq. Ž4.12. says that the most probable size, when g 2 increases, tends to a
limiting size Ž R` s '3 . slightly larger than the minimal one Ž R min s 1. corresponding
to zero entropy. ŽNote that even for the formal asymptotic value R` s '3 , the reduction
in entropy due to the factor 1 y Ry2 is only 2r3.. On the other hand, the fractional
self-gravity GN MrR ) Žwhich measures the gravitational deformation away from flat
space., or the corresponding term in Eq. Ž4.4., continues to increase with g 2 as
l l2
s . Ž 4.14 .
R) 1 q '1 q 3 l2
The right-hand side of Eq. Ž4.14. becomes unity for l s '5 s 2.236. The picture
suggested by these results is that the string smoothly contracts, as g increases, from its
initial random walk size down to a limiting compact state of size slightly larger than l s .
For some value of l of order unity Žmay be between 1 and 2; indeed, even for l s 1 the
size R ) s 2 and the self-gravity lrR ) s 0.5 suggest one may still trust a compact
string description. the self-gravity of this compact string state will become so strong that
one expects it to collapse to a black hole state. We recall that, as emphasized in Refs.
w5,7–11,20x, when l ; 1, the mass of the string state matches Žin order of magnitude.
that of a ŽSchwarzschild. black hole with Bekenstein–Hawking entropy equal to the
string entropy S.
4.2. d s 4
When d s 4, the argument above Eq. Ž4.9. suggests that the random-walk size
remains the most probable size up to g 2 Q g 02 ; My1 , i.e. up to l Q 1. A more accurate
approximation to the most probable size R ) , when l - 1, is obtained by minimizing

exactly V Ž R ., Eq. Ž4.7.. This yields
1r4
R Ž)ds4. , M 1r2 Ž 1 y l . , when l-1 . Ž 4.15 .
This shows that the size will decrease, but one cannot trust this estimate when l 1y.
To study more precisely what happens when l ; 1 we must take into account the more
™
exact factorized form Ž4.4.. Let us now neglect the R 2rM 2 term and consider the
approximation
1 l
S Ž ds4. Ž M , R . , a0 M 1 y 2 1 q 2 .
ž /ž / Ž 4.16 .
R R
The right-hand side of Eq. Ž4.15. has a maximum only for l ) 1, in which case
1r2
2l
R Ž)ds4. , ž / when l)1 . Ž 4.17 .
ly1
If we had taken into account the full expression Ž4.4. the two results Ž4.15., Ž4.17., valid
on each side of l s 1, would have blended in a result showing that around15 l s 1 the
most probable size continuously interpolates between R rw and a size of order l s . Note
that, according to Eq. Ž4.17., as l becomes 4 1, R Ž)ds4. tends to a limiting size Ž R`
s '2 . slightly larger than R min s 1 Žcorresponding to zero entropy.. When l ) 1 the
fractional self-gravity of the compact string states reads
l ly1
s . Ž 4.18 .
R 2) 2
As in the case d s 3, one expects that for some value of l strictly larger than 1, the
self-gravity of the compact string state will become so strong that it will collapse to a
black hole state. Again the mass, size and entropy match Žin order of magnitude. those
of a black hole when l ; 1. The only difference between d s 4 and d s 3 is that the
transition to the compact state, though still continuous, is sharply concentrated around
l s 1 instead of taking place over the extended range My1 r2 Q l Q 1.
4.3. d 0 5
When d 0 5, the argument around Eq. Ž4.8. shows that the random walk size R rw
, 'M is a consistent local maximum of the entropy in the whole domain g 2 < g 02 , i.e.
for l ' g 2 M < M Ž dy4.r2 , which allows values l 4 1. However, a second, discon-
nected maximum of the entropy, as function of R, could exist. To investigate this we
consider again Ž4.4., when neglecting the R 2rM 2 term Žbecause we are interested in
other possible solutions with small sizes.,
SŽ M, R.
, Ž1 yx . Ž1 ql xn . 'sŽ x . , Ž 4.19 .
a0 M
where we have defined x ' Ry2 and n ' Ž d y 2.r2. By studying analytically the
maxima and inflection points of sŽ x ., one finds that, in the present case where
15
The transition takes place in the range < l y1 < ; My2 r3 corresponding to R ) ; M 1r3 .
n s Ž d y 2.r2 ) 1, there are two critical values l1 - l2 of the parameter l ' g 2 M.

The first one,
ny1 1r2 2
nq1 nq1 ny1
l1 s ž ny1 / , R 1 s xy1
1
r2
s ž ny1 / ) 1, s1 s 1 y ž nq1 / ,
Ž 4.20 .
corresponds to the birth Žthrough an inflection point. of a maximum and a minimum of
the function sŽ R .. Because s1 - 1 is strictly lower than the usual random walk
maximum with sŽ R rw . , 1 y 2rM , 1, the local maximum near R ; 1 of the entropy,
which starts to exist when l ) l1 , is, at first, only metastable with respect to R rw .
However, there is a second critical value of l, l2 ) l1 , defined by
n ny1 n 1r2
l2 s n žn / , R 2 s xy1
2
r2
s ž
) 1, /
s2 s 1 . Ž 4.21 .
y1 ny1
When l ) l2 the local maximum near R ; 1 of the entropy has sŽ R . ) 1, i.e. it has
become the global maximum of the entropy, making the usual random walk local
maximum only metastable. Therefore,when l ) l2 the most probable string state is a
™
very compact state of size comparable to l s . Formally, this new global maximum exists
for any l R 1 and tends, when l `, toward the limiting location R` s ŽŽ n q 1.rn .1r2
) 1, i.e. slightly Žbut finitely. above the minimum size R s 1. However, as in the cases
d ( 4, the self-gravity of the stable compact string state will become strong when l R 1,
so that it is expected to collapse Žfor some l c ) l 2 . to a black hole state. As in the cases
d ( 4, the mass, size and entropy of this compact string state match those of a black
hole. The big difference with the cases d ( 4 is that the transition between the Žstable.
random walk typical configuration and the Žstable. compact one is discontinuous. Our
present model suggests that Žwhen n ' Ž d y 2.r2 ) 1. a highly excited single string
system can exist, when l ) l 2 , in two different stable typical states: Ži. a dilute state of
typical size R rw , 'M and typical mean density r ; MrR rw d
; My n < 1, and Žii. a
condensed state of typical size R ; 1 and typical mean density using l ; 1.: r ; M ;
Ž
gy2 4 1. We shall comment further below on the value r ; gy2 of the dense state of
string matter.
5. Discussion
Technically, the main new result of the present work is the Ždimension independent.
estimate16 cŽ R . s 1 y c1rR 2 , with c1 , Ž3r4. l s2 s 3 a Xr2, of the factor giving the
decrease in the entropy Ž2p ŽŽ d y 1. a Xr6.1r2 M . of a narrow band of very massive
Žopen17 . string states M 4 Ž a X .y1 r2 , when considering only string states of size R
16
In spite of our efforts in Section 2, this result remains non-rigorous and open to O Ž1r d . fractional
corrections because of the difficulty to define a good quantum operator representing the mean radius of a
string state.
17
For technical simplicity, we have restricted our attention to open bosonic strings. We could have dealt
with closed bosonic strings by doubling the oscillators, but the level matching condition, NL s NR , would have
complicated the definition of the grand canonical ensemble we used. We expect that our results Žwhich are
semi-classical. apply Žwith some numerical changes. to open or closed superstrings.
Žmodulo some fractionally small grand-canonical-type fluctuations.. We have also

justified Žby dealing explicitly with relativistic and quantum effects in a semi-classical
approximation. and refined Žby computing the numerical coefficient, Eq. Ž3.27.. the
naive estimate, d M s yc d GN M 2rR dy 2 , of the mass shift of a massive string state due
to the exchange of long range fields Žgraviton, dilaton and axion.. ŽThe exchange of
these fields is expected to be the most important one both because very excited string
states tend to be large, and because the corresponding interactions are attractive and
cumulative with the mass..
Conceptually, the main new result of this paper concerns the most probable state of a
very massive single18 self-gravitating string. By combining our estimates of the entropy
reduction due to the size constraint, and of the mass shift we come up with the
expression Ž4.4. for the logarithm of the number of self-gravitating string states of size
R. Our analysis of the function SŽ M, R . clarifies the correspondence w7,9–11,20x
between string states and black holes. In particular, our results confirm many of the
results of Ref. w20x, but make them Žin our opinion. physically clearer by dealing directly
with the size distribution, in real space, of an ensemble of string states. When our results
differ from those of Ref. w20x, they do so in a way which simplifies the physical picture
and make even more compelling the existence of a correspondence between strings and
hysteresis, with a critical value g 02 ; My1 r2 for the string ™

black holes. For instance, Ref. w20x suggested that in d s 5 there was a phenomenon of
black hole transition, and
a different critical value g c2 ; My1 < g 02 for the inverse transition: black hole ™
string. Also, Ref. w20x suggested that in d ) 6, most excited string states would never
form black holes. The simple physical picture suggested19 by our results is the
following: In any dimension, if we start with a massive string state and increase the
string coupling g, a typical string state will, eventually, become more compact and will
end up, when l c s g c2 M ; 1, in a ‘‘condensed state’’ of size R ; 1, and mass density
r ; gy2c . Note that the basic reason why small strings, R ; 1, dominate in any
dimension the entropy when l ; 1 is that they descend from string states with bare mass
M0 , M Ž1 q lrR dy 2 . ; 2 M which are exponentially more numerous than less con-
densed string states corresponding to smaller bare masses.
The nature of the transition between the initial ‘‘dilute’’ state and the final ‘‘con-
densed’’ one depends on the value of the space dimension d. ŽAs explained below Eq.
Ž4.4., we henceforth set to unity, by suitable redefinitions of l s , R and g, the
coefficients c1 , c 2 and c 3 .. In d s 3, the transition is gradual: when l - My1 r2 the size
of a typical state is R Ž)ds3. , M 1r2 Ž1 y M 1r2 lr8., when l ) M 1r2 the typical size is
R Ž)ds3. , Ž1 q Ž1 q 3 l2 .1r2 .rl. In d s 4, the transition toward a condensed state is still
continuous, but most of the size evolution takes place very near l s 1: when l - 1,
R Ž)ds4. , M 1r2 Ž1 y l.1r4 , and when l ) 1, R Ž)ds4. , Ž2 lrŽ l y 1..1r2 , with some smooth
18
We consider states of a single string because, for large values of the mass, the single-string entropy
approximates the total entropy up to subleading terms.
19
Our conclusions are not rigorously established because they rely on assuming the validity of the result
Ž4.4. beyond the domain Ž Ry2 <1, g 2 Mr R dy2 <1. where it was derived. However, we find heuristically
convincing to believe in the presence of a reduction factor of the type 1y Ry2 down to sizes very near the
string scale. Our heuristic dealing with self-gravity is less compelling because we do not have a clear signal of
when strong gravitational field effects become essential.
blending between the two evolutions around < l y 1 < ; My2 r3. In d 0 5, the transition is
discontinuous Žlike a first-order phase transition between, say, gas and liquid states..
Barring the consideration of metastable Žsupercooled. states, on expects that when
l s l2 , n nrŽ n y 1. ny1 Žwith n s Ž d y 2.r2., the most probable size of a string state
will jump from R rw Žwhen l - l2 . to a size of order unity Žwhen l ) l2 ..
Let us, for definiteness, write down in more detail what happens in d s 3. After
maximization over R, the entropy of a self-gravitating string is given, when My3 r2 <
g 2 < My1 , by
2
S Ž M . s S Ž M , R ) Ž M . . , a 0 M 1 q 14 Ž g 2 M . . Ž 5.1 .
By differentiating S with respect to M, one finds the temperature of the ensemble of
highly excited single string states of mass M:
2
T , THag 1 y 34 Ž g 2 M . ,
ž / Ž 5.2 .
with THag ' ay1 0 . Eq. 5.2 explicitly exhibits the modification of the Hagedorn tempera-
Ž .
ture due to self-gravity Žin agreement with results of w20x obtained by a completely
different approach.. Note that, both in Eqs. Ž5.1. and Ž5.2., the self-gravity modifications
are fractionally of order unity at the transition g 2 M ; 1.
One can think of the ‘‘condensed’’ state of Žsingle. string matter, reached Žin any d .
when l ; 1, as an analog of a neutron star with respect to an ordinary star Žor a white
dwarf.. It is very compact Žbecause of self-gravity. but it is stable Žin some range for g .
under gravitational collapse. However, if one further increases g or M Žin fact,
l s g 2 M ., the condensed string state is expected Žwhen l reaches some l3 ) l2 ,
l3 s O Ž1.. to collapse down to a black hole state Žanalogously to a neutron star
collapsing to a black hole when its mass exceeds the Landau–Oppenheimer–Volkoff
critical mass.. Still in analogy with neutron stars, one notes that general relativistic
strong gravitational field effects are crucial for determining the onset of gravitational
collapse; indeed, under the ‘‘Newtonian’’ approximation Ž4.4., the condensed string
state could continue to exist for arbitrary large values of l.
It is interesting to note that the value of the mass density at the formation of the
condensed string state is r ; gy2 . This is reminiscent of the prediction by Atick and
Witten w35x of a first-order phase transition of a self-gravitating thermal gas of strings,
near the Hagedorn temperature 20 , towards a dense state with energy density r ; gy2
Žtypical of a genus-zero contribution to the free energy.. Ref. w35x suggested that this
transition is first-order because of the coupling to the dilaton. This suggestion agrees
with our finding of a discontinuous transition to the single string condensed state in
dimensions 0 5 ŽRef. w35x work in higher dimensions, d s 25 for the bosonic case.. It
would be interesting to deepen these links between self-gravitating single string states
and multi-string states.
Assuming the existence Žconfirmed by the present work. of a dense state of
self-gravitating string matter with energy density r ; gy2 , it would be fascinating to be
able to explore in detail Žwith appropriate, strong gravity tools. its gravitational
20
Note that, by definition, in our single string system, the formal temperature T s Ž E SrE M .y1 is always
near the Hagedorn temperature.
dynamics, both in the present context of a single, isolated object Ž‘‘collapse problem’’.,
and in the cosmological context Žproblem of the origin of the expansion of the universe..
Let us come back to the consequences of the picture brought by the present work for
the problem of the end point of the evaporation of a Schwarzschild black hole and the
interpretation of black hole entropy. In that case one fixes the value of g Žassumed to be
< 1. and considers a black hole which slowly looses its mass via Hawking radiation.
When the mass gets as low as a value 21 M ; gy2 , for which the radius of the black hole
is of order one Žin string units., one expects the black hole to transform Žin all
dimensions. into a typical string 22 state corresponding to l s g 2 M ; 1, which is a
dense state Žstill of radius R ; 1.. This string state will further decay and loose mass,
predominantly via the emission of massless quanta, with a quasi thermal spectrum with
temperature T ; THagedorn s ay1 0
Žsee Eq. Ž3.29. and Refs. w9,10,34x. which smoothly
matches the previous black hole Hawking temperature. This mass loss will further
decrease the product l s g 2 M, and this decrease will, either gradually or suddenly,
cause the initially compact string state to inflate to much larger sizes. For instance, if
d 0 4, the string state will quickly inflate to a size R ; 'M . Later, with continued mass
loss, the string size will slowly shrink again toward R ; 1 until a remaining string of
mass M ; 1 finally decays into stable massless quanta. In this picture, the black hole
entropy acquires a somewhat clear statistical significance Žas the degeneracy of a
corresponding typical string state. only when M and g are related by g 2 M ; 1. If we
allow ourselves to vary Žin a Gedanken experiment. the value of g this gives a potential
statistical significance to any black hole entropy value S BH Žby choosing g 2 ; Sy1
BH . We
.
do not claim, however, to have a clear idea of the direct statistical meaning of S BH when
g 2 S BH 4 1. Neither do we clearly understand the fate of the very large space Žwhich
could be excited in many ways. which resides inside very large classical black holes of
radius R BH ; Ž g 2 S BH .1rŽ dy1. 4 1. The fact that the interior of a black hole of given
mass could be arbitrarily large 23 , and therefore arbitrarily complex, suggests that black
hole physics is not exhausted by the idea Žconfirmed in the present paper. of a reversible
transition between string-length-size black holes and string states.
On the string side, we also do not clearly understand how one could follow in detail
Žin the present non-BPS framework. the ‘‘transformation’’ of a strongly self-gravitating
string state into a black hole state.
Finally, let us note that we expect that self-gravity will lift nearly completely the
degeneracy of string states. ŽThe degeneracy linked to the rotational symmetry, i.e.
2 J q 1 in d s 3, is probably the only one to remain, and it is negligible compared to the
string entropy.. Therefore we expect that the separation d E between subsequent Žstring
and black hole. energy levels will be exponentially small: d E ; D M expŽySŽ M ..,
21
Note that the mass at the black hole™ string transition is larger than the Planck mass MP ; Ž GN .y1 r2 ;
™
y1 y1
g by a factor g 41.
22
A check on the single-string dominance of the transition black hole string is to note that the single
string entropy ; Mr M s is much larger than the entropy of a ball of radiation Srad ; Ž RM . d rŽ dq1. with size
R; R BH ; l s at the transition.
23
For example, in the Oppenheimer–Snyder model, one can join an arbitrarily large closed Friedmann dust
universe, with hyperspherical opening angle 0 ( x 0 (p arbitrarily near p , onto an exterior Schwarzschild
space-time of given mass M.
where D M is the canonical-ensemble fluctuation in M. Such a d E is negligibly small

compared to the radiative width G ; g 2 M of the levels. This seems to mean that the
discreteness of the quantum levels of strongly self-gravitating strings and black holes is
very much blurred, and difficult to see observationally.
Acknowledgements
This work has been clarified by useful suggestions from M. Douglas, K. Gawedzki,
M. Green, I. Kogan, G. Parisi, A. Polyakov, and Žlast but not least. M. Vergassola. We
wish also to thank A. Buonanno for collaboration at an early stage and D. Gross, J.
Polchinski, A. Schwarz, L. Susskind and A. Vilenkin for discussions. T.D. thanks the
Theory Division of CERN, Gravity Probe B ŽStanford University., and the Institute for
Theoretical Physics ŽSanta Barbara. for hospitality. Partial support from NASA grant
NAS8-39225 is acknowledged. G.V. thanks the IHES for hospitality during the early,
crucial stages of this work.
References
w1x S. Fubini, G. Veneziano, Nuovo Cim. A 64 Ž1969. 811.

w2x K. Huang, S. Weinberg, Phys. Rev. Lett 25 Ž1970. 895.
w3x J.D. Bekenstein, Phys. Rev. D 7 Ž1973. 2333.
w4x S.W. Hawking, Comm. Math. Phys. 43 Ž1975. 199.
w5x M. Bowick, L. Smolin, L.C.R. Wijewardhana, Gen. Rel. Grav. 19 Ž1987. 113.
w6x G. Veneziano, Europhys. Lett. 2 Ž1986. 199.
w7x L. Susskind, hep-thr9309145, unpublished.
w8x G. Veneziano, in: Hot Hadronic Matter: Theory and Experiments, Divonne, June, 1994, ed. J. Letessier,
H. Gutbrod, J. Rafelsky, NATO-ASI Series B: Physics, Vol. 346 ŽPlenum, New York, 1995. p. 63.
w9x E. Halyo, A. Rajaraman, L. Susskind, Phys. Lett. B 392 Ž1997. 319.
w10x E. Halyo, B. Kol, A. Rajaraman, L. Susskind, Phys. Lett. B 401 Ž1997. 15.
w11x G.T. Horowitz, J. Polchinski, Phys. Rev. D 55 Ž1997. 6189.
w12x A. Strominger, C. Vafa, Phys. Lett. B 379 Ž1996. 99.
w13x A. Sen, Mod. Phys. Lett. A 10 Ž1995. 2081.
w14x C.G. Callan, J.M. Maldacena, Nucl. Phys. B 472 Ž1996. 591.
w15x J.C. Breckenridge et al. Phys. Lett. B 381 Ž1996. 423.
w16x P. Salomonson, B.S. Skagerstam, Nucl. Phys. B 268 Ž1986. 349.
w17x P. Salomonson, B.S. Skagerstam, Physica A 158 Ž1989. 499.
w18x D. Mitchell, N. Turok, Phys. Rev. Lett. 58 Ž1987. 1577.
w19x D. Mitchell, N. Turok, Nucl. Phys. B 294 Ž1987. 1138.
w20x G.T. Horowitz, J. Polchinski, Phys. Rev. D 57 Ž1998. 2557.
w21x B. Sathiapalan, Phys. Rev. D 35 Ž1987. 3277.
w22x I.A. Kogan, JETP Lett. 45 Ž1987. 709.
w23x J.J. Atick, E. Witten, Nucl. Phys. B 310 Ž1988. 291.
w24x J. Scherk, Rev. Mod. Phys. 47 Ž1975. 123.
w25x M. Karliner, I. Klebanov, L. Susskind, Int. J. Mod. Phys. A 3 Ž1988. 1981.
w26x L.D. Landau, E.M. Lifshitz, Statistical Physics, Part 1, 3rd edition ŽPergamon, Oxford, 1980..
w27x H. Yamamoto, Prog. Theor. Phys. 79 Ž1988. 189.
w28x K. Amano, A. Tsuchiya, Phys. Rev. D 39 Ž1989. 565.
w29x B. Sundborg, Nucl. Phys. B 319 Ž1989. 415.
w30x B. Sundborg, Nucl. Phys. B 338 Ž1990. 101.

w31x A. Buonanno, T. Damour, Phys. Lett. B 432 Ž1998. 51.
w32x V. Alessandrini, D. Amati, M. Le Bellac, D. Olive, Phys. Rep. C 1 Ž1971. 269.
w33x M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, Vol. 1 ŽCambridge Univ. Press, Cambridge,
1987..
w34x D. Amati, J.G. Russo, Phys. Lett. B 454 Ž1999. 207.
w35x J.J. Atick, E. Witten, Nucl. Phys. B 310 Ž1988. 291.
B ™X l l s
q y
decays in supersymmetry
E. Lunghi a , A. Masiero, a , I. Scimemi b, L. Silvestrini c
a
SISSA-ISAS, Via Beirut 2-4, Trieste, Italy and INFN, Sezione di Trieste, Trieste, Italy
b
Dep. de Fisica Teorica, UniÕ. de Valencia, c. Dr. Moliner 50, E-46100, Burjassot, Valencia, Spain
c
Physik Department, Technische UniÕersitat ¨ Munchen,
¨ D-85748 Garching, Germany
Received 15 June 1999; received in revised form 4 October 1999; accepted 27 October 1999
Abstract
™ ™
We study the semileptonic decays B X s eqey, B X s mqmy in generic supersymmetric
extensions of the Standard Model. SUSY effects are parameterized using the mass insertion
™
approximation formalism and differences with the Constrained MSSM results are pointed out.
Constraints on SUSY contributions coming from other processes Že.g. b sg . are taken into
account. Chargino and gluino contributions to photon and Z-mediated decays are computed and
non-perturbative corrections are considered. We find that the integrated branching ratios and the
asymmetries can be strongly modified. Moreover, the behavior of the differential forward–back-
ward asymmetry remarkably changes with respect to the Standard Model expectation. q 2000
Elsevier Science B.V. All rights reserved.
PACS: 11.30.Pb; 13.20.He

Keywords: Semileptonic rare B decays; Supersymmetry
1. Introduction
One of the features of a general low energy supersymmetric ŽSUSY. extension of the
Standard Model ŽSM. is the presence of a huge number of new parameters. FCNC and
CP-violating phenomena constrain strongly a big part of the new parameter space.
However, there is still room for significant departures from the SM expectations in this
interesting class of physical processes. It is worthwhile to check all these possibilities on
the available data and on those processes that are going to be studied in the future. In
this way it is possible to indicate where new physics effects can be revealed as well as to
establish criteria for model building.
™
In this work we want to investigate the relevance of new physics effects in the
semileptonic inclusive decay B X s l q l y. This decay is quite suppressed in the
Standard Model; however, new B-factories should reach the precision requested by the
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 9 2 - 6
E. Lunghi et al.r Nuclear Physics B 568 (2000) 120–144 121
SM prediction w1x and an estimate of all possible new contributions to this process is
compelling.
Semileptonic charmless B decays have been deeply studied. The dominant perturba-
tive SM contribution has been evaluated in Ref. w2x and later two-loop QCD corrections
have been provided w3–7x. The contribution due to cc resonances to these results are
included in the papers listed in Refs. w8–11x. Long distance corrections can have a
different origin according to the value of the dilepton invariant mass one considers.
O Ž1rm2b . corrections have been first calculated in Ref. w12x and recently corrected in
Refs. w13–16x. Near the peaks, non-perturbative contributions generated by cc reso-
nances by means of resonance-exchange models have been provided in Refs. w13–
15,17–20x. Far from the resonance region, instead, Ref. w21x Žsee also Ref. w22x. estimate
cc long-distance effects using a heavy quark expansion in inverse powers of the
charm-quark mass Ž O Ž1rm2c . corrections..
An analysis of the SUSY contributions has been presented in Refs. w23–28x where the
authors estimate the contribution of the Minimal Supersymmetric Standard Model
ŽMSSM.. They consider first a universal soft supersymmetry breaking sector at the
Grand Unification scale ŽConstrained MSSM. and then partly relax this universality
condition. In the latter case they find that there can be a substantial difference between
the SM and the SUSY results in the Branching Ratios and in the forward–backward
asymmetries. One of the reasons of this enhancement is that the Wilson coefficient
™
in some region of the parameter space while respecting constraints coming from b sg .
The recent measurements of b sg w29x have narrowed the window of the possible
™
C 7 Ž MW . Žsee Section 2 for a precise definition. can change sign with respect to the SM
values of C 7 Ž MW . and in particular a sign change of this coefficient is no more allowed
™
in the Constrained MSSM framework. Hence, on one hand it is worthwhile considering
B X s l q l y in a more general SUSY framework then just the Constrained MSSM,
and, on the other hand, the above-mentioned new results prompt us to a reconsideration
of the process. In Ref. w30x the possibility of new-physics effects coming from
gluino-mediated FCNC is studied. Effects of SUSY phases in models with heavy first
and second generations sfermions have been recently discussed in Ref. w31x.
We consider all possible contributions to charmless semileptonic B decays coming
from chargino–quark–squark and gluino–quark–squark interactions and we analyze
both Z-boson and photon mediated decays. Contributions coming from penguin and box
diagrams are taken into account; moreover, corrections to the MIA results due to a light
t̃ R are considered. A direct comparison between the SUSY and the SM contributions to
the Wilson coefficients is performed. Once the constraints on mass insertions are
established, we find that in generic SUSY models there is still enough room to see large
deviations from the SM expectations for branching ratios and asymmetries. For our final
computation of physical observables we consider NLO QCD evolution of the coeffi-
cients and non-perturbative corrections Ž O Ž1rm2b ., O Ž1rm2c ., . . . ., each in its proper
range of the dilepton invariant mass.
Because of the presence of so many unknown parameters Žin particular in the scalar
mass matrices. which enter in a quite complicated way in the determination of the mass
eigenstates and of the various mixing matrices it is very useful to adopt the so-called
‘‘Mass Insertion Approximation’’ ŽMIA. w32x. In this framework one chooses a basis for
fermion and sfermion states in which all the couplings of these particles to neutral
122 E. Lunghi et al.r Nuclear Physics B 568 (2000) 120–144
gauginos are flavor diagonal. Flavor changes in the squark sector are provided by the
non-diagonality of the sfermion propagators. The pattern of flavor change is then given
by the ratios
2
d i fj s
žm ˜/
f
ij AB
, Ž 1.
Ž . AB Msq2
˜
where Ž m if j . A2 B are the off-diagonal elements of the f˜s u,d
˜ ˜ mass squared matrix that
mixes flavor i, j for both left- and right-handed scalars Ž A, B s left,right. and Msq is
the average squark mass Žsee e.g. Ref. w33x.. The sfermion propagators are expanded in
terms of the d s and the contribution of the first two terms of this expansion are
considered. The genuine SUSY contributions to the Wilson coefficients will be simply
proportional to the various d s and a keen analysis of the different Feynman diagrams
involved will allow us to isolate the few insertions really relevant for a given process. In
this way we see that only a small number of the new parameters is involved and a
general SUSY analysis is made possible. The hypothesis regarding the smallness of the
d s and so the reliability of the approximation can then be checked a posteriori.
Many of these d ’s are strongly constrained by FCNC effects w33–39x or by vacuum
stability arguments w40x. Nevertheless it may happen that such limits are not strong
enough to prevent large contributions to some rare processes. For instance it has been
recently found in Ref. w41x that the off-diagonal squark mass matrix elements can
enhance rare kaon decays by roughly an order of magnitude with respect to the SM
result.
The paper is organized as follows. In Section 2 we define the operator basis, the basic
formulae for the BR, the forward–backward asymmetry and the non-perturbative
corrections. Section 3 and Section 4 treat chargino and gluino contributions in the mass
insertion approximation. The light t˜R corrections are presented in Section 5. Constraints
on d ’s are discussed in Section 6 and final results and conclusions are drawn in Sections
7 and 8.
2. Operator basis and general framework
™
The effective Hamiltonian for the decay B X s l q l y in the SM and in the MSSM
is given by Žneglecting the small contribution proportional to K u)s K u b .
4GF 8 a 10
Heff s y
'2 K t)s K t b Ý Ci Ž m . Qi q 4p Ý C˜i Ž m . Q i , Ž 2.
is1 is9
where
Q1 s sL a gm bL a c L b g m c L b ,
Q2 s sL a gm bL b c L b g m c L a ,
Q3 s sL a gm bL a Ý q L b g mq L b ,
qsu , . . . , b
Q4 s sL a gm bL b Ý q L b g mq L a ,
qsu , . . . , b
Q5 s sL a gm bL a Ý q R b g mq R b ,
qsu , . . . , b
Q6 s sL a gm bL b Ý q R b g mq R a ,
qsu , . . . , b
e
Q7 s m b sL s mn bR Fmn ,
16p 2
gs
Q8 s m b sL T as mn bR Gmn
a
,
16p 2
Q9 s Ž sL gm bL . lg m l,
Q10 s Ž sL gm bL . lg m g 5 l, Ž 3.
K is the CKM matrix and q LŽ R. s 12 Ž1 . g 5 . q. This Hamiltonian is known at next-to-
leading order both in the SM w3–7x and in the MSSM w26–28x. We find that the most
general low-energy SUSY Hamiltonian also contains the operators
e
QX7 s m s s mn bL Fmn ,
8p 2 b R
QX9 s Ž sR gm bR . lg m l,
QX10 s Ž sR gm bR . lg m g 5 l. Ž 4.
However, it is shown in the following sections that the contribution of these operators is
negligible and so they are not considered in the final discussion of physical quantities.
SUSY contributions to other operators are negligible because they influence our
observables at an higher perturbative order.
With these definitions the differential branching ratio and the forward–backward
asymmetry can be written as
d G ŽB ™X l q y
™ X en .
1 s l .
RŽ s. '
GŽB c ds
2
a2 K t)s Ž1 y s.
s 2 Ž 1 q 2 s . ž < C˜9eff Ž s . < 2 q < C˜10 < 2 / q 4 Ž 1 q 2rs . < C 7 < 2
4p K cb f Ž z. kŽ z.
q12Re C 7) C˜9eff Ž s . , Ž 5.
1
Hy1 d cos u
™ X l l . SgnŽcosu .
d 2G Ž B s
q y
d G Ž B™X l l .
d cos u d s
A FB Ž s . ' 2 q y
1 s
Hy1 d cos u
d cos u d s
3Re w C˜10 )
Ž s C˜9eff Ž s . q 2C7 . x
sy
2
, Ž 6.
Ž 1 q 2 s . Ž < C˜9eff Ž s . < 2 q < C˜10 < 2 . q 4 1 q ž / < C 7 < 2 q 12Re w C 7) C˜9eff Ž s . x
s
where s s Ž p l qq p l y . 2rm2b , u is the angle between the positively charged lepton and
the B flight direction in the rest frame of the dilepton system, f Ž z . and k Ž z . are the
phase space factor and the QCD correction factor Ž z s m crm b . that enter G Ž B X c en .
and can be found in Refs. w7,42,43x. C˜9eff Ž s . includes all the contributions of the
™
operators Q1 y Q6 and Q8 and its complete definition for the SM and MSSM can be
found again in Refs. w3–7,26x.
In the literature the energy asymmetry is also considered w26x but it is easy to show
that these two kinds of asymmetries are completely equivalent; in fact a configuration in
the dilepton c.m.s. in which l q is scattered in the forward direction kinematically
implies E l q- E l y in the B rest frame Žsee for instance Refs. w13–15x..
It is worth underlying that integrating the differential asymmetry given in Eq. Ž6. we
do not obtain the global forward–backward asymmetry which is by definition
1 ™ X l l . SgnŽ cosu .
d 2G Ž B s
q y
N Ž l™
q
. y N Ž l§
q Hy1 d cos uHd s
d G Ž B™X l l .
. d cos u d s
s Ž 7.
N Ž l™
q
. q N Ž l§
q
. 1
2
s
q y
H d cos uHd s
y1 d cos u d s
where l ™q
and l § q
stand respectively for leptons scattered in the forward and
backward directions.
To this extent it is useful to introduce the following quantity:
Hy11 d cos u
d 2G Ž B ™X l s
q y
l .
Sgn Ž cos u .
™
d cos u d s
A FB Ž s . '
1 d 2G Ž B Xs l q l y .
H d cos u d s
H
y1 d cos u d s
y3Re w C˜10 )
Ž s C˜9eff Ž s . q2C7 . x Ž1y s . 2
s ,
2
Hd s Ž1y s .2 Ž 1q2 s . Ž < C˜9eff Ž s . < 2 q < C˜10 < 2 . q4 1q ž < C 7 < 2 q12Re w C 7) C˜9eff Ž s . x
/
s
Ž 8.
whose integrated value is given by Eq. Ž7..
Eqs. Ž5. and Ž6. have been corrected in order to include several non-perturbative
effects. First of all O Ž1rm2b . effects have been estimated by w13–15x
d 1r m 2b R Ž s .
2
3 l2 a2 K t)s 1
s
ž Ž 1 y 15s 2 q 10 s 3 . ž < C˜9eff Ž s . < 2 q < C˜10 < 2 /
2 m2b 4p 2
K cb f Ž z. kŽ z.
< C7 < 2 gŽ z.
y Ž 5 q 6 s y 7s 2 . Re C 7) C˜9eff Ž s .
y4 Ž 6 q 3s y 5s . 3
s / q
f Ž z.
RŽ s. ,
Ž 9.
d 1r m 2b A FB Ž s .
3 l2 Re C˜10
)
s Ž 9 q 14 s y 15s 2 . C˜9eff Ž s . q 2 Ž 7 q 10 s y 9 s 2 . C 7
s
2 m2b Ž 1 q 2 s . < C˜9eff Ž s . < 2 q < C˜10 < 2 q 4 Ž 1 q 2rs . < C 7 < 2 q 12Re C 7) C˜9eff Ž s .
ž /
3 l2 g Ž z . 4l1 s d 1r m 2b R Ž s .
q A FB Ž s .
ž 2 m2b f Ž z .
q
3m2b Ž 1 y s . 2
y
RŽ s. / , Ž 10 .
where
g Ž z . s 3 y 8 z 2 q 24 z 4 y 24 z 6 q 5 z 8 q 24 z 4 log w z x .
l1 and l2 are the two parameters that appear in the Heavy Quark Expansion ŽHQE..
While the value of l2 is quite well established Ž l2 s Ž MB2) y MB2 .r4., l1 is not yet
well known. In Refs. w44–46x l1 is estimated as l1 s y0.52 " 0.12 GeV 2 and in Ref.
w47x l1 s y0.10 " 0.05 GeV 2 . In what follows we consider the weighted average of the
two results l1 s y0.16 GeV 2 . As was pointed out by the authors of Refs. w13–15x these
corrections are no more valid near the endpoint region, s 1, where they diverge
because of the breaking down of the HQE. Following some recent analyses we have
™
stopped the BR’s corrections given in Eq. Ž9. at s s 0.78 Žsee Ref. w16x. and the ones in
Eq. Ž10. at s s 0.7 Žsee Refs. w13–15x..
In order to account for the corrections to the parton model approximation in the high
s region Refs. w13–15x and Ref. w16x adopt two different approaches. The former
considers a Fermi-motion model and the latter invokes the Heavy Hadron Chiral
Perturbation Theory ŽHHChPT.. A discussion about the usefulness of the Fermi-motion
model for semileptonic charmless B-decays is beyond the scope of this paper. In order
to have a model-independent description of the high energy region of the spectrum we
have considered the HHChPT corrections.
™ ™
For 0.88 - s - smax s Ž MB y MK . 2rm2b s 0.99 the branching ratio is dominated by
the exclusive decays B K l q l y and B Kp l q l y; in Ref. w16x it is shown that the
contribution of the latter is completely negligible. In the following we report the
expression of the branching ratio, valid in the interval of the spectrum given above,
computed in the HHChPT framework
t Ž Bd . GF2 MB5 a 2 f 1 Ž s m2brMB2 .

™ X en . 192p
2
R Ž s . Ch s 3
< K t)s K t b < 2 2 ½f 2
q Ž s . C˜9EP
ž
RŽ B c 4p 2
2 2
q C˜10 2
/ qa Ž s. m
T
2
b C 7 y 2 fq Ž s . aT Ž s . m b Re C 7) C˜9EP 5, Ž 11 .
where
g fB 1
aT s Ž 12 .
fp Õ P pK q Ž MB ) y MB . q Ž MB s y MB .
and the definitions of f 1 , fq, C˜9EP , Õ P pK can be found in Ref. w16x. Moreover, we have
put g s 0.5 according to the theoretical estimate given in Ref. w48x. In the intermediate
region 0.78 - s - 0.88 we have interpolated the obtained results. The form factors fq
and aT can be computed also using other methods ŽQCD sum rules w49–51x, light-cone
QCD sum rules w52,53x, QCD relativistic potential model. w54x but the HHChPT
approach is preferable in the endpoint region of the spectrum.
The asymmetries receive no contribution from the single kaon mode B K l q l y
and the endpoint of their spectrum is fixed, instead, by B Kp l q l y at s s Ž MB y MK ™ ™
y Mp . 2rm2b s 0.93. In the region 0.7 - s - 0.93 we use the parton model result because
these asymmetries have not been computed yet in the HHChPT framework.
Finally, the following O Ž1rm2c . corrections occur for s - 0.2 Žsee e.g. Ref. w21x.
2 2
8 l2 a K t)s Ž1 ys.
d 1r m 2c R Ž s . s y C2
9m 2c 4p 2 K cb f Ž z. kŽ z.
1 q 6 s y s2
q C˜9eff ) Ž 2 q s .
=Re F Ž s . C 7)
ž s / , Ž 13 .
d 1r m 2c A FB Ž s .
3 l2
s C2 Ž 1 q 3s .
9m2c
Re F Ž s . C˜10
)
= .
Ž 1 q 2 s . ž < C˜9eff Ž s . < 2 q < C˜10 < 2 / q 4 Ž 1 q 2rs . < C 7 < 2 q 12Re C 7) C˜9eff
Ž 14 .
The O Ž1rm2b . and O Ž1rm2c . corrections to A FB can be easily computed because of
the following relation:
RŽ s.
A FB s A FB . Ž 15 .
H d s RŽ s.
All the effects coming from the mass insertion approximation can be included in
formulae Ž5. – Ž14. writing the coefficients C 7 , C˜9eff Ž s ., C˜10 as
C 7 s C 7SM q C 7diag q C 7MI ,
SM diag MI
C˜9eff Ž s . s Ž C˜9eff Ž s . . q Ž C˜9eff . q Ž C˜9eff . ,
C˜10 s C˜10
SM
q C˜10
diag
q C˜10
MI
Ž 16 .
where all the contributions are evaluated at the MB scale and the various Cidiag
summarize all the contributions coming from graphs including SUSY Higgs bosons and
sparticles in the limit in which we neglect all the mass insertion contributions Žthey
would be the only SUSY diagrams if the scalar mass matrices were diagonalized by the
same rotations as those needed by the fermions.. The explicit expressions for Cidiag can
be found in Ref. w26x.
™
Fig. 1. Some of the relevant penguin diagrams for b s l q l y. The bubble indicates a mass insertion.
Diagrams A,B are based on chargino interaction. Diagrams Žc. and Žd. consider gluino interactions.
™
The Feynman diagrams with MI relevant for b s l q l y are drawn in Figs. 1 and 2.
We have considered gluino-like and chargino-like contributions with both single and
double mass insertions.
Both photons and Z-bosons can mediate the decay. Usually one finds that Z-boson
contributions are dominant in those graphs where an ‘‘explicit SUŽ2.L breaking’’ is
provided, i.e. both left and right squarks are present in the same loop. In the latter cases
the photon cannot feel any gauge-symmetry-breaking and its contribution to the Wilson
coefficients is suppressed by a factor m2ZrMsq2 with respect to the Z-boson one. For
Msq ; 300 GeV, this factor amounts to about an order of magnitude. On the other hand
if the graph does not give any SUŽ2.L explicit breaking we are in the opposite situation
and the Z-boson contribution is suppressed by a factor m2brm2Z ; 3 = 10y3 . Moreover, a
general feature of g-mediated four fermion contributions is that, for high average squark
masses, they decouple much faster than in the Z-boson case. This can be understood
simply using dimensional arguments. While Wilson coefficients for Z-boson mediated
four-fermion interactions are proportional to DrŽ m2Z Msq2 ., the same coefficients must be
proportional to DrMsq4 for the g Ž D here is a generic off-diagonal element of the
sparticle squared mass matrices and it cannot rise as fast as Msq2 for high values of Msq ..
Thus photon graphs can compete with Z-boson graphs if the sparticle spectrum is not
too heavy.
Fig. 2. Relevant box diagram for b ™s l q y

l . The bubble indicates a mass insertion.
Table 1
Central values of physical constants used in the phenomenological analysis
mt 173.8 GeV
mb 4.8 GeV
mc 1.4 GeV
ms 125 MeV
MB 5.27 GeV
a sŽ mZ . 0.119
1r a e l Ž m Z . 128.9
sin2u W 0.2334
Finally the value of the physical constants we use is reported in Table 1.

In the following subsections we describe in detail the contributions of each graph of
Figs. 1 and 2.
3. Chargino interactions
In the weak eigenstate basis the chargino mass matrix is given by
M2 '2 MW sin b
Mx s
ž '2 MW cos b m / ,
where the index 1 of rows and columns refers to the wino state, the index 2 to the
Ž 17 .
higgsino state, m is the Higgs quadratic coupling and M2 the soft SUSY breaking wino
mass. In order to define the mass eigenstates the unitary matrices U and V which
diagonalize Mx are introduced,
diag Ž Mx 1 , Mx 2 . s U ) Mx Vq. Ž 18 .
After the rotation to mass eigenstates it is always possible to speak of wino–quark–squark
or higgsino–quark–squark interactions. In order to identify the wino and higgsino states
from the chargino ones it is sufficient to pick up the right elements from the U and V
matrices. To be clear we write them explicitly for the cases of interest in the super-CKM
basis. The wino–quark–squark, d LWu, ˜ ˜ vertex is Ž d and u˜ are a generic down-quark and
up-squark .
ig 2 Ý K k)j Ý Va1 d j L x a u˜ k L q h.c., Ž 19 .
j, ks1,2,3 as1,2
˜ ˜ vertex is
and the higgsino–quark–squark, dHu,
i Ý K k)j Ý Ž lu . k Va2 d j L xa u˜ k R y Ž ld . k Ua2) d j R xa u˜ k L q h.c., Ž 20 .
j, ks1,2,3 as1,2
where K is the CKM matrix,
u
'2 M u M u g2
g 2 s ersin u W , l s s
Õsin b '2 sin b MW
and
'2 M d
ld s
Õcos b
are the Yukawa matrices for the up- and down-quarks.
Chargino graphs can contribute to the decay via both single and double insertions
Žsee Figs. 1 and 2.. The double insertion is particularly convenient if the corresponding
d ’s are not very constrained w41x. In the following subsections we examine both cases.
In the case of a single insertion approximation both g- and Z-mediated decays are
considered.
In all what follows our results for the integrals are written in terms of the functions
j
1 1 yiŽ1yy.
Pi jk Ž a,b . ' H0 H0dx dy k
. Ž 21 .
Ž 1 y y q axy q b Ž 1 y x . y .
To get a feeling with numbers it is sufficient to say that for a s b s 1,
Pi jk Ž 1,1 . s bE w 1 q i ,1 q j x , Ž 22 .
where bE is the Euler b-function.
3.1. Single mass insertion – Z
The Z-boson mediated decay can proceed in two ways depending on the type of
chargino–quark–squark vertices we consider. If an explicit SUŽ2.L breaking on the
squark line of Fig. 1a is required we must take both an higgsino and a wino vertex. In
this way we get a contribution to the Wilson coefficients
C9 u
l t K c)s 1
y 2
s C10 s Ž d 23 . LR
1 y 4sin u W g 2 K t)s 4sin2u W
= Ý
i , js1,2
½ (
Vi1Vj2) Ui1) Uj1 x i x j P112 Ž x i , x j . q Vi1) Vj1 P111 Ž x i , x j .
1
y d i j P021 Ž x i , x j . ,
2 5 Ž 23 .
where x i s Mx2irMsq2 . This diagram, however, is exactly null in the limit in which U, V
approximate the identity matrix and so it is negligible for high M2 .
With two wino–quark–squark vertices we obtain
C9
y s C10
1 y 4sin2u W
u
K c)s 1
s yŽ d 23 . LL
K t)s 4sin2u W
Ý
i , js1,2
½ (
Vi1Vj1) Ui1) Uj1 x i x j P112 Ž x i , x j .
qVi1) Vj1 P111 Ž x i , x j . y d i j P021 Ž x i , x j . , 5 Ž 24 .
where we have retained only the contribution which arises because of the explicit
SUŽ2.L breaking Žwith a double wino–higgsino mixing in the wino line.; in fact Eq. Ž24.
is null in the limit of a diagonal chargino mass matrix.
Graphs with two higgsino–quark–squark vertices are suppressed with respect to these
ones by Yukawa or CKM factors.
3.2. Single mass insertion – g
The contributions of the g penguin with two wino vertices are
MW2 1 K c)s
u
C 7 s y Ž d 23 . LL Ý Vi1Vi1) 32 P222 Ž x i , x i . q P132 Ž x i , x i . 4 ,
Msq2 3 K t)s is1,2
u
MW2 2 K c)s
C9 s y Ž d 23 . LL Ý Vi1Vi1) P312 Ž x i , x i .
Msq2 3 K t)s is1,2
q 13 P042 Ž x i , x i . q x i P313 Ž x i , x i . 4 ,
MW2 1 K c)s m s
CX7 s y Ž d 23
u
. LL Ý Vi1Vi1) 32 P222 Ž x i , x i . q P132 Ž x i , x i . 4 . Ž 25 .
Msq2 3 K t)s m b is1,2
The contributions of the g penguin with an higgsino and a wino vertices are
MW2 K c)s lt
C7 s Ý Vi2) Vi1 12 P222 Ž x i , x i . q 13 P132 Ž x i , x i . 4 Ž d 23u . L R
Msq2 K t)s is1,2 g2
Mx i l b
qUi2) Vi1 P212 Ž x i , x i . q 23 P122 Ž x i , x i . 4 Ž d 23u . L L ,
mb g2
u
MW2 2 l t K c)s
C9 s Ž d 23 . LR Ý Vi2) Vi1 P312 Ž x i , x i .
Msq2 3 g 2 K t)s is1,2
q 13 P042 Ž x i , x i . q x i P313 Ž x i , x i . 4 ,
MW2 1 l t K c)s m s
CX7 s Ž d 23
u
. LR Ý Vi2) Vi1 32 P222 Ž x i , x i . q P132 Ž x i , x i . 4 .
Msq2 3 g 2 K t)s m b is1,2
Ž 26 .
3.3. Single mass insertion – box
Finally we compute the contributions which come from chargino box diagrams of
Fig. 2.
In the wino exchange case the result is
u
K c)s MW2 1
C9 s yC10 s Ž d 23 . LL Ý Ž Vi1) Vj1Vi1Vj1) . f Ž x i , x j , xn˜ . , Ž 27.
K t)s Msq2 sin2u W i , js1,2
where
f Ž x i , x j , xñ .
2
1 1 1 1 yz Ž 1 y z .
s
2
H0 dxH0 dyH0 dz 2
y Ž 1 y z . q xn˜ Ž 1 y y . Ž 1 y z . q z Ž x i x q x j Ž 1 y x . .
Ž 28 .
and xn˜ s Mn˜2rMsq2 .
If the wino–bottom–stop vertex is replaced by a higgsino–bottom–stop one we
obtain
u
K c)s MW2 lt
C9 s yC10 s y Ž d 23 . LR Ý Ž Vi1) Vj1Vi1Vj2) . f Ž x i , x j , xn˜ . .
K t)s Msq2 g 2 sin2u W i , js1,2
Ž 29 .
3.4. Double mass insertion – Z
It was recently pointed out w41x that a double mass insertion can provide a great
enhancement of the SUSY contribution to the decay width, at least in the K-system case,
if the d ’s are not very constrained.
For B decay we obtain contributions from this graph to C9 and C10 ,
C9
y s C10
1 y 4sin2u W
Ž d 23u . L R Ž d 33u . L R K c)s

sy
4sin2u W K t)s
Ý
i , js1,2
½ (
Vi1Vj1) Ui1) Uj1 x i x j P123 Ž x i , x j .
di j
q 12 Vi1) Vj1 P122 Ž x i , x j . y
3 5
P032 Ž x i , x j . . Ž 30 .
4. Gluino interactions
The main contribution of this kind of interactions comes from the graphs drawn in
Fig. 1c,d. In what follows we analyze the single and double mass insertion cases.
4.1. Single mass insertion – g
The corrections to the coefficients in the photon mediated decay case are
'2 1 Nc2 y 1 pa s ms 1
C7 s
Msq2 GF 3 2 Nc K t)s K t b ž Ž d 23d . L L q Ž d 23d . R R m
b
/ 4 P132 Ž x , x .
d
Mgl
q Ž d 23 . R L P122 Ž x , x . ,
mb
'2 1 Nc2 y 1 pa s ms 1
C 7X s
Msq2 GF 3 2 Nc K t)s K t b ž Ž d 23d . R R q Ž d 23d . L L m
b
/ 4 P132 Ž x , x .
d
Mgl
q Ž d 23 . L R P122 Ž x , x . ,
mb
'2 1 Nc2 y 1 pa s 1 d
C9 s y P042 Ž x , x . Ž d 23 . LL,
Msq2 GF 3 2 Nc K t)s K t b 3
'2 1 Nc2 y 1 pa s 1 d
C9X s y P042 Ž x , x . Ž d 23 . RR. Ž 31 .
Msq2 GF 3 2 Nc K t)s K t b 3
The term proportional to the gluino mass in Eq. Ž31. seems to be strongly enhanced
™
with respect to the others. However, the mass insertion which enters the diagram is also
strongly constrained from b sg w33x.
4.2. Single mass insertion – Z
The only relevant contributions to the Z-boson mediated decay width come from
diagrams in which the Z feels directly the breaking of SUŽ2.L . According to the
argument of Section 2 all the diagrams that do not respect this condition are suppressed
with respect to the photon mediated ones and can be neglected. However, for penguins
containing a gluino, an explicit SUŽ2.L breaking can be provided only with a double MI.
If only one MI is considered, Z-mediated decays are completely negligible with respect
to the g-mediated ones.
4.3. Double mass insertion – Z
For completeness we report here also the result obtained performing a double mass
insertion in the gluino penguin,
C9 Ž d 33d . L R Ž d 23d . R L Nc2 y 1 a s
y s C10 s P032 Ž x , x . ,
1 y 4sin2u W K t b K t)s 2 Nc 12 a
C9X X Ž d 33d . R L Ž d 23d . L R Nc2 y 1 a s

y s C10 s P122 Ž x , x . . Ž 32 .
1 y 4sin2u W K t b K t)s 2 Nc 12 a
5. Light t˜R effects
In the mass insertion approximation framework we assume that all the diagonal
entries of the scalar mass matrices are degenerate and that the off diagonal ones are
sufficiently small. In this context we expect all the squark masses to lie in a small region
around an average mass which we have chosen not smaller than 250 GeV. Actually
there is the possibility for the t˜R to be much lighter; in fact the lower bound on its mass
is about 70 GeV. For this reason it is natural to wonder how good the MIA is when a t˜R
explicitly runs in a loop.
Among those we have computed, the diagrams interesting in this effect are the
u .
chargino penguins and box with the Ž d 23 L R insertion. To compute the light-t˜R contribu-
tion we adopt the approach presented in Ref. w55x. There the authors consider an
expansion valid for unequal diagonal entries which gives exactly the MIA in the limit of
complete degeneration.
The new expressions for the contributions to the coefficients C9 and C10 are the
following.
Ø Chargino Z–penguin with both an higgsino and a wino vertex:

C9 u
l t K c)s 1
y 2
s C10 s Ž d 23 . LR
1 y 4sin u W g 2 K t)s 4sin2u W
= Ý
i , js1,2
½ (
Vi1Vj2) yUi1) Uj1 x i x j j Ž x i , x j , x t˜R .
q 12 Vi1) Vj1 k Ž x i , x j , x t˜R . y 12 d i j k Ž x i , x t˜R ,1 . ,

5 Ž 33 .
where x t˜R s Ž m t˜R rMsq . 2 and the functions jŽ x, y, z . and k Ž x, y, z . can be found in
Ref. w55x.
Ø Chargino g –penguin with both an higgsino and a wino vertex:
u
MW2 2 l t K c)s
C9 s Ž d 23 . LR Ý Vi1Vi2) g 7 Ž x i , x t˜R . ,
Msq2 3 g 2 K t)s is1,2
u
MW2 1 l t K c)s
C 7 s Ž d 23 . V V)
Ý g 1 Ž x i , x t˜R . , Ž 34 .
LR
Msq2 6 g 2 K t)s is1,2 i1 i2
where
1
fi Ž x . y f Ž x y.
y i
gi Ž x , y . s ,
1yy
52 y 153 x q 144 x 2 y 43 x 3 q Ž 36 y 54 x q 12 x 3 . log Ž x .
f7 Ž x . s 4
,
6 Ž y1 q x .
y8 q 3 x q 12 x 2 y 7x 3 q 6 x Ž y3 q 2 x . log Ž x .
f 1Ž x . s 4
. Ž 35 .
6 Ž y1 q x .
Ø Chargino box with an higgsino vertex:
C9 s yC10
1 K c)s MW2 lt
s Ž d 23u . L R Ý Ž Vi1) Vj1Vi1Vj2) . k Ž x i , x j , xn˜ , x t˜ . ,
4 K t)s Msq2 g 2 sin2u W i , js1,2
R
Ž 36 .
where k Ž x, y, z,t . is defined in Ref. w55x.
All the above formulas reduce exactly to those presented in Section 3 in the limit
x t̃ R s 1.
6. Constraints on mass insertions
In order to establish how large the SUSY contribution to B X s l q l y can be, one
can compare, coefficient per coefficient, the MI results with the SM ones taking into
™
account possible constraints on the d ’s coming from other processes.
The most relevant d ’s interested in the determination of the Wilson coefficients C 7 ,
u . Ž u. Ž u. Ž d. Ž d.
C9 and C10 are Ž d 23 L L , d 23 L R , d 33 R L , d 23 L L and d 23 L R .
Ø Vacuum stability arguments regarding the absence in the potential of color and
charge breaking minima and of directions unbounded from below w40x give
Ž di3u . L R ( m t
(2 M 2 2
u˜ q 2 M l˜
,2
mt
. Ž 37 .
Msq2 Msq
For Msq ( 300 GeV this is not an effective constraint on the mass insertions.
d,u .
Ø A constraint on Ž d 23 L L can come from the possible measure of D M B s.
d .2 Ž
In fact the gluino–box contribution to D MB s w39x is proportional to Ž d 23 L L see for
w x.
instance Ref. 39 . A possible experimental determination of D MB s, say
D MB s - 30 psy1 Ž 38 .
would imply that
Ž d 23d . L L - 0.5 Ž 39 .
for squark masses about 250 GeV. Moreover the LL up- and down-squark soft
breaking mass matrices are related by a Cabibbo–Kobayashi–Maskawa rotation
2 2
Ž Msqd . L L s K † Ž Msqu . L L K Ž 40 .
so that the limit Ž39. would be valid for the up sector too:
Ž d 23u . L L - 0.5 . Ž 41 .
™
Ø Some constraints come from the measure of B X sg . The branching ratio of this
process depends almost completely on the Wilson coefficients C 7 and C 7X which are
™
d . Ž u.
proportional respectively to Ž d 23 L R or R L and d 23 L L . The most recent CLEO esti-
mate of the branching ratio for B X s g is w29x
B exp Ž B ™ sg . s Ž 3.15 " 0.35 " 0.32 " 0.26. = 10 y4
. Ž 42 .
where the first error is statistical, the second is systematic and the third comes from
the model dependence of the signal. The limits given at 95% C.L. are w29x
2.0 = 10y4 - B exp Ž B ™sg . - 4.5 = 10y4 . Ž 43 .
We can define a C 7eff Ž MB . as
2
C 7eff Ž MB . s
Bexp Ž B ™ sg . , Ž 44 .
2
Ž K t)s K t brK cb . Ž 6 a Frp g Ž z .
where F can be found for instance in Refs. w57,58x. Considering the experimental
limits we find at 95% C.L.
0.28 - < C 7eff Ž MB . < - 0.41. Ž 45 .
2
Actually < C 7eff Ž MB .< s < C 7 Ž MB .< 2
q < CX7 Ž MB .< 2
and the constraint given in Eq. Ž45.
should be shared between the two coefficients. However, in order to get the
maximum SUSY contribution, we observe that in physical observables CX7 does not
interfere with C 7 , the CX7 C9 term is suppressed by a factor m srm b with respect to the
C 7 C9 one and CX7 C9X is numerically negligible Žin fact C9X is much smaller than C9 ..
For these reasons we choose to fulfill the constraint of Eq. Ž45. with C 7 Ž MB . alone.
The bounds Ž45. are referred to the coefficient evaluated at the MB scale while we
are interested to the limits at the much higher matching scale. After the RG evolution
has been performed we find that for an average squark mass lower than 1 TeV, the
MIA contribution alone with a suitable choice of d ’s, can always fit the experimental
constraints.
Thus, since we are interested in computing the maximum enhancement Žsuppression.
SUSY can provide, we can choose the total C 7eff Ž MB . anywhere inside the allowed
region given in Eq. Ž45. still remaining consistent with the MIA.
d . y2
The limit we get for Ž d 23 L R is of order 10 and this rules out Z-mediated gluino
penguins contributions to C9 and C10 .
u .
For what concerns Ž d 23 L L we find that the constraint changes significantly according
eff Ž
to the sign of C 7 MB .. In this case it is important to consider both the positive and
negative region as this delta can give a non-negligible contribution to C9 and C10 .
The limits depend on the choice of the parameters in the chargino sector; the
numerical results given below are computed for Msq , 250 GeV, m , y160 GeV,
Mñ , 50 GeV, tan b , 2 Žin Section 7 we will show that these are the conditions
under which we find the best SUSY contributions.. Considering the positive interval
u . <Ž u . <
we find y0.7 - Ž d 23 L L - y0.5 while in the negative one d 23 L L - 0.1.
Ø Finally a comment on the d ’s coming in graphs with a double MI is in order.
d .
Given the constraints on Ž d 23 L R one can see that the gluino-penguins with a double
d .
MI give negligible contributions to the final results even if Ž d 33 R L is of order O 1 .
Ž .
u
A Ž d 33 .R L of order O Ž1., can give rise to light or negative squark mass eigenstates. In
particular a light t˜L would contribute too much to the r W -parameter. Eventual
model-dependent cancelation can provide an escape to these constraints. In any case
the numerical value of these contributions is not particularly important for the
determination of physical observables. Since we want to provide a model-indepen-
dent analysis we prefer not to consider in our final computation these double
insertion graphs and we present them only for completeness.
Contributions with three mass insertions are suppressed due to small loop integrals
and to the various constraints on the deltas.
7. Results
The results of the calculations of Sections 3 and 4 are presented in Figs. 3 and 4 and
in Tables 2 and 3. While the gluino sector of the theory is essentially determined by the
knowledge of the gluino mass Ži.e. Mgl ., the chargino one needs two more parameters
Ži.e. M2 , m and tan b .. Moreover it is a general feature of the models we are studying
the decoupling of the SUSY contributions in the limit of high sparticle masses: we
expect the biggest SUSY contributions to appear for such masses chosen at the lower
bound of the experimentally allowed region. On the other hand, these considerations
u . Ž u. Ž
Fig. 3. Ž d 23 L R above and d 23 L L below contributions to C 9 coming from chargino diagrams as functions
Ž . .
of m expressed in GeV . Msq is fixed to 250 GeV while tan b varies between 2 and 30.
Ž .
u . Ž u. Ž
Fig. 4. Ž d 23 L R above and d 23 L L below contributions to C10 coming from chargino diagrams as functions
Ž . .
of m Žexpressed in GeV.. Msq is fixed to 250 GeV while tan b varies between 2 and 30.
suggest us to constrain the three parameters of the chargino sector by the requirement of
the lighter eigenstate not to have a mass lower than the experimental bound of about 70
Table 2
Contributions to the coefficients C 7 , C9 and C10 from diagrams involving gluino loops. Mgl and Msq both
vary between 250 GeV and 1000 GeV. Exchanging L with R in the mass insertions we get the contributions of
X X X
gluino diagrams to C 7 , C9 and C10 . For further explanations see the caption in Table 3
Diagram Msq Mgl C7 C9
d . Ž d. d .
g˜g y1 ins 250 250 y0.192Ž d 23 L L y33.4 d 23 L R y0.513Ž d 23 LL
d . d . d .
250 500 y0.125 d 23 L L y31.2Ž d 23
Ž LR y0.189 d 23
Ž LL
d . d . d
500 500 y0.0449Ž d 23 LL y15.6 Ž d 23 L R y0.12Ž d 23 .L L
d . Ž d. d .
250 1000 y0.0344Ž d 23 L L y10.3 d 23 L R y0.0463Ž d 23 LL
d . d . d .
500 1000 y0.0291Ž d 23 LL y14.5 Ž d 23 LR y0.0439Ž d 23 LL
d . d . d .
1000 1000 y0.0105 d 23 L L y7.26 d 23
Ž Ž LR y0.0279Ž d 23 LL
C10 C9
d . Ž d . d . Ž d .
˜
gZy2 ins 250 250 y10.2Ž d 23 L R d 33 R L 0.763Ž d 23 L R d 33 R L
d . Ž d . d . Ž d .
250 500 y17.3Ž d 23 L R d 33 R L 1.29Ž d 23 L R d 33 R L
d . Ž d . d . Ž d .
d . Ž d . d . Ž d .
d . Ž d . d . Ž d .
d . Ž d . d . Ž d .
Table 3
Contributions to the coefficients C 7 , C9 and C10 from diagrams involving chargino loops. We assume
m sy160 GeV, M2 s 50 GeV, tan b s 2, mn˜ s 50 GeV, Mt˜R s90 GeV while Msq varies between 250 GeV
and 1000 GeV. In the first column we indicate the number of mass insertions present in each squark line,
which charginos are present at the vertexes and the kind of graph computed Žg-penguin, Z-penguin or box
diagram.
Diagram Msq C7 C9
˜ ˜g y1 ins
WW 250 u .
0.35Ž d 23 LL
u .
1.4Ž d 23 LL
u . u .
500 0.12Ž d 23 LL 0.76Ž d 23 LL
u . u .
1000 0.033Ž d 23 LL 0.32Ž d 23 LL
˜ ˜g y1 ins
HW 250 u .
y2.1Ž d 23 Ž u.
L L y0.25 d 23 L R
u .
y0.71Ž d 23 LR
u . u . u .
500 y1.1Ž d 23 LL y0.27 Ž d 23 LR y0.87Ž d 23 LR
u . Ž u. u .
1000 y0.45Ž d 23 L L y0.27 d 23 L R y0.93Ž d 23 LR
C10 C9
˜˜
WWZy2 ins 250 u . Ž u .
1.4Ž d 23 L R d 33 R L y0.092Ž d 23u . Ž u .
L R d 33 R L
u . Ž u . u . Ž u .
500 1.8Ž d 23 L R d 33 R L y0.12Ž d 23 L R d 33 R L
u . Ž u . u . Ž u .
1000 2.1Ž d 23 L R d 33 R L y0.14Ž d 23 L R d 33 R L
˜˜
WHZy1 ins 250 y8.4Ž d 23u .
LR
u .
0.56Ž d 23 LR
u . u .
500 y11.Ž d 23 LR 0.74Ž d 23 LR
u . u .
1000 y13.Ž d 23 LR 0.84Ž d 23 LR
˜˜
WWZy1 ins 250 u .
y0.91Ž d 23 LL
u .
0.06Ž d 23 LL
u . u .
500 y0.47Ž d 23 LL 0.031Ž d 23 LL
u . u .
1000 y0.19Ž d 23 LL 0.013Ž d 23 LL
box W˜ y1 ins 250 u .
2.7Ž d 23 LL
u .
y2.7Ž d 23 LL
u . u .
500 1.3Ž d 23 LL y1.3Ž d 23 LL
u . u .
1000 0.55Ž d 23 LL y0.55Ž d 23 LL
˜ ˜ y1 ins
box HW 250 u .
y0.97Ž d 23 LR
u .
0.97Ž d 23 LR
u . u .
500 y1.1Ž d 23 LR 1.1 d 23 L R
Ž
u u
1000 y1.2Ž d 23 .L R 1.2Ž d 23 .L R
GeV w56x. The remaining two-dimensional space has yet no constraint. For these reasons
we scan the chargino parameter space by means of scatter plots for which Msq s
250 GeV, Mñ s 50 GeV, 80 GeV ( < m < ( 300 GeV and 2 ( tan b ( 30; for every choice
of these two parameters, M2 is determined imposing to the lighter eigenstate a mass of
about 70 GeV. In the plots we sum all contributions coming from different graphs
proportional to a common mass insertion Žthe actual values of the coefficients are
obtained multiplying the points in the plots by the MI..
In the tables we report the contribution of each diagram and the explicit dependence
on the mass insertion parameters. We evaluate the coefficients varying Msq and Mgl
between 250 GeV and 1 TeV. The other parameters in Table 3 are fixed from the scatter
plots in order to give the best SUSY contributions to C9 and C10 .
Thus, with m , y160, Mgl , Msq , 250 GeV, Mn˜ , 50 GeV, Mt˜R s 90 GeV, tan b ,
2 one gets
C9MI Ž MB . s y1.2 Ž d 23
u
. L L q 0.69 Ž d 23u . L R y 0.51 Ž d 23d . L L
½ MI
C10 Ž MB . s 1.75 Ž d 23u . L L y 8.25 Ž d 23u . L R .
Ž 46 .
In order to numerically compare Eq. Ž46. with the respective SM results we note that
the minimum value of Ž C9eff Ž s .. SM Ž MB . is about 4 while C10 SM
s y4.6. Thus one
deduces that SM expectations for the observables are enhanced when C9MI Ž Mb . is
MI Ž
positive. Moreover the big value of C10 MB . implies that the final total coefficient
C10 Ž MB . can have a different sign with respect to the SM estimate. As a consequence of
this, the sign of asymmetries can be the opposite of the one calculated in the SM.
The diagonal contributions to C9 , C10 introduced in Section 2, and computed in the
same range of the parameters are
C9diag Ž MB . s y0.35
½ diag
C10 Ž MB . s y0.27
Ž 47 .
The sign and the value of the coefficient C 7 has a great importance. In fact the
integral of the BR Žsee Eq. Ž5.. is dominated by the < C 7 < 2rs and C 7 C9 term for low
values of s. In the SM the interference between O 7 and O 9 is destructive and this
behavior can be easily modified in the general class of models we are dealing with.
In the following, according to the discussion of Section 6, we give the configurations
of the various d ’s for which we find the best enhancements and suppressions of the SM
expectations.
Ø Best enhancement
C 7eff Ž MB . C9MI MI
C10 u, d .
Ž d 23 LL
u .
Ž d 23 LR
R 0.41 1.5 y8.3 y0.5 0.9

A FB 0.41 0.96 y2.1 y0.5 0.15
A FB 0.28 0.96 y2.1 y0.5 0.15
It is important to note that with such choices the behavior of the asymmetries in the low
s region of the spectrum is greatly modified: the coefficients of the operators Q 7 and Q9
sum up instead of cancel each other in such a way that the asymmetries are never
negative. It is also important to stress that the asymmetries get their extremal value with
u .
a rather small Ž d 23 L R : the enhancement given here will survive possible future
constraints on this insertion.
Ø Best enhancement with C 7 - 0
C10 u, d .
Ž d 23 LL
u .
Ž d 23 LR
R, y0.41 1.5 y8.3 y0.5 0.9

A FB , A FB y0.28 0.75 0.36 y0.5 y0.15
Ø Best depression
C10 u, d .
Ž d 23 LL
u .
Ž d 23 LR
R y0.28 y1.3 5.8 0.5 y0.6

A FB , A FB 0.28 y1.5 8.3 0.5 y0.9
Fig. 5. Differential branching ratio for the decay B ™ X s l q l y. The solid line corresponds to the SM
expectation; the dashed and dotted–dashed lines correspond respectively to the SUSY best enhancement
Ž C 7eff s 0.41, C9MI s1.5, C10
MI
sy8.3. and depression Ž C 7eff sy0.28, C9MI sy1.3, C10 MI
s 5.8.; the dotted line
is the maximum enhancement obtained without changing the sign of C 7 C 7 sy0.41, C9MI s1.5, C10
Ž eff MI
s
y8.3..
The plots of BRŽ s ., A FB Ž s . and A FB Ž s . are drawn in Figs. 5–8. Here both SM and
SUSY results are shown. The discontinuity in the A FB plot at s s 0.7 corresponds to the
point at which we have stopped the corrections O Ž1rm2b .. In fact, a model-independent
description of the differential asymmetry in the region 0.7 - s - 0.93 beyond the parton
model is still lacking. Further, the peak, which occurs at s s Ž2 m crm b . 2 , 0.3, is due to
the perturbative remnant of the cc resonance.
™
The integrated BR’s and asymmetries for the decays B X s eqey and B X s mqmy
in the SM case and in the SUSY one Žwith the above choices of the parameters. are
™
summarized in Table 4. There we computed the total perturbative contributions neglect-
ing the resonances; these occur in the intermediate range of the spectrum Ž Jrc at 3.1
GeV Ž s s 0.42. and c X at 3.7 GeV Ž s s 0.59. plus others at higher energies.. However,
it is possible to exclude the resonant regions from the experimental analysis by
opportune cuts and to correct the effects of their tails in the remaining part of the
spectrum.
Fig. 6. Forward–backward asymmetry Ž A FB . for the decay B ™

X s l q l y. The solid line corresponds to the
SM expectation; the dashed and dotted-dashed line corresponds to the SUSY best enhancement Ž C 7eff s 0.41,
C9MI s 0.96, C10
MI
sy2.1. and depression Ž C 7eff s.28, C9MI sy1.5, C10 MI
s8.3.; the dotted line is the
maximum enhancement obtained without changing the sign of C 7 C 7 sy0.28, C9MI s 0.75, C10
Ž eff MI
s 0.36..
Fig. 7. Forward–backward asymmetry Ž A FB . for the decay B ™ X s mqmy. The solid line corresponds to the
SM expectation; the dashed and dotted–dashed lines correspond respectively to the SUSY best enhancement
Ž C 7eff s 0.28, C9MI s 0.96, C10
MI
sy2.1. and depression Ž C 7eff s 0.28, C9MI sy1.5, C10
MI
s8.3.; the dotted line
is the maximum enhancement obtained without changing the sign of C 7 Ž C 7eff sy0.28, C9MI s 0.75,
MI
C10 s 0.36..
The results of Table 4 must be compared with the experimental best limit which reads
w59x
BR exp - 5.8 = 10y5 . Ž 48 .
A comment on the CMSSM ŽConstrained MSSM. prediction for the observables we
have computed is now necessary. An analysis on the subject is presented in Ref. w26x. In
this paper the authors show that the effect of CMSSM on the integrated BR’s,
considering only contributions to C9 and C10 , varies between a depression up to 10%
and an enhancement of few percents relative to the corresponding SM values. The
asymmetries get even smaller corrections. On the other hand, a direct computation of
C 7MS SM Ž MW . yields w26x
y0.59 - C 7MS SM Ž MW . - q0.49 in the large tan b regime,
y0.26 - C 7MS SM Ž MW . - y0.20 in the low tan b regime. Ž 49 .
It is worth noting that comparing the above intervals with the experimentally allowed
region obtained via RG evolution at the MW scale of the limits in Eq. Ž45. Žwe use only
Fig. 8. Forward–backward asymmetry Ž A FB . for the decay Bd ™X e s

q y
e ŽSee the caption of Fig. 7..
Table 4
™ ™
Integrated BR, A FB and A FB in the SM and in a general SUSY extension of the SM for the decays
B X s eq ey and B X s mqmy. The second and third columns are the extremal values we obtain with a
positive C 7eff while the fourth one is the C 7eff - 0 case. The actual numerical inputs for the various coefficients
™
can be found in the text. The BR is just the integral of RŽ s . multiplied by the BR of the semileptonic
dominant B decay ŽBRŽ B X c en . s 0.105.
Observable SM SUSY SUSYr SUSY SUSYrSM SUSY SUSYrSM
maximal SM minimal Ž C 7 - 0.
BRŽ e . 9.6=10y6 4.4=10y5 4.6 3.9=10y6 0.41 3.9=10y5 4.0
A FB Ž e . 0.23 0.33 1.5 y0.18 y0.78 0.31 1.4
A FB Ž e . 0.071 0.24 3.3 y0.19 y2.7 0.11 1.5
BRŽ m . 6.3=10y6 4.0=10y5 6.3 1.6=10y6 0.26 3.4=10y5 5.4
A FB Ž m . 0.23 0.33 1.5 y0.18 y0.78 0.31 1.4
A FB Ž m . 0.11 0.27 2.5 y0.27 y2.4 0.15 1.3
the SM contribution to C8 ; the inclusion of the MSSM corrections does not change
significantly the result.
y0.39 - C 7 Ž MW . - y0.099 and 0.66 - C 7 Ž MW . - 0.95 Ž 50 .
it is excluded that the CMSSM could drive a positive value for C 7eff Ž MB .. For what
concerns the negative interval of values of C 7eff Ž MB . we see that it can be accommo-
dated both in the CMSSM and in our framework.
Looking at Figs. 5–8. and Table 4 we see that the differences between SM and SUSY
predictions can be remarkable. Moreover a sufficiently precise measure of BR’s, A FB ’s
and A FB ’s can either discriminate between the CMSSM and more general SUSY models
or give new constraints on mass insertions. Both these kind of information can be very
useful for model building.
8. Conclusions
™ ™™ ™
In this paper an extensive discussion about SUSY contributions to semileptonic
decays B X s eqey, B X s mqmy is provided. We see that the interplay between
b sg and B X s l q l y is fundamental in order to give an estimate of the SUSY
relevance in these decays. The two kinds of decays are in fact strongly correlated.
Given the constraints coming from the recent measure of b sg and estimating all
possible SUSY effects in the MIA framework we see that SUSY has a chance to
™
strongly enhance or depress semileptonic charmless B-decays. The expected direct
measure will give very interesting information about the SM and its possible extensions.
Acknowledgements
We thank S. Bertolini and E. Nardi for fruitful discussions. I.S. wants to thank
SISSA, for support and kind hospitality during the elaboration of the first part of this
work and Della Riccia Foundation ŽFlorence, Italy. for partial support. The work of L.S.
¨ Bildung und Forschung under

was supported by the German Bundesministerium fur
contract 06 TM 874 and by the DFG project Li 519r2-2. This work was partially
supported by INFN and by the TMR–EEC network ‘‘Beyond the Standard Model’’
Žcontract number ERBFMRX CT960090..
References
w1x See, for instance, The BaBar Physics Book, SLAC-R-504 and references therein.
w2x B. Grinstein, M.J. Savage, M.B. Wise, Nucl. Phys. B 319 Ž1989. 271.
w3x G. Buchalla, A.J. Buras, Nucl. Phys. B 400 Ž1993. 225.
w4x M. Misiak, Nucl. Phys. B 393 Ž1993. 23.
w5x M. Misiak, Nucl. Phys. B 439 Ž1995. 461, Erratum.
w6x ¨ Phys. Rev. D 52 Ž1995. 186.
A.J. Buras, M. Munz,
w7x G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 Ž1996. 1125.
w8x N.G. Deshpande, J. Trampetic, K. Panrose, Phys. Rev. D 39 Ž1989. 1461.
w9x C.S. Lim, T. Morozumi, A.I. Sanda, Phys. Lett. B 218 Ž1989. 343.
w10x A.I. Vainshtein et al., Yad. Fiz. 24 Ž1976. 820, Sov. J. Nucl. Phys. 24 Ž1976. 427.
w11x P.J. O’Donnell, H.K. Tung, Phys. Rev. D 43 Ž1991. 2076.
w12x A.F. Falk, M. Luke, M.J. Savage, Phys. Rev. D 49 Ž1994. 3367.
w13x A. Ali, G. Hiller, L.T. Handoko, T. Morozumi, Phys. Rev. D 55 Ž1997. 4105.
w14x A. Ali, G. Hiller, Phys. Rev. D 58 Ž1998. 071501.
w15x A. Ali, G. Hiller, Phys. Rev. D 58 Ž1998. 074001.
w16x G. Buchalla, G. Isidori, Nucl. Phys. B 525 Ž1998. 333.
w17x ¨
F. Kruger, L.M. Sehgal, Phys. Lett. B 380 Ž1996. 199.
w18x A. Ali, T. Mannel, T. Morozumi, Phys. Rev. B 273 Ž1991. 505.
w19x M.R. Ahmadi, Phys. Rev. D 53 Ž1996. 2843.
w20x C.D. Lu,¨ D.X. Zhang, Phys. Lett. B 397 Ž1997. 279.
w21x G. Buchalla, G. Isidori, S.J. Rey, Nucl. Phys. B 511 Ž1998. 594.
w22x J.W. Chen, G. Rupak, M.J. Savage, Phys. Rev. B 410 Ž1997. 285.
w23x S. Bertolini, F. Borzumati, A. Masiero, G. Ridolfi, Nucl. Phys. B 353 Ž1991. 591.
w24x A. Ali, G.F. Giudice, T. Mannel, Z. Phys. C 67 Ž1995. 417.
w25x J. Hewet, J.D. Wells, Phys. Rev. D 55 Ž1997. 5549.
w26x P. Cho, M. Misiak, D. Wyler, Phys. Rev. D 54 Ž1996. 3329.
w27x T. Goto, Y. Okada, Y. Shimizu, M. Tanaka, Phys. Rev. D 55 Ž1997. 4273.
w28x T. Goto, Y. Okada, Y. Shimizu, Phys. Rev. D 58 Ž1998. 094006.
w29x CLEO Collaboration, CONF98-17, ICHEP98, 1011.
w30x Y.G. Kim, P. Ko, J.S. Lee, Nucl. Phys. B 544 Ž1999. 64.
w31x S. Back, P. Ko, hep-phr9904283.
w32x L.J. Hall, V.A. Kostolecki, S. Raby, Nucl. Phys. B 267 Ž1986. 415.
w33x F. Gabbiani, A. Masiero, Nucl. Phys. B 322 Ž1989. 235.
w34x J.S. Hagelin, S. Kelley, T. Tanaka, Nucl. Phys. B 415 Ž1994. 293.
w35x E. Gabrielli, A. Masiero, L. Silvestrini, Phys. Lett. B 374 Ž1996. 80.
w36x J.A. Bagger, K.T. Matchev, R. Zhang, Phys. Lett. B 412 Ž1997. 77.
w37x M. Ciuchini et al., J. High Energy Phys. 10 Ž1998. 008.
w38x R. Contino, I. Scimemi, hep-phr9809437.
w39x F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 Ž1996. 321.
w40x J.A. Casas, S. Dimopoulos, Phys. Lett. B 387 Ž1996. 107.
w41x G. Colangelo, G. Isidori, J. High Energy Phys. 09 Ž1998. 009.
w42x N. Cabibbo, L. Maiani, Phys. Lett. B 79 Ž1978. 109.
w43x C.S. Kim, A.D. Martin, Phys. Lett. B 225 Ž1989. 186.
w44x P. Ball, V.M. Braun, Phys. Rev. D 49 Ž1994. 2472.
w45x V. Eletsky, E. Shuryak, Phys. Lett. B 206 Ž1992. 191.
w46x M. Neubert, Phys. Lett. B 322 Ž1994. 419.
w47x M. Neubert, Phys. Lett. B 389 Ž1996. 727.

w48x R. Casalbuoni et al., Phys. Rep. 281 Ž1997. 145.
w49x P. Colangelo, P. Santorelli, Phys. Lett. B 327 Ž1994. 123.
w50x P. Colangelo, F. De Fazio, P. Santorelli, E. Scrimieri, Phys. Rev. D 53 Ž1996. 3672.
w51x P. Colangelo, F. De Fazio, P. Santorelli, E. Scrimieri, Phys. Rev. D 57 Ž1998. 3186, Erratum.
w52x V.M. Belyaev, A. Khodjamirian, R. Ruckl, Z. Phys. C 60 Ž1993. 349.
w53x P. Ball, J. High Energy Phys. 9809 Ž1998. 005.
w54x M. Ladisa, G. Nardulli, P. Santorelli, Phys. Lett. B 455 Ž1999. 283.
w55x A.J. Buras, A. Romanino, L. Silvestrini, Nucl. Phys. B 520 Ž1998. 3.
w56x C. Caso et al., Eur. Phys. J. C Ž1998. 1.
w57x K. Chetyrkin, M. Misiak, M. Munz,¨ Phys. Lett. B 400 Ž1997. 206.
w58x K. Chetyrkin, M. Misiak, M. Munz,¨ Phys. Lett. B 425 Ž1998. 414.
w59x CLEO Collaboration, S. Glenn et al., Phys. Rev. Lett. 80 Ž1998. 2289.
T-duality, space-time spinors and R–R fields in curved

backgrounds
S.F. Hassan
´
Centre de Physique Theorique, Ecole Polytechnique, 91128 Palaiseau, France
Received 23 July 1999; received in revised form 19 October 1999; accepted 25 October 1999
Abstract
We obtain the T-duality transformations of space-time spinors Žthe supersymmetry transforma-

tion parameters, gravitinos and dilatinos. of type-II theories in curved backgrounds with an
isometry. The transformation of the spinor index is shown to be a consequence of the twist that
T-duality introduces between the left- and right-moving local Lorentz frames. The result is then
used to derive the T-duality action on Ramond–Ramond field strengths and potentials in a simple
way. We also discuss the massive IIA theory and, using duality, give a short derivation of
‘‘mass’’-dependent terms in the Wess–Zumino actions on the D-brane world-volumes. q 2000
PACS: 11.25.-w; 04.65.qe

Keywords: T-duality; String theory; Supergravity
1. Introduction
The action of a T-duality transformation on the string world-sheet fermions can be

studied by demanding compatibility with the N s 1 world-sheet supersymmetry. This
determines the T-duality transformation of the world-sheet spinors both in flat space w1x
as well as in the presence of background fields with an isometry along which duality is
performed w2x. The effect on extended world-sheet supersymmetry has been studied in
w3,4x, and also in w2,5x when the extended supersymmetry does not respect the isometry.
In this paper, we study the action of T-duality on space-time fermions in Type-II
superstring theories with background fields and use the results to give a simple
derivation of the R–R T-duality rules.
In flat backgrounds, the action of T-duality on space-time spinors follows in a rather
fermions c " M w
1,6–9x: A T-duality with respect to X 9 sends Eq X 9 yEq X 9 and ™
straightforward way from its action on the world-sheet currents E " X M and world-sheet
E-mail address: fawad@cpht.polytechnique.fr ŽS.F. Hassan..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 8 4 - 7
146 S.F. Hassanr Nuclear Physics B 568 (2000) 145–161
cq9 ™ ycq9 , keeping all other variables unchanged. Hence, on the left-moving half of
the world-sheet theory, it can be regarded as a parity reflection along X 9 , while the
right-moving sector remains invariant. The action of such a transformation on the
left-moving Ramond ground state is represented by V 0 s G 11 G 9 as this operator sends
G 9 of the left-moving sector to Vy1 9 9
0 G V 0 s yG , consistent with the fact that G
9
is
9
the zero-mode of cq in the Ramond sector. The action of V 0 can now be absorbed in
the space-time spinors. For example, for the parameters of space-time supersymmetry
™
e " Žwhere the subscripts ‘‘"’’ refer to the world-sheet sectors in which the supersym-
metry acts., this leads to eq V 0 eq, while ey remains unchanged. One can also obtain
the action of T-duality on the gravitinos C " M from the invariance of their vertex
™ ™ ™
operators under T-duality. CqM contains the left-moving R ground state and hence
transforms as CqM V 0CqM while CyM contains a left-moving NS field cyM and
hence Cy9 yCy9 , Cyi Cyi .
In the general case of non-flat backgrounds with an isometry, say, along X 9 ,
T-duality no longer reduces to a parity transformation acting on left-moving Žor
right-moving. world-sheet variables alone. In fact, in general, it acts as a canonical
transformation affecting both left- and right-moving sectors of the world-sheet theory
w2,10,11x. Furthermore, in curved backgrounds, the relationship between the world-sheet
fermions and space-time Dirac algebra is not as straightforward as in flat space.
Therefore, it does not seem possible to obtain the T-duality action on space-time
fermions, or equivalently, on the Ramond sector, from world-sheet considerations alone.
In this paper, we study the action of T-duality on space-time spinors in type-II string
theories in the presence of NS–NS and R–R background fields. The spinors we consider
are the space-time supersymmetry transformation parameters e ", the two gravitinos
CM " and the two dilatinos l ". These results are then used to derive the T-duality rules
for the R–R fields, including the massive IIA case. Both backgrounds and spinors are
assumed to be independent of the coordinate X 9 along which T-duality is performed
Žwith the exception of type-IIB potentials dual to massive type-IIA..
The paper is organized as follows: In Section 2, the transformation of e " is obtained
by identifying a T-duality action on the local Lorentz frame associated with the
left-moving sector of the world-sheet theory. We also describe a set of variables in terms
of which the curved-space duality resembles the flat-space case. In Section 3, we
consider space-time supersymmetry transformations in type-II theories in NS–NS
backgrounds and determine the gravitino and dilatino T-duality transformations. These
are shown to be independent of R–R backgrounds. In Section 4, we use these
transformations to derive the T-duality rules for R–R field strengths and potentials,
emphasizing the locality of potentials in the massive type-IIA case. We then use
T-duality to give a simple derivation of the ‘‘mass’’-dependent terms in the Wess–
Zumino action for D-branes in massive IIA theory. Section 5 contains the conclusions.
Many of the formulas used in this paper are given in the appendices for convenience and
to insure consistency of conventions.
2. Representation of T-duality on spinors in curved backgrounds
In this section we describe how T-duality acts on the spinorial index of space-time
fermions in type-II theories with background fields Žthe extension to other string theories
S.F. Hassanr Nuclear Physics B 568 (2000) 145–161 147
is straightforward.. This fully determines the transformation of the supersymmetry

transformation parameters e ".
The action of T-duality on massless NS–NS sector fields GM N , BM N and the dilaton
f is well known w14–16x. For later reference, we write the result here in our T-duality
conventions,
G˜ 99 s Gy1
99 ,
G˜ 9 i s yGy1
99 B9 i ,
B˜9 i s yGy1
99 G 9 i ,
G˜i j s Gi j y Gy1
99 Ž G 9 i G 9 j y B9 i B9 j . ,
B˜i j s Bi j y Gy1
99 Ž G 9 i B9 j y B9 i G 9 j . ,
2 f˜ s 2 f y lnG 99 . Ž 1.
Here, M, N are space-time indices in 10 dimensions. The backgrounds are assumed to be
independent of the X 9 coordinate along which T-duality is performed, but may depend
on the remaining coordinates which we label by X i with i s 0,1, . . . ,8. Throughout this
paper, a tilde denotes a field in the T-dual theory.
Let us decompose the 10 dimensional metric of type-II theories in terms of the
vielbeins, GM N s e M aha b e b N , where a,b are Lorentz frame indices. It is known that the
M M
T-dual theory contains two possible vielbeins that we denote by e˜Žy. a and e˜Žq. a , both
giving rise to the same T-dual metric G˜ M N w2,4,12x. Explicitly,
M M N M M N
e˜Žy. a s QyN e a , e˜Žq. a s QqN e a . Ž 2.
The matrices Q " that implement T-duality on the vielbeins are given Žalong with their
inverses. by
.G 99 .Ž G . B . 9 i .Gy1
99 yGy1
99 Ž G . B . 9 i
Q "s
0 19 0 , Qy1
" s
0 19
where 1 9 denotes the identity matrix in nine dimensions. The two vielbeins in the dual
0 , Ž 3.
theory are related by a local Lorentz transformation L a b ,

M M a
e˜Žq. ˜Žy.
bse aL b , L s ey1 Qy
y1
Qq e . Ž 4.
Using the expressions for Q ", it is easy to see that the matrix L is given by
L a b s d a b y 2Gy1 a
99 e 9 e 9 b . Ž 5.
Note that det L s y1.
The appearance of two possible vielbeins in the dual theory is not an inconsequential
ambiguity and disregarding either of them will lead to an inconsistent theory. In fact, it
forces us to augment T-duality with a local Lorentz transformation acting only on the
Lorentz frame associated with the left-moving sector of the world-sheet theory. To see
this, it is useful to regard the two vielbeins in GM N s e M a ha b e b N as the wavefunctions
associated with the left-moving and right-moving world-sheet operators, respectively,
that contribute to the graviton vertex operator. Though these vielbeins may be assigned
to different world-sheet sectors, they are identical from the point of view of space-time
geometry which does not directly see the string world-sheet. However, T-duality acts
differently on the two world-sheet sectors and one may expect it to transform the
corresponding vielbeins in different ways.1 That this is the origin of the difference
between e˜Žq. and e˜Žy. can be argued as follows: The left-moving and right-moving
world-sheet sectors are interchanged under the world-sheet parity transformation s
ys which also interchanges Qq and Qy w4x and, hence, the two vielbeins in the dual
™
M
theory. This suggests that e˜Žq. a is T-dual to the vielbein associated with the left-moving
M
sector of the original world-sheet theory, while e˜Žy. a is T-dual to the one associated
with the right-moving sector. This identification also gives a heuristic understanding of
the T-duality action Ž2. on the vielbeins: Note that in flat space, e Ma ’s appear as
M
wavefunctions for states created by Fourier modes of the world-sheet fermions c " . In
curved backgrounds, T-duality transforms these fermions to c˜ " s Q " N c " w2x, which is
M M N
consistent with the mapping of their associated wavefunctions to e˜Ž " . , as given by Ž2.,
depending on the world-sheet sector they come from.
The necessity of retaining both e˜Žq. and e˜Žy. in the dual theory is not evident if we
are dealing with bosonic fields alone. However, their presence is essential to insure the
consistency of the dual theory in the presence of space-time fermions, as will be seen in
the next section.2 This implies that we have to keep track of how the vielbeins
transform, depending on the world-sheet sector they originate in. Then, to reconcile the
results with the standard formulation of gravity with one set of vielbeins, we should use
Ž4. to re-express one of the vielbeins, say e˜Žq. in terms of the other one, i.e. e˜Žy. . In
™
other words, we have to augment the T-duality action on the left-moving vielbein by a
local Lorentz transformation, e Qq e Ly1 , so that it transforms to e˜y, rather than to
e˜q. This translates to the T-duality action on the spinor index that the left-moving
Ramond sector contributes to the space-time fields. Formulating the dual theory in terms
of e˜Žy. is natural since for self-dual backgrounds, Qy in Ž3. reduces to the identity
matrix and e˜Žy. s e without further field redefinitions ŽThough this is not the case with
e˜Žq. , choosing it will also lead to a physically equivalent description..
Consider the space-time supersymmetry transformation parameters e " and the Dirac
matrices G M s e M a G a in either IIA or IIB theory. The Majorana–Weyl spinors e "
are taken to be independent of X 9 and the subscripts ‘‘"’’ refer to their world-sheet
origin and not their space-time chirality which will depend on the theory and will be
specified later. After T-duality, we will have two possible sets of G-matrices,
G˜Žq.
M M
s e˜Žq. aG
a
, G˜Žy.
M M
s e˜Žy. a
aG . Ž 6.
Keeping track of their world-sheet origin, the spinors e " in the dual theory are
1
That T-duality could transform the vielbeins associated with the left- and right-moving world-sheet sectors
in different ways, is not evident from the transformation of the metric. This is because the T-duality action on
the metric is determined by the invariance of the energy–momentum tensor and not that of the world-sheet
Lagrangian.
2
That both vielbeins necessarily appear in the dual theory also follows from the T-duality action on
complex structures associated with extended world-sheet supersymmetry, in cases where the complex
structures could be constructed in terms of target-space Killing spinors Žfor example, in non-compact
Calabi–Yau in four dimensions w17x. as discussed in w2x.
associated with the Dirac algebras generated by G˜Ž "

M
. , respectively. The two sets of
Dirac matrices are related by
G˜Žq.
M
s Vy1G˜Žy.
M
V, with Vy1G aV s L a b G b . Ž 7.
Clearly, V is the spinorial representation of the Lorentz transformation Ž4.. The form of
V , including its normalization, can be determined by the following argument: Let us
write the L a b in Ž5. as
La b s d a b y 2 v a b , with v a b s Gy1 a
99 e 9 e 9 b . Ž 8.
a b a a
One can easily verify that v b v c s v c , so that v b is a projection operator of rank
1. The operator v s v a b Ž ErE X a . dX b projects the vector G s G a ErE X a along the
r2 a Ž
isometry generator K which, normalized to unity, is given by K s Gy1 99 e 9 ErE X a ..
The projected component of G is then given by - K , v P G ) . The transformation L a b
in Ž8. changes the sign of this component, keeping other components of G unchanged.
Therefore, its spinor representation V is obtained by multiplying the projected compo-
nent with G 11 ,
V s G 11 - K , v P G ) s Gy1 (
99 G 11 G 9 , Ž 9.
as can be directly verified using Ž7.. The sign of V is not fixed by these considerations
and its arbitrariness gives rise to different T-duality conventions as we will discuss later.
Note the appearance of G 9 s G 9 M G M rather than G 9 Žas a naive generalization from
the flat-space case may suggest. in this formula. This is related to the fact that, unlike in
flat backgrounds Žor more generally, self-dual backgrounds defined by Qys 1., T-dual-
ity now mixes E " X 9 with other coordinates E " X i when regarded as a canonical
transformation in the world-sheet theory. As will be apparent in Section 4, The factor
( Gy1
99 in Ž9. is essential for giving the correct dilaton transformation, though here its
existence was dictated by different considerations.
To write the dual theory with a single Dirac algebra basis, we express G˜Žq.M
in terms
˜ M
of GŽy. using 7 , and absorb V in a redefinition of the spinor eq, with ey remaining
Ž .
unchanged. This gives the T-duality transformation rules for the space-time supersym-
metry parameters e ", which are the simplest spinorial objects in the theory, as
e˜ys ey ,
e˜qs aŽ oyf . V eq , where aŽ oyf . s "1 . Ž 10 .
Note that e˜q and eq have opposite space-time chiralities, which is the basis of IIA-IIB
interchange under T-duality. The factor aŽ oyf . Žwith ‘‘o’’ standing for original and ‘‘ f ’’
for final . reflects the arbitrariness in the sign of V . It is used to denote aŽ AyB . when
T-duality converts an original IIA theory to a final IIB theory, and aŽ ByA. Õice Õersa.
The arbitrariness in sign allows for two distinct conventions: Consider two successive
T-duality transformations along X 9. Since V˜ s V , as can be verified using Ž1., we have
˜ s y1. If we choose the convention aŽ AyB . s aŽ ByA. , then e˜qs yeq. In fact, with
VV
this convention, all left-moving Ramond states behave in this way and T-duality squares
to Žy1. FL on the spectrum Žwhere FL is the left-moving space-time fermion number..
However, since IIA and IIB are different theories, we can also choose the alternative
convention,
aŽ AyB . s yaŽ ByA. , Ž 11 .
in which case, the T-duality operation that takes IIA to IIB is the inverse of the one that
takes IIB to IIA, and the transformation squares to q1 on the spectrum. In the
following, we use the latter convention whenever a convention is explicitly specified.
The correctness of Eqs. Ž10. will be checked in the next section when we examine the
supersymmetry variations of gravitinos and dilatinos to extract their T-duality transfor-
mations.
Unlike the flat-space case, in non-self-dual backgrounds the canonical transformation
that implements T-duality acts on both world-sheet sectors. Explicitly w2x,
c˜ "Ms Q "
M N
Nc" ,
E " X˜ M s Q "
M N i M N
N E "X q c "Ei Q " N c " . Ž 12 .
These equations are non-trivial only for E " X˜ and ˜ 9
reducing to X˜ s X and
9
c" , i i
c˜ "
i
sc" i
for the rest. However, the invariance of ey may tempt one to search for
variables in terms of which T-duality in curved space has the same form as that in
flat-space, affecting only the left-moving sector. To find such variables, note that the
matrices Q ", which have a very simple upper triangular form, may be decomposed as
y1 0
Qqs A˜q Qys A˜y
y1
ž 0 /
1 9 Aq ,
y1
Ay .
Here A˜" are the same matrices as A ", but in the dual theory. These equations admit
many solutions, all with Ai" N s d Ni while A9" M are not uniquely determined Žfor
example, A9yM s A9qM s G 9 M r G 99 .. If we define new world-sheet fermionic and
(
bosonic variables,
M
S" s AM N
" Nc" ,
M
J" s AM N j M N
" N E "X q c " Ej A " N c " ,
then the canonical transformations Ž12. implementing T-duality take the flat-background
form
J˜q9 s yJq9 , S˜q9 s ySq9 ,
with Jy9 and Sy9 unchanged Ž J " i
s E " X i and S "
i
sc" i
are trivially invariant..
However, the Lagrangian in terms of the new variables does not look any simpler which
shows the basic difference between the self-dual Ž Qys 1., and the more general
non-self-dual cases, even though the transformations can be written is a similar form.
3. Action of T-duality on gravitinos and dilatinos
In this section we will derive the transformations of the type-II superstring gravitinos
C " M Žnot to be confused with the world-sheet spinors c "M . and dilatinos l " under
T-duality, by demanding compatibility between T-duality and space-time supersymme-
try. Again, the ‘‘"’’ subscripts refer to the world-sheet sector in which the spinor index
of the fermion, i.e. its Ramond component, originates and not to its space-time chirality.
All spinors are assumed to be independent of the coordinate X 9 along which T-duality
is performed. The T-duality action on these spinors is independent of the R–R fields,
which we set to zero in this section for convenience. The case of non-zero R–R fields
will be considered in the next section.
Let us first consider the supersymmetry variations of the gravitinos C " M . With e "
as the supersymmetry transformation parameters and in the absence of R–R fields, these
are given by
1
ž
d "C " M s EM q WM"a b G a b e "q . . . ,
/ Ž 13.
4
d " C. M s 0 q . . . Ž 14.
Here, ‘‘ . . . ’’ indicates the presence of 3-spinor terms that we do not write down
explicitly, but which will be automatically accounted for in our final result. WM"ab are the
torsionful spin-connections given by
1
WM"a b s wM a b . HM a b . Ž 15 .
2
The above transformations hold in both IIA and IIB theories, depending on the chirality
of the spinors. In our conventions, in IIB, e " and hence C " M have positive chirality
while in IIA, ey, CyM have positive chirality and eq, CqM have negative chirality.
The supersymmetry transformation generated by eq Ž ey . acts on the left-moving
Žright-moving. world-sheet sector by interchanging R and NS boundary conditions.
Therefore, the supersymmetry variations d " C " M convert R-NS states into NS–NS
states and do not get modified if R–R fields are switched on. Therefore, we expect that
the gravitino T-duality rules obtained from Eq. Ž13. are independent of R–R fields. The
same argument applies to dilatino T-duality rules.
Let us now consider the gravitino supersymmetry variations in the T-dual theory.
First, note that the dual theory contains two sets of torsionful spin-connections,
a a
corresponding to the two vielbeins e˜Žy. M and e˜Žq. M given by 2 . We denote these by
Ž .
W˜Žy. M a b and W˜Žq. M a b , respectively. One can verify that
" "
y y y1 N
W̃Žy. M a b s WNa b Ž Qq . M , Ž 16 .
q q y1 N
W̃Žq. M a b s WNa b Ž Qy . M . Ž 17 .
Since we have chosen to express the T-dual theory in terms of e˜Žy. , the supersymmetry
variations d " C˜ " M in the T-dual theory should be expressed in terms of W˜Žy. "
M ab
alone,
1
d " C˜ " M s E M q W˜ Ž"y . M a b G a b e˜ "q . . . .
ž / Ž 18.
4
To determine C˜ qM in terms of CqM , note that using the relation Ž4. between e˜Žq. and
e˜Žy. , we can write W˜Žq.
q ˜q
M a b in terms of WŽy. M a b as
a a
W˜Žq.
q a b ˜q c
M b Ga s WŽy. M d Ž L
y1
. c L d b Ga b q Ž Ly1 . c EM L c b Ga b
s W˜Žy.
q a y1 b y1
M b V Ga V q 4 V E M V . Ž 19 .
i i
Now, using Eqs. Ž10. and Ž16. – Ž19., along with the fact that Q " jsd j
Ž3. and
E 9 eys 0, it is easy to see that the variation Ž13. implies the one in the dual theory Ž18.
provided
N
dy C˜ yM s dy CyN Ž Qy1
q . Mq... , Ž 20.
N
dq C˜ qM s aŽ oyf . V dq CqN Ž Qy1
y . Mq... Ž 21.
Again, ‘‘ . . . ’’ denotes 3-spinor terms.
Let us now consider the supersymmetry variations of the dilatinos l " in the absence
of R–R fields,
1 1
d " l "s ž G ME M f . G M N K HM N K e "q . . . ,
/ Ž 22.
2 12
d " l .s 0 q . . . Ž 23.
These are again valid in both IIA and IIB theories. In IIB, both dilatinos have negative
chirality, while in IIA, ly has negative chirality and lq has positive chirality.
Switching on R–R fields does not affect Eq. Ž22.. In the T-dual theory, written in terms
a
of the vielbein eŽy. ˜
M , the variations d " l " are given by
1 1
d " l˜ "s ž G˜ ŽMy . E M f˜ . G˜ MŽ yN.K H˜M N K e˜ "q . . .
/ Ž 24.
2 12
Using f˜ s f y 12 lnG 99 and
G˜Ž .
MNK ˜
. HM N K s G
MNK
HM N K . 6 Gy1 .
99 G 9 Ž W9 a b G
ab
. " 6 Gy1 i
99 G E i G 99 , Ž 25 .
along with Eq. Ž10., one can see that the supersymmetry variations Ž22. and Ž24. are
compatible provided
dy l˜ ys dy lyy Gy1
99 G 9 dy Cy9 q . . . , Ž 26.
dq l˜ qs aŽ oyf . V Ž dq lqy Gy1
99 G 9 dq Cq9 . q . . . Ž 27.
Eqs. Ž20., Ž21. and Ž26., Ž27. give the T-duality transformations of the supersymme-
try variations dC " M and dl " to linear order in spinors and receive corrections cubic in
the spinors whose presence is indicated by ‘‘ . . . ’’. From these we can read off the
T-duality transformations of the gravitinos and dilatinos, in principle, only to linear
order in the spinors. However, as we will show, the linear order result is exact and in
fact, it dictates the form of the 3-spinor corrections to the T-duality maps for the
supersymmetry variations above. Thus, for the gravitinos C " M , we have the T-duality
transformations
y1 N
C̃yM s CyN Ž Qq . M ,
N
C̃qM s aŽ oyf . VCqN Ž Qy1
y . M , Ž 28 .
and for the dilatinos l " we have the transformations
l˜ ys lyy Gy1
99 G 9 Cy9 ,
l˜ qs aŽ oyf . V Ž lqy Gy1

99 G 9 Cq9 . . Ž 29 .
Here, V is given by Ž9. and, as described below Eq. Ž10., aŽ oyf . s "1 stands for
aŽ AyB . if T-duality takes us from IIA to IIB, and for aŽ ByA. if it acts the other way
round. Setting aŽ AyB . s yaŽ ByA. insures that T-duality squares to q1 on the spectrum.
That Eqs. Ž28. and Ž29. do not receive corrections can be seen as follows: To linear
order in spinors, these equations are uniquely determined by Ž20., Ž21. and Ž26., Ž27.
thus only leaving the possibility of adding corrections cubic in the spinors. The presence
of such terms, however, can be ruled out on general grounds as they would give rise to
derivative interactions for the spinors in the dual supergravity action. To rule out, in a
more concrete way, the existence of both 3-spinor corrections, as well as corrections
proportional to R–R fields, we consider the supersymmetry variations of the NS–NS
fields GM N , BM N and f given by Eqs. ŽC.6. in the appendices. These variations contain
no R–R fields and are only bilinear in spinors. Using Ž1. along with Ž28. and Ž29. one
can easily verify that these variations are consistent with T-duality. On the other hand, if
Ž28. and Ž29. contained either 3-spinor terms or R–R-dependent terms, this would not be
the case. This establishes that the spinor T-duality rules given above are exact. Note that
the NS–NS supersymmetry variations ŽC.6. are insensitive to the multiplicative factor
V . Therefore, while they can be used to rule out additional additive contributions to Ž28.
and Ž29., they cannot be used to infer the existence of V in these transformations.
For supersymmetric backgrounds, when the fermionic backgrounds C " and l "
along with their supersymmetry variations are set to zero, Eqs. Ž13. and Ž22. reduce to
the string theoretic Killing spinor equations for e ". Eqs. Ž28. and Ž29. are then trivial
for the background spinors, but can be used to obtain the T-duality transformation of the
fermionic excitations around supersymmetric backgrounds. In some cases, when the
Killing spinor itself does not transform Žas is the case with ey ., the compatibility of the
Killing spinor equation with T-duality was investigated in w12,13x.
4. R–R T-duality revisited
As shown above, the T-duality rules for space-time fermions do not depend on the
R–R fields. In this section we use these rules, along with the requirement of compatibil-
ity of T-duality with space-time supersymmetry, to determine the T-duality rules for
R–R fields and discuss some related issues. Most of the results in this section are not
new but are re-derived here in a unified and more convenient way. The IIArIIB
T-duality rules for R–R fields were derived in w18,19x by studying the supergravity
action and equations of motion Žalso see Ref. w20x. and in w21–23x by dimensional
reduction of the Wess–Zumino term in the D-brane world-volume action, both consider-
ing the bosonic sector alone. Our derivation of these rules here emphasizes the
compatibility of the T-duality conventions used for the R–R fields with those used for
the spinors. The T-duality rules relating IIB to the massive IIA theory were obtained in
w19,21,22x. Here we re-derive these rules for generic configurations, emphasizing how
potential non-localities in the T-duality rules for R–R potentials are avoided. We also
present a simple derivation of the ‘‘mass’’-dependent terms in the Wess–Zumino action
for ‘‘massive’’ IIA branes using T-duality.
In the presence of R–R backgrounds, the supersymmetry variations dq CqM and
dy CyM are still given by Ž13., while dy CqM and dq CyM are no longer zero and
receive contributions from R–R fields. The same is true for the dilatino variations
d " l .. The T-duality rules for the R–R fields can be obtained by considering any one of
these variations, say dy CqM . In type IIA theory, this variation is given by w24x Žsee
Appendix C for details.
1 1 1
dy CqM s e f F Ž 0. q G M 1 M 2 FMŽ2.1 M 2 q G M 1 M 2 M 3 M 4 FMŽ4.1 M 2 M 3 M 4 GM eyq . . . ,
8 2! 4!
Ž 30 .
where ‘‘ . . . ’’ denote 3-spinor terms as usual. F Ž0. s m is the mass parameter of massive
type-IIA theory and the field strengths F Ž n. for the massive theory are given by ŽC.3. in
the appendices. The usual massless IIA equations are obtained by setting m s 0. In
type-IIB theory the corresponding variation is given by w25x Žsee Appendix B for details.
1 1
dy CqM s y e f G M 1 F MŽ 11. q G M 1 M 2 M 3 F MŽ 31.M 2 M 3
8 3!
1
q G M 1 M 2 M 3 M 4 M 5 F MŽ 51.M 2 M 3 M 4 M 5 GM eyq . . . Ž 31.
2 Ž 5! .
It is convenient to write these two equations in the generic form
n
1 Ž y1.
dy CqM s ef Ý G M 1 . . . M n F MŽ n1.. . . M n GM eyq . . . Ž 32.
2 Ž 8. n n!
In exactly the same way as for the R–R vertex operator in flat space Žsee, for example,
Refs. w6–9x., the actual content of the above equation is determined by the chirality of
the space-time spinors: In type-IIB theory, both ey and CqM have positive chirality
and therefore the right-hand side contains only terms with even number of G-matrices
Žcorresponding to n s 1,3,5,7,9., whereas in IIA, ey and CqM have positive and
negative chiralities respectively and hence only terms with even n Ž n s 0,2,4,6,8,10.
enter the summation. Furthermore, using the G-matrix identity ŽA.2., the positive
chirality of ey implies that F Ž n. s yŽy1. nŽ ny1.r2 ) F Ž10yn.. This allows us to write the
summation in terms of F Ž n. with n ( 5 alone, recovering Ž30. and Ž31..
Let us now consider the above equation in the T-dual theory expressed in terms of the
a
vielbein e˜Žy. M,
n
1 ˜ Ž y1.
dy C˜ qM s ef Ý G˜ŽMy1.. . . M n F˜ MŽ n1.. . . M n G˜Žy. M e˜yq . . . Ž 33.
2 Ž 8. n n!
Using Eqs. Ž2., Ž6., Ž10. and Ž28., one can readily obtain the T-duality transformation for
the R–R field strengths as
n N1 Nn
F̃MŽ n.1 . . . M n s Ž y1 . aŽ oyf . Ž F9Ž nq1 . Ž ny1. y1
N1 . . . N n q nG 9w N1 FN 2 . . . N n x . Ž Qy . M1
y1
. . . Ž Qy . Mn .
Ž 34 .
where, aŽ oyf . denotes a convention-dependent sign as explained below Eq. Ž10.. Let us
now choose the convention Ž11. so that T-duality squares to 1 on R–R fields. Then,
using the form of Qy1y given in 3 , the above equation reduces to the component form
Ž .
F̃9Ž in.2 . . . i n s yaŽ AyB . F Ži 2ny1 . y1 Ž ny1.
. . . i n y Ž n y 1 . G 99 G 9w i 2 F9 i 3 . . . i n x , Ž 35 .
F˜iŽ1n. Ž nq1. ˜ Ž n.
i 2 . . . i n s yaŽ AyB . F9 i 1 . . . i n y nB9w i 1 F9 i 2 . . . i n x . Ž 36 .
aŽ AyB . is still arbitrary and could be chosen as either q1 or y1. The antisymmetriza-
tion denoted by the square bracket affects the indices i n and not the index 9. Since the
spinors were assumed to be independent of X 9 , Eq. Ž32. implies that F Ž n. should also
be independent of this coordinate.
The above T-duality rules for F Ž n. are valid for both massless and massive type-IIA
theories and can be iteratively integrated to yield the corresponding transformations for
the R–R potentials C Ž n.. Let us first consider duality between IIB and massless IIA. In
this case, F Ž0. s m s 0 and the field strengths are given by ŽB.10.. C Ž n. can be chosen to
be X 9-independent and under T-duality transform as w18–20,22,23x
C̃9Ž n. Ž ny1 . y1 Ž ny1.
i 2 . . . i n s aŽ AyB . C i 2 . . . i n y Ž n y 1 . G 99 G 9w i 2 C 9 i 3 . . . i n x , Ž 37 .
C˜iŽ1n.i 2 . . . i n s aŽ AyB . C9Ž nq1. ˜Ž n.
i 1 . . . i n y nB9w i 1C 9 i 2 . . . i n x . Ž 38 .
Let us now consider the massive-IIA case. For n s 0, Eq. Ž36. reduces to F˜ s Ž0.
yaŽ AyB . F Ž1. s yaŽ AyB . E 9 C Ž0.. As noticed in w19,21,22x, this implies that type-IIB
theory dualizes to the massive IIA theory with F˜ Ž0. s m, provided the IIB 0-form has an
X 9 dependence given by C Ž0. s yaŽ AyB . mX 9 q Cˆ Ž0., where the last term is indepen-
dent of X 9. Naively, one may expect that this X 9-dependence could lead to a similar
dependence for the IIA potentials, which should not be the case: Consider an X 9-depen-
dent function C Ž X 9 ., say, in the IIB theory leading to an X 9-dependent T-dual C˜Ž X 9 .
in IIA. Since the natural variable in the T-dual theory is X˜ 9 , which is related to X 9
through the canonical transformation Ž12., C˜ has to be expressed in terms of X˜ 9.
However, the relationship between X 9 and X˜ 9 is non-local, involving an integration
over the string world-sheet, and hence C˜ is a non-local function of X˜ 9. This problem
can be avoided if we arrange things such that the X 9-dependent C dualizes to an
X 9-independent C,˜ or Õice Õersa. Let us define
Cˆ Ž0. s C Ž0. q aŽ AyB . mX 9 ,
CˆMŽ2.1 M 2 s CMŽ2.1 M 2 q aŽ AyB . mX 9 BM 1 M 2 ,
CˆMŽ4.1 . . . M 4 s CMŽ4.1 . . . M 4 q 3aŽ AyB . mX 9 Bw M 1 M 2 BM 3 M 4 x ,
or more generally, using the notation of w22x, with C s Ý9ns 0 C Ž n.,
Cˆ s C q aŽ AyB . m X 9 e B . Ž 39 .
We give the C Ž2 p. in type-IIB a dependence on X 9 in such a way that Cˆ Ž2 p. are X 9
independent, while in type-IIA, Cˆ Ž2 pq1. s C Ž2 pq1. and are X 9 independent. Then, using
the T-duality rules for the field strengths Ž35., Ž36., along with Eqs. ŽB.10. for type-IIB
and ŽC.3. for the massive type-IIA, one can obtain the T-duality rules for the potentials.
These are still given by Ž37. and Ž38. but now with all C Ž2 p. replaced by Cˆ Ž2 p.. The X 9
independence of Cˆ Ž2 p. guarantees the X 9 independence of the IIA potentials C Ž2 pq1.,
preventing the appearance of non-localities. Note that while the massive T-duality rules,
written in terms of Cˆ Ž2 p., have the same form as the usual massless IIArIIB rules, the
two differ by m-dependent terms when written in terms of the actual R–R potentials
C Ž2 p..
The special X 9 dependence of C Ž2 p. can be easily understood when massive-IIArIIB
duality is regarded as a Scherk–Schwarz compactification to nine dimensions w19x: The
UŽ1. ; SLŽ2, R . transformation in IIB theory that gives the right X 9 dependence to C Ž0.,
by shifting it to C Ž0. y aŽ AyB . mX 9 Žcorresponding to p s s s 1, r s 0 and q s
yaŽ AyB . mX 9 in ŽB.5.., also produces the correct X 9 dependences in C Ž2. and C Ž4..
For m / 0, the Wess–Zumino terms in the IIA D-brane world-volume actions contain
m-dependent terms the forms of which were studied in w21,22x. We will now derive
these terms in a very straightforward way using T-duality: Let us start with the WZ
terms in the D-brane world-volume actions in type-IIB theory and express the potentials
C Ž2 p. in terms of Cˆ Ž2 p. as defined in Ž39.,
ŽIIB.
I WZ s Hv C e FyB s Hv Cˆ e Fy B y aŽ AyB . m
Hv X 9 eF . Ž 40 .
2p 2p 2p
When C Ž2 p. are chosen such that Cˆ Ž2 p. are X 9 independent, the dual theory is massive
ŽIIB.
type-IIA. Therefore, on dimensional reduction, I WZ should reduce to the corresponding
action for massive IIA theory, including the m-dependent terms. As mentioned earlier,
the massive T-duality rules relating Cˆ Ž2 p. and C Ž2 pq1. have exactly the same form as the
massless T-duality rules relating C Ž2 p. and C Ž2 pq1.. Therefore, the analysis for the
massless case, for example, as presented in Ref. w23x or Ref. w22x, implies that the first
term on the right-hand side of Ž40. dualizes to the standard WZ term in type-IIA which
is common between the massive and massless theories. The m-dependent terms are
contained in the second integral on the right-hand side of Ž40.. Let us identify X 9 with a
world-volume direction, say s , along which the theory is reduced. Taking F to be
Abelian Ž F s dV ., we write e F s Ý p p!1 dŽ V n F py1 . so that
1
H X 9 eFs Ý X 9 Es Ž V n F py1 . n d s q Ea Ž V n F py1 . n d x a ,
H
v2 p p p! v 2p
Ž 41 .
a 9
where x are the world-volume directions transverse to s . Since X does not depend
on x a , the second term in the integrand leads to a surface term and can be dropped. The
Ž2 p y 1.-form V n F py 1 in the first term now only has non-zero components along x a ,
and not along s . Finally, remembering that X 9 s s and dropping a surface term, the
integration over s leads to
1 py 1
ŽIIA .
I WZ s H C e FyB q aŽ AyB . m Ý V n Ž dV .
H , Ž 42 .
v 2 py1 p p! v 2 py1
which reproduces the m-dependent terms of Refs. w21,22x Žwe have ignored the D-brane
tension that can be easily inserted into the equations..
5. Conclusions
We have shown that, besides acting on the space-time indices of fields, T-duality also
has an action on the local Lorentz frame associated with the left-moving sector of the
world-sheet theory by twisting it with respect to the one associated with the right-mov-
ing sector. This twist translates to the T-duality action on the spinor index originating in
the left-moving Ramond sector, and fixes the T-duality action on the space-time
supersymmetry parameters. The gravitinos and dilatinos also contain an NS sector
contribution to their T-duality transformations which is obtained by demanding consis-
tency between T-duality and space-time supersymmetry. It is also shown that the
T-duality action on the spinors is independent of the R–R backgrounds. The result is
then used to re-derive the R–R T-duality rules. We discuss the case of the massive IIA
theory in more detail, showing that there exist variables in terms of which the massive
T-duality rules for the R–R potentials have the same form as the massless ones,
manifestly avoiding non-local relations between potentials. Using this, we give a simple
derivation of the ‘‘mass’’-dependent terms in the WZ actions for the associated
D-branes based on T-duality. In most part, we have explicitly retained the convention
dependence of the T-duality action on the Ramond sector. In one convention, T-duality
squares to 1, while in the other, it squares to Žy1. FL on the spectrum, where FL is the
left-moving space-time fermion number.
There are certain similarities between T-duality in flat and curved backgrounds. At
the world-sheet level, as we have shown, there exist variables in terms of which the
canonical transformation that implements T-duality in curved space, has the flat-space
form. One can also check that the T-duality rules for gravitinos and R–R fields in
curved backgrounds easily follow from their flat-space vertex operators, provided we
interpret these operators as curved space objects Žwhich, of course, is not really the
s
case.. For example, consider the gravitino emission operator ; SqsCqM cyM in flat
space. To interpret this as a curved-space expression, we define the spin-filed Sqs as an
s
operator that generates space-time supersymmetry transformations of CqM with param-
s
eter eq , but now in curved-space. Sq and eq will have opposite space-time chiralities
and Sq eq is invariant under T-duality. Then using the curved-space T-duality rules for
eq Ž10. and cyM Ž12. in the flat-space vertex operator, we recover the T-duality action
Ž28. on CqM . Similarly, the R–R T-duality rules can be obtained from the correspond-
X X
ing flat-space vertex operator, e f Sqs F s s SysX , where F s s is the R–R bi-spinor.
Note added
In a recent paper w26x, which appeared after this paper was completed, the authors
consider the SO Ž d,d,Z . action on R–R fields from the point of view of low-energy
effective action. One should be able to obtain the same results in our approach, after
determining the SO Ž d,d,Z . action on gravitinos, and then using space-time supersym-
metry. The results are expected to look the same as the single T-duality case with Q "
and V appropriately generalized to SO Ž d,d ..
Acknowledgements
¨
I would like to thank C. Angelantonj, I. Antoniadis, A. Dhar, K. Forger, D. Ghoshal,
D. Jatkar, B. Pioline, A. Sagnotti, and A. Sen for many useful discussions during the
course of this work.
Appendix A. G-matrix conventions
We use the metric signature y1,q 1, . . . ,q 14 and Gamma matrix conventions

G a , G b 4 s 2h a b , Ž A.1 .
a 0
so that in the Majorana–Weyl representation all G are real, with G antisymmetric
and others symmetric. We also need the identity Žwith e 01 . . . 9 s 1.
n Ž ny1 .r2
M1 . . . M n
y Ž y1 . M 1 . . . M 10
G G 11 s
'y G Ž 10 y n . ! e GM nq 1 . . . M 10 . Ž A.2 .
Appendix B. Type-IIB supergravity
For the gravitino and dilatino supersymmetry variations in type-IIB supergravity, we

start with the SUŽ1,1. invariant formulation of the theory in w25x. Using a prime to
indicate the use of the Einstein metric and after scaling the 5-from field strength to
match the standard string theory conventions, we have
i
dlX s i G X M PM e X ) y G X K L NGK L N e X q . . . , Ž B.1 .
24
1
dCMX s DM e X q Ž GMX K L NGK L N y 9G X L NGM L N . e X )
96
i
q G X K L N P QGMX FK L N P Q e X q . . . Ž B.2 .
4 Ž 480 .
Here, e X , CMX and lX are complex Weyl spinors with G 11 e X s e X , G 11C X s C X , while lX
has negative chirality, and
1 i
DM e X s E M q ž wXM a b G a b y QM e X ,/ GK L N s yea b Vqa FKbL N ,
4 2
PM s yea b Vqa E M Vqb , Q M s yi ea b Vya E M Vqb . Ž B.3 .
a
a , b s 1,2 are SUŽ1,1. indices, V " is an SUŽ1,1. matrix Ž Vy1 Vq2y Vq1 Vy2s 1. and
1 2)
FK L N s FK L N . To identify the fields in the usual string theory conventions, we go to the
SLŽ2, R . formulation by writing F a in the real basis. Then the NS–NS and R–R
2-forms BM N and CMŽ2.N are given by
ydCŽ 2 . Re Ž F 1 . F 1r'2 1 1 1
ž dB
s
/ ž1
Im Ž F .
sh
F 2r'2 / ž / , with h s
'2 ž yi qi
. /
The dilaton and the R–R scalar C 0 are identified by parameterizing the matrix V such
that
Vy1 Vq1 1 yt e i u yt eyi u
U s hV ' h
ž Vy2 Vq2 / s
(2t 2
ž e iu eyi u
,
/ Ž B.4 .
with t s C Ž0. q i ey f . We can set u s 0 by fixing the UŽ1.. In these conventions, the
SLŽ2, R . action takes the form
t ™ prtt qqsq , ž CŽ 2 .
B / ™ ž pr qs / ž CB Ž 2.
/ . Ž B.5 .
C Ž4. also transforms such that F Ž5. is invariant Žsee ŽB.10... Having identified the
dilaton, we define the string frame metric and associated spinors as
GM N s e f r2 GXM N , e s e f r8e X , l s eyf r8lX ,
i
CM s e f r8 CMX q ž GMX lX ) . /
4
Furthermore, we write the complex Weyl spinors in terms of real Majorana–Weyl

spinors, e s eqq i ey, CM s CqM q iCyM , l s lyq i lq, where the subscript ‘‘"’’ is
chosen to denote the world-sheet sector that contributes the spin content of the spinor.
The supersymmetry variations ŽB.2. then take the form Žto linear order in the spinors.
1 1
d " l "s ž G ME M f . G M 1 M 2 M 3 HM 1 M 2 M 3 e "q . . . ,
/ Ž B.6 .
2 12
1 1
d. l "s e f "G M F MŽ 1 . q
ž G M 1 M 2 M 3 F MŽ 31.M 2 M 3 e .q . . . ,
/ Ž B.7 .
2 12
1 1
d "C " M s EM q ž 4 ž wM a b .
2 / /
HM a b G a b e "q . . . , Ž B.8 .
1 1
d. C " M s
8 ž
e f .G M 1 F MŽ 11. y
3!
G M 1 M 2 M 3 F MŽ 31.M 2 M 3
1
.
2 Ž 5! . /
G M 1 . . . M 5 F MŽ 51. . . . M 5 GM e .q . . . Ž B.9 .
The R–R potentials C Ž n. are defined such that
n!
FMŽ n.1 . . . M n s n E w M 1CMŽ ny1.
2 . . . Mnx
y Hw M 1 M 2 M 3 CMŽ ny3.
4 . . . Mnx
. Ž B.10 .
3! Ž n y 3 . !
Appendix C. Type-IIA supergravity
The gravitino and dilatino supersymmetry variations in type-IIA theory are given in
w24x for massive IIA. When written in terms of appropriate variables, they lead to the
usual massless IIA equations when the mass parameter is set to zero. In the standard
string theory normalizations for the fields, these equations take the form
1 1
dlX s G X ME M f y G X M N P HM N P G 11 e X
2 12
1 3
q 5 e 5f r4 F Ž 0 . y e 3 f r4G X M NF MŽ 2N. G 11
8 2!
1
q e f r4G X M N P Q F MŽ 4N. P Q e X q . . . , Ž C.1 .
4!
1 X 1 y f r2
dCMX s E M q wM ab G
ab
q e Ž GMX NP Q
y 9d w NM G X P Q x . HN P Q G 11 e X
4 96
1 1 1
q y e 5f r4GMX F Ž 0 . y e 3 f r4 Ž GMX NP
y 14d w NM G X P x . FNŽ2.P G 11
32 2 2
1 20
q e f r4 GMX
ž NP Q R
y d w NM G X P Q R x FNŽ4.P Q R e X q . . .
/ Ž C.2 .
4 3
Here, a prime indicates the use of the Einstein metric, ‘‘ . . . ’’ denote 3-spinor terms and
the field strengths F Ž n. are given by
F Ž0. s m ,
FMŽ2.N s 2 E w M CNŽ1.x q mBM N ,
FMŽ4.N P Q s 4E w M CNŽ3.P Q x y 4 Hw M N P CQŽ1.x q 3mBw M N BP Q x . Ž C.3 .

The constant m is the mass parameter of the massive type-IIA theory and the usual
massless IIA theory is recovered by setting m s 0, in which case the these equations
take the form ŽB.10. above. The string frame metric and spinors are given by
GM N s e f r2 GXM N , e s e f r8e X , l s eyf r8lX ,
1
CM s e f r8 CMX qž GMX lX . /
4
Let us consider the above equations in terms of the positive and negative chirality
components of e and other spinors. One can then see that the type-IIA theory described
in w24x is the one in which the positive chirality component of e originates in the
left-moving world-sheet sector. However, in our conventions for T-duality, we need the
IIA in which the positive chirality component of e originates in the right-moving
world-sheet sector. This IIA theory is obtained from the one described in w24x by a
world-sheet parity transformation that reverses the signs of HM N P and F Ž2., keeping F Ž0.
and F Ž4. unchanged. Then the above equations lead to
1 3 1
d. l "s e f 5F Ž 0 . "
ž G M 1 M 2 FMŽ2.1 M 2 q G M 1 M 2 M 3 M 4 FMŽ4.1 M 2 M 3 M 4 e .q . . . ,
/
8 2! 4!
Ž C.4 .
1 1 1
d. C " M s e f F Ž 0. " G M 1 M 2 FMŽ2.1 M 2 q G M 1 M 2 M 3 M 4 FMŽ4.1 M 2 M 3 M 4 GM e .q . . .
8 2! 4!
Ž C.5 .
The variations d " l " and d " C " M are still given by Eqs. ŽB.6. and ŽB.8. though now,
ey, CyM and lq have positive chirality and eq, CqM and ly have negative
space-time chirality.
Supersymmetry Õariations of NS–NS fields:
d " GM N s 2 e " GŽ MC " N . , d " BM N s "2 C " w M GN x e " , d "f s l"e " .
Ž C.6 .
Here, GM N is the string metric and Ž . denotes symmetrization with unit weight. These
equations are valid in both IIA and IIB.
References
w1x M. Dine, P. Huet, N. Seiberg, Nucl. Phys. B 322 Ž1989. 301.

w2x S.F. Hassan, Nucl. Phys. B 460 Ž1996. 362. hep-thr9504148.
w3x I. Ivanov, B.B. Kim, M. Rocek, Phys. Lett. B 343 Ž1995. 133, hep-thr9406063.
w4x S.F. Hassan, Nucl. Phys. B 454 Ž1995. 86, hep-thr9408060.
w5x I. Bakas, K. Sfetsos, Phys. Lett. B 349 Ž1995. 448. hep-thr9502065.
w6x J. Polchinski, TASI lectures on D-branes, hep-thr9611050.
w7x J. Polchinski, S. Chaudhuri, C.V. Johnson, Notes on D-branes, hep-thr9602052.
w8x C. Bachas, Lectures on D-branes, hep-thr9806199.
w9x C. Bachas, ŽHalf. a lecture on D-branes, hep-thr9701019.
w10x A. Giveon, E. Rabinovici, G. Veneziano, Nucl. Phys B 322 Ž1989. 167.
w11x E. Alvarez, L. Alvarez-Gaume, Y. Lozano, Phys. Lett. B 336 Ž1994. 183. hep-thr9406206.
w12x E. Bergshoeff, I. Entrop, R. Kallosh, Phys. Rev. D 49 Ž1994. 6663. hep-thr9401025.
w13x E. Bergshoeff, R. Kallosh, T. Ortin, Phys. Rev. D 51 Ž1995. 3009. hep-thr9410230.
w14x T. Buscher, Phys. Lett. B 159 Ž1985. 127.
w17x E. Kiritsis, C. Kounnas, D. Lust, Int. J. Mod. Phys. A 9 Ž1994. 1361. hep-thr9308124.
w18x E. Bergshoeff, C.M. Hull, T. Ortin, Nucl. Phys. B 451 Ž1995. 547. hep-thr9504081.
w19x E. Bergshoeff, M. de Roo, M.B. Green, G. Papadopoulos, P.K. Townsend, Nucl. Phys. B 470 Ž1996. 113.
hep-thr9601150.
w20x E. Eyras, B. Janssen, Y. Lozano, Nucl. Phys. B 531 Ž1998. 275. hep-thr9806169.
w21x E. Bergshoeff, M. de Roo, Phys. Lett. B 380 Ž1996. 265. hep-thr9603123.
w22x M.B. Green, C.M. Hull, P.K. Townsend, Phys. Lett. B 382 Ž1996. 65. hep-thr9604119.
w23x J. Simon, T-duality and effective D-brane actions, hep-thr9812095.
w24x L.J. Romans, Phys. Lett. B 169 Ž1986. 374.
w25x J.H. Schwarz, Nucl. Phys. B 226 Ž1983. 269.
w26x M. Fukuma, T. Oota, H. Tanaka, Comments on T-dualities of Ramond–Ramond potentials, hep-
thr9907132.
Weyl fermions on the lattice and the non-abelian gauge

anomaly
¨
Martin Luscher
Deutsches Elektronen-Synchrotron DESY, Notkestrasse 85, D-22603 Hamburg, Germany
Received 10 May 1999; accepted 8 November 1999
Abstract
Starting from the Ginsparg–Wilson relation, a general construction of chiral gauge theories on
the lattice is described. Local and global anomalies are easily discussed in this framework and a
closed expression for the effective action can be obtained. Particular attention is paid to the
non-abelian gauge anomaly, which is shown to be related to a local topological field on the lattice
representing the Chern character in 4 q 2 dimensions. q 2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.15.Ha; 11.30.Rd

Keywords: Lattice gauge theory; Weyl fermions; Gauge anomaly
1. Introduction
In abelian chiral gauge theories the gauge anomaly is proportional to the topological
charge density and its topological significance is hence relatively easy to understand. As
has recently become clear w2–7x, the same is true on the lattice if the lattice Dirac
operator D satisfies the Ginsparg–Wilson relation w1x
g 5 D q Dg 5 s aDg 5 D. Ž 1.1 .
For any value of the lattice spacing a, this identity implies an exact symmetry of the
fermion action, which may be regarded as a lattice version of the usual chiral rotations.
Moreover the axial anomaly Žwhich coincides with the gauge anomaly in the abelian
case. arises from the non-invariance of the fermion integration measure under these
transformations and can be shown to be a topological field, i.e. the associated charge
does not change under local deformations of the gauge field.
¨
E-mail address: luscher@mail.desy.de ŽM. Luscher..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 3 1 - 2
¨
M. Luscherr Nuclear Physics B 568 (2000) 162–179 163
This has now led to a construction of abelian chiral gauge theories on the lattice,
which complies with all the basic requirements including exact gauge invariance w8x. The
fermion multiplet has to be anomaly free for this to work out, but otherwise there are no
restrictions and the anomaly cancellation can be proved on the basis of the topological
nature of the anomaly alone w9x.
If the gauge group is not abelian, the gauge anomaly assumes a more complicated
form and its topological interpretation is not immediately clear. An important clue is
provided by the Stora-Zumino descent equations w10–12x, which allow one to pass from
the Chern character in 4 q 2 dimensions Žan expression proportional to the third power
of the gauge field tensor. via the Chern–Simons term in 4 q 1 dimensions to the
anomaly in 4 dimensions. It is then possible to show w13x that the anomaly is related to
the index theorem in 4 q 2 dimensions and to the existence of certain non-contractible
two-spheres in the space of gauge orbits Žfor a review and an extensive list of references
see Refs. w14,15x..
In this paper a general formulation of chiral gauge theories on the lattice is proposed.
The basic ansatz is the same as in the case of the abelian theories considered in Ref. w8x,
but there are some new elements which make the approach more transparent. In
particular, a direct connection between the gauge anomaly and a local topological field
representing the Chern character in 4 q 2 dimensions will be established. Apart from
providing an interesting link to the earlier work on the gauge anomaly in the continuum
limit, the significance of this result is that the exact cancellation of the anomaly on the
lattice is reduced to a local cohomology problem which appears to be quite tractable.
2. Lattice action and chiral projectors
In the lattice theories studied in this paper the gauge field couples to a multiplet of
left-handed fermions, which transform according to some unitary representation R of the
gauge group G. We do not impose any restrictions on R or G at this point except that G
should be a compact connected Lie group. As usual the gauge field is represented by
link variables UŽ x, m . g G, where x runs over all lattice points and m s 0, . . . ,3 labels
the lattice axes. The lattice is assumed to be finite with periodic boundary conditions in
all directions.
As already mentioned, the use of a lattice Dirac operator D satisfying the Ginsparg–
Wilson relation is a key element of the present approach to chiral gauge theories. While
the details of the definition of D are largely irrelevant, Neuberger’s operator w5,6x is an
obvious choice in this context, since it is relatively simple and has all the required
technical properties. In particular, the locality of the operator and the differentiability
with respect to the gauge field is rigorously guaranteed if the gauge field satisfies the
bound
51 y R U Ž p . 5 - e for all plaquettes p, Ž 2.1 .
where UŽ p . denotes the product of the link variables around p and e any fixed positive
number less than 301 w16x.
In the following we shall take it for granted that the gauge field action restricts the
functional integral to this set of fields. This can be achieved through a modified
164 ¨
M. Luscherr Nuclear Physics B 568 (2000) 162–179
plaquette action, for example w8x. As far as the continuum limit in the weak coupling
phase is concerned, lattice actions of this type should be in the same universality class as
the standard Wilson action, because the bound Ž2.1. constrains the gauge field fluctua-
tions at the scale of the cutoff only and does not violate any fundamental principle such
as the locality or the gauge invariance of the theory.
Chiral fields may now be defined in a natural way following the steps previously
described in Refs. w8,17–19x. One first observes that the operator gˆ 5 s g 5 Ž1 y aD .
satisfies the relations
† 2
Ž gˆ 5 . s gˆ 5 , Ž gˆ 5 . s 1, Dgˆ 5 s yg 5 D. Ž 2.2 .
The fermion action
SF s a4 Ý c Ž x . Dc Ž x . Ž 2.3 .
x
thus splits into left- and right-handed parts if the chiral projectors for fermion and
anti-fermion fields are defined through
1 1
P̂ "s Ž 1 " gˆ 5 . , P "s Ž 1 " g 5 . , Ž 2.4 .
2 2
respectively. In particular, by imposing the constraints
P̂y c s c , c Pqs c , Ž 2.5 .
the right-handed components are eliminated and one obtains a classical lattice theory
where a multiplet of left-handed Weyl fermions couples to the gauge field in a
consistent way.
An interesting point to note here is that the space of gauge fields satisfying the bound
Ž2.1. decomposes into disconnected topological sectors w20–22x. In the non-trivial
sectors the index of the lattice Dirac operator w4,7x is in general different from zero and
it turns out that the dimensions of the spaces of left-handed fermion and anti-fermion
fields are then not the same. Fermion number violating processes can thus take place,
exactly as expected from the semi-classical approximation in continuum chiral gauge
theories w8,18x.
3. Fermion integration measure
To complete the definition of the lattice theory, the functional integration measure for
left-handed fermions needs to be specified. The principal difficulty here is that the
constraint Ž2.5. depends on the gauge field. This leads to a non-trivial phase ambiguity
in the measure and eventually gives rise to the gauge anomaly.
To make this clearer let us suppose that Õj Ž x ., j s 1,2,3, . . . , is a basis of complex-
valued lattice Dirac fields such that
P̂y Õj s Õj , Ž Õk ,Õj . s d k j , Ž 3.1 .
the bracket being the obvious scalar product for such fields. The quantum field may then
be expanded according to
c Ž x . s Ý Õj Ž x . c j , Ž 3.2 .
j
¨
where the coefficients c j generate a Grassmann algebra. They represent the independent
degrees of freedom of the field and an integration measure for left-handed fermion fields
is thus given by
Dw c x s Ł d cj . Ž 3.3 .
j
Evidently if we pass to a different basis

Õ˜ j Ž x . s Ý Õ l Ž x . Ž Qy1 . l j , c˜ j s Ý Qjl c l , Ž 3.4 .
l l
the measure changes by the factor det Q which is a pure phase factor since the
transformation matrix Q is unitary.
In the following two sets of basis vectors Õj and Õ˜ j are considered to be equivalent if
they are related to each other through Eq. Ž3.4. with det Q s 1. Choosing a fermion
integration measure amounts to specifying an equivalence class of bases. A given basis
thus represents the associated measure, but it should not be confused with the measure
which is a much simpler object. In particular, any two fermion measures coincide up to
a gauge field dependent phase factor. The question of how to fix this phase will occupy
us throughout the rest of this paper. For the time being we assume that some particular
choice has been made and proceed with the definition of the theory.
In the case of the anti-fermion fields the subspace of left-handed fields is independent
of the gauge field and one can take the same orthonormal basis Õ k Ž x . for all gauge
fields. The ambiguity in the integration measure
D c s Ł d ck , c Ž x . s Ý c k Õk Ž x . Ž 3.5 .
k k
is then only a constant phase factor. Fermion expectation values of any product O of the
basic fields may now be defined through
² O : F s D w c x D c O eyS F .
H Ž 3.6 .
In the non-trivial topological sectors a constant weight factor should be included in this
formula w8x, but for brevity this factor is omitted here since we shall almost exclusively
be concerned with the vacuum sector. The fermion partition function in this sector is
²1: F s det M , Mk j s a4 Ý Õ k Ž x . DÕj Ž x . , Ž 3.7 .
x
and correlation functions of products of fermion fields may be calculated as usual by

applying Wick’s theorem, the propagator being given by
² c Ž x . c Ž y . : F s ²1: F = Pˆy S Ž x , y . Pq , DS Ž x , y . s ay4d x y . Ž 3.8 .
Full normalized expectation values are finally obtained through
1 yS G ²
²O:s
Z
HD w U x e O :F , Ž 3.9 .
where SG denotes the gauge field action, Z the partition function and DwU x the standard
integration measure for gauge fields on the lattice.
166 ¨
4. Locality condition
We are now left with the problem to fix the phase of the fermion integration measure.
Evidently this should be done in such a way that the locality of the theory is preserved
and it also seems reasonable to demand that the measure is smoothly dependent on the
gauge field. In this section these conditions are given a precise meaning and a few key
formulae are derived which will later prove useful when we discuss the gauge anomaly.
The dependence of the fermion measure on the gauge field is best studied by
considering variations1
dh U Ž x , m . s ahm Ž x . U Ž x , m . , hm Ž x . s hma Ž x . T a , Ž 4.1 .
of the link field. Requiring the measure to be smooth means that in the neighbourhood
of any given gauge field there exists a differentiable basis Õj of left-handed fields which
represents the measure in the way explained above. The change dh Õj of the basis vectors
and the linear functional
L h s i Ý Ž Õj , dh Õj . . Ž 4.2 .
j
are then well-defined. Moreover it is easy to show that L h transforms according to
L̃ h s L h y i dhln det Q Ž 4.3 .

under basis transformations Ž3.4.. Equivalent bases thus yield the same linear functional
and L h is hence a quantity associated with the measure rather than the basis vectors Õj .
Roughly speaking it tells us how the phase of the measure changes when the gauge field
is deformed.
Starting from the definition Ž3.7., the variation of the effective action is now easily
worked out and one obtains
dh ln det M s Tr dh DPˆy Dy1 Pq y i L h .

½ 5 Ž 4.4 .
The first term in this equation is the naively expected one while the second arises from
the gauge field dependence of the measure. L h is hence referred to as the measure term
in the following. Moreover, taking the linearity of L h into account, an associated
current jmŽ x . may be defined through
L h s a 4 Ý hma Ž x . jma Ž x . . Ž 4.5 .

x
ˆ in the present approach to

As will soon become clear this current plays a central role
chiral gauge theories. In particular, we shall show in Section 6 that the measure can be
reconstructed from the current under certain conditions.
Whether a euclidean field theory is local or not is usually evident from the action.
The situation here is slightly more complicated, because the fermion integration measure
1
Without loss the gauge group G may be assumed to be a subgroup of UŽ n. for some value of n. The
generators T a are then anti-hermitian matrices and the field components hma Ž x . are real.
¨
is not a product of local measures. An important point to note however is that the
effective action is the only place where the non-trivial structure of the measure shows
up. In particular, in the vacuum sector the fermion integrals ² O : F are equal to the
partition function ²1: F times a factor which is independent of the measure.
We are thus led to require that the current jmŽ x . which is induced by the fermion
measure is a local expression in the gauge field. The measure term L h then assumes the
form of a local counterterm, i.e. the interaction vertices which arise from the gauge field
dependence of the fermion measure are local. The locality of the theory is thus preserved
and the arbitrariness in the phase of the measure is greatly reduced.
5. Gauge anomaly
Another fundamental requirement on the fermion measure is that it should not break
the gauge symmetry. In particular, the effective action should be gauge-invariant. On the
lattice the group of gauge transformations is connected and it thus suffices to consider
infinitesimal gauge transformations. One should not conclude from this that there are no
global anomalies, but as will become clear later they arise in slightly different ways than
expected from the semi-classical analysis.
Infinitesimal gauge transformations are generated by lattice fields v Ž x . with values
in the Lie algebra of the gauge group. The corresponding variations of the link field are
obtained by substituting
hm Ž x . s y=m v Ž x . Ž 5.1 .
in Eq. Ž4.1., where the gauge-covariant forward difference operator =m is given by
1 y1
=m v Ž x . s U Ž x ,m . v Ž x q a m
ˆ . U Ž x ,m . yv Ž x. Ž 5.2 .
a
Žm
ˆ denotes the unit vector in direction m .. Taking the transformation behaviour of the
Dirac operator into account,
dh D s R Ž v . , D , Ž 5.3 .
the terms on the right-hand side of Eq. Ž4.4. are easily worked out and for the gauge
variation of the effective action the result
a
dh ln det M s ia4 Ý v a Ž x . A a Ž x . y =m ) jm
½ 5
Ž x. , Ž 5.4 .
x
ia
AaŽ x. s tr g 5 R Ž T a . D Ž x , x . 4 , Ž 5.5 .
2
is thus obtained. In these equations =m ) denotes the gauge-covariant backward differ-
ence operator and DŽ x, y . the kernel representing the Dirac operator in position space.
The trace is taken over Dirac and flavour indices only.
We now show that AŽ x . converges to the covariant gauge anomaly in the classical
continuum limit. The calculation is practically the same as in the case of the axial
168 ¨
anomaly which has been studied in detail in Refs. w23–25x. One begins by representing
the link field through
1
U Ž x , m . s Pexp a ½H 0
d t Am Ž x q Ž 1 y t . a m
ˆ. ,5 Ž 5.6 .
where P implies a path-ordered exponential and AmŽ x . is an arbitrary smooth gauge

potential. Using the locality and differentiability properties of the kernel DŽ x, y .
established in Ref. w16x, it is then possible to derive an asymptotic expansion
`
AaŽ x. ;
a ™ 0 ks0
Ý a ky4 Oka Ž x . . Ž 5.7 .
The fields Ok Ž x . which occur in this series are traces of RŽT a . times a polynomial of
dimension k in Rw AmŽ x .x and its derivatives. Moreover they must have the proper
transformation behaviour under the symmetries of the lattice theory.
From Eq. Ž5.5. it follows that AŽ x . is a gauge-covariant pseudo-scalar field which
changes sign when the fermion representation R is replaced by its complex conjugate.
Taking this into account, it is easy to convince oneself that all terms Ok Ž x . with
dimension k ( 3 have to be equal to zero. In the continuum limit we are then left with
the term
A a Ž x . s c1 d Ra b cemnrs Fmnb Ž x . Frsc Ž x . q O Ž a . , Ž 5.8 .

where Fmn Ž x . is the field tensor associated with the gauge potential and
d Ra b c s 2 itr R Ž T a . R Ž T b . R Ž T c . qR Ž T c . R Ž T b . 4. Ž 5.9 .
The constant c1 s y1r128p 2 does not depend on the gauge group and can be
calculated in the UŽ1. theory with a single fermion in the fundamental representation
w23–27x. Since the gauge anomaly coincides with the axial anomaly in this case, the
number may also be inferred from the index theorem w4,7,9x.
Returning to the question posed at the beginning of this section, the results obtained
above show that the effective action is gauge-invariant if Žand only if.
a
=m ) jm Ž x . sAaŽ x . . Ž 5.10 .
In other words, the phase of the fermion measure should be chosen so that the associated
current satisfies this equation. Together with Eq. Ž4.4. the gauge invariance of the
effective action moreover implies that jmŽ x . has to be gauge-covariant.
As is well-known one cannot have both, locality and gauge invariance, unless the
anomaly cancellation condition
d Ra b c s 0 Ž 5.11 .
is fulfilled. There is more than one way to prove this in the present framework, a quick
argument being that an expansion similar to Eq. Ž5.7. must exist in the case of the
current jmŽ x . too, since it is required to be local and smoothly dependent on the gauge
field. In the continuum limit the anomaly is hence equal to the divergence of a covariant
local current which is a polynomial of dimension 3 in the gauge potential and its
¨
derivatives. This is only possible if the anomaly vanishes at a s 0, i.e. if Eq. Ž5.11.
holds.
6. Integrability condition
So far we have assumed that the current jmŽ x . is obtained from a given fermion
measure through Eqs. Ž4.2. and Ž4.5.. We now show that any prescribed current
satisfying a certain integrability condition arises from a measure in this way. The
relation between the measure and the current is thus invertible and one may adopt the
point of view that the latter is the fundamental object.
To derive the integrability condition we need to study the change of phase of the
fermion measure along smooth curves
Ut Ž x , m . , 0 ( t ( 1, Ž 6.1 .
in the space of gauge fields. As discussed in Section 4, the measure term L h tells us
how the phase varies when the gauge field is deformed in a particular direction. The
total change of phase along any given curve is thus given by the Wilson line
1 y1
W s exp i ½H 0
d t Lh ,5 ahm Ž x . s E t Ut Ž x , m . Ut Ž x , m . . Ž 6.2 .
In general this phase is non-integrable, i.e. the Wilson lines around closed curves are not
necessarily equal to 1. To work this out we introduce the projector
Pt sPˆy UsU t Ž 6.3 .

and define a unitary operator Q t through the differential equation
E t Q t s w E t Pt , Pt x Q t , Q 0 s 1. Ž 6.4 .
It is easy to prove that
Pt Q t s Q t P0 Ž 6.5 .
and Q t thus transports the projector Pt along the curve. In a few lines Žappendix A. it is
then possible to establish the identity
W sdet 1yP0qP0 Q1 4 Ž 6.6 .
for all closed curves. The Wilson loops are hence the same for all fermion measures. In
other words, they represent a geometrical invariant of the measure.
Let us now assume that jma Ž x . is an arbitrary current depending smoothly on the
gauge field. L h and the associated Wilson lines W may then be defined through Eqs.
Ž4.5. and Ž6.2. respectively. Evidently, for the current to arise from a fermion measure, a
necessary condition is that Eq. Ž6.6. holds for all closed loops. This is in fact also a
sufficient condition for the existence of a such a measure. Moreover, in each topological
sector this measure is uniquely determined, up to a constant phase factor, and smooth.
To prove these statements we consider a definite topological sector and choose an
arbitrary reference field U0 in this sector. Any other field U in the same sector can then
170 ¨
be reached through a smooth curve Ut such that U1 s U. If we define Q t through Eq.

Ž6.4. as before, a basis of left-handed fields at the point U is given by
Q1w1Wy1 if j s 1,
Õj s
½ Q 1 wj otherwise,
Ž 6.7 .
where wj denotes a fixed basis at the reference point and W the Wilson line Ž6.2.
computed from the given current. This basis is path-dependent, but the associated
measure is not, because any two curves Ut and U˜t form a closed loop and the
integrability condition Ž6.6. then implies that the unitary transformation relating the
basis vectors Õj and Õ˜ j has determinant 1. Taking this into account, it is easy to show
that the fermion measure defined by the basis Ž6.7. has all the properties mentioned
above.
The construction of chiral gauge theories on the lattice is thus reduced to the problem
of finding a local current which fulfils the integrability condition Ž6.6. and the
requirement of gauge invariance. Once this is achieved, the theory is completely
specified up to a constant phase factor in each topological sector. Note that it suffices to
define the current for all gauge fields satisfying the bound Ž2.1. since only these
contribute to the functional integral.
7. Fermion determinant
Using the results obtained in the preceding section, we are now in a position to derive
a closed expression for the fermion determinant in terms of the Dirac operator and the
current jmŽ x .. We shall then be able to make contact with Kaplan’s approach to chiral
gauge theories w28,29x and the earlier work of Alvarez-Gaume´ et al. w31–33x and Ball
and Osborn w34–37x on the effective action in the continuum theory.
Suppose U0 is an arbitrary reference field in the vacuum sector and let wj be a basis
of left-handed fermion fields at this point. As explained above, the basis Ž6.7. then
represents the fermion measure at any other point U in the vacuum sector, up to a
constant phase factor. If we insert this basis in Eq. Ž3.7., the formula
det M det M0† sdet 1yPq qPq DQ1 D†0 4 Wy1 Ž 7.1 .
is obtained, where D 0 and M0 denote the Dirac operator and fermion matrix at the
reference point 2 . All other notations are as in Eq. Ž6.7.. To fully understand this result
the following remarks may be helpful.
Ža. The integrability condition guarantees that the right-hand side of the equation does
not depend on the curve Ut which has been chosen to connect U s U1 with U0 . In other
words, the path-dependence of the determinant and the Wilson line W precisely cancel
each other. Note incidentally that the constant phase ambiguity of the measure drops out
in the product of determinants.
2
Another expression for the effective action, involving the Dirac operator and the current jmŽ x . only, may
be obtained by integrating Eq. Ž4.4. along any particular path. The idea has recently been pursued by Suzuki
w39x in abelian chiral gauge theories.
¨
Žb. Using equations Ž2.2. and Ž6.5. it is easy to check that DQ1 D†0 commutes with
g 5 . The determinant of this operator in the subspace of left-handed anti-fermion fields
coincides with the determinant on the right-hand side of Eq. Ž7.1. and the chiral nature
of the expression is thus evident.
Žc. In Section 8 we shall show that the current jmŽ x . vanishes in the classical
continuum limit if the fermion multiplet is anomaly free. The Wilson line W conse-
quently does not contribute to the fermion determinant in this limit.
Žd. Concerning the operator Q t we note that the differential equation Ž6.4. may be
rewritten in the form
E t Q t s 12 a w g 5 E t Dt , Pt x Q t , Dt sD Us U t . Ž 7.2 .
Close to the classical continuum limit, and when acting on fermion fields with
frequencies far below the lattice cutoff, the operator is hence equal to 1 up to terms of
order a. In particular, in Eq. Ž7.1. the operator Q1 only affects the contribution of the
high-frequency modes and it may, therefore, be regarded as part of the lattice regulariza-
tion prescription for the chiral determinant.
Že. So far the reference field U0 has been assumed to be fixed and the factor det M0†
on the left-hand side of Eq. Ž7.1. is then just a constant. Since U0 and U1 are
interchangeable in this equation, another option is to interpret U0 as a second gauge field
and det M0† as the determinant arising from a multiplet of right-handed fermions. In the
present framework the formulation of such left–right symmetric chiral gauge theories
thus appears to be particularly natural.
Having clarified the structure of Eq. Ž7.1., we now briefly discuss how the formula
relates to Kaplan’s approach to chiral gauge theories w28,29x. In the version proposed by
Shamir w30x, this approach starts from a gauge theory in 4 q 1 dimensions with a
multiplet of massive Dirac fermions, where the additional coordinate is assumed to
range between 0 and T with Dirichlet boundary conditions on the gauge and fermion
fields. The reduction to 4 dimensions is then achieved by noting that the Dirac operator
admits chiral surface modes whose interactions at large T are described by an effective
chiral gauge theory.
Contact with the present framework can now be made if we identify the fifth
coordinate, scaled to the range w0,1x, with the parameter t of the path Ut . The path thus
becomes a gauge field in 4 q 1 dimensions with boundary values U0 and U1. In
particular, we can compare the fermion determinant in 4 q 1 dimensions with the
determinant on the right-hand side of Eq. Ž7.1. and it is then conceivable that they agree
in the limit where the lattice spacing in the fifth dimension is sent to 0 and T to infinity.
Preliminary studies suggest that this is indeed what happens if the lattice Dirac operator
in 4 q 1 dimensions is chosen appropriately, but the details are complicated and will not
be presented here. It is interesting to note, however, that from the point of view of the
higher-dimensional theory, the Wilson line W in Eq. Ž7.1. amounts to adding a local
counterterm to the gauge field action. The term cancels the dependence of the fermion
determinant on the gauge field in the interior of the space-time volume and thus allows
one to reduce the theory to 4 dimensions, the dynamically relevant degrees of freedom
being the boundary values U0 and U1.
In the continuum limit the phase of the fermion determinant is known to be
proportional to the h-invariant of the Dirac operator in 4 q 1 dimensions w31–37x. One
172 ¨
considers the massless Dirac operator in this case and uses Pauli–Villars regularization
or analytic continuation methods to define the determinant, but otherwise the setup is the
same as the one described above. The h-invariant, Kaplan’s approach and the results
obtained here are hence closely related to each other. As discussed by Kaplan and
Schmaltz w38x, the formula for the effective action of Refs. w31–37x may in fact be
directly derived from the fermion integral in 4 q 1 dimensions.
8. Classical continuum limit
To complete the construction of the lattice theory we still need to prove that there
exists a local current jmŽ x . satisfying the requirement of gauge invariance and the
integrability condition. The aim in the following lines is to determine the general
solution of this problem in the classical continuum limit. Along the way an important
simplification is achieved by considering the integrability condition in its differential
form. Global anomalies and the relation between this equation and the gauge anomaly
are further topics which will be addressed.
The differential form of the integrability condition,
dh L z y dz L h q aL wh , z x s iTr Pˆy dh Pˆy , dz Pˆy

½ 5, Ž 8.1 .
is obtained by computing the variation of the Wilson loop and the determinant in Eq.
Ž6.6. under infinitesimal deformations of the loop Ut . The equation may also be derived
in a more direct way, starting from the representation Ž4.2. of the measure term and
making use of the identity 3
dh dz y dz dh q a d wh , z x s 0 Ž 8.2 .
and the defining properties of the basis vectors Õj .
An important point to note is that the integrability condition is a slightly stronger
constraint than its differential form. To make this completely clear, let us assume that
L h is an arbitrary linear functional satisfying Eq. Ž8.1., for all gauge fields complying
with the bound Ž2.1. and all vector fields hmŽ x . and zmŽ x .. The Wilson loop associated
with any closed curve Ut of fields is then given by
W s h det 1 y P0 q P0 Q1 4 , Ž 8.3 .
where h is invariant under continuous deformations of the curve. Evidently h is equal to
1 for all contractible loops, but in general this need not be so and Eq. Ž6.6. thus imposes
an additional constraint on L h if there are topologically non-trivial loops. Presumably
the global anomalies discovered by Witten w40,41x are related to this observation since
they arise from certain non-contractible loops in the space of gauge orbits w42x. Further
studies are however required before a definite answer to this question can be given w43x.
In the classical continuum limit the differential form of the integrability condition
reduces to a simple equation. To show this we follow the steps previously described in
Section 5, i.e. we insert the representation Ž5.6. for the link variables and assume that
3
In equations Ž8.1. and Ž8.2. the variations h and z are assumed to be independent of the gauge field.
Further terms proportional to dh z and dz h have to be included if this is not the case.
¨
hmŽ x . and zmŽ x . are restrictions to the lattice of some differentiable vector fields. The
locality properties of the Dirac operator w16x then imply
`
iTr Pˆy dh Pˆy , dz Pˆy
½ 5 ;™ Ý a ky 4 Hd 4 x Ok Ž x . , Ž 8.4 .
a 0 ks0
where the fields Ok Ž x . are traces of polynomials in Rw AmŽ x .x, RwhmŽ x .x, Rw zmŽ x .x and
their derivatives. Taking the symmetries of the expression into account, this leads to the
result

½ 5 s c Hd x d
2
4 abc
R emnrs hm
a
Ž x . zn b Ž x . Frsc Ž x . q O Ž a . , Ž 8.5 .
the notations being the same as in Section 5.
The proportionality constant c 2 in this equation is related to the coefficient c1 of the
anomaly. To work this out we consider a gauge variation hmŽ x . s y=m v Ž x . and note
that
dh Pˆys R Ž v . , Pˆy , Pˆy dz Pˆy Pˆys 0. Ž 8.6 .
It is then straightforward to establish the identity

½ 4 a
5 s ya Ý v Ž x . d A Ž x . z
a
Ž 8.7 .
x
and after substituting the asymptotic forms, Eqs. Ž5.8. and Ž8.5., the relation
c 2 s y4c1 s 1r32p 2 Ž 8.8 .
is thus obtained. In passing we remark that the anomalous conservation law Ž5.10. is
consistent with the integrability condition Ž8.1. in the sense that the combination of these
equations does not lead to further constraints on the current jmŽ x . apart from the fact
that it should transform covariantly under gauge transformations.
An important conclusion which can be drawn at this point is that jmŽ x . s 0 is an
acceptable choice of the current in the classical continuum limit if the fermion multiplet
is anomaly free. Both, the requirement of gauge invariance and the integrability
condition in its differential form, are then satisfied. There is in fact no other sensible
solution since it is impossible to construct a gauge-covariant polynomial of dimension 3
in the gauge potential AmŽ x . and its derivatives which transforms as an axial vector
current. As far as the classical continuum limit is concerned, the theory is thus
completely specified up to a constant phase factor in each topological sector.
9. Equivalent cohomology problem in 4 H 2 dimensions
Most gauge field configurations which contribute to the functional integral are not as
smooth as those considered in the classical continuum limit and a general strategy to
determine the current jmŽ x . thus needs to be developed if one is interested in
constructing the complete theory. As a first step in this direction, we here show that the
anomalous conservation law Ž5.10. and the integrability condition Ž8.1. can be mapped
to a local cohomology problem whose solution is known to all orders in the lattice
spacing.
174 ¨
To explain which type of cohomology problem we are heading to, let us consider the
pure gauge theory on R n with gauge group G and suppose q Ž z . is a gauge-invariant
polynomial in the gauge potential Aa Ž z . and its derivatives. Such fields are called
topological if
n
Hd z d q Ž z . s 0 Ž 9.1 .
for any variation d Aa Ž z . of the gauge potential with compact support. Using the
Stora-Zumino descent equations w10–12x, it is possible to prove that all topological fields
are of the form
q Ž z . s c Ž z . q Ea k a Ž z . , Ž 9.2 .
where cŽ z . is a linear combinations of Chern monomials
ca 1 . . . a 2 m t a1 . . . a m Faa11a 2Ž z . . . . Faa2mmy 1a 2 mŽ z . Ž 9.3 .
and k a Ž z . a gauge-invariant local current w44–49x. The tensor ca 1 . . . a 2 m in this expres-
sion has to be totally anti-symmetric and t a1 . . . a m should be invariant under the adjoint
action of the gauge group.
The classification of topological fields modulo divergence terms is a particular case
of a local cohomology problem in which the gauge symmetry plays an important role. ˆ
From this point of view the Chern monomials represent the non-trivial cohomology
classes. Depending on the gauge group, a basis of linearly independent Chern monomi-
als is usually not difficult to find Žsee Ref. w50x for example..
Returning to the lattice, our aim in the following paragraphs is to construct a
topological field in 4 q 2 dimensions whose cohomology class is trivial if Žand only if.
there exists a local current jmŽ x . with the required properties. The added dimensions are
continuous, i.e. we are concerned with lattice gauge fields
U Ž z , m . g G, z s Ž x ,t , s . , m s 0, . . . ,3, Ž 9.4 .
which depend on two additional real coordinates t and s. We also introduce gauge
potentials A t Ž z ., A s Ž z . along these directions and define the associated field tensor
through
Ft s Ž z . s E t A s Ž z . y Es A t Ž z . q A t Ž z . , A s Ž z . . Ž 9.5 .
Under arbitrary gauge transformations in 4 q 2 dimensions, the covariant derivative
DrA U Ž z , m . s Er U Ž z , m . q A r Ž z . U Ž z , m . y U Ž z , m . A r Ž z q a m
ˆ. Ž 9.6 .
then transforms in the same way as UŽ z, m . and a similar statement also applies to the
derivatives
DrA Pˆys Er Pˆy q R Ž A r . , Pˆy Ž 9.7 .
of the projector Pˆy Žhere and below the index r stands for t or s ..
We now consider the field
1 1
q Ž z . s yitr ½ 4
gˆ 5 DtA Pˆy , DsA Pˆy q
4
DtA Pˆy , DsA Pˆy gˆ 5
1
q R Ž Ft s . gˆ 5 Ž x , x . ,
2 5 Ž 9.8 .
¨
where w . . . x Ž x, y . denotes the kernel representing the operator enclosed in the square
bracket, at fixed t and s, in the same way as DŽ x, y . represents the Dirac operator. The
trace is taken over the Dirac and flavour indices only and q Ž z . is thus a gauge-invariant
local field in 4 q 2 dimensions. It is also not difficult to check ŽAppendix B. that q Ž z .
satisfies
a4 Ý d t d s d q Ž z . s 0
H Ž 9.9 .
x
for all local variations of the link variables UŽ z, m . and the potential A r Ž z ., i.e. it is a
topological field.
By definition q Ž z . is in the trivial cohomology class if it is equal to the divergence of
a gauge-invariant local current. We now show that this implies the existence of a local
current jmŽ x . in 4 dimensions satisfying the anomalous conservation law Ž5.10. and the
integrability condition Ž8.1.. The converse is also true, but we shall not prove this here.
So let us suppose that
q Ž z . s Em) km Ž z . q E t k s Ž z . y Es k t Ž z . , Ž 9.10 .
where kmŽ z ., k t Ž z . and k s Ž z . are gauge-invariant polynomials in
E tnEsm U Ž z , m . , E tnEsmA r Ž z . , n q m 0 1, Ž 9.11 .
with coefficients that are local fields on the lattice depending on UŽ z, m . and A r Ž z .. In
Eq. Ž9.10. the symbol Em) denotes the backward difference operator and the last two
terms have been written in the form of a curl for reasons to become clear below.
Under scale transformations of the coordinates t and s, the monomials contributing
to kmŽ z ., k t Ž z . and k s Ž z . transform homogeneously if A r Ž z . is transformed in the usual
way. It is then immediately clear that all terms on the right-hand side of Eq. Ž9.10. with
scale dimensions different from those of the left-hand side have to cancel. Without loss
we may, therefore, assume that the Ž t, s .-dimensions of kmŽ z ., k t Ž z . and k s Ž z . are Ž1,1.,
Ž1,0. and Ž0,1. respectively. Taking the gauge symmetry into account, this implies
k r Ž z . s a4 Ý l ar , m Ž w . K ra, m Ž w, z . , w s Ž y,t , s . , Ž 9.12 .
y
y1
a l r , m Ž w . s DrA U Ž w, m . U Ž w, m . , Ž 9.13 .
where K r, mŽ w, z . is a gauge-covariant local expression in the link variables.
After summing over all lattice points, Eq. Ž9.10. thus assumes the form
a4 Ý E t l as, m Ž w . j s,a m Ž w . y Es l at , m Ž w . jta, m Ž w . 4

y
1
s iTr Pˆy DtA Pˆy , DsA Pˆy y
½ 5
R Ž Ft s . gˆ 5 , Ž 9.14 .
2
jra, m Ž w . s a 4 Ý K ra, m Ž w, z . . Ž 9.15 .
x
Evidently jr, mŽ w . is a gauge-covariant local current depending on the link variables at

the given values of t and s, but not on their derivatives with respect to these
176 ¨
coordinates. Collecting all terms in Eq. Ž9.14. proportional to E t Es UŽ w, m ., we thus

conclude that
jt , m Ž w . s j s, m Ž w . ' jm Ž w . . Ž 9.16 .
Note that the t, s-dependence of jmŽ w . arises through the link variables only, i.e. the
current may be considered to be a local field in 4 dimensions which has been extended
to 4 q 2 dimensions by letting the gauge field depend on t and s.
It is now straightforward to show that this current has all the required properties. The
anomalous conservation law Ž5.10., for example, follows from Eq. Ž9.14. by choosing
the link variables to be independent of t and s. We may also set A r Ž z . s 0 in this
equation and in a few lines one then finds that the integrability condition Ž8.1. is
fulfilled. This proves that the triviality of the cohomology class of q Ž z . implies the
existence a local current satisfying Eqs. Ž5.10. and Ž8.1..
In the classical continuum limit the cohomology class of q Ž z . is easily determined.
As in the case of the anomaly discussed in Section 5, the field may be expanded in a
power series in a, the leading term being given by
1
qŽ z. s c1 d Ra b cea 1 . . . a 6 Faa1a 2 Ž z . Fab3a 4 Ž z . Fac5a 6 Ž z . q O Ž a . . Ž 9.17 .
6
The obvious notations for gauge fields in n s 6 dimensions are being used in this
equation, with space-time indices running from 0 to 5. If the fermion multiplet is
anomaly free, the expression on the right-hand side Žwhich is equal to 2p times the
Chern character w50x. vanishes and in this case q Ž z . has trivial cohomology to lowest
order in a. The same is true to any order in a, because the local fields which one
generates at the higher orders of the expansion are all topological. Recalling the theorem
quoted at the beginning of this section, this implies that they are of the form Ž9.2. with
cŽ z . s 0 since there are no Chern monomials with scale dimension greater than the
space-time dimension.
The existence of a local current satisfying Eqs. Ž5.10. and Ž8.1. is thus guaranteed to
all orders in a. Moreover all what is needed to extend this result to any fixed value of
the lattice spacing is the classification of all topological fields in 4 q 2 dimensions.
Presumably the non-trivial cohomology classes on the lattice are in one-to-one corre-
spondence with the linearly independent Chern monomials. At least this is so in the
abelian case w9x.
10. Concluding remarks
When studying chiral gauge theories one is often led to consider gauge and fermion
fields in higher dimensions. It may well be that this is just a matter of mathematical
convenience. On the other hand, the experience should perhaps be taken as an indication
that chiral gauge theories are merely effective descriptions of the low-energy modes of a
more fundamental theory in 4 q 1 or 4 q 2 dimensions. The approach of Kaplan and
Shamir w28–30x provides a concrete model for this and it would be important to work
out its relation to the framework presented in this paper in full detail, following the lines
sketched in Section 7.
¨
In the continuum limit the gauge anomaly cancels if the tensor d Ra b c vanishes. The
same is presumably true on the lattice, but a complete proof of this has only been given
in abelian theories so far w8x. For non-abelian gauge groups the current status is that the
anomaly cancellation has been established to all orders of an expansion in powers of the
lattice spacing. Moreover, as explained in Section 9, the problem has been reduced to
classifying the topological fields in 4 q 2 dimensions, which does not seem to be an
impossible task.
Global anomalies are a separate issue which requires control over the first homotopy
group of the space of lattice gauge fields satisfying the bound Ž2.1.. One may be able to
achieve this by noting that such fields are continuous on the scale of the lattice spacing
up to gauge transformations. The topology of the space of gauge orbits is hence
expected to be essentially the same as in the continuum theory.
I am grateful to Raymond Stora for guiding me through the literature on local
cohomology in gauge theories. Thanks also go to Oliver Bar ¨ and Isabel Campos for
helpful discussions on global anomalies and to Peter Weisz for a critical reading of a
first draft of this paper.
Appendix A
To prove Eq. Ž6.6. we choose a differentiable basis Õj of left-handed fields represent-

ing the fermion measure along the curve such that Õj < ts1 s Õj < ts0 Žthis is always
possible if the measure is smooth.. The measure term in Eq. Ž6.2. is then given by
L h s i Ý Ž Õj , E t Õj . . Ž A.1 .
j
Taking the properties of the operator Q t into account, we have

Õj s Q t Ý Õ l < ts0 Ž Sy1 . l j , Ž A.2 .
l
where S is some unitary transformation matrix satisfying S < ts0 s 1. When inserted in
Eq. ŽA.1. this yields
L h s yi E tln det S Ž A.3 .
and the Wilson loop W is thus equal to det S < ts1. At t s 1 the matrix S represents the
action of Q1 in the subspace of left-handed fields. In particular, its determinant
coincides with the determinant of Q1 in this subspace, i.e. with the right-hand side of
Eq. Ž6.6..
Appendix B
Starting from the definition Ž9.8. it is straightforward to show that

1 1
a4 Ý q Ž z . s iTr Pˆy E t Pˆy , Es Pˆy y
½ E t R Ž A s . gˆ 5 q Es R Ž A t . gˆ 5 5 .
x 2 2
Ž B.1 .
178 ¨
As a consequence we have
a4 Ý d t d s d q Ž z . s d t d s d iTr Pˆy E t Pˆy , Es Pˆy

H H ½ ½ 55 Ž B.2 .
x
for any local deformation of the gauge field. To evaluate the right-hand side of this
equation, we make use of the identity
Tr d Pˆy E t Pˆy Es Pˆy s 0,

½ 5 Ž B.3 .
.2
which may be established by inserting Žgˆ 5 s 1 and noting that gˆ 5 anti-commutes with
the derivatives of the projector. One then finds that the integrand is given by
E t iTr Pˆy d Pˆy , Es Pˆy

½ ½ 5 5 y E ½ iTr ½ Pˆ
s y d Pˆy , E t Pˆy 55 Ž B.4 .
and after integrating over t and s one gets zero because the variation of the gauge field
is compactly supported.
References
w1x P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 Ž1982. 2649.

w2x P. Hasenfratz, Nucl. Phys. B ŽProc. Suppl.. 63A-C Ž1998. 53.
w3x P. Hasenfratz, Nucl. Phys. B 525 Ž1998. 401.
w4x P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 Ž1998. 125.
w5x H. Neuberger, Phys. Lett. B 417 Ž1998. 141.
w6x H. Neuberger, Phys. Lett. B 427 Ž1998. 353.
w7x ¨
M. Luscher, Phys. Lett. B 428 Ž1998. 342.
w8x ¨
M. Luscher, Nucl. Phys. B 549 Ž1999. 295.
w9x ¨
M. Luscher, Nucl. Phys. B 538 Ž1999. 515.
w10x R. Stora, Continuum gauge theories, in: New Developments in Quantum Field Theory and Statistical
` 1976., ed. M. Levy,
Mechanics ŽCargese, ´ P. Mitter ŽPlenum Press, New York, 1977..
w11x R. Stora, Algebraic structure and topological origin of anomalies, in: Progress in Gauge Field Theory
` 1983., ed. G. ‘t Hooft et al. ŽPlenum Press, New York, 1984..
ŽCargese
w12x B. Zumino, Chiral anomalies and differential geometry, in: Relativity, groups and topology ŽLes Houches
1983., ed. B.S. DeWitt, R. Stora ŽNorth-Holland, Amsterdam, 1984..
w13x L. Alvarez-Gaume, ´ P. Ginsparg, Nucl. Phys. B 243 Ž1984. 449.
w14x L. Alvarez-Gaume, ´ An introduction to anomalies, in: Fundamental Problems of Gauge Field Theory
ŽErice 1985., ed. G. Velo, A.S. Wightman ŽPlenum Press, New York, 1986..
w15x R.A. Bertlmann, Anomalies in Quantum Field Theory ŽOxford Univ. Press, Oxford, 1996..
w16x ´
P. Hernandez, ¨
K. Jansen, M. Luscher, Nucl. Phys. B 552 Ž1999. 363.
w17x P. Hasenfratz, F. Niedermayer, private communication ŽFebruary 1998..
w18x F. Niedermayer, Nucl. Phys. B ŽProc. Suppl.. 73 Ž1999. 105.
w19x R. Narayanan, Phys. Rev. D 58 Ž1998. 97501.
w20x ¨
M. Luscher, Commun. Math. Phys. 85 Ž1982. 39.
w21x A.V. Phillips, D.A. Stone, Commun. Math. Phys. 103 Ž1986. 599.
w22x A.V. Phillips, D.A. Stone, Commun. Math. Phys. 131 Ž1990. 255.
w23x K. Fujikawa, Nucl. Phys. B 546 Ž1999. 480.
w24x H. Suzuki, Prog. Theor. Phys. 102 Ž1999. 141.
w25x D.H. Adams, Axial anomaly and topological charge in lattice gauge theory with overlap Dirac operator,
hep-latr9812003.
w26x Y. Kikukawa, A. Yamada, Phys. Lett. B 448 Ž1999. 265.
w27x T.-W. Chiu, T.-H. Hsieh, Perturbation calculation of the axial anomaly of Ginsparg–Wilson fermion,
hep-latr9901011.
¨
w28x D.B. Kaplan, Phys. Lett. B 288 Ž1992. 342.

w29x D.B. Kaplan, Nucl. Phys. B ŽProc. Suppl.. 30 Ž1993. 597.
w30x Y. Shamir, Nucl. Phys. B 406 Ž1993. 90.
w31x L. Alvarez-Gaume, ´ S. Della Pietra, V. Della Pietra, Phys. Lett. B 166 Ž1986. 177.
w32x L. Alvarez-Gaume, ´ S. Della Pietra, V. Della Pietra, Commun. Math. Phys. 109 Ž1987. 691.
w33x L. Alvarez-Gaume, ´ S. Della Pietra, The effective action for chiral fermions, in: Recent Developments in
Quantum Field Theory ŽNiels Bohr Centennial Conference, Copenhagen, 1985., ed. J. Ambjørn et al.
ŽNorth-Holland, Amsterdam, 1985..
w34x R.D. Ball, H. Osborn, Phys. Lett. B 165 Ž1985. 410.
w35x R.D. Ball, H. Osborn, Nucl. Phys. B 263 Ž1986. 245.
w36x R.D. Ball, Phys. Lett. B 171 Ž1986. 435.
w37x R.D. Ball, Phys. Rept. 182 Ž1989. 1.
w38x D.B. Kaplan, M. Schmaltz, Phys. Lett. B 368 Ž1996. 44.
w39x H. Suzuki, Prog. Theor. Phys. 101 Ž1999. 1147.
w40x E. Witten, Phys. Lett. B 117 Ž1982. 324.
w41x E. Witten, Nucl. Phys. B 223 Ž1983. 422.
w42x S. Elitzur, V.P. Nair, Nucl. Phys. B 243 Ž1984. 205.
w43x ¨ I. Campos, Global anomalies in chiral lattice gauge theory, hep-latr9909081.
O. Bar,
w44x F. Brandt, N. Dragon, M. Kreuzer, Phys. Lett. B 231 Ž1989. 263.
w45x F. Brandt, N. Dragon, M. Kreuzer, Nucl. Phys. B 332 Ž1990. 224.
w46x F. Brandt, N. Dragon, M. Kreuzer, Nucl. Phys. B 332 Ž1990. 250.
w47x M. Dubois-Violette, M. Henneaux, M. Talon, C.-M. Viallet, Phys. Lett. B 267 Ž1991. 81.
w48x M. Dubois-Violette, M. Henneaux, M. Talon, C.-M. Viallet, Phys. Lett. B 289 Ž1992. 361.
w49x N. Dragon, BRS symmetry and cohomology, Lectures given at Saalburg Summer School Ž1995.,
hep-thr9602163.
w50x P.B. Gilkey, Invariance Theory, the Heat Equation, and the Atiyah–Singer Index Theorem, 2nd ed. ŽCRC
Press, Boca Raton, 1995..
Constant field strengths on T 2n

1
Jan Troost
Theoretische Natuurkunde, Vrije UniÕersiteit Brussel, Pleinlaan 2, 1050 Brussels, Belgium
Received 30 September 1999; accepted 2 November 1999
Abstract
We analyse field strength configurations in UŽ N . Yang–Mills theory on T 2 n that are diagonal

and constant, extending early work of Van Baal on T 4. The spectrum of fluctuations is determined
and the eigenfunctions are given explicitly in terms of theta functions on tori. We show the
relevance of the analysis to higher dimensional D-branes and discuss applications of the results in
string theory. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 11.25.-w
1. Introduction
Constant field strengths were studied as solutions to UŽ N . Yang–Mills on T 4 fifteen

years ago in an attempt to get a handle on the mechanism of confinement w1,2x. The
spectrum of fluctuations around the background field configuration was explicitly
determined by Van Baal in terms of theta functions on the four-torus w3,4x. These results
were later used in the context of string theory in, amongst others ŽRefs. w5–7x. to
compare the spectrum of open strings ending on D-branes to the predictions from
Yang–Mills theory and the non-abelian Dirac–Born–Infeld action. The number of
massless torons also played a crucial role in the black hole entropy counting in w8,9x.
We extend the analysis of Van Baal to UŽ N . Yang–Mills solutions with constant and
diagonal field strength on general even dimensional tori. We will show how this analysis
applies to the low-energy effective action for D-branes in a flat background and with
small field strengths. Specifically, the results are relevant for higher dimensional
D-branes wrapped on even dimensional tori, e.g., D6-branes wrapped around a six-torus.
In the Yang–Mills theory, we derive the conditions on the field strengths to have a
stable configuration. We examine the criterion to have massless bosonic modes and give
explicitly the full set of fluctuations in terms of theta functions on higher tori. In the
E-mail address: troost@tena4.vub.ac.be ŽJ. Troost..

1
Aspirant FWO
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 2 1 - X
J. Troostr Nuclear Physics B 568 (2000) 180–194 181
dimensionally reduced SYM theory relevant to D-branes the spectrum of fermionic

fluctuations on T 2 n can also be determined and index theory is used to check the
number of fermionic massless modes.
The article starts out with dimensionally reducing D s 9 q 1 N s 1 UŽ N . Yang–
Mills to the low-energy ‘non-relativistic’ D-brane action. We delimit the gauge field
backgrounds we will study. Next, we compactify the theory on an even dimensional
torus and concentrate on the dependence of the fluctuations on the internal coordinates.
Here we extend the analysis of Van Baal straightforwardly to higher tori. Moreover, we
derive the spectrum of scalar and fermionic fluctuations in the given background and
discuss the zero modes of the configurations in detail. Finally, we situate some of the
applications in string theory in the context of our systematic analysis and indicate new
applications.
2. Reduced action
Van Baal studied a constant diagonal field strength configuration for UŽ N . Yang–
Mills on a four-torus w3,4x. We will later look at UŽ N . Yang–Mills on a general even
dimensional torus, but we will first show how this fits in the Yang–Mills approximation
to D-brane actions. We will closely follow the analysis and notation of w3,4x and w8,9x in
the next few sections. The bosonic part of the D s 9 q 1 N s 1 UŽ N . Yang–Mills
action can be written in terms of the ten-dimensional field strength Gmn Ž m s 0, . . . ,9.,
Gmn s Em Bn y En Bm q i Bm , Bn
as
1
S9q 1 s y d 10 x Tr Gmn
H 2
.
4
Reduced to p q 1 dimensions the action becomes Ž a s 0, . . . , p and m s p q 1, . . . ,9.
1 2 2
S pq 1 s y d pq1 x Tr Ga2b y w fm ,f n x q 2 Ž Ea fm q i w Ba ,f m x . ,
H ž / Ž 1.
4
where all fields depend only on the coordinates x b and the scalars and gauge fields are
f m s Bm Ž x b . , Ba s Ba Ž x b . . Ž 2.
The dimensionally reduced gauge transformations read
B ™VB V a
y1
y i VEa Vy1 , f ™ Vf V
m
y1
, Ž 3.
with V g UŽ N ..
3. Background and fluctuations
In this section we determine the action for the fluctuations around a general diagonal
and constant background field strength G 0 . We fix notations as follows:
Ba s Ba0 q Aa ,
0
Gab s Ea Bb0 y Eb Ba0 q i Ba0 , Bb0 ,
182 J. Troostr Nuclear Physics B 568 (2000) 180–194
Da s Ea q i Ba0 ,. ,
Gab s Ga0b q Fa b ,
Fab s Da Ab y Db Aa q i Aa , Ab , Ž 4.
where the gauge field fluctuations are denoted A, with corresponding field strength F,
and we have defined a background covariant derivative D. We choose the background
gauge fixing condition Da A a s 0. When the background field strength is diagonal and
constant Žsolving the classical equations of motion., we find the following action for the
fluctuations:
1 pq 1 0 2
Ssy Hd x Tr Ž Gab y 2 Aa D 2Aa y 4 iAa Ga0b , Ab y 2 fm D 2fm
4
2
q 2 i Ž Da Ab y Db Aa . Aa , Ab q 4 i fm Da w fm , Aa x y Aa , Ab
2 2
y 2 w Aa , f m x y w f m , f n x . Ž 5.
It will be convenient to expand the fluctuations in a Lie algebra basis for UŽ n., namely,
ŽTi .a b s d i a d i b and Ž e i j .a b s d i a d jb for i / j:
Aa s aai Ti q bai j e i j ,
fm s c mi Ti q d mi j e i j . Ž 6.
From the reality properties of the gauge fields and the scalars we find that a and c are
real, and that bai j s baji ) and d mi j s d mji ) are complex.
4. Compactification
Now we will focus on our main interest. Consider 2 n spatial dimensions of the
D-brane to be wrapped on a torus of dimension 2 n with radii R aˆ s Laˆ r2p , where
Ž aˆ s 1, . . . ,2 n.. In what follows, we only consider non-trivial field strengths in these
directions. Magnetic flux quantization2 and the fact that the background field strength is
i
diagonal implies then that we can write our background in terms of the integers nab ˆ ˆ,
i
0
nab
ˆˆ
ˆ ˆ s 2p
Gab Ti . Ž 7.
Laˆ Lbˆ
We can choose the background gauge field to be
i b ˆ
nab
ˆˆx
Baˆ0 s yp Ti . Ž 8.
Laˆ Lbˆ
Next, we substitute this form of the background into the action Ž5. and concentrate on
the terms quadratic in the fluctuations. Since we will analyse the spectrum of small
2
We restrict to configurations where the branes are wrapped only once around, for example, the odd cycles.
Generalizing this to wrapping more than once these cycles changes the quantization condition. See, for
instance, Refs. w7–9x.
fluctuations around the background, we neglect the interactions between the fluctuations
from now on. The quadratic action is
1
S Ž2. s y d pq1 x aai M0 aai q c mi M0 c mi q 2 Ý bai j) Ž Mi j da b y 4p iJai bj . bbi j
H ž ž
2 i-j
qd mi j) Mi j d mi j //, Ž 9.
where we have defined J, a measure for the difference in field strength in sector i and j
Žon brane i and brane j ., and the mass operators M0 and Mi j ,
i j
ij
nab
ˆ ˆ y naˆ bˆ
J, ab
ˆ ˆs
Laˆ Lbˆ
M0 s yEa2 ,
2 2 2 2
E0 E 2 nq1 Ep Eâ ˆ
Mi j s ž / ž y / y... y ž / ž y yp Jaî bjˆ x b / . Ž 10 .
i i i i
5. The mass operators on T 2n

From now on we will concentrate on the dependence of the fluctuations on the
coordinates of the torus, the internal coordinates x â. Specifically, we consider the
problem of diagonalizing the mass operators as operators on T 2 n. As usual, this gives
information on the spectrum of modes living in the non-compact space. The quadratic
action Ž9. can be analysed in each sector ij separately. In the following we will leave out
the indices i and j to simplify notation. We note that the analysis of the bosonic
fluctuations in the following sections is valid for general n.
It will turn out that the mass operator Ž10. can be written in a simple form.
Transforming coordinates with an element of O Ž2 n., we can bring the difference in field
strengths in sector ij in the standard form
0 f1 0 0 0
yf 1 0 0 0 0
Jab
ˆ ˆs
0
0
0
0
0
0
where f 1 0 f 2 0 . . . 0 f n 0 0.
.
.
.
0
0
0
0
yf n
0 ,
fn
0
0
Next, we introduce a complex structure on the torus as follows:
Ž 11 .
1
z s Ž z1 , z 2 , . . . , z n . s
'2 Ž x 1 y ix 2 , . . . , x 2 ny1 y ix 2 n . ,
1
A s Ž A z1 , A z 2 , . . . , A z n . s
'2 Ž A1 q iA 2 , . . . , A 2 ny1 q iA 2 n . ,
1
Ž Ez1 , . . . , Ez n . s ' Ž Ex 1 q i Ex 2 , . . . ,Ex 2 ny 1 q i Ex 2 n . . Ž 12 .
2
Further, we define the positive hermitian form H,

H Ž z ,w . s 2 Ž z 1 f 1w 1 q . . . qz n f n wn . s w† hz , Ž 13 .
h s 2 diag Ž f 1 , . . . ,f n . , Ž 14 .
and the creation and annihilation operators
1 E
ak s
i ž E zk
q p fk zk ,
/
1 E
a†k s
i ž E zk
y p fk zk ,
/ Ž 15 .
where Ž k s 1, . . . ,n.. Thus we can write a nice expression for the relevant non-trivial
mass operator Mi j Ž10.,
M s a k ,a†k 4 . Ž 16 .
Before we can diagonalize the mass operator, it is crucial to discuss the boundary
conditions the eigenfunctions have to satisfy. They encode the topological data of the
background gauge field.
6. Boundary conditions
We recall that the background field strength and gauge fields were given by
0
2p
Gab
ˆ ˆs naiˆ bˆ Ti ,
Laˆ Lbˆ
p ˆ
Baˆ0 s y naiˆ bˆ x b Ti . Ž 17 .
Laˆ Lbˆ
The transition functions V of the gauge bundle over the torus have to satisfy
ˆ ˆ bˆ
p i
Baˆ0 x b q Lbˆ s Vbˆ Ba0 Ž x b . Vby1
ž / y1 0
ˆ y i Vbˆ Ea Vbˆ s Baˆ Ž x . y n ˆT . Ž 18 .
Lâ aˆ b i
We choose them to be
Vaˆ sexp yp inaiˆ bˆ x bˆ TirLbˆ .
ž / Ž 19 .
The boundary conditions following from the background gauge field induced transition
functions read, using Eqs. Ž3. and Ž6.,
ž ˆ /
aai x b q Lbˆ s aai Ž x b . ,
ˆ
ij
bai j ž x ˆ q L ˆ / sexp ž yp in ˆ ˆ x ˆrL ˆ / b
b
b bg
g
g
ij
a Ž xb.,
c mi ž x ˆqL ˆ / sc Ž x ˆ . ,
b
b
i
m
b
ij
d mi j ž x ˆ q L ˆ / sexp ž yp in ˆ ˆ x ˆrL ˆ / d
b
b bg
g
g
ij
m Ž x bˆ . . Ž 20 .
7. Spectrum and eigenfunctions
After the preliminary work of writing the non-trivial mass operator M in a harmonic
oscillator form in terms of complex coordinates Ž16., and discussing the boundary
conditions the fluctuations have to satisfy Ž20., we determine the spectrum and the
eigenfunctions. We only discuss in detail the non-trivial case of off-diagonal modes.
Moreover, the difference between gauge field off-diagonal modes and scalar off-diago-
nal modes is a mere constant in the eigenvalues, so we can treat them in one go. We
follow the analysis of Refs. w3,4x and w8,9x. ŽSee Ref. w10x for an early mathematical
treatment.. The ground state x 0 of the mass operator in the off-diagonal sector we take
to satisfy the usual conditions
1 E p
0 s ak x 0 s
i ž Ez k
q
2
hk l zl x0 ,
/ Ž 21 .
and it has to obey the boundary conditions for the off-diagonal sector
ij
x 0 x bˆ q Lbˆ s ey p i nb̂g x x 0 Ž x bˆ . .
g
r Lg
ž / Ž 22 .
The differential equation is immediately solved in terms of the hermitian form H Ž13.
and a general holomorphic function f,
p
x 0 Ž z . s ey 2 H Ž z , z . f Ž z . , Ž 23 .
but the treatment of the boundary conditions is more involved. The boundary condition
for the fluctuation x 0 implies a non-trivial boundary condition for the holomorphic
function f Ž z .. It will be convenient to introduce some extra machinery to write these
boundary conditions in terms of objects well known in the mathematical literature w11x
on theta functions w3,4x. In terms of the hermitian form H Ž z,w ., we define an
antisymmetric form EŽ z,w .,
1
baˆ aˆ b 2 2 ˆ ˆ
H Ž z Ž x . ,w Ž y . . s x Ž yJ . ab
ˆ ˆ y q ix Jaˆ bˆ y ,
1 bˆ
E Ž z ,w . s Im H Ž z ,w . s Ž H Ž z ,w . y H Ž w, z . . s x aˆ Jab
ˆˆy . Ž 24 .
2i
We introduce the notation
1
q s Ž q1 , . . . ,qn . '
'2 Ž m1 L1 y im 2 L2 , . . . ,m 2 ny1 L2 ny1 y im 2 n L2 n . , Ž 25 .
to write the second degree bicharacter a Ž q . in the simple form
a Ž q . sexp p i ž Ý
aˆ - bˆ
maˆ naˆ bˆ mbˆ . / Ž 26 .
These objects make it easy to write down the boundary condition for the fluctuations x
for windings around the torus for any number of times in different directions,
x 0 Ž z q q . s x 0 Ž z . eyp i EŽ q , z .a Ž q . . Ž 27 .
The boundary conditions the holomorphic function f Ž23. satisfies can then finally be
written in terms of the hermitian form H and the second degree bicharacter a ,
p
f Ž z q q . s f Ž z . a Ž q . ep H Ž z , q .q 2 H Ž q , q . . Ž 28 .
Now comes the pay-off for introducing the appropriate mathematical machinery. These
holomorphic functions f are theta functions on T 2 n w11x. They span a vector space of
dimension < Pf Ž nab
ˆ ˆ . This space of theta functions is the space of ground state
.<
fluctuations around the given gauge field background. They can be written down
explicitly and we do so in the appendix.
The higher modes are given by acting with the creation operators on the ground state.
They automatically satisfy the boundary conditions. It is clear from Ž16. then that the
spectrum of off-diagonal scalar field fluctuations is given by the harmonic oscillator
formula
n
l s 2p Ý Ž 2 m iq1. f i Ž 29 .
is1
and after a further trivial diagonalization Žcompare Ž9.. for the off-diagonal gauge fields
we get the shifted spectra
n
l"
k s 2p Ý Ž 2 m iq1. f i " 2 f k . Ž 30 .
is1
8. Summary
We summarize the spectrum and eigenfunctions for diagonal, off-diagonal and gauge
field and scalar fluctuations in Table 1. We use the notations e Ž z k . s 1 Ž0,0, . . . ,1,y
'2
i, . . . ,0,0. for the eigenvectors of J Ž11., V for the volume of the torus, i, j g
1, . . . , N 4 ;ij g 1, . . . , N Ž N y 1.4 ;k g 1, . . . ,n4 ; p g Z 4 ;m g N n ;ri g 0, . . . ,e i y 14 . All
of these notations are straightforward except for the components ri and e i for which we
Table 1
Eigenfunctions and eigenvalues
Scalar Fluctuations Eigenvalues
Diagonal c li Ti 2
2p pâ
f lp,i Ý â ž /
Lâ
Off-diagonal d li j e i j
fi / 0 f lm, r,i j 2p w Ý nis1 Ž 2 m i q1 . f i x
Gauge field Fluctuations Eigenvalues
Diagonal aaiˆ Ti 2
2p pâ
fâp,i Ý â ž /
Lâ
Off-diagonal baî j e i j
fi / 0 e z kf m, r,i j 2p w Ý nis1 Ž 2 m i q1 . f i y2 f k x
e z kf m, r,i j 2p w Ý nis 1 Ž 2 m i q1 . f i q2 f k x
refer to the appendix. Suffice it to remark that the degeneracy of the off-diagonal
ij .
fluctuations is given by Pf Ž nab ˆ ˆ s Ł i e i in sector ij, and the space of theta functions is
indexed by r. We have, moreover,
1 2p i p x r L
f p ,i s
'V e
aˆ aˆ aˆ T ,
i
m1 mn
f m , r ,i j s Ž a†1 . . . . Ž a†n . r m1 ! . . . m n ! xr e i j .
( Ž 31 .
ij
We only catalogued the case where all f are different from zero and H is
non-degenerate. The eigenvalues and eigenfunctions can also be classified easily in the
other cases.
9. Stability and supersymmetry
In this section we discuss stability, supersymmetry and the occurrence of massless

bosonic modes in our configurations. From the general classification we find that there
are no tachyonic modes when f 2 q f 3 q . . . qf n 0 f 1 Žfor all i, j .. If this condition is
satisfied, stability of the gauge field configuration is insured, at quadratic level. On T 4
the condition for stability implies that the field strength is self-dual Ž f 1 s f 2 .. ŽRecall
that we still have f 1 0 f 2 0 . . . 0 f n 0 0.. The condition for stability is much more loose
when there are more than two non-trivial field strength components. Massless modes for
the gauge fields occur when the equality f 2 q f 3 q . . . qf n s f 1 is satisfied. For higher
tori they appear only in a complex combination of the gauge field components A1,2 . On
T 4 , stable configurations automatically have massless modes and they occur for the
gauge field components A1,2 and A 3,4 . For higher dimensions a stable configuration
does not necessarily have bosonic massless modes.
Turning back to our starting point, we can regard the theory we study Žfor n ( 4. as
dimensionally reduced super-Yang–Mills theory. The supersymmetry variation of the
adjoint fermions in our background with trivial scalars then reads
dc s Fab ˆˆ
ab
ˆ ˆg e . Ž 32 .
It is clear that if f 2 q f 3 q . . . qf n s f 1 , the following conditions project onto the
preserved supersymmetry parameter:
yg 12 e s g 34 e s . . . s g 2 ny12 n e . Ž 33 .
There are two other possibilities to preserve supersymmetry 3 when n s 4, namely if we
have f 1 q f 4 s f 2 q f 3 or if f 1 s f 2 , f 3 s f 4 . The following projection conditions yield
the preserved supersymmetry:
yg 12 e s g 34 e s g 56 e s yg 78 e , Ž 34 .
and
yg 12 e s g 34 e , g 56 e s yg 78 e , Ž 35 .
respectively. These last configurations do not have any bosonic massless modes. In
general, each projection condition halves the number of supersymmetries. Note that for
3
We do not consider the case where one of the field strengths vanishes, although it is easily incorporated in
our framework.
the four-torus the condition for stability, namely self-duality, coincides with the condi-
tion for preservation of supersymmetry. For higher tori this is not the case. The space of
stable gauge field configurations is for higher tori much larger than the space of
supersymmetric gauge field configurations.
10. Fermionic spectrum
Up till now we ignored the fermions in our reduction of super-Yang–Mills. In this

section we determine the spectrum of the fermionic fluctuations in the theory reduced to
p q 1 dimensions, compactified on T 2 n , and in the background gauge field configuration
Ž17.. It will turn out that the analysis is simple once the bosonic case has been treated in
detail.
The ten-dimensional Majorana–Weyl fermions are in the adjoint representation and
can be decomposed as follows:
c s c i Ti q c k l e k l , Ž 36 .
where c k l s c l k ) and k / l. The relevant Dirac equation for the fermionic modes is
easily derived from the fermionic part of the Yang–Mills action quadratic in the
fluctuations. As for the bosonic case, we can analyse the spectrum of the Dirac operator
on T 2 n to find the mass spectrum in the non-compact directions. To that end, we
analyse the equation4
Eâ ˆ
Ž i ng 1 . . . 2 n . gaˆ ž y p Jaˆkblˆ x b c k l s mc k l .
/ Ž 37 .
i
We concentrate on the non-trivial off-diagonal components in sectors kl. The standard
trick to find the spectrum of the fermionic mass operator is to square it,
2
Eâ ˆ
gaˆ ž y p Jaˆkblˆ x b / s Ý a k ,a†k 4 y 2p if 1 g 12 y . . . y 2p if n g 2 ny1,2 n ,
i k
where we used Ž11.. It is then easy to determine the spectrum of the fermionic mass
operator by projecting onto eigenspinors of g 2 ky1,2 k ,
l s Ý 2p Ž 2 m kq1 . f k " 2p f 1 " . . . " 2p f n . Ž 38 .

k
Implicitly, we have made use of the fact that the off-diagonal fermions satisfy the same
boundary conditions as the bosons 5. Drawing on the results in the previous sections, the
eigenfunctions can then also easily be determined. Remark that the fermions all have a
4
The appearance of the chirality operator is to ensure that the non-compact and the compact part of the
Dirac operator commute. It does not play a crucial role in our analysis since it does not change the Dirac
algebra w12x.
5
We do not want to break more supersymmetry by adding non-trivial monodromies on top of the ones
induced by the background.
certain helicity associated to the magnetic fields in the directions 12,34, . . . and each
component of the magnetic field is responsible for a Zeeman splitting of the energy
levels.
11. Zero modes and supersymmetry
11.1. Fermionic zero modes
In this section we take a closer look at the bosonic and fermionic zero modes that
often play a crucial role in applications. We start by describing the fermionic zero modes
in greater detail. It is clear from the analysis in the previous section that only the
following projected spinor has zero modes:
kl
c kl
cqq . . . qs Ž 1 q ig 1 g 2 . Ž 1 q ig 3 g4 . . . . Ž 1 q ig 2 ny1 g 2 n . . Ž 39 .
2n
The signs of the projection operators reverse for sector lk, since the field strengths are
opposite in that sector. Explicitly, the differential equations and boundary conditions in
sectors kl and lk read
a i cqk l. . . qs 0, Ž 40 .
a )i cyl k . . . ys 0, Ž 41 .
kl
c k l x bˆ q Lbˆ s ey p
ž / i nb̂g x g r Lg
c k l Ž x bˆ . . Ž 42 .
From these equations we determine the total number of massless fermionic modes,
namely 16 Ž kl .
2 n Ý k - l Pf naˆ bˆ complex zero modes. In certain cases, there is a shorter route to
get the number of fermionic zero modes, via index theory. As a preliminary, we indicate
the 2 n-dimensional chirality of the spinors. It is clear that
Ž 1 y i ng 1 . . . g 2 n . cqk l. . . qs 0, Ž 43 .
n
Ž 1 y Ž yi . g 1 . . . g 2 n . cyl k . . . ys 0, Ž 44 .
such that, for n s 2 k even, the zero modes have the same chirality in the two sectors,
while for n s 2 k q 1 odd, they have opposite chirality. Note that this strokes with Ž40.,
Ž41. and the fact that the complex conjugate representation of the Weyl representation is
the original one for SO Ž4 k ., and of different chirality for SO Ž4 k q 2..
11.2. Index theory check on number of zero modes
Using index theory, we can learn the difference in number of massless fermionic
modes of positive and negative chirality. If the fermionic zero modes all have the same
chirality, index theory predicts the total number of fermionic zero modes. We will use
the index theorem for the twisted spin complex on a flat manifold for the adjoint
representation w17,18x
index Ž D "= Ž adj . , D . s HMch Ž F adj . s nqy ny , Ž 45 .

where
j
Fadj
ch Ž Fadj . s Ý Tr j
j j! Ž 2p .
is the Chern class evaluated in the adjoint representation and n " is the number of
positive, negative chirality zero modes. In the background Ž17. the integral is easily
evaluated on T 2 n ,
nqy nys Ý Pf nakˆ lbˆ .
ž / Ž 46 .
kl
For d s 4 k q 2 the sum over all sectors is zero, the contributions from sector Ž kl .
Žstrings going one way. cancelling the contribution from sector Ž lk . Žstrings going the
other way.. Indeed, from the analysis in the previous section we know that the number
of zero modes of positive chirality equals the number of zero modes of negative chirality
in this case. For d s 4 k we find, in our conventions, only zero modes of positive
chirality, and index theory counts Ý k l Pf Ž nakˆ lbˆ . complex fermionic zero modes in d s 4 k.
Taking into account the multiplicity of the zero modes originating in ten dimensions, we
find 16 Ž k l . non-constant complex spinor zero mode components, as before.
2 n Ý k - l Pf naˆ bˆ
This straightforwardly extends the well-known results in four dimensions w19,20x.
11.3. Supersymmetry and massless modes
The number of massless fermionic and massless bosonic modes differs in supersym-
metric configurations, and at first sight it is difficult to see how they form a representa-
tion of supersymmetry when it is partially unbroken. Nevertheless, they do. We discuss
this slightly puzzling feature in this subsection. The unbroken supersymmetry transfor-
mations rules are given by the dimensionally reduced formulae of N s 1 SYM in ten
dimensions. In ten dimensions the formulae read
i
d Ama s egm c a , Ž 47 .
2
dc a s y 14 Fmna g mne . Ž 48 .
Consider first the case where the unbroken supersymmetry is given by
yg 12 e s g 34 e s . . . s g 2 ny12 n e . Ž 49 .
Starting out with a fermionic zero mode satisfying
kl
c kl
cqq . . . qs Ž 1 q ig 1 g 2 . Ž 1 q ig 3 g4 . . . . Ž 1 q ig 2 ny1 g 2 n . , Ž 50 .
2n
we easily see from Eqs. Ž47., Ž49., Ž50. and m s m that it will never transform into a
massless scalar. That is consistent with the spectrum we found earlier. When n 0 3, we
find only non-trivial gauge field components in the 1 and 2 direction. For gauge field
components in the other directions, the projection condition on the parameter e and the
zero mode c make sure that the variation vanishes. This is again consistent with what
we found earlier. It can easily be checked that on T 4 a similar analysis yields complex
bosonic zero modes in directions 1,2 and 3,4. Moreover, for the cases where the
supersymmetry parameter satisfies g 12 e s yg 34 e and yg 56 e s g 78 e we find that no

bosonic zero modes are generated, as expected. Finally, it can be checked using the
formulae for the bosonic zero modes that the transformation rule Ž48. is always trivial
Žat the order we are working.. Thus, the analysis of fermionic and bosonic zero modes is
perfectly consistent with supersymmetry, as it should be. Of course, one could perform a
similar consistency check for higher modes in the spectrum.
12. Applications and conclusion
The theory we have been discussing is the low-energy theory of D-branes in a trivial
background and with only small Ž‘non-relativistic’. gauge fields excited. Indeed, string
theory should be seen as the high-energy completion of the theories we studied.
Nevertheless, the approximation to string theory we discussed was often used to
understand results in string theory. We will briefly review some of the applications of
the results that we obtained that have already been made in the literature and that can be
coherently presented and extended in our framework.
In w7x the fluctuation spectrum on T 4 in the Yang–Mills approximation was
compared to the string theory fluctuation spectrum and the role of the Žnon-abelian.
Dirac–Born–Infeld action in resolving the discrepancy was clarified. The precise form
of the non-abelian Dirac–Born–Infeld action remained unclear. Our analysis could be
useful for further studying fluctuation spectra from the different points of view along the
lines of w7x, for higher branes on higher tori. Note also that the explicit form for a
tachyonic fluctuation was used in w7x to discuss tachyon condensation intuitively. That
discussion is now easily extended to higher branes.
The condition for the background gauge fields to preserve supersymmetry is familiar
in D-brane physics, especially in its T-dual form. To repeat this well-known point, it is
sufficient to give an archetypical example. Consider the following configuration: a pair
of D8-branes compactified on T 8 with a constant field strength on the first D8-brane
F12 0 F34 0 F56 0 F78 0 0. When F12 s F34 q F56 q F78 , supersymmetry is conserved.
T-dualize this configuration over directions 2,4,6,8 to obtain a pair of D4-branes at
angles. The angles are given by the following formulae6 : F12 s tan f 1 , F34 s tan f 2 ,
F56 s tan f 3 , F78 s tan f4 . Working at small angle, or taking into account the modifica-
tions the Born–Infeld action induces in the Yang–Mills theory in the spirit of w7x, we
find that the condition coincides with the well-known one for rotated branes w14–16x.
Note though that for the case n s 4 we saw a physical distinction between the case
where f 1 s f 2 q f 3 q f 4 and f 1 q f 4 s f 2 q f 3 . The difference between the two cases is
probably related to the mechanism of the creation of a string in the D0–D8 system Žor
D4–D4 system., but we do not pursue this here.
As noted before, on higher tori, the space of stable gauge field configurations is much
larger than the space of supersymmetric configurations. The same applies therefore to all
kinds of D-brane constructions in gauge theories. The results on T 4 for the Yang–Mills
theory were extensively discussed in for instance w5–7x. Stability of some special
6
We ignore some constant factors.
configurations on T 6 and T 8 was used in w13x to adhere zero-branes to six- and

eight-branes in Matrix theory. It is now obvious that these are points in a larger
parameter space of stable configurations. Note though that we only proved stability up to
quadratic order in the fluctuations w13x. It may be useful to recall that for instance the
black hole with only D0- and D6-brane charge, which has a corresponding quadratically
stable representation in the gauge theory on the D6-brane, is in fact metastable w13,23x.
Another motivation for our work can be found in the following problem. Consider the
only regular four-dimensional supersymmetric black hole that is solely made out of
D-branes w21x. Specifically, dualize to the configuration where the compact part carries
D6–D2–D2–D2-brane charges. The entropy for this black hole, calculated in supergrav-
ity, was microscopically accounted for in w21,22x up to a constant factor which it would
be interesting to determine. We can now easily find supersymmetric configurations in
the gauge theory living on the D6-branes that represent such a black hole. Next, we can
use the results derived in this article to calculate the dimension of the moduli space of
these configurations by counting massless states. Although in principle this program as
suggested in w21x looks sound, we have as yet not been able to make it work. Finally, we
indicate that the analysis of strings stretching between branes in Matrix theory is
analogous Žsee, for instance, Ref. w24x. and we believe that our systematic treatment
could be of practical use in that context too.
To sum up, we have determined the spectrum and explicit eigenfunctions of the
fluctuations around constant and diagonal field strengths in UŽ N . Yang–Mills on an
even dimensional torus, extending earlier work w3,4,8,9x on T 4 . We discussed supersym-
metry and stability of these configurations, and the counting of zero modes. The analysis
yields a systematic framework for applications in string theory.
Acknowledgements
It is a pleasure to thank Ben Craps, Frederik Denef, Marc Massar, Frederik Roose,
Alex Sevrin and especially Walter Troost for useful discussions, and Wati Taylor and
Pierre Van Baal for friendly correspondence.
Appendix A. Explicit eigenfunctions
To find the explicit form of the theta functions we need some more machinery w11x.
We go to a canonical ŽFrobenius. basis of the lattice using an SLŽ2 n,Z. transformation,
°0 0 e1
.
0
.
¶
0 0 . .
. .
0 0 0 en
ns , Ž A.1 .
ye1 0 0 0
. .
. . 0 0
. .
¢0 ye n 0 0 ß
where the e i are positive integers, each e i dividing the next e iq1. We choose a C-basis
as follows:
j Ž nq1. j Ž2 n.
z s z˜1 q . . . qzñ s Uz˜ , Ž A.2 .
e1 en
where j â denotes the canonical basis of the lattice. By X we denote the real vector
space generated by j nq 1 , . . . , j 2 n. Since Im H s E Ž24. is zero on X = X, we find that
h˜ s U † hU is real and symmetric. We define then the symmetric C-bilinear form S
uniquely associated with H
˜˜
S Ž z ,w . s yw˜ T hz Ž A.3 .
and the quasi-hermitian form Q
Q Ž z ,w . s H Ž z ,w . q S Ž z ,w . . Ž A.4 .
Moreover, we need the period matrix t defined in terms of the quasi-hermitian form Q:
i
t i j ' Q Ž j Ž i. , j Ž j. . . Ž A.5 .
2
Further definitions are required, namely the bicharacter ep i BŽ q, q . with respect to the
canonical basis j :
q s qã̂ j â ,
n
B Ž q,q . s Ý e i q˜i q˜iqn Ž A.6 .
is1
and b Ž q ., encoding the characters m and l of a :

a Ž q . s ep i BŽ q , q .b Ž q . ,
n
b Ž q . s e 2 p i Ý Ž m k q̃ kq ny l k q˜ k . . Ž A.7 .
is 1
The theta functions are then explicitly given by

u Ž z. s Ý d r ur Ž z . ,
0(r i- < e i <
y1
Ž mqr ..t Ž pqe y1 Ž mqr ..q 2 p iŽ pqe y1 Ž mqr ..PŽ z̃ql .
ur Ž z . s Ý ep iŽpq e , Ž A.8 .
pgZ n
where e1 < e 2 < e 3 . . . < e n and e is diagonal.
References
w1x G. ’t Hooft, Nucl. Phys. B 153 Ž1979. 141.

w2x G. ’t Hooft, Comm. Math. Phys. 81 Ž1981. 267.
w3x P. Van Baal, Comm. Math. Phys. 94 Ž1984. 397.
w4x P. Van Baal, Comm. Math. Phys. 85 Ž1982. 529.
w5x Z. Guralnik, S. Ramgoolam, Nucl. Phys. B 521 Ž1998. 129. hep-thr9708089.
w6x Z. Guralnik, S. Ramgoolam, Nucl. Phys. B 499 Ž1997. 241. hep-thr9702099.

w7x A. Hashimoto, W. Taylor IV, Nucl. Phys. B 503 Ž1997. 193. hep-thr9703217.
w8x M. Costa, M. Perry, Nucl. Phys. B 524 Ž1998. 333. hep-thr9712160.
w9x M. Costa, M. Perry, Nucl. Phys. B 520 Ž1998. 205. hep-thr9712026.
w10x ¨
J. Hano, in: Manifolds and Lie groups ŽBirkhauser, Basel, 1981..
w11x J. Igusa, Theta Functions ŽSpringer, Berlin, 1972..
w12x M. Green, J. Schwarz, E. Witten, Superstring Theory II ŽCambridge Univ. Press, Cambridge, 1987..
w13x W. Taylor IV, Nucl. Phys. B 508 Ž1997. 122. hep-thr9705116.
w14x M. Berkooz, M. Douglas, R. Leigh, Nucl. Phys. B 480 Ž1996. 265. hep-thr9606139.
w15x V. Balasubramanian, R. Leigh, Phys. Rev. D 55 Ž1997. 6415. hep-thr9611165.
w16x J. Polchinski, String Theory II ŽCambridge Univ. Press, Cambridge, 1998..
w17x T. Eguchi, P. Gilkey, A. Hanson, Phys. Rep. 66 Ž1980. 213.
w18x M. Nakahara, Geometry, Topology and Physics ŽIOP, 1990..
w19x A. Schwarz, Commun. Math. Phys. 64 Ž1979. 233.
w20x A. Schwarz, Phys. Lett. B 67 Ž1977. 172.
w21x V. Balasubramanian, F. Larsen, Nucl. Phys. B 478 Ž1996. 199. hep-thr9604189.
w22x V. Balasubramanian, F. Larsen, R. Leigh, Phys. Rev. D 57 Ž1998. 3509. hep-thr9704143.
w23x H. Sheinblatt, Phys. Rev. D 57 Ž1998. 2421. hep-thr9705054.
w24x P. Ho, M. Li, Y. Wu, Nucl. Phys. B 525 Ž1998. 146. hep-thr9706073.
Energy–momentum conservation and holographic S-matrix

Miao Li
Enrico Fermi Institute, UniÕersity of Chicago, 5640 Ellis AÕenue, Chicago, IL 60637, USA
Received 10 May 1999; received in revised form 4 October 1999; accepted 8 October 1999
Abstract
We investigate the consequence of the energy–momentum conservation law for the holo-
graphic S-matrix from AdSrCFT correspondence. It is shown that the conservation law is not a
natural consequence of conformal invariance in the large N limit. We predict a new singularity for
the four-point correlation function of a marginal operator. Only the two-point scattering amplitude
is explicitly calculated, and the result agrees with what is expected. q 2000 Elsevier Science B.V.
All rights reserved.
PACS: 11.25.-w; 11.25.Sq
1. Introduction and intuitive observation
Although there exist several non-perturbative formulations of M theory and string

theory in a flat space-time, it has been hard to do a quantitative calculation in the right
regime where the formulation is supposed to be valid. This difficulty has to do with the
unusual physics which is required of a holographic theory w1,2x. The most recent
proposal is an explicit ansatz for the S-matrix in the flat space limit using a convolution
of conformal correlators in the boundary conformal field theory w3,4x, for a related
discussion, see w5,6x. A few puzzling aspects are already pointed out within this context
in w7x.
The ansatz of Refs. w3,4x involves taking a peculiar high energy limit along with a
large N limit. For instance, the type IIB string theory defined on R 10 is encoded in the
super Yang–Mills theory on S 3 = R in the large N limit. Of course, the whole theory
may not exist in the large N limit when the Yang–Mills coupling constant is held fixed.
The conjecture of Refs. w3,4x rather asserts that a certain subsector must exist. If one
possesses infinite calculational power to calculate all the relevant correlation functions
in SYM, the IIB string S-matrix can be constructed non-perturbatively.
E-mail address: mli@theory.uchicago.edu ŽM. Li..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 5 6 - 2
196 M. Li r Nuclear Physics B 568 (2000) 195–207
Of course, for the time being we are not yet that powerful. On the contrary, the
Maldacena conjecture w8x has been used to make predictions about the SYM theory.
Although we do not know much about the non-perturbative S-matrix, a few principles
are certainly applicable. Traditionally, very general principles such as Lorentz invari-
ance, unitarity and analyticity constitute strong constraints on the S-matrix. We expect
that these constraints transform into the ones on a subset of correlation functions in
SYM via the holographic S-matrix ansatz. Our purpose in this paper is to point out that
the simplest consequence of Lorentz invariance, the energy–momentum conservation
law, is not a bona fide consequence of conformal invariance. Notice that the isometry of
the anti-de Sitter space is the conformal group. Taking the large radius limit, the
conformal group contracts to the Poincare´ group. Rather surprisingly, already at the
level of the four-point amplitude, implementation of energy–momentum conservation
requires the existence of a new type of singularity in the four-point correlation functions
in the large N limit. This singularity, to our knowledge, is not dictated by conformal
symmetry.
In order to extract information about physical process happening in the center of the
anti-de Sitter space, well-focused wave packets must be prepared. A precise ansatz for
an incoming particle or an outgoing particle is given in w3x. We will focus on massless
particles, for we will work with the type IIB string theory, and the only stable states are
those of the supergraviton. Denote the creation operator of an incoming particle by
av ey, where v s RE is the dimensionless energy, R is the radius of AdS 5 , and e is a
unit four-vector. This particle carries a momentum v e tangent to AdS 5 in its center. The
state is smeared over S 5, or it carries a zero momentum in the internal space. Similarly,
denote the annihilation operator of an outgoing particle by av eq. For a scalar particle,
the ansatz of w3x is
v 2
av eys vy3r2 dt d V exp y
H ½ Ž t q pr2. q < x q e < 2
2
5
yi v Ž t q pr2 . O Ž t ,x . ,
v 2
av eqs vy3r2 dt d V exp y
H ½ Ž t y pr2. q < x y e < 2
2
5
qi v Ž t y pr2 . O Ž t ,x . , Ž 1.1 .
where O is the appropriate operator corresponding to the scalar field w9–11x. For the
dilaton, it is proportional to tr F 2 , for the axion, it is proportional to tr F n F. We
assumed that O is properly normalized, so some v independent numerical factors in w3x
were dropped. The integral Hd V is over the unit S 3 which is parameterized by x.
The above ansatz clearly indicates that the incoming Žoutgoing. particle originates
Žends up. at time ypr2 Žpr2.. In the large R limit, the proper time goes to "`. The
Gaussian factor helps to focus the beam in the direction e. A S-matrix element is given
by
N ™`
S s lim Fy1 ¦Ł ai
v e iy Ł av e q
j
j ;, Ž 1.2 .
where F is a normalization factor. For a fixed string coupling constant, due to the
relation R 4 s 4p g s Na X 2 , the large R limit is achieved by taking the large N limit.
M. Li r Nuclear Physics B 568 (2000) 195–207 197
In a conformal field theory, both the two-point functions and three-point functions are
fixed up to a numerical coefficient by conformal symmetry. One would expect that the
calculation of the two-point amplitudes and three-point amplitudes using Eqs. Ž1.1.,
Ž1.2. is a simple matter. Actually, as shown in Section 2, an exact form of the
three-point amplitude can be obtained only after some tedious calculations. In this
section we will be content with a qualitative examination of these amplitudes.
The geometry S 3 = R is conformal to R 4 , so correlation functions on S 3 = R can be
obtained from those on R 4 using the conformal transformation. For instance, the
Euclidean distance between the two points x and y, r 2 s < x y y < 2 is mapped to
exp Ž t x q t y . Ž cosh Ž t x y t y . y x P y . , Ž 1.3 .
where we parametrize R 4 by the radial coordinates r s expt and the unit sphere S 3. x
and y are unit four-vectors. Now t and x parametrize S 3 = R.
In a correlation function, the extra factors such as expŽt 1 q t 2 . are removed by a
conformal factor. If F Ž ri2j . is a correlation function on R 4 , then the corresponding
correlation function on S 3 = R is obtained by simply replacing ri2j with coshŽt i y t j . y
x i P x j . To obtain the correlation function on a Minkowskian S 3 = R, we wick-rotate t
to it, and add a term i e :
ri2j scos Ž t i y t j . y x i P x j q i e , Ž 1.4 .
where the i e prescription is introduced to ensure causality in the boundary conformal
theory. To see this, assume t i y t j small, and the angle f i j between x i and x j small, we
obtain
2
y 12 Ž t i y t j . q 12 f i2j q i e , Ž 1.5 .
we see that if one uses ry2
i j as the propagator, the signal will propagate along the future
light cone for a positive energy mode of the form expŽyi v t ..
The scaling dimension of an operator corresponding to a massless scalar field is
D s 4. The two-point function is therefore
² O Ž t 1 ,x 1 . O Ž t 2 ,x 2 . : s Ž cos Ž t 1yt 2 . yx 1 P x 2qi e . y4 Ž 1.6 .
up to a normalization constant.
The two-point scattering amplitude is obtained using Ž1.2.. Without taking the large
N and high energy limit, there is no energy–momentum conservation, since as shown in
the appendix of w3x, the incoming an well as outgoing waves have finite width in both
energy and momentum. The width of v is proportional to 'v . Using v s RE, the width
'
of E is proportional to ErR and goes to zero in the large R limit. So energy–
momentum conservation has to be recovered in this limit. We shall show in the next
section that the energy conservation is always guaranteed in this limit.
However, momentum conservation rather imposes strong constraints on the behavior
of conformal correlators in the large N limit. As we will see in Section 2, the
convolution of Ž1.2. using ansatz Ž1.1. is rather subtle. To obtain the exact numerical
answer, one cannot simply replace the Gaussian distribution of Ž1.1. by a simpler one,
say a delta function. To see the momentum conservation, though, we will do this in this
section.
In the case of the two-point amplitude, replacing the Gaussian wave packets by delta
functions, we obtain
y4
Ž 1 y e1 P e 2 . . Ž 1.7 .
Together with a factor depending on v i , we will have a null result if e 1 P e 2 / 1. The
above expression is singular if e 1 P e 2 s 1 or equivalently e 1 s e 2 . Thus we hope that a
more careful calculation will result a delta function d 3 Že 1 y e 2 . for the two-point
amplitude. Here we define the delta function by
3
Hd Ž e 1 y e 2 . d V 2 s 1.
Together with the energy conservation v 1 s v 2 , or E1 s E2 , this implies the momentum

conservation E1 e 1 s E2 e 2 . The conservation is due to the fact that the two-point
conformal correlation function is singular if one point sits on the future light cone of the
other point.
The three-point correlation function of operator O is fixed by conformal invariance.
Again up to a constant, it is
y2
O Ž t 1 ,x 1 . O Ž t 2 ,x 2 . O Ž t 3 ,x 3 . :s Ž cos Ž t 1 y t 2 . y x 1 P x 2 q i e .
langleO ,
y2 y2
Ž cos Ž t 2 y t 3 . y x 2 P x 3 q i e . Ž cos Ž t1 y t 3 . y x 1 P x 3 q i e . . Ž 1.8 .
Consider the case where two states are incoming, one is outgoing. The Gaussian
wave packets force t 1,2 to center around ypr2, and t 3 to center around pr2. If we
simply replace the Gaussian wave packets by delta functions, the above expression
becomes
y2 y2 y2
Ž 1 y e1 P e 2 . Ž1ye2 Pe3. Ž 1 y e1 P e 3 . . Ž 1.9 .
This function is singular whenever e i s e j for i / j. It is most singular when all e i are
equal. Thus we expect that a more careful treatment will lead to delta functions forcing
all e i to be equal. Indeed this is the consequence of momentum conservation for a
three-point amplitude involving only massless particles: that all momenta of outlegs
must be collinear. This is most easily seen in the following physical way. If, say, the two
incoming states are not collinear, then one can go to the center of mass frame in which
the end product of the scattering can never be a single massless particle.
Mathematically, the above result, as in the two-point amplitude case, is a conse-
quence of causality in the boundary theory: Whenever two points are separated by a null
geodesics, then the correlation function becomes singular. For the three-point function, it
is most singular when they lie on the same light cone.
This raises a puzzle already at the level of the three-point amplitudes. Imagine that
there are stable massive particles. In this case the three-point correlation is still given by
a formula similar to Ž1.8. if one replaces 2 by Ž Di q D j y Dk .r2, if the particles carry
zero momentum in the internal space S 5. Again the correlation function becomes
singular when e i and e j are equal, provided Di q D j ) Dk . However, this has nothing to
do with the momentum conservation for massive particles. It is therefore quite interest-
ing to note that there are no stable massive stringy states in the type IIB theory. Thus
this puzzle is avoided. A related feature is that the conformal dimension of a stringy
state is divergent in the large N limit w9,10x, thus although the convolution Ž1.2. exists
for finite N, its large N limit does not exist.
It is more interesting to see what happens to the four-point correlation function when
the momentum conservation is imposed. The four-point correlation function of operator
O of scaling dimension 4 can not be fixed by conformal symmetry alone. Up to a
scaling factor, it is a function of two independent cross-ratios. Defining the cross-ratios
2 2 2 2
r 12 r 34 r 12 r 34
as 2 2
, bs 2 2
, Ž 1.10 .
r 13 r 24 r 23 r 14
the correlation function can be written as, for instance,

² O Ž t 1 ,x 1 . . . . O Ž t 4 ,x 4 . s F4 Ž ri2j . s Ł ry8r3
ij f Ž a,b . , Ž 1.11 .
i-j(4
where f is an undetermined function of a and b. All ri2j are as given in Ž1.4..

Unlike in some 2D conformal field theories, f Ž a,b . is not constrained in SYM as far
as we know, since no other non-trivial symmetries extending conformal invariance have
been discovered by far. The scaling factor in Ž1.11. is singular whenever two points are
separated by a null geodesics. f Ž a,b . can be singular too in this case, since one of a and
b vanishes or becomes infinity. Unlike for the two- and three-point amplitudes,
energy–momentum conservation in general does not require two points being on the
same light cone. To see the general consequence, we follow the same strategy as the
above to replace ri2j by their ‘‘on-shell’’ value:
2
r 12 s 1 y e 1 P e 2 s 2sin2 Ž f 12r2 . , 2
r 34 s 2sin2 Ž f 34r2 . ,
2
r 13 s y1 q e 3 P e 4 s y2sin2 Ž f 13r2 . , 2
r 24 s y2sin2 Ž f 24r2 . ,
2
r 23 s y2sin2 Ž f 23r2 . , 2
r 14 s y2sin2 Ž f 14r2 . , Ž 1.12 .
where we assumed that particles 1 and 2 are incoming, and particles 3 and 4 are
outgoing; f i j is the angle between momentum k i and momentum k j .
Now use the energy conservation E1 q E2 s E3 q E4 and the momentum conserva-
tion law E1 e 1 q E2 e 2 s E3 e 3 q E4 e 4 , we derive
s s 4 E1 E2 sin2 Ž f 12r2 . s 4 E3 E4 sin2 Ž f 34r2 . , Ž 1.13 .
where s is one of the Mandelstam variables. Similarly
t s y4E1 E3 sin2 Ž f 13r2 . s y4E2 E4 sin2 Ž f 24r2 . ,
u s y4E1 E4 sin2 Ž f 14r2 . s y4E2 E3 sin2 Ž f 23r2 . . Ž 1.14 .

We see that those ‘‘on-shell’’ distances in Ž1.12. are simply related to the Mandelstam
variables. However, ri2j also depends on individual energies. It appears that the only way
to eliminate the dependence on energies is to use the two cross-ratios
s2 s2
as , bs . Ž 1.15 .
t2 u2
Now since Mandelstam variables satisfy relation s q t q u s 0, we obtain

1 1
q s 1. Ž 1.16 .
'a 'b
In other words, the above relation is a consequence and energy–momentum conserva-
tion.
Our experience with the two- and three-point amplitudes tells us that in order for the
four-point scattering amplitude to obey energy–momentum conservation, the four-point
correlation function must be singular when Ž1.16. is satisfied. The correlation is also
singular whenever one of ri2j vanishes. As already explained, ri2j s 0 has nothing to do
with energy–momentum conservation, the singularity of f Ž a,b . at 1r 'a q 1r 'b s 1
must be more severe than the singularity of the correlation at ri2j s 0.
The new singularity we observed above is not dictated by conformal invariance at all.
Also, this singularity is demanded only in the large N limit, since for finite N there is
no momentum conservation in the anti-de Sitter space. As we will explain in detail in
Section 2, f Ž a,b . must be singular if
2
4 ab s Ž ab y a y b . , Ž 1.17 .
and Ž1.16. is only one of the four solutions
1 1
" s "1 Ž 1.18 .
'a 'b
to the above equation.
2. Explicit calculations
We will see that an explicit calculation based on the ansatz of w3x is quite difficult.
We will be able to obtain a closed form for the two-point amplitude, we will not be able
to complete the calculation of the three-point amplitude. However, we will show that
energy–momentum conservation is ensured in the large R limit. We discuss the
calculation of the four-point amplitude. A lot of work is left for the future.
™ First, we want to slightly simplify the calculation of a general amplitude. Shifting

t t . pr2, for an incoming particle or an outgoing one, one can always ignore the
quadratic term in t in ansatz Ž1.1., if one remembers that the integral over t will always
pick up the most important contribution around t ; 0. To see this, consider the integral
v
y t 2 qi v t
Hdt e 2 f Ž t, . . . . , Ž 2.1 .
where the dots denote other variables. f Ž t . is a periodic function of t of period 2p .

Expressing f Ž t, . . . . as
f Ž t , . . . . s d v f˜Ž v , . . . . eyi v t ,
H
we have
2p
Hdt e y
v
2
t 2 qi v t
f Ž t, . . . . s ( v
Hd v X
ey Ž v y vX .2 v f˜Ž v X , . . . . . Ž 2.2 .
The integral over v X centers around v with width dv X ; 'v , compared with the
principle value v , this deviation tends to zero in the large v limit. Thus the above
integral is approximately
2p f˜Ž v , . . . . s dt e i v t f Ž t , . . . . .
H Ž 2.3 .
Compared with this approximate value, the deviation is about
df˜Ž v , . . . .
dv , Ž 2.4 .
dv
so it can be ignored if
f˜X Ž v , . . . . rf˜Ž v , . . . . < 1r'v .
This condition is generally satisfied. Alternatively, as we already saw, the function
f Ž t, . . . . has infinitely many poles. Integral over t will pick up these poles, the Gaussian
factor expŽyŽ vr2. t 2 . helps to suppress all these poles except the one closest to zero. In
this case, the other factor expŽ i v t . is more important than the Gaussian factor, since it
oscillates fast around t ; 1r 'v .
To demonstrate the above result, let us see how the energy conservation is derived in
the large v limit. For a general scattering amplitude, we have
1 2
H Łi dt exp y Ý v t
i 2 i i qi Ý ai vi ti f Ž ti , . . . . , Ž 2.5 .
i i
where a i s 1 for an outgoing state, and it is y1 for an incoming state. Using a new set
of times: t 1 , t i s t i q t 1 , i s 2, . . . ,n, the function f Ž t i , . . . . is a function of t i only,
since the correlation function in SYM is invariant under a time translation. Performing
the integral over t 1 first, we obtain the integrand for t i ,
2p 1 1
( v
exp y
2v
Ž Ý ai vi .
2
y
2v
Ý v it i
2
y 12 Ý v it i2
qi Ý a i v it i f Ž t i , . . . . , Ž 2.6 .
where v s Ý v i , and we omitted a term

i
y
v
Ž Ý ai v i .Ž Ý v it i .
in the exponential. This is because the first exponential together with the prefactor gives
rise to a delta function in the large R limit,
2p
d Ž Ý a i Ei . , Ž 2.7 .
R
which is just the energy conservation law. This factor can be obtained without including
the Gaussian factors. The remaining Gaussian factor in Ž2.6. is positive definite in t i ,
and as we argued before, can be ignored so long if we pick up poles closest to zero in t i .
The integral over variables on S 3 is much more complicated, and since the Gaussian
factor expŽyŽ v ir2.<x i " e i < 2 . is the only non-trivial factor in the convolution, one has
to treat the convolution carefully. In the following, we will examine the two-, three- and
four-point amplitudes separately. Before doing that, let us remark that this Gaussian
factor can be replaced by
vi
< x i"e i < 2
d4 ki 2 v i.qi k i Ž e i"x i .
ey 2 s H Ž 2pv . 2
eyŽk 2 i
i
d4 ki 2 .qi v i k iP Ž e i"x i .
s H Ž 2p . 2
eyŽk 2 i ' . Ž 2.8 .
2.1. Two-point amplitude
As shown above, the integration over the ‘‘center of times’’ results in a factor
2p
d Ž E1 y E2 . . Ž 2.9 .
R
The remaining part is
y3
d 4 k1 d4 k2
F y1
N2 Ž RE1 . H Ž 2p . 2 2
exp yk 12 y 12 k 22 q i v 1 ( Ý k Pe i i F2 Ž k i . ,
Ž 2p .
Ž 2.10 .
with
F2 Ž k i . s dt d V 1 d V 2exp yi v 1 t y i v 1
H ( Ý a k Px i i i
y4
= Ž cost y x 1 P x 2 y i e . , Ž 2.11 .
where in Eq. Ž2.10. we introduced a normalization factor N2 depending on the
normalization of the operator O .
The integral over t in Ž2.11. can be performed first. It picks up a pole at
t s f 12 y i e , where x 1 P x 2 s cos f 12 . Other poles are suppressed by a Gaussian factor
we have omitted. The leading contribution is
p 1
F2 Ž k i . s v 13 d V 1 d V 2 exp yi Vf 12 y i v 1
H ( Ý a k Px i i i 4
,
3 sin Ž f 12 y i e .
Ž 2.12 .
other terms are suppressed by powers of v 1. Performing the integration over V 2 first,
we have
8p 3 32p 4 p J1 v 1 < k 1 y k 2 <
ž( /.
F2 s v 14 d V 1 e i'v 1 Ž k 2yk 1 .Px 1 s
H v 14 Ž 2.13 .
3 3 (v < k y k <
1 1 2
We will see momentarily that the fact that the above result is a function of only k 1 y k 2
ensures the momentum conservation in the large R limit.
Plugging back the above result into Ž2.10., ignoring the prefactor in Ž2.10. for the
moment, we have, after changing variables k 1 y k 2 s k 2r2, k 1 q k 2 s k 1r2
2 5p y k 12 qi v 1 k 1 P Ž e 1 ye 2 .
d 4k 2 2
v 14 4
Hd k e 1
' H 2(v ey k 2 qi'v 1 k 2PŽ e 1qe 2 . J1 2 v 1 < k 2 < .
ž( /
3 <
1 k2
Ž 2.14 .
The integral over k 1 is separated from that over k 2 . The first integral results in
2
p 2 eyv 1 < e 1ye 2 <
which in the large v limit tends to
3r2
p
p2 ž / v1
d 3 Ž e1 y e 2 . . Ž 2.15 .
Due to the delta function, the second integral in Ž2.14. is simplified. The integral over
k 2 can be separated into the radial part and the angular part, and the latter can be easily
performed. In the end, we obtain
2 8p 5 3r2 2
Ž pv 1 . d 3 Ž e 1 y e 2 . dk k J12 2 v 1 k eyk .
H ž( / Ž 2.16 .
3
In the large v 1 limit, the Bessel function J1 can be replaced by its asymptotic form,
namely
`
2
H0 dk k J ž 2(v k / e
1 1
yk 2
™ p(1v 1
2
Hdk cos ž 2(v k / e 1
yk 2
,
and since for a large v 1 , the cos factor can be replaced by its average value 1r2, the
(
value of the above integral is Ž1r4.Ž1r pv 1 .. We have checked this result by using a
formula for the integral in Ž2.16. involving the Bessel function. Substitute this into
Ž2.16., we find
6
Ž 2p .
v 1 d 3 Ž e1 y e 2 . . Ž 2.17 .
3
Together with the prefactor in Ž2.10. and the delta function in Ž2.9., the end result is
7
Ž 2p .
² av a
1 e 1y v 2 e 2q
: s Fy1 N2 E1y2d Ž E1 y E2 . d 3 Ž e 1 y e 2 . . Ž 2.18 .
3 R3
Using the identity

d 4 Ž E1 e 1 y E2 e 2 . s E1y3d Ž E1 y E2 . d 3 Ž e 1 y e 2 .
it follows that
7
Ž 2p .
² av
1 e 1y
av 2 e 2 q sF
: y1
N2 E1 d 4 Ž E1 e 1yE2 e 2 . . Ž 2.19 .
3 R3
Conservation of momentum for the two-point amplitude also ensures conservation of
energy, so we have obtained the right delta function, as expected by our intuitive
argument in Section 1. It remains to check whether the kinetic factor is also right. Note
that the first normalization factor in Ž2.19. depends on the overlap of the two wave
functions w3x, so it can depend on the energy. The other normalization factor, N2 , does
not depend on energy, although it can be a function of R and g s .
The wave function of w3x takes the form
v 2 2.
F Ž t , x . s eyi v Ž tyeP x .y 2 Ž x Hq Ž tyePx . Ž 2.20 .
near the center of AdS space. The Gaussian factor can be ignored so long if the
space-time region has a scale smaller than 1r 'v . The proper scale is Rr 'v ; 'R . So
in the large R limit the Gaussian factor can be ignored and we have a plane wave. Since
the creation operator is defined by
ˆ t FŽ t , x . q . . . ,
a s dV fE
H
so close to the center of the AdS space we have, roughly
d4 k
f̂ Ž t , x . s H E
Ž aq Ž k . eyi E tqi kP x q a Ž k . e i E tykP x . , Ž 2.21 .
where all the coordinates are the proper ones, unlike the ones in Ž2.20.. Thus we expect
that the scattering amplitude, up to a numerical factor, must be
² av
1 e 1y
av 2 e 2q : s R 5 E1 d 4 Ž E1 e 1 y E2 e 2 . , Ž 2.22 .
where the volume factor R 5 comes from the internal space S 5, since the particles have
zero momenta in the internal space. The kinetic factor is precisely the same as in Ž2.19..
Therefore, it appears that the normalization F is order 1, and N2 ; R 8 ; N 2 g s2 .
2.2. Three-point amplitude
Consider the three-point amplitude with two incoming particles. It is more convenient
to integrate out t 3 first, with t 1 s t 1 q t 3 , t 2 s t 2 q t 3 . As before a delta function
ensuring energy conservation results. The amplitude is
2p d4 ki
A 3 s N3 d Ž E1 q E2 y E3 . Ł vy3r2
i H Łi Ž 2p . 2
R i
= exp y 12 Ý k i2 q i Ý v i k i e i F3 Ž k i . ,
( Ž 2.23 .
with
y2
F3 Ž k i . s dt 1 dt 2 Ł d V i eyi v it iqi'v i k iPx icos Ž t 1 y t 2 . y x 1 P x 2 q i e .
H i
y2 y2
= cos Ž t 1 y x 1 P x 3 y i e . cos Ž t 2 y x 2 P x 3 y i e . , Ž 2.24 .
™
where we have reflected x 3 yx 3 . Denote x i P x j by cos f i j . To perform the integral
over t i first, we use the following formula:
y2
1 `
i vt " i < v <Ž f " i e .
Ž cost y cos f " i e . sy 2 H dv < v < e
sin Ž f " i e . y`
, Ž 2.25 .
which is derived from
y1
i i vt " i < v <Ž f " i e .
Ž cost y cos f " i e . s. Hd v e Ž 2.26 .
sin Ž f " i e .
by taking derivative with respect to f once. With the Fourier transform Ž2.25., the
integral over t i in Ž2.24. is readily performed, with the result
ž(
exp i v i k i P x i /
F3 Ž k i . s y H Łi d V sin Ž f i 2
12 q i e . sin Ž f 13 y i e . sin2 Ž f 23 y i e .
2
= d v < v = Ž v 1 y v . Ž v 2 q v . < e i < v < f 12 yi < v 1yv < f 13 yi < v 2qv < f 23 < .
H
Ž 2.27 .
We do not know how to carry out the calculation of the above integral. One thing is
certain, though: due to the singular behavior of the integrand, the integral is peaked
around f 12 s f 13 s f 23 s 0. We thus expect that F3 will be a function of Ý v i k i only. (
We now argue that this ensures the momentum conservation in the large R limit.
Introduce new vectors
1
l1 s
(2 v Ý (v k ,
3
i i
l s y(v rv k q (v v k ,
2 2 3 1 1 3 2
l s (v r Ž 2 v . k q (v r Ž 2 v
3 1 3 1 2 3 . k 2 y 1r'2 k 3 . Ž 2.28 .
With the above relations, it is easy to see that Ýk i2 s Ýl i2 , using the fact v 3 s v 1 q v 2 .
Now F3 is a function of l 1 only. We can perform the integral in Ž2.23. over l 2 and l 3
first. We note in particular that in the exponential in Ž2.23., the l 3 dependent part is
l3
y 12 l 32 q i
(2 v Ž Ý a v e . .
3
i i i Ž 2.29 .
It is seen that the integral over l 3 results in a delta function ensuring momentum
conservation, in the large R limit.
After some calculations, we obtain
2 11r2p 5r2 E32 d 4 l1 1 2

A 3 s N3
E13 E23
d Ž Ý a i Ei . d 4
Ž Ý ai Ei e i . H Ž 2p . 2 ey l qi'2 v
2 1 1 l 1 Pe 1 F3 Ž l 1 . .
Ž 2.30 .
2.3. Four-point amplitude
We have much less to say about the four-point amplitude. Although we trust that the
condition for energy–momentum conservation derived in the previous section is neces-
sary, we are not able to prove that it is also sufficient. As in the two- and three-point
amplitude cases, we can always write the four-point amplitude as
4 d4 ki
A 4 s N4 H is1
Ł 2
exp y 12 Ý k i2 q i Ý v i k i P e i F4 Ž k i . ,
( Ž 2.31 .
Ž 2p .
where F4 is given by a similar formula as Ž2.24.. We omitted a factor conserving
energy.
To see whether the momentum conservation is true in the large R limit, as in
Subsection 2.2, we introduce a set of new vectors
li s Ý Vi j k j , Ž 2.32 .
j
(
with V i j 4 being an orthogonal matrix. We can choose V 1 i s a i v irv , where v s Ý i v i .
If, in the large R limit, F4Ž l i . is less dependent on l 1 than on other l i , the integral over
l 1 in Ž2.31. can be performed first, thus resulting in a delta function associated with
momentum conservation.
It can be shown that the condition that 1r 'a q 1r 'b s 1 is a singularity of f Ž a,b .
is a necessary one, where f Ž a,b . is a function introduced in Ž1.11.. This can be seen
using a formula similar to Ž2.24.. Independence of various components can be used to
derive relations in Ž1.13. and Ž1.14., thus the above condition. This is just a single
condition, so it is far from clear whether it is also a sufficient condition for conserving
four-momentum.
Finally, we want to show that if 1r 'a q 1r 'b s 1 is a singularity of f Ž a,b ., then
both 1r 'a y 1r 'b s 1 and 1r 'a y 1'b s y1 are also singularities of f Ž a,b ..
f Ž a,b . is a symmetric function of a and b. To see this, we go to the Euclidean space.
™ ™
Exchange point 1 and point 2, a and b are exchanged. Exchange point 2 and point 3 we
are led to a 1ra, b bra, thus
F Ž 1ra,bra . s F Ž a,b . s F Ž b,a . . Ž 2.33 .

'
Now, 1r 'a q 1r 'b s 1 is the same as 'a y arb s 1. This means that f Ž1ra,bra.
is singular when this relation is satisfied. Renaming the variables, we conclude that
f Ž a,b . is singular when 1r 'a y 1r 'b s 1. Exchanging a with b, we deduce that
f Ž a,b . is singular when 1r 'a y 1r 'b s y1.
3. Conclusion
We have only scratched the surface of the problem of investigating the consequence
of Lorentz invariance for the holographic S-matrix and associated correlation functions
in the large N limit. For instance, more constraints can be derived from the requirement
that A 4 is a function of only s and t, apart from a kinetic factor. Even more interesting,
is the consequence of causality in the flat space limit. We leave these problems for
future investigations.
Our main result in this paper is the identification of a new singularity in the
four-point amplitude, in the large N limit. This means that the dominant contribution to
the scattering amplitude comes from around this ‘‘saddle point’’. This reminds us the
problem of sensitive initial conditions raised in w7x. It is observed there that if the two
beams aimed at the center of the AdS are emitted with time difference greater than 1rR,
then the beams will miss each other. In the large N limit, this time difference can be
arbitrarily small. It appears that a kind of sharp saddle point may help to understand this
puzzle. Presumably this is a consequence of locality in the bulk space. It remains to see
whether bulk locality together boundary conformal invariance guarantee bulk Poincaré
invariance in the large N limit.
Acknowledgements
This work was supported by DOE grant DE-FG02-90ER-40560 and NSF grant PHY
91-23780. We are grateful to R. Siebelink for collaboration during the initial stage of
this project. Useful discussions with P. Kraus have helped me going through some
difficult phases. Correspondence with J. Polchinski is acknowledged. We are also
grateful to P-M. Ho for helpful discussions. This work was completed during a visit to
National Taiwan University. We thank P-M. Ho, W-Y. P. Hwang and Y-C. Kao for
hospitality.
References
w1x G. ’t Hooft, Dimensional reduction in quantum gravity, hep-thr9310026.

w2x L. Susskind, The world as a hologram, hep-thr9409089.
w3x J. Polchinski, S-matrices from AdS space-time, hep-thr9901076.
w4x L. Susskind, Holography in the flat space limit, hep-thr9901079.
w5x V. Balasubramanian, S.B. Giddings, A. Lawrence, What do CFTs tell us about anti-de Sitter space-times,
hep-thr9902052.
w6x S.B. Giddings, The boundary S-matrix and the AdS to CFT dictionary, hep-thr9903048.
w7x J. Polchinski, L. Susskind, N. Toumbas, Negative energy, superluminosity and holography, hep-
thr9903228.
w8x J.M. Maldacena, The large N limit of superconformal field theories and supergravity, hep-thr9711200.
w9x E. Witten, Anti De Sitter space and holography, hep-thr9802150.
w10x S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from non-critical string theory,
hep-thr9802109.
w11x V. Balasubramanian, P. Kraus, A. Lawrence, Bulk vs. boundary dynamics in anti-de Sitter space-time,
hep-thr9805171.
Factorization of tree QCD amplitudes in the high-energy limit

and in the collinear limit
Vittorio Del Duca a , Alberto Frizzo b, Fabio Maltoni b
a
Istituto Nazionale di Fisica Nucleare, Sezione di Torino, Õia P. Giuria, 1, 10125 Turin, Italy
b
Dipartimento di Fisica Teorica, UniÕersita` di Torino, Õia P. Giuria, 1, 10125 Turin, Italy
Received 29 September 1999; accepted 11 October 1999
Abstract
In the high-energy limit, we compute the gauge-invariant three-parton forward clusters, which
in the BFKL theory constitute the tree parts of the NNLO impact factors. In the triple collinear
limit, we obtain the polarized double-splitting functions. For the unpolarized and the spin-corre-
lated double-splitting functions, our results agree with the ones obtained by Campbell–Glover and
Catani–Grazzini, respectively. In addition, we compute the four-gluon forward cluster, which in
the BFKL theory forms the tree part of the NNNLO gluonic impact factor. In the quadruple
collinear limit we obtain the unpolarized triple-splitting functions, while in the limit of a
three-parton central cluster we derive the Lipatov vertex for the production of three gluons,
relevant for the calculation of a BFKL ladder at NNLL accuracy. Finally, motivated by the
reorganization of the color in the high-energy limit, we introduce a color decomposition of the
purely gluonic tree amplitudes in terms of the linearly independent subamplitudes only. q 2000
PACS: 12.38.-t; 12.38.Bx; 13.85.Rm

Keywords: Perturbative QCD; BFKL; High energy
1. Introduction
QCD calculations of multijet rates beyond the leading order ŽLO. in the strong
coupling constant a s are generally quite involved. However, in recent years it has
become clear how to construct general-purpose algorithms for the calculation of multijet
rates at next-to-leading order ŽNLO. accuracy w1–11x. The crucial point is to organize
the cancellation of the infrared Ži.e. collinear and soft. singularities in a universal, i.e.
process-independent, way. The universal pieces in a NLO calculation are given by the
tree-level splitting w12–15x and eikonal w16–18x functions, and by the universal structure
of the poles of the one-loop amplitudes w1,4,19x.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 5 7 - 4
212 V. Del Duca et al.r Nuclear Physics B 568 (2000) 211–262
Eventually, the same procedure will permit the construction of general-purpose

algorithms at next-to-next-to-leading order ŽNNLO. accuracy. It is mandatory then to
fully investigate the infrared structure of the phase space at NNLO. The universal pieces
needed to organize the cancellation of the infrared singularities are given by the
tree-level double-splitting w20,21x, double-eikonal w18,22x and splitting-eikonal w20,22x
functions, by the one-loop splitting w23–27x and eikonal w23–25x functions, and by the
universal structure of the poles of the two-loop amplitudes w28x.
Another outstanding issue in QCD, at first sight unrelated to the topics discussed
above, is the calculation of the higher-order corrections to the BFKL equation w29–31x.
In scattering processes characterized by two large and disparate scales, like s, the
squared parton center-of-mass energy, and t, a typical momentum transfer, the BFKL
equation resums the large logarithms of type lnŽ srt .. The LO term of the resummation
requires gluon exchange in the cross channel, which for a given scattering occurs at
O Ž a s2 .. The corresponding QCD amplitude factorizes then into a gauge-invariant
effective amplitude formed by two scattering centers, the LO impact factors, connected
by the gluon exchanged in the cross channel. The LO impact factors are characteristic of
the scattering process at hand. The BFKL equation resums then the universal leading-
logarithmic ŽLL. corrections, of O Ž a sn ln n Ž srt .., to the gluon exchange in the cross
channel. The building blocks of the BFKL resummation are the Lipatov vertex w32,33x,
i.e. the effective gauge-invariant emission of a gluon along the gluon ladder in the cross
channel, and the gluon reggeization w29x, i.e. the LL part of the one-loop corrections to
the gluon exchange in the cross channel.
The accuracy of the BFKL equation is improved by computing the next-to-leading
logarithmic ŽNLL. corrections w34–37x, i.e. the corrections of O Ž a sn ln ny1 Ž srt .., to the
gluon exchange in the cross channel. In order to do that, the universal building blocks of
the BFKL ladder must be computed to NLL accuracy. These are given by the tree
corrections to the Lipatov vertex, i.e. the emission of two gluons w38–40x or of a qq pair
w40,41x along the gluon ladder, by the one-loop corrections to the Lipatov vertex
w42–46x, and finally by the NLL gluon reggeization w47–50x, i.e. the NLL part of the
two-loop corrections to the gluon ladder. However, to compute jet production rates at
NLL accuracy, the impact factors must be computed at NLO w51–53x. For jet production
at large rapidity intervals, they are given by the one-loop corrections w51x to the LO
impact factors, and by the tree corrections w38,39,41,54x, i.e. the emission of two partons
in the forward-rapidity region. In the collinear or soft limits, the latter reduce to the tree
splitting or eikonal functions w55x.
To further improve the accuracy of the BFKL ladder one needs to compute the
next-to-next-to-leading logarithmic ŽNNLL. corrections, i.e. the corrections of
O Ž a sn ln ny2 Ž srt .., to the gluon ladder. At present it is not known whether such
corrections can be resummed. If that is the case, the universal building blocks of a
BFKL ladder at NNLL would be: the emission of three partons along the gluon ladder,
the one-loop corrections to the emission of two partons along the ladder, the two-loop
corrections to the Lipatov vertex, and the gluon reggeization at NNLL accuracy. None
of them is known at present. In this paper we compute the gluonic NNLO Lipatov
vertex, i.e. the emission of three gluons along the ladder.
In addition, to compute jet production rates at NNLL accuracy, the BFKL ladder
should be supplemented by impact factors at NNLO. They are not known either. In this
V. Del Duca et al.r Nuclear Physics B 568 (2000) 211–262 213
paper we compute their tree components, i.e. the emission of three partons in the
forward-rapidity region. By taking then the triple collinear limit of the tree NNLO
impact factors, we obtain the polarized double-splitting functions. Summing over the
parton polarizations, we obtain the unpolarized and the spin-correlated double-splitting
functions, previously computed in Ref. w20x and w21x, respectively, in the conventional
dimensional regularization ŽCDR. scheme. Since we sum over two helicity states of the
external partons, as it is done in the dimensional reduction ŽDR. scheme w56–58x, our
results agree with the ones in the CDR scheme by setting there the dimensional
regularization scheme ŽRS. parameter e s 0.
For a scattering with production of m partons, we define the n-parton cluster, with
m ) n, as the set of n final-state partons where the distance in rapidity between any two
partons in the cluster is much smaller than the rapidity distance between a parton inside
the cluster and a parton outside. In the BFKL theory, Ž n q 1.-parton forward clusters
provide the tree parts of N n LO impact factors, while Ž n q 1.-parton central clusters
provide the tree parts of the N n LO Lipatov vertex. n-parton clusters were given also a
field-theoretical basis in terms of an effective action describing the interaction between
physical partons grouped into gauge-invariant clusters and the gluons exchanged in the
cross channel w59x. In addition to computing the three-parton forward clusters and the
three-gluon central cluster mentioned above, we compute the four-gluon forward cluster,
i.e. the purely gluonic tree part of the NNNLO impact factor. By taking then the
quadruple collinear limit, we obtain the polarized triple-splitting functions. They could
be used in a gauge-invariant evaluation of the Altarelli–Parisi evolution at three loops
w60x.
The outline of the paper is: in Section 2 we review the standard color decompositions
of the n-parton tree amplitudes, and we present a color decomposition of the gluon
amplitudes in terms of the linearly independent subamplitudes only. In Section 3 we
review the elastic scattering of two partons in the high-energy limit, which allows for the
extraction of the LO impact factors. In Section 4 we review the amplitudes for the
production of three partons, with a gauge-invariant two-parton forward cluster; from
these, we can extract the tree parts of the NLO impact factors; by taking the collinear
limit, we obtain the LO splitting functions. In Section 5 we compute the amplitudes for
the production of four partons, with a three-parton forward cluster; then we extract the
tree parts of the NNLO impact factors, and by taking the triple collinear limit we obtain
the polarized and unpolarized double-splitting functions. In Section 6.1 we compute the
amplitude for the production of five gluons, with a four-gluon forward cluster. We
extract the tree part of the gluonic NNNLO impact factor, and by taking the quadruple
collinear limit we obtain the polarized triple-splitting functions. In addition, by taking
the limit in which three gluons are emitted in the central-rapidity region, we obtain the
gauge-invariant three-gluon central cluster, i.e. the tree part of the NNLO Lipatov
vertex. In Section 7 we draw our conclusions.
2. Tree amplitudes
In this section we review the color decomposition of purely gluonic and quark–gluon
tree amplitudes. For the purely gluonic tree amplitudes, we introduce a color decomposi-
tion in terms of the linearly independent subamplitudes, Eq. Ž2.9..
2.1. Gluon amplitudes
For an amplitude with n gluons the usual color decomposition at tree level reads
w61–69x
i A Ž g 1 , . . . , g n . s ig ny 2 Ý tr Ž ls1 . . . ls n . A Ž gs 1 , . . . , gs n . , Ž 2.1 .
sgS nrZ n
where SnrZ n are the non-cyclic permutations of n elements. The dependence on the
particle helicities and momenta in the subamplitude, and on the gluon colors in the trace,
is implicit in labeling each leg with the index i. Helicities and momenta are defined as if
all particles were outgoing.
The gauge invariant subamplitudes A satisfy the relations w68,69x, proven for
arbitrary n in Ref. w70x,
A Ž 1,2, . . . ,n y 1,n . s A Ž n,1,2, . . . ,n y 1 . cyclicity

n
A Ž 1,2, . . . ,n . s Ž y1 . A Ž n, . . . ,2,1 . reflection
A Ž 1,2,3, . . . ,n . q A Ž 2,1, . . . ,n . q . . . qA Ž 2,3, . . . ,1,n . s 0 dual Ward identity
Ž 2.2 .
The above relations are sufficient to show that, for n ( 6 the number of independent
subamplitudes can be reduced from Ž n y 1.! to Ž n y 2.!. For n 0 7 it is still possible to
introduce a basis of Ž n y 2.! elements by using Kleiss–Kuijf’s relation w71x
p
A Ž 1, x 1 , . . . , x p ,2, y 1 , . . . , yq . s Ž y1 . Ý A Ž 1,2, a 4 b 4 . , Ž 2.3 .
s gOP a 4 b 4
where a i g a 4 ' x p , x py1 , . . . , x 14 , bi g b 4 ' y 1 , . . . , yq 4 and OP a 4 b 4 is the set of

permutations of the Ž n y 2. objects x 1 , . . . , x p , y 1 , . . . , yq 4 that preserve the ordering of
the a i within a 4 and of the bi within b 4 , while allowing for all possible relative
orderings of the a i with respect to the bi . The above relation has been checked up to
n s 8 in Ref. w70x, and proven for arbitrary n in Ref. w72x. Accordingly, the expression
for the summed amplitude squared can be written as
Ž ny1 . !
Ý < A Ž 1, . . . ,n . < 2 s Ý c i j A i A)j Ž 2.4 .

a1 , . . . , a n i , js1
1
< A Ž 1, s 2 , . . . , sn . < 2 q O
s Cn Ž Nc . Ý
sgS ny1
ž /Nc2
Ž 2.5 .
Ž ny2 . !
s Ý c˜i j A i A)j , Ž 2.6 .
i , js1
where c i j in Eq. Ž2.4. is

ny2 )
ci j s Ž g 2 . Ý tr Ž Pi Ž ld , . . . , ld . . tr Ž Pj Ž ld , . . . , ld . .
1 n 1 n , Ž 2.7 .
colors
with Pi the ith permutation in SnrZ n . In Eq. Ž2.5., the coefficient CnŽ Nc . is
ny 2
Ž g 2 Nc .
Cn Ž Nc . s
2n
Ž Nc2 y 1 . . Ž 2.8 .
The first term in Eq. Ž2.5. constitutes the Leading Color Approximation ŽLCA.. Up to
n s 5, the 1rNc2 corrections in Eq. Ž2.5. vanish and LCA is exact. The reduced color
matrix c˜i j in Eq. Ž2.6., has been obtained from c i j applying the linear transformations of
Eq. Ž2.3., thus the labels i, j in Eq. Ž2.6. run only on the permutations of the linearly
independent subamplitudes.
Motivated by the reorganization of the color in the high-energy limit w39,54,73,74x,
and using Eqs. Ž2.2. and Ž2.3. we rewrite Eq. Ž2.1. as
i A Ž g1 , . . . , gn .
ny 2
Ž ig .
si Ý f a1 a 2 x 1 f x 1 a 3 x 2 . . . f x ny 3 a ny 1 a n A Ž g 1 , gs 2 , . . . , gs ny 1 , g n .
2 sgS ny2
g ny 2
si
2
Ý Ž Fa 2 . . . F a ny 1 . a1 a n A Ž g 1 , gs 2 , . . . , gs ny 1 , g n . , Ž 2.9 .
sgS ny2
where Ž F a . b c ' if b ac. We have checked Eq. Ž2.9. up to n s 7. Eq. Ž2.9. enjoys several
remarkable properties. Firstly, it shows explicitly which is the color decomposition that
allows us to write the full amplitude i A in terms of the Ž n y 2.! linearly independent
subamplitudes only. In the following we shall refer to it as to a color ladder. Hence the
color matrix obtained squaring Eq. Ž2.9. yields directly the c˜i j matrix in Eq. Ž2.6.. We
have checked it against the explicit results of Ref. w75x, up to n s 5. Moreover, it is quite
suggestive to note the formal correspondence with the amplitudes with a quark–anti-
quark pair and Ž n y 2. gluons, Eq. Ž2.11., where the only difference between the two1 is
the appropriate representation for the color matrices, namely the adjoint for the n-gluon
amplitude and the fundamental for the one with the qq pair. Finally, the most relevant
applications of Eq. Ž2.9. for this work are to the study of the multi-gluon amplitudes in
the high-energy limit. As discussed in the following, the color ladder naturally arises
w73,74x in the configurations where the gluons are strongly ordered in rapidity, i.e. in the
multi-Regge kinematics. Indeed in the strong-rapidity ordering only the subamplitude
with the corresponding order in the color coefficient contributes to Eq. Ž2.9.. At NLO,
where the strong ordering is relaxed for two adjacent gluons, the leading subamplitudes
are the two which differ just by the exchange of the gluon labels in the color ladders
w54x. As we shall see this result generalizes at NNLO and beyond. Nonetheless, in the
1
The factor 1r2 in front of Eq. Ž2.9. is due to our choice for the normalization of the fundamental
representation matrices, i.e. trŽ l al b . s d a b r2.
following we have chosen to derive our results starting from Eq. Ž2.1. instead of using
directly Eq. Ž2.9.. The former, though more laborious, shows explicitly how the color
traces must be recombined to obtain the color ladder and, more importantly, allows us to
find the relations necessary to prove the factorization in the multi-collinear limits.
For the maximally helicity-Õiolating configurations, Žy,y,q, . . . ,q ., in Eq. Ž2.1. or
Eq. Ž2.9., there is only one independent colorrhelicity subamplitude, the Parke–Taylor
ŽPT. subamplitude
² i j :4
A Ž g 1 , . . . , g n . s 2 n r2 , Ž 2.10 .
²1 2: . . . ² Ž n y 1 . n:² n 1:
where the ith and the jth gluons have negative helicity. All other colorrhelicity
amplitudes can be obtained by relabelling and by use of reflection symmetry, Eq. Ž2.2.,
l
and parity inversion. Parity inversion flips the helicities of all particles, and it is
accomplished by the substitution ² i j : w j i x . Subamplitudes of non-PT type, i.e. with
three or more gluons of y helicity have a more complicated structure.
2.2. Quark–gluon amplitudes
For an amplitude with two quarks and Ž n y 2. gluons the color decomposition at tree
level is w61–69x,
ı
i A Ž q,q ; g 1 , . . . , gŽ ny2. . s ig ny2 Ý Ž ls 1 . . . ls ny 2 . j A Ž q,q ; gs 1 , . . . , gs ny 2 . ,
sgS ny2
Ž 2.11 .
where Sny 2 is the permutation group on n y 2 elements.

For the maximally helicity-Õiolating configurations, Žy,y,q, . . . ,q ., there is one
independent colorrhelicity subamplitude, the Parke–Taylor ŽPT. subamplitude
² qi :² qi :3
A Ž qq,qy ; g 1 , . . . , gŽ ny2. . s 2 Ž ny2.r2 , Ž 2.12 .
² qq :² q1: . . . ² Ž n y 2 . q :
where gluon g i has negative helicity. Helicity is conserved along the massless-fermion
line. All other colorrhelicity amplitudes can be obtained by relabelling and by use of
parity inversion, reflection symmetry and charge conjugation. In performing parity
inversion, there is a factor of y1 for each pair of quarks participating in the amplitude.
Reflection symmetry is like in Eq. Ž2.2., for gluons andror quarks alike. Charge
conjugation swaps quarks and antiquarks without inverting helicities. In particular, using
reflection symmetry and charge conjugation on Eq. Ž2.12. we obtain
² qi :3² qi :
A Ž qy,qq ; g 1 , . . . , gŽ ny2. . s 2 Ž ny2.r2 , Ž 2.13 .
² qq :² q1: . . . ² Ž n y 2 . q :
where gluon g i has negative helicity.

For an amplitude with four quarks and Ž n y 4. gluons the color decomposition at tree
level is w61x
i A Ž q1 ,q1 ;q2 ,q2 ; g 1 , . . . , gŽ ny4. .

ny4
ı ı
s ig ny 2 Ý Ý Ý Ž ls 1 . . . l s k . j 2 1 Ž l r 1 . . . l r l . j1 2
ks0 s gS k r gS l
=A Ž q1 ,q1 ;q2 ,q2 ; gs 1 , . . . , gs k ; g r 1 , . . . , g r l .

1 ı ı
y Ž l s 1 . . . l s k . j1 1 Ž l r 1 . . . l r l . j 2 2
Nc
=B Ž q1 ,q1 ;q2 ,q2 ; gs 1 , . . . , gs k ; g r 1 , . . . , g r l . , Ž 2.14 .

with k q l s n y 4, and where we suppose that the two quark pairs have distinct flavor.
The sums are over the partitions of Ž n y 4. gluons between the two quark lines, and over
the permutations of the gluons within each partition. For k s 0 or l s 0, the color strings
same term with the exchange of the quarks Ž q1 q2 ..

For the maximally helicity-violating configurations, Žy,y,q, . . . ,q ., with like-
l
reduce to Kronecker delta’s. For identical quarks, we must subtract from Eq. Ž2.14. the
helicity for all of the gluons, the A and B subamplitudes factorize into distinct
contributions for the two quark antennae w61–66,68,69x. However, as we shall see in
Section 5.4, we need the helicity configurations with two gluons of opposite helicity. For
these the above mentioned factorization does not occur.
3. The leading impact factors

We consider the elastic scattering of two partons of momenta pa and p b into two
partons of momenta paX and p bX , in the high-energy limit, s 4 < t <. Firstly, we consider
the amplitude for gluon–gluon scattering ŽFig. 1a.. Using Eqs. Ž2.1., Ž2.2., or Eq. Ž2.9.,
and Eq. Ž2.10., and Appendix B, we obtain w74x
A g g ™ g g Ž pan a , panXaX < p bnXbX , p bn b .

X 1 X
s 2 s ig f a a c C g ; g Ž pan a ; panXaX .
ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 3.1 .
t
with q s p bX q p b and t , y< q H < 2 . In Eq. Ž3.1. and thereafter we use in the argument of
Fig. 1. Ža. Amplitude for g g ™ ™

g g scattering and Žb,c. for q g q g scattering. We label the external lines
with momentum, color and helicity, and the internal lines with momentum and color.
™
an amplitude A a vertical bar to separate clusters of particles which are close in rapidity
from other clusters. The LO impact factors g ) g g, with g ) an off-shell gluon, are
p b)X H
C g ; g Ž py q
a ; pa . s 1
X C g ; g Ž py qX
b ; pb . s . Ž 3.2 .
p bX H
They conserve helicity along the on-shell gluon line and transform under parity into
their complex conjugates,
)
C g;g Ž kn 4 . s C g ; g Ž ky n 4 . . Ž 3.3 .
2
In Eq. Ž3.1. four helicity configurations are leading, two for each impact factor . The
helicity-flip impact factor C g ; g Ž pq; pXq . is subleading in the high-energy limit.
From Eqs. Ž2.11., Ž2.12., we obtain the quark–gluon q g q g scattering amplitude
in the high-energy limit w54x,
™
A q g ™ q g Ž pan a , panXaX < p bnXbX , p bn b .
n aX
1 X
s 2 s g lcaX a C q ; q Ž py
a ; panXaX . ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 3.4 .
t
A g q ™ g q Ž pan a , panXaX < p bnXbX , p bn b .

X 1
s 2 s ig f a a c C g ; g Ž pan a ; panXaX . g lcbX b C q ; q Ž py
b
n bX
; p bnXbX . , Ž 3.5 .
t
where we have labeled the incoming quarks as outgoing antiquarks with negative
™
momentum, e.g. the antiquark is pa in Eq. Ž3.4. ŽFig. 1b., and p b in Eq. Ž3.5. ŽFig. 1c..
The LO impact factors g ) q q are
1r2
p b)X H
C q ; q Ž py q
a ; pa . s yi ;
X C q ; q Ž py qX
b ; pb . s i
ž / p bX H
. Ž 3.6 .
Under parity, the functions Ž3.6. transform as

)
C q ;q Ž k n 4 . s S C q ; q Ž ky n 4 . with S s ysign Ž q 0 q 0 . , Ž 3.7 .
™
and in general an impact factor acquires a coefficient S for each pair of quarks Žsee
Section 2.. Analogously, the antiquark–gluon q g q g amplitude is
A q g ™ q g Ž pan a , panXaX < p bnXbX , p bn b .
n aX
1 X
s 2 s g lca aX C q ; q Ž py
a ; panXaX . ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 3.8 .
t
A g q ™ g q Ž pan a , panXaX < p bnXbX , p bn b .

X 1
s 2 s ig f a a c C g ; g Ž pan a ; panXaX . g lcb bX C q ; q Ž py
b
n bX
; p bnXbX . , Ž 3.9 .
t
2
All throughout this paper, we shall always write only half of the helicity configurations contributing to an
impact factor, the other half being obtained by parity.
™
where the antiquark is paX in Eq. Ž3.8. and p bX in Eq. Ž3.9., and the LO impact factors
g ) q q are
1r2
p b)X H
C q ;q
Ž py q
a ; pa
X . si ; C q ;q
Ž py qX
b , pb . s yi ž /
p bX H
. Ž 3.10 .
In the amplitudes Ž3.1., Ž3.4., Ž3.5., Ž3.8., Ž3.9., the leading contributions from all the
Feynman diagrams have been included. However, the amplitudes have the effective
form of a gluon exchange in the t-channel ŽFig. 1., and differ only for the relative color
strength in the production vertices w76x. This allows us to replace an incoming gluon
with a quark, for instance on the upper line, via the simple substitution
X
ig f a a c C g ; g Ž pan a ; panXaX . lg l cX
aa C q ; q Ž py
a
n aX
; panXaX . , Ž 3.11 .
™
and similar ones for an antiquark andror for the lower line. For example, the
quark-quark q q q q scattering amplitude in the high-energy limit is
A q q ™ q q Ž pan a , panXaX < p bnXbX , p bn b .
n aX
1
s 2 s g lcaX a C q ; q Ž py
a ; panXaX . g lcbX b C q ; q Ž py
b
n bX
; p bnXbX . . Ž 3.12 .
t
4. The next-to-leading impact factors
Let three partons be produced with momenta k 1 , k 2 and p bX in the scattering between
two partons of momenta pa and p b , and to be specific, we shall take partons k 1 and k 2
in the forward-rapidity region of parton pa , the analysis for k 1 and k 2 in the
forward-rapidity region of p b being similar. Parametrizing the momenta as in Eq. ŽA.1.,
we have
y 1 , y 2 4 y bX ; < k 1 H < , < k 2 H < , < p bX H < . Ž 4.1 .
4.1. The NLO impact factor gg ) ™ gg

We consider the amplitude for the scattering g g g g g ŽFig. 2a.. Only PT
subamplitudes contribute, thus using Eqs. Ž2.1., Ž2.2. and Ž2.10., and Appendix C, we
™
obtain w38,39x
A g g ™ 3 g Ž pan a ,k 1n 1 ,k 2n 2 < p bnXbX , p bn b .

s
s4 g3 C g ; g Ž p bn b ; p bnXbX . Ý A g ; g g Ž pan a ;ksns1 1 ,ksns2 2 .
< qH < 2 sgS 2
X X X X
=tr Ž l al ds 1 l ds 2 l bl b y l al ds 1 l ds 2 l bl b q l al bl bl ds 2 l ds 1 y l al bl bl ds 2 l ds 1 .
X X
qB g ; g g Ž pan a ;ksns1 1 ,ksns2 2 . tr Ž l al ds 1 l bl bl ds 2 y l al ds 2 l bl bl ds 1 . , Ž 4.2 .
Fig. 2. Amplitudes for the production of three partons, with partons k 1 and k 2 in the forward-rapidity region of
parton pa .
with the sum over the permutations of the two gluons 1 and 2, the LO impact factor,
C g ; g Ž p bn b ; p bnXbX ., as in Eq. Ž3.2., and
A g ; g g Ž pan a ;k 1n 1 ,k 2n 2 . s C g ; g g Ž pan a ;k 1n 1 ,k 2n 2 . An Ž k 1 ,k 2 . ,
B g ; g g Ž pan a ;k 1n 1 ,k 2n 2 . s C g ; g g Ž pan a ;k 1n 1 ,k 2n 2 . B n Ž k 1 ,k 2 . , Ž 4.3 .
with n s signŽ na q n 1 q n 2 ., and
C g ; g g Ž py q q
a ;k 1 ,k 2 . s 1,
C g ; g g Ž pq y q 2
a ;k 1 ,k 2 . s x 1 ,
C g ; g g Ž pq q y 2
a ;k 1 ,k 2 . s x 2 . Ž 4.4 .
The momentum fractions are defined as
kq
i
xi s q q i s 1,2 Ž x 1 q x 2 s 1 . , Ž 4.5 .
k1 q k 2
and the function Aq as follows:
qH x1 1
Aq Ž k 1 ,k 2 . s y'2
k1H x 2 12:
²
, ( Ž 4.6 .
with ²12: a shorthand for ² k 1 k 2 :. Using the dual Ward identity w61x, or UŽ1.
decoupling equations w18,67x, the function B n in Eq. Ž4.3., and thus the function B g ; g g ,
can be written as
B n Ž k 1 ,k 2 . s y An Ž k 1 ,k 2 . q An Ž k 2 ,k 1 . . Ž 4.7 .
™
is subleading to the required accuracy. The function An
g; g gŽ
The function C pq q q.
a ;k 1 ,k 2
has a collinear divergence as 2 k 1 P k 2 0, but the divergence cancels out in the function
B n where gluons 1 and 2 are not adjacent in color ordering w39x.
Using Eq. Ž4.7., and fixing t , y< q H < 2 , the amplitude Ž4.2. may be rewritten as
A g g ™ 3 g Ž pan a ,k 1n 1 ,k 2n 2 < p bnXbX , p bn b .

2 X
½
s 2 s Ž ig . Ý
sgS 2
f a ds 1c f c ds 2 cA g ; g g Ž pan a ;ksns1 1 ,ksns2 2 . 5
1 X X
= ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 4.8 .
™ g g is enclosed in curly brackets, and includes
t
where the NLO impact factor for g ) g
six helicity configurations.
In the multi-Regge limit y 1 4 y 2 ,

1
lim A g ; g g Ž pan a ;k 1n 1 ,k 2n 2 . s C g ; g Ž pan a ;k 1n 1 . C g Ž q1 ,k 2n 2 ,q . , Ž 4.9 .
y14y 2 t1
with q1 s yŽ pa q k 1 ., and t 1 , y< q1 H < 2 , and with LO Lipatov vertex, g ) g )

w32,33,74x,
™g
q1)H q2 H
C g Ž q1 ,kq,q2 . s '2 . Ž 4.10 .
kH
Accordingly, the amplitude Ž4.8. is reduced to an amplitude in multi-Regge kinematics

w29,74x, with the effective form of a gluon-ladder exchange in the t channel,
A g g ™ 3 g Ž pan a ,k 1n 1 < k 2n 2 < p bnXbX , p bn b .

1 X
s 2 s ig f a d1 c C g ; g Ž pan a ;k 1n 1 . ig f c d 2 c C g Ž q1 ,k 2n 2 ,q2 .
t1
1 X X
t2
with q2 s p b q p bX and t 2 , y< q2 H < 2 .
4.2. The NLO impact factor gg ) ™ qq

™
The amplitude g g q q g for the production of a qq pair in the forward-rapidity
region of gluon a ŽFig. 2c. is obtained by taking the amplitudes Ž2.11. – Ž2.13. in the
kinematics Ž4.1. w41x,
A g g ™ q q g Ž pan a ,k 1n 1 ,ky
2
n 1 < nXbX
p b , p bn b .
X
s 2 s g 2 Ž lcl a . d 2 d 1 A g ; q q Ž pan a ;k 1n 1 ,ky
2
n1
.
X 1 X X
q Ž l alc . d 2 d 1 A g ; q q Ž pan a ;ky
2
n1
,k 1n 1 . 4t ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 4.12 .
with k 1 the antiquark, the NLO impact factor g ) g ™ qq in curly brackets, and with
A g ; q q Ž pan a ;k 1n 1 ,ky
2
n1
. s C g ; q q Ž pan a ;k 1n 1 ,ky2 n 1 . An Ž k 1 ,k 2 . ,
C g q q Ž pq q y 3
a ;k 1 ,k 2 . s x 1 x 2 ,(
C g ; q q Ž pq y q
a ;k 1 ,k 2 . s (x 3
1 x2 , Ž 4.13 .
™
with momentum fractions as in Eq. Ž4.5., An in Eq. Ž4.6. and n s na . The NLO impact
factor g ) g qq allows for four helicity configurations.
In the multi-Regge limit kq q )
™
1 4 k 2 , the NLO impact factor g g qq vanishes, since
quark production along the multi-Regge ladder is suppressed.
4.3. The NLO impact factor qg ) ™ qg

™
The amplitude q g q g g for the production of a q g pair in the forward-rapidity
region of quark a ŽFig. 2b. is obtained by taking the amplitudes Ž2.11. – Ž2.13. in the
kinematics Ž4.1. w54x
A q g ™ q g g Ž py
a
n1
,k 1n 1 ,k 2n 2 < p bnXbX , p bn b .
X
n1
s 2 s g 2 Ž l d 2lc . d1 a A q ; q g Ž py a ;k 1n 1 ,k 2n 2 .
X
n1
1 X X
q Ž lcl d 2 . d1 a B q ; q g Ž py ;k 1n 1 ,k 2n 2 . 4 ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 4.14 .
™
a
t
with k 1 the final-state quark, and the NLO impact factor q g ) q g in curly brackets.
As above, the NLO impact factor includes four helicity configurations,
n1
A q ; q g Ž py
a ;k 1n 1 ,k 2n 2 . s C q ; q g Ž py a
n1
;k 1n 1 ,k 2n 2 . An Ž k 1 ,k 2 . ,
n1
B q ; q g Ž py
a ;k 1n 1 ,k 2n 2 . s C q ; q g Ž py a
n1
;k 1n 1 ,k 2n 2 . B n Ž k 1 ,k 2 . ,
C q ; q g Ž py q q
a ;k 1 ,k 2 . s yi x 1 , (
C q ; q g Ž pq y q 3
a ;k 1 ,k 2 . s i x 1 , ( Ž 4.15 .
n n 3
with A in Eq. Ž4.6., and B given by Eq. Ž4.7., with n s n 2 . As in Section 4.1, the
function B q; q g vanishes in the collinear limit.
In the multi-Regge limit kq q
1 4 k 2 the amplitude 4.14 reduces to Eq. 4.11 , with
Ž . Ž .
™
the substitution Ž3.11. for the upper line, and the LO impact factor C q; q in Eq. Ž3.6..
The treatment of the amplitude q g q g g for the production of a q g pair in the
™
forward-rapidity region of antiquark a is identical to the former, thus the NLO impact
factor q g ) q g is the same as in Eq. Ž4.14. up to inverting the color flow on the
quark line w54x. The corresponding functions A and B are the same as in Eq. Ž4.15..
4.4. NLO impact factors in the collinear limit
The collinear factorization for a generic amplitude occurs both on the subamplitude
and on the full amplitude w61x, since in Eqs. Ž2.1., Ž2.11. and Ž2.14. color orderings
where the collinear partons are not adjacent do not have a collinear divergence. Hence in
the collinear limit for partons i and j, with k i s zP and k j s Ž1 y z . P, a generic
amplitude Ž2.1. can be written as
lim A . . . d i d j . . . Ž . . . ,k in i ,k nj j , . . . . s Ý A . . . c . . . Ž . . . , P n , . . . . Splityf ™
n
fi fj
Ž k in i ,k nj j . ,
ki<< kj n
Ž 4.16 .
with f denoting the parton species. Accordingly, for k 1 s zP and k 2 s Ž1 y z . P, we can
write the amplitudes Ž4.8., Ž4.12. and Ž4.14. as
lim A f g ™ f 1 f 2 g Ž pan a ,k 1n 1 ,k 2n 2 < py
b
n n
X b,p b.
b
k1< < k 2
s A f g ™ f g Ž pan a , Pyn a < py n n

X b , p b . P Split
f ™ f1 f 2
Ž k 1n 1 ,k 2n 2 . , Ž 4.17 .
™ fg
b b na
fg
with A as in Eq. Ž3.1., Ž3.4. and Ž3.8., respectively, and where we have used
3
In this context, Eq. Ž4.7. is only a bookkeeping, since the UŽ1. decoupling equation is valid only for the
gluino-gluon subamplitudes corresponding to the quark–gluon subamplitudes used in Eq. Ž4.14..
helicity conservation in the s channel ŽSection 3.. For the collinear factors, Splityf ™
n
f1 f 2
,
we obtain
Splityg ™
n
gg
Ž k 1n 1 ,k 2n 2 . s ig f c d1 d 2 splityg ™
n
gg
Ž k 1n 1 ,k 2n 2 . ,
Splityg ™
n
qq
Ž k 1n 1 ,k 2n 2 . s g Ž lc . d 2 d 1 splityg ™
n
qq
Ž k 1n 1 ,k 2n 2 . ,
q™ q g
Splity n Ž k 1n 1 ,k 2n 2 . s g Ž ld 2 . d1 c splityq ™
n
qg
Ž k 1n 1 ,k 2n 2 . ,
q™ q g
Splity n Ž k 1n 1 ,k 2n 2 . s g Ž ld 2 . c d 1 splityq ™
n
qg
Ž k 1n 1 ,k 2n 2 . , Ž 4.18 .
with splitting factors w61,72x
splityg ™ g g Ž kq q
1
'
1 ,k 2 . s 2
(z Ž 1 y z . ²12: ,
splitqg ™ g g Ž ky q
z2
'
1 ,k 2 . s 2
(z Ž 1 y z . ²12: ,
™ g g Ž kq1 ,ky2 . s '2
2
Ž1yz .
splitqg
(z Ž 1 y z . ²12: ,
splitqg ™ q q Ž kq y
1yz
'
1 ,k 2 . s 2 ,
²12:
splitqg ™ q q Ž ky q
z
'
1 ,k 2 . s 2 ,
²12:
q™ q g
Ž kq1 ,kq2 . s splityq ™ q g Ž kq1 ,kq2 . s '2
1
splity
'1 y z ²12: ,
q™ q g
Ž ky1 ,kq2 . s splitqq ™ q g Ž ky1 ,kq2 . s '2
z
splitq Ž 4.19 .
'1 y z ²12:
and splitnf ™ f 1 f 2 Ž ky
1
n 1 yn 2 .
,k 2 obtained from splityf ™
n
f1 f 2 Ž n 1 n 2 .
k 1 ,k 2 by exchanging ² k 1 k 2 :
with w k 2 k 1 x, and multiplying by the coefficient S, Eq. Ž3.7., if the splitting factor
includes a quark pair.
Summing over the two helicity states of partons 1 and 2, we obtain a two-dimen-
sional matrix, whose entries are the Altarelli–Parisi splitting functions at fixed color and
helicity of the parent parton w12–15x
Splitlf ™ f 1 f 2 Ž k 1n 1 ,k 2n 2 . Split rf ™ f 1 f 2 Ž k 1n 1 ,k 2n 2 . e iŽ fly fr . Plfr™ f 1 f 2 ,

) X 2 g2
Ý s d cc
n 1n 2 s12
Ž 4.20 .
f ™ f1 f 2
with e iŽ flyf r . a phase, where e iŽ f qyf y. s w21xr²12:, and where by definition Pqq
f ™ f1 f 2 f ™ f1 f 2 f ™ f1 f 2
s Pyy , and Pqy s Pyq , and
g™ gg
z 1yz
Pqq s 2CA q qz Ž1yz . ,
1yz z
g™ gg
Pqy s 2CA z Ž 1 y z . ,
g ™ qq
1 2 2
Pqq s z q Ž1yz . ,
2
g ™ qq
Pqy sz Ž1yz . ,
™ ™
2
q qg q qg
1qz
Pqq s Pqq s CF ,
1yz
q™ q g q™ q g
Pqy s Pqy s0 . Ž 4.21 .
q™ q g
to zero. P q ™ g q is obtained from P q ™ q g by exchanging Ž z l 1 y z .. Since we sum

For P helicity conservation on the quark line sets the off-diagonal elements equal
over two helicity states of the external partons, Eq. Ž4.21. is valid in the-dimensional
reduction ŽDR. scheme w56–58x. Eq. Ž4.21. agrees with the corresponding spin-corre-
lated splitting functions of Ref. w77x in the DR scheme, after contracting the ones of type
P g ™ f 1 f 2 with a parent-gluon polarization as in Appendix E. The connection of Eq.
Ž4.21. with other regularization schemes ŽRS. is also given in Ref. w77x.
Averaging over the trace of P f ™ f 1 f 2 in Eq. Ž4.20., i.e. over color and helicity of the
parent parton on the left-hand side of Eq. Ž4.20., we obtain the unpolarized Altarelli–
Parisi splitting functions 4
<Splityf ™ ² P f ™ f1 f 2 : ,
1 2 g2
Ý n
f1 f 2
Ž k 1n 1 ,k 2n 2 . < 2 s Ž 4.22 .
2C nn 1n 2 s12
with C s Nc2 y 1 for a parent gluon and C s Nc for a parent quark, and where the
averaged trace of P f ™ f 1 f 2 is ² P f ™ f 1 f 2 : s tr P f ™ f 1 f 2r2 s Pqq
f ™ f1 f 2
.
5. The next-to-next-to-leading impact factors
In order to derive the next-to-next-to-leading ŽNNLO. impact factors, we repeat the

analysis of Section 4 with one more final-state parton. Let four partons be produced with
momenta k 1 , k 2 , k 3 and p bX in the scattering between two partons of momenta pa and
p b , with a cluster of three partons, k 1 , k 2 and k 3 , in the forward-rapidity region of
parton pa ,
y 1 , y 2 , y 3 4 y bX ; < k 1 H < , < k 2 H < , < k 3 H < , < p bX H < . Ž 5.1 .
5.1. The NNLO impact factor gg ) ™ ggg

We begin with the amplitude for the scattering g g g g g g ŽFig. 3a. in the
kinematics Ž5.1.. Using Eqs. Ž2.1., Ž2.2. and Ž2.10., and the subamplitudes of non-PT
™
4
Note that in the DR scheme the unpolarized splitting functions do not coincide with the azimuthally
averaged ones. The latter are given in any RS in Ref. w77x.
Fig. 3. Amplitudes for the production of four partons, with partons k 1 , k 2 and k 3 in the forward-rapidity
region of parton pa .
type, with three gluons of q helicity and three gluons of y helicity w61x, and Appendix
D, we obtain
A g g ™ 4 g Ž pan a ,k 1n 1 ,k 2n 2 ,k 3n 3 < p bnXbX , p bn b .

s
s4 g4 C g ; g Ž p bn b ; p bnXbX . Ý A g ;3 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 .
< qH < 2 sgS 3
X X X
=tr Ž l al ds 1 l ds 2 l ds 3 l bl b y l al ds 1 l ds 2 l ds 3 l bl b q l bl bl ds 3 l ds 2 l ds 1 l a
X X
yl bl bl ds 3 l ds 2 l ds 1 l a . q B g ;3 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 . tr Ž l al ds 1 l ds 2 l bl bl ds 3
X X X
yl al ds 1 l ds 2 l bl bl ds 3 q l bl bl ds 2 l ds 1 l al ds 3 y l bl bl ds 2 l ds 1 l al ds 3 . , Ž 5.2 .
with the sum over the permutations of the three gluons 1, 2 and 3, and the LO impact
factor, C g ; g Ž p bn b ; p bnXbX ., as in Eq. Ž3.2.. From the PT subamplitudes Ž2.10. we obtain the
function of Žyqqq . helicities
A g ;3 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 . s C g ;3 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 . An Ž k 1 ,k 2 ,k 3 . , Ž 5.3 .

where n s signŽ na q n 1 q n 2 q n 3 . and
qH x1 1
Aq Ž k 1 ,k 2 ,k 3 . s y2
k1H ( x 3 ²12:²23:
Ž 5.4 .
and
kq
i
xi s i s 1,2,3 Ž x 1 q x 2 q x 3 s 1. . Ž 5.5 .
kq q q
1 q k2 q k3
The functions C g ;3 g are a straightforward generalization of the functions C g ; g g defined

in Eq. Ž4.4. and read
1 na s y
C g ;3 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 . s
½ x i2 ni s y i s 1,2,3
with n s q, Ž 5.6 .
From the non-PT subamplitudes w61x we obtain the function of Žyyqq . helicities
A g ;3 g Ž py q q y
a ,k 1 ,k 2 ,k 3 .
2 s a12 s123
s
s12 < k 1 H < 2
y
s23 ž b Ž k 1 ,k 2 ,k 3 . x 1
g Ž k 1 ,k 2 ,k 3 . Ž x 1 x 2 q b Ž k 1 ,k 2 ,k 3 . Ž x 2 q x 3 . .
q
x2 x3 /
2 2
b Ž k 1 ,k 2 ,k 3 . s a12 g Ž k 1 ,k 2 ,k 3 . s123 < k 1H < 2 s12 x 12 x 2 < q H < 2
y q q ,
x2 s23 x 1 x 2 x 3 s23 Ž x 2 q x 3 .
A g ;3 g Ž py q y q
a ,k 1 ,k 2 ,k 3 .
2 s a12 s123
s
s12 < k 1 H < 2
y
s23 ž ya Ž k 1 ,k 3 ,k 2 . x 1
g Ž k 1 ,k 3 ,k 2 . Ž x 1 x 2 y a Ž k 1 ,k 3 ,k 2 . Ž x 2 q x 3 . .
q
x2 x3 /
2 2
a Ž k 1 ,k 3 ,k 2 . s a12 g Ž k 1 ,k 3 ,k 2 . s123 < k 1 H < 2 s12 x 12 x 2 < q H < 2
y q q ,
x2 s23 x 1 x 2 x 3 s23 Ž x 2 q x 3 .
2 2
2 g Ž k 2 ,k 3 ,k 1 . s123 a Ž k 2 ,k 3 ,k 1 . s a12
A g ;3 g Ž py y q q
a ,k 1 ,k 2 ,k 3 . s y
s12 s23 x 1 x 2 x 3 x2 < k3H < 2
a Ž k 2 ,k 3 ,k 1 . g Ž k 2 ,k 3 ,k 1 . s a12 s123 Ž x 2 q x 3 .
q ,
s23 x 2 x 3 < k 3 H < 2
Ž 5.7 .
with si jk s Ž pi q pj q p k . 2 the three-particle invariant, and
(x 1 k3H ž (x 1
)
qH (
q x 2 w1 2x /,
a Ž k 1 ,k 2 ,k 3 . '
s a13
Ž k 1 H qk 2 H . w 1 2 x (x 1 x 2
b Ž k 1 ,k 2 ,k 3 . ' ,
s a12
(x 1 ž(
x 2 x 3 w 1 2 x x 1 ²1 3: q x 2 ²2 3: ( /.
g Ž k 1 ,k 2 ,k 3 . ' Ž 5.8 .
s123
Using the UŽ1. decoupling equations w18,67x, the function B in Eq. Ž5.2. can be
written as
B g ;3 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 . s y A g ;3 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 .
qA g ;3 g Ž pan a ;k 1n 1 ,k 3n 3 ,k 2n 2 .
qA g ;3 g Ž pan a ;k 3n 3 ,k 1n 1 ,k 2n 2 . . Ž 5.9 .
In the triple collinear limit, k 1 < < k 2 < < k 3 , Section 5.6, the function A has a double collinear
divergence, while the function B, whose gluon 3 is not color adjacent to gluons 1 and 2,
has only a single collinear divergence.
Using Eq. Ž5.9., we can rewrite Eq. Ž5.2. as
A g g ™ 4 g Ž pan a ,k 1n 1 ,k 2n 2 ,k 3n 3 < p bnXbX , p bn b .
3 X X XX
½
s 2 s Ž ig . Ý
sgS 3
f a ds 1c f c ds 2 c f c ds 3 c A g ;3 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 . 5
1 X XX
t
where the NNLO impact factor g ) g
14 helicity configurations.
™ g g g is enclosed in curly brackets, and includes
5.2. The NNLO impact factor gg ) ™ gqq
We consider the amplitude for the scattering g g g q q g ŽFig. 3b., in the kinemat-
ics Ž5.1.. Using Eqs. Ž2.11. – Ž2.13. and the subamplitudes of non-PT type, with two
™
gluons of q helicity and two gluons of y helicity w61x, we obtain
A g g ™ g q q g Ž pan a ,k 1n 1 ,k 2n 2 ,ky
3
n 2 < nXbX
p b , p bn b .
1 X X
s 2 s I g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
. ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 5.11 .
t
with k 3 the quark, and with NNLO impact factor g g ) ™ g q q,
X
I g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
. s g 3 Ž lclald1 . d 3 d 2 A1g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky3 n 2 .
X
q Ž l alcl d1 . d 3 d 2 A 2g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
X
q Ž l d1lcl a . d 3 d 2 A 3g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
X
q Ž l d1l alc . d 3 d 2 A 4g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
X
q Ž l al d1lc . d 3 d 2 B1g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
X
q Ž lcl d1l a . d 3 d 2 B2g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
. .
Ž 5.12 .
The NNLO impact factor allows for eight helicity configurations. From the PT subam-
plitudes Ž2.12., Ž2.13. we obtain
qH 1
A1g ; g q q Ž pq q y q
a ;k 1 ,k 2 ,k 3 . s y2
k1H
x 1 x 23
²1 2 ²2 3:
:
, (
x 23 x 32
A 2g ; g q q Ž pq q y q
a ;k 1 ,k 2 ,k 3 . s 2
qH
k3H
) x1 ²1 2: ²2 3:
1
,
qH x3 1
a ;k 1 ,k 2 ,k 3 . s y2
k2 H
x 22 ( x 1 ²1 3: ²3 2:
,
qH 1
a ;k 1 ,k 2 ,k 3 . s 2
k1H
(x x 1
2
2 x3
²1 3: ²3 2:
,
qH 1
B1g ; g q q Ž pq q y q
a ;k 1 ,k 2 ,k 3 . s y2 x2 x3 ,
k3H k 1 H ²2 3:
qH 1
B2g ; g q q Ž pq q y q
a ;k 1 ,k 2 ,k 3 . s y2 x 22 , Ž 5.13 .
k2 H k 1 H ²2 3:
with momentum fractions as in Eq. Ž5.5.. The impact factors from the non-PT
subamplitudes w61x are,
A1g ; g q q Ž py q y q
a ;k 1 ,k 2 ,k 3 .
2
g Ž k 1 ,k 3 ,k 2 . s123 (x g Ž k 1 ,k 3 ,k 2 . s123 a Ž k 1 ,k 3 ,k 2 . s3 b bX
s2
½ w 1 3 x²1 2 :s23 x 1 x 2 x 3
2
q
1
)
s12²2 3 :k 1H x 3 x 3 w1 3x k 2 H (
Ž k 3 H qq H . a Ž k 1 ,k 3 ,k 2 . s3 b bX < q H < 2 x 12 x 23r2
y y
(x 2 x 3 s12 < k 1 H < 2 k 2 H (x 3 s23 < k 1 H < 2 Ž x 2 q x 3 .
x 1 s 3 b bX g Ž k 1 ,k 3 ,k 2 . s123
y
s12 s23 < k 1 H < x 3 2 (
s123 x 2 x 3 a Ž k 1 ,k 3 ,k 2 . q
(x x3
2
a Ž k 1 ,k 3 ,k 2 . Ž x 2 q x 3 .
ž
= yx 2 q
x1 /5 , Ž 5.14 .
A 2g ; g q q Ž py q y q
a ;k 1 ,k 2 ,k 3 .
° g Ž k 1 ,k 3 ,k 2 . s123
2
x2 x3 ž y w2 3x (x q w1 3x x 2 ( /
s 2~y ( 1
¢ w 1 3 x²1 2 :s23 x 1 x 2 x 3
y
x1 s12 w 2 3 x
k 2 H k 3)H y w 2 3 x x 2 x 3
ž ( /
2
(x 2 x3 (x 2 x 3 g Ž k 1 ,k 3 ,k 2 . s123 ¶•
q
(
s23 k 3)H s1 b bX x 2 x 3
q
s23
q
x 1²1 2 :²2 3 :w 1 3 x k 3)H ß,
Ž 5.15 .
a ;k 1 ,k 2 ,k 3 .
2 2
(
k 3 H x 2 a Ž k 3 ,k 1 ,k 2 . s1 b bX g Ž k 1 ,k 3 ,k 2 . s123 w 1 2 x
s2
½ k 2 H < k 2 H < 2 s23 x 3
(
y
x 1 x 2 x 3 w 1 3 x s13 s23
²1 2 :x 1 a Ž k 3 ,k 1 ,k 2 . s1 b bX(x 2 ž²1 3 :y ²1 2 :(x 2 x 3 /

q q
²1 3 :s23 ²1 3 :< k 2 H < 2 s23(x 3
w 1 3 x x 2 x 3(x 1 q Ž k 2)H x 1 y q H
)
x 2 x 3 . (x 3
q
(x 1 ²1 3 :k 2)H w2 3x 5 , Ž 5.16 .
a ;k 1 ,k 2 ,k 3 .
2 ) 2
Ž k 2 H qq H . (x 1 a Ž k 1 ,k 2 ,k 3 . Ž k 2 H qq H . q H x 1 g Ž k 1 ,k 3 ,k 2 . s123
s2
½ 2
²1 3 :< k 1 H < k 3 H (x 2
y
²2 3 :w 1 3 x < k 1 H < 2 s13(x 1 x 3 x 2
w 1 2 x g Ž k 1 ,k 3 ,k 2 . 2 s123 x 12 < q H < 2(x 2 x 3
q q
w 1 3 x s13 s23 x 1 x 2 x 3 < k 1 H < 2 s23 Ž x 2 q x 3 .
w 1 2 x x 1 s123 y Ž k 2 H qq H . x 2(x 1 x 3 w 1 3 x q s2 b bX g Ž k 1 ,k 3 ,k 2 .
y
x 2 < k 1 H < 2 s13 s23 w 1 3 x
x 1 Ž k 2 H qq H . w 1 2 x g Ž k 1 ,k 3 ,k 2 . s123 Ž x 2 q x 3 .
q
s13 < k 1H < 2 s23 x 1 x 3 x 2
( 5 , Ž 5.17 .
B1g ; g q q Ž py q y q
a ;k 1 ,k 2 ,k 3 .
y a Ž k 3 ,k 1 ,k 2 . 2 s1 b bX Ž k 2 H q q H . 2 w 1 3 x x 1 x 3 a Ž k 1 ,k 2 ,k 3 .
'
s2
½ k 3)H s23 k 2 H 'x 2 x3
q
< k1 H < 2 < k 3 H < 2 k 3 H x 2
'
w 1 3 x x 1 x 3 ² 2 3 : x 2 < k 3 H < 2 y Ž k 2 H q q H . s23 Ž x 1 q x 2 . x 3 q k 1 H k 2 H w2 3 x x 3 x 2
' ' ' '
y 2 2
'x 2
< k 1 H < < k 3 H < s23
'x 1 qH x2
y 2
< k 1 H < s23 x 2
'
)
y w1 3x k 1 H x 2 q q H x1 x 3 ' ž x2 q x3
q k2 H / 5 , Ž 5.18 .
B2g ; g q q Ž py q y q
a ;k 1 ,k 2 ,k 3 .
°yŽ k 1 H qk 2 H . ž k 1)H q w 1 2 x (x 1 x 2 / (x 2
2
s 2~
¢ < k 1 H < 2 k 2)H s3 b bX x 3(
2
( ž
k 3 H x 2 k 3)H y w 2 3 x x 2 x 3 ( /
q
k 2)H s23 s1 b bX x 33r2
x2
y
< k 1 H < < k 2 H < 2 s23 x 3
2
yk 3 H w 2 3 x s3 b bX x 1 x 2 q k 1)H q w 1 2 x x 1 x 2
ž ( /
ž (
= ²2 3 :q H k 3)H y w 2 3 x x 2 x 3 y k 3 H w 2 3 x k 2 H Ž x 2 q x 3 .
ž / /
(x 2 x 1 x 32
y < k 3 H < 2 s 3 b bX x 1 x 2 q < q H < 2 < k 2 H < 2
< k 1 H < 2 < k 2 H < 2 s23 x 33r2 x2 q x3
¶•
q k 1)H q w 1 2 x x 1 x 2 ( /Žk < <2
3 H s1 b b x 2 x 3 q k 2 H k 3 H Ž x 2 q x 3 . .
ž X
ß. Ž 5.19 .
The functions A and B for the remaining helicity configurations are derived using the
relations
A ig ; g q q Ž pan a ;kq q y g ; g qq na q y q
1 ,k 2 ,k 3 . s yA 5yi Ž pa ;k 1 ,k 3 ,k 2 . , i s 1,2,3,4,
Ž 5.20 .
Big ; g q q Ž pan a ;kq q y
1 ,k 2 ,k 3 . g ; g qq
s yB3yi Ž pan a ;kq y q
1 ,k 3 ,k 2 ., i s 1,2.
5.3. The NNLO impact factor qg ) ™ qgg

We consider the amplitude q g ™ q g g g for the production of a quark and two
gluons in the forward-rapidity region of quark a ŽFig. 3c. in the kinematics Ž5.1.. Using
Eqs. Ž2.11. – Ž2.13. and the subamplitudes of non-PT type, with two gluons of q helicity
and two gluons of y helicity w61x, we obtain
A q g ™ q 3 g Ž py a
n1
,k 1n 1 ,k 2n 2 ,k 3n 3 < p bnXbX , p bn b .
n1
1 X X
s 2 s I q ; q g g Ž py a ;k 1n 1 ,k 2n 2 ,k 3n 3 . ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 5.21 .
™
t
with k 1 the final-state quark, and the NNLO impact factor q g ) q g g,
n1
I q ;q g g
Ž py
a ;k 1n 1 ,k 2n 2 ,k 3n 3 .
X
n1
sg 3
Ý Žl ds 2
l ds 3
lc . d1 a A q ; q g g Ž py
a ;k 1n 1 ,ksns2 2 ,ksns3 3 .
sgS 2
X
n1
q Ž lcl ds 2 l ds 3 . d1 a B1q ; q g g Ž py
a ;k 1n 1 ,ksns2 2 ,ksns3 3 .
X
n1
q Ž l ds 2 lcl ds 3 . d1 a B2q ; q g g Ž py
a ;k 1n 1 ,ksns2 2 ,ksns3 3 . . Ž 5.22 .
The NNLO impact factor allows for eight helicity configurations. From the PT subam-
plitudes Ž2.12., Ž2.13. we obtain
qH x1 1
A q ; q g g Ž py q q q
a ;k 1 ,k 2 ,k 3 . s 2 i ,
k 1 H x 3 ²1 2: ²2 3: (
qH x1 x 3 1
B1q ; q g g Ž py q q q
a ;k 1 ,k 2 ,k 3 . s 2 i
k3H ( x2 k 1 H ²2 3:
,
qH x1 1
B2q ; q g g Ž py q q q
a ;k 1 ,k 2 ,k 3 . s 2 i Ž 5.23 .
k1H (x 2 k 3 H ²1 2:
and
A q ; q g g Ž pq y q q
a ,k 1 ,k 2 ,k 3 . s yx 1 A
q ;q g g
Ž py q q q
a ;k 1 ,k 2 ,k 3 . ,
Biq ; q g g Ž pq y q q
a ;k 1 ,k 2 ,k 3 . s yx 1 Bi
q ;q g g
Ž py q q q
a ;k 1 ,k 2 ,k 3 . , i s 1,2. Ž 5.24 .
The impact factors from the non-PT subamplitudes w61x are
A q ; q g g Ž py q y q
a ;k 1 ,k 2 ,k 3 .
°k ž k )
1H q w 1 2 x (x 1 x 2 / ) 2 2 2
g Ž k 1 ,k 2 ,k 3 . g Ž k 1 ,k 3 ,k 2 . s123
s 2 i~
2H
q
¢ < k 1 H < 2 w 1 2 x s3 b bX(x 2 3r2
Ž x 1 x 2 x 3 . s12 s23²2 1 :w 1 3 x
w 1 3 x²1 2 :x 1 ž (x 3 ²2 3 :y k 1 H (x 2 / q k 1 H (x 3 g Ž k 1 ,k 3 ,k 2 . s123
q
s12 s23 k 1H x 1 x 3 (
(x 1 x1 x 2 x 3
q 2
k 1)H x 3 Ž q H x 2 q k 2 H Ž x 2 q x 3 . . y < q H < 2
s23 < k 1 H < x 3 x2 q x3
¶
qx 1 yq H k 3)H x 2 q k 2 H x 3 k 1)H x 3 y w 2 3 x x 2
ž ( ž ( ( / / ß•, Ž 5.25 .
A q ; q g g Ž py q q y
a ; k1 , k 2 , k 3 .
b Ž k 1 , k 2 , k 3 . 2 s 3 b bX k 2)H '
b Ž k 1 , k 2 , k 3 . s 3 b bX k 2)H x 1 Ž x 3 q x 2 .
s2 i
½ w 1 2 x s 12 < k 1 H < 2
x 23r2
q
s 12 < k 1 H < 2 x 3 x 2 w 3 2 x
'
)
k 2)H x 13r2 q H g Ž k 1 , k 2 , k 3 . s 132
y
s 12 < k 1 H < 2 w 3 2 x ' x1 x 3 ² 1 2:
'
q s 13 2 x 3 x 2
y
'x 2
)
b Ž k 1 , k 2 , k 3 . s 3 b bX q H
q
g Ž k 1 , k 3 , k 2 . g Ž k 1 , k 2 , k 3 . 2 s 132
) 2
3r 2
x1 ' x 3 k 2)H Ž x1 x 3 x 2 . s 12 s 32² 3 1 : w 1 2 x
2
w
x 1 ys 12 < q H < x 22 x 1 x 3 q ² 1 3 : b Ž k 1 , k 2 , k 3 . s 3 b bX k 2)H Ž x 3 q x 2 . 2
' x
q
s 12 s 32 < k 1 H < 2 x 2 x 3 Ž x 3 q x 2 .
'
'x 1 s 3 b bX w w 1 2 x² 1 3 : k 2)H ' x 13 x 2 '
q b Ž k 1 , k 2 , k 3 . k 2)H s 132 x 1 x 3 y ² 1 3 : k 1)H q H
Ž )
x2 .x
q
w 1 2 x s 12 s 32 < k 1 H < 2 x 2 x 3
' 5 Ž 5.26 .
B1q ; q g g Ž py q y q
a ;k 1 ,k 2 ,k 3 .
2
Ž k 2 H qq H . w 1 3 x x 1(x 3 k 2 H w1 3x x 1 x 3
s2i
½ 2
< k 1 H < k 3 H s 2 b bX
)
q
2
²2 3 :k 1)H < k 3 H < 2(x 2
a Ž k 2 ,k 1 ,k 3 . a Ž k 3 ,k 1 ,k 2 . s12b bX w 1 3 x (x 3
y 2
q
< k3H < k 3)H k 2 H s23 (x 1 x2 < k 1 H < 2 < k 3 H < 2 s23 x 2
= y< k 3 H < 2 Ž k 2 H qq H . x 1 x 2 q < k 1 H < 2 k 2 H Ž x 1 q x 2 . x 3

x3
q k 2 H k 1)H < k 3 H < 2 x 1 q k 2 H < k 3 H < 2 w 1 2 x x 13r2 x 2 (
< k 1 H < < k 3 H < 2 s23 x 2 x 1
2
(
q H < k 3 H < 2 x 12 x 2
)
yq H
ž x2 q x3
q < k1H < 2 k 2 H Ž x1 q x 2 . x 3
/5 , Ž 5.27 .
B1q ; q g g Ž py q q y
a ;k 1 ,k 2 ,k 3 .
) 2
a Ž k 3 ,k 1 ,k 2 . a Ž k 2 ,k 1 ,k 3 . s12b bX
s2i y
½ k 3 H k 2)H < k 3 H < 2 s32 x 1 x 2 (
²1 3 :(x 3 a Ž k 1 ,k 2 ,k 3 . 2 s2 b bX (x 3
y 2 2
q
< k1H < < k 3H < k 3H < k 1 H < < k 3 H < 2 s23
2
(x x1 3
= w 1 2 x ²1 3 : Ž s2 b bX x 1 y s1 b bX x 2 . q (x 1 ž yk 1 H q H x 33r2
(x 2
2
x1 y x 2
(
q²1 3 :k 3 H x 1 Ž x 3 q x 2 . / ž )2
qH y w1 2x y
(x 1 x2
)
qH w1 2x /
< k 3H < 2qH x1 (
)
yq H x1 (x 3
ž x3 q x2
q ²1 3 :Ž s1 b bX q s2 b bX . x 3 ( /5 , Ž 5.28 .
2i ° ~Žk 2 H qq H . x 1 ž k 1)H q w 1 2 x (x 1 x 2 /
B2q ; q g g Ž py q y q
a ;k 1 ,k 2 ,k 3 . s
< k1H < ¢
2
k 3 H w 1 2 x (x 2
2
k 2 H k 1)H q w 1 2 x x 1 x 2
ž ( /
y
w 1 2 x s3 b bX(x 2
2
Ž k 2 H qq H . w 1 3 x x 1(x 3 • ¶
y
k 3 H s 2 b bX ß, Ž 5.29 .
B2q ; q g g Ž py q q y
a ;k 1 ,k 2 ,k 3 .
2i ²1 3 :(x 3 a Ž k 1 ,k 2 ,k 3 . 2 s2 b bX 2
b Ž k 1 ,k 2 ,k 3 . s3 b bX k 2)H
s
< k1H < 2 ½ k3H < k3H < 2
q
s12 w 2 1 x x 23r2
x1 ²1 2 :x 3 a Ž k 1 ,k 2 ,k 3 . s2 b bX
y
²1 2 : k 3)H (x 1 k3H
)
qH x 1 y k 3)H x 2 b Ž k 1 ,k 2 ,k 3 . s3 b bX
q 1q
ž (x 1 x2 w1 2x / (x 2
5 . Ž 5.30 .
5.4. The NNLO impact factor qg ) ™ qQQ

We consider the amplitude q g ™ qQ Q g for the production of three quarks in the
forward-rapidity region of quark a ŽFig. 3d. in the kinematics Ž5.1.. Using Eq. Ž2.14.
and the subamplitudes of non-PT type, with two gluons of opposite helicities w78x, we
obtain
A q g ™ q Q Q g Ž py
a
n1
,k 1n 1 ,k 2n 2 ,ky
3
n 2 < nXbX
p b , p bn b .
n1
1 X
s 2 s I q ; q Q Q Ž py
a ;k 1n 1 ,k 2n 2 ,ky
3
n2
. ig f b b c C g ; g Ž p bn b ; p bnXbX . , Ž 5.31 .
t
with NNLO impact factor q g ) ™q Q Q

n1
I q ; q Q Q Ž py
a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
n1
s g 3 lcd 3 a d d1 d 2 A1q ; q Q Q Ž py
a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
1 n1
y lcd1 a d d 3 d 2 A q2 ; q Q Q Ž py
a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
Nc
n1
qlcd1 d 2 d d 3 a B1q ; q Q Q Ž py
a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
y
1
Nc
lcd 3 d 2 d d1 a B2q ; q Q Q Ž py
a
n1
;k 1n 1 ,k 2n 2 ,ky
3
n2
. y dq Q Ž 1 l 3. . Ž 5.32 .
The term proportional to d q Q is due to the interference of identical quarks Ži.e. with the
same flavor and helicity . in the final state. The NNLO impact factor allows for four
helicity configurations. From the non-PT subamplitudes w78x we obtain
A1q ; q Q Q Ž py q y q
a ;k 1 ,k 2 ,k 3 .
x 1 x 3 a Ž k 1 ,k 3 ,k 2 . (x x 3
x3 < qH < 2 g Ž k 1 ,k 3 ,k 2 .
si
½( x2 < k1H < 2
q
1 2
1 y x1 s23 < k 1 H < 2

q
(x 1 x2 x 3 s23
x1 g Ž k 1 ,k 3 ,k 2 . s123
q
(x 2 x 3 s23 < k 1 H < 2 (x 1 a Ž k 3 ,k 1 ,k 2 . s a23 y
(x 1
5 ,
i g Ž k 1 ,k 3 ,k 2 .
A q2 ; q Q Q Ž py q y q
a ;k 1 ,k 2 ,k 3 . s
(x 2 x 3 s23 ž (x 1
(
q x 1 a Ž k 3 ,k 1 ,k 2 . ,
/
x 1 x 2 b Ž k 1 ,k 3 ,k 2 . (x x3
x3 < qH < 2
B1q ; q Q Q Ž py q y q
a ;k 1 ,k 2 ,k 3 . s i
½( x3 < k1H < 2
y
1 2
1 y x1 s23 < k 1 H < 2
x1 a Ž k 3 ,k 1 ,k 2 . x1
q ( x2 x3 s23
y
(x 2 x 3 s23 < k 1 H < 2
g Ž k 1 ,k 3 ,k 2 . s123
(
= x 1 a Ž k 3 ,k 1 ,k 2 . s a23 y
(x 1
5 ,
(
i x1 x2 x3
B2q ; q Q Q Ž py q y q
a ;k 1 ,k 2 ,k 3 . s
< k1H < 2 ž( x3
b Ž k 1 ,k 3 ,k 2 . q ( x2
a Ž k 1 ,k 3 ,k 2 .
/
Ž 5.33 .
with a , b ,g defined in Eq. Ž5.8. and
A qi ; q Q Q Ž pq y y q q ;qQQ
a ;k 1 ,k 2 ,k 3 . s A i Ž py q y q
a ;k 1 ,k 3 ,k 2 . , i s 1,2,
Biq ; q Q Q Ž pq y y q
a ;k 1 ,k 2 ,k 3 . s Biq ; q Q Q Ž py q y q
a ;k 1 ,k 3 ,k 2 . , i s 1,2 . Ž 5.34 .
Note that for each helicity configuration, we have the following relation between the
functions A and B:
A1q ; q Q Q q B1q ; q Q Q s A q2 ; q Q Q q B2q ; q Q Q , Ž 5.35 .
5.5. NNLO impact factors in the high-energy limit
The amplitudes Ž5.10., Ž5.11., Ž5.21. and Ž5.31. have been computed in the kinematic
limit Ž5.1., in which they factorize into an effective amplitude with a ladder structure,
made of a three-parton forward cluster and a LO impact factor connected by a gluon
exchanged in the crossed channel ŽFig. 3.. In the limits y 1 , y 2 4 y 3 or y 1 4 y 2 , y 3 ,
the amplitudes must factorize further into NLO impact factors or into NLO Lipatov
vertices for the production of two partons along the ladder. Such limits constitute then
necessary consistency checks, and we display them in this section.
Fig. 4. Limits of the amplitude for the production of three gluons in the forward-rapidity region of gluon pa ,
for y1 4 y 2 , y 3 Ža. and y1 , y 2 4 y 3 Žb..
In the limit, y 1 4 y 2 , y 3 , the NNLO impact factor, g ) g g g g, Eq. Ž5.10.,

factorizes into a NLO Lipatov vertex for the production of two gluons convoluted with a
™
multi-Regge ladder ŽFig. 4a.
3 X X XX
lim
y14y 2,y 3
½Ž ig . Ý
sgS 3
1 2 X X XX
s ig f a d1 c C g ; g Ž pan a ;k 1n 1 .
t1 ½Ž ig . Ý
sgS 2
5
f c ds 2 c f c ds 3 c A g g Ž q1 ,ksns2 2 ,ksns3 3 ,q2 . ,
Ž 5.36 .
with the NLO Lipatov vertex, g g
given by w38–40x
) )
™
g g, for the production of two gluons k 2 and k 3
q1)H q2 H x2 1
A g g Ž q1 ,kq
2 ,kq
3 ,q 2 . s 2
k2 H x 3 ²23:
, (
A g g Ž q1 ,kq y
2 ,k 3 ,q 2 .
k 2)H 1 k 32H < q1 H < 2 k 22 H < q2 H < 2 s 3 b bX k 2 H k 3 H

s y2
k2 H ½ y
s23
2
Ž ky y
2 q k3 . kq
3
q
Ž kq q
2 q k3 . ky
2
q
ky q
2 k3
Ž q2 H qk 3 H . q2 H qk 3 H ky y
2 q k3 kq q
2 q k3
q
s 3 b bX
y
s23 ky
2
k2 H y
kq
3
k 3H
5 Ž 5.37 .
with exchanged momenta in the t channel q1 s yŽ paX q pa ., q2 s p bX q p b , three-par-

ticle invariant s3 b bX s Ž k 3 q q2 . 2 , yŽ< q2 H qk 3 H < 2 q ky q.
2 k 3 , and with the mass-shell
2
conditions k i s < k i H < rk i for i s 2,3.
y q
In the collinear limit, k 2 s zP and k 3 s Ž1 y z . P, the NLO Lipatov vertex Ž5.37.

reduces to the splitting factor Ž4.19., and amplitude Ž5.10. factorizes into a multi-Regge
amplitude Ž4.11. times a collinear factor Ž4.18.
lim A g g ™ 4 g Ž pan a ,k 1n 1 < k 2n 2 ,k 3n 3 < p bnXbX , p bn b .
k2<< k3
s Ý A g g ™ 3 g Ž pan a ,k 1n 1 < P n < p bnXbX , p bn b . P Splityg ™

n
gg
Ž k 2n 2 ,k 3n 3 . ,
n
™
In the limit y 1 , y 2 4 y 3 , the NNLO impact factor in Eq. Ž5.10. factorizes into a
NLO impact factor, g ) g g g, Eq. Ž4.8., convoluted with a multi-Regge ladder ŽFig.
4b.
3 X X XX
lim
y1,y 24y 3
½Ž ig . Ý
sgS 3
2 X
½
s Ž ig . Ý
sgS 2
f a ds 1c f c ds 2 cA g ;2 g Ž pan a ;ksns1 1 ,ksns2 2 . 5
1 X XX
= ig f c d 3 c C g Ž q1 ,k 3n 3 ,q2 . , Ž 5.38 .
t1
with q1 s yŽ pa q k 1 q k 2 ., and with LO Lipatov vertex C g Ž q1 ,k 3n 3 ,q2 ., Eq. Ž4.10..
In the limit y 1 4 y 2 , y 3 , the functions A and B in Eq. Ž5.13. – Ž5.20. fulfill the
NNLO impact factor, g ) g ™

relations A 2g ; g q q s A 3g ; g q q s 0, B2g ; g q q s yA1g ; g q q, and B1g ; g q q s yA 4g ; g q q, thus the
g q q, Eq. Ž5.12., factorizes into a NLO Lipatov vertex
for the production of a qq pair convoluted with a multi-Regge ladder ŽFig. 5a.,
lim I g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky

3
n2
.
y14y 2,y 3
1 X
s ig f a d1 c C g ; g Ž pan a ;k 1n 1 . g2 Ž lclc . d 3 d 2 A q q Ž q1 ,k 2n 2 ,ky3 n 2 ,q2 .
t1
X
n2
q Ž lclc . d 3 d 2 A q q Ž q1 ,ky ,k 2n 2 ,q2 . 4, Ž 5.39 .
™ q q, for the production of a qq pair w40,41x
3
with the NLO Lipatov vertex, g ) g )
A q q Ž q1 ,kq y
2 ,k 3 ,q 2 .
kq kq <
3 q2 H
<2 ky <
3 k 3 H q1 H
<2 kq )
3 k 2 H Ž q 2 H qk 3 H .
s y2 ( ½ 2
kq
3 Ž kq2 q kq3 . s23
q
k 2 H Ž ky y
2 q k 3 . s 23
q
kq
2 s3 b b
X
Ž q2 H qk 3 H . ky2 kq3 y k 2)H k 3 H y Ž q2)H q k 3)H . k 3 H < k3H < 2

q
k 2 H s23
y
s23 5 ,
Ž 5.40 .
with q1 , q2 , and s3 b bX as in Eq. Ž5.37..
In the collinear limit, k 2 s zP and k 3 s Ž1 y z . P, the NLO Lipatov vertex Ž5.40.
reduces to the splitting factor Ž4.19., and amplitude Ž5.11. factorizes into a multi-Regge
amplitude Ž4.11. times a collinear factor Ž4.18.
lim A g g ™ g q q g Ž pan a ,k 1n 1 < k 2n 2 ,ky
3
n 2 < nXbX
p b , p bn b .
k2<< k3

n
qq
Ž k 2n 2 ,ky3 n 2 . .
n
In the limit y 2 , y 3 4 y 1 , the functions A and B in Eq. Ž5.13. – Ž5.20. fulfill the
relations A1g ; g q q s A 4g ; g q q s 0, B2g ; g q q s yA 3g ; g q q, and B1g ; g q q s yA 2g ; g q q thus the
Fig. 5. Same as Fig. 4 for the production of a quark–antiquark pair and a gluon in the forward-rapidity region
of gluon pa .
Fig. 6. Same as Fig. 4 for the production of a quark and two gluons in the forward-rapidity region of quark pa .
NNLO impact factor, Eq. Ž5.12., factorizes into a NLO impact factor, g ) g
Ž4.12., convoluted with a multi-Regge ladder ŽFig. 5b.,
™ q q, Eq.
lim I g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
y 2,y 34y 1
s g 2 Ž lcl a . d 3 d 2 A g ; q q Ž pan a ;k 2n 2 ,ky

3
n2
. q Ž lalc . d 3 d 2 A g ; q q Ž pan a ;ky3 n 2 ,k 2n 2 . 4
1 X
= ig f c d1 c C g Ž q1 ,k 1n 1 ,q2 . , Ž 5.41 .
t1
with q1 s yŽ pa q k 2 q k 3 ..
In the limit y 1 4 y 2 , y 3 , the functions A and B in Ž5.23. – Ž5.30. fulfill the relations
n1
B1q ; q g g Ž py
a ;k 1n 1 ,ksns2 2 ,ksns3 3 . s A q ; q g g Ž py
a
n1
;k 1n 1 ,ksns3 3 ,ksns2 2 . ,
n1
B2q ; q g g Ž py
a ;k 1n 1 ,k 2n 2 ,k 3n 3 . s B2q ; q g g Ž py
a
n1
;k 1n 1 ,k 3n 3 ,k 2n 2 .
n1
s y Ž A q ; q g g Ž py
a ;k 1n 1 ,k 2n 2 ,k 3n 3 .
n1
qA q ; q g g Ž py ;k 1n 1 ,k 3n 3 ,k 2n 2 . . ,
™
a
)
thus the NNLO impact factor, q g q g g, Eq. Ž5.22., factorizes into a NLO Lipatov
vertex for the production of two gluons Ž5.37. convoluted with a multi-Regge ladder
ŽFig. 6a.
n1
lim I q ; q g g Ž py
a ;k 1n 1 ,k 2n 2 ,k 3n 3 .
y14y 2,y 3
n1
s g lcd1 a C q ; q Ž py
a ;k 1n 1 .
1 2 X X XX
=
t1 ½Ž ig . Ý
sgS 2
5
f c ds 2 c f c ds 3 c A g g Ž q1 ,ksns2 2 ,ksns3 3 ,q2 . . Ž 5.42 .
In the limit y 1 , y 2 4 y 3 , the functions A and B in Ž5.23. – Ž5.30. fulfill the relations
n1
A q ; q g g Ž py
a
n1
;k 1n 1 ,k 2n 2 ,k 3n 3 . s 0,
n1
B2q ; q g g Ž py
a ;k 1n 1 ,k 2n 2 ,k 3n 3 . s yA q ; q g g Ž py
a
n1
;k 1n 1 ,k 2n 2 ,k 3n 3 . ,
n1
B2q ; q g g Ž py
a
n1
;k 1n 1 ,k 3n 3 ,k 2n 2 . ,
Fig. 7. Limit of the amplitude for the production of a quark and a quark–antiquark pair in the forward-rapidity
region of quark pa , for y1 4 y 2 , y 3 .
™
thus the NNLO impact factor, Eq. Ž5.22., factorizes into a NLO impact factor,
q g ) q g, Eq. Ž4.14., convoluted with a multi-Regge ladder ŽFig. 6b.
X
n1
lim I q ; q g g Ž py
a ;k 1n 1 ,k 2n 2 ,k 3n 3 . s g 2 Ž l d 2lc . d1 a A q ; q g Ž py
a
n1
;k 1n 1 ,k 2n 2 .
y1,y 24y 3
X
n1
q Ž lcl d 2 . d1 a B q ; q g Ž py
a ;k 1n 1 ,k 2n 2 . 4
1 X XX
= ig f c d 3 c C g Ž q1 ,k 3n 3 ,q2 . , Ž 5.43 .
t1
In the limit y 1 4 y 2 , y 3 , the function A 2 in Eq. Ž5.33. vanishes, A q;
™
qQQ
2 s 0, and
)
using Eqs. Ž5.33.-Ž5.35. the NNLO impact factor, q g qQ Q, Eq. Ž5.32., factorizes
into a NLO Lipatov vertex for the production of a qq pair Ž5.40. convoluted with a
multi-Regge ladder ŽFig. 7.
n1
lim I q ; q Q Q Ž py
a ;k 1n 1 ,k 2n 2 ,ky
3
n2
.
y14y 2,y 3
n1
1 X
s g lcd1 a C q ; q Ž py
a ;k 1n 1 . g2 Ž lclc . d 3 d 2 A q q Ž q1 ,k 2n 2 ,ky3 n 2 ,q2 .
t1
X
n2
q Ž lclc . d 3 d 2 A q q Ž q1 ,ky
3 ,k 2n 2 ,q2 . 4. Ž 5.44 .
5.6. NNLO impact factors in the triple collinear limit
In the triple collinear limit, k i s z i P, with z 1 q z 2 q z 3 s 1 a generic amplitude must

factorize as w20,21x
lim A . . . d1 d 2 d 3 . . . Ž . . . ,k 1n 1 ,k 2n 2 ,k 3n 3 , . . . .
k1< < k 2 < < k 3
s Ý A . . . c . . . Ž . . . , P n , . . . . P Splityf ™
n
f1 f 2 f 3
Ž k 1n 1 ,k 2n 2 ,k 3n 3 . . Ž 5.45 .
n
Accordingly, we must show that taking the triple collinear limit of the NNLO impact
factors, we can write the amplitudes Ž5.10., Ž5.11., Ž5.21. and Ž5.31. as
lim A f g ™ f 1 f 2 f 3 g Ž pan a ,k 1n 1 ,k 2n 2 ,k 3n 3 < py

b
n n
X b, p b.
b
k1< < k 2 < < k 3
s A f g ™ f g Ž pan a , Pyn a < py

b
n n
X b , p b . P Split
b
f ™ f1 f 2 f 3
na Ž k 1n 1 ,k 2n 2 ,k 3n 3 . , Ž 5.46 .
with f denoting the parton species, A f g ™ f g given in Eqs. Ž3.1., Ž3.4. and Ž3.8., and
with Splityf ™
n
f1 f 2 f 3
the polarized double-splitting functions.
™ ™
In the triple collinear limit, the functions A of Sections 5.1, 5.2, 5.3 and 5.4 yield a
quadratic divergence as s123 0 or si j 0 with i, j s 1,2,3. In the same limit, the
functions B have a single collinear divergence since only two out of the three partons
are color adjacent. However, terms with a single divergence when integrated over the
triple collinear region of phase space yield a negligible contribution w20x, thus we ignore
them.
It is easy to show that a function A g ;3 g , Eqs. Ž5.3. – Ž5.7., differs from its reflection
by a term which contains only a single divergence. Using this property and Eq. Ž5.9., we
obtain a dual Ward identity and a reflection identity for the functions A g ;3 g , up to singly
divergent terms,
A g ;3 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 .
q A g ;3 g Ž pan a ;k 1n 1 ,k 3n 3 ,k 2n 2 . q A g ;3 g Ž pan a ;k 3n 3 ,k 1n 1 ,k 2n 2 . s 0,
A g ;3 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 . s A g ;3 g Ž pan a ;k 3n 3 ,k 2n 2 ,k 1n 1 . . Ž 5.47 .

Using the identities Ž5.47. in the impact factor in Eq. Ž5.10., we can factorize the color
structure on a leg
3 X X XX
Ž ig . Ý f a ds 1c f c ds 2 c f c ds 3 c A g ;3 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 .
sgS 3
3
Ž ig . XX X X
s f acc Ý f c ds 1c f c ds 2 ds 3 A g ;3 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 .
3 sgS 3
XX
s igf acc ½ g2 Ý Ž Fd
sgS 2
s1
5
F ds 2 . c d 3 A g ;3 g Ž pan a ;ksns1 1 ,ksns2 2 ,k 3n 3 . , Ž 5.48 .
where Ž F a . b c ' if b ac. Thus amplitude Ž5.10. can be put in the form of Eq. Ž5.46. with
collinear factor
Splityg ™
n
3g
Ž k 1n 1 ,k 2n 2 ,k 3n 3 . s g 2 Ý Ž Fd s1 F ds 2 . c d 3 splityg ™
n
3g
Ž ksns1 1 ,ksns2 2 ,k 3n 3 . .
sgS 2
Ž 5.49 .
The splitting factors splityg ™
n
3g
are the functions A, Eqs. Ž5.3. – Ž5.7., in the triple
collinear limit, up to singly divergent terms, and thus they fulfill the identities, Eq.
Ž5.47.. The splitting factors of PT type can be soon read off from Eqs. Ž5.3. – Ž5.6., while
for the ones of non-PT type we note that the coefficients of Eq. Ž5.8. reduce to
a Ž k 1 ,k 2 ,k 3 . ™
z1 z 3
z1 q z 3
,
b Ž k 1 ,k 2 ,k 3 . ™ y (zP z 1 2
)
H
w1 2x ,
g Ž k 1 ,k 2 ,k 3 . ™ (z sz z
1 2
123
3
d Ž 1,2,3 . Ž 5.50 .
with
ž(
d Ž 1,2,3 . ' w 1 2 x z 1 ²1 3: q z 2 ²2 3: . ( / Ž 5.51 .
Thus we obtain
splityg ™ 3 g Ž kq q q
1 1
1 ,k 2 ,k 3 . s 2 ,
(z 1 z3
²1 2:²2 3:
splitqg ™ 3 g Ž ky1 ,kq2 ,kq3 . s 2 z 12 1

,
(z 1 z3
²1 2:²2 3:
splitqg ™ 3 g Ž kq y q
z 22 1
1 ,k 2 ,k 3 . s 2 ,
(z 1 z3
²1 2:²2 3:
splitqg ™ 3 g Ž kq q y
z 32 1
1 ,k 2 ,k 3 . s 2 ,
(z 1 z3
²1 2:²2 3:
splityg ™ 3 g Ž kq q y
1 ,k 2 ,k 3 .
2
2 s12 z 2 d Ž 1,2,3 . z2
s
s12 s23 Ž 1 y z1 .
q
s123
q ( z1 z 3
Ž 1 y z 3 . d Ž 1,2,3 . ,
splityg ™ 3 g Ž ky q q g™3g
1 ,k 2 ,k 3 . s splity Ž kq3 ,kq2 ,ky1 . ,
splityg ™ 3 g Ž kq y q g™3g
1 ,k 2 ,k 3 . s ysplity Ž ky2 ,kq1 ,kq3 . y splityg ™ 3 g Ž kq1 ,kq3 ,ky2 . .
Ž 5.52 .
A
In the triple collinear limit of the NNLO impact factor g g ) q q g, the functions
g g qq
™
, Eqs. Ž5.13. – Ž5.20., fulfill the relations A 2g ; g q q s yA1g ; g q q and A 4g ; g q q s yA 3g ; g q q,
and A 3g ; g q q Ž pan a ;k 1n 1 ,k 2n 2 ,ky
3
n2.
s A1g ; g q q Ž pan a ;k 1n 1 ,ky
3
n2 n2.
,k 2 . Thus amplitude Ž5.11. can
be put in the form of Eq. Ž5.46. with collinear factor
Splityg ™
n
g qq
Ž k 1n 1 ,k 2n 2 ,ky3 n 2 . s g 2 Ž lcld1 . d 3 d 2 splityg ™
n
g qq
Ž k 1n 1 ,k 2n 2 ,ky3 n 2 .
q Ž l d1lc . d 3 d 2 splityg ™
n
g qq
Ž k 1n 1 ,ky3 n 2 ,k 2n 2 . , Ž 5.53 .
with
splitqg ™ g q q Ž kq
z 23
y q
1 ,k 2 ,k 3 . s 2 ) z1
1
²1 2:²2 3:
,
splitqg ™ g q q Ž kq
z2 1
q y
1 ,k 2 ,k 3 . s 2 z 3 ( z 1 ²1 2:²2 3:
,
splityg ™ g q q Ž kq y q
1 ,k 2 ,k 3 .
2
2 d Ž 1,3,2 . w 1 2 x d Ž 1,3,2 . z 2 Ž z1 y z 3 . (
z1 z 2 w2 3x
sy
s12 s23 w 1 3 x s123
q
z3 ž (z 1
y
w1 3x z 3 /
(z 2 Ž yz 2 s13 q z 3 s23 q z1 z 2 s123 .
q ,
(z3 Ž 1 y z1 .
splityg ™ g q q Ž kq q y
1 ,k 2 ,k 3 .
2
2 d Ž 1,2,3 . w 1 3 x d Ž 1,2,3 . Ž 1 y z 3 . (z 2 z 3 s12
sy q q . Ž 5.54 .
s12 s23 w 1 2 x s123 (z 1 Ž 1 y z1 .
™
Writing the functions A, Eqs. Ž5.23. – Ž5.30., in the triple collinear limit of the NNLO
impact factor q g ) q g g, the amplitude Ž5.21. can be put in the form of Eq. Ž5.46.
with collinear factor
q™ qgg q™ qgg
Splity n Ž k 1n ,k 2n 2 ,k 3n 3 . s g 2 Ý Ž ld s2 l ds 3 . d1 c splity n Ž k 1n ,ksns2 2 ,ksns3 3 . ,
sgS 2
Ž 5.55 .
with
q™ qgg
2i 1
splity Ž kq1 ,kq2 ,kq3 . s y ,
(z 3
²1 2:²2 3:
q™ qgg
2 iz 1 1
splitq Ž ky1 ,kq2 ,kq3 . s ,
(z 3
²1 2:²2 3:
q™ qgg
splity Ž kq1 ,ky2 ,kq3 .
2
s
2i d Ž 1,3,2 . ž (z w 1 3 x q (z 2 w 2 3 x /
1
q
(z2 Ž 1 y z3 . d Ž 1,3,2.
s12 s23 w 1 3 x s123 (z3
(z 1 z 2 s12
q
1 y z1
(
q z 2 ²2 3 :w 1 3 x ,
2
q
splity ™ q g g Ž kq1 ,kq2 ,ky3 . s 2i d Ž 1,2,3 . ž (z w 1 2 x y (z 3 w 2 3 x /
1
s12 s23 w 1 2 x s123

(z 2 Ž 1 y z 3 . d Ž 1,2,3 . (z 1 z 2 s23
q q . Ž 5.56 .
(z 1 y z1
™
3
In the triple collinear limit of the NNLO impact factor q g ) q Q Q, the functions
A Ž5.33. fulfill the relation A1q; q Q Q s A q;
2
qQQ
. Thus the amplitude Ž5.31 . can be put in the
form of Eq. Ž5.46., with collinear factor
q ™ qQQ
Splity n Ž k 1n ,k 2n 2 ,k 3n 3 . s 2 g 2 lad1 c lad 3 d 2 P splityq ™
½ n
qQQ
Ž k 1n ,k 2n 2 ,k 3n 3 .
q ™ qQQ
yd q Q l ad 3 c l ad1 d 2 P splity n Ž k 3n 3 ,k 2n 2 ,k 1n 1 . , Ž 5.57 . 5
where the second term occurs for the case of identical quarks, and c is the color index of
the parent quark. The splitting factors are
q ™ qQQ
i (z 1 z2 z3 d Ž 1,3,2 .
splity
q ™ qQQ
Ž kq1 ,ky2 ,kq3 . s
s23 ž™ 1 y z1
q
s123 / ,
splitq Ž ky1 ,ky2 ,kq3 . s splityq q Q Q Ž kq1 ,ky3 ,kq2 . . Ž 5.58 .

f ™ f 1 f 2 f 3 Ž yn 1 yn 2 yn 3 . f ™ f1 f 2 f 3 Ž n 1 n 2 n 3 .
The factor splitn k 1 ,k 2 ,k 3 can be obtained from splity n k 1 ,k 2 ,k 3
in Eqs. Ž5.52., Ž5.54., Ž5.56. and Ž5.58. by exchanging ² ij : with w ji x, and multiplying by
a coefficient S, Eq. Ž3.7., for each quark pair the splitting factor includes.
Using Eq. Ž5.49. and Eqs. Ž5.52. – Ž5.58., and summing over the two helicity states of
partons 1, 2 and 3, we obtain, as in Section 4.4, the two-dimensional Altarelli–Parisi
polarization matrix at fixed color and helicity of the parent parton,
Splitlf ™ f 1 f 2 f 3 Ž k 1n 1 ,k 2n 2 ,k 3n 3 . Split rf ™ f 1 f 2 f 3 Ž k 1n 1 ,k 2n 2 ,k 3n 3 .
)
Ý
n 1n 2 n 3
Plrf ™ f 1 f 2 f 3 ,
X 4g4
s d cc 2 Ž 5.59 .
s123
f ™ f1 f 2 f 3 f ™ f1 f 2 f 3 f ™ f1 f 2 f 3 f ™ f1 f 2 f 3 . )
q ™ q f2 f3 q™ q g g q™ q Q Q
and P q1 ™ q1 q 2 q 2 , where the last splitting
where Pqq s Pyy , and Pyq s Ž Pqy . For splitting functions of
type P , namely for P , P
function is for identical quarks, helicity conservation on the quark line sets the
off-diagonal elements equal to zero.
Averaging over the trace of matrix Ž5.59., i.e. over color and helicity of the parent
parton, we obtain the unpolarized Altarelli–Parisi splitting functions w20x
<Splityf ™ ² P f ™ f1 f 2 f 3 : ,
1 4g4
Ý n
f1 f 2 f 3
Ž k 1n 1 ,k 2n 2 ,k 3n 3 . < 2 s 2 Ž 5.60 .
2C nn 1n 2 n 3 s123
with C defined below Eq. Ž4.22.. For ² P g ™ g 1 g 2 g 3 :, the sum over colors can be
immediately done using Eq. Ž2.5., and it yields
<Splityg ™
n
g1 g 2 g 3
Ž k 1n 1 ,k 2n 2 ,k 3n 3 . < 2 s 4C
C4 Ž Nc . Ý <splityg ™
n
3g
Ž ksns 1 ,ksns 2 ,ksns 3 . < 2 ,
1 2 3
sgS 3
Ž 5.61 .
with CnŽ Nc . as in Eq. Ž2.8.. Eq. Ž5.61. shows that for the purely gluonic unpolarized
Since the averaged trace of P f ™ f 1 f 2 f 3 is ² P f ™ f 1 f 2 f 3 : s tr P f ™ f 1 f 2 f 3r2 s Pqq

f ™ f1 f 2 f 3
splitting function the color factorizes.
f ™ f1 f 2 f 3
,
we have checked that for the diagonal elements, Pqq , our expressions agree with
the unpolarized splitting functions of Ref. w20x by setting there the RS parameter e s 0.
Finally, for the off-diagonal elements of the splitting functions of type P g ™ g f 2 f 3 we
obtain
2
g ™ g1 g 2 g 3
s123 2 w s 1 s 2 x zs 1 zs 2
Pqy s CA2 Ý
sgS 3 ss1s 2 ½ y
ss 1s 2Ž 1 y zs 3 . zs 3
(z s 2 zs 3 2 Ž 1 y zs 2 . zs 2
y
ss1s 3
Ds 2 Ds 3 y3 q
ž Ž 1 y zs . zs
3 3
/
zs1 zs 2 Ds22Ž 1 y 2 zs 3 . zs 1 zs 3 Ds23Ž 1 y 2 zs 2 .
with
y
Ž 1 y zs . zs 3 3
y
Ž 1 y zs . zs
2 2
5 , Ž 5.62 .
Di s w i j x z j q w i k x z k
( ( with i , j,k s 1,2,3 and j,k / i Ž 5.63 .
and
g ™ g1 q2 q3
Ž C P g ™ g 1 q 2 q 3 Ž ab . q CA Pqy
g ™ g 1 q 2 q 3 Ž nab .
1
Pqy s ., Ž 5.64 .
2 F qy
where the abelian and non-abelian terms are
g ™ g 1 q 2 q 3 Žab.
2 s123 2
Pqy s
s12 s13
½ z D y 2(z
1 1 2 z3 D2 D3 , 5
g ™ g 1 q 2 q 3 Žnab.
s123 1
Pqy s Ý
sgS 2 s1 s 2 s1 s 3
2
½ y z 1 D 12 q zs 2 zs 3 Ds 2 Ds 3
2
(
2 w s 2 s 3 x zs 2 zs 3 s1 s 2 s1 s 3 s1 s 3 2 Ds23 zs2 2 zs 3
q q yzs 2 Ds22 q
ss2 2 s 3Ž 1 y z 1 . z 1 ss 2 s 3 Ž 1 y z1 . z1
2 zs 2Ž zs 3 y z 1 .
(
y zs 2 zs 3 Ds 2 Ds 3 1 q
ž Ž 1 y z1 . z1 /5 . Ž 5.65 .
We have checked that Eqs. Ž5.62. – Ž5.65. agree with the corresponding spin-corre-
lated splitting functions of Ref. w21x after contracting them with a parent-gluon polariza-
tion as in Appendix E, and after setting the RS parameter e s 0.
6. Four-parton forward clusters
The procedure of Sections 4 and 5 can be clearly extended to n-parton forward

clusters. In a forward cluster there are one incoming and n outgoing partons. Thus, for
purely gluonic clusters there are 2 nq 1 helicity configurations. However, in the high-en-
ergy limit two of these are subleading, thus an n-gluon forward cluster contains
2Ž2 n y 1. helicity configurations. For n-parton forward clusters including q q pairs, all
the helicity configurations are leading; then an easy counting yields 2 n helicity configu-
rations for the one including a q q pair, 2 ny 1 for the one including two q q pairs, and so
on. For n s 3, we obtain the helicity configurations dealt with in Section 5.
6.1. The NNNLO impact factor gg ) ™ gggg

Here we analyse in detail the four-gluon forward cluster. We take the production of
five gluons with momenta k 1 , k 2 , k 3 , k 4 and p bX in the scattering between two partons
of momenta pa and p b , and we take partons k 1 , k 2 , k 3 and k 4 in the forward-rapidity
region of parton pa ŽFig. 8a.,
y 1 , y 2 , y 3 , y4 4 y bX ; < k 1 H < , < k 2 H < , < k 3 H < , < k 4 H < , < p bX H < . Ž 6.1 .
Using Eqs. Ž2.1., Ž2.2. and Ž2.10. and the subamplitudes of non-PT type, with four
gluons of q helicity and three gluons of y helicity w75x, we obtain
A g g ™ 5 g Ž pan a ,k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 < p bnXbX , p bn b . s

s
s4 g5 C g ; g Ž p bn b ; p bnXbX . Ý A g ;4 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 ,ksns4 4 .
< qH < 2 sgS 4
X X
=tr Ž l al ds 1 l ds 2 l ds 3 l ds4 l bl b y l al ds 1 l ds 2 l ds 3 l ds4 l bl b
X X
yl bl bl ds4 l ds 3 l ds 2 l ds 1 l a q l bl bl ds4 l ds 3 l ds 2 l ds 1 l a .
X
qB g ;4 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 ,ksns4 4 . tr Ž l al ds 1 l ds 2 l ds 3 l bl bl ds4
X X
yl al ds 1 l ds 2 l ds 3 l bl bl ds4 y l bl bl ds 3 l ds 2 l ds 1 l al ds4
X
ql bl bl ds 3 l ds 2 l ds 1 l al ds4 . q D g ;4 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 ,ksns4 4 .
X X
=tr Ž l al ds 1 l ds 2 l bl bl ds 4 l ds 3 y l al ds 3 l ds 4 l bl bl ds 2 l ds 1 . , Ž 6.2 .
with the sum over the permutations of the four gluons 1, 2, 3 and 4. From the PT
subamplitudes Ž2.10. we obtain the functions of Žyqqqq . helicities
A g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . s C g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . An Ž k 1 ,k 2 ,k 3 ,k 4 . ,
Ž 6.3 .
where
n s sign Ž na q n 1 q n 2 q n 3 q n4 .
and
qH x1 1
Aq Ž k 1 ,k 2 ,k 3 ,k 4 . s y2'2
k1H ( x 4 ²12:²23:²34:
and
kq
i
xi s , i s 1,2,3,4 Ž x 1 q x 2 q x 3 q x 4 s 1. . Ž 6.4 .
kq q q q
1 q k2 q k3 q k4
Fig. 8. Amplitude for the production of five gluons, with gluons k 1 , k 2 , k 3 and k 4 in the forward-rapidity
region of gluon pa .
As in Eq. Ž5.6., the functions C g ;4 g are
1 na s y
C g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . s
½ x i2 n i s y i s 1,2,3,4
with n s q, Ž 6.5 .
From the non-PT subamplitudes w75x we have obtained the functions of Žyyqqq .
helicities. We do not reproduce them here because they are quite lengthy. They are
available from the authors upon request.
Using the UŽ1. decoupling equations for one and two photons, the functions B and D
in Eq. Ž6.2. can be written as
B g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 .
s y A g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . q A g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 4n 4 ,k 3n 3 .
qA g ;4 g Ž pan a ;k 1n 1 ,k 4n 4 ,k 2n 2 ,k 3n 3 . q A g ;4 g Ž pan a ;k 4n 4 ,k 1n 1 ,k 2n 2 ,k 3n 3 . , Ž 6.6 .
D g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 .
s A g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . q A g ;4 g Ž pan a ;k 1n 1 ,k 3n 3 ,k 2n 2 ,k 4n 4 .
qA g ;4 g Ž pan a ;k 3n 3 ,k 1n 1 ,k 2n 2 ,k 4n 4 . q A g ;4 g Ž pan a ;k 1n 1 ,k 3n 3 ,k 4n 4 ,k 2n 2 .
qA g ;4 g Ž pan a ;k 3n 3 ,k 1n 1 ,k 4n 4 ,k 2n 2 . q A g ;4 g Ž pan a ;k 3n 3 ,k 4n 4 ,k 1n 1 ,k 2n 2 . . Ž 6.7 .

In the quadruple collinear limit, k 1 < < k 2 < < k 3 < < k 4 , Section 6.4, the function A has a
triple collinear divergence; the function B, whose gluon 4 is not color adjacent to gluons
1, 2 and 3, has only a double collinear divergence; the function D, where gluon 1 is
adjacent to 2 and gluon 3 is adjacent to 4 but the pairs are not adjacent one to another,
has two single collinear divergences.
Using Eqs. Ž6.6. and Ž6.7., we can rewrite Eq. Ž6.2. as
A g g ™ 5 g Ž pan a ,k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 < p bnXbX , p bn b .
4 X X XX XX XXX
s 2 s Ž ig . ½ sgS 4
Ý f a ds 1c f c ds 2 c f c ds 3 c f c ds 4 c
A g ;4 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 ,ksns4 4 . 5
1 X XXX
t
where the NNLO impact factor g ) g g g g g is enclosed in curly brackets, and

includes 30 helicity configurations, in agreement with the counting above.
™
6.2. NNNLO impact factors in the high-energy limit
The amplitude Ž6.8. has been computed in the kinematic limit Ž6.1., in which it
factorizes into a four-gluon cluster and a LO impact factor connected by a gluon
exchanged in the cross channel. In the limits y 1 , y 2 , y 3 4 y4 or y 1 , y 2 4 y 3 , y4 ,
or y 1 4 y 2 , y 3 , y4 , Eq. Ž6.8. must factorize further into a NNLO impact factor or into
a NLO impact factor times a NLO Lipatov vertex, or into a NNLO Lipatov vertex ŽFig.
8., respectively. While the first two limits constitute necessary consistency checks, the
last one allows us to derive the so far unknown NNLO Lipatov vertex for the production
of three gluons along the ladder.
In the limit y 1 , y 2 , y 3 4 y4 , the NNNLO impact factor, g ) g g g g g, in Eq.
Ž6.8. factorizes into a NNLO impact factor, g ) g g g g, Eq. Ž5.10., convoluted with a ™ ™
multi-Regge ladder ŽFig. 8a.
lim
y1,y 2,y 34y4
4 X X XX XX XXX
½
= Ž ig .
sgS 4
Ý f a ds 1c f c ds 2 c f c ds 3 c f c ds 4 c
3 X X XX
½
s Ž ig . Ý
sgS 3
1 XX XXX
= ig f c d4 c
C g Ž q1 ,k 4n 4 ,q2 . , Ž 6.9 .
t1
with q1 s yŽ pa q k 1 q k 2 q k 3 ., q2 s p bX q p b , and with LO Lipatov vertex

C g Ž q1 ,k 3n 3 ,q2 ., Eq. Ž4.10..
™ ™
In the limit y 1 , y 2 4 y 3 , y4 , the NNNLO impact factor in Eq. Ž6.8. factorizes into
a NLO impact factor, g ) g g g, Eq. Ž4.8., times a NLO Lipatov vertex for production
of two gluons g ) g ) g g Ž5.37., convoluted with a multi-Regge ladder ŽFig. 8b.
lim
y1,y 24y 3,y4
4 X X XX XX XXX
½
= Ž ig . Ý
sgS 4
f a ds 1c f c ds 2 c f c ds 3 c f c ds 4 c
2 X
½
s Ž ig . Ý
sgS 2
f a ds 1c f c ds 2 cA g ; g g Ž pan a ;ksns1 1 ,ksns2 2 . 5
1 2 X XX XX XXX
=
t1 ½Ž ig . Ý
sgS 2
f c ds 3 c f c ds 4 c
A g g Ž q1 ,ksns3 3 ,ksns4 4 ,q2 . , 5 Ž 6.10 .
6.3. The NNLO LipatoÕ Õertex
In the limit y 1 4 y 2 , y 3 , y4 , the NNNLO impact factor in Eq. Ž6.8. factorizes into
a NNLO Lipatov vertex convoluted with a multi-Regge ladder ŽFig. 8c.
lim
y14y 2,y 3,y4
4 X X XX XX XXX
½
= Ž ig . Ý
sgS 4
f a ds 1c f c ds 2 c f c ds 3 c f c ds 4 c
s ig f a d1 c C g ; g Ž pan a ;k 1n 1 .
1 3 X X XX XX XXX
=
t1 ½Ž ig . Ý f c ds 2 c f c ds 3 c f c ds 4 c
A3 g Ž q1 ,ksns2 2 ,ksns3 3 ,ksns4 4 ,q2 . 5 Ž 6.11 .
™
sgS 3
with the NNLO Lipatov vertex, g ) g ) g g g, for the production of three gluons k 2 ,
k 3 and k 4 enclosed in curly brackets in the right-hand side, with
x2 1 q1)H q2 H
A3 g Ž q1 ,kq q q '
2 ,k 3 ,k 4 ,q 2 . s y2 2 ( x 4 ²2 3:²3 4: k2 H
, Ž 6.12 .
A3 g Ž q1 ,ky q q
2 ,k 3 ,k 4 ,q 2 .
< q1 H < 2 k 2 H Ž k 3)H . 2 Ž q2)H q k 4)H . x 2

s 2'2
½ y
s4 b bX k 2)H s23 Ž < k 3 H < 2 x 2 q < k 2 H < 2 x 3 .
k2 H
q Ž q1)H q2)H k 3)H ²3 4:
s4 b bX ²3 4: k 2)H s23 x 3 (
q²2 4: Ž q1)H y k 2)H . Ž q2)H q k 4)H . w 2 3 x . x 3 (

yk 3)H Ž Ž q1)H y k 2)H . s23 q q1)H s34 . x 4 (
< q1 H < 2 k 2 H Ž k 3)H . 2 x 2 x 3 (
q 2
²3 4: k 2)H s23 Ž < k3H < x2 q < k2 H < 2 x3 . x4 (
1
q ž y< k 2H
< 2 k 3)H
²3 4: k 2)H s23 (x 3 x4
( ž
q x 2 yq1H Ž q1)H y k 2)H . k 3)H x 2 (
yq ž yk )
1H
)
2 H k3H (x 2 (
q ²2 3: k 3)H x 3 q k 2 H w 2 3 x x 3 (
( /
q²2 4: k 3)H x 4 q k 2)H Ž ²2 3: k 3)H q k 2 H w 2 3 x . x 3
ž (
< q2 H < 2 x 2 ²2 3: x 3 q ²2 4: x 4
ž ( ( /
q²2 4: k 3)H x 4 ( /// q
(
²2 3:²3 4: s234 x 4
2
q2 H k 2 H Ž q1)H y k 2)H . ž q (x q w 3 4 x (x /
)
2H 3 4
q
s2 a aX s4 b bX ²3 4: k (x )
2H 4
2 2
< q1 H < Ž k . k k (x
2H
) )
3H 4 H 3
q
²3 4: s23 Ž < q1 H yq2 H < 2 q s234 .Ž < k 3 H < 2 x 2 q < k 2 H < 2 x 3 . x 4 (
< q1 H < 2 k 2 H w 3 4 x ²2 4: k 4)H x 3 q ²2 3: k 3)H x 4
ž ( ( /
q 2
(x 2 x 3 x 4 ²3 4: s23 s234 Ž < q1 H yq2 H < q s234 .
w3 4x
q
²3 4: s23 s234 x 2 x 3 x 4 ( ž yq )
1H k2 H x3 x 4 ²2 3: x 3
ž (
( / ( ž
q²2 4: x 4 q q2 H x 2 k 4 H Ž q2)H q k 4)H . x 2 x 3
q Ž < k 3 H < 2 x 2 q k 2 H Ž k 2)H y q1)H . x 3 . x 4 // 5 , Ž 6.13 .
A 3 g Ž q1 , k q y q
2 ,k 3 ,k 4 ,q 2 .
< q2 H <2k3H x 2 ' 'x 3 < q 2 H < 2 Ž q1 H yk 2 H . k 2)H x 3

s2 2 ' ²2 3 : k 2 H s 34 Ž 1 y x 2 .
q
s 2 a aX k 2 H s 34 Ž 1 y x 2 .
y
< q2 H <2 'x 2 x 33
y
< q 1 H < 2 Ž k 3 H . 2 k 2)H Ž q 2)H q k 4)H . x 2
²2 3 : s 34 Ž 1 y x 2 . s 4 b bX k 2 H s 23 Ž < k 3 H < 2 x 2 q < k 2 H < 2 x 3 .
Ž k 3 H . 2 x 2 Ž y Ž q 2)H q k 4)H . w2 3 x x 3 q q 1)H Ž q 2)H x 2 q w2 4 x x 4 . .

' ' ' '
q
s4 b bX k 2 H s 23 x 3
q 1)H k 3 H w 2 4 x 2 ²2 3 : Ž q 1 H yk 2 H . x 2 q ²3 4 : k 2 H x 4
Ž ' ' .
q
k 2 H s 23 s 34 s 234 x 3 '
y
'
k 3 H k 2)H q 2)H k 2 H x 2 y q 1 H Ž q 2)H q k 4)H . x 2 q k 2 H w 2 4 x x 4
Ž ' ' .
s 2 a aX s4 b bX k 2 H x 2 '
y
' Ž ' '
k 2)H x 3 yq 1 H q 2 H k 4)H x 2 x 3 y Ž q 1 H yk 2 H . k 3 H w 3 4 x x 2 x 4 q q 2 H k 2 H w 2 4 x x 3 x 4 ' .
'
s 2 a aX k 2 H s 34 x 2 x 4
< q 1 H < 2 Ž k 3 H . 2 k 2)H x 2 x 3 '

q
²3 4 : k 2 H s 23 Ž < k 3 H < 2 x 2 q < k 2 H < 2 x 3 . x 4
'
2
< q2 H < w 2 4 x x Ž y ² 2 3 : 'x q ² 3 4 : 'x .
3 2 4
q
² 2 3 : s s 'x34 234 4
2
q2 H w 2 4 x 'x Ž ²3 4 : Ž q
3 q k . 'x y ²2 3 : k 'x .
)
2H
)
4H 2
)
2H 4
q
s s s 'x x23 34 234 2 4
< q 1 H < 2 Ž k 3 H . 3 k 2)H k 4)H x 2

y
'x 3 x4 ²3 4 : k 2 H s 23 Ž < q 1 H yq 2 H < 2 q s 234 .Ž < k 3 H < 2 x 2 q < k 2 H < 2 x 3 .
< q 1 H < 2 Ž k 3 H . 2 w 2 4 x ² 3 4 : k 4)H x 2 y ² 2 3 : k 2)H x 4

Ž ' ' .
y
x 3 x 4 ²3 4 : k 2 H s 23 s 234 Ž < q 1 H yq 2 H < 2 q s 234 .
'
1
q Ž x 3 Ž yx 2 Ž k 3 H q 2 H k 3)H w2 4 x q k 3 H w2 3 xw3 4 x .
'
k 2 H s 23 s 34 x 2 x 4 x 3
qq 2 H k 3 H k 2)H w 3 4 x x 2 x 3 y q 2 H < k 2 H < 2 w 2 4 x x 3

' .
Ž '
qq 1)H k 3 H x 2 q 2 H w 2 4 x x 3 q k 3 H w 3 4 x x 2 x 3 q k 3 H w 2 4 x x 4 .. , Ž 6.14 .
A3 g Ž q1 ,kq q y
2 ,k 3 ,k 4 ,q 2 .
k 2)H
s 2'2
s2 a aX ²2 3: k 2 H s34 x 2 ' Žq 1 H q2 H
²2 3: k 3)H x 2 '
'
y²2 4:w 3 4 x x 2 Ž q1 H y k 2 H .Ž q2 H q k 4 H . q k 2 H x 3 Ž q2 H s23 q Ž q2 H q k 4 H . s34 . . '
q1)H Ž q2 H q k 4 H . 2 Ž q1 H k 3)H 'x 2 y k 2 H w2 3x x 3 ' .
q
s2 a aX s4 b bX ²2 3: k 2 H x 3 '
< q2 H < 2 k 3 H x 2 x 3 ' < q1 H < 2 Ž k 4 H . 2 Ž ²3 4: k 3)H x 2 q²2 4: k 2)H x 3
' ' .
q q
²2 3: k 2 H s34 Ž 1y x 2 . x 4 x 3 ²2 3:²3 4: k 2 H s234 Ž < q1 H y q2 H < 2 q s234 .
'
< q2 H < 2 Ž q1 H y k 2 H . k 2)H x 3
q
s2 a aX k 2 H s34 Ž 1y x 2 .
q
s4 b bX k 4 H x 2 ' 'x 3 '
q Ž q2 H q k 4 H . Ž Ž ²3 4: k 3)H q k 4 H w 3 4 x . x 2 q²2 4: k 2)H x 3 ' . 'x 4
²2 3: k 2 H s34 x 4
< q2 H < 2 w 2 3 x Ž ²2 4: x 2 q²3 4: x 3

' ' . 'x 4
< q2 H < 2 x 2 x 3 x 4
'
q q
²2 3: s34 s234 ²2 3: s34 Ž 1y x 2 .
w2 3x ²2 4:< k 2 H < 2 ²3 4: k 2 H k 3)H

q
²2 3: k 2 H s34 s234 x 4 ž < q1 H < 2 Ž k 4 H . 2 x 2 q q 2 H
ž 'x 2
q
'x 3
/(
x 43
y q1)H k 2 H k 4 H Ž k 4 H q q2 H x 4 .
/ , Ž 6.15 .
where in Eqs. Ž6.13. – Ž6.15. we have used the three-particle invariants, s2 a aX s Ž k 2 y q1 . 2

and s4 b bX s Ž k 4 q q2 . 2 .
Eq. Ž6.11. must not diverge more rapidly than 1r< qi H < for < qi H < 0, with i s 1,2, in
order for the related cross section not to diverge more than logarithmically. Since Eq.
™
Ž6.11. is proportional to 1r< qi H < 2 , the NNLO Lipatov vertex must be at least linear in
< qi H <,
lim A3 g Ž q1 ,k 1n 1 ,k 2n 2 ,k 3n 3 ,q2 . s O Ž < qi H < . ,
< qiH< ™0 i s 1,2, Ž 6.16 .
which is fulfilled by Eqs. Ž6.12. – Ž6.15..
As a consistency check on Eq. Ž6.11., in the further limits y 2 4 y 3 , y4 or y 2 , y 3
4 y4 , the NNLO Lipatov vertex in Eq. Ž6.11. must factorize into a NLO Lipatov vertex
convoluted with a multi-Regge ladder,
3 X X XX XX XXX
lim
y 24y 3,y4
½Ž ig .
sgS 3
Ý f c ds 2 c f c ds 3 c f c ds 4 c
A3 g Ž q1 ,ksns2 2 ,ksns3 3 ,ksns4 4 ,q2 . 5
X
s ig f c d 2 c C g Ž q1 ,k 2n 2 ,q12 .
1 2 X XX XX XXX
=
t 12 ½Ž ig . Ý
sgS 2
f c ds 3 c f c ds 4 c
A g g Ž q12 ,ksns3 3 ,ksns4 4 ,q2 . , 5 Ž 6.17 .
with q12 s q1 y k 2 , and

3 X X XX XX XXX
lim
y 2,y 34y4
½Ž ig .
sgS 3
Ý f c ds 2 c f c ds 3 c f c ds 4 c
A3 g Ž q1 ,ksns2 2 ,ksns3 3 ,ksns4 4 ,q2 . 5
2 X X XX
½
s Ž ig .
sgS 2
Ý f c ds 2 c f c ds 3 c A g g Ž q1 ,ksns2 2 ,ksns3 3 ,q12 . 5
1 XX XXX
= ig f c d4 c
C g Ž q12 ,k 4n 4 ,q2 . , Ž 6.18 .
t 12
with q12 s q2 q k 4 .
In the triple collinear limit, k 2 s z 2 P, k 3 s z 3 P and k 4 s z 4 P, with z 2 q z 3 q z 4 s 1,
the coefficients of the NNLO Lipatov vertex Ž6.12. – Ž6.15. reduce to the splitting
functions Ž5.52., and amplitude Ž6.11. factorizes into a multi-Regge amplitude Ž4.11.
times a double-collinear factor Ž5.49.
lim A g g ™ 5 g Ž pan a ,k 1n 1 < k 2n 2 ,k 3n 3 ,k 4n 4 < p bnXbX , p bn b .
k2 < < k3< < k4

n
3g
Ž k 1n 1 ,k 2n 2 ,k 3n 3 . .
n
6.4. NNNLO impact factors in the quadruple collinear limit
In the quadruple collinear limit, k i s z i P, with z 1 q z 2 q z 3 q z 4 s 1 a generic

amplitude is expected to factorize as
lim A . . . d1 d 2 d 3 d 4 . . . Ž . . . ,k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 , . . . .
k1< < k 2 < < k 3< < k 4
s Ý A . . . c . . . Ž . . . , P n , . . . . P Splityf ™
n
f1 f 2 f 3 f4
Ž k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . . Ž 6.19 .
n
Accordingly, we show that we can write Eq. Ž6.8. as

lim A g g ™ 5 g Ž pan a ,k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 < py
b
n n
X b,p b.
b
k1< < k 2 < < k 3< < k 4
s A g g ™ g g Ž pan a , Pyn a < py

b
n n
X b , p b . P Split
b
g™4g
na Ž k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . , Ž 6.20 .
by taking the quadruple collinear limit of the NNNLO impact factor.
In the quadruple collinear limit, the functions A g ;4 g of Eq. Ž6.5. yield a cubic
divergence as s1234 s Ž k 1 q k 2 q k 3 q k 4 . 2 0 or si jk 0, or si j 0 with i, j,k s
1,2,3,4. Analogously to Section 5.6, a function A g ;4 g differs from its reflection by a
™ ™ ™
term which contains only a quadratic divergence in the vanishing invariants. Using this
property and Eqs. Ž6.6. and Ž6.7., we obtain a reflection identity and dual Ward
identities, up to quadratically divergent terms,
A g ;4 g Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . s yA g ;4 g Ž pan a ;k 4n 4 ,k 3n 3 ,k 2n 2 ,k 1n 1 . , Ž 6.21 .
A g ;4 g
Ž pan a ;k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4
. qA Ž . g ;4 g
pan a ;k 1n 1 ,k 2n 2 ,k 4n 4 ,k 3n 3
qA g ;4 g
Ž pa ;k 1 ,k 4 ,k 2 ,k 3 . q A Ž pa ;k 4 ,k 1 ,k 2 ,k 3n . s 0,
n n n n an 1 g ;4 g
4 n n 2 n n
3 a 4 1 2 3
Ž 6.22 .
A g ;4 g Ž pan ;k 1n ,k 2n ,k 3n ,k 4n . q A g ;4 g Ž pan ;k 1n ,k 3n ,k 2n ,k 4n .
a 1 2 3 4 a 1 3 2 4
q A g ;4 g Ž pan ;k 3n ,k 1n ,k 2n ,k 4n . q A g ;4 g Ž pan ;k 1n ,k 3n ,k 4n ,k 2n .
a 3 1 2 4 a 1 3 4 2
q A g ;4 g Ž pan ;k 3n ,k 1n ,k 4n ,k 2n . q A g ;4 g Ž pan ;k 3n ,k 4n ,k 1n ,k 2n . s 0 .
a 3 1 4 2 a 3 4 1 2
Ž 6.23 .
We note, however that the last identity is not independent from the first two. Using the
identities Ž6.21. – Ž6.23. in Eq. Ž6.8., we can factorize the color structure on a leg
4 X X XX XX XXX
Ž ig . Ý f a ds 1c f c ds 2 c f c ds 3 c f c ds 4 c
A g ;4 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 ,ksns4 4 .
sgS 4
XXX
s igf acc ½ g3 Ý Ž Fd
sgS 3
s1 F ds 2 F ds 3 . c d 4 A g ;4 g Ž pan a ;ksns1 1 ,ksns2 2 ,ksns3 3 ,k 4n 4 . 5 Ž 6.24 .
thus amplitude Ž6.8. can be put in the form of Eq. Ž6.20. with collinear factor
Splityg ™
n
4g
Ž k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 .
sg3 Ý Ž Fd s1 F ds 2 F ds 3 . c d 4 splityg ™
n
4g
Ž ksns1 1 ,ksns2 2 ,ksns3 3 ,k 4n 4 . . Ž 6.25 .
sgS 3
The splitting factors splityg ™

n
4g
are the functions A of Section 6.1 in the quadruple
collinear limit, up to quadratically divergent terms, and thus they fulfill the identities,
Eqs. Ž6.21. – Ž6.23.. The splitting factors of PT type can be soon read off from Eqs.
Ž6.3. – Ž6.5., while the ones of non-PT type can be given in terms of three functions of
the collinear momenta,
splityg ™ 4 g Ž kq q q q
1 1
'
1 ,k 2 ,k 3 ,k 4 . s 2 2 ,
(
z 1 z 4 1 2 2 3:²3 4:
² :²
splitqg ™ 4 g Ž ky q q q
z 12 1
'
1 ,k 2 ,k 3 ,k 4 . s 2 2 ,
(z 1 z4
²1 2:²2 3:²3 4:
splitqg ™ 4 g Ž kq y q q
z 22 1
'
1 ,k 2 ,k 3 ,k 4 . s 2 2 ,
(z 1 z4
²1 2:²2 3:²3 4:
splitqg ™ 4 g Ž kq q y q
z 32 1
'
1 ,k 2 ,k 3 ,k 4 . s 2 2 ,
(z 1 z4
²1 2:²2 3:²3 4:
splitqg ™ 4 g Ž kq q q y
z 42 1
'
1 ,k 2 ,k 3 ,k 4 . s 2 2 ,
(z 1 z4
²1 2:²2 3:²3 4:
splityg ™ 4 g Ž ky q q q
1 ,k 2 ,k 3 ,k 4 . s yB
B1 Ž 4,3,2,1 . ,
splityg ™ 4 g Ž kq y q q
1 ,k 2 ,k 3 ,k 4 . s B1 Ž 4,3,1,2 . q B1 Ž 4,1,3,2 . q B1 Ž 1,4,3,2 . ,
splityg ™ 4 g Ž kq q y q
1 ,k 2 ,k 3 ,k 4 . s yB
B1 Ž 1,2,4,3 . y B1 Ž 1,4,2,3 . y B1 Ž 4,1,2,3 . ,
splityg ™ 4 g Ž kq q q y
1 ,k 2 ,k 3 ,k 4 . s B1 Ž 1,2,3,4 . ,
splityg ™ 4 g Ž ky y q q
1 ,k 2 ,k 3 ,k 4 . s yB
B2 Ž 4,3,2,1 . ,
splityg ™ 4 g Ž ky q y q
1 ,k 2 ,k 3 ,k 4 . s B3 Ž 1,2,3,4 . ,
splityg ™ 4 g Ž ky q q y
1 ,k 2 ,k 3 ,k 4 . s yB
B3 Ž 1,2,4,3 . q B2 Ž 3,2,4,1 . q B2 Ž 3,2,1,4 . ,
splityg ™ 4 g Ž kq y y q
1 ,k 2 ,k 3 ,k 4 . s yB
B2 Ž 1,4,2,3 . q B3 Ž 3,4,2,1 . y B2 Ž 4,1,2,3 . ,
splityg ™ 4 g Ž kq y q y
1 ,k 2 ,k 3 ,k 4 . s yB
B3 Ž 4,3,2,1 . ,
splityg ™ 4 g Ž kq q y y
1 ,k 2 ,k 3 ,k 4 . s B2 Ž 1,2,3,4 . Ž 6.26 .
with
2 '2 z 3 d Ž 1,2,3 . (z 2 z 3 z 4 ²1 2:
B1 Ž 1,2,3,4 . s y y
²1 2: ²2 3: s34 (z 1 Ž z 3 q z 4 . w1 2x Ž 1 y z1 . Ž z 3 q z 4 .
z3
q ( z1 z 4
Ž 1 y z 4 . e Ž 1,2,3,4 .
2
e Ž 1,2,3,4 . Ž ²2 4: w 1 2 x q ²3 4: w 1 3 x . w 3 4 x
y
s1234 s234
2
z 2 e Ž 1,2,3,4 . w 2 3 x
q ( z1 s234
(z 4 ž(
²1 2: w 2 3 x z 1 ²1 4: y Ž 2 y z 1 . e Ž 1,2,3,4 . /
q , Ž 6.27 .
Ž 1 y z1 . s234
2'2 z2 z3( (z 2 Ž z1 q z 2 . d Ž 1,2,3 .

B2 Ž 1,2,3,4 . s q
w 3 4 x s23 Ž 1 y z1 . (z 4 (z 1 z4 s12
2
d Ž 1,2,3 . e Ž 1,2,3,4 . w 3 4 x
y
(
s1234 z 4 s12 s123
d Ž 1,2,3 . ž (z 1 ( / (
d Ž 3,4,1 . q z 2 d Ž 3,4,2 . q z 1 z 2 z 4 s12 s34
q
(
s1234 z 4 ²1 2: w 3 4 x
(
z 1 z 2 d ) Ž 3,4,2 . z 1²3 4: w 1 2 x Ž ²1 3: w 2 3 x q ²1 4: w 2 4 x .
q y ,
Ž 1 y z1 . s234 s1234 s234
Ž 6.28 .
B3 Ž 1,2,3,4 .
'
2'2 z2 z2 z3
ž' z1 ²2 3: s34 'z 3
²1 2: s23
s
s12 s23 s34 'z 1 Ž z1 q z 2 . 1y z 4
y
z3 q z4 /
z2 '
z 2 Ž z 3 e Ž 1,2,4,3 . s23 y²2 3:²3 4:w 2 4 x .
y ( z1 z 4
Ž 'z 2 z 3 ²1 2:y z 4²1 3: . d Ž2,4,3. y Ž z1 q z 2 .
Ž 'z 2 ²1 2:d Ž2,4,3. 2q'z1 ²1 3: Ž z 2 ²1 2:²2 3:q'z 3 'z 4 ²1 3:²3 4:q z 4²1 4:²3 4: . w2 4x 2 .
y
'z 1 s1234
z 2 s23 s34 d ) Ž 1,3,2 . s12 d Ž 2,4,3 . 2 e Ž 2,3,4,1 . z3

q
'z 1 Ž 1y z 4 . w 2 3 x s123
q
'
s1234 z1 s234
q ( z1
z 2 ²1 3: s234
s34
q 'z 4
²1 3:2 w 2 4 x Ž e Ž 1,2,4,3 . w 2 3 x y e Ž 2,3,4,1 . w 1 2 x .
s1234 s123
z 2 ²1 3:2 s34
y ( z1 s123
Ž y'z 2 e ) Ž2,3,4,1. q'z1 z 4 w2 4x . Ž 6.29 .
with d Ž1,2,3. as in Eq. Ž5.51., and
(
e Ž 1,2,3,4 . s z 1 ²1 4: q z 2 ²2 4: q z 3 ²3 4: . ( ( Ž 6.30 .
As in Section 5.6, summing over the helicities of gluons 1, 2, 3 and 4, one can obtain
the two-dimensional polarization matrix,
Splityg ™ Ž k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . Splityg ™
)
Ý n
4g
n
4g
Ž k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 .
n 1n 2 n 3 n4
Plrg ™ 4 g ,
X 8 g6
s d cc 3 Ž 6.31 .
s1234
g™4g g™4g g™4g g ™ 4 g .)

where Pqq s Pyy , and Pyq s Ž Pqy . Averaging then over the trace of
matrix Ž6.31., i.e. over color and helicity of the parent gluon, one can obtain the
unpolarized Altarelli–Parisi gluon triple-splitting function
<Splityg ™ ²P g™4g: ,
1 8 g6
Ý n
4g
Ž k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . < 2 s 3 Ž 6.32 .
2 Ž Nc2 y 1 . nn 1n 2 n 3 n4 s1234
with ² P g ™ 4 g : s Pqq
g™4g
. As in Section 5.6, the sum over colors can be done using Eq.
Ž2.5., and we obtain
<Splityg ™
n
4g
Ž k 1n 1 ,k 2n 2 ,k 3n 3 ,k 4n 4 . < 2
s 4C
C5 Ž Nc . Ý <splityg ™
n
4g
Ž ksns 1 ,ksns 2 ,ksns 3 ,ksns4 . < 2 , Ž 6.33 .
1 2 3 4
sgS 4
with C5 Ž Nc . as in Eq. Ž2.8.. It is then clear that for the splitting functions P g ™ n g , with
n ) 4, the color will not factorize since LCA, Eq. Ž2.5., is not exact any more. We do
g™4g g™4g
not compute here Pqq and Pqy , all the information about them being already
contained in Eqs. Ž6.26. – Ž6.30..
7. Conclusions
In this paper, the structure of QCD amplitudes in the high-energy limit and in the
collinear limit has been explored beyond NLO. We have computed forward clusters of
three partons and four gluons, which in the BFKL theory constitute the tree parts of
NNLO and NNNLO impact factors for jet production. In the BFKL theory the NNLO
impact factors could be used to compute jet rates at NNLL accuracy. In Sections 5.1,
5.2, 5.3 and 5.4, we have computed the tree parts of the NNLO impact factors for all the
parton flavors. On these we have performed in Section 5.5 a set of consistency checks in
the high-energy limit, and we have obtained in the triple collinear limit ŽSection 5.6. the
polarized, the spin-correlated and the unpolarized double-splitting functions. The last
two agree with previous calculations by Catani–Grazzini and Campbell–Glover, respec-
tively. They can be used to set up general algorithms to compute jet rates at NNLO.
From the four-gluon forward cluster we have obtained in Section 6.1 the tree part of
the purely gluonic NNNLO impact factor. In the quadruple collinear limit, this yields
ŽSection 6.4. the purely gluonic unpolarized triple-splitting functions. They could be
used to compute the three-loop Altarelli–Parisi evolution, or to compute jet rates at
NNNLO. In addition, by separating a central cluster of three gluons out of the
four-gluon forward cluster, we have computed the emission of three gluons along the
ladder, Eqs. Ž6.11. – Ž6.15., which contributes to the NNLO Lipatov vertex. This
constitutes one of the universal building blocks in an eventual construction of a BFKL
resummation at NNLL accuracy.
Finally, inspired by the color structure in the high-energy limit, we have found a
compact color decomposition of the tree multigluon amplitudes in terms of the linearly
independent subamplitudes only, Eq. Ž2.9.. It would be interesting to analyse whether
this structure generalizes to multigluon amplitudes at one loop, and beyond.
The decomposition in rapidity of amplitudes in terms of gauge-invariant parton
clusters performed in this work suggests naturally a modular decomposition of a generic
multiparton amplitude, where each module is an n-parton cluster. Such an approxima-

tion could be tested against existing approximations of multiparton amplitudes w79,80x.
In the high-energy limit, the cluster decomposition seems superior, in that it does not use
only PT-type subamplitudes, like the Kunszt–Stirling approximation w79x, and within a
cluster it is not limited to collinear kinematics, like the Maxwell approximation w80x.
Acknowledgements
We would like to thank Stefano Catani, Walter Giele, David Kosower and Zoltan
Trocsanyi for discussions. We are particularly grateful to Lance Dixon for his valuable
insight.
Appendix A. Multiparton kinematics
We consider the production of n partons of momentum pi , with i s 1, . . . ,n and

n 0 2, in the scattering between two partons of momenta pa and p b5 .
Using light-cone coordinates p "s p 0 " pz , and complex transverse coordinates
p H s p x q ip y , with scalar product 2 p P q s pqqyq pyqqy p H q H
) )
yp H q H , the 4-
momenta are
pa s Ž pq q q
a r2,0,0, pa r2 . ' Ž pa ,0;0,0 . ,
p b s Ž py y y
b r2,0,0,y p b r2 . ' Ž 0, p b ;0,0 . ,
pi s Ž Ž pq y q y
i q pi . r2,Re w pi H x ,Im w pi H x , Ž pi y pi . r2 .
' Ž < pi H < e y i , < pi H < eyy i ; < pi H <cos f i , < pi H <sin f i . , Ž A.1 .
where y is the rapidity. The first notation in Eq. ŽA.1. is the standard representation
p m s Ž p 0 , p x , p y , p z ., while in the second we have the q and - components on the left
of the semicolon, and on the right the transverse components. In the following, if not
differently stated, pi and pj are always understood for 1 ( i, j ( n. From momentum
conservation,
n
0s Ý pi H ,
is1
n
pq
a sy Ý pqi ,
is1
n
py
b sy Ý pyi , Ž A.2 .
is1
5
By convention we consider the scattering in the unphysical region where all momenta are taken as
outgoing, and then we analytically continue to the physical region where pa0 - 0 and p b0 - 0. Thus partons are
ingoing or outgoing depending on the sign of their energy. Since the helicity of a positive-energy Žnegative-en-
ergy. massless spinor has the same Žopposite. sign as its chirality, the helicities assigned to the partons depend
on whether they are incoming or outgoing. Our convention is to label outgoing Žpositive-energy. particles with
their helicity; so if they are incoming the actual helicity and charge is reversed.
the Mandelstam invariants may be written as

si j s 2 pi P pj s pq y y q ) )
i p j q pi p j y pi H p j H y pi H p j H
so that
n
s s 2 pa P p b s Ý pq y
i pj ,
i , js1
n
s ai s 2 pa P pi s y Ý pyi pqj ,
js1
n
s b i s 2 p b P pi s y Ý pqi pyj . Ž A.3 .
js1
Massless Dirac spinors c " Ž p . of fixed helicity are defined by the projection
1 " g5
c "Ž p. s c Ž p. , Ž A.4 .
2
with the shorthand notation
c " Ž p . s < p " :, c "Ž p. s² p "< ,
² pk : s ² p y < k q : s cy Ž p . cq Ž k . ,
w pk x s ² p q < k y : s cq Ž p . cy Ž k . . Ž A.5 .
Using the chiral representation of the g-matrices,
0 I 0 ys i
g 0s ž I 0/, g is
ž si 0
,
/ Ž A.6 .
and the normalization condition
² p " < gm < p " : s 2 pm , Ž A.7 .
if 6
and the complex notation p H s < p H < e , the spinors for the momenta ŽA.1. are
(p q
i
0
0
cq Ž pi . s
0
(p
(y p
y
i
0
0
e if i ,
q
a
cy Ž pi . s
0
( y yi f i ,
pi e
y pq
0
i(
0
cq Ž pa . s i
0 0
0
0
0
, cy Ž pa . s i
0
0
y y pq
(
a
0
,
0
cq Ž p b . s yi
0 ( y py
b
0
0
,
0
cy Ž p b . s yi
y py
( b
0
. Ž A.8 .
6
The spinors of the incoming partons must be continued to negative energy after the complex conjugation.
'
For instance, cq Ž pa . s iŽ y pq
a ,0,0,0 .
.
Using the above spinor representation, the spinor products for the momenta ŽA.1. are
pq pq
² pi p j : s pi H ( j
pq
i
y pj H ) i
pq
j
,
ypq
² pa pi : s yi ( a
pq
i
pi H ,
² pi p b : s i y py
( q
b pi ,
² pa p b : s y'sˆ , Ž A.9 .
2
where we have used the mass-shell condition < pi H < s pq y
i pi .
Note that in the present
convention the spinors ŽA.8. and the spinor products ŽA.9. differ by phases with respect
to the same in Ref. w74x.
We consider also the spinor products ² pi q < g P p k < pj q :, which in the spinor
representation ŽA.8. take the form
² pi q < g P p k < p j q :
1
s Ž pqi pqj pyk y pqi pj H pk)H y pi)H pqj pk H qpi)H pj H pqk . , ;k ,
(pq q
i pj
ypq
² pi q < g P p j < p a q : s i ( a
pq
i
Ž pqi pyj y pi)H pj H . , ; j,
ypy
² pi q < g P pj < p b q : s yi ( b
pq
i
Ž ypqi pj)H q pi)H pqj . , ; j. Ž A.10 .
The spinor products fulfill the identities Ž i ' pi , j ' pj .

² ij : s y² ji : ,
w ij x s y w ji x ,
² ij :) s sign Ž pi0 pj0 . w ji x ,
)
Ž ² i q < g m < j q : . s sign Ž pi0 pj0 . ² j q < g m < i q : ,
² ij : w ji x s 2 pi P pj s ŝi j
² i q < ku < j q : s w ik x ² kj : ,
² i y < ku < j y : s ² ik : w kj x ,
² ij :² kl : s ² ik :² jl : q ² il :² kj : ,
w ij xw kl x s w ik xw jl x q w il xw kj x Ž A.11 .
and if Ý nis1 pi s 0 then
n
Ý w ji x ² ik : s 0 . Ž A.12 .
is1
Throughout the paper the following representation for the gluon polarization is used:
² p " < gm < k " :
em" Ž p,k . s "
'2 ² k . < p " : , Ž A.13 .
which enjoys the properties

em" ) Ž p,k . s em. Ž p,k . ,
em" Ž p,k . P p s em" Ž p,k . P k s 0 ,
pm kr q pr km
Ý emn Ž p,k . ern ) Ž p,k . s ygm r q , Ž A.14 .
ns" pPk
where k is an arbitrary light-like momentum. The sum in Eq. ŽA.14. is equivalent to use
an axial, or physical, gauge.
Appendix B. Multi-Regge kinematics
In the multi-Regge kinematics, we require that the gluons are strongly ordered in
rapidity and have comparable transverse momentum,
y 1 4 . . . 4 yn ; < p 1 H < , . . . , < pn H < . Ž B.1 .
Momentum conservation ŽA.2. then becomes
n
0s Ý pi H ,
is1
pq q
a , yp 1 ,
py y
b , ypn . Ž B.2 .
The Mandelstam invariants ŽA.3. are reduced to,
s s 2 pa P p b , pq y
1 pn ,
s ai s 2 pa P pi , ypq y
1 pi ,
sb i s 2 p b P pi , ypq y
i pn ,
si j s 2 pi P pj , < pi H < < pj H < e < y iyy j < Ž B.3 .

to leading accuracy. The spinor products ŽA.9. become
pq
² pi p j : , y ) i
pq
j
pj H for yi ) yj ,
pq
² pa pi : , yi ( a
pq
i
pi H ,
² pi p b : , i pq( y
i pn ,
² pa p b : , y pq y
(
1 pn . Ž B.4 .
Appendix C. NLO multi-Regge kinematics
We consider the production of n partons of momenta p 1 , . . . , pn , with partons 1 and 2

in the forward-rapidity region of parton pa ,
y 1 , y 2 4 y 3 4 . . . 4 yn ; < p 1 H < , < p 2 H < , . . . , < pn H < . Ž C.1 .
Momentum conservation ŽA.2. becomes
n
0s Ý pi H ,
is1
pq q q
a , y p1 q p 2
Ž .,
py y
b , ypn . Ž C.2 .
The spinor products ŽA.9. become
² pa p b : s y's , y (Ž p q q
1 q p2 . py
n ,
ypq pn H
² pa pn : s yi ( a
pq
n
pnH , i
< pn H <
² pa p b : ,
ypq pq q
1 q p2
² pa p k : s yi ( a
pq
k
p k H , yi ( pq
k
pk H , k s 1, . . . ,n y 1,
² p k p b : s i y py
( q q y
(
b p k , i p k pn , k s 1, . . . ,n y 1,
² pn p : s i(y p
b
y q
b pn , i < pn H < ,
pq pq pq
² p k pn : s p k H ( n
pq
k
y pn H ( k
pq
n
, ypn H ( k
pq
n
, k s 1, . . . ,n y 1,
pq pq
² p1 p 2 : s p1 H ( 2
pq
1
y p2 H ( 1
pq
2
,
pq pq pq
² p k pi : s p k H ( i
pq
k
y pi H ( k
pq
i
, ypi H ( k
pq
i
, k s 1,2;i s 3, . . . ,n y 1 .
Ž C.3 .
which differ by phases with respect to the same spinor products in Ref. w39x because of
the convention for the spinor representation we use in Section A.
Appendix D. NNLO multi-Regge kinematics
The extension to the production of n partons of momenta p 1 , . . . , pn , with partons 1,

2 and 3 in the forward-rapidity region of parton pa ,
y 1 , y 2 , y 3 4 y4 4 . . . 4 yn ; < p 1 H < , < p 2 H < , . . . , < pn H < , Ž D.1 .
is straightforward. We mention it here because by taking the further limit y 1 4 y 2 , y 3 ,
one obtains the kinematics of the NLO Lipatov vertex ŽSection 5.5..
With Eq. ŽD.1., momentum conservation ŽA.2. becomes

n
0s Ý pi H ,
is1
pq q q
Ž q
a , y p1 q p 2 q p 3 .,
py y
b , ypn . Ž D.2 .
The spinor products ŽA.9. become
² pa p b : s y's , y (Ž p q q q
1 q p2 q p3 . py
n ,
ypq pn H
² pa pn : s yi ( a
pq
n
pnH , i
< pn H <
² pa p b : ,
ypq pq q q
1 q p2 q p3
² pa p k : s yi ( a
pq
k
p k H , yi ( pq
k
pk H , k s 1, . . . ,n y 1,
² p k p b : s i y py
( q q y
b p k , i p k pn , ( k s 1, . . . ,n y 1,
² pn p : s i(y p
b
y q
b pn , i < pn H < ,
pq pq pq
² p k pn : s p k H ( n
pq
k
y pn H ( k
pq
n
, ypn H ( k
pq
n
, k s 1, . . . ,n y 1,
pq pq pq
² p k pi : s p k H ( i
pq
k
y pi H ( k
pq
i
, ypi H ( k
pq
i
,
k s 1,2,3; i s 4, . . . ,n y 1 , Ž D.3 .
while the others spinor products remain unchanged. The spinor products ŽD.3. general-
ize straightforwardly to the kinematics Ž6.1..
Appendix E. The Sudakov parametrization
We want to elucidate the relationship between our parametrization of the momenta

and the one of Ref. w21x. Recalling the last of Eqs. ŽA.1., we can write,
x i Pq < pi H < 2
pi s Ž 1,0,0,1 . q Ž 0,Re w pi H x ,Im w pi H x ,0 . q Ž 1,0,0,y 1 . , Ž E.1 .
2 2 x i Pq
where P m is the sum of the three momenta, the x i are the momentum fractions and we
used the mass-shell condition pq y < <2
i pi s pi H . This is exactly what is obtained from the
general Sudakov parametrization of Ref. w21x,
2
kH i nm
pim s x i p m m
qkH iy Ž E.2 .
xi 2 pPn
through the following choices for the lightlike vectors:
Pq
pms Ž 1,0,0,1 . and n m s Ž 1,0,0,y 1 . , Ž E.3 .
2
and the identification

m
kH i s Ž 0,Re w pi H x ,Im w pi H x ,0 . . Ž E.4 .
The spin-correlated splitting functions of Ref. w21x are expressed in terms of the vectors
k˜ i defined as k˜ im s k H
m m
i y z i P H , where, as in our case, the z i variables represent the
momentum fractions in the collinear limit. In order to compare Eq. Ž5.59. with the
spin-correlated splitting functions of Ref. w21x, we must project the latter onto the
helicity basis, namely to contract them with the polarization vector, Eq. ŽA.13.,
1
em" Ž P , n . s
'2 Ž 0,1,. i ,0 . . Ž E.5 .
The contraction of the k˜ im vectors with eq is
zi
k˜ i P eqs
2
(w i j x z j q w i l x zl ,
ž ( ( / Ž E.6 .
with i, j,l s 1,2,3 and j,l / i, with the analogous expressions for ey obtained by
For the off-diagonal terms, P g ™ g f 2 f 3 , we find a relative minus sign between the
complex conjugation.
results of Ref. w21x and ours, which, however, has no physical relevance.
References
w1x W.T. Giele, E.W.N. Glover, Phys. Rev. D 46 Ž1992. 1980.
w2x W.T. Giele, E.W.N. Glover, D.A. Kosower, Nucl. Phys. B 403 Ž1993. 633. hep-phr9302225.
w3x S. Keller, E. Laenen, Phys. Rev. D 59 Ž1999. 114004. hep-phr9812415.
w4x Z. Kunszt, D.E. Soper, Phys. Rev. D 46 Ž1992. 192.
w5x S. Frixione, Z. Kunszt, A. Signer, Nucl. Phys. B 467 Ž1996. 399. hep-phr9512328.
w6x Z. Nagi, Z. Trocsanyi, Nucl. Phys. B 486 Ž1997. 189. hep-phr9610498.
w7x S. Frixione, Nucl. Phys. B 507 Ž1997. 295. hep-phr9706545.
w8x S. Catani, M.H. Seymour, Phys. Lett. B 378 Ž1996. 287. hep-phr9602277.
w9x S. Catani, M.H. Seymour, Nucl. Phys. B 485 Ž1997. 291. hep-phr9605323.
w10x S. Catani, M.H. Seymour, Nucl. Phys. B 510 Ž1998. 503.
w11x D.A. Kosower, Phys. Rev. D 57 Ž1998. 5410. hep-phr9710213.
w12x V.N. Gribov, L.N. Lipatov, Yad. Fiz. 15 Ž1972. 781, Sov. J. Nucl. Phys. 46 Ž1972. 438.
w13x L.N. Lipatov, Yad. Fiz. 20 Ž1974. 181, Sov. J. Nucl. Phys. 20 Ž1975. 95.
w14x G. Altarelli, G. Parisi, Nucl. Phys. 126 Ž1977. 298.
w15x Yu.L. Dokshitzer, Zh. Eksp. Teor. Fiz. 73 Ž1977. 1216, Sov. Phys. JETP 46 Ž1977. 641.
w16x D.R. Yennie, S.C. Frautschi, H. Suura, Ann. Phys. 13 Ž1961. 379.
w17x A. Bassetto, M. Ciafaloni, G. Marchesini, Phys. Rep. 100 Ž1983. 201.
w18x F.A. Berends, W.A. Giele, Nucl. Phys. B 313 Ž1989. 595.
w19x Z. Kunszt, A. Signer, Z. Trocsanyi, Nucl. Phys. B 420 Ž1994. 550. hep-phr9401294.
w20x J.M. Campbell, E.W.N. Glover, Nucl. Phys. B 527 Ž1998. 264. hep-phr9710255.
w21x S. Catani, M. Grazzini, Phys. Lett. B 446 Ž1999. 143. hep-phr9810389.
w22x S. Catani, M. Grazzini, preprint, hep-phr9908523.
w23x Z. Bern, G. Chalmers, Nucl. Phys. B 447 Ž1995. 465. hep-phr9503236.
w24x Z. Bern, V. Del Duca, C.R. Schmidt, Phys. Lett. B 445 Ž1998. 168. hep-phr9810409.
w25x Z. Bern, V. Del Duca, W.B. Kilgore, C.R. Schmidt, Phys. Rev. D 60 Ž1999. 116001. hep-phr9903516.
w26x D.A. Kosower, Nucl. Phys. B 552 Ž1999. 319. hep-phr9901201.
w27x D.A. Kosower, P. Uwer, preprint, hep-phr9903515.
w28x S. Catani, Phys. Lett. B 427 Ž1998. 161. hep-phr9802439.
w29x V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Zh. Eksp. Teor. Fiz. 71 Ž1976. 840, Sov. Phys. JETP 44 Ž1976.
443.
w30x V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Zh. Eksp. Teor. Fiz. 72 Ž1977. 377, Sov. Phys. JETP 45 Ž1977.
199.
w31x Ya.Ya. Balitsky, L.N. Lipatov, Yad. Fiz. 28 Ž1978. 1597, Sov. J. Nucl. Phys. 28 Ž1978. 822.
w32x L.N. Lipatov, Yad. Fiz. 23 Ž1976. 642, Sov. J. Nucl. Phys. 23 Ž1976. 338.
w33x L.N. Lipatov, Nucl. Phys. B 365 Ž1991. 614.
w34x V.S. Fadin, L.N. Lipatov, Phys. Lett. B 429 Ž1998. 127. hep-phr9802290.
w35x G. Camici, M. Ciafaloni, Phys. Lett. B 412 Ž1997. 396. hep-phr9707390.
w36x G. Camici, M. Ciafaloni, Phys. Lett. B 417 Ž1998. 390, Erratum.
w37x G. Camici, M. Ciafaloni, Phys. Lett. B 430 Ž1998. 349. hep-phr9803389.
w38x V.S. Fadin, L.N. Lipatov, Yad. Fiz. 50 Ž1989. 1141, Sov. J. Nucl. Phys. 50 Ž1989. 712.
w39x V. Del Duca, Phys. Rev. D 54 Ž1996. 989. hep-phr9601211.
w40x V.S. Fadin, L.N. Lipatov, Nucl. Phys. B 477 Ž1996. 767. hep-phr9602287.
w42x V.S. Fadin, L.N. Lipatov, Nucl. Phys. B 406 Ž1993. 259.
w43x V.S. Fadin, R. Fiore, A. Quartarolo, Phys. Rev. D 50 Ž1994. 5893. hep-thr9405127.
w44x V.S. Fadin, R. Fiore, M.I. Kotsky, Phys. Lett. B 389 Ž1996. 737. hep-phr9608229.
w45x V.S. Fadin, Proc. LAFEX International School on High-Energy Physics ŽLISHEP 98., Rio de Janeiro,
Brazil, hep-phr9807528.
w46x V. Del Duca, C.R. Schmidt, Phys. Rev. D 59 Ž1999. 074004. hep-phr9810215.
w47x V.S. Fadin, R. Fiore, M.I. Kotsky, Phys. Lett. B 359 Ž1995. 181.
w48x V.S. Fadin, R. Fiore, M.I. Kotsky, Phys. Lett. B 387 Ž1996. 593. hep-phr9605357.
w49x V.S. Fadin, R. Fiore, A. Quartarolo, Phys. Rev. D 53 Ž1996. 2729. hep-phr9506432.
w50x J. Blumlein,
¨ V. Ravindran, W.L. van Neerven, Phys. Rev. D 58 Ž1998. 091502. hep-phr9806357.
w51x V. Del Duca, C.R. Schmidt, Phys. Rev. D 57 Ž1998. 4069. hep-phr9711309.
w52x M. Ciafaloni, Phys. Lett. B 429 Ž1998. 363. hep-phr9801322.
w53x M. Ciafaloni, D. Colferai, Nucl. Phys. B 538 Ž1999. 187. hep-phr9806350.
w54x V. Del Duca, Proc. of Les Rencontres de Physique de la Vallee d’Aoste, La Thuile, ed. M. Greco ŽINFN
Press, Italy, 1996. hep-phr9605404.
w55x V. Del Duca, Proc. LAFEX International School on High-Energy Physics ŽLISHEP98., Rio de Janeiro,
Brazil, hep-phr9902211.
w56x Z. Bern, D.A. Kosower, Nucl. Phys. B 379 Ž1992. 451.
w57x W. Siegel, Phys. Lett. B 84 Ž1979. 193.
w58x D.M. Capper, D.R.T. Jones, P. van Nieuwenhuizen, Nucl. Phys. B 167 Ž1980. 479.
w59x L.N. Lipatov, Nucl. Phys. B 452 Ž1995. 369.
w60x P. Uwer, talk at the Europhysics Conference on High Energy Physics ŽEPS-HEP99., Tampere, Finland.
w61x M. Mangano, S.J. Parke, Phys. Rep. 200 Ž1991. 301.
w62x J.E. Paton, H.-M. Chan, Nucl. Phys. B 10 Ž1969. 516.
w63x P. Cvitanovic, P.G. Lauwers, P.N. Scharbach, Nucl. Phys. B 186 Ž1981. 165.
w64x D.A. Kosower, B.-H. Lee, V.P. Nair, Phys. Lett. B 201 Ž1988. 85.
w65x M. Mangano, Nucl. Phys. B 309 Ž1988. 461.
w66x D. Zeppenfeld, Int. J. Mod. Phys. A 3 Ž1988. 2175.
w67x Z. Bern, D.A. Kosower, Nucl. Phys. B 362 Ž1991. 389.
w68x F.A. Berends, W.T. Giele, Nucl. Phys. B 294 Ž1987. 700.
w69x M. Mangano, S.J. Parke, Z. Xu, Nucl. Phys. B 298 Ž1988. 653.
w70x F.A. Berends, W.T. Giele, Nucl. Phys. B 306 Ž1988. 759.
w71x R. Kleiss, H. Kuijf, Nucl. Phys. B 312 Ž1989. 616.
w72x Z. Bern, L. Dixon, D.C. Dunbar, D.A. Kosower, Nucl. Phys. B 425 Ž1994. 217. hep-phr9403226.
w73x V. Del Duca, Phys. Rev. D 48 Ž1993. 5133.
w75x F.A. Berends, W.T. Giele, H. Kuijf, Nucl. Phys. B 333 Ž1990. 120.
w76x B.L. Combridge, C.J. Maxwell, Nucl. Phys. B 239 Ž1984. 429.
w77x S. Catani, M.H. Seymour, Z. Trocsanyi, Phys. Rev. D 55 Ž1997. 6819. hep-phr9610553.
w78x H. Kuijf, Ph.D. Thesis Žunpublished..
w79x Z. Kunszt, W.J. Stirling, Phys. Rev. D 37 Ž1988. 2439.
w80x C.J. Maxwell, Phys. Lett. B 192 Ž1987. 190.
NNLO evolution of deep-inelastic structure functions:

the non-singlet case
W.L. van Neerven, A. Vogt
Instituut-Lorentz, UniÕersity of Leiden, P.O. Box 9506, 2300 RA Leiden, The Netherlands
Received 27 July 1999; accepted 20 October 1999
Abstract
We study the next-to-next-to-leading order ŽNNLO. evolution of flavour non-singlet quark

densities and structure functions in massless perturbative QCD. Present information on the
corresponding three-loop splitting functions is used to derive parametrizations of these quantities,
including Bjorken-x dependent estimates of their residual uncertainties. Compact expressions are
also provided for the exactly known, but rather involved two-loop coefficient functions. The size
of the NNLO corrections and their effect on the stability under variations of the renormalization
scale are investigated. The residual uncertainty of the three-loop splitting functions does not lead
to appreciable effects for x ) 10y2 . Inclusion of the NNLO contributions reduces the main
theoretical uncertainty of a s determinations from non-singlet scaling violations by more than a
factor of two. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 12.38.Bx; 13.60.Hb

Keywords: Deep-inelastic lepton–hadron scattering; Structure functions; Parton densities; Higher-order QCD
corrections
1. Introduction
More than thirty years after the pioneering experiments at SLAC w1–3x, structure
functions in deep-inelastic lepton–hadron scattering ŽDIS. remain among the most
important probes of perturbative QCD and of the partonic structure of hadrons. Indeed,
experiments have proceeded towards very high accuracy and a greatly extended kine-
matic coverage during the past two decades w4x. Moreover, the forthcoming luminosity
upgrade of the electron–proton collider HERA at DESY will allow for accurate
measurements up to very high resolution scales Q 2 , 10 4 GeV 2 , thus considerably
increasing the lever arm for precise determination of the scaling violations, i.e. the
Q 2-dependence, of the structure functions. An accurate knowledge of the parton
densities will also be indispensable for interpreting many results at the future Large
Hadron Collider at CERN.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 6 8 - 9
264 W.L. Õan NeerÕen, A. Vogt r Nuclear Physics B 568 (2000) 263–286
Given the non-perturbative Bjorken-x dependence of the structure functions at one

scale, the scaling violations can be calculated in the QCD-improved parton model in
terms of a power expansion in the strong coupling constant a s . The next-to-leading
order ŽNLO. ingredients for such analyses are available since 1980 for unpolarized
structure functions in massless perturbative QCD w5x. Yet the corresponding results for
the next-to-next-to-leading order ŽNNLO. are not complete at present, due to the
enormous complexity of the required loop calculations. Of the components entering the
NNLO description, the three-loop b-function Žgoverning the scale dependence of the
strong coupling constant. w6,7x and the two-loop contributions to the coefficient func-
tions Žconnecting the structure functions to the parton densities. w8–13x have been
derived. However, only partial results have been obtained so far for the three-loop terms
of the splitting functions Žgoverning the scale-dependence of the quark and gluon
densities., most notably the lowest even-integer Mellin moments of those combinations
relevant to unpolarized electromagnetic deep-inelastic scattering w14–16x.
Standard global analyses of deep-inelastic scattering and related processes, like the
Drell–Yan process for which two-loop coefficient functions have also been calculated
w17–19x, have thus been restricted to NLO up to now w20–25x. This level of accuracy is
however not sufficient to make full use of present and forthcoming data, as the
theoretical uncertainties of the NLO results, for instance on the strong coupling constant,
already now tend to exceed the corresponding experimental errors. Therefore first
approximate NNLO analyses have been performed recently of data on neutrino-nucleon
w26–28x and electron Žmuon.-proton w29x DIS structure functions, directly using the
results of Refs. w14–16x via integer Mellin-N techniques. However, these techniques lack
some flexibility, e.g., they cannot incorporate additional information on the x-depen-
dence of the two-loop coefficient functions w9–13x and of the three-loop splitting
functions w30–35x. Hence we pursue an alternative approach which allows for incorpo-
rating the NNLO corrections into programs using standard x-space w20–23x or equiva-
lent complex-N techniques w24,25,36–38x. Its most important ingredients are compact
approximate x-space expressions for the three-loop splitting functions including quanti-
tative estimates of their present uncertainty. In the present article, we deal with the
important flavour non-singlet case. The flavour-singlet quantities will be discussed in a
subsequent publication.
This paper is organized as follows. In Section 2 we recall the general formalism for
the scale dependence Ž‘evolution’. of non-singlet quark densities and structure functions
in massless perturbative QCD. The a s-expansions are explicitly given up to NNLO for
arbitrary choices of the renormalization and mass-factorization scales. In Section 3 we
present accurate, compact parametrizations of the exactly known w8–13x, but rather
involved x-dependence of the two-loop coefficient functions. In Section 4 we employ
the present constraints w14–16,30,33x on the three-loop non-singlet splitting functions for
deriving approximate expressions for their x-dependence. The remaining uncertainties
are quantified. All these results are put together in Section 5 to study the impact of the
NNLO contributions on the evolution of the various non-singlet parton densities and
structure functions. Here we also discuss the implications on determinations of a s from
DIS structure functions. Finally our findings are summarized in Section 6. Mellin-N
space expressions for our parametrizations of the two-loop coefficient functions of
Section 3 can be found in Appendix A.
W.L. Õan NeerÕen, A. Vogt r Nuclear Physics B 568 (2000) 263–286 265
2. The general formalism
We set up our notations by recalling the NNLO evolution equations for non-singlet
parton densities and structure functions. The number distributions of quarks and
antiquarks in a hadron are denoted by qi Ž x, m2f , m2r . and qi Ž x, m2f , m 2r ., respectively,
where x represents the fraction of the hadron’s momentum carried by the parton. m r and
m f stand for the renormalization and mass-factorization scales, and the subscript i
indicates the flavour of the Žanti-.quark, with i s 1, . . . , Nf for Nf flavours of effectively
massless quarks.
The scale dependence of non-singlet combinations of these quark densities is
governed by the Žanti-.quark Žanti-.quark splitting functions. Suppressing the depen-
dence on x, m r and m f for the moment, the general structure of these functions,
constrained by charge conjugation invariance and flavour symmetry, is given by
Pq i q k s Pq i q k s d i k PqVq q PqSq ,
Pq i q k s Pq i q k s d i k PqVq q PqSq . Ž 2.1 .

In an expansion in powers of the strong coupling constant a s the flavour-diagonal
Ž‘ valence’. quantity PqVq starts at first order, while PqVq and the flavour-independent
Ž‘sea’. contributions PqSq and PqSq are of order a s2 . A non-vanishing difference
PqSq y PqSq occurs for the first time at third order. This general structure leads to three
independently evolving types of non-singlet distributions: The evolution of the flavour
asymmetries
"
q NS ,i ks q i " q i y Ž q k " q k . Ž 2.2 .
"
and of linear combinations thereof, hereafter generically denoted by q NS, is governed by
"
P NS s PqVq " PqVq . Ž 2.3 .
The sum of the valence distributions of all flavours,
Nf
V
q NS s Ý Ž qr y qr . , Ž 2.4 .
rs1
evolves with
V
P NS s PqVq y PqVq q Nf PqSq y PqSq .
ž / Ž 2.5 .
V
The first moments of Py
NS and P NS vanish,
1 Ny 1
1 Ny1 V
H0 dx x Py
NS s H0 dx x P NS s 0 for N s 1, Ž 2.6 .
V
since the first moments of qy NS and q NS reflect conserved additive quantum numbers.
The difference PqSq y PqSq is unknown except for the first moment, which vanishes by
virtue of Eqs. Ž2.3., Ž2.5. and Ž2.6.. However, the size of the two-loop contributions to
PqVq and PqSq relative to the corresponding term of PqVq suggests that this difference is
negligibly small at moderate and large x. Hence we shall use the approximation
V
P NS s Py
NS Ž 2.7 .
V
for the rest of this article, i.e. we henceforth treat q NS as a ‘–’-quantity.
Restoring the dependence on the fractional momentum x and the renormalization and
mass-factorization scales m r and m f , our evolution equations thus read
d m2f
dln m2f
"
q NS "
Ž x , m2f , m2r . s P NS ž
a s Ž m2r . ,
m2r / "
m q NS Ž m2f , m2r . Ž x . . Ž 2.8 .
Here m stands for the Mellin convolution in the momentum variable,

1 dy x
w a m b x Ž x . sH
x y
aŽ y . b ž /
y
. Ž 2.9 .
"
The expansion of P NS up to the third order ŽNNLO. in a s ' a srŽ4p . takes the form
m2f
"
P NS
ž
x , a s Ž m2r . ,
m2r /
m2f
s a s Ž m2r . P NS
Ž0. Ž1."
Ž x . q a2s Ž m2r . PNS
ž Ž 0.
Ž x . y b 0 PNS Ž x . ln
m2r /
m2f
ž
Ž 2."
qa3s Ž m2r . PNS Ž 0.
Ž x . y b 1 PNS Ž1."
Ž x . q 2 b 0 PNS Ž x . 4 ln
m2r
m2f
q b 02 PNS
Ž 0.
Ž x . ln2
m2r / q... Ž 2.10 .
The one- and two-loop functions P NS Ž0. Ž .

x are known for a long time w5x;
Ž1." Ž .
x and P NS
Ž2." Ž .
the three-loop quantities P NS x are the subject of Section 4. The relevant coefficients
of the QCD b-function,
da s
s b Ž a s . s y Ý a lq2
s bl , Ž 2.11 .
dln m2r ls0
are given by w5–7x

b 0 s 11 y 23 Nf ,
b 1 s 102 y 383 Nf ,
b 2 s 2857 5033 325 2

2 y 18 Nf q 54 Nf . Ž 2.12 .
The first two coefficients b 0 and b 1 are scheme independent in massless QCD; the
result given for b 2 refers to the MS renormalization scheme employed throughout this
paper.
"
The non-singlet structure functions Fa,NS, a s 1,2,3, are in Bjorken-x space obtained
by convoluting the solution of Eq. Ž2.8. with the corresponding coefficient functions:
Q 2 m2f
ha Fa",NS Ž x ,Q 2 . s Ca",NS a s Ž m2r . ,
ž ,
m2f m2r
"
m q NS
/Ž m2f , m2r . Ž x . Ž 2.13 .
with h1 s 2, h 2 s 1rx, h 3 s 1, and

Q 2 m2f
ž
Ca",NS x , a s Ž m 2r . , ,
m2f m2r /
Q2
s d Ž 1 y x . q a s Ž m2r . c aŽ,NS
1.
ž Ž 0.
Ž x . q PNS Ž x . ln
m2f / q a 2s Ž m 2r .
žŽ 2."
c aŽ,NS Ž x.
Ž1." Ž 0. 1.
Q2 Q2
q P NS Ž x. q P NS m c aŽ,NS Ž x . 4 ln y b 0 c Ž1.
a ,NS Ž x . ln
m2f m2r
Q2
q 12 PNS
Ž 0. Ž 0.
m PNS Ž 0.
Ž x . y b 0 PNS Ž x . 4 ln2
m2f
Q2 m2f
yb 0 PNS
Ž0.
Ž x . ln
mf 2
ln
m2r / q...
"
Ž 2.14 .
Here an overall electroweak charge factor has been absorbed into q NS . The first-order
coefficients c a,NS x can be found in Ref. w5x; the two-loop quantities c Ž2."
Ž1. Ž .
a,NS x
Ž .
computed in Refs. w8–11x are discussed in Section 3.
It is often convenient, especially in the non-singlet sector considered here, to express
the scaling violations of the structure functions in terms of these structure functions
"
themselves. The expansion coefficients of the corresponding kernels Ka,NS in
d Q2
dlnQ 2
ha Fa",NS Ž x ,Q . s Ka",NS a s Ž m2r . ,
2
ž m2r / m ha Fa",NS Ž Q 2 . Ž x . Ž 2.15 .
are built up of factorization-scheme invariant combinations of the splitting functions

Ž l ." Ž . l ." Ž .
P NS x and the coefficient functions c Ža,NS x . Up to third order this expansion reads
Q2
"
KNS
ž
x , a s Ž m 2r . ,
m2r /
Q2
s a s Ž m2r . P NS
Ž0. Ž1."
Ž x . q a2s Ž m2r . PNS
ž 1.
Ž x . y b 0 c aŽ,NS Ž 0.
Ž x . y b 0 PNS Ž x . ln
m2r /
qa3s Ž m2r . PNS
ž
Ž 2." 2."
Ž x . y 2 b 0 c aŽ,NS Ž x.
1. 1. 2 Ž0. 2
Q2
qb 0 c aŽ,NS m c aŽ,NS Ž x . y b 1 c Ž1.
a ,NS Ž x . q b 0 P NS Ž x . ln
m2r
Q2
y Ž1."
2 b 0 P NS Ž x. 1.
y b 0 c aŽ,NS Ž x.4 Ž 0.
q b 1 PNS Ž x . 4 ln
m2r / q... Ž 2.16 .
This approach removes the dependence of the finite-order predictions on the factoriza-
tion scheme and the scale m f , thus allowing for an easier control of the theoretical
uncertainties.
3. The two-loop non-singlet coefficient functions

The O Ž a s2 . contributions CaŽ2. to the coefficient functions for the structure functions
F2 , FL s F2 y 2 xF1 and xF3 were calculated some time ago in Refs. w8–13x. The
resulting expressions are rather lengthy and involve higher transcendental functions.
Hence it is convenient to employ more compact, if approximate, parametrizations of
these quantities. This holds in particular if one uses the moment-space technique
w36–38x, which requires the analytic continuation of all ingredients to complex Mellin-N.
The reader is referred to Refs. w39,40x for a more rigorous approach to the moment-space
expressions for CaŽ2.. Those parts of the coefficient functions arising from m r / Q and
m f / m r in Eq. Ž2.14. are simple convolutions of the well-known lower-order anomalous
dimensions and Wilson coefficients. The same applies to the terms induced by usual
scheme transformations, e.g., that from the MS to the DIS factorization scheme. For
explicit expressions see Refs. w12,13x. Hence the parametrizations can be restricted to the
MS scheme, and to lnŽ m 2r rQ 2 . s lnŽ m2r rm2f . s 0.
Our procedure for deriving compact approximate expressions for c Ž2. a,NS x is as
Ž .
follows: We keep the q-distribution parts, defined by
1 1
H0 dx aŽ x . qb Ž x . s H dx a Ž x . b Ž x . y b Ž 1 . 4 ,
0
Ž 3.1 .
exactly Žup to a truncation of the numerical coefficients.. The integrable x - 1 terms are
fitted to the exact results for 10y6 ( x ( 1–10y6 . Finally the coefficients of d Ž1 y x .
are slightly adjusted from their exact values using the lowest integer moments. The
resulting parametrizations deviate from the exact results by no more than a few permille.
a,NS x - 1 themselves as well as for the convolutions with typical
This holds for the c Ž2. Ž .
hadronic x-shapes. The adjustment of the d Ž1 y x . pieces is important for the latter
agreement.
The non-singlet coefficient function entering the electromagnetic F2 can be written as
1
Ž2.q
c 2,NS Ž x. s ž 14.2222 L31 y 61.3333 L21 y 31.105 L1 q 188.64 /
1yx q
y 17.19 L31 q 71.08 L21 y 660.7 L1 q L1 L0 Ž y174.8 L1 q 95.09 L 0 .
y 2.835 L30 y 17.08 L20 q 5.986 L0 y 1008 x y 69.59

1
y 338.046 d Ž 1 y x . q Nf
1yx ½ž
1.77778 L21 y 8.5926 L1
q6.3489 / y 1.707 L21 q 22.95 L1 q L1 L 0 Ž 3.036 L0 q 17.97 .

q
q2.244 L20 q 5.770 L0 y 37.91 x y 5.691 q46.8405 d Ž 1 y x . 5

Ž 3.2 .
with
L1 ' ln Ž1yx . , L0 ' ln x.
Ž2.y Ž .
For c 2, NS x ,
relevant for the charged-current case, the second and third line of this
expression have to be replaced by
y17.19 L31 q 71.08 L21 y 663.0 L1 q L1 L0 Ž y192.4 L1 q 80.41 L0 .
y3.748 L30 y 19.56 L20 y 1.235 L0 y 1010 x y 84.18 y 337.994 d Ž 1 y x . . Ž 3.3 .
The corresponding parametrizations for FL read
2 2
c LŽ2.q
,NS Ž x . s 13.62 L1 y 55.79 L1 y 150.5 L1 L 0 q Ž 26.56 x y 0.031 . L 0
y 14.85 L0 q 97.48 x y 40.41 y 0.164 d Ž 1 y x .

q 16
27 Nf 6 xL1 y 12 xL 0 y 25 x q 6 4 Ž 3.4 .
and
2 2
c LŽ2.y
,NS Ž x . s 13.30 L1 y 59.12 L1 y 141.7 L1 L 0 q Ž 23.29 x y 0.043 . L 0
y 22.21 L0 q 100.8 x y 52.27 y 0.150 d Ž 1 y x .

q 16
27 Nf 6 xL1 y 12 xL 0 y 25 x q 6 4 . Ž 3.5 .
For the Nf parts, which are identical in Eqs. Ž3.4. and Ž3.5., we have taken the exact
expression from Ref. w8x.
Also for the charged-current non-singlet structure function F3 there are two combina-
tions which differ at O Ž a s2 .. The first one, entering F3n N q F3n N , can be written as
1
Ž2.y
c 3,NS Ž x. s ž 14.2222 L31 y 61.3333 L21 y 31.105 L1 q 188.64 /
1yx q
y 15.20 L31 q 94.61 L21 y 409.6 L1 y 147.9 L21 L 0 y 3.922 L30
y 33.31 L20 y 67.60 L 0 y 576.8 x y 206.1 y 338.635 d Ž 1 y x .

1
q Nf ½ž
1yx
1.77778 L21 y 8.5926 L1 q 6.3489
q
/
q0.042 L31 y 0.808 L21 q 25.00 L1 q 9.684 L1 L 0
q2.207 L20 q 8.683 L0 y 14.97 x y 6.337 q46.857 d Ž 1 y x . . 5

Ž 3.6 .
For the other combination c 3,Ž2.q
NS x ,
corresponding to
Ž . F3n N y F3n N , the second and third
lines of this result have to be replaced by
y15.20 L31 q 94.61 L21 y 396.1 L1 y 92.43 L21 L 0
y3.049 L30 y 30.14 L20 y 79.14 L0 y 467.2 x y 242.9 y 338.683 d Ž 1 y x . .
Ž 3.7 .
The complex Mellin moments of these results, c Ž2."
a,NS N ,
can be readily obtained.
Ž .
They do not involve special functions beyond the logarithmic derivatives of the
G-function. The explicit expressions can be found in Appendix A.
4. The three-loop non-singlet splitting functions
Only partial results are presently available for the O Ž a s3 . terms P Ž2. Ž x . of the
splitting functions. In the non-singlet sector, the current information comprises
Ø the lowest five even-integer moments of P NS Ž2.q

calculated in Refs. w14–16x, while for
Ž2.y
PNS only the first moment Žs 0 in MS. is known;
Ø the complete O Ž Nf2 . piece Židentical for the ‘q’ and ‘y’ combinations. determined
via an all-order leading-Nf approach in Ref. w30x;
Ø the most singular small-x terms Ž; ln4 x . of P NS Ž2.q
and PNSŽ2.y
inferred in Ref. w33x
from the leading small-x resummation of the non-singlet evolution kernels w41x.
Ž1." Ž . w
The two-loop results P NS x 42x and c Ž2." Ž . w9–11x furthermore indicate that the
a,NS x
Ž2.q Ž . Ž2.y Ž .
difference PNS x y PNS x is negligibly small at large x. Finally present knowl-
edge complies with the conjecture w43x that the splitting functions do not receive
contributions of the form wln l Ž1 y x .rŽ1 y x .xq with l ) 0 in the MS factorization
scheme, unlike the coefficient functions discussed above.
In what follows we employ this information for approximate reconstructions of
Ž2."
P NS Ž x . s P0Ž2." Ž x . q Nf P1Ž2." Ž x . q Nf2 P2Ž2. Ž x . . Ž 4.1 .
Our approach is to fix the coefficients of suitably chosen basis functions by the above
constraints. The spread of the result due to ‘reasonable’ variations in the choice of those
functions then provides a measure of the residual uncertainty. Specifically we employ
the ansatz
A i ,1
Pi" Ž x . s q A i ,2 d Ž 1 y x . q A"
i ,3 f 1 Ž x .
Ž1yx .q
q A" " "
i ,4 f m Ž x . q A i ,5 f 0 Ž x . q fas Ž x . Ž 4.2 .
™ ™
1
for the Nf-independent Ž i s 0. and NF Ž i s 1. terms in Eq. Ž4.1.. Here f 1 and f 0
represent contributions which, while being integrable, peak at x 1 and x 0,
respectively. f m stands for a part with a rather flat x-dependence. As for the illustrations
in Refs. w14–16x, these contributions are build up of powers of lnŽ1 y x ., x, and ln x.
Finally fas allows to account for known leading small-x terms. Equating the second to
tenth even moments of Eq. Ž4.2. to the results of Refs. w14–16x yields five linear
equations which can be solved for the coefficients Aq Ž2.y
i, j . The case of P NS is treated
afterwards by taking over A i,1 and A i,2 from the ‘q’-combinations, as already indicated
in Eq. Ž4.2., and adjusting the remaining coefficients as discussed below.
Ž2." Ž .
Before addressing PNS x we demonstrate our procedure by applying it to a known
Ž1.q Ž . w
result, the Nf s 0 part P0Ž1.q of the NLO splitting function P NS x 42x. In this case
2
the leading small-x contributions are ln x and ln x, while the integrable terms most
peaked at large-x read x 2 and lnŽ1 y x .. Disregarding small-x constraints in this
example, we thus choose
f 1Ž x . s x 2 or ln Ž 1 y x . ,
fm Ž x . s 1 or x ,
f 0 Ž x . s ln x or ln2 x ,
fas Ž x . ' 0. Ž 4.3 .
The resulting eight approximations are compared to the exact result in Fig. 1 for x - 1.
The latter curve runs inside the uncertainty band over the full x-range. The moments
tightly constrain P0Ž1.q Ž x . for x R 0.15, the total spread in our approach being about 5%
at x , 0.3. The coefficients of the common leading large-x terms are
AŽ1.
0,1 s 65.13 . . . 68.74 Ž exact 66.47 .
AŽ1.
0,2 s 61.85 . . . 79.64 Ž exact 69.00 . , Ž 4.4 .
and the first moments read
P0Ž1.q Ž N s 1 . s y2.404 . . . 0.400 Ž exact y1.127 . . Ž 4.5 .
‘Unreasonable’ combinations in the sense of Eq. Ž4.2., like lnŽ1 y x ., x 2 , and 1 Ži.e. no
f 0 . or 1, ln x, and ln2 x Ži.e. f 1 missing., can lead to considerably worse approximations.
Now we turn to P0Ž2.q Ž x .. The additional loop or emission may, besides adding two
powers of ln x, lead to two additional large-x logarithms with respect to P0Ž1.q Ž x . Žthe
transition from one-loop to two-loop yields however only a term ln1 Ž1 y x . .. Hence we
put
f 1 Ž x . s x 2 or ln Ž 1 y x . or ln2 Ž 1 y x . or ln3 Ž 1 y x . ,
f m Ž x . s 1 or x ,
f 0 Ž x . s ln x or ln2 x ,
fas Ž x . s 23 CF3 Ž ln4 x q lln3 x . . Ž 4.6 .
Besides l s 0 we also include l s y4 and l s 8 for f 0 s ln x. Subleading small-x
Ž0.q Ž1.q
terms of this order of magnitude are suggested by the expansion of P NS and P NS in
moment-space around N s 0 w44x. Thus we consider 32 combinations, 8 of which are
rejected as they fail to fulfill the further ad hoc, but mild constraint
100 P0Ž1.q Ž N s 1 . ( P0Ž2.q Ž N s 1 . ( y40 P0Ž1.q Ž N s 1 . Ž 4.7 .
on the perturbative expansion of the first moment. The x - 1 behaviour of the remaining
24 function is displayed in Fig. 2; their 1rŽ1 y x .q coefficients span the range
1138 ( AŽ2. 0,1 ( 1625 Ž 1347 . . Ž 4.8 .
The bracketed number applies if combinations with f 1Ž x . s ln3 Ž1 y x . are disregarded.
Ž1.q
Fig. 1. Approximations for the N f -independent part of PNS , derived from the lowest even-integer moments
by means of Eqs. 4.2 and 4.3 , compared to the exact result.
Ž . Ž .
Ž2.q
Fig. 2. Approximations for the N f -independent part of P NS , denoted by P0Ž2.q in Eq. Ž4.1., as derived from
the five lowest even-integer moments by means of Eqs. Ž4.2., Ž4.6. and Ž4.7.. The full lines represent those
functions selected for further consideration.
Due to the larger function pool of Eq. Ž4.6., the large-x uncertainty band of Fig. 2 is
some factor of three wider than that for P0Ž1.q Ž x . in Fig. 1, reaching a total spread of
about 15% at x , 0.3. Moreover P0Ž2.q Ž x . is rather unconstrained at small x Q 10y2 by
present information, as the leading small-x term w33x does not dominate over less
singular contributions at practically relevant values of x. However, physical quantities
are only affected by the splitting functions via convolutions with smooth non-perturba-
tive initial distributions which ‘wash out’ the oscillating large-x differences of Fig. 2 to
a large extent. Furthermore the convolutions receive important contributions from the
Žwell-constrained. large-x region of P NS Ž x . even at very small x. The above ‘bare’
uncertainty is thus considerably reduced over the full x-range. This effect is illustrated
in Fig. 3, where four representative approximate results for P0Ž2.q are convoluted with a
simple, but typical input shape. The total spread after this convolution is as small as
0.3% for 0.2 Q x Q 0.9, and becomes large only at x Q 0.02.
Fig. 3. The convolution of the approximations ‘A’ – ‘D’ of P0Ž2.q selected in Fig. 2 with a shape typical of
hadronic non-singlet initial distributions.
The uncertainty band of Fig. 3 is rather completely covered by the results ‘A’ and
‘B’. Hence our final estimates for P0Ž2.q Ž x . and its remaining uncertainty are given by
1
P0,Ž2.q
A Ž x . s 1347.207 q 2283.011 d Ž 1 y x . y 722.137 ln2 Ž 1 y x .
Ž1yx .q
y 1236.264 y 332.254 ln x q 1.580 Ž ln4 x y 4 ln3 x . , Ž 4.9 .
1
P0,Ž2.q
B Ž x . s 1137.897 q 1099.754d Ž 1 y x . y 2975.371 x 2
Ž1yx .q
y 125.243 y 64.105 ln2 x q 1.580 ln4 x. Ž 4.10 .
1
The average 2
w Ž2.q Ž . Ž2.q Ž . x
P0, A x q P0, B x represents our central result.
The Nf1-term P1Ž2.q is the leading radiative correction to P1Ž1.q Ž x ., which is in turn
only slightly more complicated than the one-loop non-singlet splitting function w42x.
Hence it is natural to adopt here the ansatz Ž4.3. employed for the two-loop Nf0-piece in
our above illustration. The resulting eight approximations for P1Ž2.q Ž x . are displayed for
x - 1 in the left part of Fig. 4 Ždashed curves.. Their spread at large x is similar to that
obtained for P0Ž1.q in Fig. 1. The leading large-x coefficients fall into the range
y190 ( AŽ2. 1,1 ( y180. Ž 4.11 .
Ž2.q Ž .
The uncertainty of the complete result for P NS x is dominated by the spread of the
above Nf-independent contribution, as estimated by the difference between Eqs. Ž4.9.
and Ž4.10.. This is also true at small x, despite the fact that the band in Fig. 4 is
presumably an underestimate in this region, as a possible term ; ln3 x has been
disregarded. Hence it is sufficient, at the present stage, to keep only the Nf0-contribution
Ž2.q
to the error band of P NS and to employ just one representative for P1Ž2.q. Our choice,
an average of two typical results with and without a lnŽ1 y x . term, reads
1
P1Ž2.q Ž x . s y184.4098 y 180.6971 d Ž 1 y x . y 98.5885 ln Ž 1 y x .
Ž1yx .q
q 205.7690 x 2 q 6.1618 q5.0439 ln2 x Ž 4.12 .
and is also shown in the left part of Fig. 4 Žsolid curve..
Fig. 4. Left: approximations to the Nf1 part of P NS

Ž2.q
, obtained from the five lowest even-integer moments
using Eqs. Ž4.2. and Ž4.3.. Right: approximate results for the Nf0 and Nf1 terms of the three-loop splitting
Ž2.y
function PNS .
As mentioned before the Nf2-piece in Eq. Ž4.1. is exactly known from Ref. w30x. After
transformation to x-space, this contribution reads
1 64
P2Ž2. Ž x . s
81 ž y
Ž1yx .q
y 204 q 192 z Ž 3 . y 320 z Ž 2 . d Ž 1 y x . q 64
x ln x
q
1yx /
Ž 96 ln x q 320. q Ž 1 y x . Ž 48 ln2 x q 352 ln x q 384. ,
Ž 4.13 .
where z Ž l . denotes Riemann’s z-function.
Ž2.y Ž .
Finally we consider PNS x . Here our treatment is inevitably more approximate.
According to the expectations given at the beginning of this section, we take over the
1rŽ1 y x .q and d Ž1 y x . terms of the ‘q’-combinations in Eqs. Ž4.9., Ž4.10., and
Ž4.12.. The remaining coefficients are Žafter inserting the appropriate Nf0 leading
small-x piece w33x. determined by the first, eighth and tenth moments of Refs. w14–16x,
Ž2.q Ž .
assuming that the difference to PNS x is negligible for the latter two, entirely large-x
dominated quantities. The results are shown in the right half of Fig. 4. The uncertainty
band for P0Ž2.y Ž x . is about 50% wider than that for P0Ž2.q Ž x . around x s 0.3, reflecting
the lack of precise knowledge of the intermediate-N moments, but smaller at small-x, as
this region plays a much greater role for the first moment known here from Eq. Ž2.6.,
than for the second moment in the ‘q’-case. Our parametrizations spanning the present
uncertainty are given by
1
P0,Ž2.y
A Ž x . s 1347.207 q 2283.011 d Ž 1 y x . y 722.238 ln2 Ž 1 y x .
Ž1yx .q
y 1234.756 y 327.479 ln x q 1.432 Ž ln4 x y 4 ln3 x . , Ž 4.14 .
1
P0,Ž2.y
B Ž x . s 1137.897 q 1099.754 d Ž 1 y x . y 2954.684 x 2 y 143.709
Ž1yx .q
y 2.761 ln2 x q 1.432 ln4 x Ž 4.15 .
Žalso here the average represents the central result., supplemented by
1
P1Ž2.y Ž x . s y184.4098 y 180.6971 d Ž 1 y x . y 98.5722 ln Ž 1 y x .
Ž1yx .q
q 205.3670 x 2 q 6.5740 q3.5474 ln2 x. Ž 4.16 .
For the latter expression an average has been calculated in the same manner as for
P1Ž2.q .
5. Numerical results
We are now ready to consider the numerical impact of the NNLO terms on the
evolution of the non-singlet parton densities and structure functions. Before doing so,
however, it is worthwhile to look at the perturbative running of a s underlying these
considerations. In the left part of Fig. 5 the a s-expansion Ž2.11. of the b-function is
Fig. 5. Left: The perturbative expansion of the QCD b-function up to order a s5 , for four flavours in the MS
renormalization scheme. Right: Illustration of the resulting scale dependence of a s , using a variable Nf as
detailed in the text. m2r is given in GeV 2 .
shown for Nf s 4 flavours. Besides the contributions of Eq. Ž2.12. relevant for NNLO
calculations, also the contribution yb 3 a s5 of Ref. w45x has been included. If one uses
the effect of this four-loop ŽN 3 LO. term as an estimate of the residual error of the
expansion, the resulting uncertainty amounts to 0.08%, 0.35%, 1.1% and 2.5% for a s s
0.12, 0.20, 0.30 and 0.40, respectively. The effects are somewhat larger Žsmaller. for
Nf s 3 Ž Nf s 5.. The consequences of this expansion on the scale dependence of a s are
illustrated in the right part of Fig. 5. For this illustration we have used Eq. Ž2.11. with
Nf s 3 at m r ( m c s 1.5 GeV, Nf s 4 between m c and m b s 4.5 GeV, and Nf s 5 for
m r ) m b , assuming that a s Ž m2r . is continuous at these thresholds. If a s is fixed to 0.115
at m r s MZ , then the four-loop effect reaches 0.1% Ž1%. only at m2r s 20 GeV 2 Ž1.5
GeV 2 ., respectively. Clearly the truncation of the series Ž2.11. after three terms does not
introduce a significant theoretical uncertainty in the kinematic regime of deep-inelastic
scattering.
For illustrations of the scale dependence of the parton densities and structure
functions, initial distributions have to be chosen at some reference scales, in the
following denoted by m2f ,0 and Q02 , respectively, in Eqs. Ž2.8. and Ž2.15.. We will
employ the function
3
f s x 0.5 Ž 1 y x . Ž 5.1 .
for all six quantities
" " "
f s xq NS Ž x , m2f ,0 . , F2,NS Ž x ,Q02 . and xF3,NS Ž x ,Q02 . . Ž 5.2 .
Eq. Ž5.1. represents a simple model shape which incorporates the most important
features of non-singlet x-distributions of nucleons. The same input is used in all cases,
as this allows for a direct comparison of the effects of the various kernels in Eqs. Ž2.10.
and Ž2.16.. The overall normalization of f is irrelevant for the logarithmic derivatives
considered below. Our initial scales are specified via
a s Ž m2r s m2f ,0 . s a s Ž m2r s Q02 . s 0.2, Ž 5.3 .
irrespective of the order of the expansion. For a s Ž MZ2 . s 0.114 . . . 0.120 this choice
corresponds to m2f ,0 s Q02 , 25–50 GeV 2 , a Q 2-region typical for fixed-target DIS. If
not explicitly indicated otherwise, the results will be given for Nf s 4 massless flavours.
2.
The evolution of qq NS x, m f is illustrated in Fig. 6 for the standard choice m r s m f of
Ž
the renormalization scale. In this case the perturbative expansion appears to be very well
convergent: Except for the region around x , 0.07 where the scale derivative is very
2
small, the NNLO corrections for q˙q q
NS ' dln q NS rdln m f are as small as about 2%, while
the NLO contributions typically amount to 10–20%. The residual uncertainty of the
three-loop splitting functions of Section 4 leads to a noticeable effect only for x Q 0.02,
and even at x , 10y3 this effect does not exceed "2% with respect to the central result
1
2 NNLOA q NNLOB
Ž . not shown in the figure. Over the full x-range the NNLO
corrections are comparable to the dependence on the number of active flavours: If Nf is
increased Ždecreased. to Nf s 5 Ž Nf s 3., q˙q NS is decreased increased by about 2%,
Ž .
respectively.
Another way to assess the reliability of perturbative calculations is to investigate the
stability of the results under variations of the renormalization scale m r . In Fig. 7 the
consequences of varying m r over the rather wide range 18 m 2f ( m2r ( 8 m2f are displayed
for six representative values of x. The relative scale uncertainties of the average results,
estimated by
D q˙q
NS
2 1 2 2 2 1 2 2
max q˙q ˙q
NS Ž x , m r s 4 m f . . . 4m f . y min q NS Ž x , m r s 4 m f . . . 4m f .
' 2 1 2 2
2 < average q˙q
NS Ž x , m r s 4 m f . . . 4m f .
<
Ž 5.4 .
are shown in the left part of Fig. 8. Also this estimate leads to about 2% for the NNLO
uncertainty, an improvement by more than a factor of three with respect to the
2
Fig. 6. The perturbative expansion of the scale derivative, q˙q q
NS ' dln q NS r dln m f , for a non-singlet ‘q’-combi-
nation of quark densities at m r s m f . The initial conditions are as specified in Eqs. Ž5.1. – Ž5.3.. Here and in
what follows the subscripts A and B indicate the approximations for the three-loop splitting functions derived
in Section 4.
2
Fig. 7. The dependence of the NLO and NNLO predictions for dln qq
NS r dln m f on the renormalization scale
m r for six typical values of x.
corresponding NLO result. Even as low as x , 10y3 the NNLO calculation, despite its
approximation uncertainty increasing towards small x, is superior to the NLO.
Finally the evolution of ‘y’-combinations qyNS is illustrated in the right part of Fig.
8. For x ) 0.1 the difference to the ‘q’-case discussed so far is negligible at NLO as
Fig. 8. Left: The renormalization scale uncertainty of the NLO and NNLO predictions for the scale derivative
of qq ˙q
NS, as obtained from the quantity D q NS defined in Eq. 5.4 . Right: The NNLO effects on the evolution of
Ž .
qy
NS for the standard scale choice m r s m f , together with a comparison of the NLO partonic NSq and NSy
evolutions.
Fig. 9. The perturbative expansion of the scale derivative, F˙2,qNS' dln F2,NS
q
r dlnQ 2 , for a non-singlet structure
function at m r s Q. The initial conditions are as specified in Eqs. Ž5.1. – Ž5.3., the leading-order curve is
identical to that of Fig. 6. Also shown Žright part. is the effect of omitting the contribution from the three-loop
Ž2.q
splitting function P NS .
well as at NNLO. At small x the NLO predictions differ by up to 2%. As expected from
the discussion in Section 4, the residual uncertainty of the NNLO result is considerably
less pronounced at small x in the ‘y’-case, but somewhat larger for 0.01 Q x Q 0.1.
We now turn to the evolution of the non-singlet structure functions. The physical
scale derivative F˙2,qNS' dln F2,NS
q
rdlnQ 2 is shown in the left part of Fig. 9 for m r s Q.
Ž1,2.q Ž .
Besides the splitting functions P NS x the effect of which has been illustrated in Fig.
Ž2.q Ž .
6, here also the coefficient functions c 2,Ž1.NS Ž x . and c 2,NS x enter the NLO and NNLO
evolution kernels as detailed in Eq. 2.16 . These additional terms considerably increase
Ž .
the Q 2-dependence at large x, as can be seen by comparing Figs. 6 and 9. For example,
the NNLO corrections rise from 4% at x s 0.5 to about 7, 11 and 21% at x s 0.65, 0.8
and x s 0.95, respectively. The corresponding NLO contributions amount to 24, 30, 37
and 51% of the LO results. Unlike for the parton densities, the NNLO corrections to the
structure functions are larger than the Nf -dependence at large x: If Nf is increased
Ždecreased. to Nf s 5 Ž Nf s 3., F˙2qNS is decreased Žincreased. between 3.5% and 7% for
0.5 ( x ( 0.95, respectively.
The worse convergence of the expansion at large x is due to the large soft-gluon
contributions wln k Ž1 y x .rŽ1 y x .xq, k s 1, . . . ,2 l y 1, to c 2,NSŽ l ." Ž .
x which are conjec-
Ž l .q Ž .
tured to be absent w43x in the MS splitting functions PNS x . Consequently, as shown
in the right part of Fig. 9, keeping only the coefficient-function contributions in the
O Ž a s3 . term of Eq. Ž2.16. yields a very good approximation at large x. In fact P NS Ž2.q
contributes less than 2% to the total NNLO derivative F˙2, NS at x ) 0.2. The residual
q
Ž2.q
uncertainty of PNS , given by the difference NNLOA – NNLOB , is thus completely
negligible in this region.
The dependence of F˙2,qNS on the renormalization scale m r is presented in Figs. 10 and
11 Žleft part., analogously to the partonic case Žsee Eq. Ž5.4.. in Figs. 7 and 8 using
max w F˙2q,NS Ž x , m2r s 14 Q 2 . . . 4Q 2 . x ymin w F˙2q,NS Ž x , m2r s 14 Q 2 . . . 4Q 2 . x

D F˙2q,NS ' . Ž 5.5 .
2 < average w F˙2q,NS Ž x , m2r s 14 Q 2 . . . 4Q 2 . x <
Fig. 10. The dependence of the NLO and NNLO predictions for dln F2,qNS r dlnQ 2 on the renormalization scale
m r for six typical values of x.
The slower large-x convergence of the a s series for F˙2,NS

q
is obvious from these results
as well, e.g., no extremum close to m r s Q is obtained for x s 0.8. The NNLO
uncertainties as estimated using Eq. Ž5.5. read 3%, 4.5% and 7% for x s 0.5, 0.65 and
0.8. The corresponding NLO results are 8.5%, 10.5% and 12%, respectively. The
accuracy of the Q 2-slope predictions is thus improved by a factor 2–3 except for very
large x.
q
Fig. 11. Left: The m r-uncertainty of the scale derivative of F2,NS, as estimated by D F˙2,NS
q
defined in Eq. Ž5.5..
Note that the absolute values of F˙2,qNS are very small for 0.05- x - 0.15. Right: The NNLO effects on the
evolution of F2,yNS for m r s Q, together with a comparison of the NSq and NSy evolutions for F2,NS at NLO.
As for the parton densities shown in Fig. 8, the evolution of F2,yNS illustrated in the
right part of Fig. 11 is indistinguishable from that of F2,qNS at x ) 0.1, while being better
constrained at NNLO at very small x. For x , 10y3 , the 2.5%–3.5% positive effect of
Ž2.y
P NS in Fig. 8 is overcompensated by the coefficient-function contributions. This effect
also occurs for F˙2,qNS not displayed at small x. In both cases the NLO corrections are
"
smaller than for q˙NS , resulting in a better small-x NLO renormalization-scale stability of
F˙2, NS as can be seen by comparing the left parts of Figs. 11 and 8.
"
The scaling violations of F3,yNS are presented in Fig. 12 for the medium- to large-x
" Žl.
region. Since the soft-gluon terms wln k Ž1 y x .rŽ1 y x .xq are identical in c 2,NS and
™
c 3, NS , the results for F˙3,NS and F˙2,NS agree Žfor identical initial distributions as assumed
" Žl.
here. as x 1. However, the different regular terms lead to noticeable differences

already at medium x, reaching 5% and 10% at x , 0.4 and 0.3, respectively. At small x
the corrections are considerably larger for F˙3, NS than for F˙2,NS , resulting in scale
uncertainties of about 10% at NLO and 4% at NNLO for 10y3 Q x Q 10y2 for the
former quantity.
Finally we turn to the determination of a s from scaling violations of non-singlet
structure functions. Here we address the uncertainties Da s which arise from the
truncation of the perturbation series, confining ourselves to the region x R 0.25 of
considerable negative scale derivatives F˙2 ,NS . In this region the results for F˙3,NS are
rather similar to those for F˙2, NS and need not to be considered separately. We also
disregard the negligible large-x differences between the scaling violations of F2,qNS and
F2,yNS and between the NNLOA and NNLOB calculations. Our procedure for estimating
Da s is as follows. For each x we determine those scales m r,min and m r,max which led to
the minimal and maximal NLO and NNLO results for < F˙2, NS < used in Eq. Ž5.5.. The
value of a s Ž Q02 . is then adjusted to obtain, at these values of x and m r , the same results
for F˙2, NS as found for m r s Q0 and a s Ž Q02 . s 0.2 ŽFig. 9, left part.. The latter
standard-scale results thus play the role of the experimental results for F˙2, NS in
determinations of Da s in data fits.
Fig. 12. The scale derivative F˙3,

y y 2
NS' dln F3,NS r dlnQ at m r s m f . The initial conditions are as given in Eqs.
Ž5.1. – Ž5.3., the leading-order curve is the same as in Figs. 6 and 9. Also shown Žright part. is the ratio of
F˙3,yNS to the corresponding result for F2,NS
q
.
The resulting upper and lower limits for a s Ž Q 02 . are shown in the left part in Fig. 13.
Due to the increase of the higher-order corrections towards large x discussed above, the
uncertainty Da s rises with increasing x. As available experimental DIS results are
restricted to x ( 0.85 w4x, we choose a value x , 0.55 for estimating the x-averaged
uncertainties given by the differences to the reference result a s Ž Q02 . s 0.2. This
procedure yields
q0.020
Da s Ž Q02 , 25–50 GeV 2 . NLO s , Ž 5.6 .
y0.012
q0.008
Da s Ž Q02 , 25–50 GeV 2 . NNLO s . Ž 5.7 .
y0.004
Often results and uncertainties for a s from different processes and observables are
compared after evolution to a common reference scale, conventionally chosen as the
Z-boson mass MZ . Adopting Q 02 s 30 GeV 2 Žand Nf s 5 for Q 0 ( m r ( MZ . for
definiteness, one obtains the error bands displayed in the right part of Fig. 13 and
q0.006 q0.0025
Da s Ž MZ2 . NLO s , Da s Ž MZ2 . NNLO s . Ž 5.8 .
y0.004 y0.0015
As expected from our previous discussions below Eq. Ž5.5., the NNLO calculation
reduces the theoretical uncertainty under consideration by a factor of about 2.5.
In a data analysis, also the NLO and NNLO central values for a s Ž Q 02 . will be
different, since the NNLO scaling violations are stronger over most of the large-x region
as shown in Fig. 9. A simple estimate analogous to that for Da s yields
a s Ž MZ2 . NNLO y a s Ž MZ2 . NLO , y0.002. Ž 5.9 .
Due to the strong x-dependence of the NNLOrNLO ratio, this estimate is less reliable
than Eq. Ž5.8., its uncertainty amounts to about "0.001. Nevertheless it is interesting to
note that Eq. Ž5.9. agrees with the findings of Refs. w26–28x from analyses of data on
Fig. 13. Right: The x-dependent theoretical uncertainty of the determination of a s from the scale derivative of
F2, NS at Q02 s 25– 50 GeV 2 , estimated by the m r-variation 14 Q02 ( m2r ( 4Q 02 . The scales leading to a maximal
Žminimal. < F˙2 NS < are denoted by m r,max Ž m r,min .. Left: The resulting error band for a s Ž MZ2 . using Q 02 s 30
GeV 2 and Nf s 5.
Ž2.
F3 . The three-loop splitting function P NS contribute only about y0.0007 to the shift
Ž5.9. of the NNLO result.
6. Summary
We have investigated the effect of the NNLO perturbative QCD corrections on the
scale dependence of flavour non-singlet quark densities and structure functions. For this
purpose, and for application in further analyses, we have derived compact parametriza-
Ž2."
tions of the corresponding three-loop splitting functions PNS and the two-loop
coefficient functions c a,NS, a s 2,3, L. The latter quantities are exactly known w8–13x;
Ž2."
their analytic x-dependent expressions are however rather cumbersome and not readily
transformed to moment space w39,40x. Our parametrizations of c Ž2." a,NS x and their Mellin
Ž .
transforms thus provide a convenient technical tool. They agree to the exact results up to
a few permille or less over the full x-range, thus introducing a negligible error of well
below 0.1% after insertion into the perturbative expansions.
As only partial results are presently available for the three-loop splitting functions
w14–16,30,33x, our parametrizations of P NS Ž2." Ž .
x serve the additional purpose of provid-
ing quantitative estimates of their x-dependent residual uncertainties. The function
x , relevant to the evolution of flavour asymmetries like u q u y w d q d x, is well
Ž2.q Ž .
PNS
constrained at large x by the lowest even-integer moments of Refs. w14–16x, the spread
reaching about "7% at x , 0.3. On the other hand, PNS Ž2.q Ž .
x is very weakly con-
y2
strained for x Q 10 so far, despite the fact that the leading small-x term is known
w33x. The quantity P NSŽ2.y Ž .
x , entering the evolution of the quark–antiquark differences,
is somewhat better Žworse. constrained at small x Žmedium x ., respectively, than
Ž2.q Ž .
PNS x . As the splitting functions enter parton densities and structure functions only
via convolutions with smooth non-perturbative initial distributions, these ‘bare’ uncer-
tainties are very much reduced for physical quantities over the whole x-range. For
example, the spread of P NS Ž2.q
leads to effects of less than "0.2% at x R 0.2 after
convolution with typical nucleonic input shapes. In this region the present uncertainties
Ž2."
of PNS are thus rendered absolutely negligible, leading to effects even below 0.01%
after insertion into the perturbation series. Their impact becomes significant only for
x Q 10y2 , without seriously impairing the NNLO calculations even down to x , 10y3 .
"Ž
The perturbative expansion for the scale dependence dln q NS x, m2f .rdln m2f of the
non-singlet combinations of quark densities appears to be very well convergent. For
a s s 0.2, corresponding to scales of about 25–50 GeV 2 , the NNLO effects of PNS Ž2."
are
on the level of 2% rather uniformly in x. This result is to be compared to the NLO
corrections which amount to 10–20%. Also the variation of the renormalization scale
leads to effects of about "2% at NNLO. Corrections of this size are comparable to the
dependence of the predictions on the number of quark flavours, rendering a proper
treatment of charm effects w46,47x rather important even for large-x non-singlet quanti-
ties.
Especially at x ) 0.5, the higher-order corrections are much larger for the scale
" Ž
derivative dln Fa,NS x,Q 2 .rdlnQ 2 , a s 2,3, of the non-singlet structure functions. This
enhancement is an effect of the coefficient functions containing large wln k Ž1 y x .r
Ž1 y x .xq soft-gluon terms, which are conjectured to be absent in the MS splitting
functions w43x. For example, the NLO and NNLO effects reach about 37% and 11% of
the respective lower-order results at x s 0.8 for a s s 0.2 and four flavours. The NNLO
calculations thus represent a distinct improvement, reducing also the renormalization-
scale dependence of the predictions by a factor of two to three, e.g., to about "7% at
x s 0.8. Accordingly the inclusion of the NNLO corrections into fits of data on
non-singlet scaling violations is expected to yield, besides a slight lowering of the
central values for a s Ž MZ2 . by roughly 0.002, a considerable reduction of the Žso far
dominant. theoretical error due to the truncation of the perturbation series,
q0.0025
Da s Ž MZ2 . NNLO s .
y0.0015 mr
These estimates are compatible with the results of the fits of F3, NS-data performed in
Refs. w26–28x, where an alternative, integer-moment based approach to the calculation
of the scaling violations has been pursued.
FORTRAN subroutines of our parametrizations of c Ž2." Ž2." Ž .
a,NS x , a s 2,3, L, and P NS
Ž . x
can by obtained via email to neerven@lorentz.leidenuniv.nl or avogt@lorentz.
leidenuniv.nl.
Acknowledgements
This work has been supported by the European Community TMR research network
‘Quantum Chromodynamics and the Deep Structure of Elementary Particles’ under
contract No. FMRX–CT98–0194.
Appendix A. Third-order quantities in Mellin-N space
The Mellin transforms of the approximate NNLO expressions of Section 3 and

Section 4 are given in terms of the integer-N sums Sl Ž N . and their analytic continua-
tions
N l
1 Ž y1.
Sl ' Sl Ž N . s Ý l
sz Ž l. y c Ž ly1. Ž N q 1 . . Ž A.1 .
ks1 k Ž l y 1. !
Here z Ž1. stands for the Euler–Mascheroni constant, and z Ž l ) 1. for Riemann’s
z-function. The l th logarithmic derivative c Ž ly1. of the G-function can be readily
evaluated using the asymptotic expansion for Re N ) 10 together with the functional
equation.
Due to the simplicity of our parametrizations for the three-loop splitting functions,
only the most simple Mellin transforms occur for these quantities. Therefore we are able
to dispense with details here. The Mellin-N dependence of the exactly known Nf2-piece
can be found in Ref. w30x.
The moments the non-singlet ‘q’-coefficient function Ž3.2. entering F2e.m . are given
by
Ž2.q
c 2,NS Ž N . s q3.55555 Ž 6 S4 q 8 S3 S1 q 3 S22 q 6 S2 S12 q S14 .
q 20.4444 Ž 2 S3 q 3 S2 S1 q S13 . y 15.5525 Ž S2 q S12 . y 188.64 S1

165.356 15.38 9.7467 358.503 2.9678
q
N
S3 y
N ž 2
y
N / S2 q
N
S2 S1 q
N
S13
174.8 9.7467 190.18 116.734

q ž N2
q
N / ž S12 y
N3
y
N
S1 /
17.01 34.16 306.849 72.5824 1008
q y q y y y 338.044
N4 N3 N 2
N Nq1
½
q Nf y0.59259 Ž 2 S3 q 3 S2 S1 q S13 . y 4.2963 Ž S2 q S12 .
6.072 6.072 18.0408

y6.3489 S1 y
N
S3 y
N ž 2
y
N / S2
6.072 17.97 14.3574 0.07078 4.488

y ž N 3
y
N 2
q
N / S1 q
N
S12 q
N3
4.21808 21.6028 37.91
q
N2
y
N
y
Nq1
q 46.8406 . 5 Ž A.2 .
For the charged current ‘y’-combination the third to fifth line of this result are,
according to Eq. Ž3.3., replaced by
229.916 31.58 9.7467 393.703 2.9678
q
N
S3 q
Nž 2
q
N / S2 q
N
S2 S1 q
N
S13
192.4 9.7467 160.82 61.1321

q ž N2
q
N / ž S12 y
N3
y
N / S1
22.488 39.12 265.774 164.777 1010

q 4
y 3
q 2
y 337.992.y Ž A.3 . y
N N N N Nq1
The first two lines and the sixth line of Eq. ŽA.2. stem from the universal q-distribution
parts of Eq. Ž3.2.. They are exact up to a truncation of the numerical factors w9–11x.
The corresponding N-space results for the coefficient functions Ž3.4. and Ž3.5. for FL
read
136.88 13.62 2 55.79 150.5 0.062
c LŽ2.q
,NS Ž N . s y
N
S2 q
N
S1 q
N
y
N 2
S1 y
N3 ž /
14.85 207.153 53.12 97.48
q 2
q q 3
q y 0.164
N N Ž N q 1. Nq1
16 6 6 6 25
q Nf
27 ½ y
Nq1
S1 q
N
q
Ž N q 1.
2
y
Nq1 5 Ž A.4 .
and
128.4 13.30 59.12 141.7 0.086
c LŽ2.y
,NS Ž N . s y S2 q S12 q ž y 2 / S1 y
N N N N N3
22.21 180.818 46.58 100.8

q 2
q q 3
q y 0.150
N N Ž N q 1. Nq1
16 6 6 6 25
q Nf
27 ½ y
Nq1
S1 q
N
q
Ž N q 1.
2
y
Nq1 5 . Ž A.5 .
Here the Nf parts represent an exact result w8x.

The ‘y’-coefficient function Ž3.6. for F3 , occurring in the n q n sum, leads to
Ž2.y
c 3,NS Ž N . s q3.55555 Ž 6 S4 q 8 S3 S1 q 3 S22 q 6 S12 S2 q S14 .
q20.4444 Ž 2 S3 q 3 S2 S1 q S13 . y 15.5525 Ž S2 q S12 . y 188.64 S1
297.756 147.9 33.2767 298.733 0.9778

q
N
S3 q ž N 2
q
N / S2 q
N
S2 S1 q
N
S13
147.9 33.2767 45.8683 23.532 66.62 67.6

q ž N 2
q
N / S12 y
N
S1 q
N 4
y
N 3
q
N2
373.029 576.8
y y y 338.625 q Nf y 0.59259 Ž 2 S3 q 3 S2 S1 q S13 .
N Nq1
0.042
y 4.2963 Ž S2 q S12 . y 6.3489 S1 y Ž 2 S3 q 3 S2 S1 q S13 .
N
0.96978 9.684 16.4074 10.6538 4.414
q S12 q ž 2
y / S1 q S2 q
N N N N N3
8.683 15.9177 14.97
y 2
y y q 46.856 4 . Ž A.6 .
N N Nq1
For the ‘q’-combination of Eq. Ž3.7. entering F3n N y F3n N one has to replace the third to
fifth line of the above result by
186.816 92.43 33.2767 187.793 0.9778
q
N
S3 q ž N 2
q
N / S2 q
N
S2 S1 q
N
S13
92.43 33.2767 123.121

q ž N 2
q
N / S12 q
N
S1
18.294 60.28 79.14 276.473 467.2

q 4
y 3
q 2
y y y 338.681. Ž A.7 .
N N N N Nq1
References
w1x D.H. Coward et al., Phys. Rev. Lett. 20 Ž1968. 292.

w2x E.D. Bloom et al., Phys. Rev. Lett. 23 Ž1969. 930.
w3x H. Breitenbach et al., Phys. Rev. Lett. 23 Ž1969. 935.
w4x Particle Data Group, C. Caso et al., Eur. Phys. J. C 3 Ž1998. 1, and references therein.
w5x W. Furmanski, R. Petronzio, Z. Phys. C 11 Ž1982. 293, and references therein.
w6x O.V. Tarasov, A.A. Vladimirov, A.Yu. Zharkov, Phys. Lett. B 93 Ž1980. 429.
w7x S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 303 Ž1993. 334.
w8x J. Sanchez Guillen et al., Nucl. Phys. B 353 Ž1991. 337.
w9x E.B. Zijlstra, W.L. van Neerven, Phys. Lett. B 272 Ž1991. 127.
w12x E.B. Zijlstra, W.L. van Neerven, Nucl. Phys. B 383 Ž1992. 525.
w13x E.B. Zijlstra, thesis, Leiden University 1993.
w14x S.A. Larin, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 427 Ž1994. 41.
w15x S.A. Larin, P. Nogueira, T. van Ritbergen, J.A.M. Vermaseren, Nucl. Phys. B 492 Ž1997. 338.
w16x T. van Ritbergen, thesis, Amsterdam University 1996.
w17x R. Hamberg, W.L. van Neerven, T. Matsuura, Nucl. Phys. B 359 Ž1991. 343.
w18x R. Hamberg, thesis, Leiden University 1991.
w19x W.L. van Neerven, E.B. Zijlstra, Nucl. Phys. B 382 Ž1992. 11.
w20x A.D. Martin, R.G. Roberts, W.J. Stirling, Phys. Lett. B 387 Ž1996. 419.
w21x A.D. Martin, R.G. Roberts, W.J. Stirling, R.S. Thorne, Eur. Phys. J. C 4 Ž1998. 463.
w22x CTEQ Collaboration, H.L. Lai et al., Phys. Rev. D 55 Ž1997. 1280.
w23x CTEQ Collaboration, H.L. Lai et al., Michigan State University preprint MSU-HEP-903100, hep-
phr9903282.
w24x ¨ E. Reya, A. Vogt, Z. Phys. C 67 Ž1995. 433.
M. Gluck,
w25x ¨ E. Reya, A. Vogt, Eur. Phys. J. C 5 Ž1998. 461.
M. Gluck,
w26x A.L. Kataev, A.V. Kotikov, G. Parente, A.V. Sidorov, Phys. Lett. B 388 Ž1996. 179.
w27x A.L. Kataev, A.V. Kotikov, G. Parente, A.V. Sidorov, Phys. Lett. B 417 Ž1998. 374.
w28x A.L. Kataev, G. Parente, A.V. Sidorov, ICTP preprint ICr99r51, hep-phr9905310.
w29x J. Santiago, F.J. Yndurain, Madrid University preprint FTUAM-99-8, hep-phr9904344.
w30x J.A. Gracey, Phys. Lett. B 322 Ž1994. 141.
w31x J.F. Bennett, J.A. Gracey, Nucl. Phys. B 517 Ž1998. 241.
w32x S. Catani, F. Hautmann, Nucl. Phys. B 427 Ž1994. 475.
w33x ¨
J. Bumlein, A. Vogt, Phys. Lett. B 370 Ž1996. 149.
w34x ¨
J. Blumlein, A. Vogt, Phys. Rev. D 58 Ž1998. 014020.
w35x ¨
J. Blumlein, V. Ravindran, W.L. van Neerven, A. Vogt, Proc. of DIS 98, Brussels, April 1998, ed. Gh.
Coremans, R. Roosen ŽWorld Scientific, Singapore, 1998. p. 211, hep-phr9806368.
w36x M. Diemoz, F. Ferroni, E. Longo, G. Martinelli, Z. Phys. C 39 Ž1988. 21.
w37x ¨ E. Reya, A. Vogt, Z. Phys. C 48 Ž1990. 471.
M. Gluck,
w38x Ch. Berger, D. Graudenz, M. Hampel, A. Vogt, Z. Phys. C 70 Ž1996. 77.
w39x ¨
J. Blumlein, S. Kurth, DESY preprint 97-160, hep-phr9708388.
w40x ¨
J. Blumlein, S. Kurth, Phys. Rev. D 60 Ž1999. 014018.
w41x J. Kirschner, L.N. Lipatov, Nucl. Phys. B 213 Ž1983. 122.
w42x G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 Ž1980. 27.
w43x A. Gonzales-Arroyo, C. Lopez, F.J. Yndurain, Nucl. Phys. B 126 Ž1979. 161.
w44x ¨
J. Blumlein, S. Riemersma, A. Vogt, Nucl. Phys. B ŽProc. Suppl.. 51C Ž1996. 30, hep-phr9608470.
w45x T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 Ž1997. 379.
w46x E. Laenen, S. Riemersma, J. Smith, W.L. van Neerven, Nucl. Phys. B 392 Ž1993. 162.
w47x M. Buza, Y. Matiounine, J. Smith, W.L. van Neerven, Eur. Phys. J. C 1 Ž1998. 301, and references
therein.
The semiclassical gluon distribution at next-to-leading order

H.G. Dosch, A. Hebecker, A. Metz, H.J. Pirner
¨ Theoretische Physik der UniÕersitat
Institut fur ¨ Heidelberg, Philosophenweg 16 & 19,
D-69120 Heidelberg, Germany
Received 29 September 1999; accepted 4 November 1999
Abstract
The interaction of the partonic fluctuation of the virtual photon in deep inelastic scattering with
soft color fields describing the hadron is treated in an eikonal approximation. It is known that, in
this approach, the small-x limit of the leading-order gluon distribution xg Ž x,Q 2 . is a constant
characterizing the averaged local field strength in the target. Matching the next-to-leading order
calculation in this semiclassical framework with the one-loop parton model result, we obtain the
next-to-leading order contribution to xg Ž x,Q 2 .. It shows a lnŽ1rx . enhancement at small x and is
sensitive to the large distance structure of the target. The final expression is a simple integral over
non-abelian eikonal factors measuring the target color field. We derive a quantitative relation
between the short-distance cutoff of this integral and the scale of the gluon distribution function in
the MS scheme. Our calculation demonstrates that higher order contributions can be systematically
included in the semiclassical approach. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
The small-x limit of deep inelastic scattering ŽDIS. remains one of the most
challenging problems on the interface of perturbative and non-perturbative QCD. It can
be expected that both the perturbative physics described by QCD resummation tech-
niques w1–11x and the non-perturbative soft dynamics underlying the growth of total
hadronic cross sections w12–17x are important for a complete picture of the small-x limit
of structure functions. In this paper, we systematically relate models for the soft color
fields, which we consider to be a promising tool for the characterization of the hadron,
to the hadronic gluon distribution, which is the basic object of the perturbative
treatment.
We are interested in a region of x and Q 2 where, although x is very small, the
concept of parton distributions and the conventional DGLAP evolution equations w1–4x
are applicable. For the description of the DIS process, the gluon distribution g Ž x, m2 .,
which dominates in the small-x region, has to be given at some scale m2 - Q 2 . Thus, the
problem of the non-perturbative structure of the target hadron has to be addressed.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 2 2 - 1
288 H.G. Dosch et al.r Nuclear Physics B 568 (2000) 287–304
Our basic starting point is the idea of describing high-energy processes in QCD by
studying the eikonalized interaction of energetic partons with soft color fields w18,19x. In
this approach, which has been used for the description of DIS in w20x, soft color fields in
the center-of-mass frame of the collision mediate the interaction between the energetic
projectile and target quarks. Progress towards the description of the energy dependence
of cross sections in the framework of soft color field dynamics was reported, e.g., in
w21,22x.
A closely related approach to both inclusive and diffractive DIS, which treats the
target proton as a superposition of soft color fields, was developed in w23–25x. In fact,
this semiclassical approach reproduces the treatment of w18–20x if the dynamics of the
underlying color field is modelled on the basis of the stochastic vacuum w26,27x and a
phenomenological ansatz for the constituent-quark wave function of the proton is made.
For different approaches to the description of the target color fields see, e.g., Refs.
w28,29x. However, in the following we are completely general and use no model-specific
features of the target color field configurations or, equivalently, the wave functional of
the proton.
In the present paper, the gluon distribution g Ž x, m 2 . at x < 1 and m2 4 L2 Žwhere
L is a soft hadronic scale. is calculated for a target given by soft color field
configurations. Here ‘soft’ means that all momentum components of the field are O Ž L..
Following w30x, a scalar ‘photon’, coupled directly to the gluon field, is used as a
convenient theoretical tool for extracting the gluon distribution. The leading order
calculation gives a constant for x g Ž x, m2 ., which is a measure for the averaged gluon
field strength in the target. This is in agreement with the seminal paper of Mueller w30x
and with w25x, the spirit of which we follow closely.
In our opinion, it is crucial, both from a theoretical and a phenomenological
perspective, to demonstrate the viability of the approach at higher orders. However,
already at next-to-leading order the gluon distribution is scheme dependent and a careful
matching of the partonic calculation Žwe use the MS scheme. and the semiclassical
calculation is necessary to obtain an unambiguous result. With this result, we treat
problems that were not addressed in the one-loop calculations of w30,31x, where the
scheme dependence was not discussed. In fact, the problems of regularization and
scheme dependence arise immediately if one attempts to translate the one-loop, uninte-
grated gluon distribution of w31x into a correction to the leading term in the spirit of w25x
Žsee Ref. w32x for a comparison of the results of w25x and w31x in the case of the quark
distribution.. As emphasized in w25x, where diffractive and inclusive quark and gluon
distributions were calculated in the semiclassical approach, the inclusive gluon distribu-
tion dominates the small-x region. Therefore, we expect that our next-to-leading order
result is the dominant correction relevant as input for the next-to-leading order DGLAP
evolution.
In our approach, the most intricate part is the semiclassical calculation at next-to-
leading order. Working in Feynman gauge, we employ the optical theorem and calculate
the forward scattering amplitude. In the high-energy limit, certain diagrams can be
dropped. The remaining contribution is given in the form of a two-gluon production
cross section. In this way, the identification with the parton model result becomes simple
since the dangerous high-mass region, where the semiclassical approximation fails,
cancels explicitly.
H.G. Dosch et al.r Nuclear Physics B 568 (2000) 287–304 289
Let us note that, for a soft hadron governed by the single scale L, the perturbative
expansion of the gluon distribution makes no sense. However, following ideas of w33x
Žsee also the recent calculations of Refs. w34–36x., we can always assume that we are
dealing with a very large target, in which case the gluon distribution becomes calculable
without losing the interest of being genuinely non-perturbative in its origin Žsee Ref. w37x
for a discussion of the new hard scale in a framework close to the present paper.. It
remains to be seen in how far this large-target approach will allow for a description of
the qualitative features in the realistic proton case.
The paper is organized as follows. In Section 2 the scattering of a scalar photon off
the target color field is calculated in the parton model and the semiclassical approach.
The comparison of the two results gives rise to the leading-order semiclassical expres-
sion for the gluon distribution. In Sections 3 and 4, the semiclassical and parton model
calculations, respectively, are carried out at next-to-leading order. The extraction of the
next-to-leading order contribution to the gluon distribution from the comparison of the
semiclassical and the parton model results is the subject of Section 5. Section 6 contains
our conclusions, and a number of technical details of the calculations are outlined in
Appendices A–D.
2. The leading-order result
In the following analysis we use a scalar ‘photon’ coupled directly to the gluon field
as a convenient theoretical tool for extracting the gluon distribution w30x. To be precise,
the real ‘photon’ field x couples to the field strength tensor Fmn via the interaction
Lagrangian
l
LI s y x trFmn F mn . Ž 1.
2
The leading-order amplitude for the scattering of the ‘photon’ off a classical color
field is given by the diagram in Fig. 1.
Let q and k be the momenta of the incoming virtual ‘photon’ Ž q 2 s yQ 2 . and the
outgoing gluon respectively. We define the light-cone components of a vector p by
p "s p 0 " p 3 and work in a frame where the plus components of q and k are large. In
the high-energy limit, the amplitude T a corresponding to Fig. 1 is given in the rest
frame of the proton by
i 2p d Ž k 0 y q0 . T a Ž D H . s yi l Ž 12 kq A˜y
a
Ž D. . Ž e H
)
DH . . Ž 2.
Here D s k y q, the field A˜ is the Fourier transform of the external color field A, and e
is the polarization vector of the produced gluon. The evaluation of the r.h.s. of Eq. Ž2. in
Fig. 1. The process x ™ g in an external color field with one gluon exchange.
Fig. 2. The process x ™ g in an external color field. The dot symbolizes the initial x gg vertex.
the high-energy limit shows that T a does not depend on Dq and Dy due to the
softness of the external field. It is convenient to consider the impact parameter space
amplitude
i l q0
T˜ a Ž x H . s Hdx q tr T aŽ eH
) A
E H . Ay Ž xq , x H . Ž 3.
2CA
Žwith E H ' ErE x H ., which is related to the amplitude in Eq. Ž2. by a Fourier
transformation in transverse space. The xy dependence of A A is irrelevant in the
high-energy limit. Here A A s A b T b and T b are the generators of SUŽ Nc . in the adjoint
representation, which we use throughout this paper; CA s Nc .
Resumming the gluon exchange to all orders means that the fast gluon created at the
initial x gg vertex acquires a non-abelian eikonal factor while travelling through the rest
of the external field. This is illustrated in Fig. 2.
The transition to the resummed amplitude of Fig. 2 is realized by the substitution
A
E H Ay ™U A
Ž` , x q . Ž x H . E H AyA UŽÀ,†xq . Ž x H . Ž 4.
in Eq. Ž3., where
ig `
UŽÀ, xq . Ž x H . s Pexp y H A
dxq Ay Ž xq , x H . . Ž 5.
2 x q
The operator P denotes path ordering along xq.

From this, the total cross section for the scattering of the virtual photon off the color
field target, i.e. the semiclassical Ž sc . leading-order result, can be derived. Using the
identity
`
A
2
Hy`dx q UŽ` , x q . Ž x H . Ž E H AyA Ž xq , x H . . UŽ Axq ,y `. Ž x H . s y E H UŽÀ,y `. Ž x H . ,
ig
Ž 6.
one finds
sscŽ0. Ž x ,Q 2 .
2
l2 E l2
s Hd 2
xH WxAH Ž yH. y H s0 ' Hd 2
x H < E H WxAH Ž 0 H . < 2 ,
4 g 2 CA E yH 4 g 2 CA
Ž 7.
where the index Ž0. stands for ‘leading order’. Here
WxAH Ž y H . s U A Ž x H . U A † Ž x H qy H . y 1 Ž 8.
and
Fig. 3. The partonic process x g ™ g at leading order.

U A Ž x H . s UŽÀ,y `. Ž x H . . Ž 9.
In Eq. Ž7., the summation over color indices is implicit. In the following it is always
assumed that an appropriate averaging over the color fields underlying the basic quantity
W A is performed Žcf. Ref. w25x..
The leading-order semiclassical gluon distribution is obtained by comparing this
result with a conventional partonic calculation, where the target is described by a gluon
distribution. The relevant diagram is shown in Fig. 3, and the corresponding parton
model Ž pm. cross section reads
2
pl2
spŽ0.
m Ž x ,Q . s xg Ž0. Ž x ,Q 2 . . Ž 10 .
4
Identifying the cross sections of Eqs. Ž7. and Ž10., one obtains
1 1
xg Ž0. Ž x ,Q 2 . s 2 Hd 2
x H < E H WxAH Ž 0 H . < 2 . Ž 11 .
2p a s 2CA
This result has been derived in w25x identifying the scaling violations of F2 with the
gluon distribution. Note the color factor 1r2CA in Eq. Ž11. due to the adjoint
™
representation, which we use throughout the paper. As expected, this leading-order
gluon distribution xg Ž0. Ž x,Q 2 . is constant for x 0 and shows no scaling violations
beyond those induced by the explicit a s factor on the r.h.s. of Eq. Ž11.. In Ref. w20x the
value of lim x ™ 0 xg Ž x,Q 2 . has been given in the stochastic vacuum model.
3. Semiclassical calculation at next-to-leading order
™
The leading ln Q 2 calculation of the process x gg in the semiclassical approach
reproduces the conventional gluon–gluon splitting function, as shown in w25x. Here, we
need the complete next-to-leading order total cross section for the scattering of a x
particle off an external color field. This calculation can be simplified significantly if one
starts with a discussion of all diagrams contributing to the forward amplitude. The total
cross section follows from the imaginary part of this amplitude. Furthermore, it is
convenient to begin by considering the two-gluon exchange approximation.
Clearly, the leading order diagram for the forward amplitude is simply the square of
Fig. 1. At next-to-leading order, all the diagrams in Fig. 4 have to be considered.
In the high-energy limit, the diagrams Žd., Že. Žwith their mirrored analogues. and
diagram Žf. do not contribute. This can be understood intuitively by recalling that the
gluon field is localized in a given region of space. Therefore, in the limit of infinite plus
momentum, the right-moving gluonic degrees of freedom have no time for a virtual
fluctuation between their first and second interaction with the external field Žsee
Appendix A for a more technical argument.. A related discussion in the case of particle
radiation in high-energy scattering of external fields can be found in w38x.
Fig. 4. Two-gluon exchange diagrams for the forward scattering of the scalar ‘photon’ x off a color field. The
mirrored analogues of Žc., Žd. and Že. and the analogue of graph Ža. with the two t channel gluons attached to
different horizontal lines are not shown.
What remains to be calculated is the imaginary part of diagram Ža. Žand the
corresponding diagram where the two t channel gluons are attached to different lines.,
™
diagram Žb. and diagram Žc. Žwith its mirrored analogue.. This corresponds to calculat-
ing the cross section for x gg from the amplitude defined by the three diagrams in
Fig. 5. Note that this is a significant simplification since the amplitude in Fig. 5 has no
contributions arising from cutting diagrams Žd., Že. and Žf. of Fig. 4.
The next step is to resum the interaction with the external field to all orders, i.e. to
repeat the step leading from Fig. 1 to Fig. 2. The result is illustrated in Fig. 6. Here, the
first diagram corresponds to the fluctuation of the incoming ‘photon’ into a gg pair
before the target and the subsequent scattering of the gluons off the color field, treated in
the eikonal approximation. In an expansion in powers of the external field, the leading
term reproduces the first two diagrams of Fig. 5. The second diagram of Fig. 6
corresponds to the creation of the two gluons in the space-time region of the external
field, via a x ggg vertex. The two fast gluons then acquire non-abelian eikonal phases
while travelling through the rest of the color field. An expansion in powers of the
external field generates the third diagram of Fig. 5.
The amplitude of Fig. 6 can be calculated using the methods of w23,24x Žsee also Ref.
w39x.. Some details relevant to this particular process are discussed in Appendix B. The
result can be given in the form
l2 dkXH2
sscŽ1. Ž x ,Q 2 . s 6
da
H H a Ž1ya .
32 Ž 2p .
2
N 2d i j q 2 k i k j
= d xHH 2
Hd 2
kH ˜ Ž
WxAH kXH y k H . , Ž 12 .
N 2qkH
2
where a and 1 y a are the longitudinal momentum fractions carried by the two gluons,
N 2 s a Ž1 y a . Q 2 , k H and kXH are the transverse momenta of one of the two gluons
Fig. 5. The simplified amplitude for x ™ gg, which can be used for the calculation of the total cross section.
Fig. 6. The relevant contributions to the amplitude in the next-to-leading order semiclassical calculation. The
dot in the second diagram symbolizes the initial x ggg vertex.
before and after the interaction with the external field respectively, and W˜ A is the
Fourier transform of the function defined in Eq. Ž8.. Summation over the indices i, j and
over the color indices of W˜ A is implicit.
The full next-to-leading order semiclassical result is given by the sum of Eqs. Ž7. and
Ž12.:
ssc Ž x ,Q 2 . s sscŽ0. Ž x ,Q 2 . q sscŽ1. Ž x ,Q 2 . . Ž 13 .
4. Parton model result at next-to-leading order
Working in d s 4 q e dimensions, the amplitude corresponding to Fig. 7, which

includes a factor 1rŽ2 q e . for initial state gluon polarization, a factor 1r2 for identical
final state particles, and the color factor CA s Nc , reads
CA 1 2e
< Tx g ™ g g < 2 s l2d g d2 ½ Ž sˆ4qtˆ4quˆ 4qQ 8 . y Q2 . 5 Ž 14 .
2 ˆˆˆ
2 stu 2qe
Here l d s lmye r2 and g d s g my e r2 are the d-dimensional couplings, and s, ˆ t,ˆ uˆ are
the usual Mandelstam variables of a 2-to-2 scattering process.
™
From this squared amplitude, the total partonic cross section sˆ Ž1. Ž z,Q 2 . for the
process x g gg is obtained by standard methods. The variable z is defined by
z s Q 2rŽ Q 2 q sˆ.. Alternatively, it is given by z s xry, where y is the fraction of the
target momentum carried by the struck gluon. Combining this with the leading order
result, one can write
1 dz
sp m Ž x ,Q 2 . s Hx Ž s 0 d 4 d Ž 1 y z . q sˆ Ž 1 . Ž z ,Q 2 . . yg b Ž y . , Ž 15 .
z
where s 0 d4 s pl2dr4 Žcf. Eq. Ž10... Note that in the following we only calculate terms
enhanced by powers of lnŽ1rx .. Therefore virtual corrections, which affect only the
endpoint z s 1 and do not produce such terms, do not contribute.
Fig. 7. Conventional partonic amplitude for the process x g ™

gg. The crossed contribution belonging to the
last of the three graphs and the ghost diagrams that have to be added for covariant polarization summation are
not shown.
Working in the MS scheme, we renormalize the bare gluon distribution g b Ž x .

according to
as 1 dz 2
g b Ž x . s g Ž x , m2 . y H
2p x z
Pg g Ž z . ½ e 5
q g E y ln4p g Ž y, m2 . . Ž 16 .
The result can finally be given in the form
1 dz as Q2
sp m Ž x ,Q 2 . s s 0 ds44 x Hx z ž d Ž1yz . q
2p
Pg g Ž z . ln
m2
q CMSg Ž z .
/
=g Ž y, m2 . , Ž 17 .
where the integrand is only complete in the region z - 1. In Eq. Ž17. Pg g Ž z . Žwith
z - 1. is the usual gluon–gluon splitting function. The coefficient function C MS g Ž z .,
characteristic of the process under consideration, has been derived to be
1yz 11CA
CMSg Ž z . s Pg g Ž z . ln y . Ž 18 .
z 6 z Ž1yz .
5. Extracting the gluon distribution
As in the leading-order case discussed in Section 2, the gluon distribution is extracted

from the next-to-leading order semiclassical calculation by identifying the next-to-lead-
ing order total cross sections of the semiclassical and the parton model approach given
by Eqs. Ž13. and Ž17.: ssc Ž x,Q 2 . s sp mŽ x,Q 2 ..
We write the gluon distribution xg Ž x, m2 . entering Eq. Ž17. as
xg Ž x , m2 . s xg Ž0. Ž x , m2 . q xg Ž1. Ž x , m 2 . , Ž 19 .
2. 2.
where xg x, m is given by Eq. Ž11. and xg x, m is a higher-order correction.
Ž0. Ž Ž1. Ž
Inserting Eq. Ž19. into Eq. Ž17., both g Ž0. and g Ž1. are kept in the d Ž1 y z . term, but
only the leading-order distribution g Ž0. is kept in the a s contribution. Now, identifying
ssc and sp m , the leading order parts of both cross sections cancel and one finds
xg Ž1. Ž x , m2 .
2
1 dkXH2 N 2d i j q 2 k i k j
s 7 Hd aH a Ž 1 y a . Hd 2
xH Hd 2
kH ˜ Ž
WxAH kXH y k H .
4 Ž 2p . N 2qkH
2
as 1 Q2
y Hx dz Pg g Ž z . ln q CMSg Ž z . yg Ž0. Ž y, m2 . . Ž 20 .
2p m2
When evaluating the r.h.s. of this equation, all terms that are not enhanced by lnŽ1rx .
will be dropped. In particular, we can use
2CA 2CA 1
Pg g Ž z . , and C MS
g Ž z. , ž ln y 11
12 . / Ž 21 .
z z z
Note that in Eq. Ž20. we are only interested in the leading twist contribution of the
semiclassical cross section which will be determined in Eq. Ž22..
The integral involving the function W˜ A is conveniently rewritten using the integra-
tion variable z s Q 2rŽ Q 2 q M 2 ., where M 2 s sˆ s kXH2 rŽ a Ž1 y a .. is the invariant
mass of the two gluons in the final state. We introduce a parameter mX such that, for a
soft hadronic scale L governing the behaviour of W˜ A , one has L2 < mX 2 Q Q 2 . This
allows us to decompose the integral according to Žsee Appendix C for details.
2
1 dkXH2 N 2d i j q 2 k i k j
7
da
H H a Ž1ya . H d2 xH H d2kH W˜xAH Ž kXH y k H .
4 Ž 2p . N 2qkH
2
1 1 dz Q2 2 1 dz X2
s
4p 3 H z
ln
zm X2 Hd 2
x H < E H WxAH Ž 0 H . < 2 q
p
H Hm dk f Ž k . ,
z 0
X2
H
X2
H
Ž 22 .
where the non-perturbative W A dependence is encoded in the function f :
d2 yH d2 zH X
f Ž kXH2 . s H Ž 2p . 2 2
H Ž 2p . 2 2
Hd 2
x H tr WxAH Ž y H . WxAH† Ž z H . e i k H Ž y H yz H .
yH zH
2
2Ž y H z H .
=
ž 2 2
yH zH
y1 .
/ Ž 23 .
Here and in the following we disregard all terms suppressed by powers of L2rmX 2 . The
motivation for writing the integral in the form given in Eq. Ž22. is the explicit separation
of the ln Q 2 term, which is multiplied by the short-distance specific function E H WxAH
Ž0 H ..
Note that we have not specified the lower bound of the z integrations in Eq. Ž22..
Clearly, the kinematical limit is z s x since the invariant mass of the produced gg pair
can not be larger then the total center-of-mass energy available. However, no such
bound appears explicitly in the semiclassical treatment since the classical color field
behaves like an infinitely heavy target. The physical cutoff is provided by the break-
down of the semiclassical approximation for z ; x, i.e. y ; 1.
Therefore, in order to obtain the leading logarithm, the second term on the r.h.s. of
Eq. Ž22. has to be treated by applying the substitution
H
1 dz
z
™ 1
ln .
x
Ž 24 .
For the first term on the r.h.s. of Eq. Ž22., only the short distance structure of the
external field matters. According to Eq. Ž11., this short distance behaviour is character-
ized by the leading order gluon distribution yg Ž0. Ž y, m2 .. Thus, we substitute
1 1 dz Q2
H ln Hd 2
x H < E H WxAH Ž 0 H . < 2
2CA z z mX 2
™H x
1 dz
z
ln
Q2
z mX 2
2p 2a s Ž m2 . yg Ž0. Ž y, m2 . Ž 25 .
™
and assume that the behaviour of the phenomenological gluon distribution at z x, i.e.
y s xrz 1, correctly accounts for the region where the semiclassical treatment is no
™
longer valid. In fact, when inserting Eq. Ž22. with the substitutions Eq. Ž24. and Ž25.
into Eq. Ž20., the details of the large-y behaviour of yg Ž0. Ž y, m2 . do not matter since the
ln2 Ž1rx . enhanced contributions from the semiclassical and the partonic calculations
cancel. Thus, we obtain
1 as m2 2 X2
xg Ž1. Ž x , m2 . sln
x p
CA ln
ž m X2
q 11
12 xg
/Ž0.
Ž x , m2 . q Hm dk f Ž k .
p 0
X2
H
X2
H , Ž 26 .
where the mX 2 dependence cancels between the two terms. Therefore, we can set
mX 2 s m2 expw11r12x and write
2 1 m 2 exp w11r12 x
xg Ž1. Ž x , m2 . s ž /Hln dkXH2 f Ž kXH2 . . Ž 27 .
p x 0
Note that the correction xg Ž1. Ž x, m2 . shows scaling violations consistent with the
Altarelli–Parisi evolution at small x and a logarithmic small-x enhancement that is
sensitive to the non-perturbative, large-size structure of the target.
It would be interesting to evaluate Eq. Ž27. in the framework of the model of the
stochastic vacuum of w19x or of the large hadron model employed in w25x. Note in
particular that, following w25x, the large-Nc expression
2 2 2
2
Hd x H tr Ž WxAH Ž y H . WxAH† Ž z H . . s V Nc2 1yeya y H yeya z H qeya Ž y H yz H .
Ž 28 .
can be used in Eq. Ž23.. Here V is the geometrical cross section of the target, the impact
parameter dependence of the target thickness Žwhich would be reflected in an impact
parameter dependence of the parameter a. is neglected, and averaging over all relevant
color field configurations is assumed. With this model, the unintegrated, i.e. kXH
dependent, version of xg Ž1. Ž x, m2 . reproduces the recent result of Mueller Žcf. Eq. Ž44.
of Ref. w31x.. The result presented in Eq. Ž27. extends the discussion given in w31x by
carefully matching the semiclassical and the parton model calculations. This gives rise to
a precise definition of the cutoff of the kXH integration in Eq. Ž27. and to the
interpretation of the result as a correction to the leading order gluon distribution of Eq.
Ž11..
Eqs. Ž23. and Ž27. can be significantly simplified Žcf. Appendix D for details. leading
to the main result of our paper:
2
1 1 ` dy H
xg Ž1. Ž x , m2 . s
p3 ž /H
ln
x r Ž m .2
4
yH ½ y d 2 x H trWxAH Ž y H . ,
H 5
1
y2 g E
2
4e 12
r Ž m. s . Ž 29 .
m2
It is easy to see how such a formula comes about: if the cutoff m2 in Eq. Ž27. could be
taken to infinity, the integration dkXH2 s d 2 kXH rp would give rise to the d-function
d 2 Ž y H yz H . Žcf. Eq. Ž23.., and either the y H or the z H integration could be trivially
performed. The remaining integration, say the y H integration, is now divergent at small
y H , showing that the cutoff in Eq. Ž27. can, in fact, not be removed. The integrand of
this divergent y H integration is precisely the one of Eq. Ž29. and, as can be seen by
closer inspection of the relevant integrals Žcf. Appendix D., it is possible to translate the
upper cutoff of the kXH integration into a lower cutoff of the y H integration. This is the
origin of the compact formula in Eq. Ž29..
Given a specific model for the gluon fields of the target that allows for the calculation
of the fundamental quantity WxAH Ž y H ., Eq. Ž29. can be used to improve the leading
order semiclassical result of Eq. Ž11..
With the large-Nc expression of w25x,
y2 d 2 x H tr WxAH Ž y H . s d 2 x H tr Ž WxAH Ž y H . WxAH† Ž y H . .

H H
2
s 2 V Nc2 1 y eya y H , Ž 30 .
one can see the qualitative agreement of Eq. Ž29. with Eqs. Ž49. – Ž51. of Ref. w30x. The
progress of the present investigation is the careful matching of the semiclassical and the
parton model treatment, giving rise to the precise definition of the cutoff of the y H
integration in terms of the scale of the gluon distribution.
6. Summary and conclusions
The semiclassical approach has been successfully applied to different kinds of

high-energy reactions. In DIS, the interaction of the energetic partons of the photon with
the target can be calculated in the eikonal approximation and is essentially given by a
Žnon-perturbative. Wegner–Wilson loop, which measures an integral of the field strength
of the target.
This approach, though limited to low scales m2 , can predict input distributions for the
DGLAP equation. Of course, in order to make numerical estimates of the parton
densities, the evaluation of the Wegner–Wilson loop in a specific non-perturbative
model is necessary.
In this respect the gluon distribution is of particular interest as it dominates DIS at
low x. To leading order, the distribution x g Ž0. Ž x, m 2 . is a constant characterizing the
averaged local field strength in the target. Therefore, a calculation at next-to-leading
order is mandatory to obtain a non-trivial energy dependence of the gluon density.
Using a scalar ‘photon’ Ždenoted by x ., which couples directly to the gluon in a
gauge invariant way, we derive the gluon density by matching the semiclassical and the
™
parton model approach. At leading order, we have to equate the cross section for the
™
transition x g in a soft external field with the cross section of the process x g g as
™
given in the parton model. At next-to-leading order, the cross section for the transition
™
x gg in a soft external field has to be equated with the parton model cross section of
the process x g gg. The a s correction of the gluon distribution shows a logarithmic
enhancement, i.e. x g Ž1. Ž x, m2 . A lnŽ1rx ..
In our approach we can not go beyond the leading-lnŽ1rx . approximation. This is a
fundamental limitation of the semiclassical treatment and many related approaches,
which has its origin in the artificial separation of the QCD dynamics into the soft
degrees of freedom of the target and the high-energy modes of the projectile. Integrating
over the fluctuations of the projectile, one has to drop the soft modes since their
interaction can not be treated in the eikonal approximation.
Parton distributions at next-to-leading order are scheme dependent. Our final result in
Eq. Ž29. Žtogether with the leading order contribution of Eq. Ž11.. provides the gluon
density in the MS scheme. The short-distance cutoff in Eq. Ž29. is quantitatively related
to the scale of the gluon distribution. This enables us to obtain numerical predictions for
the gluon distribution at next-to-leading order in any non-perturbative approach describ-
ing the soft color field of the proton.
Acknowledgements
One of us ŽA.H.. would like to thank W. Buchmuller ¨ and T. Gehrmann for the
fruitful collaboration on a previous paper, during which essential ideas underlying this
present investigation took shape.
Appendix A. Suppression of intermediate virtual fluctuations in the high-energy

limit
To illustrate the vanishing of diagrams Žd., Že. and Žf. of Fig. 4 in the high-energy
limit, consider first the simpler case where the horizontal gluon lines are replaced by
scalar lines. The scalar version of diagram Fig. 4d. with the naming of momenta used in
this appendix is shown in Fig. 8.
In the high-energy limit, the soft external field can not change the momentum plus
component of the fast right-moving particles essentially, kq, qq. Furthermore, the
dependence of the external field vertex on the minus component of the transferred
momentum can be neglected. Thus, the minus component integrations in the diagram
Fig. 8 take the form
I s dky dly
H H
1
= 2 2 2
. Ž A.1 .
w kq ky y k H q ie xw lq ly y l H q ie x w Ž k y l . q Ž k y l . yy Ž k y l . H q ie x
Note first that this expression vanishes unless 0 ( lq( kq, in which case the two poles
in the complex ly plane lie on different sides of the integration contour. However, this
means that both kq and Ž k y l .q have to be positive, so that now both poles in the
complex ky plane are on the same side of the integration contour. This demonstrates
that the integral I vanishes in any case.
Fig. 8. Scalar version of diagram Fig. 4d..

The same argument applies to the scalar analogues of diagrams Figs. 4e,f, since there
the momentum k also flows through more than one propagator between the two external
field vertices. This results in the presence of several poles in ky, all on the same side of
the integration contour, and therefore in the vanishing of the integral.
All that has been said above applies to the gluonic case as well, as long as there is no
additional minus component dependence introduced by the numerator factors. Such a
dependence would prevent us from closing the integration contour. Working in Feynman
gauge and applying the decomposition of the gluon propagator given in Appendix B to
the lines with momentum k, one can easily convince oneself that, in the high-energy
limit, there is indeed no dependence on the minus components.
Thus, we are justified in disregarding diagrams Figs. 4d–f when calculating the
forward scattering amplitude at next-to-leading order.
Appendix B. Details of the next-to-leading order semiclassical calculation
The left-hand diagram of Fig. 6 contains three contributions: two of them describe the
interaction of a single outgoing gluon with the color field, whereas in the third case both
gluons interact with the field. Here we consider the latter case as an example, and
outline some crucial steps of the evaluation of the corresponding amplitude which we
call T g g .
Working in Feynman gauge and denoting the momenta of the gluons before Žafter.
the interaction by k and p Ž kX and pX ., respectively, the amplitude reads
d4 k yig mr
i2pd Ž kX0 q pX0 y q0 . Ti gj g s H Ž 2p . e a)
4 Ž i.
X X
Ž p . Va m Ž p , p . Hrs Ž p,k .
p2
yig sn
Vnb† Ž kX ,k . eŽbj.) Ž kX . ,
= Ž B.1 .
k2
where the indices i, j characterize the polarization of the gluons in the final state. The
quantity H represents the x gg vertex,
Hrs Ž p,k . s i l Ž Ž kp . g rs y kr ps . , Ž B.2 .
while V Ž pX , p . and V Ž kX ,k ., which will be explicitly given below, are the effective
vertices for the scattering of the fast gluons off the external field.
To simplify the calculation we exploit an alternative representation of the metric
tensor appearing in the gluon propagators. For instance, the tensor g mr of the propaga-
tor yig mrrp 2 is decomposed according to
2 n mp r p mn r nmn r
g mr s y Ý eŽmi. Ž p . eŽrj. Ž p . q q y 2
p2 , Ž B.3 .
is1 np np Ž np .
with the light like vector n s Ž nq,ny,n H . s Ž0,2,0 H .. A possible choice of the polariza-
tion vectors eŽ i.Ž p . in Eq. ŽB.3., which in particular satisfy eŽ i.Ž p . p s 0, is given by
2 p H eŽ i .H
ž
eŽ i. Ž p . s 0,
pq /
,e Ž i . H , Ž B.4 .
with the transverse basis eŽ1.H s Ž1,0. and eŽ2.H s Ž0,1..

In the high energy limit, the t channel exchange of gluons leads to an amplitude
which is proportional to qq. As a consequence, in order to obtain the leading part of the
amplitude, in Eq. ŽB.3. only the term containing the polarization vectors has to be
considered. Because of gauge invariance of the x gg vertex the second term on the r.h.s.
of Eq. ŽB.3. vanishes. The contribution of the remaining two terms is at most
proportional to Ž qq . 0 and hence negligible.
The effective vertices are now multiplied by the polarization vectors of two on-shell
gluons; these vertices are governed by the non-abelian eikonal factor defined in Eqs. Ž9.
and Ž5.. One finds
eŽai. Ž pX . Va m Ž pX , p . eŽmj. Ž p . s 2pd Ž pX0 y p 0 . 2 p 0

2
= Ũ A Ž pXH y p H . y Ž 2p . d 2 Ž pXH y p H . d i j ,
Ž B.5 .
and an analogous expression for the second vertex. The contribution proportional to
d 2 Ž pXH y p H . subtracts the term in U˜ A which contains no interaction, while d i j
indicates conservation of the gluon helicity. In Ref. w39x the interested reader can find
more details of the derivation of Eq. ŽB.5..
Writing the loop integration in terms of light cone variables, d 4 k s
Ž1r2. dkq dky d 2 k H , and using the approximation d Ž kX0 y k 0 . , 2 d Ž kq
X
y kq . we arrive
at
l 1
Ti gj g s y 3
2
qq dky d 2 k H
H a Ž 1 y a . Ž Q 2 q k 2 q p 2 . di j q 2 k i k j
2 Ž 2p . k p2
2
2
= Ũ A Ž pXH y p H . y Ž 2p . d 2 Ž pXH y p H .
2
= Ũ A † Ž kXH y k H . y Ž 2p . d 2 Ž kXH y k H . . Ž B.6 .
The diagram on the r.h.s. of Fig. 6 cancels the k 2 and p 2 term in the expression
Ž Q 2 q k 2 q p 2 . of Eq. ŽB.6.. Now, the ky integration can be performed by closing the
integration contour in the lower half of the complex ky plane.
The resulting expression contains terms proportional to U˜ A U˜ A †, U˜ A , and U˜ A †, as
well as a constant term. The two additional contributions, where only a single gluon
interacts with the external field and which belong to the left-hand side of Fig. 6, contain
terms proportional to U˜ A and U˜ A † and a constant term. If one adds these contributions,
the color field dependence of the total amplitude Ti j turns out to be
4
U˜ A Ž pXH y p H . U˜ A † Ž kXH y k H . y Ž 2p . d 2 Ž pXH y p H . d 2 Ž kXH y k H .
X X
s d 2 x H eyi x H Ž k H qp H . W˜xAH Ž kXH yk H . .
H Ž B.7 .
This provides us with the final result of the total amplitude,
yi l qq N 2d i j q 2 k i k j X X
Ti j s 2 Hd 2
kH 2 2 Hd 2
x H eyi x H Ž k H qp H . W˜xAH Ž kXH yk H . .
2 Ž 2p . N qkH
Ž B.8 .
The cross section in Eq. Ž12. is obtained by means of the standard formula for
scattering off an external field,
1 d 3 kX d 3 pX 1
sscŽ1. Ž x ,Q 2 . s H 3 X 3
2pd Ž kX0 q pX0 y q0 . < Ti j < 2 , Ž B.9 .
2 Ž 2p . 2 k 0 Ž 2p . 2 pX0 2 q0
where the two identical particles in the final state require the factor 1r2 in front of the
integral.
Appendix C. Evaluation of the integral with the function W A
In this appendix we present some details of the derivation of Eq. Ž22.. To introduce
the integration variable z we exploit the relation kXH2 rŽ a Ž1 y a .. s M 2 s Q 2 Ž1 y z .rz.
Therefore the kXH integration appearing on the l.h.s. of Eq. Ž22. can be replaced
according to
H a Ž1ya . ™Q H
dkXH2 2
1 dz
, Ž C.1 .
z2
where just the leading contribution at small z is kept. Concerning the lower bound of
the z integration we refer to the discussion given in Section 5.
Subsequently, it is convenient to divide the a integration of Eq. Ž22. in two parts by
introducing an arbitrary parameter mX which fulfills the condition L2 < mX 2 Q Q 2 . To be
specific we separate the symmetric and asymmetric gluon configurations in the integra-
tion by means of
X2 2 X2 2
Hd a sHm1yrMm rM d a q 2H0m rM d a ,
X2 2
Ž C.2 .
where the symmetric Žasymmetric. configurations give rise to the first Žsecond. term on
the r.h.s. of Eq. Ž22.. For a given value of M 2 the symmetric contribution contains large
transverse momenta kXH while the asymmetric term is entirely soft.
To extract the leading twist of the symmetric hard part we use
N 2d i j q 2 k i k j
Hd 2
kH W˜xAH Ž kXH y k H . s Ci j, a d 2 k H k a W˜xAH Ž yk H .
H
N 2qkH
2
q higher twist , Ž C.3 .

with
kXi d ja q kXj d i a kXa Ž N 2d i j q 2 kXi kXj .

Ci j, a s 2
ž N 2 q kXH2
y
Ž N 2 q kXH2 .
2
/ .
The simple relations

4 z Ž1yz . Ž1qz2 .
Ci j, a Ci j, b s da b , Ž C.4 .
N2
and
2
4 2
Hd 2
k H k aW˜xAH Ž yk H . s Ž 2p . E H WxAH Ž 0 H . Ž C.5 .
are important for the further evaluation of the hard contribution. In writing Eq. ŽC.4. we
have anticipated the symmetric kXH integration and employed the substitution kXa kXb
kXH2 d a br2. Finally, performing the a integration and keeping only the contribution that
™
is dominant at small z, the first term on the r.h.s. of Eq. Ž22. is obtained.
In the case of the asymmetric soft contribution, which is independent of Q 2 and
therefore leading twist, the a integration is replaced by an integral over kXH . Since a is
small we have Q 2a , kXH2 z, leading to the replacement
X2 2 z X2
H0m rM d a , Q H0m dk 2
X2
H . Ž C.6 .
The remaining step to obtain the second term on the r.h.s. in Eq. Ž22. is the k H
integration. Keeping in mind that for the soft term we can neglect N 2 this integration
can easily be done with the aid of
ki kj 2p 2 yi y j
Hdk 2
H 2
kH
e i k H y Hs 2
yH ž d i jy 2
yH / . Ž C.7 .
Appendix D. Simplification of the final expression
In this appendix, the details of the derivation of Eq. Ž29., which represents a
particularly simple form of our final result, are described.
It is convenient to introduce a real parameter e ) 0 and to write the gluon distribution
given by Eqs. Ž23. and Ž27. in the form
2 1
xg Ž1. Ž x , m2 . s 2 ž /
ln I0 Ž 0, mX . , Ž D.1 .
p x
where
Ie Ž a,b . ' Ha -k X2
d 2 kXH fe Ž kXH2 . , Ž D.2 .
2 2
H-b
d2 yH d2 zH X
fe Ž kXH2 . ' H Ž 2p . 2 2y e
H Ž 2p . 2 2y e
h Ž y H , z H . e i k H Ž y H yz H .
yH zH
2
2Ž y H z H .
=
ž 2 2
yH zH
y1 ,
/ Ž D.3 .
and
h Ž y H , z H . ' d 2 x H tr WxAH Ž y H . WxAH† Ž z H . .

H Ž D.4 .
Furthermore, we have
I0 Ž 0, mX . s lim Ie Ž 0,` . y Ie Ž mX ,` . .
e™0 Ž D.5 .
The kXH integration in the definition of Ie Ž0,`. is easily performed giving a d-function
of y H yz H . The result is
d2 yH
Ie Ž 0,` . s H Ž 2p . 2 4y2 e
hŽ y H , y H . . Ž D.6 .
yH
Since
h Ž y H , z H . , C y H Pz H for < y H < , < z H < < 1rL , Ž D.7 .

where C is a constant, one can write
C r 2e 1 2
dy H
Ie Ž 0,` . s q H 4
hŽ y H , y H . q OŽ e . . Ž D.8 .
2p 2 e 4p r -y 2 2
H
yH
The parameter r has been introduced to separate the small-distance from the large-dis-
tance part of the y H integration.
In the kXH integration defining Ie Ž mX ,`., the momentum variable kXH is always large.
Therefore, the result is only sensitive to the small distance structure of hŽ y H , z H ., and
we can write
d2 yH d2 zH X
Ie Ž mX ,` . s Hm -kX2 X2
d 2 kXH H Ž 2p . 2 2y e
H Ž 2p . 2 2y e
C y H Pz H e i k H Ž y H yz H .
H yH zH
2
2Ž y H z H .
=
ž 2 2
yH zH
y1 .
/ Ž D.9 .
This integral can be calculated using standard methods:

2
1ye
C m Xy2 e G Ž1qe . G ž /
Ie Ž mX ,` . s
2p 2
2e
G 1yž
e
2 /
2
0 OŽ e 2. .
Ž 1qe . qO Ž D.10 .
When inserting Eqs. ŽD.8. and ŽD.10. in Eq. ŽD.5., the poles in e cancel. Setting the
parameter r, which so far was arbitrary, to
1
y2 g E
1y2 g E 12
2
4e 4e
r Ž m. s s , Ž D.11 .
mX 2 m2
the result of Eq. Ž29. follows. Here g E is Euler’s constant.

References
w1x V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 Ž1972. 438.
w2x V.N. Gribov, L.N. Lipatov, Sov. J. Nucl. Phys. 15 Ž1972. 675.
w3x G. Altarelli, G. Parisi, Nucl. Phys. B 126 Ž1977. 298.
w4x Yu.L. Dokshitzer, Sov. Phys. JETP 46 Ž1977. 641.
w5x E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 44 Ž1976. 443.
w6x E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Sov. Phys. JETP 45 Ž1977. 199.
w7x Y.Y. Balitsky, L.N. Lipatov, Sov. J. Nucl. Phys. 28 Ž1978. 822.
w8x J. Bartels, Nucl. Phys. B 151 Ž1979. 293.
w9x J. Bartels, Nucl. Phys. B 175 Ž1980. 365.
w10x A.H. Mueller, Nucl. Phys. B 415 Ž1994. 373.
w11x N.N. Nikolaev, B.G. Zakharov, Z. Phys. C 64 Ž1994. 631.
w12x A. Donnachie, P.V. Landshoff, Nucl. Phys. B 244 Ž1984. 322.
w13x A. Donnachie, P.V. Landshoff, Phys. Lett. B 296 Ž1992. 227.
w14x A. Donnachie, P.V. Landshoff, Phys. Lett. B 437 Ž1998. 408.
w15x J.R. Cudell, A. Donnachie, P.V. Landshoff, Phys. Lett. B 448 Ž1999. 281.
w16x ¨
K. Golec-Biernat, M. Wusthoff, Phys. Rev. D 59 Ž1999. 014017.
w17x U. D’Alesio, A. Metz, H.J. Pirner, Eur. Phys. J. C 9 Ž1999. 601.
w18x O. Nachtmann, Ann. Phys. 209 Ž1991. 436.
w19x ¨
H.G. Dosch, A. Kramer, Phys. Lett. B 252 Ž1990. 669.
w20x H.G. Dosch, T. Gousset, H.J. Pirner, Phys. Rev. D 57 Ž1998. 1666.
w21x I. Balitsky, Phys. Rev. D 60 Ž1999. 014020.
w22x J. Jalilian-Marian, A. Kovner, H. Weigert, Phys. Rev. D 59 Ž1999. 014015.
w23x ¨
W. Buchmuller, M.F. McDermott, A. Hebecker, Nucl. Phys. B 487 Ž1997. 283.
w24x ¨
W. Buchmuller, M.F. McDermott, A. Hebecker, Nucl. Phys. B 500 Ž1997. 621, ŽE..
w25x ¨
W. Buchmuller, T. Gehrmann, A. Hebecker, Nucl. Phys. B 537 Ž1999. 477.
w26x H.G. Dosch, Phys. Lett. B 190 Ž1987. 177.
w27x H.G. Dosch, Yu.A. Simonov, Phys. Lett. B 205 Ž1988. 339.
w28x F. Hautmann, Z. Kunszt, D.E. Soper, Phys. Rev. Lett. 81 Ž1998. 3333.
w29x Yu.V. Kovchegov, L. McLerran, Phys. Rev. D 60 Ž1999. 054025.
w30x A.H. Mueller, Nucl. Phys. B 335 Ž1990. 115.
w31x A.H. Mueller, preprint CU-TP-937, hep-phr9904404.
w32x ¨
W. Buchmuller, talk at New Trends in HERA Physics, Ringberg Workshop 1999, hep-phr9906546.
w33x L. McLerran, R. Venugopalan, Phys. Rev. D 49 Ž1994. 2233.
w34x Yu.V. Kovchegov, Phys. Rev. D 54 Ž1996. 5463.
w35x Yu.V. Kovchegov, Phys. Rev. D D 55 Ž1997. 5445.
w36x Yu.V. Kovchegov, A.H. Mueller, Nucl. Phys. B 529 Ž1998. 451.
w37x A. Hebecker, H. Weigert, Phys. Lett. B 432 Ž1998. 215.
w38x S.J. Brodsky, P. Hoyer, Phys. Lett. B 298 Ž1993. 165.
w39x A. Hebecker, preprint HD-THEP-99-12, hep-phr9905226.
HqHy pair production at the Large Hadron Collider

A.A. Barrientos Bendezu´ a , B.A. Kniehl b,1
a
¨ Theoretische Physik, UniÕersitat
II. Institut fur ¨ Hamburg, Luruper Chaussee 149, 22761 Hamburg, Germany
b
Max-Planck-Institut fur¨ Physik (Werner-Heisenberg-Institut), Fohringer
¨ Ring 6, 80805 Munich, Germany
Received 19 August 1999; accepted 22 November 1999
Abstract
We study the pair production of charged Higgs bosons at the CERN Large Hadron Collider in
the context of the minimal supersymmetric extension of the standard model. We compare the
contributions due to qq annihilation at the tree level and gg fusion, which proceeds at one loop.
At small or large values of tan b , HqHy production proceeds dominantly via bb annihilation,
due to Feynman diagrams involving neutral CP-even Higgs bosons and top quarks, which come in
addition to the usually considered Drell–Yan diagrams. In the case of gg fusion, the squark loop
contributions may considerably enhance the well-known quark loop contributions. q 2000 Else-
vier Science B.V. All rights reserved.
PACS: 12.60.Jv; 13.85.-t; 14.80.Cp
1. Introduction
One of the prime objectives of the CERN Large Hadron Collider ŽLHC. is the search
for spin-zero particles which remain in the physical spectrum after the elementary-par-
ticle masses have been generated through the Higgs mechanism of electroweak symme-
try breaking w1x. Should the world be supersymmetric, then the Higgs sector is more
complicated than in the standard model ŽSM., which predicts just one scalar Higgs
boson. The Higgs sector of the minimal supersymmetric extension of the SM ŽMSSM.
consists of a two-Higgs-doublet model Ž2HDM. and accommodates five physical Higgs
bosons: the neutral CP-even h 0 and H 0 bosons, the neutral CP-odd A0 boson, and the
charged H "-boson pair. At the tree level, the MSSM Higgs sector has two free
parameters, which are usually taken to be the mass m A of the A0 boson and the ratio
tan b s Õ 2rÕ1 of the vacuum expectation values of the two Higgs doublets.
1
¨ Theoretische Physik, Universitat
Permanent address: II. Institut fur ¨ Hamburg, Luruper Chaussee 149,
22761 Hamburg, Germany.
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 3 2 - 4
306 ´ B.A. Kniehl r Nuclear Physics B 568 (2000) 305–318
A.A. Barrientos Bendezu,
Fig. 1. Tree-level Feynman diagrams for bb ™H q

Hy in the MSSM.
The discovery of the H " bosons would prove wrong the SM and, at the same time,
give strong support to the MSSM. The logistics of the H "-boson search at the LHC
™ ™ ™
may be summarized as follows. For H "-boson masses m H - m t y m b , the dominant
production mechanisms are gg,qq tt followed by t bHq w1x.2 The dominant decay
mode of H " bosons in this mass range is Hq tnt unless tan b - m crmt f 1 w1x. In (
contrast to the SM top-quark events, this signature violates lepton universality, a
criterion which is routinely applied in ongoing H "-boson searches at the Fermilab pp
™ ™
collider Tevatron w2,3x. For larger values of m H , the most copious sources of H "
™
bosons are provided by gb tHq w4–7x, gg tbHq w8x, and qb q X bHq w9x. The
preferred decay channel is then Hq tb, independently of tan b w1x. An interesting
™
alternative is to produce H " bosons in association with W . bosons, so that the
leptonic decays of the latter may serve as a trigger for the H "-boson search. The
™
dominant subprocesses of W " H . associated production are bb W " H . at the tree
level and gg W "H . at one loop, which were investigated for m b s 0 and small
™
values of tan b Ž0.3 ( tan b ( 2.3. in Ref. w10x and recently, without these restrictions, in
Refs. w11,12x. A careful signal-versus-background analysis, based on the analytic results
of Ref. w11x, was recently reported in Ref. w13x.
™
In this paper, we investigate HqHy pair production in the MSSM. At the tree level,
this proceeds via qq annihilation, qq Hq Hy, where q s u,d, s,c,b. The Drell–Yan
process, where a photon and a Z-boson are exchanged in the s channel Žsee upper
Feynman diagram in Fig. 1. has been studied by a number of authors w14–16x. As
pointed out in Ref. w11x, in the case q s b, there are additional Feynman diagrams
involving the h 0 and H 0 bosons in the s channel and the top quark in the t channel Žsee
middle and lower diagrams in Fig. 1.. As we shall see later on, for small or large values
of tan b , their contribution greatly exceeds the one due to the Drell–Yan process, which
is independent of tan b . To our knowledge, these additional diagrams have not been
2
Here and in the following, the charge-conjugate processes will not be explicitly mentioned.
´ B.A. Kniehl r Nuclear Physics B 568 (2000) 305–318
A.A. Barrientos Bendezu, 307
™
considered elsewhere in the literature. At one loop, HqHy pair production receives an
additional contribution from gg fusion, gg HqHy. Although the cross section of gg
fusion is suppressed by two powers of a s relative to the one of qq annihilation, it is
expected to yield a comparable contribution at multi-TeV hadron colliders, due to the
overwhelming gluon luminosity. In the 2HDM, gg fusion is mediated by heavy-quark
loops Žsee upper two rows in Fig. 2. w17–19x. We calculated these QCD contributions
and found full agreement with the analytical and numerical results presented in Ref.
w19x. In the MSSM, there are additional QCD contributions induced by squark loops Žsee
lower three rows in Fig. 2., which we shall present here.
As for bb annihilation, it should be noted that the treatment of bottom as an active
™ ™
flavour inside the colliding hadrons leads to an effective description, which comprises
™
contributions from the higher-order subprocesses gb Hq Hy b, gb Hq Hy b, and
™
gg Hq Hy bb. If all these subprocesses are to be explicitly included along with
bb Hq Hy, then it is necessary to employ a judiciously subtracted parton density
™
function ŽPDF. for the bottom quark in order to avoid double counting w4–7,20,21x. The
evaluation of bb Hq Hy with an unsubtracted bottom PDF is expected to slightly
overestimate the true cross section w4–7,20,21x. For simplicity, we shall nevertheless
adopt this effective approach in our analysis, keeping in mind that a QCD-correction
factor below unity is to be applied.
The circumstance that the spectrum of states is more than doubled if one passes from
the SM to the MSSM gives rise to a proliferation of parameters, which weakens the
predictability of the theory. A canonical method to reduce the number of parameters is
to embed the MSSM into a grand unified theory ŽGUT., e.g., a suitable supergravity
ŽSUGRA. model, in such a way that it is recovered in the low-energy limit. The MSSM
Fig. 2. One-loop Feynman diagrams for gg ™Hq

Hy in the MSSM.
thus constrained is described by the following parameters at the GUT scale, which come
in addition to tan b and m A : the universal scalar mass, m 0 ; the universal gaugino mass,
m1r2 ; the trilinear Higgs-sfermion coupling, A; the bilinear Higgs coupling, B; and the
Higgs-higgsino mass parameter, m. Notice that m A is then not an independent parameter
anymore, but it is fixed through the renormalization group evolution. The number of
parameters can be further reduced by making additional assumptions. Unification of the
t-lepton and b-quark Yukawa couplings at the GUT scale leads to a correlation between
m t and tan b . Furthermore, if the electroweak symmetry is broken radiatively, then B
and m are determined up to the sign of m. Finally, it turns out that the MSSM
parameters are nearly independent of the value of A, as long as < A < Q 500 GeV at the
GUT scale. Further details on the SUGRA-inspired MSSM scenario may be found in
Ref. w22x and the references cited therein.
™
This paper is organized as follows. In Section 2, we shall list analytic results for the
™
tree-level cross section of qq Hq Hy, including the Yukawa-enhanced contributions
for q s b, and the squark loop contribution to the gg HqHy amplitude in the MSSM.
The relevant Higgs-squark coupling constants and the squark loop form factors are
™
relegated to Appendices A and B, respectively. In Section 3, we shall present quantita-
tive predictions for the inclusive cross section of pp HqHyq X at the LHC adopting
the SUGRA-inspired MSSM. Section 4 contains our conclusions.
2. Analytic results
We start by defining the kinematics of the inclusive reaction AB HqHyq X, ™

where A and B are colliding hadrons, which are taken to be massless. Let 'S be the
energy of the initial state and y and p T the rapidity and transverse momentum of the
Hq boson in the centre-of-mass Žc.m.. system of the collision. By four-momentum
(
conservation, mT cosh y ( 'S r2, where mT s m2H q p T2 is the transverse mass of the
H " bosons, with mass m H . The hadron A is characterized by its PDF’s Fa r AŽ x a , M a .,
where x a is the fraction of the four-momentum of A which is carried by the Žmassless.
parton a Ž pa s x a pA ., M a is the factorization scale, and similarly for B. The Mandel-
stam variables s s Ž pa q p b . 2 , t s Ž pa y pH q . 2 , and u s Ž p b y pH q . 2 at the parton level
are thus related to S, y, and p T by s s x a x b S, t s m2H y x a'S mT expŽyy ., and
™
u s m 2H y x b'S mT expŽ y ., respectively. Notice that sp T2 s tu y m 4H . In the parton model,
the differential cross section of AB HqHyq X is given by
d 2s
dy dpT2
Ž AB ™H H qX .
q y
sÝ
a, b
1
Hx dx
a
a Fa r A Ž x a , M a . Fb r B Ž x b , Mb .
xb s
m2H y t
ds
dt
Ž ab ™H H
q y
., Ž 1.
where x a s m T expŽ y .rw'S y m T expŽyy .x and x b s x a m T expŽyy .rw x a 'S y

mT expŽ y .x. The parton-level cross section is calculated from the ab Hq Hy transi- ™
tion-matrix element T as d srdt s < T < 2rŽ16p s 2 ., where the average is over the spin
and colour degrees of freedom of the partons a and b.
We now turn to the specific subprocesses ab HqHy. We work in the MSSM,™

adopting the Feynman rules from Refs. w23–26x. For convenience, we introduce the
short-hand notations sw s sin uw , c w s cos uw , sa s sin a , ca s cos a , sb s sin b , cb s
cos b , s2 b s sinŽ2 b ., c2 b s cosŽ2 b ., s "s sinŽ a " b ., and c "s cosŽ a " b ., where uw
is the electroweak mixing angle, a is the mixing angle that rotates the weak CP-even
Higgs eigenstates into the mass eigenstates h0 and H 0 , and tan b s Õ 2rÕ1 is the ratio of
the vacuum expectation values of the two Higgs doublets. We neglect the Yukawa
couplings of the first- and second-generation quarks. We treat the b and b quarks as
active partons inside the colliding hadrons A and B. This should be a useful picture at
the energies of interest here, 'S ) 2 m H . For consistency with the underlying infinite-
momentum frame, we neglect the bottom-quark mass, m b . However, we must not
suppress terms proportional to m b in the Yukawa couplings, since they generally
dominate the related m t-dependent terms if tan b is large enough, typically for tan b
R m trm b f 6. This is obvious for the Hy bt vertex, which has the Feynman rule
(
w23–26x
i2y1 r4 GF1r2 m t cot b Ž 1 q g 5 . q m b tan b Ž 1 y g 5 . , Ž 2.
where GF is Fermi’s constant and we have neglected the Cabibbo-Kobayashi–Maskawa
mixing, i.e. Vt b s 1.
™
The tree-level diagrams for bb Hq Hy in the MSSM are depicted in Fig. 1. The
resulting parton-level cross section reads
ds
dt
Ž bb ™H H q y
.s
GF2
3p s
< S < 2 q 4 p T2 Ž < V q Tq < 2 q < A q Ty < 2 . , Ž 3.
where
m b m 2t
S s g H q H y h 0 g h 0 b b Ph 0 Ž s . q g H q H y H 0 g H 0 b b PH 0 Ž s . y 2
,
2 mW Ž m2t y t .
V s g H q H y Z ÕZ b b PZ Ž s . q g H q H y g Õg b b Pg Ž s . ,
A s g H q H y Z aZ b b PZ Ž s . ,
m2t cot 2b " m2b tan2b
T "s 2
, Ž 4.
8 mW Ž m2t y t .
with couplings
m Z sq c 2 b m b sa
g H q H y h 0 s mW syy , g h0 b b s ,
2 cw 2 mW cb
m Z cq c2 b m b ca
g H q H y H 0 s ymW cyq , gH 0 bb s y ,
2 cw 2 mW cb
c w2 y sw2 Ib y 2 sw2 Q b Ib
g Hq Hy Z s y , ÕZ b b s y , aZ b b s y ,
2 cw 2 cw 2 cw
g H q H y g s ysw , Õg b b s ysw Q b , Ž 5.
weak isospin Ib s y1r2, and electric charge Q b s y1r3. Here,

1
PX Ž s . s Ž 6.
s y m2X q im X GX
™
recover the well-known Drell–Yan cross section of qq Hq Hy w14–16x from Eq. Ž3.
by putting S s T "s 0 and substituting b q. The approximation S s T "s 0 is justi-
™
is the propagator function of particle X, with mass m X and total decay width G X . We
fied for the quarks of the first and second generations, q s u,d, s,c, because S and T "
are then suppressed by the smallness of the corresponding Yukawa couplings. However,
the S and T " terms give rise to sizeable contributions in the case of q s b, especially at
small or large values of tan b . The full cross section of qq annihilation is obtained by
complementing the bb-initiated cross section of Eq. Ž3. with the Drell–Yan cross
sections for q s u,d, s,c.
™
The one-loop diagrams for gg HqHy in the MSSM are displayed in Fig. 2. As for
the quark loops, our analytical results fully agree with those listed in Ref. w19x, and there
is no need to repeat them here. In the squark case, the T-matrix elements corresponding
to the triangle and box diagrams are found to be
2
GF m W a s Ž mr .
T˜^ s ´mc Ž pa . ńc Ž p b . A1mn F˜^ ,
'2 p
2
GF m W a s Ž mr .
T˜I s ´mc Ž pa . ńc Ž p b . A1mn F˜I q A 2mn G˜ I ,
ž / Ž 7.
'2 p
respectively, where a s Ž m r . is the strong coupling constant at renormalization scale m r ,
´mc Ž pa . is the polarization four-vector of gluon a and similarly for gluon b, it is summed
over the colour index c s 1, . . . ,8,
2
A1mn s g mn y pan p bm ,
s
2 m 2H u y m 2H t y m2H
A 2mn s g mn q
p T2 ž s
pan p bm q
s
pan pHmq q
s
p bm pHn q q pHmq pHn q ,
/
Ž 8.
and the form factors F˜^ , F˜I , and G˜ I are listed in Appendix B. Due to Bose
l
symmetry, T˜^ and T˜I are invariant under the simultaneous replacements m n and l
™
pa p b . Consequently, F˜^ , F˜I , and G˜ I are symmetric in t and u.
The parton-level cross section of bb Hq Hy including both quark and squark
contributions is then given by
™H H
2
ds GF2 a s2 Ž m r . 2
2 mW
dt
Ž gg q y
.s
256 Ž 2p .
3 Ý Q
C^ Q
F^ q FI y
s
ž F˜ ˜
^ q FI /
Qst , b
2 2
2 mW 2
q GI y G˜ I q H I , Ž 9.
s
Q Q
where the generalized coupling C ^ and the form factors F^ , FI , G I , and H I may
be found in Eq. Ž8. and Appendix A of Ref. w19x, respectively.
3. Phenomenological implications
We are now in a position to explore the phenomenological implications of our results.

The SM input parameters for our numerical analysis are taken to be GF s 1.16639 = 10y5
GeVy2 w27x, mW s 80.394 GeV, m Z s 91.1867 GeV, m t s 174.3 GeV w28x, and m b s 4.7
GeV. We adopt the lowest-order set CTEQ5L w29x of proton PDF’s. We evaluate
a s Ž m r . from the lowest-order formula w27x with n f s 5 quark flavours and asymptotic
scale parameter LŽ5. w x
QC D s 146 MeV 29 . We identify the renormalization and factoriza-
tion scales with the H H invariant mass, m2r s M a2 s Mb2 s s. For our purposes, it is
q y
useful to replace m A by m H , the mass of the H " bosons to be produced, in the set of
MSSM input parameters. We vary tan b and m H in the ranges 1 - tan b - 40 f m trm b
and 120 GeV - m H - 550 GeV, respectively. As for the GUT parameters, we choose
™
Fig. 3. Total cross sections s Žin fb. of pp Hq Hy q X via qq annihilation Ždashed line. and gg fusion
Žsolid line. at the LHC as functions of tan b for m H s 200 GeV. For comparison, also the Drell–Yan
contribution to qq annihilation Ždotted line. and the quark loop contribution to gg fusion Ždot-dashed line. are
shown.
m1r2 s 100 GeV, A s 0, and m - 0, and tune m 0 so as to be consistent with the desired
value of m H . All other MSSM parameters are then determined according to the
SUGRA-inspired scenario as implemented in the program package SUSPECT w30x. We
do not impose the unification of the t-lepton and b-quark Yukawa couplings at the GUT
scale, which would just constrain the allowed tan b range without any visible effect on
the results for these values of tan b .
™
We now study the total cross section of pp HqHyq X at the LHC, with c.m.
energy 'S s 14 TeV. In Fig. 3 the full contributions due to qq annihilation Ždashed
line. and gg fusion Žsolid line. are displayed as functions of tan b for m H s 200 GeV.
For comparison, also the Drell–Yan contributions to qq annihilation for q s u,d, s,c,b
Ždotted line. and the quark loop contribution to gg fusion Ždot-dashed line., which is the
full one-loop result for gg-fusion in the 2HDM, are shown. In the case of qq
annihilation, as anticipated in Section 1, the Yukawa-enhanced contribution for q s b
greatly enhances the conventional Drell–Yan cross section for large values of tan b , by
more than a factor of three for tan b s 40. The expected enhancement for small tan b is
invisible, since solutions with tan b Q 2 are excluded in the SUGRA-inspired MSSM for
our choice of input parameters. As for gg fusion, the dot-dashed line nicely agrees with
Fig. 6 of Ref. w19x. The quark loop contribution exhibits a minimum at tan b
™
Fig. 4. Total cross sections s Žin fb. of pp Hq Hy q X via qq annihilation Ždashed lines. and gg fusion
Žsolid lines. at the LHC as functions of m H for tan b s1.5, 6, and 30.
(
f m trm b f 6. This may be understood by observing that the average strength of the
(
Hy bt coupling in Eq. Ž2., proportional to m 2t cot 2b q m 2b tan2b , is then minimal w19x.
Passing from the 2HDM to the MSSM, we need to coherently add the squark loop
contribution according to Eq. Ž9.. We observe that this leads to a significant rise in cross
section, by up to 50%, unless tan b is close to 10. Nevertheless, the full tree-level cross
section is dominant for all values of tan b .
In Fig. 4 the m H dependence of the full qq-annihilation Ždashed lines. and gg-fusion
cross sections Žsolid lines. is studied for tan b s 1.5, 6, and 30. As we have already seen
in Fig. 3, qq annihilation always dominates. Its contribution modestly exceeds the one
due to gg fusion, by a factor of three or less, if tan b f 1.5 or 30 and m H R 200 GeV,
but it is more than one order of magnitude larger if m H Q m t . The gg-fusion contribu-
tion is greatly suppressed if tan b f 6, independently of m H . For all values of tan b , the
latter exhibits a dip located about m H s m t , which arises from resonating top-quark
propagators in the quark box form factors. ŽIn the case of tan b s 1.5, this dip lies in the
excluded m H range.. This feature may also be seen in Fig. 5 of Ref. w19x, where the
quark loop contribution is shown separately. We note in passing that we also find good
agreement with that figure.
For a comparison with future experimental data, the qq-annihilation and gg-fusion
™
channels should be combined. From Fig. 4 we read off that the total cross section of
pp Hq Hyq X at the LHC is predicted to be 180 fb Ž0.9 fb. in the considered MSSM
scenario if tan b s 30 and m H s 120 GeV Ž500 GeV.. If we assume the integrated
luminosity per year to be at its design value of L s 100 fby1 for each of the two LHC
experiments, ATLAS and CMS, then this translates into about 36 000 Ž180. signal
events per year.
4. Conclusions
We studied the hadroproduction of HqHy pairs within the MSSM, adopting a
SUGRA-inspired scenario. We included the contributions from qq annihilation and gg
fusion to lowest order and provided full analytic results. Our analysis reaches beyond
previous studies w14–19x in two important respects. In the case of qq annihilation, we
demonstrated that previously neglected Yukawa-type contributions in the bb channel
lead to a substantial increase in cross section if tan b is large, by more than a factor of
three for tan b s 40. In the case of gg fusion, we upgraded a previous result w19x, which
we confirmed, from the 2HDM to the MSSM by including the contributions induced by
virtual squarks. As a result, the gg-fusion cross section may be significantly enhanced,
by up to 50%, depending on tan b . Should the MSSM be realized in nature, then HqHy
pair production will provide a copious source of charged Higgs bosons at the LHC, with
an annual yield of up to 36 000 signal events, which amounts to 72 000 charged Higgs
bosons per year.
5. Note added in proof
™
After this paper was submitted, a preprint w31x appeared, in which the cross section of
pp HqHyq X via gluon fusion was studied for the LHC. The analytic results for the
squark loop contributions presented there agree with our results.
Acknowledgements
B.A.K. thanks the Theory Group of the Werner-Heisenberg-Institut for the hospitality
extended to him during a visit when this paper was finalized. The work of A.A.B.B. was
¨
supported by the Friedrich-Ebert-Stiftung through Grant No. 219747. The II. Institut fur
Theoretische Physik is supported by the Bundesministerium fur ¨ Bildung und Forschung
under Contract No. 05 HT9GUA 3, and by the European Commission through the
Research Training Network Quantum Chromodynamics and the Deep Structure of
Elementary Particles under Contract No. ERBFMRXCT980194.
Appendix A. Higgs-squark couplings
In this appendix, we collect the couplings of the h 0 , H 0 , and H " bosons to the
squarks q˜i , with q s t,b and i s 1,2, which are relevant for our analysis. Defining the
mixing matrix which rotates the left- and right-handed squark fields, q˜L and q˜R , into the
mass eigenstates q˜i as
cos uq˜ sin uq˜

M q̃ s
ž ysin uq˜ cos uq˜ / , Ž A.1 .
we have w23–26x
g h 0 t˜1 t˜1 g h 0 t˜1 t˜2

ž g h 0 t˜2 t˜1 g h 0 t˜2 t˜2 /
m Z sq Ž It3 y sw2 Q t . m 2t ca m t Ž m sa q A t ca .
y y
s M t˜
y
cw
m t Ž m sa q A t ca .
2 mW sb
mW sb 2 mW sb
m Z sq sw2 Q t
cw
y
m2t ca
mW sb
0 T
Ž M t˜ . ,
g h 0 b˜ 1 b˜ 1 g h 0 b˜ 1 b˜ 2
ž g h 0 b˜ 2 b˜ 1 g h 0 b˜ 2 b˜ 2 /
m Z sq Ž Ib3 y sw2 Q b . m2b sa m b Ž m ca q A b sa .
q
sM b
˜
cw
m b Ž m ca q A b sa .
2 mW cb
mW cb 2 mW cb
m Z sq sw2 Q b
cw
q
m 2b sa
mW cb
0 T
Ž M b˜ . ,
g H 0 t˜1 t˜1 g H 0 t˜1 t˜2

ž g H 0 t˜2 t˜1 g H 0 t˜2 t˜2 /
m Z cq Ž It3 y sw2 Q t . m2t sa m t Ž m ca y A t sa .
y y
s M t˜
cw
m t Ž m ca y A t sa .
2 mW sb
mW sb
y
2 mW sb
m Z cq sw2 Q t
cw
y
m 2t sa
mW sb
0 Ž M t˜ . ,
T
g H 0 b˜ 1 b˜ 1 g H 0 b˜ 1 b˜ 2
ž g H 0 b˜ 2 b˜ 1 g H 0 b˜ 2 b˜ 2 /
m Z cq Ž Ib3 y sw2 Q b . m2b ca m b Ž m sa y A b ca .
y y
sM b
˜
cw
m b Ž m sa y A b ca .
2 mW cb
mW cb
y
2 mW cb
m Z cq sw2 Q b
cw
y
m2b ca
mW cb
0 Ž M b˜ . ,
T
g H q t˜1 b˜ 1 g H q t˜1 b˜ 2
ž g H q t˜2 b˜ 1 g H q t˜2 b˜ 2 /
2
ymW s2 b q m2t cot b q m2b tan b m b Ž m q A b tan b .
s M t̃
'2 mW
m t Ž m q A t cot b .
'2 mW
'2 mW
m t m b Ž tan b q cot b .
'2 mW 0 Ž M b˜ . ,
T
g H q H y t˜1 t˜1 g H q H y t˜1 t˜2

ž g H q H y t˜2 t˜1 g H q H y t˜2 t˜2 /
c 2 b It3 Ž 1 y 2 c w2 . y sw2 Q t m 2b tan2b
y 0
sM t̃

= Ž M t̃ . ,
T
2 c w2
0
2
2 mW
c 2 b sw2 Q t
2 c w2
y
m2t cot 2b
2
2 mW
0
g H q H y b˜ 1 b˜ 1 g H q H y b˜ 1 b˜ 2
ž g H q H y b˜ 2 b˜ 1 g H q H y b˜ 2 b˜ 2 /
c 2 b Ib3 Ž 1 y 2 c w2 . y sw2 Q b m2t cot 2b
y 0
sM b̃

= Ž M b̃ . .
T
2 c w2
0
2
2 mW
c 2 b sw2 Q b
2 c w2
y
m2b tan2b
2
2 mW
0
Ž A.2 .
The corresponding Feynman rules emerge for the trilinear Žquartic. couplings by
multiplication with ig Ž ig 2 ., where g is the SUŽ2. coupling constant. Similar relations
apply for the squarks of the first and second generations, which are also included in our
analysis. However, in these cases, we neglect terms which are suppressed by the
smallness of the corresponding light-quark masses.
Appendix B. Squark loop form factors
In this appendix, we express the squark triangle and box form factors, F˜^ , F˜I , and
G̃ I , in terms of the standard scalar three- and four-point functions,
2
C0 Ž p 12 , Ž p 2 y p 1 . , p 22 ,m20 ,m12 ,m22 .
d4q 1
s H ip Ž q 2 2 2 2
,
y m20 q i e . Ž q q p1 . y m12 q i e Ž q q p 2 . y m22 q i e
D 0 Ž p12 , Ž p 2 y p1 . 2 , Ž p 3 y p 2 . 2 , p 32 , p 22 , Ž p 3 y p1 . 2 ,m20 ,m12 ,m22 ,m23 .
d4q 1
s H ip 2 ,
Ž q 2 y m20 q i e . w Ž q q p1 . 2y m12 q i e xw Ž q q p 2 . 2y m22 q i e xw Ž q q p 3 . 2y m23 q i e x
Ž B.1 .
which we evaluate numerically with the aid of the program package FF w32x. We have
F˜^ s g H q H y h 0 Ph 0 Ž s . g h 0 t˜1 t˜1 F1 Ž s,m t˜1 . q g h 0 t˜2 t˜2 F1 Ž s,m t˜2 . q g h 0 b˜ 1 b˜ 1 F1 Ž s,m b˜ 1 .
qg h 0 b˜ 2 b˜ 2 F1 Ž s,m b˜ 2 . q g H q H y H 0 PH 0 Ž s . g H 0 t˜1 t˜1 F1 Ž s,m t˜1 .
qg H 0 t˜2 t˜2 F1 Ž s,m t˜2 . q g H 0 b˜ 1 b˜ 1 F1 Ž s,m b˜ 1 . q g H 0 b˜ 2 b˜ 2 F1 Ž s,m b˜ 2 .
y g H q H y t˜1 t˜1 F1 Ž s,m t˜1 . y g H q H y t˜2 t˜2 F1 Ž s,m t˜2 . y g H q H y b˜ 1 b˜ 1 F1 Ž s,m b˜ 1 .
y g H q H y b˜ 2 b˜ 2 F1 Ž s,m b˜ 2 . ,
lm .
F˜I s g H2 q t˜1 b˜ 1 F2 Ž h, s,t ,u,m t˜1 ,m b˜ 1 . q Ž m t˜1 b˜ 1
qg 2
F Ž h, s,t ,u,m ,m . q Ž m l m
H q t˜1 b˜ 2 2 t˜1 b˜ 2 t˜1 b˜ 2 .
qg 2
F Ž h, s,t ,u,m ,m . q Ž m l m
H q t˜2 b˜ 1 2 t˜2 b˜ 1 t˜2 b˜ 1 . q g H2 q t˜2 b˜ 2
= F Ž h, s,t ,u,m ,m . q Ž m l m . ,
t˜2 b˜ 2 t˜2 b˜ 2 Ž B.2 .
™ F . Here, we have
2
where h s m2H . G˜ I is obtained from F˜I by substituting F2 3

introduced the auxiliary functions
F1 Ž s,m q˜ . s 2 q 4 m2q˜ C0 0,0, s,m 2q˜ ,m2q˜ ,m2q˜ ,

ž /
2
F2 Ž h, s,t ,u,m t˜ ,m b˜ . s y
s
Ž t1C03 q u1C05 . q pT2 D10 q 2 m2t˜ Ž D10 q D 03 q D 05 . ,
1
F3 Ž h, s,t ,u,m t˜ ,m b˜ . s
sp T2
s y Ž t q u . C02 q t 2 D03 q u 2 D05 y 2 tt 1C04
y2 uu1C06 q Ž t 2 q u 2 y 2 h 2 . C07 q 2 m2t˜ s Ž C01 y C02
qpT2 D 10 y tD 02 y uD 04 . y t 12 D 03 y u12 D 05 q sm2t˜ Ž m 2t˜ y m2b˜ .
= Ž 2 D 10 q D 02 q D 03 q D 04 q D 05 . 4 , Ž B.3 .
where t 1 s t y h, u1 s u y h, and
C01 s C0 Ž 0,0, s,m 2b˜ ,m2b˜ ,m2b˜ . ,
C02 s C0 Ž 0,0, s,m 2t˜ ,m2t˜ ,m2t˜ . ,
C03 s C0 Ž h,0,t ,m2b˜ ,m2t˜ ,m2t˜ . ,
C04 s C0 Ž h,0,t ,m2t˜ ,m2b˜ ,m 2b˜ . ,
C05 s C0 Ž h,0,u,m2b˜ ,m2t˜ ,m 2t˜ . ,
C06 s C0 Ž h,0,u,m 2t˜ ,m2b˜ ,m 2b˜ . ,
C07 s C0 Ž h,h, s,m 2t˜ ,m2b˜ ,m2t˜ . ,
D 10 s D 0 Ž h,0,h,0,t ,u,m2b˜ ,m 2t˜ ,m 2t˜ ,m2b˜ . ,
D 02 s D 0 Ž h,h,0,0, s,t ,m 2b˜ ,m 2t˜ ,m2b˜ ,m 2b˜ . ,
D 03 s D 0 Ž h,h,0,0, s,t ,m 2t˜ ,m2b˜ ,m2t˜ ,m 2t˜ . ,
D 04 s D 0 Ž h,h,0,0, s,u,m2b˜ ,m2t˜ ,m 2b˜ ,m2b˜ . ,
D 05 s D 0 Ž h,h,0,0, s,u,m2t˜ ,m2b˜ ,m 2t˜ ,m 2t˜ . . Ž B.4 .

References
w1x Z. Kunszt, F. Zwirner, Nucl. Phys. B 385 Ž1992. 3, and references cited therein.
w2x CDF Collaboration, F. Abe et al., Phys. Rev. D 54 Ž1996. 735.
w3x B. Bevensee Žfor the CDF and D0 Collaborations., Report No. FERMILAB–Conf–98r155–E ŽMay
1998., to be published in the proceedings of 33rd Rencontres de Moriond: QCD and High Energy
Hadronic Interactions, Les Arcs, France, 21–28 March 1998.
w4x J.F. Gunion, H.E. Haber, F.E. Paige, W.-K. Tung, S.S.D. Willenbrock, Nucl. Phys. B 294 Ž1987. 621.
w5x R.M. Barnett, H.E. Haber, D.E. Soper, Nucl. Phys. B 306 Ž1988. 697.
w6x F.I. Olness, W.-K. Tung, Nucl. Phys. B 308 Ž1988. 813.
w7x V. Barger, R.J.N. Phillips, D.P. Roy, Phys. Lett. B 324 Ž1994. 236.
w8x J.L. Diaz-Cruz, O.A. Sampayo, Phys. Rev. D 50 Ž1994. 6820.
w9x S. Moretti, K. Odagiri, Phys. Rev. D 55 Ž1997. 5627.
w10x D.A. Dicus, J.L. Hewett, C. Kao, T.G. Rizzo, Phys. Rev. D 40 Ž1989. 787.
w11x A.A. Barrientos Bendezu, ´ B.A. Kniehl, Phys. Rev. D 59 Ž1998. 015009.
w12x A.A. Barrientos Bendezu, ´ B.A. Kniehl, Preprint No. DESY 99-121, KEK-TH-638 and hep-phr9909502
ŽSeptember 1999., to appear in Phys. Rev. D.
w13x S. Moretti, K. Odagiri, Phys. Rev. D 59 Ž1999. 055008.
w14x E. Eichten, I. Hinchliffe, K. Lane, C. Quigg, Rev. Mod. Phys. 56 Ž1984. 579.
w15x E. Eichten, I. Hinchliffe, K. Lane, C. Quigg, Rev. Mod. Phys. 58 Ž1986. 1065, Erratum.
w16x N.G. Deshpande, X. Tata, D.A. Dicus, Phys. Rev. D 29 Ž1984. 1527.
w17x S.S.D. Willenbrock, Phys. Rev. D 35 Ž1987. 173.
w18x J. Yi, M. Wen-Gan, H. Liang, H. Meng, Y. Zeng-Hui, J. Phys. G 24 Ž1998. 83.
w19x A. Krause, T. Plehn, M. Spira, P.M. Zerwas, Nucl. Phys. B 519 Ž1998. 85.
w20x D.A. Dicus, S. Willenbrock, Phys. Rev. D 39 Ž1989. 751.
w21x D.A. Dicus, C. Kao, Phys. Rev. D 41 Ž1990. 832.
w22x A. Djouadi, J. Kalinowski, P. Ohmann, P.M. Zerwas, Z. Phys. C 74 Ž1997. 93.
w23x J.F. Gunion, H.E. Haber, Nucl. Phys. B 272 Ž1986. 1.
w24x J.F. Gunion, H.E. Haber, Nucl. Phys. B 402 Ž1993. 567, Erratum.
w25x J.F. Gunion, H.E. Haber, Nucl. Phys. B 278 Ž1986. 449.
w26x J.F. Gunion, H.E. Haber, G. Kane, S. Dawson, The Higgs Hunter’s Guide ŽAddison-Wesley, Redwood
City, 1990..
w27x C. Caso et al. ŽParticle Data Group., Eur. Phys. J. C 3 Ž1998. 1.
w28x D. Abbaneo et al. ŽThe LEP Collaborations ALEPH, DELPHI, L3, OPAL, the LEP Electroweak Working
Group and the SLD Heavy Flavour and Electroweak Groups., Report No. CERN-EPr99-15 ŽFebruary
1999..
w29x H.L. Lai et al., CTEQ Collaboration, Report No. MSU-HEP-903100 and hep-phr9903282 ŽMarch 1999..
w30x A. Djouadi, J.-L. Kneur, G. Moultaka, Report No. PMr98-27 and GDR-S-017 Ž1998..
w31x O. Brein, W. Hollik, Preprint No. KA-TP-11-1999 and hep-phr9908529 ŽAugust 1999., to appear in Eur.
Phys. J. C.
w32x G.J. van Oldenborgh, Comput. Phys. Commun. 66 Ž1991. 1.
Two-point functions at two loops in three flavour chiral

perturbation theory q
a,b
´
Gabriel Amoros , Johan Bijnens a , Pere Talavera a
a
¨
Department of Theoretical Physics 2, Lund UniÕersity, SolÕegatan 14A, S22362 Lund, Sweden
b
Department of Physics, P.O. Box 9, FIN-00014 UniÕersity of Helsinki, Helsinki, Finland
Abstract
The vector and axial-vector two-point functions are calculated to next-to-next-to-leading order
in Chiral Perturbation Theory for three light flavours. We also obtain expressions at the same
order for the masses, mp2 , m 2K and mh2 , and the decay constants, Fp , FK and Fh . We present some
numerical results after a simple resonance estimate of some of the new O Ž p 6 . constants. q 2000
PACS: 12.39.F; 14.40.Aq; 12.38.Lg
1. Introduction
With the new collider facilities, upcoming experiments will bring higher statistics
data samples into the low energy regime. Due to their accuracy higher order calculations
are needed to update the theoretical prediction for the measurements. In this frame the
calculation of the two-point functions at next-to-next-to-leading order ŽNNLO. at low
energies has become necessary. These provide us with the pseudoscalar masses and the
decay constants, which are needed input for most other quantities. The two-point Green
functions are also basic tools in the study of the strong interaction. They form the basis
for a series of very useful sum rules starting with the Weinberg w1x and DMO w2x sum
rules Žwe refer to Refs. w3,4x and references therein for a more complete discussion about
sum rules..
In this work we are concerned with the low energy regime of QCD. We will study
the two-point functions with Chiral Perturbation Theory ŽCHPT., valid for energies
below the first resonance Ž E ; mr . and describing the strong interactions using the
pseudoscalar octet as the basic fields. This is by now a fairly developed field. We refer
to Ref. w5x for reviews and various abstracts on recent works.
q
Work supported in part by TMR, EC-Contract No. ERBFMRX-CT980169 ŽEURODAFNE..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 7 4 - 4
320 ´ et al.r Nuclear Physics B 568 (2000) 319–363
G. Amoros
For a future study of various sum rules, we present the vector and axial-vector
two-point functions at NNLO in three flavour CHPT, in the limit of unbroken isospin.
Four of the six basic correlators have been calculated earlier w6,7x and we have fully
confirmed their results in the vector–vector two-point function and partially 2 in the
axial-vector case. The other two are new and complete the three flavour basis. As a
byproduct we also give the masses and decay constants to NNLO.
The interest in the NNLO calculation is beyond the precise measurement of the
couplings and masses of the effective theory, it allows to test the convergence of the
theory and provides a more stringent check on the principle of resonance saturation of
the constants in the low energy chiral Lagrangian. While this principle worked well at
O Ž p 4 . w8,9x only a few tests at O Ž p 6 . have been done. In this paper we estimate some
of the constants appearing to two loops and check their effect on the full O Ž p 6 . results
including the loop contributions. We use the MS-subtraction scheme and the recent
classification for the O Ž p 6 . Lagrangian w10x.
Some applications to chiral sum rules for the isospin and hypercharge cases can be
found in Refs. w11,12x. We intend to return to that subject in a future publication.
The paper is organized as follows. In Section 2 the two-point Green functions are
defined, followed by a short overview of CHPT in Section 3. The vector two-point
function is relatively easy since it only involves products of one-loop integrals. Its
calculation is described and results given in Section 4. The masses and decay constants
can be calculated in two ways. The masses can be obtained from the zero of the inverse
pseudoscalar propagator or from the pole in the axial-vector two-point function. The
decay constants can be directly determined from the residue of the poles in the
axial-vector two-point function or through the definition with the axial-vector–pseudo-
scalar function. We have checked that both methods agree and the first is described in
Section 5. Finally the axial-vector two-point functions are presented in Section 6. The
new O Ž p 6 . constants appearing are estimated in Section 7 on the basis of Resonance
Dominance. In Section 8 some results are presented, postponing a more detailed analysis
w13x. And finally in Section 9 we discuss our main results.
We refer the lengthiest expressions and the more technical discussion of the loop
integrals and renormalization to appendices. In Appendix A.1 and Appendix A.2 we
give the full expressions for the masses and the decay constants. In Appendix B we
display the the axial-vector two-point function components. The loop integrals are
collected in Appendix C.
2. Definition of two-point functions
We calculate in CHPT the two-point functions of vector and axial-vector currents.

The quark currents are defined by
Vmi j Ž x . s q igm q j and Ami j Ž x . s q igm g 5 q j , Ž 1.
2
The authors of Refs. w6,7x use a different method to perform the sunset integrals making comparison of
those parts difficult. We agree on all the parts we could check without converting their sunset functions to
ours.
´ et al.r Nuclear Physics B 568 (2000) 319–363
G. Amoros 321
where the indices i and j run over the three light quark flavours, u, d and s. Working in
the isospin limit all SUŽ3. currents can be constructed using isospin relations from
1
Vmp Ž x . s 11 22
'2 Ž Vm Ž x . y Vm Ž x . . ,
1
Vmh Ž x . s 11 22 33
'6 Ž Vm Ž x . q Vm Ž x . y 2Vm Ž x . . ,
VmK Ž x . s Vm31 Ž x . . Ž 2.
We refer to these as isospin, hypercharge and kaon respectively. E.g. the electromag-
netic current corresponds to
e e
Vmem s Vmp Ž x . q h
'2 '6 Vm Ž x . . Ž 3.
We calculate the two-point functions defined as

†
PmnV a Ž q . ' i d 4 x e i qP x ²0 < T Ž Vma Ž x . Vna Ž 0 . . <0: ,
H Ž 4.
for a s p ,h , K. All other vector two-point functions can be constructed from these using
isospin relations. Lorentz-invariance allows to express them in a transverse, P Ž1., and a
longitudinal, P Ž0., part
PmnV a s Ž qm qn y q 2 gmn . P VŽ1. 2 2
a Ž q . q qm qn P V a Ž q . .
Ž0.
Ž 5.
Similar definitions and comments apply for the axial-vector currents.
The currents obey Ward identities and other symmetry relations; for m u s m d the
vector currents are conserved,
2 2
P VŽ0.
p Ž q . s P Vh Ž q . s 0 .
Ž0.
Ž 6.
In the SUŽ3. V limit, the three vector two-point functions reduce to the same expression.
The same holds for the axial-vector currents. In the addition the last ones are conserved
only when the relevant quark masses vanish.
3. Chiral perturbation theory
Effective theories are a general tool in understanding a wide range of physical

processes, from high energy physics to superconductivity. In that frame Chiral Perturba-
tion Theory is a successful theory describing the strong interaction at low energy. It is
based on the existence of a mass-gap in the hadronic spectrum, at low energies only the
low mass states can be excited. Those are the Goldstone boson particles and are the only
states that are actually predicted from first principles in QCD w14x. For the present status
of the field we refer to the listing of review articles and the abstracts in Ref. w5x and to
some recent lectures w15,16x.
For constructing the effective action, the high energy states of the theory should be
integrated out, thus the Lagrangian describing processes at low energy consists of a
series of operators involving only Goldstone boson particles. These general operators
G. Amoros
should share the same symmetries as the basic underlying theory w17,18x, in particular
Lorentz invariance, local chiral symmetry, parity and charge conjugation. Thereby in the
following we use that the generating functional of both theories, QCD and CHPT, are
the same at low energies w18x
½H
exp iZ 4 s Dq Dq D Gm exp i d 4 x LQCD s DUexp i d 4 x Leff .
H 5 H ½H 5 Ž 7.
Following this philosophy, the QCD effective Lagrangian is given by a non-linear

realization of chiral symmetry Žsee Ref. w19x and references therein.. The lowest order in
an expansion by quark masses and external momenta is
F02
L2 s ² Dm UD m U † q U †x q x †U : , Ž 8.
4
where
2
U Ž f . s u Ž f . s exp Ž i'2 FrF0 . , Ž 9.
parametrizes the pseudo-Goldstone bosons and
x s 2 B0 Ž s q ip . Ž 10 .
is given in terms of the scalar and pseudoscalar sources s and p. ² X : s trflavour Ž X . and
U, s, p are matrices in flavour space. Both B0 and F0 are constants not restricted by
symmetry. They are related with the quark condensate and the meson decay constant
respectively. To respect local invariance the external sources are incorporated through
the covariant derivatives
DmU s Em U y irm U q iUlm , Ž 11 .
and the field strength tensors
FLmn s E m l n y E n l m y i w l m ,l n x , FRmn s E m r n y E n r m y i w r m ,r n x . Ž 12 .
As mentioned above the purpose of this paper is to compute two-point Green
functions, the most straightforward way is to incorporate classical sources in the
effective action. This reduces the calculation of any n-point Green function to the
evaluation of functional derivatives acting on the generating functional
dn
G Ž n. Ž x 1 , . . . , x n . s Zw J x . Ž 13 .
d j Ž x 1 . PPP d j Ž x n . Js0
This allows for instance, to relate the chiral condensate with the constant B0 by taking
the derivative with respect to the scalar sources. This formalism is not only suitable for
an easy calculation but also allows to incorporate the electromagnetic, weak interactions
and the symmetry breaking through the quark masses via the following identifications:
rm s Õm q am , lm s Õm y am , ssMq . . . , Ž 14.
where M stands for the diagonal quark mass matrix, M s diag Ž m u ,m d ,m s . .
G. Amoros 323
To get the desired chiral order – O Ž p 6 . – in our calculation we will deal with three
kind of contributions: Ži. tree, one- and two-loop diagrams involving vertices of L 2 , Žii.
tree and 1-loop graphs with vertices from L 2 and from L4 given by
L4 s L1² =m U † = m U :2 q L2 ² =m U † =n U :² = mU † =n U : q L3 ² =m U † = m U =n U † =n U :
q L 4 ² =m U † = m U :² x †U q x U † : q L5 ² =m U † = m U Ž x †U q x U † . :
q L6 ² x †U q x U † :2 q L 7 ² x †U y x U † :2 q L8 ² x †Ux †U q x U †x U † :
y iL9 ² FRmn =m U =n U † q FLmn =m U † =n U : q L10 ²U †FRmn UFL mn :
q H1² FRmn FR mn q FLmn FL mn : q H2 ² x †x : Ž 15 .

and finally Žiii. the tree graphs of L6 . The latter was first classified in Ref. w20x and
recently in Ref. w10x a more restrictive general set was found. We borrow in the
following our notation from the last reference. Notice that the terms H1 and H2 in Eq.
Ž15. have no direct physical meaning, their value depends on the precise way in which
the currents are defined in QCD. But once a consistent definition of a QCD current has
been given they are defined unambiguously.
For later use we define the following quantities:
Õ mn s Ž FRmn q FLmn . r2 ,
a mn s Ž FRmn y FLmn . r2 ,
um s i u† Ž Em y irm . u y u Ž Em y ilm . u† 4 ,
Gm s 12 u† Ž Em y irm . u q u Ž Em y ilm . u† 4 ,
xqs u†x u† q u x † u ,
=m X s Em X q Gm X y X Gm . Ž 16 .
4. The vector two-point functions
Within the framework of previous sections we can start to calculate the vector–vector
two-point functions. The first contribution appears at O Ž p 4 .. It has been calculated for
the isospin case in Ref. w21x in SUŽ2. CHPT. The extension to O Ž p 6 . in SUŽ3. CHPT
has been done in Ref. w6x for the isospin and the hypercharge case. We have reproduced
their results and in addition we present the kaon vector two-point function as well here.
The calculation to O Ž p 6 . presents no new difficulties besides being rather tedious
since only products of one-loop integrals appear.
The O Ž p 4 . contributions come from diagrams Ža. – Žc. in Fig. 1. Diagrams Žd., Že.,
Žm. and Žn. can be calculated directly or using wave-function renormalization and mass
corrections. We have checked that both approaches give the same result. As a conse-
G. Amoros
Fig. 1. Diagrams for the vector two-point function. The crossed circle stands for the O Ž p 4 . vertex insertion.
Wiggly lines are the external vector currents. Dots are O Ž p 2 . vertices and a square is a O Ž p 6 . vertex. The
solid lines are meson propagators.
quence the final result contains no three-propagator integrals. They cancel when the
O Ž p 4 . result is expressed in terms of the physical masses using
E
f Ž m2i 0 ;q 2 . s f Ž m 2i ;q 2 . q Ý Ž m2j0 y m 2j . f Ž m2i 0 ;q 2 . , Ž 17 .
j E m2j 0 m 2i 0 sm 2i
where the m2i 0 are the bare masses and the m 2i the next-to-leading order masses. In
addition we replace F0 by Fp and all masses by their physical ones in the O Ž p 6 .
expression.
There are no one-particle reducible contributions to the vector two-point functions.
We have performed the following checks:
™
1. In the isospin and hypercharge case the longitudinal part vanishes.
2. In the SUŽ3. limit, i.e. mp2 s m 2K s mh2 m2chiral , all two-point functions are equal.
3. The SUŽ3. breaking effect in the form-factors appears only in second order in the
quark masses, i.e. order Ž m s y m ˆ . 2 , as required by the Ademollo–Gatto theorem
w22,23x.
4. All divergences with a non-analytical dependence on masses or q 2 cancel and the
lnŽ4p . and g terms can be absorbed in the counter-terms as well. Both of these
follow from general renormalizability theorems.
G. Amoros 325
5. The remaining divergences are in agreement with those from the general calculation
using heat-kernel methods w24x and with the double logarithms of Ref. w25x.
The result can be expressed simply in terms of the finite functions

m 2i m2i
mi s ln ,
32p 2 m2
1 l0 q2
BVi j Ž q 2 . s
q2 ½ B22 Ž m2i ,m2j ,q 2 . y
64p 2 ž m2i q m 2j y
3 / q 12 Ž m i q m j . , 5
Ž 18 .
for i, j s p , K,h . The definitions of B22 Ž m 2i ,m2j ,q 2 . and l0 can be found in Appendix
C.1. Notice that BVi i Ž q 2 . is regular at q 2 s 0 for i s j.
For the isospin transverse part we find
2 r r pp 2 KK 2
P VŽ1.
p Ž q . s y4 Ž 2 H1 q L10 . y 8 BV Ž q . y 4 BV Ž q .
1 2 pp
q ½ y16 Ž2 q B V Ž q 2 . q q 2 BVK K Ž q 2 . . Lr9
Fp2
2
q16 q 2 Ž 2 mp q m K . Ž Lr9 q Lr10 . q 4 q 2 Ž 2 BVp p Ž q 2 . q BVK K Ž q 2 . . 5
y 32 mp2 C61
r
y 32 Ž mp2 q 2 m2K . C62
r
y 8 q 2 C93
r
. Ž 19 .
The hypercharge transverse part is given by
2 r r KK 2
1 2 KK
P VŽ1.
h Ž q . s y4 Ž 2 H1 q L10 . y 12 BV Ž q . q ½ y48q B V Ž q 2 . Lr9
Fp2
2
q48 m K Ž Lr9 q Lr10 . q 36 q 2 Ž BVK K Ž q 2 . . 5 y 32 m C 2
h
r
61
y 32 Ž mp2 q 2 m 2K . C62
r
y 8 q 2 C93
r
. Ž 20 .
The longitudinal part vanishes for the above two. These results agree with those
obtained in Ref. w6x when the differences in subtraction schemes are taken into account.
The expressions for the kaon two-point functions are new and are somewhat longer.
The transverse part is given by
P VŽ1.K Ž q 2 . s y4 Ž 2 H1r q Lr10 . y 6 Ž BVp K Ž q 2 . q BVh K Ž q 2 . .

1 8
q
Fp2 ½ y
q2
Ž m2K y mp2 . Ž 3 mp y 2 m K y mh . Lr5
q12 Ž mp q 2 m K q mh . Ž Lr9 q Lr10 . y 24 q 2 Ž BVp K Ž q 2 .

3
qBVh K Ž q 2 . . Lr9 y Ž 5mp mp y 4m K m K y 3 mh mh y 4mp m K
4 q2
G. Amoros
2
y6 mp mh q 12 m K mh . q 9 q 2 Ž BVp K Ž q 2 . q BVh K Ž q 2 . . 5
32 2
y
q2
Ž mp2 y m2K . r
C38 y 32 m2K C61
r
y 32 Ž mp2 q 2 m 2K . C62
r
8 2
y
q2
Ž mp2 y m2K . C91r y 8 q 2 C93r . Ž 21 .
This two-point function has also a longitudinal part
3 2 2
P VŽ0.K Ž q 2 . s 4 Ž m2K y mp2 . J p K Ž q 2 . q Ž m2K y mh2 . J h K Ž q 2 .
ž /
2q
1
q Fp ½ž
q 2 2 8 Ž m 2K y mp2 . Ž 3 mp y 2 m K y mh .
2
2
q12 Ž m2K y mp2 . J p K Ž q 2 . q 12 Ž m 2K y mh2 . J h K Ž q 2 . Lr5 /
3
q 4Ž 5mp mp y 4mp m K y 6 mp mh y 4m K m K q 12 m K mh y 3 mh mh .
y Ž y5mp q 2 m K q 3 mh . Ž m2K y mp2 . J p K Ž q 2 .
3
4
y 94 Ž mp y 2 m K q mh . Ž m2K y mh2 . J h K Ž q 2 . q 32 Ž Ž m2K y mp2 .

2 2
= J p K Ž q 2 . . y 169 Ž m 2K y mp2 . J p K Ž q 2 . q Ž m2K y mh2 . J h K Ž q 2 .
ž /
1 2
q 2 y 38 Ž mp2 q m2K . Ž Ž m2K y mp2 . J p K Ž q 2 . .
ž
q
q 34 Ž mp2 q m 2K .Ž m2K y mp2 . Ž m2K y mh2 . J p K Ž q 2 . J h K Ž q 2 .
2
y Ž 38 mp2 y 138 m2K . žŽm 2 2
K y mh . Jh K Ž q 2 . / /
9 2 2
2 2
y
16 q 4 žŽm 2
K y mp . J p K Ž q 2 . q Ž m2K y mh2 . J h K Ž q 2 . /5
8 2
q
q2
Ž mp2 y m2K . Ž 4C38r q C91r . . Ž 22 .
Notice that the Ademollo–Gatto theorem w22,23x is explicitly satisfied.
All divergences have been absorbed in the coefficients of the O Ž p 6 . Lagrangian by
setting
5 5
m4e Ž 4C38 q C91 . s 4C38 r r
q C91 q l2 y l Lr ,
32 F02
3F02 1 5
3
m4e C61 s C61
r
y l Ž Lr q Lr10 . ,
8 F02 1 9
1
m4e C62 s C62
r
y l Ž Lr q Lr10 . ,
8 F02 1 9
1 1
m4e C93 s C93
r
y l2 q l Lr , Ž 23 .
32 F0 2
2 F02 1 9
which agrees with the calculation of Ref. w24x.
G. Amoros 327
5. Masses and decay constants
5.1. Masses
The definition of mass is the position of the pole in a two-point Green function
GŽ p 2 ,m2 . that contains the relevant particle as a possible intermediate state. The
axial-vector two-point function is a suitable candidate to obtain the masses for the
pseudoscalar mesons. The general structure of this two-point function is shown in terms
of one-particle-irreducible Ž1PI. diagrams in Fig. 2. However the meson-propagator or
the two-point function of meson fields itself is a simpler set with the same pole. We
denote the sum of 1PI graphs by i P Ž p 2 ,m2i0 , F0 .. The set of diagrams contributing to
P Ž p 2 ,m2i0 , F0 . is depicted in Fig. 3c. The propagator is
n
i ` i
GŽ p 2
,m 2i0 , F0 . s p 2 y m2 Ý
0 ns0
ž iP Ž p 2
,m2i0 , F0 . p 2 y m2
0
/
i
s , Ž 24 .
p 2 y m 20 q P Ž p 2 ,m2i0 , F0 .
where m20 stands for the lowest order mass and m2i0 collectively denotes the various
lowest order masses. The physical mass is given by the zero of the denominator once the
external legs are on mass-shell
m2phys y m20 q P Ž m2phys ,m2i0 , F0 . s 0 . Ž 25 .

We replace the masses m i0 by their physical masses and F0 by Fp . It is sufficient to use
the NLO formulae for these in P Ž p 2 ,m2i0 , F0 . since it is already of O Ž p 4 .. This leads to
Ž4 . Ž6 .
m 2phys s m20 q Ž m2 . q Ž m2 . , Ž 26 .
where the bare masses appear only at the leading order and superscripts refer to the
chiral order.
The resulting formulae for the pion, kaon and eta masses are gathered in Appendix
A.1. The formulae for the pion and eta mass agree with those of Ref. w7x Žthe explicit
Fig. 2. The diagrams contributing to the axial-vector two-point function. The filled circles indicate the
one-particle-irreducible diagrams. Solid lines are pseudoscalar meson propagators and the wiggly lines indicate
insertions of an axial-vector current. For the inverse propagator the wiggly lines are meson legs and for the
decay constant the right wiggly line is a meson leg while the left remains an axial current.
G. Amoros
Fig. 3. The set of diagrams contributing to the 1PI quantities. Ža. axial-vector–axial-vector, Žb. axial-vector–
pseudoscalar, Žc. pseudoscalar–pseudoscalar. Žd. – Žl. the respective diagrams when the dashed lines are
replaced with the external legs of Ža., Žb. or Žc.. A line is a meson propagator, a wiggly line an external source,
a dot a vertex of O Ž p 2 ., a crossed circle a vertex of O Ž p 4 . and a crossed box a vertex of O Ž p 6 ..
formulae only appear in the preprint version. in the way described earlier, see Footnote
2, while the kaon result is new.
Notice that the precise expression for the O Ž p 6 . is dependent on the choice of the
O p 4 . expression3 , using the Gell-Mann–Okubo relation at O Ž p 4 ., produces differ-
Ž
ences at O Ž p 6 ..
The masses depend on seven combinations of the O Ž p 6 . constants. All the relevant
checks described in Section 4 were done.
5.2. Decay constants
The pseudoscalar decay constants, Fa , are defined by

²0 < A a Ž 0 . m < f a Ž p . : s i'2 pm Fa . Ž 27 .
They can be obtained directly from the definition or by the residue of the pole in the
axial-vector two-point function. We have calculated them first using their definition and
verified that the calculation via the two-point function yields the same results.
This calculation involves the use of the expression for P Ž p 2 ,m2i0 , F0 . obtained earlier
to get the wavefunction renormalization in addition to those diagrams of Fig. 3b for the
matrix element itself.
We can then write the results in the form
Fa s F0 Ž 1 q FaŽ 4 . q FaŽ 6 . . . Ž 28 .
Similarly to the masses, the precise form of FaŽ6. depends on the choice of FaŽ4.. For Fp
and Fh we have checked the double logarithms with those presented in Ref. w7x and the
result for FK is new.
3
This is why our expressions have some differences with those of Ref. w7x even after correcting for the
renormalization scheme.
G. Amoros 329
The explicit formulae are rather long and can be found in Appendix A.2. The relevant
checks described in Section 4 were performed.
6. The axial-vector two-point functions
The axial-vector two-point functions to lowest order are quite simple and all three
reduce to
2 F02
PmnA a Ž q . s 2 F02 gmn y qm qn . Ž 29 .
q 2 y m20 a
The NLO corrections only introduce minor changes. The decay constants change to Fa ,
the masses to the physical ones and there is an additional contribution from the O Ž p 4 .
constants to P AŽ1. Ž 2. r r Ž 6 . corrections
a q s 4 L10 y 8 H1 . In fact a very large part of the O p
is of a similar nature. We thus define
2 Fa2
PmnA a Ž q . s 2 Fa2 gmn y qm qn q Ž qm qn y gmn q 2 . Pˆ AŽ1. 2
a Žq .
q 2 y m2a
q qm qn Pˆ AŽ0. 2
a Žq . . Ž 30 .
The function ˆ P AŽ1.
a q
Ž 2. can be fully calculated from diagram Ža. of Fig. 2. These are
depicted in more detail in Fig. 3d–l and discussed in Appendix B.1.
All the diagrams in Fig. 2 contribute to Pˆ AŽ0. Ž 2 . even though most of their contents
a q
actually go into the redefinitions of the respective decay constants and masses. The full
result is given in Appendix B.2. The results fulfill the same checks as in Section 4. We
call Pˆ AŽ0.
a q
Ž 2 . and Pˆ AŽ1.
a q
Ž 2 . the longitudinal and transverse remainder respectively.
7. Estimates of some O (p 6 ) constants
In this section we estimate some of the O Ž p 6 . constants that appear in the results.
We assume saturation by the lightest vector, axial-vector and scalar mesons, extending
the formalism used in Refs. w8,9x to the present case.
For the spin-1 mesons we use the realization where the vector contribution to the
chiral Lagrangian starts at O Ž p 6 .. Keeping only the relevant terms for our calculation
we have
fV
L V s y 14 ² Vmn V mn y 2 M V2 Vm V m : y ² Vmn fqmn : q . . . , Ž 31.
2'2
where
mn
Vmn s =m Vn y =n Vm , f" s u Ž Õmn y amn . u† " u† Ž Õmn q amn . u ,
and the same holds for the axial-vector with the label change V A and fq fy. Vm
and Am are three-by-three matrices in flavour space and describe the full vector and
l l
axial-vector nonets, thus we assume a nonet symmetry throughout the rest of this
section. The rest of the notations was already presented in Section 3.
G. Amoros
After integrating out the vectors the terms contributing to the two-point Green
functions at O Ž p 6 . are
f V2 f A2
LV , A s 2
² fqmn =m=l fqln : q ² fymn =m=l fyln : . Ž 32 .
4 MV 4 MA2
In the scalar case, the Lagrangian reads
LS s 12 ² = mS=m S y MS2 S 2 : q c d ² Su m um : q c m² S xq :
dm
q ² S 2xq : q cg ² Sfq mn fqmn : q cgX ² Sfy mn fymn : . Ž 33 .
2
After integrating out the scalars, the O Ž p 6 . contribution we are interested in comes from
the terms
yc m2 d m c m2 c d cm d m
LS s 4
² xq = 2xq : q ² xq3 : q ² xq2 um u m :
2 MS 2 MS4 MS4
c m cg c m cgX
q ² xq fq mn fqmn : q ² xq fy mn fymn : , Ž 34 .
MS2 MS2
obtained after the shift of the vacuum expectation value and using the equation of
motion for the scalars. Note that only the relevant terms are written and, as in the vector
and axial case, a full nonet of scalars is assumed in S.
As input parameters we use
M V s Mr s 0.77 GeV, f V s 0.20, MA s M a1 s 1.23 GeV,
f A s 0.10, MS s 0.93 GeV,
c m s 0.042 GeV, c d s 0.032 GeV, cg s 19 P 10y3 GeVy1 ,
cgX ; cg and d m s y2.4 . Ž 35 .
™ ™
MS and d m are obtained from the masses of the scalars K 0) Ž1430. and a 0 Ž980.. The
value cg is obtained from G Ž f 0 gg . s 0.56 " 0.11 keV and is compatible also with
G Ž a0 gg . s 0.30 " 0.10 keV. The values of c m and c d are obtained forcing the
™
saturation of some of the O Ž p 4 . constants by the scalars w8,9x and are compatible with
those obtained in G Ž a 0 ph .. cgX value cannot be determined from data at present, we
assume a value similar to cg .
Using the notation of Ref. w24x for the O Ž p 6 . terms, the spin-1 Lagrangian yields
f V2 f A2
r
C87 ; 18
ž M V2
y
MA2 / ; 7.6 P 10y3 GeVy2 , Ž 36 .
r
f V2
C93 ;y ; y17 P 10y3 GeVy2 , Ž 37 .
4 M V2
G. Amoros 331
and the scalar Lagrangian estimates
r
cd cm d m
C14 ; ; y4.3 P 10y3 GeVy2 , Ž 38 .
MS4
r
d m c m2
C19 ; ; y2.8 P 10y3 GeVy2 , Ž 39 .
2 MS4
r
c m2
C38 ; ; 1.2 P 10y3 GeVy2 , Ž 40 .
2 MS4
r
c m cg
C61 ; ; 1.9 P 10y3 GeVy2 , Ž 41 .
MS2
r
c m cgX
C80 ; ; 1.9 P 10y3 GeVy2 . Ž 42 .
MS2
We stress that the goal of this section is to roughly estimate the values of the O Ž p 6 .
constants. A real determination would imply the use of chiral sum rules or other
processes to fix them. It is worth mentioning that the result in Eq. Ž37. is the same if we
use an antisymmetric formalism for the vector Lagrangian, and is in agreement with the
result extracted from the experimental data using sum rules for the vector–vector
two-point functions w11x. The generalization to the three flavours introduces a new
relation of O Ž p 6 . constants, 4C38
r r
q C91 , due to the explicit chiral symmetry breaking in
the kaon Green function.
Finally, we remark that the precise value of the constants can have an important
variation depending of the input values in Eq. Ž35.. Consequently, although the values
cited in this section are used for the numerical results, with the understanding that the
other counter-terms are set to zero, we have to keep in mind that these values could
overestimate the physical ones. The latter is especially true for d m since the K 0) and a 0
mass difference appears unnaturally large.
8. Some numerical results
We defer a more accurate comparison with experimental data to Ref. w13x, but we
would like to present some results using our explicit expressions. We use the values for
the Cir obtained in the previous section and two sets of the O Ž p 4 . constants. They only
differ in the values of Lr1 , Lr2 and Lr3 . Set A is obtained from the fit of the unitarized K l4
calculation while set B refers to K l4 and pp data at one loop accuracy w26x. We give
both sets to show an example of the variation with the O Ž p 4 . constants. We do not
show results for varying the other Lri but this results in a similar variation in size of the
O Ž p 6 . results. The explicit values we use, at m s 0.77 GeV, are
Lr1 s 0.37 P 10y3 , Lr2 s 1.35 P 10y3 , Lr3 s y3.5 P 10y3 , set A ,
Lr1 s 0.60 P 10y3 , Lr2 s 1.50 P 10y3 , Lr3 s y3.3 P 10y3 , set B ,
G. Amoros
Lr4 s y0.3 P 10y3 , Lr5 s 1.4 P 10y3 , Lr6 s y0.2 P 10y3 ,

Lr7 s y0.4 P 10y3 , Lr8 s 0.9 P 10y3 , Lr9 s 6.9 P 10y3 ,
Lr10 s 1r4 Ž f V2 y f A2 . s y7.5 P 10y3 ,
H1r s 1r8 Ž f V2 q f A2 . s 6.25 P 10y3 . Ž 43 .
For H1r ,
which can not be obtained experimentally, we take the value from the Meson
Saturation Model. Because the vector contribution should cancel for the axial-vector
two-point function, we use the same model value for Lr10 .
The rest of the quantities we use are
mp s 0.135 GeV, m K s 0.495 GeV,
mh s 0.548 GeV and Fp s 0.0924 GeV. Ž 44 .
They seem reasonable averages of the various isospin related ones.
8.1. The Õector two-point functions
In Figs. 4 and 5 we plot the real part of the three vector–vector two-point functions
choosing set A inputs.
In all the three cases the slopes are given mainly by the O Ž p 6 . constants as estimated
above. We have shown these contributions in the curves labelled ŽCT.. Essentially – and
r
with exception of C93 – the main effect of varying the input parameters is to shift the
plots vertically. We see that the loop effects are larger in the isospin case and smaller for
both, the hypercharge and kaon. In the chiral limit all three cases reduce to the same,
Fig. 4. Different contributions to the real part of the vector–vector two-point function in the isospin and the
hypercharge case. The label CT indicates the contribution from only the counter-terms Lri at O Ž p 4 . and Cir at
O Ž p 6 .. p 4 gives the O Ž p 4 . result. VMD indicates the vector model of Eq. Ž45..
G. Amoros 333
Fig. 5. Different contributions to the real part of the kaon vector–vector two-point function. The superscripts L
ŽT. refers to the longitudinal Žtransverse. part. For comparison we also plot the isospin case. CT as in Fig. 4.
and differences are related to the breaking of the SUŽ3. symmetry. For the isospin, the
two-pion channel produces the notable difference with the counter-term contribution,
while for the hypercharge and kaon the smaller difference is explained by the larger
masses in the loops and some explicit breaking of the symmetry through the quark
masses in the counter-term contributions.
In Fig. 4 we also plotted the case with a complete saturation by the vector meson
ŽVMD.
2 f V2 Ž 1 y q 2rM V2 .
Re Ž P V . ; 2
, Ž 45 .
Ž 1 y q 2rMV2 . q G 2rM V2
with G s 0.150 GeV. The conclusion is that models with only vectors explain the main
part of the two-point function, however an important contribution coming from the
two-pion intermediate states is present. The curve including only the counter-terms –
isospin ŽCT. – can also be obtained with the first two terms of the expansion in the
previous formula with G s 0 and considering the tiny modification due to the scalars.
8.2. Masses and decay constants
We continue our discussion with the masses and decay constants. We have summa-
rized our numerical results in Table 1 using the values for the O Ž p 4 . constants quoted
above. As one sees in columns three to six, both masses and decay constants have
substantial O Ž p 6 . loop contributions. In addition the pure polynomial piece at O Ž p 6 .
tends to have the opposite sign and is very large using our model dependent estimates.
This only reinforces the statement in Section 7 of the lack of knowledge in the scalar
G. Amoros
Table 1
O Ž p 4 . and O Ž p 6 . contributions to the masses and decay constants. The numbers in parentheses are the
extended double log approximations of Ref. w25x. Columns three to six are the O Ž p 6 . loop contributions only.
The contribution from the Cir is listed separately in the last column.
OŽ p4 . set A set A set A set B Cir
m ŽGeV. 0.77 0.77 0.5 1.0 0.77 –
Fp r F0 0.068 y0.101 y0.066 y0.100 y0.172 y0.001
Ž Fp r F0 . Ž0.013. Žy0.050.
FK r Fp 0.216 0.055 y0.023 0.100 0.035 y0.450
Ž FK r Fp . Ž0.08. Ž0.06.
Fh r Fp 0.312 0.092 y0.011 0.150 0.065 y0.600
mp2 r mp2 phy y0.039 0.214 0.132 0.238 0.355 y0.003
m2K r m2K phy y0.003 0.241 0.246 0.194 0.423 y0.873
mh2 r mh2 phy y0.045 0.312 0.234 0.273 0.521 y2.428
sector. The terms containing d m are the only ones contributing in this subsection and
seem severely overestimated even though they are of a size expected by naiÕe
dimensional analysis.
In order to have a full presentation of the O Ž p 6 . contributions a refit of all O Ž p 4 .
coefficients using the full O Ž p 6 . expressions would be needed. We postpone this till
after the main other processes are also calculated to this order given the dependence of
the O Ž p 6 . contributions on Lr1 –Lr3 . As an example using set A at m s 0.77 GeV
otherwise but shifting Lr5 to 1.0 P 10y3 reproduces the experimental value of FK rFp
when setting d m s 0.
For the decay constants the O Ž p 6 . contributions to the ratios are smaller than the
O Ž p 4 ., not including the estimates from scalar exchange to the O Ž p 6 . constants.
To judge the effect of the O Ž p 6 . contributions on determining the quark mass ratios
we use the lowest order, O Ž p 4 . and O Ž p 6 . formulae in terms of physical quantities to
obtain the lowest order masses using Eq. Ž26.. This leads to
2 ms 2 m 20 K y m 20p
°25.9 Ž p . ,2
s ~
s 24.9 Ž p 4 . , and
m20p
mu q md
¢24.1 Ž p . ,6
2 ms 3m20h y m20p
°24.2 Ž p . ,2
s ~
s 24.4 Ž p 4 . , Ž 46 .
2 m 20p
mu q md
¢23.3 Ž p . ,6
using the results from set A at m s 0.77 GeV and Cir s 0. These ratios can be compared
with 2 m srŽ m u q m d . s 25.5 w27,28x, 22.8 w29x; extracted from QCD sum rules and
lattice calculations respectively.
The emerging conclusions about the convergence of the chiral series should be very
cautious since a full study includes also the effect of the O Ž p 6 . constants. However,
while the corrections calculated are significant they do not show evidence of a
breakdown of the chiral expansion for the quantities presented here.
G. Amoros 335
Fig. 6. Different contributions to the real part of the remainder axial–axial two-point function. The superscripts
L ŽT. refers to the remainder longitudinal Žtransverse. part. CT as in Fig. 4.
8.3. The axial-Õector two-point functions
In Fig. 6 we plotted the dependence on momenta of the real part for the remainders of
the axial-vector two-point functions for the three cases under study. Because both the
longitudinal and transverse remainders have poles at q 2 s 0, we show the combination
Pˆ AŽ1. q Pˆ AŽ0.. A priori we would expect a different behaviour for the isospin case due to
the three pion channel. However there is virtually no effect because the imaginary part is
very small in the energy region we are considering in agreement with the dominance of
the a1 axial meson. For the other two cases even the three pseudoscalar channel is far.
The curves are thus very linear. The vertical shifts are due to the explicit breaking from
the quark masses. The O Ž p 6 . contributions are rather small, the scale in the plot should
be compared with <2 Fp2rq 2 < which is larger than 0.07 for the entire region plotted.
9. Summary and conclusions
In this paper we have calculated to NNLO in CHPT the vector and axial-vector
two-point functions in the isospin limit and in the complete three flavour basis.
In the vector–vector case, we confirm previous results for the isospin and hyper-
charge w6x.
For the axial-vector case, besides the cancellation of the non-analytic poles, we obtain
the same double and simple poles that appear with the use of the heat kernel expansion
w24x. We also agree with the double logarithms, appearing in previous work w7x, for the
isospin and hypercharge cases. All these checks give us confidence about our result.
G. Amoros
We have also given expressions to NNLO for the masses and decay constants. The
Lagrangian at O Ž p 6 . contains a rather large number of free constants. We have
estimated some of them using a simple resonance estimate and used this to present some
first numerical results for the two-point functions, masses and decay constants. We also
studied somewhat the m-dependence of the final result.
Although the corrections are significant they do not show evidence of a breakdown of
the chiral expansion. For instance, our estimates of the quark mass ratios are in
agreement with previous determinations. However, the sensitivity to the input values,
indicate that the O Ž p 4 . constants need to be refitted using the full O Ž p 6 . expressions
and that better estimates of the O Ž p 6 . constants are necessary.
Acknowledgements
We thank Ll. Ametller for a careful reading of the manuscript. The work of P.T. was
supported by the Swedish Research Council ŽNFR..
Appendix A. Explicit results for the masses and decay constants
A.1. Masses
The masses are split as follows:

Ž4 . Ž6 . Ž6 .
m2a s m20 a q Ž m2a . q Ž m2a . CT q Ž m2a . loops , Ž A.1 .
6.
with m 0 a the contribution from the bare masses. In the O Ž p we have explicitly
separated the chiral loop contribution from the model dependent counter-terms.
In a previous step the masses are obtained in terms of only the bare masses Žquark
masses., we rewrite the O Ž p 4 . contribution with the physical masses implying a
modification of the O Ž p 6 . terms. In the O Ž p 6 . we can safely replace bare masses with
physical masses.
For the pion we obtain
m 20p s B0 Ž m u q m d . s 2 B0 m
ˆ,
Fp2 Ž4 .
Ž mp2 . s 8 Ž mp2 q 2 m2K . Ž 2 Lr6 y Lr4 . q 8 mp2 Ž 2 Lr8 y Lr5 . q mp y 13 mh ,
mp2
Ž A.2 .
in agreement with Ref. w19x. The O Ž p 6 . contributions are
Ž6 .
Fp2 Ž mp2 . CT s y32 mp6 C12
r
y 32 mp6 C13
r
y 16 mp6 C14
r
y 16 mp6 C15
r
y 48 mp6 C16
r
y 16 mp6 C17
r
q 48 mp6 C19
r
q 80 mp6 C20
r
q 48 mp6 C21
r
q 32 mp6 C31
r
q 32 mp6 C32
r
y 64 mp4 m2K C13
r
y 32 mp4 m2K C15
r
q 64 mp4 m2K C16
r
q 192 mp4 m 2K C21

r
q 64 mp4 m 2K C32
r
y 64 mp2 m 4K C16
r
q 64 mp2 m4K C20

r
q 192 mp2 m 4K C21
r
, Ž A.3 .
G. Amoros 337
Ž6 . 1
Fp4 Ž mp2 . Loops s 2 Ž y527r1296mp6 p 2 y 3217r1728mp6
Ž 16p 2 .
y37r324mp4 m2K p 2 y 139r216mp4 m 2K y 11r36mp2 m4K p 2
1
y15r16mp2 m 4K . q Ž Ž y2 mp4 y 2 mp2 m2K . mp y 2 mp2 m2K mK
16p 2
q4mp6 Lr1 q Ž 74r9mp6 y 16r9mp4 m 2K q 104r9mp2 m4K . Lr2
q Ž 56r27mp6 y 16r27mp4 m 2K q 86r27mp2 m4K . Lr3 .
y Ž 589r36mp2 q 2 m2K . mp2 y 4 mp2 mp m K q 5r3mp2 mp mh
y 112 mp4 mp Lr1 y 64 mp4 mp Lr2 y 56 mp4 mp Lr3
q Ž 144mp4 q 80 mp2 m2K . mp Lr4 q 96 mp4 mp Lr5
q Ž y256mp4 y 160mp2 m 2K . mp Lr6 y 176mp4 mp Lr8
q Ž y2r3mp4 my2 2 2 2 2 2 r
K y 7mp . m K y 4r3mp m K mh y 128mp m K m K L1
y 32 mp2 m2K m K Lr2 y 40 mp2 m2K m K Lr3 q Ž 16 mp4 q 160mp2 m2K .

=m K Lr4 q Ž 16 mp4 q 32 mp2 m2K . m K Lr5 q Ž y32 mp4 y 192 mp2 m2K .
=m K Lr6 q Ž y32 mp4 y 64 mp2 m 2K . m K Lr8 q Ž y5r9mp4 mhy2
y29r36mp2 . mh2 q Ž 32r3mp4 y 128r3mp2 m2K . mh Lr1
q Ž 8r3mp4 y 32r3mp2 m 2K . mh Lr2 q Ž 8r3mp4 y 32r3mp2 m2K . mh Lr3
q Ž y16mp4 q 48 mp2 m 2K . mh Lr4 q Ž y64r9mp4 q 64r9mp2 m2K .
=mh Lr5 q Ž 64r3mp4 y 160r3mp2 m2K . mh Lr6 q Ž y128r3mp4
q128r3mp2 m2K . mh Lr7 y 16r3mp4 mh Lr8 q Ž y128mp6
y512 mp4 m2K y 512 mp2 m 4K . Lr42 q Ž y256mp6 y 384mp4 m2K
y128mp2 m4K . Lr4 Lr5 q Ž 512 mp6 q 2048mp4 m2K q 2048mp2 m4K . Lr4 Lr6
q Ž 512 mp6 q 768mp4 m 2K q 256mp2 m4K . Lr4 Lr8 y 128mp6 Lr52
q Ž 512 mp6 q 768mp4 m 2K q 256mp2 m4K . Lr5 Lr6 q 512 mp6 Lr5 Lr8
q Ž y512 mp6 y 2048mp4 m 2K y 2048mp2 m4K . Lr62
q Ž y1024mp6 y 1536mp4 m2K y 512 mp2 m4K . Lr6 Lr8 y 512 mp6 Lr82
q 5r6 H F Ž mp2 ,mp2 ,mp2 ;mp2 . mp4 y 5r8 H F Ž mp2 ,m 2K ,m2K ;mp2 . mp4
q 1r18 H F Ž mp2 ,mh2 ,mh2 ;mp2 . mp4 q H F Ž m2K ,mp2 ,m2K ;mp2 . mp2 m2K
y5r6 H F Ž m 2K ,m2K ,mh2 ;mp2 . mp4 y 1r8 H F Ž mh2 ,m2K ,m2K ;mp2 . mp4
q 1r2 H F Ž mh2 ,m2K ,m2K ;mp2 . mp2 m2K q H1F Ž mp2 ,m2K ,m2K ;mp2 . mp4
q 2 H1F Ž m 2K ,m2K ,mh2 ;mp2 . mp4 q 3 H21
F
Ž mp2 ,mp2 ,mp2 ;mp2 . mp4
F
y 3r8 H21 Ž mp2 ,m2K ,m2K ;mp2 . mp4 q 3 H21F Ž m2K ,mp2 ,m2K ;mp2 . mp4
F
q 9r8 H21 Ž mh2 ,m2K ,m2K ;mp2 . mp4 . Ž A.4 .
Where the definitions of HiF Ž m12 ,m22 ,m23 ;q 2 . appear in Appendix C.
G. Amoros
For the kaon we obtain

m20 K s B0 Ž m s q m
ˆ. ,
Fp2 Ž4 . mp2 q 3mh2
m2K
Ž m2K . s 8 Ž mp2 q 2 m 2K . Ž 2 Lr6 y Lr4 . q 8 m 2K Ž 2 Lr8 y Lr5 . q
6 m2K
mh ,
Ž A.5 .
Ž6 .
Fp2 Ž m2K . CT s y16mp4 m2K C14
r
y 48 mp4 m 2K C16
r
q 16mp4 m 2K C17
r
q 48 mp4 m 2K C19
r
q 48mp4 m2K C20
r
q 48 mp4 m2K C21
r
y 32 mp2 m4K C13
r
q 32 mp2 m 4K C14
r
y 16 mp2 m4K C15
r
q 64 mp2 m4K C16
r
y 32 mp2 m4K C17
r
y 96 mp2 m 4K C19
r
y 32 mp2 m4K C20
r
q 192 mp2 m4K C21
r
q 32 mp2 m 4K C32
r
y 32 m6K C12
r
y 64 m6K C13
r
y 32 m6K C14
r
y 32 m6K C15
r
y 64 m6K C16
r
q 96 m6K C19
r
q 128m6K C20
r
q 192 m6K C21
r
q 32 m6K C31
r
q 64m6K C32
r
,
Ž A.6 .
Ž6 . 1
Fp4 Ž m2K . Loops s 2 Ž y1r8mp4 m2K p 2 y 13r24mp4 m2K y 73r648mp2 p 2
Ž 16p 2 .
y19r108mp2 m 4K y 763r1296m6K p 2 y 4709r1728m6K .

1
q
16p 2
Ž y3r2 m4K mp q Žy3r2 mp2 m2K y 3r2 m4K . mK
q Ž y1r2 mp2 m2K y m 4K . mh q 4 m6K Lr1 q Ž 56r9mp4 m2K
y16r9mp2 m4K q 122r9m6K . Lr2 q Ž 41r27mp4 m2K
y4r27mp2 m 4K q 89r27m6K . Lr3 . q Ž y2 mpy2 m4K y 27r8m2K . mp2
y 3m2K mp m K q Ž 1r3mp2 y 41r12 m2K . mp mh y 96 mp2 m2K mp Lr1
y 24 mp2 m2K mp Lr2 y 30 mp2 m2K mp Lr3 q Ž 136mp2 m2K q 64 m4K .

=mp Lr4 q Ž 24 mp2 m 2K q 32 m4K . mp Lr5 q Ž y176mp2 m2K y 128m4K .
=mp Lr6 q Ž y48mp2 m 2K y 64 m4K . mp Lr8 q Ž y3r2 mp2
y251r18m 2K . m 2K y 8r3m2K m K mh y 144m4K m K Lr1 y 72 m 4K m K Lr2
y 60 m 4K m K Lr3 q Ž 16 mp2 m2K q 160m4K . m K Lr4 q 64m4K m K Lr5
q Ž y32 mp2 m2K y 256m 4K . m K Lr6 y 128m4K m K Lr8
q Ž y43r288mp4 mhy2 y 25r32 mp2 y 5r9m2K . mh2 q Ž 32r3mp2 m2K

G. Amoros 339
y128r3m 4K . mh Lr1 q Ž 8r3mp2 m2K y 32r3m4K . mh Lr2

q Ž 14r3mp2 m2K y 56r3m4K . mh Lr3 q Ž y8mp2 m 2K q 64 m4K . mh Lr4
q Ž 16r9mp4 y 8r3mp2 m2K q 224r9m4K . mh Lr5 q Ž 16r3mp2 m 2K
y256r3m 4K . mh Lr6 q Ž y64r3mp4 q 64 mp2 m2K y 128r3m4K . mh Lr7
q Ž y32r3mp4 q 112r3mp2 m2K y 64m 4K . mh Lr8 q Ž y128mp4 m2K
y512 mp2 m 4K y 512 m6K . Lr42 q Ž y128mp4 m2K y 256mp2 m4K
y384m6K . Lr4 Lr5 q Ž 512 mp4 m 2K q 2048mp2 m4K q 2048m6K . Lr4 Lr6
q Ž 128mp4 m 2K q 384mp2 m 4K q 1024m6K . Lr4 Lr8 q Ž y64mp2 m4K
y64m6K . Lr52 q Lr5 Lr6 Ž 256mp4 m2K q 512 mp2 m 4K q 768m6K .
q Lr5 Lr8 Ž 128mp2 m 4K q 384m6K . q Ž y512 mp4 m2K y 2048mp2 m4K
y2048m6K . Lr62 q Ž y256mp4 m2K y 768mp2 m4K y 2048m6K . Lr6 Lr8
y512 m6K Lr82 q 3r4H F Ž mp2 ,mp2 ,m2K ;m 2K . mp2 m2K
q 3r8 H F Ž mp2 ,mp2 ,m2K ;m2K . m4K q 1r4H F Ž mp2 ,m2K ,mh2 ;m2K . m4K
y 3r32 H F Ž m2K ,mp2 ,mp2 ;m2K . m 4K q 9r16 H F Ž m2K ,mp2 ;mh2 ;m2K .
=m4K q 3r4H F Ž m2K ,m 2K ,m 2K ;m2K . m4K

q 181r288 H F Ž m2K ,mh2 ,mh2 ;m2K . m4K
y 3r2 H1F Ž mp2 ,mp2 ,m2K ;m2K . m4K y 3r2 H1F Ž m2K ,mp2 ,mh2 ;m2K . m 4K
y 5r4H1F Ž m 2K ,mh2 ,mh2 ;m2K . m4K q 9r4H21

F
Ž mp2 ,mp2 ,m2K ;m2K . m4K
F
y 9r32 H21 Ž m2K ,mp2 ;mp2 ,m2K . m4K
F
q 27r16 H21 Ž m2K ,mp2 ,mh2 ;m2K . m4K
F
q 9r4H21 Ž m2K ,m2K ,m2K ;m2K . m4K
F
q 27r32 H21 Ž m2K ,mh2 ,mh2 ;m2K . m4K . Ž A.7 .
For the eta we obtain
m20h s B0 23 Ž 2 m s q m
ˆ.,
Ž4 .
Fp2 Ž mh2 . s y8mh2 Ž mp2 q 2 m 2K . Lr4 q 83 mh2 Ž mp2 y 4 m2K . Lr5
y 163 Ž mp4 y 2 mp2 m 2K y 8 m 4K . Lr6 q 163 Ž 3mp4 y 8 mp2 m2K q 8 m4K . Lr8
y mp2 mp q Ž 32 mp2 q 2 mh2 . m K q Ž 97 mp2 y 169 m2K . mh

2
q 128 2 2 r
3 Ž m K y mp . L 7 , Ž A.8 .
G. Amoros

Ž6 .
Fp2 Ž mh2 . CT s 32r27mp6 C12
r
y 32r9mp6 C13
r
q 16r3mp6 C14
r
y 16r9mp6 C15
r
q 16 mp6 C16
r
q 16r3mp6 C17
r
q 128r9mp6 C18
r
y 16 mp6 C19
r
q 16 mp6 C20
r
y 16 mp6 C21
r
y 32r3mp6 C31
r
q 32 mp6 C32
r
y 128r9mp4 m2K C12
r
q 64r3mp4 m2K C13

r
y 320r9mp4 m 2K C14
r
q 32r3mp4 m2K C15
r
y 256r3mp4 m2K C16

r
y 320r9mp4 m2K C17
r
y 256r3mp4 m 2K C18
r
q 192 mp4 m2K C19

r
q 64 mp4 m 2K C20
r
q 128mp4 m2K C31
r
y 64r3mp4 m2K C32

r
q 512r3mp4 m2K C33
r
q 512r9mp2 m4K C12
r
q 640r9mp2 m4K C14

r
q 320r3mp2 m4K C16
r
q 640r9mp2 m 4K C17
r
q 128mp2 m4K C18

r
y 384mp2 m4K C19
r
y 192 mp2 m4K C20
r
q 192 mp2 m4K C21

r
y 256mp2 m4K C31
r
y 256r3mp2 m4K C32
r
y 1024r3mp2 m4K C33

r
y 2048r27m6K C12
r
y 1024r9m6K C13
r
y 512r9m6K C14
r
y 512r9m6K C15
r
y 256r3m6K C16
r
y 512r9m6K C17
r
y 512r9m6K C18
r
q 256m6K C19
r
q 256m6K C20
r
q 256m6K C21
r
q 512r3m6K C31
r
q 512r3m6K C32
r
q 512r3m6K C33
r
, Ž A.9 .
Ž6 . 1
Fp4 Ž mh2 . Loops s 2 Ž y91r11664mp6 p 2 y 1781r15552 mp6
Ž 16p 2 .
y269r486mp4 m2K p 2 y 4133r1296mp4 m2K q 133r108mp2 m 4K p 2
q367r48mp2 m 4K y 1091r729m6K p 2 y 7567r972 m6K .

1
q
16p 2
Ž mK Ž 2r3mp4 y 4r3mp2 m2K y 16r3m4K .
qLr1 Ž y4r27mp6 q 16r9mp4 m2K y 64r9mp2 m4K q 256r27m6K .
qLr2 Ž y58r27mp6 q 88r9mp4 m2K y 88r9mp2 m4K q 544r27m6K .
qLr3 Ž y20r27mp6 q 32r9mp4 m2K y 34r9mp2 m4K q 152r27m6K . .
q mp2 Ž 43r12 mp2 y 3m 2K . q mp m K Ž y8r3mp2 y 32r3m2K .
q mp mh Ž y35r9mp2 q 64r9m 2K . q mp Lr1 Ž 32 mp4 y 128mp2 m2K .
q mp Lr2 Ž 8 mp4 y 32 mp2 m2K . q mp Lr3 Ž 8 mp4 y 32 mp2 m2K .
q mp Lr4 Ž y160r3mp4 q 144mp2 m2K q 256r3m4K .
q mp Lr5 Ž 16r9mp4 y 64r3mp2 m 2K q 512r9m 4K . q mp Lr6 Ž 64 mp4
y544r3mp2 m 2K y 512r3m 4K . q mp Lr7 Ž y384mp4 q 1664r3mp2 m2K
G. Amoros 341
y512r3m4K . q mp Lr8 Ž y144mp4 q 640r3mp2 m2K y 512r3m4K .
qm2K Ž y4mp4 my2 2 2 2

K q 43r3mp y 212r9m K . q m K mh Ž 28r9mp
y64r9m2K . q m K Lr1 Ž 128r3mp2 m 2K y 512r3m 4K .

q m K Lr2 Ž 32r3mp2 m 2K y 128r3m4K . q m K Lr3 Ž 56r3mp2 m 2K
y224r3m 4K . q m K Lr4 Ž y16r3mp4 y 32r3mp2 m 2K q 256m4K .
q m K Lr5 Ž 80r9mp4 y 32 mp2 m2K q 1024r9m4K . q m K Lr6 Ž 32r3mp4
q64r3mp2 m2K y 256m4K . q m K Lr7 Ž y512r3mp4 q 1280r3mp2 m2K
y256m 4K . q m K Lr8 Ž y224r3mp4 q 704r3mp2 m 2K y 256m4K .
q mh2 Ž y5r9mp4 mhy2 q 55r12 mp2 y 25r3m2K .
q mh Lr1 Ž y16r3mp4 q 128r3mp2 m2K y 256r3m4K .
q mh Lr2 Ž y16r3mp4 q 128r3mp2 m 2K y 256r3m 4K .
q mh Lr3 Ž y8r3mp4 q 64r3mp2 m2K y 128r3m 4K .
q mh Lr4 Ž 32r3mp4 y 112r3mp2 m2K q 128r3m4K .
q mh Lr5 Ž 496r27mp4 y 1472r27mp2 m2K q 1408r27m4K .
q mh Lr8 Ž y592r9mp4 q 2176r9mp2 m 2K y 256m4K .
qLr42 Ž 128r3mp6 y 512 mp2 m4K y 2048r3m6K . q Lr4 Lr5 Ž 256r9mp6

y128r3mp4 m2K y 128mp2 m4K y 5632r9m6K . q Lr4 Lr6 Ž y512r3mp6
q2048mp2 m 4K q 8192r3m6K . q Lr4 Lr7 Ž 1024mp6 y 3072 mp2 m4K
q2048m6K . q Lr4 Lr8 Ž 1024r3mp6 y 256r3mp4 m2K y 3328r3mp2 m4K
q7168r3m6K . q Lr52 Ž y128r9mp6 q 256r3mp4 m2K
y256r3mp2 m4K y 1024r9m6K . q Lr5 Lr6 Ž y1024r9mp6
q256r3mp4 m2K q 1280r3mp2 m4K q 10240r9m6K .
q Lr5 Lr7 Ž 1024mp6 y 4096r3mp4 m 2K y 1024r3mp2 m 4K
q2048r3m6K . q Lr5 Lr8 Ž 3584r9mp6 y 6656r9mp4 m 2K
y512r9mp2 m4K q 8192r9m6K . q Lr62 Ž 512r3mp6 y 2048mp2 m4K
y8192r3m6K . q Lr6 Lr7 Ž y4096r3mp6 q 4096mp2 m4K
y8192r3m6K . q Lr6 Lr8 Ž y1024r3mp6 q 512r3mp4 m2K
q3584r3mp2 m4K y 4096m6K . q Lr7 Lr8 Ž y4096r3mp6
G. Amoros
q4096r3mp4 m2K q 4096r3mp2 m4K y 4096r3m6K .
qLr82 Ž y512 mp6 q 2048r3mp4 m 2K q 2048r3mp2 m4K
y4096r3m6K . q 1r6 H F Ž mp2 ,mp2 ,mh2 ;mh2 . mp4
q 1r8 H F Ž mp2 ,m 2K ,m2K ;mh2 . mp4
q 3r2 H F Ž mp2 ,m2K ,m2K ;mh2 . mp2 m2K
y 1r2 H F Ž m 2K ,m 2K ,mh2 ;mh2 . mp4 q 4 H F Ž m 2K ,m2K ,mh2 ;mh2 . mp2 m 2K
y 64r9H F Ž m2K ,m2K ,mh2 ;mh2 . m 4K q 1r8 H F Ž mh2 ,m2K ,m2K ;mh2 . mp4
y H F Ž mh2 ,m2K ,m2K ;mh2 . mp2 m2K q 2 H F Ž mh2 ,m2K ,m2K ;mh2 . m4K
q 49r486 H F Ž mh2 ,mh2 ,mh2 ;mh2 . mp4
y 112r243H F Ž mh2 ,mh2 ,mh2 ;mh2 . mp2 m2K
q 128r243H F Ž mh2 ,mh2 ,mh2 ;mh2 . m 4K q H1F Ž mp2 ,m2K ,m2K ;mh2 . mp4
y4H1F Ž mp2 ,m 2K ,m2K ;mh2 . mp2 m2K q 2 H1F Ž m2K ,m 2K ,mh2 ;mh2 . mp4
y 40r3H1F Ž m2K ,m2K ,mh2 ;mh2 . mp2 m2K
q 64r3H1F Ž m2K ,m2K ,mh2 ;mh2 . m4K q 3r8 H21

F
Ž mp2 ,m2K ,m2K ;mh2 . mp4
F
y 3 H21 Ž mp2 ,m2K ,m2K ;mh2 . mp2 m2K q 6 H21F Ž mp2 ,m2K ,m2K ;mh2 . m4K
F
q 3r8 H21 Ž mh2 ,m2K ,m2K ;mh2 . mp4 y 3 H21F Ž mh2 ,m2K ,m2K ;mh2 . mp2 m2K
F
q 6 H21 Ž mh2 ,m2K ,m2K ;mh2 . m4K . . Ž A.10 .
A.2. Decay constants
The decay constants are given by
Fa s F0 1 q FaŽ 4 . q Ž FaŽ 6 . . CT q Ž FaŽ 6 . . loops .

ž / Ž A.11 .
For the pion we obtain
Fp2 FpŽ4. s 4 Ž mp2 q 2 m 2K . Lr4 q 4 mp2 Lr5 y 2 mp y m K , Ž A.12 .

4. 6.
in agreement with Ref. w19x for the O Ž p contribution. The O Ž p contribution is
Ž6 .
Fp2 Ž Fp . CT s 8 mp4 C14
r
q 8 mp4 C15
r
q 24 mp4 C16
r
q 8 mp4 C17
r
q 16 mp2 m 2K C15
r
y 32 mp2 m2K C16

r
q 32 m4K C16
r
, Ž A.13 .
G. Amoros 343
1
Fp4 Ž FpŽ6. . loops s 2 Ž 35r288mp4 p 2 q 41r128mp4 q 1r144mp2 m2K p 2
Ž 16p 2 .
y5r32 mp2 m2K q 11r72 m 4K p 2 q 15r32 m4K .

1
q
16p 2
Ž mp Ž1r2 mp2 q m2K . q Ž1r4mp2 q m2K . mK
q Ž 1r3mp2 y 1r3m2K . mh y 2 mp4 Lr1 q Ž y37r9mp4 q 8r9mp2 m2K
y52r9m 4K . Lr2 q Ž y28r27mp4 q 8r27mp2 m 2K y 43r27m 4K . Lr3
q Ž 8 mp4 q 20mp2 m 2K q 8 m4K . Lr4 q Ž 8 mp4 q 4 m 4K . Lr5
q Ž y16mp4 y 40 mp2 m 2K y 16 m4K . Lr6 q Ž y16mp4 y 8 m4K . Lr8 .
q Ž 7r8 q mpy2 m 2K . mp2 q 2r3 mp mh q 56 mp2 mp Lr1 q 32 mp2 mp Lr2
q 28 mp2 mp Lr3 q 48 m 2K mp Lr4 q 12 mp2 mp Lr5 q Ž y32 mp2
y64m 2K . mp Lr6 y 32 mp2 mp Lr8 q Ž 3 y 1r2 mp2 my2 2
K . mK
y 2r3 m K mh q 64 m2K m K Lr1 q 16 m 2K m K Lr2 q 20 m2K m K Lr3

q Ž 12 mp2 y 8 m2K . m K Lr4 q Ž y4mp2 q 8 m2K . m K Lr5 q Ž y16mp2
y32 m2K . m K Lr6 y 16m2K m K Lr8 q 3r8 mh2 q Ž y16r3mp2
q64r3m2K . mh Lr1 q Ž y4r3mp2 q 16r3m 2K . mh Lr2 q Ž y4r3mp2
q16r3m2K . mh Lr3 q Ž 8r3mp2 y 32r3m 2K . mh Lr4 y 4r3mp2 mh Lr5
q Ž y8mp4 y 32 mp2 m2K y 32 m4K . Lr42 q Ž y16mp4 y 32 mp2 m2K .

X
=Lr4 Lr5 y 8 mp4 Lr52 q 5r12 H F Ž mp2 ,mp2 ,mp2 ;mp2 . mp4
y 1r2 H F Ž mp2 ,mp2 ,mp2 ;mp2 . mp2
X
y 5r16 H F Ž mp2 ,m2K ,m2K ;mp2 . mp4
q 1r16 H F Ž mp2 ,m 2K ,m2K ;mp2 . mp2
X
q 1r36 H F Ž mp2 ,mh2 ,mh2 ;mp2 . mp4
X
q 1r2 H F Ž m 2K ,mp2 ,m2K ;mp2 . mp2 m2K
y 1r2 H F Ž m2K ,mp2 ,m 2K ;mp2 . m2K
X
y 5r12 H F Ž m2K ,m2K ,mh2 ;mp2 . mp4
X
y 1r16 H F Ž mh2 ,m2K ,m2K ;mp2 . mp4
X
q 1r4H F Ž mh2 ,m2K ,m 2K ;mp2 . mp2 m2K
q1r16 H F Ž mh2 ,m2K ,m 2K ;mp2 . mp2

G. Amoros
y 1r4H F Ž mh2 ,m2K ,m 2K ;mp2 . m2K

X
q 1r2 H1F Ž mp2 ,m 2K ,m2K ;mp2 . mp4
X X
q H1F Ž m 2K ,m 2K ,mh2 ;mp2 . mp4 q 3r2 H21
F
Ž mp2 ,mp2 ,mp2 ;mp2 . mp4
X
F
y 3r16 H21 Ž mp2 ,m2K ,m2K ;mp2 . mp4
X
F
q 3r2 H21 Ž m2K ,mp2 ,m2K ;mp2 . mp4
X
F
q 9r16 H21 Ž mh2 ,m2K ,m2K ;mp2 . mp4 . Ž A.14 .
X
The functions HiF Ž m12 ,m22 ,m23 ;q 2 . are defined in Appendix C.
For the kaon we obtain
Fp2 FKŽ4. s 4 Ž mp2 q 2 m 2K . Lr4 q 4 m2K Lr5 y 34 mp y 32 m K y 34 mh . Ž A.15 .
The O Ž p 6 . contribution is given by
Fp2 Ž FKŽ6. . CT s 8 mp4 C14
r
q 24 mp4 C16
r
y 8 mp4 C17
r
y 16 mp2 m 2K C14
r
q 8 mp2 m 2K C15
r
y 32 mp2 m 2K C16
r
q 16mp2 m 2K C17
r
q 16 m4K C14
r
q 16 m4K C15
r
q 32 m4K C16
r
, Ž A.16 .
1
Fp4 Ž FKŽ6. . Loops s Ž 1r16mp4 p 2 q 13r48mp4 q 1r32 mp2 m2K p 2
Ž 16p 2 .
y29r192 mp2 m 2K q 3r16m4K p 2 q 197r384m4K .

1
q
16p 2
Ž 15r16m2K mp q Ž3r4mp2 q 1r8m2K . mK
q Ž 1r12 mp2 q 41r48m2K . mh y 2 m4K Lr1
q Ž y28r9mp4 q 8r9mp2 m 2K y 61r9m4K . Lr2
q Ž y41r54mp4 q 2r27mp2 m2K y 89r54m4K . Lr3
q Ž 2 mp4 q 14 mp2 m 2K q 20 m4K . Lr4
q Ž 10r3mp4 y 8r3mp2 m 2K q 34r3m4K . Lr5
q Ž y4mp4 y 28 mp2 m2K y 40 m4K . Lr6
q Ž y16mp4 q 32 mp2 m 2K y 16 m4K . Lr7
q Ž y12 mp4 q 16 mp2 m 2K y 28 m4K . Lr8 .
q mp2 Ž 9r32 q 3r8mpy2 m2K . q 3r8 mp m K
q Ž 13r16q 3r4mp2 mhy2 . mp mh q 48 mp2 mp Lr1
q 12 mp2 mp Lr2 q 15mp2 mp Lr3 q Ž y15mp2 q 18 m2K . mp Lr4
q Ž 6 mp2 y 3m2K . mp Lr5 q Ž y12 mp2 y 24 m2K . mp Lr6
y 12 mp2 mp Lr8 q m2K Ž 27r8 q 3r4mp2 my2
K .
G. Amoros 345
q Ž y29r8y 1r2 mp2 mhy2 . m K mh q 72 m2K m K Lr1

q 36 m2K m K Lr2 q 30 m2K m K Lr3 q Ž 18 mp2 q 4 m2K . m K Lr4
q 6 m2K m K Lr5 q Ž y24mp2 y 48 m2K . m K Lr6
y 24 m2K m K Lr8 q Ž 17r8 q 1r32 mp2 mhy2 . mh2
q Ž y16r3mp2 q 64r3m 2K . mh Lr1 q Ž y4r3mp2 q 16r3m2K . mh Lr2
q Ž y7r3mp2 q 28r3m 2K . mh Lr3 q Ž 35r3mp2 q 22r3m2K . mh Lr4
q Ž y2 mp2 q 23r3m2K . mh Lr5 q Ž y12 mp2 y 24 m2K . mh Lr6
q Ž y18mp4 mhy2 q 42 mp2 y 24 m2K . mh Lr7
q Ž y6mp4 mhy2 q 18 mp2 y 24 m2K . mh Lr8
q Ž y8mp4 y 32 mp2 m2K y 32 m 4K . Lr42
q Ž y16mp2 m2K y 32 m4K . Lr4 Lr5 y 8 m4K Lr52
X
q 3r8 H F Ž mp2 ,mp2 ,m2K ;m2K . mp2 m 2K
X
q 3r16 H F Ž mp2 ,mp2 ,m2K ;m 2K . m4K
X
y 3r8 H F Ž mp2 ,mp2 ,m2K ;m2K . mp2
X
q 1r8 H F Ž mp2 ,m2K ,mh2 ;m 2K . m4K
X
y 3r64H F Ž m2K ,mp2 ,mp2 ;m 2K . m4K
q 3r64H F Ž m 2K ,mp2 ,mp2 ;m2K . m2K
X
q9r32 H F Ž m 2K ,mp2 ,mh2 ;m2K . m4K
y 9r32 H F Ž m 2K ,mp2 ,mh2 ;m2K . m2K
X
q 3r8 H F Ž m2K ,m2K ,m2K ;m2K . m4K
y 3r8 H F Ž m2K ,m2K ,m 2K ;m2K . m2K
X
q 181r576 H F Ž m2K ,mh2 ,mh2 ;m2K . m4K
y 9r64H F Ž m2K ,mh2 ,mh2 ;m2K . m 2K
X
y 3r4H1F Ž mp2 ,mp2 ,m2K ;m2K . m4K
X
y 3r4H1F Ž m2K ,mp2 ,mh2 ;m2K . m4K
X
y 5r8 H1F Ž m2K ,mh2 ,mh2 ;m2K . m4K
X
F
q 9r8 H21 Ž mp2 ,mp2 ,m2K ;m2K . m4K
X
F
y 9r64H21 Ž m2K ,mp2 ,mp2 ;m2K . m4K
X
F
q 27r32 H21 Ž m2K ,mp2 ,mh2 ;m2K . m4K
X
F
q 9r8 H21 Ž m2K ,m2K ,m2K ;m2K . m4K
X
F
q 27r64H21 Ž m2K ,mh2 ,mh2 ;m2K . m4K . Ž A.17 .
G. Amoros
For the eta we obtain

Fp2 FhŽ4. s 4 Ž mp2 q 2 m 2K . Lr4 q 43 Ž 4 m2K y mp2 . Lr5 y 3 m K , Ž A.18 .
in agreement with Ref. w19x for the O Ž p 4 . contribution. The O Ž p 6 . contribution is
Fp2 FhŽ6.
ž / CT
s 8 mp4 C14
r
y 8r3mp4 C15
r
q 24 mp4 C16
r
q 8 mp4 C17
r
q 64r3mp4 C18
r
y 64r3mp2 m 2K C14
r
q 16r3mp2 m2K C15
r
y 32 mp2 m2K C16
r
y 64r3mp2 m 2K C17
r
y 128r3mp2 m 2K C18
r
q 64r3m 4K C14
r
q 64r3m 4K C15
r
q 32 m 4K C16
r
q 64r3m4K C17
r
q 64r3m 4K C18
r
,
Ž A.19 .
1
Fp4 FhŽ6.
ž / s Ž 5r96mp4 p 2 q 91r384mp4 q 1r48mp2 m2K p 2
Loops
Ž 16p 2 .
y11r96mp2 m 2K q 5r24m4K p 2 q 49r96m4K .

1
q
16p 2
Ž Ž3r4mp2 q 3m2K . mK y m2K mh
q Ž y2r9mp4 q 16r9mp2 m2K y 32r9m 4K . Lr1
q Ž y29r9mp4 q 16r9mp2 m 2K y 68r9m4K . Lr2
q Ž y10r9mp4 q 8r9mp2 m2K y 19r9m 4K . Lr3
q Ž 12 mp2 m 2K q 24 m4K . Lr4 q 12 m4K Lr5
q Ž y24mp2 m 2K y 48 m4K . Lr6
y24m4K Lr8 . q 9r8 mp2 q 48 mp2 mp Lr1 q 12 mp2 mp Lr2
q 12 mp2 mp Lr3 y 24mp2 mp Lr4 y 4 mp2 mp Lr5
q Ž 3 q 3r2 mp2 my2 2 2 r 2 r
K . m K y 2 m K mh q 64 m K m K L1 q 16 m K m K L 2
q 28m 2K m K Lr3 q Ž 36 mp2 q 40 m2K . m K Lr4

q Ž 4 mp2 q 56r3m 2K . m K Lr5 q Ž y48mp2 y 96 m 2K . m K Lr6
y 48m 2K m K Lr8 q 9r8 mh2 q Ž y8mp2 q 32 m2K . mh Lr1
q Ž y8mp2 q 32 m2K . mh Lr2 q Ž y4mp2 q 16 m2K . mh Lr3
q Ž 8r3mp2 y 32r3m2K . mh Lr4 q Ž 28r9mp2 y 64r9m2K . mh Lr5
q Ž y8mp4 y 32 mp2 m2K y 32 m4K . Lr42 q Ž 16r3mp4 y 32r3mp2 m2K
y128r3m4K . Lr4 Lr5 q Ž y8r9mp4 q 64r9mp2 m2K y 128r9m4K . Lr52
X
q 1r12 H F Ž mp2 ,mp2 ,mh2 ;mh2 . mp4
X
q 1r16 H F Ž mp2 ,m 2K ,m 2K ;mh2 . mp4
X
q 3r4H F Ž mp2 ,m2K ,m 2K ;mh2 . mp2 m 2K
y 9r16 H F Ž mp2 ,m2K ,m 2K ;mh2 . mp2
G. Amoros 347
X
y 1r4H F Ž m2K ,m 2K ,mh2 ;mh2 . mp4
X
q 2 H F Ž m2K ,m 2K ,mh2 ;mh2 . mp2 m2K
X
y 32r9H F Ž m2K ,m2K ,mh2 ;mh2 . m 4K
X
q 1r16 H F Ž mh2 ,m 2K ,m 2K ;mh2 . mp4
X X
y 1r2 H F Ž mh2 ,m2K ,m2K ;mh2 . mp2 m2K q H F Ž mh2 ,m2K ,m2K ;mh2 . m4K
q 3r16 H F Ž mh2 ,m 2K ,m2K ;mh2 . mp2 y 3r4H F Ž mh2 ,m2K ,m2K ;mh2 . m2K
X
q49r972 H F Ž mh2 ,mh2 ,mh2 ;mh2 . mp4
X
y 56r243H F Ž mh2 ,mh2 ,mh2 ;mh2 . mp2 m 2K
X
q 64r243H F Ž mh2 ,mh2 ,mh2 ;mh2 . m4K
X
q 1r2 H1F Ž mp2 ,m2K ,m2K ;mh2 . mp4
X
y 2 H1F Ž mp2 ,m2K ,m 2K ;mh2 . mp2 m2K
X
q H1F Ž m 2K ,m 2K ,mh2 ;mh2 . mp4
X
y 20r3H1F Ž m 2K ,m2K ,mh2 ;mh2 . mp2 m 2K
X
q 32r3H1F Ž m 2K ,m2K ,mh2 ;mh2 . m 4K
X
F
q 3r16 H21 Ž mp2 ,m2K ,m2K ;mh2 . mp4
X
F
y 3 fsr2 H21 Ž mp2 ,m2K ,m2K ;mh2 . mp2 m2K
X X
F
q 3 H21 Ž mp2 ,m2K ,m2K ;mh2 . m4K q 3r16 H21F Ž mh2 ,m2K ,m2K ;mh2 . mp4
X
F
y 3r2 H21 Ž mh2 ,m2K ,m2K ;mh2 . mp2 m2K
X
F
q 3 H21 Ž mh2 ,m2K ,m2K ;mh2 . m4K . Ž A.20 .
Appendix B. Axial-vector two-point functions
B.1. The remainder transÕerse part
The remaining transverse pieces are given by

Ž4 . Ž6 . Ž6.
Pˆ AŽ1. 2 ˆ Ž1. 2
a Ž q . s Ž PA a Ž q . . q Ž Pˆ AŽa1 . Ž q 2 . . CT q Ž Pˆ AŽa1 . Ž q 2 . . loops . Ž B.1 .
as p , K ,h
The O Ž p 4 . contribution is the same for all three two-point functions

Ž4 . Ž4 . Ž4 .
Ž Pˆ AŽp1 . Ž q 2 . . s Ž Pˆ AŽK1 . Ž q 2 . . s Pˆ AŽh1 . Ž q 2 . ž / s 4 Lr10 y 8 H1r . Ž B.2 .
6.
The O Ž p contribution can be written in a somewhat simpler fashion by using the
function
X
H T Ž m12 ,m22 ,m23 ;q 2 ;m24 . s 3m44 H21
F
Ž m12 ,m22 ,m23 ;m24 . y q 2 H21F Ž m12 ,m22 ,m23 ;q 2 . .
Ž B.3 .
G. Amoros
The isospin remainder is

Ž6 .
Ž q 2Pˆ AŽp1. Ž q 2 . . CT s y8mp4 C91r y 32 mp2 C12r q 2 y 32 mp2 C13r q 2 y 32 mp2 C80r q 2
y 32 mp2 C81
r
q 2 y 64 m2K C13
r
q 2 y 64 m2K C81
r
q2
r
y 16C87 q 4 y 8C93
r
q4 , Ž B.4 .
Ž6 . 1
Ž q 2 Fp2 Pˆ AŽp1 . . loops s 2 Ž 3r32 mp4 q 1r6mp2 q 2p 2 q 1r3mp2 q 2
Ž 16p 2 .
q1r12 m2K q 2p 2 y 5r24m2K q 2 q 3r32 q 4 .

1
q
16p 2
Ž 2 q 2mp q q 2m K . q 4 my2 2 2 r 2
p q mp y 32 L10 q mp
X
2 2 r 2 4 F 2 2 2 2
q 2 my2
K q m K y 16 L10 q m K q 5r3mp H Ž mp ,mp ,mp ;mp .
y 2 mp2 H F Ž mp2 ,mp2 ,mp2 ;mp2 . q 2 mp2 H F Ž mp2 ,mp2 ,mp2 ;q 2 .

X
y 5r4mp4 H F Ž mp2 ,m 2K ,m 2K ;mp2 .
q 1r4mp2 H F Ž mp2 ,m2K ,m2K ;mp2 .
y 1r4mp2 H F Ž mp2 ,m2K ,m2K ;q 2 .
X
q 1r9mp4 H F Ž mp2 ,mh2 ,mh2 ;mp2 .
X
q 2 mp2 m 2K H F Ž m2K ,mp2 ,m2K ;mp2 .
y 2 m2K H F Ž m2K ,mp2 ,m2K ;mp2 .
q 2 m2K H F Ž m2K ,mp2 ,m2K ;q 2 .
X
y5r3mp4 H F Ž m2K ,m2K ,mh2 ;mp2 .
X
q mp2 H F Ž mh2 ,m 2K ,m2K ;mp2 . y H F Ž mh2 ,m2K ,m2K ;mp2 .
ž
qH F Ž mh2 ,m2K ,m 2K ;q 2 . Ž y1r4mp2 q m2K . /
X X
q 2 mp4 H1F Ž mp2 ,m 2K ,m2K ;mp2 . q 4 mp4 H1F Ž m2K ,m2K ,mh2 ;mp2 .
q 2 H T Ž mp2 ,mp2 ,mp2 ;q 2 ;mp2 .
y 1r4H T Ž mp2 ,m2K ,m2K ;q 2 ;mp2 .
q 2 H T Ž m2K ,mp2 ,m2K ;q 2 ;mp2 .
q 3r4H T Ž mh2 ,m2K ,m2K ;q 2 ;mp2 . . Ž B.5 .
The remainder for the kaon two-point function is
Ž6 .
Ž q 2Pˆ AŽK1. Ž q 2 . . CT s y32 mp2 C13r q 2 y 32 mp2 C81r q 2 y 8 m4K C91r y 32 m2K C12r q 2
y 64 m2K C13
r
q 2 y 32 m 2K C80
r
q 2 y 64m2K C81
r
q2
r
y 16C87 q 4 y 8C93
r
q4 , Ž B.6 .
G. Amoros 349
Ž6 . 1
Ž q 2 Fp2 Pˆ AŽK1. Ž q 2 . . loops s 2 Ž 1r24mp2 q 2p 2 y 5r48mp2 q 2 q 3r32 m4K
Ž 16p 2 .
q5r24m 2K q 2p 2 q 11r48m2K q 2 q 3r32 q 4 .

1
q
16p 2
Ž 3r4 q 2mp q 3r2 q 2mK q 3r4 q 2mh .
q 3r2 mpy2 q 2mp2 y 12 Lr10 q 2mp q 3my2 2 2
K q mK
y 24 Lr10 q 2m K q 3r2 mhy2 q 2mh2 y 12 Lr10 q 2mh

X
q 3r4m2K Ž 2 mp2 q m 2K . H F Ž mp2 ,mp2 ,m2K ;m2K .
y 3r2 mp2 H F Ž mp2 ,mp2 ,m2K ;m 2K .
q 3r2 mp2 H F Ž mp2 ,mp2 ,m2K ;q 2 .
X
q 1r2 m4K H F Ž mp2 ,m2K ,mh2 ;m 2K .
X
y 3r16m4K H F Ž m 2K ,mp2 ,mp2 ;m 2K .
q 3r16m2K H F Ž m2K ,mp2 ,mp2 ;m2K .
y 3r16m2K H F Ž m2K ,mp2 ,mp2 ;q 2 .
X
q 9r8m4K H F Ž m 2K ,mp2 ,mh2 ;m2K .
y 9r8m2K H F Ž m2K ,mp2 ,mh2 ;m2K .
q 9r8m2K H F Ž m2K ,mp2 ,mh2 ;q 2 .
X
q3r2 m4K H F Ž m 2K ,m 2K ,m 2K ;m 2K .
y 3r2 m2K H F Ž m 2K ,m2K ,m2K ;m2K .
q 3r2 m2K H F Ž m 2K ,m2K ,m2K ;q 2 .
X
q 181r144m 4K H F Ž m2K ,mh2 ,mh2 ;m2K .
y 9r16m2K H F Ž m2K ,mh2 ,mh2 ;m2K .
q 9r16m2K H F Ž m2K ,mh2 ,mh2 ;q 2 .
X
y 3m 4K H1F Ž mp2 ,mp2 ,m2K ;m2K .
X
y 3m 4K H1F Ž m 2K ,mp2 ,mh2 ;m2K .
X
y 5r2 m4K H1F Ž m2K ,mh2 ,mh2 ;m2K .
q 3r2 H T Ž mp2 ,mp2 ,m 2K ;q 2 ;m2K .
y 3r16 H T Ž m 2K ,mp2 ,mp2 ;q 2 ;m2K .
q 9r8 H T Ž m2K ,mp2 ,mh2 ;q 2 ;m2K .
q 3r2 H T Ž m2K ,m 2K ,m2K ;q 2 ;m 2K .
q 9r16 H T Ž m 2K ,mh2 ,mh2 ;q 2 ;m2K . . Ž B.7 .
G. Amoros
And finally the remainder for the hypercharge is

Ž6 .
ž q Pˆ
2 Ž1.
Ah Ž q 2 . / CT s 256r3mp4 C37r y 8 mp4 C91r y 512r3mp2 m2K C37r
q 64r3mp2 m2K C91
r
q 32r3mp2 C12
r
q 2 y 32 mp2 C13
r
q2
q 32r3mp2 C80
r
q 2 y 32 mp2 C81
r
q 2 q 256r3m 4K C37
r
y 64r3m 4K C91
r
y 128r3m 2K C12
r
q 2 y 64m2K C13
r
q2
y 128r3m2K C80
r
q 2 y 64 m2K C81
r
q 2 y 16C87
r
q 4 y 8C93
r
q4 ,
Ž B.8 .
Ž6 . 1
2
žq F 2 ˆ
Ž1.
p P Ah Ž q 2 . / loops s Ž 1r96mp4 y 1r12 mp2 m2K y 1r4mp2 q 2
2 2
Ž 16p .
q1r6m4K q 1r4m2K q 2p 2 q 3r8m2K q 2 q 3r32 q 4 .

1
q 3q 2m K q 6 my2 2 2 r 2
K q m K y 48 L10 q m K
16p 2
X
q 1r3mp4 H F Ž mp2 ,mp2 ,mh2 ;mh2 . q Ž mp4 q 9r4mp2 mh2 .
X
=H F Ž mp2 ,m2K ,m2K ;mh2 . y 9r4mp2 H F Ž mp2 ,m2K ,m2K ;mh2 .
q 9r4mp2 H F Ž mp2 ,m2K ,m2K ;q 2 . q Ž ymp4 q 8 mp2 m 2K

X
y128r9m4K q 9r4mh4 . H F Ž m2K ,m2K ,mh2 ;mh2 .
y9r4mh2 H F Ž mh2 ,m2K ,m2K ;mh2 . y H F Ž mh2 ,m2K ,m2K ;q 2 .

ž /
q Ž 49r243mp4 y 224r243mp2 m2K q 256r243m4K .
X X
=H F Ž mh2 ,mh2 ,mh2 ;mh2 . y 6 mp2 mh2 H1F Ž mp2 ,m2K ,m2K ;mh2 .
X
q 4 Ž y3mp2 q 8 m2K . mh2 H1F Ž m2K ,m2K ,mh2 ;mh2 .
q 9r4H T Ž mp2 ,m 2K ,m2K ;q 2 ;mh2 .
q 9r4H T Ž mh2 ,m 2K ,m2K ;q 2 ;mh2 . . Ž B.9 .
B.2. The remainder longitudinal part
In that case the O Ž p 4 . contribution vanishes so we have that

Ž6 . Ž6 .
Pˆ AŽ0. 2 ˆ Ž 0. 2 ˆ Ž 0. 2
a Ž q . s Ž P A a Ž q . . CT q Ž P A a Ž q . . loops . Ž B.10 .
as p , K ,h
G. Amoros 351
The expansion of the resummed self-energy around the relevant pseudoscalar mass
leads in general to rather high derivatives and produces naturally the combinations
1
HiL Ž m12 ,m22 ,m23 ;q 2 ;m24 . s 2 Ž
q 2 HiF Ž m12 ,m 22 ,m 23 ;q 2 .
Ž q 2 y m24 .
yq 2 HiF Ž m12 ,m22 ,m23 ;m24 . y m24 Ž q 2 y m 24 .
X
=HiF Ž m12 ,m22 ,m23 ;m24 . . Ž B.11 .
for HiF s H F, H1F, H21
F4
and
1
H M Ž m12 ,m22 ,m23 ;q 2 ;m 24 . s 2 Ž H F Ž m12 ,m22 ,m23 ;q 2 .
2
Žq y m24 .
yH F Ž m12 ,m22 ,m23 ;m 24 . y Ž q 2 y m24 .
X
=H F Ž m12 ,m22 ,m23 ;m24 . . . Ž B.12 .
2
All of these functions are regular at q s m24 .
The longitudinal isospin remainder is
Ž6 .
Ž q 2Pˆ AŽp0 . Ž q 2 . . CT s 8 mp4 C91r , Ž B.13 .
Ž6 . y3r32 mp4
Ž q 2 Fp2 Pˆ AŽp0. Ž q 2 . . loops s 2 2
q 1r3H L Ž mp2 ,mp2 ,mp2 ;q 2 ;mp2 . mp4
Ž 16p .
y 2 H M Ž mp2 ,mp2 ,mp2 ;q 2 ;mp2 . mp6
q H L Ž mp2 ,m2K ,m 2K ;q 2 ;mp2 . mp4
q 1r4H M Ž mp2 ,m 2K ,m2K ;q 2 ;mp2 . mp6
y 1r9H L Ž mp2 ,mh2 ,mh2 ;q 2 ;mp2 . mp4
y 2 H M Ž m 2K ,mp2 ,m 2K ;q 2 ;mp2 . mp4 m2K
q 5r3H L Ž m 2K ,m2K ,mh2 ;q 2 ;mp2 . mp4
y 3r4H M Ž mh2 ,m 2K ,m 2K ;q 2 ;mp2 . mp4 mh2
y 2 H1L Ž mp2 ,m2K ,m 2K ;q 2 ;mp2 . mp4
y 4 H1L Ž m2K ,m 2K ,mh2 ;q 2 ;mp2 . mp4

L
y 6 H21 Ž mp2 ,mp2 ,mp2 ;q 2 ;mp2 . mp4
L
q 3r4H21 Ž mp2 ,m2K ,m2K ;q 2 ;mp2 . mp4
L
y 6 H21 Ž m2K ,mp2 ,m2K ;q 2 ;mp2 . mp4
L
y 9r4H21 Ž mh2 ,m2K ,m2K ;q 2 ;mp2 . mp4 . Ž B.14 .
G. Amoros
The kaon remainder is

Ž6 .
Ž q 2Pˆ AŽK0. Ž q 2 . . CT s 8 m4K C91r , Ž B.15 .
Ž6 . y3r32 m4K
Ž q 2 Fp2 Pˆ AŽK0. Ž q 2 . . loops s 2
y 3r2 H M Ž mp2 ,mp2 ,m2K ;q 2 ;m2K . mp2 m4K
Ž 16p 2 .
q 3r16 H M Ž m2K ,mp2 ,mp2 ;q 2 ;m2K . m6K
y 3r4H L Ž mp2 ,mp2 ,m2K ;q 2 ;m2K . m4K
y 1r2 H L Ž mp2 ,m2K ,mh2 ;q 2 ;m2K . m4K
y 9r8 H M Ž m2K ,mp2 ,mh2 ;q 2 ;m2K . m6K
y 3r2 H M Ž m 2K ,m2K ,m2K ;q 2 ;m2K . m6K

y 25r36 H L Ž m 2K ,mh2 ,mh2 ;q 2 ;m2K . m4K
y 9r16 H M Ž m 2K ,mh2 ,mh2 ;q 2 ;m2K . m6K
q 3 H1L Ž mp2 ,mp2 ,m2K ;q 2 ;m2K . m4K

q 3 H1L Ž m 2K ,mp2 ,mh2 ;q 2 ;m2K . m4K
q5r2 H1L Ž m 2K ,mh2 ,mh2 ;q 2 ;m2K . m 4K

L
y 9r2 H21 Ž mp2 ,mp2 ,m2K ;q 2 ;m2K . m4K
L
q 9r16 H21 Ž m2K ,mp2 ,mp2 ;q 2 ;m2K . m4K
L
y 27r8 H21 Ž m2K ,mp2 ,mh2 ;q 2 ;m2K . m4K
L
y 9r2 H21 Ž m2K ,m2K ,m2K ;q 2 ;m2K . m4K
L
y 27r16 H21 Ž m2K ,mh2 ,mh2 ;q 2 ;m2K . m4K . Ž B.16 .
Finally the hypercharge remainder is
Ž6 .
ž q Pˆ
2 Ž 0.
Ah Ž q 2 . / CT s y256r3mp4 C37r q 8 mp4 C91r q 512r3mp2 m2K C37r
y 64r3mp2 m2K C91
r
y 256r3m4K C37
r
q 64r3m4K C91
r
, Ž B.17 .
Ž6 . 1
2
žq F 2 ˆ
Ž 0.
p P Ah Ž q 2 . / loops s 2 Ž y1r96mp4 q 1r12 mp2 m2K y 1r6m4K .
Ž 16p 2 .
y 1r3H L Ž mp2 ,mp2 ,mh2 ;q 2 ;mh2 . mp4
y H L Ž mp2 ,m 2K ,m2K ;q 2 ;mh2 . mp4
y 9r4H M Ž mp2 ,m2K ,m 2K ;q 2 ;mh2 . mp2 mh4
q H L Ž m2K ,m2K ,mh2 ;q 2 ;mh2 . Ž mp4 y 8 mp2 m2K q 128r9m4K .

G. Amoros 353
y 9r4H M Ž mh2 ,m2K ,m2K ;q 2 ;mh2 . mh6
q 1r243H L Ž mh2 ,mh2 ,mh2 ;q 2 ;mh2 . Ž y49mp4 q 224mp2 m2K
y256m 4K . q 6 H1L Ž mp2 ,m2K ,m 2K ;q 2 ;mh2 . mp2 mh2
q H1L Ž m2K ,m2K ,mh2 ;q 2 ;mh2 . 4 Ž 3mp2 y 8 m 2K . mh2

L
y 27r4H21 Ž mp2 ,m2K ,m2K ;q 2 ;mh2 . mh4
L
y 27r4H21 Ž mh2 ,m2K ,m2K ;q 2 ;mh2 . mh4 . Ž B.18 .
Appendix C. Loop integrals
We use dimensional regularization here throughout in d dimensions with d s 4 y 2 e .
C.1. One-loop integrals
We need integrals with one, two and three propagators in principle. These we define
by
1 d dq 1
A Ž m2 . s H d
. Ž C.1 .
i Ž 2p . q y m2
2
We also use below

1 d dq 1
A Ž n,m2 . s H d n , Ž C.2 .
i Ž 2p . Žq 2
y m2 .
which can be obtained by derivation w.r.t. m2 of AŽ m2 ..
The two propagator integrals we encounter are
1 d dq 1
B Ž m12 ,m 22 , p 2 . s H d
,
i Ž 2p . Ž q 2
y m12 . Ž Ž q y p . 2 y m22 .
1 d dq qm
Bm Ž m12 ,m22 , p . s H d
s pm B1 Ž m12 ,m22 , p 2 . ,
i Ž 2p . Ž q 2
y m12 . Ž Ž q y p . 2 y m22 .
1 d dq qm qn
Bmn Ž m12 ,m 22 , p .s i H d
Ž 2p . Ž q 2 y m12 . Ž Ž q y p . 2 y m22 .
s pm pn B21 Ž m12 ,m 22 , p 2 . q gmn B22 Ž m12 ,m22 , p 2 . . Ž C.3 .
All the cases with three propagator integrals that show up can be absorbed into the
two-propagator ones by moving to the real masses rather than the lowest order masses.
This provided in fact a consistency check on the calculations.
G. Amoros
The explicit expressions are well known,

m2
A Ž m2 . s
16p 2
l0 y ln Ž m2 . q e Ž 12 C 2 q 12 q 121 p 2 q 12 ln2 Ž m2 . y C ln Ž m2 . . 4
qO Ž e 2 . ,
1 m12 ln Ž m12 . y m 22 ln Ž m 22 .
B Ž m12 ,m 22 , p 2 . s
16p 2 ž l0 y
m12 y m22 / q J Ž m12 ,m22 , p 2 .
qO Ž e . ,
1 1 m12 x q m22 Ž 1 y x . y x Ž 1 y x . p 2
J Ž m12 ,m 22 , p 2 . s y
16p 2
H0 dx ln ž m12 x q m22 Ž 1 y x . / , Ž C.4 .
C s lnŽ4p . q 1 y g and l0 s 1re q C. The function J Ž m12 ,m22 , p 2 . develops an imagi-

nary part for p 2 G Ž m1 q m 2 . 2 . Using D s m12 y m 22 , S s m12 q m22 and n 2 s p 4 q m14 q
m42 y 2 p 2 m12 y 2 p 2 m22 y 2 m12 m 22 it is given by
2
D S m12 Ž p2 q n . y D2
n
2
Ž 32p . J Ž m12 ,m22 , p 2 . s2q ž y
p2
q
D / ln 2 y 2 ln
m2 p 2
Ž p2 y n . y D2
.
Ž C.5 .
The two-propagator integrals can all be reduced to B and A via
1
B1 Ž m12 ,m 22 , p 2 . s y Ž A Ž m12 . y A Ž m22 . q Ž m22 y m12 y p 2 . B Ž m12 ,m22 , p 2 . . ,
2 p2
1
B22 Ž m12 ,m 22 , p 2 . s Ž A Ž m22 . q 2 m12 B Ž m12 ,m22 , p 2 .
2 Ž d y 1.
q Ž m22 y m12 y p 2 . B1 Ž m12 ,m22 , p 2 . . ,
1
B21 Ž m12 ,m22 , p 2 . s Ž A Ž m22 . q m12 B Ž m12 ,m22 , p 2 . y dB22 Ž m12 ,m22 , p 2 . . . Ž C.6 .
p2
The basic method used here is the one from Passarino and Veltman w30x.
C.2. Sunset integrals
In this appendix we discuss the nontrivial two-loop integrals that show up in this
calculation. They have been treated in several places already, in general and for various
special cases. We use here a method that is a hybrid of various other approaches. We
only cite the literature actually used. We define
1 d dq ddr X
²² X :: s 2 H Ž 2p . d d
,
i Ž 2p . Žq 2
y m12 .Ž r 2
y m 22 . Ž Ž q q r y p . 2 y m23 .
Ž C.7 .
G. Amoros 355
H Ž m12 ,m22 ,m23 ; p 2 . s ²²1:: ,
Hm Ž m12 ,m 22 ,m 23 ; p 2 . s ²² qm :: s pm H1 Ž m12 ,m22 ,m23 ; p 2 . ,
Hmn Ž m12 ,m22 ,m23 ; p 2 . s ²² qm qn ::s pm pn H21 Ž m12 ,m 22 ,m 23 ; p 2 .
q gmn H22 Ž m12 ,m 22 ,m 23 ; p 2 . . Ž C.8 .

By redefining momenta the others can be simply related to the above three. In particular
²² rm :: s pm H1 Ž m22 ,m12 ,m 23 ; p 2 . ,
²² rm rn :: s pm pn H21 Ž m22 ,m12 ,m 23 ; p 2 . q gmn H22 Ž m 22 ,m12 ,m 23 ; p 2 . ,
²² qm rn :: s ²² rm qn :: ,
²² qm rn :: s pm pn H23 Ž m12 ,m22 ,m23 ; p 2 . q gmn H24 Ž m12 ,m 22 ,m 23 ; p 2 . , Ž C.9 .

with
2 H23 Ž m12 ,m22 ,m23 ; p 2 . s yH21 Ž m12 ,m22 ,m23 ; p 2 . y H21 Ž m22 ,m12 ,m23 ; p 2 .
q H21 Ž m23 ,m12 ,m22 ; p 2 . q 2 H1 Ž m12 ,m22 ,m23 ; p 2 .
q 2 H1 Ž m22 ,m12 ,m 23 ; p 2 . y H Ž m12 ,m 22 ,m23 ; p 2 . ,
2 H24 Ž m12 ,m22 ,m23 ; p 2 . s yH22 Ž m12 ,m 22 ,m23 ; p 2 . y H22 Ž m22 ,m12 ,m23 ; p 2 .
q H22 Ž m23 ,m12 ,m22 ; p 2 . . Ž C.10 .

The first two follow from interchanging q and r and the third from the fact that it is
proportional to gmn or pm pn , which are both symmetric in m and n . The last one
follows from
Ž qm rn q rm qn . s Ž qm q rm y pm . Ž qn q rn y pn . y qm qn y rm rn y pm pn
q 2 pm Ž qn q rn . q 2 pn Ž qm q rm . Ž C.11 .
and redefining momenta and masses on the r.h.s. In addition we have the relation
p 2 H21 Ž m12 ,m22 ,m23 ; p 2 . q dH22 Ž m12 ,m22 ,m23 ; p 2 .
s m12 H Ž m12 ,m22 ,m23 ; p 2 . q A Ž m22 . A Ž m23 . , Ž C.12 .

which allows to express H22 in a simple way in terms of H21. There is also a relation
between H1 and H,
H1 Ž m12 ,m22 ,m23 ; p 2 . q H1 Ž m22 ,m12 ,m23 ; p 2 . q H1 Ž m23 ,m12 ,m22 ; p 2 .
s H Ž m12 ,m22 ,m23 ; p 2 . , Ž C.13 .

which allows to write H1Ž m 2 ,m2 ,m2 ; p 2 . s 1r3H Ž m2 ,m2 ,m2 ; p 2 . in the case of equal
masses. The function H is fully symmetric in m12 ,m22 and m23 , while H1 , H21 and H22
are symmetric under the interchange of m 22 and m23 .
G. Amoros
We do not explicitly evaluate the integrals analytically. H, H1 and H21 are all finite
after two subtractions. We therefore evaluate them as follows Ž Hi stands for H, H1 and
H21 .
E
Hi Ž m12 ,m22 ,m23 ; p 2 . s Hi Ž m12 ,m 22 ,m 23 ;0 . q p 2 Hi Ž m12 ,m 22 ,m 23 ;0 .
E p2
q Hi Ž m12 ,m22 ,m23 ; p 2 . . Ž C.14 .

The functions Hi are finite in 4 dimensions and can be evaluated by their dispersive
representation w31,32x or below threshold by the methods of Ref. w32x.
The value at zero and its derivative there have been derived essentially using the
methods of Ref. w33x except that we use a slightly simpler procedure than the recursion
relations given there.
First we define the intermediate integrals
I Ž n1 ,n 2 ,n 3 .
1 d dq ddr 1
s 2 H Ž 2p . d d n1 n2 n3 , Ž C.15 .
i Ž 2p . Ž q 2 y m12 . Ž r 2 y m22 . Ž Ž q q r . 2 y m23 .
which show up in the momentum expansion of Hi . The I Ž n1 ,n 2 ,n 3 . with one of the
n i s 0 are separable and are e.g.
I Ž n1 ,n 2 ,0 . s A Ž n1 ,m12 . A Ž n 2 ,m22 . . Ž C.16 .

All the others can be derived by taking derivatives of I Ž1,1,1. w.r.t. the masses m12 ,m22
and m23 . The function I Ž1,1,1. is taken from Ref. w33x, note that our definition differs by
overall factors from theirs,
I Ž 1,1,1 .
y1 G 2 Ž1qe . 2e
1
s
Ž 16p . 2 2 2Ž 1 y e . Ž 1 y 2 e .
Ž 4p . ½ y
e2
Ž m12 q m22 q m23 .
2
q
e
Ž m12 ln 1 q m22 ln 2 q m23 ln 3 . q m12 Ž ln 2 ln 3 y ln 1Ž ln 1 q ln 2 q ln 3 . .
qm22 Ž ln 3 ln 1 y ln 2 Ž ln 1 q ln 2 q ln 3 . . q m23 Ž ln 1 ln 2 y ln 3 Ž ln 1 q ln 2 q ln 3 . .
qC Ž m12 ,m22 ,m23 . q O Ž e . . 5 Ž C.17 .
In ŽC.17. we used ln i s lnŽ m2i . and the function C Ž m12 ,m22 ,m 23 .. The expression for C is
somewhat dependent on the relation between the various masses. Using
2
l m s l Ž m12 ,m22 ,m23 . s Ž m12 y m22 y m23 . y 4 m22 m23 , Ž C.18 .
G. Amoros 357
we have for the case l m F 0 w33x

C Ž m12 ,m22 ,m23 .
(
s 2 y l m Cl 2 Ž 2arccos z 1 . q Cl 2 Ž 2arccos z 2 . q Cl 2 Ž 2arccos z 3 . 4 , Ž C.19 .
with
ym12 q m22 q m23 ym22 q m23 q m12 ym23 q m12 q m22
z1 s , z2 s , z3 s .
2 m2 m3 2 m 3 m1 2 m1 m 2
Ž C.20 .
The case m1 q m 2 F m 3 , with l m G 0, is
C Ž m12 ,m22 ,m23 .

m12 m22 p2
( ½
s y l m 2ln x 1 ln x 2 y ln
m32
ln
m23
q
3
y 2Li 2 Ž x 1 . y 2Li 2 Ž x 2 . , 5
Ž C.21 .
with
m 23 q m12 y m22 y l m ( m 23 q m 22 y m12 y l m (
x1 s , x2 s . Ž C.22 .
2 m23 2 m 23
The cases m1 q m 3 F m 2 and m 2 q m 3 F m1 can be obtained from the last one by
relabelling masses. Li 2 Ž x . is the dilogarithm defined by
1 dt
Li 2 Ž x . s y H0 ln Ž 1 y xt . , Ž C.23 .
t
and Cl 2 Ž x . is Clausen’s function defined by
x t ix
px x2
Cl 2 Ž x . s y H0 dt ln 2sin 2 s yi Ž Li 2 Ž e . y Li 2 Ž 1 . . y i qi . Ž C.24 .
2 4
Notice that C Ž m12 ,m22 ,m 23 . is fully symmetric w.r.t. the three masses. The I Ž n1 ,n 2 ,n 3 .
for general n i can be obtained by taking derivatives of I Ž1,1,1.. The relation
E
C Ž m12 ,m22 ,m23 .
E m12
m12 y m22 y m23 m14 m 22 y m23 m 22
s C Ž m12 ,m22 ,m23 . y ln q ln , Ž C.25 .
lm m22 m23 m12 m23
allows an easy evaluation of all needed derivatives and is equivalent to the recursion
relations used in Ref. w33x.
In order to express the Hi functions at zero and the derivatives w.r.t. p 2 at zero the
easiest is to shift momenta to q˜ s q y p in the integral and then Taylor-expand using
i
1 1 y2 q˜ P p y p 2
2
Ž q˜ q p . y m12
s
q˜ 2 y m12
Ý
is0,`
ž q˜ 2 y m12 / . Ž C.26 .
G. Amoros
™
The integrals can then be done using q˜m qñ gmn q 2rd and equivalent identities for the
higher orders. We have run this procedure to higher orders then necessary to check the
cancellations of infinities there. This results in
H Ž m12 ,m22 ,m23 ;0 . s I Ž 1,1,1 . ,
E 4yd 4
2
H Ž m12 ,m 22 ,m 23 ;0 . s I Ž 2,1,1 . q m12 I Ž 3,1,1 . ,
Ep d d
d y 2 2
H1 Ž m12 ,m22 ,m23 ;0 . s I Ž 1,1,1 . y m12 I Ž 2,1,1 . . ,
d d
E 1
H Ž m2 ,m 2 ,m 2 ;0 . s Ž Ž 2 y d . Ž d y 4. I Ž 2,1,1 .
E p2 1 1 2 3 d Ž d q 2.
q8 Ž d y 4 . m12 I Ž 3,1,1 . y 24 m14 I Ž 4,1,1 . . ,
1
H21 Ž m12 ,m 22 ,m 23 ;0 . s Ž d Ž d y 2. I Ž 1,1,1 .
d Ž d q 2.
y4 Ž d y 2 . m12 I Ž 2,1,1 . q 8 m14 I Ž 3,1,1 . . ,
E
H21 Ž m12 ,m22 ,m23 ;0 .
E p2
1
s
d Ž d q 2. ½Ž yd 2 q 10 d y 48 . I Ž 2,1,1 .
q12 Ž d y 10 . m12 I Ž 3,1,1 . y 72 m14 I Ž 4,1,1 .

192
q
dq4
Ž I Ž 2,1,1 . q 3m12 I Ž 3,1,1 . q 3m14 I Ž 4,1,1 . q m16 I Ž 5,1,1 . . . 5
Ž C.27 .
Evaluating these expressions then leads to
2
Ž 16p 2 . H Ž m12 ,m22 ,m23 ;0 . s 12 l2 Ž m12 q m22 q m23 . q 12 l1 Ž m12 Ž 1 y 2ln 1 .
qm22 Ž 1 y 2ln 2 . q m23 Ž 1 y 2ln 3 . .
y 12 C Ž m12 ,m22 ,m23 . q m12 Ž 121 p 2 q 23 y ln 1
q 12 Ž yln 2 ln 3 q ln 1 ln 4 . . q m22 Ž 121 p 2 q 32 y ln 2
q 12 Ž yln 1 ln 3 q ln 2 ln 4 . . q m23 Ž 121 p 2 q 32 y ln 3
q 12 Ž yln 1 ln 2 q ln 3 ln 4 . . q O Ž e . , Ž C.28 .
2 E m12 m 22 m33
Ž 16p 2 . H Ž m12 ,m 22 ,m 23 ;0 . s y 14 l1 q C Ž m12 ,m22 ,m23 . q 18
E p2 l2m
m12 dm m22
q
2 lm
ln 1 q
2 lm
Ž m22 y m12 y m23 . ln 2
m23
q
2 lm
Ž m23 y m12 y m22 . ln 3 q O Ž e . , Ž C.29 .
G. Amoros 359
2 m22 q m23
Ž 16p 2 . H1 Ž m12 ,m22 ,m23 ;0 . s l2 q 18 l1 Ž 2 m12 q m22 Ž 1 y 4ln 2 .
4
m12 dm
qm23 Ž 1 y 4ln 3 . . q 14 y1 q
ž lm /
=C Ž m12 ,m 22 ,m 23 . q m12 Ž 38 y 12 ln 1 .
p2
q m 22
ž 24
q 169 y 14 ln 2 q 14 Ž ln 2 ln 4 y ln 1 ln 3 .
/
p2
q m 23
ž 24
q 169 y 14 ln 3 q 14 Ž ln 3 ln 4 y ln 1 ln 2 .
/
qO Ž e . , Ž C.30 .
2 E m14 m22 m23 dm
Ž 16p 2 . 2
H1 Ž m12 ,m22 ,m23 ;0 . s y 121 l1 q C Ž m12 ,m22 ,m23 .
Ep l3m
m12 dm m14 ln 1 m 22 m23

y
6 lm
q
6 lm ž 1 q 12
lm /
m22 ln 2
q
6 lm ž m 22 y m23 y 2 m12
6 m12 m23 m23 ln 3

y
lm
Ž dm q 2 m22 . / q
6 lm ž m23 y m22
6 m12 m22
y2 m12 y
lm
Ž dm q 2 m23 . / q 727 q O Ž e . ,
Ž C.31 .
2 2
m22 q m23
Ž 16p . H21 m12 ,m22 ,m23 ;0
Ž .s l2 q 361 l1 Ž 3m12 q m22 Ž 2 y 12ln 2 .
6
1
qm23 Ž 2 y 12ln 3 . . q Ž yl2m q l m m12 dm
6 l2m
q2 m14 m22 m 23 . C Ž m12 ,m22 ,m23 .
m12 ln 1 dm
q m12
ž 17
72 y 13 ln 1 q
6 lm /
p2 m12
q m22
ž 36
q 19 1
54 y 9 ln 2 y
6 lm
Ž dm q 2 m23 . ln 2
G. Amoros
p2
q 16 Ž ln 2 ln 4 y ln 1 ln 3 . q m23
/ ž 36
q 19 1
54 y 9 ln 3
m12
y
6 lm
Ž dm q 2 m22 . ln 3 q 16 Ž ln 3 ln 4 y ln 1 ln 2 . /
qO Ž e . , Ž C.32 .
2 E m12
Ž 16p 2 . H21 Ž m12 ,m22 ,m23 ;0 . s y 241 l1 q 288
17
y Ž dm q 2 m12 .
E p2 24l m
5m14 m 22 m23 m16 m22 m23 5m22 m23
y
6 l2m
q
l3m ž 1q
lm /
m16 dm m22 m23
=C Ž m12 ,m 22 ,m 23 . q
12 l2m ž
ln 1 1 q 30
lm /
m14 m 22 m43
q ln 3 52 Ž m23 y m12 y m22 .
l3m
m23 m12 m23
q ln 3 Ž m23 y m22 . q ln 3 Ž y3m14
12 l m 12 l2m
y12 m12 m22 q 5m12 m23 q 3m42 y m22 m23 y 2 m43 .

m14 m 23 m 42
q ln 2 52 Ž m22 y m12 y m 23 .
l3m
m22 m12 m22
q ln 2 Ž m22 y m23 . q ln 2 Ž y3m14
12 l m 12 l2m
y12 m12 m23 q 5m12 m22 q 3m43 y m22 m23 y 2 m42 .
qO Ž e . . Ž C.33 .
Here we used
ln 4 s ln 1 q ln 2 q ln 3 , dm s m12 y m22 y m23 ,
2
l2 s l20 q Ž ln Ž 4p . q 1 y g . , l1 s l0 q ln Ž 4p . q 1 y g ,
1
l0 s q ln Ž 4p . q 1 y g . Ž C.34 .
e
These are appropriate for MS subtraction as is customary in CHPT w34x.
Below threshold the methods of w32x lead to a two-integral representation of the finite
part
2 2 2 2
½ H , H , H 5 Ž m ,m ,m ; p .
1 21 1 2 3
m 22 m23
s
`
HŽ m qm .
2 3
2
ds (ž
l 1,
s s
,
/ =
1
H0 dx K Ž x , s , p
2
2
. 1, x , x 2 4 , Ž C.35 .
G. Amoros 361
with
K2 Ž x , s , p 2 .
1 m12 Ž 1 y x . q s x y x Ž 1 y x . p 2 p2 x Ž1 y x .
s
Ž 16p 2 .
2 ž ln
m12 Ž 1 y x . q s x
q
m12 Ž 1 y x . q s x /
Ž C.36 .
and
2
lŽ x , y, z . s Ž x y y y z . y 4 yz , Ž C.37 .
¨ ´ function.
the Kallen
The dispersive representation
B Ž m22 ,m23 ,t .
Ž dy3 .r2
p Ž1yd .r2 2 3y2 d ` m22 m23 1
s
G Ž Ž d y 1 . r2 .
H Ž m 2qm 3 . 2 žž
d s l 1,
s s
,
// s Ž dy4.r2
syt
,
Ž C.38 .
has been used here instead of the simpler case with equal masses used in Ref. w32x.
Above threshold, s G Ž m1 q m 2 q m 3 . 2 the functions Hi develop imaginary parts and
they can then be evaluated from their dispersive representation
s2 ` dz Im Hi Ž z .
Hi Ž m12 ,m22 ,m23 ; s . s H . Ž C.39 .
p Ž m qm qm .1 2 3
2 z2 zys
The imaginary parts are given by Žin d s 4.
2 2 2
½
Im H , H1 , H21 5 Ž m ,m ,m ; s .
1 2 3
y1 1 E1max E1 4 E12 y m12

s
16 Ž 2p . 3
Hm
1
dE1 1, ½ 's , 3s 5Ž E2max y E2min . , Ž C.40 .
with
1
E1max s
2's
Ž s q m12 y Ž m 2 q m 3 . 2 . ,
m223 s s q m12 y 2's E1 ,
1
E2max y E2min s
m223 's (lŽ s,m ,m . (lŽ m
2
1
2
23
2 2 2
23 ,m 2 ,m 3 . . Ž C.41 .
In the text we use mainly

X
the finite functions HiF Ž m12 ,m22 ,m 23 ;q 2 . and their derivatives
with respect to q , Hi m1 ,m22 ,m23 ;q 2 .. These correspond to using Eq. ŽC.14. and
2 F Ž 2
setting l2 and l1 to zero in Eqs. ŽC.28. – ŽC.33..

G. Amoros
Appendix D. Regularization and renormalization
In this paper we have employed the version of Modified Minimal Subtraction ŽMS.
that is customary in CHPT. The precise procedure has been discussed in great detail in
Ref. w34x.
The procedure used in Ref. w6x corresponds to subtracting only the l0 terms present
in all the integrals, including those in l1 and l 2 .
As mentioned in Ref. w34x a Taylor expansion of the p 4 coefficients introduces in
principle new parameters d i via the Laurent-expansion of the L i s a ire q LM S
i q ed i
q . . . We have checked that the terms involving d i take the form of a local action for
the quantities considered in this manuscript, thus they can be absorbed in the p 6 LECs
as proven in general in Ref. w24x.
We have defined
y2 e
y1 y2 e
y1
Li ' Ž m c . ž 2
32p e
Gi q Lri Ž m . s Ž m .
/ ž 32p 2
Gi l0 q Lri Ž m . q O Ž e . .
/
Ž D.1 .
In the main text we have suppressed the explicit m-dependence of the Lri . The
coefficients Gi are given in Ref. w19x and ln c s y1r2ŽlnŽ4p . y g q 1.. The order e
term in the last part of Eq. ŽD.1. has been used as well to check the explicit
cancellations of lnŽ4p . and g in all expressions.
Similarly the coefficients in the p 6 Lagrangian are used to absorb the remaining
infinities via
y4 e
g2i g1i
Ci ' Ž m c . 2 ž
q q Cir Ž m . /
e e
s my4 e Ž g 2 i l 2 q g 1 i l1 q Cir Ž m . q O Ž e . . . Ž D.2 .
Dropping the terms with l0 , l1 , l 2 , replacing the Ci by Cir
in the main text and
subtracting the terms proportional to C, l0 , l1 and l2 in the expressions for the
integrals given in the preceding appendices, gives the results in the MS scheme.
References
w1x S. Weinberg, Phys. Rev. Lett. 18 Ž1967. 507.

w2x T. Das, V.S. Mathur, S. Okubo, Phys. Rev. Lett. 19 Ž1967. 859.
w3x S. Narison, QCD spectral sum rules, World Scientific Lecture Notes in Physics, Vol. 26 ŽWorld
Scientific, Singapore, 1989..
w4x E. de Rafael, An introduction to sum rules in QCD: Course, Les Houches 1997, hep-phr9802448.
w5x J. Bijnens, U.-G. Meißner, Miniproceedings of the meeting on Chiral Effective Theories, Bad Honnef,
Germany, 30 Nov. – 4 Dec. 1998, hep-phr9901381.
w6x E. Golowich, J. Kambor, Nucl. Phys. B 447 Ž1995. 373, hep-phr9501318.
w7x E. Golowich, J. Kambor, Phys. Rev. D 58 Ž1998. 36004, hep-phr9710214.
w8x G. Ecker et al., Nucl. Phys. B 321 Ž1989. 311.
w9x G. Ecker et al., Phys. Lett. B 223 Ž1989. 425.
w10x J. Bijnens, G. Colangelo, G. Ecker, J. High Energy Phys. 2 Ž1999. 20, hep-phr9902437.
w11x E. Golowich, J. Kambor, Phys. Rev. D 53 Ž1996. 2651, hep-phr9509304.
w12x E. Golowich, J. Kambor, Phys. Rev. Lett. 79 Ž1997. 4092, hep-phr9707341.
G. Amoros 363
w13x G. Amoros,´ J. Bijnens, P. Talavera, hep-phr9912398.

w14x C. Vafa, E. Witten, Nucl. Phys. B 234 Ž1983. 173.
w15x G. Ecker, Chiral symmetry, Schladming 1998, hep-phr9805500.
w16x A. Pich, Effective field theory: Course, Les Houches 1997, hep-phr9806303.
w17x S. Weinberg, Physica A 96 Ž1979. 327.
w18x H. Leutwyler, Ann. Phys. ŽNY. 235 Ž1994. 165, hep-phr9311274.
w19x J. Gasser, H. Leutwyler, Nucl. Phys. B 250 Ž1985. 465.
w20x H.W. Fearing, S. Scherer, Phys. Rev. D 53 Ž1996. 315, hep-phr9408346.
w21x J. Gasser, H. Leutwyler, Ann. Phys. ŽNY. 158 Ž1984. 142.
w22x M. Ademollo, R. Gatto, Phys. Rev. Lett. 13 Ž1964. 264.
w23x R.E. Behrends, A. Sirlin, Phys. Rev. Lett. 4 Ž1960. 186.
w24x J. Bijnens, G. Colangelo, G. Ecker, hep-phr9907333.
w25x J. Bijnens, G. Colangelo, G. Ecker, Phys. Lett. B 441 Ž1998. 437, hep-phr9808421.
w26x J. Bijnens, G. Colangelo, J. Gasser, Nucl. Phys. B 427 Ž1994. 427, hep-phr9403390.
w27x J. Prades, Nucl. Phys. Proc. Suppl. 64 Ž1998. 253, hep-phr9708395.
w28x M. Jamin, Nucl. Phys. Proc. Suppl. 64 Ž1998. 250, hep-phr9709484.
w29x ´
V. Gimenez et al., Nucl. Phys. B 540 Ž1999. 472, hep-latr9801028.
w30x G. Passarino, M. Veltman, Nucl. Phys. B 160 Ž1979. 151.
w31x P. Post, J.B. Tausk, Mod. Phys. Lett. A 11 Ž1996. 2115, hep-phr9604270.
w32x J. Gasser, M. Sainio, Eur. Phys. J. C 6 Ž1999. 297, hep-phr9803251.
w33x A.I. Davydychev, J.B. Tausk, Nucl. Phys. B 397 Ž1993. 123.
w34x J. Bijnens et al., Nucl. Phys. B 508 Ž1997. 263, hep-phr9707291.
¨
The sphaleron rate: Bodeker’s leading log
Guy D. Moore
Department of Physics, UniÕersity of Washington, Seattle WA 98195-1560, USA
Received 15 October 1999; accepted 24 November 1999
Abstract
¨
Bodeker has recently shown that the high temperature sphaleron rate, which measures baryon
number violation in the hot standard model, receives logarithmic corrections to its leading
parametric behavior; G s k X wlogŽ m Drg 2 T . q O Ž1.xŽ g 2 T 2rm2D . a W
5 4
T . After discussing the physi-
cal origin of these corrections, I compute the leading log coefficient numerically: k X s 10.8 " 0.7.
The log is fairly small relative to the O Ž1. ‘‘correction;’’ so non-logarithmic contributions
dominate at realistic values of the coupling. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
It has been known for some time now that baryon number is not a conserved quantity
in the minimal standard model w1x. It is violated non-perturbatively because of the
anomaly, the chiral couplings of fermions to SUŽ2. weak, and the topologically
non-trivial vacuum structure of SUŽ2.. However, as is characteristic of a non-perturba-
tive process in a weakly coupled theory, the rate of violation is so tiny that it is
completely irrelevant phenomenologically. Certainly, if baryon number is also violated
due to high dimension operators descended from some GUT, baryon number violating
decay rates due to the GUT mechanism will greatly exceed the electroweak rate, of
order m1W expŽy4pra W . - 10y1 70 GeV even before accounting for additional suppres-
sion from powers of small CKM matrix elements and a high power of Ž m protonrmW ..
However, as first realized by Kuzmin, Rubakov, and Shaposhnikov in 1985, the
efficiency of standard model baryon number violation is very much higher at finite
temperature w2x. A perturbative estimate w3x based on a saddle-point expansion about
Klinkhamer and Manton’s sphaleron w4x indicated that the rate is more than enough to
erase any relic abundance of baryon number left over from the GUT scale, unless the
E-mail address: guymoore@phys.washington.edu ŽG.D. Moore..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 4 6 - 4
368 G.D. Moorer Nuclear Physics B 568 (2000) 367–404
combination B y L, baryon minus lepton number, is non-zero; this quantity is preserved

by electroweak physics. It also may be that the baryon number abundance in the
universe was produced by electroweak physics, which motivates the more careful study
of electroweak baryon number violation at finite temperature.
Recently our understanding of thermal baryon number violation has improved,
though it is not complete. We understand that the efficiency of baryon number violation
can in most relevant circumstances be related by a fluctuation dissipation argument to
the diffusion constant for Chern–Simons number w5–7x. This diffusion constant, called
the sphaleron rate, is defined as
² Ž NCS Ž t . y NCS Ž 0 . . 2 :
V™ ` t™ `
G ' lim lim
Vt
, Ž 1.
where the expectation values refer to a trace over the equilibrium thermal density matrix.
The quantity G is the topological susceptibility of the electroweak sector at finite
temperature, in Minkowski time.
We know that, when the electroweak phase transition is first-order, the value of G
jumps discontinuously between the phases; it is quite small in the broken phase and
much larger in the symmetric phase. An old power counting argument says that the
symmetric phase rate should be of order G ; a W4 T 4 with an order unity coefficient. This
argument relied, correctly, on the natural non-perturbative length scale in the hot plasma
being 1rŽ a W T .. One then assumes that the natural time scale is the same; on-dimen-
sional grounds the space-time rate of NCS diffusion must then be of order Ž a W T . 3a W T.
However, Arnold, Son, and Yaffe have shown that at leading parametric order, the
natural time scale is not 1rŽ a W T ., but 1ra W2 T w8x, up to possible logarithmic
corrections, which the authors did not consider. More recently, Bodeker ¨ has demon-
strated that logarithmic corrections to their argument do occur w9x.
This says nothing about the numerical value of the sphaleron rate; it could be
parametrically a W5 T 4 but numerically irrelevantly small. Ambjørn and Krasnitz pre-
sented numerical evidence that it was large, by considering classical, thermal Yang–Mills
theory on the lattice w10x. Their definition of NCS was not topological and could
therefore suffer from potentially severe lattice artifacts, but Turok and I studied the same
system with a topological definition of NCS and verified that G is substantial w11x. The
sphaleron rate we found, expressed in physical units, was lattice spacing dependent,
which turns out to be a prediction of the arguments of Arnold, Son, and Yaffe; the extra
power of a W arises from the interaction between the infrared fields and ultraviolet
excitations, as I will discuss more below, and on the lattice it becomes one power of the
lattice spacing a.
In fact the lattice spacing dependence of G for pure classical lattice Yang–Mills
theory only fits G A a if there are substantial corrections to scaling, which has led some
to call into question whether the Arnold, Son, and Yaffe’s analysis is correct. In this
paper I will assume that it is, as seems justified on theoretical grounds and numerical
evidence from classical Yang–Mills theory ‘‘enhanced’’ with added degrees of freedom
which reproduce the hard thermal loop effects w13x. I will return to the large corrections
to scaling in the classical lattice theory in Subsection 5.3.
Because the interactions between infrared and ultraviolet excitations are important to
setting the sphaleron rate, it is somewhat difficult to actually extract G at leading order
G.D. Moorer Nuclear Physics B 568 (2000) 367–404 369
in a W for the continuum quantum theory, at the physical value of a W or even in the
parametric small a W limit. Hu and Muller ¨ proposed a technique based on including the
UV physics, lost to the lattice regulation, by introducing new degrees of freedom which
influence the IR fields in the same way w12x. They implemented and applied the
technique jointly with me w13x. As I will discuss later, this technique still suffers from
some poorly controlled systematics, which are related to the logarithmic corrections
discovered by Bodeker;¨ in fact, beyond the leading log the infrared physics the
technique will simulate is not rotationally invariant.
A good first step to answering the remaining questions about the sphaleron rate is to
determine it at leading logarithmic order in a W . Bodeker
¨ has demonstrated that this can
be done within an effective theory which is completely UV well behaved; in fact it is
nothing but the Langevin equation for 3-D Yang–Mills theory w9x. The leading log
behavior is probably not very useful by itself, for estimating G at the physical value of
a W . Leading log expansions often miss large constant corrections; we know for instance
that the O Ž g 2 T . contribution to the Debye mass has a much larger constant contribution
than the leading log might suggest w14,15x. However, it is still useful to know the leading
log; for instance, its size is related to the severity of the systematic problems with the
method of Hu and Muller, ¨ and it might in principle be useful for extrapolating lattice
results which correspond to an inappropriate value of a W back to the correct value.
The purpose of this paper is to determine the coefficient of the leading log behavior
of G ; namely, to find k X defined through
mD g 2T 2
G s k X log
ž qO Ž 1 . / a W5 T 4 q Ž higher order . . Ž 2.
g 2T m2D
ŽMy logs are always natural logs.. Neglecting the O Ž1. means that logŽ1rg . is treated as
much larger than any order unity constant, a rather extreme interpretation of the
perturbative expansion. Probably this expansion is completely unjustified, but as I said
there are still important things to be learned from making it. Until Section 5 I will not
worry about whether the expansion in logŽ1rg . 4 1 is justified; the goal is simply to
determine k X . I will also work in Yang–Mills theory, which is appropriate at leading
order only for temperatures well above the equilibrium temperature. I will mention how
to include the Higgs field in Section 5.
A summary of the paper is as follows. Section 2 will discuss at an intuitive, physical,
but non-technical level why the rate has the parametric form I show; where the extra a
comes from and particularly why there is a log. The section provides two apparently
different arguments, one in terms of conductivities and scattering processes for hard
particles and one in terms of hard thermal loops ŽHTL’s. and Wilson lines; the two are
of course equivalent. Some salient details about the Wilson line are in Section 3, which
is more technical. This section also shows how the numerical model of Hu, Muller ¨ and
¨
Moore fails beyond leading log order. Section 4 studies Bodeker’s effective theory for
extracting the leading log, numerically. Since the Langevin dynamics are simple but
numerically costly, the emphasis is on controlling systematic errors. Many of the details
appear either in previous papers or Appendix A. For the reader’s ease I present the
answer now: k X s 10.8 " 0.7. The dominant error here is statistical; systematic errors
are completely under control. In Section 5 I discuss the meaning of the result, and try to
estimate what the sphaleron rate is for the realistic values of m 2D and g 2 by using this
¨
result to extrapolate the results of Hu, Muller, and Moore to the right value of the log.
My estimate is G , 20 a W5 T 4 for m2D s Ž11r6. g 2 T 2 and g 2 s 0.4; however there are
uncontrolled systematics which may be as large as 30%. I also discuss corrections to the
approximation that g < 1; the largest of these will be of order 10%. The conclusion
concludes. There is also a technical appendix which discusses the match between lattice
and continuum Langevin time scales; the match between continuum and Langevin time
scales is computed at O Ž a. and the match between lattice Langevin and heat bath time
scales is determined by a measurement.
5
2. The physics behind a W log(m D r g 2 T)
2.1. Note on the classical approximation
To a very good approximation the behavior of infrared fields in thermal Yang–Mills

or Yang–Mills Higgs theory at weak coupling is that of classical fields. The ‘‘old’’
parametric estimate, G A a W4 T 4 , relies on the fact that the only length scale available in
classical Yang–Mills theory is 1rg 2 T. It is also assumed that the only time scale is the
same, in which case any essentially infrared space-time rate must go as g 8 T 4 .
This argument relies on a decoupling between the infrared and ultraviolet fields, since
the ultraviolet fields Žby which I mean fields of wave number k ; T . do not behave
classically. It is known that this decoupling is very accurate for thermodynamic
variables, except that the A 0 component of the gauge field receives a Debye mass. In
fact the validity of the decoupling is equivalent to the quality of the-dimensional
reduction approximation w16–19x, which has been discussed extensively w20,21x.
However, the decoupling does not extend to dynamics; the generalization of the
Debye mass to unequal times are the hard thermal loops, which significantly affect the
infrared dynamics. The hard thermal loops are precisely that set of diagrams which are
linearly divergent within the classical theory. This linear divergence is cut off at the
ultraviolet scale k ; p T where the theory ceases to behave classically; so the size of
these effects depends essentially on the way the IR classical theory is regulated.1 Thus it
is only correct to say that the IR fields behave classically if we mean that they behave
like the IR fields of a classical theory regulated in some way which correctly reproduces
the hard thermal loops. In nature that regulator is quantum mechanics, but we might be
able to find some other appropriate regulator in a numerical setting.
2.2. Argument in terms of conductiÕity and scatterings
Here I will give an argument for the a W5 logŽ1rg .T 4 law based on Lenz’s Law and
the conductivity of the plasma. The argument has very recently been made quantitative
w22x, but I will present it at the qualitative, intuitive level.
1
I should mention parenthetically that the classical theory also contains quadratic and cubic divergences in
the energy density, but these do not affect the IR dynamics responsible for baryon number violation.
To see how hard thermal loops influence the sphaleron rate, I first make the point that
the sphaleron rate is set by the evolution of very soft infrared fields, where by very soft I
mean fields with wave number k ; g 2 T. Parametrically shorter wave lengths do not
contribute appreciably because the probability for non-perturbative physics to occur at
such scales is exponentially suppressed. According to standard sphaleron type argu-
ments, the contribution from the scale k ; g 2y d T is suppressed by of order
expŽykrg 2 T . s expŽygyd .. Even the scale k ; g 2 T logŽ1rg . gives a contribution
suppressed by a power of g.
The second point is that it is physics in the transverse sector which matters, and in
particular, diffusion of NCS requires the evolution of magnetic fields. To see this, first
go back to the definition of NCS ,2
t2 3
g2
NCS Ž t 2 . y NCS Ž t 1 . s Ht dtHd x Eia Bia Ž x . . Ž 3.
1
8p 2
Now the B field is always transverse, meaning that D P B s 0, by the Bianchi identity;
and so only the transverse part of the E field contributes. The Bianchi identity also
states that D = E s yw Dt , B x. Since the relevant part of the electric field is transverse,
it will in general have non-zero covariant curl. For E to be non-zero and to remain the
same sign for long enough to give a non-trivial contribution to HEia Biadt, there must then
be time evolution of infrared magnetic fields.
At this point it is useful to recall how infrared magnetic fields evolve in the abelian
theory, on wave lengths longer than the Debye screening length. The answer is familiar
plasma physics; the plasma is very conducting, and a conducting medium resists changes
in magnetic fields by Lenz’s Law. In the limit of infinite conductivity the magnetic
fields are perfectly frozen; for finite conductivity the time scale for their evolution scales
with conductivity. A magnetic field of wave number k, with well more than its mean
thermal energy density but much less energy density than is contained in the bulk
plasma, decays according to
dA i 1
w D0 , A i x s
dt
sy
s Ž k,v<k .
Ž k 2di j y k i k j . A j , Ž 4.
in the parametric limit that the decay time is well longer than 1rk, which is satisfied at
all length scales parametrically longer than the Debye screening length. The characteris-
tic decay time of a magnetic field in the plasma is then t s s Ž k, v < k .rk 2 . Of course I
have only written the dissipative part of the magnetic field evolution equation; there
must also be a noise term which is uniquely specified by the requirement that the
thermodynamics of the IR magnetic fields are correct.
Note that the conductivity is wave number dependent. It has a good infrared limit
which is achieved for length scales larger than the mean scattering length l free of a
2
Although I set the speed of light cs1, I typically write expressions non-covariantly with a positive space
metric, which is convenient in the finite temperature context because the thermal bath establishes a preferred
frame.
current carried by hard particles. In the abelian theory a particle’s charge is preserved
when it undergoes a scattering, so this length scale is the mean length for large angle
scattering, parametrically l free ; 1ra 2 T times logarithmic corrections. If every particle
had the same free path and a scattering perfectly randomized its momentum, the
conductivity on scales longer than l free would be
m2D l free
ss , k < 1rl free . Ž 5.
3
When scattering processes are more complicated this formula defines an effective value
of l free . For length scales well between the Debye screening length and l free , where
scatterings of the charge carriers can be neglected, the conductivity is related to the
Debye length through
p m2D
ss , m D 4 k 4 1rl free . Ž 6.
4k
ŽIn both expressions the Debye length m D , which is O Ž gT ., is just keeping track of the
number density, charge, and ²1rE : of the particles.The derivations of each expression
assume ultrarelativistic dispersion relations for the charge carriers.. This expression
follows from the form of the transverse self-energy and the fact that the conductivity we
are discussing in this case is just a special case of the HTL self-energy, s Ž k, v < k . s
Im P T Ž k, v .rv . This is the connection between this ‘‘conductivity’’ picture and the
hard thermal loops.
The behavior of the electroweak gauge fields for scales parametrically between the
non-perturbative scale k s g 2 T and the Debye scale k s gT is the same as in the abelian
theory at leading parametric order.3 Since the mean free path of a hard excitation to
undergo any scattering is ; 1rŽ g 2 T logŽ1rg .. w24x, the conductivity for k s g 2y d T,
with 0 - d - 1, is s s p m2D r4 k ; g d T, and the time constant associated with the decay
of a magnetic field is t ; 1rŽ g 4y 3 d T .. Although the scale k ; g 2 T does not fit within
™
the range of validity of this argument it cannot be that the decay rate for a magnetic field
with k ; g 2 T differs from the d 0 limit by any non-zero power of g. Hence the
relevant time scale for the dynamics of non-perturbative infrared magnetic fields in
weakly coupled, hot Yang–Mills theory is O ŽŽ g 4 T .y1 ., up to corrections at most
logarithmic in g. This is a paraphrase of the argument of Arnold, Son, and Yaffe w8x,
who however neglected the possibility of logarithmic corrections.
Logarithmic corrections do in fact occur. While in the abelian theory the electrical
conductivity only reaches a long wave length limit at a scale set by a hard particle’s free
path for large angle scattering, in the non-abelian theory that limit is set by the free path
3
Some years ago Ambjørn and Olesen argued that non-abelian fields obey an anti-Lenz’s Law w23x,
apparently in contradiction to the argument presented here. Their work refers to the non-abelian interactions
between W and Z fields in the presence of strong Želectromagnetism. magnetic fields at zero temperature. It
may have some bearing on the mutual interactions of the fields at the g 2 T scale, but the response of the harder
modes to the very soft fields is at leading order the Lenz law type behavior seen in the abelian theory.
for any scattering. The reason is that when a colored particle undergoes a scattering,
however small the transfer momentum, its color is changed. This degrades the color
current even if it does not degrade the momentum carried by the particle substantially.
The total rate for any scattering to occur is twice the damping rate, which has been
computed for hard particles at leading log by Pisarski. The damping rate for an adjoint
charged particle of any spin is w24x
Ng 2 T mD
gs log q O Ž 1. , Ž 7.
4p g 2T
™
and for a fundamental representation particle it is the same with N Ž N 2 y 1.r2 N.
Note the log, which arises from an integral over exchange momenta, running from the
gT to the g 2 T scale; also note that the result is independent of the particle’s momentum,
provided it is harder than the gT scale. A collision largely but incompletely randomizes
a particle’s charge, and so on scales longer than ; 1rŽ g 2 T logŽ1rg .. the electrical
conductivity of the plasma is ; m2D r3g , up to non-logarithmic corrections. To find the
numerical constant one must determine how thoroughly a scattering randomizes a
particle’s charge, which depends on the representation of the particle. Also, the particle’s
charge is not destroyed, just transferred to another particle; one must check whether this
induces any important currents. ŽIt turns out not to, because a particle is as likely to
scatter from a charge carrier moving in one direction as in the exact opposite direction..
The calculation is quite non-trivial but it has been done recently by Arnold, Son, and
Yaffe w22x, who show that, at leading log, the conductivity is simply m 2D r3g , with g
given in Eq. Ž7.. This is independent of the group representation of the particles carrying
the current.
Performing an extreme parametric expansion, logŽ1rg . 4 1, the scale set by particle
mean free paths and the non-perturbative scale are well separated, and the fields with
k ; g 2 T see a k-independent conductivity. ŽSee Fig. 1.. Hence the relevant infrared
dynamics for transverse modes is, at leading logarithmic order,
3 Ng 2 T log Ž 1rg .
w D0 , A i x s Dj Fji q noise , Ž 8.
4p m2D
where the transverse part of the noise is fixed by the requirement that the thermodynam-
ics come out right. The longitudinal part of the noise generates time-dependent gauge
rotations of the A fields, which are irrelevant to Chern–Simons number; so we may
choose the amplitude of the longitudinal part of the noise to be whatever we want. It is
most convenient to choose it to be of the same magnitude as the transverse part, in
¨
which case we reproduce the effective theory of Bodeker, which is also the Langevin
equation for 3-D Yang–Mills theory.
While this derivation has presented the ideas in an intuitive way it is scarcely
rigorous, so I will now approach the problem a little more formally by looking at the
hard thermal loop effective theory.
2.3. HTL effectiÕe theory and the Wilson line
¨
Now I will go through the argument for Bodeker’s effective theory also from the
point of view of the effective HTL theory for the infrared modes. The idea of the HTL
Fig. 1. ‘‘It’s quite simple, really . . . .’’ A scorecard of the scales involved in the problem and the
approximations which are valid in each. None of these scales are distinct if we do not take g <1.
effective infrared theory is that one can construct an effective theory for the modes with
k < T, valid at leading order in g, by integrating out all degrees of freedom with k ; T.
The procedure for separating the degrees of freedom and regulating the effective theory,
cut off at some scale between gT and T, is left unspecified and does not affect the result
at leading order.
The resulting effective theory is a classical theory for the remaining modes, but with
a non-local ‘‘HTL’’ effective action correction. The non-local effective action was first
derived in w25–27x. In the current context, since in the long term we have in mind a
position space regulation of the IR theory, it is most convenient to write the effective
action in real space, as was first done by Huet and Son w28x:
a a
y w D 0 , Ei x Ž x ,t . q Dj , Fji Ž x ,t .
m 2D yi y j
s j ia Ž x ,t . q Hd 3
y U a b Ž Ž x ,t . , Ž x q y,t y y . . Ejb Ž x q y,t y y . . Ž 9.
4p y4
Here U a b ŽŽ x,t .,Ž x q y,t y y .. is the adjoint parallel transporter along the straight,
lightlike path between the points Ž x,t . and Ž x q y,t y y .. Note that the electric field on
the right-hand side is at the retarded time t y y. The noise j ia is Gaussian with a
non-vanishing two-point function for lightlike separated points,
m 2D T yi y j
² j ia Ž x ,t . j jb Ž x q y,tX . : s d Ž < t y tX < y < y < . U a b Ž Ž x ,t . , Ž x q y,tX . . .
4p y4
Ž 10 .
For k < gT, the effective action simplifies somewhat. The coefficient of the term on
the right in Eq. Ž9., which contains one time derivative Žsince Ei s w D 0 , A i x., is large
enough that the time evolution is determined by this term and we can neglect the w D 0 , E x
term on the left. Further, the time scale associated with the fields’ evolution is longer
than the time retardation appearing in the non-local HTL action, and we are justified to
neglect that retardation. The result is
a m 2D yi y j
Dj , Fji Ž x. s j ia Ž x. q Hd 3
y U a b Ž x , x q y . Ejb Ž x q y . , Ž 11 .
4p y4
and the noise correlator is simplified by dropping the retardation there as well Žthough to
get its magnitude right we must remember that there are two contributions from
d Ž< t y tX < y < y <... If in addition k 4 g 2 T logŽ1rg . then it is possible to choose a gauge
such that the parallel transporter is close to the identity and at leading order it can be
ignored. In this case we can recover Eq. Ž6. by expanding w Dj , Fji x to leading order in A
and Fourier transforming.
Huet and Son argue that, to model the O Ž g 2 T . modes alone, we can use Eq. Ž11.,
interpreted as an expression for fields at the g 2 T scale alone with all shorter wave
length scales integrated out. The reasoning is that the higher modes only appear in the
equation in w Dj , Fji x. But w Dj , Fji x is a thermodynamical quantity depending only on the
transverse fields, and as already stated, the UV causes perturbatively small corrections to
this sector, which can be ignored at leading order. The power counting arguments are
laid out explicitly in w29x.
However, this misses one key issue. Can we neglect the influence of the modes with
g 2 T < k < gT on the adjoint parallel transporter U a b Ž x, x q y . when y ; 1rŽ g 2 T .?
The answer is, no. When evaluating an adjoint parallel transporter for a path of length
; 1rŽ g 2 T ., essential contributions arise from all scales intermediate between gT and
g 2 T. This problem has recently been addressed by Arnold and Yaffe, in the context of
studying O Ž g 2 T . corrections to the Debye screening length w30x. They show that, in
SUŽ N . pure gauge theory, any two point correlator of equal-time adjoint fields at
separation y, connected by a straight adjoint Wilson line, falls off at least as rapidly as
Ng 2 T m reg
exp Ž yyrl . , l y1
s
4p ž log
g 2T
qK , / Ž 12 .
where m reg is a UV regulator and K is a constant of order unity. The constant is

evaluated for SUŽ2. in w15x; for the regulation appropriate to determining the Debye
mass, m reg s m D and K s 6.7 " 0.3. Non-logarithmic corrections are large in this case;
I will return to this point later. The value of the constant K includes non-perturbative
physics at the g 2 T scale, but the logarithm arises perturbatively from scales intermedi-
ate between g 2 T and m reg . Thus, to study physics on the length scale 1rg 2 T, we must
include the influence of scales between gT and g 2 T on the parallel transporter in Eq.
Ž12..
Note the sneaky way the modes with g 2 T < k < gT have entered the dynamics of
the g 2 T modes. Their direct influence on the interactions between the g 2 T modes is
indeed small, as Son showed w29x; but they change the way that the g 2 T modes interact
¨
with the hard modes, in a way which turns out to be important. Bodeker emphasizes this
viewpoint in his original derivation of the effective theory for the leading log w9x.
Using the correction to the Wilson line from the intermediate momentum modes, I
¨
can now establish Bodeker’s effective theory for the k ; g 2 T modes. The modes with
k 4 g T change quickly compared to the g 2 T scale modes, so the g 2 T fields see an
2
average over the more UV scales. In particular the parallel transporter relevant for the
g 2 T modes’ evolution is the average of the parallel transporter over realizations of the
g 2 T < k < gT fields. As I demonstrate in Subsection 3.1, averaging over realizations
of k 4 g 2 T modes leads to exponential damping of the parallel transporter, for y of
order 1rŽ g 2 T logŽ m D rg 2 T ... In the Coulomb gauge,4
ab
UCoulomb Ž x , x q y . , d a bexp Ž yyrl. , Ž 13 .
with l the same as in Eq. Ž12.. If we are permitted to expand in logŽ m D rg 2 T . 4 1,
then the integrand on the right-hand side of Eq. Ž11. has already fallen away before y
comes on order 1rŽ g 2 T .; therefore the approximation which gives Eq. Ž13. is valid, up
to log corrections, throughout the range of y which dominates the contribution to the
integral. The effective theory for the g 2 T modes Žfixing the gauge freedom on scales
more UV than g 2 T to Coulomb gauge. is therefore
yi y j a
j ia q d 3 yeyŽ y r l.
H Eja Ž x q y . s Dj , Fji Ž x. . Ž 14 .
y4
Since the integral is dominated by y ; 1rŽ g 2 T logŽ m D rg 2 T .., and for the IR fields
of interest E varies only on the 1rg 2 T scale, it is permissible at leading log order to
pull the E field out of the integral. The integral is then quite simple; performing it gives
m2D l
Dj , Fji s Eia q j ia ,
3
2Tm2D l
² j ia Ž x ,t . j jb Ž y,tX . : s d Ž x y y . d Ž t y tX . d i j d a b , Ž 15 .
3
where the form for the noise correlator also follows from the approximation for the
parallel transporter. ŽAlternately, one can always recover the form of j by insisting that
¨
the thermodynamics come out correctly.. This is Bodeker’s effective theory, though it
remains to establish that m reg should be m D . I discuss this more in Section 3.
3. Wilson line, more carefully
Now it is time to look more carefully at the Wilson line appearing in Section 2, first
to verify the claims there, second to show the connection to the argument involving
scatterings of the hard particles, and third because it is relevant to the analysis of the
¨
results of Hu, Muller, and Moore.
4
The use of Coulomb gauge becomes problematic when considering length scales l 01r g 2 T, and for
considering any unequal-time correlator if the total volume of space considered is V 4 Ž g 2 T .y3 ; however
this is not relevant because technically we are only applying Coulomb gauge to modes with k 4 g 2 T in order
to integrate them out and establish an effective theory. The gauge fixing of the IR effective theory, i.e. of the
problematic k ; g 2 T modes, has not been specified. Also, at leading log the same results would be obtained
in Landau gauge.
3.1. Exponential decay of the Wilson line
What we want to know about is the Wilson line between lightlike separated points a
spatial distance l apart, with l 4 1rm D but less than 1rŽ g 2 T . by at least a logarithmic
factor. Actually, we want to know the average of the Wilson line over realizations of the
modes with k R g 2 T logŽ1rg ., since these fields fluctuate faster than the O Ž g 2 T . fields,
which therefore see the average over realizations, up to corrections subdominant in
1rlogŽ1rg ..
The Wilson line is given by
l
U s Pexp H0 ig Ž A q A . Ž z ,t s z . dz ,
0 z Ž 16 .
with A in the adjoint representation. I will only consider the transverse contributions
here, the longitudinal ones are subdominant. I use Coulomb gauge, in which the A 0
contribution arises entirely from the longitudinal modes and the A z only arises from the
transverse modes. Further, I will work at leading parametric order, by which I mean that
higher point correlators and vertex insertions are ignored, and combinations of A fields
are evaluated assuming A is Gaussian by applying Wick’s theorem. Of course I will
include the hard thermal loop corrections to the A field propagators. These approxima-
tions are justified at leading log down to k ; g 2 T logŽ1rg ., which is all we need. ŽIf I
were interested in lengths l ; 1rg 2 T rather than logarithmically shorter, most of the
approximations I make would break down completely..
All terms with odd powers of A vanish on averaging over realizations, while even
terms look like Žapplying Wick’s theorem between lines 1 and 2.
n
` Ž y. g 2 n
²U : s Ý Hdz1 . . . dz2 n² A z Ž z1 . . . . A z Ž z2 n . :
ns0 Ž 2 n. !
n
` 1 yg 2 l
s Ý
ns0 n! ž 2
H0 dz dz ² A Ž z . A Ž z
1 2 z 1 z 2 .: /
yg 2 l
s exp
ž 2
H0 dz dz ² A Ž z . A Ž z
1 2 z 1 z 2 /
.: , Ž 17 .
so the average over realizations is the identity times the exponential of the two-point
contribution. ŽNote that for any given realization the Wilson line has unit modulus. But
the average over realizations does not, its modulus falls exponentially with distance..
To evaluate this we need the two-point correlator,
d3k dv T ki k j X
² A i Ž x ,t . A j Ž y,tX . : s e i v Ž t yt . eyi kPŽ yyx . .
H Ž 2p . 3
2p v ž
r Ž k , v . di j y
k2 /
Ž 18 .
Here r Ž k, v . is the spectral density, which is the magnitude of the discontinuity in the
propagator 1rŽ v 2 y k 2 y P T Ž k, v .. across the real v axis on analytic continuation
from Euclidean Žimaginary. v ,
2 y1
r Ž k , v . s 2Im Ž Ž v q i e . y k 2 y P T Ž k , v q i e . . , Ž 19 .
and Trv is the classical approximation for 1 q nŽ v ., with nŽ v . the Bose distribution
function.
So far I have suppressed group indices, but when they and the integrals over z are
evaluated we get
Tr ²U : Ng 2 d3k d v Tr Ž k , v . k 2x q k 2y 4sin2 Ž k z y v . lr2
log sy H Ž 2p . 3
.
Tr 1 2 2p v k2 Ž kzyv .
2
Ž 20 .
The equal-time correlator of the A field goes as Trk 2 plus subleading corrections,
for all k 4 g 2 T ; so
d v Tr Ž k , v . T
H 2p s . Ž 21 .
v k2
What matters now is where r is concentrated. In the regime k < m D , almost all the
contribution to Eq. Ž21. is from < v < < k, see Ref. w24x. This is just the statement that
these modes evolve on time scales slower than 1rk, as I have already discussed. Hence,
in evaluating the low k contribution to Eq. Ž20. I can set k z y v , k z , and then perform
the integral over v , giving
Ng 2 T < < d3k 1 k 2x q k 2y 4sin2 k z lr2
y
2
H k (m D
3
Ž 2p . k
2
k2 k z2
. Ž 22 .
The k z integral is completely well behaved; in fact the large l limit of

4sin2 Ž k z lr2.rk z2 is 2p l d Ž k z .. This represents the fact that only modes with k z - 1rl
have A of the same phase all along the Wilson line; other modes’ contributions
destructively interfere in the integral along the Wilson line. The integral over the other
two directions is dominated by a logarithm arising from scales intermediate between
k H ; 1rl and k H s m D , where the approximation for r breaks down. Performing the
integral over k H first gives
Ng 2 lT mD sin2 Ž k z lr2 .
y
4p 2 ž log
l
q O Ž 1. /H Ž k z lr2.
2
d Ž k z lr2 .
Ng 2 lT mD
sy
4p ž log
l
q O Ž 1. . / Ž 23 .
In the opposite limit, k 4 m D , the excitations obey normal vacuum ultrarelativistic

dispersion relations to good approximation, so the spectral density is approximately
p
r Ž k 4 mD , v . , Ž d Ž kyv . yd Ž kqv . . . Ž 24 .
k
The large k contribution is then
Ng 2 d3k k 2x q k 2y T 4sin2 Ž Ž < k < y k z . lr2
y Hm 3
. Ž 25 .
2 D Ž 2p . k2 k2 Ž< k<ykz .
2
Now the term with sin2 in it is forcing < k < s k z ; only modes propagating along the
direction of the Wilson line keep in phase, others destructively interfere. But the
polarization vector of such a mode is close to orthogonal to the Wilson line; because
2
k 4 1rl, for any k for which < k < y k z - 1rl, the factor k H rk 2 will be near zero. Thus,
the only modes which avoid destructive phase interference are polarized in the wrong
direction to contribute significantly. Continuing to carry out the integral by defining
x s k zrk, we get
Ng 2 T 1 2
4sin2 Ž lk Ž 1 y x . r2 .
y
2p 2
Hy1dxHm dk Ž 1 y x .
Ž1yx .
2
D
Ng 2 T
;y Ž log Ž lm D . q O Ž 1. . , Ž 26 .
2p 2 m D
which is O Ž g .. The hard modes do not contribute at leading order to the Wilson line
along a lightlike path.
I have not treated the modes with k ; m D , which are more complicated because this
is where r does not fit into either limiting category. But they turn out to give a result
smaller by a logarithm than that from the modes with k < m D . The final result is that,
on averaging over realizations of the modes with k greater than g 2 T by at least a
logarithm, the Wilson line in Coulomb gauge is
Ng 2 T mD
²U : s 1 exp Ž ylrl . , ly1 s
4p ž log
g 2T /
q O Ž 1. . Ž 27 .
3.2. Relation to scattering
Now look at Eq. Ž20. again. Remember that the Wilson line is representing the
trajectory of a hard particle, with p much greater than any momentum scale which gives
a leading order contribution to the integral. We want to interpret Eq. Ž20. as l times the
rate for the particle to undergo a scattering involving the transfer of a soft field, times a
Žrepresentation-dependent. group theory factor which tells how thoroughly the scattering
randomizes the particle charge. To see the relation, take the large l limit. Then
4sin2 ŽŽ k z y v . lr2.rŽ k z y v . 2 s 2p l d Ž k z y v ., and the decay rate per unit length of
the Wilson line is
Ng 2 d3k d v Tr Ž k , v . k 2x q k 2y
H Ž 2p . 3
2pd Ž k z y v . . Ž 28 .
2 2p v k2
Recall that the hard particle starts out with pz 4 < k <, p x s p y s 0. If it emitted a particle
of wave number k, its momentum would change to p y k and its energy would change
by yk P pˆ s yk z plus a correction of order k 2r< p <, which is negligible by assumption.
The delta function is just the energy conserving delta function appearing in the
expression for the rate of the process shown in Fig. 2a, the emission of a soft gluon by
an adjoint charged, hard mode.
The rate for this process has been considered by Pisarski w24x; his expression Ž4.4.
does not look quite the same as Eq. Ž28. for the following reasons. First, his definition of
Fig. 2. Ža. Emission of a gluon below the light cone; Žb. Self-energy diagram which, cut, describes the
emission; Žc. The cut must split an HTL insertion, so the physical process is scattering with a small exchange
momentum.
the spectral density differs from the one used here by a factor of 2p . Second, he has
allowed the hard particle to be a little off shell, which is important in the part of the
integral with k ; g 2 T logŽ1rg . but not for higher k. ŽNote however that my large l
approximation breaks down in exactly this regime.. For momenta k 4 g 2 T logŽ1rg . we
may integrate over v t in his expression and obtain Eq. Ž28.. The difference caused by
k ; g 2 T logŽ1rg . does not change the leading log, but would be important in investigat-
ing corrections to the leading log. Also note that Pisarski considers the contribution of
longitudinal momenta and finds that they do not give a log, only a constant times Ng 2 T.
The process considered involves the emission of a soft mode with < k < ) < v <. The
reason that such emission is possible at all is because of the hard thermal loop correction
to the gauge propagator. The rate is the imaginary part obtained by cutting the one-loop
self-energy diagram, Fig. 2b. The cut gives a non-zero result when it goes through a
HTL self-energy insertion, which can be interpreted as diagram Žc. in the figure. Hence
the Wilson loop calculation is proportional to the rate for scatterings of one hard particle
off all other hard particles by exchange of soft intermediaries.
Pisarski calculates the total rate for a particle to undergo a collision, and because of
the conventional definition of the damping rate, his damping rate is actually half of the
scattering rate. Also, his result does depend on the representation of the particle
undergoing the scattering. The rate the Wilson line calculation determines is actually the
rate of color randomization, not of collisions; there is a representation-dependent
correction between the two, which depends purely on color factors at the vertex where
the particle of interest interacts with the soft background field. The total collision rate of
a particle is proportional to the group factor
Tr T a T a
, Ž 29 .
Tr 1
where T a is in whatever representation the particle is in. The original disturbance of the
particle distribution from equilibrium is caused by an electric field, which is an adjoint
object; the disturbance to the single particle density matrix is proportional to E b T b. The
color randomization per collision is
Tr 1 T aT b T a
1y , Ž 30 .
Tr T a T a Tb
where the second term tells how much the color after the collision is aligned with the
color before. Multiplying by the total collision rate gives a color randomization rate
A Nr2, independent of representation.
¨
3.3. Subleading corrections in the method of Hu and Muller
Now I will discuss the relation between the quantum theory and the technique
proposed by Hu and Muller, ¨ refined and implemented jointly with myself, concentrating
on whether the behavior is the same at next to leading logarithmic order.
First I should explain why the technique is necessary. Traditionally people have tried
to determine the sphaleron rate by studying classical Yang–Mills Žor Yang–Mills Higgs.
theory regulated on a spatial lattice. However, if we just study the classical theory on the
lattice, using the lattice spacing as the UV cutoff, the hard lattice modes generate HTL
effects which do not look like Eq. Ž9. and in fact are not rotationally invariant w31x. Even
in the leading log approximation this is a problem, because it means that, where Eq. Ž15.
has Ei , we will get E rescaled by a rotationally non-invariant factor, determined in w32x.
A proposal by Arnold to fix this problem, staying within lattice classical theory, by
making the dispersion relations turn up very steeply w32x, does not work because the
hard modes are then Landau damped 5. The only alternatives I am aware of involve
adding new degrees of freedom which influence the IR classical fields in a way
equivalent to correct hard thermal loops. Two such proposals exist in the literature. One
¨
is due to Bodeker, McLerran, and Smilga w31x, more recently discussed by Iancu w33x. I
will not discuss it since no one has yet specified a complete discrete numerical
implementation. The other idea was proposed by Hu and Muller ¨ w12x; the details of the
implementation were worked out and applied jointly with me w13x.
We also simulated the classical IR physics by studying non-perturbatively the
classical system regulated on a spatial lattice, thereby treating the left-hand side of Eq.
Ž9. fully non-perturbatively. To include the HTL effective action, the right-hand side of
the expression, we added to the classical lattice system a large number of adjoint
charged classical particles. They take coordinate positions in the continuous space in
which the lattice fields sit, obey ultrarelativistic dispersion relations, and interact with
the lattice fields when they cross the dual planes to lattice links. There are two parts to
the interaction with the lattice fields.
First, the particles ‘‘kick’’ lattice electric fields, and their momenta receive a similar
‘‘kick’’. The kicks and the approximately random distribution of the charges performs
the noise, and a correlation between past fields and the ‘‘kick’’ the gauge field receives,
arising from a change in the particle trajectory from the ‘‘kick’’ it received, accounts for
the non-local term. The size of each kick is proportional to a charge Q which is made
small so the particles individually interact weakly.
Second, the particles’ adjoint charges are parallel transported by the gauge field
connection. Since the gauge fields are only defined on the lattice links, the Wilson line
used for the parallel transportation of a particle is the sequence of lattice links which
maintains the shortest distance to the actual path of the particle, as illustrated in Fig. 3.
The exact rule used to choose the sequence of links is that a link is used if the Wilson
™ ™
5
The problem with Arnold’s proposal is that it gives a Lorentz non-invariant hard mode dispersion relation
under which 1 2 and 2 1 processes are kinematically allowed. They are efficient, so the hard excitations
have a mean free path for hard scatterings ;1r g 2 T. I have Arnold’s agreement on this point.
Fig. 3. A simple 2-D example of how a Wilson line Ždiagonal. on a lattice Ždashed lines. is replaced by the
sequence of links which stay closest to it Žsolid, jagged line..
line penetrates the face of the dual lattice corresponding to that link. Parallel transporting
the particle’s charge ensures that we are including the parallel transporter in the
right-hand side of Eq. Ž9.; it is also absolutely necessary to make the update rule
manifestly gauge covariant. The full update is described in excruciating detail in w13x.
The Debye mass depends on the density ² n: and charge Q of the added particles as
m D A Q 2 ² n:, so by changing the number of particles one can tune the HTL strength to
2
be whatever is needed while keeping Q small. If there were only IR fields, meaning
™ ™
fields with k < 1ra Ž a the lattice spacing. then the behavior would correctly reproduce
Eq. Ž9., at least in the limit Q 0, ² n: ` with m2D fixed. This is also discussed in
w13x.
™ ™
The lattice theory involves two scales, a and m D ; to get the leading parametric
behavior we must seek the limits a 0, m D ` Žif we think of the length scale
1rŽ g 2 T . as remaining fixed.. There are two ways we could go about doing this. One
corresponds to using a as a cutoff between the gT and the T scales, meaning that we
maintain a < 1rm D but a 4 g 2 Trm2D . The latter condition is essential to make sure
that ‘‘bad’’ hard thermal loops arising from modes with k ; 1ra, which have the wrong
dependence on k, v and in particular are not rotationally invariant w31x, are subdominant
to ‘‘good’’ hard thermal loops due to the particles. The other way of taking the limits is
to make m D 4 1ra, so the wave number Ž1ra. falls between the gT and g 2 T scales.
In each approach, the model correctly generates the effective HTL dynamics up to
power corrections in Ž g 2 aT ., Ž Qrg 2 aT ., g 2 Trm D , and Ž g 2 T .rŽ am2D ., except perhaps
for the behavior of the parallel transporter. The contribution to the parallel transporter
from IR fields is correct because for a gauge field of wave number k < 1ra the
replacement of the straight line path with the ‘‘jagged’’ path actually used Žsee again
Fig. 3. gives the right behavior up to corrections suppressed by O Ž k 2 a 2 .. The question
is, how does the UV contribute to the parallel transporter?
In the case where we make m D 4 1ra, all of the lattice degrees of freedom have
overdamped evolution. None of them propagate and there are no problems from hard
thermal loops arising from classical lattice degrees of freedom with k ; 1ra. Except for
the parallel transporter, the systematics are then best under control. However, the log
appearing in the parallel transporter is obviously cut off by the inverse lattice spacing,
logŽ m D rg 2 T . becomes C q logŽ1rg 2 aT . with C a coefficient to be determined by
matching. Since we know that the lattice regulator is not rotationally invariant we expect
that C will be direction dependent, leading to a rotationally non-invariant correction.

However, this correction is only at next to leading order in an expansion in logŽ1rg 2 aT ..
In the other case, m D < 1ra, all of the physics which sets both upper and lower
limits on the log in Eq. Ž20. lie at k < 1ra, and naively we should get the correct
behavior at leading order in g. We do not, however, because the most UV lattice modes
have dispersion relations which ‘‘turn over,’’ k ) v . Diagram Ža. from Fig. 2 can occur
ˇ
with the emitted gluon on shell. In fact this is just Cerenkov radiation, which is
permitted because the group velocity of the most UV lattice modes is subluminal, while
the hard particles move at the speed of light. Hence, while there is no contribution to the
parallel transporter from modes with 1ra 4 k 4 m D , there is a contribution from
k ; 1ra, and again we get a rotationally non-invariant correction, logŽ m D rg 2 T .
becomes C q logŽ m D rg 2 T ..
To determine the value of C we need to repeat the arguments leading to Eq. Ž20. but
using lattice gauge fields multiplied along a ‘‘jagged’’ path which stays closest to the
straight line path under consideration. For a Wilson line of length l s La in a direction
with unit vector pi , with pi in the first octant, the equivalent of Eq. Ž20. is
Tr ²U : Ng 2 pra d3k ` d v Tr Ž k , v .
log sy Hypra Ž 2p . Hy` 2p3
FŽ k,v . , Ž 31 .
Tr 1 2 v
FŽ k,v .
w Lp xqx x
1 nxyx
sÝ
e
H0 ½
dx dy dz a e x Ý
n xs1
expi v
ž px
y k x Ž n x y 0.5 .
2
nxyx nxyx
yk y y q
px
py y k z z q
px
pz
/ žÝ Ý /5
q
ny
,
nz
similar , Ž 32 .
and the values in other octants follow from cubic symmetry. Here the sum over e is
over a basis of the two unit vectors satisfying Ý i e i sinŽ k i ar2. s 0. The integral over
Ž x, y, z . appearing in the definition of F averages over starting positions for the Wilson
line within a lattice cell, and square brackets always mean the argument is rounded
down to an integer. The rounding down to an integer makes it very difficult to evaluate
the expression analytically. It is also tricky to find r Ž k, v . on the lattice, away from the
m D 4 1ra or m D < 1ra limits.
In the infrared, meaning k < 1ra, the sum in F is well approximated by an integral,
recovering Eq. Ž20.. In the UV, while Eq. Ž31. is manifestly cubic symmetric, it is not
spherically symmetric: the k integral is over a cubic range, and neither F Ž k, v . nor
r Ž k, v . are rotationally invariant. The integral has been computed by Arnold and Yaffe
in the large L limit for the special case that pi lies in a lattice direction and the spectral
density is concentrated at v s 0 w30x, although they did not present their calculation in
this context. I will now compute it in the opposite limit, m D < 1ra, but still along a
lattice direction. If L is large and pi lies in a lattice direction Žsay, the z direction. then
F Ž k, v . , 2p l d Ž k z y v mod Ž2pra... For m D < 1ra the dispersion relation is the free
lattice dispersion relation v 2 a 2 s Ý i Ž2 y 2cosŽ k i a.. and the spectral density r all lies
on shell, and the integral reduces to
Ng 2 lT pra d3k k˜ 2x q k˜ 2y 2p
Hypra Ž 2p . 3 2 ž (
2pd k z y k˜ 2 mod / , Ž 33 .
2 Ž k̃ 2 . a
where k˜ i s 2sinŽ k i ar2.ra. Evaluating numerically gives 0.0466539Ng 2 lT, which is

0.586 times the coefficient of logŽ m D rg 2 T .. This demonstrates that there are extra UV
contributions arising from k ; 1ra in the case m D < 1ra, which though not terribly
large are not negligible.
The upshot is that however m D is tuned, the UV lattice degrees of freedom will
introduce a rotationally non-invariant contribution to the radiative correction to the
Wilson line which determines the parallel transport of a particle’s charge. This spoils the
rotational invariance of the IR HTL effective theory at next to leading order in logŽ1rg ..
The best we could do would be to average the value of Eq. Ž31. over angles. Then an
appropriate choice of lattice spacing a in the regime with m D 4 1ra could match the
value of the logŽ m D rg 2 T . as closely as possible to the quantum theory value. This is
the best option I see for determining the sphaleron rate beyond leading order, but it does
not eliminate all systematic errors even at leading parametric order in g.
I have not shown that the same problem will arise for other possible lattice
implementations of HTL effects; but since such implementations must generally involve
parallel transportation on the lattice I expect that the problem discussed here is general.
Certainly, when proposing some other numerical implementation of HTL’s, the burden
of proof must lie on the side of showing that problems from rotational non-invariance of
Wilson lines do not arise.
4. Numerics
Now that I have discussed the establishment of the effective theory, Eq. Ž15., I will
discuss how to make a lattice model of that effective theory and how I compute the
leading log coefficient of the sphaleron rate by using that model. First, define a
Langevin time t related to the time appearing in Eq. Ž15. via
3 3 Ng 2 T log Ž m D rg 2 T .
dt ' dt s dt . Ž 34 .
m2D l 4p m2D
Note that t has dimensions of length squared, not length. The effective infrared theory
then has the familiar form
a a
w Dt , A i x Ž x ,t . s Dj , Fji Ž x ,t . q j ia Ž x ,t . ,
² j ia Ž x ,t . j jb Ž y,t X . : s 2Td Ž x y y . d Ž t y t X . , Ž 35 .
which is a Langevin equation. Studying it, I will find the diffusion constant for
Chern–Simons number
² Ž NCS Ž t . y NCS Ž 0 . . 2 :
G Langevin s lim lim
V ™ ` t™ ` Vt
, Ž 36 .
numerically. To do that I find the diffusion constant per lattice site per a 2 of Langevin
time, and multiply by ay5 . On-dimensional grounds G Langevin must be of order a W5 T 5.
Using Eq. Ž34., the relation between G Langevin and k X defined in Eq. Ž2. is
3 N G Langevin
kX s . Ž 37 .
4p a W5 T 5
Before getting further it is worth commenting that the IR behavior of Eq. Ž35. is
insensitive to the UV regulation and the limit in which that regulation is removed, exists.
¨
Though this statement appears banal, it makes Bodeker’s effective theory completely
different from the classical Hamiltonian dynamics. The essential difference is that there
are no long time scale correlations for the UV fields in the Langevin evolution; a mode
with wave number k gets randomized in Langevin time t ; 1rk 2 , which is much faster
than the natural time scale for the evolution of IR fields. Hence the IR fields see the
average over all excitations of the UV fields. The influence of the UV must be purely
thermodynamic, and we know from the super-renormalizability of 3-D Yang–Mills
theory that the thermodynamic influence of the UV on the transverse sector is well
behaved. For a more rigorous presentation of the argument see Ref. w22x, who show that
any purely dissipative update algorithm will give a good continuum limit. For the
Hamiltonian system, on the other hand, UV modes are propagating; the unequal-time
correlator behaves like cosŽ kt . rather than like expŽyk 2t .. The existence of long time
scale correlations of the UV fields is what makes the HTL effects important to the IR
dynamics. Because of this difference, we can expect a good small lattice spacing a limit
to exist for the Langevin time dynamics, and it is worth it to try to control systematic
errors.
I discuss the continuum t , spatial lattice implementation of Eq. Ž35. in Appendix A;
here I will just mention how I discretize the time update. I define a time step Dt s a2D,
D a pure number much less than 1. The fields will be well defined at times n Dt , n an
integer. The noise is constant in each interval w n Dt ,Ž n q 1. Dt .; its value at each point,
direction, and Lie algebra direction is drawn from the Gaussian distribution with mean
(
value 2Tra3Dt ; its value at each point, direction, and Lie algebra direction, and in each
time interval, is independent. To determine the fields at time Ž n q 1. Dt from the fields
at time n Dt I use the following second order algorithm: I compute dArdt Ž n Dt q 0.
and use it to predict AŽŽ n q 1. Dt .; then I average the values of dArdt at the starting
point and at the predicted end point, and use this average to update AŽ n Dt . to time
Ž n q 1. Dt . The step size errors are O Ž D 2 .. In particular, if I were studying the free
theory a mode with wave number k would be updated with step size errors , Ž k 2Dtr2. 2 .
The direct errors in the update of the IR fields are tiny. However, the most UV modes
are only updated correctly if D < 1. The UV fields influence the IR modes radiatively,
so we do need D to be small; but the radiative corrections are suppressed by O Ž g 2 aT .,
and it is not too difficult to get D small enough to make step size errors subdominant to
statistics.
Unfortunately the above update is quite inefficient. However, there is a much more
efficient algorithm for dissipatively updating the fields, the heat bath algorithm. Rather
than applying a very small step of Langevin update to each lattice link in parallel, the
idea is to go through the links of the lattice at random, performing a complete heat bath
update of each link. The relation between Langevin time and the number of links
updated is discussed in Appendix A.2; in particular it is possible to make a very accurate

match between the Langevin and heat bath time scales by measuring the autocorrelations
of some IR observable. I apply the small D limit as part of the matching, so that is taken
care of.
Appendix A, together with previous work w34,35x, shows how to control lattice
spacing systematics so they first appear at O Ž a 2 .. Also, it is possible to define a lattice
measurable to use for NCS which is topological and will eliminate systematic errors in G
due to the definition of NCS . In fact, two fairly efficient techniques are available w11,36x;
here I will use the method developed in w36x. It remains to take the large volume and
time limits. It was shown in w10x that finite volume systematics are negligible on cubic
toroidal lattices larger than 8rg 2 T on a side. To be doubly sure, I have used a lattice
16rg 2 T on a side; as a check I measure G also on a lattice of half this size to check
that the result is the same. Taking the infinite time limit is tied up with the problem of
converting a Langevin time series for NCS into a measurement of G . I use the same
analysis techniques as w11x.
To verify good control of lattice spacing systematics I have made measurements of G
at three lattice spacings, a s 2r3 g 2 T Ž bL s 6., a s 2r5g 2 T Ž b s 10., and a s
2r7g 2 T Ž bL s 14.. The results are presented in Table 1. Finite volume and spacing
systematics are under control. In particular, varying the lattice spacing by over a factor
of two leads to corrections smaller than the statistical errors. This makes large a
extrapolation unnecessary, which is very important, since numerical cost rises as Ž1ra. 5.
If I had used naive rather than radiatively corrected relations between lattice and
continuum parameters, then the value of k X at the three lattice spacings would be 17.7,
15.0, and 13.3 respectively. Such strong dependence is because converting G from
lattice to continuum units involves the fifth power of a, and the radiative corrections are
at leading order a shift in the meaning of a from the naive value, by roughly 10%, 6%,
and 4% for the three lattice spaces used. At two loops there is a further shift, estimated
to be of order Žand probably less than. the square of the first order shift w34,35x; around
1%, 0.36%, and 0.2% respectively. The latter two are negligible compared to the
statistical errors, even after taking account of the fifth power dependence. It is less clear
how to estimate the importance of O Ž a 2 . non-renormalizable operators; but if I estimate
all O Ž a2 . errors by using the three lattice spacings to extrapolate to a s 0, assuming
errors proportional to a2 , the result lies within the error bars of the two finer lattice data,
and the error in the extrapolation is dominated by the error in the finer lattice data.
Hence I will adopt the middle lattice spacing result and its statistical error bars as the
best estimate.
Table 1
X
Results for k at two lattice spacings and two lattice volumes. The results show excellent spacing and volume
independence.
Lattice spacing a Volume Langevin time k X " Statistical error
2r3 g 2 T Ž8r g 2 T . 3 290000 a2 10.44"0.23
2r3 g 2 T Ž16r g 2 T . 3 49500 a2 10.30"0.21
2r5g 2 T Ž16r g 2 T . 3 21000 a2 10.70"0.67
2r7g 2 T Ž16r g 2 T . 3 42000 a2 10.26"0.79
5. Interpretation and systematic corrections
The numerical result is that

mD g 2T 2
ž
G s Ž 10.8 " 0.7 . log q O Ž 1. / a W5 T 4 q Ž higher order in g . , Ž 38 .
g 2T m 2D
but it remains to determine or estimate the rate at the realistic standard model value of
m D , m2D s Ž11r6. g 2 T 2 with g 2 , 0.4. For this value, logŽ m D rg 2 T . , 1.5, and the
O Ž1. correction may be quite important.
In fact we might expect that the log needs to be quite large before it dominates the
O Ž1. ‘‘correction’’. The leading log behavior is based on the hard particles propagating
only a short distance before undergoing a collision which randomizes their charge. This
‘‘short’’ distance is 4prŽ Ng 2 T logŽ m D rg 2 T .., with N s 2 since we are in SUŽ2.
theory. The leading log contribution to the free path is ; Ž6rg 2 T .rlogŽ m D rg 2 T .. For
comparison, above I confirm Ambjørn and Krasnitz’ result w10x that a lattice only 8rg 2 T
across is already large enough to give continuum like behavior for the sphaleron rate.
The non-perturbative length scale characterizing baryon number violating processes
must be shorter than this, perhaps by a factor of two. Hence the log will need to be quite
big before the ‘‘short’’ distance really is short compared to the scale which is setting the
physics. This supports the expectation that there will be large corrections to the leading
log.
5.1. Estimate using Laine and Philipsen’s results
As I noted already in Subsection 2.3, the log arises from the behavior of a Wilson
line, and the same Wilson line appears in the definition of the Debye mass beyond
leading order. In this context the value for the O Ž1. correction to the leading log
behavior has been found by Laine and Philipsen. In that case, wlogŽ m D rg 2 T . q O Ž1.x
has O Ž1. s 6.7. If the same number held for the sphaleron rate, then using the standard
model value of m D to evaluate the log, the leading behavior would be G s
89Ž g 2 T 2rm2D . a W5 T 4 s 48 a W5 T 4 . This is a crude way of estimating the non-leading
corrections, though, and I do not take it too seriously. In particular there is evidence that
the length scale relevant for baryon number violating processes is longer than the
, 1rg 2 T Laine and Philipsen find for the O Ž g 2 T . correction to Debye mass; the
baryon number violation rate on a cubic toroidal lattice 3rg 2 T across is over 1000 times
slower than for a large volume w36x, so physics must be going on involving lengths at
least half as long as 3rg 2 T.
¨
5.2. Estimate using Hu, Muller, and Moore’s results
Another way of trying to determine the subleading corrections is to use the value of
¨
the leading log coefficient to correct my results with Hu and Muller. There, we used a
technique which included hard thermal loops, but in a way which does not correctly
reproduce the subtleties of the Wilson line responsible for the logarithmic dependence of
G on m D . As discussed in Subsection 3.3, the log arises because excitations more UV
than the g 2 T scale make the Wilson line effectively randomize the charge of a
propagating particle, and there is a log in the reciprocal length for randomization. The
best approach would be to compute the angle averaged value of the reciprocal length, in
the quantum theory and in the lattice theory actually studied. Unfortunately, so far I can
only compute the reciprocal length in the lattice theory in one direction, and only in the
cases m D < 1ra or m D 4 1ra. The data in w13x are taken at m D ; 1ra, not 4 1ra,
so it is at least reasonable do the match using the small m D a approximation. I will make
do with the log evaluated in the one direction where I can do the integral, which gives a
difference in logs between the two theories of logŽ m D Žlattice.rm D Žcontinuum. q 0.59.
I will use the datapoint from that paper taken with the largest value of m D , because
the ‘‘wrong’’ lattice mode induced HTL’s really are strongly subdominant to the
‘‘right’’ particle induced HTL’s for this case. Redoing the match between the lattice and
continuum time scales, which was performed wrongly there because we did not have the
O Ž a. calculation performed in Appendix A of this paper, revises the result from
G s 53 " 5Ž g 2 T 2rm2D . a W5 T 4 down to 50 " 5Ž g 2 T 2rm2D . a W5 T 4 . This result was ob-
tained at m D , 4 g 2 T, so the difference of logs between the lattice theory where this
number was computed and the quantum theory at the physical value of m 2D is about 0.63
due to the values of m D plus 0.59 due to UV contributions present on the lattice but not
in the continuum theory. Using the determined coefficient of the leading log term, I
should correct the diffusion constant we found downwards by 1.22 = 10.8, giving
37 " 5, with only the statistical error bar shown. ŽSubstituting in m2D s 11 g 2 T 2r6 gives
G s 20 " 3 a W5 T 4 ..
Using k X to correct the old data in this way assumes that
mD d m 2D
G s k Xa W5 T 4 .
dm D ž 2
g T 2 / Ž 39 .
We really only know that this true in the large Ž m D rg 2 T . limit, where the leading log
expansion is valid. It may have quite non-negligible corrections at realistic values of
m D , which would appear in a systematic expansion in logŽ m D rg 2 T . as inverse powers
of the log; the O Ž1. correction to the leading log behavior would really be O Ž1. s C1 q
C2 ŽlogŽ m D rg 2 T ..y1 q . . . Intuitively I expect that the real m D dependence will be
weaker than the leading log suggests Žmeaning C2 ) 0., though I cannot give a cogent
argument to show this is so. In this case I have performed an overcorrection, and the real
rate would be higher. For now I will accept the corrected answer as the best current
guess, but I take a systematic error bar of order 30% to cover both the rotational
non-invariance not handled correctly in the correction, and the fact that the correction
may be an overestimate at realistic m D .
5.3. Corrections which are formally parametrically suppressed
There is a further cause of systematic error in the determination of the sphaleron rate,
arising from corrections to the g < 1 approximation. I will mention the two such
corrections which I think are the largest; fortunately they have opposite sign and the
optimistic can hope that they largely cancel.
One problem is that the parametric argument for the importance of hard thermal loops
assumes m D 4 g 2 T, and it is not clear realistically whether this is obtained w14x. There
is evidence that it is not. In particular, while the sphaleron rate in classical Yang–Mills
theory depends on the lattice spacing in a way consistent with the Arnold-Son-Yaffe
prediction G A my2 w x
D A a, the corrections to the linear dependence are large 11 . Of
course, some dependence is expected, since we know now that the scaling behavior
should be not G A a but G A aŽlogŽ1ra. q O Ž1..; but we can determine the coefficient
of the logŽ1ra. term by using the results of this paper. According to Arnold w32x, we can
relate the classical lattice results to continuum ones, approximately, by replacing
m2D , 0.684 g 2 Tra. In this case the a dependence of G in classical, lattice Yang–Mills
theory should be
X 2 y1 r2 g 2 aT
Gon the lattice s 0.465k log ž Žg aT . q O Ž 1. / a W4 T 4 , Ž 40 .
4
with k X the same as the one we find but the O Ž1. correction different. This formula
makes it possible to correct the data in w11x to remove the logarithmic dependence on a,
for instance by adjusting the data so they all correspond to g 2 aT s 1r4, which is the
value for the finest lattice used there. I have done so, and the result is plotted in Fig. 4.
I fit the data to the form Gra s C1 q C2 a to find the corrections to the small spacing
limit which do not arise from the log. The fit is very good, see Fig. 4, but the coefficient
C2 is quite substantial. The physical value of HTL strength, using m D s Ž11r6. g 2 T 2
and g 2 s 0.4, corresponds to g 2 aTr4 s 0.037, where the correction C2 a accounts for
an 11% shift from the small a limit, with the actual value falling below the large m D
extrapolation. The correction reduces the sphaleron rate. Given the other systematics in
play it is probably not fair to call this a measurement of the correction to the m D 4 g 2 T
limit; rather I will call it an estimate to tell how large the systematic error is.
Another correction which is parametrically suppressed but not necessarily very small
arises from QCD scatterings of quarks. Quarks are responsible for almost half of m2D ,
and hence almost half of all HTL effects. But quarks scatter strongly. While a strong
scattering does not disturb a quark’s electroweak charge, it does change its direction of
flight, whereas the calculation of the HTL’s is made assuming particles maintain straight
line trajectories. If the free path for strong scattering were ( 1rg 2 T this would make an
O Ž1. correction to the quarks’ influence on IR physics. Actually the quark free path for
Fig. 4. Small lattice spacing extrapolation of data in pure classical lattice Yang–Mills theory, taken from Ref.
w11x. The data have been corrected to absorb the leading log dependence on m2D A1r a determined here. They
show a substantial linear correction to the predicted G A a behavior. This is evidence of non-negligible
corrections to the parametric m D 4 g 2 T limit.
large angle scattering is parametrically order Ž a s2 logŽ1rg s .T .y1 and the correction is
formally O Ž a s2ra W . ; g 2 ; but numerically this might not be small. The effect of this
correction is to increase the sphaleron rate. It might be possible to include this correction
within the context of the ‘‘particles’’ approach by calculating more precisely the size
and frequency of quark scatterings and adding them to the dynamics of the particle
degrees of freedom.
5.4. Including the Higgs
I should also mention that the evaluation of the leading log coefficient made here was
within pure SUŽ2. Yang–Mills theory, without a Higgs boson. In the context of
baryogenesis we actually need to know the sphaleron rate in the presence of at least one
Higgs boson, in the symmetric phase and roughly at the equilibrium temperature. The
Higgs field’s evolution is not overdamped, because the hard thermal loops for a scalar
field are nothing but a mass squared correction. Hence the k ; g 2 T modes of the Higgs
fields evolve on the time scale 1rg 2 T, which is parametrically faster than the gauge
fields. Therefore, on the 1rg 4 T time scale, the gauge fields only see the thermodynamic
average over all Higgs fields in the fixed gauge field background6 . This could be
simulated by including the Higgs field in the Hamiltonian, and evolving it with heat bath
dynamics, but updating the Higgs fields much more often than the gauge fields. Then we
would have to extract the limit as the Higgs field update is made infinitely faster than
the gauge field update. I have not yet attempted to do this, but I anticipate that for
parameters which make the phase transition strongly first order, the effect should be a
slight reduction of the leading log coefficient for the sphaleron rate.
6. Conclusions
In the formal small g limit, in which an expansion in logŽ1rg . 4 1 is justified, the

sphaleron rate in SUŽ2. Yang–Mills theory is
mD g 2T 2
ž
G s Ž 10.8 " 0.7 . log qO Ž 1 . / a W5 T 4 . Ž 41 .
g 2T m 2D
The value 10.8 " 0.7 is clean; the errors are dominated by statistics, systematics are well
under control.
However, interpreting this result to get the sphaleron rate at the realistic values of
m2D s Ž11r6. g 2 T 2 and g 2 , 0.4 is very problematic, because the O Ž1. correction is not
subdominant. The reason is that the expansion in large logŽ1rg . corresponds to treating
the length scale 2prw g 2 T logŽ m D rg 2 T .x as much shorter than the scale relevant for
baryon number violating processes, which is around ; 4rg 2 T. The best current
estimate for the O Ž1. correction comes from using the result of this paper to correct
¨
previous results of Hu, Muller, and myself, which effectively were using the wrong
value for the log. The result at the physical values for m D and g 2 is G , Ž20 " 3. a W5 T 4 ,
with only statistical errors quoted. Unfortunately the correction procedure is not well
under control; I estimate that the systematic errors should be taken to be at least twice as
6
Dam Son pointed this out to me.
large as the statistical ones. There are also corrections to the g < 1 approximation,
probably at around the 10% level.
Besides making it possible to correct previous results, the leading log approximation
¨
and Bodeker’s effective theory provide a clean test-bed for determining how good an
approximation it is to treat the symmetric phase of Yang–Mills Higgs theory using just
Yang–Mills theory. Checking how much difference the Higgs field makes is an
interesting project for the future. It would also be interesting to study the dependence of
k X on the number of colors, since the sphaleron rate in the strong sector is also
phenomenologically interesting for baryogenesis.
Acknowledgements
¨
I am grateful to Peter Arnold, Dietrich Bodeker, Dam Son, and Larry Yaffe for
illuminating discussions.
Appendix A. O(a) lattice time scale renormalization
In this appendix I compute at O Ž a. the lattice to continuum correction to the

Langevin time scale, and the relation between the continuum Langevin time scale and
the amount of heat bath applied. The calculation of the Langevin time scales will be
quite technical and will depend to some extent on previous work relating the lattice and
continuum theories at the thermodynamic level, found in w34,35x. I will relate the heat
bath depth to the continuum Langevin time scale by making a non-perturbative lattice
measurement comparing it to the lattice Langevin time scale, and then using the analytic
relation between this and the continuum Langevin time scale. It would also be possible
in principle to make the connection directly, but the analysis is difficult and I am lazy.
A.1. Lattice and continuum LangeÕin time scales
Ref. w34,35x explains how the lattice to continuum relation for thermodynamical
properties of 3-D gauge or gauge-Higgs theory can be studied as an expansion in g 2 aT,
where a is the lattice spacing and g 2 is the coupling in 4-D notation. The combination
g 2 T is the coupling constant of the 3-D theory. Since it is dimensionful, a perturbative
matching between continuum and lattice theories must be an expansion in g 2 Ta on
dimensional grounds. In Yang–Mills Higgs theory with fundamental or adjoint scalar
fields the leading terms can behave as 1ra because the theory contains dimension 2
operators, but in strict Yang–Mills theory the lowest order operator is dimension 4 and
the leading corrections are O Ž a.. Further, the only O Ž a. correction is a rescaling of the
coupling, equivalent to a rescaling of the physical length scale, and it only arises from
one-loop diagrams, and has been computed w34,35x.
Arnold, Son, and Yaffe have recently demonstrated that, because of super-renormal-
izability and general considerations worked out by Zinn-Justin and Zwanziger w37x, the
same applies to the dynamics under Langevin dynamics w22x. In particular, any high
dimension corrections which can appear in Eq. Ž15. would change the dynamics at
O Ž a2 ., simply because there are no gauge invariant, P-even dimension-5 operators in
Yang–Mills theory. The only possible O Ž a. correction, besides the thermodynamic one
already mentioned, is a rescaling of the Langevin time scale, which will arise exclu-
sively at one loop. This opens the possibility of computing the O Ž g 2 aT . corrections
between the continuum and lattice Langevin dynamics, by performing a one-loop match.
It also means the Langevin dynamics can be replaced by heat bath dynamics at the cost
of another O Ž g 2 aT . shift, which I will measure, rather than compute, in Appendix A.2.
In the continuum theory and in temporal gauge the Langevin evolution is
dA ai Ž x . EH
sy a q j ia Ž x . ,
dt E Ai Ž x .
² j ia Ž x ,t . j jb Ž y,t X . : s 2Td i j d a b d 3 Ž x y y . d Ž t y t X . . Ž A.1 .
The Langevin equation for the lattice theory must be written in terms of the unitary
parallel transporter matrices Ui Ž x . and their derivatives. By definition the matrix
Ui Ž x . g SUŽ N . is the matrix such that, if F is a fundamental representation object
transforming as an object at point x q ai,ˆ Ui Ž x .F is the parallel transport to point x;
thus Ui Ž x . ‘‘lives on’’ the link between the site x and the site x q ai.ˆ Writing DLa for
™
the left acting derivative,
DLaU s igaT aU , DLa F Ž U . s F Ž U U q DLaU . y F Ž U . , Ž A.2 .
the Langevin equation for U is
dUi Ž x .
s yDLaUi Ž x . Ž bL TDLa H KS q j ia Ž x . . ,
dt
8 di j da b d x y
² j ia Ž x . j jb Ž y . : s d Žtyt X . . Ž A.3 .
g 2 a 4bL
Here T a is a fundamental representation Lie algebra generator with the standard
normalization; H KS is the Kogut–Susskind Hamiltonian, the sum over elementary
plaquettes of the trace of the plaquette,
H KS s Ý 1 y 12 Tr UI ; Ž A.4 .
I
and bL is the inverse temperature in lattice units. At leading perturbative order
bL s 4rŽ g 2 aT ., but it receives radiative corrections, computed in w34,35x, which shift it
by a constant. What I write here as bL is bnaive in the notation of w34,35x, but in the
body of the paper I have always used the O Ž a. improved definitions when I refer to a or
bL . The combination bL H KS equals HrT of the continuum theory, up to radiative
corrections and high dimension operators which correct infrared physics at O Ž a 2 .. The
radiative corrections are absorbed up to errors which are O Ž a 2 . by the one-loop
radiative correction to bL , which I will present in due course.
Our task is to compute to O Ž a. the relation between the Langevin time scale t in the
continuum and lattice cases. The correction at loop order l will be O Ž g 2 aT . l , with the
g 2 T from loop counting and the a to balance dimensions; so we need only go to
one-loop order. Part of the correction is from the shift in b already mentioned, but there
are also corrections from the relation between U and A and from radiative differences
between the lattice D a and the continuum derivative. I will be satisfied to perform the
calculation in a particular gauge, strict Coulomb gauge, without checking for the gauge
independence of the result. By strict Coulomb gauge I mean that at every Langevin time
the 3-D configuration is in 3-D Landau gauge. I should also fix a global time-dependent
gauge ambiguity; but this is irrelevant at the level of perturbation theory.
In Coulomb gauge the U matrices are all close to the identity matrix, with a departure
of order by1L
r2
; so it makes sense to define a lattice gauge field A ai through
Ui Ž x . 'exp Ž igaT aA ai . . Ž A.5 .
The A field so defined has the same normalization as the continuum A field at tree
level, but there are radiative corrections. The value of the radiative correction to A is
obtained in the theory with an added scalar field by matching the one-loop values of the
gauge-scalar three-point vertex at small transfer momentum. There is a contribution
from one-loop vertex corrections and from one-loop scalar wave function corrections,
and neither depends on the number of scalar particles in the theory, so the answer is the
same in the pure gauge theory. The one-loop correction was computed as a byproduct in
w34,35x in a general gauge, for SUŽ2.. The result, in Coulomb gauge but for SUŽ N .
gauge theory, is
Ng 2 aT 1 S j
A ai Ž continuum. s 1q ž 4 /ž 18 4p
q3
4p / A ai Ž lattice. . Ž A.6 .
The numerical constants S and j were first defined in w38x, and their numerical values
are S s 3.175911536 and j s 0.152859325.
Next I need to find the relation between applying the lattice and continuum
derivatives. For this purpose it is actually more convenient to write the update rule in
terms of center acting derivatives,
DCa U s U 1r2 Ž igaT a . U 1r2 , Ž A.7 .
because the formulation will then be parity symmetric. Naively one would expect that if
we set dUirdt s Eia DCa U, that dA ai rdt s Eia, and indeed this is correct at leading order
in A. But beyond leading order there are corrections; expanding both sides gives
dUi d
s exp Ž igaT aA ai .
dt dt
s igaT a
dA ai
dt
y
g 2 a2 T b T c
2
A bi
dAci
dt ž
q Ž b c. l /
ig 3a3 T d T e T f dA if
y
6 ž A di A ei
dt /
q permutations q O Ž a 4 . , Ž A.8 .
g 2 a2
Eia D aU s 1 q igaT aA ai y T b T cA bi Aci q O Ž a3 .
2
g 2 a2
= igaT d Eid 1 q igaT eA ei y T f T hA if A hi q O Ž a3 . . Ž A.9 .
2
We must expand to O Ž a 2 . corrections because ² A2 : ; 1ra. Equating the expressions,
after some work we obtain
dA ai d
g 2 a2
Ž lattice . s E i d ad y f f A b AcqO Ž a 3 . . Ž A.10 .
dt 24 a b e c d e i i
To save some writing I left off marking that all A, E above are in position space and at
the same coordinate position. The above correction is strictly a lattice effect and the
equivalent continuum relation is dA ai rdt s Eia.
Naively I should now equate Eia with ybL DCa H KS q j , evaluate the ŽUV dominated.
mean value of the correction between the lattice and continuum relations, and thereby
determine the rescaling of the Langevin time. This would not be strictly correct, though,
since the Langevin equation, Eq. ŽA.3., gives time evolution in temporal gauge. The
temporal gauge evolution breaks the Coulomb gauge condition, and if we are to do the
calculation in Coulomb gauge we must make a time-dependent gauge change to
maintain the Coulomb condition at all Langevin times. There is some danger that the
gauge changing will also lead to a radiative correction to the time relation; it turns out it
does not, but I will go through the calculation anyway to show that it does not, since the
cancellations may be special to strict Coulomb gauge.
The Coulomb gauge condition is DL P A a Ž x . s 0, which means
Ý Aai Ž x . y Aai Ž x y aiˆ . s 0 . Ž A.11 .

i
This does not look cubic invariant because our labeling associates Ui , and hence A i , on
the link between x and x q aiˆ with the site x; it might better be associated with the
ˆ which would make the cubic invariance more obvious.
point x q air2,
To maintain this Coulomb condition, we must apply a time-dependent gauge transfor-
mation at each site. The difference between Coulomb gauge and temporal gauge
satisfying the Coulomb gauge condition at t s 0 will be a gauge transformation by
L s T expHigT a G a Žt . dt ., where the T means the exponential should be time ordered
with respect to t . The value of G is fixed by the requirement that the Coulomb
condition remain true; we must choose G so the departure from the gauge condition due
to evolution of the fields and due to the action of G cancel. In an infinitesimal time
interval dt the gauge change alters U through
d Ui Ž x . s Ž 1 y igT a G a Ž x . dt . Ui Ž x . Ž 1 q igT b G b Ž x q aiˆ . dt . y Ui Ž x . .
Ž A.12 .
Expanding U on each side, the G contribution to the time evolution of A is
dA ai G d Ž x q aiˆ . y G d Ž x . g 2 a2
dt
Ž from G . s
a ž da d q
12
f a b e f c d e A bi Ž x . Aci Ž x .
/
G c Ž x q aiˆ . q G c Ž x .
y g f a b c A bi Ž x . . Ž A.13 .
2
Now we need to substitute this expression, and the relation between E and dArdt ,
into Eq. ŽA.11. to determine the relation between E and G. The result, Fourier
transformed to momentum space, is
ig 2 a2
0 s ik˜ i Eia Ž k . y k˜ i f a b e f c d e b c
Hlm A Ž l . A Ž m . E
i i i
d
Ž k y l y m . y k˜ 2 G Ž k .
24
y igf a b c k˜ i cos Ž al ir2 . A bi Ž k y l . G c Ž l .

Hl
g 2 a2
y k˜ i k˜ i f a b e f c d e b
i
c
Hlm A Ž l . A Ž k y l y m . G
i
d
Ž m. . Ž A.14 .
12
Here summation over vector indices is implied in terms where the index appears at least
2 times. I have used the conventional shorthand k˜ i s Ž2ra.sinŽ k i ar2., k˜ 2 s Ý i k˜ i2 , and
Hl s Hd 3 lrŽ2p . 3, where the range of the integration is wypra,prax 3. Indices are
summed whenever the index appears an even number of times, which can be more than
twice in cubic symmetric but rotationally non-symmetric expressions; but if an index
appears an even number of times on each side of a q or y sign I am re-using it, it is an
independent index in each expression. To get the continuum version of this expression,
drop the O Ž a2 . terms and set k˜ s k, cosŽ kar2. s 1.
We can determine G perturbatively in g by expanding Eq. ŽA.14. in powers of g.
The result to O Ž g 2 . is
k˜ i Eia k˜ i l˜j
GaŽ k. si q gf a b c Hl k˜ cos Ž l˜i ar2 . A bi Ž k y l . Ejc Ž l .
k˜ 2
l˜
2 2
k˜ i l˜j m̃ k
y ig 2 f a b e f c d e Hlm ˜ cos Ž l˜i ar2 . cos Ž m
˜ j ar2 . A bi Ž k y l .
k l˜ m̃
2 2 2
ig 2 a 2 k˜ i
= Acj Ž l y m . Ekd Ž m . y fa b e fc d e b c
Hlm A Ž l . A Ž m . E
i i i
d
Ž k y l y m.
24 k˜ 2
ig 2 a 2 k˜ i k˜ i m̃ j
y fa b e fc d e Hlm A bi Ž l . Aci Ž k y l y m . Ejd Ž m . . Ž A.15 .
12 k˜ 2 m̃2
This in turn must be substituted into Eq. ŽA.13. to find the true value of dArdt in
Coulomb gauge,
dA ai Ž k .
dt
k˜ i k˜ j l˜k k˜ i k˜ j
ž
s di j y
k˜ 2 / Eja q igf a b c Hl l˜ 2 ž di j y
k˜ 2 /
cos Ž l j ar2 . A bj Ž k y l . Ekc Ž l .
y1 k˜ i k˜ j
q g 2 a2 f a b e f c d e Hlm 24 ž di j y
k˜ 2 / A bj Ž l . Acj Ž m . Ejd Ž k y l y m .
1 ˜ i m˜ k d i j
m k˜ i k˜ j k˜ j m
˜k l˜k m
˜l
y
12 ž ˜
m 2
y
k˜ 2 m
˜2 / A bj Ž k y l y m . Acj Ž l . Ekd Ž m . y
l˜2 m
˜2
k˜ i k˜ j
ž
= di j y
k̃ 2 / cos Ž l j ar2 . cos Ž m k ar2 . A bj Ž k y l . Ack Ž l y m . Eld Ž m . .
Ž A.16 .
Below I will be interested in the case where E is uncorrelated with A, in which case
the mean value of the term linear in A vanishes and we can substitute the leading order
Landau gauge A field correlator,
k˜ i k˜ j y1
Ž k˜ 2 .
² A ai Ž k . A bj Ž l . : s d a b d Ž k q l . d i j y
ž k̃ 2 / , Ž A.17 .
into the remaining terms. The contributions from the terms with Ž1r12. and Ž1. in front
vanish, only the term with the leading Ž1r24. coefficient contributes. It requires that we
perform the integral
1 S
a2 Hl l˜ 2
sa , Ž A.18 .
4p
which is the definition of the constant S which appeared earlier. We also need the
integral
l˜12 a S
a2 Hl 2
s . Ž A.19 .
Ž l˜ 2 . 3 4p
The result is
dA ai Ž k . k˜ i k˜ j 1 Ng 2 aT S
dt ž
Ž lattice. s d i j y
k˜ 2 / 1q
9 4 4p
Eja Ž k . , Ž A.20 .
valid when E is uncorrelated with gauge fields. The first factor projects out the
transverse component of E and is responsible for maintaining Coulomb gauge. The
continuum expression is the same but with a set to zero.
Using the previously established relation between lattice and continuum gauge field
normalization, the relation for the continuum normalized gauge field is
dA aa Ž k . Ž contin. Ng 2 aT 1 S j k˜ i k˜ j
dt Ž latt .
s 1q
4 ž 6 4p
q3
4p /ž di j y
k˜ 2 / Eja Ž k . , Ž A.21 .
valid for the gauge field response to the random force part of the Langevin equation. I
will not attempt to study the response of A from the Hamiltonian gradient part of the
Langevin equation, since it would involve understanding the radiative corrections to
D a H and would lead to an E which might be correlated with A fields. Since we know
that the Langevin equation correctly thermalizes the lattice system when we use the
radiatively corrected value of bL , it is sufficient to study the response to the random
force alone to determine the rescaling of the Langevin time scale.
Note that the correction appearing in Eq. ŽA.21. is precisely one quarter of the
radiative wave function correction for an adjoint scalar field in 3-D lattice gauge theory
when the scalar self-coupling vanishes, see Refs. w34,35x. This is perhaps not too
surprising. In the real time theory, the time evolution is generated by the electric fields,
which appear in the thermodynamics as the A 0 field, an adjoint scalar with zero
self-coupling. In w11x Turok and I speculated incorrectly that the time scale correction
would not contain any large tadpole corrections; but Eq. ŽA.21. is one quarter the adjoint
scalar wave function correction, which does contain tadpoles. Our incorrect guess was
based on analyzing the abelian theory, where the A 0 field turns out to be a free field.
The correction found here vanishes in the abelian case, for compact or non-compact
implementations.
Now I will finish the relation between time scales. I have just shown that the
response of the A field to the random force, shifting its normalization to correspond to
the continuum theory normalization, is
dA ai Ž x .
s Ž 1 q corr . j , Ž A.22 .
dt Ž latt .
where Ž1 q corr. is the quantity in brackets in Eq. ŽA.21.. But the autocorrelator of j is
not the same as it would be in the continuum, because of the radiative corrections to bL .
The mean square change in A ai over a Langevin time dt Žlatt. is, using Eq. ŽA.3.,
2Tdt Ž latt . 4
² Ž d A ai Ž x . . 2 : s Ž 1qcorr. 2 3 2
. Ž A.23 .
a g aTbL
However, bL s 4rg 2 aT only at leading order in bL . Beyond leading order, in the pure
gauge theory, it is w34,35x
bL g 2 aT g 2 aT N 2y2 37N j
4
y 1 ' Zy1
g y1s
4 ž 3N
q
3 4p / . Ž A.24 .
The continuum theory Langevin equation would cause a mean square change to A of
2Tdt Ž contin.
² Ž d A ai Ž x . . 2 : s , Ž A.25 .
a3
so the relation between time scales is
dt Ž contin. 2
g 2 aT N S 19 j N 2y2
dt Ž latt .
s Ž 1 q corr . Z g s 1 q
4 ž 3 4p
y
3 4p
y
3N / .
Ž A.26 .
The numerical value of this expression for N s 2 is 1 y 0.3189Ž g 2 aTr4..

I have now related the lattice and continuum Langevin time scales at O Ž a.. It is also
easy to show that, for the Hamiltonian system, the correction between the lattice and
continuum time scales is exactly half as large. However, this is less useful in that case
because, while this correction is technically correct for determining the time falloff of
correlators over very short time scales, the IR dynamics on longer time scales receive
HTL corrections which depend on the lattice spacing as 1ra q O Ž1.. For the technique
¨
of Hu and Muller, HTL’s are included by adding ‘‘particle’’ degrees of freedom. In w13x
we work out the correction for time scales in the limit m2D 4 g 2 Tra, in an approxima-
tion which corresponds to setting Ž1 q corr. s 1. The correction found here changes our
result there from being t lattrtcontin s Zy2
g to being Zy2
g
Ž1 q corr.y2 . I used this correc-
tion in Section 5.
A.2. Lattice LangeÕin time and depth of heat bath
The effect of the heat bath algorithm on the infrared degrees of freedom is equivalent
to Langevin evolution. I will show this at tree level, which gives a tree relation between
Langevin time and the number of heat bath updates applied. The only O Ž a. correction to
this relation possible is an O Ž a. shift in time scales between the algorithms; to find the
magnitude of the shift I make a direct numerical measurement. I do not attempt a
perturbative calculation of the relation between continuum Langevin time and amount of
heat bath applied, beyond leading order.
First I will show that the effect of the heat bath algorithm on the IR degrees of
freedom is equivalent to the Langevin algorithm, and I find the relation between the
Langevin time scale to the number of links updated by heat bath, at leading order. The
way the heat bath algorithm works is as follows:
Ž1. pick a link on the lattice at random.

Ž2. replace the connection U on the link with the one which minimizes the
Hamiltonian.
Ž3. multiply this link by a random SUŽ2. element chosen from a distribution
centered on the identity, with a weight function dependent only on the arc
length from the identity, not the direction. The weight is chosen to correctly
reproduce the thermal ensemble on this link holding others fixed; it is
approximately but not exactly Gaussian.
This is a heat bath update; for a more precise description see Ref. w39x. The first part,
the quench of the link, serves to perform the yE HrE A part of the Langevin update, and
the multiplication by a random SUŽ2. element reproduces the noise part of the Langevin
update.
To see the relation between the heat bath and the Langevin update at leading order it
is sufficient to consider the linearized theory, that is, to expand the Hamiltonian to
quadratic order in the gauge field A. On an N = N = N toroidal lattice the relation
between the connections U, the gauge field A i Ž x ., and the Fourier transform of the
gauge field AŽ k, s . Ž s a polarization index. is Žwriting all Lorentz and group index sums
explicitly, there is no implicit summation convention in what follows.
g 2 a2
Ui Ž x . s exp žÝ igaT aA ai /
Ž x. ,1q Ý igaT aA ai Ž x. y Ý Aai Ž x . Aai Ž x . ,
a a 8 a
Ž A.27 .
ˆ .. ,
A a Ž k , s . s Ny3r2 Ý e i Ž s,k . A ai Ž x . exp Ž yik P Ž x q air2 Ž A.28 .
x ,i
ˆ .. ,
A ai Ž x . s Ž longitudinal piece . q Ny3r2 Ý A a Ž k , s . e i Ž s,k . exp Ž ik P Ž x q air2
k,s
Ž A.29 .
where e i Ž s,k . is a transverse polarization vector, satisfying
Ý e i Ž s,k . e i Ž sX ,k . s ds, s X Ž A.30 .
i
and
Ý e i Ž s,k . k˜ i s 0 ; Ž A.31 .
i
there are two such states for each k. The sum over k includes all k of form Ž2praN . n,
with n a triple of integers each in the range 0 ( n i - N. Only the transverse fields are of
interest here, at the order we are working the longitudinal part is pure gauge. Its
behavior depends on our choice of gauge fixing, and has no influence on the transverse
fields Žor on physics.. ŽThis is not true at higher order in g, where the longitudinal terms
are responsible for the Faddeev–Popov determinant.. At the level we are working the
value in thermal equilibrium of A is given by
) T
² Ž A a Ž k , s . . A b Ž kX , sX . : s d k , k X d a b d s, sX . Ž A.32 .
k̃ 2
Now let us analyze how the fields change when the heat bath algorithm is applied to
a link Ž x,i .. The terms in the Hamiltonian containing the link x,i are
H 1 2
T
> 2
a T
Ý
a, j/i
1
2 ž A Ž x . q A Ž x q ajˆ . y A Ž x q aiˆ . y A Ž x . /
a
j
a
i
a
j
a
i
2
q 12 yA aj Ž x y ajˆ . q A ai Ž x y ajˆ . q A aj x y a Ž iˆq jˆ . y A ai Ž x .
ž ž / / ,
Ž A.33 .
and ‘‘quench’’ part of the heat bath algorithm will replace A ai Ž x . with the value which
minimizes this expression,
1
A ai Ž x ,after . s
4
Ý ž Aaj Ž x . q Aai Ž x q ajˆ . y Aaj Ž x q aiˆ . y Aaj Ž x y ajˆ .
j/i
qA ai Ž x y ajˆ . q A aj x y a Ž iˆq jˆ .
ž // . Ž A.34 .
Using equation ŽA.29. and Eq. ŽA.31., and adding a term j a to represent the noise
which will be added, this is
A ai Ž x . Ž after .
a2 k˜ 2
s A ai Ž x ,before . q j a y Ý ˆ .. .
e i Ž s,k . A a Ž k , s . exp Ž ik P Ž x q iar2
k,s N 3r2
Ž A.35 .
Now we should compute the size of j a. Because the Hamiltonian is expanded only to
quadratic order, the noise is Gaussian, of amplitude set by the size of the quadratic in
A i Ž x . term in H, which from Eq. ŽA.33. is Ž2ra2 T .Ý a A ai Ž x . A ai Ž x .. The amplitude of
the noise j a is then Žno sum on a. ² j aj a : s a 2 Tr4.
Next we will see what impact this update has had on the Fourier mode AŽ k, s ..
Combining Eqs. ŽA.28. and ŽA.35., we find
e i2 Ž s,k . a2 k˜ 2
A a Ž k , s,after . s A a Ž k , s,before . 1 y
ž 4N 3 /
e i Ž s,k .
q ˆ ..j a
exp Ž yik P Ž x q air2
N 3r2
a 2 l˜ 2e i Ž s,k . e i Ž sX ,l .
y Ý A a Ž l, sX ,before .
X
Ž s ,l ./ Ž s, k . 4N 3
ˆ .. .
= exp Ž i Ž k y l . P Ž x q air2 Ž A.36 .
The Ž k, s . term in the sum is removed and included instead in the first term. It is
responsible for the damping term in the Langevin equation. Both the noise term j and
the final term, which I will call the ‘‘cross-talk’’ term, are responsible for the noise term
in the Langevin equation.
To measure the magnitude of the Langevin damping term, we must compute the
correlator ² A) Ž k, s,before. AŽ k, s,after.:. Because ² AŽ k, s . AŽ l, sX .: s 0 s ² AŽ l, sX . j :,
we get
a2 k˜ 2e i2 Ž s,k .
² A) Ž k , s,before . A Ž k , s,after . : s ² A Ž k , s,before . 2 : 1
ž 4N 3 / . Ž A.37 .
It is also important to make sure that the mean square value of A a Ž k, s . is unchanged
by the update, which is the requirement that the noise have the right amplitude. Here we
get a little surprise; squaring Eq. ŽA.36.,
² Ž A a Ž k , s,after . . 2 :
2
a2 k˜ 2e i2 Ž s,k . e i2 Ž s,k .
² Ž A a Ž k , s,before . 2 : q ²Ž j a . 2 :
s 1y
ž 4N 3 / N3
2
1 a 2 l˜ 2
² A a Ž l, sX . A a Ž l, sX . :e i2 Ž s,k . e i2 Ž sX ,l .
q
N6
Ý
Ž l , sX ./ Ž k , s .
ž / 4
. Ž A.38 .
Using Eq. ŽA.32. and taking N 3 4 1, this becomes
² Ž A a Ž k , s,after . . 2 :
2
a2 Te i2 Ž s,k . 1 1 1
s ² Ž A a Ž k , s,before . :
N 3 ž y
2
q
4
q
16 N 3
Ý e i2 Ž sX ,l . a2 l˜ 2
X / .
l, s
Ž A.39 .
At leading order in large N, the sum gives 4 N 3. Therefore the mean size of A a Ž k, s . is
unchanged, which means that we have the correct amount of noise. We see that fully
half of the noise actually arises from ‘‘cross-talk’’ between the mode of interest and
extremely UV modes, with the other half arising from the noise explicitly appearing in
the algorithm.
To be Langevin the noise must have zero unequal-time correlation. This is the case
for j by explicit construction, but we need to check it for the ‘‘cross-talk’’ noise. For
the heat bath algorithm to act like a Langevin algorithm, the ‘‘cross-talk’’ contribution
to AŽ k, s . from updating the Ž x,i . link must be independent of that from the Ž y, j . link,
at least on averaging over the choice of Ž y, j . Žwhich is indeed chosen randomly in the
algorithm I use.. This is the case; from Eq. ŽA.36., the correlation between the
‘‘cross-talks’’ is
a4 l˜2 m̃2
e i Ž s,k . e j Ž s,k . e i Ž sX ,l . e j Ž sXX ,m . ² A a Ž l, sX .
X
Ý Ý
XX
Ž l , s ./ Ž k , s . Ž m , s ./ Ž k , s .
ž 16 N 6 /
= A a Ž m, sXX . :exp ik P x y y q a Ž iˆy jˆ. r2 exp Ž yil P Ž x q air2
ž ž ˆ . //
a 4 l˜2
ˆ .. s
qim P Ž y q ajr2 Ý e i Ž s,k . e j Ž s,k . e i Ž sX ,l . e j Ž sX ,l .
Ž l , sX . / Ž k , s . 16 N 6
= exp i Ž k y l . P x y y q a Ž iˆy jˆ. r2

ž ž // , Ž A.40 .
which suffers from a rapidly oscillating phase. The expression is typically smaller in
magnitude by Ny3 r2 compared to the corresponding term in Eq. ŽA.38. and its average
over Ž y, j . is strictly zero. Hence there is no unequal-time correlation in the ‘‘cross-talk’’
part of the noise. Note also that the cross-talk is very strongly UV dominated, which
means that there will be no hidden correlations in the IR effective evolution because of
it, at least at O Ž a. and probably higher; it is also fortunate because the UV is most
quickly randomized.
What we have shown is that applying the heat bath update is equivalent to damping
the A fields and applying noise. In particular, its influence on the IR degrees of freedom
is equivalent to that of Langevin dynamics. Applying the heat bath algorithm to many
links in succession, the rate at which a mode is damped Žand the amount of noise it
receives. is given by Eq. ŽA.37. Žand Eq. ŽA.39.. after averaging over the direction i
Žsince each direction is bathed with equal frequency.. Using Ý i e i2 Ž s,k . s 1, we find that
it takes 12 N 3 heat bath updates to perform the equivalent of a 2 of Langevin update.
This gives the tree relation between the algorithms. We can then define a heat bath time
in terms of the number of links we have updated, t Žheatbath. s a 2 Žnumber of links
updated.rŽ12 N 3 .. At leading order in a weak field expansion this is the same as
Langevin time but we expect subleading corrections.
Now we must push the analysis beyond tree level. Since the influence of the heat
bath algorithm on the IR degrees of freedom Žin fact, all degrees of freedom. ‘‘looks
like’’ Langevin dynamics, the analysis of w37x applies; up to high dimension operator
corrections, which by power counting appear first at O Ž a2 ., the algorithms are related,
in the presence of interactions, by a rescaling of all parameters. Since each algorithm
gives correct thermodynamic behavior Žafter the O Ž a. correction already discussed is
™
applied., the only remaining correction would be a rescaling of the time scale, which
must be at worst O Ž a. since it vanishes as g 2 0 Žin which limit the calculation just
presented is exact., and on-dimensional grounds any O Ž g 2 . correction must be O Ž g 2 aT ..
We could in principle determine this O Ž a. correction by an analytic computation,
extending the one just presented to second order in g. Instead I compute the subleading
effects by the following strategy. I choose some infrared measurable O , and measure it
at each lattice point at a closely spaced series of Langevin times. I do the same using the
heat bath algorithm. Then I compare the unequal-time correlator or autocorrelator
C Žt y t X . s ² O Ž x,t . O Ž x,t X .:, where the average is over the ensemble of Langevin
trajectories, or in practice over coordinates and times in a single very long Langevin
trajectory. To match the time scales, we see what rescaling of the heat bath time scale is
needed to make the autocorrelator match the autocorrelator for the Langevin case. Any
IR measurable will do because we know that the algorithms both behave as Langevin
algorithms on the IR degrees of freedom, so the only O Ž a. difference would be a time
rescaling which will be of the same amplitude for any unequal-time observable.
I should explain that the reason this is worth doing at all is that, first, the heat bath
algorithm is much faster and does not suffer from step size errors like the Langevin
algorithm, and second, there are infrared measurables other than the topological density,
for which the autocorrelation statistics improve much more quickly. If the latter were not
true we would spend as much computer time making the match between techniques as it
would take to do the measurement of G by the Langevin method.
The measurable I choose is a fundamental representation Wilson loop after some
amount of cooling, specifically a 4 = 4 square Wilson loop after t s 3.125a 2 of gradient
flow cooling under H KS . This is an infrared measurable because such a large Wilson
loop samples mostly the infrared gauge fields, and because the cooling removes most of
the UV fluctuations anyway.
Incidentally, it is not too hard to compute the leading order perturbative prediction for
this quantity. The mean trace of an l = l Wilson loop in SUŽ N . gauge theory after a
length t of gradient flow cooling, neglecting lattice artifacts, is
g 2T d3k 16sin2 Ž k x lr2 . sin2 Ž k y lr2 . k 2x q k 2y

N y Tr Ul=l s Ž N 2 y 1 . H Ž 2p . 3
4 k 2x k 2y k2
= exp Ž y2 k 2t . q O Ž g 4 . . Ž A.41 .
™
In the t 0 limit the integral has logarithmic UV divergences but for finite t it has a
well defined value and is dominated by the infrared, k Q ty1 r2 . Of course, for the
Wilson loops under consideration here, perturbation theory will be unreliable because
the length scales involved are close to the scale where perturbation theory breaks down
completely. It might be interesting to see whether the infrared fields are stronger or
weaker than at leading order in perturbation theory, though.
I measured the same site, unequal-time Wilson loop correlator by measuring each
4 = 4 Wilson loop in an even sublattice every a 2r2 of Langevin time, for a series of
Langevin trajectories each about 200 a2 long, with 50 a 2 Langevin time between
trajectories to eliminate correlations between trajectories. For each trajectory I deter-
mined the autocorrelator averaged over volume and time. The autocorrelation function
C Ž Dt . looks something like an exponential tail but is not well fit by one; there is some
small amount of much longer time scale correlation caused by the slow evolution of the
most infrared gauge fields. To compare the Langevin and heat bath time scales I
averaged C Ž Dt . over data sets for each update method and determined the rescaling of
the Langevin time which minimized the difference between the results,
Dtmax 2
x 2s H0 Ž C Langevin Ž Dt . y rC C heat Ž rt Dt . . d Dt , Ž A.42 .
where I allow a rescaling both of the autocorrelation time and of the overall magnitude
of C. I chose Dtmax to be enough that the autocorrelator has fallen about 1 y ey1 of the
way to its large Dt limit, but the result is quite insensitive to the specific choice. The
coefficient rt at the extremum of x 2 gives the rescaling of the Langevin time scale. The
multiplicative rescaling of C is necessary because there are small, very long time scale
correlations in the measurable which can effectively shift one data set somewhat with
respect to the other. As a check I compared the first half of the Langevin data I took
with the second half. I find that rt s 1 within a small tolerance the same size as the
jackknife error bars, but rC differs from 1 by a few percent.
For Ž4rg 2 aT . s 6 I ran each update procedure on a 24 3 lattice for t , 9000 a2 . The
rescaling of the time scales was
Dt heat s 1.098 " 0.007Dt Langevin , Ž A.43 .
with the error bar determined by the jackknife method.
The Langevin step size used here was D s 0.05. The definition of D and the
second-order algorithm I used are in Section 4 in the body of the paper. To check for
step size errors I evolved a trajectory for half as much Langevin time, using D s 0.025.
The rescaling between this trajectory and the heat bath was 1.099 " 0.010 and the
rescaling between Langevin evolutions with the two step sizes was 1.001 " 0.009, so
Langevin step size errors are negligible at D s 0.05.
I also measured NCS during the D s 0.05 Langevin trajectory; the ratio of the
diffusion constants for the heat bath and Langevin algorithms was
G heat
Ž unrescaled. s 1.114 " 0.058 , Ž A.44 .
G Langevin
which is compatible with the measured difference in time scales, but with much larger
error bars. The autocorrelator of the Wilson loop develops good statistics more quickly
than the diffusion constant for NCS .
Incidentally the mean value of the Wilson loop trace was ²2 y Tr U4= 4 : s 0.261,
while the leading order perturbative prediction is 0.096. The infrared of Yang–Mills
theory has more excitation than leading order perturbation theory predicts, by quite a bit
on the scale of a square Wilson loop 8rŽ3 g 2 T . on a side.
For Ž4rg 2 aT . s 10 I used a 40 3 lattice but only t s 2000 a 2 . The rescaling of the
time scales was smaller as expected, 1.065 " 0.016. Assuming the subleading correction
to be purely O Ž a., we would have guessed from the bL s 6 result that the rescaling
factor would be 1.059 " 0.004, which is within error. The rescaling at both lattice
spacings are within error of being t Langevinrt heat s Z g , and I speculate that this is the
correct analytic relation at O Ž a.. For the finest lattice spacing, I have simply extrapo-
lated the medium spacing data assuming a pure O Ž a. form for the correction.
The deviation of the mean value of the Wilson loop measurable from perturbation
theory is also smaller on the finer lattice; the value is 0.110, while the perturbative
estimate is 0.058. This is also expected, since a 4 = 4 Wilson loop is smaller in physical
units in this case, and perturbation theory works better as the length scale becomes
smaller.
References
w1x G. ’t Hooft, Phys. Rev. Lett. 37 Ž1976. 8.

w2x V. Kuzmin, V. Rubakov, M. Shaposhnikov, Phys. Lett. B 155 Ž1985. 36.
w3x P. Arnold, L. McLerran, Phys. Rev. D 36 Ž1987. 581.
w4x F. Klinkhamer, N. Manton, Phys. Rev. D 30 Ž1984. 2212.
w5x S. Khlebnikov, M. Shaposhnikov, Nucl. Phys. B 308 Ž1988. 885.
w6x E. Mottola, S. Raby, Phys. Rev. D 42 Ž1990. 4202.
w7x V. Rubakov, M. Shaposhnikov, Phys. Usp. 39 Ž1996. 461, Usp. Fiz. Nauk 166 Ž1996. 493.
w8x P. Arnold, D. Son, L. Yaffe, Phys. Rev. D 55 Ž1997. 6264.
w9x ¨
D. Bodeker, Phys. Lett. B 426 Ž1998. 351.
w10x J. Ambjørn, A. Krasnitz, Phys. Lett. B 362 Ž1995. 97.
w11x G.D. Moore, N. Turok, Phys. Rev. D 56 Ž1997. 6533.
w12x ¨
C. Hu, B. Muller, Phys. Lett. B 409 Ž1997. 377.
w13x ¨
G.D. Moore, C. Hu, B. Muller, Phys. Rev. D 58 Ž1998. 045001.
w14x K. Kajantie, M. Laine, J. Peisa, A. Rajantie, K. Rummukainen, M. Shaposhnikov, Phys. Rev. Lett. 79
Ž1997. 3130.
w15x M. Laine, O. Philipsen, Nucl. Phys. B 523 Ž1998. 267.
w16x P. Ginsparg, Nucl. Phys. B 170 Ž1980. 388.
w17x T. Applequist, R. Pisarski, Phys. Rev. D 23 Ž1981. 2305.
w18x S. Nadkarni, Phys. Rev. D 27 Ž1983. 917.
w19x N.P. Landsman, Nucl. Phys. B 322 Ž1989. 498.
w20x K. Farakos, K. Kajantie, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 425 Ž1994. 67.
w21x K. Kajantie, M. Laine, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 458 Ž1996. 90.
w22x P. Arnold, D. Son, L. Yaffe, Phys. Rev. D 59 Ž1999. 105020.
w23x J. Ambjørn, P. Olesen, Acta Phys. Polon. B 21 Ž1990. 109.
w24x R. Pisarski, Phys. Rev. D 47 Ž1993. 5589.
w25x J. Taylor, S. Wong, Nucl. Phys. B 346 Ž1990. 115.
w26x J. Frenkel, J. Taylor, Nucl. Phys. B 374 Ž1992. 156.
w27x E. Braaten, R. Pisarski, Phys. Rev. D 45 Ž1992. 1827.
w28x P. Huet, D. Son, Phys. Lett. B 393 Ž1997. 94.
w29x D. Son, UWrPT-97-19, hep-phr9707351.
w30x P. Arnold, L. Yaffe, Phys. Rev. D 52 Ž1995. 7208.
w31x ¨
D. Bodeker, L. McLerran, A. Smilga, Phys. Rev. D 52 Ž1995. 4675.
w32x P. Arnold, Phys. Rev. D 55 Ž1997. 7781.
w33x E. Iancu, Phys. Lett. B 435 Ž1998. 152.
w34x G.D. Moore, Nucl. Phys. B 493 Ž1997. 439.
w35x G.D. Moore, Nucl. Phys. B 523 Ž1998. 569.
w36x G.D. Moore, Phys. Rev. D 59 Ž1999. 014503.
w37x J. Zinn-Justin, D. Zwanziger, Nucl. Phys. B 295 Ž1987. 297.
w38x K. Farakos, K. Kajantie, K. Rummukainen, M. Shaposhnikov, Nucl. Phys. B 442 Ž1995. 317.
w39x A. Kennedy, B. Pendleton, Phys. Lett. B 156 Ž1985. 393.
M-theory resolution of four-dimensional cosmological

singularities via U-duality
A. Feinstein a , M.A. Vazquez-Mozo
´ b,c
a
´
Dpto. de Fısica ´
Teorica, UniÕersidad del Paıs´ Vasco, Apdo. 644, E-48080 Bilbao, Spain
b
Instituut Õoor Theoretische Fysica, UniÕersiteit Õan Amsterdam, Valckenierstraat 65, 1018 XE Amsterdam,
The Netherlands
c
Spinoza Instituut, UniÕersiteit Utrecht, LeuÕenlaan 4, 3584 CE Utrecht, The Netherlands
Received 9 June 1999; received in revised form 18 August 1999; accepted 1 November 1999
Abstract
We consider cosmological solutions of string and M-theory compactified to four dimensions by

giving a general prescription to construct four-dimensional modular cosmologies with two
commuting Killing vectors from vacuum solutions. By lifting these solutions to higher dimensions
we analyze the existence of cosmological singularities and find that, in the case of non-closed
Friedmann-Robertson-Walker universes, curvature singularities are removed from the higher-di-
mensional model when only one of the extra dimensions is time-varying. By studying the moduli
space of compactifications of M-theory resulting in homogeneous cosmologies in four dimensions
we show that U-duality transformations map singular cosmologies into non-singular ones. q 2000
PACS: 11.25.-w; 11.25.Mj; 11.25.Sq; 04.20.Ex; 04.20.Jb

Keywords: M-theory cosmology; M-theory; String cosmology; Cosmological singularities; Modular cosmol-
ogy
1. Introduction and motivation
It is well known that string physics has changed the way we now look at Cosmology.
In particular, string theory leads to consider multidimensional cosmological scenarios in
a natural way, since superstrings can only be consistently quantized in ten dimensions.
M-theory has confirmed this trend extending cosmology to the realm of the eleventh
dimension Žsee for example w1–8x..
E-mail addresses: wtpfexxa@lg.ehu.es ŽA. Feinstein., vazquez@wins.uva.nl,

´
M.Vazquez-Mozo@phys.uu.nl ŽM.A. Vazquez-Mozo ..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 9 7 - 5
406 ´
A. Feinstein, M.A. Vazquez-Mozor Nuclear Physics B 568 (2000) 405–420
Higher dimensional cosmology is not new. Before the advent of string theory the
Kaluza–Klein paradigm had already put forward the idea that four-dimensional space-
time should not be something taken for granted in cosmology Ža collection of the most
relevant articles on the subject can be found in w9x; see also Ref. w10x.. In its original
version, the Kaluza–Klein program aimed to describe all four-dimensional matter fields
as purely gravitational Žor supergravitational. degrees of freedom in 4 q N dimensions.
Although this idea met serious obstacles when trying to account for the existence of
chiral matter in four dimensions w11,12x, it has been partially incorporated in string
theory where the presence of ten-dimensional chiral matter fermions solves the problem.
When looking at the four-dimensional physics, the extra internal dimensions leave
their blueprints in the form of a plethora of massless scalar and vector fields. On general
grounds, the presence of these massless scalars poses a serious problem in trying to
extract realistic cosmological models from string theory. The situation is complicated by
the fact that most of these fields represent flat directions of the superpotential that are
not lifted by quantum corrections as long as enough supersymmetry is preserved. In
many occasions, however, the study is restricted to the simplest cosmological models by
looking just at the effects of the dilaton and tensor fields, for example, and ignoring the
remaining moduli fields altogether 1 w13,44–50x.
The ultimate motivation of stringrM-theory cosmology is of course to solve the
problem of cosmological singularities, explain the initial conditions in cosmology and
the dimensionality of space-time. Close to t s 0, quantum gravity effects become
dominant and hopefully will smear the semiclassical singularity, thus opening a new
window to the study of the early universe. Although a full quantum cosmology
description of the early universe is still missing, the semiclassical analysis supported by
the use of stringy symmetries has been useful in getting a flavor of the physics close to
the Big-Bang singularity w14,22–31x. On the other hand, from a Kaluza–Klein perspec-
tive it is possible that the existence of the initial singularity might just be the result of
integrating out the physics associated with higher dimensions w15–17x and that it could
be removed already in the semiclassical approximation.
In this paper we will combine these two ideas and investigate the effect of extra
dimensions on the initial cosmological singularity by studying a family of Friedmann-
Robertson-Walker ŽFRW. cosmologies coupled to a number of scalar fields which we
associate with compactification moduli. We find that in the particular case of open or
flat Ž k s y1,0. universes in which only one of the extra dimensions has a non-trivial
dynamics, the Big-Bang singularity might be just an artifact of the Kaluza–Klein
reduction which may be removed when going to higher dimensions. On the other hand,
in the case of spatially closed solutions Ž k s 1. the Big-Bang or the Big-Crunch
singularity is postponed in the higher-dimensional model, again when only one of the
extra dimensions is time-varying. It is interesting that by switching on more than one
dynamical internal dimension the regularization of curvature singularities is definitively
spoiled in all cases. When interpreting, in the open and flat cases, this higher dimen-
sional cosmologies as solutions in M-theory we find that the U-duality group G 7 acts on
1
From the M-theory perspective, on the other hand, the dilaton itself is just a moduli associated with the
compactification of the eleventh dimension. The reason to separate it from other fields is that in the weakly
coupled string limit of M-theory its compactification scale g s l s is much smaller than the string scale l s .
´
A. Feinstein, M.A. Vazquez-Mozor Nuclear Physics B 568 (2000) 405–420 407
the moduli space of metrics by relating singular geometries with non-singular ones. This
seems to indicate that a certain class of cosmological singularities in M-theory can be
physically resolved in terms of an equivalent dual regular background.
The plan of the article is the following. In Section 2 we will review the properties of
moduli fields arising from compactifications of Ž4 q N .-dimensional Einstein gravity on
a N-dimensional straight torus and provide a general algorithm to construct Gowdy-type
cosmologies in the presence of moduli fields. We will apply in Section 3 this algorithm
to generate modular FRW cosmologies and, after undoing the Kaluza–Klein reduction,
will study the structure of singularities of the higher dimensional ‘‘parent’’ metric. In
Section 4 the analysis will be focused on M-theory compactifications to four dimensions
and the action of U-duality on the moduli space of solutions. Finally, conclusions will be
summarized in Section 5.
2. Cosmologies coupled to scalar fields versus dimensional reduction
2.1. Scalar fields from dimensional reduction
Scalar fields appear naturally in the old Kaluza–Klein program or its stringrM-the-
ory versions. Here, we will consider the dimensional reduction of a Ž4 q N .-dimensional
metric on an N-dimensional straight torus T N s ŽS 1 . N using the ansatz 2
4
N N ci
2
2 ci '3 2
ds4qN s ey 3 ' Ý ds42 q Ý e Ž dw i . . Ž 1.
is 1
is1
The metric functions c i Ž x m . Ž m s 0, . . . ,3. should be thought of as the components of

the metric in the internal N-dimensional torus. Notice that this ansatz is invariant under
the permutation of the c i fields.
From the four-dimensional point of view, the fields c i Ž x m . are scalars. Their
dynamical equations are obtained by demanding that the Ž4 q N .-dimensional metric Ž1.
is a vacuum solution of the Einstein equations. Writing the Einstein–Hilbert action for
Ž1. we find that
N N
H (
S s d 4 x y g Ž4. R Ž 4 . y 2
ž Ý Em ci E m ci q 23 Ý Em ci E m cj
is1 i-j
/ , Ž 2.
leading to the result that the breathing modes of the higher dimensional metric appear as
mixed scalar in the four-dimensional action. To eliminate this mixing we may perform a
diagonalization in field space by defining the new fields w i through the relation
c i s Di j w j , Ž 3.
2
For a generic analysis of the structure of the dimensionally reduced action, see Refs. w37,38x.
408 ´
where Di j g GLŽ N,R. is given by

°m y1 r2
1 my1r2
2 my1r2
3 ... my1r2
Ny1
¶
my1r2
N
ymy1r2
1 my1r2
2 my1r2
3 ... my1r2
Ny1 my1r2
N
0 y2 my1r2
2 my1r2
3 ... my1r2
Ny1 my1r2
N
Ds .
.
.
.
.
.
.
.
.
.
Ž 4.
. . . . .
0 0 0 ... my1 r2
Ny 1 my1r2
N
¢ 0 0 0 ... y Ž N y 1 . my1r2
Ny 1 N
ß
my1r2
with
m n s 23 n Ž n q 1 . , n s 1, . . . , N y 1,
m N s 31 N Ž N q 2 . ,
which is not an O Ž N . transformation. The new scalar fields w i will be propagation
eigenstates and there will be no classical mixing among those. Their dynamics will be
governed by the action
N
H (
S s d 4 x y g Ž4. R Ž 4 . y 2 Ý Em wi E m wi . Ž 5.
is1
In our analysis we have assumed that all scalar fields have a geometric origin as
moduli of dimensional reduction. This point of view is very much appropriate for
M-theory where the dilaton is on the same footing with all other scalar as compactifica-
tion moduli. Nonetheless, in those regimes of M-theory that can be described in terms of
a weakly coupled string theory the dilaton field plays a privileged role as the field whose
vacuum expectation value determines the string coupling constant. In this case, dimen-
sional reduction from ten dimensions will produce again a number of scalar fields in the
lower-dimensional theory. The dynamics of those moduli can be extracted again from
the Kaluza–Klein ansatz
6
2
2
ds10 s ds42 q Ý e 2'2 s Ž dw i .
i , Ž 6.
is1
where now the ten-dimensional metric is no longer a vacuum solution of Einstein
equation but rather a solution of dilaton gravity instead. The resulting four-dimensional
action in string frame is w21x
6
H (
Sstring s d 4 x y g Ž4. ey2 f R Ž 4 . q 4Em f E mf y 2 Ý Em si E msi , Ž 7.
is1
where the four-dimensional dilaton f is defined in terms of the ten-dimensional one as
1 6
f s fŽ10. y Ý si . Ž 8.
'2 is1
We see from Ž7. that the kinetic term for the fields si is diagonalized in the
four-dimensional action, although it is conformally coupled to the dilaton field. This
conformal coupling can be removed as usual by a conformal transformation of the
metric. It is however when we re-express the four-dimensional dilaton in terms of the
ten-dimensional one via Ž8. that the mixing between the different si appears. In order to
recover the result from the compactification of a vacuum solution of M-theory ŽEq. Ž2.
´
with N s 7. we would need to lift the solution Ž6. of ten-dimensional dilaton gravity to
a vacuum solution in eleven dimensions.
2.2. Four-dimensional modular cosmologies from Õacuum solutions
In the following we will be interested in finding exact solutions to the field equations
derived from the four-dimensional action Ž5., that we can write in the manifestly
O Ž N .-invariant form as
S s d 4 x y g R y 2 EmF T E mF
H ' Ž 9.
where we have defined
w1
.
wN 0
F ' .. .
The matter energy–momentum tensor for Ž9. can be written as a sum of the correspond-
ing stress-energy tensors for each scalar field 3
N
Tmn s Ý TmnŽ i. s 2 Ž EmF T EnF y 12 gmn Es F T E sF . . Ž 10 .
is1
Let us concentrate our attention on Gowdy-type cosmologies with line element
p Žt , z .
ds 2 s e f Ž t , z . Ž ydt 2 q dz 2 . q K Ž t , z . e dx 2 q eypŽ t , z . dy 2 . Ž 11 .
At first sight, the Gowdy-type coordinates seem an unnecessary complication, since we
will be mostly dealing with cases where both geometries and scalar fields are homoge-
neous. The telling point to use them, nevertheless, is that: Ži. The scalar field equation
are linear in the metric functions. Žii. The evolution of the transversal metric functions
K Ž t, z . and pŽ t, z . is decoupled from scalar field dynamics, the longitudinal function
f Ž t, z . being the only one influenced by the presence of the scalar fields. And Žiii. that
due to the presence of a six-parameter isometry group G6 which includes the three-di-
mensional group G 3 acting simply transitively on the three-dimensional surfaces of
constant curvature in FRW models, the above line element, which has a G 2 isometry
group, naturally includes all three FRW geometries.
Let f Ž t, z . s f Ž t, z . vac such that Ž11. is a solution of the Einstein vacuum equations.
In this case the functions pŽ t, z . and K Ž t, z . satisfy the following conditions:
E E
K Ž t , z . p˙ Ž t , z . y K Ž t , z . pX Ž t , z . s 0
Et Ez
and
K¨ Ž t , z . y K XX Ž t , z . s 0.
The idea now is to solve Einstein equations with the energy–momentum tensor Ž10.. As
it happens with a single scalar field Žsee for example w18x., the transverse part
3
We have normalized the energy–momentum tensor in such a way that the Einstein equations are
Gmn sTmn .
410 ´
characterized by the metric functions K Ž t, z . and pŽ t, z . will be left unchanged. On the

other hand, the longitudinal function f Ž t, z . vac is replaced by
f Ž t , z . vac ™f Ž t, z. vac q f Ž t , z . sc
and the equations for f Ž t, z . sc can be written from the Einstein equations as
2K
f˙Ž t , z . sc s Ž K X Tt z y K˙ Tt t . ,
K y K˙ 2
X2
2K
f X Ž t , z . sc s Ž K X Tt t y K˙ Tt z . ,
K y K˙ 2
X2
where Tmn are the components of the energy–momentum tensor. Substituting Ž10. we
finally get
2K N N N
f˙Ž t , z . sc s 2 KX Ý w˙ i wiX y K˙ Ý w˙ i2 q Ý wiX 2
K X 2 y K˙ 2 is1
ž is1 is1
/ ,
2K N N N
f X Ž t , z . sc s KX Ý w˙ i2 q Ý wiX 2 y 2 K˙ Ý w˙ i w iX .
K X 2 y K˙ 2 ž is1 is1
/ is1
Ž 12 .
The structure of this expression Žsum over each scalar field. is the result of the fact that
Einstein equations are linear in the energy–momentum tensor and the energy–momen-
tum tensor itself is a sum of the contributions from each scalar field. Expressions Ž12.
are invariant under the global O Ž N . symmetry of Ž5., as one expects, since a rotation of
the fields by an element of this group does not modify the geometry. In addition, the
scalar fields w i must satisfy the wave equation
E E
K Ž t , z . w˙ i Ž t , z . y K Ž t , z . w iX Ž t , z . s 0, i s 1, . . . , N. Ž 13 .
Et Ez
Eqs. Ž12. can be used to generalize the algorithm of generation of string cosmologies
given in Ref. w13x. Taking w i s si Ž i s 1, . . . ,6. and w 7 s f , the four dimensional
dilaton, we generate exact solutions of the Einstein equations with seven scalar fields
that, after the conformal transformation by e 2 f will provide us with solutions to the
equations derived from the string theory action Ž7.. After this, we can further transform
the resulting four-dimensional metric by O Ž2,2;R. w13x or SLŽ2,R. w51x to generate other
fields, the moduli si remaining invariant under these transformations.
In addition to this, we can use this generation technique to directly construct
four-dimensional modular cosmologies representing toroidal compactifications of cos-
mological solutions of M-theory. In the following, we will particularize our analysis to a
certain class of these cosmological solutions that render homogeneous cosmologies in
four dimensions and use them to study how four-dimensional physics can be regularized
in eleven dimensions. This family of solutions will also be useful to study the moduli
space of M-theory on T 7.
´
3. FRW cosmologies with moduli fields
We will now apply what we have learnt to the construction of rolling moduli
cosmologies in four dimensions w32–35x. We start with a vacuum solution and apply the
algorithm Ž12. with a collection of N time-dependent scalar fields of the form
w i Ž t . s qi w 0 Ž t . , i s 1, . . . , N Ž 14 .
with qi a N-tuple of real numbers and w 0 Ž t . a particular homogeneous solution to the
wave Eq. Ž13.. The numbers qi 4 play now the role of coordinates in the moduli space of
solutions. From Ž12. we get the following equations for f Ž t, z . sc
N 2 KK˙
f˙Ž t , z . sc s y
ž /
Ý qi2
is1 K X 2 y K˙ 2
w˙ 02 ,
N 2 KK X
f X Ž t , z . sc s
žÝ /
is1
qi2
K X 2 y K˙ 2
w˙ 02 . Ž 15 .
Notice that in the family of solutions under consideration, the O Ž N . global symmetry of
the Einstein-Klein–Gordon action Ž5. acts linearly on the qi ’s and that the numerical
prefactor in Ž15. is just the O Ž N .-invariant metric on the moduli space.
In the generic case, the resulting family of metrics will always have strong curvature
singularities at some values of the time coordinate t. In what follows we will study what
happens to these singularities from a higher-dimensional point of view, when we undo
the Kaluza–Klein reduction.
3.1. Open FRW
Let us start with the following solution to the vacuum Einstein equations in four
dimensions
y1 r2
2
ds vac s Ž sinh2 t . Ž cosh4t y cosh4 z . 3 4 Ž ydt 2 q dz 2 .
q 12 sinh2 tsinh2 z Ž tanh z dx 2 q cotanh z dy 2 . Ž 16 .
and ‘‘dress’’ it with the homogeneous scalar fields Ž14. taking
'3
w0 Ž t . s log tanht.
2
The solution coupled to the N scalar fields is
1 3
Ž3 l y1 . Ž1y l .
ds 2 s Ž sinh2 t . 2
Ž cosh4t y cosh4 z . 4
Ž ydt 2 q dz 2 .
q 12 sinh2 tsinh2 z Ž tanh z dx 2 q cotanh z dy 2 . , Ž 17 .
where we have defined l as the O Ž N .-invariant quadratic form
N
l' Ý qi2 . Ž 18 .
is1
The interesting feature of these solutions is that the dynamics of the scalar fields is
only relevant at early times, saturating to a constant when t `. One would expect this
sort of behavior in ‘‘realistic’’ models for the modular fields in stringrM-theory
™
412 ´
cosmology. The scalar fields are supposed to play an important role in the evolution
during the early epoch, but are expected to be frozen at a certain vacuum expectation
value after supersymmetry breaking. From that time on, the dynamics of the universe is
dominated by matter or radiation.
The case with l s 1 is especially interesting, since we recover an open FRW
universe
dsl2s1 s sinh2 t Ž ydt 2 q dz 2 . q 12 sinh2 tsinh2 z Ž tanh z dx 2 q cotanh z dy 2 . , Ž 19 .

'3 N
wi Ž t . s
2
qi log tanht , with Ý qi2 s 1.
is1
It is straightforward to check that the metric Ž19. has a cosmological singularity at t s 0

where the curvature scalar blows up. Actually, the physical properties of the singularity
are independent of the particular values qi take, as long as the condition Ž18. with l s 1
is satisfied. This is in accordance with the requirement of O Ž N . global invariance of the
theory.
Let us now ‘‘re-construct’’ the Ž4 q N .-dimensional theory from which Ž19. is being
obtained by dimensional reduction. Here we have to keep in mind that the scalar fields
associated with the scale factors of the internal torus are related to a linear combination
of the original fields w i Ž t ., as shown above. This means that the higher dimensional
metric will be
N N
2 1y Ý p Ž cosht . 1q Ý p Ž ydt 2 q dz 2 q sinh2 z dx 2 q cosh2 z dy 2 .

i i
ds4q N s2 Ž sinht . is 1 is1
N
2
q Ý tanh2 p i t Ž dw i . , Ž 20 .
is1
where we have introduced the new constants pi s Di j q j , with Di j the matrix Ž4.. We
can now rewrite the condition l s 1 in terms of the new moduli space coordinates pi 4
as
N N
l' Ý pi2 q 23 Ý pi pj s 1. Ž 21 .
is1 i-j
The new parameters just provide a non-orthogonal system of coordinates in moduli
exponents in the N-dimensional internal torus when t 0. ™

space. Actually we see from Ž20. that the pi ’s can be thought of as a kind of Kasner
The metric Ž20. with l s 1 is a vacuum solution of Einstein equations in 4 q N

dimensions Žin fact it is such for any l.. Naively, the singularity seems to be at t s 0, as
it was in the original four-dimensional space-time. However, it may happen that this
apparent singularity is just the result of choosing a singular coordinate system. To clarify
™
this, we can evaluate the square of the Riemann curvature tensor Ra b c d R a b c d, which is
non-vanishing, in the t 0 limit
Ra b c d R a b c d ; C Ž pi . t 2Ž Sy3. q O w t 2Ž Sy2. x , Ž 22 .
´
N
where we have written S s Ý is1 pi and C Ž pi . is defined by
N
2
C Ž pi . s 163 Ž S y 1 . Ž S 2 y 2 S q 5 . q Ý pi2 3 q pi2 q Ž S y 3 . Ž pi q S .
is1
N
q Ý pi p j .
i-j
™
First, let us notice that S ( '3 l - 3 so the leading term in Ž22. will always diverge at
t 0 limit. Thus, in general, we will have curvature singularities in 4 q N dimensions
in this limit. However, we find the surprising result that whenever pi s 1, pj s 0 Ž j / i .
C Ž pi . s 0 and actually the curvature invariant is regular for all times and given by 4
18
Ra b c d R a b c d s . Ž 23 .
cosh8 t
This corresponds to having only one dynamically non-trivial cycle whose scale factor
evolves with time. In the four-dimensional metric, on the other hand, due to the
non-diagonal relation between pi 4 and qi 4 , in general all scalar fields will have a
non-trivial dynamics. Note, however, that the dimensionally reduced model Ž17. is
generally inhomogeneous for generic l, the regular higher-dimensional solution falling
into a homogeneous and isotropic class. One might wonder whether higher-dimensional
regularity and lower-dimensional isotropy are physically linked somehow.
3.2. Closed FRW
We now pass to discuss a four-dimensional metric with closed spatial sections. The
vacuum seed metric is obtained from Ž16. by replacing hyperbolic functions by their
trigonometric counterparts. We again couple the space-time to N scalar fields of the
form Ž14. now with
'3
w0 Ž t . s log tant.
2
Imposing as before l s 1, we are left with the family of four-dimensional metrics
dsl2s1 s sin2 t Ž ydt 2 q dz 2 . q 12 sin2 t sin2 z Ž tan z dx 2 q cotan z dy 2 . ,
'3 N
wi Ž t . s
2
qi log tan t , with Ý qi2 s 1. Ž 24 .
is1
The coordinate transformation from the standard closed FRW metric to Ž24. are given in
the Appendix of Ref. w18x, the coordinates z, x and y used here being just Euler angles
for S 3. In spite of the similarities with the flat case, the structure of cosmological
singularities is now richer. Studying the scalar curvature R we find that the metric Ž24.
has curvature singularities at t s l p2 with l being an integer. We have a bouncing
universe that evolves from a Big-Bang singularity at t s l p into a Big-Crunch at
t s Ž2 l q 1. p2 .
4
Actually, it is easy to realize that, provided this curvature invariant is finite, so are all other scalars
™
constructed from contractions of any number of Riemann tensors. In our case this follows from the fact that
R 1a;1 b is finite when t 0, where the index 1 denotes the dynamical internal dimension w 1.
414 ´
In going to 4 q N dimensions, we get

N N
2 1y Ý p Ž cos t . 1q Ý p Ž ydt 2 q dz 2 q sin2 z dx 2 q cos 2 z dy 2 .

i i
ds4q N s2 Ž sint . is 1 is1
N
2
q Ý tan2 p i t Ž dw i . , Ž 25 .
is1
where the same definition for the pi as in Subsection 3.1, satisfying the condition Ž21.,
has been used. From inspection of Ž25. we might be tempted to infer that the higher
™
dimensional theory will be singular again when t s l p2 . Computing the curvature
invariant when t 0q we get Eq. Ž22., as in the previous case. However, if we now
compute the same invariant close to the Big-Crunch singularity Ž t py
. we find ™
2
abcd y2Ž Sq3. y2 Ž Sy2.
Ra b c d R ; C Ž ypi . t qOw t x.
The situation is somewhat different as compared to the open case. There we found that
when the condition l s 1 is saturated by a single pi s 1, the initial Big-Bang singularity
disappears altogether by going to higher dimensions. Here we find that we do not get rid
of all curvature singularities, but only of ‘‘half’’ of them. The resulting geometry in this
case in only singular when t s p2 Ž2 l q 1., whereas those at t s p l are smeared in
4 q N dimensions. The curvature invariant now is given by Ž23. with the hyperbolic
cosine replaced by a trigonometric one. The reverse situation happens when we take
pi s y1 and pj s 0 Ž j / 0.: we avoid singularities located at t s p2 Ž2 l q 1., the
geometry being the same as for the previous case but now with t t q p2 .
In Ref. w17x a similar situation was noticed for a closed FRW cosmology that could
™
be resolved into a five-dimensional black hole interior with just one curvature singular-
ity in the past or in the future. The difference with the model analyzed here is that in the
case at hand our higher-dimensional universe still has a finite life, although it is doubled
with respect to the one of the four-dimensional geometry.
3.3. Flat FRW
Finally, let us briefly analyze the case of cosmological models with flat spatial
sections. The four-dimensional metric and scalar fields system is given by
ds 2 s 2 t Ž ydt 2 q dz 2 q dx 2 q dy 2 . ,
'3 N
wi Ž t . s
2
qi log t , with Ý qi2 s 1, Ž 26 .
is1
which is singular when t s 0. The higher-dimensional version of this metric according

to the ansatz Ž1. is
N N
2 1y Ý p i 2
ds4q N s2t Ž ydt 2 q dz 2 q dx 2 q dz 2 . q
is 1
Ý t 2 p Ž dw i . .
i
Ž 27 .
is1
We can transform this solution into a standard Kasner form by rewriting it in co-moving
time coordinates Žand re-scaling spatial coordinates. to give
N
2a0 2
2
ds4q 2
N s ydT q T Ž dX 2 q dY 2 q dZ 2 . q Ý T 2 a Ž dW i .
i , Ž 28 .
is1
´
where
N
1y Ý pj
js1 2 pi
a0 s N , ai s N Ž i s 1, . . . , N . , Ž 29 .
3y Ý pj 3y Ý pj
js1 js1
which can be easily shown to satisfy the Kasner equalities

N N 6 Ž l y 1.
3a0 q Ý a i s 1, 3 a 02 q Ý a i2 s 1 y 2
s 1.
N
is1 is1
ž 3y Ý pi
is1
/
Here, we have implemented l s 1 to get the second condition.
With this expression for the metric we can easily compute the curvature invariant,
with the result
4 N N
Ra b c d R a b c d s
t4
6 a 02 Ž 1 y a 0 . y 3 a 04 q Ý a i2 Ž a i y 1. 2 q Ý a i2a j2 .
is1 i-j
Using the definition of the Kasner exponents in terms of the original pi it is possible to
check that the only case in which we will have a regular geometry at t s 0 will again
occur if pi s 1 with all other pj Ž j / i . vanishing. In this case the resulting metric is
flat. Actually it can be seen to be Ž4 q N .-dimensional Minkowski space-time in Rindler
coordinates 5. The non-trivial topology of the internal dimensions will hinder global
identification of the manifold with static Minkowski solution.
4. M-theory connections
Let us particularize the study of the cosmological models of the previous section to
the case with N s 7 in which they can be interpreted as four-dimensional cosmologies
arising from a compactification of M-theory on T 7 s ŽS 1 . 7 with vanishing three-form. In
describing the moduli space of cosmological solutions coupled to scalar fields of the
form Ž14. we have two different possibilities. In the four-dimensional theory it seems
natural to take the coordinates qi Ž i s 1, . . . ,7. which characterize the seven independent
scalar fields coupled to gravity. Due to the global O Ž7. symmetry of the low-energy
action, there is a natural choice for the moduli space metric
7
I4 s Ý qi2 .
is1
The resulting geometry will only depend on qi through the quadratic form I4 as seen in
Eq. Ž15..
5
In Refs. w15,16x an inverse procedure was evoked by dimensionally reducing from a five-dimensional flat
model to four dimensions to argue that the cosmological singularity could be an artifact of dimensional
reduction.
416 ´
On the other hand, in eleven-dimensional language it seems more appropriate to

change coordinates in moduli space and use instead of qi 4 the new parameters pi 4
which determine the scale factors of the internal compactified dimensions and are
related to the original coordinates by a non-orthogonal linear transformation, pi s Di j q j .
If we now express the moduli space metric in the new coordinates, we find 6
7 7
I11 s Ý pi2 q 23 Ý pi pj . Ž 30 .
is1 i-j
Incidentally, this moduli space metric is the same as the one obtained in w14x Žsee also
Refs. w19,20x. for the compactifications of M-theory on a seven-torus with A M N P s 0
using group theoretical considerations. The O Ž7. global symmetry will act linearly on
the pi ’s leaving I11 invariant. It is however important to notice that the eleven-dimen-
sional geometry do transform under O Ž7., in contrast to the four-dimensional one that
was a singlet under the action of this group. In eleven dimensions the only transforma-
tions that leave invariant the space-time metric are permutations of the pi ’s, which
generates the Weyl subgroup of the mapping class group of the seven-dimensional torus,
SLŽ7,Z..
We will consider the family of M-theory metrics obtained by taking N s 7 in Ž20..
These line elements are vacuum solution of eleven-dimensional supergravity everywhere
in the submanifold of the moduli space defined by I11 s 1. We can ask now about the
action on this metric of G 7 ' Weylw E7Ž7.ŽZ.x, the subgroup of U-duality transformations
preserving the straight torus T 7 s ŽS 1 . 7 and the vanishing of the three-form w14,19,20x,
which is generated by permutations of the pi ’s and the so-called 2r5 transformation
Ž p1 , . . . , p 7 . ™ ž p y 23s , p y 23s , p y 23s , p q 3s , . . . , p q 3s /

1 2 3 3 7 Ž 31 .
with s s p 1 q p 2 q p 3 . It is easy to show that G 7 leaves invariant the bilinear form I11
and it is thus a discrete finite subgroup of O Ž7. connecting physically equivalent
M-theory vacua.
In the following we will restrict our analysis to the open FRW metric of Subsection
3.1 although most of our results can be extended to the flat case as well. Acting with the
elements of G 7 on the different solutions characterized by pi 4 we can look how
sensitive M-theory physics is to the semiclassical geometry. The most striking fact we
find is that the U-duality group maps solutions with a Big-bang initial singularity into
geometries that are regular for all values of the cosmic time. For example, the metric
2
2
ds11 s 2 Ž sinht cosh2 t . Ž ydt 2 q dz 2 q sinh2 z dx 2 q cosh2 z dy 2 .
3
5 7
2 2 y 43 2
q Ž tanht . 3
Ý Ž dw i . q Ž tanht . Ý Ž dw i .
is1 is6
6
Since we are just performing a change of coordinates, I4 s I11. We will use however different notation to
indicate the coordinates used to write the quadratic form.
´
which is singular at t s 0 Ž Ra b c d R a b c d ; ty16r3 . can be mapped by G 7 into one of the

non-singular M-theory cosmologies using the following sequence of permutations and
2r5 transformations
Ž 13 , 13 , 13 , 13 , 13 ,y 23 ,y 23 . ™ Ž ,y
1
3
2
3 ,y 23 , 13 , 13 , 13 , 13 . ™ Ž 1,0,0,0,0,0,0. .
2r5
We have concluded that using an U-duality transformation, which is an exact symmetry

of M-theory, we can transform a singular background into a non-singular one. The
obvious bottom line seems to be that, at least in the low-energy limit, M-theory physics
is insensitive to a certain class of cosmological singularities.
In the case of the open Ž k s y1. solution, however, the asymptotic form of the
™
metrics at large times is to a great extend left invariant by the action of G 7 . The general
feature of all solutions Ž20. when t ` is that the open three dimensional space-time
inflates while the internal torus reach a constant volume, namely
Vol 3D
;t 3,
Vol T 7
with t the co-moving time. Physically, what we have is a cosmological solution that
evolves from an eleven-dimensional regime in which all scale factors are of the same
size into an asymptotic state with describes a ‘‘large’’ four-dimensional expanding
universe and a ‘‘small’’ static seven-dimensional torus Žcf. Refs. w36,42,43x..
In the case of the flat solution, on the other hand, the large-time behavior of the
metric is not universal. As we saw in Subsection 3.3, in this case we can rewrite the
metric in the Kasner form Ž28. with exponents given by Ž29.. The U-duality group G 7
acts on these exponents in a rather complicated way through the transformation of the
moduli space coordinates pi 4 Žcf. Ref. w14x.. Since the sum Ý7is1 pi is not invariant
under the 2r5 transformation, the asymptotic form of the metrics Ž27. or Ž28. will be
sensitive to rotations by elements of the U-duality group. In addition to this, the internal
manifold will not remain ‘‘small’’ in general. We will have the usual behavior of any
Kasner universe, with at least one contracting direction and a number of expanding ones.
The main characteristic of the non-singular M-theory cosmologies that we have
obtained is that in all cases there is only one time-varying coordinate, while the
remaining ones are constant during the evolution of the Universe. In trying to make a
string theory interpretation of such solutions it seems natural to identify the dynamical
coordinate with the eleventh dimension associated with the dilaton field. If we do so and
perform a dimensional reduction of the open Ž k s y1. regular eleven-dimensional
solution down to ten dimensions, the resulting metric and dilaton in string frame are7
6
2 2
dsstring s sinh2 t Ž ydt 2 q dz 2 q sinh2 z dx 2 q cosh2 z dy 2 . q tanht Ý Ž dw i . ,
is1
f Ž t . s 32 log tanht
which corresponds to a string background with three open spatial dimensions and
another six compactified on a torus whose dynamics is characterized by a global
breathing mode Ža situation extensively studied in the Kaluza–Klein and string cosmol-
2 4
7 2
The appropriate Kaluza–Klein ansatz in this case is ds11 s ey 3 f ds10
2
q e 3 f dw 2 .
418 ´
™
ogy literature w22–35,39–41x.. The geometry is singular when t 0 but, on the other, at
large times we find again an inflating three-dimensional space together with a frozen
internal torus, now also with a constant dilaton. Notice, however, that M-theory
U-duality maps these kind of singular string cosmologies into the regular ones that are
obtained by identifying the dilaton field with one of the static circles. These transforma-
tions involve the interchange of the dilaton with other moduli fields and thus are
intrinsically non-perturbative from the point of view of string theory.
Incidentally, let us remark that because of the invariance of I11 ' l under O Ž7., the
family of solutions Ž17. is stable under the U-duality group which thus maps solutions
into solutions Žcf. Ref. w14x.. This is what one should expect, since G 7 is a symmetry of
M-theory on a straight torus with a vanishing three-form.
5. Concluding remarks
In this paper we have been concerned with four-dimensional cosmologies that arise
from the compactification of stringrM-theory on a straight torus. We have given a
general algorithm to generate these modular cosmologies in four dimensions with two
commuting isometries. Using this technique we have also constructed FRW metrics and
observed that, although they are singular in four dimensions, in the open Ž k s y1. and
flat Ž k s 0. cases their lifted higher-dimensional vacuum images have a regular curva-
ture invariant Ra b c d R a b c d. In the case of closed FRW cosmologies Ž k s 1., singulari-
ties are not removed but the lifetime of the higher-dimensional universe Ži.e. the time
elapsed from the Big-Bang to the Big-Crunch. is twice that of their four-dimensional
versions.
It is important to stress, however, that in general the homogeneous character of the
scale factors of the extra dimensions derived from Ž14. does not ensure the homogeneity
of the reduced four-dimensional cosmology. Given the absence of a singularity-free
higher-dimensional ‘‘parent’’ cosmology for the four-dimensional inhomogeneous solu-
tions, it is remarkable that the regularity condition for the space-time in higher
dimensions translates itself into isotropy of the four-dimensional solution. Thus, the
initially isotropic universe could be naturally chosen by the requirement of regularity of
the higher-dimensional space-time.
We have also studied four-dimensional modular cosmologies related to toroidal
compactifications of stringrM-theory. In the case of M-theory on a straight seven-torus
with vanishing three-form we have found open and flat singular metrics that can be
rotated into regular ones by the elements of the U-duality group G 7 , thus indicating that
the physics of a certain class of singular universe can be described in terms of a dual
regular geometry. In the case of homogeneous open universes, however, the large time
asymptotic behavior of the metric is insensitive to the action of G 7 . In all cases we find
an inflating three-dimensional space with k s y1 and an internal torus whose size
asymptotically reaches a constant value. It is remarkable that U-duality is able to relate
geometries with such a different behavior near t s 0 without modifying the dynamics of
the universe at large times. Incidentally, this stabilization of the internal dimensions is
produced by the presence of the positive spatial curvature and it is absent in the case of
the flat solutions.
´
Because of the O Ž7. invariance of the dimensionally reduced cosmologies, all the
eleven-dimensional metrics labeled by pi 4 and the same value of I11 will produce
exactly the same geometry in four dimensions, although it will be coupled to scalar
fields with different amplitudes in each case. In particular, using the classical symmetry
of the four-dimensional action any solution in our family with I4 s 1 can be rotated into
one with a regular higher-dimensional ‘‘parent’’ metric. However, since M-theory in
this background is not invariant under the full group O Ž7., one expects that quantum
effects will break this global symmetry down to the U-duality group G 7 which preserves
the lattice of charges. The situation is completely analogous to the classical invariance of
the four-dimensional effective action under O Ž6,6;R. w21x, a symmetry that is broken
down to the T-duality group O Ž6,6;Z. by string effects.
Finally, it would be interesting to check whether our results can be extended to
compactifications of M-theory on a generic seven-torus with a non-vanishing three-form.
From our analysis it is clear at least that there will be solutions of this kind that can be
transformed into non-singular universes by U-duality, as can be seen by acting on the
regular metric with the Borel generators of E7Ž7.ŽZ. w38x.
Acknowledgements
We thank J.L.F. Barbon ´ and J.L. Manes

˜ for interesting discussions. M.A.V.-M.
wishes also to thank K. Skenderis and E. Verlinde for discussions on Ref. w14x. The
work of A.F. has been supported by Spanish Science Ministry Grant 172.310-0250r96
and the University of the Basque Country Grant UPV 172.310-EB150r98 and that of
M.A.V.-M. by FOM ŽFundamenteel Ordenzoek der Materie. and by the University of
the Basque Country Grant UPV 063.310-EB187r98.
References
w1x ¨ S. Mukherji, C.N. Pope, K.-W. Xu, Phys. Rev. D 55 Ž1997. 7926. hep-thr9610107.
H. Lu,
w2x A. Lukas, B.A. Ovrut, D. Waldram, Nucl. Phys. B 495 Ž1997. 365. hep-thr9610238.
w3x A. Lukas, B. Ovrut, Phys. Lett. B 437 Ž1998. 291. hep-thr9709030.
w4x N. Kaloper, I. Kogan, K.A. Olive, Phys. Rev. D 57 Ž1998. 7340. hep-thr9711027.
w5x K. Benakli, Int. J. Mod. Phys. D 8 Ž1999. 153. hep-thr9804096.
w6x A. Lukas, B.A. Ovrut, D. Waldram, Phys. Rev. D 60 Ž1999. 086001. hep-thr9806022.
w7x H.S. Reall, Phys. Rev. D 59 Ž1999. 103506. hep-thr9809195.
w8x M. Maggiore, A. Riotto, Nucl. Phys. B 548 Ž1999. 427. hep-thr9811089.
w9x T. Appelquist, A. Chodos, P.G.O. Freund, Modern Kaluza–Klein Theories ŽAddison-Wesley, New York,
1987..
w10x J.M. Overduin, P.S. Wesson, Phys. Rep. 283 Ž1997. 303. hep-thr9805018.
w11x E. Witten, Nucl. Phys. B 186 Ž1981. 412.
w12x E. Witten, Fermion quantum numbers in Kaluza–Klein theory, Proceedings of the Shelter Island II
Conference ŽM.I.T. Press, 1985..
w13x ´
A. Feinstein, R. Lazkoz, M.A. Vazquez-Mozo, Phys. Rev. D 56 Ž1997. 5166. hep-thr9704173.
w14x T. Banks, W. Fischler, L. Motl, J. High Energy Physics 01 Ž1999. 019. hep-thr9811194.
w15x G.W. Gibbons, P.K. Townsend, Nucl. Phys. B 282 Ž1987. 610.
w16x G.W. Gibbons, G.T. Horowitz, P.K. Townsend, Class. Quant. Grav. 12 Ž1995. 297. hep-thr9410073.
w17x F. Larsen, F. Wilczek, Phys. Rev. D 55 Ž1997. 4591. hep-thr9610252.
420 ´
w18x M. Carmeli, C. Charach, A. Feinstein, Ann. of Phys. 150 Ž1983. 392.

w19x O. Aharony, Nucl. Phys. B 476 Ž1996. 470. hep-thr9604103.
w20x S. Elitzur, A. Giveon, D. Kutasov, E. Rabinovici, Nucl. Phys. B 509 Ž1998. 122. hep-thr9707217.
w21x J. Maharana, J.H. Schwarz, Nucl. Phys. B 390 Ž1993. 3. hep-thr9207016.
w22x E. Alvarez, Phys. Rev. D 31 Ž1985. 418.
w23x R. Brandenberger, C. Vafa, Nucl. Phys. B 316 Ž1989. 191.
w24x K.A. Meissner, G. Veneziano, Phys. Lett. B 267 Ž1991. 33.
w25x K.A. Meissner, G. Veneziano, Mod. Phys. Lett. A 6 Ž1991. 3397. hep-thr9110004.
w26x M. Gasperini, J. Maharana, G. Veneziano, Phys. Lett. B 296 Ž1992. 51. hep-thr9209052.
w27x M. Gasperini, G. Veneziano, Astropart. Phys. 1 Ž1997. 317. hep-thr9211021.
w28x N. Kaloper, R. Madden, K.A. Olive, Nucl. Phys. B 452 Ž1995. 677. hep-thr9506027.
w29x M. Gasperini, M. Maggiore, G. Veneziano, Nucl. Phys. B 949 Ž1997. 315. hep-thr9611039.
w30x R. Brandenberger, R. Easther, J. Maia, J. High Energy Physics 08 Ž1998. 007. gr-qcr9806111.
w31x D.A. Easson, R. Brandenberger, JHEP 09 Ž1999. 003. hep-thr9905175.
w32x M. Mueller, Nucl. Phys. B 337 Ž1990. 37.
w33x G. Veneziano, Phys. Lett. B 265 Ž1991. 287.
w34x A.A. Tseytlin, C. Vafa, Nucl. Phys. B 372 Ž1992. 443. hep-thr9109048.
w35x A.A. Tseytlin, Phys. Lett. B 334 Ž1994. 315. hep-thr9404191.
w36x A. Chodos, S. Detweiler, Phys. Rev. D 21 Ž1980. 2167.
w37x ¨ C.N. Pope, Nucl. Phys. B 465 Ž1996. 127. hep-thr9512012.
H. Lu,
w38x N.A. Obers, B. Pioline, Phys. Rep. 318 Ž1999. 113. hep-thr9809039.
w39x P.G.O. Freund, Nucl. Phys. B 209 Ž1982. 146.
w40x D. Sahdev, Phys. Lett. B 137 Ž1984. 155.
w41x S. Randjbar-Daemi, A. Salam, J. Straathdee, Phys. Lett. B 135 Ž1984. 388.
w42x J. Demaret, J.-L. Hanquin, Phys. Rev. D 31 Ž1985. 258.
w43x J. Demaret, J.-L. Hanquin, M. Henneaux, P. Spindel, Nucl. Phys. B 252 Ž1985. 538.
w44x E.J. Copelan, A. Lahiri, D. Wands, Phys. Rev. D 50 Ž1994. 4868. hep-thr9406216.
w45x N.A. Batakis, A.A. Kehagias, Nucl. Phys. B 449 Ž1995. 248. hep-thr9502007.
w46x N.A. Batakis, Phys. Lett. B 353 Ž1995. 450. hep-thr9503142.
w47x N.A. Batakis, Phys. Lett. B 353 Ž1995. 39. hep-thr9504057.
w48x J.D. Barrow, K.E. Kunze, Phys. Rev. D 55 Ž1997. 623. hep-thr9608045.
w49x J.D. Barrow, K.E. Kunze, Phys. Rev. D 56 Ž1997. 741. hep-thr9701085.
w50x J.D. Barrow, M.P. Dabrowski, Phys. Rev. D 55 Ž1997. 630. hep-thr9608136.
w51x D. Clancy, A. Feinstein, J. Lidsey, R. Tavakol, Phys. Rev. D 60 Ž1999. 043503. gr-qcr9901062.
Big bang nucleosynthesis: an accurate determination of light

element yields
S. Esposito, G. Mangano, G. Miele, O. Pisanti
Dipartimento di Fisica, UniÕersita´ di Napoli ‘‘Federico II’’, and INFN, Sezione di Napoli, Mostra
D’Oltremare Pad. 20, I-80125 Naples, Italy
Received 16 June 1999; accepted 5 November 1999
Abstract
We report the results of a new accurate evaluation of light nuclei yields in primordial
nucleosynthesis. All radiative effects, finite nucleon mass, thermal and plasma corrections are
included in the proton to neutron conversion rates. The relic densities of 4 He, D and 7 Li have been
numerically obtained Õia a new updated version of the standard BBN code. In particular the
theoretical uncertainty on 4 He is reduced to the order of 0.1%. q 2000 Elsevier Science B.V. All
rights reserved.
PACS: 98.80.Cq; 95.30.Cq; 11.10.Wx; 13.40.Ks
1. Introduction
Big Bang Nucleosynthesis ŽBBN. still represents one of the key subject of modern
cosmology even if its clear understanding traces back to over 25 years ago w1–3x. The
reason for this relies on the fact that BBN is one of the most powerful tools to study
fundamental interactions, since light nuclei abundances are crucially depending on many
elementary particle properties. As a well-known example, the 4 He abundance is strongly
affected by the number of effectiÕe neutrino degrees of freedom, but others fascinating
phenomena such as neutrino degeneracy or oscillation phenomena can be studied too,
using the universe few seconds after the bang as a laboratory.
In the recent years, the experimental accuracy of the light primordial nuclei abun-
dances, mainly the one of 4 He, underwent a sort of revolution. The qualitative results of
not too many years ago, suggesting that the 4 He mass fraction Y4 was of the order of
0.25, recently turned in measurements with accuracies of the order of one percent. A
similar good improvement has been obtained in both deuterium ŽD. and 7 Li abundances
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 2 5 - 7
422 S. Esposito et al.r Nuclear Physics B 568 (2000) 421–444
Y2 ' DrH and Y7 '7 LirH. In particular for D, measurements in distant Quasars
Absorption line Systems ŽQAS. now represent a reliable estimate of the primordial value
for Y2 , which is only lowered by subsequent stellar processing. Paradoxically, the
refinement of these experimental techniques, due to the uncertainties in the models
describing stellar activity, is at the basis of large discrepancies between different set of
results. Such discrepancies are possibly of systematic origin, or may reveal new aspects
of cosmological evolution of the universe. The observations of Y4 from regression to
zero metallicity in blue compact galaxies in two independent surveys still produce two
incompatible results, a low value w4–6x,
Y4Ž l . s 0.234 " 0.002 " 0.005, Ž 1.1 .

and a significantly higher one w7,8x,
Y4Ž h. s 0.243 " 0.003. Ž 1.2 .

A similar situation occurs in D measurements, where observations in different QAS,
both at red shift larger than 3, give two results at bias for one order of magnitude w9–17x
Y2Ž l . s Ž 3.4 " 0.3 . = 10y5 , Ž 1.3 .

Y2Ž h. s Ž 1.9 " 0.4 . = 10y4 . Ž 1.4 .
For the 7 Li abundance, the almost constant Spite plateau observed in the halo of POP II
stars w18,19x
Y7Ž l . s Ž 1.6 " 0.36 . = 10y1 0 , Ž 1.5 .

is generally considered a reliable estimate of primordial abundance. Nevertheless, the
observation of stars similar to the ones contributing to the Spite plateau, but with no
traces of 7 Li w18,20x, seems to imply the presence of a depletion mechanism. A recent
analysis based on a sample of 41 stars does not find any evidence of a depletion
mechanism or post-BBN creation and yields the primordial abundance w21x
Y7Ž h. s Ž 1.73 " 0.21 . = 10y1 0 . Ž 1.6 .

A brief summary of the complete experimental situation on primordial abundances can
be found in Ref. w22x.
Probably, future measurements or a better understanding of the present data will
clarify the systematics. Nevertheless, what is emerging from the above results is that the
4
He data are reaching a precision of the order of a few percent. This fact requires a
similar effort in the theoretical analysis, in order to reduce the uncertainty on the
predictions at least at the same level of magnitude. In a previous paper w23x we
performed a thorough analysis of all corrections to the protonrneutron conversion rates,
™ e q p,
ne q n y
™ e q n,
q
e q p ™ n q n,
ne qp
n ™ e q n qp,
Ž a. Ž d.
y y
e q n ™ n qp, e q n qp ™ n,
Ž b. e Ž e. e Ž 1.7 .
q y
Ž c. e Ž f. e
S. Esposito et al.r Nuclear Physics B 568 (2000) 421–444 423
which fix at the freeze out temperature ; 1 MeV the neutron to proton density ratio.
The Born rates, obtained in the tree level V y A limit and with infinite nucleon mass,
have been corrected to take into account basically three classes of relevant effects:
Ži. electromagnetic radiative corrections, which largely contribute to the rates of the
fundamental processes, in particular in the low temperature regime, T ( 0.1 MeV;
Žii. finite nucleon mass corrections, which are of the order of TrMN or m erMN ,
with m e , MN the electron and nucleon mass, respectively;
Žiii. plasma effects, proportional to the surrounding plasma temperature, which both
affect the microscopic process rates Ža. – Žf., as well as the neutrino to photon
temperature ratio through e ", g equations of state.
The other main source of theoretical uncertainty comes from the partial knowledge of
nuclear rates relevant for nuclei formation. Their numerical expressions, obtained by a
convolution of the experimental data with a Boltzmann distribution, are affected by
uncertainties of the order of 10% Žsee references quoted in w24x.. More crucially, in
many cases, these fits are known to well describe the data in a temperature interval
which is only partially overlapping the one relevant for BBN, 0.01 MeV ( T ( 10 MeV.
However, both a Monte Carlo analysis to sample the error distribution of the reaction
cross sections w25,26x, and a more recent method based on linear error propagation w27x,
show that, in particular for 4 He mass fraction, the effect is at most as large as the one
due to the uncertainty on neutron lifetime tn , and smaller than 1%. Therefore it is
theoretically justified to look, as in w23x, for all sources of theoretical uncertainty up to
this level of precision. The situation gets worse with D and 7 Li, where the uncertainties
due nuclear reactions can be as large as Ž10–30.% w27x.
This paper represents the natural companion to w23x. We have built a new updated
version of the standard BBN code, which is available since many years w1–3,24x, where
all corrections Ži. – Žiii. have been included. In particular we have also included the
l
modified e ", g equations of state due to electromagnetic mass renormalization. In
Section 2 we review the corrections to n p Born rates, while in Section 4 we discuss
the numerical method we have used to integrate the set of equations relevant for BBN,
which are described in Section 3. The numerical results for light nuclei abundances, as
functions of the final baryon to photon density ratio, h , the number of effective neutrino
degrees of freedom, Nn , and the neutron lifetime, tn , are reported in Section 5, where
they are discussed and compared with the experimental data. We have also performed a
fit of these abundances with a precision of the order of 0.1% in the interesting range for
the parameters h , Nn and tn . Finally in Section 6 we give our conclusions.
2. Corrections to n l p Born rates

As is well known, the key parameter in determining the primordial 4 He mass fraction,
Y4 , is the value of the neutron to proton density ratio at the freeze-out temperature
T ; 1 MeV, since almost all residual neutrons are captured in 4 He nuclei due to its large
binding energy per nucleon. In order to make an accurate theoretical prediction for Y4 it
is necessary, though not sufficient, to have a reliable evaluation of the rates for the
processes Ž1.7.. An effort in this direction has been pursued in the last ten years by
many authors. Recently, the entire set of corrections to the Born rates v B at the level of
1% accuracy have been recalculated in w23,28x, with quite compatible results. In this
section we shortly summarize the main corrections Dvrv B coming from considering
radiative, finite nucleon mass, and thermal effects. This short review is here included for
the sake of completeness and to fix the notation. A detailed discussion of the subject can
be found in our paper w23x.
2.1. The Born rates
Let us consider as an example, the thermal averaged rate per nucleon for the neutron
decay process Že.. In the simple V y A tree level, and in the limit of infinite nucleon
mass, which we will refer to as Born approximation, one has
vB Ž n ™ e q n qp . s
y
e
GF2 Ž C V2 q 3CA2 .
2p 3
`
X X 2
H0 d < p < < p < q02 Q Ž q0 . 1 y Fn Ž q0 .
= 1 y Fe Ž pX0 . , Ž 2.1 .
where GF is the Fermi coupling constant, C V and CA the nucleon vector and axial
coupling. In our notation pX and pX0 are the electron momentum and energy, and q0 the
neutrino energy. The integration limits are imposed by the Q-function, q0 0 0. For
reaction Že. we have q0 s Mn y M p y pX0 ' D y pX0 . The Fermi statistical distributions
for e " and neutrinos in the comoÕing frame, neglecting chemical potentials, are
X y1 y1
Fe Ž pX0 . s w e b < p 0 < q 1 x , Fn Ž q0 . s w e bn < q 0 < q 1 x , Ž 2.2 .
with b s 1rT and bn s 1rTn .1 All other rates for processes Ža. – Žd., Žf. can be simply
obtained from Ž2.1. properly changing the statistical factors and the expression for q0
w23x.
The accuracy of Born approximation can be tested by comparing, for example, the
prediction for neutron lifetime with the experimental value tnex s Ž886.7 " 1.9. s w29x.
Using C V s 0.9751 " 0.0006 and CArC V s 1.2601 " 0.0025 w29x, Eq. Ž2.1. in the
vanishing density limit gives tn , 961 s. Therefore, to recover the experimental value, a
correction of about 8% is expected to come from radiative andror finite nucleon mass
effects. In the same way these corrections are also expected to contribute to all six
processes Ža. – Žf. relevant for BBN. In addition to these, microscopic n p reactions
taking place in the early universe, also feel the presence of the surrounding plasma of g
l
and e " pairs in thermodynamical equilibrium. Emission and absorption of real g or e "
from the thermal bath can be taken into account using the finite temperature field theory
in the real time formalism. This has been considered by several authors w30–45x, and
recently in w23x.
1
The ratio Tn r T is fixed by entropy conservation and using the neutrino decoupling temperature w23x Žsee
Section 2.4.1..
2.2. Electromagnetic radiatiÕe corrections
It is customary to separate the electromagnetic radiative corrections to the Born

amplitudes for processes Ž1.7. in outer and inner terms. The first ones involve the
nucleon as a whole and consist in a multiplicative factor to the modulus squared of
transition amplitude of the form
a
1q g Ž pX0 ,q0 . . Ž 2.3 .
2p
The function g Ž pX0 ,q0 . w46x depends on electron and neutrino energies and describes the
deformation in the electron spectrum. Its effect on a freely decaying neutron is such to
reduce the Born prediction for the lifetime of about 1.6%.
Inner corrections are sensible to nucleon structure details, and thus much more
difficult to handle. They have been estimated in w47x, studying corrections to the quark
weak currents. Translating the quark-based description in the hadronic language, the
inner corrections result in the additional multiplicative factor
a MZ Mp
1q
2p ž 4 ln
Mp
qln
MA
q2C qA g ,
/ Ž 2.4 .
where the first term is the short-distance contribution and A g s y0.34 is a perturbative
QCD correction. The other two terms are related to the axial-induced contributions, with
MA s 400–1600 MeV a low energy cut-off in the short-distance part of the g W box
diagram, and C related to the remaining long distance term.
The global effect of these two kind of corrections, improved by resumming all
leading logarithmic corrections a n ln n Ž MZ . w48x, is Õia the multiplicative factor
a Mp a Ž Mp .
G Ž pX0 ,q0 . s 1 q g Ž pX0 ,q0 . qA g
2p ž ln
MA /
q2C q
2p
S Ž M p , MZ . ,
Ž 2.5 .
where a Ž m . is the QED running coupling constant defined in the MS scheme and
SŽ M p , MZ . a short distance rescaling factor, defined in w23x.
Another effect to be considered, which can be in fact as large as few percents of the
Born rates, is the so-called Coulomb correction, due to the rescattering of the electron in
the field of the proton and leading to the Fermi function for Coulomb scattering
pX0
F Ž pX0 . , 1 qap X .
ž / Ž 2.6 .
<p <
However, this effect is only present when both electron and proton are in either the
initial or final states, namely it only corrects the amplitudes of processes Ža., Žb., Že. and
Žf..
One may wonder if including the effects given by Ž2.5. and Ž2.6. the theoretical
prediction for neutron decay is now compatible with the experimental results. Evaluating
numerically the integral over the phase space one finds tnth s 893.9 s, still at variance
with the experiment. Even adding all known sub-leading effects the agreement does not
really improve w49x. As in Ref. w23x we take the point of view of rescaling all the rates
Ž1.7., after including finite nucleon mass corrections Žsee Section 2.3., by the constant
factor 1 q dt s tnth rtnex s 1.008, which should be regarded as an energy independent

correction to the weak process rates. This renormalization of the coupling guarantees the
correct prediction for tn .
l
In Fig. 1 we report the Born rates v B for n p processes, while in Fig. 2 we plot
the corresponding radiative corrections, Dv Rrv B . Their effect is particularly large, up
to ; 8%, at low temperature.
2.3. Finite nucleon mass corrections
l
There are three additional contributions to the n p rates which appear when one
relaxes the approximation of infinitely massive nucleons. The leading effects are
proportional to m erMN or TrMN , which, in the temperature range relevant for BBN,
can be as large as the radiative corrections considered in Section 2.2. This has been first
pointed out in w50,51x and then also numerically evaluated in w23x. At order 1rMN there
are new couplings appearing in the expression of the weak hadronic current, the larger
one coming from the weak magnetic moments of nucleons
GF f 2 X n X
Jmwm s i
'2 MN u p Ž p . smn Ž p y q . u n Ž q . , Ž 2.7 .
where, from CVC, f 2 s Vu d Ž m p y m n .r2 s 1.81Vu d . Both scalar and pseudoscalar con-
tributions can be shown to be much smaller and negligible for the accuracy we are
interested in. At the same order in inverse nucleon mass power it is also necessary to
include the deformation of the allowed phase space for the relevant scattering and decay
Fig. 1. The total Born rates, v B , for n ™ p Žsolid line. and p ™ n transitions Ždashed line.. This notation is
adopted hereafter.
Fig. 2. The radiative corrections to Born rates, Dv R r v B , for n l p transitions.

processes, due to nucleon recoiling. The sum of these two corrections for n
with respect to the Born values, Dv M rv B , is plotted in Fig. 3.
l p rates
Fig. 3. The finite nucleon mass corrections to Born rates, Dv M r v B , for n l p transitions.
The third effect is due to the initial nucleon thermal distribution. In the infinite
nucleon mass limit, the average of weak rates over nucleon distribution is in fact trivial,
since the nucleon is at rest in any frame. For finite MN , by considering only 1rMN
terms, the effect of the thermal average over the thermal spreading of the nucleon
velocity produces a purely kinetic correction Dv K , whose expression can be reduced to
a one-dimensional integral over electron momentum which can be numerically evalu-
ated. The explicit expression, which we do not report for brevity can be found in
Subsection 4.2 and Appendix C of Ref. w23x. The ratios Dv K rv B for n p are
reported in Fig. 4. Their size is rapidly growing with temperature, since they are
l
proportional to the ratio TrMN .
2.4. Thermal-radiatiÕe corrections
l
The n p rates, calculated as the processes would occur in vacuum, get slight
corrections from the presence of the surrounding plasma of e " pairs and g . These are
the so-called thermal-radiatiÕe effects.
To compute these corrections one may use the standard real time formalism for finite
temperature field theory. According to this scheme, field propagators get additional
contributions proportional to the number density of that particular specie in the
surrounding medium. For g and e " we have
i d Ž k2 . i
i Dgmn Ž k . s y 2
q 2p b <k0<
g mn s y q 2p d Ž k 2 . B Ž k 0 . g mn ,
k e y1 k2
Ž 2.8 .
i
i S e Ž pX . s X y 2p d Ž pX 2 y m2e . Fe Ž pX0 . Ž pu X q m e . . Ž 2.9 .
pu y m e
The entire set of thermal corrections Dv T R , at first order in its typical scale factor, i.e.
a Trm e , have been computed by several authors w30–45x with quite different results. We
have recently performed this lengthy calculation in w23x, to which we refer for all details,
and we have found a good agreement with the original result of Ref. w30x, namely that
they contribute to correct the Born rates only for less than 1%.
2.4.1. RadiatiÕe corrections on neutrino temperature

By assuming a sharp neutrino decoupling at TD s 2.3 MeV w52x, the ratio TnrT can be
evaluated using entropy conservation w23x. This leads to the expression
° I Ž xg . q 2 I Ž x e .
1r3
Tn
T
s~ž¢ I Ž xgD . q 2 I Ž x eD . / T ( TD
, Ž 2.10 .
1 T ) TD
with
` 1r2 y1
IŽ x. s H0 Ž y 2 q 2 yx . Ž 4 y 2 q 8 yx q 3 x 2 . exp Ž x q y . " 1 dy. Ž 2.11 .
According to our notation, x a ' maRrT and x aD ' maRrTD with a s g ,e Žq or y in the
above integrand is for fermions or bosons, respectively.. Note that mgR and m eR are the
Fig. 4. The kinetic corrections to Born rates, Dv K r v B , for n l p transitions.

effective masses that photons and e " acquire in the heat bath due to their interactions
with the background plasma Žsee Appendix A for details..
Fig. 5. The thermal-radiative corrections to Born rates, Dv T R r v B , for n l p transitions.

Fig. 6. The total corrections to Born rates for n l p transitions.

In w23x the neutrino temperature Tn as a function of photon temperature T was
evaluated by using in Ž2.10. the approximated expressions mgR , 0 and m eR , m e q
a T 2rm e . This simplified expression for Tn has been used in all previous sections in
order to obtain the Born rates and their corrections as a function of T only. The
difference between the neutrino temperature evaluated with the correct renormalized
masses ŽA.1., ŽA.2., and the one obtained with the simplified expressions results to be
smaller than 0.01%. The corresponding effect on the rates Ž1.7. due to this small change
in neutrinorphoton temperature ratio, which can be seen as a further sub-leading
thermal-radiative correction to Born rates, can be neglected at the level of precision we
are interested in.
All thermal-radiative corrections to Born rates Dv T Rrv B are reported in Fig. 5. As
evident from this plot, around the freeze-out temperature T ; 1 MeV, Dv T R only
contribute as ; 0.4% to the total rates. Thus they are clearly subdominant. Note that
changing TD in the range Ž2–3. MeV only affects TnrT for less than 0.2%.
2.5. The total rates for n l p reactions

l
In Fig. 6 we report the total corrections Dv to Born rates v B . In order to use the
corrected n p rates v s v B q Dv in the BBN code, it is useful to fit their expres-
sions as a function of the adimensional inverse photon temperature z ' m erT,
vŽ n ™ p . s t1 expŽ yq
n
ex np z.
13
Ý a l zyl ,
ls0
0.01 MeV ( T ( 10 MeV, Ž 2.12 .
°1
v Ž p ™ n . s~ t
13
ex
exp Ž yq p n z . Ý bl zyl 0.1 MeV ( T ( 10 MeV
Ž 2.13 .
¢ n
0
ls1
0.01 MeV ( T - 0.1 MeV
with
a0 s 1 a1 s 0.160615 a 2 s 0.456817 = 10 1
a3 s y0.401109 = 10 2 a 4 s 0.137254 = 10 3 a5 s y0.583644 = 10 2
a6 s 0.655938 = 10 2 a 7 s y0.162185 = 10 2 a8 s 0.371109 = 10 1
a9 s y0.378497 a10 s 0.223840 = 10y1 a11 s 0.723091 = 10y5
a12 s y0.462476 = 10y4 a13 s 0.186287 = 10y5 qn p s 0.340994,
Ž 2.14 .
b 1 s 0.199695 = 10 2 b 2 s y0.671993 = 10 2 b 3 s 0.109230 = 10 3

b4 s y0.295891 = 10 1 b5 s 0.407831 = 10 2 b6 s y0.225830 = 10 1
b 7 s 0.146751 b 8 s y0.185408 = 10y2 b 9 s y0.205210 = 10y3
b 10 s 0.158424 = 10y5 b 11 s 0.369573 = 10y6 b 12 s y0.130731 = 10y9
b 13 s y0.329060 = 10y9 q p n s 2.89858.
Ž 2.15 .
The fit has been obtained requiring that the fitting functions differ by less than 0.1%
™™
from the numerical values, while it is also a good approximation to consider a vanishing
rate v Ž p n. for T ( 0.1 MeV, see Eq. Ž2.13., since it is a rapidly decreasing function
with T 0.
3. The set of equations for BBN
Let us consider Nnuc species of nuclides, whose number densities, X i s n irn B , are
normalized with respect to the total baryon density n B . The different nuclides are
ordered in the following way: n, H, D, 3 H, 3 He, 4 He, 6 Li, 7 Li, 7 Be, . . . Žfor the complete
list see Ref. w24x.. Denoting with RŽ t . the universe scale factor, the BBN set of
equations as functions of R, n B , T, X i , and of the electron chemical potential
f e ' m erT reads 2
Ṙ 8p
R
s ( 3 MP2
w rg q r e q rn q rB x 1r2 , Ž 3.1 .
2
We are using natural units " s cs k B s1.
n˙ B R˙ 24p
nB
s y3
R
sy ( MP2
w rg q r e q rn q rB x 1r2 , Ž 3.2 .
me p 2 nB
L ž T
, fe s / 2 T3
Ý Zj X j ,
j
Ž 3.3 .
Ṙ
T˙s y 3
½ R
rg qpg qr e qp e qQ Ž T y TD . Ž rn qpn . qp B
Er e Ef R˙ Ef e
q
Ef e ž j
e
Ý E X X˙j y 3 R n B
j E nB / q nB Ý Ž D M j q 32 T . X˙j
j
5
y1
d rg Er e Er e Ef e d rn
= ½ dT
q
ET
q
Ef e E T
q Q Ž T y TD .
dT
q nB 3
2 Ý Xj
j
5 , Ž 3.4 .
X lNl X kNk X iN i X jN j
Ni G k l ™ i j y Gi j ™ k l
Ẋ i s Ý
j, k ,l
ž Nl ! Nk ! Ni ! Nj ! / ' Gi Ž X j . . Ž 3.5 .
In the previous relations r and p denote the energy density and the pressure of an
arbitrary particle specie. The function LŽ z, y . in Ž3.3. is defined as
1 ` ey eyy
LŽ z , y . ' Hz dx x 'x 2 y z 2 ž y / , Ž 3.6 .
2 e xqe y e x q eyy
i, j,k,l s 1, . . . , Nnuc , and the ith nuclide, with charge and atomic number Ž Zi , A i ., has
mass Mi and mass excess Ž Mu is the atomic mass unit.
D Mi s Mi yA i Mu . Ž 3.7 .
Moreover, in Ž3.5. we are considering in the sum a reaction between Ni nuclides of type
i and Nj of type j which results in Nl nuclides of type l and Nk of type k, with its
reverse reaction. The energy density and the pressure of baryons take the form
r B s n B Mu q Ý Ž D M j q 32 T . X j , Ž 3.8 .
j
pB s nB T Ý Xj . Ž 3.9 .
j
Eq. Ž3.1. is easily recognized as the Friedmann equation where we have neglected for
simplicity the cosmological constant. Eq. Ž3.2. rules the scaling on n B , whereas Ž3.3.
states the neutrality of primordial plasma. From entropy conservation one gets Ž3.4., and
Ž3.5. are the Boltzmann equations for the Nnuc nuclide number densities. Note that the
presence of the Q-function in Ž3.4. is connected with neutrino decoupling at T s TD .3
3
We have assumed that all neutrinos decouple at the same temperature TD . Actually muon and tau
neutrinos decouple at a slightly larger temperature of 3.5 MeV, but nevertheless our approximation is largely
consistent with the required precision on 4 He yields.
In the set of equations Ž3.1. – Ž3.5. one can safely substitute Eq. Ž3.3. with an
analogous relation, obtained expanding the l.h.s. of Ž3.3. with respect to f e ,
` ex
L Ž z , y . , y H dx x 'x 2 y z 2 2
' y f y1 Ž z . . Ž 3.10 .
z Ž e x q 1.
In this case, Eq. Ž3.3. provides an explicit expression for f e s f e ŽT,n B , X j .,
p2 me nB
fe ,
2
f ž /
T T3
Ý Zj X j .
j
Ž 3.11 .
The consistency of this approach has been tested by means of an iterative check.
The set of Eqs. Ž3.1. – Ž3.5. can be transformed in a set of Nnuc q 1 differential
equations with the dimensionless variable z s m erT as the evolution parameter. For
numerical reasons, it is also better to turn the variable n B into the dimensionless quantity
hˆ ' n BrT 3 , which varies more slowly with z than n B . In terms of these new variables
the BBN set of equations becomes
dhˆ 3hˆ
s 1 y Hˆ z ,h,
ˆ X j G z ,h,
ž ˆ Xj / ž / , Ž 3.12 .
dz z
dX i Gî
ˆ Xj
s G z ,h,
ž / , Ž 3.13 .
dz z
where the function G is
ž ˆ Xj
G z ,h, /
° Erâ 3
hˆ Ý X j
¶
s ~ a
ž
Ý 4 râ y z E z / q 4Q Ž z D y z . rˆn q
2 j
•.
4 3
¢3 Ý Ž râ q pˆ a . q 3 Q Ž z D y z . rˆn q hˆ Ý X j Hˆ q hˆ Ý ž z D Mˆ j q
2 / ˆß
Gj
a j j
Ž 3.14 .
In the previous equations z D s m erTD , a s e,g , and we have considered the dimension-
less Hubble parameter Hˆ s Hrm e ,
1r2
8p m e 1 3
Hˆ z ,h,
ž ˆ Xj s / ( 3 MP z 2
rˆg q rˆ e q rˆn q hˆ zMˆ u q Ý z D Mˆ j q
ž j
ž 2 / /Xj ,
Ž 3.15 .
and the quantities Mˆ u s Murm e , D Mˆ j s D M jrm e , Gˆj s Gjrm e . Energy densities and
pressures have been adimensionalized dividing by T 4 . In Eqs. Ž3.12. and Ž3.13. we have
neglected the terms containing the derivatives of chemical potential. In Appendix A we
report the expressions for pˆ a and râ evaluated taking into account the g and e "
electromagnetic mass renormalization. As already mentioned in Section 2, this effect,

changing the g and e " equations of state, slightly modifies the TnrT ratio too. In order
to speed up the numerical computation a fit of pˆ a and râ as functions of z has been
performed and is also reported in Appendix A.
The initial value for Ž3.12. is provided in terms of the final baryon to photon density
ratio h according to the equation
2 z Ž 3. 11 2 z Ž 3 .
ĥ in s 2
hin s h. Ž 3.16 .
p 4 p2
The condition of Nuclear Statistical Equilibrium ŽNSE., which is satisfied with high
accuracy at the initial temperature Tin s 10 MeV, is then fixing the initial nuclide
relative abundances. From NSE one gets, for an arbitrary ith nuclide with g i internal
degrees of freedom,
A iy1 3
gi 8 Tin Ž A iy1 . Bi
X i Ž Tin . s
2 ž ( /
z Ž 3.
p
A3r2
i ž /
MN
2
h A iy1 X pZ i X nA iyZ i exp ½ 5
Tin
,
Ž 3.17 .
where Bi denotes the binding energy.
4. Numerical method
The most critical part of the BBN code concerns the method of numerical resolution
of the set of differential Eqs. Ž3.12., Ž3.13.. In fact, since at high temperatures nuclear
reactions proceed in both forward and reverse directions with almost equal rapidity, the
r.h.s. of Ž3.13. results to be a small difference of large numbers. When this occurs the
numerical problem is said to be stiff. As a consequence, the step size is limited more
severely by the requirement of stability than by the accuracy of the numerical technique.
In other words, to preserve integration stability it is required to use a shorter step size
than what would be dictated by accuracy only. In order to manage the problem, we use a
NAG routine implementing a method belonging to the class of Backward Differentiation
Formulas ŽBDFs. w53x. This is quite a new approach for BBN codes. In fact the standard
code w1–3,24x uses instead the implicit differentiating method Žbackward Euler scheme.
w53x for writing the r.h.s. of Ž3.13. and a Runge–Kutta solver.
Few comments on the different numerical methods are in order. Let us consider the
differential equation
dy Ž t .
sf Ž t, yŽ t. . . Ž 4.1 .
dt
In the Runge–Kutta methods the solution at t iq1 is completely determined by its value
at t i Žone-step methods., namely the solver has no memory. A different approach is
provided by a wide class of numerical methods referred to as multistep methods like
BDFs. Here, the values of the solution at t k Ž k s i,i y 1, . . . ,i y p ., y Ž t k . ' y k , previ-
ously computed, and the unknown value y Ž t iq1 . ' yiq1 , are interpolated by a polyno-
mial, P Ž t; yiq1 , yi , . . . ., in order to approximate the solution and its derivative. Substi-
tuting in the differential equation,
dP
Ž t iq1 ; yiq1 , yi , . . . . , f Ž t iq1 , yiq1 . , Ž 4.2 .
dt
one obtains a family of BDFs,
Ž t iq1 y t i . f Ž t iq1 , yiq1 . , P Ž t iq1 ; yiq1 , yi , . . . .
y P Ž t i ; yiq1 , yi , . . . . s a 0 yiq1 q a 1 yi q . . . . Ž 4.3 .
Two methods can be used for solving the previous equation in the implicit case, a 0 / 0:
functional iteration and Newton’s method. In the former case some initial guess is taken
for yiq1 and refined by iteration. In the latter case, one linearizes Eq. Ž4.3. by expanding
f around yi . The new point, yiq1 , is then found by inverting a matrix, in a way similar
to the backward Euler scheme. The NAG routine implements both methods and
incorporates an error control test, which drives the step-size adjustment.
The nuclear reaction network used in the code includes all the 88 reactions between
the 26 nuclides present in the standard code w1–3,24x. We used the same nuclear rate
data of the standard code, which are collected and updated in w54x. In order to reduce the
computation time one can also choose a reduced network, made of the 25 reactions
Table 1
The reduced network of nuclear reactions
l
l
Ž1. n p
l
3
Ž2. T He
l
Ž3. pq n Dqg
l
Ž4. nq D T qg
l
Ž5. nq3 He 4
Heqg
l
Ž6. nq6 Li 7
Liqg
l
Ž7. nq3 He Tq p
l
Ž8. nq7 Be 7
Liq p
l
Ž9. nq6 Li 4
HeqT
l
Ž10. nq7 Be 4
Heq4 He
l
3
Ž11. pq D Heqg
l
4
Ž12. pqT Heqg
l
Ž13. pq6 Li 7
Beqg
l
Ž14. pq6 Li 4
Heq3 He
l
Ž15. pq7 Li 4
Heq4 He
l
Ž16. Dq4 He 6
Liqg
l
Ž17. T q4 He 7
Liqg
l
3
Ž18. Heq4 He 7
Beqg
l
3
Ž19. Dq D Heq n
l
Ž20. Dq D Tq p
l
4
Ž21. DqT Heq n
l
Ž22. Dq3 He 4
Heq p
l
3
Ž23. Heq3 He 4
Heq pq p
l
Ž24. Dq7 Li 4
Heq4 Heq n
Ž25. Dq7 Be 4
Heq4 Heq p
Table 2
The predictions on light element abundances obtained with the numerical code for h s 5=10y1 0 and Nn s 3.
In the lower rows the effect of the various corrections is reported
Y2 Y3 Y4 Y7
y4 y4
v Tot 0.3638=10 0.1175=10 0.2446 0.2814=10y9
vB 0.3727=10y4 0.1184=10y4 0.2550 0.2873=10y9
Dv R y2.3% y2.8% y3.8% y1.9%
Dv M 0.2% 0.1% 0.3% 0.2%
Dv K 0.2% 0.1% 0.3% 0.2%
Dv T R y0.6% y0.1% y0.7% y0.4%
dt v T y0.3% y0.1% y0.6% y0.3%
between 9 nuclides listed in Table 1. Using the complete network we have verified that
the reduced one affects the abundances for no more than 0.01 %, while it greatly reduces
the evaluation time.
5. Results on light element abundances
l
The reliable numerical code just discussed can now be used to study the effect of the
different corrections to n p Born rates on light elements abundances. By definition
X3 X5 M6 X6 X8
Y2 s , Y3 s , Y4 s , Y7 s . Ž 5.1 .
X2 X2 Ý Mj X j X2
j
l
In the first two rows of Table 2 are shown the predictions for Y2 , Y3 , Y4 and Y7 at
h s 5 = 10y1 0 , corresponding to the complete n p rates, v Tot , and to the Born
™
approximation, v B .4 As is clear from Table 2, the main effect of the corrections, which
results into the enhancement of the n p conversion rate, is to allow a smaller number
of neutrons to survive till the onset of nucleosynthesis. This ends up in a smaller fraction
of elements which fix neutrons with respect to hydrogen.
The effects on light element yields due to the various corrections with respect to the
Born predictions are also reported in Table 2. For all nuclides the pure radiative
correction Dv R provides the dominant contribution, while the finite nucleon mass
effects, the kinetic and the thermal-radiative ones almost cancel each other. Finally the
last row reports the further contribution due to the additional term required to recover
the experimental neutron lifetime w23x.
If we make use of the results of w27x to quantify the uncertainties coming from
nuclear reaction processes, we observe that only for Y4 the radiative correction affects
the Born result by an amount larger than the theoretical uncertainties, including nuclear
reactions. For 4 He mass fraction in fact, the theoretical uncertainty due to nuclear
4
Note that, according to our notation, with v B we denote the pure Born predictions for n p rates l
without any constant rescaling of coupling to account for the experimental value of neutron lifetime Žsee
Section 2.2..
Fig. 7. The 4 He mass fraction, Y4 , versus h. The three solid lines are, from larger to lower values of Y4 , the
predictions corresponding to Nn s 3 and tnex s888.6 s, 886.7 s, 884.8 s, respectively. Analogously, the dashed
lines correspond to Nn s 4 and the dotted ones to Nn s 2. The dotted and dashed horizontal band are the
experimental values of Refs. w4–6x and w7,8x, respectively, with 1 s interval.
reaction rates is estimated to be of the order of 0.1% and thus comparable with the
uncertainty due to the experimental error on neutron lifetime. For D, 3 He and 7 Li the
uncertainty due to the poor knowledge of nuclear reaction rates is estimated to be of the
l
order of Ž10–30.% w27x, thus completely covering any radiativerthermal correction on
n p rates.
In Fig. 7 the predictions on Y4 are shown versus h for Nn s 2,3,4 and for a 1 s
variation of tnex . The two experimental estimates for the primordial 4 He mass fraction,
Y4Ž l . and Y4Ž h., as horizontal bands, are also reported. Fig. 8 and Fig. 9 show the
predictions for D and 7 Li abundances. Note that, due to the negligible variation of Y2
and Y7 on small tn changes, no splitting of predictions for 1 s variation of tnex is
present.
A fit, up to 1% accuracy, of the relevant observables Y2 , Y3 , Y4 and Y7 as a function
of x s log 10 Ž 10 10h . , Nn and tn has been performed. The following expressions have
been obtained:
4
10 3 P Y2 s Ý a i x i q a5 Ž Nn y 3. exp a6 x q a7 x 2 4 , Ž 5.2 .
is0
4
10 5 P Y3 s Ý a i x i q a5 Ž Nn y 3. exp a6 x 4 , Ž 5.3 .
is0
Fig. 8. The quantity Y2 versus h is reported. The notation used is the same of Fig. 8. Due to the negligible
dependence of Y2 on small variations of tnex no splitting of lines is present. The horizontal bands dashed and
dotted are the experimental values of Refs. w9–17x.
Fig. 9. The quantity Y7 versus h. The notation used is the same of Fig. 8. There is no splitting of lines related
to Dtn , due to the negligible dependence of Y7 on small variations of tnex . The horizontal bands dashed and
dotted are the experimental values of Refs. w18,19x and w21x, respectively.
Table 3
Values of the fit coefficients Ž5.2. – Ž5.5. for light element abundances
Coeff. 10 3PY2 10 5PY3 10PY4 10 9PY7
a0 0.4854 3.325 2.209 0.5419
a1 0.2919 0.1496 0.5548 y0.5981
a2 y0.3516 1.597 y0.6491 y1.914
a3 0.5048 y1.923 0.7661 4.521
a4 y0.4269 1.312 y0.5366 0.1587
a5 0.7772=10y1 0.1782 0.1614 y0.3256
a6 y4.397 y1.705 0.2059=10y2 y4.102
a7 0.5925 – 0.1300 5.072
a8 – – y0.4156=10y4 y1.209
a9 – – 0.7433=10y2 y0.6269
5
10 P Y4 s Ý a i x i q a6 Ž t y te x . q a7 Ž Nn y 3. q a8 x Ž t y te x . q a9 x Ž Nn y 3. ,
is0
Ž 5.4 .
3 4
10 9 P Y7 s Ý
is0
a i x i q a 4 Ž Nn y 3 . q a5 x Ž Nn y 3 . exp ½Ýis1
5
a5qi x i , Ž 5.5 .
where te x s 886.7 s and the values of the fit coefficients are reported in Table 3.
Neutrino decoupling has been shown by many authors w55–61x to be a process which
still takes place when e " pairs annihilate. This implies that neutrinos are in fact slightly
X
Fig. 10. The ratio Ž Y4 yY4 .r Y4 versus log 10 Ž10 10h . for Nn s 3 and tn s885.3 s w22x Žsee Section 5..
Fig. 11. The likelihood distributions for the light element yields Y2 , Y4 , Y7 are shown as functions of Nn and
log 10 Ž10 10h ., normalized to unity in correspondence of the experimental values. From left to right and from
top to bottom the following cases are considered: Ža. high D, low 4 He; Žb. high D, high 4 He; Žc. low D, low
4
He; Žd. low D, high 4 He. The plots for cases Žc. and Žd. are rescaled by a factor 25 and 100 times,
respectively, compared to the one of Ža. and Žb..
reheated during this annihilation process and their final distribution in momentum space
shows an interesting non-equilibrium shape. In Ref. w62x it is estimated that the effect on
Y4 due to the inclusion of this slight neutrino heating is very small, d Y4 ; 1.5 = 10y4 , in
the whole range 10y1 0 ( h ( 10y9 . We have included this constant correction to Y4
prediction.
From the fit reported in Eq. Ž5.4. it is easy to quantify the theoretical error on Y4 .
Since this is basically due to the uncertainty on tn we have
D Y4 Ž a6 q a8 x . Dtn
s ( 0.1%. Ž 5.6 .
Y4 10Y4
In Fig. 10, our prediction for Y4 of Eq. Ž5.4. with Nn s 3 and tn s 885.3 s, is compared
with an analogous fit, Y4X , performed in w22x. The agreement between the two expressions
obtained by independent codes is up to 1% in the relevant range for h.
In Fig. 11 we present the results of a likelihood analysis of the theoretical predictions

obtained with the numerical code for the four combinations of experimental results: Ža.
high D, low 4 He; Žb. high D, high 4 He; Žc. low D, low 4 He; Žd. low D, high 4 He, and
using the low value for 7 Li abundance. In particular, we plot the product of gaussian
distribution for D, 4 He and 7 Li centered around the measured values and with their
corresponding experimental errors,
2 2
y Ž Y2 Ž Nn , x . y Y2ex . y Ž Y4 Ž Nn , x . y Y4ex .
L Ž Nn , x . s exp
½ 2 s 22 5 ½
exp
2
2 s42 5
y Ž Y7 Ž Nn , x . y Y7ex .
=exp
½ 2 s 72 5 . Ž 5.7 .
Notice that all functions have been normalized to unity in the maximum. As is clear
from the plots, the analysis prefers the high value of D Žplots Ža. and Žb... In both cases
the distributions are centered in the range x g 0.2–0.4, but at Nn ; 3 for low 4 He and
Nn ; 3.5 for high 4 He. For low D the compatibility with experimental data is worse.
Note that Žc. and Žd. distributions have been multiplied by a factor of 25 and 100
respectively and centered in the range x g 0.6–0.8, and at Nn ; 2 for low 4 He and
Nn ; 3 for high 4 He. The better agreement at 1 s of the data sets Ža. and Žb. with the
theoretical predictions is basically due to the effect of 7 Li data which corresponds to
values for h compatible with low D data of Žc. and Žd. at 2 s only. It should be
mentioned however that these results only take into account experimental errors, so that
the confidence level regions in the Nn –x plane would be broader by convoluting the
considered distributions with the ones containing the theoretical error.
6. Conclusions
In this paper a detailed study of the effects on light element yields of the radiative,
finite nucleon mass, thermal and plasma corrections to Born rates Ž1.7. has been carried
out. The aim of such an analysis was to reduce the error on, basically, Y4 to less than
1%, which is motivated by the most recent experimental determinations for 4 He
abundance. This accurate analysis has been performed using an update version of the
BBN standard code w1–3,24x. A different numerical approach, based on BDF techniques,
has been implemented to solve the stiff Boltzmann equations for nuclei densities. The
numerical results for 4 He mass fraction almost confirm the computation reported in Ref.
w22x, while the theoretical error, also including the propagation of uncertainties on
nuclear processes, as estimated in w27x, is of the order of 0.1%. Our analysis shows that
the preferred experimental values are high value for D and low one for 4 He, in which
case the distribution is centered at x ; 0.3 and Nn ; 3.
Appendix A. Radiative corrections to e ", g equations of state
In an accurate description of the primordial plasma it is important to consider the

electromagnetic correction to the e " and g equations of state induced by the e " and g
mass renormalization.
As is well known, the photon renormalized mass, up to first-order correction in the

electromagnetic coupling constant a , reads w63x
1r2
2 a ` 'x 2 yz 2
mgR Ž z . , m e
z
( p
Hz dx
1qe x
, Ž A.1 .
and for e " w30–45x
a p2 ` z2 dx
½
m eR Ž z , y . , m e 1 q
pz 2
3
q
ž
Hz 2'x 2
yz2 q
( 2
2 y yz 2
log L
/ 1qe x 5
Ž A.2 .
,
where z ' m erT, y ' EerT and

2
x 2 y 2 y z 2 q 'x 2 y z 2 y 2 y z 2
ž ( /
L Ž x , y, z . s 2
. Ž A.3 .
x2 y2 y ž z y 'x
2 2
yz2 (y 2
yz2 /
Note that in Eqs. ŽA.1. and ŽA.2. one can neglect the contribution of electron chemical
potential, f e ' m erT, due to its small value.
By using ŽA.1. and ŽA.2. in the expressions of rg , pg , r e , p e one gets the latter
quantities as functions of z only. Since the e " and g energy densities and pressures
have to be used in a BBN code, in order to speed up the computation one can fit these
quantities as function of z and use these fits in the evolution equations. The fitted
expressions for the dimensionless electron energy density and pressure, rˆ e s r erT 4 and
pˆ e s prT 4 , in the range z g w0.05,8.52x Ž rˆ e s pˆ e s 0 for z ) 8.52., result to be
rˆ e Ž z . s 1.145 q 0.33981 = 10y1 z y 0.14543 z 2 q 0.25507 = 10y1 z 3
y 0.54168 = 10y3 z 4 y 0.11263 = 10y3 z 5 y 0.29742 = 10y5 z 6

q 0.38331 = 10y6 z 7 q 0.45263 = 10y7 z 8 q 0.19241 = 10y8 z 9
y 0.96597 = 10y1 0 z 10 y 0.19505 = 10y1 0 z 11 y 0.14079 = 10y1 2 z 12 ,
Ž A.4 .
pˆ e Ž z . s Ž 0.3786 q 0.19126 = 10y1 z y 0.63895 = 10y1 z 2 q 0.32085 = 10y1 z 3
y0.48501 = 10y2 z 4 y 0.16611 = 10y3 z 5 q 0.82922 = 10y4 z 6

q0.79884 = 10y5 z 7 y 0.60619 = 10y6 z 8 y 0.19568 = 10y6 z 9
2
y0.10921 = 10y7 z 10 q 0.38564 = 10y8 z 11 . ey0 .13145 z . Ž A.5 .
Moreover, in the considered temperature range, one can show that rˆg s rgrT 4 only
varies between 0.6580 and 0.6573, and pˆ g s pgrT 4 between 0.2193 and 0.2187. Thus,
for simplicity, one can assume rˆg constant and equal to the average value 0.6577, and to
0.2190 for pˆ g .
References
w1x R.V. Wagoner, W.A. Fowler, F. Hoyle, Astrophys. J. 148 Ž1967. 3.

w2x R.V. Wagoner, Astrophys. J. Suppl. 18 Ž1969. 247.
w3x R.V. Wagoner, Astrophys. J. 179 Ž1973. 343.
w4x B.E.J. Pagel, E.A. Simonson, R.J. Terlevich, M. Edmunds, MNRAS 255 Ž1992. 325.
w5x E. Skillman, R.C. Kennicutt, Astrophys. J. 411 Ž1993. 655.
w6x E. Skillman, R.J. Terlevich, R.C. Kennicutt, D.R. Garnett, E. Terlevich, Astrophys. J. 431 Ž1994. 172.
w7x Y.I. Izotov, T.X. Thuan, V.A. Lipovetsky, Astrophys. J. 435 Ž1994. 647.
w8x Y.I. Izotov, T.X. Thuan, V.A. Lipovetsky, Ap. J.S. 108 Ž1997. 1.
w9x R.F. Carswell, M. Rauch, R.J. Weymann, A.J. Cooke, J.K. Webb, MNRAS 268 Ž1994. L1.
w10x A. Songaila, L.L. Cowie, C. Hogan, M. Rugers, Nature 368 Ž1994. 599.
w11x M. Rugers, C.J. Hogan, Astrophys. J. 111 Ž1996. 2135.
w12x R.F. Carswell et al., MNRAS 278 Ž1996. 518.
w13x E.J. Wampler et al., Astron. Astrophys. 316 Ž1996. 33.
w14x J.K. Webb, R.F. Carswell, K.M. Lanzetta, R. Ferlet, M. Lemoine, A. Vidal-Madjar, D.V. Bowen, Nature
388 Ž1997. 250.
w15x D. Tytler et al., astro-phr9810217 Ž1998..
w16x D. Tytler, M. Fan, X.-S. Burles, Nature 381 Ž1996. 207.
w17x S. Burles, D. Tytler, Astrophys. J. 460 Ž1996. 584.
w18x J.A. Thorburn, Astrophys. J. 421 Ž1994. 318.
w19x P. Molaro, F. Primas, P. Bonifacio, Astron. Astrophys. 295 Ž1995. L47.
w20x S.G. Ryan, J.E. Norris, T.C. Beers, Astrophys. J. 506 Ž1998. 892.
w21x P. Bonifacio, P. Molaro, Mon. Not. Roy. Astron. Soc. 285 Ž1997. 847.
w22x S. Sarkar, astro-phr9903183.
w23x S. Esposito, G. Mangano, G. Miele, O. Pisanti, Nucl. Phys. B 540 Ž1999. 3.
w24x L. Kawano, preprint FERMILAB-Pub-88r34-A.L. Kawano, preprint FERMILAB-Pub-92r04-A.
w25x L.M. Krauss, P. Romanelli, Astrophys. J. 358 Ž1990. 47.
w26x P.J. Kernan, L.M. Krauss, Phys. Rev. Lett. 72 Ž1994. 3309.
w27x G. Fiorentini, E. Lisi, S. Sarkar, F.L. Villante, Phys. Rev. D 58 Ž1998. 063506.
w28x R.E. Lopez, M.S. Turner, Phys. Rev. D 59 Ž1999. 103502.
w29x C. Caso et al., Eur. Phys. J. C 3 Ž1998. 1.
w30x D.A. Dicus, E.W. Kolb, A.M. Gleeson, E.C.G. Sudarshan, V.L. Teplitz, M.S. Turner, Phys. Rev. D 26
Ž1982. 2694.
w31x J.L. Cambier, J.R. Primack, M. Sher, Nucl. Phys. B 209 Ž1982. 372.
w32x J.F. Donoghue, B.R. Holstein, R.W. Robinett, Ann. Phys. ŽN.Y.. 164 Ž1985. 23.
w33x J.F. Donoghue, B.R. Holstein, Phys. Rev. D 28 Ž1983. 340.
w34x J.F. Donoghue, B.R. Holstein, Phys. Rev. D 29 Ž1984. 3004.
w35x A.E. Johansson, G. Peresutti, B.S. Skagerstam, Nucl. Phys. B 278 Ž1986. 324.
w36x W. Keil, Phys. Rev. D 40 Ž1989. 1176.
w37x R. Baier, E. Pilon, B. Pire, D. Schiff, Nucl. Phys. B 336 Ž1990. 157.
w38x W. Keil, R. Kobes, Physica A 158 Ž1989. 47.
w39x M. LeBellac, D. Poizat, Z. Phys. C 47 Ž1990. 125.
w40x T. Altherr, P. Aurenche, Phys. Rev. D 40 Ž1989. 4171.
w41x R.L. Kobes, G.W. Semeneff, Nucl. Phys. B 260 Ž1985. 714.
w42x R.L. Kobes, G.W. Semeneff, Nucl. Phys. B 272 Ž1986. 329.
w43x R.F. Sawyer, Phys. Rev. D 53 Ž1996. 4232.
w44x I.A. Chapman, Phys. Rev. D 55 Ž1997. 6287.
w45x S. Esposito, G. Mangano, G. Miele, O. Pisanti, Phys. Rev. D 58 Ž1998. 105023.
w46x A. Sirlin, Phys. Rev. 164 Ž1967. 1767.
w47x W.J. Marciano, A. Sirlin, Phys. Rev. Lett. 56 Ž1986. 22.
w48x W.J. Marciano, A. Sirlin, Phys. Rev. Lett. 46 Ž1981. 163.
w49x D.M. Wilkinson, Nucl. Phys. A 377 Ž1982. 474.
w50x D. Seckel, preprint BA-93-16, hep-phr9305311.
w51x R.E. Lopez, M.S. Turner, G. Gyuk, Phys. Rev. D 56 Ž1997. 3191.
w52x K. Enqvist, K. Kainulainen, V. Semikoz, Nucl. Phys. B 374 Ž1992. 392.

w53x B.P. Flannery, W.H. Press, S.A. Teukolsky, W.T. Vetterling, Numerical Recipes in Fortran ŽCambridge
Univ. Press, Cambridge..
w54x For the expressions of the rates see the Web sites: http:rrwww.phy.ornl.govrastrophysicsr
datardata.html or http:rrpntpm.ulb.ac.berNacrerbarre – database.htm.
w55x M.A. Herrera, S. Hacyan, Ap. J. 336 Ž1989. 539.
w56x N.C. Rana, B. Mitra, Phys. Rev. D 44 Ž1991. 393.
w57x S. Dodelson, M.S. Turner, Phys. Rev. D 46 Ž1992. 3372.
w58x A.D. Dolgov, M. Fukugita, Phys. Rev. D 46 Ž1992. 5378.
w59x N.Y. Guedin, O.Y. Guedin, astro-phr9712199.
w60x S. Hannestad, J. Madsen, Phys. Rev. D 52 Ž1995. 1764.
w61x A.D. Dolgov, S.H. Hansen, D.V. Semikoz, Nucl. Phys. B 503 Ž1997. 426.
w62x B. Fields, S. Dodelson, M.S. Turner, Phys. Rev. D 47 Ž1993. 4309.
w63x K. Ahmed, S.S. Masood, Ann. Phys. 207 Ž1991. 460.
Constrained quantization of open string in background B field

and non-commutative D-brane
Chong-Sun Chu a , Pei-Ming Ho b
a
ˆ CH-2000 Neuchatel,
Institute of Physics, UniÕersity of Neuchatel, ˆ Switzerland
b
Department of Physics, National Taiwan UniÕersity, Taipei 10764, Taiwan, China
Abstract
In a previous paper we provided a consistent quantization of open strings ending on D-branes

with a background B field. In this letter, we show that the same result can also be obtained using
the more traditional method of Dirac’s constrained quantization. We also extend the discussion to
the fermionic sector. q 2000 Elsevier Science B.V. All rights reserved.
1. Introduction
It was shown by Connes, Douglas and Schwarz w1x that for the Matrix model
compactified on a torus, the three-form C field background should be incorporated in
the supersymmetric Yang–Mills theory ŽSYM. by deforming the base space into a
quantum space. Due to the correspondence between the DLCQ M-theory and the
D-brane world-volume field theory w2x, this means that the D-brane field theory lives on
a non-commutative space in the presence of a NS–NS B field background w3–5x. In a
previous paper we quantized an open string ending on a D-brane with a constant NS–NS
B field background, and we found that the space-time coordinates of the open string
endpoint are non-commutative. The non-commutativity precisely agrees with previous
results w1x, but in addition we showed that the compactness of the torus is not necessary
for the non-commutativity, and in general the deformation is parametrized by F s B y F
instead of B.
The quantization of an open string that we did in Ref. w6x was carried out explicitly in
terms of a mode expansion of the generic solution of the equations of motion for the
space-time coordinates X m. We used a heuristic argument in the derivation of commuta-
tion relations among the mode coefficients, and then check that it is a consistent
quantization of the classical theory.
E-mail addresses: cschu@sissa.it ŽC.-S. Chu., pmho@phys.ntu.edu.tw ŽP.-M. Ho..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 8 5 - 9
448 C.-S. Chu, P.-M. Ho r Nuclear Physics B 568 (2000) 447–456
In this paper we carry out the quantization again by following the standard quantiza-
tion procedure of Dirac for systems with constraints. The constraints here are, of course,
the boundary conditions of the open string ending on a D-brane. We find that there is a
single ambiguous expression which needs to be regularized. We obtain our previous
results by choosing a natural regularization. We also find the commutation relations for
the fermionic fields using supersymmetry.
2. Necessity of non-commutativity
We first review briefly the setting. Consider a fundamental string ending on a

D p-brane. This can be in type IIsuperstring, type 0 superstring, or in the bosonic string
theory. The bosonic part of the action takes the same form w7,8x
1
SB s 2
Hd s g a b Gmn Ea X m Eb X n q e a b Bmn Ea X m Eb X n
4pa X S
1
q EESdt A Ž X . E X i, Ž 1.
2pa X i t
where A i , i s 0,1, . . . , p, is the UŽ1. gauge field living on the D p-brane. We use the
convention h ab s diagŽy1,1. and e 01 s 1 as in w6x. Here the string background is
Gmn s hmn , F s constant, H s dB s 0. Ž 2.
Adding the fermionic sector does not affect the result and will be considered later. With
slight modification, the considerations in this paper can also be applied to study open
string ending on a D-brane in type I string theory.
If both ends of a string are attached to the same D p-brane, the last term in Ž1. can be
written as
y1 2 ab
HSd se Fi j Ea X i Eb X j . Ž 3.
4pa X
Furthermore, consider the case B s Ý i,p js0 Bi j dX i dX j, then the action Ž1. can be written
as
1
SB s y dt L s
H d 2s g a bhmn Ea X m Eb X n q e a bFi j Ea X i Eb X j .
H Ž 4.
4pa X
Here
F s B y dA s B y F Ž 5.
is the modified Born–Infeld field strength and x 0a is the location of the D-brane. Indices
are raised and lowered by hi j s Žy,q, . . . ,q ..
One obtains the equations of motion
Ž Et2 y Es2 . X m s 0 Ž 6.
and the boundary conditions at s s 0,p :
Es X i q Et X j Fj i s 0, i , j s 0,1, . . . , p, Ž 7.
X a s x 0a , a s p q 1, . . . , D. Ž 8.
C.-S. Chu, P.-M. Ho r Nuclear Physics B 568 (2000) 447–456 449
The constraint Ž8. is standard. We will be mainly interested in the constraint Ž7.. As
demonstrated in w6x, the BC Ž7. implies that
2pa X P k Ž t ,0 . Fk i s yEs X j Ž t ,0 . M ji , Ž 9.
where P k is the canonical momentum
2pa X P k Ž t , s . s Et X k q Es X j Fj k , Ž 10 .
and Mi j s hi j y Fi kFk j . It follows that
2pa X P k Ž t ,0 . , P j Ž t , s X . Fk i s yEs X k Ž t , s . , P j Ž t , s X . s s0
Mk i , Ž 11 .
and
2pa X P k Ž t ,0 . , X j Ž t , s X . Fk i s yEs X i Ž t , s . , X j Ž t , s X . s s0
. Ž 12 .
These simple relations show that the standard canonical commutation relations for
F s 0,
X i Ž t , s . , Pj Ž t , s X . s i d jid Ž s , s X . , Ž 13 .
Pi Ž t , s . , Pj Ž t , s X . s 0, Ž 14 .
X i Ž t , s . , X j Ž t , s X . s 0, Ž 15 .
are no longer valid when F / 0. They are not compatible with the boundary condition
Ž7. when F / 0. In particular, without doing any further calculations, one can see
already from Ž12. that the string coordinates must necessarily be non-commutative
somewhere along the string. A consistent quantization was therefore proposed in w6x and
it was shown that the canonical commutation relations are modified to
P i Ž t , s . , P j Ž t , s X . s 0, Ž 16 .
X kl
y1
X k Ž t , s . , X l Ž t , s X . s "2p i a Ž M F . , s s s X s 0 or p ,
0,½ otherwise,
Ž 17 .
X i Ž t , s . , P j Ž t , s X . s ih i jd˜ Ž s , s X . , Ž 18 .
where d˜ Ž s , s X . is the delta function on w0,p x with vanishing derivative at the boundary,
e.g. Es d˜ Ž s , s X . s 0 for s s 0,p . See Ref. w6x for its explicit form. Thus we see that the
string becomes non-commutative at the endpoint, i.e. the D-brane becomes non-com-
mutative. It was further shown that for the case F s B, Ž17. agrees with the results
obtained from other considerations w9,10x. The result of our quantization was later
confirmed in w11x using standard string perturbation theory.
Although our quantization in w6x is entirely consistent and produces results in
agreement with other considerations, the argument there was mainly based on intuitive
guess and consistency argument. It may perhaps be more satisfactory to give a
‘‘derivation’’ of that result based on more traditional method. In the next section, we
will show that the result in w6x can also be obtained from the standard constraint
quantization of Dirac. Consequences of the quantization Ž16. – Ž18. in the boundary state
formalism Žclosed string. and D-brane physics are under investigation w12x; as well as
the quantization of charged open string w13–15x and its possible consequences1.
3. Dirac quantization of the bosonic sector
In this section, we compute the Dirac bracket starting from the standard Poisson
brackets
Ž X i Ž s . , Pj Ž s X . . s d jid Ž s , s X . , Ž 19 .
Ž Pi Ž s . , Pj Ž s X . . s 0, Ž 20 .
Ž X i Ž s . , X j Ž s X . . s 0. Ž 21 .
See Refs. w19,20x for a review of the Dirac procedure of constrained quantization. Since
Es X i q Et X j Fj i s 2pa X P j Fj i q Es X j M ji , Ž 22 .
the boundary condition is a constraint in the phase space
F i Ž 0 . s F i Ž p . s 0, i s 0,1, . . . , p, Ž 23 .
where we have introduced the notation
F i Ž s . s 2pa X P j Fj i q Es X j M ji . Ž 24 .
Using the Hamiltonian
1 2 2
HB s
4pa X
Hd s Ž Ž E t X . q Ž Es X . . , Ž 25 .
it is easy to show that one requires also the constraints at s s 0,p 2

Es2 n F i Ž s . s 0, Es2 nq1 P i Ž s . s 0, n s 0,1, . . . , Ž 26 .
and that these are all the second class constraints. We will denote them by f Ž a k n .,
a s 1,2; k s 0,1, . . . , p; n s 0,1, . . . ,
f Ž1 k n . s Es2 n F k , f Ž2 k n . s Es2 nq1 P k . Ž 27 .
These constraints are consistent with the explicit Fourier mode expansion of the fields
X i and P i.
1
See in particular the following papers for some recent developments: Ref. w16x studies the closed string
interaction in the presence of non-commutative D-branes; the papers w17x and w18x study the effects of the
B-field in the boundary state formalism.
2
We thank A. Bilal for a helpful discussion about the issue of equations of motion in determining the
secondary constraints.
One can then compute the Poisson matrix C Ž a k n .Ž b l m . of the constraints. The basic
ones are the C Ž a k 0.Ž b l 0. ones:
Ž F k Ž s . ,F l Ž s X . . s 2pa X Ž F M . k l Es d Ž s , s X . q Es X d Ž s X , s . , Ž 28 .
Ž F k Ž s . , Es P l Ž s X . . s M k l Es Es d Ž s , s X . ,
X X Ž 29 .
Ž Es P k Ž s . , Es P l Ž s X . . s 0
X Ž 30 .
and in general
C Ž a k n .Ž b l m . Ž s , s X . s Ž f Ž a k n . Ž s . , f Ž b l m . Ž s X . . s Es2 n Es2Xm C Ž a k 0.Ž b l 0. Ž s , s X . .
Ž 31 .
A distinct feature of our case is that the constraints Ž27. are imposed only at the
boundary of the open string. As a result, the Dirac bracket should be given by
Ž AŽ s . , B Ž s X . . s Ž AŽ s . , B Ž s X . . y Ý Ž AŽ s . , f Ž a k n . Ž s . .
U XX
s XXs XXX
=CŽ a k n .Ž b l m . Ž s XX , s XXX . Ž f Ž b l m . Ž s XXX . , B Ž s X . . , Ž 32 .

XX XXX
where in place of an integral, we have a sum s , s over the endpoints 0,p . We also
adopt the Einstein summation convention for the indices k,l,n,m unless otherwise stated.
With an obvious labelling of the columns and rows of the matrix, the inverse
CŽ a k n .Ž b l m .Ž s XX , s XXX . is given by
CŽ a k n .Ž b l m . Ž s XX , s XXX .
0 y Ž My1 . k l R n m Ž s XX , s XXX .
s
ž Ž My1 . k l R n m Ž s XX , s XXX . 2pa X Ž F My1 . k l Sn m Ž s XX , s XXX . / Ž 33 .
with R,S satisfying

Ý Es2 nq1 Es2 mq1d Ž s , s XX . R m k Ž s XXs XXX . s d kn dss
XX XXX , Ž 34 .
s XX
Ý Es2 n Es2 m XX Es d Ž s , s XX . q Es XX d Ž s XX , s . R m k Ž s XX , s XXX .

XX
s
s Ý Es2 nq1Es2XX mq 1d Ž s , s XX . Sm k Ž s XX , s XXX . . Ž 35 .

s XX
It follows immediately from the triangular form of CŽ a k n .Ž b l m . that

Ž P iŽ s . , P jŽ s X . . U
s0 Ž 36 .
is not modified. To proceed with the computation of the rest of the Dirac brackets
among X i and P i , one may try to invert the relations Ž34., Ž35. to determine the form
of R and S and then use them in Ž32.. However, the explicit form of R,S obtained this
way are highly singular and generally contain ambiguities. Fortunately, there is a trick to
bypass these steps. We will now show that it is in fact possible to compute the desired
brackets using only the defining relations Ž34. and Ž35.. The detailed form of R,S is not
needed.
We first compute Ž X i Ž s ., X j Ž s X .. , the definition Ž32. gives

U
ij
Ž X iŽ s . , X jŽ s X . . U
s y2pa X Ž My1F .
= Ý Es2XXnd Ž s , s XX . R n m Ž s XX , s XXX . Es2XXXmq1d Ž s X , s XXX .
XX XXX
s s
yEs2XXnq1d Ž s , s XX . Sn m Ž s XX , s XXX . Es2XXXmq1d Ž s X , s XXX .

qEs2XXnq1d Ž s , s XX . R n m Ž s XX , s XXX . Es2XXXmd Ž s X , s XXX . . Ž 37 .
iŽ jŽ X U X
It is easy to see that Ž X s ., X s .. s 0 for s , s not both at the endpoints. Now we
multiply Ž35. for the case n s 0 by Ý k Ýs XXX Es2XXXkq1d Ž s X , s XXX . and integrate over s . We
obtain
Ý Es2XXmd Ž s , s XX . R m k Ž s XX , s XXX . Es2XXXkq1d Ž s X , s XXX . < sss
s0
p
XX XXX
s s
s Ý Es2XXmq1d Ž s , s XX . Sm k Ž s XX , s XXX . Es2XXXkq1d Ž s X , s XXX . < sss p

s0 , Ž 38 .
s XXs XXX
where we have used Hd sEs XX d Ž s XX , s . s Es XX Hd sd Ž s XX , s . s 0 to get rid of the second

term on the left-hand side of Ž35.. Therefore the first and second term in Ž37. cancel
each other in Ž X i Žp . y X i Ž0., X j Ž s X .. and we are left with
U
ij
Ž X i Ž p . y X i Ž 0. , X j Ž s X . . U
s y2pa X Ž My1F . Ý
XX XXX
Es2XXnq1d
s s
= Ž s , s . R n m Ž s , s XXX . Es2XXXmd Ž s X , s XXX . < sss

XX XX
s0
p
ij
s y2pa X Ž My1F . P Ž ds X 0 q ds X p . , Ž 39 .
where we have used in the last step above
Ý Es2XXnq1d Ž s , s XX . R n m Ž s XX , s XXX . Es2XXXmd Ž s X , s XXX . < sss p
s0 s ds X 0 q ds X p . Ž 40 .
s XXs XXX
This can be obtained by taking the n s 0 case of Ž34., multiplying it with

Ý k Ýs XXX Es2XXXk d Ž s X , s XXX . and integrating over s ; and note that
Hs Ý d ss XXX d Ž s X , s XXX . s H Ý dss XXX d Ž s X , s . s Ý ds Xs XXX s ds X 0 q ds Xp . Ž 41 .

XXX s s XXX
s s XXX
Formally, if one exchange the order of integration and the sum in the above, one gets
zero times a delta function. Therefore Ž41. calls for a better justification. One way to
justify it is to use a lattice regularization by replacing the interval w0,p x by a lattice of M
equidistant points with spacing e s prM,
d Ž s ,s X . ™ 1e d ss X , Hd s ™ e Ý .
s
Ž 42 .
Since we know w6x from the D-brane field theory considerations that
U U
Ž X i Ž 0. , X j Ž 0. . s yŽ X i Ž p . , X j Ž p . . , Ž 43 .
it follows from Ž39. that the non-trivial Dirac bracket must be
ij
s y Ž X i Ž p . , X j Ž p . . s y2pa X Ž My1F . .
U U
Ž X i Ž 0. , X j Ž 0. . Ž 44 .
Next, we compute Ž X i Ž s ., P j Ž s X .. . It is straightforward to obtain

U
Ž X iŽ s . , P jŽ s X . . U
s h i j Ž d Ž s ,s X . y QŽ s ,s X . . , Ž 45 .
where
QŽ s ,s X . s Ý Es2XXmq1d Ž s , s XX . R m k Ž s XX , s XXX . Es2XXXkq1d Ž s XXX , s X . . Ž 46 .
s XXs XXX
Using the definition Ž34. with n s 0, it is now easy to show that

Es Q Ž s , s X . s ds 0 P Es d Ž s , s X . s s0 q dsp P Es d Ž s , s X . ss p Ž 47 .
and hence d˜ Ž s , s X . ' d Ž s , s X . y QŽ s , s X . has the desired property of having a vanish-
ing derivative when one of its arguments is at the boundary; and d˜ Ž s , s X . is just
d Ž s , s X . when both arguments are away from the boundary. The Fourier expansion of
d˜ Ž s , s X . can be found in w6x.
Needless to say, some of the steps presented above are a little formal. The consistent
quantization using modes w6x provides a more concrete basis of our calculation here and
can be viewed as a complementary viewpoint of the same quantization. From a practical
point of view, the modes commutation relations are more useful to perturbative string
calculations.
4. Inclusion of fermions
In a generic background, the fermionic part of a RNS open string 3 gets additional
couplings, for example
yi i ab
HB i jc e ra Eb c j . Ž 48 .
4pa X
The situation is much simpler for our case with the flat background Ž2. and the complete
supersymmetric Lagrangian can be easily written down. The equation of motion is not
modified
Ecqi s Ecyi s 0, Ž 49 .
where E s Et q Es , E s Et y Es . The open string world-sheet in the type IIand type 0 string
theories has the supersymmetry
d X m s ec m , dc m s yi r a Ea X me Ž 50 .
with the preserved supersymmetry parametrized by
eqs ley , l s "1. Ž 51 .
3
We use the convention of two-dimensional spinor algebra in w 21 x . In particular,
cq 0 1 0 1
cs
cy ž / ž
, r0s
y1 0
, r1 s
1 0
. / ž /
The supersymmetric boundary condition compatible with Ž7. is 4

j j
cqi Ž h q F . i q lcyi Ž h y F . i s 0, i , j s 0,1, . . . , p, Ž 52 .
cqa q lcya s 0, a s p q 1, . . . ,9, Ž 53.
for s s 0,p .
The constrained quantization on the fermions can be similarly performed as in the
bosonic case above. However, since we know that c i and X i are related by supersym-
metry and we have already derived the commutation relations for the X i ’s, it is perhaps
more interesting not to repeat the Dirac quantization here, but instead to utilize the
power of world-sheet supersymmetry to derive the Dirac bracket for the c i ’s directly.
This approach is expected to be particularly useful in more complicated cases. For
example in the case w22x where the background is not constant but weak enough so that
one can still determine the leading non-trivial commutation relation for the open string;
and in the case of supersymmetric AdS k = S k background w12x relevant for the AdSrCFT
correspondance.
It is easy to see that Ž c m Ž s ., X n Ž s X .. s 0. The reason is that the original Poisson
U
m n X
bracket Ž c Ž s ., X Ž s .. is zero and the constraints never mix the bosonic and
fermionic fields, so this Poisson bracket is not modified by the Dirac procedure.
Applying Ž50., we obtain
U U U
i Ž E X m , X n . q Ž cqm , cqn . y l Ž cqm , cyn . P ey
0sd Ž c m, Xn . s
U
ž P ley / . Ž 54 .
Now we will concentrate on the modified boundary condition Ž52.. The Dirac bracket
relations for the tranverse directions can be obtained simply by setting F s 0 in the
following. Substituting Ž52. into the above, we obtain
U
yi U
Ž cqi , cqj . s
2
Ž E X i, X k . Žh yF . k j,
U
yi U
Ž cyi , cyj . s
2
Ž E X i, X k . Žh q F . k j. Ž 55 .
At s s 0,p , the use of Ž9. allows one to express Es X i and Et X i in terms of P k
Es X i s y2pa X P k Ž F My1 . k i , Et X i s 2pa X P k Ž My1 . k i . Ž 56 .
Thus
ij
ŽE X iŽ s . , X jŽ s X . . s 2pa X Ž My1 Ž h q F . . d˜ Ž s , s X . ,
U
Ž 57 .
ij
ŽE X iŽ s . , X jŽ s X . . s 2pa X Ž My1 Ž h y F . . d˜ Ž s , s X . ,
U
Ž 58 .
and so Ž55. implies
Ž cqi Ž s . , cqj Ž s X . . s Ž cyi Ž s . , cyj Ž s X . . s yipa Xh i jd˜ Ž s , s X . ,

U U
Ž 59 .
4
We thank V. Schomerus for a useful email exchange about this boundary condition. A different boundary
condition was used in w6x which corresponds to a non-supersymmetric D-brane.
with s and s X at the boundary. It is also easy to see that in the interior of the string, the
Dirac bracket is not modified and takes the same form as Ž59.. Therefore the fermion
commutator is not modified Žexcept for the modification of the delta function. by the
presence of F and takes the standard form.
5. Remarks
In the quantization procedure of Dirac, Eq. Ž41. is the only ambiguity. But as
mentioned in Section 2, it is inconsistent to say that Ž41. vanishes. Also note that the
ambiguity resides only in the commutation relations for the endpoint coordinates, the
quantization for the space-time coordinates in the interior of the opens string has no
ambiguity, and their commutation relations are the same as if F s 0. The lattice
regularization we choose there gives the same result we obtained in w6x. It leads to the
conclusion that a D-brane in the B field background has a non-commutative world-
volume in the sense that its world-volume theory is the SYM theory living on a
non-commutative space.
Upon quantization, the Dirac bracket becomes the Žanti.commutator of the operators.
From the point of view of the open string, the modifications to the commutation
relations is ‘‘not very much’’. But as discussed in w6x, since the endpoint of the open
string is living on the D-brane, the D-brane world-volume becomes a non-commutative
one. It is well known that a non-commutative manifold can be described in the dual
language of the algebra of functions living on it, one immediately arrives at the
conclusion that the D-brane world-volume field theory is a non-commutative one, i.e.
one with a modified multiplication. In fact it is easy to check that our modified
commutators Ž44. are equivalent to the following Moyal product w23,24x defined on the
D-brane world-volume w1x,
E E
f Ž j . w g Ž j X . s exp Q i j f Ž j . g Ž j X . < js j X ,
ž Ej Ej X j
i / Ž 60 .
where j i , i s 0, . . . , p are the D-brane world-volume coordinates and

ij
Q i j s "ipa X Ž F My1 . . Ž 61 .
The non-commutative D p-brane world-volume theory takes functionally the same
form as the SYM 9q 1 theory dimensionally reduced to p q 1 dimensions, with the usual
product of fields replaced by the Moyal product Ž60., and with the space-time integral
replaced by a cyclic trace. It has been argued w5x that in the DLCQ limit of e 0 with ™
m 2
X ;e , B ; 1re , Ž 62 .
aX ;e , g s ; e 3r2 , Ž 63 .
where g s is the string coupling constant, the SYM Lagrangian is exact and hence
deserves attention.
When this work is nearly finished, we learned that another group w25x has also
employed Dirac’s procedure for the same problem, but their final result is different from
ours. The reason is presumably that they chose a different regularization.
Acknowledgements
We thank A. Bilal, M. Cederwall, Y.-C. Kao, D. Matalliotakis, R. Russo and V.

Schomerus for helpful discussions. The work of C.S.C. is supported by the Swiss
National Science Foundation. The work of P.M.H. is supported in part by the National
Science Council, Taiwan, R.O.C.
References
w1x A. Connes, M.R. Douglas, A. Schwarz, J. High Energy Phys. 02 Ž1998. 003. hep-thr9711162.
w2x N. Seiberg, Phys. Rev. Lett. 79 Ž1997. 3577. hep-thr9710009.
w3x M.R. Douglas, C. Hull, J. High Energy Phys. 2 Ž1998. 8. hep-thr9711165.
w4x C. Hofman, E. Verlinde, U-duality of Born–Infeld on the non-commutative two-torus, hep-thr9810116.
w5x N. Seiberg, talk given at the Conference New Ideas in Particle Physics and Cosmology, Univ. Penn., May
19-22, 99; transpancies available at http:rrdept.physics.upenn.edurparticle – meetingrseiberg.html.
w6x C.-S. Chu, P.-M. Ho, Nucl. Phys. B 550 Ž1999. 151. hep-thr9812219.
w7x J. Dai, R.G. Leigh, J. Polchinski, Mod. Phys. Lett. A 4 Ž1989. 2073.
w8x R. Leigh, Mod. Phys. Lett. A 4 Ž1989. 2767.
w9x Y.-K.E. Cheung, M. Krogh, Nucl. Phys. B 528 Ž1998. 185.
w10x T. Kawano, K. Okuyama, Phys. Lett. B 433 Ž1998. 29.
w11x V. Schomerus, D-branes and deformation quantization, hep-thr9903205.
w12x C.-S. Chu, P.-M. Ho, work in progress.
w13x A. Abouelsaood, C.G. Callan, C.R. Nappi, S.A. Yost, Nucl. Phys. B 280 Ž1987. 599.
w14x C. Bachas, Phys. Lett. B 296 Ž1992. 77. hep-thr9209032.
w15x C. Bachas, Phys. Lett. B 374 Ž1996. 37. hep-thr9511043.
w16x S. Hyun, Y. Kiem, S. Lee, C.-Y. Lee, Closed string interacting with non-commutative D-branes,
hep-thr9909059.
w17x N. Ishibashi, A relation between commutative and non-commutative descriptions of D-branes, hep-
thr9909176.
w18x K. Okuyama, A path integral representation of the map between commutative and non-commutative
gauge fields, hep-thr9910138.
w19x P.A.M. Dirac, Lectures on Quantum Mechanics, Yeshiva University, 1964.
w20x A. Hanson, T. Regge, C. Teitelboim, Constrained Hamiltonian Systems, RX-748, 1976. Contribution to
the Lincei Interdisciplinary Centre for Mathematical Sciences and their Applications, no. 22.
w21x ˝ S. Theisen, Lectures on String Theory ŽSpringer, Berlin, 1989..
D. Lust,
w22x C.-S. Chu, P.-M. Ho, Y.-C. Kao, World-volume uncertainty relations for D-branes, hep-thr9904133.
w23x J.E. Moyal, Proc. Camb. Phil. Soc. 45 Ž1949. 99.
w24x F. Bayen, M. Flato, C. Fronsdal, A. Lichnerowicz, D. Sternheimer, Ann. Phys. 111 Ž1978. 61.
w25x F. Ardalan, H. Arfaei, M.M. Sheikh-Jabbari, Dirac quantization of open strings and non-commutativity in
branes, hep-thr9906161.
Integration of the SL Ž2,R. rU Ž1. gauged WZNW model with

periodic boundary conditions
a b
¨
Uwe Muller , Gerhard Weigt
a
¨ Physik, Johannes-Gutenberg-UniÕersitat,
Institut fur ¨ Staudinger-Weg 7, D-55099 Mainz, Germany
b
Deutsches Elektronen-Synchrotron DESY Zeuthen, Platanenallee 6, D-15738 Zeuthen, Germany
Received 15 June 1999; accepted 9 November 1999
Abstract
Gauged WZNW models are integrable conformal field theories. We integrate the classical
SLŽ2,R.rUŽ1. theory with periodic boundary conditions, which describes closed strings moving in
a curved target-space geometry. We calculate its Poisson bracket structure by solving an initial
state problem. The results differ from previous field-theoretic calculations due to zero-modes. For
a future exact canonical quantization the physical fields are Žnon-locally. transformed onto
canonical free fields. q 2000 Elsevier Science B.V. All rights reserved.
Keywords: Conformal field theory; Integrability; Black hole
1. Introduction
The SLŽ2,R.rUŽ1. gauged WZNW model has attracted much interest in the past
w1–8x before it was recognized that this non-linear theory is classically integrable w9x.
More generally, we could prove that integrability holds for any gauged WZNW theory
w10x. This was known for nilpotent gauging only, which yields Toda theories w7,11,12x.
So far we have completely integrated the non-linear equations of motion of the classical
SLŽ2,R.rUŽ1. model for a field theoretic case with asymptotic boundary conditions
w10x.
In this paper we solve the classical SLŽ2,R.rUŽ1. theory for periodic boundary
conditions. As a conformal field theory this model describes a closed string moving in
the background of a black hole target-space metric w3x. It is especially interesting for
quantization; quantum mechanical deformations of its metric and a correlated dilaton
w13x were obtained in some perturbative manner w4–6x. However, these calculations were
based on an incomplete effective action w10,14,15x.
Our intention is to provide a different understanding of such quantum mechanical
results. Starting with an entirely classical approach w1,10x, we expect that the exact
classical solution of this theory will facilitate its exact canonical quantization. We
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 2 4 - 5
458 ¨
U. Muller, G. Weigtr Nuclear Physics B 568 (2000) 457–471
calculate, as in the field theoretic case, the Poisson bracket structure of the theory by
solving an initial state problem and look for a canonical transformation of the physical
fields onto canonical free fields. But the results for periodic boundary conditions cannot
simply be inferred from the field theoretic ones because additional zero-modes become
important.
To make this paper self-contained, we mention in Section 2 some of our earlier
results; more details are found in Ref. w10x. First we define the theory, give its general
solution and inspect the symmetry properties. Section 3 solves an initial state problem
which allows us to calculate in Section 4 the Poisson bracket structure of the model. A
free-field realization of these brackets is given in Section 5. The summary provides a
canonical transformation of the physical fields onto the free fields. Some technical
details are found in two appendices.
2. The SL(2,R) r U(1) theory
The exact action of the SLŽ2,R.rUŽ1. gauged WZNW theory written in light-cone
coordinates z s t q s , z s t y s
1 2
S w r ,t x s HM Ž E rE r q tanh r E tE t . dz dz
z z z z Ž 1.
g2
was derived entirely classically and in a gauge-invariant manner w1,10x. M has cylindri-
cal topology where the space-like submanifolds are topologically equivalent to a circle
M s R = S1 , i.e. 0 ( s ( 2p , y` - t - `, Ž 2.
The physical fields r Ž s ,t ., t Ž s ,t ., which represent the position of a closed bosonic
string in the target-space at proper time t , are subject to the boundary conditions
r Ž s q 2p ,t . s r Ž s ,t . , t Ž s q 2p ,t . s t Ž s ,t . q 2p w, w g Z. Ž 3.
The t coordinate is an angular variable given modulo 2p only, and the winding number
w tells us how often the string surrounds the coordinate origin. The string moves in the
curved metric of a Euclidean black hole w3x
ds 2 s dr 2 q tanh2 r dt 2 , Ž 4.
and the dynamics is given by the equations of motion
sinh r
Ez Ez r s Ez t Ez t ,
cosh3 r
1
Ez Ez t s y Ž E z rE z t q E z t E z r . . Ž 5.
sinh r cosh r
These equations are integrable because they have a Lax pair representation
Ez y C, Ez y C s Ez C y Ez C y w C,C x s 0. Ž 6.
C and C take values in the Lie algebra of the group SLŽ2,R. w10x
C s C aT a , C s C aT a Ž a s 1,2,3 . Ž 7.
¨
U. Muller, G. Weigtr Nuclear Physics B 568 (2000) 457–471 459
with
0 1 1 0 0 y1
T1s ž ,/ T2s ž , / T 3 s T 1T 2 s ž /. Ž 8.
y1 0 0 y1 y1 0
One can check that
C1 s y 12 tanh2 r Ez t , C2 s C3 s 0, C1 s 12 tanh2 r Ez t ,
1 1
C2 s y Ez Ž sinh rcos t . , C3 s Ez Ž sinh r sin t . Ž 9.
cosh r cosh r
makes the flatness condition Ž6. equivalent to the equations of motion Ž5.. But unlike
Toda theories w16,17x there is at present no general method to integrate a Lax pair
following from a non-nilpotent gauged WZNW model like Ž6. directly. We found the
general solution of Ž5. in w9x by analysing non-abelian Toda theories w7,8x as
i X
sinh2 r s XX , t s iŽ B y B . q ln Ž 10 .
2 X
with the definitions
BX BX
XsAq Ž 1 q AA . , XsAq Ž 1 q AA . . Ž 11 .
AX AX
A s AŽ z ., B s B Ž z ., A s AŽ z . and B s B Ž z . are complex Žanti-.chiral parameter
functions and AX Ž z ., etc. derivatives. However, as we shall see we must restrict this
solution in order to render r and t real. Straightforward substitution shows that the
solution Ž10. fulfills the equations of motion Ž5.. But it will become obvious from the
investigation of initial-value problems of Section 3 that the solution Ž10. exhausts the
entire solution space Žexcluding singular solutions..
The theory is also characterized by conservation laws. The equations of motion Ž5.
guarantee, in particular, conservation and chirality of the energy–momentum tensor Žwe
shall omit the anti-chiral parts whenever possible.
1
T ' Tz z s
g2
Ž Ž Ez r . 2 q tanh2 r Ž Ez t . 2 . , Tz z s 0, Ž 12 .
and in addition of parafermionic observables w1,9,18x

1
V "s e " i n Ž Ez r " itanh r Ez t . , Ž 13 .
g2
where n is defined by
Ez n s Ž 1 q tanh2 r . Ez t , Ez n s Ž 1 y tanh2 r . Ez t. Ž 14 .
Since the integrability condition of these equations corresponds to one of the equations
of motion Ž5., the general solution Ž10. integrates Eqs. Ž14. to
i
n s t q i Ž B q B . q iln Ž 1 q AA . y ln Ž 1 q XX . q n 0 . Ž 15 .
2
The main purpose of this paper is to calculate the Poisson bracket structure of the
model, assuming canonical Poisson brackets for the physical fields r,t and their
460 ¨
conjugate momenta. Therefore, we have to find A, B, A and B as functions of these

physical variables. In principle, they are given by solving an initial state problem
defined by a second-order differential equation of the Gelfand-Dikii type
yXX y Ž Ez VyrVy . yX y g 2 Ty s 0, Ž 16 .
because its two independent solutions y 1 , y 2 are related to our parameter functions
y1 s e B , y2 s A e B , Ž 17 .
and the coefficients of Ž16. are functions of r, t and their derivatives. This differential
equation simply follows from the conserved quantities Ž12., Ž13. and Ž17. as an identity.
But the functions A, B, A and B are not uniquely determined by this procedure.
Because the solution Ž10. is invariant under the GLŽ2,C. transformations
A ™ T w A x s aAcA qy db ,
B ™ T w B x s B q ln Ž cA q d . ,
A ™T w Ax s
dA y c
,
bA q a
B ™ T w B x s B q ln Ž bA q a . ,
a yb
ž c d /
g GL Ž 2,C . , Ž 18 .
they are only given by the physical fields up to four complex constants. We shall fix this
arbitrariness in Section 3.
The monodromy properties of the functions A, B, A, B are, as well, determined by
GLŽ2,C. transformations
pA Ž z . y q
A Ž z q 2p . s T X A Ž z . s ,
rA Ž z . q s
B Ž z q 2p . s T X B Ž z . s B Ž z . q ln Ž rA Ž z . q s . ,
sA Ž z . y r
A Ž z y 2p . s T X A Ž z . s ,
qA Ž z . q p
B Ž z y 2p . s T X B Ž z . s B Ž z . q ln Ž qA Ž z . q p . ,
p yq
ž r s /
g GL Ž 2,C . . Ž 19 .
We should remark here that the GLŽ2,C. transformations act, indeed, in two different
manners. We also find that n is not periodic modulo 2p
n Ž s q 2p ,t . s n Ž s ,t . q 2p w q iln Ž ps q qr . . Ž 20 .
Therefore, the conserved quantities V " and V " are not periodically defined. We can
describe their periodicity behaviour by the conserved total momentum of the field t
1 2p 2p
Pt s 2 H0 tanh2 r t˙ d s X s H0 p t Ž s X ,t . d s X . Ž 21 .
g
¨
Using Eqs. Ž14., instead of Ž20. we obtain for n and n the periodicity relations in terms
of Pt
sq2 p X
n Ž s q 2p ,t . y n Ž s ,t . s Hs n Ž s X ,t . d s X s 2p w q g 2 Pt ,
sq2 p X
n Ž s q 2p ,t . y n Ž s ,t . s Hs n Ž s X ,t . d s X s 2p w y g 2 Pt . Ž 22 .
So we have
2 2
V " Ž z q 2 p . s e " ig Pt
V" Ž z . , V " Ž z y 2 p . s e " ig Pt
V" Ž z . , Ž 23 .
and we can define, up to a constant normalization, new periodic conserved quantities
2 2
W "' e . ig P t z rŽ2 p .
V" , W "' e " ig P t z rŽ2 p .
V". Ž 24 .
Comparing Ž20. and Ž22., the real-valued momentum Pt becomes
g 2 Pt s iln Ž ps q qr . , Ž 25 .
so that the monodromy transformations Ž19. are restricted to those with unit determinant,
< ps q qr < s 1. Ž 26 .
3. The solution of initial value problems
It is advantageous to use in the following calculations Kruskal coordinates:

u s sinh r e i t , u s sinh r eyi t . Ž 27 .
The general solution Ž10. can then be parameterized most symmetrically by the solution
of the Gelfand–Dikii equations y k Ž z ., y k Ž z .
y 1 yX1 q y 2 yX2 y 1 yX1 q y 2 yX2
us , us . Ž 28 .
y 1 yX2 y yX1 y 2 y 1 yX2 y yX1 y 2
We shall restrict ourselves, furthermore, to regular solutions
y 1 yX2 y yX1 y 2 / 0, y 1 yX2 y yX1 y 2 / 0 ; z , z , Ž 29 .
which guarantee that the Gelfand–Dikii equations are not singular, because their
coefficients, determined by
1 yXX1 yX2 y yX1 yXX2 eyi n 0
Ts , Vys Ž y1 yX2 y yX1 y 2 . , Ž 30 .
g 2 y 1 yX2 y yX1 y 2 g2
are non-singular. The GLŽ2,C. invariance of the general solution Ž28. now takes the
form
ž / ™ž
y2
y1
a
c
yb
d /ž / y2
y1
,
ž / ™ž
y2
y1
d
b
yc
a /ž y2
y1/,
a yb
ž c d /
g GL Ž 2,C . . Ž 31 .
462 ¨
It determines the solutions y k Ž z ., y k Ž z . in terms of the physical fields u, u, just as

before, up to four indeterminate integration constants, provided we have chosen for the
physical fields the initial values at ‘time’ t 0
u Ž s ,t 0 . s u 0 Ž s . , u Ž s ,t 0 . s u 0 Ž s . , u˙ Ž s ,t 0 . s u1 Ž s . ,
u̇ Ž s ,t 0 . s u1 Ž s . . Ž 32 .
But in contrast to that, the chiral and anti-chiral second-order Gelfand–Dikii differential
equations allow together eight integration constants for their four independent solutions.
This puzzle can be solved as follows: differentiating the general solution Ž28., four
first-order differential equations result
Ez u Ez u
yX1 s Ž uy1 y y 2 . , yX2 s Ž uy 2 q y1 . ,
1 q uu 1 q uu
Ez u Ez u
yX1 s Ž uy1 y y 2 . , yX2 s
Ž uy 2 q y1 . . Ž 33 .
1 q uu 1 q uu
The elimination of the anti-chiral functions y k Ž z . yield, again, the Gelfand–Dikii
equations
yXXk y Ž Ez VyrVy . yXk y g 2 Tyk s 0, Ž 34 .
and correspondingly the anti-chiral equations result. In case, we look now for solutions
of these equations which fulfill besides the initial state conditions Ž32. the linear
differential Eqs. Ž33. too, the number of integration constants is reduced from eight to
the four of the GLŽ2,C. invariance group. Fixing this GLŽ2,C. invariance the parameter
functions y k , y k are determined, in principle, from the physical fields u, u uniquely.
Since the coefficients of the Gelfand–Dikii equations Ž34. are periodic functions of
z, apart from y k Ž z . also the functions y k Ž z q 2p . are solutions of these equations. They
are given by Ž19. as linear combinations of the y k Ž z .
y 2 Ž z q 2p . y2 Ž z . p yq
ž y 1 Ž z q 2p . / ž /
sM
y1Ž z .
, Ms ž r s /
g GL Ž 2,C . . Ž 35 .
Under the GLŽ2,C. transformation N, the monodromy transformation M changes

according to
M ™ NMN y1
. Ž 36 .
In case that
2
Ž 1 . Ž trM . / 4det M or Ž 2 . M s a12, Ž 37 .
M can be brought to diagonal form,
X X
ey a 0X eya 0
Ms
0 ž ea /
, Ms
0 ž ea
X
/ Ž 38 .
Žthe non-diagonalizable cases can be obtained by a limiting procedure..
The periodicity conditions Ž35. then simplify to
X X
y 1 Ž z q 2p . s e a y 1 Ž z . , y 2 Ž z q 2p . s ey a y 2 Ž z . ,
X X
y 1 Ž z y 2p . s eya y 1 Ž z . , y 2 Ž z y 2p . s e a y 2 Ž z . . Ž 39 .
¨
These properties already restrict the possible GLŽ2,C. transformations Ž31. to scalings
with only two free parameters a and d
ž /™ž / ž / ™ž /
y2
y1
ay 2
dy 1
,
y2
y1
dy 2
ay 1
. Ž 40 .
This means that we have implicitly fixed two of the four integration constants. Any
function invariant under the scalings Ž40. can now be determined uniquely by Ž39., in
particular, the periodic quotients
yXk Ž z .
hk Ž z . ' , k s 1,2. Ž 41 .
yk Ž z .
They determine the solution of Ž34. by
2p z cQ
ln y 1 Ž z . s 12 H0 h1 Ž zX . h Ž z y zX . dzX q a1 q ,
2p 2
2p z cQ
ln y 2 Ž z . s 12 H0 h 2 Ž zX . h Ž z y zX . dzX q a 2 q Ql y . Ž 42 .
2p 2
Ql, cQ are integration constants defined by the Eqs. ŽA.10. of Appendix A, and hŽ z .
denotes the periodic saw-tooth function
z z
h Ž z . s e 2p Ž z . y s 2 n q 1 y for 2p n - z - 2p Ž n q 1 . , n g Z.
p p
Ž 43 .
Here e 2 p Ž z . is the stair-step function
e 2 p Ž z . s 2 n q 1 for 2p n - z - 2p Ž n q 1 . , n g Z, Ž 44 .
and
2p
ak s H0 hk Ž z . dz Ž 45 .
areX the zero-modes of the fields hk Ž z .. ŽIn passing we mention that Ž39., Ž42. imply
e a s e a 1 , and we define a X ' a 1.. But we have to stress here in particular that this
result does not deliver hk or y k explicitly as functions of u, u. However, it will be
sufficient in order to calculate their Poisson brackets.
4. The Poisson brackets
We calculate Poisson brackets by assuming canonical Poisson brackets of the

physical fields, which are obtained from the action Ž1.. For the Kruskal coordinates we
get the following non-vanishing expressions:
u Ž s ,t . ,u˙ Ž s X ,t . 4 s u Ž s ,t . ,u˙ Ž s X ,t . 4 s 2g 2 Ž 1 q uu . d 2p Ž s y s X . ,
u˙ Ž s . ,u˙ Ž s X . 4 s 2g 2 Ž uu
˙ y uu˙ . d 2 p Ž s y s X . , Ž 46 .
464 ¨
where d 2 p is the periodic d-function defined by

`
d 2p Ž s y s X . ' Ý d Ž s y s X q 2p n . . Ž 47 .
nsy`
This allows us to calculate Poisson brackets of all quantities explicitly expressed in

terms of the physical fields. Here we want to determine the Poisson brackets of the
parameter functions y k Ž z .. We saw in Section 3 that the hk are uniquely defined by the
initial state conditions, but they were not given explicitly as functions of the physical
fields. We shall show that the Poisson brackets of the hk , and the y k can, nevertheless,
be derived.
First, we have to determine the Poisson brackets of the hk Ž z ., and those of the y k
then follow by means of Ž42.. We calculate the variations dhk Ž z . as functions of the
varied physical fields and momenta by varying the Gelfand–Dikii equations
d yXXk y Ž Ez VyrVy . d yXk y g 2 Td y k s d Ž Ez VyrVy . yXk q g 2d Tyk . Ž 48 .
X X
We vary, as well, the boundary conditions Ž39., eliminate da and da , and get
subsidiary conditions for Eqs. Ž48.
yXk Ž z . d y k Ž z q 2p . y y k Ž z . d yXk Ž z q 2p . s yXk Ž z q 2p . d y k Ž z .
y y k Ž z q 2p . d yXk Ž z . . Ž 49 .
They restrict the general solution of Ž48., which is defined by a special solution of Ž48.
and the general solution of Ž34., to
2p
d yk Ž z . s H0 V k Ž z , zX . Ž d Ž E VyrVy . Ž zX . yXk Ž zX . q g 2d T Ž zX . y k Ž zX . . dzX
q d Ck y k Ž z . ,
y 2 Ž zX . y1 Ž z . E Ž z , zX . y e 2 p Ž z y zX .
V 1 Ž z , zX . ' ,
y 1 Ž zX . yX2 Ž zX . y y 2 Ž zX . yX1 Ž zX . 2
y1 Ž zX . y 2 Ž z . E Ž zX , z . q e 2 p Ž z y zX .
V 2 Ž z , zX . ' ,
y 1 Ž zX . yX2 Ž zX . y y 2 Ž zX . yX1 Ž zX . 2
a1 y a2
exp ž e 2 p Ž z y zX . / y 2 Ž z . y1 Ž zX .
X 2
EŽ z , z . ' a1 y a2 . Ž 50 .
sinh y1 Ž z . y 2 Ž zX .
2
The variations d Ck correspond to the undetermined scalings Ž40.. They cannot simply
be set to zero because we are integrating Ž50. over non-periodic functions of zX , and
without the term d Ck y k Ž z . these integrals would depend on a shift of the integration
range.
Since the functions hk Ž z . do not depend on the scalings Ž40., their variations
yXk Ž z . d yXk Ž z . yXk Ž z . d y k Ž z .
dhk Ž z . s d s y 2 Ž 51 .
yk Ž z . yk Ž z . yk Ž z .
¨
show, indeed, the cancellation of d Ck in

2p
dhk Ž z . s H0 v k Ž z , zX . Ž d Ž E VyrVy . Ž zX . hkX Ž zX . q g 2d T Ž zX . hk Ž zX . . dzX ,
h1 Ž z . y h 2 Ž z .
v 1 Ž z , zX . ' 12 E Ž z , zX . ,
h1 Ž zX . y h 2 Ž zX .
h1 Ž z . y h 2 Ž z .
v 2 Ž z , zX . ' y 12 E Ž zX , z . . Ž 52 .
h1 Ž zX . y h 2 Ž zX .
The integrands of Ž52. are periodic functions of the integration variable, and Ž52.
determines the Poisson brackets of hk Ž z .. The non-vanishing ones are given by
X
g2
h1Ž z . ,h 2 Ž z . 4 s Ž h1 Ž z . y h 2 Ž z . . E Ž z , zX . Ž h1 Ž zX . y h 2 Ž zX . .
2
y g 2 Ž h1 Ž z . y h 2 Ž z . . d 2 p Ž z y zX . . Ž 53 .
The Poisson brackets of the functions ln y k result, finally, by means of Ž42.
ln y1Ž z . ,ln y1Ž zX . 4 s 0,
g2 z y zX g2
ln y1Ž z . ,ln y 2 Ž zX . 4 s ž e 2 p Ž z y zX . y / y E Ž z , zX .
2 2p 2
g2 2p
qq H dz E Ž z , zX . ,
8p 0
g2
ln y1Ž z . ,ln y1Ž zX . 4 s y Ž z y zX . ,
4p
g2 2p h2 Ž z .
ln y1Ž z . ,ln y 2 Ž zX . 4 s y H dz E Ž z , zX . ,
8p 0 h1 Ž z .
X
g2 2p h2 Ž z .
ln y 2 Ž z . ,ln y 2 Ž z . 4 s y H dz E Ž z , zX .
8p 0 h1 Ž z .
g2 2p h 2 Ž zX .
q H0 dzX E Ž zX , z . ,
8p h1 Ž zX .
g2 g2 2p
ln y 2 Ž z . ,ln y 2 Ž zX . 4 s y Ž z y zX . y H dzX E Ž zX , z .
4p 8p 0
g2 2p
q H d z E Ž z , zX . . Ž 54 .
8p 0
In distinction to the field theoretic results of Ref. w10x we observe here a structurally
changed non-local realization of the algebra. This is due to zero-modes which arise
additionally in the periodic case. It might be surprising that the algebra treats y 1 and y 2
Žas well as y 1 and y 2 . asymmetrically. A symmetric treatment of the functions y k and
y k is presented in Appendix B. But it turns out that the algebra Ž54. is more appropriate
for transforming the y k , y k onto canonical free fields.
466 ¨
5. The canonical transformation onto periodic free fields
There are several methods to find relations between y k Ž z ., y k Ž z . and chiral,

respectively anti-chiral components f k Ž z ., f k Ž z . of canonical free fields Ž k s 1,2.
c k Ž s ,t . s f k Ž z . q f k Ž z . . Ž 55 .
Sometimes it will be useful to have the mode expansions in mind, e.g.,
qk pk wX i aŽnk .
fk Ž z . s q ž q /
d k ,2 z q Ý eyi n z . Ž 56 .
2 4p 2g '4p n/0 n
wX is an integer ‘winding number’.
Here we simply assume that there is a free-field theory with a corresponding
free-field energy-momentum tensor which is canonically related to our SLŽ2,R. model.
Then the easiest and most straightforward approach identifies the energy–momentum
tensors of both theories
2 2
1 yXX1 yX2 y yX1 yXX2
T Ž z . s Ž Ez f 1 . q Ž Ez f 2 . s . Ž 57 .
g 2 y 1 yX2 y yX1 y 2
Furthermore, we assume that the free fields c 1 , c 2 are local expressions of the
parameter functions y k .
It is appropriate to introduce complex free fields
c s c1 q i c2 , c s c1 y i c2 , Ž 58 .
which factorize the components of the energy–momentum tensor
T Ž z . s Ez cEz c , T Ž z . s Ez cEz c . Ž 59 .
Eq. Ž55. gives a corresponding chiral decomposition of c and c
c Ž s ,t . s f Ž z . q x Ž z . , c Ž s ,t . s x Ž z . q f Ž z . , Ž 60.
with
f Ž z . s f 1Ž z . q i f 2 Ž z . , f Ž z . s f 1Ž z . y i f 2 Ž z . ,
x Ž z . s f 1Ž z . y i f 2 Ž z . , x Ž z . s f 1Ž z . q i f 2 Ž z . . Ž 61 .
The most general solution of this problem depends on several complex constants, and
it is given by w15x
1 a yX1 q b yX2 C
fs
gC ž ln
y 1 yX2 y yX1 y2
qD ,
/ xs
g
ln Ž a y 1 q b y 2 . ,
1 a yX1 q b yX2 C
fs
gC ž ln
y 1 yX2 y yX1 y2
qD ,
/ xs
g
ln Ž a y 1 q b y 2 . . Ž 62 .
™ ™
But the constants can be further restricted. Taking into consideration the invariance of
Ž59. under f e i df , x eyi dx , we can choose C real positive. Of course, the physics
should not depend on the choice of the branch of the logarithm. This implies C s 1.
¨
Furthermore, f 2 is defined modulo 2prg only Ži.e. f 2 takes values on a circle with
radius 1rg .
2p 2p
f2 ' f2 q , f2 ' f2 q . Ž 63 .
g g
Up to winding contributions the free fields c k are assumed periodic
2 p wX
c 1 Ž s q 2p ,t . s c 1 Ž s ,t . , c 2 Ž s q 2p ,t . s c 2 Ž s ,t . q , wX g Z,
g
Ž 64 .
and we obtain
pk p wX
f k Ž z q 2p . y f Ž z . s q d k ,2 ,
2 g
pk p wX
f k Ž z q 2p . y f Ž z . s y d k ,2 , p k g R. Ž 65 .
2 g
But this is consistent with Ž62. and Ž39. only if one of the pairs Ž a , a . and Ž b , b . is
™
Ž0,0.. Choosing b s b s 0, the rescaling Ž40. now allows us to set a s a s 1, and
considering the invariance under c k c k q const., we can without loss of generality fix
D s D s 0. Thus the solution Ž62. simplifies finally to
1 yX1 1
f s ln X X , x s ln y 1 ,
g y1 y 2 y y1 y 2 g
1 yX1 1
fs ln , xs ln y 1 . Ž 66 .
g y 1 yX2 y yX1 y2 g
As expected, from the non-local Poisson bracket relations Ž54. we get for the fields f k ,
f k , indeed, the local free-field Poisson brackets
dk l sys X
fk Ž t q s . ,fl Ž t q s X . 4 s y e 2p Ž s y s X . y
ž , /
4 2p
X
dk l X
sys X
f
k Ž t y s . , f l Ž t y s . 4 4 2p
s e ž
Ž s y s . y
2p
, /
d
f k Ž t q s . , f l Ž t y s X . 4 s y 8pk l Ž s q s X . . Ž 67 .
Solving now Ž66. for y k , y k , their non-local free-field representation result
y 1 Ž z . s exp Ž gx Ž z . . ,
exp Ž gx Ž z . . 2p g p1
y2 Ž z . s y H dzXgx X Ž zX . exp y ž e 2 p Ž z y zX . y 2gf 1 Ž zX . ,
/
2sinh Ž g p r2 . 01 2
y 1 Ž z . s exp Ž gx Ž z . . ,
exp Ž gx Ž z . . 2p g p1
y2 Ž z . s y H d zXgx X Ž zX . exp y ž e 2 p Ž z y zX . y 2gf 1 Ž zX . ,
/
2sinh Ž g p r2 . 01 2
Ž 68 .
468 ¨
where the zero-mode momentum is given by

2p
p1 s H0 c˙ 1 Ž t , s . d s . Ž 69 .
We have checked that the free-field Poisson brackets Ž67. yield, conversely, the
non-local Poisson brackets of the y k Ž z ., y k Ž z . Ž54., and we could show that these
results also follow from the Gelfand–Dikii equations Ž34., in case, their coefficients are
expressed in terms of the free fields and the initial state problem is solved anew.
This proves that these free-field transformations of the physical fields r, t, or u, u are
canonical transformations, and they are one to one.
6. Summary
We have completely integrated the periodic SLŽ2,R.rUŽ1. gauged WZNW theory

and calculated its symplectic structure. This allows us to relate this model canonically to
a free-field theory. The results could be summarized in terms of local Backlund ¨
transformations which are identical to those of the non-periodic case w10x. Instead, we
give here the complete canonical transformation of the fields uŽ s ,t ., uŽ s ,t . onto the
free fields
i
u s e g Ž f q x . Ž 1 q FF . y 14 ey g Ž f q x . q Ž e g Žf y f . F q ey g Ž xy x . F . ,
2
i
u s e g Ž f q x . Ž 1 q FF . y 14 ey g Ž f q x . y Ž e g Ž xy x . F q ey g Žf y f . F . . Ž 70 .
2
This transformation is non-locally defined by
1 2p X g p1
F Ž z. sy Hdz gf X2 Ž zX . exp y ž e Ž z y zX . y 2gf 1 Ž zX . , /
2sinh Ž g p 1r2 . 0 2 2p
1 2p g p1
F Ž z. sy H d zXgf X2 Ž zX . exp y
ž e 2 p Ž z y zX . y 2gf 1 Ž zX ./ .
2sinh Ž g p r2 . 0
1 2
Ž 71 .
As in Liouville theory this structure might require quantum mechanical deformations.
But we might be confronted, as well, with further unusual problems related to the
quantization of the parafermionic structure of the theory, which is classically defined by
non-linear Poisson brackets. Using the freedom of normalization of Ž24. Žhere we take
into consideration the full q zero mode of the free field. for the periodic case the
parafermions fulfill
W " Ž z . ,W " Ž zX . 4 s g 2 W " Ž z . W " Ž zX . h Ž z y zX . ,
1
W " Ž z . ,W. Ž zX . 4 s yg 2 W " Ž z . W. Ž zX . h Ž z y zX . q d 2X p Ž z y zX .
g2
ip 2
q d 2 p Ž z y zX . ,
2pg
p 2 ,W " Ž zX . 4 s .2 ig W " Ž zX . . Ž 72 .
¨
In our opinion this may be a good starting point for a quantization. Therefore, it remains
a challenge to implement the exact canonical quantization of the SLŽ2,R.rUŽ1. model
on the basis of our results.
Acknowledgements
We would very much like to thank C. Ford and G. Jorjadze for reading the
manuscript and for useful discussions.
Appendix A. The definition of the integration constants
Here we explain the relations Ž42. in more detail.

Using chirality, the functions hk Ž z . of Ž41. can be integrated to
2p z
ln y k Ž z . s 12
H hk Ž zX . h Ž z y zX . dzX q a q Dk . Ž A.1 .
0 2p k
hŽ z . and a k are defined by Ž43., Ž45.. D k are the integration constants under discussion.
Let us consider the canonical free field Ž58., Ž66.
1 yX1 Ž z . 1
c Ž s ,t . s ln q ln y 1 Ž z . . Ž A.2 .
g y1Ž z . yX2 Ž z. y yX1 Ž z . y2 Ž z . g
Eqs. Ž41. and Ž33. allow one to replace the functions yXk Ž z ., y k Ž z . and y 1Ž z . by hk Ž z ..
Then, c is completely given in terms of hk Ž z . and the physical fields u, u
1 u Ez u y h 2 Ž 1 q uu .
c Ž s ,t . s
g
ln
ž Ez u Ž h 1 y h 2 .
h1 .
/ Ž A.3 .
Similarly, Eqs. Ž33. yield

Ez u y h 2,1 u
h1,2 s Ez u, Ž A.4 .
u Ez u y h 2,1 Ž 1 q uu .
which shows that we could express c , as well, in terms of hk Ž z . and u, u.
On the other hand, substituting ŽA.1. into ŽA.2. we obtain
2p t
c Ž s ,t . s 12 H h1 Ž t q s X . q h 2 Ž t y s X . h Ž s y s X . d s X q Ž a y a2 .
0 2p 1
q i s m q D1 y D 2 q lŽ z . . Ž A.5 .
Here
lŽ z . ' lnh1 Ž z . y ln Ž h 2 Ž z . y h1 Ž z . . , Ž A.6 .
and m, defined by
a 1 q a 2 s 2p i m, Ž A.7 .
is an integer due to the periodicity of y 1Ž z .ry 2 Ž z . Žcf. Ž39. with the Žanti-.chiral ŽA.1...
D 1 y D 2 is uniquely determined by the constant zero-modes of c and l
D 1 y D 2 s cQ y Ql . Ž A.8 .
470 ¨
Since the two parts of ŽA.2. are the chiral and anti-chiral components of c , and
distributing the zero-mode cQ half and half to these components, by comparison we
obtain
cQ cQ cQ cQ
D1 s q Ql , D1 s q Ql , D2 s y , D2 s y . Ž A.9 .
2 2 2 2
This immediately implies Ž42.. We give, finally, the explicit expressions for the
constants cQ and Ql in terms of the hk
1 2p
cQ s H Ž c Ž s ,t . y cP Ž s ,t . . d s , with
2p 0
s 2p Ec Ž s X ,t . t 2p Ec Ž s X ,t .
c P Ž s ,t . s H ds X q H ds X ,
2p 0 Es X 2p 0 Et
1 2p Pl
Ql s H
2p 0 žŽ. l z y
2p /
z dz , with
2p X 2p h1 Ž z . hX2 Ž z . y h1X Ž z . h 2 Ž z .
Pl s H0 l Ž z . dz s H0 dz , Ž A.10 .
h1 Ž z . Ž h1 Ž z . y h 2 Ž z . .
which also determine their commutation relations.
Appendix B. Symmetric Poisson brackets
are symmetric under the exchange 1 2.

Using the shorthand notation
l
In this appendix we define functions y˜ k Ž z . and y˜ k Ž z . with Poisson brackets which
1 2p t
kQ s H ln y 1 Ž t q s . y ln y 2 Ž t y s . y Ž a1 y a2 . y i s m ds ,
2p 0 2p
1 2p t
kQ s H ln y 1 Ž t y s . y ln y 2 Ž t q s . y Ž a1 y a2 . y i s m ds ,
2p 0 2p
Ž B.1 .
these functions are defined by
kQ cQ kQ cQ
ln y˜ 1 Ž z . s ln y 1 Ž z . q y , ln y˜ 2 Ž z . s ln y 2 Ž z . q y ,
2 2 2 2
kQ cQ kQ cQ
ln y˜ 1 Ž z . s ln y 1 Ž z . q y , ln y˜ 2 Ž z . s ln y 2 Ž z . q y . Ž B.2 .
2 2 2 2
¨
The symmetric Poisson brackets are
X
g2 X
z y zX g2
ln y˜ 1Ž z . ,ln y˜ 2 Ž z . 4 s ž e 2p Ž z y z . y / y E Ž z , zX .
2 2p 2
2 2
g 2p g 2p
qq H dz E Ž z , zX . q H dzX E Ž z , zX .
8p 0 8p 0
g2 2p 2p
y 2 H0 H0 dz dzX Ž E Ž z , zX . y E Ž zX , z . . ,
32p
y˜ 1Ž z . , y˜ 1Ž zX . 4 s y˜ 2 Ž z . , y˜ 2 Ž zX . 4 s y˜ 1Ž z . , y˜ 2 Ž zX . 4 s y˜ 1Ž z . , y˜ 2 Ž zX . 4 s 0,
ln y˜1Ž z . ,ln y˜1Ž zX . 4

g2 g2 2p g2 2p
sy Ž z y zX . q H dzX E Ž z , zX . y H d z E Ž zX , z .
4p 8p 0 8p 0
g2 2p 2p
y 2 H0 H0 dz dzX Ž E Ž z , zX . y E Ž zX , z . . ,
32p
ln y˜ 2 Ž z . ,ln y˜ 2 Ž zX . 4
g2 g2 2p g2 2p
sy Ž z y zX . q H dz E Ž z , zX . y H d zX E Ž zX , z .
4p 8p 0 8p 0
g2 2p 2p
y 2 H0 H0 dz dzX Ž E Ž z , zX . y E Ž zX , z . . . Ž B.3 .
32p
References
w1x K. Bardakci, M. Crescimanno, E. Rabinovici, Nucl. Phys. B 344 Ž1990. 344.

w2x I. Bars, D. Nemeschansky, Nucl. Phys. B 348 Ž1991. 89.
w3x E. Witten, Phys. Rev. D 44 Ž1991. 314.
w4x R. Dijkgraaf, E. Verlinde, H. Verlinde, Nucl. Phys. B 371 Ž1992. 269.
w5x I. Bars, K. Sfetsos, Phys. Rev. D 48 Ž1993. 844.
w6x A.A. Tseytlin, Nucl. Phys. B 399 Ž1993. 601.
w7x J.-L. Gervais, M. Saveliev, Phys. Lett. B 286 Ž1992. 271.
w8x A. Bilal, Nucl. Phys. B 422 Ž1994. 258.
w9x ¨
U. Muller, G. Weigt, Phys. Lett. B 400 Ž1997. 21.
w10x ¨
U. Muller, G. Weigt, Comm. Math. Phys. 205 Ž1999. 421.
w11x J. Balog, L. Feher, P. Forgacs, L. O’Raifeartaigh, A. Wipf, Phys. Lett. B 227 Ž1989. 214.
w12x J. Balog, L. Feher, P. Forgacs, L. O’Raifeartaigh, A. Wipf, Annals Phys. 203 Ž1992. 269.
w13x C.G. Callan, D. Friedan, E.J. Martinec, M.J. Perry, Nucl. Phys. B 262 Ž1985. 593.
w14x T.H. Buscher, Phys. Lett. B 201 Ž1988. 466.
w15x ¨
U. Muller, PhD Thesis at the Humboldt University Berlin, 1998, written in German, unpublished.
w16x A.N. Leznov, M.V. Saveliev, Lett. Math. Phys. 6 Ž1982. 505.
w17x A.N. Leznov, M.V. Saveliev, Comm. Math. Phys. 89 Ž1983. 59.
w18x V.A. Fateev, A.B. Zamolodchikov, Zh. Eksp. Teor. Fiz. 89 Ž1985. 380.
Nuclear Physics B 568 wPMx Ž2000. 475–502
Superconformal hypermultiplets
Bernard de Wit a,b, Bas Kleijn a , Stefan Vandoren c
a
Institute for Theoretical Physics, Utrecht UniÕersity, 3508 TA Utrecht, The Netherlands
b
¨ GraÕitationsphysik, Albert Einstein Institut, Am Muhlenberg
Max Planck Institut fur ¨ 1,
D-14476 Golm, Germany
c
C.N. Yang Institute for Theoretical Physics, SUNY, Stony Brook, Stony Brook, NY 11794-3840, USA
Abstract
We present theories of N s 2 hypermultiplets in four space-time dimensions that are invariant

under rigid or local superconformal symmetries. The target spaces of theories with rigid
¨
superconformal invariance are Ž4 n.-dimensional special hyper-Kahler manifolds. Such manifolds
can be described as cones over tri-Sasakian metrics which itself are SpŽ1. fibrations over
quaternionic manifolds. The latter manifolds appear in the coupling of hypermultiplets to N s 2
supergravity. We employ local sections of an SpŽ n. = SpŽ1. bundle in the formulation of the
Lagrangian and transformation rules, thus allowing for arbitrary coordinatizations of the hyper-
¨
Kahler and quaternionic manifolds. q 2000 Elsevier Science B.V. All rights reserved.
PACS: 04.65.qe; 11.30.Pb

¨
Keywords: N s 2 Conformal supergravity; Hypermultiplets; Hyper-Kahler and quaternionic geometry
1. Introduction
It is well known that in theories with rigid N s 2 supersymmetry the hypermultiplet

action takes the form of a supersymmetric sigma model with scalars that parametrize a
¨
hyper-Kahler manifold w1x. In the case of local supersymmetry the scalar fields
parametrize a quaternionic manifold of negative curvature w2x. In this paper we study
actions for hypermultiplets invariant under rigid or local N s 2 superconformal symme-
tries. This study is both motivated by recent interest in superconformal theories w3x and
E-mail addresses: b.dewit@phys.uu.nl ŽB. de Wit., b.kleijn@phys.uu.nl ŽB. Kleijn.,

vandoren@insti.physics.sunysb.edu ŽS. Vandoren..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 2 6 - 9
476 B. de Wit et al.r Nuclear Physics B 568 [PM] (2000) 475–502
by efforts to find alternative and hopefully more convenient formulations of the

hypermultiplet actions. The N s 2 superconformal algebra in four dimensions contains
the bosonic subalgebra associated with SOŽ4,2. = SUŽ2. = UŽ1., together with 8 real
supersymmetry and 8 real ‘special’ supersymmetry transformations, called Q- and
S-supersymmetry, respectively. Requiring that the action is invariant under rigid super-
conformal transformations leads to extra constraints on the target-space geometry w4x.
¨
For instance, these manifolds admit a so-called hyper-Kahler potential whose derivative
defines a conformal homothetic Killing vector and the three complex structures rotate
under the action of the SUŽ2. group, which must be contained as a factor in the isometry
group of the manifold. Spaces that satisfy these constraints will be called special
¨
hyper-Kahler manifolds1. By using the superconformal multiplet calculus w5,6x one can
then obtain corresponding quaternionic sigma models coupled to N s 2 supergravity.
Because of the gauge degrees of freedom associated with the dilatations and the SUŽ2.
¨
transformations, a Ž4 n.-dimensional special hyper-Kahler manifold leads to a Ž4 n y 4.-
dimensional quaternionic manifold. At the time this construction was applied to only flat
¨
hyper-Kahler ¨
spaces or hyper-Kahler quotients thereof. The coupling to supergravity
¨
then leads to a quaternionic projective space and its quaternionic hyper-Kahler quotients
w6x. But it has been known for some time that there exist quaternionic spaces that can
couple to supergravity which are not in this class but can be described in the context of
the formalism of w2x. Some of them have also been obtained explicitly in the context of
harmonic superspace w7x. Therefore it is imperative to apply the superconformal ap-
¨
proach to more general special hyper-Kahler sigma models. This application is the main
topic of our paper, where, in order to avoid introducing an infinite number of fields, we
will no longer insist on off-shell supersymmetry for the hypermultiplets.
Already quite some time ago the very same option was discussed by Galicki w8x.
Rather than starting from the superconformally invariant hypermultiplets, he described
the geometry of these more general hyper-Kahler ¨ spaces using a result of Swann w9x,
who has proven that any quaternionic manifold has a corresponding special hyper-Kahler¨
manifold which admits a quaternionically extended homothety and which has three
complex structures that rotate under an isometric SUŽ2. action. And indeed, the
¨
hyper-Kahler manifolds that he discusses have many properties in common with the
hypermultiplet actions discussed in w4x. Moreover it is known that a special hyper-Kahler
¨
manifold is a cone over a so-called tri-Sasakian manifold, so that there exists a beautiful
¨
relation between quaternionic manifolds, tri-Sasakian manifolds and special hyper-Kahler
manifolds Žfor a recent review, see Ref. w10x.. The tri-Sasakian manifolds have also
appeared recently in the context of supergravity compactifications and the ADSrCFT
correspondence w11x.
In this paper we follow the original superconformal approach and start with the
¨
Ž4 n.-dimensional special hyper-Kahler manifolds as they were formulated in w4x. We
establish that these spaces are indeed cones over Ž4 n y 1.-dimensional tri-Sasakian
spaces Žthis feature was also discussed in w12x.. The special hyper-Kahler
¨ manifolds have
1
¨
Note that hyper-Kahler manifolds that are in the image of the c-map are sometimes called special, because
¨
of the underlying special geometry features. We stress that the usage of the term special hyper-Kahler in this
paper has no relationship to special geometry.
B. de Wit et al.r Nuclear Physics B 568 [PM] (2000) 475–502 477
only a restricted holonomy group contained in SpŽ n y 1.; locally they are a product of a
flat four-dimensional space and a Ž4 n y 4.-dimensional quaternionic space. After gaug-
ing away the degrees of freedom associated with the dilatations and the SUŽ2.
transformations, the quaternionic space remains when coupling to supergravity. We
present the full Lagrangian and transformation rules for the supersymmetric non-linear
sigma models based on special hyper-Kahler ¨ spaces, including the option of gauged
isometries. Furthermore we construct local SpŽ n. = SpŽ1. sections of the so-called
associated quaternionic bundle which is known to exist for any special hyper-Kahler ¨
manifold w9x. It turns out that the use of these sections greatly simplifies the formulation
of the transformation rules and the Lagrangian. In this way our general results remain
closely in line with the results of Ref. w6x; the formulae are identical up to modifications
by connections and covariant tensors. When the sections are trivial, so that the
connections can be put to zero and the tensors become constant, they can be identified
with the hypermultiplet scalar fields and one directly recovers the results of Ref. w6x.
Guided by supersymmetry we thus make contact with the mathematical results quoted
above and we construct the general action and transformation rules in a new form.
The last topic is to couple these supersymmetric non-linear sigma models to
supergravity, using the conformal multiplet calculus. In addition to presenting the
corresponding field theory, we exhibit how the quaternionic manifold emerges in the
coupling. This manifold can now be encoded in terms of SpŽ n. = SpŽ1. sections that are
projective with respect to quaternionic multiplication.
Our results could facilitate the study of type-II string compactifications on Calabi–Yau
three-folds. These lead to four-dimensional models with both vector multiplets and
hypermultiplets. While the moduli space of the vector multiplet scalars is described in
¨
terms of a special Kahler geometry and is well understood, much less is known about
the full quaternionic hypermultiplet moduli space. It is known that at string tree level the
quaternionic manifolds are obtained from a special Kahler ¨ manifold via the c-map w13x.
One would like to understand what the corrections are to the classical hypermultiplet
moduli space coming from both string perturbation theory and non-perturbative effects
w14–17x. With rigid conformal symmetry, the results of this paper could also be helpful
in the description of cone branes w11x.
This paper is organized as follows. In Section 2 we briefly summarize some essential
features of hypermultiplet Lagrangians with gauged target-space isometries. For hyper-
multiplets there exists no unconstrained off-shell formulation in terms of a finite number
of degrees of freedom, hence the supersymmetry algebra will only be realized up to the
field equations of the hypermultiplet fermions. This is in contrast with the vector
multiplets, introduced to gauge the isometries, and the superconformal theory itself, for
which off-shell formulations exist. As a result of the latter, the algebra of gauged
isometries and of the superconformal transformations, including certain field-dependent
structure constants, is completely fixed and not affected by the presence of hypermulti-
plets. Section 3 deals with rigidly superconformal hypermultiplets, where we find the
¨
constraints on the hyper-Kahler manifold imposed by superconformal invariance. Sub-
section 3.1 defines the superconformal transformation rules, the second one deals with
¨
the hyper-Kahler potential and the construction of local SpŽ n. = SpŽ1. sections, and the
third one gives the Lagrangian and the transformation rules. The geometry of special
¨
hyper-Kahler manifolds is explained in Section 4. We first discuss the cone structure of
¨
these hyper-Kahler manifolds which lead to a tri-Sasakian space. The latter is indeed an
SpŽ1. fibration over a smaller space, which we prove to be quaternionic. This quater-
nionic space couples to supergravity, as we then show in Section 5. Here we present the
action for the hypermultiplets associated with a special hyper-Kahler¨ target space
coupled to conformal supergravity and exhibit how the target-space metric becomes
quaternionic.
2. Preliminaries
In this section we summarize hypermultiplet Lagrangians in flat space-time. As is

well known, these constitute N s 2 supersymmetric non-linear sigma models with a
¨
hyper-Kahler target space w2x. The holonomy group is contained in SpŽ n. and it is this
group that is relevant for the hypermultiplet fermions. In Subsection 2.1 we discuss the
supersymmetry transformations, the Lagrangian and the target-space geometry. In
Subsection 2.2 we present possible extensions related to gauged target-space isometries,
which will involve couplings to vector multiplets associated with the gauge algebra.
2.1. Hypermultiplet non-linear sigma models
We will base ourselves on the formulation of hypermultiplet Lagrangians of w18x.

With respect to the results of w2x this formulation differs in that it incorporates both a
¨
metric g A B for the hyper-Kahler target space and a metric Ga b for the fermions. Here
we assume that the n hypermultiplets are described by 4 n real scalars f A , 2 n
positive-chirality spinors z a and 2 n negative-chirality spinors z a . Hence target-space
indices A, B, . . . take values 1,2, . . . ,4 n, and the indices a , b , . . . and a , b , . . . run from
1 to 2 n. The chiral and antichiral spinors are related by complex conjugation Žso that we
have 2 n Majorana spinors. under which indices are converted according to a a ,
while SUŽ2. indices i, j, . . . s 1,2 are raised and lowered. An explicit fermionic metric
l
Gab can be avoided as it can always be converted to a constant diagonal matrix by a
similarity transformation. But retaining a fermionic metric is, for example, important in
obtaining transparent transformation rules under symplectic transformations induced by
the so-called c-map from the electric-magnetic duality transformations on a correspond-
ing theory of vector multiplets. In formulations based on N s 1 superfields Žsuch as in
w19x. one naturally has a fermionic metric but of a special form.
The Lagrangian and transformation rules are subject to a number of equivalence
™™
transformations, two of which are associated with the target space. One set consists of
the target-space diffeomorphisms f f X Ž f .. The other refers to reparametrizations of
the fermion ‘frame’ of the form z a S a b Ž f . z b, and corresponding redefinitions of
™
other quantities carrying indices a or a . For example, the fermionic metric transforms
as Gab w Sy1 xg a w Sy1 x d b Ggd . There are connections, GA a b , associated with these
fermionic redefinitions, which appear in the Lagrangian and supersymmetry transforma-
tion rules. Finally, there are chiral SUŽ2. ( SpŽ1. redefinitions of the supercharges,
which in the rigidly supersymmetric case must be constant and are therefore trivial. In
the locally supersymmetric case this will be different and in the latter part of this paper
we will have to deal with local SUŽ2..
The supersymmetry transformations are parametrized in terms of certain f-dependent

quantities g A and VA according to
d Q f A s 2 g i Aa e zi a q gaA i e i z a ,
ž /
d Q z a s VAai Euf Ae i y d Q f A GA a
b zb ,
d Q z a s VAi a Euf Ae i y d Q f A GA a
b zb . Ž 2.1 .
A
In principle g and VA each denote Ž4 n. = Ž4 n. complex quantities, but as we shall see
below, these quantities are related and satisfy a pseudoreality condition. As it turns out
they will play the role of the quaternionic Žinverse. vielbeine of the target space.
Observe that the supersymmetry variations are consistent with a UŽ1. chiral invariance
under which the scalars remain invariant, while the fermion fields and the supersymme-
try transformation parameters transform nontrivially. This group will be denoted by
UŽ1. R to indicate that it is a subgroup of the automorphism group of the supersymmetry
algebra. In Section 3 we will see that this UŽ1. will correspond to one of the conformal
gauge transformations. However, for generic g A and VA , the SUŽ2. R ( SpŽ1. part of the
automorphism group cannot be realized consistently on the fields. This would require
the presence of an SUŽ2. isometry in the target space. In the above, we merely used that
z a and z a are related by complex conjugation.
The Lagrangian takes the following form:
1 1
L s y g A B Em f AE mf B y Ga b z a Du z b q z b Du z a y Wa bgd z agm z b z gg mz d ,
ž /
2 4
Ž 2.2 .
where we employed the covariant derivatives
Dm z a s Em z a q Em f A GA a
b zb , Dm z a s Em z a q Em f A GA a
b zb . Ž 2.3 .
Besides the Riemann curvature R A B C D we will be dealing with another curvature
R A B a b associated with the connections GA a b , which takes its values in spŽ n. (
uspŽ2 n;C.. The tensor W is defined by
1
Wabgd s R A B e g g i Aa gbi B Ged s R A B C D g i Aa gbi B g jCg gd j D , Ž 2.4 .
2
and will be discussed shortly in more detail.
The target-space metric g A B , the tensors g A , VA and the fermionic hermitian metric
Gab Ži.e. satisfying Ž Ga b . ) s Gba . are all covariantly constant with respect to the
Christoffel connection and the connections GA a b . Furthermore we note the following
relations:
g i Aa VBj a q gaA j VBai s d i j d BA ,
g A B g iBa s Ga b VAbi , VAi a g j Ab s d ji d a b . Ž 2.5 .

From them one derives a number of useful relations, such as
gAja VBai s g B i a VAj a s yg Bj a ViaA q d i j g A B . Ž 2.6 .

The following three bilinears define antisymmetric covariantly constant target-space

tensors:
JAi jB s gA k a ´ kŽ i VBj. a , Ž 2.7 .

¨
that span the complex structures of the hyper-Kahler target space. They satisfy
) 1
Ž Ji j . A B ' Ž JAi jB . s ´ i k ´ jl JAk Bl , J Ai jC JCk Bl s
2
´ iŽ k´ l . j g A B q ´ Ž iŽ k JAl .Bj. .
Ž 2.8 .
In addition we note the following useful identities:
1
gA i a VBj a s ´ i k JAk Bj q g A B dij , JAi jB g aBk s yd kŽ i´ j.l
gA l a . Ž 2.9 .
2
We also note the existence of covariantly constant antisymmetric tensors,
1 1
Vab s ´ i j g A B g i Aa g jBb , V a b s ´ i j g A B VAi a VBj b , Ž 2.10 .
2 2
satisfying Vag V gb s yda b . Their complex conjugates satisfy
)
Vab ' Ž Va b . s G ga V gd G db . Ž 2.11 .
The tensor V can be used to define a reality condition on V and g ,
´ i j Va b VAj b s g A B g iBa s Ga b VAbi . Ž 2.12 .

This equation leads to
g A B VAai VBbj s ´ i j V a b , g A B g i Aa g jBb s ´ i j Va b . Ž 2.13 .

Another convenient identity is given by
1
VAi a Va b VBj b s ´ i j g A B y JAi jB . Ž 2.14 .
2
The existence of the covariantly constant tensors implies a variety of integrability
conditions which have a number of consequences for the various curvature tensors
w2,18x. First of all the covariant constancy of g A implies
R A B C D gaC i gbD j s y´ i j Vag R A B g b . Ž 2.15 .

Observe that the right-hand side is manifestly antisymmetric in w ij x and symmetric in
Ž ab .. This implies that the Riemann tensor can be written with tangent-space indices
according to
1
R A B C D gaA i gbB j ggC k gdD l s ´ i j ´ k l Wa bgd , Ž 2.16 .
2
where, as a result of the cyclicity property of the Riemann tensor, Wabgd is symmetric in
all four indices. This tensor is linearly related to the tensor Ž2.4. upon multiplication
with the tensors G and V . In terms of Wabgd the curvatures read
1
RA BCD s ´ i j ´ k l VAai VBbj VCgk VDd l Wa bgd ,
2
1
Vae R A B e b s y ´ i j VAgi VBdj Wa bgd . Ž 2.17 .
2
The above results are all derived from the requirement of supersymmetry. To
characterize the geometry of the target space, one could start from the non-singular VAi a
and a non-singular skew-symmetric tensor Vab that is covariantly constant with respect
to a symplectic connection GA a b . Subsequently one notes that ´ i j Va b VAj b and the
inverse of VA , denoted by g iBa , are linearly related by a symmetric matrix g A B .
Requiring that this matrix is real we can identify it with the target-space metric while the
ensuing reality constraint on the VA enables their identification as the corresponding
quaternionic vielbeine. This information is sufficient for deriving all the algebraic
identities listed above. The vielbeine and the symplectic connection then allow the
definition of an affine target-space connection, with respect to which the vielbeine are
covariantly constant thus leading to a generalized vielbein postulate. All of above results
then follow upon assuming that the target space has no torsion so that the affine
connection and the Christoffel connection coincide.
2.2. Gauged target-space isometries
The equivalence transformations of the fermions and the target-space diffeomor-

phisms do not constitute invariances of the theory. This is only the case when the metric
g A B and the SpŽ n. = SpŽ1. one-form Via Žand thus the related geometric quantities. are
left invariant under Ža subset of. them. Therefore these are related to isometries of the
¨
hyper-Kahler space. We can then elevate such invariances to a group of local Ži.e.
space-time-dependent. transformations, by introducing the required gauge fields in the
form of vector multiplets. Such gauged isometries have been studied earlier in the
literature w7,19–22x and the purpose of our discussion here is to incorporate them into
the formulation used in this paper.
We consider scalar fields transforming under a certain isometry Žsub.group G
characterized by a number of Killing vectors k IA Ž f ., with parameters u I. Hence under
infinitesimal transformations,
dG f A s g u I k IA Ž f . , Ž 2.18 .
where g is the coupling constant and the k IA Ž f . satisfy the Killing equation,
DA k IB q DB k I A s 0 . Ž 2.19 .
The isometries constitute an algebra with structure constants f I J K ,
k IBE B k JA y k JBE B k IA s yf I J K
k KA . Ž 2.20 .
Our definitions are such that the gauge fields that are needed once the u I become
space-time dependent, transform according to dG WmI s Em u I y gf JK I WmJ u K . The Killing
equation generally implies the following property:
DA DB k IC s R B C A E k IE . Ž 2.21 .
Quantities that carry SpŽ n. indices, such as VAai ,
are only required to be invariant
under isometries up to fermionic equivalence transformations. Thus yg Ž k IB E B VAai q
EA k IB VBai . must be cancelled by a suitable infinitesimal rotation on the index a . Here we
assume that the effect of the diffeomorphism is entirely compensated by a rotation that
affects the indices a . In principle, one can also allow for a compensating SpŽ1.
transformation acting on the indices i, j, . . . . However, the latter transformations must be
constant, so they will generically not appear here. This is equivalent to requiring that the
isometry group will commute with supersymmetry.
Let us parametrize the compensating transformation acting on the SpŽ n. indices by
dG z a s g w t I y k IA GA x a b z b , where the Ž f-dependent. matrices t I Ž f . remain to be
determined,
a
yk IB E B VAai y EA k IB VBai q Ž t I y k IB GB . b VAbi s 0 . Ž 2.22 .
Obviously similar equations apply to the other geometric quantities, but as those are not
independent we do not need to consider them. Using the covariant constancy of VA , we
derive from Ž2.22.,
a
Ž tI . b VAbi s DA k IB VBai , Ž 2.23 .
so that
a 1
Ž tI . bs VAai gbB i DB k IA . Ž 2.24 .
2
Target-space scalars will satisfy algebraic identities, such as
g g g
Ž tI . a Ggb q Ž t I . b Gag s Ž t I . wa Vb xg s 0 . Ž 2.25 .
This establishes that the field-dependent matrices t I take values in spŽ n.. From Ž2.19.
and Ž2.21., it easily follows that
a
DA t I b skI
B
RA B a b , Ž 2.26 .
for any infinitesimal isometry. From the group property of the isometries it follows that
the matrices t I satisfy the commutation relation
a a
w tI , t J x b s fI J
K
Ž tK . b qkI
A
k JB R A B a b , Ž 2.27 .
which takes values in spŽ n.. The apparent lack of closure represented by the presence of
the curvature term is related to the fact that the coordinates f A on which the matrices
depend, transform under the action of the group. One can show that this result is
consistent with the Jacobi identity.
Furthermore we derive from Ž2.22. that the complex structures JAi jB are invariant
under the isometries,
k ICEC JAi jB y 2 E w A k IC JBi jxC s 0 . Ž 2.28 .
This means that the isometries are tri-holomorphic. From Ž2.28. one shows that
EAŽ JBi Cj k IC . y E B Ž JAC
ij
k IC . s 0, so that, locally, one can associate three Killing potentials
Žor moment maps. PIi j to every Killing vector, according to
EA PIi j s JAi jB k IB . Ž 2.29 .
Observe that this condition determines the moment maps up to a constant. Up to
constants one can also derive the equivariance condition,
JAi jB k IA k JB s yf I J K
PKi j , Ž 2.30 .
which implies that the moment maps transform covariantly under the isometries,
dG PIi j s u J k JA EA PIi j s yf JI K
PKi j u J . Ž 2.31 .
Summarizing, the invariance group of the isometries acts as follows:
a
dG f s g u I k IA , dG z a s g Ž u I t I . b z b y dG f AGA a
b zb . Ž 2.32 .
When the parameters of these isometries become space-time dependent we introduce
corresponding gauge fields and fully covariant derivatives,
Dm f A s Em f A y gWmI k IA , Dm z a s Em z a q Em f A GAa b z b y gWm a
bz
b
,
Ž 2.33 .
where Wm a b s WmI Ž t I . a b . The covariance of Dm z a depends crucially on Ž2.26. and
Ž2.27.; after some calculation one finds
a
dG Dm z a s g Ž u I t I . b Dm z b y dG f AGA a
b Dm z b . Ž 2.34 .
The gauge fields WmI are accompanied by complex scalars X , spinors and auxiliary I
V iI
fields Yi Ij , constituting off-shell N s 2 vector multiplets. For our notation of vector
multiplets, the reader may consult w18x.
The minimal coupling to the gauge fields requires extra terms in the supersymmetry
transformation rules for the hypermultiplet spinors as well as in the Lagrangian, in order
to regain N s 2 supersymmetry. The extra terms in the transformation rules are
d QX z a s 2 gX I k IA VAai ´ i je j , d QX z a s 2 gX I k IA VAa i ´ i j e j . Ž 2.35 .

These terms can be conveniently derived by imposing the commutator of two supersym-
metry transformations on the scalars, as this commutator should yield the correct
field-dependent gauge transformation.
We distinguish three additional couplings to the Lagrangian. The first one is
quadratic in the hypermultiplet spinors and reads
LgŽ1. s gX I gaA i ´ i jgbB j DB k A I z az b q h.c.s 2 gX I t Ig a Vbg z az b q h.c. Ž 2.36 .

The second one is proportional to the vector multiplet spinor V I and takes the form
LgŽ2. s y2 gk IA VAai Va b z bV I i q h.c.s 2 gk IA gAi a ´ i j z aV I j q h.c. Ž 2.37 .

Finally there is a potential given by
Lgscalar s y2 g 2 k IA k JB g A B X I X J q g PIi j Yi Ij , Ž 2.38 .
where PIi j is the triplet of moment maps on the hyper-Kahler

¨ space. These terms were
determined both from imposing the supersymmetry algebra and from the invariance of
the action. To prove Ž2.38., one has to make use of the equivariance condition Ž2.30..
Actually, gauge invariance, which is prerequisite to supersymmetry, already depends on
Ž2.31..
3. Rigidly superconformal hypermultiplets
¨
In this section we determine the restrictions on the hyper-Kahler geometry that follow
from imposing invariance under rigid superconformal transformations. As we already
mentioned in Section 1, the corresponding spaces, called special hyper-Kahler ¨ mani-
folds, have an intriguing geometrical structure. In Section 5 we will obtain the coupling
of hypermultiplets to conformal supergravity. A crucial element in the construction of
this coupling is that the full superconformal theory is known in an off-shell form, so that
the superconformal algebra remains unaffected in the presence of matter fields. Our goal
is more modest in this section where we only consider rigid superconformal transforma-
tions. This aspect does not play a role for the derivation of the superconformal
transformations on the hypermultiplets and the results of this section describe the
situation that would arise when freezing all the fields of conformal supergravity to zero
in a flat space-time metric. In that case the superconformal transformations acquire an
explicit but fixed dependence on the space-time coordinates parametrized by a finite
number of space-time-independent parameters Žthis is explained, for instance, in w23x..
In Subsection 3.1 we impose the superconformal algebra on the fields and find the
transformation rules as well as a number of important results for the complex structures
and the moment maps associated with possible isometries. In Subsection 3.2 we derive
¨
the existence of a hyper-Kahler potential and reformulate the theory in terms of local
sections of an SpŽ n. = SpŽ1. bundle. Then, in Subsection 3.3 we present the Lagrangian
and the transformation rules in terms of these local sections.
3.1. Superconformal transformations
We start by implementing the N s 2 superconformal algebra w5x on the hypermulti-

plet fields. We assume that the scalars are invariant under special conformal and special
supersymmetry transformations, but they transform under Q-supersymmetry and under
the additional bosonic symmetries of the superconformal algebra, namely chiral wSUŽ2.
= UŽ1.xR and dilatations denoted by D. At this point we do not assume that these
transformations are symmetries of the action and we simply parametrize them as
follows:
i
df A s u D k DA Ž f . q u UŽ1. k UŽ1.
A
Ž f . q Ž u SUŽ2 . . k ´ jk k iAj Ž f . , Ž 3.1 .
A
where the k are left arbitrary. Note that k iAj Ž f .
is assigned to the same symmetric
pseudoreal representation of SUŽ2. as the complex structures, while u SUŽ2 . is antihermi-
tian and traceless.
An important difference with the situation described in Section 2, is that in the
conformal superalgebra the dilatations and chiral transformations do not appear in the
commutator of two Q-supersymmetries, but in the commutator of a Q- and an

S-supersymmetry. To evaluate the S-supersymmetry variation of the fermions, we
assume that d S f A s d K z a s 0 and covariantize the derivative in the fermionic transfor-
mations with respect to dilatations. Subsequently we impose the commutator,
w d K Ž L K ., d Q Ž e .x s yd S Ž Lu K e . on the spinors. This expresses the S-supersymmetry
variations in terms of k DA ,
d S Ž h . z a s ViaA k DA h i , d S Ž h . z a s VAi a k DA hi . Ž 3.2 .

With this result we first evaluate the commutator of an S- and a Q-supersymmetry
transformation on the scalars. This yields
d S Ž h . , d Q Ž e . f A s Ž e ihi q e ih i . k DA q 2 Ji k A
B ´ k j e ihj y e jh i k DB .
ž / Ž 3.3 .
This result can be confronted with the corresponding expression from the N s 2
superconformal algebra, which reads
d S Ž h . , d Q Ž e . s d M Ž 2h is a be i q h.c. . q d D Ž hi e i q h.c. . q d UŽ1. Ž ihi e i q h.c. .
q d SUŽ2 . Ž y2h ie j y h.c.; traceless . . Ž 3.4 .

A
Comparison thus shows that k UŽ1. vanishes and that the SUŽ2. vectors satisfy
k iAj s Ji j A
B k DB . Ž 3.5 .
Now we proceed to impose the same commutator on the fermions, where on the
right-hand side we find a Lorentz transformation, a UŽ1. transformation and a dilatation,
if and only if we assume the following condition on k DA :
DA k DB s dAB . Ž 3.6 .
The geometric significance of these results will be discussed in later subsections. Here
we note that Ž3.6. suffices to show that the kinetic term of the scalar fields is invariant
under dilatations, provided one includes a space-time metric or, in flat space-time,
includes corresponding scale transformations of the space-time coordinates. Neverthe-
less, observe that k DA is not a Killing vector of the hyper-Kahler
¨ space, although it still
satisfies Ž2.21., but an example of a conformal homothetic Killing vector. Another
consequence is that the SUŽ2. vectors k iAj , as expressed by Ž3.5., are themselves Killing
vectors, because their derivative is proportional to the corresponding antisymmetric
complex structure
DA k Bi j s yJAi jB . Ž 3.7 .
¨
From this it follows that the Kahler two-forms are exact, provided that the Killing
vectors are globally defined. The product rule of the SUŽ2. Killing vectors can now be
worked out and one finds
k B i j E B k A k l y k B k l E B k A i j s 2 k A Ž iŽ k ´ l . j. , Ž 3.8 .
which is indeed in accord with the SUŽ2. structure constants.
From the w d S , d Q x commutator we also establish the fermionic transformation rules

under the chiral transformations and the dilatations,
d SUŽ2 . z a q d SUŽ2 . f A GA a
b z bs0 ,
1
d UŽ1. z a q d UŽ1. f A GA a
b z b s y i u UŽ1. z a ,
2
3
d D z a q d D f A GA a
bz
b
s uD z a . Ž 3.9 .
2
Note that the UŽ1. transformation further simplifies because d UŽ1. f A s 0.
To establish that the model as a whole is now invariant under the superconformal
transformations it remains to be shown that the tensor VAai is invariant under the
diffeomorphisms generated by k iAj , k UŽ1.
A
and k DA up to compensating transformations
that act on the SpŽ n. = SpŽ1. indices in accordance with the transformations of the z a
given above and the symmetry assignments of the supersymmetry parameters e i. To
A
emphasize the systematics we ignore the fact that k UŽ1. actually vanishes and we write
yk kBl E B VAai y EA k kBl VBai y k kBl GB a b

b VA i q ydŽjk ´ l .i VAaj s 0 ,
1 1
B
yk UŽ1. E B VAai y EA k UŽ1.
B
VBai q y i dba y k UŽ1.
B
GB a
b VAbi q i d i j VAaj s 0 ,
2 2
3 1
yk DB E B VAai y EA k DB VBai q dba y k DB GB a
b VAbi q y d i j VAaj s 0 . Ž 3.10 .
2 2
In these equations the first two terms on the left-hand side represent the effect of the
isometry or dilatation, the third term represents a uniform scale and chiral UŽ1.
transformation on the indices associated with the SpŽ n. tangent space, and the last terms
represent an SUŽ2., a UŽ1. and a scale transformation, respectively, on the indices
associated with SpŽ1.. Eq. Ž3.10. should be regarded as a direct extension of Ž2.22..
We close with a few comments. First of all, the SUŽ2. isometries induce a rotation on
the complex structures,
k kCl E C JAi jB y 2 E w A k kCl JBi jxC s y2 Jk lCw A J Bi jC Ži j. m
x s 2 d Ž k ´ l . m JA B , Ž 3.11 .
¨
as should be expected. Under dilatations, the Kahler two-forms JA B scale with weight
two, whereas the complex structures J A B are invariant.
Secondly, one can verify that the isometries introduced in Subsection 2.2 commute
with scale transformations, provided that
k IA s k DB DB k IA . Ž 3.12 .
This leads to another identity,
g A B k IA k DB s 0 . Ž 3.13 .
In particular these results hold for the SUŽ2. Killing vectors and imply, in addition, that
the latter commute with the tri-holomorphic isometries. To see this, one writes k iBj DB k I A
as k DB DA E B PI i j using Ž3.5., Ž2.29. and the fact that the complex structures are
covariantly constant. Interchanging the order of the derivatives and extracting the
complex structure then gives
k iBj DB k IA s JiAj B k IB , Ž 3.14 .
which implies that the tri-holomorphic Killing vectors commute with SUŽ2.. From the
above equations one can derive the following result for the variation of the moment
maps under a dilatation:
k DA EA PIi j s JAi jB k DA k IB s yk Ai j k IA s 2 PIi j , Ž 3.15 .
i.e. they scale with conformal weight 2. Here we have adjusted an integration constant in
PIi j in the last equation. Combining the above equation with previous results, one
establishes that the moment maps transform under SUŽ2. according to
k kAl EA PIi j s 2 dŽŽki ´ l . m PIj. m . Ž 3.16 .
The latter expresses the fact that the moment maps form a triplet under SUŽ2.. It is then
easy to check that the action is invariant under dilatations, UŽ1. and SUŽ2..
¨
3.2. Hyper-Kahler potential and Sp(n) = Sp( 1 ) sections
The existence of the homothetic Killing vector satisfying Ž3.6. has important conse-
quences for the geometry. First of all Ž3.6. implies that k DA can Žlocally. be expressed in
terms of a potential x , according to k D A s EA x . Up to a suitable additive integration
constant, one can then show that w12x
1
x Žf. s g A B k DA k DB . Ž 3.17 .
2
Observe that x is positive for a space of positive signature. A second Žcovariant.
derivative acting on x yields the metric, and therefore a third derivative vanishes,
DA DB x s g A B , DA DB DC x s 0. Ž 3.18 .
The first condition expresses the fact that the metric is the second Žcovariant. derivative
of some function, somewhat analogous to the Kahler ¨ ¨
potential in Kahler metrics, but
now written in real coordinates. A Kahler¨ potential is guaranteed to exist for any
¨
hyper-Kahler space, but the potential x does not always exist. In the literature x is
¨
sometimes called the hyper-Kahler potential Žsee, e.g., Refs. w8,9x.. This means that x
¨
serves as a Kahler potential for each of the three complex structures, as follows from
1
2
Ž dAC q J L A C . Ž dBD y J L B D . DC DD x s JALB , Ž 3.19 .
where J L s Ž s 2 s L . i j J i j and L s 1,2,3 is kept fixed.
¨
The hyper-Kahler potential x is invariant under isometries, as follows directly from
Ž3.13.. In particular it is invariant under the SUŽ2. isometry; explicitly,
i i
dx s Ž u SUŽ2 . . k ´ jk k iBj E B x s Ž u SUŽ2 . . k ´ jk Ji j A B k DB k DA s 0 , Ž 3.20 .
where we made use of Ž3.5.. However, it is not invariant under dilatations,
dx s k DB E B x s 2 x . Ž 3.21 .
Another interesting consequence of the homothety is that it enables a reformulation in

terms of local sections of an SpŽ n. = SpŽ1. bundle. The existence of such a so-called
associated quaternionic bundle is known from general arguments w9x. These sections are
defined from the S-supersymmetry variation of the hypermultiplet spinors Žc.f. Ž3.2..,
A i a Ž f . ' k DB Ž f . VBai Ž f . . Ž 3.22 .
They satisfy a quaternionic pseudo-reality condition
)
Ai a ' Ž A i a . s ´ i j V a b G bg A j g , Ž 3.23 .
as follows from Ž2.12.. Using Ž3.6. one proves that the covariant derivative of A i a
reproduces the quaternionic vielbeine,
DB A i a s VBai , ´ i j Va b DB A j b s g B C g iCa . Ž 3.24 .
¨
One easily verifies that the hyper-Kahler potential x can be written as
1 1 1
xs g A B k DA k DB s G ba A i aAi b s ´ i jVa b A i aA j b , Ž 3.25 .
2 2 2
or
Vab A i aA j b s ´ i j x . Ž 3.26 .
We also note the following identity:
J i j B C DC A k a s yd kŽ i´ j.l DB A l a . Ž 3.27 .
Furthermore we have
R A B a b A i b s R A B a b Vag A i g s 0, Ž 3.28 .
a
which is a consequence of DA DB A i s 0 and the symplectic nature of the curvature
R A B a b . This implies that the generic holonomy group is now reduced from SpŽ n. to
SpŽ n y 1.. Also, using Ž3.5., Ž3.26. and Ž3.27., one finds
k DB DB A i a s A i a , k i j B DB A k a s d kŽ i´ j.l A l a , Ž 3.29 .
so that
1
Vab A i a DB A j b s ´ i j k D B q k i j B . Ž 3.30 .
2
Applying a second derivative DA to the above relation gives
1
Vab DA A i a DB A j b s ´ i j g A B y Ji j A B . Ž 3.31 .
2
Note that the quantities in Ž3.31. have weight 2 under the homothety. For future use we
also recall some earlier results, but now expressed in terms of the local sections,
g A B DA A i a DB A j b s ´ i j V ab ,
g A B DA A i a DB A j b s d i j G a b ,
R A B g a Vgb DC A i a DD A j b ´ i j s R A B C D . Ž 3.32 .
3.3. The hypermultiplet action and transformation rules
In this subsection, we write the hypermultiplet action and transformation rules in

terms of the sections A i a Ž f . introduced in Ž3.22.. The complete Lagrangian, including
the terms associated with gauged isometries, can be written as
1
L s y Gab Dm A i b D mAi a y Ga b z a Duz b q z b Duz a ž /
2
1
y Wabgd z agm z b z gg mz d
4
q 2 g X g a Vbg z az b q 2 g z a Va b V i b g A i g q h.c.
1
q 2 g 2 Gab Ai a X b g X g d A i d q g A i a Va b Y i j b g A j g , Ž 3.33 .
2
where the covariant derivatives are defined by
Dm A i a s Em A i a q Em f A GAa b A i b y g Wm a
b Ai b ,
Dm z a s Em z a q Em f A GAa b z b y g Wm a
b zb , Ž 3.34 .
and we have used Lie-algebra valued vector multiplet fields associated with gauged
isometries, Wm a b , X a b , Y i j a b and V i a b Žfor the precise definition, see below., In
addition to the equation in Subsection 3.2 we made use of the identities,
k IA VAai s k IA DA A i a s t Ia b A i b ,
1 1 b
PI i j s y k A i j k IA s y Va b A i a Ž t I . g A j g . Ž 3.35 .
2 2
The first relation follows from Ž2.23. and Ž3.12., and for the second equation we made
use of the last equality in Ž3.15..
The action may be compared to the one in w6x Žmore precisely, to the part that
pertains to the rigidly supersymmetric Lagrangian.. However, in that reference, the A i a
are identical to the coordinate fields, whereas in the present more general case they are
local sections as explained in Subsection 3.2. Because the target-space manifold is not
flat, we encounter a non-trivial metric in Ž3.33. as well as non-trivial SpŽ n. connections
in the covariant derivatives Ž3.34.. Furthermore, the generators t I Ž f . associated with the
isometries are not constant, but depend on the scalar fields as we indicated before. This
means that the Lie-algebra valued vector multiplet fields associated with the gauged
isometries depend also on the hypermultiplet scalars. Their definitions are
a I a
Wm b s Wm tI Ž f . b ,
a a
X a b s X I tI Ž f . b , X a b s X I tI Ž f . b ,
a
Y i ja b s Y I i j tI Ž f . b ,
a a
V i a b s V I i tI Ž f . b , Vi a
b s Vi
I
tI Ž f . b . Ž 3.36 .
Nevertheless, the correspondence with the formulation in w6x will be helpful later on
when evaluating the coupling to conformal supergravity.
In order to obtain the transformation rules of the SpŽ n. = SpŽ1. sections under
dilations, SUŽ2. and isometry transformations, we use the general relation
d A i a s df B E B A i a s df B VBai y df BGB a
b Ai b . Ž 3.37 .
Using Ž2.9., Ž2.23. and Ž3.12., we then find for a combined dilatation, chiral transforma-
tion and target-space isometry, that
d A i a s u D A i a q Ž u SUŽ2 . . i j A j a q gu I t I a
b A i b y df AGA a
b Ai b . Ž 3.38 .
This result should be combined with that for the fermions, derived in Section 2,
3 1
dz a s u D z a y i u UŽ1. z a q gu I t I a b z b y df AGA a b z b . Ž 3.39 .
2 2
Similarly we determine the transformations under Q- and S-supersymmetry,
d A i a s 2 e i z a q 2 ´ i j G a bVbg e jz g y d Q f BGB a
b Ai b ,
dz a s Du A i ae i y d Q f BGB a
b z b q 2 g X a b A i b ´ i je j q A i a h i ,
dz a s Du Ai ae i y d Q f BGB a
b z b q 2 g X a b Ai b ´ i j e j q Ai a hi . Ž 3.40 .
Again, we stress that, apart from the SpŽ n. connection Žand a slight change in
notation., these transformation rules are identical to the ones specified for a flat target
space w6x, where the local sections can be identified directly with the target-space
coordinates.
Finally, we recall that it is straightforward to write down actions for the vector
multiplets that are invariant under rigid N s 2 superconformal transformations. Those
are based on a holomorphic function that is homogeneous of degree two w24x.
4. Cone structure and quaternionic geometry
¨
In this section we discuss the properties of the special hyper-Kahler space. We will
show how this space can be described as a cone over a tri-Sasakian manifold. The latter
spaces Žwhich are of dimension 4 n y 1. are characterized by the existence of three Ž1,1.
tensors and three Killing vectors that are subject to certain conditions. A manifold is
¨
tri-Sasakian if and only if its cone is hyper-Kahler. Tri-Sasakian spaces are Einstein and
take the form of an SpŽ1. fibration over a quaternionic space. This quaternionic space is
the one that appears in the coupling of hypermultiplets to supergravity Žfor more details,
see Ref. w10x, where the relation between special hyper-Kahler,
¨ tri-Sasakian and quater-
nionic spaces is reviewed from a more mathematical viewpoint..
We start by noting that the Riemann tensor vanishes upon contraction with any one of
the four vectors Ž k DA ,k iAj ., i.e.
R A B C E k DE s 0 , R A B C E k iEj s 0. Ž 4.1 .
The first equation Ž4.1. is derived by antisymmetrizing the second equation Ž3.18. in the
indices w AB x. The second Ž4.1. follows from inserting Ž3.7. into Ž2.21.. Incidentally,
Ž4.1. implies that the Ricci tensor has at least four null vectors. However, in the case at
¨
hand this poses no extra restrictions as hyper-Kahler spaces are Ricci-flat. The above
results can also derived from the fact that the SpŽ n. holonomy group is reduced to
SpŽ n y 1., c.f. Ž3.28.. This follows from applying Ž3.30..
We recall that these four vectors are orthogonal Žcf. Ž3.5., Ž3.17..,
k DA k D A s 2 x , k iAj k Ak l s dŽki d j.l x , k DA k Ai j s 0 . Ž 4.2 .
¨
This implies that the hyper-Kahler manifold is locally a product R 4 = Q 4 ny4 , where R 4
denotes a flat four-dimensional space. By decomposing R 4 as Rq= S 3 , we can write the
¨
hyper-Kahler manifold as a cone over a so-called tri-Sasakian manifold; the latter is then
a fibration of SpŽ1. over Q 4 ny4 . Hence the manifold can be written as 2 Rq=wSpŽ1. =
Q 4 ny4 . Spaces with a homothety can always be described as a cone. This becomes
manifest when decomposing the coordinates f A into coordinates tangential and orthog-
onal to the Ž4 n y 1.-dimensional hypersurface defined by setting x to a constant. The
line element can then be written in the form w12x,
dx 2
ds 2 s q 2 x h a b Ž x . dx a dx b , Ž 4.3 .
2x
where the x a are the coordinates associated with the hypersurface 3. In the present case
this hypersurface must be a tri-Sasakian space and the hyper-Kahler ¨ space is therefore a
cone over the tri-Sasakian space.
The purpose of the remainder of this section is to establish that Q 4 ny4 is a
quaternionic manifold. In Section 5 we show how Q 4 ny4 arises in the coupling of
hypermultiplets to supergravity. The tangent space of the hyper-Kahler ¨ space can be
decomposed into the four directions along Ž k DA ,k iAj ., and a Ž4 n y 4.-dimensional space
Q 4 ny4 that is locally orthogonal to that. Tensors that vanish upon contraction with
Ž k DA ,k iAj . will be called horizontal.
Let us introduce a vector field VA i j which will serve as a connection for SpŽ1. in a
way that will become clear shortly,
ki j A
VA i j s s Ji j A B E B ln x . Ž 4.4 .
x
This vector field is invariant under target-space dilatations and gauge isometries, i.e.
d D VA i j s k DB E B VA i j q EA k DB V B i j s 0 ,
dG VA i j s k IB E B VA i j q EA k IB V B i j s 0, Ž 4.5 .
and rotates under target-space SUŽ2., as follows from
d VAi j s k B k l E B VAi j q EA k B k l V Bi j s 2 ´ Ž iŽ k VAl . j. . Ž 4.6 .
2
Strictly speaking it is SpŽ1.rZ 2 where SpŽ1. is the group that acts on the quaternionic vielbeine and on
the sections introduced in the previous chapter.
3
In terms of a radial variable r 2 s 2 x , this yields the usual form of a cone metric
ds s dr 2 q r 2 h a b Ž x . dx adx b .
2
With VA i j we associate an SpŽ1. curvature tensor,

R A B i j ' EA V B i j y E B VA i j y ´ k l Ž VA i k V B jl q VA jk V B i l .
s xy1Dab DA A i a DB A j b q DA A j a DB A i b
, Ž 4.7 .
where we have used the definition
1
Dab s Va b y ´ k l Vag A k g
ž / žV bd Al d .
/ Ž 4.8 .
x
Observe that Dab is a projection operator, i.e. it satisfies Da b V bgDgd s yDa d , and it
projects onto the Ž2 n y 2.-dimensional subspace orthogonal to the A i a ,
Dab A i b s 0. Ž 4.9 .
Note that we have k DB DB Da b s k iBj DB Da b s 0, so that Da b is invariant under dilata-
tions and SUŽ2. transformations. One can also show that Dab DB A i b is horizontal, i.e.
k DB Da b DB A i b s k iBj Da b DB A i b s 0 . Ž 4.10 .
The identity Ž4.7. can be generalized to
1 1
xy1Dab DA A i a DB A j b s ´ i j GA B q R A B i j , Ž 4.11 .
2 2
where
GA B s xy1´ i j Da b DA A i a DB A j b . Ž 4.12 .
Observe that both GA B and R A B i j are of zero weight under the homothety and are
horizontal, i.e. they vanish upon contraction with any of the four vectors Ž k DA ,k iAj ., and
are thus orthogonal to the corresponding Žlocal. four-dimensional subspace.
The tensor GA B will provide a metric for Q 4 ny4 . The relation between GA B and the
¨
hyper-Kahler metric g A B is given by
1 1
gABs k DA k DB q k A i j k Bi j q x GA B
2x x
1
s k D A k D B q x VA i j V Bi j q GA B , Ž 4.13 .
2x
where we have used Ž3.30. and Ž3.31.. Observe that this relation reflects both the cone
¨
structure of the hyper-Kahler space and the SpŽ1. fibration of the tri-Sasakian space. It is
not possible to give an explicit expression for the inverse metric, at least not in general,
but this is not really needed in view of the horizontality of GA B . When acting on
horizontal tensors, x g A B acts as the inverse metric in view of the identity
GAC g C D GD B s xy1 GA B . Ž 4.14 .
b
We already showed that Dab DB A i was horizontal, and conversely, the horizontal
projection GA B g B C DC A i a is in the Ž2 n y 2.-dimensional eigenspace projected onto by
Dab . Therefore Da b DB A i b is a candidate for the quaternionic vielbein associated with
Q 4 ny4 and Dab projects onto the tangent space of Q 4 ny4 . More precisely, we introduce
the following related sets of 4 n y 4 vectors:
1 1
V̂Aai ' y V a b Dbg VAgi s y V a b Dbg DA A i g ,
x ' x '
1
gÂ i a ' GA B g iBa s ´ i j Da b Vˆ Aj b , Ž 4.15 .
'x
which satisfy algebraic relations that are completely analogous to those satisfied by the
¨
quaternionic vielbeine of the hyper-Kahler space. In particular we note that VÂ and gÂ
are each other’s inverse in the reduced 4 n y 4.-dimensional space,
Ž
x g A B gˆ Ai a VˆBbj s d ji Dag V bg , ´ i j Da b VÂai VˆBbj s GA B , Ž 4.16 .

bg
where Dag V is the identity matrix projected onto the Ž2 n y 2.-dimensional subspace.
The significance of these results will become clear in due course.
Subsequently we note that there exists an identity similar to Ž4.13. which relates the
complex structures to the field strength R A B i j ,
1 1
Ji j A B s y k D w A ki jB x q ´ k l kk iw A kl jB x y x RA B i j . Ž 4.17 .
x 2
This motivates us to introduce the following tensors:
JAi Bj s JAi j C GC B . Ž 4.18 .
A straightforward calculation using Ž4.17. shows that they satisfy
1
JA B i j s y R A B i j , Ž 4.19 .
2
so that the JA B i j are antisymmetric, horizontal and scale invariant. Furthermore these
tensors satisfy the product rule
ij
1
x JAC g C D JDk Bl s ´ iŽ k´ l . j GA B q ´ Ž iŽ k JAl .Bj. , Ž 4.20 .
2
which is similar to Ž2.8.. The tensors JA B i j are candidate almost-complex structures in
the horizontal subspace Q 4 ny4 . Under SUŽ2. target-space transformations they rotate
into each other according to
k kCl EC JA B i j q EA k kCl JC B i j q E B k kCl JAC i j s 2 ´Ž iŽ k JA B l . j. . Ž 4.21 .
Given a horizontal tensor HA B . . . that is invariant under the homothety and the SUŽ2.
target-space transformations, then the covariant derivative of such a tensor is no longer
horizontal. This can be cured by making use of a modified covariant derivative DÂ ,
defined so that the following properties hold:
k DA DˆC HA B . . . s k DC DˆC HA B . . . s 0,
k iAj DˆC HA B . . . s k iCj DˆC HA B . . . s 0. Ž 4.22 .
The modified derivative is obtained by using a modified target-space connection,
GÂ B C
s GA B C
y dŽCA E B . ln x q 2 V Ž A i j J Bi j.C . Ž 4.23 .
Because the modification is symmetric in Ž A, B ., the connection remains torsion free.

Observe that DÂŽ x g B C ., DÂ k DB and DÂ k iBj should be zero when contracted with a
horizontal tensor. This is obviously the case as can be seen from the formulae
DÂ Ž x g B C . s ydAŽ B k DC . q 2 J Ai jŽ B k iCj . ,
1
D̂A k DB s xy1 y k D A k DB q k i j A k i j B ,
ž /
2
1 1 B
D̂A k iBj s xy1 y k D A k iBj q k i j A k DB y k Ak l ´ k Ž i k j . l .
ž / Ž 4.24 .
2 2
The above construction can be generalized to tensors H that carry also SUŽ2. indices,
indicating that they transform covariantly under target-space SUŽ2. transformations, e.g.
as in k kAl EA H i s dŽik ´ l . j H j in the simplest case. Then one can show that the derivatives
of these tensors are still horizontal, provided one covariantizes DÂ and includes an
SUŽ2. connection VA i j . The crucial identity for showing this is k iAj VAk l s dŽki d j.l .
With respect to the new connection, GA B is covariantly constant,
D̂C GA B s 0, Ž 4.25 .
so that the new connection must be just the Christoffel connection associated with GA B .
Likewise the tensors JA B i j are covariantly constant modulo a rotation that involves the
SpŽ1. connection,
DˆC JA B i j s 2 V C kŽ i JA B j.l ´ k l . Ž 4.26 .

Note that the terms on the right-hand side covariantize the derivative on the left-hand
side with respect to SUŽ2.. Hence the tensors JA B i j define three almost-complex
structures in the space Q 4 ny4 which are covariantly constant up to an SpŽ1. rotation
proportional to the SpŽ1. connections. This implies that Q 4 ny4 is a quaternionic space
Žsee, e.g., Refs. w25–27x..
To verify this result, let us compute the Riemann tensor associated with the new
connection Ž4.23..
Rˆ A B C D s R A B C D y GCw A d BDx q R A B i j J Ci j D y R Cw A i j J Bi jxD . Ž 4.27 .

Observe that the right-hand side is not horizontal, but by construction Žvia the Ricci
identity. is horizontal when acting on a horizontal tensor with lower index D. Hence,
when lowering the index by contraction with the metric GD E one must obtain a
horizontal tensor. This is confirmed by explicit construction,
Rˆ A B C D ' Rˆ A B C E GE D s xy1 R A B C D q GDw A GB xC q R A B i j JCi Dj y R Cw A i j JBi xjD .

Ž 4.28 .
By virtue of Ž4.19. Rˆ A B C D has all the symmetry properties of a Riemann tensor.

Observe that the explicit factor of xy1 arises because the original curvature of the
¨
hyper-Kahler manifold is defined by lowering the upper index by means of the metric
g D E . Furthermore it satisfies the Bianchi identity Dˆ w A Rˆ B C x D E s 0.
Let us now calculate the Ricci tensor, which is symmetric by virtue of Ž4.25.,
Rˆ A B s x Rˆ AC B D g C D s y2 Ž n q 1 . GA B . Ž 4.29 .
¨
Observe that we used that the original hyper-Kahler manifold was Ricci flat and that
GA B g A B x s 4Ž n y 1.. We may also verify the expressions for the SpŽ1. holonomy
Rˆ A B C D g C E g D Fx 2 JEF
ij
s y4 Ž n y 1 . JAi Bj , Ž 4.30 .
where we used that the original hyper-Kahler¨ manifold has zero SpŽ1. holonomy. These
are the expected results w28,29x for a Ž4 n y 4.-dimensional quaternionic manifold with
SpŽ1. curvature given by Ž4.19..
This completes the discussion of target-space properties. We now return to aspects
¨
related to the SpŽ n. bundle over the special hyper-Kahler space. First of all we consider
a modification of the connection GA a b such that the modified derivative of a tensor
that is orthogonal to A i a remains orthogonal. This requires that this derivative acting on
A i a must be proportional to A i a itself. When combining this with a few other obvious
requirements 4 ,
a a
2
ĜA b s GA b y ´ i j A i Ž a DA A j g . q A i a A j g VAi j Vgb . Ž 4.31 .
x
With this modification, the tensors Vab and Ga b remain covariantly constant. The
presence of the term proportional to VAi j is required to preserve covariance with respect
to target-space SUŽ2. transformations. This term also ensures that the modification is
horizontal. With the modified connection we establish the required result,
1
D̂A A i a s EA ln x A i a q VA i k A l a ´ k l , Ž 4.32 .
2
where the last term can be interpreted as an SUŽ2. covariantization of the derivative on
the left-hand side. The result Ž4.32. suffices to show that the modified derivative of a
tensor that is orthogonal to A i a , will remain orthogonal. It is now obvious that the
projection operator Dab is covariantly constant under the modified derivative
D̂A Da b s 0. Ž 4.33 .
Including the modified connections GÂ B C and GÂ a b as well as the SUŽ2. connec-
tion VAi j, one can explicitly verify that DÂVBai is equal to 12 EA ln x VBai , up to terms that
4
In determining the precise modifications of the various connections, we were also guided to some extent
by supersymmetry. However, this aspect is postponed to Section 5, where we outline the significance of the
results of this section in the context of the coupling of hypermultiplets to supergravity.
are proportional to A k a. This implies that the quaternionic vielbeine introduced in Ž4.15.
are covariantly constant with respect to the new connections, so that we have
DÂ Ž A i ar x . s DÂVˆBai s DÂ gˆ B i a s 0 .

' Ž 4.34 .
This result leads to two integrability relations
Rˆ A B a b A i b y R A B i k A j a ´ k j s 0 ,
Rˆ A B C D gâ D i q Rˆ A B b
a gˆ Cib q R iAkB gˆ Cla ´ k l s 0 . Ž 4.35 .
Here Rˆ A B a b is the curvature associated with the new connection Ž4.31.. We can
explicitly evaluate this tensor,
2 1
R̂ A B a b s R A B a b q V agDgd ´ i j DA A i Ž d DB A j e . Deb y R Ai jB A i aA j g Vgb ,
x x
Ž 4.36 .
which indeed satisfies the first integrability relation. Note that all expressions appearing
in Ž4.36. are horizontal.
Now we recall that for a special hyper-Kahler¨ manifold the tensor Wabgd defined in
Ž2.16. satisfies the constraint
Wabgd A i d s 0 . Ž 4.37 .
With this in mind we write the new curvature tensors as follows:
1
Rˆ A B C D s ´ i j ´ k l VÂai VˆBbj VˆCgk VˆDd l Wâ bgd q GDw A GB xC
2
y 2 JAi Bj JC D i j q 2 JCwi j A JB x D i j ,
1
Va e Rˆ A B e b s y´ i j VÂgi VˆBdj Wâ bgd q 2 Da Ž g D d . b
2
q xy1 R iAjB A i gA j d Vga Vdb , Ž 4.38 .
where
Ŵabgd ' x Wa bgd . Ž 4.39 .
One can now verify that these curvatures satisfy also the second integrability condition
Ž4.35.. We will return to this and related issues in Section 5.
We close this section with a brief discussion of the isometries. For every tri-holomor-
¨
phic Killing vector of the special hyper-Kahler manifold we construct a corresponding
vector in the horizontal manifold Q 4 ny4 by the projection
kˆ I A s GA B k IB . Ž 4.40 .
By explicit calculation one can then show that DÂ kˆ IB q Dˆ B kˆ I A s 0, so that we have a
corresponding Killing vector in the horizontal space and thus an isometry. Observe that
¨
the SUŽ2. isometries of the special hyper-Kahler manifold do not generalize in this way,
because the corresponding kˆ I A would simply vanish. This is not so surprising, as the
SUŽ2. acts on the corresponding tri-Sasakian space through its SpŽ1. fibre.
To study whether these isometries are tri-holomorphic in the horizontal subspace, we

first raise the index according to
kˆ IA s x g A B GB C k IC s k IA q 2 PÎi j k iAj . Ž 4.41 .
where PÎ i j s xy1 PI i j . The transformation of the almost complex structures in the
horizontal subspace is then governed by the expression,
kˆ IC EC JA B i j y 2 E w A kˆ IC JB xC i j s k IC EC JA B i j y 2 E w A k IC JB xC i j
q 2 PÎk l Ž k kCl EC JA B i j y 2 E w A k kCl JB xC i j . . Ž 4.42 .

The first line on the right-hand side is zero, as follows from Ž4.17. and the fact that the
isometries are tri-holomorphic and commute with dilatations and SUŽ2. in the special
¨
hyper-Kahler space. The second line is equal to 4 JA B kŽ i ´ j.l PÎk l by virtue of Ž4.21.. We
can now elevate the derivatives on the left-hand side to SUŽ2. covariant derivatives. In
this way we find
DÂ JB C i j kˆ IC y Dˆ B JAC i j kˆ IC s y2 R A B kŽ i ´ j.l PÎk l ,
ž / ž / Ž 4.43 .
where we used the horizontality of kˆ IC and the Bianchi identity for Žor the covariant
constancy of. R A B i j A JA B i j . The solution is given by
JA B i j kˆ IB s DÂ PÎ i j , Ž 4.44 .

which can also be verified by explicit calculation. By substituting previous results one
verifies directly the modified equivariance condition,
JA B i j kˆ IA kˆ JB s yf I J K PˆK i j q 4 ´ k l PÎ kŽ i PˆJ j.l . Ž 4.45 .

The above results are in complete agreement with the moment map construction for
quaternionic manifolds w21,30x. The fact that the isometries generated by kˆ IA act
consistently on horizontal tensors is ensured by the following identities which follow
from explicit calculation:
k DB Dˆ B kˆ IA s k iBj Dˆ B kˆ IA s 0 . Ž 4.46 .
Finally the algebra of the isometries is governed by
kˆ IB E B kˆ JA y kˆ JB E B kˆ IA s yf I J K
kˆ KA q 2 JB C i j kˆ IB kˆ CJ k A i j . Ž 4.47 .
Hence the algebra of isometries is satisfied up to SUŽ2..
5. Locally superconformal hypermultiplets
In this last section we consider the coupling of the hypermultiplets to superconformal

gravity. To that order we introduce the Weyl multiplet, which contains the gauge fields
associated with the superconformal symmetries as well as some extra matter fields w5x.
The bosonic gauge fields are the vielbeine ema , the spin-connection vma b , the dilatational
gauge field bm , the gauge field associated with special conformal boosts fma and the
gauge fields associated with SUŽ2. = UŽ1., denoted by Vmi j Žantihermitian. and Am . The
fermionic gauge fields are the gravitino fields cmi and the fields fmi associated with
S-supersymmetry. Finally, the matter fields are Ta b i j Žantisymmetric and self-dual in
Lorentz indices and antisymmetric in SUŽ2. indices., a spinor x i and a real scalar D.
The fields vma b , fma and fmi are not independent and can be expressed in terms of the
other fields. We refer to w5,6x for more details on the notation and conventions.
The transformation rules have been given in previous sections, but will change in the
context of local supersymmetry. The most obvious change concerns the replacement of
the derivatives by derivatives that are covariant with respect to the additional gauge
symmetries. The derivatives covariant with respect to the bosonic gauge symmetries for
the scalar fields, the sections and the fermion fields, read
1
Dm f A s Em f A y bm k DA q Vmi k ´ jk k iAj y g WmI k IA ,
2
1
Dm A i a s Em A i a y bm A i a q Vm i j A j a y g Wm a b A i b q Em f A GAa b A i b ,
2
1 3 1
Dm z a s Em z a y vma b ga b z a y bm z a q iAm z a
4 2 2
a b A a b
y g Wm b z qEm f GA b z , Ž 5.1 .
where we have also included the terms related to possible gauged isometries. All
covariantizations follow straightforwardly from the formulae presented in Subsection 3.3
and from the gauge field conventions given in w5,6x. Observe that the derivative in Em f A
multiplying the connection GA a b does not require an additional covariantization.
The transformation rules under Q- and S-supersymmetry are now as follows:
df A s 2 g i Aa e zi a q gaA i e i z a ,
ž /
d A i a s 2 e i z a q 2 ´ i j G a bVbg e jz g y d Q f BGB a
b Ai b ,
dz a s Du A i ae i y d Q f BGB a
b z b q 2 g X a b A i b ´ i je j q A i a h i ,
dz a s Du Ai ae i y d Q f BGB a
b z b q 2 g X a b Ai b ´ i j e j q Ai a hi , Ž 5.2 .
where we have made use of the supercovariant derivatives Žwe also give the supercovari-
ant derivative of z a which is not needed above.,
Dm f A s Dm f A y g i Aa cmi z a y gaA i cm i z a ,
Dm A i a s Dm A i a y cm i z a y ´ i j G a bVbg cmj z g ,
1 1
Dm z a s Dm z a y Du A i acmi y
A af i . Ž 5.3 .
2 2 i m
We have verified that no further modifications of the fermionic transformation rules
beyond those given above are possible, assuming that the bosonic transformation rules
remain the same. One of the underlying reasons for the absence of additional terms may
be that the above rules are already consistent with rigid supersymmetry and with the
¨
case of a flat hyper-Kahler manifold which was taken as a starting point in w6x. All
additional modifications would thus have to vanish in the corresponding limits, while at
the same time one must preserve covariance under target-space diffeomorphisms and
fermionic frame reparametrizations. Therefore the possible modifications should be
proportional to the target-space curvature times the superconformal fields and, as it turns
out, it is difficult if not impossible to see how such terms could emerge. Given the fact
that the transformation rules take the same form, we expect the same situation for the
Lagrangian, where, again, it is difficult to construct suitable modifications that would
vanish in the appropriate limits.
Motivated by these considerations, we write down the Lagrangian by converting and
covariantizing the relevant equation Ž3.28. in Ref. w6x. Here we suppress the hypermulti-
plet auxiliary fields, as we no longer insist on off-shell supersymmetry for the hypermul-
tiplets. The result reads as follows, where the derivatives are all fully covariantized:
1 3
ey1 L s ´ i j Vab A i a D a Da q D A j bž /
2 2
1
q Vab 2 g 2 ´ i j A i a X b g X g d A j d q g A i a Y i j b g A j g
2
1
y ž z ay cmig mA i a /ž Gba Du z b
2
3 1
qVab ž ´ jkx j A k b y ´ jk Ta b jk g a bz b /
2 4
b
yg Vab V i ž g Ai g q 2 X b g z g / / q h.c.
1
q g Vab A i a V i b g z g q h.c. . Ž 5.4 .
2
After substituting the expressions for the dependent gauge fields fmi and fma in terms of
the other fields and dropping a total derivative, we write the Lagrangian as follows:
1 1 1
ey1 L s y Gab Dm A i b D mAi a q R Ga b A i bAi a q D Ga b A i bAi a
2 12 4
1
y Gab z a Duz b q z b Duz a y Wa bgd z agm z b z gg mz d
ž /
4
1 1
q Gab y žA i bAi a ey1´ mnrscm jgn Dr csj q A i bAi a c j m g mx j
12 8
1 1
y A k bA k a cmi cn j Timn ia b
j y A z xi q V ag Ggl z bg a b Ta b i j ´ i jz l
48 16
2 1
qz bg m Du Ai acm i y Ai a z b g mn Dm cn i q Ai a z bg a b Ta b i jg mcmj
3 24
1
y ey1´ mnrs cmign cr j A i b Ds A j a
4
1
y z bg mg ncm i cniz a q ´ i j V a r Grl cn j z l q h.c.
ž / / . Ž 5.5 .
2
Here we did not include the terms related to gauged isometries. To incorporate those one
includes the relevant terms into the covariant derivatives and adds the following
g-dependent terms to the Lagrangian:
1
ey1 Lg s 2 g 2 Ga b Ai a X b g X g d Adi q g A i a Va b Y i j b g A j g
2
q g 2 X g a z az b Vbg q 2 Va b z a V i b d A i d y 2 cmig mz b X a b Vag A i g
1 1
y cmig mV k a b Vag A i g A k b y cmig mncnk A k b
Vag A i g X a b q h.c. .
2 2
Ž 5.6 .
As mentioned above, these results are in agreement with the action presented in
Subsection 2.2 as well as with the results of w6x in the appropriate limits. In addition we
performed a number of independent checks on Ž5.5. and Ž5.6.. For instance, because the
superalgebra closes only modulo the field equations for the fermion fields z a and z a ,
we have calculated these field equations from the supersymmetry transformation rules
Ž5.2.. As it turns out the result is in agreement with the field equations derived from the
action.
The above action is invariant under all superconformal symmetries. In particular the
scalar fields are subject to dilatations and to SUŽ2. transformations. Ignoring the
contributions from the vector multiplets, which are essential for obtaining the complete
and consistent action for Poincare´ supergravity coupled to vector multiplets and hyper-
multiplets, but which do not affect the target-space geometry of the hypermultiplets, we
express the bosonic terms in scale-invariant quantities, by introducing a normalized
section
Aî a s xy1r2 A i a , Ž 5.7 .
which satisfies Vab ˆ a ˆ b
A i A j s ´ i j . Similarly we redefine the various other fields, such
as the vierbeine, spin connection, etc., by a x-dependent scale transformation. The result
for the bosonic terms then takes the form
1 1
L s y e Vab ´ i j Dm Aî a D mAˆj b y R y D , Ž 5.8 .
2 3
where R is the Ricci scalar of the space-time. Suppressing possible gauged isometries
for convenience, this results in
1 1 1
L s y e Vab ´ i j Em f A DA Aî a q Vm i k Aˆk a E mf B DB Aˆj b q V m j lAˆl b
ž /ž /
2 2 2
1 1
q e Rq e D . Ž 5.9 .
6 2
The field equations for the SUŽ2. gauge fields Vm i j yield,
Vm i j s y2 Em f A VA i k ´ k j . Ž 5.10 .
This result can be substituted back into the Lagrangian, which then reads
1 1 1
L s y e GA B Em f A E mf B q e R q e D , Ž 5.11 .
2 6 2
so that the target-space metric GA B corresponds indeed to the quaternionic space which
we constructed in Section 4. The terms with the Ricci scalar and the auxiliary field D
combine with similar terms from the Lagrangian of the vector multiplets to give the
Einstein–Hilbert action.
The material derived in Section 4 now fits in nicely with what is known about the
general coupling of hypermultiplets to supergravity w2x. First of all, the quantity Dab
projects out precisely the S-invariant hypermultiplet spinors which thus describe 2 n y 2
physical spinors after modding out the S-supersymmetry. Hence, the non-linear sigma
model comprises precisely the expected 4 n y 4 scalars and 2 n y 2 spinors. The relevant
quaternionic vielbeine have already been defined in Ž4.15., but can equally well be
obtained from working out the above Lagrangian after removing the appropriate gauge
degrees of freedom We will list a number of relevant identities, which all follow from
Section 4,
1 1
Dab VÂai VˆBbj s ´ i j GA B q RA Bi j ,
2 2
GA B gâA i gˆbB j s ´ i j Da b ,
R A B i j gâA k gˆbB l s 2 d iŽ kd jl . Da b . Ž 5.12 .

The second integrability condition Ž4.35. can be rewritten as
Rˆ A B C D gâC i gˆb D j s y´ i j Dag Rˆ A B g b y Da b R iAjB , Ž 5.13 .

which gives the decomposition of the Riemann tensor into an SpŽ n y 1. and an SpŽ1.
curvature. Of course, this relation is already incorporated into the expression Ž4.38. and
its correctness can also be verified directly. The curvature Rˆ A B a b satisfies Žc.f. Ž4.36..,
Vag Rˆ A B g b s Vag R A B g b y 2 ´ i j gÂŽi a gˆ Bjb . q Vag Vbd Aî g A j d R iAjB . Ž 5.14 .

Upon projection with D, the last term vanishes and one finds an identity that is well
known from the literature.
Hence we see that all aspects of quaternionic geometry that arise in the coupling of
hypermultiplets to supergravity are correctly reproduced. Our results provide an elegant
extension of the work reported in w6x and give a unified prescription for all hypermulti-
plet couplings to supergravity. Although this is in principle straightforward, it remains to
work out the details of the Lagrangian and transformation rules after removing the gauge
degrees of freedom associated with S-supersymmetry.
Acknowledgements
We acknowledge useful discussions with J. Figueroa-O’Farrill, G. Gibbons and A.

Swann. B.d.W. is grateful to the Alexander von Humboldt-Stiftung for supporting his
stay at the AEI as part of the Humboldt Award program. S. V. thanks PPARC for
financial support during his stay at the University of Swansea, and the Institute for
Theoretical Physics in Utrecht and the AEI for their hospitality. This work is supported
in part by the European Commission TMR programmes FMRX-CT96-0012, in which
the Albert Einstein Institut and the University of Wales in Swansea participate, and
ERBFMRX-CT96-0045, in which Utrecht University participates.
References
w1x L. Alvarez-Gaume, ´ D.Z. Freedman, Comm. Math. Phys. 80 Ž1981. 443.

w2x J. Bagger, E. Witten, Nucl. Phys. B 222 Ž1983. 1.
w3x A. Aharony, S.S. Gubser, J. Maldacena, H. Ooguri, Y. Oz, Large-N field theories, string theory and
gravity, hep-thr9905111.
w4x B. de Wit, B. Kleijn, S. Vandoren, in: Supersymmetries and Quantum Symmetries, Proc. Int. Sem.
Dubna, 1997, ed. J. Wess, E.A. Ivanov, Lecture Notes in Physics, Vol. 524 ŽSpringer, Berlin, 1999., p.
37, hep-thr9808160.
w5x B. de Wit, J.W. van Holten, A. Van Proeyen, Nucl. Phys. B 167 Ž1980. 186.
w6x B. de Wit, P. Lauwers, A. Van Proeyen, Nucl. Phys. B 255 Ž1985. 569.
w7x J.A. Bagger, A.S. Galperin, E.A. Ivanov, V.I. Ogievetsky, Nucl. Phys. B 303 Ž1988. 522.
w8x K. Galicki, Class. Quantum Grav. 9 Ž1992. 27.
w9x A. Swann, Math. Ann. 289 Ž1991. 421.
w10x C.P. Boyer, K. Galicki, Three-Sasakian Manifolds, to appear in: Essays on Einstein Manifolds, ed. M.
Wang, C. LeBrun, hep-thr9810250.
w11x B. Acharya, J. Figueroa-O’Farrill, C. Hull, B. Spence, Adv. Theor. Math. Phys. 2 Ž1998. 124.
hep-thr9808014.
w12x G.W. Gibbons, P. Rychenkova, Phys. Lett. B 443 Ž1998. 138, hep-thr9809158.
w13x S. Cecotti, S. Ferrara, L. Girardello, Int. J. Mod. Phys. A 4 Ž1989. 2475.
w14x A. Strominger, Phys. Lett. B 421 Ž1998. 139, hep-thr9706195.
w15x E. Kiritsis, C. Kounnas, Nucl. Phys. B 422 Ž1995. 472, hep-thr9501020.
w16x I. Antoniadis, S. Ferrara, R. Minasian, K.S. Narain, Nucl. Phys. B 507 Ž1997. 571, hep-thr9707013.
w17x H. Gunther,
¨ C. Herrmann, J. Louis, Quantum Corrections in the Hypermultiplet Moduli Space, in: Proc.
of the 32nd Symposium Ahrenshoop, Buckov, Fortsch. Phys. 48 Ž2000. 119, hep-thr9901137.
w18x J. De Jaegher, B. de Wit, B. Kleijn, S. Vandoren, Nucl. Phys. B 514 Ž1998. 553, hep-thr9707262.
w19x C.M. Hull, A. Karlhede, U. Lindstrom,¨ M. Rocek, ˇ Nucl. Phys. B 266 Ž1986. 1.
w20x G. Sierra, P.K. Townsend, Nucl. Phys. B 233 Ž1984. 289.
w21x R. D’Auria, S. Ferrara, P. Fre,
´ Nucl. Phys. B 359 Ž1991. 705.
w22x L. Andrianopoli, M. Bertolini, A. Ceresole, R. D’Auria, S. Ferrara, P. Fre,
´ T. Magri, J. Geom. Phys. 23
Ž1997. 111, hep-thr9605032.
w23x B. de Wit, Conformal invariance in extended supergravity, in: Supergravity ’81, ed. S. Ferrara, J.G.
Taylor ŽCambridge Univ. Press, Cambridge, 1982..
w24x B. de Wit, A. Van Proeyen, Nucl. Phys. B 245 Ž1984. 89.
w25x K. Yano, Differential Geometry on Complex and almost Complex Spaces ŽPergamon, New York, 1965..
w26x K. Yano, M. Kon, Structures on manifolds ŽWorld Scientific, Singapore, 1984..
w27x S. Kobayashi, K. Nomizu, Foundations of Differential Geometry ŽWiley, New York, 1969..
w28x D.V. Alekseevskii, Funct. Anal. Appl. 2 Ž1968. 97.
w29x S. Ishihara, J. Diff. Geom. 9 Ž1974. 483.
w30x K. Galicki, Commun. Math. Phys. 108 Ž1987. 117.
Chiral extensions of the WZNW phase space, Poisson–Lie

symmetries and groupoids
J. Balog a , L. Feher
´ b,1, L. Palla c
a
Research Institute for Nuclear and Particle Physics, Hungarian Academy of Sciences, H-1525 Budapest 114,
POB 49, Hungary
b
´
Institute for Theoretical Physics, Jozsef Attila UniÕersity, H-6726 Szeged, Tisza Lajos krt 84-86, Hungary
c
¨ ¨ UniÕersity, H-1117, Budapest,
Institute for Theoretical Physics, Roland EotÕos
´ ´ P. setany
Pazmany ´ ´ 1 A-ep,´ Hungary
Abstract
r
The chiral WZNW symplectic form V chir is inverted in the general case. Thereby a precise
r
relationship between the arbitrary monodromy dependent 2-form appearing in V chir and the
exchange r-matrix that governs the Poisson brackets of the group valued chiral fields is
established. The exchange r-matrices are shown to satisfy a new dynamical generalization of the
classical modified Yang–Baxter ŽYB. equation and Poisson–Lie ŽPL. groupoids are constructed
that encode this equation analogously as PL groups encode the classical YB equation. For an
arbitrary simple Lie group G, exchange r-matrices are found that are in one-to-one correspon-
dence with the possible PL structures on G and admit them as PL symmetries. q 2000 Elsevier
Science B.V. All rights reserved.
PACS: 11.25.Hf; 11.10.Kk; 11.30.Na

Keywords: WZNW model; Exchange algebra; Poisson–Lie symmetry
1. Introduction
This paper contains a systematic analysis of the classical phase space that arises from
the chiral separation of the degrees of freedom in the Wess–Zumino–Novikov–Witten
ŽWZNW. model w1x. The WZNW model occupies a central position in conformal field
theory w2x. Various structures that emerged from its study play an increasingly important
ˆ in other areas of theoretical physics and in mathematics as well w3x. Among these
role
1
Corresponding author.
´ ..
E-mail address: lfeher@sol.cc.u-szeged.hu ŽL. Feher
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 7 3 8 - 5
504 J. Balog et al.r Nuclear Physics B 568 [PM] (2000) 503–542
structures are the quadratic exchange algebras that encode the Poisson brackets ŽPBs. of
the chiral group valued fields, g C Ž x C . for C s L, R, which yield the general solution of
the WZNW field equation as g Ž x L , x R . s g LŽ x L . gy1
R
Ž x R .. These exchange algebras
were investigated intensively at the beginning of the decade w4–15x motivated by the
idea to understand the quantum group properties of the WZNW model w16x directly by
means of canonical quantization w17–21x. In accordance with the general philosophy of
quantum groups w22x, the Poisson–Lie ŽPL. symmetries of the chiral fields should be the
most relevant in this respect.
The chiral WZNW Poisson structures found in the literature have the form
m 1
½g C 5
Ž x . , gC Ž y . s Ž gC Ž x . m gC Ž y . . Ž rˆ q 12 Iˆsign Ž y y x . . ,
kC
0 - x , y - 2p , Ž 1.1 .
ˆ
where I is given by the quadratic Casimir of the simple Lie algebra, G , of the WZNW
ˆ The choice of the PB is
group, G, and the interesting object is the ‘exchange r-matrix’ r.
¨
highly non-unique due to the fact that the g C are determined by the physical field g
only up to the gauge freedom g C g C p for any constant p g G. In general rˆ may
depend on the monodromy matrix M of the chiral fields, which satisfy g C Ž x q 2p . s
g C Ž x . M. There are two qualitatively different cases that correspond to building the
WZNW field out of chiral fields with diagonal monodromy Ž‘Bloch waves’. or out of
fields with generic monodromy.
For Bloch waves w8–10x, the Poisson structure is essentially unique and the associated
r-matrix is a solution of the so-called classical dynamical Yang–Baxter ŽCDYB.
equation, which has recently received a lot of attention Žsee e.g. the review in Ref. w23x..
For chiral fields with generic monodromy, it has been argued in w11,14x that the
possible exchange r-matrices should correspond to certain local differential 2-forms r
on open domains Gˇ ; G, whose exterior derivative is the 3-form that occurs in the
WZNW action. The precise connection between r and rˆ has not been elaborated, and in
most papers dealing with generic monodromy actually only those very special cases
were considered for which rˆ is a monodromy independent constant. In these cases rˆ is
necessarily a solution of the classical modified YB equation on the Lie algebra G with a
certain definite normalization ŽEq. Ž3.71. with Ž3.64... This is a nice situation since if
the same r-matrix is used to equip G with a PL structure, then the gauge action of G on
the chiral WZNW field defines a PL symmetry. However, this mechanism of PL
symmetry is not available in the physically most interesting case of a compact Lie
group, because the relevant normalization admits no constant r-matrix for a compact G .
Thus, in addition to the problem to understand the case of a general r , an interesting
question is whether the exchange r-matrix can be chosen for a compact group in such a
way to admit a PL symmetry on the chiral WZNW phase space.
In this paper we study the family of chiral exchange algebras Ž1.1. concentrating on
the case of generic monodromy Žfor a related investigation of Bloch waves, see Ref.
w24x.. Our main results are the following.
v First, we establish the relationship between the 2-form r introduced in w11,14x

and the corresponding exchange r-matrix in the general case. The result is given
by Eq. Ž3.36. with the notations in Ž3.35., Ž3.21., Ž3.2..
J. Balog et al.r Nuclear Physics B 568 [PM] (2000) 503–542 505
v Second, we point out a dynamical generalization of the modified YB equation,

Eq. Ž3.63., whose solutions are the exchange r-matrices for generic mon-
odromy.
v Third, we present explicitly a subfamily of the exchange r-matrices whose
members are in one-to-one correspondence with the possible PL structures on G
and admit them as PL symmetries. These exchange r-matrices, given by Ž5.18.
with Ž5.22., contain the constant r-matrices studied earlier, and in another
remarkable special case of them the gauge action of the group G becomes a
classical symmetry, i.e. PL symmetry with the zero Poisson structure on G.
They work for any Žcompact or not. simple Lie group.
v Fourth, we construct a family of PL groupoids that encode the dynamical YB
equation Ž3.63. analogously as PL groups encode the classical YB equation.
This result generalizes a construction of w25x from diagonal to generic mon-
odromy.
The above mentioned results have been announced in w26x without proofs. In addition
to their detailed account, several other technical results can be found in this paper. The
systematic exposition of the subject and the numerous examples that we present may be
useful as a starting point for future studies.
The organization of the rest of the paper is as follows. In Section 2 a necessary
review of the chiral separation of the WZNW phase space is presented. Section 3
contains a detailed account of the inversion of the possible symplectic structures on the
chiral WZNW phase space, leading to the exchange algebra Ž1.1.. Here many interesting
additional issues are considered as well. In Section 4 an alternative, shorter but less
rigorous, derivation of the general exchange algebra is given, and a quick derivation of
the exchange algebra of Bloch waves is also included. Section 5 is devoted to a general
outline of the PL symmetries of the exchange algebra, and in particular to the exchange
r-matrices for which the standard gauge action of G on the chiral WZNW field yields
such a symmetry. Section 6 deals with the interpretation of the chiral WZNW Poisson
structures in terms of PL groupoids. The paper ends with a discussion, and there are also
two appendices containing some examples and the details of a proof.
2. The WZNW solution space and its chiral extension
In this section we review the structure of the WZNW Hamiltonian system concentrat-
ing on the possible symplectic forms on the chiral extension of its solution space, which
are examined throughout the paper. The presentation closely follows the line of thought
found in w14x.
We consider a simple, real or complex, Lie algebra, G , with a corresponding
connected Lie group, G, and identify the phase space of the WZNW model associated
with the group G as
M s T ) G˜ s Ž g , JL . < g g G,
½ ˜ JL g G˜ , 5 Ž 2.1 .
where G˜ s C S ,G . is the loop group and G˜ s C S , G . is its Lie algebra. The
`Ž 1 `Ž 1
isomorphism of the cotangent bundle T ) G˜ with G˜ = G˜ is established by means of

˜ The elements g g G˜ Žresp. JL g G˜ . are modeled as 2p-periodic
right-translations on G.
G-valued Žresp. G-valued. functions on the real line R. The phase space is equipped
with the symplectic form
2p k 2p X
V ksd H0 d s Tr Ž JL dg gy1 . q H0 d s Tr Ž dg gy1 . n Ž dg gy1 . Ž 2.2 .
2
with some constant k . Here prime denotes derivative with respect to the space variable,
s g R, and for any A, B g G Tr Ž AB . denotes a fixed multiple of the Cartan–Killing
scalar product. If g and the ‘left-current’ JL serve as coordinates on M , then the
‘right-current’ is given by
JR s ygy1 JL g q k gy1 g X , Ž 2.3 .
and in the alternative variables Ž g, JR . the symplectic form reads
2p k 2p X
V k s yd H0 d s Tr Ž JR gy1 dg . y H d s Tr Ž gy1dg . n Ž gy1 dg . . Ž 2.4 .
2 0
Although the expression of V k appears rather formal at first sight, it can be used to
unambiguously associate Hamiltonian vector fields and PBs with a set of admissible
functions, which include, for example, the Fourier components of g, JL and JR . We do
not elaborate the precise meaning of the symplectic form here, since this is a standard
matter in the context of the full WZNW model, but will face the analogous issue in the
chiral context later, where it is much less understood. The only point that we wish to
note is that in the case of a complex Lie algebra the admissible functions depend
holomorphically on the matrix elements of g, JL , JR in the finite-dimensional irre-
ducible representations of G, and G˜ = G˜ is then a model of the holomorphic cotangent
bundle.
The phase space M represents the initial data for the WZNW system, whose
dynamics is generated by the Hamiltonian
1 2p
H WZ NW s H0 d s Tr Ž JL2 q JR2 . . Ž 2.5 .
2k
Denoting time by t and introducing lightcone coordinates as
x L :s s q t , x R :s s y t ,
E E
EL s s 12 Ž Es q Et . , ER s s 12 Ž Es y Et . , Ž 2.6 .
E xL E xR
Hamilton’s equation can be written in the alternative forms w1x
kE L g s JL g , E R JL s 0 m kE R g s gJR , E L JR s 0. Ž 2.7 .
Let M sol be the space of solutions of the WZNW system. M sol consists of the
smooth G-valued functions g Ž s ,t . which are 2p-periodic in s and satisfy E R Ž E L g gy1 .
s 0. The general solution of this evolution equation can be written as
g Ž s ,t . s g LŽ x L . gy1
R Ž xR . , Ž 2.8 .
where Ž g L , g R . is any pair of G-valued, smooth, quasiperiodic function on R with equal
monodromies, i.e. for C s L, R one has g C Ž x C q 2p . s g C Ž x C . M with some C-inde-
pendent M g G. To elaborate this representation of the solutions in more detail, we

define the space Mˆ:
Mˆ:s Ž g L , g R . < g L , R g C` Ž R,G . , g L , R Ž x q 2p . s g L , R Ž x . M , M g G 4 . Ž 2.9 .
There is a free right-action of G on Mˆ given by
G 2 p: Ž g L , g R . ¨ Ž g p, g L R p. . Ž 2.10 .
™
Notice that Mˆ is a principal fibre bundle over M sol with respect to the above action of
G. The projection of this bundle, q : Mˆ M sol , is given by
q :Ž gL , gR . ¨ g s g L gy1
R i.e. g Ž s ,t . s g LŽ x L . gy1
R Ž xR . . Ž 2.11 .
We can identify M with M sol by associating the elements of the solution space with
their initial data at t s 0. Formally, this is described by the map
i : M sol ™M ,
i: M sol
2 g Ž s ,t . ¨ Ž g Ž s ,0. , J Ž s . s Ž kE g g
L L
y1
. Ž s ,0 . . g M . Ž 2.12 .
k.
Obviously, i )Ž
V is then the natural symplectic form on the solution space. Explicitly,
2p
Ž i )V k . Ž g . s yk d H d s Tr Ž gy1E R g gy1dg .
ž 0
1 2p
q
2
H0 d s Tr Ž gy1 dg . n Es Ž gy1 dg .
/ Ž 2.13 .
™M
ts0
Regarding now M sol as the base of the bundle q : Mˆ sol

, we obtain a closed 2-form,
Vˆ k , on Mˆ,
)
V̂ k :s q ) Ž i )V k . s Ž i (q . V k . Ž 2.14 .
By substituting the explicit formula Ž2.11. of q , one finds that
Vˆ k Ž g L , g R . s k L V chir Ž g L . q k R V chir Ž g R . , Ž 2.15 .
where
k L :s k , k R :s yk , Ž 2.16 .
and V chir is the so-called chiral WZNW 2-form:
2p X
V chir Ž g C . s y 12 H0 dx C Tr Ž gy1 y1
C dg C . n Ž g C dg C .
y 12 Tr Ž Ž gy1 y1
C dg C . Ž 0 . n dMC MC . ,
MC s gy1
C Ž x . g C Ž x q 2p . . Ž 2.17 .
This crucial formula of Vˆ was first obtained by Gawedzki w11x.
k
It is clear from its definition that d Vˆ k s 0, but Vˆ k is not a symplectic form on Mˆ,
™
since it is degenerate. Of course, its restriction to any Žlocal. section of the bundle
q : Mˆ M sol is a symplectic form, since such sections yield Žlocal. models of M sol. On
the other hand, one can check that V chir has a non-vanishing exterior derivative w11x:
d V chir Ž g C . s y 16 Tr Ž My1 y1 y1
C dMC n MC dMC n MC dMC . . Ž 2.18 .
Although this cancels from d Vˆ k , since ML s MR for the elements of Mˆ, it makes the
chiral separation of the WZNW degrees of freedom a very non-trivial and interesting
problem.
The problem of the chiral separation can be described as follows w14x. First, recall
that the chiral currents JC Ž C s L, R . generate two commuting copies of the non-twisted
affine Kac–Moody ŽKM. algebra of G and the WZNW field Ž2.8. is a KM primary field
under the Poisson bracket defined by the symplectic form on M sol. In fact, by defining
Fourier components as
2p
JCa , n :s H0 dx C eyi n x C Tr Ž T a JC . Ž x C . Ž 2.19 .
using a basis 2 T a of G , it can be derived from Ž2.13. that the currents satisfy
JCa , m , JCb , n 4 s fgab JCg , mqn q 2 ipkC m dm ,y n I ab , JLa , m , JRb , n 4 s 0, Ž 2.20 .
and
g Ž x L , x R . , JLa , n 4 s eyi n x T a g Ž x L , x R . ,
L
g Ž x L , x R . , JRa , n 4 s yeyi n x g Ž x L , x R . T a .
R
Ž 2.21 .
Second, the currents almost completely determine the chiral WZNW fields g C , and thus
also g s g L gy1
R , by means of the differential equations
k C EC g C s JC g C for C s L, R. Ž 2.22 .
Thus it appears an interesting possibility to construct the WZNW model as a reduction
of a simpler model, in which the left and right-moving degrees of freedom would be
separated in terms of completely independent chiral fields g L and g R regarded as
fundamental variables. It is clear that the solution space of such a chirally extended
model must be a direct product of two identical but independent spaces, i.e. it must have
the form
Mˆ ext :s ML = MR Ž 2.23 .
with
MC :s g C < g C g C` Ž R,G . , g C Ž x q 2p . s g C Ž x . MC , MC g G 4 . Ž 2.24 .
Ideally, the space Mˆ should be endowed with such a symplectic structure, ˆ
ext
that k
V ext ,
reduces to Vˆ k on the submanifold Mˆ; Mˆ ext defined by the periodicity constraint
ML s MR . Ž 2.25 .
ˆ k
It is easy to see that these requirements force V ext to have the following form:
Vˆ ext
k r
Ž g L , g R . s k L V chir r
Ž g L . q k R V chir Ž gR . , Ž 2.26 .
where
r
V chir Ž g C . s V chir Ž g C . q r Ž MC . Ž 2.27 .
with some 2-form r depending only on the monodromy of g C . Since in the extended
2
We have I a b :s Tr ŽT a T b . and w T a ,T b x s fgab T g with summation over coinciding indices.
r .
model the factors Ž MC , k C V chir should be symplectic manifolds separately, we have to
satisfy the condition
r
d V chir s y 16 Tr Ž MCy1 dMC n MCy1dMC n MCy1 dMC . q d r Ž MC . s 0. Ž 2.28 .
The problem now arises from the well-known fact that no globally defined smooth
2-form exists on G that would satisfy this condition for all MC g G.
There are two rather different wayouts from the above difficulty w14x. The first is to
restrict the possible domain of the monodromy matrix MC to some open submanifold in
G on which an appropriate 2-form r may be found. We refer to a choice of such a
domain and a r as a chiral extension of the WZNW system, and will explore the
structure of the associated PB in the subsequent sections.
The second possibility is to restrict the domain of the allowed monodromy matrices
much more drastically from the beginning, in such a way that after the restriction d V chir
vanishes, whereby the difficulty disappears. For example, one may achieve this by
restricting the monodromy matrices to vary in a fixed maximal torus of G, which
amounts to constructing Ža subset of. the solutions of the WZNW field equation in terms
of chiral ‘Bloch waves’. This second possibility is especially natural in the case of
compact or complex Lie groups, for which there is only one maximal torus up to
™
conjugation. The restriction to Bloch waves is equivalent to a partial Žand local. gauge
fixing of the bundle q : Mˆ M sol. The resulting symplectic form is studied in detail in
w24x.
3. The chiral WZWN phase space
We here investigate the structure of the chiral WZNW phase space MC introduced in
Section 2. The analysis is the same for both chiralities, C s L, R, and we simplify our
notation by putting Mchir for MC and g, M, J, k for g C , MC , JC , k C , respectively. We
assume that the monodromy matrix M is restricted to some open submanifold Gˇ ; G on
which a smooth 2-form r is chosen in such a way that Ž2.28. holds. The domain in
Mchir that corresponds to M g Gˇ is denoted by Mˇchir . It turns out that kV chir
r
, defined by
ˇ
Ž2.27. with Ž2.17., is non-degenerate if G is appropriately chosen Žso that Eq. Ž3.34. has
a smooth, unique solution., and we shall describe the general features of the PBs on
Mˇchir associated with this symplectic form. We will then consider examples, in particular
the choices of r introduced in w14x that lead to Poisson–Lie symmetry on the full Mchir .
3.1. Lie algebraic and differential geometric conÕentions
r
Before we can turn to the task of inverting kV chir , we need to set up some
conventions.
An element A g G has the components Aa s Tr Ž ATa . and A a s Tr Ž AT a . with
respect to dual bases Ta and T b of G :
Tr Ž Ta T b . s dab , I ab s Tr Ž T a T b . , Ia b s Tr Ž Ta Tb . . Ž 3.1 .
We will use I ab and Iab to raise and lower Lie algebra indices. Given a matrix Qa b ,
we can define an operator Q g EndŽ G . and an element Qˆ g G m G by
Q Ž A . s Ta Q abAb , Qˆ s Q ab Ta m Tb . Ž 3.2 .
The matrix I ab defines the identity operator I, and Iˆs Ta m T a is the ‘tensor Casimir’.
For any M g G, the matrix of the linear operator Ad M on G , which we write as
ŽAd M .Ž A. s MAMy1 , will be denoted as
Wab Ž M . s Tr Ž Ta MTb My1 . . Ž 3.3 .

The property ŽAd M .ŽAd M˜ . s ŽAd MM˜ . yields the matrix multiplication rule
Wag Ž M . I gu Wub Ž M˜ . s Wa b Ž MM˜ . , Ž 3.4 .
and we also have
Wba Ž M . s Wa b Ž My1 . s Way1
b Ž M., Wa b Ž M . I bg Wgu
y1
Ž M . s Iau . Ž 3.5 .
Acting on a smooth function c on G, we introduce the differential operators
d d
Ž La c . Ž M . :s c Ž e tTa M . , Ž Ra c . Ž M . :s c Ž Me tTa . , Ž 3.6 .
dt ts0 dt ts0
and their linear combinations

Da" :s Ra " La . Ž 3.7 .
™
For some purposes we will use a representation L:G GLŽ V . of G on a finite-dimen-
sional vector space V. The corresponding representation of G is denoted by the same
letter, and we put
M L :s L Ž M . for M g G, T L :s L Ž T . for T g G . Ž 3.8 .
Such representations will be used Že.g. the irreducible ones. that
Tr Ž Ta Tb . s cL tr Ž TaLTbL . Ž 3.9 .
holds with a constant cL , and we will then write
Tr Ž AB . :s cL tr Ž AB . , ;A, B g End Ž V . . Ž 3.10 .
On the left-hand side of Eq. Ž3.9. tr is a fixed Žrepresentation independent. multiple of
the Cartan–Killing form of G , while t r is matrix trace over the representation space V.
Remember that the phase space Mchir is parametrized by the G-valued field g Ž x .,
which is assumed to be smooth in x and is subject to the monodromy condition
g Ž x q 2p . s g Ž x . M M g G. Ž 3.11 .
X y1 Ž
The corresponding chiral current, J Ž x . s k g Ž x . g x . g G , is thus a smooth, 2p-
periodic function of x. To define tangent vectors at g g Mchir , we first have to consider
smooth curves on the phase space. Such a curve is given by a function g Ž x,t . g G,
which is smooth in x, t and satisfies the deformed monodromy condition
g Ž x q 2p ,t . s g Ž x ,t . M Ž t . , M Ž t . g G. Ž 3.12 .
To make sure that the curve passes through g g Mchir at t s 0, we require g Ž x,0. s g Ž x ..
A vector X w g x at g g Mchir is obtained as the velocity to the curve at t s 0, encoded by
the G-valued, smooth function
d
j Ž x . :s gy1 Ž x . g Ž x ,t . . Ž 3.13 .
dt ts0
It is useful to note that, due to the analogous property of g, the function j on R may be
reconstructed from its restriction to w0,2 p x. The monodromy properties of j Ž x . can be
derived by taking the derivative of Ž3.12.,
j X Ž x q 2p . s My1j X Ž x . M. Ž 3.14 .
This can be solved in terms of a smooth, 2p-periodic G-valued function, X J Ž x ., and a
constant Lie algebra element, j 0 , as follows:
x
y1
j Ž x. sj0q H0 dy g Ž y. XJ Ž y. g Ž y. . Ž 3.15 .
¨¨
A vector field X on Mchir is an assignment, g X w g x, of a vector to every point
g g Mchir . It acts on a differentiable function, g F w g x, on Mchir by the definition
d
XŽ F.w gxs F w gt x , g t Ž x . s g Ž x ,t . . Ž 3.16 .
dt ts0
Since a vector X w g x can be parametrized by j Ž x ., which, in turn, can be parametrized
¨
by the pair Ž j 0 , X J Ž x . . , we can specify a vector field by the assignments g j 0 w g x g G
and g X J w g x g G˜ . Of course, the evaluation functions F x w g x:s g Ž x . and F x w g x:s
¨
J Ž x . are differentiable with respect to any vector field, and their derivatives are given by
XŽ g Ž x . . s g Ž x . j Ž x . , XŽ J Ž x . . s k X J Ž x . , Ž 3.17 .
which clarifies the meaning of X J as well. It is also obvious from Ž3.11. that the
monodromy matrix yields a G-valued differentiable function on Mchir , g M s
gy1 Ž x . g Ž x q 2p ., whose derivative is characterized by the G-valued function
¨
X Ž M . My1 s Mj Ž x q 2p . My1 y j Ž x . . Ž 3.18 .
Having defined vector fields, one can now introduce differential forms as usual. We
only remark that by Ž3.17. evaluation 1-forms like dg Ž x ., dJ Ž x . or Ž gy1 dg .X Ž x . are
perfectly well defined:
dg Ž x . Ž X . s X Ž g Ž x . . s g Ž x . j Ž x . , dJ Ž x . Ž X . s X Ž J Ž x . . s k X J Ž x . ,
X
Ž gy1 Ž x . dg Ž x . . Ž X . s j Ž x . . X
Ž 3.19 .
3.2. Admissible Hamiltonians and Hamiltonian Õector fields
Now we turn to the following problem. For a fixed Žscalar. function F on the phase
space Mˇchir , we are looking for a corresponding vector field, Y F, satisfying
r
X Ž F . s kV chir Ž X ,Y F . Ž 3.20 .
for all vector fields X. Notice that Y F does not necessarily exist for a given F. We say
that F is an element of the set of admissible Hamiltonians, denoted as H, if the
We first compute kV chirr Ž

¨
corresponding Hamiltonian vector field, Y F, exists. Our purpose is to characterize H and
to describe the mapping H2 F Y F.
X,Y . for two vector fields X and Y. Let X be parametrized
by j Ž x . and further by the pair Ž j 0 , X J Ž x . . . The analogous parametrization for Y is
r
given by h Ž x . and the pair Ž h 0 ,YJ Ž x . . . Recall that kV chir is defined by Eqs. Ž2.17.,
Ž2.27., and parametrize r now as
r Ž M . s 12 q ab Ž M . Tr Ž Ta My1dM . n Tr Ž Tb My1dM . . Ž 3.21 .
The q ab , q a b s yq ba , are smooth functions on the domain Gˇ ; G, such that d V chir
r
s0
ˇ
on Mchir . A simple calculation, using partial integrations and 3.18 , gives that
Ž .
r
kV chir Ž X ,Y . s kV chir Ž X ,Y . q kr Ž X ,Y . , Ž 3.22 .
where
2p
V chir Ž X ,Y . s H0 dx Tr Ž X J Ž x . g Ž x . Ž h Ž x . y 12 My1 Y Ž M . . gy1 Ž x . .
y 12 Tr Ž j 0 Ž My1 Y Ž M . q Y Ž M . My1 . . Ž 3.23 .

and
2p
r Ž X ,Y . s H0 dx Tr Ž X J Ž x . g Ž x . B Y Ž M . gy1 Ž x . .
q Tr Ž j 0 Ž B Y Ž M . y MB Y Ž M . My1 . . Ž 3.24 .
with
B Y Ž M . :s q ab Ž M . TaTr Ž Tb My1 Y Ž M . . . Ž 3.25 .
Of course, all the expressions in the above formulae are functions of g g Mˇchir .
Let us now suppose that F g H and apply the above formula to Y:s Y F. Then the
form of the right-hand side of Ž3.20. implies that there must exist a smooth G-valued
function on R, A F Ž x ., and a constant Lie algebra element, a F, such that for any vector
field X
2p
X Ž F . sk H0 dx Tr Ž X J Ž x . A F Ž x . . q k Tr Ž j 0 a F . . Ž 3.26 .
¨ ¨
This means that F g H must have an exterior derivative parametrized by the assignments
g A F Ž x .w g x and g a F w g x. On the other hand, if F is such that Ž3.26. holds, then
we may try to solve Ž3.20. for the Hamiltonian vector field. Using the parametrization of
Y by h Ž x ., this leads to the following two equations:
h Ž x . y 12 My1 Y Ž M . q B Y Ž M . s gy1 Ž x . A F Ž x . g Ž x . , Ž 3.27 .
and
My1 Y Ž M . q Y Ž M . My1 y 2 B Y Ž M . q 2 MB Y Ž M . My1 s y2 a F . Ž 3.28 .
X
Now it is clear that h x can be directly read off from 3.27 . From the identity
Ž . Ž .
hX Ž x . s gy1 Ž x .YJ Ž x . g Ž x ., we then obtain that
X 1
YJ Ž x . s A F Ž x . q AF Ž x . , J Ž x . . Ž 3.29 .
k
This not only gives us the explicit formula of YJ , but also means that any F g H must be
such that the right-hand side of Ž3.29. defines a 2p-periodic smooth function of x.
Incidentally, this is equivalent to the monodromy condition Ž3.14. applied to the
Hamiltonian vector field Y. To proceed further, we use
Y Ž M . My1 s Mh Ž 2p . My1 y h Ž 0 . . Ž 3.30 .
This implies that Eqs. Ž3.27. and Ž3.28. are not linearly independent, and they can be
simultaneously solved for h Ž x . only if one has
a F s gy1 Ž 0 . A F Ž 0 . y A F Ž 2p . g Ž 0 . . Ž 3.31 .
We now also see that the pair Ž A F,a F . is uniquely determined for any F g H. Indeed,
the restriction of A F to w0,2p x is completely fixed by Ž3.26., and is uniquely extended to
a function on R on account of the periodicity of the expression in Ž3.29..
To summarize, we have shown that every element F g H must satisfy the three
conditions 3 expressed by Ž3.26., the periodicity of YJ Ž x . in Ž3.29., and Ž3.31.. Con-
versely, it turns out that these conditions characterize H. In fact, if these conditions are
satisfied then the solution of Ž3.27., Ž3.28. for h is given by
h Ž x . s gy1 Ž x . A F Ž x . g Ž x . y 12 a F q r Ž M . Ž a F . , Ž 3.32 .
where
r Ž M . Ž a F . s Ta r ab Ž M . abF Ž 3.33 .
ab
and the matrix r is defined as the solution of the linear equation
r ab q Ž Wg a y 2 qg a y 2 q au Wgu . r gb s 12 I a b y 12 W ba q q a b y qg a W bg . Ž 3.34 .
y1 Ž FŽ
This formula of h Ž x . s g x .Y g Ž x .. is one of the main results in this paper.
Some remarks are here in order. First, in Ž3.34. we suppressed the M-dependence of
the various matrices like qg a Ž M . s q ba Ž M . Ibg and Wg a s Tr ŽTg MT a My1 .. Second,
in terms of the notations given at the beginning of the section, r Ž M . is the linear
operator on G associated with the matrix r ab Ž M .. By introducing now the operators
r " Ž M . and q " Ž M . that correspond to the matrices
ab
r" Ž M . s r a b Ž M . " 12 I a b , ab
q" Ž M . s q a b Ž M . " 12 I a b , Ž 3.35 .
Eq. Ž3.34. can be rewritten, in fact, in the following equivalent form:
qq Ž M . ( ry Ž M . s qy Ž M . (Ad Ž My1 . ( rq Ž M . . Ž 3.36 .
The solution can be formally written as
y1
r Ž M . s 12 Ž qq Ž M . y qy Ž M . (Ad Ž My1 . .
( Ž qq Ž M . q qy Ž M . (Ad Ž My1 . . . Ž 3.37 .
This shows that one must define the domain Gˇ in such a way that the inverse operator
above exists, which is always possible since it becomes the identity operator at
M s e g G. Then it is easy to see that Ž3.37. yields a smooth, antisymmetric matrix
ˇ By choosing Gˇ ; G appropriately, hence we may indeed
function r ab Ž M . on G.
3
These conditions do not depend on the 2-form r , a reason for this is described at the end of Subsection
3.5.
associate with the smooth 2-form r on Gˇ a unique, smooth map Gˇ 2 M

EndŽ G .. Third, we will see in Subsection 3.3 that the object
¨rŽ M . g
rˆ Ž M . s r ab Ž M . Ta m Tb g G n G Ž 3.38 .
appears in the classical exchange algebra that encodes the Poisson brackets correspond-
r
ing to the symplectic form kV chir and it satisfies a dynamical generalization of the
¨
r
modified classical Yang–Baxter equation. Incidentally, kV chir is symplectic Ži.e. non-
degenerate. in the sense that it permits to unambiguously determine the map H2 F Y F,
as we just saw, and H will turn out to contain a ‘complete set of functions’ on Mˇchir .
Finally, as for the derivation of Eq. Ž3.34., note that one may arrive at the special
form of the integration constant in formula Ž3.32. of h Ž x . by the expectation of a
classical exchange algebra type PB for the field g Ž x ., or simply by inspecting the
equations that result if one writes h Ž x . s gy1 Ž x . A F Ž x . g Ž x . q constant. After introduc-
ing Ž3.32. as an ansatz, it is not difficult to verify that Ž3.27. and Ž3.28. reduce to Ž3.34..
3.3. Elements of H and their Poisson brackets
Below we describe a large set of functions that are admissible Hamiltonians and
apply the result in Ž3.32. to find their Hamiltonian vector fields. We shall also discuss
the interpretation of these Hamiltonian vector fields in terms of PBs, in particular we
shall see that the field g Ž x . is subject to a quadratic exchange algebra.
Let us first study functions that depend on g only through the current J s k g X gy1 .
y
Of course, the evaluation functions Fay w g x s Ja Ž y . do not belong to H, since A Fa Ž x . in
Ž3.26. would not be a smooth function of x. Therefore we consider the ‘smeared out’
version
2p
Fm :s H0 dx Tr Ž m Ž x . J Ž x . . , Ž 3.39 .
where m Ž x . is a G-valued, smooth, 2p-periodic test function. In this case we find that
A Fm Ž x . s m Ž x . , a Fm s 0. Ž 3.40 .
The conditions expressed by Ž3.26., Ž3.29. and Ž3.31. are trivially satisfied and thus
Fm g H. The parameter h Ž x . of the Hamiltonian vector field Y Fm is h Ž x . s
gy1 Ž x . m Ž x . g Ž x ., whence
Y Fm Ž g Ž x . . s m Ž x . g Ž x . ,
Y Fm Ž J Ž x . . s m Ž x . , J Ž x . q kmX Ž x . , Y Fm Ž M . s 0. Ž 3.41 .
This shows in particular that Fm generates an infinitesimal action of the loop group on
the phase space with respect to which g Ž x . is an affine KM primary field, and the KM
current J Ž x . transforms according to the coadjoint action of the Žcentrally extended.
loop group, as expected. Naturally, the local functionals of J defined as the integral
over w0,2 p x of any differential polynomial in the components of J, with periodic,
smooth test function coefficients, also belong to H; the corresponding Hamiltonian
vector fields are easy to determine.
Now we study some non-local functionals of the current. Let E g G denote the path
ordered exponential integral of J Ž x . over w0,2p x. More precisely, we put E :s EŽ2p .,
where EŽ x . g G is defined as the solution of
k EX Ž x . s J Ž x . E Ž x . with E Ž 0 . :s e g G. Ž 3.42 .
Let w be an arbitrary smooth function on G. Introduce a corresponding function, F , on
the phase space by
F w g x :s w Ž E . . Ž 3.43 .
From the well-known formula of the variation of EŽ x ., we obtain that
1
AF Ž x . s Ž E Ž x . T a Ey1 Ž x . . Ž Ra w . Ž E . , aF s 0, Ž 3.44 .
k
where Ra is defined in Ž3.6.. It follows that the conditions imposed by Ž3.26. and Ž3.29.
are satisfied, actually from Ž3.29. we get Y F Ž J Ž x .. s 0. However, F does not belong
to H in general. By means of Ž3.44. we get that
1
AFa Ž 0 . y AFa Ž 2p . s Ž Day w . Ž E . , Ž Days Ra y La . . Ž 3.45 .
k
Because of the condition Ž3.31., this means that F g H precisely if w is an inÕariant
function on G with respect to the adjoint action of G on G. Examples of invariant
functions are furnished by the trace of E k Ž k s 1,2, . . . . in some representation. That
only the invariant functions of E are admissible is a well-known result in the context of
current algebras, where they provide the Casimir functions of J. In our context, we
obtain from the above that for an invariant function w
1
YF Ž g Ž x. . s g Ž x . T a Ž Ra w . Ž M . , Y F Ž J Ž x . . s 0, Y F Ž M . s 0.
k
Ž 3.46 .
To derive these, we used that, since g and E satisfy the same differential equation,
g Ž x . s EŽ x . g Ž0.. Hence M s gy1 Ž0. E g Ž0., and for an invariant function
gy1 Ž 0 . T a g Ž 0 . Ž Ra w . Ž E . s T a Ž Ra w . Ž M . . Ž 3.47 .
The monodromy matrix M is not a function of J alone, but its invariant functions
coincide with those of E , and we have just seen that these functions belong to H. Let us
now take an arbitrary smooth function, c , on G and associate with it a function, C , on
the chiral WZNW phase space by C w g x:s c Ž M .. Using Ž3.18. and the definition of the
parameter of a vector field, Eq. Ž3.15., one gets
1 1
AC Ž x . s Ž g Ž x . T a gy1 Ž x . . Ž Ra c . Ž M . , aC s T a Ž Day c . Ž M . .
k k
Ž 3.48 .
CŽ
It follows that C g H. For the Hamiltonian vector field we obtain Y J Ž x .. s 0 and
1 1
gy1 Ž x . Y C Ž g Ž x . . s T a Ž Daq c . Ž M . q Ta r a b Ž M . Ž Dby c . Ž M . .
2k k
Ž 3.49 .
Let us elaborate this for the functions defined by the matrix elements of M in some
representation L of G. We denote these matrix elements as MiLj and denote by g iLj Ž x .
the corresponding matrix element of g Ž x .. Now we shall use rˆŽ M . in Ž3.38. and
ˆ
rˆ" Ž M . s rˆ Ž M . " 12 I, M1 s M m 1, M2 s 1 m M. Ž 3.50 .
Then Ž3.49. can be rewritten in the tensorial form
L 1 L
Y M k l Ž g iLj Ž x . . s
k
Ž g Ž x . m M Qˆ Ž M . . i k , jl , Ž 3.51 .
where
Qˆ Ž M . :s rˆq Ž M . y My1
2 rˆy Ž M . M 2 Ž 3.52 .
is taken in the corresponding representation of G , and our notation is Ž K m L. i k, jl s
K i j L k l . Furthermore, we obtain
L 1 L
Y M k l Ž MiLj . s
k
Ž M m M Dˆ Ž M . . i k , jl Ž 3.53 .
with
Dˆ Ž M . :s Qˆ Ž M . y My1 ˆ
1 Q Ž M . M1 . Ž 3.54 .
We shall comment on the interpretation of these equations later on.
The PB of two smooth functions F1 and F2 on a finite-dimensional smooth
symplectic manifold is defined by the standard formula
F1 , F2 4 s Y F 2 Ž F1 . s yY F1 Ž F2 . s V Ž Y F 2 ,Y F1 . , Ž 3.55 .
where Y F i is the Hamiltonian vector field associated with Fi by the symplectic form V .
The so obtained Poisson algebra is closed under pointwise multiplication of the
functions as well as under the PB. One may formally apply the same formula in the
infinite-dimensional case to the admissible smooth functions that possess a Hamiltonian
vector field. However, it then may be a very non-trivial problem to fully specify the set
of functions that form a closed Poisson algebra, and are a complete set in the sense that
they separate the points of the phase space. In our case, it is clear from the foregoing
formulae that the products of local functionals of the current J and the smooth functions
of the monodromy matrix M form two subsets of H that are separately closed under the
PB. Moreover, these two subsets commute with each other under the PB Žthey should
clearly be each others centralizer in an appropriate Poisson algebra.. But they do not
form a complete set of functions on our phase space, since the fundamental field g Ž x .
cannot be completely reconstructed out of J Ž x . and M.
™
Let us again consider a representation L:G GLŽ V . of G. Since the evaluation
functions Fi xj w g x s g iLj Ž x . are not elements of H, we smear out the local field and define
2p
Ff w g x :s H0 dx Tr Ž f Ž x . g L Ž x . . , Ž 3.56 .
where f :R ¨ EndŽV . is a smooth test function. It is then easy to see from Ž3.26. that
1 2p
A Ff Ž x . s g Ž x . T a gy1 Ž x . Hx dy Tr Ž f Ž y . g L Ž y . TaL . for x g w 0,2 p x ,
k
1 2p
a Ff s Ta H0 dyTr Ž f Ž y . g L Ž y . TaL . . Ž 3.57 .
k
By inspecting the condition that Y Ff Ž J Ž x .. in Ž3.29. must be periodic, we find that
Ff g H for those f that satisfy
f Ž k . Ž 0 . s f Ž k . Ž 2p . s 0 , k s 0,1, . . . Ž 3.58 .
Ff
Assuming that this holds, the Hamiltonian vector field Y is obtained from Ž3.32. as
1 2p
gy1 Ž x . Y Ff Ž g Ž x . . s Ta Hx dy Tr Ž TaLf Ž y . g L Ž y . . y 12 a Ff q r Ž M . Ž a Ff . ,
k
x g w 0,2 p x . Ž 3.59 .
This permits the following interpretation. Let us define the ‘Poisson bracket’ of the
evaluation functions by the equality
2p 2p m
Y Ff Ž Fx . :s Fx , Ff 4 :s H0 H0 dx dy Tr 12 x Ž x . m f Ž y . g L Ž x . , g L Ž y .
ž ½ 5/,
Ž 3.60 .
where Tr 12 means the normalized trace over V m V and
L m
½g Ž x. , g LŽ y. 5 i k , jl
s g iLj Ž x . , g kLl Ž y . 4 . Ž 3.61 .
With these definitions, formula Ž3.59. of the Hamiltonian vector field is equivalent to
m 1 L
½g L
Ž x . , g L Ž y . s Ž g L Ž x . m g L Ž y . . Ž rˆ Ž M . q 12 Iˆsign Ž y y x . . ,
5 k
0 - x , y - 2p . Ž 3.62 .
Ff Ž
Indeed, upon integration the right-hand side of Ž3.60. equals Y Fx . for any functions
f and x subject to Ž3.58.. This equation has the form of a quadratic exchange algebra
type PB for the field g Ž x .. Such a classical exchange algebra is usually regarded as a
classical analogue of a quantum group symmetry in the chiral WZNW model, but
observe that in general our r-matrix is monodromy dependent.
The admissible Hamiltonians of type Fm , C and Ff that we studied in the above
should together generate a closed Poisson algebra. Although at present we cannot fully
characterize the set of elements that belong to this algebra, we wish to point out that the
Jacobi identity for three functions of type Ff , in any Poisson algebra that contains them,
is equivalent to the following equation for rˆŽ M .:
rˆ12 Ž M . ,rˆ23 Ž M . q Qa b Ž M . T1a R b rˆ23 Ž M . q cycl. perm.s y 14 fˆ. Ž 3.63 .
Here
fˆs fabg T a m T b m T g Ž 3.64 .
and the cyclic permutation is over the three tensorial factors with rˆ23 s r ab Ž1 m Ta m Tb .,
T1a s T a m 1 m 1 and so on. Furthermore, we use the components of Qˆ s Qa b T a m T b
given by Ž3.52., for which
Qab R b s 12 Daq q ra b Dby . Ž 3.65 .

Eq. Ž3.63. can be viewed as a dynamical generalization of the classical modified YB
equation, to which it reduces if the r-matrix is a monodromy independent constant. Of
course, Ž3.63. is satisfied for any rˆŽ M . that arises as a solution of Ž3.34., since the
r
Jacobi identity is guaranteed by d V chir s 0.
For later reference, let us comment here on the analogue of Eq. Ž3.63. that appears in
connection with chiral WZNW Bloch waves. The space of Bloch waves 4 in question is
defined as
M Bloch :s b g C` Ž R,G . < b Ž x q 2p . s b Ž x . e v , v g A ; H 4 , Ž 3.66 .

where A is a certain domain in a Cartan subalgebra H of G . There is a natural
symplectic form on this space, which is induced by the embedding M Bloch ; Mchir and is
given by kV Bloch with
2p X
V Bloch Ž b . s y 12 H0 dx Tr Ž by1 db . n Ž by1 db . y 12 Tr Ž Ž by1 db . Ž 0 . n d v . .
Ž 3.67 .
It is known w8–10x Žfor a proof in the spirit of the present paper, see Ref. w24x. that the
PBs associated with Ž3.67. are encoded by the following classical exchange algebra:
m 1
½ b Ž x . , b Ž y . 5 s k Ž b Ž x . m b Ž y . . Ž Rˆ Ž v . q 1
2 Iˆsign Ž y y x . . ,
0 - x , y - 2p , Ž 3.68 .
1
Rˆ Ž v . s Ý < a < 2 coth Ž 12 a Ž v . . Ea m Ey a . Ž 3.69 .
4 a
The domain A is chosen so that a Ž v . f i2p Z for any root a and the root vectors Ea
satisfy the normalization Tr Ž Ea Ey a . s 2 < a < 2. As was first pointed out in w8x, the
Jacobi identity of the PB in this case gives rise to the equation
E
Rˆ12 Ž v . , Rˆ23 Ž v . q Ý H1k Rˆ23 Ž v . q cycl. perm.s y 14 fˆ, Ž 3.70 .
k Ev k
where v k s Tr Ž v Hk . with respect to a basis Hk of H whose dual basis is denoted as

H k . The same classical dynamical YB equation appears in other contexts w27–31x, too,
and has recently received lot of attention w23,25,32,33x.
4
In this context G is either a complex simple Lie algebra or its normal or compact real form.
3.4. Constant exchange r-matrices
r
We have seen that any symplectic structure kV chir on Mˇchir gives rise to a PB of the
form Ž3.62. governed by an ‘exchange r-matrix’ rˆŽ M ., which is a solution of Eq.
Ž3.63.. Those cases for which the exchange r-matrix is M-independent have already
been discussed by Falceto–Gawedzki w14x and others. The main point w14x here is that
one can construct an appropriate 2-form r out of any constant, antisymmetric solution r̂
of the modified classical YB equation,
rˆ12 ,rˆ23 q rˆ13 ,rˆ23 q rˆ12 ,rˆ13 s y 14 fˆ, Ž 3.71 .
r
and then the same rˆ appears in the exchange algebra determined by kV chir . We below
present the construction of w14x, showing that in our formalism it is easy to give a
complete proof as well.
All antisymmetric solutions of Ž3.71. are known. In fact, Belavin and Drinfeld w34x
classified the solutions in the case of a complex simple Lie algebra and their solutions
belong also to the normal real form. For other real forms very few solutions survive Žsee
Theorem 3.3 in w35x.. There is no solution for the compact real form, because of the
negative sign of the coefficient on the right-hand side. To explain the mechanism w14x
whereby constant exchange r-matrices appear in the chiral WZNW model, we first need
to recall a few standard results on Eq. Ž3.71., which can be found in the reviews Žsee,
e.g., Refs. w14,36x..
In association with a solution of Ž3.71., rˆ s r ab Ta m Tb g G n G , one has the
constant linear operators r and r "s r " 12 I. It follows from Ž3.71. that the formula
w A, B x r s r Ž A . , B q A,r Ž B . , A, B g G , Ž 3.72 .
™
defines a new Lie bracket on the linear space G ; the Lie algebra Ž G ,w , x r . is denoted as
Gr . Then r ": Gr G are Lie algebra homomorphisms,
r " Ž A . ,r " Ž B . s r " Ž w A, B x r . . Ž 3.73 .
Any A g G can be decomposed as
A s Aqy Ay with A " :s r " Ž A . . Ž 3.74 .
As a consequence of Ž3.73., one has the following equality of linear operators on G :
exp Ž ad A " . ( r "s r " (exp Ž ad r A . ;A g G . Ž 3.75 .
Here ad A and ad r A are defined by Žad A.Ž B . s w A, B x and Žad r A.Ž B . s w A, B x r for
any A, B g G . Note also that there exists a neighbourhood of the unit element in G, now
denoted as Gˇ ; G, whose elements, M g G, ˇ admit a unique decomposition in the form
y1
M s Mq My with M "s e G " , Ž 3.76 .
where G varies in a neighbourhood of zero in G . On Gˇ one has
My1 dM s My Ž Mq
y1 y1
dMqy My y1
dMy . My . Ž 3.77 .
ˇ By means of Ž3.73., Ž3.77. leads to
Let X be an arbitrary vector field on G.
My1 y1 y1
" dM " Ž X . s Ž My M dM Ž X . My . " . Ž 3.78 .
On the right-hand side the subscript refers to the decomposition Ž3.74..

According to Ref. w14x, the definition
r Ž M . :s 12 Tr Ž My1 y1
q dMqn My dMy . Ž 3.79 .
yields a 2-form on Gˇ for which
d r s 16 Tr Ž My1 dM n My1 dM n My1dM . ˇ

on G. Ž 3.80 .
It is straightforward to verify Ž3.80. by using Ž3.78., Ž3.73. and the antisymmetry of the
r-matrix, which imply e.g. that Tr Ž Aq w Bq ,Cq x . s Tr Ž Ay w By ,Cy x . for any A, B,C
g G.
Coming now to the main point, let us define Mˇchir ; Mchir to be the submanifold
where the monodromy matrix is restricted to G. ˇ Thanks to Ž3.80., r in Ž3.79. yields a
r
symplectic form kV chir by Ž2.27.. It is stated in w14x Žwithout a proof. that the PB Ž3.62.
on Mˇchir that results is in this case governed by the same constant r-matrix rˆ out of
which r Ž3.79. has been constructed. Our formalism permits us to prove this important
result as follows.
First, we need to rewrite the 2-form r in Ž3.79. in the notation used in Ž3.20.. With
the aid of Ž3.78., we obtain that
q ab Ž M . s W ag Ž My . rgu W ub Ž My
y1
., Ž 3.81 .
where we employ the notation of Ž3.3.. Here rab refers to the solution of Ž3.71. that we
used to define r , and we have to show that this monodromy independent r-matrix also
solves the defining equation of the exchange r-matrix, Eq. Ž3.36..
As a consequence of Ž3.81., for the operators r " and q " that appear in Ž3.36. we
have
y1
q " Ž M . s Ad Ž My . ( r " (Ad Ž My .. Ž 3.82 .
y1 Ž
By using this together with M s Mq My 3.76. and Ad M "s exp Ž ad G " . , the desired
identity Ž3.36. becomes equivalent to
rq (exp Ž yad Gy . ( rys ry (exp Ž yad Gq . ( rq . Ž 3.83 .
Because of Ž3.75., the last equation is in turn equivalent to
rq ( ry (exp Ž yad r G . s ry ( rq (exp Ž yad r G . , Ž 3.84 .
which is obviously valid since the operators rq and ry commute. This proves that the
constant r-matrix underlying r in Ž3.79. does indeed coincide with the exchange
r
r-matrix associated by Ž3.36. with the corresponding symplectic form kV chir .
A well-known feature of a constant exchange r-matrix is that it naturally admits a
Poisson–Lie action of the group G on Mchir . Observe that if rˆ in Ž3.62. is independent
of M, then the Poisson structure on Mˇchir smoothly extends to a Poisson structure on the
full Mchir . At the same time, one can give G the structure of a PL group by the
definition w37,38x
m 1
½ h , h5 s k w rˆ,h m h x h g G. Ž 3.85 .
Then one obtains a natural Žleft. PL action of G on Mchir by the map

Mchir = G 2 Ž g ,h . ¨ gh y1
g Mchir . Ž 3.86 .
Indeed, this is a Poisson map if Mchir = G carries the direct product of the exchange
¨
algebra PB on Mchir and the Sklyanin bracket Ž3.85. on G. In the present case, the
meaning of Eqs. Ž3.51.-Ž3.54. is that Mchir 2 g M s gy1 Ž x . g Ž x q 2p . provides the
‘non-abelian’ momentum map w14,36x for this PL action. ŽOf course, an equivalent right
PL action is obtained by replacing hy1 with h and using the opposite of the PB on G..
The above mechanism cannot be used to define a PL symmetry on Mchir in the case
of a compact Lie group G, since Ž3.71. has no solution for a compact G . This is
somewhat puzzling since as a quantum field theory the WZNW model is usually
considered in the compact domain, where various manifestations of quantum group
symmetries were found in the literature w16x. Later we shall see that PL symmetries can
be defined on the chiral WZNW phase space by certain mechanisms different from the
one described above, and they work in the compact case too.
3.5. A parametrization of Mchir and classical G-symmetry
We below introduce local coordinates on the chiral WZNW phase space consisting of
a periodic G-valued field and the logarithm, G , of the monodromy matrix. This will
r
lead us to realize the existence of a special choice of r such that with respect to kV chir
G generates an infinitesimal symplectic action of G on Mˇchir , i.e. a classical G-symme-
try. The parametrization will also shed a new light on the notion of admissible
Hamiltonians.
If the monodromy matrix M is near to e g G, then the chiral WZNW field can be
uniquely parametrized as
g Ž x . s hŽ x . e x G , Ž 3.87 .
¨
where hŽ x . is a G-valued, smooth, 2p-periodic function and G varies in a neighbour-
hood of zero in G , Gˇ ; G , for which the map Gˇ 2 G M s e 2 p G g Gˇ ; G is a
diffeomorphism. We may identify a domain in Mchir with the corresponding space of
parameters,
Mˇchir s G˜ = Gˇ s Ž h, G . 4 . Ž 3.88 .
An easy computation gives the following formula for V chir Ž2.17. in this parametriza-
tion:
0
V chir Ž h, G . s V chir Ž h, G . y r 0 Ž G . , Ž 3.89 .
where
2p X 2p
0
V chir Ž h, G . s y 12 H0 dx Tr Ž hy1 dh n Ž hy1 dh . . q d
H0 dx Tr Ž G hy1 dh . ,
Ž 3.90 .
2p
r 0 Ž G . s y 12 H0 dx Tr Ž d G n de x G eyx G . . Ž 3.91 .
Taking into account that M s e 2 p G , it is not difficult to verify that

d r 0 Ž G . s 16 Tr Ž My1 dM n My1dM n My1dM . . Ž 3.92 .
0
Upon comparison with Ž2.18., this implies that d V chir s 0,
which one can check directly
as well. Recalling Eq. Ž2.28., we then notice that the 2-form r in Ž2.27. in this case can
be parametrized by an arbitrary closed 2-form b on Gˇ as
r Ž G . s r0 Ž G . q b Ž G . , d b Ž G . s 0. Ž 3.93 .
r 0 r0 0
By Ž2.27. we thus have V chir s V chir qb,
in particular V chir s V chir
In order to .
determine the exchange r-matrix, rˆ0 , corresponding to r 0 , we note that the integral
defining r 0 can be computed in closed form. In fact, the linear operator, q0 , associated
with its matrix in Ž3.21. according to Ž3.2. turns out to be 5
2 Y q eyYY y e Y
q0 s with Y :s 2p Ž ad G . . Ž 3.94 .
2 Ž e Y y 1 . Ž 1 y eyYY .
Then from Eq. Ž3.36. we find the linear operator version, r 0 , of the exchange r-matrix
as
1
r 0 s 12 coth Ž Yr2 . y . Ž 3.95 .
Y
By means of Ž3.62. this r-matrix defines one of the possible monodromy dependent
exchange algebras for the chiral WZWN field, and it also represents a non-trivial
solution of Ž3.63.. In the knowledge of the r-matrix, the PBs containing M can be
determined straightforwardly from Ž3.51., Ž3.53.. In particular, it is easy to see that the
Hamiltonian vector field generated by the function Ga s Tr ŽTa G . gives rise to the PBs
g Ž x . , Gã 4 s yg Ž x . Ta , ½ G˜ , G˜ 5 s f
a b
g
ab G˜g for Gã :s y2pkGa . Ž 3.96 .
0
This means that in the case of the symplectic form kV chir the logarithm of M generates
a classical G-symmetry on Mˇchir . Indeed, the momentum map corresponding to this
symmetry is just G˜ . A classical G-symmetry is sometimes called ‘abelian’ to contrast it
with a proper Ž‘non-abelian’. PL symmetry, for which the symmetry group itself is
endowed with a non-zero PB w36x.
The above construction is valid for any simple Lie group. Perhaps even more
surprisingly than the possibility to define a classical G-symmetry on Mˇchir for any G , in
Section 5 it will turn out that the symplectic structure on Mˇchir can be chosen so as to be
compatible with any prescribed PL structure on G.
Now we explain how the characterization of the admissible Hamiltonians found in
Subsection 3.2 appears in the coordinates Ž h, G .. For this, we first remark that on
account of Ž3.88. it is natural to represent a vector field X on Mˇchir as
X s Ž X h , XG . , ˜
X h g ThG, XG g G Ž 3.97 .
y1 ˜
with h X h g G . By regarding h and G as evaluation functions, we may write
X h s X Ž h. and XG s X Ž G .. Of course, the derivative X Ž F . of function F on Mˇchir with
5
The expressions in Eqs. Ž3.94., Ž3.95. are defined by the power series expansions of the corresponding
2 2 kB
complex analytic functions around zero. For instance w39x, 2 r 0 sÝ`ks1 Ž 2 k .2!k Ž 21 Y . 2 ky1.
respect to X is still defined by means of a curve to which X is tangent. Let us call a

function F periodically differentiable if its derivative with respect to any X exists and
has the form
2p
X Ž F . s ² dF , X : s Tr Ž dG F XG . q H0 Tr Ž Ž hy1 d h F .Ž hy1 X h . . Ž 3.98 .
with the exterior derivative
dF s Ž d h F ,dG F . , ˜
d h F g Th) G, dG F g G , Ž 3.99 .
where h d h F g G˜ by the natural identifications. Our point then is that the periodically
y1
differentiable functions coincide precisely with the admissible Hamiltonians. To prove

this, recall that the definition of an admissible Hamiltonian was that its derivative exists
with respect to any vector field and has the form Ž3.26., where A F assigns to any
g g Mˇchir a smooth G-valued function on R such that Ž3.29. gives a 2p-periodic
function, and a F is given by Ž3.31.. If we now suppose that the derivative of a function
F has the form in Ž3.26., where A F Ž x . is a smooth function on R, then by inserting the
parametrization Ž3.87. and performing a few partial integrations we obtain
2p
XŽ F. s H0 dx Tr Ž X Ž k g X gy1 . A F . q k Tr Ž gy1 Ž 0 . X Ž g Ž 0 . . a F .
2p X
s H0 dx Tr Ž Ž X Ž h . hy1 . Ž yk Ž A F . y A F , k g X gy1 ..
2p 2p
ž
q k Tr X Ž G . H0 dx Ž gy1A F g . q k/ H0 dx Tr Ž eyx G X Ž e x G . , G
= Ž gy1 Ž x . A F Ž x . g Ž x . . . q k Tr Ž Ž gy1 Ž 0 . Xg Ž 0 . . Ž a F q gy1 Ž 0 .

= A F Ž 2p . y A F Ž 0 . g Ž 0 . . . Ž 3.100 .
for any vector field X on Mˇchir . Clearly, this formula can be rewritten in the form of
Ž3.98. if and only if
a F q gy1 Ž 0 . A F Ž 2p . y A F Ž 0 . g Ž 0 . s 0, Ž 3.101 .
F X
y Ž d h F . hy1 s k Ž A . q A F , k g X gy1 Ž 3.102 .
and
2p
Tr Ž X Ž G . dG F . s k H0 dx Tr Ž Ž X Ž G . q eyx G X Ž e x G . , G . Ž gy1A F g . .
2p X
sk H0 dx Tr Ž Ž eyx G X Ž e x G . . Ž gy1A F g . Ž x . .
2p
ž
s k Tr X Ž G . H0 dx Ž hy1A F h . Ž x . . / Ž 3.103 .
The last equality follows by an easy calculation, and it implies that

2p
dG F s k H0 Ž hy1A F h . Ž x . . Ž 3.104 .
We conclude that a function F for which X Ž F . has the form of the first line in Ž3.100.
is periodically differentiable if and only if Ž3.101. is satisfied and the right-hand side of
Ž3.102. defines a smooth, 2p-periodic function. In particular, all admissible Hamiltoni-
ans are periodically differentiable. Conversely, every periodically differentiable function
is admissible in the sense of Subsection 3.2 since it is possible to uniquely determine
Ž A F,a F . in terms of Ž d h F,dG F . in such a way Ž3.98. is converted into the first line of
Ž3.100.. To achieve this, if Ž d h F,dG F . are given, one has to solve the differential
equation Ž3.102. together with the condition in Ž3.104. for A F Ž x ., and then one may
define a F by Ž3.101.. It is not difficult to show that, since G is restricted to Gˇ , Ž3.104.
has a unique solution for the initial value A F Ž0. of A F Ž x ., which completes the proof.
4. Kinematical derivation of the Poisson brackets
In this section we rederive the PBs of the chirally extended WZNW model using
purely ‘kinematical’ considerations. Instead of explicitly inverting the symplectic form,
we postulate the natural properties of the chiral extension Žwhich we have established in
the symplectic formalism. and this way we can reproduce the quadratic exchange
algebras Ž3.62. and Ž3.68. almost effortlessly. This is especially so in the case of Bloch
waves, where the dynamical r-matrix Ž3.69. is determined algebraically. The subsequent
considerations are complementary to the symplectic approach presented in Section 3
and, for Bloch waves, in w24x. The kinematical derivation sheds a new light on the origin
of the chiral exchange algebras, and some issues are also easier to discuss in this
approach.
We have seen in Section 2 that it is very natural to extend the WZNW phase space as
M f M sol ™M ext
s MˇL = MˇR , Ž 4.1 .
ˇ ˇ
where ML and MR are two identical copies of the chiral phase space characterized by
the smooth, quasiperiodic chiral fields g C Ž x q 2p . s g C Ž x . MC , C s L, R. Since the
separation of the chiral degrees of freedom is an essential feature of the WZNW model,
we shall assume that MˇL and MˇR are independent and they are equipped with the same
symplectic structure Žup to an overall sign difference, see Eq. Ž2.16... The corresponding
Poisson algebra will be supposed to contain the important Hamiltonians Fm , C and Ff
studied in Subsection 3.3. The further main assumptions of the kinematical approach are
that the constraints ML s MR are first class and the corresponding gauge transforma-
tions 6 operate according to Ž2.10. so that the WZNW solution g Ž s ,t . s g LŽ x L . gy1
R
Ž xR.
is gauge invariant. These assumptions, together with simple properties of the original
WZNW phase space M sol , allow us to reproduce the PBs Ž3.62. and Ž3.68..
4.1. The chiral Poisson brackets for generic monodromy
From now on we mainly concentrate on the chiral half of the problem and, for
notational simplicity, omit the subscript C, wherever it is possible.
6
In the rest of the paper, when considering gauge transformations or G-symmetries, we shall always
assume that the domain of allowed monodromy matrices, G;G, ˇ consists of full conjugacy classes in G. If this
did not hold, everything would still be true for infinitesimal gauge transformations or G-symmetries.
We start by noting that because the Fourier components of the left KM current belong
to the space of admissible Hamiltonians, acting on the left chiral field they must
generate the transformation
2p
JTn , g Ž x . 4 s yT eyi n x g Ž x . , where JTn :s H0 dx eyi n x Tr Ž TJ Ž x . . , Ž 4.2 .
which means that the chiral field is a KM primary field. This crucial relation can be
obtained by first noting that on the submanifold of M ext defined by imposing the
constraint ML s MR Ž2.21. holds for the product Ž2.8. that gives the WZNW solution.
The gauge invariance of the solution field and the fact that the different chiral pieces
completely Poisson commute then allow us to derive Ž4.2.. Of course, an analogous
relation is valid for the right-moving fields.
An other quantity which, by assumption, belongs to the space of admissible Hamilto-
nians is the monodromy matrix M. From Ž4.2. it follows that
JTn , M 4 s 0 , Ž 4.3 .
which is obvious because M is invariant under the KM transformations. For later use we
note that a quantity which Poisson commutes with the Fourier components of the KM
current must be a function of the monodromy matrix M. ŽThis is most easily seen by
using the parametrization Ž3.87...
We wish to determine the PBs of the ‘smeared out’ field7
2p
Ff s H0 dx Tr Ž f Ž x . g Ž x . . , Ž 4.4 .
where the matrix valued test function f Ž x . satisfies Ž3.58.. Note that while in the
symplectic approach of Section 3 the fact that Ff is an admissible Hamiltonian follows
from the properties of the symplectic form, here it is an additional assumption. In order
to compute the PBs with Ff it is enough to find
Bf Ž x . :s Ff , g Ž x . 4 Ž 4.5 .
Ff Ž
for 0 ( x ( 2p Ž Bf Ž x . corresponds to yY g Ž x .. in Subsection 3.3.. To constrain
Bf Ž x . we apply Ž4.2. to Ff and then using the fact that Ff is an admissible Hamiltonian
we obtain the local form
Ff , J Ž x . 4 s Tr Ž f Ž x . T a g Ž x . . Ta . Ž 4.6 .
Comparing Ž4.5. and Ž4.6. leads to the differential equation
k BfX Ž x . y J Ž x . Bf Ž x . s Tr Ž f Ž x . T a g Ž x . . Ta g Ž x . , Ž 4.7 .
whose solution is
1 2p
Bf Ž x . s H0 dy sign Ž x y y . Tr Ž f Ž y . g Ž y . T a . g Ž x . Ta q g Ž x . Uf , Ž 4.8 .
2k
where
Uf s Ufa Ta Ž 4.9 .
is a constant element of the Lie algebra. If we consider Ff̃ , an other Hamiltonian of type
7
Here some representation L of G is used like in Ž3.56., but henceforth L is omitted from all notations.
Ff belonging to a smearing function f˜ Ž x ., and use the antisymmetry of the PB Ff , Ff˜ 4 ,

we obtain
Ufafã s yUfãfa , Ž 4.10 .
where
2p
fa :s H0 dx Tr Ž f Ž x . g Ž x . Ta . Ž 4.11 .
and fã is defined analogously. Eq. Ž4.10. implies that Ufa is a linear combination of the
integrals fa of the form
1
Ufa s y r a bfb , Ž 4.12 .
k
where r ab s yr ba . By means of Ž4.12., Ž4.8. becomes equivalent to the classical
exchange algebra
m 1
½ g Ž x . , g Ž y . 5 s k g Ž x . m g Ž y . Ž rˆq 1
2 Iˆsign Ž y y x . . , 0 - x , y - 2p ,
Ž 4.13 .
where the r-matrix rˆ s r ab Ta m Tb is an x-independent constant. Of course, as before,
Ž4.13. has to be interpreted in the distributional sense.
Although x-independent, the r-matrix can still depend on the phase space. This latter
dependence can be restricted by Poisson commuting JTn with Ž4.13. and applying the
Jacobi identity. In this way we get
JTn ,r ab 4 s 0 , Ž 4.14 .
which implies, as explained earlier, that rˆ must be a function of M alone.
Next we consider the PBs of the monodromy matrix. Using Ž4.5., Ž4.8. and Ž4.12. we
get
Ff , M 4 s gy1 Ž x . Ž Bf Ž 2p . y Bf Ž 0. M .
1
s fa Ž 12 Ž MT a q T a M . q r a b Ž MTb y Tb M . . Ž 4.15 .
k
and because M is an admissible Hamiltonian this implies the local form
m 1
½ g Ž x . , M 5 s k g Ž x . m M Qˆ Ž M . , Ž 4.16 .
where Qˆ is given by Ž3.52.. We obtain in a similar way that
m 1
½ M , M 5 s k M m M Dˆ Ž M . Ž 4.17 .
with Dˆ defined in Ž3.54..
To ascertain that our construction is self-consistent, we now show that
C s ML y MR f 0 Ž 4.18 .
is a first class constraint on M ext and the WZNW solution is gauge invariant. In fact, C
is first class since
m m m
½C , C5 s ½ M L 5 ½
, ML q MR , MR 5
1 1
s Ž ML m ML . Dˆ Ž ML . y Ž MR m MR . Dˆ Ž MR . f 0 .
k k
Similarly, the gauge invariance of g Ž s ,t . in Ž2.8. can be shown as follows:
m
k g Ž s ,t . , C s Ž g LŽ x L . m ML . Qˆ Ž ML . Ž gy1
½ 5 R Ž x R . m 1.
y Ž g LŽ x L . m MR . Qˆ Ž MR . Ž gy1
R Ž x R . m 1.
f Ž g LŽ x L . m M . Ž Qˆ Ž ML . y Qˆ Ž MR . . Ž gy1
R Ž x R . m 1. f 0 .
Here the notation f indicates ‘weak’ equality, i.e. equality on the constrained
manifold, and we used the assumption that MˇL and MˇR are independent and carry
opposite PBs.
For later use we mention an additional consistency check. Consider the path ordered
integral E s EŽ2p . defined in Ž3.42.. Since E and M are related by conjugation, their
invariants coincide:
´ N s Tr Ž E N . s Tr Ž M N . s m N , Ž 4.19 .
and since the PBs of the Hamiltonians ´ N can be calculated unambiguously using the
KM algebra only, the following relation must hold:
N
g Ž x . ,m N 4 s g Ž x . , ´ N 4 s g Ž x . T a Tr Ž M N Ta . . Ž 4.20 .
k
It is easy to verify by using Ž3.52. in Ž4.16. that Ž4.20. is indeed satisfied.
To summarize, by postulating the ML s MR constraint to be first class, as well as the
admissible Hamiltonian nature of the Fourier components of the current, the smeared out
chiral field and the monodromy matrix, we have established that the possible extensions
of the WZNW phase space correspond to the quadratic exchange algebra Ž4.13. with
some monodromy dependent exchange r-matrix. Of course the classical exchange
algebra can only provide a valid PB if it satisfies the Jacobi identity. This is guaranteed,
by construction, if the r-matrix is obtained as a solution of Ž3.36.. In the present
approach, we have to impose the Jacobi condition and, as mentioned in Subsection 3.3,
this leads to the dynamical YB equation Ž3.63.. The chiral extensions of the WZNW
model are thus characterized by the solutions of this equation.
Most known solutions of Ž3.63. are local in that they are defined only on a proper,
open submanifold Gˇ ; G. This is obviously the case for the solutions obtained by
solving Ž3.36. for the r-matrix, starting from a q-matrix representing by Ž3.21. a local
solution of Ž2.28.. On the other hand, as the example of constant r-matrices shows, there
are global solutions as well. Since constant solutions exist for non-compact groups only,
an interesting open question is whether there exist globally defined exchange r-matrices
also for compact groups. We hope to return to this question in a future study.
We end this subsection by discussing a generalization of the ‘gauge’ freedom Ž2.10.
™
of the chiral extension. It is clear that the gauge transformed chiral fields given by
gŽ x. Ž ¶g . Ž x . s g Ž x . p Ž M . Ž 4.21 .
with an arbitrary monodromy dependent group element p M correspond to the same
Ž .
point in the physical phase space after the projection Ž2.11., provided we apply the same
gauge transformation to both chiral fields. In other words, both the original fields, g Ž x .,
and the gauge transformed fields, g˜ Ž x .:s g Ž x . pŽ M ., are smooth quasiperiodic G-val-
ued ‘coordinates’ of the same point in the physical phase space.
The multiplication law for two elements, ¶1 ,¶2 , of this huge gauge group is given by
p 12 Ž M . s p 2 Ž M . p 1 Ž ¶2 M . , Ž 4.22 .
where p 1 , p 2 and p 12 correspond to ¶1 ,¶2 and ¶12 :s ¶1¶2 :s ¶1 (¶2 , respectively. Here
¶M :s py1 Ž M . Mp Ž M . Ž 4.23 .
is how the monodromy matrix itself is transformed under a gauge transformation. To
qualify as an element of the gauge group we must also require that the inverse
transformation exists. This is equivalent to requiring that the group valued function,
pŽ M ., corresponding to the inverse element, ¶y1 , exists and solves
p Ž M . p Ž ¶M . s e . Ž 4.24 .
In terms of the new ‘coordinates’ g˜ Ž x . defined by Ž4.21. the exchange algebra has
the same form as Ž4.13. with a gauge transformed exchange r-matrix, r. ˜ˆ On account of
the Leibniz rule of the PB, one finds that
r̂˜ s py1 Ž M . m py1 Ž M . Ž rˆ Ž M . q Q a b Ta m Ab y Ab m Ta
qD ab Aa m Ab . p Ž M . m p Ž M . , Ž 4.25 .
where Aa :s Ž Ra p . p and r˜ˆ should be expressed as a function of the gauge
y1
transformed monodromy M:s ˜ ¶M. Since the Jacobi identity of the exchange algebra
Ž4.13. is independent of the coordinates used, it is clear that the solutions of the
dynamical YB equation Ž3.63. are transformed into each other by the elements of the
gauge group and therefore can be classified up to gauge transformations.
4.2. Diagonal monodromy
Below we briefly outline a kinematical derivation of the PBs on the space of chiral
WZNW Bloch waves, M Bloch defined in Ž3.66.. In order to emphasize their diagonality,
we denote the monodromy matrices of the Bloch waves, bŽ x ., here by D,
k
D s e v Hk , Ž 4.26 .
where Hk are the basis elements of a splitting Cartan subalgebra of G . We will also use
the derivatives E k s E Ev k.
The assumptions and the main steps of the construction Žwith obvious modifications
for the diagonal case. are the same as discussed above for the general case. Now we
obtain a classical exchange algebra of the form
m 1
½b Ž x . , b Ž y . s b Ž x . m b Ž y . Ž rˆ Ž v . q 12 Iˆsign Ž y y x . . , 0 - x , y - 2p .
5 k
Ž 4.27 .
The main difference comes from requiring Ž4.20., because, unlike in the general case
where it was a consequence of the exchange algebra, here it completely determines the
PBs of the monodromy matrix:
m 1 k
½ bŽ x . , D5 s k Ž bŽ x . m D. Ž H m H k .. Ž 4.28 .
On the other hand, the analogue of Ž4.16. implied by Ž4.27. now reads explicitly as
m 1
½ b Ž x . , D 5 s k Ž b Ž x . m D . ž rˆ Ž v . q 1
2 Iˆy Ž 1 m Dy1 . Ž rˆ Ž v . y 12 Iˆ. Ž 1 m D . .
/
Ž 4.29 .
The comparison of the last two equations fixes the exchange r-matrix almost com-
pletely:
rˆ Ž v . s Rˆ Ž v . q Xˆ Ž v . , Ž 4.30 .
where RˆŽ v . is given by Ž3.69. and XˆŽ v . is an antisymmetric purely Cartan piece,
Xˆ Ž v . s X k l Ž v . Hk m Hl , X k l Ž v . s yX l k Ž v . . Ž 4.31 .
Thus we have determined the exchange r-matrix algebraically up to the Cartan piece.
The Jacobi identity takes the following form for the diagonal case:
rˆ12 Ž v . ,rˆ23 Ž v . q H1k E k rˆ23 Ž v . q cycl. perm.s y 14 fˆ. Ž 4.32 .

This is the celebrated CDYB equation, whose neutral solutions have been classified in
w25x. The r-matrix Ž4.30. satisfies Ž4.32. if
E k X l m q E l X m k q Em X k l s 0. Ž 4.33 .
Therefore there exists a ‘gauge vector’ Vk Ž v . such that
X k l Ž v . s E k Vl Ž v . y E l Vk Ž v . . Ž 4.34 .
With the help of Vk Ž v . we can introduce the gauge transformed chiral field
k
b˜ Ž x . s b Ž x . eyV k Ž v . H , Ž 4.35 .
which has diagonal monodromy and satisfies the ‘standard’ exchange algebra
m 1
½ b˜ Ž x . , b˜ Ž y . 5 s k b˜ Ž x . m b˜ Ž y . Ž Rˆ Ž v . q 1
2 Iˆsign Ž y y x . . , 0 - x , y - 2p .
Ž 4.36 .
It has been mentioned that the PB Ž4.36. follows w24x from the symplectic form
kV BlochŽ b˜ . on M Bloch given by Ž3.67.. Upon the substitution Ž4.35., the symplectic form
gets shifted by the exact 2-form k X k l Ž v . d v k n d v l and the shifted symplectic form
corresponds of course to the exchange algebra Ž4.27. with rˆŽ v . in Ž4.30.. It is also easy
to see that the family of the symplectic forms k Ž V Bloch Ž b . q X k l Ž v . v k n v l . on
r
M Bloch results as the reduction of the family of symplectic forms kV chir to diagonal
monodromy.
5. Exchange r-matrices with Poisson–Lie symmetry
We studied in Subsection 3.4 the Poisson–Lie action of the group G on the chiral
ˆ of the exchange
phase space for the special case of a constant r-matrix playing the role
r-matrix. As mentioned there, this PL action is restricted to the case of complex or real,
non-compact groups, since there is no constant solution of Ž3.63. for compact groups. In
this section we consider a set of more general PL actions which also work for the
physically most interesting case of compact groups.
It is clear from the examples studied so far that the PL action we are looking for is a
kind of ‘hidden’ symmetry, extending and centralizing the Kac–Moody symmetries in
the total symmetry group of the chiral WZNW model. More precisely, we require that
v the KM currents are invariant under the PL action;

v the PL action commutes with the KM transformations.
It is not difficult to see that because of the above two requirements the PL action as a
non-linear group action on the chiral phase space has to be a special case of the gauge
™ ¶ g Ž x . s g Ž x . p Ž M ,h.
transformations discussed in Subsection 4.1:
gŽ x. h ;h g G. Ž 5.1 .
The G-valued function pŽ M,h. is chosen so that the group multiplication law ¶h¶k s ¶h k
is satisfied:
p Ž M ,k . p Ž ¶k M ,h . s p Ž M ,hk . , Ž 5.2 .
y1 Ž
where the induced action on the monodromy matrix is ¶k M s p M,k . MpŽ M,k .. One
must also require that pŽ M,e . s e.
The simplest case is the ‘standard’ Žleft. action
§ h g Ž x . s g Ž x . hy1 Ž 5.3 .
corresponding to pŽ M,h. s hy1 . A family of ‘trivial’ actions is obtained by conjugating
the standard action in the gauge group by an arbitrary element ¶ Žsee Ž4.21..:
§˜ h s ¶y1 § h¶. Ž 5.4 .
In terms of the corresponding G-valued functions we have
s˜Ž M ,h . s p Ž M . hy1 p Ž h P ¶M P hy1 . . Ž 5.5 .
In practice it is useful to consider the infinitesimal version of Ž5.1.. For the
a
parametrization h s e u Ta we have
¶h g Ž x . s g Ž x . y u a Xa g Ž x . q O Ž u 2 . , Ž 5.6 .
where the infinitesimal generators are of the form
Xa g Ž x . s yz b a Ž M . g Ž x . Tb Ž 5.7 .
b
with some monodromy dependent coefficients z a Ž M ., and satisfy the commutation
relations w Xa , Xb x s fa b g Xg . This latter equation can be expressed as a requirement on
the coefficients z b a as
z l a Dlyz v b y z l b Dlyz v a q fk l vz k a z l b q fa b lz v l s 0 . Ž 5.8 .
b b
Clearly the simplest solution of Ž5.8. is the standard one, z a s yd a.
The infinitesimal generators are also useful in studying the question of trivialization.
The infinitesimal form of Ž5.5. is
pTa q z l a Ž Dly q Tl . p s 0 . Ž 5.9 .

It is easy to see that the consistency conditions of this set of partial differential equations
are precisely the Eqs. Ž5.8., but it is not clear if all possible non-linear actions ¶h can be
trivialized in the form Ž5.4. or not.
The next question is which of the above non-linear group actions are Poisson–Lie?
Following w36x, we recall that a Lie group Žor algebra. action characterized by infinitesi-
mal generators Xa is Poisson–Lie, if for any pair of phase space functions F1 , F2 the
PBs satisfy the relations
Xa Ž F1 . , F2 4 q F1 , Xa Ž F2 . 4 y Xa Ž F1 , F2 4 . s yf˜bg a Xb Ž F1 . Xg Ž F2 . ,
Ž 5.10 .
where the pair of structure constants Ž fab g , f˜bg a . together define a Lie bi-algebra. Now
applying this to F1 s g Ž x ., F2 s g Ž y . and parametrizing the exchange r-matrix as
r ab s k klz a k z b l Ž 5.11 .
we find that the infinitesimal action Ž5.7. will be PL if
z a k z b l z s g Dsy k k l q k k s fsg l q k sl fsg k q k f˜k l g s 12

ž / Ž Dq az b g y Dq bz a g . .
Ž 5.12 .
It is well known that all simple PL groups are coboundary. This means that the PL
structure on G is given by the Sklyanin bracket,
m 1
½ h , h5 s k w Rˆ ,h m h x , Ž 5.13 .
for which the structure constants of the induced dual Lie algebra are
1
f˜bg a s Ž R sb fsa g q R gs fsa b . , Ž 5.14 .
k
where Rˆ s R ab Ta m Tb g G n G is some constant, antisymmetric r-matrix. The r-matrix
that occurs here is an arbitrary solution of the Žmodified. classical YB equation,
Rˆ12 , Rˆ 23 q cycl. perm.s yn 2 fˆ Ž 5.15 .

with some constant parameter yn 2 . This parametrization of the right-hand side will
prove useful below. The value n s 0 is also allowed and for real Lie groups n 2 is of
course real. From the classification w34,35x of the solutions we recall that for a compact
simple Lie algebra n must be purely imaginary or zero, while for the maximally
non-compact Žsplit. real forms n is real.
Let us introduce K ab by writing
k ab Ž M . s K ab Ž M . q R ab . Ž 5.16 .
Then Ž5.12. can be reduced to

z a k z b l Ž z s g Dsy K k l q K k s fsg l q K sl fsg k
. s Ž Dq az b g y Dq bz a g . .
1
2
Ž 5.17 .
This depends on how the infinitesimal generator Xa is parametrized in terms of z b a ,
but the explicit reference to the dual structure constants Ž5.14. has disappeared.
From now on we will concentrate on the simplest case corresponding to the standard
action z b a s yd b a . In this case
r ab Ž M . s K ab Ž M . q R ab Ž 5.18 .
and Ž5.17. reduces to
Day K s w K ,T
Ta x , where Ta :s ad Ta , Ž 5.19 .
l l
that is, Ž Ta .s s fsa . Eq. Ž5.19. is the infinitesimal form of
Kˆ Ž hMhy1 . s Ž h m h . Kˆ Ž M . Ž hy1 m hy1 . , Ž 5.20 .
which expresses the equivariance of Kˆ Ž M . s K ab Ž
M .Ta m Tb under the action Ž5.3..
One may also verify directly that the standard left action of G equipped with the PB
Ž5.13. is PL for the exchange r-matrix Ž5.18. if Kˆ Ž M . is equivariant.
So far we have established that the action Ž5.3. of G on Mˇchir is PL if r Ž M . is the
sum of a constant r-matrix R and an equivariant piece K Ž M .. Of course, the exchange
r-matrix Ž5.18. has to be a solution of the dynamical YB equation Ž3.63.. Using Ž5.19.,
Ž3.63. can be rewritten as
bg
yŽ K T aK . q 12 Dq a K bg q cycl. perm.s Ž n 2 y 14 . f a bg Ž 5.21 .
in this special case, where the cyclic permutations are over the upper indices a , b ,g .
The search for solutions of Ž5.21. is made feasible by the observation that any analytic
function of Y Ždefined in Ž3.94.. is equivariant. We show in Appendix B that a solution
of Ž5.21. is given by the analytic function
K s 12 coth Ž Yr2 . y n coth Ž n Y . . Ž 5.22 .
This formula is valid on a domain Gˇ around e g G where the exponential parametriza-
tion is applicable and the power series of the above expression converges. The
derivation of the exchange r-matrix given by Ž5.18., Ž5.22., which is compatible with
the standard action of the PL group G equipped with the PB Ž5.13., is one of our main
results.
We conclude this section with a couple of remarks related to the above r-matrices.
We first note that for n s 0 Ž5.22. is understood as the appropriate limit and therefore
for R s 0 we recover from Ž5.18. the exchange r-matrix r 0 Ž3.95. together with the
classical G-symmetry discussed in Subsection 3.5. We can also have n s 0 in corre-
spondence with any antisymmetric solution R / 0 of the classical YB equation. For
n s 1r2 we get K s 0 and the dynamical r-matrix Ž5.18. then reduces to one of the
constant r-matrices treated in Subsection 3.4. For compact groups all solutions Ž5.22.
are strictly dynamical Žnon-constant., since in this case yn 2 0 0. Finally, we remark
that the existence of a suitable local 2-form r corresponding to the r-matrix Ž5.18.,
Ž5.22. is guaranteed by the solvability of Ž3.36..
We have seen that the classical G-symmetries discussed in Subsection 3.5 and the
special PL symmetries treated in Subsection 3.4 are generated by momentum maps.
Without going into details, we wish to mention that it is possible to show the existence
of a non-abelian momentum map also in the general case of the above PL symmetries.
The momentum map is given by a function mŽ M . on Mˇchir Ždepending on the
monodromy matrix only., which takes its values in the dual PL group G ) in such a way
that for all phase space functions F
Xa F s Ž my1 m, F 4 . a , Ž 5.23 .
where the Ž .a component on the right-hand side is evaluated in the dual Lie-algebra,
G ) . Moreover, the G )-valued momentum map satisfies the PBs
m
½ m , m5 s h )
Ž m. m m m , Ž 5.24 .
where the Poisson tensor h ) Ž m. g G ) n G ) is chosen so that Ž5.24. defines the

Poisson structure of the dual PL group. ŽFor an explanation of these notions, see e.g.
Ref. w36x..
It is in principle possible to use the momentum map construction also backwards. If
there is a G ) -valued function m on the phase space satisfying Ž5.24., then using Ž5.23.
to define an infinitesimal generator Xa one obtains that
v the infinitesimal generators represent the Lie algebra, w Xa , Xb x s fa b g Xg ,

v the Lie algebra action is PL, i.e. Ž5.10. holds.
Let now suppose that a compact simple Lie algebra G acts on a phase space as a
classical symmetry generated by a G ) -valued, equivariant Ž‘abelian’. momentum map.
In this situation one can always define also an infinitesimal PL action of the group G
equipped with the so-called standard PL structure. This is a consequence of the fact w40x
that there exists a diffeomorphism between G ) and G ) that converts the natural linear
Poisson structure on G ) into the standard PL structure on G ) . Applying this to the
classical G-symmetry studied in Subsection 3.5, we can thus find a map,
G ) 2 G˜¨ mŽ G˜ . g G , )
Ž 5.25 .
where G˜ : Mˇ ™ G is given by Ž3.96., such that mŽ G˜ . satisfies Ž5.24. with respect to
chir
)
)
the standard PL structure. The resulting G -valued momentum map then generates a PL
action on the phase space Mˇchir as outlined above.
This somewhat surprising construction is not specific to the chiral WZNW phase
space, since it can be used whenever one has a classical G-symmetry based on a
compact simple Lie algebra. When applying it to the chiral WZNW phase space, the Lie
algebra action Ž5.23. constructed using the momentum map Ž5.25. will be different from
the standard one Ž5.3.. It is an interesting question whether this Lie algebra action can be
gauge transformed to the standard form and, if the answer is positive, to find the
corresponding gauge transform of the r-matrix r 0 Ž3.95.. We wish to discuss this in a
future publication.
6. Interpretation in terms of Poisson–Lie groupoids
The CDYB equation Ž4.32. can be regarded as the guarantee of the Jacobi identity in
a PL groupoid w25x. Below we show that Eq. Ž3.63. admits an analogous interpretation.
For this, we introduce a family of PL groupoids in such a way that a subfamily of these
is naturally associated with the possible PBs on the chiral WZNW phase space.
Remarkably, these groupoids are finite-dimensional Poisson manifolds that encode
practically all information about the infinite-dimensional chiral WZNW PBs.
Roughly speaking, a groupoid is a set, say P, endowed with a ‘partial multiplication’
that behaves similarly to a group multiplication in the cases when it can be performed.
To understand the following construction one does not need to know details of the
notion of a groupoid Žsee, e.g., Ref. w41x., since we shall only use the most trivial
example of such a structure, for which
PsS=G=Ss Ž M F, g , M I .4 , Ž 6.1 .
where G is a group and S is some set. The partial multiplication is defined for those
triples Ž M F, g, M I . and Ž M F, g, M I . for which M I s M F, and the product is
Ž M F , g , M I . Ž M F , g , M I . :s Ž M F , gg , M I . for M I s M F . Ž 6.2 .
In other words, the graph of the partial multiplication is the subset of
P=P=Ps Ž M F, g , M I .4 = ½Ž M F
,g,MI . =5 ½ ž Mˆ F
, gˆ , Mˆ I /5 Ž 6.3 .
defined by the constraints
M I sM F, Mˆ F s M F , Mˆ I s M I , gˆ s gg , Ž 6.4 .
where the hatted triple encodes the components of the product. A PL groupoid w42x P is
a groupoid and a Poisson manifold in such a way that the graph of the partial
multiplication is a coisotropic submanifold of P = P = Py, where Py denotes the
manifold P endowed with the opposite of the PB on P. In other words, this means that
the constraints that define the graph are first class. This definition reduces to that of a
PL group in the particular case for which the set S in Ž6.1. consists of a single point.
In the interpretation of Ž3.70. given in w25x the groupoid P is of the form above with
S taken to be a domain in the dual of a Cartan subalgebra of a simple Lie group G. By
thinking about a generic monodromy matrix, we now take P to be
P s Gˇ = G = G,
ˇ Ž 6.5 .
where Gˇ is some open domain in G. On this P, we postulate a PB , 4P defined, by
using the usual tensorial notation, as follows:
k g 1 , g 2 4 P s g 1 g 2 rˆ Ž M I . y rˆ Ž M F . g 1 g 2 ,
k g 1 , M2I 4 P s g 1 M2I Qˆ Ž M I . ,
k g 1 , M2F 4 P s M2FQˆ Ž M F . g 1 ,
k M1I , M2I 4 P s M1I M2I Dˆ Ž M I . ,
k M1F , M2F 4 P s yM1F M2FDˆ Ž M F . ,
k M1I , M2F 4 P s 0. Ž 6.6 .
Here k is an arbitrary constant included for comparison purposes. The ‘structure

ˆ Qˆ , Dˆ are G m G valued functions on G;
functions’ r, ˇ in components
rˆ Ž M . s r ab Ž M . Ta m Tb , Qˆ Ž M . s Q ab Ž M . Ta m Tb ,
Dˆ Ž M . s D ab Ž M . Ta m Tb . Ž 6.7 .
It is quite easy to verify that a PB given by the ansatz Ž6.6. always yields a PL groupoid,
since the constraints in Ž6.4. will be first class for any choice of the structure functions.
Of course, the structure functions must satisfy a system of equations in order for the
above ansatz to define a PB. The antisymmetry of the PB is ensured by
rˆ s yrˆ21 Ž rˆ21 :s r ab Tb m Ta . , Dˆ s yDˆ21 , Ž 6.8 .
while the Jacobi identity is, in fact, equivalent to the following system of equations:
rˆ12 ,rˆ13 q Qa b T1a R b rˆ23 q cycl. perm.s m fˆ, m s constant, Ž 6.9 .
Dˆ12 , Dˆ13 q Da b T1a R bDˆ23 q cycl. perm.s 0, Ž 6.10 .
rˆ12 ,Qˆ 13 q Qˆ 23 q Qˆ 13 ,Qˆ 23 q Da b T3a R b rˆ12
q Qab T1a R bQˆ 23 y T2a R bQˆ 13 s0,

ž / Ž 6.11 .
Qˆ 12 q Qˆ 13 , Dˆ23 q Qˆ 12 ,Qˆ 13 q Qa b T1a R bDˆ23
q Dab T3a R bQˆ 12 y T2a R bQˆ 13 s0.

ž / Ž 6.12 .
Observe that the left-hand side of Ž6.9. is of the same form as that of Ž3.63., but in the
groupoid context on the right-hand side we have an arbitrary constant m. The
derivation of the above equations from the various instances of the Jacobi identity is not
difficult. What is somewhat miraculous is that one does not obtain more equations than
these. This is actually ensured by our choice of the relationship between the PBs that
involve M I and those that involve M F. As an illustration, let us explain how Ž6.9. is
derived. By evaluating
g 1 , g 2 4 P , g 3 4 P q cycl. perm.s 0, Ž 6.13 .
one obtains that this is equivalent to
g 1 g 2 g 3 rˆ12 ,rˆ13 q Qa b T1a R b rˆ23 q cycl. perm. Ž M I .
ž /
s ž rˆ12 ,rˆ13 q Qa b T1a R b rˆ23 q cycl. perm. Ž M F . g 1 g 2 g 3 .
/ Ž 6.14 .
This holds if and only if the expression in the parenthesis is a constant, Ad-invariant
element of n 3 Ž G ., and m fˆ is the only such element for a simple Lie algebra G .
We have seen that the chiral WZNW PBs are encoded by Eqs. Ž3.62., Ž4.16. and
Ž4.17., where Qˆ and Dˆ are defined by Ž3.52. and Ž3.54. respectively in terms of a
solution rˆ of Ž3.63.. Now our point is the following: A PL groupoid can be naturally
associated with any Poisson structure on the chiral WZNW phase space by taking the
ˆ Qˆ , Dˆ that arises in the WZNW model to be the structure functions of a PL
triple r,
groupoid according to (6.6).
It can be checked that the Jacobi identities of the PL groupoid Ž6.9. – Ž6.12. are
ˆ Qˆ , Dˆ that arises in the WZNW model. This actually follows
satisfied for any triple r,
without any computation since, indeed, the Jacobi identities of the chiral WZNW PBs in
Ž3.62., Ž4.16., Ž4.17. lead to the same equations, with m s y 14 , and they are satisfied
r
since they follow from the symplectic form kV chir .
Among the ‘chiral WZNW PL groupoids’ described above there are those special
cases for which Kˆ s rˆ y Rˆ satisfies the equivariance condition Ž5.20. in relation with
some constant r-matrix Rˆ subject to Ž5.15.. In these cases, we equip the group G s h4
with the Sklyanin bracket opposite to that in Ž5.13.,
m 1
½ h , h s w h m h, Rˆ x ,
5 Ž 6.15 .
k
¨ Ž M , gh,h M h. g P
and consider its commuting right and left actions on P given respectively by the maps
P = G 2 Ž Ž M F , g , M I . ,h . F y1 I
Ž 6.16 .
G = P 2 Ž h, Ž M , g , M . . ¨ Ž hM h ,hg , M . g P .
and
F I F y1 I
Ž 6.17 .
Then it is not difficult to verify that these are both Poisson maps, i.e. they define two PL
actions of the PL group G Žwith Ž6.15.. on the PL groupoid P. In the final analysis, this
is a consequence of the fact that, as explained in Section 5, in the present situation we
have a PL action of G on the chiral WZNW phase space whose Poisson structure is
encoded by Ž P, , 4P .. Here Gˇ ; G must be Ad -invariant, see footnote 6.
In w25x PL groupoids are associated with arbitrary subalgebras K ; G , although the
corresponding dynamical r-matrices are described only if K is a Cartan subalgebra.
The K s G special case of their groupoids is in fact equivalent to our PL groupoid
whose structure function is the r-matrix in Ž3.95.. Their PL groupoids are different from
ours in general.
7. Conclusion
In this paper we explored the Poisson structures on the chiral WZNW phase space of
group valued quasiperiodic fields with generic monodromy. We have shown that the
possible PBs are defined by the exchange r-matrices that are solutions of Ž3.63.. This
equation can be viewed as an analogue of the celebrated CDYB equation Ž3.70., since
ˆ for chiral WZNW fields with diagonal monodromy. An
the latter plays a similar role
analysis of chiral WZNW Bloch waves and their classical Wakimoto realizations in the
spirit of the present paper is contained in w24x.
We have given an interpretation of our dynamical YB equation Ž3.63. in terms of a
family of PL groupoids, whose further study may be fruitful. In this respect, the most
interesting open questions appear to be to quantize these PL groupoids and to find
applications for them outside the chiral WZNW context. It is known that Eq. Ž3.70.
admits interesting applications in the field of integrable systems w31x.
We also investigated the PL symmetries of the exchange algebra Ž3.62.. We have
found that for any PL structure on the WZNW group G there is a corresponding choice
of the exchange r-matrix such that the standard gauge action of G on the chiral fields
becomes a PL action. It would be desirable to understand if this result has any analogue
at the level of the quantized Žchiral. WZNW model.
Acknowledgements
This investigation was supported in part by the Hungarian National Science Fund
ŽOTKA. under T019917, T030099, T025120, M028418 and by the Ministry of Educa-
tion under FKFP 0178r1999, FKFP 0596r1999.
Appendix A. Exchange r-matrices for SU(2)
In this appendix we present an explicit, local formula for the most general exchange
r-matrix on the simplest compact Lie group G s SUŽ2.. The formula ŽA.15. below is
valid in a neighbourhood of the unit element. It illustrates some general results obtained
in Section 3, and it may prove useful in a future study of the question whether globally
defined exchange r-matrices exist for SUŽ2. or not.
As a basis for the Lie algebra suŽ2., we choose the generators T a :s 21i sa , where the
sa Ž a s 1,2,3. are the usual Pauli matrices. We parametrize the matrices r a b Ž M . and
q ab Ž M . that appear in Ž3.36. in terms of 3-component vectors as
r a b s e a b cR c , q a b s e a b c Qc , Ž A.1 .
where e a b c is the totally antisymmetric tensor for which e 123 s 1. Furthermore, we
identify the SUŽ2. group manifold with S 3 ; R 4 by writing M g SUŽ2. according to
M s x 0 s 0 q ix a sa , x 02 q x a x a s 1, Ž A.2 .
whereby x 0 , x a define smooth functions on SUŽ2. Ž s 0 is the 2 = 2 unit matrix.. It is
then a matter of straightforward calculation to translate Eq. Ž3.37. into the formula
R a s 14 Dy1 Ž x a q 4 x 0 Q a q 4e a b c x b Q c . , D:s x 0 y 4 x a Q a , Ž A.3 .

which is valid on a neighbourhood of the unit element where D / 0. By assumption, on
this neighbourhood the Q a are smooth functions subject to
3n
d r Ž M . s 16 tr Ž My1 dM .
for r Ž M . s 21 q a b Ž M . tr Ž Ta My1 dM . n tr Ž Tb My1 dM . . Ž A.4 .

As discussed in Subsection 3.3, ŽA.4. implies that R a defined by ŽA.1., ŽA.3. yields a
solution of the dynamical YB equation Ž3.63., which for SUŽ2. can actually be written
in the form
2 R a R a q 12 D aq R a q e a b c R c Dby R a s y 18 . Ž A.5 .
Observe that we cannot have a constant solution since the R a must be real. It is also
worth noting that locally we have the inverse of ŽA.3. given by
Q a s 14 D˜ y1 Ž yx a q 4 x 0 R a q 4e a b c x b R c . , ˜ x0 q 4 x a R a .
D:s Ž A.6 .
This formula defines via ŽA.1. a solution of ŽA.4. out of any solution of ŽA.5..
We shall now derive the general local solution of ŽA.5. by making explicit the
general local solution Ž3.93. of ŽA.4. that we have obtained in Subsection 3.5 for any
group. For this we need the exponential parametrization of SUŽ2.,
a
M s e 2 p G s e 2 p G aT with < G < 2 s Ga Ga - 1, Ž A.7 .
which covers the domain SUŽ2. _ ys 0 4 . Upon comparison with ŽA.2., we get
Ga
x 0 s cos Ž p < G < . , xa sy sin Ž p < G < . . Ž A.8 .
<G <
We shall also use the following expressions for the powers of the operator ad G . For the
odd powers, we have
2 nq1 n
Ž ad G . s Ž y1 . < G < 2 n Ž ad G . , n 0 0,
Ž ad G . Ž T a . s w G ,T a x s e a b c Tb Gc . Ž A.9 .
For the even powers, defining the matrix of Žad G . n by Žad G . n ŽT b . s wŽad G . n x a b T a,
we have
2n n Ga G b
Ž ad G . bs Ž y1. < G < 2 n d a b y
ž / , n 0 1. Ž A.10 .
a < G <2
Using these relations, we can rewrite the formulae Ž3.94. and Ž3.95. as follows:
2p < G < y sin Ž 2p < G < .
q0a b s e a b c Q cŽ0. with Q cŽ0. s Gc Ž A.11 .
8 < G <sin2 Ž p < G < .
and
1 Gc 1
r 0a b s e a b c R Ž0.
c with R Ž0.
c s
4 <G < ž cot Ž p < G < . y
p <G < / . Ž A.12 .
One may check that Eqs. ŽA.3. – ŽA.6. hold for these expressions, which represent
smooth functions on SUŽ2. _ ys 0 4 .
To obtain the most general 2-form r on SUŽ2. _ ys 0 4 that satisfies ŽA.4., we have
to add an arbitrary closed 2-form to the 2-form, r 0 , that corresponds to the matrix q0ab.
In fact, the result can be written as
sin Ž 2p < G < . y 2p < G <
r Ž G . s d Ga n d G b e cb a Gc
ž 2< G <3
q Uc Ž G . ,
/ Ž A.13 .
where UaŽ G . is a smooth ‘ vector function’ in the interior of the unit ball, < G < - 1,
which is divergence free, i.e.
3 E Ua Ž G .
Ý s 0. Ž A.14 .
as1 EGa
We then have to rewrite this 2-form in the manner indicated by the second parts of ŽA.1.
and ŽA.4.. By means of ŽA.3., this will provide us with the general local solution of
ŽA.5.. By performing the necessary Žrather tedious. calculations, in this way we obtain
the following formula:
Ga 1 1
Ra s < <
4 cot Ž p G . y
<G < < G < Ž 4p y 2 G P U .
Ga 1
q Ž G P U . y < G < Ua , Ž A.15 .
<G < Ž 4p y 2 G P U . 2 p < G <
where G P U s GaUa . This expression is valid on the open subset of SUŽ2. that excludes
ys 0 and the points where G P U s 2p . In particular, R a is smooth in a neighbourhood
of the unit element, for which G s 0, since UaŽ G . is smooth there by assumption. One
can also verify explicitly that on its domain of validity R a solves the dynamical YB
equation ŽA.5. for any divergence free UaŽ G .. For this verification, one needs to spell
out ŽA.5. more explicitly. For instance, if one uses the x a in ŽA.2. as coordinates around
s 0 g SUŽ2., then ŽA.5. becomes
(1 y x b xb E Ra E Ra E Ra
2 Ra Ra y q2 xa Rb s y 18 .
y2 Rb xb Ž A.16 .
2 E xa E xb E xa
In summary, we have derived the form of the most general exchange r-matrix in a
neighbourhood of the unit element of SUŽ2.. The solution ŽA.15. may in general
develop singularities away from the unit element, and it is an open question if globally
smooth solutions of ŽA.5. exist on SUŽ2. or not.
Appendix B. Analytic solution of the dynamical YB equation

In this appendix we show that the equivariant analytic function Ž5.22. is a solution of
the dynamical YB equation Ž5.21..
We will use the power series expansion of the coth function:
1 `
coth z s q Ý a r z 2 rq1 . Ž B.1 .
z rs0
Here the coefficients a r can be expressed in terms of the Bernoulli numbers w39x. They
can also be computed using the recursion relation
1 my1
a 0 s 13 , am s y Ý aa , m s 1,2, . . . Ž B.2 .
2 m q 3 rs0 r my1yr
Using the properties of the operator Dqa Ž3.7. and the definition of K Ž5.22. the
first two terms of Ž5.21. can be expanded as
` 1
bg
Ž K T a K . s Ý a r a s 2 rq2 y n 2 rq2
ž /
r , ss0 2
1 bg
= 2 sq2 y n 2 sq2 Ž Y 2 sq1T a Y 2 rq1 .
ž / Ž B.3 .
2
and
1 qa bg
` 1 2r Y Y al bg
2 D K s Ý a r 2 rq2 y n
rs0
ž
2
2 rq2
Ý 2 coth 2 Ž Y sTl Y 2 rys . .
/ ž
ss0
/
Ž B.4 .
Here, in writing the second equality we exploited that in the parametrization of M

introduced in Subsection 3.5 one readily obtains by writing Y s X aTa that
ab
Y
L aŽX b . s
ž 1 y exp Ž yY
Y.
exp Ž yY
Y.
/ ,
ab
Y
a
R ŽX .s b
ž exp Ž Y . y 1
exp Ž Y .
/ . Ž B.5 .
It is clear that all terms in Ž5.21. are built from powers of Y and structure constants. It
will prove useful to contract all the indices with Lie algebra generators and thus
reformulate Ž5.21. as an equation in the triple tensor product of the Lie algebra. We
introduce the notation
ak bl gs
² k ,l,m: :s Ž Y k . Ž Y l. Ž Y m. fkls Ta m Tb m Tg . Ž B.6 .
Cyclic permutation of the indices now corresponds to cyclic permutation of the tensor
factors and we also introduce the symbol
w k ,l,m x :s ² k ,l,m: q ² l,m,k : q ² m,k ,l : . Ž B.7 .
We expand both sides of Ž5.21. in powers of n and every coefficient of n also in
powers of Y . In our notation this latter expansion corresponds to putting together all
terms with a fixed total N s k q l q m, and Eq. Ž5.21. requires separately the equality of
all such terms on the two sides.
We start with the n 0 terms. We find that the N s 0 piece is satisfied identically,
while for N s 2 m q 2 Ž m s 0,1, . . . . we get
2 mq2
a mq 1 Ý Ž y1. sq 1 w0, s,2 m q 2 y s x y Ý a r a s w 0,2 s q 1,2 r q 1 x
ss0 rqssm
2r
q Ý ar ak Ý Ž y1. sq 1 w2 k q 2, s,2 r y s x s 0 . Ž B.8 .
rqksm ss0
Similarly the N s 0 piece of the n 2 term vanishes identically, while for N s 2 r q 2

Ž r s 0,1, . . . . we obtain
w 2 r q 2,0,0 x q w 0,1,2 r q 1 x q w 0,2 r q 1,1 x s 0 . Ž B.9 .
2 mq4
Finally, from the n terms Žfor m s 0,1, . . . . we get contributions with N s Ž2 m
q 2.th powers of Y ,
2 mq2
s
a mq 1 Ý Ž y1. w0, s,2 m q 2 y s x y Ý a r a s w 0,2 s q 1,2 r q 1 x s 0,
ss0 rqssm
Ž B.10 .
and also terms with N s Ž2 m q 4 q 2 r .th powers of Y Žfor r s 0,1, . . . .,
w 0,2 m q 3,2 r q 1 x q w 0,2 r q 1,2 m q 3 x

2 mq2
s
q Ý Ž y1. w2 r q 2, s,2 m q 2 y s x s 0 . Ž B.11 .
ss0
Before proceeding we note that using the Jacobi identity for the structure constants
we can write down the following identity:
² k q 1,l,m: q ² k ,l q 1,m: q ² k ,l,m q 1: s 0 . Ž B.12 .
Now it is easy to see that ŽB.9. is a special case and ŽB.11. is a simple consequence
of the above identity. In fact, to prove ŽB.11. we group w0,2 m q 3,2 r q 1x with the first
Žw0,2 r q 1,2 m q 3x with the last. m q 1 terms of the sum:
m
s
Z1 s w 0,2 m q 3,2 r q 1 x q Ý Ž y1. w2 r q 2, s,2 m q 2 y s x ,
ss0
m
Z2 s w 0,2 r q 1,2 m q 3 x q Ý Ž y1. j w2 r q 2,2 m q 2 y j, j x ,
js0
then in both groups, for odd s Ž j . we use ŽB.12. to write

s
Ž y1. w 2 r q 2, s,2 m q 2 y s x
s w 2 r q 1, s q 1,2 m q 2 y s x q w 2 r q 1, s,2 m q 3 y s x ,
j
Ž y1. w 2 r q 2,2 m q 2 y j, j x
s w 2 r q 1,2 m q 3 y j, j x q w 2 r q 1,2 m q 2 y j, j q 1 x .
In this form, as a consequence of ŽB.12., both Z1 and Z2 cancel ‘telescopically’ almost

completely; for odd m they give
Z1 ¨ w2 r q 1,m q 1,m q 2x , Z2 ¨ w2 r q 1,m q 2,m q 1x , Ž B.13 .

while for even m we get
Z1¨ w2 r q 2,m,m q 2x q w2 r q 1,m,m q 3x ,

Z ¨ w 2 r q 2,m q 2,m x q w 2 r q 1,m q 3,m x .
2 Ž B.14 .
One readily verifies, using again ŽB.12., that these remaining terms give zero in both
cases with the ‘central’ element of the sum Žy1. mq 1w2 r q 2,m q 1,m q 1x.
As for ŽB.8., it should be perfectly possible to prove it using the properties of the
coefficients a r , but in the present context there is no need to prove it independently
since we know that the n s 0 case coincides with the solution Ž3.95. and therefore it
must satisfy Ž3.63..
Thus we are left with ŽB.10.. Using ŽB.8., we can rewrite it as
2r
s
2 Ý a r a k w 0,2 r q 1,2 k q 1 x q Ý ar ak Ý Ž y1. w2 k q 2, s,2 r y s x s 0
rqksm rqksm ss0
Ž B.15 .
and in this form we see that it trivially follows from ŽB.11. and the index symmetry of
ak ar .
References
w1x E. Witten, Commun. Math. Phys. 92 Ž1984. 455.

w2x P. Di Francesco, P. Mathieu, D. Senechal,
´ ´ Conformal Field Theory ŽSpringer, Berlin, 1996..
w3x P. Etingof, I. Frenkel, A. Kirillov Jr., Lectures on Representation Theory and Knizhnik–Zamolodchikov
Equations ŽAMS, 1998..
w4x O. Babelon, Phys. Lett. B 215 Ž1988. 523.
w5x B. Blok, Phys. Lett. B 233 Ž1989. 359.
w6x L. Faddeev, Commun. Math. Phys. 132 Ž1990. 131.
w7x A. Alekseev, S. Shatashvili, Commun. Math. Phys. 133 Ž1990. 353.
w8x J. Balog, L. Da̧browski, L. Feher,
´ Phys. Lett. B 244 Ž1990. 227.
w9x M. Chu, P. Goddard, I. Halliday, D. Olive, A. Schwimmer, Phys. Lett. B 266 Ž1991. 71.
w10x O. Babelon, F. Toppan, L. Bonora, Commun. Math. Phys. 140 Ž1991. 93.
w11x K. Gawȩdzki, Commun. Math. Phys. 139 Ž1991. 201.
w12x T. Itoh, Y. Yamada, Prog. Theor. Phys. 85 Ž1991. 751.
w13x V.A. Fateev, S.L. Lukyanov, Int. J. Mod. Phys. A 7 Ž1992. 853.
w14x F. Falceto, K. Gawȩdzki, J. Geom. Phys. 11 Ž1993. 251.
w15x A. Alekseev, I.T. Todorov, Nucl. Phys. B 421 Ž1994. 413.
w16x C. Gomez,
´ M. Ruiz-Altaba, G. Sierra, Quantum Groups in Two-Dimensional Physics ŽCambridge Univ.
Press, Cambridge, 1996., and references therein.
w17x M. Chu, P. Goddard, Phys. Lett. B 337 Ž1994. 285.
w18x M. Chu, P. Goddard, Nucl. Phys. B 445 Ž1995. 145.
w19x P. Furlan, L.K. Hadjiivanov, I.T. Todorov, Nucl. Phys. B 474 Ž1996. 497.
w20x P. Furlan, L.K. Hadjiivanov, I.T. Todorov, Int. J. Mod. Phys. A 12 Ž1997. 23.
w21x L. Caneschi, M. Lysiansky, Nucl. Phys. B 505 Ž1997. 701.
w22x V.G. Drinfeld, in: Proc. Int. Congr. Math. Berkeley, 1986, pp. 798–820.
w23x P. Etingof, O. Schiffmann, Lectures on the dynamical Yang–Baxter equations, math. QAr9908064.
w24x J. Balog, L. Feher,
´ L. Palla, Classical Wakimoto realizations of chiral WZNW Bloch waves, hep-
thr9910112.
w25x P. Etingof, A. Varchenko, Commun. Math. Phys. 192 Ž1998. 77.
w26x J. Balog, L. Feher,
´ L. Palla, Phys. Lett. B 463 Ž1999. 83.
w27x J.-L. Gervais, A. Neveu, Nucl. Phys. B 238 Ž1984. 125.
w28x E. Cremmer, J.-L. Gervais, Commun. Math. Phys. 134 Ž1990. 619.
w29x G. Felder, C. Wieczerkowski, Commun. Math. Phys. 176 Ž1996. 133.
w30x G. Felder, in: Proc. Int. Congr. Math. Zurich,
¨ 1994, pp. 1247–1255.
w31x J. Avan, O. Babelon, E. Billey, Commun. Math. Phys. 178 Ž1996. 281.
w32x Z.J. Liu, P. Xu, Dirac structures and dynamical r-matrices, math. DGr9903119.
w33x J.H. Lu, Classical dynamical r-matrices and homogeneous Poisson structures on Gr H and K r T,
mathr9909004.
w34x A.A. Belavin, V.G. Drinfeld, Funct. Anal. Appl. 16 Ž1982. 159.
w35x M. Cahen, S. Gutt, J. Rawnsley, Contemp. Math. 179 Ž1994. 1.
w36x O. Babelon, D. Bernard, Commun. Math. Phys. 149 Ž1992. 279.
w37x E. Sklyanin, J. Sov. Math. 19 Ž1982. 1546.
w38x V.G. Drinfeld, Sov. Math. Dokl. 27 Ž1983. 68.
w39x I.S. Gradshteyn, I.M. Ryzhik, Tables of Integrals, Series and Products, ed. A. Jeffrey, Fifth Edition
ŽAcademic Press, 1996..
w40x V.L. Ginzburg, A. Weinstein, J. Am. Math. Soc. 5 Ž1992. 445.
w41x K. Mackanzie, Lie Groupoids and Lie Algebroids in Differential Geometry ŽCambridge Univ. Press,
Cambridge, 1987..
w42x A. Weinstein, J. Math. Soc. Jpn. 4 Ž1988. 705.
Symmetry breaking boundaries

II. More structures; Examples
¨
Jurgen Fuchs, Christoph Schweigert
¨ Theoretische Physik, ETH Honggerberg,
Institut fur ¨ ¨
CH-8093 Zurich, Switzerland
Received 24 August 1999; accepted 20 October 1999
Abstract
Various structural properties of the space of symmetry breaking boundary conditions that
preserve an orbifold subalgebra are established. To each such boundary condition we associate its
automorphism type. We show that correlation functions in the presence of such boundary
conditions are expressible in terms of twisted boundary blocks which obey twisted Ward
identities. The subset of boundary conditions that share the same automorphism type is controlled
by a classifying algebra, whose structure constants are shown to be traces on spaces of chiral
blocks. T-duality on boundary conditions is not a one-to-one map in general. These structures are
illustrated in a number of examples. Several applications, including the construction of non-BPS
boundary conditions in string theory, are exhibited. q 2000 Elsevier Science B.V. All rights
reserved.
PACS: 11.25.Hf; 11.25.-w

Keywords: Boundary conditions; Conformal field theory; Simple currents; Symmetry breaking
1. Introduction
The study of conformally invariant boundary conditions in two-dimensional confor-

mal field theory is of considerable interest both for applications in condensed matter
physics and in string theory. In such applications typically only the Žsuper-.conformal
symmetry needs to be preserved by the boundaries, while the rest of the chiral bulk
symmetries A may be broken.
In w1x we have studied conformally invariant boundary conditions for an arbitrary
conformal field theory that preserve a Žconsistent. subalgebra A of A, such that
AsAG
is the subalgebra that is fixed under a finite abelian group G of automorphisms of A.
We have shown that such boundary conditions are governed by a classifying algebra
C Ž A ., in the sense that the reflection coefficients w2,3x – the data that characterize the
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 6 9 - 0
boundary condition– are precisely the one-dimensional irreducible representations of
C Ž A ..
This paper is a continuation to w1x. A very brief summary of the pertinent results of
w1x is as follows. The algebra C Ž A . is a commutative associative semisimple algebra.
Thus its regular representation is fully reducible, and the structure constants are
expressible through the corresponding diagonalizing matrix S˜ by an analogue of the
Verlinde formula. This matrix S, ˜ in turn, can be expressed in terms of various quantities
that are already known from the chiral conformal field theory associated to A Žsee
formula ŽA.6...
The conformal field theory with chiral algebra A can be obtained from the A-theory
as an orbifold by the group G; conversely, the original A-theory is recovered from the
A-theory as an integer spin simple current extension, with the group G of simple
currents being the character group of the orbifold group, G s G ) . Denoting the labels
for the primary fields of the A-theory by l and those for the primary A-fields by l, a
natural basis of C Ž A . is labeled by pairs Ž l, w ., where l refers to a A-primary in the
untwisted sector and w g Sl) is a group character, while the boundary conditions are
labeled by pairs w r , cˆr x consisting of an arbitrary primary label r of the A-theory and a
character cˆr g Ul) . Here the stabilizer Sl and the untwisted stabilizer Ul are sub-
groups of the simple current group G s G ) ; Sl consists of all simple currents in G
that leave l fixed, and Ul is the subgroup of Sl on which a certain alternating
bi-homomorphism Fl, which can be defined through the modular properties of one-point
chiral blocks on the torus, is trivial. For the precise meaning of these terms we refer to
w1x Žin particular appendix A. and to w4x. Quite generally, the number of basis elements
of C Ž A . – or equivalently, the number of independent boundary blocks, i.e. chiral
blocks for one-point correlation functions of bulk fields on the disk – and the number of
boundary conditions have to be equal. It is a rather non-trivial result of the analysis that
this equality indeed holds for the two sets of labels, so that in particular the matrix
S˜' S˜Ž l , w .,w r , cˆr x is a square matrix.
The results established in w1x clearly demonstrate an unexpectedly nice behavior of
the space of conformally invariant boundary conditions. In the present paper we show
that indeed this space is endowed with even more structure. Section 2 deals with the
implementation of the orbifold group on the representation spaces of the chiral algebra
A; this discussion does not yet involve boundary conditions at all, but the results will be
needed in the sequel. Sections 3–8 are devoted to the discussion of several additional
generic features of symmetry breaking boundary conditions. We start in Section 3 by
analyzing the notion of automorphism type, which in the present setting is a derived
concept that arises as a direct consequence of the general structure and does not need be
introduced by hand. Next we show that boundary conditions of definite automorphism
type can be naturally formulated with the help of twisted boundary blocks, which satisfy
twisted Ward identities ŽSection 4.. Furthermore, to the boundary conditions of fixed
automorphism type one can associate their own classifying algebra, which is an invariant
subalgebra of the total classifying algebra C Ž A . and whose structure constants can be
understood in terms of suitable traces on chiral blocks; this is done in Section 5. It is
also shown that the individual classifying algebra for automorphism type g only
depends on the automorphism g, but not on the specific orbifold subalgebra A, i.e. not
on the group G containing g.
Afterwards, in Section 6, we turn to a detailed study of the dependence of the
classifying algebra on the chosen torus partition function, which leads to the concept of
T-duality of boundary conditions. First we show that only the ‘difference’ between an
automorphism characterizing the torus partition function and the automorphism type of a
boundary condition is observable, and then we discuss aspects of T-duality among
Žfamilies of. boundary conditions for fixed choice of the torus partition function. We
emphasize that T-duality on boundary conditions is not a one-to-one map, in general. In
Section 7 we establish an action of the orbifold group G on the space of boundary
conditions, which implies a certain ‘homogeneity’ among the boundary conditions for
fixed G-orbit w r x. Finally we introduce in Section 8 the concept of a uniÕersal
classifying algebra, which governs all conformally invariant boundary conditions at the
same time, and discuss the possibility to obtain this algebra by a suitable projective
limit.
In Sections 9 and 10 we turn our attention to a specific class of boundary conditions
to which we refer as involutary, namely those where the orbifold group is Z 2 . We first
address some general features and then, in Section 10, analyze several classes of
examples that are of particular interest. Afterwards, in Section 11 we provide several
classes of examples with more complicated orbifold groups, in which for instance
untwisted stabilizer subgroups occur that are proper subgroups of the full stabilizers.
Finally, some of the pertinent formulae from Ref. w1x that will be needed in the sequel
are collected in Appendix A.
2. The action of the orbifold group on A-modules
The elements g of the orbifold group G are automorphisms of the chiral algebra A.
In the sequel we will have to deal with various subgroups of G and their properties. We
first observe that every automorphism g of A can be implemented on the physical
A-modules Hl by maps
Qg ' QgŽ l. : Hl ™ Hg wl , Ž 2.1 .
which obey the g-twisted intertwining property
Qg Y s g Ž Y . Qg for all Y g A , Ž 2.2 .
and the maps Qg are defined by this property up to a scalar multiple. ŽFor a concrete
realization of these maps in WZW theories, see Ref. w5x.. In general, such an implemen-
tation QgŽ l. maps a given space Hl to some other A-module Hg wl , thereby organizing
the primary fields of the A-theory into orbits, much like the simple current group
G ( G ) organizes w6x the A-primaries into orbits.
To every A-primary l we can associate the stabilizer
Sl :s g g G < g wl s l4 , Ž 2.3 .
which is a subgroup of G whose elements constitute endomorphisms Hl ™ Hl .
Stabilizers of A-primaries on the same G-orbit are identical Žin the more general case of
non-abelian G, they are conjugate subgroups.; the vacuum has a maximal stabilizer,
SV s G. Also, A-modules on the same G-orbit are isomorphic as A-modules. The
endomorphisms Qg for g g Sl provide us with an action of Sl on Hl which is, in
general, only projective, and hence determines a two-cocycle El of Sl or, more
precisely Žin agreement with the fact that the maps Qg are defined only up to
normalization., the cohomology class of El. We denote by Ul the subgroup of Sl that
corresponds to the regular elements of the associated twisted group algebra C El Sl , i.e.
Ul :s g g Sl < El Ž g , g X . s El Ž g X , g . for all g X g Sl 4 . Ž 2.4 .
The scalar factors in the definition of the implementers Qg can be chosen in such a
way that the maps Qg with g g Ul provide us with a honest representation of Ul on
Hl. It follows that the A-modules Hl can be decomposed as
Hl ( [ VJˆ m Hl ,Jˆ , Ž 2.5 .
ĴgUl)
where the spaces VJˆ are projective Sl-modules and the spaces Hl,Jˆ are A-modules. We
make the mild technical assumption that all these modules VJˆ and Hl ,Jˆ are irreducible;
this holds true in all known examples, and is rigorously proven for the vacuum V w7,8x
as well as w9x for other A-modules, including twisted sectors. In the case of the vacuum,
no multiplicities appear in this decomposition; thus the action of G is genuine and we
have UV s SV s G.
Since by construction g leaves the subalgebra A of A fixed, the maps Qg are
ordinary intertwiners for A; hence in the decomposition Ž2.5. they act solely on the
degeneracy space VĴ . Moreover, by the general properties of twisted group algebras
Žcompare Appendix B of w1x., all the spaces VĴ have the same dimension < Sl <r< Ul < , and
(
the basis of the twisted group algebra C El Sl can be chosen in such a way that every
g g Ul is implemented as a diagonal matrix acting on VJˆ.
The result Ž2.5. should be compared to the similar decomposition Ž3.9. of Ref. w1x
that arises from the simple current point of view, i.e.
Hl ' Hw l , cˆ x s [ Vcˆ m HJ l . Ž 2.6 .
Jg GrSl
Here the spaces Vĉ are projective Sl-modules, with corresponding cocycle Fl of Sl ,
while the spaces HJ l are A-modules; by assumption, the latter modules are irreducible
Žthis assumption is indeed satisfied for all cases we know of.. Simple current theory
w4,6,10,11x shows that in the decomposition Ž2.6. isomorphic A-modules appear pre-
cisely as a consequence of fixed point resolution; therefore the multiplicity of HJ l in
this decomposition is given by < Ul) <. On the other hand, as seen above, elements of the
orbifold group G that are not in the stabilizer Sl relate isomorphic A-modules. Thus we
can identify the groups Ul) and GrSl . To make this manifest, we dualize the exact
sequence 0 ™ Ul ™ G and complete it to an exact sequence
0 ™ Sl ™ G s G ) ™ Ul) ™ 0 . Ž 2.7 .
Conversely, in the decomposition of a A-module Hl there appear < Ul) <
many irre-
ducible A-modules. In simple current language, the number of irreducibles is just the
length of the orbit, and hence again we can dualize 0 ™ Ul ™ G and complete it to
0 ™ Sl ™ G s G ) ™ Ul) ™ 0 . Ž 2.8 .
As a consequence, the cardinalities of the respective subgroups of G and G are related
by
< Sl < < Ul < s < G < s < G < s < Sl < < Ul < , Ž 2.9 .
and hence in particular the dimensions of the degeneracy spaces in the decompositions
Ž2.5. and Ž2.6. coincide:
2 2
Ž dim Vcˆ . s < Sl <r< Ul < s < Sl <r< Ul < s Ž dim VJˆ . . Ž 2.10 .
We abbreviate these dimensions by
dl :s dim Vcˆ s dim VJˆ . Ž 2.11 .
There is also a manifest relationship between the groups Sl and GrUl . First we
realize that the implementation of G on the whole G-orbit of l provides us with a
two-cocycle of G with values in UŽ1. whose restriction to Sl = Sl coincides with El ; we
again denote this cocycle by the symbol El. Given such a cocycle, its commutator
cocycle El, which is defined by
El Ž g , g X . :s El Ž g , g X . rE El Ž g X , g . Ž 2.12 .
X
w x
for all g, g g G, constitutes 1 a bi-homomorphism on G = G which is alternating in
the sense that ElŽ g X , g . s ElŽ g, g X . ) . Now let us characterize for every J g Sl an
element h J of G by the property that
El Ž h J , g . s J Ž g . for all g g G ; Ž 2.13 .
such a group element h J exists because, owing to the exactness of the sequence Ž2.8.,
for every J g Sl we have JŽ g . s 1 for all g g Ul , and Ž2.13. characterizes h J
uniquely up to an element of Ul. Furthermore, as a consequence of the character
property of El in the first argument, we have
El Ž h JJX , g . s El Ž h J h JX , g . Ž 2.14 .
for all g g G, which tells us that h J h JX s h JJX modulo Ul . It follows that the mapping
J l h J Ul Ž 2.15 .
constitutes an isomorphism between Sl and GrUl . It is worth noting that this
isomorphism is logically independent from the isomorphism between G )rS Sl and Ul)
that exists according to the sequence Ž2.8..
Again this result has an obvious dual analogue. To work this out we recall from
Appendix A of w1x that the commutator cocycle Fl of Fl , defined by FlŽJ,L. s
FlŽJ,L.rF FlŽL,J. for J,L g Sl , possesses a natural extension. Fl is an alternating
bi-homomorphism on Sl = Sl , while its extension is a bi-homomorphism on G = Sl ;
by imposing the alternating property it can be further extended to a bi-homomorphism
on G = G , still to be denoted by Fl. By the character property of Fl in the second
argument, we can then associate to every g g Sl an element K g ' K Žgl. of G by
stipulating that
Fl Ž K g ,L . s g Ž L . for all L g G , Ž 2.16 .
which determines K g uniquely up to elements of Ul . The character property of Fl in the
first argument implies
Fl Ž K g g X ,L . s Fl Ž K g K g X ,L . , Ž 2.17 .
so that K g K g X s K g g X modulo Ul . Thus the map
g l K g Ul Ž 2.18 .
between Sl and GrU
Ul is an isomorphism.
For later reference we also mention another isomorphism that is similar to Ž2.18..
Namely, we now consider a character gˆ of Sl rather than a character g of G . Then the
requirement that
ˆ gˆ ,L s g Ž L .
Fl Kž / for all L g Sl Ž 2.19 .
ˆ gˆ g Sl uniquely modulo Ul , and
determines K
ˆ gˆ Ul
gˆ l K Ž 2.20 .
is an isomorphism between those elements of Sl) which are the identity on Ul Žand
hence can be regarded as restrictions of elements of Sl to Sl . and SlrU Ul . Furthermore,
combining the prescriptions above we learn that the restrictions of the relevant bi-homo-
morphisms to the stabilizer groups are closely related. Indeed, denoting by gˆ the
restriction of a given element g of Sl to Sl , we have
ˆ gˆX Ž g . s gˆ Kˆ gˆX s Fl Kˆ gˆ ,Kˆ gˆX
El Ž g , g X . s K ž / ž / Ž 2.21 .
ˆ gˆ ,Kˆ gˆX g Sl.
with g, g X g Sl and K
3. Automorphism types
ˆ
We have demonstrated in w1x that the reflection coefficients R wŽ lr,, cwr.;x V , i.e. the operator
product coefficients in the expansion ŽA.5. of a bulk field approaching the boundary, are
equal to one-dimensional irreducible representations R w r , cˆr x of the classifying algebra
C Ž A ., evaluated at the basis element F˜ Ž l, w . of C Ž A .. Thus they are given by Žsee
formula Ž5.48. of Ref. w1x.
ˆ
R wŽ lr ,, cwr.;x V s R w r , cˆr x F˜ Ž l , w . s S˜Ž l , w .,w r , cˆr xrS˜V ,w r , cˆr x .
ž / Ž 3.1 .
Moreover, it follows from the sum rule ŽA.7. and the fact that C Ž A . is semisimple, that
the reflection coefficients even provide all inequivalent irreducible C Ž A .-representa-
tions. The isomorphism classes of irreducible C Ž A .-representations are in one-to-one
correspondence with the conformally invariant boundary conditions that preserve the
orbifold subalgebra A s A G of the chiral algebra A. In this section we discuss some
implications of this basic result.
Let us associate to each boundary condition r ' w r , cˆr x the collection of all mon-
odromy charges QJ Ž r ., J g G , of r . The monodromy charges do not depend on the
choice of a representative of the G-orbit w r x, and via the prescription
g r Ž J . :s exp Ž 2p iQJ Ž r . . Ž 3.2 .
for all J g G , they furnish a character
gr g G ) Ž 3.3 .
of the simple current group. ŽThis G-character should not be confused with cˆr , which is
a character of the subgroup Ur : G .. The group G ) s Ž G ) . ) can be naturally
identified with the orbifold group, G ) ' G, and hence the quantity g r can be regarded
as an element of G.
To proceed, we observe that because of the simple current symmetry
S˜JŽ l , w .,w r , cˆ x s g r Ž J . S˜Ž l , w .,w r , cˆ x Ž 3.4 .
that was established in formula Ž5.16. of Ref. w1x, we have
S˜JŽ l , w .,w r , cˆr x S˜Ž l , w .,w r , cˆr x
R w r , cˆr x F˜ JŽ l , w . s
ž / s gr Ž J. s g r Ž J . P R w r , cˆr x F˜ Ž l , w .
ž /
S˜V ,w r , cˆr x S˜V ,w r , cˆr x
Ž 3.5 .
for every simple current J g G . Now the reflection coefficients constitute the main
ingredient in the relation between the boundary blocks B ˜ Ž l, w . Ždefined in formula ŽA.1..
and the boundary states Bw r , cˆ x. When using the notation introduced in Ž3.2., the precise
relationship, established in formula Ž6.4. of Ref. w1x, reads
ˆ r , cˆ x w r , cˆ x
R wŽ lr ,, cw .;x V ²CV ˜ Ž l,w . .
w
Bw r , cˆ x s [ [ :B Ž 3.6 .
l wg Sl)
g l'1
Thus the observation Ž3.5. tells us that the boundary blocks B ˜ Ž l, w . for primary fields l
of the A-theory that lie on one and the same G-orbit contribute to the boundary states
Bw r , cˆ x with a fixed relative phase, which is determined by the element g r of the orbifold
group. Put differently, in the presence of the boundary condition r ' w r , cˆ x the
reflection of a bulk field at the boundary is twisted by the action of the group element
g r g G.
This observation suggests that, in the terminology of w12x, the orbifold group element
g r provides us with the automorphism type1 of the boundary condition r . To establish
that this is indeed the case, we insert the expression Ž3.1. for the reflection coefficients
and the explicit values of the one-point correlators of the boundary vacuum fields
w r , cˆ xw r , cˆ x
CV into formula Ž3.6., so as to arrive at
Bw r , cˆ x s [ [ S˜Ž l , w .,w r , cˆ x B˜ Ž l , w . . Ž 3.7 .
l wg Sl)
g l'1
Next we split the summation over all untwisted l into a summation over G-orbits and
one within orbits, and the summation over Sl) into one over Ul) and one over
Sl) rUUl) . To this end we choose Žonce and for all. arbitrarily a set l ( 4 of representa-
tives of the set of G-orbits and a set w( 4 of representatives of the classes of Sl) rU
Ul) ;
more precisely, the symbol l ( will refer to the chosen representative of the orbit w lx,
1
The corresponding term in w13x is the gluing automorphism. The information contained in a boundary
condition that goes beyond the automorphism type was referred to as the Chan – Paton type in w12x. Thinking
in analogy with the general analysis of modular invariant partition functions on the torus, it may seem to be
more suggestive to take the fusion rule automorphism g w as a starting point for the description of the
automorphism type w12x. However, several different automorphism types may give rise to one and the same
permutation g w. For instance, in the case of WZW theories, automorphisms of the underlying finite-dimen-
sional compact simple Lie algebra g provide us with an automorphism type g, but whenever that
automorphism of g is inner, the associated map g w is just the identity. Another example is given by the inner
automorphisms of the rational free boson theories whose fixed point algebras correspond to the boson theory
compactified at an integral multiple of the original radius; this will be discussed in Subsection 11.1.
and w( g Sl) to the chosen representative of the class in Sl)rU
Ul) that restricts to
wˆ g Ul , i.e. satisfies w( < Ul s wˆ . Then 3.7 becomes
) Ž .
Bw r , cˆ x s [ [ )
[ S˜JŽ l ( , w ( .,w r , cˆ x
˜ JŽ l ( , w ( .
B
w l ( x wˆ g Ul Jg GrUl
g l'1
s [ [ S˜Ž l ) ( , w ( .,w r , cˆ x [ ˜ JŽ l ( , w ( . .
g r Ž J. B Ž 3.8 .
w l ( x wˆ g Ul Jg GrUl
g l'1
Here in the second line we have used the simple current relation Ž3.4., as well as the fact
that by this identity the matrix element S˜Ž l, w .,w r , cˆ x vanishes when g r ŽJ. / 1 for any
J g Ul. The latter observation shows that only those boundary blocks B ˜ Ž l, w . contribute
to the boundary state Bw r , cˆ x for which the character g r is equal to one on the whole
untwisted stabilizer Ul : G , which in turn implies that g r g G is actually an element
of Sl : G, i.e.
g r g Sl . Ž 3.9 .
Thus as a character of G the function g r factorizes to a character of GrU Ul ; according
to the sequence Ž2.7., the latter group can be identified with Sl) . As we will see later,
this result is perfectly natural.
We now concentrate on the J-summation for fixed values of l ( and w( . To proceed,
we need a few further tools. First, it turns out to be useful to introduce for every g g Sl
and every J g G the endomorphism
Og ,J :s dly1 r2 Ý g Ž L . OJL w ( Ž 3.10 .
Lg SlrUl
of Vĉ , where Oc are the endomorphisms defined by ŽA.3.. Inserting that formula for Oc
and interchanging the order of summations, the maps Og ,J can also be written in the
form
g Ž LX . Fl Ž L,LX .
) )
Og ,J s dly2 Ý w( Ž L . Rwˆ Ž L . Ý
X
Lg SlrUl J L g Sl rUl
)
s Ý
)
ˆg
w( Ž L . Rwˆ Ž L . P d L ,Kˆ g sJ w( K ž / ˆg ,
Rwˆ K
ž / Ž 3.11 .
Lg SlrUl J
where K ˆ g g Sl is as defined by formula Ž2.19., with gˆ g Sl) given by gˆ s g < Sl.

ŽBecause of g g Sl, the character gˆ g Sl) is the identity on Ul : Sl , as required in
ˆ g is defined only up to elements of Ul; but the result Ž3.11. for
Ž2.19.. Also recall that K
Og ,J is independent of the choice of representative..
With the help of formula Ž3.11. one checks that
)
Og ,JL s g Ž L . Og ,J for all L g Sl , Ž 3.12 .
and that
ˆ g ,Kˆ g X Og g X ,J s El Ž g , g X . Og g X ,J
Og ,J Og X ,J s Fl K ž / Ž 3.13 .
for all g, g X g Sl, i.e. the endomorphisms Og ,J with g g Sl furnish a projective
representation of the stabilizer Sl Žin the last equality we have used the identity Ž2.21...
Let us also see to which extent these results depend on the choice of representative w( .
Any other representative is of the form w( c with c g Sl) and c < Ul s id; thus upon
choosing a different representative the endomorphisms Og ,J get replaced by
)
O˜g ,J s c K
ˆgž / Og ,J . Ž 3.14 .
On the other hand, the two-cocycle that characterizes the relevant representation of Sl
does not depend on this choice. Indeed, as a consequence of Kˆ g g X s Kˆ g Kˆ g X modulo Ul
the maps O˜g ,J satisfy
)
O˜g ,J O˜g X ,J s c K
ˆ g Kˆ g X
ž / ˆ g ,Kˆ g X O˜g g X ,J .
Og ,J Og X ,J s Fl K ž / Ž 3.15 .
Next we also choose a set J( 4 of representatives for the classes in GrS
Sl and write,
for every g g Sl,
Qg ' QgŽ l. :s [ g Ž J( . Ž Og ,J ( m id . ( PJ ( l ( , Ž 3.16 .
J(g GrSl
where Og ,J ( is defined in Ž3.10. and Pm is the projector from the A-module Hl to its
isotypical component of type Hm . Owing to the identity Ž3.12., Qg is in fact indepen-
dent of the choice of representatives J( Žwhereas Og ,J ( again depends on that choice.,
and from Ž3.13. it follows that
ˆ g ,Kˆ g X Qg g X
Qg Qg X s Fl K ž / Ž 3.17 .
X
for all g, g g Sl. Let us also note that when specializing to the A-vacuum sector,
where UV s SV s G, we simply have K ˆ g s 1 and hence Og ,J s id for all J g G , so that
ŽV .
Qg s [J g G g J PJ .
Ž .
With these results at hand, we can now address the J-summation that appears in
formula Ž3.8.. Inserting the definition ŽA.1. of the boundary blocks B,˜ we find that
[ ˜ JŽ l ( , w ( . s
g Ž J. B [ g Ž J( . [ ˜ Ž J(l ( , J
g Ž L. B w(.
(L
Jg GrUl J(g GrSl Lg Sl rUl
s Nl dly2 [ g Ž J( . [ g Ž L . b( (
J(g GrSl Lg Sl rUl
w( Ž LX . Rwˆ Ž LX . m id m B J ( l ( , Ž 3.18 .
)
= ž X
[
L g SlrUl J L
(
/
Vcˆl m Vcˆlq ™ C is the non-degenerate linear form defined in Ž4.21. of Ref. w1x.
where b( :V
Performing the L-summation, this becomes
[ ˜ JŽ l ( , w ( . s Nl
g Ž J. B [ g Ž J( . b ( (
Jg GrUl J(g GrSl
)
= ž w ž Kˆ /
J( ( g
ˆ g m id m B J ( l (s Nl BlŽ g .
Rwˆ K
ž / /
Ž 3.19 .
with
BlŽ g . :s Bl ( Ž Qg m id . . Ž 3.20 .
Here Bl s BlŽ1. are the boundary blocks of the A-theory, which are given by expression
Ž4.25. of Ref. w1x, and we have used that via the identity ŽA.4. they can be written in the
form
Bl ' Bw l , wˆ x s Nly1 [ [ B˜ Ž Jl ( ,w .
wg Sl) Jg GrSl
w% ŵ
s Nly1 [ [ B˜ Ž Jl ( ,Lw ( .
Jg GrSl Lg Sl rUl
s dly1 r2 [ b( ( ž Ý /
OL w ( m id m B J l s [ b( m B J l . Ž 3.21 .
Jg GrSl Lg Sl rUl Jg GrSl
At this point it is worth realizing that the twisted intertwining property Ž2.2. of Qg is
formulated independently of the subalgebra A, and hence Qg only depends on the
automorphism g itself, but not on the particular orbifold group G containing g we are
considering. By the result Ž3.20., this independence on the choice of G then holds for
the quantities BlŽ g ., too.
Inserting the result Ž3.19. into formula Ž3.8., we finally see that
Br s [ Nl [ S˜Ž l ( , w ( ., r
BlŽ g r . . Ž 3.22 .
w l( x wˆ g Ul)
g l'1
Thus, in summary, the boundary state Br can be entirely constructed from the
information contained in the boundary blocks B w l , wˆ x together with the action of Sl they
carry and in the character g r of G. As we will see, this implies that g r indeed
constitutes the automorphism type of the boundary condition r . We have also seen that
only those boundary blocks B ˜ Ž l, w . contribute to the boundary state Br for which the
stabilizer Sl contains g r which, incidentally, shows that the factorization to a character
Ž
of GrU Ul is a rather natural property of the elements of Sl .. This should be regarded as
a selection rule on the possible boundary blocks that show up in the boundary state; the
concrete form of this selection rule is completely determined by the automorphism type
of the boundary condition.
Our derivation also demonstrates that in the case of our interest one can associate an
automorphism of the chiral algebra A to eÕery boundary condition. This comes as a
result of our analysis and does not have to be put in as an assumption. In contrast, when
the subalgebra A that is preserved by a boundary condition is not an orbifold subalgebra
A G , then the boundary condition need not necessarily possess an automorphism type.
Indeed, as we will see in Section 11.3, most conformally invariant boundary conditions
of the Z 2-orbifold of a free boson, compactified at a rational radius squared, do not
possess an automorphism type.
4. Twisted blocks and twisted Ward identities

In the discussion of automorphism types above we have introduced, for every
primary label l ' w l, cˆ x of the A-theory and for every element g g Sl of the orbifold
group that stabilizes l, the map Qg Ž3.16. as well as the linear form BlŽ g . s Bl (ŽQg m id .
Ž3.20. on the tensor product A-module Hl m Hlq. We will refer to the latter linear
forms as g-twisted boundary blocks. As demonstrated in the previous section, every
boundary state Br is a linear combination of those twisted boundary blocks BlŽ g r . for
which the twist g r lies in Sl .
What still remains to be established is that the notation for the maps Qg that was
introduced in Ž3.16. is in agreement with the use of the same notation in Section 2, i.e.
that these maps satisfy the g-twisted intertwining property Ž2.2.. To address this issue,
we need some information about the representation Rl of A on the subspaces in the
decomposition Ž2.6. of a A-module Hl ' Hw l ( , cˆ x into irreducible A-modules. When a
vertex operator algebra is extended by a finite group of simple currents, one obtains a
so-called abelian intertwining algebra w14x. In the cases of our interest this extended
algebra is, by assumption, again a vertex operator algebra, and various aspects of its
representation theory have been studied in w14,15x. A complete description of the action
of the vertex operator algebra on all modules Hl is not yet available, but the known
results naturally suggest the following structure. On the A-module Vĉ m HJ ( l ( : Hl an
element YJ Ž z . ' Y Ž ÕJ ; z . of A with ÕJ g HJ : HV is represented by
R w l ( , cˆ x Ž YJ . s Rwˆ Ž JX . m [ RJ ( l ( Ž YJ . . Ž 4.1 .
J(g GrSl
Here JX is defined by J s JX J(X with J(X the representative for a class in GrS Sl Žsee before
formula Ž3.16.., and RJ ( l (Ž YJ . is a map from HJ ( l ( to HJX( J ( l ( . To proceed we use the
commutation properties
ˆ g s Fl JX ,Kˆ g Rwˆ Kˆ g Rwˆ Ž JX . s g Ž JX . Rwˆ Kˆ g Rwˆ Ž JX .
Rwˆ Ž JX . Rwˆ K
ž / ž / ž / )
ž / Ž 4.2 .
and
[ ž id m R Ž Y . / ( P
J(l ( J J ( l ( s RJ ( l (
XX XX Ž YJ .
J(g GrSl
s PJX( JXX( l ( ( [ ž id m R Ž Y . / J(l ( J Ž 4.3 .

J(g GrSl
as well as 2
Ž Rwˆ Ž JX . m id . ( PJ X XX
( J(l ( ž ˆ g PJX( JXX( l (( Ž Rwˆ Ž JX . m id . .
s Fl J(X ,K / Ž 4.4 .
Then we obtain
ˆg
Rl Ž YJ . ( QgŽ l. s g Ž JX . Fl J(X ,K
)
ž / [
XX
J(g GrSl
)
ˆg
= g Ž J(XX . JXX( w( K ž / ž R ˆ ž Kˆ / m id / ( P
w g J ( J ( l ( ( Rl
X XX Ž YJ .
X )
g Ž J(X .
)
sg ŽJ .
XX
[
J(g GrSl
)
ˆg
= g Ž J(XX . JXX( w( K ž / ž R ˆ ž Kˆ / m id / ( P
w g J ( l ( ( Rl
XX Ž YJ .
)
s g Ž J. P QgŽ l. ( Rl ( Ž YJ . ' QgŽ l. ( Rl y1
Ž g Ž YJ . . . Ž 4.5 .
Here in the last expression the element g of the orbifold group is to be regarded as an
automorphism of the chiral algebra A, while in the intermediate steps it is interpreted as
2
This relation is needed for compatibility with the fact that while the isomorphism class of the projective
Sl-representation Rwˆ is the same for all values of l within the class w l ( x, its explicit realization can depend
on l, and this dependence should precisely be characterized by Fl.
a character of the simple current group G s G ) . Formula Ž4.5. reproduces the twisted
intertwiner property Ž2.2. and hence is the desired result.
We also remark that according to Section 2 the twisted intertwiners carry a projective
representation that is characterized by the cohomology class of the cocycle El or,
equivalently, by the commutator cocycle El. On the other hand, according to relation
Ž3.17. the concrete realization Ž3.16. of the twisted intertwiners can be characterized by
the commutator cocycle Fl. This is compatible because of the identity Ž2.21..
Now for every field Y Ž z . s Ý n g Z Yn zyn y D Y of conformal weight D Y in the chiral
algebra A, the ordinary boundary block BlŽ1. of the A-theory obeys the Ward identity
appropriate for a two-point block on P 1. That is,
BlŽ1. ( Ž Rl Ž Yn . m 1 q z Y 1 m Rl Ž Yyn . . s 0 Ž 4.6 .
with z Y s Žy1 . DYy1
. When combined with the definition Ž3.20. of the twisted boundary
blocks, the twisted intertwiner property Ž4.5. therefore allows us to write Žsuppressing
from now on the representation symbol Rl .
BlŽ g . ( Ž Yn m 1 . s Bl ( Ž Qg m id . ( Ž Yn m 1 . s Bl ( Ž g Ž Yn . m 1 . ( Ž Qg m id .
s yz Y Bl ( Ž 1 m g Ž Yyn . . ( Ž Qg m id .
s yz Y BlŽ g . ( Ž 1 m g Ž Yyn . . . Ž 4.7 .

Thus the twisted boundary states satisfy a twisted Ward identity
BlŽ g . ( Ž Yn m 1 q z Y 1 m g Ž Yyn . . s 0 . Ž 4.8 .
It is worth noting that we have defined the twisted Ward identities Ž4.8. through an
automorphism of A that only acts on the second factor of the tensor product. One could
easily generalize this by considering the action of two different automorphisms g L and
g R on the two factors, according to
BlŽ g . ( Ž g LŽ Yn . m 1 q z Y 1 m g R Ž Yyn . . s 0 . Ž 4.9 .
But obviously what matters is only the combination gy1 L g R . Thus we can describe the
space of automorphism types either as the coset space Ž G = G .rG or as the group G. In
fact, the map Ž g L , g R . ¨ gy1
L g R provides a natural bijection between the two sets which,
in the case when G is a Lie group, is even an isomorphism of smooth manifolds. ŽFor
theories of free bosons, the description in terms of Ž G = G .rG has been established in
w16x..
5. The classifying algebra C (g) for automorphism type g
5.1. IndiÕidual classifying algebras for fixed automorphism type
We now restrict our attention to the collection of boundary conditions that possess
some fixed automorphism type g. According to the results of Section 8 the correspond-
ing boundary states can all be written as linear combinations of the twisted boundary
blocks Ž3.20. with fixed g. This suggests to study analogous elements of the classifying
algebra; accordingly we introduce for every l with g g Sl the linear combination
< Ul <
g Ž J . F˜ JŽ l ( , c ( .
)
FlŽ g . :s Ý Ž 5.1 .
< G< Jg GrUl
of basis elements of the classifying algebra C Ž A ., where by l ( and c( Žwhich satisfies

c( % cˆ( for l ' w l ( , cˆ( x . are the representatives introduced in the paragraph before
formula Ž3.8.. To proceed, we also note the relation
g Ž J . S˜JŽ l , c ., r s Ž < G <r< Ul < . d g , g r S˜Ž l , c ., r

)
Ý Ž 5.2 .
Jg GrUl
that follows for every g g Sl with the help of the simple current symmetry Ž3.4..
ŽWhen g is not in Sl, then according to the remarks after Ž3.8. this expression
vanishes..
Using the Verlinde-like formula that expresses the structure constants of C Ž A . in
˜ the result Ž5.2. allows us to compute the product of
terms of the diagonalizing matrix S,
two elements of C Ž A . of the form Ž5.1. as
X XX
FlŽ g .wFlŽX g . s d g , g X Ý NlŽ ,gl.X lFlŽXXg . Ž 5.3 .
XX
l
with3
XX XX y1
NlŽ ,gl.X l ' NlŽ ,gl.X l Ž A . :s Ý S˜Ž l ( , c ( ., r
S˜Ž lX( , c X( ., r S˜Ž)lXX( , c XX( ., r S˜V , r
ž / . Ž 5.4 .
r
g rsg
This means that the elements FlŽ g . for all l with g g Sl span not only a subalgebra,
but even an ideal of the classifying algebra C Ž A .. We call this ideal of C Ž A . the
indiÕidual classifying algebra for automorphism type g and denote it by C Ž g . Ž A .. Also,
by construction the FlŽ g . are linearly independent, and hence they furnish a basis of
C Ž g . Ž A ., i.e. for every fixed g g G we have
C Ž g . Ž A . s span C FlŽ g . < Sl 2 g 4 . Ž 5.5 .

Clearly, C Ž g . Ž A . is again semisimple; its one-dimensional irreducible representations
are obtained by restriction to C Ž g . Ž A . of those one-dimensional irreducible representa-
tions Rr of C Ž A . which satisfy g r s g.
Moreover, the following counting argument shows that together the elements FlŽ g .
for all l and all g g G span all of C Ž A .. Namely, associated to every G-orbit w lx of
A-primaries there are < Ul < many A-primaries w l, cˆ x, and each of them gives rise to < Sl <
many basis elements FlŽ g .. On the other hand, each such G-orbit contains < G <r< Sl <
many A-primaries, each of them leading to < Sl < many basis elements F˜ Ž l , f . of C Ž A ..
3
Note that this expression is only defined when g g Sl l SlX l SlXX .
With the help of the identities Ž2.9. among the sizes of the various subgroups it thus
follows that
Ý < Sl < s Ý < Sl < . Ž 5.6 .
l l
Q G Ž l .s0
As a consequence we have indeed – as algebras over C, and with the distinguished

bases related by Ž5.1. – an isomorphism
CŽ A. ( [ CŽ g.Ž A . . Ž 5.7 .
ggG
This decomposition may be regarded as expressing the fact that in the situation of our
interest every boundary condition has an automorphism type. Put differently, the algebra
C Ž A . provides a unified description of the boundary conditions for the < G < different
automorphism types that correspond to the elements of G.
In the special case of trivial automorphism type, g s 1, we can use the result w1x that
S˜Ž l, c ., r s Sl , r for all r with g r s 1 to see that the ideal C Ž1. Ž A . is nothing but the
fusion rule algebra of the A-theory, so that we recover the known results w3,17x for
boundary conditions that do not break any of the bulk symmetries. It should also be
noticed that the precise form of the structure constants of the ideals C Ž g . Ž A . does
depend on the choice of representatives l ( and c( Žexcept when g s 1, where
independence of this choice follows as a consequence of the simple current relation
Ž3.4... This is, however, perfectly fine, because the twisted boundary block depends on
the choice of representatives as well, and in fact in a manner so as to cancel the overall
dependence in all physically meaningful quantities like one-point correlators for bulk
fields on the disk.
5.2. Independence of C ( g ) on the orbifold group
A remarkable feature of the formula Ž5.4. for the structure constants of C Ž g . Ž A . is

that the automorphism type g just enters by restricting the range of summation. This
suggests that the individual classifying algebra for automorphism type g in fact does
not depend on the specific orbifold subalgebra A, i.e. on the group G that contains the
automorphism g, but rather on g alone. Below we will show that this is indeed the case,
and actually the statement already applies to the relevant entries of the diagonalizing
matrix S.˜ Now recall from Section 3 Žsee the remarks after formula Ž3.21.. that a similar
statement applies to the twisted boundary blocks BlŽ g .. When combined with the present
result, then according to formula Ž3.22. this independence property of the blocks BlŽ g .,
which are chiral quantities, extends to the Žnon-chiral. boundary states Br .
To prove the independence of C Ž g . Ž A . on the choice of orbifold group G containing
g, it is convenient to study the orbifold with respect to the cyclic group
GX :s ² g : ( Z <GX < Ž 5.8 .
that is generated by g. We first note that the diagonalizing matrix S˜X for the total
classifying algebra C Ž AX . of all boundary conditions preserving AX s A ² g : ; A is
given by the expression ŽA.6. which contains contributions involving the various
matrices S J Žthough in this special cyclic case the formula simplifies.. However, in the
expression for the structure constants of C Ž g . Ž AX . only those entries S˜ŽX lX , w .,w r X , cˆ x X appear
for which g g SlX X , and these entries turn out to be particularly simple. Indeed, since g
generates ² g :, the latter property means that the stabilizer is maximal, SlX X s GX ; by the
duality relation UlXX ( GX ) rSlX X ) this implies SlXX s UlXX s 1. Similarly, because of g r X s g
the monodromy charges of r X with respect to the extension from AX to A have
denominator < GX <, so that r X cannot be a fixed point under any simple current in GX ) ,
which implies that also SrXX s UrXX s 1. Thus we have to deal with full GX ) -orbits only; in
particular the simple current summation in the formula for S˜X reduces to the term with
J s 1:
< GX <
˜
SlX X ,w r X , cˆ x X s X X X X
SlX X , r X s < GX < SlX X , r X . Ž 5.9 .
(< S < < U < < S < < U <
lX lX rX rX
To proceed, we note that the algebra AX s A ² g : can be obtained from A s A G as a

simple current extension by the subgroup
G XX :s J g G < g Ž J . s 1 4 Ž 5.10 .
XX X
of the simple current group G s G ) . ŽThe subgroup G has index < G < in G ; in fact, the
factor group GrG GXX is cyclic of order < GX <.. In particular, the modular S-matrix SX of the
X
A -theory can be expressed through quantities of the A-theory as
< G XX <
SlX X , mX s XX Ý cˆlX Ž J . SlJ , m cˆm)X Ž J . Ž 5.11 .
(< S lX
< < UlXXX < < SmXXX < < UmXXX < Jg U
XX XX
U
ll m
X X
with lX ' w l, cˆlX x XX and mX ' w m , cˆmX x XX .

The AX-theory has a simple current of the form
XX
Jg X s Jg P G XX ' Jg , Ž 5.12 .
where Jg g G is a simple current of the A-theory that is characterized by the property
that < GX < is the smallest positive integer m such that ŽJg . m lies in G XX , so that Jg X has
order < GX <. By this property of Jg and the definition Ž5.10. of G XX it follows that
g r Ž Jgm KXX . s g Ž Jgm . / 1 Ž 5.13 .
XX XX X
for every K g G and every m s 1,2, . . . , < G < y 1. This means that none of the
monodromy charges QJ gm K XX Ž r . vanishes, so that r cannot be a fixed point with respect
to any of these simple currents Jgm KXX . It follows that the stabilizer Sr is contained in
G XX , which in turn implies that SrXXX s Sr and hence also UrXXX s Ur . As a consequence,
the boundary labels w r , cˆr x of the matrix S˜ that satisfy g r s g are precisely the same as
those appearing in formula Ž5.11. for SX , and the simple current summation in the
expression for the corresponding entries of S˜ only runs over elements of G XX . Using also
the fact that SlXXX s Sl l G XX , we can therefore write
<G<
S˜Ž l , cl.,w r , cˆr x s Ý cl Ž J . Sl ,Jr cˆr) Ž J .
(< S < < U < < S < < U <
l l r r Jg SllUr
< GX < < G XX <

s Ý clX Ž J . SlJ, r cˆr)X Ž J . . Ž 5.14 .
( < Sl < < Ul < < SrXXX < < UrXXX < Jg S XX
U XX
ll
X X
r
Furthermore, because of g r s g the simple current relation Ž3.4. for S˜ implies that these
entries of S˜ are identical for all l on one and the same G XX-orbit. We may therefore
equate the expression Ž5.14. with its average over SlXX . The simple current relation for
the matrices S J then amounts to a restriction of the summation to UlXX , so that
< GX < < G XX <
S˜Ž l , cl.,w r , cˆr x s XX < Ý clX Ž J . Sl ,Jr cˆr)X Ž J .
(< S < < U < < S
l l rX
< UrXXX < Jg UlXXX lUrXXX
< SlXXX < < UlXXX <

s < GX < ( < Sl < < Ul <
SwX l , cˆlX x XX ,w r , cˆrX x XX . Ž 5.15 .
To analyze the prefactor appearing here, we first remark that the index of the subgroup
SlXXX in Sl is some divisor nl of < GX <. There is a simple current Jl g Sl which plays an
analogous role for the embedding SlXXX : Sl as Jg plays for the embedding of G XX in G ,
i.e. nl is the smallest power such that Jlnl is in SlXXX , and the elements of Sl are of the
form Jlm KXX with KXX g SlXXX and m s 1,2, . . . ,nl y 1. Moreover, from the fact that
g g Sl it follows with the help of duality that Jlm f Ul for all m s 1,2, . . . ,nl y 1, and
hence we have Ul : SlXX . Thus when forming the untwisted stabilizer associated to SlXXX
one does not lose any elements of Ul, so that Ul : UlXXX . Observing that Ul is precisely
the kernel of the group homomorphism from UlXXX to C that maps KXX g UlXXX to the nlth
root of unity FlŽJl ,K XX ., it follows that the index of Ul in UlXXX is nl . Thus we have
< Sl <r< SlXXX < s nl s < UlXXX <r< Ul < , Ž 5.16 .
so that Ž5.15. reduces to
S˜Ž l , cl.,w r , cˆr x s < GX < SwX l , cˆlX xXX ,w r , cˆrX x .

XX
Ž 5.17 .
Comparison with Ž5.9. then shows that the relevant matrix elements of S˜ are identical
to those of S˜X . We conclude that
C Ž g . Ž A G . s C Ž g . Ž A ² g : . s :C
CŽ g. Ž 5.18 .
for every finite abelian orbifold group G with G 2 g. It is worth pointing out that the
fact that the individual classifying algebra does not depend on the preserved subalgebra
˜
constitutes another quite non-trivial check of our ansatz for the diagonalizing matrix S.
Ž g.
We also learn that the structure constants of the algebra C read
X X X 2
SlX X , r X SmX X , r X SnX )X , r X
NlŽXg, m. Xn s NlŽXg, m. Xn ŽA ² g:
. s<G < Ý X . Ž 5.19 .
w rX x X
SV X
,r
X
g rXsg
By inserting a suitable projector, they can rewritten in the form
X SlX X , r X SmX X , r X SnX )X , r X X X

NlŽXg, m. Xn s
X
Ý Ý g Ž J . gr Ž J .
X X
X
X
X )
X
SV X X
s
X
Ý g Ž JX . NlX J, mn
X
X X , Ž 5.20 .
J gG r ,r J gG
i.e. as a linear combination of fusion coefficients of the A ² g :-theory.

It is reasonable to expect that similar considerations apply to orbifold subalgebras
with respect to non-abelian groups G, too, so that in particular also in that case every
boundary condition possesses a definite automorphism type. Assuming this to be true,
the classifying algebras for fixed automorphism type studied here should coincide with
their analogs in the non-abelian case. In other words, the set of boundary conditions will
be exhausted by those boundary conditions that are already known from the cyclic
groups ² g : for all g g G. ŽOn the other hand, the detailed structure of the classifying
algebra, which involves the distinguished basis F˜ 4 , will still be more involved..
5.3. Relation with traces on bundles of chiral blocks
The independence on the specific group G finds its natural explanation in the fact
that these numbers are interpretable as traces of appropriate maps on bundles of chiral
blocks w12x. Namely, since for every l we are given the twisted intertwiner maps
QgŽ l. : Hl ™ Hg wl Ž2.1., we also have the corresponding tensor product maps
Qg ' QgŽ,lg1,l. 2. ..,.g. l m . :s QgŽ l1 . m QgŽ l 2 . m . . . m QgŽ l m . Ž 5.21 .
on tensor products of A-modules. In view of the definition w18,19x of chiral blocks
Vl1l 2 .. . l m as singlets Žwith respect to a suitable block algebra. in the algebraic dual
Ž Hl m Hl m . . . m Hl . ) of these tensor products, the twisted intertwining property
1 2 m
together with the fact that the automorphism g respects the grading of the chiral algebra
implies the existence of a linear map
Qg) ' Qg), gŽ,l.1. .l,2g. . . l m . : Vl1 l 2 . . . l m ™ Vg wl1 g wl 2 . . . g wl m Ž 5.22 .
between spaces of chiral blocks. When g wl i s l i for all i s 1,2, . . . ,m, this linear map
Qg) is an endomorphism so that one can compute its trace; we will be interested in the
traces of three-point blocks. ŽA concrete description of the block algebras and hence of
the maps Qg) is so far only available for the case of WZW theories, where the situation
can be analyzed in terms of the horizontal subalgebra of the relevant affine Lie algebra..
Now in the case of our interest, where all simple current stabilizers in the AX-theory are
trivial, the fusion rules of the A-theory can be expressed through the modular S-matrix
of the AX-theory as
n X 2
1 SlX X , r X SmX X , r X SnX )X , r X
Nl , m s < G < Ý < SrXX < SVX X , r X
w rX x X
X
Q G X Ž r .s0
rX .
SlX X , r X SmX X , r X SnX )X , r X X
s Ý Ý e2p i Q JX Ž
X s Ý NlX X J, nmX Ž 5.23 .
X
J gG
X
rX SV X
,r
X X
J gG
X
This indicates that the chiral three-point blocks of our interest can be decomposed into
the direct sum of spaces of chiral blocks of the AX-theory. Such a decomposition should
in fact be expected on general grounds, and the chiral blocks of the AX-theory should fit
together to sub-bundles of the bundles of chiral blocks of the A-theory. Now when
restricted to irreducible AX-modules the maps Qg are ordinary intertwiners, and as a
consequence the map Qg) acts on the subspaces of chiral blocks as a multiple of the
X
identity. On the subspace of dimension NlŽXg, m. XJn the map Qg) should therefore act with
eigenvalue g ŽJ.. It thus follows that upon choosing representatives of the G X-orbits, the
trace of this map is precisely given by the number Ž5.20.:
X
tr Vlmn Qg) Ž lmn . s NlŽXg, m. Xn Ž A ² g : . . Ž 5.24 .
For g / 1 these numbers do depend on the choice of representatives lX of the orbits

l ' w lX x, in agreement with the fact that the maps Qg) are defined only up to a phase.
6. T-duality
In all the considerations so far, we have required that the torus partition function, and
correspondingly the pairing of the labels l and l˜ of the bulk fields fl, l˜ , is given by
charge conjugation, which we denote by sc ) : l˜ s sc ) Ž l. ' lq. In this section we
analyze what happens when a different torus partition function is chosen.4 To this end
we first have to state what we mean by a chiral algebra A and its automorphisms in
more concrete terms than was done so far. In mathematical terms, a chiral algebra5 is a
Õertex operator algebra w21–23x; the relevant data are therefore the vector space HV ,
the vacuum vector V g HV , the Virasoro element Õ Vir g HV , and a ‘ vertex operator
map’ Y. The latter realizes the state–field correspondence, i.e. associates to every
Õ g HV a field operator Y Ž Õ; z . Žtechnically, a linear map from HV to EndŽ HV . m
Cww z, zy1 xx with z a formal variable., e.g. the energy–momentum tensor T to the
Virasoro element, Y Ž Õ Vir ; z . s T Ž z .. By an automorphism of a vertex operator algebra
A we mean an invertible linear map
s: HV ™ HV Ž 6.1 .
that is compatible with state–field correspondence, i.e. satisfies
sy1 Y Ž s Õ ; z . s s Y Ž Õ ; z . Ž 6.2 .
for all Õ g HV . ŽLet us stress that – unlike e.g. in w24x – at this point we do not require
that the map s leaves the Virasoro element fixed.. As already outlined in Section 2,
each such map is accompanied by a permutation s ) of the label set I:s l4 of
A-primaries and by twisted intertwiners Qs ' QsŽ l. : Hl ™ Hs ) l between the corre-
sponding irreducible A-modules.
4
It is by no means necessary that the permutation that Žfor maximally extended chiral algebra. characterizes
the torus partition function is equal to the permutation g w defined via the automorphism type– if such an
automorphism type exists at all – of a boundary condition. While both mappings are associated to the
transition from chiral conformal field theory Ži.e. conformal field theory on a complex curve. to full
conformal field theory Žconformal field theory on a real two-dimensional surface., they refer to such a
transition for two different world-sheets – the disk and the torus, respectively– which are not related by any
factorization rules. As a consequence, they can be chosen independently.
5
Not to be confused with the global object for which the term chiral algebra has also been used in the
recent mathematical literature w20x, which we prefer to call a block algebra.
To proceed, we introduce two particular subgroups of the group G of all maps f
from I to AutŽ A .. For every f g G and every l g I, the image
fl :s f Ž l . : HV ™ HV Ž 6.3 .
is an automorphism in AutŽ A .. We denote by GZ the subgroup of all those elements f
of G for which the map 6
p f) : I ™ I, l ¨ fl) Ž l . Ž 6.4 .
preserves conformal weights modulo integers, i.e. fulfills Dp f) Ž l . s Dl mod Z for all
l g I, and is an automorphism of the fusion rules, i.e. satisfies
Np f) Ž l .,p f) Ž m .,p f) Ž n . ' N fl) Ž l ., fm) Ž m ., fn) Ž n . s Nl , m , n Ž 6.5 .
for all l, m , n g I. ŽExamples for such automorphisms are those induced by simple
currents, see e.g. Refs. w6,25x.. Every element of GZ gives rise to a modular invariant
torus partition function
)
Z f Ž t . :s Ý xl Ž t . xp ) q
f Žl .
Žt . . Ž 6.6 .
lgI
The second subgroup of G of our interest consists of constant maps f whose image
– to be denoted by g f – is an automorphism of A that leaves the Virasoro element
fixed, g f Ž Õ Vir . s Õ Vir Žand hence in particular obeys g f) Ž V . s V and preserves confor-
mal weights exactly, not only modulo integers.. We denote the subgroup of those maps
of this kind that also lie7 in GZ by GB . Every automorphism g f of this subgroup of GZ
with f g GB can be used to define conformally invariant boundary conditions. Further-
more, for every f g GB and every f X g GZ the modular invariant partition functions Z f X
and Z f f X are physically indistinguishable, i.e. upon a suitable relabelling of the fields all
correlation functions in the associated conformal field theories coincide. Accordingly, it
is appropriate to refer to GB as the T-duality group of the theory. For every chiral
algebra A the T-duality group GB contains in particular the map f c whose image is the
charge conjugation automorphism sc , i.e. f c Ž l. s sc for all l g I, and p f)c Ž l. s sc Ž l.
s lq.
The two theories with partition functions Z f X and Z f f X being indistinguishable, in
particular the respective sets of all conformally invariant boundary conditions must be
the same. It is worth investigating this correspondence in some detail. Let us denote for
any f g GB and any f X g GZ by
X X
CŽ gf ; f . 'CŽ gf ;f .Ž A² g: . Ž 6.7 .
the classifying algebra for boundary conditions of automorphism type g f for a confor-
mal field theory with torus partition function Z f X . Recall from Section 5 that this algebra
Žas well as its distinguished basis. can be constructed by starting with any arbitrary finite
6
Note that typically several distinct f g G will give rise to one and the same permutation p f) .
7
In fact one should expect that the property of inducing a fusion rule automorphism need not be required
independently, but is satisfied automatically as a consequence of the consistency of the relevant orbifold
theory. This has been demonstrated in the case of order-2 automorphisms in w5x. Indeed, consistency of the
orbifold theory requires that the S-matrix of the A-theory behaves with respect to the permutation g w that is
induced by the non-trivial element g of Z 2 via the maps Qg as Sl , g wm s S g wl , m . When combined with the
Verlinde formula, this implies that g w furnishes an automorphism of the fusion rules of the A-theory.
abelian group G containing g. As also discussed there, for the case X
where f X s f c
corresponds to charge conjugation, the structure constants of C Ž g f ; f . are given by the
traces over the linear maps induced by Qg f on the three-point chiral blocks:
NlŽ ,gmf ;, nf c . s tr Vl , m ,n Qg)f , g f , g f Ž l m n . .

ž / Ž 6.8 .
In the general case, what should matter are not the individual maps f and f X by
themselves, but rather only the information on how the pairing described by the
boundary conditions relates to the pairing in the torus partition function. In other words,
a simultaneous
X
action of an element of GB on both labels of the classifying algebra
C Ž g f ; f . will not change the situation in an observable manner. A well-known example
of this effect is seen in the theory of a free boson compactified at radius R, where the
T-duality map R ¨ 2rR Žwhich corresponds to f ¨ ffc . amounts to exchanging Dirich-
let and Neumann conditions Žcompare subsection 10.1.. ŽFor orbifolds of free bosons,
T-duality has been studied recently e.g. in w26x..
In short, for every f XX g GB there is an isomorphism
XX X
C Ž g f fX ; f f . ( C Ž g f ; f
X X .
Ž 6.9 .
of classifying algebras. With the help of these isomorphisms X
we deduce that the
structure constants of the classifying algebra C Ž g f ; f c f . are given by the traces
X
NlŽ ,gmf ;, nf c f . s tr Vl , m ,n Qg)f gyf X 1 , g f gyf X 1 , g f gyf X 1 Ž l m n . .
ž / Ž 6.10 .
In this formula it is manifest that simultaneous application of a T-duality transformation
f g GB on the bulk modular invariant and on the automorphism type of the boundary
conditions does yield isomorphic classifying algebras.
Building on this result, we expect that formula Ž6.10. extends naturally to arbitrary
elements f of GZ , i.e. to the case that f is not necessarily constant on I. We are thus led
to conjecture that in this case the relation Ž6.10. gets generalized to 8
X
NlŽ ,gmf ;, nf c f . s tr Vl , m ,n Qg)f flXy 1 , g f fmXy 1 , g f fnXy 1 Ž l m n . .
ž / Ž 6.11 .
In the special case of Dodd-type modular invariants Žwhich correspond to a simple
current automorphism for an order-2 simple current of half-integral conformal weight.,
the relation Ž6.11. is already known w27,28x to hold.
Our observations imply in particular that the automorphism type g r of a boundary
condition is not an obserÕable concept. What is observable is the product g r)p f)y 1 , i.e.
the ‘difference’ between an automorphism in the torus partition function and on the
boundary. ŽIn other words, one should regard automorphism types as elements of the
‘affinum’ – see Section 7 below – of the group GB rather than as elements of GB
itself.. Similarly, also the difference g Ž g X .y1 of two automorphism types is observable;
e.g. the annulus partition function will be different in the situation where one deals with
two boundary conditions of distinct automorphism type as compared to the situation
where the automorphism type of both boundary conditions is the same.
8
One might have expected that here products of Q-maps rather than the Q-map for the product of
automorphisms appears. But when the group GZ is realized projectively, this would lead to inconsistencies.
When GZ is realized genuinely, then the two descriptions are equivalent.
While in the considerations above the T-duality transformations had to be applied to
the torus partition function and to the boundary conditions simultaneously, there is also a
slightly different notion of T-duality for boundary conditions, to which we now turn our
attention. Namely, we keep the bulk theory fixed, and ask whether boundary conditions
of different automorphism type can be associated to each other through a suitable
element of the T-duality group. While in many cases this question turns out to have an
affirmative answer, the relationship in question is between families of boundary
conditions rather than between individual boundary conditions. In short, T-duality on
boundary conditions is not a one-to-one map, in general. Closer inspection shows that
the relevant families of boundary conditions can be understood as orbits of boundary
conditions with respect to some suitable symmetry. Here by the term symmetry we mean
a bijection g of the space of boundary conditions of a given automorphism type g such
that the annulus amplitudes coincide, i.e.
A g Ž r . g Ž rX . Ž t . s A r rX Ž t . Ž 6.12 .
X
for all r , r with automorphism type g r s g s g r X .
Let us study the presence of such a symmetry first in the example of the critical
three-state Potts model. It is known w29x that the duality symmetry of this model maps
the free boundary condition to any of the three fixed boundary conditions, and indeed
this is a specific case of the general duality w30x between free and ‘configurational’
boundary conditions of lattice spin models. Similarly, the new boundary condition
discovered in w29x gets mapped to any of the three mixed boundary conditions. In this
case the symmetry group H s Z 3 of the fixed or mixed boundary conditions is directly
inherited from the lattice realization of the Potts model. We also observe that precisely
those boundary conditions for which this Z 3-symmetry is spontaneously broken are
grouped in non-trivial orbits. Furthermore, one readily checks that for orbits that are
related by the T-duality p T) s sc ) , the sum rule
Npy1
)
T Ž r.
Npy1
) X
T Žr .
Ý A gp T) Ž r . g X p T) Ž r X . s Nry1 Nry1
X Ý A g Ž r . g XŽ rX . Ž 6.13 .
X
g ,g gH g ,g XgH
for the annulus coefficients holds, where Nr is the order of the stabilizer of the H-action
on r . Roughly speaking, the sum rule Ž6.13. tells us that T-dual orbits give rise to an
equal number of open string states on the boundary.
This pattern can be detected in various other examples as well. For instance, in the
theory of a single uncompactified free boson, there is a single Neumann boundary
condition, whereas the Dirichlet boundary conditions are labeled by a position in R,
which should be interpreted as the affine space over the group H s R of translations.
This group is spontaneously broken for Dirichlet boundary conditions, and it is
straightforward to check that relations Ž6.12. and Ž6.13. are satisfied in this case as well.
Another class of examples is provided by boundary conditions in WZW theories that
break the bulk symmetry via an inner automorphism of the underlying simple Lie
algebra g. In these cases the boundary conditions of each automorphism type are labeled
by the same set, namely by the primary labels of the original theory Žsee w5x for the case
of automorphisms of order two.. The group H is in this case realized by the action of
simple currents of the original theory, which account for the different possible choices of
the shift vector that characterizes the inner automorphism of g. Again the validity of
relations Ž6.12. and Ž6.13. is easily verified.
We would like to emphasize, though, that the existence of such T-duality relations is
in fact a quite special feature of an individual model. In general we do not expect that
boundary conditions of different automorphism type are related in such a manner. For
instance, in the case of boundary conditions of WZW theories that break the bulk
symmetry via an outer automorphism of g, no relations of the form Ž6.12. or Ž6.13. are
known to us.
7. Boundary homogeneity
In this section we exhibit another general aspect of the space of conformally invariant
boundary conditions that preserve only a subalgebra A of A. Namely, we show that the
orbifold group G is realized as a group of symmetries on the space of boundary
conditions. These symmetries permute the boundary conditions within each set
w r , cˆr x < cˆr g Ur) 4 with fixed G-orbit w r x; the permutation is the same for all G-orbits.
This behavior, to which we refer as boundary homogeneity, is similar to the so-called
fixed point homogeneity that is present in simple current extensions, and indeed the
arguments closely resemble the ones needed in the latter context w4x.
We start from the observation that the orbifold group G can be identified with the
dual G ) of the simple current group G s G ) , and consider some character C of G . For
every field r of the A-theory and every character cˆ g Ur) we then define a new
character C cˆ g Ur) by
C
cˆ Ž J . :s C Ž J . cˆ Ž J . Ž 7.1 .
for all J g Ur . This is indeed again a character of Ur . Moreover, manifestly the group
law of G is reproduced for different choices of C , so that our construction provides an
action of G on the group Ur) . Typically G does not act freely, but it does act
transitively.
The next step is to realize that this prescription supplies us with a well defined action
on the space of boundary conditions. This is not entirely trivial, because the labels for
the boundary conditions are not pairs of orbits, but rather are obtained by the equiva-
lence relation ŽA.9. of Ref. w1x, i.e. Ž l, cˆl . ; J Ž l, cˆl . ŽJwl, J cˆl ., which involves a
non-trivial manipulation of the characters. However, this complication does not do any
harm, because just as for the action of C g G ) it consists of a multiplication, so that
the two operations commute. We write
C
r :s r ,Ccˆ for r s r , cˆ . Ž 7.2 .
X
Similarly, when we extend by some smaller group G ; G , we can define the analogous
object C w r , cˆ X x X :s w r ,Ccˆ X x X .
Let us now explain in which sense the elements of G are to be regarded as
symmetries. Using the explicit expression Ž6.29. of Ref. w1x for the annulus coefficients,
we can establish the identity
X X
ˆ ˆ
ACC wwsr 1,,ccˆs1xxC w r 2 , cˆ 2 x s Awwsr ,1c, csˆ 1x x w r 2 , cˆ 2 x Ž 7.3 .
between annulus coefficients. Thus when we also act on the corresponding chiral labels
of the open string states in the annulus amplitude, we can absorb the transformation into
a relabelling of the summation variables, so as to conclude that the annulus partition
function is invariant. We expect that this extends indeed to a full-fledged symmetry at
the level of correlation functions, where again one has to act on the insertions on the
boundary.
As a consequence, the boundary conditions that correspond to one and the same
G-orbit w r x should in fact better be labeled by the elements of what may be called the
affinum over the character group Ur) rather than by Ur) itself. Here by the term
affinum over a group G we refer to the elementary structure of a set A G that carries a
free and transitive action of G.9 Conversely, the group G can be identified with the
quotient of A G = A G by the equivalence relation Ž p,q . ; Ž p h ,q h . for all h g G. For
every p g A G we are given an identification g l p g between the group G and its
affinum A G , but there does not exist any canonical identification. Roughly speaking, in
the structure of the affinum one ignores the special role played by the identity element;
thus an affinum and its group are related in much the same way as an affine space AV is
related to the corresponding vector space V, which can be identified with its group of
translations. Indeed, the group of automorphisms Žthat is, bijections intertwining the
action of G. of an affinum A G is precisely G. It follows that any two affina over G are
isomorphic; but the isomorphism is never canonical; it is always determined only up to
the isomorphism group G.
We should admit that, even though we avoided using the term, the concept of an
affinum is already implicit at several other places of our discussion of boundary
conditions. For instance, the quantities cl and cˆr that appear in the definition ŽA.6. of
the matrix S˜ are best regarded as elements of the affina of the respective character
groups, because w31x the matrices S J are only defined up to certain changes of basis in
the space of one-point blocks on the torus and because such a change amounts to a
relabelling of the characters.
We finally add a comment on how this symmetry property of the boundary conditions
looks like in the case of the three-state Potts model. In this case it exchanges
simultaneously two fixed and two mixed boundary conditions, while it leaves the third
fixed and mixed boundary conditions invariant. In the Potts model we actually have yet
another symmetry on the space of boundary conditions, the action of a Z 3 group. These
two symmetries combine to the symmetric group S3 .
8. The universal classifying algebra
The decomposition Ž5.7. of the classifying algebra C Ž A . into individual classifying

algebras for fixed automorphism types implies that for every subgroup H of the orbifold
group G one has C Ž A H . : C Ž A G .. Indeed, C Ž A H . is just the direct sum of the ideals
C Ž g . Ž A . : C Ž A G . for all g g H. It is also known w8x that the mapping H ¨ A H
provides a bijection between the set of subgroups of G and the set of consistent chiral
9
Thus there exists a map A=G ™ A acting as Ž p, g . ¨ p g such that p g s p if and only if g s e,
p s Ž p g . h , and such that for each pair Ž p,q . g A= A there is a unique g g G with p g s q. When G is
gh
non-abelian, one must distinguish between left and right actions, and hence left and right affina.
subalgebras of A that contain A G.10 In our situation the orbifold group G is finite and
abelian, but the latter result continues to hold for all finite groups, and there is even a
version for compact Lie groups w9,33,34x.11
In this section we would like to address the issue of consecutive breakings of bulk
symmetries in more generality, which leads us in particular to introduce the notion of a
uniÕersal classifying algebra. While our detailed studies have so far been restricted to
cases where the preserved subalgebra A of the bulk symmetries satisfies A s A G for
some finite abelian group G, it is reasonable to expect that several features of our
analysis will persist for general A. In particular, it should again be possible to determine
a classifying algebra, provided that the following two pieces of information are
available:
1. the decomposition of A-modules into direct sums of irreducible A-modules;

2. concrete expressions relating the bundles of chiral blocks of the A-theory with those
of the A-theory.
We also expect that the statements about inclusions of classifying algebras valid for
the case of finite abelian orbifold groups generalize as follows. To every inclusion
A ® A of preserved bulk symmetry algebras there is associated a projection of the
corresponding classifying algebras; the classifying algebra for A is a suitable quotient
of the one for A. More generally, for every chain of inclusions
A ® AX ® A Ž 8.1 .
of symmetry algebras, there should exist a corresponding chain of projections
p pX
C Ž A . ∏ C Ž AX . ∏ C Ž A . Ž 8.2 .
for the associated classifying algebras. As a consequence, every irreducible representa-
tion of C Ž AX . gives rise to an irreducible representation of C Ž A .. This makes sense
indeed: an irreducible representation of C Ž AX . corresponds to a boundary condition that
preserves AX and thus, a fortiori, also preserves the smaller algebra A; it should
therefore correspond to some irreducible representation of C Ž A .. Relation Ž8.2. clearly
holds when both A s A G and AX s A H are orbifold subalgebras for abelian orbifold
groups with H : G. ŽMoreover, the projections are compatible with the distinguished
bases of the algebras, compare e.g. the arguments leading to formula Ž5.18. for the case
H s ² g :.. More generally, one can hope to obtain this way also quantitative informa-
tion on solvable orbifold groups.
This way the following picture emerges. The set M of all consistent subalgebras of a
given chiral algebra A that possess the same Virasoro element as A is partially ordered
10
A similar Galois correspondence has been established in the context of braided monoidal )-categories in
w32x.
11
Assuming that the statement holds for every group within a certain class X, it follows in particular that
whenever there exists at least one boundary condition that does not possess an automorphism type, then the
Virasoro algebra by itself cannot be the orbifold subalgebra of A with respect to any group that belongs to X.
by inclusion. It is reasonable to expect that M is even an inductive system; that is, given
any two consistent subalgebras A 1 and A 2 of A, one can find another consistent
subalgebra A 3 that is contained in their intersection,
A 3 : A1 l A 2 . Ž 8.3 .
Note that this implies in particular that we do not have to make the assumption that the
intersection of all consistent subalgebras of A contains a consistent subalgebra; rather,
one only needs to deal with intersections of finitely many subalgebras.
Assuming that also in general the inclusion A 1 ; A 2 implies that the classifying
algebra for A 2 is a quotient of the one for A 1 , one arrives at a projective system Ž A i .
of classifying algebras. Taking the projectiÕe limit over this system, we then arrive at a
uniÕersal classifying algebra
C ` :s lim A i . Ž 8.4 .
§
ŽThe projective limit of the closely related structure of a fusion ring has been studied in
w35x.. By construction, this algebra C ` governs all conformally invariant boundary
conditions. In other words, it is the classifying algebra C Ž V ir. for the case where the
preserved subalgebra just consists of the Virasoro algebra. The universal classifying
algebra can be found explicitly in simple models, e.g. for the free boson compactified on
a circle or for the Z 2-orbifold of these theories Žsee Subsection 11.1..
The construction of a surjective homomorphism p : C Ž A . ™ C Ž AX . that maps the
elements of the distinguished basis B of C Ž A . to elements of the distinguished basis
BX of C Ž AX . is in fact straightforward as far as the l-part of the labels Ž l, w . for B are
concerned. In contrast, concerning the w-part one faces complications which stem from
the absence of a simple relationship between UlX and Ul Žthis fact had also to be taken
into account in the manipulations that were necessary to establish integrality of the
annulus coefficients, see Subsection 6.4 of w1x.. As a matter of fact, for a complete
discussion of this issue even in the case of abelian orbifold subalgebras additional
simple current technology is required that goes beyond the results of w4x. In particular it
will be necessary to implement the powerful results that have recently been obtained in
w36x.
We also note that for consistency, along with the projection p there should come an
injection i from the set of boundary conditions that preserve AX to those that preserve
A, in such a way that the diagonalizing matrices for C Ž A . and for C Ž AX . are related as
S˜Ž l , w ., i Žw r , cˆ xX . A S˜pX ŽŽ l , w .. ,w r , cˆ x X . Ž 8.5 .

Again the explicit construction of i proves to be difficult; it will not be pursued here.
9. Involutary boundary conditions
In this section we focus our attention to the situation where the symmetries A s A G
that are preserved by the boundary conditions form a subalgebra that is fixed by an
involution √ . In other words, for such inÕolutary boundary conditions the orbifold
group G is just the Z 2 group consisting of √ and the identity. For the associated Z 2
orbifolds, a lot of information is available Žsee e.g. Refs. w5,37x. The vacuum sector of
the A-theory decomposes into subspaces as
HV ( HV [ HJ Ž 9.1 .
with J a simple current of order two, and the automorphism √ acts as
√ < HV s id HV , √ < H J s yid H J . Ž 9.2 .

Involutary subalgebras are very special indeed. Several of the structures that required a
detailed discussion in the general case are realized rather trivially here. For instance, as
cyclic groups have trivial second cohomology, all untwisted stabilizers are equal to the
full stabilizers; this already simplifies various equations considerably.
The reason why we nevertheless study this simple situation in quite some detail is
that it is realized in various interesting systems. We will soon display several of these
examples, but first we summarize some generic features of the Z 2 case; in particular we
will study the individual classifying algebra for automorphism type √ .
9.1. EÕen and odd boundary conditions
In the case at hand, the orbifold group G and the simple current group G s G ) are
both isomorphic to Z 2 . Thus in particular the exponentiated monodromy charge g takes
its value in Z 2 , so that there are just two automorphism types of boundary conditions.
We refer to those boundary conditions whose automorphism type is given by the identity
as eÕen boundary conditions while those with automorphism type √ will be called odd.
The Chan–Paton types for the even boundary conditions are labeled by the primary
fields l of the A-theory, while those for the odd boundary conditions are labeled by the
orbits of primary fields of the orbifold theory Žwhose chiral algebra is A s A Z 2 . with
monodromy charge Q J s 1r2. The simple current group is G s V ,J4 , so in particular
its orbits w lx either have length two Ži.e. have stabilizer Sl s V 4. or length one Ži.e. are
fixed points, with Sl s Z 2 .. Only fields l with vanishing monodromy charge can be
fixed points; therefore fixed points cannot give rise to odd boundary conditions.
The even boundary conditions preserve, of course, all bulk symmetries. The relevant
boundary blocks e Bl can therefore be expressed in terms of the boundary blocks of the
A-theory as
e
1
Bl s žB [ B / ,
l Jl Ž 9.3 .
< Sl <
and the Ward identities that come from a field Y in the chiral algebra A look like
e
Bl ( Ž Yn m 1 q z Y 1 m Yyn . s 0 Ž 9.4 .
with z Y s Žy1. DYy1 . As for any boundary conditions that preserve all of A, the
classifying algebra C Ž A . for the even boundary conditions is the fusion algebra of the
A-theory, with structure constants expressible via the Verlinde formula in terms of the
modular transformation matrix S of the A-theory.
As established in Subsection 4, in the case of odd boundary conditions, the Ward
identities Ž9.4. get replaced by those for twisted boundary blocks; here they read
o
Bl ( Ž Yn m 1 q z Y 1 m √ Ž Yyn . . s 0 . Ž 9.5 .
o
The odd boundary blocks Bl satisfying these constraints are ‘differences’
o
Bl s Bl [ Ž yB J l . Ž 9.6 .
of the boundary blocks of the A-theory. They are thus related to the ordinary boundary
blocks as in formula Ž3.20., i.e. we have
o
Bl se Bl ( Ž Q√ m id . , Ž 9.7 .
where the maps Q√ satisfy the √-twisted intertwining property Q√ Y s √ Ž Y . Q√
Žwhich together with their action on the highest weight vector determines them
uniquely..
9.2. The classifying algebra
Let us now display the total classifying algebra C Ž A . which governs even and odd
boundary conditions simultaneously. We already know what the labels for the basis of
C Ž A . and for its one-dimensional irreducible representations look like. Moreover, in the
formula ŽA.6. for the diagonalizing matrix S, ˜ in the Z 2 case only two different matrices
appear, namely the modular S-matrix S ' S V of the orbifold theory and the matrix
˘ S J for the simple current J. The length Nl of an orbit of J can be either one or two;
S:s
it will be convenient to use different symbols for labels for fixed points and those for
length-two orbits; we choose roman letters f, g, . . . for the former and greek letters
a , b , . . . from the beginning of the alphabet for the latter. Also, for simplicity we will
use one and the same symbol c to refer to a G-character and to its value on the
non-trivial element J g G , which can be either of "1. For the subsets of the whole set
I s l4 of primary labels of the A-theory that consist of the labels for full orbits and for
fixed points we write I( and If , respectively, i.e.
I( :s m g I < Nm s 2 4 , If :s m g I < Nm s 1 4 . Ž 9.8 .
We also introduce the corresponding subsets
I( :s a g I < a s a ,J a 4 g I( 4 , If :s f < Ž f , c . g If 4 Ž 9.9 .
of the label set I of the orbifold theory. With these notations the dimension of C Ž A .
reads dim C Ž A . s < I( < q 2 < If <. By the sum rule ŽA.7. this must equal the number of
boundary conditions, i.e. the number of Z 2-orbits of the A-theory, counted with their
stabilizer:
dim C Ž A . s 12 < I( < q 12 < I1r2 < q 2 < If < . Ž 9.10 .
This tells us that for every Z 2-orbifold the number < I( < of length-2 Q s 0 orbits
coincides with the total number < I1r2 < of Q s 1r2 orbits. On the other hand the number
< If < of fixed points, which necessarily have Q s 0, can be arbitrary. Also note that
dim C Ž A . ' < I < s < I( < q < If <, i.e. there are always at least as many even as there are odd
boundary conditions. The numbers of odd and even boundary conditions are equal
precisely in those cases where there are no fixed points, which happens precisely when
the associated automorphism of the fusion rules is the identity.
Further, the entries of the diagonalizing matrix S˜ read explicitly
Sã ,w r x s 2 Sa , r , S˜Ž f , c .,w r x s S f , r for Nr s 2 ,
Sã ,w g , c X x s Sa , g , S˜Ž f , c .,w g , c X x s 12 S f , g q cc X S˘f , g .

ž / Ž 9.11 .
This leads to the formulæ
˜ a , b ,g s 2 Na , b ,g ,
N ˜ a , b ,Ž f , c . s Na , b , f ,
N
˜ a ,Ž f , c .,Ž g , c X . s 12 Na , f , g q cc X N˘ f , a , g se Na ,Ž f , c .,Ž g , c X . ,
N ž /
˜ Ž f , c .,Ž g , c X .,Ž h , c XX . se NŽ f , c .,Ž g , c X .,Ž h , c XX .
N Ž 9.12 .
e
for the structure constants with three lower indices. Here N and N denote the fusion
coefficients of the A-theory and of the A-theory, respectively. Moreover, we have
˘
introduced twined fusion coefficients, defined with the help of the twined S-matrix S,
according to
S˘f , h Sr , h S˘g), h
N̆ f , r , g :s Ý . Ž 9.13 .
hgIf SV , h
Recall from Subsection 5.3 that these twined fusion coefficients are the traces of the
action of the outer automorphisms associated to J on bundles of chiral blocks.
Furthermore, from equation Ž9.12. we read off that the matrix C B s SS ˜˜t furnishes a
conjugation on the basis labels Žnot just an involution as in the generic case.. We have
CaB, b s 2 da , bq , CaB,Ž f , c . s 0 , CŽBf , c .,Ž g , c X . s d f , gq C hB,ŽcfX . ; Ž 9.14 .

in particular, conjugation on the fixed points is exactly as in the A-theory. It follows that
the structure constants of the classifying algebra read
˜ a , b g s Na , b g ,
N ˜ a , b Ž f , c . s Na , b f ,
N
X X
˜ a ,Ž f , c . g s 12 e Na ,Ž f , c . g ,
N ˜ a ,Ž f , c . Ž g , c . se Na ,Ž f , c .
N Ž g ,c .
,
XX XX
˜ Ž f , c .,Ž g , c X . g s 12 e NŽ f , c .,Ž g , c X . g ,
N ˜ Ž f , c .,Ž g , c X . Ž h , c . se NŽ f , c .,Ž g , c X . Ž h , c . .
N
Ž 9.15 .
9.3. The indiÕidual classifying algebra for odd boundary conditions
The basis F˜ Ž l , c . < l g I( , c g Sl4 of the classifying algebra C Ž A . is mapped by

Ž5.1. to the distinguished basis of the fusion algebra of the A-theory. In the present
situation we find
Fa s 12 Fã q F˜ J a
e
ž / and F Ž f , c . s F˜ Ž f , c . .
e
Ž 9.16 .
As we have seen, this provides an algebra homomorphism w from the classifying
algebra C Ž A . to the fusion algebra of A, which is just the classifying algebra C Ž A . for
the even boundary conditions, i.e. C Ž A . se C Ž A . s span C eFl4 . On the other hand, the
kernel of w is an ideal of C Ž A . which provides us with a classifying algebra
o
C Ž A . ' C Ž√ . Ž A . for the odd boundary conditions.
Let us analyze this classifying algebra o C Ž A . in more detail. Only G-orbits of length
two contribute. On each such orbit we choose one distinguished representative a . We
can and will assume that these choices are made in such a manner that, first, for the
vacuum orbit the orbifold vacuum V is taken as the representative, and second, on
conjugate orbits one chooses conjugate representatives. Then the set of elements
o
Fa :s 12 Fã y F˜ J a ,
ž / Ž 9.17 .
o
with a the distinguished representative, form a basis for C Ž A .. In this basis the
structure constants of o C Ž A . are given by the following traces on spaces of chiral
blocks Žwhich are integers.:
o
Na , b g :s Na , b g y NaJ , b g . Ž 9.18 .
As an ideal of C Ž A ., the algebra o C Ž A . inherits several properties of C Ž A .: it is
semisimple, commutative and associative; it is unital, the unit element being oFV s F˜ V
y F˜ J ; and it has a conjugation which is evaluation at the unit element. The dimension
of o C Ž A . is
dim o C Ž A . ' < I( < s < I1r2 < . Ž 9.19 .
o
Since C Ž A . is semisimple, it must possess a diagonalizing matrix. Indeed, by the
Verlinde formula for the orbifold fusion rules Na , b g we can write the structure constants
Ž9.18. as
o
Sa , m Sb , m
Na , b g s Ý žS ) )
g , m y SJg , m /. Ž 9.20 .
mgI SV , m
Here a priori the summation is over all sectors of the orbifold theory. But only the
twisted fields give a non-vanishing contribution, and in that case the two terms with g
and Jg are equal. Labeling these twisted fields by roman letters a,b, . . . and calling the
corresponding index set I1r2 , we thus obtain o Na , b g s 2Ý d g I1r 2 Sa , d Sb , d Sg), drSV , d .
Moreover, elements d on the same Z 2-orbit give identical results, hence we may rewrite
this formula as a sum over orbits d:s d,Jd4 . Denoting the set of these orbits by I1r2 , so
that
I1r2 :s a g I < a s a,Ja4 g I1r2 ,
½ 5 Ž 9.21 .
we arrive at
o
Sa , d Sb , d Sg), d
Na , b g s 4 Ý . Ž 9.22 .
dgI1r2 SV , d
We can interpret this result as stating that the structure constants o N are governed by the
matrix o S with entries
o
Sa , b :s 2 Sa , b with a g I( , b g I1r2 ; Ž 9.23 .
this is the diagonalizing matrix for o C Ž A .. Also note that owing to o SV , b s 2 SV , b ) 0
for all b g I1r2 , this matrix shares the positivity property of modular S-matrices Žthis
does not already follow from the commutativity and semisimplicity of o C Ž A ...
Moreover, by combining unitarity of S with the simple current symmetry SJm, n s
Žy1. 2 Q J Ž n.Sm, n one obtains
Ý S a),l S a , m s 12 Ž d l , m y d l ,Jm . , Ý Sn),l Sn , m s 12 Ž d l , m q d l ,Jm . ,

agI1r2 ngI( jIf
Ž 9.24 .
with the help of which one can show that the matrix Ž9.23. is unitary, which in turn tells
us once again that o S is a square matrix.
The structure constants o Na , b g as defined by Ž9.18. do depend on the choice of
representatives that has been made. Indeed, upon replacing g by J g , o Na , b g goes to
minus itself. But this does not change the classifying algebra o C Ž A ., since the choice of
a sign constitutes a one-cocycle on o C Ž A ., which in turn can be absorbed by choosing a
correlated sign for the boundary blocks. More concretely, for the full Z 2-orbits we have
the isomorphism Ha ( Ha [ HJ a ( H√ w a of A-modules, so that the associated even
boundary blocks are precisely e Ba s Ba [ B J a , as given by formula Ž9.3., where
B m : Hm m Hm q™ C are the orbifold boundary blocks. In contrast, there are two
possible choices of odd boundary blocks, namely o Ba s Ba [ ŽyB J a . as in Ž9.6. as well
as
o
B√ w a s B J a [ Ž yBa . s yo Ba . Ž 9.25 .
o o
By construction both of these linear forms Ba and B√ w a on Ha m Haq satisfy the
appropriate Ward identities, but of course we must keep just one out of the two. The
right prescription is to keep o Bb when the label b g I( is the chosen representative for
the orbit b g I( . ŽAlso, positivity of mixed annulus amplitudes is guaranteed only with
this choice, see below.. In short, the label of the orbifold boundary state that appears
with a positive sign in the boundary block Ž9.6. is the one that is to be chosen as the
representative of the orbit.
9.4. Annulus amplitudes
Using our general results from Section 6 of w1x it is also straightforward to calculate
all annulus amplitudes. For the case of two even boundary conditions we obtain
a
Aa b s Ý žN b ,m q Nb ,Jm a xm Ž i tr2 . ,
/
mgI(jIf
A a Ž f ,c . s Ý N f , m a xm Ž i tr2 . ,
mgI(jIf
1
A Ž f ,c . Ž g ,c X . s Ý žN g ,m
f ˘ g ,m
q cc X N f
/ x Ž i tr2. .
m Ž 9.26 .
2 mgI(jIf
Similarly, for two odd boundary conditions the annulus amplitudes are
1
A a bŽ t. s Ý N N a q Nb ,Jm a xm Ž i tr2. .
2 mgI jI m b , m ž / Ž 9.27 .
( f
Finally, for mixed annuli, i.e. annuli with one even and one odd boundary, we find
1
A a mŽ t . s Ý Nm Nm ,c a q Nm ,Jc a xc Ž i tr2 . .
ž / Ž 9.28 .
2 cgI1r2
Thus in particular we correctly establish the Z 2 selection rule

Asr 1 r 2 s 0 whenever QJ Ž r 1 . q QJ Ž r 2 . q QJ Ž s . g Z q 1r2 , Ž 9.29 .
which is analogous to the selection rule that is valid for the orbifold fusion rules.
For every triple r 1 r 2 , s the annulus coefficient Asr 1 r 2 is manifestly a non-negatiÕe
integer, in agreement with the physical meaning of the annulus amplitude as a partition
function. We would like to stress that this coefficient is not only a non-negative integer,
but in addition has a natural representation theoretic interpretation, namely as the
dimension of a space of chiral blocks. The expressions N g , m f " N ˘ g , m f , for example,
w x
that appear in the last line of 9.26 , are equal 38 to the ranks of two invariant
Ž .
subbundles of bundles of chiral blocks. This observation also seems to be relevant for a
better understanding of the multiplicities of boundary fields which are counted by these
dimensions Žthat is, there is a separate boundary field Csr 1 , r 2 ; l for every l s
1,2, . . . ,Asr 1 r 2 . and for which a satisfactory understanding is unfortunately so far still
lacking. Inspecting the calculations more closely, one also observes that the coefficients
Aca m check the correctness of the sign convention in the definition of the odd boundary
blocks that was discussed after formula Ž9.25.; with a different prescription, some of
these coefficients would become negative.
With the results above, it is easily verified that the annulus coefficients Asr 1 r 2 satisfy
the relations that are expected on the basis of factorization arguments,12 i.e. that they are
‘associative’ in the sense that
sq q
Ý Asr 1 r 2 A r 3 r 4s Ý Asr 1 rq
3
Asrq2 r 4 Ž 9.30 .
s s
for all possible values of the r i , and that they are ‘complete’ in the sense that
A r1 A r 2 s Ý Mr1 , r 2 r3
A r3 , Ž 9.31 .
r3
where the A’s are regarded as matrices in their two lower indices. The non-vanishing
coefficients M are
Ml , m n se Nl , m n ,
M a , b n s 12 Nn N a , b n q N a ,Jb n ,
ž /
2
M a , m c s 14 Ž Nm . žN a,m
c
q NJa , m c .
/ Ž 9.32 .
Note that the matrix M V is the identity matrix.
12
As already pointed out in Subsection 6.5 of w1x for the case of general orbifold group G, a rigorous
derivation of these relations is, however, not yet available.
10. Examples for involutory boundary conditions
We now present several classes of examples with involutory boundary conditions.

We do not intend to be exhaustive, but concentrate on particularly interesting conformal
field theories.
10.1. Dirichlet and Neumann conditions for the free boson
A simple and well-known realization of involutory boundary conditions is encoun-

tered for the c s 1 theory of a single free boson, compactified on a circle of radius R
s '2 N , where N is a positive integer – respectively, at the T-dual radius TŽ R . s 2rR
'
s 2rN N . This theory has 2 N primary fields, which at the chiral level we label by
integers l mod 2 N ; their u Ž1. charge modulo '2 N is ql s lr '2 N . At radius '2 N
the theory has diagonal torus partition function, i.e. the primaries are of the form Ž l, l.,
while for radius 2rN' N one deals with the charge conjugation invariant, i.e. with
primaries Ž l,y l.. We will follow our general convention to describe the boundary
conditions for the case of the charge conjugation invariant; the corresponding results for
the diagonal invariant can be deduced via T-duality, as explained in Section 6.
The chiral algebra A of the free boson theory consists of operators of the form
P Ž E X . exp Ž i n'2 N X . , Ž 10.1 .
where n g Z and P Ž E X . is a Žnormally ordered. polynomial in the u Ž1. current
j s i E Xr '2 N . This algebra has an obvious involutory automorphism, which in terms
of the Fubini–Veneziano field X is expressed as
X ¨ yX Ž 10.2 .
and has the physical meaning of charge conjugation. It maps the u Ž1. current to minus
itself and exchanges the fields expŽ"i n'2 N X .; thus its fixed point algebra consists of
the even polynomials in j combined with the operators cosŽ n'2 N X . with n g Z and
odd polynomials in j combined with sinŽ n'2 N X .. This algebra A is just the chiral
algebra of the Z 2-orbifold of the free boson.
We recall w37x that this Z 2-orbifold has N q 7 primary fields. First, one has the
vacuum V and a simple current J of conformal weight DJ s 1, which comes from the
u Ž1. current of the A-theory. Besides these two fields there is one other length-two orbit
c 1, c 2 4 of monodromy charge zero Žcoming from the self-conjugate field with l s N
of the A-theory., as well as N y 1 fixed points w l with l s 1,2, . . . , N y 1, and
finally there is one pair of twisted fields for each of the two self-conjugate fields of the
A-theory, which are denoted by s , s X 4 and t ,t X 4 .
The number of even boundary conditions is equal to the number 2 N of primary
fields of the boson theory. According to our general prescription they are labeled by the
orbits of the orbifold and the characters of their stabilizers, i.e. there is one even
boundary condition for each of V ,J4 and c 1 , c 2 4 and two for each of the fixed points
wl. ŽIn the language of the circle theory, the latter correspond to the two primary fields
of opposite charge "l.. There are just two odd boundary conditions, corresponding to
the two J-orbits of twisted fields. The classifying algebra o C Ž A . for the odd boundary
conditions turns out to be just the group algebra of Z 2 , and the total classifying algebra
C Ž A . is isomorphic to the direct sum of this CZ 2 and the fusion rule algebra CZ 2 N of
the boson theory.
As is already apparent from Ž10.2., the odd boundary conditions are nothing but
Neumann conditions for the free boson X Žwith definite rational values of the Wilson
line., while the odd ones are Dirichlet conditions, with definite rational values of the
position of the D0-brane, namely at j R with j any of the 2 N th roots of unity. ŽRecall
that this formulation refers to the case where the torus partition function is the charge
conjugation invariant; for the true diagonal invariant, in agreement with T-duality the
role of Dirichlet and Neumann conditions get interchanged.. The fact that the Wilson
line Žfor Neumann conditions. respectively the position of the brane Žfor Dirichlet
conditions. are restricted to a discrete set of values is of course a consequence of the fact
that in the situation considered here the preserved bulk symmetries correspond to a
rational conformal field theory. By breaking more bulk symmetries one arrives at more
general possibilities. In particular, as will be seen in Subsection 11.1 below, one may
consider orbifolds that correspond to a change of the radius of the circle and thereby
arrive at Wilson lines and brane positions at arbitrary points of the circle.
10.2. s l (2) WZW theories
Another example is given by s l Ž2. WZW theories at levels k g 4Z. In this case the
full conformal field theory based on the diagonal modular invariant for the orbifold
theory can be realized as a sigma model on the group manifold SUŽ2., while the
diagonal modular invariant for the A-theory corresponds to a sigma model on the
non-simply connected group manifold SOŽ3.. In the usual notation w39x these are the
theories labeled A kq 1 and D k r2q2 , respectively.
The non-trivial simple current J g G has conformal weight DJ s kr4. Labeling the
sectors of the A kq 1 model by their highest s l Ž2.-weights and taking as representatives
of Z 2-orbits those with smaller weight, the various label sets look as follows:
I( s 0,2,4, . . . , 2k y 2, 2k q 2, 2k q 4, . . . ,k 4 , If s 2k 4 ,
I1r2 s 1,3,5, . . . ,k y 1 4 ,
I( s 0,2,4, . . . , 2k y 2 4 , If s Ž 2k , c . < c s "1 4 ,
I1r2 s 1,3,5, . . . , 2k y 1 4 . Ž 10.3 .
Thus < I( < s < I1r2 < s kr4 and < If < s 2.
For the modular matrix S 'eS of the D k r2q2-model, a natural ordering of the labels
in I ' I( j If is
m s 0,2,4, . . . , 2k y 2, Ž 2k ,q . , Ž 2k ,y . , Ž 10.4 .
both for rows and for columns. With this labeling the general formula for the modular
S-matrix of a simple current extension Žestablished in w4x and displayed also in Appendix
A of w1x. yields
°2 S m ,n for m , n s 0,2, . . . , 2k y 2 ,
S m ,n for m s 0,2, . . . , 2k y 2, n s Ž 2k , c .
e ~
Sm , n s orm s Ž 2k , c . , n s 0,2, . . . , 2k y 2 , Ž 10.5 .
1
q cc X I for m s Ž 2k , c . , n s Ž 2k , c X . .
¢
1
2
ž' kr2 q 1 /
Here S is the S-matrix of the A kq 1-model, i.e.
(
S m , n s 2r Ž k q 2 . sin Ž Ž m q 1 . Ž n q 1 . pr Ž k q 2 . . ,
and we introduced the number
1 for k g 8Z ,
I:s ½ i ' 'y 1 for k g 8Z q 4 ,
Ž 10.6 .
which is nothing but the Žone-by-one. matrix S J. In contrast, for the diagonalizing
matrix o S of o C Ž A . there will typically not exist any preferred ordering of the labels
a g I( Žfor the rows of o S . and a g I1r2 Žfor the columns.; amusingly, in the present
case it is possible to order rows and columns in such a fashion that o S is symmetric. We
have
8 Ž a q 1. Ž b q 1. p
o
Sa , b s 2 Sa , b s ( kq2
sin
ž kq2 / . Ž 10.7 .
Now ordering again I( according to the value of the weight, i.e. as a s 0,2,4, . . . ,kr2
y 2, symmetry is achieved when I1r2 is ordered by taking first the weights of the form
4 j q 1 in ascending order and afterwards the weights of the form 4 j q 3 in descending
order, i.e. a s 1,5,9,13, . . . ,15,11,7,3. Doing so, the matrix Ž10.7. simply becomes the
matrix with entries
& 8 Ž 2 p y 1. Ž 2 q y 1. p
o
Sp , q s ( kq2
sin
ž kr2 q 1 / , Ž 10.8 .
where the integers p and q run from 1 to kr4. In particular, inspection shows that
o
C Ž A . coincides with the fusion algebra w40x of the Ž 2k q 1,2. Virasoro minimal models.
Via the simple current symmetries of the matrices S and S J, the diagonalizing matrix
˜
S of the classifying algebra is already completely determined by eS and o S. Concretely,
when we choose the ordering of the rows of S˜ as
Ž l , w . s 0 , 2 , 4 , . . . , 2k y 2 , Ž 2k ,q . , Ž 2k ,y . , k , k y 2 , . . . , 2k q 2 Ž 10.9 .
and the ordering of the columns according to
Ž r , c . s 0 , 2 , 4 , . . . , 2k y 2 , Ž 2k ,q . , Ž 2k ,y . , 1 , 3 , . . . , 2k y 1 , Ž 10.10 .
then S˜ is block diagonal of the form
e eo
S S
S˜s ž oe o /, Ž 10.11 .
S S
and the off-diagonal blocks are related to the diagonal ones by
eo
S < full s yo S , eo S < fixed s 0 , oe S seS < full , Ž 10.12 .
< <
where the symbols full and fixed stand for restriction of the rows to those corresponding
to full orbits and to fixed points, respectively.
10.3. Relation with incidence matrices of graphs
In the s l Ž2. case under consideration, it is not too difficult to establish that – just
like the algebra o C Ž A . – also the total classifying algebra C Ž A . constitutes a structure
that has been encountered in conformal field theory before. Indeed, we will construct an
isomorphism between C Ž A . and an algebra that appears in the work of Pasquier et al.
To simplify some of the formulæ below, let us write k s 4 l with l g Z. Then a
2 l q 2-dimensional associative algebra with structure constants
ˆ r , s t :s Ý Sˆr ,u Sˆs,u Sˆt),urSˆ1,u

N Ž 10.13 .
u
for r, s,t g 1,2, . . . ,2 l q 24 has been considered in w41–43x and been called the
Pasquier algebra associated to the situation of our interest. In formula Ž10.13., Sˆ is the
matrix with entries
° ž'1
2
1
qI
/ for r s 2 l q 1 , s s l q 1 or r s s s 2 l q 2 ,
2 lq1
1 1
2 ž' 2 lq1
yI
/ for r s 2 l q 1 , s s 2 l q 2 or r s 2 l q 2 , s s l q 1 ,
~' l
Ŝr , s :s
1
2 q1
Ž y1 . Ž ry 1 . r2 Ž 1 y Ž y1 . .
r
for r s 1 ,2 , . . . ,2 l , s s l q 1 ,2 l q 2 , Ž 10.14 .
1
for r s 2 l q 1 ,2 l q 2 , s / l q 1 ,2 l q 2 ,
'4 l q 2
1 Ž2 l y r q 1 . Ž 2 s y 1 . p
¢' 4 lq2
P 2 cos
ž 4 lq2 / otherwise ,
with I as in Ž10.6.. Note that the matrix Nˆ2 with entries Ž Nˆ2 .st s N
ˆ 2, s t is nothing but the
incidence matrix of the graph D 2 lq2 . Conversely, up to a phase the columns of Sˆ are
uniquely determined by the two requirements that Sˆ is unitary and diagonalizes Nˆ2 –
with the exception, however, of the columns numbered l q 1 and 2 l q 2, which both
are eigenvectors to the same eigenvalue zero. For the latter two columns, in formula
Ž10.14. Žunlike in Table 2 of Ref. w41x. we have chosen specific linear combinations that
are singled out by the property that all structure constants N ˆ r, s t are non-negative
integers. We also remark that the matrix Ž10.14. is unitary, but it is not symmetric, nor
can it be made symmetric by any re-ordering of rows and columns,13 and finally that
rq 1
Sˆr ,2 lq2 ys s Ž y1 . Sˆr , s for r / 2 l q 1,2 l q 2 , s / l q 1,2 l q 2 . Ž 10.15 .
By inspecting formulæ Ž10.14. and Ž10.11., we observe the relation
´l
SˆrŽw r , c x.,sŽŽ l , w .. s S˜Ž l , w .,w r , c x Ž 10.16 .
(N l
between the matrices Sˆ and S. ˜ Here Nl is the length of the G-orbit through l Ži.e.
Nl s 2 except for N2 l s 1., ´ is the sign factor
1 for l s 2 l ,
´l :s ½ Ž y1.
lr2
otherwise ,
Ž 10.17 .
13
However, as follows from the identifications below, when combining suitable re-orderings with rescalings,
certain submatrices of Sˆ become symmetric.
and we introduced bijections r and s between the respective index sets of Sˆ and S˜ which
act as
rq1° for r / 2 l ,
~
r Ž w r , c x . :s 2 l q 1 for Ž r , c . s Ž 2 l ,q . , Ž 10.18 .
¢
2 lq2 for Ž r , c . s Ž 2 l ,y .
and
°lr2 q 1 for l / 2 l ,
~
s Ž Ž l , w . . :s l q 1 for Ž l , w . s Ž 2 l ,q . , Ž 10.19 .
¢2 l q 2 for Ž l , w . s Ž 2 l ,y . ,
respectively. In particular, the diagonalizing matrices Ž10.5. and Ž10.7. for the even and
odd boundary conditions obey
mrNm
e
(
Sm , n s Nm Ž y1 . SˆrŽ n .,sŽ m . Ž 10.20 .
and
ar2
o
Sa , b s '2 Ž y1 . Sˆbq1 ,sŽ a . , Ž 10.21 .
respectively.
Using the fact that S is symmetric, the result Ž10.16. tells us that up to normalizations
of the rows of S, ˜ transposition, and reordering of the rows and columns, the two
matrices S˜ and Sˆ coincide. As all these manipulations can be absorbed into a basis
transformation, it follows that the two algebras that via these matrices are associated to
the simple current extension of s l Ž2. l , i.e. the classifying algebra C Ž A . and Pasquier’s
algebra are isomorphic.
For concreteness, let us also display explicitly these matrices in the simplest case, i.e.
for l s 1:
°1 '2 1 '2 ¶
'3 0 y'3 0
1 1 1
Sˆs 1 ' '
'6 '2 Ž 1 q i 3 . 1
'2 Ž 1 y i 3 . ,
1 1
¢1 '
'2 Ž 1 y i 3 . 1
'2 Ž 1 q i 3 .ß
'
°1 1 1 y'3 ¶
1 1
1
1
2
Ž 1 q i'3 . 2
Ž 1 y i'3 . 0
S˜s . Ž 10.22 .
'3 1 1
1
2
Ž 1 y i'3 . 2
Ž 1 q i'3 . 0
¢1 1 1 '3 ß
10.4. Virasoro minimal models
The unitary minimal models of the Virasoro algebra are labeled by m s 2,3, . . . ; they
have conformal central charge c s c m :s 1 y 6rŽ m q 1.Ž m q 2.. Via their realization as
a coset theory s l Ž2. my 1 [ s l Ž2.1rs l Ž2. m , the s l Ž2. WZW situation of the previous
subsection gives rise to similar effects in these minimal models. The requirement that
the level must be divisible by four translates to the condition m g 4Z j Ž4Z q 1. on the
label m. In these cases the chiral algebra of the A-theory is obtained from A, which is
just the Virasoro algebra, via extension by the field J with label lJ s Ž m,1. ŽKac table
notation., which has conformal weight D J s mŽ m y 1.r4.
In these cases we even know that the Virasoro algebra A is the only consistent
subalgebra of A, simply because no other unitary conformal field theories exist at the
same value of c. Thus our methods supply us14 with all conformally invariant boundary
conditions of the A-theory. In particular there are precisely two automorphism types, the
even boundary conditions which preserve the full bulk symmetry A, and the odd
boundary conditions which preserve only the Virasoro subalgebra.
The primary fields of the A-theory with central charge c m are labeled by l ' Ž r,r X .
with
1(r(m , 1 ( rX ( m q 1 , Ž 10.23 .
X X
modulo the identification Ž r,r . ; Ž m q 1 y r,m q 2 y r ., so that there is a total of
mŽ m q 1.r2 sectors. We first look at the cases with m s 4 l for some l g Z ) 0 . Then
the A- and A-theory are commonly w39x denoted by Ž A 4 l , A 4 lq1 . and Ž A 4 l , D 2 lq2 .,
respectively. ŽIn the simplest of these, obtained for l s 1, the A-theory is the tetracriti-
cal Ising model Ž A 4 , A 5 . while the A-theory is the three-state Potts model Ž A 4 , D4 ... We
have
I( s Ž r ,r X . < r s 1,3,5, . . . ,4 l y 1, r X s 1,3,5, . . . ,2 l y 1,2 l q 3, . . . ,4 l q 1 4 ,
If s Ž r ,2 l q 1 . < r s 1,3,5, . . . ,4 l y 1 4 ,
I1r2 s Ž r ,r X . < r s 2,4,6, . . . ,4 l , r X s 2,4,6, . . . ,4 l 4 , Ž 10.24 .
and hence < I( < s < I1r2 < s 4 l 2 , < If < s 2 l . Thus we obtain 2 l Ž l q 2. even and 2 l 2 odd
boundary conditions, and hence a total of 4 l Ž l q 1. conformally invariant boundary
conditions.
The similar series with m s 4 l q 1 for l g Z ) 0 can be treated analogously. The A-
and A-theory are now known under the names Ž A 4 lq1 , A 4 lq2 . and Ž D 2 lq2 , A 4 lq2 .,
respectively. We have
I( s Ž r ,r X . < r s 1,3,5, . . . ,2 l y 1,2 l q 3, . . . ,4 l q 1 , r X s 1,3,5, . . . ,4 l q 1 ,
If s Ž 2 l q 1,r X . < r X s 1,3,5, . . . ,4 l q 1 ,
I1r2 s Ž r ,r X . < r s 2,4,6, . . . ,4 l , r X s 2,4,6, . . . ,4 l q 2 , Ž 10.25 .
so that < I( < s < I1r2 < s 2 l Ž2 l q 1., < If < s 2 l q 1, and the number of even and odd
boundary conditions is Ž2 l q 1.Ž l q 2. and l Ž2 l q 1., respectively.
14
At least modulo what goes w44x under the name of ‘complex charges’.
Via the coset construction, it is possible to express all the ingredients in the formula
for S˜ through quantities of the underlying s l Ž2. WZW models, so that again the
classifying algebra C Ž A . can easily be obtained explicitly. We refrain from displaying
any details, which are not too illuminating. We would like to mention, however, that
these results are in perfect agreement with the findings of w45,46x. In the latter papers,
various statements were encoded in the language of graphs; the following remarks allow
to make contact to that point of view.
The total number of conformally invariant boundary conditions is
< I( < q 2 < If <
°4 l Ž l q 1. s 1 P rankŽ A . P rank Ž D 2 lq2 . for m s 4 l ,

4l
s~
2
1
¢2Ž l q 1. Ž 2 l q 1. s 2 P rankŽ A 4 lq2 . P rank Ž D 2 lq2 . for m s 4 l q 1 .
Ž 10.26 .
Regarding the graphs A 4 l and D 2 lq2 Ži.e. the Dynkin diagrams of the respective simple
Lie algebras. as bi-colored, starting with Žsay. a black node, this can be understood as
follows. The even boundary conditions are in one-to-one correspondence with pairs of
black nodes from the ‘product’ of the two graphs, while odd boundary conditions are in
one-to-one correspondence with pairs of white nodes. Mixed pairs of nodes do not
correspond to boundary conditions, which accounts for the factor of 1r2 in Ž10.26.. The
latter selection rule may be implemented by a suitable folding of the A 4 l graph. The
resulting graph has a loop, hence in particular it is no longer bi-colorable; pairs of nodes
from D 2 lq2 and the folded graph are then in one-to-one correspondence with all
conformally invariant boundary conditions, including both even and odd ones.
Further, let us denote by EŽ r˜ . the r˜ th exponent of the Lie algebra D 2 lq2 . For every
r˜ s 1,2, . . . ,2 l q 2, the integer EŽ r˜ . lies in the label set of the A 4 lq1 graph; indeed, the
exponents correspond precisely to the black nodes of A 4 lq1 , with the middle node
appearing twice. We can therefore define for every s s 1,2, . . . ,4 l q 1 a matrix Vs
through
Ž Vr . s˜t̃ :s Ý SrŽ ,EŽ

A. Ž D. Ž D.)
˜ S s,u
u.
Ž A.
˜ ˜ S t˜,ũ rS1 ,EŽ u.
˜ , Ž 10.27 .
ũ
where S Ž A. and S Ž D. are the unitary diagonalizing matrices for the graphs A 4 lq1 and
D 2 lq2 . Thus S Ž A. is nothing but the modular S-matrix of the s l Ž2.4 l WZW model.15
By direct calculation one checks that the matrices Ž10.27. furnish a representation of the
fusion ring of s l Ž2.4 l . Further, it can be shown that V2 coincides with the incidence
15
On the other hand, the matrix S Ž D . defined this way is by far not unique; in particular, the incidence
matrix of D 2 lq2 has an eigenvalue with multiplicity 2. For a generic choice of diagonalizing matrix the
numbers N˜ r,˜ s˜ t̃ Ž4.6. will not be integral. But w45x there is a unique choice such that these numbers are integral.
matrix of D 2 lq2 . By the explicit form of the s l Ž2.4 l fusion rules, this implies that one
may equivalently define the matrices Vs inductively via V1:s 1, V2 :s incidence matrix
of D 2 lq2 and
Vs :s V2 Vsy1 y Vsy2 for s s 3,4, . . . ,4 l q 1 . Ž 10.28 .
One also has the matrix equation
s̃
˜ r˜ s Ý Ž Vq . r̃ N˜ s˜
Vq N Ž 10.29 .
s̃
˜ s̃ are the ‘graph fusion
for all q s 1,2, . . . ,4 l q 1 and all r˜ s 1,2, . . . ,2 l q 2, where N
Ž D.
matrices’ associated to S , i.e. the matrices with entries
Ñr˜ , s˜ t̃ :s Ý SrŽ˜ ,u
D. Ž D. Ž D.)
˜ S s,u
Ž D.
˜ ˜ S t̃ ,ũ rS1,u˜ Ž 10.30 .
ũ
˜ ˜s 1,2, . . . ,2 l q 2.
˜ s,t
for r,
In w45,46x also the E-type series of modular invariants for minimal models were
discussed in the framework of graphs, and cyclic groups larger than Z 2 have been
addressed in w47x. It will certainly be interesting to compare the results obtained there
with the ones that can be derived by the methods of the present paper, and in particular
to study how the graph oriented approach deals with the case of non-cyclic abelian
groups, where non-trivial two-cocycles appear.
11. More examples
In this section we present a few further examples, in which the orbifold group G is
larger than Z 2 . We first discuss two examples of direct relevance to string theory.
Afterwards we turn to some specific examples in which the effects of non-trivial
two-cocycles can be analyzed in detail.
11.1. General cyclic groups
Let us study a situation of immediate interest in which the orbifold group is cyclic
and hence has trivial second cohomology, so that the untwisted and full stabilizers
coincide. We start with the c s 1 theory of a free boson, compactified on a circle of
radius R s '2 N with N g Z ) 0 . The chiral algebra A of this theory consists of
operators of the form Ž10.1.. Inspection shows that for every m g Z ) 0 there is a
subalgebra A Ž m. of A which is obtained by restricting the value of n in Ž10.1. to be a
multiple of m. The algebra A Ž m. is nothing other than the chiral algebra of another free
boson theory, with the free boson X compactified on a circle of radius m'2 N .
Therefore it is a consistent subalgebra in the sense that it allows for the construction of
chiral blocks which obey factorization rules and have a Knizhnik–Zamolodchikov
connection. The orbifold group is G s Z m ; its generator acts on an operator of the form
Ž10.1. by multiplication with the phase expŽ2p i nrm.; in terms of the Fubini–Veneziano
field X this means
X ¨ X q 2pr Ž m'2 N . . Ž 11.1 .
This leaves the u Ž1. current j invariant and multiplies expŽi n'2 N X . by the phase
expŽ2p i nrm.. ŽTogether with the automorphism X ¨ yX Ž10.2., the transformation
Ž11.1. generates the dihedral group Dm ..
In this example no fixed points are present, so that it is most straightforward to write
down the classifying algebra. The A Ž m.-theory has 2 m2N primary fields, which may be
labeled by integers l mod 2 m 2N ; their u Ž1. charge is ql s lrm'2 N . The twist sector
is determined by the value of l mod m; in particular, those l s ml which are multiples
of m are in the untwisted sector and label a basis of the classifying algebra C Ž A Ž m. ..
The multiplication in C Ž A Ž m. . is just given by the restriction of the fusion product of the
A Ž m.-theory to the untwisted sector, i.e.
XX
˜ l ,lX l s d lqlXqlXX mod 2 m N .
N Ž 11.2 .
Ž m. .
Thus C Ž A is the group algebra of the cyclic group Z m N . The reflection coefficients
R ma ; V must therefore obey the relation R am l 1 ; V R am l 2 ; V s R amŽ l 1ql 2 .; V , where the addition is
modulo 2 m N . The solutions to this requirement are
R am l ; V s exp Ž 2p i mlar2 m2N . s exp Ž p i larm N . Ž 11.3 .
with a g Z. Moreover, a must be taken only modulo 2 m N , and hence the possible
values of a are in one-to-one correspondence with the Z m -orbits of the A Ž m.-theory, in
accordance with the general theory.
This result allows for the following geometric interpretation. After performing a
suitable T-duality transformation, we can assume that we are dealing with a Dirichlet
boundary condition, so that we can characterize the boundary state by the position a of a
point-like defect on the circle. Breaking the bulk symmetry to the subalgebra A Ž m.
correspond to defects located at 2 m N th roots of unity on the unit circle. In general, we
expect that boundary conditions breaking more chiral symmetries of the bulk correspond
to more generic locations in the space of boundary conditions.
In the case at hand we can also study explicitly the projective limit of classifying
algebras that was used in the definition ofX the universal classifying algebra in Section 8.
Manifestly, when mX divides m, then A Ž m . is a subalgebra of A Ž m.. In more fancy terms
this can be expressed as follows. Divisibility introduces a partial ordering on the set I
non-negative integers; the subalgebras A Ž m. form an inductive system over I. The
inductive limit consists just of the subalgebra A Ž`. of uncharged elements of A.
Moreover, there is a natural projection relating the classifying algebras:
X
C Ž A Ž m. . ( CZ m N ™ CZ mX N ( C Ž A Ž m . . . Ž 11.4 .
There then exists a projective limit
Ẑ N :s lim Z m N Ž 11.5 .
§
of the classifying algebras, from which every classifying algebra C Ž A Ž m. . can be

obtained as a quotient. The group algebra of the infinite group Zˆ N is the projective limit
of the classifying algebra. ŽFor N s 1 this group is a well-known object; it appears as
the absolute Galois group of any finite field, or as the Galois group of the infinite
extension QŽW .rQ, where W is the group of all roots of unity..
Up to this point we were exclusively considering subalgebras that are rational in the
sense that the number of primary fields of the orbifold is finite. When we allow also for
boundary conditions that preserve non-rational subalgebras, then the situation simplifies
considerably: A Ž`. is then an allowed subalgebra, and the projective limit becomes
isomorphic to the group algebra of Z.16 In this case the irreducible representations of the
classifying algebra, and thus the corresponding boundary conditions, are labeled by the
group UŽ1., the dual group of Z. Recall that depending on whether the trivial or the
charge conjugation modular invariant has been chosen in the bulk, the elements of this
UŽ1. group correspond either to values of the Wilson lines or to positions of D0-branes.
11.2. Simple current symmetries in string theory
Our second example concerns the construction of perturbative superstring theories.

Such theories contain fermionic degrees of freedom, and one must impose several
projections to obtain consistency. All these projections can be formulated in terms of
simple currents Žsee e.g. Refs. w6,48x.. First, one has to impose alignment of the
fermionic degrees of freedom; the space-time fermions, the fermionic degrees of the
internal theory and the superghosts have to be either all in Neveu–Schwarz or all in the
Ramond sector.17 This is accomplished by enlarging the chiral algebra by all bilinears of
the supercurrents of the world sheet theory. We assume that the theory has a space-time
sector containing D free bosons and D free fermions with a supercurrent Jst , which is a
simple current of conformal weight D s 3r2 and of order two. Similarly, there are
supercurrents Jint for the inner sector and Jsgh for the superghosts. The relevant simple
current group G is Z 2 = Z 2 , consisting of the identity and the three non-trivial currents
Ž Jst , Jint ,1 . , Ž Jst ,1, Jsgh . and Ž 1, Jint , Jsgh . . Ž 11.6 .
This extension ensures supersymmetry of the world sheet theory. Boundary conditions
that break this symmetry seem to be unacceptable, since they would spoil the consis-
tency of the theory; they lead to a tachyonic spectrum and other undesired features.
These remarks refer to world sheet supersymmetry and hence apply to all string theories
built from N s 1 superconformal field theories.
When we have superconformal theories with extended N s 2 superconformal symme-
try on the world sheet, another simple current extension allows to build space-time
supersymmetric theories. To this end one imposes the GSO projection, which can be
achieved by enlarging the chiral algebra by another integer spin simple current JGSO .
This simple current JGSO implements the spectral flow in the world sheet theory Žsee
e.g. Refs. w6,48x.; it has non-trivial components both in the space-time and in the
superghost sector. The order M of JGSO is essentially the common denominator of the
u Ž1. charges in the Ramond sector; thus it depends on the specific N s 2 model under
consideration.
JGSO can be described more explicitly as follows. The u Ž1. current
J Ž z . s Jst Ž z . q Jint Ž z . Ž 11.7 .
of the N s 2 algebra is the sum of a component Jst in the space-time sector and Jint in
the inner sector; it can be expressed in terms of a standard free boson as J s cr3 i E X ' '
16
This might look confusing at first sight; but indeed the projective limit does depend on the category in
which it is taken, i.e. on the selection of the objects – here boundary conditions – one considers.
17
For a discussion of boundary states in the Žsuper-.ghost sector, see e.g. Ref. w49x.
'5 i E X. The spectral flow simple current is then realized as the Ramond ground state
expŽiŽ'5 r2. X . of conformal weight D s 5r8. It has to be combined with its counter-
part in the superghost sector. Expressing the superghost in terms of a free boson F , the
spectral flow operator is expŽiFr2. which has conformal weight y5r8, so that the total
simple current
'5
JGSO s exp i
ž 2 /
X Ž z . exp Ž i 12 F Ž z . . Ž 11.8 .
has integral conformal weight. In total, the projections in the construction of a

superstring theory require a simple current extension by the abelian group
Z2 = Z2 = ZM . Ž 11.9 .
In this extension typically fixed points do occur, as well as untwisted stabilizers that
differ from the full stabilizers.
In contrast to the situation with supersymmetry on the world sheet, it is definitely of
interest to study boundary conditions that do not preserve all space-time supersymme-
tries. A particular example is given by BPS conditions, but our formalism allows to
describe also boundary conditions in which all space-time supersymmetries are broken.
For reviews of non-BPS states and their conformal field theory description, we refer to
Refs. w50,51x.
Let us mention that the formalism developed in the present work has another
application in string theory. Namely, every superconformal field theory with N s 2
supersymmetry has an automorphism of order two that reverses the sign of the u Ž1.
current J of the N s 2 algebra and exchanges the two supercurrents G " of charge "1.
Accordingly, when studying boundary conditions that correspond to this automorphism
we get two automorphism types; they are usually called ‘A-type’ and ‘B-type’ Žsee e.g.
Ref. w52x.. According to the general results in Section 6, T-duality interchanges these
two automorphism types. Notice that both types of boundary conditions are encom-
passed by a single classifying algebra.
11.3. The Z 2 orbifold of the free boson and fractional branes
Another illustrative example for our formalism is provided by the Z 2-orbifold of a

free boson, compactified at a rational radius squared. For concreteness, we again restrict
our attention to the case when R 2 s 2 N with N g Z ) 0 , which corresponds to the
diagonal modular invariant. The boundary conditions that preserve all bulk symmetries
are in one-to-one correspondence to the labels of primary fields, for which we will use
the convention of w37x. These boundary conditions can be given the following interpreta-
tion.
In the untwisted sector of the orbifold there are N y 1 primaries Fq , q s 1,2, . . . , N
y 1, of conformal weight D q s q 2r4N N , as well as two pairs 1, J Žof conformal weight
0 and 1. and c 1, c 2 Žwith Dc s Nr4. which each combine to a single primary field of
the underlying circle theory. The boundary conditions labeled by the Fq constitute
D0-branes sitting at j R with j one of the 2 N th roots ep i q r N , 0 - q - N , of unity;
those labeled by 1 and J describe D0-branes which are both localized at one orbifold
point, and those labeled by c 1 and c 2 are D0-branes localized at the other orbifold
point. The latter boundary conditions deserve particular attention, as in these cases the
position in target space is not sufficient to describe uniquely the boundary conditions.
Rather, an additional discrete label is needed. Now the primary fields 1, J, c 1 and c 2
have quantum dimension 1, while the primaries Fq all have quantum dimension 2. It is
also known that in a string compactification the Ramond–Ramond charge is propor-
tional to certain Žgeneralized. quantum dimensions. Thus we can conclude that the
D-branes sitting at orbifold points have a Ramond–Ramond charge that is only half the
one of D-branes sitting at smooth points; accordingly w53x they are referred to as
fractional branes.
In terms of boundary states Žor, equivalently, reflection coefficients. this behavior is
explained as follows. For boundary conditions that preserve the full bulk symmetry, the
reflection coefficients are Žratios of. elements of the modular matrix S. The pairs 1, J
and c 1, c 2 form full orbits of the order-2 simple current J, while the fields Fq are
fixed points of J. Since the monodromy charge with respect to J is 0 for fields in the
untwisted sector and 1r2 in the twisted sector, the standard simple current relation
S J l, m s e 2 p i Q J Ž m .Sl, m implies that in the boundary conditions labeled by Fq boundary
blocks of the twisted sector do not appear. Indeed, because of JFq s Fq we have
S q, m ' S J q, m s yS q, m for all m in the twisted sector. In contrast, orbits of full length do
appear, but with opposite sign for the two primaries l and Jl on the orbit. Briefly, the
ambiguity for boundary conditions that are localized at the orbifold points reflects the
fact that on a disk with such a boundary condition bulk fields in the twisted sector can
acquire a non-vanishing one-point function and that two values with opposite sign are
possible for this correlation function.
The twisted sector contributes four more primaries s 1,2 and t 1,2 of conformal
weight 1r16 and 9r16, respectively. It turns out that the corresponding boundary
conditions are not localized and are thus Neumann-like. It is therefore tempting to
identify them with the four different types of Z 2-equivariant line bundles over the circle.
To gain insight in symmetry breaking boundary conditions, we need consistent
subalgebras A of the chiral algebra A of the Z 2-orbifold. Examples are easily obtained
by observing that the vector space that underlies the chiral algebra A can be decom-
posed according to the absolute value of the u Ž1. charge in the circle theory. This
quantity is well-defined because we only have to deal with fields in the untwisted sector
and only states with opposite charge are identified. In the decomposition all multiples of
'N appear:
HV s [ Hnu Ž1.'N . Ž 11.10 .
ngZ 00
This decomposition does not constitute a grading of A over the additive group Z 0 0 ,
because the fusion structure within the chiral algebra reads w q1 x w w q2 x s w q1 q
q2 x q w < q1 y q2 <x. Still these fusions imply that for every integer l 0 2 the subspace
HVŽ l . :s [ HnulŽ1.'N Ž 11.11 .

ngZ 00
provides a subalgebra of the chiral algebra, and it is in fact a consistent subalgebra

because it is nothing but the chiral algebra of the Z 2-orbifold of a free boson at radius
R Ž l . s l R.
Moreover, inspection shows that, except for l s 2, we are not dealing with an
orbifold subalgebra A G of A. Indeed, for l 0 3 there does not exist any automorphism
of A for which the fixed point set is HVŽ l .. ŽNote that for generic N the orbifold theory
also has very few automorphisms of the fusion rules that preserve the conformal
weights.. In particular, this situation is not covered by our formalism. However, the
specific conformal field theory in question is simple enough to allow for a direct
construction of the corresponding boundary states w54x. They are not related to automor-
phisms of the chiral algebra, and thus it is in general not possible to associate an
automorphism type to such boundary conditions. Hence they provide a simple counter
example to the common misconception that every boundary condition should possess a
definite automorphism type.
In the particular case l s 2, our techniques can be applied to obtain still more
boundary conditions. We denote primary fields in the orbifold theory at radius R:s 2 R
by analogous labels as above, with an additional bar. The boundary conditions then
correspond to orbits of primaries with respect to the simple current c 1 , which has order
two. We only describe the orbits of non-trivial automorphism type. The orbits
Fq ,F4 N yq 4 for q s 1,3, . . . ,2 N y 1 give D0-branes localized at R times a 4NN th root
of unity that is not a 2 N th root of unity. From the other orbits in the untwisted sector of
the A-theory we recover the other boundary conditions in the untwisted sector that were
described earlier. In the twisted sector of the A-theory we find two orbits that are fixed
points, s 1 and t 1 ; they give rise to four boundary conditions that preserve all bulk
symmetries, corresponding to s 1,2 and t 1,2 in the A-theory. Finally, the remaining
orbit s 2 ,t 2 4 gives rise to a Neumann-like boundary condition which breaks the
symmetries down to HVŽ2..
We conclude our discussion with the remark that, unlike for the free boson case, the
automorphism type of a boundary condition in the Z 2-orbifold of a free boson does not
allow any longer to distinguish Dirichlet and Neumann boundary conditions. Still, all
Dirichlet boundary conditions come from the untwisted sector while all Neumann
boundary conditions come from the twisted sector.
11.4. Examples with genuine untwisted stabilizer
We now consider in detail an example where an untwisted stabilizer occurs that is a

proper subgroup of the full stabilizer. It is worth emphasizing that the situation with
genuine untwisted stabilizers arises rather naturally in string compactifications. This is a
consequence of the following elementary fact that applies to any tensor product of three
or more subtheories with three simple currents Jî Ž i s 1,2,3. of half-integral conformal
weight in three distinct subtheories. In the Gepner construction of superstring vacua such
subtheories can be factors in the inner sector, but also the conformal field theories
describing space-time fermions or the Žsuper-.ghosts. Generic examples for simple
currents with D g Z q 1r2 are the various components of the supercurrent on the
world sheet, but typically in such compactifications other simple currents of this type are
present as well.
It has been shown in w4x that in WZW theories the commutator cocycle obeys
Fl Ž J,K . s exp 2p i Ž DJ y DJŽ K . . , Ž 11.12 .

where DJ is the conformal weight of J, while DJŽK . is the conformal weight of the
projection of J in what is called the fixed point theory w6x with respect to K. This
formula also applies to those simple currents of coset theories which come from simple
currents of the underlying WZW models.18 In the situation at hand, each of the simple
currents Jî is projected to the identity primary field of the fixed point theory, so that
Ž11.12. implies that FlŽJî ,Jî . s y1. On the other hand, for i / j the two currents are
fields in distinct subtheories, so that FlŽJî ,Jˆj . s 1. Now out of the three currents Jî we
can form three simple currents Ji of the tensor product theory by setting J1:s Jˆ2 Jˆ3 and
cyclic. These have integral conformal weight and hence – unlike the original currents Jî
themselves – can be used to extend the chiral algebra, by a Z 2 = Z 2 group. By the
bihomomorphism property of Fl we then see that Fl is non-trivial on this group:
1 for i s j ,
Fl Ž Ji ,J j . s ½ y1 for i / j .
Ž 11.13 .
As a consequence, the untwisted stabilizer is a proper subgroup of the full stabilizer.

We will now study the effect of a non-trivial untwisted stabilizer in the situation that
the A-theory is a WZW theory, based on some affine Lie algebra g s g Ž1. and that the
orbifold theory is a WZW theory as well, now based on gX s gX Ž1. with gX ; g. Such
WZW orbifolds have been studied in w5,55x. Here we are interested in conformally
invariant boundary conditions. Thus the Virasoro algebras of g and gX must coincide;
this is precisely the case when gX ® g is a conformal embedding w56,57x.
Many, though not all, conformal embeddings can be understood in terms of simple
currents. An example where this is possible is the following. For any three odd positive
integers d1 ,d 2 ,d 3 there is a conformal embedding
gX :s so Ž d1 . 1 [ so Ž d 2 . 1 [ so Ž d 3 . 1 ® so Ž d1 q d 2 q d 3 . 1 s :g . Ž 11.14 .
The theory based on g can be obtained as an extension of the gX-theory by a simple
current group G s Z 2 = Z 2 , consisting of the fields
Ž o,o,o . , Ž o,v,v . , Ž v,o,v . and Ž v,v,o . , Ž 11.15 .
where o and v refer to the singlet and vector representation of soŽ d i ., respectively. There
is only a single fixed point, namely the tensor product Ž ß, ß, ß . of three soŽ d i . spinor
representations; it has stabilizer S s G . The conformal weight of the vector simple
current v at level 1 is 1r2; according to the general arguments presented at the
beginning of this subsection, the untwisted stabilizer is therefore trivial, U s Žo,o,o.4 .
ŽThe value w S : U x s 2 2 of the index is in agreement with the fact that the ground state
degeneracy of the irreducible spinor representation of the soŽ d1 q d 2 q d 3 . theory is
twice as large as the one of Ž ß, ß, ß ., namely 2 P 2 Ž d1y1 .r2 2 Ž d 2y1 .r2 2 Ž d 3y1 .r2 ..
As a side remark, we mention that these theories can be realized in terms of free
fermions. Thus the effect of a genuine untwisted stabilizer can occur even in free
conformal field theories. This is in fact not too surprising. As we have explained, the
presence of untwisted stabilizers is related to the fact that the orbifold group acts only
projectively on certain sectors of the theory; this is also known w37x to be true for the
18
In coset theories there can, however, exist additional simple currents which arise from resolving fixed
points with integral quantum dimension.
action of the three polyhedral groups on the free boson Žcompactified at the self-dual
radius. that gives rise to the exceptional c s 1 theories.
In the case at hand, the relevant automorphisms of the chiral algebra A can be
understood in terms of the finite-dimensional Lie groups G and GX associated to g and
gX . Namely, for every boundary block Bl: Hl m Hlq that preserves the full bulk
symmetry and every element g g G the combination BlŽ g . :s Bl (Žg m id. provides us
with a twisted boundary block. The corresponding automorphism on the affine Lie
algebra g is the inner automorphism that acts on the modes Jna of g as Jna ¨ Žg J agy1 . n .
This automorphism preserves the symmetries in the subalgebra gX if and only if g is in
the centralizer CG Ž GX . of GX in G, and it acts trivially if and only if g is even in the
center ZŽ G . of G. Thus the non-trivial twists are those by elements in the group
CG Ž GX .rZŽ G ..
In the case of our interest, the relevant embedding on the level of Lie groups reads
GX :s Ž Spin Ž d1 . = Spin Ž d 2 . = Spin Ž d 3 . . r Ž Z 2 = Z 2 .
® Spin Ž d1 q d 2 q d 3 . s :G ; Ž 11.16 .
X X
note that the subgroup G is not simply connected. For determining CG Ž G .rZŽ G . it is
instructive to consider first the embedding
G˜ X :s SO Ž d1 . = SO Ž d 2 . = SO Ž d 3 . ® SO Ž d1 q d 2 q d 3 . s :G˜ Ž 11.17 .
that is obtained from Ž11.16. by dividing out the center of SpinŽ d1 q d 2 q d 3 .. Then the
center of the G˜ is trivial; moreover, using the matrix realization of these groups, one
shows that the centralizer CG̃ Ž G˜ X . is a Z 2 = Z 2 group consisting of the unit matrix
1 d1qd 2qd 3 and the diagonal matrices
1 d1 0 0 y1 d1 0 0
M˜ 23 :s
0
0
y1 d1
1 d2
0
0
0
1 d3 0 ,
0
M˜ 13 :s
0
0
1 d2
0
0
y1 d 3 0 ,
M˜ 12 :s
0
0
y1 d 2
0 1 d3
0
0 .
Already at this stage we can conclude that the centralizer CG X Ž G . is an extension of this
Ž 11.18 .
Z 2 = Z 2 group by the center Z 2 of SpinŽ d1 q d 2 q d 3 .. To decide which extension we

are dealing with, we introduce gamma matrices g i that satisfy the Clifford relations
g i ,g j 4 s 2 d i j. We can then determine the two lifts Mi"j of the matrices M˜ i j to
SpinŽ d1 q d 2 q d 3 . from the requirement that M12 g i Ž M12 .y1 s yg i for 1 ( i ( d1 q d 2
and M12 g i Ž M12 .y1 s g i for i ) d1 q d 2 . It is easy to verify that
M12" s " g 1g 2 . . . g d1qd 2 , "

M23 s " g d1q1g d1q2 . . . g d1qd 2qd 3 ,
Ž 11.19 .
M13" s " g 1g 2 . . . g d1 g d1qd 2q1 . . . g d1qd 2qd 3 .
It can be checked that these matrices commute with all elements in GX . They form a
group of order 8; the structure of this group depends on the values of d1 ,d 2 and d 3 .
First, when all d i leave the same rest modulo 4Z, then the group is isomorphic to the
eight-element generalized quaternion group; this group has one two-dimensional and
four one-dimensional irreducible representations. Otherwise, i.e. when only two of the
d i leave the same rest modulo 4Z, the group is isomorphic to the dihedral group D4 of
8 elements. This group has four one-dimensional and one two-dimensional irreducible
representation, too. In both cases the centralizer is non-abelian.
Let us present some more details. The boundary blocks are labeled by the fields of
monodromy charge zero and characters of their Žfull. stabilizers; thus there are 8
boundary blocks coming from full orbits; we label them lexicographically,
B 1 :s B ooo , B 2 :s B oov , B 3 :s B ovo , B 4 :s B ovv ,
B 5 :s B voo , B 6 :s B vov , B 7 :s B vvo , B 8 :s B vvv . Ž 11.20 .
Note that the blocks numbered as 1,4,6,7 come from the vacuum of the A-theory, while
the others come from the field that carries the vector representation of soŽ d1 q d 2 q d 3 ..
In addition we have 4 boundary blocks coming from the fixed point Ž ß, ß, ß .. They
correspond to the four irreducible characters c of Z 2 = Z 2 ; we label them as
B 9 :s Bqqqq , B 10 :s Bqqyy , B11 :s Bqyqy , B 12 :s Bqyyq ,
Ž 11.21 .
where the " labels indicate the values "1 of c on the four elements of Z 2 = Z 2 , in the
lexicographic order chosen in formula Ž11.15..
The boundary conditions are labeled by the orbits and characters of their untwisted
stabilizers. Thus in addition to the three boundary conditions that preserve all of A,
there are three conditions from the length-4 orbits and two conditions for each of the
three length-2 orbits which have stabilizer Z 2 . We label them according to
B1 ˆ
s Ž ooo. , Ž ovv. , Ž vov. , Ž vvo. 4 , B7 ˆ
s Ž o ßß . , Ž vßß . 4 with c s1 ,
B2 ˆ
s Ž oov. , Ž ovo. , Ž voo. , Ž vvv. 4 , B8 ˆ
s Ž o ßß . , Ž vßß . 4 with c sy1 ,
B3 ˆ
s Ž ßßß . 4 , B9 ˆ
s Ž ßo ß . , Ž ß vß . 4 with c s1 ,
B4 ˆ
s Ž ßoo. , Ž ßov . , Ž ß vo . , Ž ß vv. 4 , B10 ˆ
s Ž ßo ß . , Ž ß vß . 4 with c sy1 ,
B5 ˆ
s Ž o ßo . , Ž o ß v . , Ž vßo . , Ž vß v . 4 , B11 ˆ
s Ž ßßo . , Ž ßß v . 4 with c s1 ,
B6 ˆ
s Ž oo ß . , Ž ovß . , Ž vo ß . , Ž vvß . 4 , B12 ˆ
s Ž ßßo . , Ž ßß v . 4 with c sy1 .
Ž 11.22 .
With this numbering, the diagonalizing matrix S˜ looks as follows:
° 1 1 '2 '2 '2 '2 1 1 1 1 1 1 ¶
1 1 '
- 2 '2 '2 '
- 2 y1 y1 y1 y1 1 1
1 1 -'2 '2 -'2 '2 y1 y1 1 1 y1 y1
1 1 '2 '2 -'2 -'2 1 1 y1 y1 y1 y1
1 1 -'2 -'2 '2 '2 1 1 y1 y1 y1 y1
1 1 1 '2 -'2 '2 -'2 y1 y1 1 1 y1 y1

S̃ s
2 1 1 '2 -'2 -'2 '2 y1 y1 y1 y1 1 1
. Ž 11.23 .
1 1 -'2 -'2 -'2 -'2 1 1 1 1 1 1
'2 '
- 2 0 0 0 0 b 23 yb 23 b 13 yb 13 b 12 yb 12
'2 -'2 0 0 0 0 b 23 yb 23 yb 13 b 13 yb 12 b 12
'2 -'2
¢
'2 -'2
0
0
0
0
0
0
0
0
yb 23
yb 23
b 23
b 23
b 13
yb 13
yb 13
b 13
yb 12
b 12
b 12
yb 12
ß
Here we have put
bi j :s '2 iyŽ d iqd j .r2 Ž 11.24 .
for i, j g 1,2,34 . We note that
2
˜˜† s 4 P 1 s S˜†S˜ and
SS ˜˜t .
Ž SS s 16 P 1 , Ž 11.25 .
˜˜t is a permutation if and only if precisely one out of the expressions
but SS
2 2 2 2 2 2
2 q b 23 q b 13 q b 12 , 2 q b 23 y b 13 y b 12 ,
2 2 2 2 2 2
2 y b 23 q b 13 y b 12 , 2 y b 23 y b 13 q b 12 Ž 11.26 .
is non-vanishing, which are precisely those cases where after division by 2 these
numbers furnish a character of Z 2 = Z 2 .
Let us now add some general remarks. Every G-module can be decomposed into
irreducible modules of the product CG Ž GX . = GX . The fact that the group CG Ž GX . is
non-abelian implies that it has higher-dimensional irreducible representations, which in
turn means that higher-dimensional degeneracy spaces appear; this way we recover a
generic feature of genuine untwisted stabilizers. Conversely, we are led to the following
conjecture. Let GX ® G be an embedding of reductive compact Lie groups such that the
associated embedding of affine Lie algebras is a simple current extension. Then the
centralizer CG Ž GX . is non-abelian if and only if at some value of the level a genuine
untwisted stabilizer appears. For instance, there is a conformal embedding Ž D4 . 2 ®
Ž E7 .1 which again is a simple current extension by a Z 2 = Z 2 group. As E7 is an
exceptional Lie algebra, the relevant centralizer, namely the one of SOŽ8.rZ 2 in E7 , is
now difficult to compute, but in any case the analysis of the boundary conditions
indicates that this centralizer is non-abelian. For completeness, we mention that in the
E7 case there are eight boundary conditions preserving the affine D4 subalgebra, and the
matrix S˜ reads
°1 1 1 1 1 1 1 ¶
1
1 1 1 1 y1 y1 y1 y1
1 1 y1 y1 1 1 y1 y1
1 1 1 y1 y1 y1 y1 1 1
S˜s . Ž 11.27 .
'2 1 y1 i yi i yi i yi
1 y1 i yi yi i yi i
¢ 1
1
y1
y1
yi
yi
i
i
i
yi
yi
i
yi
i
ß
i
yi
On the other hand, the automorphisms themselves are classified by the group
CG Ž GX .rZŽ G ., which in both cases considered above is Z 2 = Z 2 . Thus they precisely
correspond to the automorphism types that are predicted by the general analysis.
Acknowledgements
We would like to thank Valya Petkova and Patrick Dorey for discussions, and Peter
Bantay and Bert Schellekens for helpful correspondence.
Appendix A. Collection of formulae
Here we collect a few basic formulae from w1x that are used in this paper. The
equation numbers are the same as in w1x.
v The boundary blocks B ˜ Ž l, w . are the linear forms
B̃Ž l , c . :s Nm dly1 r2 bc m Bl Ž A.1 .

on
Hl m Hlq' [ Vcˆ m HJ l ( m [ Vcˆ X m HJ lq( .

Jg GrSl Jg GrSl
Here bc :VVcˆl m Vcˆlq ™ C is a linear form on the degeneracy spaces, Bl : Hl m Hl ™ C is

an ordinary boundary block of the A-theory, dl s < Sl <r< Ul < , and Nl is a phase that is
(
left undetermined Žin the first place, this normalization is introduced as Nl, c , but as
shown in w1x it depends only on the primary label l s w l, cˆ x of the A-theory.. The
linear form bc can be written as
bc s b( ( Ž Oc m id . Ž A.2 .
with b( defined by b( Ž Õ m w . s BlŽ Õ m p( m w m q( .rBlŽ p( m q( . Žwhere p( g Hl

and q( g Hlq are any vectors such that BlŽ p( m q( . / 0. and
)
Oc :s dly3r2 Ý c Ž J . Rcˆ Ž J . , Ž A.3 .
Jg SlrUl
where Rĉ denotes the irreducible representation of the twisted group algebra that is
labeled by cˆ $ c . The Oc with c % cˆ Žthat is, c < Ul s cˆ . form a partition of unity:
Ý Oc s dl1r2 1 dl . Ž A.4 .
cg Sl)
c% cˆ
The operator product expansion that describes the excitation on the boundary caused by
a bulk field approaching it reads
y2 Dlq D
fŽ l , c .,Ž lq , cq . Ž re i s . s Ý Ž 1 y r 2 . m R aŽ l , c .; m Cma a Ž e i s .
m
qdescendants for r ™ 1 . Ž A.5 .

The diagonalizing matrix S˜ of C Ž A . can be written as
<G<
S˜Ž l , cl.,w r , cˆr x :s cl Ž J . cˆr Ž J . SlJ , r ,
)
1r2 Ý Ž A.6 .
SlUl Sr Ur Jg SllUr
where the matrices S J represent the modular S-transformation on the one-point chiral
blocks with insertion J on the torus. The result that S˜ is a square matrix is equivalent to
the sum rule
Ý < Sl < s Ý < Ur < . Ž A.7 .
l w rx
Q G Ž l .s0
References
w1x J. Fuchs, C. Schweigert, Symmetry breaking boundaries I. General theory, hep-thr9902132, Nucl. Phys.
B, in press.
w2x D.C. Lewellen, Nucl. Phys. B 372 Ž1992. 654.
w3x G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 381 Ž1996. 97.
w4x J. Fuchs, A.N. Schellekens, C. Schweigert, Nucl. Phys. B 473 Ž1996. 323.
w5x L. Birke, J. Fuchs, C. Schweigert, Symmetry breaking boundary conditions and WZW orbifolds, Adv.
Theor. Math. Phys. 3 Ž1999. No. 3.
w6x A.N. Schellekens, S. Yankielowicz, Int. J. Mod. Phys. A 5 Ž1990. 2903.
w7x C. Dong, H. Li, G. Mason, Int. Math. Res. Notices 18 Ž1996. 913.
w8x C. Dong, G. Mason, Duke Math. J. 86 Ž1997. 305.
w9x C. Dong, G. Mason, On the operator content of nilpotent orbifold models, hep-thr9412109.
w10x G. Moore, N. Seiberg, Phys. Lett. B 220 Ž1989. 422.
w11x A.N. Schellekens, S. Yankielowicz, Nucl. Phys. B 327 Ž1989. 673.
w12x J. Fuchs, C. Schweigert, Nucl. Phys. B 530 Ž1998. 99.
w13x A. Recknagel, V. Schomerus, Nucl. Phys. B 531 Ž1998. 185.
w14x C. Dong, J. Lepowsky, Generalized Vertex Algebras and Relative Vertex Operators ŽBirkhauser,
¨ Boston,
1993..
w15x C. Dong, H. Li, G. Mason, Commun. Math. Phys. 180 Ž1996. 671.
w16x M.B. Green, M. Gutperle, Nucl. Phys. B 460 Ž1996. 77.
w17x J.L. Cardy, Nucl. Phys. B 324 Ž1989. 581.
w18x K. Ueno, in: Physics and Geometry, ed. J.E. Andersen, H. Pedersen, A. Swann ŽMarcel Dekker, New
York, 1997. p. 603.
w19x A. Beauville, in: Hirzebruch 65 Conference on Algebraic Geometry wIsrael Math. Conf. Proc. 9x, ed. M.
Teicher ŽBar-Ilan University, Ramat Gan, 1996. p. 75.
w20x D. Gaitsgory, in: Quantum Fields and Strings: A Course for Mathematicians, ed. P. Deligne, P. Etingof,
D.S. Freed, L. Jeffrey, D. Kazhdan, J. Morgan, D.R. Morrison, E. Witten ŽAmerican Mathematical
Society, Providence, 1999..
w21x I.B. Frenkel, J. Lepowsky, A. Meurman, Vertex Operator Algebras and the Monster ŽAcademic Press,
New York, 1988..
w22x I.B. Frenkel, Y. Huang, J. Lepowsky, Memoirs Amer. Math. Soc. 104 Ž1993. 1.
w23x V.G. Kac, Vertex Algebras for Beginners ŽAmerican Mathematical Society, Providence, 1996..
w24x C. Dong, H. Li, G. Mason, Modular-invarianceof trace functions in orbifold theory, q-algr9703016.
w25x B. Gato-Rivera, A.N. Schellekens, Nucl. Phys. B 353 Ž1991. 519.
w26x I. Brunner, R. Entin, C. Romelsberger,
¨ J. High Energy Phys. 9906 Ž1999. 16.
w27x G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 356 Ž1995. 230.
w28x J. Fuchs, C. Schweigert, Phys. Lett. B 414 Ž1997. 251.
w29x I. Affleck, M. Oshikawa, H. Saleur, Boundary critical phenomena in the three-state Potts model, preprint
cond-matr9804117.
w30x K. Druhl,
¨ H. Wagner, Ann. Phys. 141 Ž1982. 225.
w31x P. Bantay, Int. J. Mod. Phys. A 13 Ž1998. 199.
w32x M. Muger,
¨ Galois theory for braided tensor categories and the modular closure, math. CTr9812040.
w33x C. Dong, G. Mason, J. Algebra 214 Ž1998. 92.
w34x A. Hanaki, M. Miyamoto, D. Tambara, Duke Math. J. 97 Ž1999. 541.
w35x J. Fuchs, C. Schweigert, Commun. Math. Phys. 185 Ž1997. 641.
w36x A.N. Schellekens, Nucl. Phys. B 558 Ž1999. 484.
w37x R. Dijkgraaf, C. Vafa, E. Verlinde, H. Verlinde, Commun. Math. Phys. 123 Ž1989. 485.
w38x J. Fuchs, C. Schweigert, The action of outer automorphisms on bundles of chiral blocks, hep-thr9805026,
Commun. Math. Phys., in press.
w39x A. Cappelli, C. Itzykson, J.-B. Zuber, Nucl. Phys. B 280 Ž1987. 445.
w40x W. Wang, Duke Math. J. ŽInt. Math. Res. Notes. 71 Ž1993. 197.
w41x V. Pasquier, Nucl. Phys. B 285 Ž1987. 162.
w42x P. di Francesco, J.-B. Zuber, Nucl. Phys. B 338 Ž1990. 602.
w43x V.B. Petkova, J.-B. Zuber, Nucl. Phys. B 438 Ž1995. 347.
w44x A. Sagnotti, Ya.S. Stanev, Fortschr. Phys. 44 Ž1996. 585.
w45x R.E. Behrend, P.A. Pearce, J.-B. Zuber, J. Phys. A 31 Ž1998. L763.
w46x R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Phys. Lett. B 444 Ž1998. 163.
w47x V.B. Petkova, J.-B. Zuber, in: GROUP21 Physical Applications and Mathematical Aspects of Geometry,
Groups, and A’s, Vol. 2, ed. H.-D. Doebner, W. Scherer, C. Schulte ŽWorld Scientific, Singapore, 1997.
p. 627.
w48x C. Schweigert, Theor. Math. Phys. 98 Ž1994. 326.
w49x P. Di Vecchia, M. Frau, I. Pesando, S. Sciuto, A. Lerda, R. Russo, Nucl. Phys. B 507 Ž1997. 259.
w50x A. Sen, Non-BPS states and branes in string theory, hep-thr9904207.
w51x A. Lerda, R. Russo, Stable non-BPS states in string theory: a pedagogical review, hep-thr9905006.
w52x H. Ooguri, Y. Oz, Z. Yin, Nucl. Phys. B 477 Ž1996. 407.
w53x D.-E. Diaconescu, M.R. Douglas, J. Gomis, J. High Energy Phys. 9802 Ž1998. 13.
w54x M. Oshikawa, I. Affleck, Phys. Rev. Lett. 77 Ž1996. 2604.
w55x V.G. Kac, I.T. Todorov, Commun. Math. Phys. 190 Ž1997. 57.
w56x A.N. Schellekens, N.P. Warner, Phys. Rev. D 34 Ž1986. 3092.
w57x F.A. Bais, P. Bouwknegt, Nucl. Phys. B 279 Ž1987. 561.
$
On the construction and solution of Uq Ž gl Ž2,1;C ..-symmetric
models
J. Gruneberg
¨ Theoretische Physik, UniÕersitat
Institut fur ¨ zu Koln,
¨ Zulpicher
¨ Straße 77, 50937 Cologne, Germany
Received 16 September 1999; accepted 22 October 1999
Abstract
$ X
The Uq Ž gl Ž2,1;C.. symmetric R-matrix, acting on the direct product of a four-dimensional
representation of this algebra with itself, constructed by Gould et al., is generalized to the case of
two different four-dimensional not necessarily unitary representations and to the case of a product
of these with the fundamental one. The latter is used in the algebraic Bethe ansatz solution of
some models. Together with the R-matrix acting on the product of the fundamental representation
with itself, these fundamental R-matrices form the starting point of an infinite fusion hierarchy of
exactly solvable models. A number of these are constructed explicitly. Some Bethe ansatz
equations for the eigenvalues of their transfer matrices are derived. q 2000 Elsevier Science B.V.
All rights reserved.
PACS: 75.10.Jm
Keywords: Graded Yang–Baxter equation; Quantum algebras; Bethe ansatz; Fusion hierarchies
1. Introduction
An important class of solvable models in statistical mechanics are the Bethe ansatz
w1x integrable ones. Baxter realized w2x that the unifying structure behind these models is
the famous Yang–Baxter equation ŽYBE.
X XX X XX X XX XX X
VV VV
R 12 Ž u,Õ . R13 Ž u,w . R V23V Ž Õ,w . s R V23V Ž Õ,w . R13
VV VV
Ž u,w . R12 Ž u,Õ . . Ž 1.1a .
X
V,V X and V XX are three in general different spaces. The operators RVV Žu. act in the direct
product V m V X . Both sides of Ž1.1a. are assumed to act in the three-fold product
V m V X m V XX . The lower indices i,j g 1,2,3 on the R-operators denote the two factors in
this product, on which the corresponding R-operator acts non-trivially. In general the
E-mail address: jg@thp.uni-koeln.de ŽJ. Gruneberg..
PII: S 0 5 5 0 - 3 2 1 3 Ž 9 9 . 0 0 6 7 5 - 6
J. Grunebergr Nuclear Physics B 568 [PM] (2000) 594–640 595
spectral-parameters u,v,w are complex. In most cases R-operators are meromorphic

functions of the difference of these arguments, leading to
X XX X XX X XX XX X
VV VV
R 12 Ž u y Õ . R13 Ž u . R V23 V Ž Õ . s R V23V Ž Õ . R13
VV VV
Ž u . R12 Ž uyÕ. . Ž 1.1b .
This equation appeared earlier in the works of Onsager w3x and Yang w4,5x. In the present
context Ž1.1. was introduced by Baxter. His excellent book w6x may serve as an account
of this development.
Given a set of solutions to the difference type equation Ž1.1b. with a set of possibly
different local quantum spaces V Ž n. in place of V XX , the monodromy matrix TâV Ž1, N < u.
and transfer matrix tˆ V Ž N < u. are defined by
ŽN.
TâV Ž 1, N < u . :s R Va NV
Ž1.
Ž u y d Ž N . . = . . . = R Va1V Ž u y d Ž1. . , Ž 1.2a .
tˆ V Ž N < u . :s tr V TâV Ž 1, N < u . 4 . Ž 1.2b .
Here Rˆ VV
an
Žn.
Žu. is assumed to act trivially on V Žm. with m / n. Usually d Žn. g C are
considered. If Xall d Žn. vanish and the V Žn. are identical, tˆ V ŽN <u. is called homogeneous.
Provided RVV Žu. is invertible almost everywhere, from Ž1.1b. and Ž1.2a. and Ž1.2b.
follows w6x
X !
tˆ V Ž N < u . ,tˆ V Ž N < Õ . s 0 . Ž 1.2c .
X
V
Expanding tˆ Ž N < Õ . in Õ yields in general an infinite number of operators commuting
with tˆ V Ž N < u., indicating the diagonalizability of tˆ V Ž N < u. w6x.
tˆ V Ž N < u. can be used as row-to-row transfer matrix of some two-dimensional
classical model with local interactions on a torus w6x. The logarithmic derivative of some
tˆ V Ž N < u. with respect to u defines a one-dimensional local quantum Hamiltonian Hˆ on
a ring w6x. The diagonalization of tˆ V Ž N < u. yields immediately the partition function of
the classical model, as well as all eigenvectors and eigenvalues of H.
Baxter’s discovery has motivated the search for solutions of Ž1.1.. The theory of
quantum groups, mainly due to Drinfel’d, provides all solutions, if V,V X and V XX are
finite-dimensional carrier spaces of representations of simple Lie Žsuper.algebras and
their standard q-deformations w7x.
Finite-dimensional simple Lie superalgebras, have been classified by Kac w8x. The
$
physical importance of solutions to Ž1.1b., based on non-fundamental representations,
X
was realized, when Gould et al. constructed such an Uq Ž gl Ž2,1;C..-symmetric R-matrix
w9x and showed that it is related to the integrable 2-parameter Hubbard type model
treated by Bariev et al. w10x.
Their solution will be generalized to the direct product of two different representa-
tions, leading to another Hamiltonian limit. Two related R-matrices will also be
constructed. Simplicity allows to obtain them by applying Jimbo’s well-known method
w14x, without reference to the powerful, but complicated techniques, developed by Gould
and coworkers w9,12,13x. The corresponding transfer matrices Ž1.2b. can be diagonal-
ized, using results of a previous paper w11x. They are the missing pieces in a puzzle,
leading to an infinite hierarchy of solvable models and complete sets of eigenvalue
equations for the corresponding transfer matrices Ž1.2b.. The general method, based on
the fusion procedure of Kulish and Sklyanin w15x, will be explained, and explicit
solutions will be given.
596 J. Grunebergr Nuclear Physics B 568 [PM] (2000) 594–640
2. Definitions
Finite-dimensional basic, classical, simple Lie superalgebras were classified by Kac

w8x. A readable account of his theory may be found in Cornwell’s book w16x, from which
most of the notation is borrowed. More details on affinization may be found in the book
by Kac w21x. Quantization of these relations is due to Drinfel’d w7x. A good account of
these developments has been given by Chari and Presley w17x. The missing Serre
relations, needed for the quantization of affine simple Lie superalgebras, were systemati-
cally constructed by Yamane w18,19x.
Basis and cobasis of a d-dimensional complex vector space Vd , will be denoted by
i:, ² j < i : s di j Ž 2.1a .
for i,j s 1, . . . ,d. A Z 2-grading is imposed on it via
p Ž i . s 0 for i s 1, . . . ,d 0 ,
p Ž i . s 1 for i s d 0 q 1, . . . ,d 0 q d1 Ž 2.1b .
and extended to Vd by linearity. d 0 and d 1 are strictly positive and sum up to
d 0 q d 1 s d 0 2. V is therefore a direct sum, i.e. it carries a Žd 0 ,d 1 .-grading:
Vd s VdŽ0.
0
[ VdŽ1.
1
. Ž 2.1c .
Here VdŽpp. are the non-trivial d p-dimensional even and odd subspaces spanned by even
Ž p s 0. and odd Ž p s 1. basis elements. Elements of Vd , which are either in Vd or in
0
Vd1 are called homogeneous. The notion of grading Ž2.1b. is naturally extended to these.
Local operators acting in Vd are expressed in the basis
e i j s i :² j . Ž 2.2a .
A grading is assigned to this basis according to
p Ž e i j . s p Ž i . q p Ž j . mod 2 Ž 2.2b .
and extended by linearity to EndŽ Vd ., which is also a graded vector space. Finite direct
products of spaces Vd , can be defined straight forwardly and are denoted by
VdŽŽ NN .. m . . . m VdŽ1.
Ž1.
. Ž 2.3a .
Vectors out of these product spaces and operators acting on them are conveniently
expressed in the product basis, which is ordered like Ž2.3a.. Grading modifies the
multiplication rules on the 2-fold product Ž2.3a., i.e.
ˆ
Ž aˆ m bˆ . Ž cˆ m dˆ . s Ž y1. p Ž b . p Ž ĉ . Ž ac ˆ ˆ.
ˆˆ . m Ž bd Ž 2.3b .
for ˆb,dˆ g EndŽVdŽ1. Ž1.
ˆ ˆ g EndŽVdŽ2.
. and a,c
Ž1.
. homogeneous,
ˆ
Ž aˆ m bˆ . Ž c :m w :. s Ž y1. p Ž b . p Ž c :. Ž aˆ c :. m Ž bˆ w :. Ž 2.3c .
for aˆ g EndŽVd Ž2. ., ˆb g EndŽVd Ž1. . and c :g Vd Ž2. , w :g Vd Ž1. homogeneous. The cobasis
Ž2. Ž1. Ž2. Ž1.
is defined via
†
Ž i :m j :. s Ž y1. p i p j ² i m ² j .
Ž . Ž .
Ž 2.3d .
Eqs. Ž2.3b. – Ž2.3d. are extended naturally to N-fold products Ž2.3a..
$
Here Yamane’s construction w18–20x of Uq ŽslŽ1. Ž2,1;C.. will be used: First a complex
vector space E f C 3 with a non-degenerate, symmetric, bilinear form ² P ,P :: E m E C, ™
is defined. The linear independent subset P ; E
P s a0 , a1 , a2 4 Ž 2.4a .
contains the simple, positiÕe roots a i . A symmetric bilinear form ² P ,P : is defined on P
according to
² a 0 , a 0 : :s 0 , ² a 0 , a 1 : :s y1 , ² a 0 , a 2 : :s 1,
² a 1 , a 1 : :s 2 , ² a 1 , a 2 : :s y1 , ² a 2 , a 2 : :s 0 Ž 2.4b .
and extended per linearity to E. Also, a parity function p: P ™ 0,14 is defined on P
according to
p Ž a1 . s 0 , pŽ a0 . s pŽ a2 . s 1 . Ž 2.4c .
It is well known that the choice of simple roots is not unique. To be precise, Ž2.4b. and
Ž2.4c. correspond to the distinguished choice preferred by mathematicians; see e.g. Ref.
w16x. The triple ŽE, P ,p. is called a datum w19x. To each g g E an unique element Hg
from the dual space H :s E ) can be related via
d Ž Hg . s² d ,g : ;d g E . Ž 2.4d .
Let " be an indeterminate. q is defined formally as
q :s e " . Ž 2.5a .
Let Cww " xx denote the C-algebra of formal power series in ". The datum ŽE, P ,p.
characterizes an associatiÕe, Z 2-graded Hopf algebra A" over the ring of formal power
series Cww " xx with generators Hg for each g g E, and Ea i , Fa i with i s 0,1,2. The
parities of the generators are defined as
p Ž Hg . s 0, p Ž Ei . s p Ž Fi . :s p Ž a i . . Ž 2.5b .
On the superalgebra A" , the supercommutator wP,P x is defined as
w a,b x :s ab y Ž y1. p Ž a . p Ž b . ba Ž 2.5c .

for all homogeneous a,b g A" . For the generators Hg g H , Ei , Fi with i s 0,1,2 the
multiplication is defined via the relations
w H , H X x s 0 ;H , H X g H , Ž 2.5d .
Hg , Ei s²g , a i : Ei ;g g E , Ž 2.5e .
Hg , Fj s y²g , a i : Fi ;g g E , Ž 2.5f .
sinh Ž "Ha i .
Ei , Fj s d i j Ž 2.5g .
sinh Ž " .
with ² a i , a j : from Ž2.4b.. It is convenient to define
Kg s e " Hg ;g g E. Ž 2.6 .
The Hopf algebra structure maps are coproduct D: A" A" m A" , antipode S: A" ™ ™A
™
and co-unit ´ : A" C. They are explicitly defined via
"
D Ž H . s H m I q I m H ;H g H ,
D Ž Ei . s Ei m I q K a i m Ei ,
D Ž Fi . s Fi m Ky1
a i q I m Fi ,
S Ž H . s yS Ž H . ;H g H ,
S Ž Ei . s yKy1
a i Ei , S Ž Fi . s yFi K a i ,
´ Ž Hg . s ´ Ž Ei . s ´ Ž Fi . s 0 , Ž 2.7 .
where I is the identity, and provide the Z 2-graded Hopf algebra structure w7,17x. A"
allows for a decomposition
A" s Ny [ C w " x H [ Nq , Ž 2.8 .
2 2
where Nq and Ny are the subalgebras generated by Ei 4is0 and Fi 4is0 respectively.
Obviously Nq( Ny and C " H is a Cartan subalgebra.
ww xx
According to Yamane w18,19x Uq Ž sl Ž1. Ž2,1;C.. is the superalgebra A" , defined above,
i.e. by Ž2.4., Ž2.5., with the Hopf algebra structure Ž2.7. on which the standard Serre
relations, appearing already in the paper of Palev and Tolstoy w24x,
w Ei , Ei x s w Fi , Fi x s 0 for i s 0,1,2, Ž 2.9a .
E1 , w E1 , E 2 x s F1 , w F1 , F2 x s0 Ž 2.9b .
and Yamanes extra Serre relations w18x
E0 , E2 , E 0 , w E 2 , E1 x s E2 , E0 , E 2 , w E 0 , E1 x ,
F0 , F2 , F0 , w F2 , F1 x s F2 , F0 , F2 , w F0 , F1 x
Ž 2.10 .
are imposed. In Ž2.9b. and Ž2.10. wwP,P xx is defined via
pŽ a i . pŽ a j .
Ei , E j s Ei E j y Ž y1 . q ² a i , a j : E j Ei Ž 2.11 .
for the generators Ei , E j g Nq and similarly for Fi , Fj g Ny. For non-graded simple
affine Lie algebras Gabber and Kac proved the Serre theorem w22x. The q-deformed
relations appear naturally in Drinfel’d’s theory w7x. For affine simple Lie superalgebras
and their q-deformed counterparts Serre relations, were systematically constructed by
Yamane w18,19x. $
Uq Ž sl Ž2,1;C.. is the subalgebra of Uq Žsl Ž1. Ž2,1;C.., obtained from the latter by
omitting the generators with index 0 in the defining relations. The elements
EH :s w E1 , E2 x , FH :s w F1 , F2 x Ž 2.12a .
are canonically related to the highest root,
a H :s a 1 q a 2 , Ž 2.12b .
of Uq Ž sl Ž2,1;C... The notion of grading Ž2.5b. is naturally extended to the set Dq of all
non-zero, positiÕe roots of Uq Ž sl Ž2,1;C.., i.e. Dqs DŽ0.
q q
j DŽ1. , with
q q
DŽ0. s a14 , DŽ1. s a2 , aH 4 . Ž 2.13 .
$ Ž1.
$
Ž2,1;C.. is related to Uq sl Ž1. Ž2,1;C.. by
Finally U Ž gl Ž
$ $
q
Uq Ž gl Ž1. Ž 2,1;C . . s Uq Ž sl Ž1. Ž 2,1;C . . [ C w " x Hu , Ž 2.14a .

where Hu is an even central element, i.e. pŽ Hu . :s 0,
w Hu , Ei x s w Hu , Fi x s 0 for i s 0,1,2,
w Hu , H x s 0 ;H g H . Ž 2.14b .
So Hu may be included in the basis of H Ž2.4. and a trivial root a u s 0 can be
assigned to this element. The non-affine UqŽ gl Ž2,1;C.. is of course given as
Uq Ž gl Ž 2,1;C . . s Uq Ž sl Ž 2,1;C . . [ C w " x Hu . Ž 2.14c .
Obviously a decomposition Ž2.8. exists for all these quantum enÕeloping superalgebras.
It is well known that by assigning for instance some real value to " and formally taking
™
the limit " 0, the definition of the quantum uniÕersal enÕeloping superalgebras
Uq Ž g . of some Lie superalgebra g, as given above, omitting the Hopf algebra structure
Ž2.7., reduce to the proper definitions of its universal enveloping superalgebra UŽ g ..
3. Representations
The R-matrices to be constructed here, act on direct products of different carrier

spaces for representations of Uq Ž gl Ž2,1;C... It is therefore necessary to investigate these
spaces and their direct products.
™
The graded complex vector spaces Vd Ž2.1. are used as carrier spaces for the
finite-dimensional matrix representations G Vd :Uq Ž gl Ž2,1;C.. EndŽ Vd .. Operators from
EndŽ Vd . are expressed in the basis Ž2.2b.. They can also be written as a matrix S d , i.e.
A B
Sd s ž C D
,/ Ž 3.1a .
where A and D are eÕen d 0 m d 0 and d 1 m d 1 matrices, while B and C are the
complementary odd submatrices. The supertrace of S d Ž3.1a. is defined as
s-tr S d 4 s tr A4 y tr D 4 . Ž 3.1b .
Here it is useful to regard q g C in Ž2.5a. or
q s e 2h Ž 3.2a .
with h g C. For this choice some of the formal power series " may not converge. To
avoid this difficulty the generators Ha i g H can be eliminated in favor of the K a i Ž2.6..
All finite-dimensional irreducible representations of glŽN,1;C. for N 0 1 can be conti-
nously deformed into corresponding ones of UqŽ gl ŽN,1;C.., iff q is generic, that is not a
root of unity, which will be excluded here. Except if stated otherwise, h is assumed to
be real and positive and the results of Kac w16,23x remain valid. This was proved by
Palev and Tolstoy w24x. For a g A"
q a y qya sinh Ž 2h a .
w ax q s y1
s Ž 3.2b .
qyq sinh Ž 2h .
™
is defined as usual. Since sl Ž2,1;C. is basic w8,16x, a generalized, supersymmetric
Killing form B: H m H C is defined for any G Vd Ž2.1c. by
B Ž H , H X . s s-tr Vd G VdŽ H . G VdŽ H X . 4 ;H , H X g H . Ž 3.3a .
This definition could be extended to the full algebra. Here it is used to define the simple
roots a i Ži s 1,2.:
a i Ž H . s B Ž Ha i , H . ;H g H . Ž 3.3b .
The non-degenerate bilinear form ² P ,P : is realized as
² a , b :s B Ž Ha , Hb . ;a , b g E . Ž 3.3c .
™
Any representation G Vd can be made diagonal for all H g H by a convenient choice of
the basis for its carrier space Vd . The weights l i : H C are defined by
G VdŽ H . i :s l i Ž H . i :. Ž 3.4a .
A highest weight Õector c :g Vd , corresponding to the highest weight L Vd , is defined
by
G VdŽ a . c :s 0 ;a g Nq , Ž 3.4b .
where the decomposition Ž2.8. is applied to UqŽslŽ2,1;C... Being a functional on H , the
2
L Vd can be expressed in the basis a i 4is1 , i.e.
LVd s m 1 a 1 q m 2 a 2 . Ž 3.4c .
Note that m 1 , m 2 g C. The highest weights for any graded, irreducible representation are
unique and every other weight li of this representation can be written as
l i s LVd y p 1 a 1 y p 2 a 2 , Ž 3.4d .
where p 1 and p 2 are non-negative integers as usual. Two graded, irreducible representa-
tions are equivalent, iff they share the same highest weight w16x. This can be formulated
in a similar manner for lowest weights.
Uq Ž gl Ž2,1;C.. is naturally realized by its V3-module
V3at s Ž V3 , G V 3 . Ž 3.5a .
with Ž2,1.-grading on V3 Ž2.1c.. The generators are
G V 3Ž Ha 1 . s e11 y e 22 ,
G V 3Ž Ha 2 . s e 22 q e33 ,
G V 3Ž Hu . s e11 q e22 q e33 s IV 3 ,
G V 3Ž E1 . s e12 , G V 3Ž E2 . s e 23 ,
†
G V 3Ž Fi . s G V 3Ž Ei . for i s 1,2 . Ž 3.5b .
They even coincide with the ones in the fundamental representation of sl Ž2,1;C. w16x.
The defining relations of Uq Ž gl Ž2,1;C.., as given in Section 2, can be checked easily.
Ž G V ,V3 . is irreducible. The weights for the V3-module Ž3.5. are listed in Appendix A.
3
Note that an equivalent representations G VX d can be obtained from any graded
representation G Vd by means of a similarity transformation,
G VX dŽ a . s S d ( G VdŽ a . ( Sy1
d ;a g Uq Ž sl Ž 2,1;C . . . Ž 3.6 .
S d is block-diagonal, i.e. B and C vanish in Ž3.1a., in order to conserve grading. Apart
from a trivial exchange of even and odd subspace in Ž2.1., these are the only
equivalence transformations w16x.
V4 with a Ž2,2.-grading Ž2.1c. is the carrier space of another parameter dependent
representation, G V4 of Uq Ž gl Ž2,1;C..: The module is denoted by
V4t Ž C . :s Ž V4 , G V4 . . Ž 3.7a .
The generators are given explicitly as
G V4Ž Ha 1 . s e 33 y e 44 ,
G V4Ž Ha 2 . s C Ž e11 q e33 . q Ž C q 1 . Ž e22 q e44 . ,
G V4Ž Hu . s e11 q e 22 q e 33 q e44 ,
C Cq1
G V4Ž E1 . s e 34 , G V4Ž E2 . s yk q 2 e 31 q m q 2
e 42 ,
C Cq1
) y 2 ) y 2
G V4Ž F1 . s e 43 , G V4Ž F2 . s yk q e13 q m q e 24 . Ž 3.7b .
The complex numbers k and k ) as well as m and m ) are not necessarily complex
conjugate to each other, but related to the parameter C by
w C x q s k )k , w C q 1 x q s m )m . Ž 3.7c .
Obviously q-numbers Ž3.2b. enter Ž3.7c.. Using Ž3.7., Ž2.5c. and Ž3.3., the defining
relations, stated in Section 2, can be checked immediately. It can easily be seen that
different choices of k , k ) and m , m ) subject to Ž3.7c. can be related by similarity
transformations Ž3.6..The representation G V4 is irreducible, iff C / 0,y 1. The weights
ŽA.2. are listed in Appendix A, in particular
LV4 s yC a 1 y 2C a 2 . Ž 3.7d .
™
In the limit h 0 Ž3.2., G V4 is well known, because it contains the lowest-dimensional
typical representation of sl Ž2,1;C. w16,23x. Eq. Ž3.7. can be derived from this limit w24x.
An irreducible representation G Vd is called typical, if any reducible, graded represen-
tation G with the same highest weight LVd Ž3.4b. can be put into the form of a direct
sum of G Vd and some other representation w16,23x. Since Nq( Ny, highest weight can
also be replaced by lowest weight in this definition.
q
For h s 0 Kac proved w16,23x that G Vd is typical, iff there is no a g DŽ1. , such that
q Ž
2 a f D 2.13 , for which
.
² LV q r , a :s 0
d
Ž 3.8a .
with Ž2.4b. and

1 1
rs Ý ay Ý a Ž 3.8b .
2 q
ag DŽ0. 2 q
a g DŽ1.
holds. In particular for UŽ sl Ž2,1;C.. a representation G Vd can only be typical if d 0 s d1

in Ž2.1c.. This remains true, in particular for h real, positive. Using Ž3.7d. and Ž2.4b.,
the criterion Ž3.8. can be checked immediately for G V4 . Note that G V 3 Ž3.5. is atypical.
Reducibility does not imply total reducibility here.
Let k , k ) and m , m ) be complex conjugates respectively. So C is real and the
representation Ž3.7. is unitary. For reasons, to be explained later, a rescaling
k ™q y1 r2
k, k) ™q 1r2 )
k Ž 3.9 .
and similarly for m , m is applied. This can be regarded as a slight change of the
)
definition Ž3.7.. Representations of Uq Ž gl Ž2,1;C.. on tensor product spaces Ž2.3., are

constructed via the coproduct Ž2.7.. u1 :s 1 :m 1 : is a highest weight vector Ž3.4b. of
G V 3mV4 with
L s yC a 1 y Ž 2C q 1 . a 2 ,
while < Õ 2 : :s <2: m <3: is a lowest weight vector with
l2 s y Ž C q 2. a 1 y 2 Ž C q 2. a 2 .
Comparing with ŽA.2., it is clear that
G V 3mV4 ( G V 8 [ G V4 ,
with u1 :g V8 , Õ 2 :g V4 and, inspecting ŽA.2.,
t
žV ,G / (V
4 V4 4 Ž C q 1. . Ž 3.10a .
V8 carries a Ž4,4.-grading. G V8 with the highest weight L is irreducible for C / 0,y 1
and according to Kac’s criterion Ž3.8. typical. The special choice of parameters Ž3.9.
ensures that V3 m V4 s V8 [ V4 holds as a vector space direct sum. Finally G V8 with the
highest weight L is irreducible iff C / 0,y 1 and according to Kac’s criterion Ž3.8.
typical. The module is denoted by
t
ž V , G / :s V
8 V8 8 ŽC. . Ž 3.10b .
Therefore, in a symbolic, but convenient notation,
V3at m V4t Ž C . ( V8t Ž C . [ V4t Ž C q 1 . . Ž 3.10c .
The weight vectors ŽA.4., ŽA.5. and ŽA.9., expanded in the product basis of V3 m V4 , are
listed in Appendix A.
Ž V4X , G V X . ( V4t Ž C X . is defined to be the module V4t Ž C . with the parameters in Ž3.7b.
4
replaced
X X
by primed X
ones,
X
subject to the irreducibility condition C X / 0,y 1. In addition,
k Ž ., k Ž .) and mŽ ., mŽ .) will be taken to be complex conjugates to each other, without an
additional rescaling Ž3.9.. It is again easy to see that u˜ 1 :s 1 :m 1 : is a highest weight
vector Ž3.4b. of G V4mV4X with
L s yŽ C q CX . a1 y 2Ž C q CX . a 2
while w˜ 2 :s 2 :m 2 : is a lowest weight vector, corresponding to

l s y Ž C q C X q 2. a 1 y 2 Ž C q C X q 2. a 2 .
So from ŽA.2. it is clear that a decomposition
G V4mV4X ( G V˜4Ž1. [ G V˜8 [ G V˜4Ž2.
with
ž V˜ Ž1.
4 , G V˜4Ž1.
t
/ (V ŽCqC . ,
4
X
ž V˜ Ž 2.
4 , G V˜4Ž2.
t
/ ( V Ž C q C q 1.
4
X
Ž 3.11a .
exists, which also holds as a vector space direct sum. Explicit realizations of these
modules, ŽA.4. and ŽA.5., are given in Appendix A. Using the explicit realization of
ŽV &
˜ 8 , G ., also listed there ŽA.11., the relation
V8
ž Ṽ , G& / ( V
8 V8 8
t
Ž C q CX . Ž 3.11b .
is realized immediately. Similarly to Ž3.10. this yields
V4t Ž C . m V4t Ž C X . ( V4t Ž C q C X . [ V8t Ž C q C X . [ V4t Ž C q C X q 1 . . Ž 3.11c .
G V 3mV 3 can be analyzed similarly, leading to
V3at m V3at ( V5at [ V4t Ž 1 . . Ž 3.12 .
The weights ŽA.8. and an explicit realization of the weight vectors ŽA.7. on V3 m V3 of
V5at :s Ž Vˇ5 , G Vˇ5 . are listed in Appendix A. The analysis of representations here is solely
based on Kac’s criterion Ž3.8.. It requires only a little effort to construct the modules as
given in Appendix A. The Queensland group developed more sophisticated techniques
w9,12,13x, which are useful in order to deal with more complicated situations.
$
If G V is any finite-dimensional graded representation of Uq Ž gl Ž2,1;C.., an eÕalua-
tion representation G Vx , of Uq Ž gl Ž1. Ž2,1;C.. is constructed via
G Vx Ž Ha 0 . s K y IV m G V Ž Ha H . ,
G Vx Ž Ha i . s IV m G V Ž Ha i . for i s 1,2 ,
G Vx Ž Hu . s IV m G V Ž Hu . ,
G Vx Ž E0 . s x m G V Ž EH . ,
G Vx Ž Ei . s IV m G V Ž Ei . for i s 1,2 ,
G Vx y1
Ž F0 . s x m G V Ž FH . ,
G Vx Ž Fi . s IV m G V Ž Fi . for i s 1,2 Ž 3.13a .
for the generators. Here x is a letter like q in Section 2 and K a central element. For
q s 1, this is the well-known loop extension of UŽ gl Ž2,1;C.., see e.g. Refs. w16,23x.
$
Quantization of the latter w17x can be omitted by directly checking Yamanes defining
relations of UqŽ gl Ž1. Ž2,1;C.., as stated in Section 2.The representation Ž3.13a. is not
$
faithful, since the deriÕatiÕe, well known for q s 1 w16,23x, has been omitted. The
X
restricted superalgebra is called UqŽ gl Ž2,1;C...The evaluation representation is obtained
from Ž3.13a. by setting K equal to zero and the loop-variable x equal to a complex
number, i.e.
x s e2 u . Ž 3.13b .
Note that Ž3.13a. is not a Hopf-algebra automorphism. Also Ž3.13. is neither a highest
nor lowest weight representation. However, it was discovered by Drinfel’d w25x and
Jimbo w26x that this representation is related to the YBE Ž1.1b.. In particular u from
Ž3.13b. becomes a spectral parameter of difference type.
4. Fundamental R-matrices
An alternative definition of the coproduct D Ž2.7. is

DX s s ( D . Ž 4.1a .
Here s is defined on any Z 2-graded Hopf algebra A via
pŽ a. pŽ b .
s Ž a m b . s Ž y1 . bma
X
$ X
Ž 4.1b .
for all homogeneous a,b g A. Since D / D for q / 1, Uq gl 2,1;C.. is non-cocommu-
Ž Ž
tatiÕe. The antipode S has to be modified accordingly.
However, Drinfel’d realized that Uq Ž g . for g a finite or affine basic Lie Žsuper.alge-
bra can be represented as a quantum double D, or rather as its quotient by some central
element w7,17x. From his construction follows the existence of an invertible operator R,
the uniÕersal R-matrix, defined on D m D, with the properties
R ( D s DX ( R ,
Ž D m I . R s R13 R23 ,
Ž I m D . R s R13 R12 ,
Ž ´mI . Rs1sŽ Im´ . R ,
Ž S m I . R s Ry1 s Ž I m Sy1 . R . Ž 4.2a .
Here Rij means that R acts non-trivially on the ith and jth factor of the threefold
tensorproduct D m D m D. A Hopf-algebra A, equipped with such an R g A m ˆ A,
which is the completion of A m A within Drinfel’d’s quantum double construction, is
called quasi-triangular or a quantum group w7x. An immediate consequence of Ž4.2a. is
R12 R13 R23 s R23 R13 R12 , Ž 4.2b .
which is Drinfel’d’s form of the YBE w7x. The connection to the R-matrices appearing in
Ž1.1. is given by
X
R Vd VdX Ž u . s Ž G Vd m G VdXX . Ž t x m I . R X xs e 2 u
. Ž 4.3a .
Here G Vd and G VdX X are finite-dimensional representations of some finite Lie Žsuper.alge-
bra g, R X is the universal R-matrix for the restricted Žsuper.algebra UqŽgˆ X ., connected to
$
the evaluation representations Ž3.13. of the untwisted, affine UqŽgˆ Ž1. . and tx is the
X
Drinfel’d automorphism, defined on the generators of UqŽ gl Ž2,1;C.. by
t x Ž Ei . s x d 0 i Ei , t x Ž Fi . s xyd 0 i Fi ,
t x Ž H . s H ;H g H . Ž 4.3b .
Note that in Ž1.1b. the graded multiplication rule Ž2.3b. has to be applied. It is also
convenient to define
X
D Vd VdX :s Ž G Vd m G VdXX . ( D .
For all finite-dimensional quantized Lie superalgebras, e.g. for Uq Ž gl Ž2,1;C.. the
universal R-matrix has been constructed by Koroshkin and Tolstoy w28x. They also
provided a construction scheme for R X w29x, but the result X
would be of no use here,
because it is extremely cumbersome to evaluate X R Vd VdX Ž u. from Ž4.3.. However, the
existence of R X guarantees the existence of R Vd VdX Ž u. g EndŽ Vd m VdXX . for any pair of
irreducible, finite-dimensional representations G Vd , G VdXX of Uq Ž g ., with g finite, which
allow an extension Ž3.13. to the affinization Uq Ž gˆ X ..
The graded YBE itself was X
™
introduced by Kulish and Sklyanin long ago w27x. The
elementary intertwiner Pˇ Vd VdX :Vd m VdX VdXX m Vd is defined by its action
X
Pˇ V d V dX pŽ i . pŽ j .
i :m j :s Ž y1 . j :m i : Ž 4.4a .
on the canonical basis Ž2.3. of Vd m VdXX . Obviously
X X
Pˇ VdX Vd ( Pˇ Vd VdX s IVdmVdXX s IVd m IVdXX .
A more general intertwiner is defined as
X X X
RˇVd VdX Ž u . :s Pˇ Vd VdX ( R Vd VdX Ž u . . Ž 4.4b .
Immediately from Ž1.1b. and Ž4.4b. follows w27x
X XX XX X X XX X XX
Rˇ12 Ž u y Õ . RˇV23d VdX Ž u . Rˇ12 Ž Õ . s RˇV23d VdX Ž Õ . Rˇ12 Ž u . RˇV23dX VdX Ž u y Õ . .
V dX V dX X X V d V dX V d V dX X X
Ž 4.4c .
The interpretation as consistency equations for intertwiners is obvious. Moreover in this
form of the YBE additional signs Ž2.3b X
. due to grading vanish. These signs can be
VV Ž .
absorbed into the definition of R u in order to obtain a new solution to the
non-graded YBE Ž1.1b., see Ref. w27x. X
Having assured the existence of R V 3V 3 Ž u., R V 3V4 Ž u. and R V4 V4 Ž u. the simplest way of
constructing them is Jimbo’s method w14x, see also the book by Ma w30x: Using Ž4.1.,
Ž4.3. and Ž4.4. the first equation of Ž4.2a. can be written as
X X X X
RˇVd VdX Ž u . D Vd Vd Ž a . s D VdX Vd Ž a . Rˇ Vd VdX Ž u . Ž 4.5a .
for any a g UqŽglŽ2,1;C.. and almost all u g C. Using Ž3.13. this is done more explicitly
for E0 and F0 , i.e.
X
Ř Vd VdX Ž u . e 2 u G VdŽ Fa H . m IVdXX q G Vd Ž Ky1
a H . m G V dX Ž Fa H .
X
X
s G VdXX Ž Fa H . m IVd q e 2 u G VdXX Ž Ky1 ˇ V d V dX Ž u .
a H . m G V d Ž Fa H . R Ž 4.5b .
and
X
Ř Vd VdX Ž u . ey2 u G VdŽ Ea H . m G VdXX Ž K a H . q IVd m G VdXX Ž Ea H .
X
s G VdXX Ž Ea H . m G VdŽ K a H . q ey2 u IVdXX m G VdŽ Ea H . Rˇ Vd VdX Ž u . . Ž 4.5c .
X
Eqs. Ž4.5. are Jimbo’s equations. They determine Rˇ V d V dX Ž u. almost uniquely w14x. It is
easy to see that they imply
X X
RˇVd Vd Ž 0 . A IVdmVd , RˇVdX Vd Ž u . RˇVd VdX Ž yu . A IVdmVdXX .
X
Therefore a natural normalization of RˇVd VdX Ž u. is
RˇVd Vd Ž 0 . s IVdmVd , Ž 4.6a .

X X
RˇVdX Vd Ž u . RˇVd VdX Ž yu . s IVdmVdXX . Ž 4.6b .
Eq. Ž4.6a. is known as the standard initial condition,
X
while Ž4.6b. is the unitarity
ˇ VdVdX Žu. has the general form
condition. From Ž4.5. follows immediately that R
X X
RˇVd VdX Ž u . s N Ž u . P Ý r i Ž u . PiVd VdX , Ž 4.6c .
i
X
where PiVdVdX are elementary intertwiners, acting on the irreducible modules appearing
in the direct sum decompositions of Vd m VdXX . The r i Žu. are rational functions in x s e 2u
and NŽu. is an overall normalization factor with
r i Ž 0 . s 1 and r i Ž u . P r i Ž yu . s 1 ,
N Ž 0 . s 1 and N Ž u . P N Ž yu . s 1 . Ž 4.6d .
Eq. Ž4.6. is more or less a direct consequence of Schur’s Lemma w14x, which is
applicable, although there is a subtlety in the application to graded carrier spaces, see
e.g. Ref. w16x.
From Drinfel’d’s definition of the coproduct D Ž2.7. it is obvious that given the jth
weight vector u j Ž q . : from the carrier space VdŽii. ; Vd m VdXX of some irreducible
X
module, the corresponding weight vector uXj Ž q . : of V Ž i. ; VdXX m Vd is obtained as
X
uXj Ž q . :s Pˇ V d V dX
u j Ž qy1 . :. Ž 4.7a .
X X X X
Using Ž2.3c. and assuming q to be real, positive and k Ž ., k Ž .) as well as mŽ ., mŽ .) inX
Ž3.7. to be complex conjugates to each other, respectively. Jimbo’s intertwiners PiVdVdX
are given as
X
dim Ž V Ži. .
PiVd VdX s " Ý js1
uXj Ž q . :² u j Ž q . . Ž 4.7b .
X
The sign is of course fixed by the normalization condition on r in Ž4.6d.. The PiVd VdX
fulfill the following identity per construction:
X X X
PiVdX Vd ( PjVd VdX s d i j PîVd VdX . Ž 4.8a .
Here Pˆ i
X
VdVdX
: Vd m VdXX ™V Ži.
d i ; Vd m Vd
X
X is a projector:
X X X X
Pî
V d V dX
( Pˆj V d V dX
s d i j Pî V d V dX
, Ý PîV V d dX s IVdmVdXX .
i
X X
PiVdX Vd can be calculated directly from PiVdVdX via
X X X X
P VdX Vd s Pˇ V dX V d
( P Vd VdX ( Pˇ V dX V d
. Ž 4.8b .
From the normalization in Ž4.6. follows immediately
Ý PiV V s IV mV d d
d d
. Ž 4.8c .
i
X
The PiVdX Vd have been calculated for V3 = V4 Ž3.10., V4 = V4X Ž3.11. and V3 = V3 Ž3.12.
in Appendix B. For V3 = V4 the normalization N Ž u. in Ž4.6. is fixed by N Ž u. s ry1
2
Ž u.:
RˇV 3V4 Ž u . s r Ž u . P1V 3V4 q P2V 3V4 Ž 4.9a .
V3V4 V V
with P1 from ŽB.1. and P2 3 4 from ŽB.7.. This ansatz fulfills Ž4.5. per construction.
In order to determine r Žu. one has to solve simply one out of eleven non-trivial
components of, e.g., Eq. Ž4.5c.. This yields
sinh Ž h Ž C q 2 . q u .
r Ž u. s . Ž 4.9b .
sinh Ž h Ž C q 2 . y u .
It is however a valuable test to check all non-trivial equations from Ž4.5b. and Ž4.5c.,
which has been done.
Similarly for V4 m V4X the normalization is fixed by N Ž u. s ry1 2
Ž u.. So the ansatz
Ž4.6. reads
X X X X
RˇV4 V4 Ž u . s f Ž u . P1V4 V4 q P2V4 V4 q g Ž u . P3V4 V4 Ž 4.10a .
X
V4V4
with Pi for i s 1,2,3 from B.2 , B.8 and B.3 . From 4.5b and 4.5c follows
Ž . Ž . Ž . Ž . Ž .
sinh Ž h Ž C q C X . q u .
f Ž u. s ,
sinh Ž h Ž C q C X . y u .
sinh Ž h Ž C q C X q 2 . y u .
g Ž u. s . Ž 4.10b .
sinh Ž h Ž C q C X q 2 . q u .
Again all non-trivial components of Ž4.5b. and Ž4.5c. have been checked.
Actually f Ž u., g Ž u. have been known before w13x. They can be calculated in terms of
the Žsecond-order. Casimir operator, see e.g. Ref. w16x of the representations in Ž3.11.,
without knowledge of the intertwiners. The formula is apparently due to Reshetikhin
w32,33x and explained in detail in the textbook by Ma w30x. This formalism was further
developed by Delius et al. w12x and used by Bracken et al. to calculate f Ž u. and g Ž u.
w13x. In the limiting case V4 s V4X the intertwiners RˇV4 V4 Ž u. have been calculated by
Gould et al. w9x. For further details see Appendix B.
For completeness RˇV 3V 3 is given as
sinh Ž 2h y u . V V
Ř V 3V 3 s P1V 3V 3 q P 3 3 Ž 4.11 .
sinh Ž 2h q u . 2
with PiV 3V 3 for i s 1,2 from ŽB.4. and ŽB.5..
From the definition Ž4.4b. follows immediately
X X X
R Vd VdX Ž u . s Pˇ VdX Vd ( RˇVd VdX Ž u . .
This can be used to obtain the final results:
R V 3V4 Ž u . s r Ž u . e11 m Ž e11 q e 33 . q e 22 m Ž e11 q e 44 .
q a 0 Ž u . e11 m Ž e 22 q e 44 . q e 22 m Ž e 22 q e 33 . q e 33
m b 0 Ž u . e11 y e 22 q g 0 Ž u . Ž e 33 q e44 . q d 1 Ž u . e12 m e43
q d 2 Ž u . e 21 m e 34 y ´ 1 Ž u . w e13 m e 23 q e 23 m e 24 x
q ´ 2 Ž u . w e 31 m e 32 q e 32 m e 42 x y z 1 Ž u . e13 m e 41 y qy1 e 23 m e 31
q z 2 Ž u . w e 31 m e14 y q e 32 m e13 x , Ž 4.12 .
with r Ž u. from Ž4.9b. and the remaining coefficients ŽC.1. listed in Appendix C. The
q s 1-limit of this expression was obtained by Kulish in w34x. From Ž4.10. follows
X
R V4 V4 Ž u . s f Ž u . e11 m e11 q g Ž u . e 22 m e 22 y e33 m e33 y e44 m e44 q r5 e22 m e11
q r5X e11 m e 22 y r 10 Ž e 33 m e44 y e44 m e33 . y r 7 Ž e33 q e44 . m e11
y r 7X e11 m Ž e 33 q e44 . y r 9 Ž e 33 q e 44 . m e 22 y r 9X e 22 m Ž e 33 q e44 .
q r 1 e 21 m e12 q r 1X e12 m e 21 y r4 e 43 m e 34 y r4X e34 m e43
q r 2 Ž e 31 m e13 q e 41 m e14 . y r 2X Ž e13 m e 31 q e14 m e 41 .
q r 3 Ž e 32 m e 23 q e 42 m e 24 . y r 3X Ž e 23 m e 32 q e 24 m e 42 .
y r6 Ž e 24 m e13 y qy1 e 23 m e14 . q r 8X Ž e42 m e 31 y q e 32 m e 41 .
q r6X Ž e13 m e 24 y q e14 m e 23 . y r 8 Ž e 31 m e42 y qy1 e 41 m e32 .
Ž 4.13 .
X
with f Ž u. and g Ž u. from Ž4.10. and riŽ . for i s 1, . . . ,10 ŽC.2. also listed in Appendix
C. Note that the rescaled parameters Ž3.9. have been used in Ž4.9. but not in Ž4.12., in
order to unify the notation of Ž4.12. and Ž4.13..
Applying the same operation to Ž4.11. yields
R V 3V 3 Ž u . s e11 m e11 q e 22 m e 22 y d Ž u . e 33 m e 33
q c Ž u . e11 m Ž e 22 q e 33 . q e22 m Ž e11 q e 33 . q e33 m Ž e11 q e22 .
q a Ž u . w e 21 m e12 q e 31 m e13 q e32 m e23 x
q b Ž u . w e12 m e 21 y e13 m e 31 y e 23 m e32 x Ž 4.14 .
with aŽ u.,bŽ u.,cŽ u. and dŽ u. listed in ŽC.3. in Appendix C. Ž4.14. was known to Perk
and Schultz w35x. It is the standard q-deformation of the Yh Ž gl Ž2,1;C..-symmetrical
R-matrix listed by Kulish and Sklyanin w27x. Comparing Ž4.14. with the general
$
Yh Ž gl Ž m,n;C..-symmetrical R-matrix given there, it can immediately be generalized to
X
an Uq Ž gl Ž m,n;C..-symmetrical R-matrix R V mq nV mq n Ž u..
Also R V4 V 3 Ž u. can be obtained trivially from Ž4.12. via
X X X X X X
R VdX Vd Ž u . s Pˇ V dX V d
( R Vd VdX Ž u . ( Pˇ V dX V d
s Rˇ Vd VdX Ž u . ( Pˇ V dX V d
, Ž 4.15 .
X
V d V dX Ž
which holds for arbitrary R u.. This yields
R V4 V 3 Ž u . s r Ž u . Ž e11 q e 33 . m e11 q Ž e11 q e44 . m e 22 .
q a 0 Ž u . Ž e 22 q e 44 . m e11 q Ž e 22 q e 33 . m e 22
q b 0 e11 y e 22 g 0 Ž u . Ž e 33 q e 44 . m e 33 q d 1 Ž u . e 43 m e12
q d 2 Ž u . e 34 m e 21 q ´ 1 Ž u . w e 23 m e13 q e 24 m e 23 x
y ´ 2 Ž u . w e 32 m e 31 q e 42 m e 22 x q z 1 Ž u . e 41 m e13 y qy1 e 31 m e 23
y z 2 Ž u . w e14 m e 31 y q e13 m e 32 x , Ž 4.16 .
with r Ž u. from Ž4.9b. and the remaining coefficients from ŽC.1. in Appendix C.
X
$
Directly from the existence of the universal R-matrix for Uq Ž gl Ž2,1;C.. w29x, and
Drinfel’d’s YBE Ž4.2b. follows that Ž4.12. together with Ž4.14., both constructed as
solutions to Jimbo’s equations Ž4.5., fulfill the YBE
V 3V 3 V 3V4
R 12 Ž u y Õ . R13 Ž u . R V233V4 Ž Õ . s R V233V4 Ž Õ . R13
V 3V4 V 3V 3
Ž u . R12 Ž uyÕ. . Ž 4.17a .
It does not matter whether the rescaling Ž3.9. is applied or not, or whether k and k )
from Ž3.7. are complex conjugates of each other. The only restrictions on k , k ) , m and
m ) , appearing in Ž4.12., are indeed given by Ž3.7c.. Different choices of these
parameters, preserving Ž3.7c., are related by Ž3.6.. It is well known, see e.g. Ref. w27x,
that this leads to an invariance of the YBE Ž1.1b. under
X
R Vd Vd Ž u . ™ Ž S mS d
X
d
X
X
. R Vd VdX Ž u . Ž Sy1 X
d m Ž Sd .
X
y1
.,
X
where S d acts on G Vd and S dX on G VdX X . Note that C is allowed to be complex.Similarly
the R-matrix Ž4.13. fulfills the YBE
X XX X XX X XX XX X
V4 V4 V4 V4
R 12 Ž u y Õ . R13 Ž u . R V234 V4 Ž Õ . s R V234 V4 Ž Õ . R13
V4 V4 V4 V4
Ž u . R12 Ž uyÕ. Ž 4.17b .
and together with Ž4.12.:
X X X X
V 3V 4 V 3V4
R 12 Ž u y Õ . R13 Ž u . R V234 V4 Ž Õ . s R V234 V4 Ž Õ . R13
V 3V 4 V 3V4
Ž u . R12 Ž uyÕ. . Ž 4.17c .
The argument given above applies also to the other V4-representations in Ž4.17b.,
Ž4.17c.. The YBE Ž1.1b. with V s V X s V XX s V3 was known to Perk and Schultz w35x.
5. Fusion hierarchies and eigenvalue equations
An infinite number of R-matrices corresponding to larger irreducible representations

can be obtained from the fundamental ones by the fusion procedure of Kulish and
Sklyanin w15x. It is applicable iff a decomposition
Vd m VdXX ( VdŽ1.
1
[ VdŽ2.
2
Ž 5.1a .
X
VdVdX
exists, such that VdŽ1.
1
is irreducible and the corresponding R-matrix R Žu. with a
convenient normalization like 4.6 at some point u s u 0 becomes degenerate, i.e.
Ž .
X X X
R Vd VdX Ž u 0 . A Pˇ V dX V d
( P1Vd VdX , Ž 5.1b .
X X
where Pˇ V dX V d
( P1 V d V dX
maps Vd m VdXX into VdŽ1.
1
; Vd m VdXX . Following Kulish and
Sklyanin, it is useful to define
X XX XX X XX
X V d V dX X X
ŽV dmV dX .V dX
R 12 ,3 Ž u . :s R13 Ž u q u1 . R V23dX VdX Ž u y u 2 . , Ž 5.2a .
XXX
where Vd is arbitrary and u 1 q u 2 s u 0 . From Ž1.1b. and Ž4.8. follows
X
X X X XX X X
0 s Ž Pˇ Ž u . Ž Pˇ
V dX V d X
( P1Vd VdX . 12 R 12,3
ŽV dmV dX .V dX V dX V d
( P2Vd VdX . 12 .
This is equivalent to the triangularity condition
X X XX X
Ž Pˆ1V V . 12 R12,3
d dX ŽV mV d dX .V dX
X
Ž u . Ž Pˆ2Vd VdX . 12 s 0 . Ž 5.2b .
X
Here Pˆ i
VdVdX
for i s 1,2 are the proper projectors from Vd m Vd on VdŽi.i Ž4.8a.. Note that X
X
only this property is necessary for fusion. In particular there is no need for an underlying
group structure. This is illustrated e.g. in a very recent preprint by Maassarani w31x.
Using Ž5.2b., Ž4.8. and the YBE Ž1.1b. again, it can be concluded that
XX X X XX X
Ž u . :s Ž Pˆ1Vd VdX . 12 R12,3 Ž u . Ž Pˆ1Vd VdX . 12
Ž1.
V d V dX X ŽV dmV dX .V dX X
R Ž12.3
1 Ž 5.2c .
is indeed an R-matrix, i.e. a solution to the YBE, acting in VdŽ1.
1
m VdXXXX , with VdŽ1.
1
; Vd m
X
VdX . Similarly from the definition
XX X XX XX X
X
V dX ŽV dmV dX . V dX V d X
V d X V dX X
R 1,23 Ž u . :s R13 Ž u q u1 . R12 Ž u y u2 . Ž 5.3a .
and the corresponding triangularity condition
X XX X X
Ž Pˆ1V V . 23 R1,23
d dX V ŽV mVdX
X
d dX .
Ž u . Ž Pˆ2Vd VdX . 23 s 0 Ž 5.3b .
the R-matrix acting in VdXXXX m VdŽ1.
1
with VdXXXX arbitrary and VdŽ1.
1
; Vd m VdXX can be derived
analogously
XX X XX X X
X
V dX V d
R 1Ž23.
Ž1.
ˆ1Vd VdX . 23 R1,23
1 Ž u . :s Ž P
V dX ŽV dmV dX .
Ž u . Ž Pˆ1Vd VdX . 23 .
X
Ž 5.3c .
The results Ž5.2. and Ž5.3. can be combined to
X X X X X X
V d V dX Vd Vd
ŽV dmV dX .ŽV dmV dX .
R 12 ,34 Ž u . :s R14 Ž u q u 0 . R13 Ž u . R V24dX VdX Ž u . R V23dX Vd Ž u y u 0 . , Ž 5.4a .
which fulfills the triangularity conditions
X X X X
Ž Pˆ1V V . 12 R12,34
d dX ŽV mV .ŽV mV .
d dX
Ž u . Ž Pˆ2V V . 12 s 0,
d dX d dX
X X X X
Ž Pˆ1V V . 34 R12ŽV ,34mV .ŽV mV . Ž u . Ž Pˆ2V V . 34 s 0

d dX d dX d dX d dX
Ž 5.4b .
and leads to the proper R-matrix
X X X X X X
Ž1.
Vd Vd
R Ž12.Ž34.
1
Ž1.
ˆ1Vd VdX . 12 Ž Pˆ1Vd VdX . 34 R12,34
1 Ž u . :s Ž P
ŽV dmV dX .ŽV dmV dX .
Ž u . Ž Pˆ1Vd VdX . 12 Ž Pˆ1Vd VdX . 34
Ž 5.4c .
mapping VdŽ1. 1
m VdŽ1.
1
onto VdŽ1.1
m VdŽ1. 1
with VdŽ1. 1
; Vd m VdXX . In Ž5.4. all quantities act
naturally on the 4-fold tensor product Ž Vd m VdXX . m Ž Vd m VdXX ., as indicated. Further
generalizations are obvious.
From Ž3.10. and Ž3.12. R V5V 3 Ž u., R V 8V 3 Ž u., R V 5V4 Ž u. and R V 8V4 Ž u. can be calculated
from Ž5.2., while R V 3V 5 ŽXu., R V 3V 8 Ž u., R V4 V5 Ž u. and R V4 V 8 Ž u. can be calculated from Ž5.3.
and R V 5V 5 Ž u. and R V 8V 8 Ž u. can be obtained from Ž5.4., using the results of Section 4.
Note however that R V 5V5 Ž u. is a straightforward generalization of R V 3V 3 Ž u. Ž4.14.. An
infinite family of R-matrices R V 2 nq 1V 2 nq 1 Ž u. can be constructed along these lines, where
X
$
V2 nq1 is atypical. This is very similar to the non-graded Uq Ž gl Ž2;C..-case, see e.g. Ref.
w17x. However, only the physically interesting modules V3at and V5at will be treated here.
All these R-matrices could of course be constructed from Ž4.5.. The fusion procedure
in addition allows to obtain Bethe ansatz expressions for the eigenvalues, provided some
simple ones are known to serve as starting points for these fusion hierarchies.
at
For V s V3at and V Ž n. s V4t Ž C Ž n. . for n s 1, . . . , N the eigenvalues of tˆ V 3 Ž1.2b. can
be obtained e.g. by using the nested algebraic Bethe ansatz, due to Kulish and
Reshetikhin w39x: Applying the N-fold direct product V l w : of lowest weight vectors
2 :n to the monodromy matrix Ž1.2a., represented as a matrix in auxiliary space V ( V3at ,

one obtains
vŽN1 . Ž u . 0 0
V 3at
Tâ Ž 1, N < u . V l w :s
0
Bˆ1 Ž u y h .
v NŽ2. Ž u .
Bˆ2 Ž u y h .
0
v NŽ3. Ž u. 0 a
V l w :. Ž 5.5 .
Here the v NŽ i. Ž u. are some functions of the general form

N
Ž n.
v NŽ i. Ž u . s Ł v iV Ž u y d Ž n. . , Ž 5.6a .
ns1
Žn.
where d Žn. are the inhomogeneities from Ž1.2a.. v Vi Žu. are the local Õacuum ampli-
tudes. They are indexed according to the basis of V, in which the corresponding
diagonal elements are taken. Bˆ i Žu. are some operators acting on quantum space. The
actual eigenvectors can be written as
< l1 , . . . , l M < m 1 , . . . , m m : s F a1 . . . a M Ž m 1 , . . . , m m . Bâ Ž l1 . . . . Bâ Ž l M . V :,

1 M
Ž 5.6b .
where l1 , . . . , l M and m 1 , . . . , mm are two sets of spectral parameters determined from

the Bethe equations. It is preferable to take products of local highest or lowest weight
vectors as pseudo-vacuum V :, since in this case the remaining eigenvectors can be
obtained from these by applying Žquantum-.group shift operators. For details see Ref.
w39x. The essence of the method is that eigenvalues and eigenvectors can be obtained
algebraically from the fundamental commutation relations
X X X X
R Va1Va 2Ž u y Õ . TâV1 Ž 1, N < u . TâV2 Ž 1, N < Õ . s TâV2 Ž 1, N < Õ . TâV1 Ž 1, N < u . R Va1Va 2Ž u y Õ . ,
Ž 5.6c .
which are a consequence of the YBE Ž1.1b.. For V s V X s V3at and V :s V l w : the
Bethe equations read w11x
N sinh Ž la y d Ž i. q Ž C q 1 . h . m sinh Ž la q m k q h .
Ł sinh Ž l Ž i.
s Ł Ž 5.7a .
is1 a yd y Ž C q 1. h . ks1 sinh Ž la q m k y h .
for a s 1, . . . ,M and
M sinh Ž l j q mb q h . m sinh Ž mb y m k q 2h .
Ł s Ł Ž 5.7b .
js1 sinh Ž l j q mb y h . ks1, k/ b sinh Ž mb y m k y 2h .
for b s 1, . . . ,m. Here it is convenient to use an analytical normalization. This means

here and in the following all R-matrices from Section 4 are multiplied with a factor,
which simply absorbs all analytical denominators due to the normalization condition in
at
Ž4.6a. in all coefficients as listed in Appendix C. After that the eigenvalues of tˆ V 3 Ž N < u.
to Ž5.6b. read
at
LVN3 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j q h . m sinh Ž u y m k y 2h .
s v NŽ1. Ž u .
½ Ł Ł
js1 sinh Ž u y l j y h . ks1 sinh Ž u y m k .
m sinh Ž u y m k q 2h . M sinh Ž u y l j q h .
qŁ
ks1 sinh Ž u y m k . 5 y v NŽ 3 . Ž u . Ł
js1 sinh Ž u y l j y h .
Ž 5.8a .
with the local vacuum amplitudes as in Ž5.5.

t t
v 1V4 ŽC . Ž u . s v V2 4 ŽC . Ž u . s sinh Ž u q Ch . ,
t
v V3 4 ŽC . Ž u . s sinh Ž u y Ž C q 2 . h . . Ž 5.8b .
The analytic parameterization assures that the Bethe equations Ž5.7. are also analyticity
conditions for the eigenvalue Ž5.8..
For V s Vt 4 Ž C X . as auxiliary space the recipe of Kulish and Reshetikhin does not
work – mainly because Ž5.5. is more complicated. In a similar context this was first
realized by Maassarani, who also provided arguments to conjecture the eigenvalue
equation w40x. For a rigorous derivation it is convenient to use Ž5.6c. with V s V3at and
V X s V4t Ž C X . and the eigenvectors Ž5.6b. to obtain
t X
LVN4 ŽC . Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j q Ž C X q 1 . h .
s v NŽ1. Ž u . Ł
js1 sinh Ž u y l j y Ž C X y 1 . h .
q v NŽ2. Ž u . Ł
js1 sinh Ž u y l j y Ž C X q 1 . h .
M sinh Ž u y l j q Ž C X q 1 . h . m sinh Ž u y m k y Ž C X y 2 . h .
y v NŽ3. Ž u. ½ Ł
Ł
ks1 sinh Ž u y m k y C Xh .
M sinh Ž u y l j q Ž C X q 1 . h . m sinh Ž u y m k y Ž C X q 2 . h .
qŁ
Ł
ks1 sinh Ž u y m k y C Xh . 5 .
Ž 5.9a .
For details see Ref. w11x. The local vacuum amplitudes are
t
v 1V4 ŽC . Ž u . s sinh Ž u q Ž C y C X . h . sinh Ž u q Ž C y C X q 2 . h . ,
t
v V2 4 ŽC . Ž u . s sinh Ž u y Ž C q C X q 2 . h . sinh Ž u y Ž C q C X . h . ,
t t
v V3 4 ŽC . Ž u . s v4V4 ŽC . Ž u . s sinh Ž u q Ž C y C X . h . sinh Ž u y Ž C q C X . h . . Ž 5.9b .
Naturally the Bethe equations Ž5.7. are also analyticity conditions for Ž5.9.. Note that
Ramos and Martins developed another Bethe ansatz scheme w41x, which was established
as a general, constructive method recently w42x and could be used here. With u 0 s u1 q u 2
the fusion relations Ž5.2. for R-matrices can be extended trivially to transfer matrices
X 1 2
tˆ Vd Ž N < u q u1 . P tˆ VdX Ž N < u y u 2 . s tˆ Vd1 Ž N < u . q tˆ Vd 2 Ž N < u . . Ž 5.10 .
The transfer matrices on the right-hand side are fixed only up to a shift in the spectral
parameter u and a normalization factor.
From Ž3.12. and Ž4.14. the conditions Ž5.1. are fulfilled with u 0 s 2h. The analytical
normalization yields
at at
tˆ V 3 Ž N < u q h . P tˆ V 3 Ž N < u y h .
N
t at
s tˆ V4 Ž1. Ž N < u . q
ž /
Ł sinh Ž u y dŽ i . q Ž C q 1. h . tˆ V Ž N < u . .
is1
at
5
From this Ž5.8a. and Ž5.9a. the eigenvalue for tˆ V5 Ž N < u. in the eigenvectors Ž5.6b. can
be calculated immediately:
at
LVN5 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j q 2h . m sinh Ž u y m k y 3h .
s v NŽ1. Ž u .
½Ł Ł
js1 sinh Ž u y l j y 2 h . ks1 sinh Ž u y m k q h .
M sinh Ž u y l j . m sinh Ž u y m k q 3h . sinh Ž u y m k y 3h .
qŁ Ł
js1 sinh Ž u y l j y 2 h . ks1 sinh Ž u y m k q h . sinh Ž u y m k y h .
m sinh Ž u y m k q 3h .
qŁ
ks1 sinh Ž u y m k y h . 5
M sinh Ž u y l j q 2h . m sinh Ž u y m k y h .
y v NŽ 4 . Ž u .
½Ł Ł
js1 sinh Ž u y l j y 2 h . ks1 sinh Ž u y m k q h .
M sinh Ž u y l j . m sinh Ž u y m k q 3h .
qŁ Ł
js1 sinh Ž u y l j y 2 h . ks1 sinh Ž u y m k q h . 5 Ž 5.11a .
with the local vacuum amplitudes

t t t
v 1V4 ŽC . Ž u . s v V2 4 ŽC . Ž u . s v V3 4 ŽC . Ž u . s sinh Ž u q Ž C y 1 . h . ,
t t
v4V4 ŽC . Ž u . s v 5V4 ŽC . Ž u . s sinh Ž u y Ž C q 3 . h . . Ž 5.11b .
The latter are indexed according to the weight vector basis ŽA.7. given in Appendix A.
The identification of these terms is unique. Combining the fusion procedure with some
¨ where able to conjecture the
reasonable analyticityX assumptions Frahm and Pfannmuller
t
expression for LVN4 ŽC . in the q s 1-limit w43x. Tsuboi extended their arguments to the
trigonometric case w44,45x. Since Ž5.9a. has been derived algebraically w11x the deriva-
tion of Ž5.11. does not involve any assumption.
Similarly from Ž3.10. and Ž4.12. the conditions Ž5.1. are fulfilled with u 0 s yŽ C X q
2.h. This yields
at t X
tˆ V 3 Ž N < u y Ž C X q 1 . h . P tˆ V4 ŽC . Ž N < u q h .
N X
t
s
ž Ł sinh Ž u y dŽ i . q Ž CŽ i . y C q 1. h . tˆ V ŽC . Ž N < u .
is1
X
/ 8
N X
t
y
ž is1 /
Ł sinh Ž u y dŽ i . y Ž CŽ i . q C y 1. h . tˆ V ŽC q1. Ž N < u . .
X 4
t X
Again from this, Ž5.8a. and Ž5.9a. the eigenvalue for tˆ V 8 ŽC . to the eigenvectors
Ž5.6b. can be calculated:
t X
LVN8 ŽC . Ž u < l1 , . . . , l M < m 1 , . . . , m m .
s v NŽ1. Ž u . Ł
M sinh Ž u y l j y C Xh . m sinh Ž u y m k y Ž C X q 3 . h .
=
½Ł js1 sinh Ž u y l j y Ž C X q 2 . h .
Ł
ks1 sinh Ž u y m k y Ž C X q 1 . h .
m sinh Ž u y m k y Ž C X y 1 . h .
qŁ
ks1 sinh Ž u y m k y Ž C X q 1 . h . 5
q v NŽ 3 . Ž u . Ł
js1 sinh Ž u y l j y Ž C X q 2 . h . ½ Ł
ks1 sinh Ž u y m k y Ž C X y 1 . h .
M sinh Ž u y l j y C Xh . m sinh Ž u y m k y Ž C X y 3 . h .
qŁ
Ł
ks1 sinh Ž u y m k y Ž C X y 1 . h . 5
M sinh Ž u y l j y C Xh . sinh Ž u y l j q Ž C X q 2 . h .
y v NŽ 5 . Ž u. Ł
js1 sinh Ž u y l j y Ž C X q 2 . h . sinh Ž u y l j y Ž C X y 2 . h .
y v NŽ 6 . Ž u. ½ Ł
Ł
M sinh Ž u y l j y Ž C X . h . sinh Ž u y l j q Ž C X q 2 . h .
qŁ
js1 sinh Ž u y l j y Ž C X q 2 . h . sinh Ž u y l j y Ž C X y 2 . h .
m sinh Ž u y m k y Ž C X q 3 . h . sinh Ž u y m k y Ž C X y 3 . h .
=Ł
ks1 sinh Ž u y m k y Ž C X q 1 . h . sinh Ž u y m k y Ž C X y 1 . h .
M sinh Ž u y l j q Ž C X q 2 . h . m sinh Ž u y m k y Ž C X y 3 . h .
qŁ
Ł
ks1 sinh Ž u y m k y Ž C X q 1 . h . 5 .
Ž 5.12a .
The local vacuum amplitudes indexed corresponding to ŽA.9. are

t t
v 1V4 ŽC . Ž u . s v V2 4 ŽC . Ž u . s sinh Ž u q Ž C y C X y 1 . h . sinh Ž u q Ž C y C X q 3 . h . ,
t t
v V3 4 ŽC . Ž u . s v4V4 ŽC . Ž u . s sinh Ž u y Ž C q C X q 3 . h . sinh Ž u q Ž C q C X . h . ,
t
v 5V4 ŽC . Ž u . s sinh Ž u y Ž C q C X q 3 . h . sinh Ž u q Ž C y C X q 3 . h . ,
t t t
v6V4 ŽC . Ž u . s v V7 4 ŽC . Ž u . s v V8 4 ŽC . Ž u .
s sinh Ž u q Ž C y C X y 1 . h . sinh Ž u y Ž C q C X y 1 . h . . Ž 5.12b .
The above remarks apply similarly to this result. In some sense the representa- V4t Ž C .
tion is an extension of the V3at representation, to an equivalent of which it is reduced for
C s 0. This becomes obvious in the present Bethe ansatz construction: For and
V s V Ž n. s V3at for n s 1, . . . , N in Ž1.2a. it is essentially the same: The global lowest
weight vector V l w :, which is a direct product of the local ones 3 :n acts according to
at
Ž5.5. on TâV 3 Ž1, N < u.. This implies that the algebra from Ž5.6c. is also the same. In
particular the eigenvectors are given by Ž5.6b., though with different operators Bî Ž u.. It
turns out that the local vacuum amplitudes for V3at and V4t Ž C X . as auxiliary spaces are
just the C Ž n. s 0 limits from Ž5.8. and Ž5.9.. In particular Ž5.8. in this limit was first
derived algebraically by Schultz w46x. It has been recalculated partially with trivial
modifications like the one leading to Ž5.8. several times in the literature. Clearly the
corresponding fusion results are also the C Ž n. s 0 limits of Ž5.11. and Ž5.12., so that they
can be omitted. It should be noted that this allows to use a mixed quantum space, built
up from local V3at- and V4t Ž C Ž n. .-representations – this results just in the substitution of
the corresponding local vacuum amplitudes in the equations given here. Unfortunately
the V5at- and V8t Ž C Ž n. .-representations do not share this property!
Ž n.
Note that the ratios of the local vacuum amplitudes v iV Ž u. from Ž5.6a. and the
positions of the Bî Ž u.-operators in the lowest fundamental representation, e.g. like in
Ž5.5., determine the algebra via Ž5.6c. and therefore the structure of higher eigenvalue
equations for the same eigenvector like Ž5.6b. completely.
For V5at as auxiliary space the structure is different: The direct product of two lowest
weight vectors 3 : from V3at or from V3at and V4t Ž C . Ž 2 :. lies in a V4t Ž1.-module, ŽA.3.,
and in the V4t Ž C q 1.-module ŽA.4., respectively, while the direct product of highest
weight vectors – in both cases 1 : – lies entirely in the V5at-module, ŽA.7., and
Vt 8 Ž C .-module ŽA.9., respectively. From Ž3.12. and Ž5.1. the second fusion relation
Ž5.3c. can be used with u 0 s 2h from Ž4.14. in order to obtain the local vacuum
at
amplitudes v iV 5 Ž u. for auxiliary spaces V s V3at ,V4t Ž C X . in Ž1.2. as
at at at
v iV 5 Ž u . , v iV 3 Ž u q h . P v iV 3 Ž u y h . , Ž 5.13 .
This equation holds as an equality only modulo a common factor in all vacuum
amplitudes.
From the definition Ž1.2a. the action of the monodromy matrix on the direct product
V hw : of local highest weight vectors uˇ 1 :n from ŽA.7. reads
vŽN1 . Ž u . Bˆ1 Ž u q h . Bˆ2 Ž u q h .
V 3at
Tâ Ž 1, N < u . V hw :s
0
0
v NŽ2. Ž u .
0
0
v NŽ3. Ž u. 0 a
V hw : , Ž 5.14 .
which is apparently different from Ž5.5. and requires a different Bethe ansatz, although
the eigenvectors take the form Ž5.6b., due to the convenient choice of Bî Ž u. in Ž5.14..
For V Ž n. s V5at for n s 1, . . . , N in Ž1.2a. the action on the direct product of highest
weight vectors 1 :n is precisely the same. From the previous remarks it is sufficient to
solve this problem and replace the local vacuum amplitudes according to Ž5.13.. Note
that V Ž n. s V3at ,V5at can be mixed arbitrarily in Ž1.2.. This contributes just the corre-
sponding local vacuum amplitude to the eigenvalue equation via Ž5.6a.. The Bethe
equations are
Ž n.
N v 1V Ž la y d Ž i. y h .
Ł Ž n.
is1 v V2 Ž la y d Ž i. y h .
M sinh Ž la y l j q 2h . m sinh Ž la y m k q h .
s Ł Ł Ž 5.15a .
js1, j/ a sinh Ž la y l j y 2 h . ks1 sinh Ž la y m k y h .
for a s 1, . . . ,M and
M sinh Ž l j y mb q h .
Ł s1 Ž 5.15b .
js1 sinh Ž l j y mb y h .
for b s 1, . . . ,m. Following Refs. w39,46x the eigenvalues read
at
LVN3 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j y h .
s v NŽ1. Ž u . Ł
js1 sinh Ž u y l j q h .
M sinh Ž u y l j q 3h . m sinh Ž u y m k .
q v NŽ2. Ž u .
½ Ł
js1 sinh Ž u y l j q h .
y1
5 Ł
ks1 sinh Ž u y m k q 2h .
,
Ž 5.16a .
with local vacuum amplitudes, which enter Ž5.15a., from
at
v 1V 3 Ž u . s sinh Ž u q 2h . ,
at at
v V2 3 Ž u . s v V3 3 Ž u . s sinh Ž u . ,
at
at at
v V2 5 Ž u . s v V3 5 Ž u . s sinh Ž u y h . . Ž 5.16b .
$
In contrast to Ž5.7. the second Žnested. Bethe ansatz is graded and involves the
X
Uq Ž gl Ž1,1;C..-symmetric R-matrix. At least superficially any similarity to the equiva-
lent Bethe ansatz based on Ž5.7. is lost. However, the equivalence of eigenvalue
equations can be shown using standard methods from the theory of complex functions.
For V s V4t Ž C X . in Ž1.2b. it is convenient to apply the strategy from w11x. The
calculations are involved, but since the method has been explained in detail there, it is
sufficient to state the result:
t X
LVN4 ŽC . Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j q Ž C X y 1 . h . m sinh Ž u y m k y C Xh .
s v NŽ1. Ž u . 1 y Ł
½ js1 sinh Ž u y l j q Ž C X q 1 . h . 5 Ł
ks1 sinh Ž u y m k q C Xh .
q v NŽ2. Ž u . 1 y Ł
½ js1 sinh Ž u y l j q Ž C X q 1 . h . 5
m sinh Ž u y m k y C Xh .
=Ł Ž 5.17a .
ks1 sinh Ž u y m k q Ž C X q 2 . h .
at at
v 1V 3 Ž u . s v V3 3 Ž u . s sinh Ž u q Ž C X q 2 . h . ,
at at
v V2 3 Ž u . s v4V 3 Ž u . s sinh Ž u q C Xh . ,
at at
v 1V 5 Ž u . s v V3 5 Ž u . s sinh Ž u q Ž C X q 3 . h . ,
at at
v V2 5 Ž u . s v4V 5 Ž u . s sinh Ž u q Ž C X y 1 . h . . Ž 5.17b .
The fusion is completely analogous to the previous one:
N
at at at
tˆ V 3 Ž N < u q h . P tˆ V 3 Ž N < u y h . s
ž /
Ł sinh Ž u y dŽ i . q h . tˆ V Ž N < u .
is1
5
N
t
q
ž is1 /
Ł sinh Ž u y dŽ i . y h . tˆ V Ž1. Ž N < u .
4
Ž n.
for V s V3at
and similarly for V Ž n.
s V5at
for n s 1, . . . , N, which will be left for the
reader. From this, Ž5.16a. and Ž5.17a. follows
at
LVN5 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j y 2h . m sinh Ž u y m k y h .
s v NŽ1. Ž u . Ł Ł
js1 sinh Ž u y l j q 2 h . ks1 sinh Ž u y m k q h .
M sinh Ž u y l j y 2h . m sinh Ž u y m k q h .
q v NŽ2. Ž u . Ł Ł
js1 sinh Ž u y l j . ks1 sinh Ž u y m k q 3h .
M sinh Ž u y l j q 4h . M sinh Ž u y l j q 4h .
=
½ Ł
js1 sinh Ž u y l j q 2h . 5
y 1 q v NŽ3. Ž u .
½ Ł
js1 sinh Ž u y l j .
M sinh Ž u y l j y 2h . m sinh Ž u y m k y h .
yŁ
js1 sinh Ž u y l j . 5 Ł
ks1 sinh Ž u y m k q 3h .
Ž 5.18a .

at
at at
v V2 3 Ž u . s v4V 3 Ž u . s sinh Ž u q h . ,
at at
v V3 3 Ž u . s v 5V 3 Ž u . s sinh Ž u y h . ,
at
v 1V 5 Ž u . s sinh Ž u q 4h . sinh Ž u q 2h . ,
at at
v V2 5 Ž u . s v4V 5 Ž u . s sinh Ž u . sinh Ž u q 2h . ,
at at
v V3 5 Ž u . s v 5V 5 Ž u . s sinh Ž u . sinh Ž u y 2h . . Ž 5.18b .
at
These v iV5 Ž u. can also be obtained directly from Ž4.16., Ž4.13. and ŽA.7., which X
is a
t
valuable check. For V Ž n. s V3at the corresponding eigenvalue equations for tˆ V 8 ŽC . Ž N < u.
can be obtained from the fusion relation
at t X
tˆ V 3 Ž N < u y Ž C X q 1 . h . P tˆ V4 ŽC . Ž N < u q h .
X
N X
at t
s tˆ V 8 ŽC .
Ž N < u. y ž is1
X
/
Ł sinh Ž u y dŽ i . y Ž C q 1. h . tˆ V ŽC q1. Ž N < u . .
4
For V Ž n. s V5at the relation is similar and shall be left to the reader. Note that the actual
local quantum spaces V3at and V5at can be mixed in the fusion relations too. This leads of
course to ‘‘mixed prefactors’’.
For V s V3at and V Ž n. s V4t Ž C Ž n. . in Ž1.2. the action of the monodromy matrix Ž1.2a.
on the direct product V hw : of local highest weight vectors 1 : can be obtained trivially
and leads again to a different pattern:
vŽN1 . Ž u . 0 Bˆ1 Ž u q 2h .
V 3at
Tâ Ž 1, N < u . V hw :s
0
0
v NŽ2. Ž u .
0
Comparing with Ž5.14., one concludes that the corresponding Bethe ansatz is entirely
Bˆ2 Ž u q 2h .
v NŽ3. Ž u. 0 a
V hw :. Ž 5.19 .
different and local quantum spaces cannot be mixed. However, the eigenvectors still
take the form Ž5.6b. and the whole calculation is very similar to the lowest weight Bethe
ansatz. In particular the Bethe equations read
N sinh Ž la y d Ž i. y Ch . m sinh Ž la q m k q h .
Ł sinh Ž l Ž i.
sŁ Ž 5.20a .
is1 a yd q Ch . ks1 sinh Ž la q m k y h .
for a s 1, . . . ,M and
M sinh Ž l j q mb q h . m sinh Ž mb y m k q 2h .
Ł s Ł Ž 5.20b .
js1 sinh Ž l j q mb y h . ks1, k/ b sinh Ž mb y m k y 2h .
for b s 1, . . . ,m. For V s V3at in Ž1.2. the corresponding ‘‘highest weight eigenvalue
equations’’ read
at
LVN3 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j . m sinh Ž u q m k q 3h .
s v NŽ1. Ž u .
½ Ł Ł
js1 sinh Ž u y l j q 2 h . ks1 sinh Ž u q m k q h .
m sinh Ž u q m k y h . M sinh Ž u y l j .
qŁ
ks1 sinh Ž u q m k q h . 5 y v NŽ 3 . Ž u . Ł
js1 sinh Ž u y l j q 2h .
Ž 5.21a .
with local vacuum amplitudes from

t t
v 1V4 ŽC . Ž u . s v V2 4 ŽC . Ž u . s sinh Ž u q Ž C q 2 . h . ,
t
v V3 4 ŽC . Ž u . s sinh Ž u y Ž C y 2 . h . . Ž 5.21b .
Similarly for V s V4t Ž C X . and V Ž n. s V4t Ž C Ž n. . in Ž1.2. the eigenvalue equations for the
highest weight pseudo-vacuum V hw : can be written as
t X
LVN4 ŽC . Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j y C Xh . M sinh Ž u y l j y C Xh .
s v NŽ1. Ž u. Ł q v NŽ2. Ž u. Ł
js1 sinh Ž u y l j q C Xh . js1 sinh Ž u y l j q Ž C X q 2 . h .
M sinh Ž u y l j y C Xh . m sinh Ž u q m k q Ž C X q 1 . h .
y v NŽ3. Ž u. ½Ł js1 sinh Ž u y l j q Ž C X q 2 . h .
Ł
ks1 sinh Ž u y m k q Ž C X y 1 . h .
M sinh Ž u y l j y C Xh . m sinh Ž u y m k q Ž C X y 3 . h .
qŁ
js1 sinh Ž u y l j q C Xh .
Ł
ks1 sinh Ž u y m k q Ž C X y 1 . h . 5 Ž 5.22a .

t
v 1V4 ŽC . Ž u . s sinh Ž u q Ž C q C X . h . sinh Ž u q Ž C q C X q 2 . h .
t
v V2 4 ŽC . Ž u . s sinh Ž u y Ž C y C X . h . sinh Ž u y Ž C y C X y 2 . h . ,
t t
v V3 4 ŽC . Ž u . s v4V4 ŽC . Ž u . s sinh Ž u y Ž C y C X . h . ,sinh Ž u q Ž C q C X q 2 . h . .
Ž 5.22b .
Compare this with the ‘‘lowest weight eigenvalue equation’’ Ž5.9., which actually
represents the same set of eigenvalues. Note that the C Ž n. s 0 limit of Ž5.22. does not
yield Ž5.17. with V Ž n. s V3at , which is a consequence of the differences between Ž5.19.
and Ž5.14.. The ‘‘highest weight Bethe ansatz’’ allows us to mix local V3at and V5at
modules in the global quantum space, while the ‘‘lowest weight Bethe ansatz allowed us
to construct the latter from an arbitrary mixture of V3at and V4t Ž C Ž n. . modules.
Nevertheless the fusion procedure for the ‘‘highest weight ansatz’’ with V4t Ž C X . as
local quantum space is completely analogous to the ‘‘lowest weight ansatz’’ and reads
at at
tˆ V 3 Ž N < u q h . P tˆ V 3 Ž N < u y h .
N
t at
s tˆ V4 Ž1. Ž N < u . q
ž /
Ł sinh Ž u y dŽ i . q Ž CŽ i . q 1. h . tˆ V Ž N < u . .
is1
at
5
From this the corresponding eigenvalue equation for tˆ V5 Ž N < u. with V Ž n. s V4t Ž C Ž n. .
follows as
at
LVN5 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j y h . m sinh Ž u q m k q 4h .
s v NŽ1. Ž u .
½ Ł Ł
js1 sinh Ž u y l j q 3h . ks1 sinh Ž u q m k q 2 h .
M sinh Ž u y l j q h . m sinh Ž u q m k q 4h .
qŁ Ł
js1 sinh Ž u y l j q 3h . ks1 sinh Ž u q m k q 2 h .
sinh Ž u q m k y 2h . m sinh Ž u q m k y 2h .
=
sinh Ž u q m k .
qŁ
ks1 sinh Ž u q m k q 2h . 5
M sinh Ž u y l j y h . m sinh Ž u q m k q 2h .
y v NŽ 4 . Ž u .
½ Ł Ł
js1 sinh Ž u y l j q 3h . ks1 sinh Ž u q m k .
M sinh Ž u y l j q h . m sinh Ž u q m k y 2h .
qŁ Ł
js1 sinh Ž u y l j q 3h . ks1 sinh Ž u q m k . 5 Ž 5.23a .

t t t
v 1V4 ŽC . Ž u . s v V2 4 ŽC . Ž u . s v V3 4 ŽC . Ž u . s sinh Ž u q Ž C q 3 . h . ,
t t
v4V4 ŽC . Ž u . s v 5V4 ŽC . Ž u . s sinh Ž u y Ž C y 3 . h . . Ž 5.23b .
X
$ $
t
Of course this can be done for ˆ t NV 8 ŽC . Ž N < u.
with V Ž n.
analogously. The s V4t Ž C Ž n. .
X X
w x
abstract results of Tsuboi 44,45 for Uq gl m,n , specialized to Uq gl Ž2,1.. allow us to
Ž Ž .. Ž
obtain the eigenvalues given above, without any Bethe ansatz calculation, if the notation
is adjusted to the one used here. On the other hand, the present method could be used
without difficulties to obtain eigenvalues corresponding to higher dimensional atypical
representations V2atnq1 with n g N, which is omitted here, because their physical
applications are not very realistic at all. Frahm has done this in the q s 1-limit w47x. The
calculations of Tsuboi and Frahm basically rely on Schulz’s result w46x, respectively
Lai’s and Sutherland’s coordinate Bethe ansatz calculations w48,49x on the fundamental
representations and the fusion procedure explained above. Instead of calculating missing
eigenvalues algebraically like in w11x they obtain the eigenvalues for V different from
V3at in Ž1.2. by analytical considerations in the spirit of Reshetikhin’s analytical Bethe
ansatz w50x. This is admittedly very efficient and surely correct in the simple cases
considered so far, but seems to fail in more complicated situations in particular for
V s V Ž n. s V8t Ž C . in Ž1.2. w47x, which will be solved in the following. The reason seems
to be that their method in contrast to Reshetikhin’s original work requires some intuitive
input, which is difficult to fix especially in the latter case. This also prevents a proof of
the resulting equations, which remained in the state of a conjecture.
For V s V3at ,V4t Ž C X . and V Ž n. s V t Ž C Ž n. . in Ž1.2. the local vacuum amplitudes
V 8t ŽC . Ž .
vi u corresponding to the highest weight vector u1 : from ŽŽA.9.. can be obtained
via
t at t
v iV 8 ŽC . Ž u . , v iV 3 Ž u y Ž C q 1 . h . P v iV4 ŽC . Ž u q h . , Ž 5.24 .
which is the analog of 5.13 . However, in order to demonstrate the generality of the
Ž .
present approach the lowest weight vector u 4 : from ŽA.9. will be taken as local
pseudo-vacuum here. Since this is not a direct product of lowest weight vectors, one has
to calculate the action of the direct product V l w : of these vectors for V s V3at and
V Ž n. s V8t Ž C Ž n. . in Ž1.2a. directly from Ž5.2c. with u1 s yŽ C q 1.h , u1 s yh by
means of ŽA.9., Ž4.12. with ŽC.1. and Ž4.14. with ŽC.3.. This yields
vŽN1 . Ž u . 0 0
V 3at
Tâ Ž 1, N < u . V l w :s
Bˆ1 Ž u .
Bˆ2 Ž u .
v NŽ2. Ž u .
Bˆ3 Ž u q h .
0
v NŽ3.
which is apparently different from the known cases Ž5.5., Ž5.14. and Ž5.19., where at
Ž u. 0 a
V l w :, Ž 5.25 .
least some v NŽ i. Ž u. coincide. One might look for a different pseudo-vacuum – not
necessarily a product of highest or lowest weight vectors, in which Ž5.25. takes one of
these forms. An inspection of all local basis vectors from ŽA.9. shows not only that such
a vector does not exist, but also that the appearance of three non-trivial off-diagonal
operators Bî Ž u. is generic.
The general eigenvectors are too complicated to be discussed here, but it is easy to
verify by a direct algebraical calculation that
l1 , . . . , l M :1 s Bˆ1 Ž l1 . . . . Bˆ1 Ž l M . V l w : , Ž 5.26a .
which is totally symmetric in the Bethe parameters l1 , . . . , lM and
m 1 , . . . , m m :2 s Bˆ3 Ž m 1 . . . . Bˆ3 Ž m m . V l w : , Ž 5.26b .
V at3 Ž
which is totally antisymmetric in m 1 , . . . , m m , are eigenvectors of tˆ N < u., provided
these parameters solve the Bethe equations
N sinh Ž la y d Ž i. y Ž C q 1 . h .
Ł sinh Ž l Ž i.
is1 ayd y Ž C y 1. h .
M sinh Ž la y l j y 2h . m sinh Ž la y m k y h .
s Ł Ł Ž 5.27a .
js1, j/ a sinh Ž la y l j q 2 h . ks1 sinh Ž la y m k q h .
for a s 1, . . . M and
N sinh Ž mb y d Ž i. q Ž C q 2 . h . M sinh Ž l j y mb q h .
Ł sinh Ž i.
sŁ for b s 1, . . . ,m .
is1 Ž mb y d y Ž C q 2. h . js1 sinh Ž l j y mb y h .
Ž 5.27b .
3
The corresponding eigenvalue equations for tˆ Vat Ž N < u. with V Ž n. s V8t Ž C Ž n. . read
at
LVN3 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j q 2h . M sinh Ž u y l j y 2h .
s v NŽ1. Ž u . Ł q v NŽ2. Ž u . Ł
js1 sinh Ž u y l j . js1 sinh Ž u y l j .
m sinh Ž u y m k q h . m sinh Ž u y m k q h .
=Ł y v NŽ3. Ž u . Ł Ž 5.28a .
ks1 sinh Ž u y m k y h . ks1 sinh Ž u y m k y h .
t
v 1V 8 ŽC . Ž u . s sinh Ž u q Ž C q 1 . h . sinh Ž u y Ž C q 1 . h . ,
t
v V2 8 ŽC . Ž u . s sinh Ž u q Ž C q 1 . h . sinh Ž u y Ž C y 1 . h . ,
t
v V3 8 ŽC . Ž u . s sinh Ž u y Ž C q 3 . h . sinh Ž u y Ž C y 1 . h . . Ž 5.28b .
Similarly one has to calculate the action of V l w : on the monodromy matrix Ž1.2a. for
V s V4t Ž C X . and V Ž n. s V8t Ž C Ž n. . from Ž5.2c. with u 0 s yŽ C q 1.h , u1 s yh , by means
of ŽA.9., Ž4.16. with ŽC.1. and Ž4.13. with ŽC.2., which yields
vŽN1 . Ž u . 0 0 0
at
TâV 3 Ž 1, N < u . V l w :s
)
)
)
v NŽ2.
0
0
Ž u. )
v NŽ3. Ž u .
)
)
0
v Ž4. Ž u .
0 a
V l w :,
Ž 5.29 .
where a star denotes some non-trivial off-diagonal entry. SoX the method from Ref. w11x
4
can be applied in order to calculate the eigenvalues of tˆ V t ŽC . Ž N < u. corresponding to the
simple eigenvectors Ž5.26. algebraically:
t X
LVN4 ŽC . Ž u < l1 , . . . , l M < m 1 , . . . , m m .
m sinh Ž u y m k q Ž C X q 1 . h .
s v NŽ1. Ž u. Ł
q v NŽ2. Ž u . Ł
M sinh Ž u y l j y Ž C X y 2 . h .
y v NŽ3. Ž u . Ł
js1 sinh Ž u y l j y C Xh .
=Ł
M sinh Ž u y l j y Ž C X q 2 . h .
y v NŽ4. Ž u . Ł
js1 sinh Ž u y l j y C Xh .
=Ł Ž 5.30a .

t
v 1V 8 ŽC . Ž u . s sinh Ž u q Ž C y C X q 1 . h . sinh Ž u q Ž C y C X q 3 . h . ,
t
v V2 8 ŽC . Ž u . s sinh Ž u y Ž C q C X q 3 . h . sinh Ž u y Ž C q C X y 1 . h . ,
t
v V3 8 ŽC . Ž u . s sinh Ž u q Ž C y C X q 1 . h . sinh Ž u y Ž C q C X q 1 . h . ,
t
v4V 8 ŽC . Ž u . s sinh Ž u q Ž C y C X q 1 . h . sinh Ž u y Ž C q C X y 1 . h . . Ž 5.30b .
It is straightforward to obtain the fusion relation:
at at
tˆ V 3 Ž N < u y h . tˆ V 3 Ž N < u q h .
N
t
s
žŁ Ž Ž
is1
N
/
sinh u y CŽ i . q 2 . h y dŽ i . . sinh Ž u y Ž CŽ i . y 2 . h y dŽ i . . tˆ V4 Ž1. Ž N < u .
at
q
žŁ Ž Ž is1
5
/
sinh u q C Ž i . q 2 . h y d Ž i . . sinh Ž u y C Ž i . h y d Ž i . . tˆ V5 Ž N < u . .
The corresponding eigenvalue equation for tˆ Vat Ž N < u. with V Ž n. s V8t Ž C Ž n. . can be
written as
at
LVN5 Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j q 3h .
s v NŽ1. Ž u . Ł
js1 sinh Ž u y l j y h .
M sinh Ž u y l j q 3h . sinh Ž u y l j y 3h .
q v NŽ2. Ž u . Ł
js1 sinh Ž u y l j q h . sinh Ž u y l j y h .
m sinh Ž u y m k . M sinh Ž u y l j y 3h .
=Ł q v NŽ3. Ž u . Ł
ks1 sinh Ž u y m k y 2h . js1 sinh Ž u y l j q h .
m sinh Ž u y m k q 2h . M sinh Ž u y l j q 3h .
=Ł y v NŽ4. Ž u . Ł
m sinh Ž u y m k . M sinh Ž u y l j y h .
m sinh Ž u y m k q 2h .
=Ł Ž 5.31a .
ks1 sinh Ž u y m k y 2h .
t
v 1V 8 ŽC . Ž u . s sinh Ž u q Ch . sinh Ž u y Ž C q 2 . h . ,
t
v V2 8 ŽC . Ž u . s sinh Ž u q Ch . sinh Ž u y Ch . ,
t
v V3 8 ŽC . Ž u . s sinh Ž u q Ch . sinh Ž u y Ž C y 2 . h . ,
t
v4V 8 ŽC . Ž u . s sinh Ž u y Ch . sinh Ž u y Ž C q 4 . h . ,
t
v 5V 8 ŽC . Ž u . s sinh Ž u y Ž C y 2 . h . sinh Ž u y Ž C q 4 . h . . Ž 5.31b .
Finally the following fusion relation holds

at t X
tˆ V 3 Ž N < u q Ž C X q 1 . h . tˆ V4 ŽC . Ž N < u y h .
N X
t
s
ž Ł sinh Ž u y dŽ i . y Ž C y C q 2. h . tˆ V ŽC . Ž N < u .
is1
X
/ 8
N
y
ž Ł sinh Ž u y dŽ i . y Ž C y C y 2. h . sinh Ž u y d Ž i. y Ž C q C q 2. h .
is1
X
X X
/
t
=tˆ V4 ŽC q1 . Ž N < u . .
t X
The corresponding eigenvalues of tˆ V 8 ŽC . Ž N < u. follow as
t X
LVN8 ŽC . Ž u < l1 , . . . , l M < m 1 , . . . , m m .
M sinh Ž u y l j q Ž C X q 3 . h . m sinh Ž u y m k q C Xh .
s v NŽ1. Ž u . Ł Ł
js1 sinh Ž u y l j q Ž C X q 1 . h . ks1 sinh Ž u y m k y C Xh .
M sinh Ž u y l j q Ž C X y 1 . h . m sinh Ž u y m k q Ž C X q 2 . h .
js1 sinh Ž u y l j q Ž C X q 1 . h . ks1 sinh Ž u y m k y C Xh .
M sinh Ž u y l j q Ž C X q 3 . h . m sinh Ž u y m k q C Xh .
q v NŽ3. Ž u. Ł Ł
js1 sinh Ž u y l j q Ž C X q 1 . h . ks1 sinh Ž u y m k y Ž C X q 2 . h .
M sinh Ž u y l j q Ž C X y 1 . h . m sinh Ž u y m k q Ž C X q 2 . h .
js1 sinh Ž u y l j q Ž C X q 1 . h . ks1 sinh Ž u y m k y Ž C X q 2 . h .
M sinh Ž u y l j q Ž C X q 3 . h . sinh Ž u y l j y Ž C X y 1 . h .
y v NŽ5. Ž u. Ł
js1 sinh Ž u y l j q Ž C X q 1 . h . sinh Ž u y l j y Ž C X q 1 . h .
m sinh Ž u y m k q C Xh . M sinh Ž u y l j q Ž C X q 3 . h .
ks1 sinh Ž u y m k y C Xh . js1 sinh Ž u y l j q Ž C X q 1 . h .
sinh Ž u y l j y Ž C X q 3 . h . m sinh Ž u y m k q C Xh .
= Ł
sinh Ž u y l j y Ž C X q 1 . h . ks1 sinh Ž u y m k y Ž C X q 2 . h .
M sinh Ž u y l j q Ž C X y 1 . h . sinh Ž u y l j y Ž C X y 1 . h .
y v NŽ7. Ž u . Ł
js1 sinh Ž u y l j q Ž C X q 1 . h . sinh Ž u y l j y Ž C X q 1 . h .
m sinh Ž u y m k q Ž C X q 2 . h . M sinh Ž u y l j q Ž C X y 1 . h .
=Ł y v NŽ8. Ž u. Ł
ks1 sinh Ž u y m k y C Xh . js1 sinh Ž u y l j q Ž C X q 1 . h .
sinh Ž u y l j y Ž C X q 3 . h . m sinh Ž u y m k q Ž C X q 2 . h .
= Ł Ž 5.32a .
sinh Ž u y l j y Ž C X q 1 . h . ks1 sinh Ž u y m k y Ž C X q 2 . h .

t
v 1V 8 ŽC . Ž u .
s sinh Ž u q Ž C y C X q 2 . h . sinh Ž u q Ž C y C X . h . sinh Ž u y Ž C y C X . h . ,

t
v V2 8 ŽC . Ž u .
s sinh Ž u q Ž C y C X q 2 . h . sinh Ž u q Ž C y C X . h . sinh Ž u y Ž C y C X q 2 . h . ,

t
v V3 8 ŽC . Ž u .
s sinh Ž u y Ž C q C X q 4 . h . sinh Ž u y Ž C q C X . h . sinh Ž u y Ž C y C X . h . ,

t
v4V 8 ŽC . Ž u .
s sinh Ž u y Ž C q C X q 2 . h . sinh Ž u y Ž C q C X . h . sinh Ž u y Ž C y C X q 2 . h . ,

t
v 5V 8 ŽC . Ž u .
s sinh Ž u q Ž C y C X . h . sinh Ž u y Ž C y C X . h . sinh Ž u y Ž C q C X q 2 . h . ,

t
v6V 8 ŽC . Ž u .
s sinh Ž u q Ž C y C X . h . sinh Ž u y Ž C y C X . h . sinh Ž u y Ž C q C X . h . ,

t
v V7 8 ŽC . Ž u .
s sinh Ž u q Ž C y C X . h . sinh Ž u y Ž C y C X q 2 . h . sinh Ž u y Ž C q C X q 2 . h . ,

t
v V8 8 ŽC . Ž u .
s sinh Ž u q Ž C y C X . h . sinh Ž u y Ž C q C X . h . sinh Ž u y Ž C y C X q 2 . h . .

Ž 5.32b .
The completely analogous calculations for the highest weight Bethe ansatz starting
from Ž5.24. will be left for the reader.
In principle all results concerning non-fundamental representations in auxiliary space
in this section could be achieved by directly applying the method from Ref. w11x, without
reference to the fusion procedure. The latter however reduces the effort considerably, so
that it would be unwise to disregard this tool. On the other hand, rigorous results from
fusion require a complete input, which in general can be provided only by calculations
similar to the one in w11x. The R-matrices derived in Section 4 provide the elementary
building blocks of the fusion hierarchy that have to be diagonalized by direct methods in
order to obtain the consecutive results almost trivially by fusion. This justifies to call
them fundamental.
¨ in
Finally it should be noted that completeness of all extremal weight Bethe ansatze
the sense that missing states can be produced from them by applying Uq Ž gl Ž2,1;C..
generators, and equivalence of highest and lowest weight eigenvalue equations can be
proved by well known standard arguments not to be repeated here.
6. Conclusions
$ X
Uq Ž gl Ž2,1;C.. symmetry has been explained in Section 2; some low-dimensional
representations, relevant for physical applications, have been given in Section 3; the
fundamental R-matrices have been constructed in Section 4. Eigenvalue equations of the
corresponding transfer matrices have been obtained in Section 5 by combining Bethe
ansatz calculations with the fusion procedure. This procedure can be iterated to higher
dimensional representations along the lines presented here. Partially this can be read off
from the results of Tsuboi, who worked with a purely analytical ansatz w44,45x. Several
conclusions may be drawn: Baxter noted w6x that all coefficients of R-matrices,
calculated in Sections 4 and 5 generically, are meromorphic functions in q, u and here
also in C. Therefore they can be analytically continued to almost all complex q-, u- and
C-values. All singularities can be removed by multiplying the R-matrices with the
vanishing denominators, as was done in Section 5. The remainingX zeroes indicate that
generically irreducible representations become reducible, i.e. Rˇ Vd VdX Ž u. Ž4.4b. is not an
intertwiner at these points, see e.g. Ref. w17x. The restriction to real q and C values in
Section 3 is due to the fact that an Euclidean metric has been chosen in Ž2.1a.. A simple
‘‘deformation’’ of the latter, e.g. absorbing some factors in the definition of vectors and
1-forms, leads to different expressions for the weight vector bases in Appendix A, but to
the same intertwiners listed in Appendix B. Therefore the product decompositions in
Section 3, hold for all generic q and C values, if a convenient metric has been chosen
for Vd and VdXX . The q-deformed Clebsch–Gordan coefficients
X
appearing implicitly in
Appendix A are artificial. The intertwiners RˇVd VdX Ž u. in the form Ž4.6. provide a
complete, metric-free characterization of G Vxd m G VydXX Ž3.13.. In addition ratios of Drin-
fel’d polynomials, which characterize the structure of the tensor products involved in the
$
construction of Ž1.2. appear on the left-hand side at least once in any set of Bethe
X
equations. Tsuboi formulated his results on Uq Ž gl Ž m,n;C.. in terms of roots and Dynkin
diagrams, not used here, which is a further step in establishing this viewpoint w44,45x.
However, many details are not understood at present and some work by mathematicians
would be appreciated.
The following physical aspect is easier to understand. Define
X X X X X X
R˜V4 V4 Ž C,C X . :sR V4 V4 Ž u . us0 s P
ˇ V4 V4 ( P1V4V4 q P2V4V4 q P3V4V4 . Ž 6.1a .
This is a purely UqŽglŽ2,1;C..-symmetric R-matrix. Ž4.17c. with u s v s 0 can be
written as
X XX X XX
R˜12
V4 V4
Ž C,C X . R˜13
V4 V4
Ž C,C XX . R˜V234 V4 Ž C X ,C XX .
X XX XX X
s R˜V234 V4 Ž C X ,C XX . R˜13
V4 V4
Ž C,C XX . R˜12
V4 V4
Ž C,C X . , Ž 6.1b .
which is the general form Ž1.1a. of the YBE. Note that C s C does imply V4 ( V4X only
X
up to a ‘‘deformation of the metric’’. This and not the strict identification is however
already sufficient to establish the Hamiltonian limit. Choosing all parameters in V4 and
t X
V4X with C :s C Ž n. s C X real, the logarithmic derivative of tˆ V4 ŽC . Ž N < u. with respect to
u at u s 0 yields the Hamiltonian of the 2-parameter model with periodic boundary
conditions, which was established previously by Bariev et al. w10x, only for U 0 y1
ŽHubbard parameter. w9x, while the eigenvalues from Ž5.9. or its highest weight
counterpart are continuous at this point. However, staggering the V Ž n. ( V4t Ž C . conve-
niently allows to continue the Hamiltonian to U ( y1, as is expected for physical
reasons. In addition phase factors can be assigned to particle operators this way, which
are known not to change the eigenvalues. Note that this Hamiltonian is not Hermitian in
general and may find applications in non-equilibrium physics w51x.
Bariev et al. found that the q-deformed Bariev model, whose Hamiltonian violates
parity invariance admits long-range pair correlations, which are regarded as a one-di-
mensional analog of superconductivity w10x. Using X
C X as a spectral parameter, as
V 4t ŽC . Ž
indicated above, the logarithmic derivative of tˆ N <0. with respect to C X at C X s C
turns out to violate parity too and can be regarded as a conserved 2-particle current w51x.
It is plausible that long range pair correlations are due to the existence of this conserved
quantity, which trivially commutes with the Hamiltonian.
at
For V s V Ž n. s V5at in Ž1.2. the logarithmic derivative of tˆ V5 Ž N < u. with respect to u
at u s 0 yields a generalization of the spin-1 generalization of the tJ model’’, treated by
¨ and Tsvelick w52x. Its eigenvalues can be obtained from Ž5.18.. The
Frahm, Pfannmuller
quantization and diagonalization of higher dimensional atypical representations V2atnq1
with n g N is a simple generalization of the V3at and V5at cases. Tsuboi has shown, how
to obtain the eigenvalues w44,45x. It seams however difficult to find a realistic physical
application for these. X
t
On the other hand, the conventional Hamiltonian limit of tˆ V 8 ŽC . Ž N < u., diagonalized
in Ž5.32., seems very promising. Obviously the lowest weight Bethe ansatz Ž5.25. does
not allow to incorporate lower dimensional local quantum spaces. The same is true for
the highest weight ansatz only mentioned in passing. Preliminary results are available
and will be published elsewhere w53x. The existence of a second Hamiltonian limit
analogous to Ž6.1. is obvious. Its physical meaning is still unclear at the moment.
In general the solution of the underlying statistical model provides not only the
integrals of motion for the corresponding Hamiltonian, but is also the starting point for
the calculation of thermodynamical properties in the exact approach via the Trotter-
Suzuki mapping pioneered by Klumper ¨ w54,55x, which is now well developed in order to
treat graded models, related to limiting cases of the ones treated here, by Juttner, ¨
¨
Klumper and Suzuki w56,57x, culminating in Ref. w58x. The author hopes to come back to
some of these interesting problems in the near future.
Acknowledgements
This work has been performed within the research program of the Sonderforschungs-
¨ ¨ .. The author thanks J. Zittartz and A. Klumper
bereich 341 ŽKoln-Aachen-Julich ¨ for
¨
continuous support, G. Juttner, Y. Kato, K. Klauck, A. ¨
Klumper, J. Suzuki, A. Zvyagin,
and especially A. Fujii for stimulating discussions and encouragement. Special thanks
¨
goes to A. Klumper for carefully reading the manuscript and useful suggestions,
incorporated in the final version.
Appendix A. Irreducible modules
The weights for the modules Ž Vd , G Vd . are defined in Section 3, Eq. Ž3.4.. All weight
vectors are chosen to be orthonormal. To achieve this, q is assumed to be real and
positive, while k , k ) and m , m ) are taken to be complex conjugate to each other.
For the V3-module Ž3.5. the basis vectors Ž2.1a. are weight vectors Ž3.4a.. The
corresponding weights are
LV 3 s l1 s ya 2 ,
l2 s LV 3 y a 1 ,
l3 s LV 3 y a 1 y a 2 . Ž A.1 .
G V4 Ž3.7. is irreducible, iff C / 0,y 1, which is assumed from here on.
For the V4-module again the basis vectors Ž2.1a. are weight vectors Ž3.4a. and the
corresponding weights are
LV4 s l1 s yC a 1 y 2C a 2 ,
l2 s LV4 ya 1 y 2 a 2 ,
l3 s LV4 ya 2 ,
l4 s LV4 ya 1 y a 2 . Ž A.2 .
Ž V4 , G V . can be characterized by fixing C Ž3.7a..
4
The elements G V4 of the weight vector basis of
Ž V4 , G V . ( V4t Ž C q 1.
4
are denoted by Õi :, i : Ž2.1a. and expressed in the product basis Ž2.3. of V3 m V4 as

Cq1 1y C
1 y
Õ 1 :s
(w C q 2x q
½ q 2 1 :m 4 :y q 2
5
2 :m 3 :y k q 3 :m 1 : ,
Õ 2 :s 3 :m 2 : ,
Cq1 1
1 y
Õ 3 :s
(w C q 2x q
½ q 2 2
1 :m 2 :y m q 3 :m 3 : , 5
Cq1 1
1 y
Õ4 :s
(w C q 2x q
½ q 2 2 :m 2 :y m q 2 3 :m 4 : . 5 Ž A.3 .
The elements of the weight vector basis of
ž V˜
4
Ž1.
, G V˜4Ž 1 . ( V4t Ž C q C X .
/
are u˜ i :, i :. In the product basis of V4 m V4X , they read
u˜ 1 :s 1 :m 1 : ,
X
u˜ 2 :s w C x q w C q 1 x q qy2 C q w C X x q w C X q 1 x q q 2 C
1
X y
q w 2 x q w C q 1 x q w C X q 1 x q q C yC 4 2
½
= k X )mX ) q C 1 :m 2 :q k )m ) qyC 2 :m 1 :
C yC
X
1 1
y
yk k q ) X) 2
q 2 3 :m 4 :y q 2 4 :m 3 :
5 ,
X
C C
1 y 2
u˜ 3 :s
(w C x qq
yC
q w CX x q q C
X ½ kq 2 1 :m 4 :q k q X)
5
4 :m 1 : ,
X
C C
1 y 2
u˜ 4 :s
(w C x qq
yC
q wC xq q X C
X ½ kq 2 1 :m 3 :q k X ) q 5
3 :m 1 : . Ž A.4 .
Similarly the elements of the weight vector basis to
ž V˜
4
Ž 2.
, G V˜4Ž 2 . ( V4t Ž C q C X q 1 .
/
are w˜ i :, i : or explicitly
X
w˜ 1 :s w C x q w C q 1 x q qy2 Ž Cq1 . q w C X x q w C X q 1 x q q 2 Ž C q1 .
1
X y
q w 2 x q w C q 1 x q w C X q 1 x q q C yC 4 2
½
= km qy Ž Cq1 . 1 :m 2 :q km
X
X X C q1
q 2 :m 1 :
C yC 1
1
qmm q X 2
qy 2 3 :m 4 :y q 2 4 :m 3 :
5 ,
1
w˜ 2 :s 2 :m 2 : w˜ 3 :s X
(w C q 1x qq
yC y1
q w C X q 1 x q q C q1
X
Cq1 C q1
y
= mq½ 2 3 :m 2 :q m q X 2
5
2 :m 3 : w̃4 :
1
s X
(w C q 1x qq
yC y1
q w C X q 1x q q C q1
X
Cq1 C q1
y
½
= mq 2 4 :m 2 :q m q X 2
2 :m 4 : . 5 Ž A.5 .
The weight vector basis of
ž Vˇ , G ˇ / ( V Ž 1.
4 V4 4
t
are Õˇi :, i :, or expanded in the product basis from V3 m V3 ,

1 1
1 y
Õˇ 1 :s
(w2x q
½ q 2 1 :m 2 :y q 2 2 :m 1 : , 5
Õˇ 2 :s 3 :m 3 :,
1 1
1 y
Õˇ 3 :s
(w2x q
½ q 2 2
1 :m 3 :y q 3 :m 1 : , 5
1 1
1 y
Õˇ4 :s
(w2x q
½ q 2 2 :m 3 :y q 2 3 :m 2 : . 5 Ž A.6 .
The Ž3,2.-graded carrier space Vˇ5 of the atypical representation G Vˇ5 is spanned by
weight vectors, expanded in the product basis Ž2.3. of V3 m V3
uˇ 1 :s 1 :m 1 :,
1 1
1 y
uˇ 2 :s
(w2x q
½ 2
q 1 :m 2 :q q 2
5
2 :m 1 : ,
uˇ 3 :s 2 :m 2 :,
1 1
1 y
uˇ 4 :s
(w2x q
½ 2
q 1 :m 3 :q q 2
5
3 :m 1 : ,
1 1
1 y
uˇ 5 :s
(w2x q
½ q 2 2 :m 3 :q q 2
5
3 :m 2 : . Ž A.7 .
The corresponding weights are

LV5 s l1 s y2 a 2 ,
l2 s LV5 y a 1 ,
l3 s LV5 y 2 a 1 ,
l4 s LV5 y a 1 y a 2 ,
l5 s L V 5 y 2 a 1 y a 2 . Ž A.8 .
Similarly the Ž4,4.-graded carrier space V8 is spanned by the following weight vectors,
expanded in the product basis of V3 m V4 :
u1 :s 1 :m 1 :,
u 2 :s 2 :m 1 :,
1
1 Cq1
2
u 3 :s
(w C q 2x q
½ y
q 2 3 :m 3 :q m q 1 :m 2 : ,
)
5
1
1 Cq1
2
u 4 :s
(w C q 2x q
½ y
q 2 3 :m 4 :q m q 2 :m 2 : , )
5
u 5 :s 1 :m 3 :,
1
u 6 :s
(w2x q w C q 2x q
1
C 1
½
= w2x q q y
2 3 :m 1 :q k q q ) y
2
2
1 :m 4 :y q 2 :m 3 :
5 ,
1
1 1
u 7 :s
(w2x q
½ q 2 1 :m 4 :q qy 2 2 :m 3 : , 5
u 8 :s 2 :m 4 :. Ž A.9 .
The corresponding weights are
LV 8 s l1 s yC a 1 y Ž 2C q 1 . a 2 ,
l2 s LV 8 y a 1 ,
l3 s LV 8 y a 1 y 2 a 2 ,
l4 s LV 8 y 2 a 1 y 2 a 2 ,
l5 s L V 8 y a 2 ,
l6 s l 7 s L V 8 y a 1 y a 2 ,
l8 s L V 8 y 2 a 1 y a 2 . Ž A.10 .
Obviously Ž V8 , G V 8 . can also be labeled by C Ž3.10a.. Finally the weight vector basis of
ž V˜ , G ˜ / ( V Ž C q C . ,
8 V8 8
t X
expanded in the product basis V4 m V4X , is given by

X
C C
1 y 2
Õ˜ 1 :s
(w C x qq
yC
q wC xq q X C
X
½ kq 2 1 :m 3 :y k X ) q 5
3 :m 1 : ,
X
C C
1 y 2
Õ˜ 2 :s
(w C x qq
yC
q w CX x q q C
X
½ kq 2 1 :m 4 :y k X ) q 5
4 :m 1 : ,
1
Õ˜ 3 :s X
(w C q 1x qq
yC y1
X
Cq 1 C q1
y
½
= m q ) 2
2 :m 3 :y m q X 2
5
3 :m 2 : ,
1
Õ˜4 :s X
(w C q 1x qq
yC y1
X
Cq 1 C q1
y
= m) q ½ 2
2 :m 4 :y mX q 2
4 :m 2 : , 5
Õ˜5 :s 3 :m 3 :,
X X 2
Õ˜6 :s w 2 x q q Cy C Ž w C x q w C q 1 x q qy2 Cy1 y w C X x q w C X q 1 x q q 2 C q1 .
½
X X 2
q Ž w C x q qyC q w C q 1 x q qyCy1 q w C X x q q C q w C X q 1 x q q C q1 .
1
y
= Ž w C q 1x q w C X x q q w C x q w C X q 1x q . 5 2
X
Cy C
X
=
½ Ž w C x q w C q 1x q q
1
y2 Cy1
1
X
y w C x q w C q 1x q q X 2 C q1
.q 2
y
= q 2 3 :m 4 :y q 2 4 :m 3 :
X X
q Ž w C x q qyC q w C q 1 x q qyC y1 q w C X x q q C q w C X q 1 x q q C q1 .
= m )k X 2 :m 1 :y kmX ) 1 :m 2 :
1 1
5 ,
1 y
Õ˜ 7 :s
w2x q ½ 2
q 3 :m 4 :q q 2 4 :m 3 : , 5
Õ˜ 8 :s 4 :m 4 :. Ž A.11 .
Appendix B. Elementary intertwiners

X
The elementary intertwiners P Vd VdX are obtained from the results in Appendix A by
Ž4.7. as explained in Section 4. They fulfill the identities Ž4.8. per construction.
For V4 ; V3 m V4 ŽA.3. the intertwiner P2V 3V4 follows from Ž4.7. with the minus sign.
1
P2V 3V4 s w e me qe me qe me qe me x
w C q 2 x q 21 12 22 22 32 23 41 14
wC xq w C q 1x q
y e13 m e 31 y e 23 m e32 y w e me qe me x
w C q 2x q w C q 2 x q 33 33 43 34
Cq 2
1 y1
k y
y q e 31 m e 24 q qe42 m e13 y q 2
w C q 2x q w C q 2x q
1 1
y 2
= q 2e
11 m e 34 y q e12 m e 33
Cq2 1 1
k) 2 2
y
2
y q q e 34 m e11 y q e 33 m e 21
w C q 2x q
Cq 2
m) 2
q q e 23 m Ž e13 q e 24 .
w C q 2x q
Cq 2
m y
q 2
q Ž e31 q e42 . m e32 . Ž B.1 .
w C q 2x q
For V˜4Ž1. ; V4 m V4X ŽA.4. the same now with a positive sign in Ž4.7. procedure leads to
X X
X k )m )k m X
P1V4 V4 s e11 m e11 q q CqC e11 m e 22
w C q C X x q w C q C X q 1x q
kmk X )mX ) X
q X X
qyŽ CqC . e 22 m e11
w C q C x q w C q C q 1x q
X
Cq C
k )k X 2
q q e11 m Ž e 33 q e 44 .
w C q CX x q
Cq C X
kk X ) y
q 2
q Ž e33 q e44 . m e11
w C q CX x q
w C x q w CX x q
q
Ž qy1 e33 m e44 q qe44 m e33 .
1
q
= w C x q w C q 1 x q e12 m e 21 q w C X x q w C X q 1 x q e 21 m e12 4
w C x q w CX x q
y Ž e me qe me .
w C q C X x q w C q C X q 1 x q 34 43 43 34
1 X
q w C x q Ž e31 m e13 q e41 m e14 .
w C q CX x q
1
y w C x q Ž e13 m e 31 q e14 m e 41 . 4 q
w C q C x q w C q C X q 1x q
X
° CqC
X
1 1
=~q
y 2
¢
2 X
m k wC xq q
)
ž 2e
13 m e 24 y q e14 m e23 /
1 1
y 2
yk m w C X x q q
) X
ž 2e
31 m e 42 y q e 41 m e 32 /
CqC X 1
y 1
2
qq 2 X)
km w C x q q X
ž y
2 e 23 m e14 y q e 24 m e13 /
1
1
2
¶•
X)
ymk w C x q q ž y
2 e 32 m e 41 y q e 42 m e 31 /ß Ž B.2 .
and similarly for V˜4Ž2. ; V4 m V4X ŽA.5.:

X X
X k )m )k m X
P3V4 V4 s X X
qyŽCqC q2. e11 m e 22
w C q C q 1x q w C q C q 2x q
kmk X )mX ) X
q X X
q Cq C q2 e 22 m e11
w C q C q 1x q w C q C q 2x q
X
Cq C q2
mmX ) 2
q q e 22 m Ž e 33 q e 44 . q e 22 m e 22
w C q C X q 2x q
Cq C Xq2
m )mX y
q 2
q Ž e33 q e44 . m e22
w C q C X q 2x q
w C q 1x q w C X q 1x q
q
w C q C X q 1x q w C q C X q 2x q
Ž qy1 e33 m e44 q qe44 m e33 .
1
q
w C q C q 1x q w C q C X q 2x q
X
= w C X x q w C X q 1 x q e12 m e 21 q w C x q w C q 1 x q e 21 m e12 4
w C q 1x q w C X q 1x q
y Ž e me qe me .
w C q C X q 1 x q w C q C X q 2 x q 34 43 43 34
1 X
y w C q 1x q Ž e32 m e23 q e42 m e24 .
w C q C X q 2x q
1
y w C q 1 x q Ž e 23 m e 32 q e 24 m e 42 . 4 q
w C q C q 1x q w C q C X q 2x q
X
° CqC Xq2 1 1
=~q
y y
2
¢ 2 X
m k w C q 1 x q q e14 m e 23 y q
) X
ž 2
e13 m e 24 /
1 1
yk m w C q 1 x q q e 41 m e 32 y q 2 e31 m e42
) X
ž 2
/
X
CqC q2 1 1
y
2 2
qq X)
km w C q 1 x q q e24 m e13 y q ž 2
e 23 m e14 /
1
y
1
2
¶•
ž
ymk X ) w C X q 1 x q q 2 e42 m e 31 y q e 32 m e 41 / ß. Ž B.3 .
For Vˇ4 ; V3 m V3 ŽA.6. the result is now again with a minus sign in Ž4.7.
1
P2V 3V 3 s e 33 m e 33 y w e12 m e21 y e13 m e31 q e21 m e12 y e23 m e32
w2x q
q
qe31 m e13 q e 32 m e 23 x q e 22 m e11 q e33 m Ž e11 q e22 .
w2x q
qy1
q e11 m Ž e 22 q e 33 . q e 22 m e 33 . Ž B.4 .
w2x q
The intertwiner P1V 3V 3 for Vˇ5 ; V3 m V3 ŽA.7. is simply related to P2V 3V 3 ŽB.4. by Ž4.8c.,
i.e.
P1V 3V 3 s IV 3mV 3 y P2V 3V 3 . Ž B.5 .
In addition, the limit q ™ 1 is particularly simple, namely
P1V 3V 3 1
qs 1 s 2 žI V 3mV 3 q P
ˇ V 3V 3 / ,
P2V 3V 3 1
qs 1 s 2 žI V 3mV 3 y P
ˇ V 3V 3 / . Ž B.6 .
which was known to Kulish and Sklyanin long ago w27x. The intertwiner for V8 ; V3 m V4
ŽA.9. is obtained from Ž4.7. with a positive sign:
P1V 3V4 s e11 m e11 q e12 m e 21 q e 31 m e13 q e 42 m e24

1
y e 33 m e 33 q e 43 m e34 q w 2 x q e13 m e31
w C q 2x q
w C q 1x q
q w e me qe me qe me x
w C q 2 x q 21 12 22 22 32 23
1
y q Cq 1 e 31 m e 24 q qyCy1 e 42 m e13
w C q 2x q
Cq2 1 1
k 2
y 2
q q 2e
q 11 m e 34 y q e12 m e 33
w C q 2x q
Cq2 1 1
k) y y
2
q q 2 q 2 e 34 m e11 y q e 33 m e 21
w C q 2x q
Cq 2
m) 2
y q e 23 m Ž e13 q e 24 .
w C q 2x q
Cq 2
m y
q 2
y Ž e31 q e42 . m e32 . Ž B.7 .
w C q 2x q
X
Finally P2V4 V4 corresponding to V˜8 ; V4 m V4X ŽA.11. is given by
X
P2V4 V4 s e 33 m e33 q e 44 m e 44
w2x q X X
q k )m )km e11 m e 22 q kmk X )mX ) e 22 m e11
ag
Cq C X
k )k X CqC
X kk X ) y
q e11 Ž e33 q e 44 . q q 2
q Ž e33 q e44 . m e11
a a
Cq C Xq2
mmX ) y
q q 2 e 22 m Ž e 33 q e 44 .
g
X
Cq C q2
m )mX 2
q q Ž e33 q e44 . m e22
g
q d 2 qy1 qy1 d 2q
q
ž w2x q
y
w 2 x q ag / e 33 m e 44 q
ž w2x q
y
w 2 x q ag / e 44 m e 33
w2x q X X
y w C q 1x q w C x q e12 m e21 q w C x q w C q 1x q e21 m e12 4
ag
1 X
q w C x q Ž e13 m e31 q e14 m e41 .
a
y w C x q Ž e 31 m e13 q e 41 m e14 . 4
1 X
q w C q 1x q Ž e23 m e32 q e24 m e42 .
g
y w C q 1 x q Ž e 32 m e 23 q e 42 m e 24 . 4
1 d2
q
ž w2x q
q
w 2 x q ag / w e34 m e43 q e43 m e34 x
1
1
1
y
ag ½ m k´ q
) X
y
2e
13 m e 24 y q
1
2
e14 m e 23
1
y
X) 2 2
ymk h q e 42 m e 31 y q e 32 m e 41
1 1
y 2
X) 2e
ykm ´ q 23 m e14 q q e 24 m e13
1 1
y
X
qk mh q e 41 m e 32 y q
) 2 2
e 31 m e 42
5 , Ž B.8 .
where the following abbreviations have been used:

a s w C q CX x q ,
b s w C q C X q 1x q ,
g s w C q C X q 2x q ,
2 2 2 2 2
d 2 s w C q 1 x q q w C X q 1 x q q w C x q q w C X x q ,y w C q C X q 1 x q y 1,
X
Cq C q2 Cq C Xq2
y
X 2 2
´s wC xq q y wC xq q ,
X
Cq C q2 Cq C Xq2
y
h s w C x q q 2 y w CX x q q 2 . Ž B.9 .
X
For V4 s V4 3.7 , which can be obtained as a continuous limit, the identity 4.8c can be
Ž . Ž .
checked immediately, i.e.
P1V4 V4 q P2V4 V4 q P3V4 V4 s IV4mV4 .
This was used by Gould et al. w9x to calculate P2V4 V4 from their knowledge of P1V4 V4 and
P3V4 V4 . In the general case there is – to the author’s knowledge – no way to omit the
tedious direct calculation of ŽB.8.. X X
The projector properties Ž4.8a. of PiV4 V4 ( PiV4 V4 can be checked immediately.
Note that also q was assumed to be real and positive in Appendix A, it can be
analytically continued to arbitrary complex values here, except
X
for those isolated values,
at which denominators vanishes or the rank of one PiVd VdX is reduced.
Appendix C. Coefficients of R-matrices
The elements of the R-matrix Ž4.12. are explicitly given by r Ž u. from Ž4.9b. and
1
a 0 Ž u . :s w C q 1x q r Ž u . y 14 ,
w C q 2x q
1
b 0 Ž u . :s w2x q r Ž u . y w C x q 4 ,
w C q 2x q
1
g 0 Ž u . :s r Ž u . y w C q 1x q 4 ,
w C q 2x q
1
d 1 Ž u . :s r Ž u . qyC y1 q q 4 ,
w C q 2x q
1
d 2 Ž u . :s r Ž u . q Cq 1 q qy1 4 ,
w C q 2x q
C C
m) y y1 q1
x ½ 5
´ 1 Ž u . :s rŽ . u q 2 qq 2 ,
wCq2 q
C C
m q1 y y1
x ½ 5
2 2
´ 2 Ž u . :s rŽ . u q qq ,
wCq2 q
Cq1 Cq 1
k) y
z 1 Ž u . :s
w C q 2x q ½ rŽ . u q 2 qq 2
5 ,
Cq 1 Cq1
k y
x ½ 5
2 2
z 2 Ž u . :s rŽ . u q qq . Ž C.1 .
wCq2 q
X
The coefficients of R V4 V4 Ž u. Ž4.12. are explicitly given by f Ž u., g Ž u. Ž4.10b. and
X X
k )m )k m X X
r1 s
abg
Ž g q Cq C f Ž u . q w2x q b q a qyCyC y2 g Ž u . . ,
kmk X )mX ) X X
r 1X s
abg
Ž g qyC yC f Ž u . q w2x q b q a q CqC q2 g Ž u . . ,
X
Cq C Cq C X
k )k X 2
y
r2 s ž
a
q f Ž u. q q 2
X
/ ,
X) Cq C X Cq C
kk y
2
r 2X s
a ž /
q 2 f Ž u. q q ,
X
X) CqC Xq2 Cq C q2
mm y
2
g ž /
r3 s q 2 qq g Ž u. ,
X
X Cq C q2 Cq C Xq2
X
mm )
2
y
g ž
r3 s q qq 2 g u Ž ./ ,
qy1 X
r4 s 1 q
abg
w C x q Ž w C x qg f Ž u . y w C q 1 x q b .
q w C X q 1x q Ž w C q 1x q a g Ž u . y w C X x q b . 4 ,
q
r4X s 1 q w C x q Ž w C X x qg f Ž u . y w C q 1 x q b .
abg
q w C X q 1x q Ž w C q 1x q a g Ž u . y w C X x q b . 4 ,
1 X
r5 s
abg
Ž w C x q w C q 1 x qg f Ž u . y w 2 x q w C x q w C q 1 x q b
q w C X xw C X q 1 x q a g Ž u . . ,
1
r5X s X X
Ž w C x q w C q 1 x qg f Ž u . y w 2 x q w C x q w C q 1 x q b X
abg
q w C xw C q 1 x q a g Ž u . . ,
X 1
Cq C q1 CqC Xq1
m )k X 2 y
r6 s
abg žw x
C qg q 2
f Ž u . y b´ q y w C q 1 x q a q X 2
X
/
g Ž u. ,
X
CqC q1 1 Cq C q1
kmX ) y y
2
X
r6 s
abg žw x C X
qg q
2 f Ž u . q b´ q y w C q 1x q a q 2
g Ž u. , /
1 X
r7 s
a
Ž w C x q y w C x q f Ž u. . ,
1
r 7X s X
Ž w C x q y w C x q f Ž u. . ,
a
X 1
Cq C q1 CqC Xq1
k )mX 2 y
r8 s
abg žw X
C x qg q 2
f Ž u . y bh q y w C q 1 x q a q 2
X
/
g Ž u. ,
CqC Xq1 1 Cq C q1
mk X ) y y
2
X
r8 s
abg žw C x qg q 2 f Ž u . q bh q X
y w C q 1x q a q 2
g Ž u. ,/
1 X
r9 s
g
Ž w C q 1x q y w C q 1x q g Ž u . . ,
1
r 9X s X
Ž w C q 1x q y w C q 1x q g Ž u . . ,
g
1 X
r 10 s
abg
w C x q Ž w C q 1 x q b y w C x qg f Ž u . .
q w C X q 1x q Ž w C X x q b y w C q 1x q a g Ž u . . 4 . Ž C.2 .
where the abbreviations ŽB.9., as introduced in Appendix B, have been used for brevity.
Finally the coefficients of Ž4.14. are explicitly given as
sinh Ž 2h .
a Ž u . :s eyu ,
sinh Ž 2h q u .
sinh Ž 2h .
b Ž u . :s eu,
sinh Ž 2h q u .
sinh Ž u .
c Ž u . :s ,
sinh Ž 2h q u .
sinh Ž 2h y u .
d Ž u . :s . Ž C.3 .
sinh Ž 2h q u .
References
w1x H.A. Bethe, Z. Phys. 71 Ž1931. 205.

w2x R.J. Baxter, Ann. Phys. 70 Ž1972. 323.
w3x L. Onsager, Phys. Rev. 65 Ž1944. 117.
w4x C.N. Yang, Phys. Rev. Lett. 19 Ž1967. 1312.
w5x C.N. Yang, Phys. Rev. 168 Ž1968. 1920.
w6x R.J. Baxter, Exactly solved Models in Statistical Mechanics ŽAcademic Press, London, 1982..
w7x V.G. Drinfel’d, Quantum Groups in: Proc. International Congress of Mathematicians, Berkeley, 1986.
w8x V.G. Kac, Adv. Math. 26, 8-96.
w9x M.D. Gould, K.E. Hibberd, J.R. Links, Y.-Z. Zhang, Phys. Lett. A 212 Ž1996. 156.
w10x ¨
R.Z. Bariev, A. Klumper, J. Zittartz, Europhys. Lett. 32 Ž1995. 85.
w11x J. Gruneberg, to appear in Comm. Math. Phys.
w12x G.W. Delius, M.D. Gould, Y.-Z. Zhang, Nucl. Phys. B 432 Ž1994. 377.
w13x A.J. Bracken, G.W. Delius, M.D. Gould, Y.-Z. Zhang, J. Phys. A 27 Ž1994. 6551.
w14x M. Jimbo, Comm. Math. Phys. 102 Ž1986. 537.
w15x P.P. Kulish, E.K. Sklyanin, Quantum Spectral Transform Method – Recent Developments, in: Integrable
Quantum Field Theories, ed. J. Hietarina, C. Montonen, Lecture Notes in Physics 151 ŽSpringer, Berlin,
1981..
w16x J.F. Cornwell, Group Theory in Physics, Vol. 3, Supersymmetry and Infinite-dimensional Algebras
ŽAcademic Press, London, 1989..
w17x V. Chari, A. Presley, A Guide to Quantum Groups ŽCambridge Univ. Press, Cambridge, 1994..
w18x H. Yamane, Publ. Proc. Japan Acad. Ser. A 70 Ž1994. 31.
w19x H. Yamane, q-algr9603015 Ž1996..
w20x H. Yamane, Publ. RIMS Kyoto Univ. 30 Ž1994. 15.
w21x V.G. Kac, Infinite-dimensional Lie Algebras, 3rd ed. ŽCambridge Univ. Press, Cambridge, 1990..
w22x O. Gabber, V.G. Kac, Bull. Amer. Math. Soc. 5 Ž1981. 185.
w23x V.G. Kac, in: Differential Geometrical Methods in Mathematical Physics, ed. K. Bleuler, H.R. Petry, A.
Reed Ž1978. 597–626.
w24x T.D. Palev, V.N. Tolstoy, Commun. Math. Phys. 141 Ž1991. 549.
w25x V.G. Drinfel’d, Sov. Math. Dokl. 32 Ž1985. 254.
w26x M. Jimbo, Lett. Math. Phys. 10 Ž1985. 63.
w27x P.P. Kulish, E.K. Sklyanin, J. Sov. Math. 19 Ž1982. 1596.
w28x S.M. Koroshkin, V.N. Tolstoy, Commun. Math. Phys. 141, 599-607.
w29x S.M. Koroshkin, V.N. Tolstoy, in: Proc. International Workshop on Mathematical Physics ‘Quantum
Symmetries’ ed. H.-D. Doebner, V.K. Dobrev Ž1993. 336–351.
w30x Z.-Q. Ma, Yang–Baxter Equation and Quantum Enveloping Algebras ŽWorld Scientific, Singapore,
1993..
w31x Z. Maassarani, sol-int9903003.
w32x N.Y. Reshetikhin, Quantized universal enveloping algebras, the Yang–Baxter equation and invariants of
links I, LOMI-preprint E-4-87 Ž1987..
w33x N.Y. Reshetikhin, Quantized universal enveloping algebras, the Yang–Baxter equation and invariants of
links II, LOMI-preprint E-17-87 Ž1987..
w34x P.P. Kulish, J. Sov. Math. 35 Ž1986. 2648.
w35x J.H.H. Perk, C.L. Schultz, Phys. Lett. A 84 Ž1981. 407.
w36x E.K. Sklyanin, L.A. Takhtajan, L.D. Faddeev, Theor. Math. Phys. 40 Ž1980. 688.
w37x L.A. Takhtajan, L.D. Faddeev, Russ. Math. Surveys 34 Ž1979. 11.
w38x L.D. Faddeev, Sov. Sci. Rev. C 1 Ž1980. 107.
w39x P.P. Kulish, N.Y. Reshetikhin, JETP 80 Ž1981. 158.
w40x Z. Maassarani, J. Phys. A 28 Ž1995. 1305.
w41x J. Ramos, M. Martins, J. Phys. A 30 Ž1997. L195.
w42x M. Martins, solv-int 9901002 and private communication.
w43x ¨
H. Frahm, M.P. Pfannmuller, Nucl. Phys. B 479 Ž1996. 575.
w44x Z. Tsuboi, Physica A 252 Ž1998. 565.
w45x Z. Tsuboi, J. Phys. A Math. Gen. 31 Ž1998. 5485.
w46x C.L. Schultz, Physica A 122 Ž1983. 71.
w47x H. Frahm, private communication.
w48x C.K. Lai, J. Math. Phys. 15 Ž1974. 1675.
w49x B. Sutherland, Phys. Rev. B 12 Ž1975. 3795.
w50x N.Y. Reshetikhin, Sov. Phys. JETP 57 Ž1983. 691.
w51x J. Gruneberg, unpublished.
w52x ¨
H. Frahm, M.P. Pfannmuller, A.M. Tsvelick, Phys. Rev. Lett. 81 Ž1998. 2116.
w53x J. Gruneberg, to be published.
w54x ¨
A. Klumper, Ann. Physik 1 Ž1992. 540.
w55x ¨
A. Klumper, Z. Phys. B 91 Ž1993. 507.
w56x ¨
G. Juttner, ¨
A. Klumper, J. Suzuki, Nucl. Phys. B 487 Ž1997. 650.
w57x ¨
G. Juttner, ¨
A. Klumper, J. Suzuki, J. Phys. A 30 Ž1997. 1181.
w58x ¨
G. Juttner, ¨
A. Klumper, J. Suzuki, Nucl. Phys. B 522 Ž1998. 471.
Cumulative Author Index B561–B568
Abou Zeid, M. B561 Ž1999. 293 Becchi, C. B562 Ž1999. 412

Ahn, C. B565 Ž2000. 210 ´
Belanger, G. B568 Ž2000. 3
Ahn, C. B565 Ž2000. 611 Belli, P. B563 Ž1999. 97
Akhundov, A.A. B563 Ž1999. 82 Bellini, M. B563 Ž1999. 245
Alimohammadi, M. B565 Ž2000. 469 Bennett, J.F. B563 Ž1999. 390
´ B.
Alles, B562 Ž1999. 581 Bergshoeff, E. B564 Ž2000. 29
´ B.
Alles, B563 Ž1999. 213 Berkovits, N. B565 Ž2000. 333
Aloisio, R. B564 Ž2000. 489 Berkovits, N. B567 Ž2000. 61
Altland, A. B562 Ž1999. 445 Bernabei, R. B563 Ž1999. 97
´
Alvarez, E. B566 Ž2000. 363 ´ J.
Bernabeu, B563 Ž1999. 82
Amoros,´ G. B568 Ž2000. 319 Bershadsky, M. B567 Ž2000. 61
Anastasiou, C. B565 Ž2000. 445 Bijnens, J. B568 Ž2000. 319
Andreev, O. B561 Ž1999. 413 Bilal, A. B562 Ž1999. 181
Angelantonj, C. B566 Ž2000. 126 ´
Binetruy, P. B565 Ž2000. 269
Anselmi, D. B567 Ž2000. 331 Blom, J. B563 Ž1999. 506
Antonelli, V. B568 Ž2000. 40 Bobeth, C. B567 Ž2000. 153
Antoniadis, I. B565 Ž2000. 123 ¨
Bodeker, D. B566 Ž2000. 402
Aoki, H. B565 Ž2000. 176 Boer, D. B564 Ž2000. 471
Arfaei, H. B561 Ž1999. 57 Bogaerts, J. B562 Ž1999. 277
Arkani-Hamed, N. B567 Ž2000. 189 Bonelli, G. B564 Ž2000. 86
Ayala, A. B564 Ž2000. 204 Bonora, L. B564 Ž2000. 86
Azcoiti, V. B564 Ž2000. 489 Bosch, S. B565 Ž2000. 3
Boudjema, F. B568 Ž2000. 3
Babu, K.S. B566 Ž2000. 33 Brambilla, N. B566 Ž2000. 275
Bagger, J. B563 Ž1999. 3 Branson, T.P. B563 Ž1999. 603
Balachandran, A.P. B566 Ž2000. 441 Brax, Ph. B561 Ž1999. 77
Balog, J. B568 Ž2000. 503 Brecher, D. B566 Ž2000. 151
Bandos, I. B565 Ž2000. 291 Brouwer, P.W. B565 Ž2000. 653
Bardakci, K. B563 Ž1999. 437 Buras, A.J. B565 Ž2000. 3
Bardakci, K. B564 Ž1999. 285 Buras, A.J. B566 Ž2000. 3
Barkema, G.T. B565 Ž2000. 521 Burgio, G. B566 Ž2000. 547
´ A.A.
Barrientos Bendezu, B568 Ž2000. 305
Barvinsky, A.O. B561 Ž1999. 159 Callan, Jr. C.G. B565 Ž2000. 157
Bassett, B.A. B561 Ž1999. 188 Camino, J.M. B562 Ž1999. 103
Basu-Mallick, B. B566 Ž2000. 511 Caracciolo, S. B562 Ž1999. 581
Batista, E. B566 Ž2000. 441 Cardoso, G.L. B567 Ž2000. 87
Baver, E. B561 Ž1999. 473 Cardy, J. B565 Ž2000. 487
Beasley, C. B566 Ž2000. 599 Cardy, J. B565 Ž2000. 506
642 CumulatiÕe Author Index B561–B568
Cardy, J. B565 Ž2000. 521 Dvali, G. B562 Ž1999. 158

Caselle, M. B562 Ž1999. 549 Dvali, G. B563 Ž1999. 63
Chaichian, M. B567 Ž2000. 360
Chamblin, H.A. B562 Ž1999. 133 Echols, R. B564 Ž2000. 225
Chan, G. B564 Ž2000. 503 Edelstein, J.D. B561 Ž1999. 273
Chen, H. B561 Ž1999. 3 Ellis, J. B563 Ž1999. 107
Cheng, H.-C. B563 Ž1999. 21 Engelhardt, M. B567 Ž2000. 249
Cheung, Y.-K.E. B564 Ž2000. 259 Engels, J. B564 Ž2000. 303
Cho, G.-C. B565 Ž2000. 483 Erdmenger, J. B562 Ž1999. 315
Chryssomalakos, C. B567 Ž2000. 293 Erdmenger, J. B565 Ž1999. 363
Chu, C.-S. B562 Ž1999. 181 Esposito, S. B568 Ž2000. 421
Chu, C.-S. B558 Ž1999. 447 Evans, J.M. B561 Ž1999. 385
Codirla, C. B561 Ž1999. 43 Eyras, E. B564 Ž2000. 29
Colangelo, G. B566 Ž2000. 3
Collina, R. B562 Ž1999. 412 Faessler, A. B565 Ž2000. 38
Corcella, G. B565 Ž2000. 227 Fateev, V.A. B565 Ž2000. 611
Costa e Silva, I.P. B566 Ž2000. 441 ´ L.
Feher, B568 Ž2000. 503
Feinstein, A. B568 Ž2000. 405
Dai, C.J. B563 Ž1999. 97 Feng, J.L. B563 Ž1999. 3
Damour, T. B568 Ž2000. 93 Fleischer, J. B566 Ž2000. 423
Danielsson, U.H. B563 Ž1999. 279 Forrester, P.J. B563 Ž1999. 547
D’Appollonio, G. B565 Ž2000. 123 Fosco, C.D. B563 Ž1999. 533
Das, A. B561 Ž1999. 357 Franz, M. B562 Ž1999. 213
David, J.R. B564 Ž2000. 103 Frau, M. B564 Ž2000. 60
David, J.R. B564 Ž2000. 128 Frau, M. B565 Ž2000. 397
´
De Azcarraga, J.A. B567 Ž2000. 293 Freedman, D.Z. B562 Ž1999. 330
Deffayet, C. B565 Ž2000. 269 Freedman, D.Z. B562 Ž1999. 353
Degrassi, G. B567 Ž2000. 3 Freedman, D.Z. B562 Ž1999. 395
Del Duca, V. B566 Ž2000. 252 Freund, M. B562 Ž1999. 29
Del Duca, V. B568 Ž2000. 211 Fring, A. B567 Ž2000. 409
Delfino, G. B565 Ž2000. 521 Frixione, S. B568 Ž2000. 60
Demichev, A. B567 Ž2000. 360 Frizzo, A. B568 Ž2000. 211
De Pietri, R. B566 Ž2000. 547 ¨
Frohlich, J. B567 Ž2000. 231
´
Derkachov, S.E. B566 Ž2000. 203 Fuchs, J. B568 Ž2000. 543
De Wit, B. B567 Ž2000. 87 Fukui, T. B562 Ž1999. 477
De Wit, B. B568 Ž2000. 475 Furusaki, A. B565 Ž2000. 653
D’Hoker, E. B562 Ž1999. 330
D’Hoker, E. B562 Ž1999. 353 Gabadadze, G. B562 Ž1999. 158
D’Hoker, E. B562 Ž1999. 395 Galante, A. B564 Ž2000. 489
D’Hoker, E. B564 Ž1999. 503 Gallot, L. B564 Ž2000. 60
Di Carlo, G. B564 Ž2000. 489 Gambino, P. B567 Ž2000. 3
Dick, K. B562 Ž1999. 29 Ganor, O.J. B564 Ž2000. 259
Dienes, K.R. B567 Ž2000. 111 ´ Perez,
Garcıa ´ M. B564 Ž2000. 159
Di Francesco, P. B567 Ž2000. 515 Gepner, D. B561 Ž1999. 467
Dimopoulos, S. B567 Ž2000. 189 Gepner, D. B561 Ž1999. 473
Dine, M. B564 Ž2000. 225 Gerganov, B. B567 Ž2000. 391
Dittmaier, S. B565 Ž2000. 69 Geyer, B. B564 Ž2000. 517
Di Vecchia, P. B565 Ž2000. 397 Gherghetta, T. B567 Ž2000. 111
D’Olivo, J.C. B564 Ž2000. 204 Ghosh, S. B561 Ž1999. 451
Dorey, P. B563 Ž1999. 573 Gilkey, P.B. B563 Ž1999. 603
Dosch, H.G. B568 Ž2000. 287 Glover, E.W.N. B565 Ž2000. 445
Dudas, E. B565 Ž2000. 123 Goeke, K. B562 Ž1999. 213
Dudas, E. B567 Ž2000. 111 ´
Gomez, C. B566 Ž2000. 363
Dunne, G.V. B564 Ž2000. 591 ´
Gomez–Reino, M. B561 Ž1999. 273
CumulatiÕe Author Index B561–B568 643
´
Gomez Dumm, D. B563 Ž1999. 82 Jamin, M. B565 Ž2000. 3
´
Gonzalez-Arroyo, A. B564 Ž2000. 159 Janik, R.A. B565 Ž2000. 193
Gorbahn, M. B565 Ž2000. 3 Jegerlehner, F. B566 Ž2000. 423
Gould, M.D. B566 Ž2000. 529 Johansen, A. B562 Ž1999. 291
Gracey, J.A. B563 Ž1999. 390 Julia, B. B563 Ž1999. 448
Graf, G.M. B567 Ž2000. 231
Gray, J.P. B564 Ž2000. 225 Kaiser, D.I. B561 Ž1999. 188
Greene, B.R. B566 Ž2000. 599 Kakushadze, Z. B562 Ž1999. 78
Grianti, F. B563 Ž1999. 97 Kakushadze, Z. B562 Ž1999. 158
Grignani, G. B561 Ž1999. 243 Kaloper, N. B567 Ž2000. 189
Grillo, A.F. B564 Ž2000. 489 Kamani, D. B561 Ž1999. 57
Grojean, C. B561 Ž1999. 77 Kamimura, K. B561 Ž1999. 341
Grojean, C. B671 Ž1999. 133 Karabali, D. B566 Ž2000. 331
Gruneberg, J. B568 Ž2000. 594 Karsch, F. B564 Ž2000. 303
Guan, X.-W. B562 Ž1999. 433 Kawai, H. B565 Ž2000. 176
Guasch, J. B562 Ž1999. 3 Kawano, T. B565 Ž2000. 427
¨
Guijosa, A. B565 Ž2000. 157 Keski-Vakkuri, E. B563 Ž1999. 279
Guitter, E. B567 Ž2000. 515 Khalil, S. B564 Ž2000. 19
¨
Gursoy, U. B561 Ž1999. 473 Kilgore, W.B. B566 Ž2000. 252
Gusynin, V.P. B563 Ž1999. 361 Kim, C. B564 Ž2000. 430
Kim, C. B565 Ž2000. 611
Hagiwara, K. B565 Ž2000. 483 Kim, C. B566 Ž2000. 331
Halbersma, R. B564 Ž2000. 29 Kim, H.-C. B562 Ž1999. 213
Halyo, E. B561 Ž1999. 17 Kim, J.E. B564 Ž2000. 3
Hambye, T. B564 Ž2000. 391 Kinar, Y. B566 Ž2000. 103
Harikumar, E. B565 Ž2000. 385 King, S.F. B562 Ž1999. 57
Hasler, D. B567 Ž2000. 231 Kirsten, K. B563 Ž1999. 603
Hassan, M. B561 Ž1999. 385 Kitanine, N. B567 Ž2000. 554
Hassan, S.F. B568 Ž2000. 145 Kitazawa, Y. B565 Ž2000. 176
Hatsuda, M. B561 Ž1999. 341 Kleijn, B. B568 Ž2000. 475
Hauer, T. B562 Ž1999. 497 Kniehl, B.A. B563 Ž1999. 200
Hauer, T. B567 Ž1999. 61 Kniehl, B.A. B568 Ž1999. 305
Haug, O. B565 Ž2000. 38 ¨
Kohler, G.O. B564 Ž2000. 391
Hautmann, F. B563 Ž1999. 153 Korchemsky, G.P. B566 Ž2000. 203
He, H.L. B563 Ž1999. 97 Korff, C. B567 Ž2000. 409
Hebecker, A. B568 Ž2000. 287 Kosower, D.A. B563 Ž1999. 477
Henneaux, M. B563 Ž1999. 448 Kovalenko, S. B565 Ž2000. 38
Hikami, K. B566 Ž2000. 511 Kraus, P. B563 Ž1999. 259
Ho, P.-M. B568 Ž2000. 447 Kristjansen, C. B567 Ž2000. 515
Hofman, C. B561 Ž1999. 125 Krogh, M. B564 Ž2000. 259
Hoppe, J. B567 Ž2000. 231 Kruczenski, M. B563 Ž1999. 279
Hull, C.M. B561 Ž1999. 293 Kuang, H.H. B563 Ž1999. 97
Hull, C.M. B564 Ž1999. 29 Kummer, W. B565 Ž2000. 291
Kunszt, Z. B563 Ž1999. 153
Ignesti, G. B563 Ž1999. 97 Kuraev, E.A. B568 Ž2000. 40
Ilieva, N. B565 Ž2000. 629 Kurihara, Y. B565 Ž2000. 49
Incicchitti, A. B563 Ž1999. 97 Kuroda, M. B565 Ž2000. 49
Ishibashi, N. B565 Ž2000. 176 Kyae, B. B564 Ž2000. 3
Isidori, G. B566 Ž2000. 3
Iso, S. B565 Ž2000. 176 Lalak, Z. B563 Ž1999. 107
Izquierdo, J.M. B567 Ž2000. 293 Langlois, D. B565 Ž2000. 269
Langmann, E. B563 Ž1999. 506
¨
Jager, S. B565 Ž2000. 3 Larsen, F. B563 Ž1999. 259
Jakob, R. B564 Ž2000. 471 Lautenbacher, M.E. B565 Ž2000. 3
Lazaroiu, C.I. B566 Ž2000. 599 Montecchia, F. B563 Ž1999. 97

Leader, E. B562 Ž1999. 291 Moore, G.D. B568 Ž2000. 367
Lechtenfeld, O. B566 Ž2000. 489 ´
Morales-Tecotl, H.A. B566 Ž2000. 547
Lee, J. B563 Ž1999. 125 Moss, I.G. B565 Ž2000. 345
Lee, J.S. B564 Ž2000. 3 Mountain, A.J. B561 Ž1999. 385
Lee, K. B563 Ž1999. 461 Mourad, J. B567 Ž2000. 133
Lee, S. B563 Ž1999. 125 Mudry, C. B565 Ž2000. 653
Lee, S. B563 Ž1999. 349 Mulders, P.J. B564 Ž2000. 471
Leibbrandt, G. B566 Ž2000. 373 ¨
Muller, U. B568 Ž2000. 457
Leontaris, G.K. B567 Ž2000. 32 ¨
Mulsch, D. B564 Ž2000. 517
Lerda ,, A. B564 Ž2000. 60 Mussardo, G. B567 Ž2000. 454
Lerda, A. B565 Ž2000. 397
Li, M. B568 Ž2000. 195 Nagao, T. B563 Ž1999. 547
Li, T. B564 Ž2000. 441 Nair, V.P. B566 Ž2000. 331
Liao, S.-B. B567 Ž2000. 493 Nandy, S. B561 Ž1999. 451
Liccardo, A. B565 Ž2000. 397 Nardelli, G. B567 Ž2000. 583
Lindner, M. B562 Ž1999. 29 Nason, P. B565 Ž2000. 245
Litim, D.F. B562 Ž1999. 237 Naud, J.D. B565 Ž2000. 572
Lozano, Y. B564 Ž2000. 29 Nest, M. B562 Ž1999. 567
Lu, L. B561 Ž1999. 3 Nesti, F. B564 Ž2000. 86
u ukaszuk, L.
L B562 Ž1999. 291 Nilles, H.P. B561 Ž1999. 30
Lunghi, E. B568 Ž2000. 120
¨
Luscher, M. B568 Ž2000. 162
Okawa, Y. B566 Ž2000. 348
Ma, J.M. B563 Ž1999. 97 Okuyama, K. B565 Ž2000. 427
Maartens, R. B561 Ž1999. 188 Oleari, C. B565 Ž2000. 245
MacKay, N.J. B561 Ž1999. 385 Oleari, C. B565 Ž2000. 445
Maillet, J.M. B567 Ž2000. 554 Olechowski, M. B561 Ž1999. 30
Maltoni, F. B566 Ž2000. 252 Olsen, K. B566 Ž2000. 562
Maltoni, F. B568 Ž2000. 211
Manashov, A.N. B566 Ž2000. 203 Palla, L. B568 Ž2000. 503
Mandal, G. B564 Ž2000. 128 Park, J.-S. B561 Ž1999. 125
Mangano, G. B568 Ž2000. 421 Park, Q.-H. B563 Ž1999. 461
Manuel, C. B562 Ž1999. 237 Paschos, E.A. B564 Ž2000. 391
March-Russell, J. B567 Ž2000. 189 Pastor, S. B566 Ž2000. 92
Martins, M.J. B562 Ž1999. 433 Pati, J.C. B566 Ž2000. 33
Mas, J. B561 Ž1999. 273 Patra, B.K. B564 Ž2000. 145
Matchev, K.T. B563 Ž1999. 21 Pelissetto, A. B562 Ž1999. 581
Mathur, S.D. B562 Ž1999. 330 Peloso, M. B567 Ž2000. 583
Mathur, S.D. B562 Ž1999. 353 Pena, C. B564 Ž2000. 159
Matusis, A. B562 Ž1999. 330 Penati, S. B567 Ž2000. 454
Matusis, A. B562 Ž1999. 353 Penin, A.A. B563 Ž1999. 200
Meissner, K.A. B561 Ž1999. 30 Pepe, M. B562 Ž1999. 581
Melles, M. B564 Ž2000. 325 Pepe, M. B563 Ž1999. 213
Mendel, E. B562 Ž1999. 567 ´ Bueno, J.C.
Perez B567 Ž2000. 293
Menon, V.J. B564 Ž2000. 145 Pernice, S.A. B561 Ž1999. 357
Metsaev, R.R. B563 Ž1999. 295 Perry, M.J. B561 Ž1999. 43
Metz, A. B568 Ž2000. 287 Perry, M.J. B566 Ž1999. 151
Miele, G. B568 Ž2000. 421 Peschanski, R. B565 Ž2000. 193
Mielke, E.W. B564 Ž2000. 185 Petkou, A. B561 Ž1999. 100
Mikhailov, A.Yu. B564 Ž2000. 259 Pineda, A. B566 Ž2000. 275
Miransky, V.A. B563 Ž1999. 361 Pirner, H.J. B568 Ž2000. 287
Misiak, M. B567 Ž2000. 153 Pisanti, O. B568 Ž2000. 421
Mohaupt, T. B567 Ž2000. 87 Plesser, M.R. B566 Ž2000. 599
CumulatiÕe Author Index B561–B568 645
Pokorski, S. B563 Ž1999. 107 Sfetsos, K. B561 Ž1999. 316

Polonsky, N. B563 Ž1999. 3 Shafi, Q. B564 Ž2000. 19
Polonyi, J. B567 Ž2000. 493 Shaikhatdenov, B.G. B568 Ž2000. 40
Ponkratenko, O.A. B563 Ž1999. 97 Sharpe, E. B561 Ž1999. 433
Prata, J.N. B565 Ž2000. 641 Shiiki, N. B565 Ž2000. 345
ˇ
Presnajder, P. B567 Ž2000. 360 Shin, H.J. B563 Ž1999. 461
Prosperi, D. B563 Ž1999. 97 Shovkovy, I.A. B563 Ž1999. 361
Pryadko, L.P. B565 Ž2000. 572 Sibold, K. B565 Ž2000. 363
Siebelink, R. B563 Ž1999. 259
Silva, S. B563 Ž1999. 448
´
Rakos, A. B562 Ž1999. 497
Silvestrini, L. B565 Ž2000. 3
Ramallo, A.V. B562 Ž1999. 103
Rastelli, L. B562 Ž1999. 330
Rastelli, L. B562 Ž1999. 353
Simons, B.D. B562 Ž1999. 445
Rastelli, L. B562 Ž1999. 395
Singh, C.P. B564 Ž2000. 145
Reall, H.S. B562 Ž1999. 133
Sivakumar, M. B565 Ž2000. 385
Refolli, A. B564 Ž2000. 241
Skenderis, K. B561 Ž1999. 100
Reinhardt, H. B567 Ž2000. 249
Smirnov, A.Yu. B563 Ž1999. 63
Rey, S.-J. B564 Ž2000. 430
Smirnov, V.A. B566 Ž2000. 469
Rey, S.-J. B565 Ž2000. 210
Rim, C. B565 Ž2000. 611 ` J.
Sola, B562 Ž1999. 3
Soldan, P.H. B564 Ž2000. 391
Ritz, A. B566 Ž2000. 311
Sondhi, S.L. B565 Ž2000. 572
Rizos, J. B567 Ž2000. 32
Sonnenschein, J. B566 Ž2000. 103
Romanino, A. B562 Ž1999. 29
Soper, D.E. B563 Ž1999. 153
Romanino, A. B566 Ž1999. 3
Sorin, A. B566 Ž2000. 489
Rupp, C. B565 Ž2000. 363
Soto, J. B566 Ž2000. 275
Sridhar, K. B568 Ž2000. 3
Sagnotti, A. B565 Ž2000. 123 Stirling, W.J. B564 Ž2000. 325
Saleur, H. B567 Ž2000. 602 Strickland, M. B567 Ž2000. 493
´
Sanchez de Santos, J.M. B562 Ž1999. 103 Strigazzi, P. B564 Ž2000. 60
Santamaria, A. B563 Ž1999. 82 Stuart, R.G. B564 Ž2000. 343
Santiago, J. B563 Ž1999. 45 Suranyi, P. B565 Ž2000. 487
Savoy, C.A. B561 Ž1999. 77 Szabo, R.J. B566 Ž2000. 562
Savvidy, G.K. B561 Ž1999. 117
Savvidy, K.G. B561 Ž1999. 117 Tada, T. B565 Ž2000. 176
Scheideler, T. B564 Ž2000. 303 Talavera, P. B568 Ž2000. 319
Schildknecht, D. B565 Ž2000. 49 Tamburini, F. B561 Ž1999. 188
Schlottmann, P. B565 Ž2000. 535 Tarasov, O.V. B566 Ž2000. 423
Schlottmann, P. B565 Ž2000. 555 Tateo, R. B562 Ž1999. 549
Schreiber, E. B566 Ž2000. 103 Tateo, R. B563 Ž1999. 573
Schubert, C. B564 Ž2000. 591 Taylor, T.R. B562 Ž1999. 78
Schulz, B.J. B567 Ž2000. 409 Teotonio-Sobrinho, P. B566 Ž2000. 441
Schunck, F.E. B564 Ž2000. 185 Terras, V. B567 Ž2000. 554
Schweigert, C. B568 Ž2000. 543 Terzi, N. B564 Ž2000. 241
Scimemi, A.M.I. B568 Ž2000. 120 Thirring, W. B565 Ž2000. 629
Scrucca, C.A. B564 Ž2000. 555 Thomas, S. B563 Ž1999. 107
Segura, J. B566 Ž2000. 92 Tomasiello, A. B564 Ž2000. 86
Seide, S. B562 Ž1999. 524 Torres, M. B564 Ž2000. 204
Sekiya, S. B561 Ž1999. 341 Tretyak, V.I. B563 Ž1999. 97
Semenoff, G.W. B561 Ž1999. 243 Trinchero, R.C. B563 Ž1999. 533
Semikoz, V.B. B566 Ž2000. 92 Troost, J. B568 Ž2000. 180
Serone, M. B564 Ž2000. 555
Sevrin, A. B562 Ž1999. 277 Umeda, Y. B565 Ž2000. 483
Seymour, M.H. B565 Ž2000. 227 Urban, J. B567 Ž2000. 153
Urrutia, L.F. B566 Ž2000. 547 Wadia, S.R. B564 Ž2000. 103
Uwer, P. B563 Ž1999. 477 Wadia, S.R. B564 Ž2000. 128
Weigt, G. B568 Ž2000. 457
Vaidya, S. B564 Ž2000. 128 Wetterich, C. B562 Ž1999. 524
Vainshtein, A. B566 Ž2000. 311 Wilczek, F. B566 Ž2000. 33
Vairo, A. B566 Ž2000. 275 Williams, J. B566 Ž2000. 373
Valle, J.W.F. B566 Ž2000. 92 Woynarovich, F. B562 Ž1999. 497
Van Baal, P. B564 Ž2000. 159
Van der Loo, S. B562 Ž1999. 277 Xiong, Z. B561 Ž1999. 3
Van der Schaar, J.P. B564 Ž2000. 29
Vandoren, S. B568 Ž2000. 475
Van Gils, S. B562 Ž1999. 277 Yamada, Y. B566 Ž2000. 642
Van Neerven, W.L. B568 Ž2000. 263 Yang, B. B565 Ž2000. 611
Van Ritbergen, T. B564 Ž2000. 343 Yang, S.-K. B566 Ž2000. 642
Vassilevich, D.V. B563 Ž1999. 603 Yau, S.-T. B567 Ž2000. 231
´
Vazquez-Mozo, M.A. B568 Ž2000. 405 ´ F.J.
Yndurain, B563 Ž1999. 45
Veneziano, G. B568 Ž2000. 93 Yung, A. B562 Ž1999. 191
Veretin, O.L. B566 Ž2000. 469
Vergados, J.D. B565 Ž2000. 38 Zanon, D. B564 Ž2000. 241
Vergara, J.D. B566 Ž2000. 547 Zdesenko, Yu.G. B563 Ž1999. 97
Vinti, S. B562 Ž1999. 549 Zhang, Y.-Z. B566 Ž2000. 529
Vogelsang, W. B568 Ž2000. 60 Zhukov, S. B567 Ž2000. 61
Vogt, A. B568 Ž2000. 263 Zvyagin, A.A. B565 Ž2000. 555
Volkov, M.S. B566 Ž2000. 173 Zwiebach, B. B567 Ž2000. 61

Nucl - Phys.B v.568 PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Nucl - Phys.B v.568 PDF

Загружено:

Авторское право:

Доступные форматы

Nuclear Physics B 568 Ž2000.

SUSY Higgs at the LHC: Large stop mixing effects and

PACS: 12.60.Jv; 14.80.Cp; 14.80.Ly

The most popular alternative to the Standard Model, S M , is supersymmetry which

2. A warm up: Variation with M A , the no mixing case

tan a tan b s y Ž 1 q r . with r < 1, Ž 2.5 .

3. Mixing in the stop sector

˜ U2˜3 R q m2t q 23 MZ2 sin2u W cos Ž 2 b . ,

t˜1 s cos u t˜ t˜L q sin u t˜ t˜R . Ž 3.3 .

3.1. The t˜1 t˜1 h Õertex

3.2. pp ™ t˜ t˜ h at the LHC

3.3. The t˜2 t˜1 h,A,H Õertex and t˜2 ™ t˜ h,H,A

The t˜1 t˜2 h vertex may be cast into

coupling does not depend on the stop mixing angle:

3.4. Constraints from low Higgs masses, Dr and CCB

A2t q 3 m2 - 7.5 MQ2˜ 3 q Mt˜2R .

3.5. tan b s 2.5

Fig. 9. As in Fig. 7 but for R g ggg versus s Ž t˜1 t˜1 h. Žfb..

3.5.3. Stop mixing with a low MA

Fig. 14. As in Fig. 13 but for R g ggg versus s Ž pp ™ t˜ t˜ h..

3.6.3. Lifting the degeneracy in the third family scalar masses

w29x Z. Kunszt, Nucl. Phys. B 247 Ž1984. 339.

w78x Z. Hioki, Prog. Theor. Phys. 73 Ž1985. 1283.

Radiative large-angle Bhabha scattering in collinear kinematics

PACS: 12.20.ym; 12.20.Ds

The process of electron–positron scattering is commonly used for luminosity mea-

In paper w9x we considered Bhabha scattering to O Ž a . order exactly improved by the

2. Born expressions in collinear kinematics

Let us begin revising the radiative Bhabha scattering process

with the properties

4. Virtual and soft photon emission in k 1 I p 1 kinematics

Ø single photon exchange between electron and positron lines ŽG,L-type.;

For L-type FD Žsee Fig. 1Ž3,4.. the initial spinor uŽ p 1 . is replaced by

The B-type FD shown in Fig. 1Ž7. with uncrossed legs gives

=W Ž E11 y E1 q xE1 k y xEk . . Ž 19 .

s13 y u13 p 1 t 3 y u13

P s1 s 13 Ž L s1 y ip . y 59 , F s x 1 A 2 q t 1 x 1 Ž J11 y J1 q xJ1 k y xJk . ,

Ý < M < 2hardqsoft s Ý < M < 2B wsoft Ž k 1 I p1 . , Ž 22 .

q 12 Ž L2s q L2s1 q L2t q L2t 1 y L2u y L2u1 . q ln y 1 Ž L u1 y L s1 y L t 1 .

The complete expression for the correction in the case k 1 I p 1 reads

photons valid to a logarithmic accuracy in the form

6. Two hard photon emission and results in LLA

with the non-singlet structure function D Ž z . w21x

For the case k 1 I pX1 the correction is found to be

mere consequence of a complicate kinematics of 2 3 type hard subprocess Žsee w14x.; ™

For the vector integral coefficients we get

ys1 L2t q 4 u1 L s1 Lr q 2 s1 L s1 L t . Ž A.4 .

The relevant quantities for tensor B-type integrals are:

w1x A.B. Arbuzov et al., Phys. Lett. B 383 Ž1996. 238.

w12x A.B. Arbuzov et al., Phys. Atom. Nucl. 60 Ž1997. 591.

Isolated-photon production in polarized pp collisions

We perform a detailed study of the production of isolated prompt photons in polarized

PACS: 13.88.qe; 13.85.Qk; 12.38.Bx

ton reaction is twofold: first, it brings us back to the dependence on non-perturbative

basic idea of determining D g in ™™ ™

photons in Ref. w41,42x and for ™ ™

perform the first measurements on ™™ ™

2. Isolated photons in perturbative QCD

q dx 1 dx 2 dzf aŽ A. Ž x 1 , mXF . f bŽ B . Ž x 2 , mXF .

2.2. NLO computer codes

In order to give phenomenological predictions, we will use two different computer

is based on an inclusive calculation of the contributing partonic subprocesses ab g X

3. Inclusive isolated-photon observables