Nucl - Phys.B v.642 PDF

Nuclear Physics B 642 (2002) 3–10
www.elsevier.com/locate/npe
The structure of the pion and nucleon, and leading

neutron production at HERA
Garry Levman
Department of Physics, University of Toronto, Toronto, ON M5S 1A7, Canada
Received 28 May 2002; accepted 7 August 2002
Abstract
Attention is paid to recent results from the ZEUS Collaboration on the photo- and electro-
production of leading neutrons in e+ p collisions at HERA. Some implications for the structure of
the pion and the nucleon are discussed.
 2002 Elsevier Science B.V. All rights reserved.
PACS: 12.40.Nn; 13.60.Hb; 13.60.Rj
Keywords: Nucleon; Pion; Meson exchange; Structure
1. Introduction
The ZEUS Collaboration has published data on leading neutron production in neutral
current e+ p collisions at HERA [1]. The data have been anticipated because of the light
they cast on the structure of the pion and nucleon.
In [1] the ZEUS Collaboration uses an effective flux method [2] to determine the photon–
pion total cross section at high W , and the deep inelastic structure function of the pion,
F2π (x, Q2 ).1 In this note, the method is described; its assumptions and inherent difficulties
are discussed; and the problem of error is confronted. Implications of the ZEUS data are
examined critically for the role of meson exchange in the production of leading neutrons
and for the structure of the pion.
E-mail address: levman@physics.utoronto.ca (G. Levman).

URL address: http://www.elsevier.com/locate/npe.
1 The notation used follows [1]. W is a center of momentum frame energy. x is a Bjorken scaling variable.
Q2 is the virtuality of the exchanged photon.
0550-3213/02/$ – see front matter  2002 Elsevier Science B.V. All rights reserved.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 8 5 - 5
RAPID COMMUNICATION
4 G. Levman / Nuclear Physics B 642 (2002) 3–10
2. Meson exchange picture of leading baryon production
Leading baryons are those produced with small transverse momentum, pT , and carrying
a large fraction xL , of the incoming proton’s energy. They are thought to represent
the conserved baryon current because the squared 4-momentum transfer, t, between the
incoming proton and outgoing baryon is small (Fig. 1 of [1]).
The production data for leading baryons can be understood in terms of meson
exchanges. Consider the specific case γp → Xn. The cross section for the production of
a leading neutron is given by a sum over all possible meson exchanges between the photon
and proton which lead to an outgoing neutron at the proton vertex (Fig. 1 of [1]):

σγp→Xn xL , t, W 2 ∼ aγ ,p (t) fi/p (xL , t)σγ i (1 − xL )W 2 , t , (1)
i
where i = π, ρ, or a2 meson; σγ i is the total γ i cross section; fi/p is the splitting function
for p → n via i exchange; and aγ ,p is a form factor which accounts for rescattering of
the neutron (absorption) due to the finite size of the projectile. Projectiles other than the
photon, and produced baryons other than the neutron are treated similarly. For example,
in the reaction pp → Xp the ω and f2 mesons, and the pomeron P also contribute. There
are complications. The direct production of baryons which decay, such as ∆ → nπ , leads
to the indirect production of nucleons; there can be two-meson exchange, and different
exchanges can interfere; the form factor, aγ ,p , can depend on xL and perhaps on the
exchanged meson i. The simple factorization assumed in Eq. (1) is only an approximation.
3. Leading neutron production and pion exchange
In hadro-production experiments (projectile = p, π ) leading neutron production is

believed to be dominated by pion exchange. The spectrum of leading neutrons in pp
collisions at Fermilab and the ISR is fairly well represented by a single term (one-meson-
exchange) of the form
1 g2 −2t
σpp→Xn = (1 − xL )1−2t σπp (1 − xL )W 2 , t = 0 , (2)
4π 4π (t − mπ )
2 2
where mπ is the mass of the pion, g 2 /4π = 14.5, and σπp is the measured, on-mass-shell,
πp total cross section. The splitting function
1 g2 −2t
feff (xL , t) = (1 − xL )1−2t (3)
4π 4π (t − m2π )2
is very close to that expected from pure one-pion exchange. The result is simple, but
surprising. Note in Eq. (3) the absence of
• an absorptive factor a(t);

• an off-mass-shell correction, that is, a t dependence to σ (πp);
• contributions from ρ and a2 exchange, and ∆ production.
RAPID COMMUNICATION
G. Levman / Nuclear Physics B 642 (2002) 3–10 5
The self-consistency of the meson exchange picture requires the presence of such
‘backgrounds’. In addition, the value, 14.5, of the pπ 0 p coupling constant is larger than
recent estimates [3,4] which suggest that g 2 /4π lies in the range (13.5–14.1).
The ZEUS Collaboration has reviewed (see Section 10.1 of [1]) the expected back-
grounds to one-pion exchange which are known both from experimental measurements
(e.g., ∆ production) and from theoretical calculations (e.g., absorption). The effects are
large, roughly 20–30%, apparently confounding the experimental observation embodied in
Eqs. (2) and (3); however,
• ρ, a2 exchange and ∆ production increase neutron production, while absorptive effects

decrease neutron production
• absorption preferentially removes ρ and a2 compared to π exchange because a(t)
decreases with increasing t and the contribution of the higher mass mesons increases
relative to the pion at high t.
The conclusion to draw is that simple one-pion exchange agrees well with the hadro-
production data because of a fortuitous near cancellation of absorptive effects and
background contributions. The large value of the effective coupling constant (14.5) implies
that the cancellation is not perfect, and that there is a residual 5% contribution to the
normalization. feff is not the true pion flux, but rather an effective pion flux in the presence
of competing and ‘compensating’ processes.
4. The effective flux
The effective flux of pions in the proton (effective splitting function) can be defined by
experiment as the deconvolution of the measured semi-inclusive differential cross section
for pp → Xn and the measured πp total cross section,
dσpp→Xn
feff (xL , t) ≡ /σπp , (4)
dxL dt
where deconvolution is written as division. By construction and definition feff corrects

one-pion exchange in pp collisions for ρ and a2 exchange, ∆ production, rescattering and
off-mass-shell effects, etc. From Eq. (1) it is seen that a ‘model’ for feff is

feff ∼ ap,p (t) fi/p (xL , t)σip (1 − xL )W 2 , t /σπp (1 − xL )W 2 , t = 0 .
i
It is important to emphasize: experiments find that a convenient parameterization of feff is

given by Eq. (3). Because of the universal behavior of hadronic cross sections feff is also
expected to work for πp and γp collisions.
RAPID COMMUNICATION
5. The structure of the pion and the effective flux
It has been proposed [8,9] to measure the structure function of the pion F2π at HERA
using virtual γ ∗ π collisions assuming pion exchange dominates the interaction γ ∗ p →
Xn, just as it does in hadro-production.
A vital consistency check for any extraction of F2π using the electro-production of
leading neutrons is the demonstration that the photo-production of leading neutrons is in
accord with the hadro-production measurements. The importance of the photo-production
data arises because:
• W is the single relevant leptonic variable (Q2 = 0);

• the W dependence of measured hadronic cross sections closely follows a simple
universal behavior; at high W the hadronic total cross sections behave as a power
law W 2(αP −1) , where αP is a constant independent of the projectile and target;
• the photo-production (Q2 < 0.02 GeV2 ) and transition region (0.1 < Q2 < 0.74 GeV2 )
data from HERA are in good agreement with vector dominance in which the photon
behaves like an exchanged vector meson and the collision is hadronic [7,10].
In contrast to photo-production, the situation in electro-production is less well

understood. The additional degree of freedom quantified by Q2 adds complication.
Rescattering, background composition and levels, and factorization properties can vary
with Q2 . Moreover, in deep inelastic scattering (Q2 4 GeV2 ) the chief interest lies
in measuring the W (that is x, x ≈ Q2 /W 2 at low x) dependence of the cross section.
In summary, a prerequisite for a believable determination of F2π using high Q2 electro-
production data is a consistent determination of σγ π using photo-production data.
A standard technique for determining σ (γ π) involves fitting the observed differential
cross section for γp → Xn while simultaneously correcting for background exchanges,
Delta production, absorption, and off-mass-shell effects [11]. This procedure has the
important advantage of providing an estimate of the size of the contributing sub-processes
and giving a transparent estimate of the error. It has the disadvantage that there is a large
number of effects to be accounted for, with a corresponding large degree of freedom. The
parameters must be guessed, fit, or fixed by experiment.
The effective flux is used instead to directly determine σ (γ π) without determining
the individual sub-processes. This has the advantages of simplicity and directness, but the
disadvantage that no information is obtained about the relative importance of the various
processes which contribute. The error is difficult to estimate (see Section 6) because the
amount of pion exchange present is not unambiguously determined. If pion exchange does
not contribute, the result, although well defined, is meaningless.
The arguments which follow (in Sections 5.1 and 5.2) give the essence of what obtains
from the ZEUS measurement.
RAPID COMMUNICATION
5.1. σ (γ π) at high W
The γ π total cross section is obtained from the photo-production data using
dσ (γp → Xn)/dxL
σ (γ π) = ,
feff
where the variable t has been integrated out in the numerator. Substitution for feff from its
definition in Eq. (4) shows that σ (γ π) can be determined from (double) ratios
dσ (γp → Xn)/dxL
σ (γ π) σ (γp) σ (πp)
= (5)
σ (γp) dσ (pp → Xn)/dx L σ (pp)
σ (pp)
of measured quantities.
The ZEUS Collaboration has observed that the relative rate of neutron production in
photo-production at HERA is half that of pp collisions. It follows from Eq. (5) that
σ (γ π)/σ (γp) is half σ (πp)/σ (pp). Therefore, as ZEUS deduces directly,
σ (γ π)
σ (γp)/3
rather than two-thirds as expected from Regge factorization or the counting of valence
quarks (the Additive Quark Model).
An important consistency check, which the ZEUS Collaboration has performed (Figs. 6
and 8 of [1]), is that the neutron energy and angular distributions are described by feff .
5.2. F2π at low x
The double ratio (Eq. (5)) is robust for hadro- and photo-production; however, for
electro-production, absorptive rescattering decreases as Q2 increases, as discussed in [5]
and [6] (see also Fig. 9 of [1]). At low Q2 (Q2 1 GeV2 ), F2π and σ (γ ∗ π) are related by
Q2
F2π
σ (γ ∗ π) (6)
4π 2 α
which, in analogy with ρ dominance in γp interactions [7], can be written as
Q2 m2ρ
F2π
σ (γ π), (7)
4π 2 α m2ρ + Q2
where mρ is the mass of the ρ meson.
Eqs. (6) and (7) connect the photo-production γ π cross section to the transition region
γ ∗ π cross section; together with the corresponding equations for γ (∗) p interactions,
σ (γ ∗ π) σ (γ ∗ p)

(8)
σ (γ π) σ (γp)
obtains. In the kinematic region 0 < Q2 < 4 GeV2 the ratio of neutron production (i.e.,
tagged divided by all) increases. To maintain Eq. (8) an absorptive correction needs to be
RAPID COMMUNICATION
applied. Rescattering decreases, and the effective flux must increase correspondingly in
order to maintain consistency between photo-production, the transition region, and deep
inelastic scattering. The ZEUS data require for deep inelastic scattering (DIS) feff →
DIS = 1.3f .
feff eff
The measurements show that 2.5% of deep inelastic scattering events, independent of
x and Q2 , have a leading neutron with 0.64 < xL < 0.82 and pT < 0.66xL GeV (Fig. 11
of [1]). The constancy of the ratio of neutron production in DIS as a function of x and
p p
Q2 implies that F2π (x, Q2 ) = kF2 ((1 − xL )x, Q2 ) = kF2 (0.27x, Q2) with a constant of
proportionality k given by
0.025
k= DIS (x , t) dx dt
= 0.25,
feff L L
where the integration is over the xL and t (pT ) range covered by the ZEUS measurement.
p
Because F2 (x) ∝ (1/x)λ with λ ≈ 0.2 [7]
p
F2π x, Q2 ≈ F2 x, Q2 /3.
6. Discussion
For the determination of F2π the effective flux method requires little theoretical
input. Only the general concepts of meson exchange theory are needed. The data,
hadro-production, photo-production and electro-production, determine everything. Note
especially that
• one-pion-exchange is not assumed. It is only assumed that pion exchange dominates

neutron production;
• the contribution of backgrounds is not neglected. The method accounts for ρ and a2
exchange and for indirect neutron production through ∆ production and decay;
• the exact composition and relative contribution of the backgrounds is unimportant;
• absorptive effects are measured;
• the consistency of the photo-production (Q2 < 0.02 GeV2 ), transition region (0.1 <
Q2 < 0.74 GeV2 ), and DIS (Q2 > 4 GeV2 ) analyses requires and defines an absorptive
correction.
If absorption approximately compensates for background exchanges in hadro- and photo-

production, as the data suggest, then the 30% rise in neutron ratio between photo-
production and DIS is a measure of the relative contribution of the backgrounds.
The effective flux gives a normalization of σ (γ π) which differs by a factor of two
from that expected by Regge factorization or valence quark counting. The disagreement
follows directly from the experimental observation that the relative rate of neutrons in
photo-production is half that observed in hadro-production. Can feff , which is determined
by pp collisions, be wrong by a factor of two when used for photo-production?
Background exchanges and absorptive effects are of the same magnitude, but contribute
with opposite signs. The ZEUS leading neutron data show that the relative size of
RAPID COMMUNICATION
absorptive effects is approximately 20–30%. Background processes must contribute at

approximately this level, in agreement with estimates from phenomenological studies
(reviewed in Section 10.1 of [1]). With this in mind, a conservative estimate for the error
on using feff for photo-production is obtained by assuming that pion exchange contributes
50% and that the backgrounds change by 50% on moving from pp to γp collisions.
The change in feff is then 25%. Suppose that the low neutron rate is not due to a small
photon–pion cross section, but rather due to changes in absorption or backgrounds. Then
absorption must increase or backgrounds decrease strongly in the photon–proton system
compared to the proton–proton system.
If the meson exchange picture underestimates the normalization of σ (γ π) by a factor
of two, the theory is wrong, misapplied, or there is a significant missing ingredient. Then
one cannot extract with any confidence the x and Q2 behavior of F2π .
If the photo-production result is disbelieved, the ZEUS data can be taken as evidence
that meson exchange plays little role in the production of leading nucleons at HERA. One
p
then argues that the ZEUS measurement is merely a reflection of F2 , arising because of
factorization (limiting fragmentation) at the proton vertex and baryon conservation.
On the other hand, the experimental evidence for the meson exchange picture is good
(see Ref. [1]). One can accept the picture and the normalization of σ (γ π) and F2π implied
by the ZEUS data. In this case some conjectures can be make:
• the x dependence of F2 for all hadrons is similar at low x and is determined mainly by
the QCD evolution equations, only weakly by the valence structure of the hadron;
• the number of partons at low x in the pion is 1/3 that of the proton; since the charged
radius of the pion is 2/3 the proton’s, the volume density of partons in the pion is
approximately the same as in the proton;
• there is a large probability for the proton to be found in a meson–nucleon or meson-
Delta Fock state (the nucleon within the nucleon); the proton is a loosely bound
meson–nucleon system composed of infinitely bound partons;
• the quark–antiquark sea of a hadron is generated mainly by valence–valence interac-
tions (three for the proton and one for the pion), and not by self interactions.
Moreover, there is a significant violation of the quark counting rules and Regge
factorization.
7. Conclusions
In the context of the meson exchange picture of leading baryon production, the leading
neutron data from ZEUS, taken together with the hadro-production data from Fermilab and
the ISR, make a statement about the normalization of σ (γ π). With the assumption that
the x and Q2 dependence of the difference between pion exchange and the backgrounds
is weak, F2π can be extracted from the data. The normalization is the firmer of the
determinations. If the normalization fixed by the photo-production data is incorrect, then
an extraction of F2π is doubtful since the self-consistency of the picture is questionable.
In particular, the x and Q2 dependence extracted from the data is questionable because
RAPID COMMUNICATION
the high Q2 region is less well understood than photo-production, both experimentally and
phenomenologically.
If the determination of F2π using the effective flux method is correct, then the ZEUS
data have important implications for the structure of the nucleon and the pion.
Acknowledgements
I thank the ZEUS Collaboration for many useful discussions and for its efforts in
publishing the inclusive leading neutron data. I thank John Martin and Malcolm Derrick
for helpful comments.
References
[1] ZEUS Collaboration, S. Chekanov et al., Nucl. Phys. B 637 (2002) 3.

[2] G. Levman, Leading Nucleon Production at HERA, in: Proceeding of the Workshop on Lepton Scattering,
Hadrons and QCD University of Adelaide March 26–April 6, 2001;
G. Levman, J. Phys. G 28 (2002) 1079.
[3] J.J. de Swart, M.C.M. Rentmeester, R.G.E. Timmermans, "N Newslett. 13 (1997) 96.
[4] T.E.O. Ericson, B. Loiseau, A.W. Thomas, Nucl. Phys. A 663 (2000) 541;
B. Loiseau, T.E.O. Ericson, A.W. Thomas, "N Newslett. 15 (1999) 162.
[5] N.N. Nikolaev, J. Speth, B.G. Zakharov, hep-ph/9708290.
[6] U. D’Alesio, H.J. Pirner, Eur. Phys. J. A 7 (2000) 109.
[7] ZEUS Collaboration, J. Breitweg et al., Eur. Phys. J. C 7 (1999) 609.
[8] H. Holtmann, G. Levman, N.N. Nikolaev, A. Szczurek, J. Speth, Phys. Lett. B 338 (1994) 363.
[9] B. Kopeliovich, B. Povh, I. Potashnikova, Z. Phys. C 73 (1996) 125.
[10] ZEUS Collaboration, S. Chekanov et al., Nucl. Phys. B 627 (2002) 3.
[11] B.G. Zakharov, V.N. Sergeev, Sov. J. Nucl. Phys. 38 (1983) 947;
B.G. Zakharov, V.N. Sergeev, Sov. J. Nucl. Phys. 39 (1984) 448.
Classical open-string field theory: A∞ -algebra,

renormalization group and boundary states
Toshio Nakatsu
Department of Physics, Graduate School of Science, Osaka University, Toyonaka, Osaka 560, Japan
Received 24 July 2001; received in revised form 29 April 2002; accepted 14 June 2002
Abstract
We investigate classical bosonic open-string field theory from the perspective of the Wilson
renormalization group of world-sheet theory. The microscopic action is identified with Witten’s
covariant cubic action and the short-distance cut-off scale is introduced by length of open-string
strip which appears in the Schwinger representation of open-string propagator. Classical open-string
field theory in the title means open-string field theory governed by a classical part of the low energy
action. It is obtained by integrating out suitable tree interactions of open-strings and is of non-
polynomial type. We study this theory by using the BV formalism. It turns out to be deeply related
with deformation theory of A∞ -algebra. We introduce renormalization group equation of this theory
and discuss it from several aspects. It is also discussed that this theory is interpreted as a boundary
open-string field theory. Closed-string BRST charge and boundary states of closed-string field theory
in the presence of open-string field play important roles.
1. Introduction
In this paper we investigate classical bosonic open-string field theory from the
perspective of the Wilson renormalization group of world-sheet theory. The microscopic
action is identified with Witten’s covariant cubic action [1] of open-string field and the
short-distance cut-off scale parameter is introduced by length of open-string strip which
appears in the Schwinger representation of open-string propagator.
E-mail address: nakatsu@het.phys.sci.osaka-u.ac.jp (T. Nakatsu).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 4 9 5 - 9
14 T. Nakatsu / Nuclear Physics B 642 (2002) 13–90
Low energy action at the cut-off scale ζ is obtained by integrating out all the
contributions of open-string interactions at length scales less than ζ . It becomes the cubic
action as ζ goes to zero, and provides a macroscopic description of open-string field
theory as ζ goes to ∞. The low energy action is of non-polynomial type similarly to
the case of closed-string field theory [2,3] and has an expansion by open-string loops.
Classical open-string field theory in the title means open-string field theory governed by
the classical part of the low energy action. It is understood to be obtained by integrating out
the tree contributions. n-valent tree interactions of open-strings beyond the cubic become
elementary at the cut-off scale ζ (> 0) and contribute to the action. Geometric structure
of these interactions can be handled by homotopy associative algebra (A∞ -algebra) [4].
Study of classical open-string field theory turns out to be deeply related with deformation
theory of A∞ -algebra which is recently developed by Fukaya et al. [5,6].
In Sections 4 and 5 we describe classical open-string field theory from the viewpoint of
the deformation theory. Differential graded algebra in the microscopic description is open-
string gauge algebra [1]. It is non-commutative associative algebra. The gauge algebra is
deformed in the macroscopic description or flows in the sense of renormalization group, to
A∞ -algebra which is non-commutative and non-associative. Our description is based on
the Batalin–Vilkovisky formalism [7]. It was elegantly used in quantization of open-string
field theory [8,9]. It also played an important role in the construction of closed-string field
theory [2].
Off-shell open-string fields at different length scales are related with one-another by
renormalization group flows. In Section 6 we introduce renormalization group equation of
classical open-string field theory. This is a rational concept since the action is the classical
part of low energy effective action. Our treatment of renormalization group is based on that
given in [10,11].
Nevertheless it may be important to explain the above idea in the context of field
theories and to make a comparison between two. For this purpose let us recall briefly
the renormalization group treatment in field theories. We consider a real scalar field
theory in d-dimensions (2 < d 4). We use the momentum representation. Let ϕp be a

momentum mode of ϕ(x) = (2π)1 d/2 dp eipx ϕp . We put ϕ ≡ {ϕp }. Let Λ be a UV cut-off.
We put ϕΛ ≡ {ϕp }|p|Λ . Let S[ϕΛ0 : Λ0 ] be an Euclidean action at Λ0 . For Λ < Λ0
the Wilsonian action S[ϕΛ : Λ] is defined in principle by the following integration on
ϕp (Λ < |p| Λ0 ) in the original action:

dϕp e−S ϕΛ0 :Λ0
= const. × e−S ϕΛ :Λ
. (1.1)
Λ<|p|Λ0
The integration on the high energy modes changes the values of the coupling constants or
generates several interactions at Λ.
It is convenient to describe the cut-off theory by using a regularized propagator ∆(p :
2 /Λ2 )
Λ). We may put ∆(p : Λ) = K(p p 2 +m2
, where K is a cut-off function and m is a bare mass.
We put S(Λ) be a set of (running) coupling constants

S(Λ) = Sn (p1 , . . . , pn : Λ) n0 , (1.2)
T. Nakatsu / Nuclear Physics B 642 (2002) 13–90 15
where Sn (p1 , . . . , pn : Λ) are symmetric functions with respect to pa . The relevant action
has the following form:

1
S[ϕ : S(Λ) : Λ] = dp ∆(p : Λ)−1 ϕp ϕ−p
2
∞ n n
1
+ dpa Sn (p1 , . . . , pn : Λ) ϕ pa
n!
n=0 a=1 a=1

1
≡ dp ∆(p : Λ)−1 ϕp ϕ−p + Sint [ϕ : Λ]. (1.3)
2
Let Z(S(Λ) : Λ) be the partition function

Dϕ e−S[ϕ:S (Λ):Λ]
Z(S(Λ) : Λ) = , Dϕ ≡ dϕp , (1.4)
Dϕ e−S[ϕ:∅:Λ] p
where the denominator is a simple Gaussian integral. The above construction of the
Wilsonian action implies
d
Λ Z(S(Λ) : Λ) = 0, (1.5)
dΛ
which states that the coupling constants flow so that the partition function is actually
independent of Λ. This equation can be brought to a perturbatively tractable form by using
the expression

δ 1
Z = exp −Sint exp dp ∆(p : Λ)jp j−p .
δj 2 j =0
After some manipulation we can rewrite equation (1.5) in the following form:

dSint 1 ∂∆(q : Λ) ∂Sint ∂Sint ∂ 2 Sint
0= Dϕ Λ − dq Λ − e−S[ϕ] .
dΛ 2 ∂Λ ∂ϕq ∂ϕ−q ∂ϕq ∂ϕ−q
(1.6)
This shows that the flow of S(Λ) needs to satisfy the Wilson–Polchinski’s renormalization
group equation [10,11]

dSint 1 ∂∆(q : Λ) ∂Sint ∂Sint ∂ 2 Sint
Λ [ϕ : S(Λ)] = dq Λ − [ϕ : S(Λ)].
dΛ 2 ∂Λ ∂ϕq ∂ϕ−q ∂ϕq ∂ϕ−q
(1.7)
In the text we regulate open-string propagator utilizing the Schwinger representation.
The counterpart in the field theory is
∞ 2 2
−s(p 2 +m2 )
exp − p Λ+m
2
∆(p : Λ) = ds e = . (1.8)
p 2 + m2
1/Λ2
Advantage in taking this regularization is that the action (1.3) becomes equivalent to the
2 +m2
following one after the similarity transformation, ϕp → exp p 2Λ 2 ϕp :

1

S[ϕ : S(Λ) : Λ] = dp p2 + m2 ϕp ϕ−p
2
∞ n
1
n pa2 + m2
+ dpa exp −
n! 2Λ2
n=0 a=1 a=1

n
× Sn (p1 , . . . , pn : Λ) ϕ pa
a=1

1
≡ dp (p2 + m2 )ϕp ϕ−p +
Sint [ϕ : S(Λ) : Λ]. (1.9)
2
Instead of the propagator the interaction vertices acquire the regularization factors (external
legs in the Feynman diagrams). The effective action of open-string field in the text is
expressed in a way analogous to this form. The renormalization group equation (1.7) can
be written down in terms of Sint as follows:

d
Sint 1 1 ∂ Sint ∂
Sint ∂ 2
Sint
Λ [ϕ : S(Λ) : Λ] = dq 2 − [ϕ : S(Λ) : Λ]
dΛ 2 Λ ∂ϕq ∂ϕ−q ∂ϕq ∂ϕ−q
∂
Sint
+Λ [ϕ : S(Λ) : Λ]. (1.10)
∂Λ
The evolution of the coupling constants Sn (p1 , . . . , pn : Λ) can be read from this equation:
∂Sn
Λ (p1 , . . . , pn : Λ)
∂Λ
− q +m
2 2
e Λ2
= dq S|I |+1 (q, PI : Λ)S|J |+1 (−q, PJ : Λ)
Λ2
I,J

− Sn+2 (q, −q, p1, . . . , pn : Λ) , (1.11)
for n = 0, 1, 2, . . . . Here I, J are the multi-indices satisfying I ∪ J = {1, . . . , n}.

Let us take a closer look on the evolution of the coupling constants. To be definite, we
start at Λ0 with the condition
4

S4 (p1 , p2 , p3 , p4 : Λ0 ) = v0 δ pa , and all other Sk vanish. (1.12)
a=1
[Note. To study the renormalizability of φ 4 -theory in d-dimensions (i.e., to search an

existence of the limit Λ0 → ∞), other coupling constants S2 (p, q : Λ0 ) = (η0 p2 +
u0 )δ d (p + q) are required. We take the above condition just to explain the underlying
idea in the text.] Solutions of the evolution equation (1.11) admit to have perturbative
expansions. At Λ = Λ0 e−ζ (ζ > 0), solution in this particular example has the expansions
as depicted in Fig. 1. The Feynman graphs are interpreted as world-lines in the figure
n-valent vertex at Λ = Λ0 e−ζ 4-valent vertex at Λ0
Fig. 1. Perturbative expansion of the solution: the Schwinger parameters si (i = 1, 2, 3) are integrated over the
region (1/Λ0 )2 < si (eζ /Λ0 )2 . The dotted lines are deleted external legs.
and the Schwinger parameter s of e−s(p +m ) is regarded as length of the world-line. The
2 2
higher vertices are generated at Λ by the contribution of graphs all the internal propagators
ζ
in which have length ( Λ10 )2 < s ( Λe 0 )2 . The appearance of loops is originated in the term
(∂ 2
Sint )/(∂ϕq ∂ϕ−q ) of the renormalization group equation. The proliferation of trees is due
to the term (∂ Sint /∂ϕq )(∂
Sint/∂ϕ−q ).
At this present stage we introduce a concept of classical renormalization group flow. It
is defined by dropping out the loop effect in the original equation:

d
Sint 1 1 ∂
Sint ∂
Sint
Λ [ϕ : S(Λ) : Λ] = dq 2
[ϕ : S(Λ) : Λ]
dΛ 2 Λ ∂ϕq ∂ϕ−q
∂
Sint
+Λ [ϕ : S(Λ) : Λ]. (1.13)
∂Λ
The evolution of the coupling constants can be read as
−q
2 +m2

∂Sn e Λ2
Λ (p1 , . . . , pn : Λ) = dq S|I |+1 (q, PI : Λ)S|J |+1 (−q, PJ : Λ),
∂Λ Λ2
I,J
(1.14)
for n = 0, 1, 2, . . . . With the same initial condition at Λ0 the evolution of the coupling
constants becomes that obtained by neglecting the loops in the previous solution.
In the case of open-string field theory we take a route which is slightly different from
the above field theory argument. This is partially because in open-string field theory there
are no coupling constants except the string field itself. It indicates that wave function
renormalizations suffice for the renormalization group argument. In fact, the classical
effective action S[Φ : ζ ], where Φ is open-string field and ζ is the cut-off scale, is given
in Section 5 and shown to satisfy the master equation, {S, S} = 0. This confirms the above
observation. Since, as one can find in the standard argument on renormalizability of gauge
field theories, the master equation ensures that the theories maintain gauge symmetries in
each step of renormalization group transformation. It is useful to point out that Eq. (1.5)
can be expressed as
dΓ
Λ [ϕ : S(Λ) : Λ] = 0, (1.15)
dΛ
where Γ is the effective action. Guided by this expression we then define the renormaliza-
tion group equation of classical open-string field theory by
dS
[Φ(ζ ) : ζ ] = 0. (1.16)
dζ
Description of renormalization group equation of string-field theory along this line was
first given for closed-string in [12]. The renormalization group equation turns out to give
an infinitesimal variation of the A∞ -algebra but its precise relation with the deformation
theory is still left unclear. Open-string field theory analogue of Eq. (1.13) is derived in
Eq. (6.7) as

dΦ ∂S[Φ : ζ ] ∂Sint [Φ : ζ ]

∂Sint [Φ : ζ ]

ω , = ω b0 + L0 Φ, . (1.17)
dζ ∂Φ ∂Φ ∂Φ
Here ω is the odd symplectic structure of open-string field theory and the ket vectors are
the hamiltonian vectors of the classical effective action and its interaction part. They are
explained in Section 5 and used to develop the Batalin–Vilkovisky formalism of open-
string field theory.
World-lines in field theories are replaced by world-sheets of strings in string field
theories. The open-string world-sheets are identified with ribbon graphs while the world-
lines are with graphs. The tree ribbon graphs are mapped to 2-disk by the Mandelstam
mapping. This cause a great difference with field theory. By this mapping the effective
action S[Φ : ζ ] acquires an interpretation in terms of two-dimensional σ -model. And the
classical renormalization group flow can be identified with renormalization group flow of
the underlying two-dimensional σ -model. This aspect is studied in the next section.
We start Section 7 with a further investigation on the renormalization group flow

by imposing the Siegel gauge condition on open-string field Φ. We express the
renormalization group equation in a form dT i /dζ = β i (T , ζ ), where T i are the
coefficients in a suitable expansion of open-string field Φ (or the spacetime variables).
The beta functions are shown in Proposition 7.2 to have the form
∂S
βi = g ij , (1.18)
∂T j
j
where gij is roughly the renormalized BPZ metric of two-dimensional field theories. Since
T i are open-string field Φ itself this expression implies that zeros of the beta functions
are nothing but classical solutions of open-string field theory. Contrary to our naive
expectation, the beta functions depend on the cut-off scale parameter ζ explicitly. This
originates in our regularization scheme of open-string field theory. Open-string field theory
is formulated using two-dimensional conformal field theory [13]. But the regularization
we choose is simply to put a restriction on length of open-string evolution, which is
a regularization of one-dimension. Actually, we have two length scales. The missing
scale is length of open-string itself. ζ is the ratio of these two length scales and is a
dimensionless parameter. In the end of Section 7 the regularization employed so far is
examined from the perspective of world-sheet boundary theories. It corresponds to a point-
splitting regularization of short-distance on the boundary when ζ is sufficiently large. The
point-splitting is prescribed by the boundary length scale ∼ e−ζ . The action S[Φ : ζ ] is
interpreted as an analogue of a generating function of all correlation function of a world-
sheet boundary theory regularized by the point-splitting method.
Our discussions in Sections 4–7 are based on a conjecture given in the end of Section 3.
The conjecture is related with construction of open-string n-valent vertices at the cut-off
scale ζ for n 3. Section 3 besides Appendix A are devoted to test the conjecture.
One of the motivations of this paper is Sen’s conjecture [14] on open-string tachyon
condensation. The conjecture has been studied by using open-string field theory in two
different formulations. One [15] is based on Witten’s covariant cubic open-string field
theory and the other [16,17] is based on the so-called boundary open-string field theory
[18]. Relation between these two treatments has not been clarified yet. In Section 8 we
investigate their relation. It is discussed that the macroscopic open-string field theory
studied in this paper is interpreted as a boundary open-string field theory. Boundary states
of closed-string field theory in the presence of open-string field and the closed-string BRST
charge play important roles in the discussion. In particular, the following equality is found:

Qc B[Φ : ζ ] = δBRS B[Φ : ζ ] . (1.19)
The ket vector is a boundary state of closed-string in the presence of open-string field Φ
and is given in Eq. (8.12). Qc is the (free) closed-string BRST charge. δBRS is the BRST
transformation of open-string field Φ which is determined by the action S[Φ : ζ ] in the
framework of the Batalin–Vilkovisky formalism.
We provide a brief review on the cubic open-string field theory in Section 2 to make the
paper self-contained as far as possible.
2. Microscopic open-string field theory
In this section we give a brief review on Witten’s covariant bosonic open-string field
theory [1]. It is described along the work of LeClair, Peskin and Preitschopf [13]. Our
goal is Definition 2.1, where the covariant action is given. This section is also devoted to a
preparation for later discussions. Basic machinery and concepts which will be used in the
subsequent sections are explained here.
Bosonic string field theory is formulated using a two-dimensional conformal field theory
(2D CFT). This conformal field theory consists of the matter sector described by Xµ (0
µ 25) and the ghost sector described by world-sheet reparametrization ghosts (b, c).
Xµ and (b, c) are respectively grassmann-even and -odd variables of string coordinates
(Xµ , b, c). The matter and ghost sectors are 2D CFTs of central charge respectively equal
to 26 and −26. The total conformal field theory therefore has central charge equal to
zero. Consider the conformal field theory formulated in the z-plane with z=eτ +iσ . Mode
expansion of the string coordinates on a unit disk |z| 1 is
α−n
µ
Xµ (z) = x µ − ipµ ln z − i zn ,
n
n=0

b(z) = b−n zn−2 , c(z) = c−n zn+1 , (2.1)
n n
where we set string length scale ls equal to one. Their first quantization gives the following
(anti-)commutation relations:
µ ν µ ν
x , p = iηµν , αn , αm = nηµν δn+m,0 ,
{bn , cm } = δn+m,0 , (2.2)
where ηµν = diag(−1, 1, . . . , 1) is the Minkowski metric of R1,25 .
For open-string, the
σ of z = eτ +iσ originally runs only over 0 σ π . The conformal field theory may be
formulated on the upper half-plane Im z 0. But we can extend the σ in the equations to
run over 0 σ 2π just as for closed-string.
2.1. Open-string Hilbert space
Open-string Hilbert space H is the tensor product Hmatter ⊗ Hghost , where Hmatter and
Hghost are, respectively, the Fock spaces of the matter and ghost CFTs. It consists of the
following vectors:
µ µ
H = b−n1 . . . b−np c−m1 . . . c−mq α−l11 . . . α−lrr |k , (2.3)
µx
where |k ≡ eik |0. We introduce the SL2 -invariant vacuum |0 by the conditions
µ
µ
αnµ |0 = 0 for n 0 α0 ≡ pµ ,
cn |0 = 0 for n 2, bn |0 = 0 for n −1. (2.4)
The open-string Hilbert space is Z-graded by ghost number G. The string coordinates have
the ghost numbers, G(Xµ ) = 0, G(b) = −1 and G(c) = 1. The vacuum state |0 is set to
have no ghost number and to be a grassmann-even vector. The BRST charge Q acts on
this Hilbert space. It is a grassmann-odd operator and has the ghost number equal to one.
The vacuum state becomes a BRST-invariant vector. The BRST charge obeys the usual
relations

Q2 = 0, Q, b(z) = T (z), (2.5)
where T is the total energy–momentum tensor.
The dual Hilbert space H∗ ≡ Hmatter ∗ ⊗ Hghost∗ consists of the following vectors:

H∗ = k|αl1 1 . . . αlr r cm1 . . . cmq bn1 . . . bnp ,
µ µ
(2.6)
where k| ≡ 0|eik xµ . We introduce the state 0| as the SL2 -invariant vacuum of H∗ by
µ
imposing the conditions:
0|αnµ = 0 for n 0,
0|cn = 0 for n −2, 0|bn = 0 for n 1. (2.7)
These conditions also make 0| a BRST-invariant state. Dual pairing between H and H∗ is
prescribed based on k |c−1 c0 c1 |k = δk +k,0 , where c±1,0 is the ghost zero modes on CP1 .
The pairing between any two vectors can be computed by using the (anti-)commutation
relations (2.2) and taking account of the conditions (2.4) and (2.7). For the consistency the
SL2 -invariant vacuum 0| must be grassmann-odd. We set 0| to have the ghost number
equal to −1.
2.2. BPZ conjugation
The Belavin–Polyakov–Zamolodchikov (BPZ) conjugation is a general operation of

two-dimensional conformal field theories. First, we explain this operation in a generic CFT
and then apply it to the case of open-string field theory.
Let U0,∞ be two-disks on CP1 , respectively, given by |z| 1 and |z| 1. For the
convenience of later application to open-string field theory, local coordinates around 0
and ∞ are chosen to be z0 ≡ z and z∞ ≡ −1/z. The coordinate transform is given by a
map I : z∞ → z0 = −1/z∞ . We have a set of local field operators on each coordinate
patch. Let O[(U0 , z0 )] and O[(U∞ , z∞ )] be sets of local field operators associated with
the patches (U0 , z0 ) and (U∞ , z∞ ). Quantum field ϕ of the theory on CP1 is considered
as a collection of ϕ (0) ∈ O[(U0 , z0 )] and ϕ (∞) ∈ O[(U∞ , z∞ )]. These two field operators
are not independent. In particular, when ϕ is primary, their relation becomes simple. It is
given by
ϕ (0) (z0 ) (dz0 )∆ = ϕ (∞) (z∞ )(dz∞ )∆ , (2.8)

where ∆ is the conformal dimension of ϕ.
We attach the Hilbert space H (representation of the Virasoro algebra) to each
coordinate patch in the operator formalism. Let H(0) and H(∞) be the Hilbert spaces
attached to (U0 , z0 ) and (U∞ , z∞ ). Consider expansions of the primary field ϕ at 0

and ∞. They are given by ϕ (0)(z0 ) = n ϕn(0) z0−n−∆ and ϕ (∞) (z∞ ) = n ϕn(∞) z∞
−n−∆ .
(0) (∞)
The coefficients ϕn and ϕn are operators which generate the building blocks of H(0)
and H(∞) . The transformation (2.8) determines a map between ϕn(0) and ϕn(∞) ,
T (0)
ϕn(∞) −→ ϕn(∞) ≡ (−)n+∆ ϕ−n . (2.9)
It can be generalized to any product of operators by letting (ϕn(∞)

1
. . . ϕn(∞) (∞) T
l ) = (ϕn1 ) · · ·
T
(ϕnl ) . This generalization induces a linear map T : H(∞) → (H(0) )∗ by

(∞) T
(∞) (∞) T
|A∞ = OA |0∞ −→ 0 AT | ≡ 0 0| OA , (2.10)
where |A is arbitrary state of H, and OA is an operator which gives |A when it acts on the
vacuum. The map (2.10) is the BPZ conjugation in the operator formalism. In particular
case of open-string field theory the BPZ conjugation becomes as follows:
µ µ
H |A = b−n1 . . . b−np c−m1 . . . c−mq α−l11 . . . α−lrr |k
µ1 T µ T
−→ AT | = k|b−n T
1
T
. . . b−n cT . . . c−m
p −m1
T
q
α−l1 · · · α−lrr ∈ H∗ , (2.11)
where the conjugations of the oscillator modes are given by
µ T µ
αn = (−)n+1 α−n , bnT = (−)n b−n , cnT = (−)n−1 c−n . (2.12)
Since the BRST current and total energy–momentum tensor are primary fields of ∆ = 1
and 2, the BPZ conjugates of the BRST charge and Virasoro generators become
QT = −Q, LTn = (−)n L−n . (2.13)
The linear map T gives a map at the level of local field operators. One-to-one
correspondence between states and local field operators is known in two-dimensional
(∞)
conformal field theories. We have a local field operator ϕA (z∞ ) for arbitrary state
(∞)
|A∞ ∈ H . The correspondence may be seen by limz∞ →0 ϕA (z∞ )|0∞ = |A∞ . For
(∞)
the BPZ conjugate 0 AT | ∈ (H(0) )∗ , we have another local field operator, which we call
I [ϕA ](0)(z0 ). The correspondence may be seen by limz0 →∞ 0 0|I [ϕA ](0)(z0 ) = 0 AT |.
Thus we obtain [13] a linear map I : O[(U∞ , z∞ )] → O[(U0 , z0 )]. (We use the same name
as the coordinate transform.) We also obtain the following commutative diagram:
T
H(∞) −→ (H(0) )∗
↓ ↓ (2.14)
I
O[(U∞ , z∞ )] −→ O[(U0 , z0 )]
The BPZ conjugation gives a non-degenerate pairing between H(∞) and H(0) , which we
call the BPZ pairing. It is defined by 0 AT |B0 for any two states |A∞ and |B0 . The BPZ
pairing is equal to the following two-point function on the z-plane:

AT |B = I [ϕA ](∞)ϕB (0) . (2.15)
In the particular case of open-string field theory, existence of the ghost zero-modes c±1,0
indicates the following selection rule:
AT |B = 0 ⇒ G(A) + G(B) = 3. (2.16)
2.3. Reflector ω12 |
Reflector ω12 | is a vector of (H⊗2 )∗ and is determined by the condition

ω12 A1 |B2 = AT |B. (2.17)
The subscripts in the LHS of the equation label the open-string Hilbert spaces appearing
in the tensor product, i.e., ω12 | ∈ (H(1) ⊗ H(2) )∗ and |A1 |B2 ∈ H(1) ⊗ H(2) . The
Hilbert spaces H(1) and H(2) are understood to be attached respectively to the coordinate
patches (U∞ , z∞ ) and (U0 , z0 ). By the condition (2.17) the reflector enjoys the following
properties:
(1) T (2)
ω12 O = ω12 (O ) (O is arbitrary operator), (2.18)

ω12 k1 = 2 k|. (2.19)
The superscripts of O in Eq. (2.18) indicate the Hilbert spaces on which the operator acts.
If we take the BRST charge Q as O in Eq. (2.18), we obtain the BRST invariance of the
reflector
(1)
ω12 Q + Q(2) = 0. (2.20)
If we regard the reflector as an element of Hom(H(1) , (H∗ )(2)), it gives the BPZ
conjugation

ω12 A1 = 2 AT |. (2.21)
This is shown by using the properties (2.18) and (2.19) as follows:
(2) (2)
ω12 A1 = ω12 OA(1)|01 = ω12 OAT |01 = ω12 01 OAT
(2)
= 2 0| OAT = 2 AT |. (2.22)
The reflector has the following oscillator representation [8]:
(2) (1) (2)
ω12 = (1)
1 k|c−1 ⊗ 2 − k|c−1 c0 + c0
k
∞

(−)n+1 µ(1) (2)
× exp αn αnµ + (−) cn b n + cn b n
n+1 (1) (2) (2) (1)
, (2.23)
n µ
n=1
µ
where the oscillators αn , bn and cn act on the bras according to their superscripts. The bra
k| is a grassmann-odd vector with the ghost number −1. Thus the reflector is a grassmann-
odd vector and has the ghost number equal to one.
As usual in the description of many body system, we identify H(1) ⊗ H(2) with
H ⊗ H(1) by imposing the relation
(2)
|A1 |B2 = (−)G(A)G(B)|B2 |A1 . (2.24)

This identification is used implicitly throughout this paper. We can see that the reflector is
symmetric under the exchange of open-string indices:

ω12 = ω21 . (2.25)
This actually means the identity, ω12 |A1 B2 = ω21 |A1 |B2 for any two states. By the
equivalence relation (2.24) we can identify ω21 |A1 |B2 with (−)G(A)G(B)ω21 |B2 |A1 .
This turns out to be (−)G(A)G(B)B T |A. Thus Eq. (2.25) means AT |B = (−)G(A)G(B)
B T |A.
2.4. Inverse reflector |S12
The inverse reflector |S12 ∈ H⊗2 is introduced as the BPZ-conjugate of the reflector. It
enjoys the same properties as (2.18) and (2.19),

O (1) S12 = (O T )(2) S12 (O is arbitrary operator), (2.26)

1 k S12 = |k2 . (2.27)
The first equation implies the BRST invariance of the inverse reflector
(1)
Q + Q(2) S12 = 0. (2.28)
If we regard the inverse reflector as an element of Hom((H∗ )(1) , H(2) ), it gives

T
1 A S12 = (−) |A2 .
G(A)
(2.29)
This can be shown by using (2.26) and (2.27) similarly to the case of Eq. (2.21).
Oscillator representation of the inverse reflector becomes
∞
(−)n+1 µ(1) (2) (1) (2) (2) (1)
S12 = exp α−n α−nµ + (−)n c−n b−n + c−n b−n
n µ
n=1
(1) (2) (1)
× c0 + c0 c1 |k1 ⊗ c1(2)| − k2 . (2.30)
k
If one takes the conjugation of Eq. (2.30), one finds that it becomes the representation
(2.23). The ket |k is a grassmann-even vector with no ghost number. Hence the inverse
reflector is a grassmann-odd vector and has the ghost number equal to three. We also see
that it is anti-symmetric under the exchange of open-string indices:

S12 = − S21 . (2.31)
Let us explain why the BPZ-conjugate of the reflector is called inverse reflector. Take
arbitrary state A ∈ H(1) . We examine the state ω12 |S23 |A1 ∈ H(3) . It turns out to be
A ∈ H(3) by Eqs. (2.21) and (2.29) as follows:

ω12 S23 |A1 = (−)G(A) ω12 A1 S23

= (−)G(A) 2 AT S23
= |A3 . (2.32)
If one regards ω12 |S23 as an element of Hom(H(1) , H(3)), it can be expressed in a
convenient form

ω12 S23 = 3P1 , (2.33)
where 3P1 is the identity which maps |A1 to |A3 .
√ of fi (Vi ) on the z-plane. They exactly

Fig. 2. Images z-plane. Their boundaries are the bold solid lines.
√ cover the √
A, B = ±1/ 3; C± = ±i; f1 (0) = 0, f2 (0) = − 3, f3 (0) = 3.
2.5. Symmetric 3-vertex 1 2 3|
Let Vi for i = 1, 2, 3 be unit disks |vi | 1 on the vi -plane. Let fi for i = 1, 2, 3 be

holomorphic maps from Vi to the z-plane of the following forms:
f1 = F, f2 = S ◦ F, f3 = S ◦ S ◦ F, (2.34)
where F and S are meromorphic functions given by
1+iv 2/3 √
1− 1−iv 2z − 2
1 3
F (v) = i 1+iv 2/3 , S(z) = √ . (2.35)
1 + 1−iv 2 z+ 2
3 1
F (v) has branch points at v = i, ∞. For the description we need three Riemann sheets. It
is understood in (2.35) that the cut of F is taken so that unit disk |v| 1 is on a single
1/2 −√3/2
sheet. S(z) is the projective action of √ . {1, S, S 2 } is a Z3 subgroup of SL2 (R).
3/2 1/2
The images fi (Vi ) are depicted in Fig. 2. In open-string field theory it may be convenient
to use the w- and u-planes instead of the z- and v-planes by
1 + iz 1 + iv
z → w = , v → u = . (2.36)
1 − iz 1 − iv
The unit disks and upper half-planes are mapped respectively to the right half-planes and
unit disks by these maps. The unit disk Vi on the vi -plane is mapped to the right half-plane
on the ui -plane. If one regards fi as holomorphic maps from the right half-plane (on the
Fig. 3. Images of fi (Vi ) on the w-plane. They exactly cover the w-plane. Their boundaries are the bold solid
lines. A, B = e±(π i)/3 ; f1 (0) = 1, f2 (0) = e2π i/3 , f3 (0) = e4π i/3 .
ui -plane) to the w-plane, F and S in (2.35) are given by 1

2π i
F (u) = u2/3 , S(w) = e− 3 w. (2.37)
The images fi (Vi ) on the w-plane are depicted in Fig. 3.
Let H(i) for i = 1, 2, 3 be the open-string Hilbert spaces attached to the coordinate
patches (Vi , vi ). Open-string trivalent vertex 1 2 3| is a vector of (H(1) ⊗ H(2) ⊗ H(3) )∗ .
We formulate this vector along the line given in [13]. Let U and V be, respectively,
unit disks |z| 1 and |v| 1. Let O[(U, z)] and O[(V , v)] be the sets of local field
operators associated with the coordinates patches (U, z) and (V , v). Let H(U,z) and H(V ,v)
be the Hilbert spaces attached to these coordinates patches. We regard F in (2.35) as a
holomorphic map from V to the z-plane. This holomorphic map induces a linear map
from H(V ,v) to H(U,z) . Since F is biholomorphic at F (0) = 0, the vacuum state of H(V ,v)
should be mapped to the vacuum state of H(U,z) . For a primary field ϕ, the relation between
ϕ U (z) ∈ O[(U, z)] and ϕ V (v) ∈ O[(V , v)] is given by
ϕ U (z)(dz)∆ = ϕ V (v)(dv)∆ . (2.38)

Analogously to the map T in the BPZ conjugation, this relation is used to introduce a
linear map between the operators acting on H(U,z) and H(V ,v) . We write it as O (V ,v) →
F [O](U,z) . Thus we obtain the induced map, which we also call F , as follows:
F
H(V ,v) −→ H(U,z)
(V ,v)
(2.39)
|A(V ,v) = OA |0(V ,v) −→ |F [A](U,z) ≡ F [OA ](U,z) |0(U,z).
1 F in (2.37) shows that the cut can be taken along the negative real line in the u-plane. In the v-plane it is the
line on the imaginary axis starting at i.
Fig. 4. Witten’s trivalent vertex. The indices 1, 2 and 3 label three open-strings.
Using the correspondence between states and local field operators, the induced map
(2.39) is equivalently described as a linear map between O[(V , v)] and O[(U, z)],
F
O[(V , v)] −→ O[(U, z)] (2.40)
V
ϕA −→ F [ϕA ]U ≡ ϕFU[A] .
Let Ui for i = 1, 2, 3 be unit disks on the z-plane centered, respectively, at z = S i−1 (0).
We put zi ≡ z − S i−1 (0). Let O[(Ui , zi )] be the sets of local field operators associated
with the coordinate patches (Ui , zi ). We regard S 1−i as a holomorphic map from Ui to U .
Following the same argument as above, we obtain linear maps S 1−i between O[(Ui , zi )]
and O[(U, z)],
S 1−i
O[(Ui , zi )] −→ O[(U, z)]
U (2.41)
ϕA i −→ S 1−i [ϕA ]U ≡ ϕSU1−i [A] .
We can now introduce open-string trivalent vertex. For each holomorphic map fi in
(2.34), we consider S 1−i F , which is a combination of the maps (2.40) and (2.41). It is
understood as a map from O[(Vi , vi )] to O[(U, z)] as follows:
F S 1−i
O[(Vi , vi )] −→ O[(Ui , zi )] −→ O[(U, z)]. (2.42)
The trivalent vertex 1 2 3| is given [13] by
√ √
1 2 3 A1 |B2 |C3 = F [ϕA ](0)S 2 F [ϕB ] − 3 SF [ϕC ] 3 , (2.43)
where the RHS is the three-point function on the z-plane. One often depicts the trivalent
vertex as Fig. 4.
The most significant property of the trivalent vertex is the BRST invariance,

1 2 3 Q(1) + Q(2) + Q(3) = 0. (2.44)
Let us show Eq. (2.44). We first compute 1 2 3|(Q(1) |A1 )|B2 |C3 as follows:

1 2 3 Q(1) |A1 |B2 |C3 = F [ϕQA ]S 2 F [ϕB ] SF [ϕC ]

= F [δBRS ϕA ]S 2 F [ϕB ]SF [ϕC ]

= δBRS F [ϕA ] S 2 F [ϕB ]SF [ϕC ] , (2.45)
where δBRS ϕA is the BRST transform of ϕA and is equal to ϕQA . We also use the relation
F [δBRS ϕ] = δBRS F [ϕ]. It is ensured by the scalar property of the BRST transformation.
If we take the Hamiltonian interpretation of the correlation function and use the BRST
invariance of the vacua, we can further compute Eq. (2.45) as follows:

δBRS F [ϕA ] S 2 F [ϕB ]SF [ϕC ]

= 0|δBRS F [ϕA ] S 2 F [ϕB ]SF [ϕC ]|0

= (−)G(A)+10|F [ϕA ]δBRS S 2 F [ϕB ] F S[ϕC ]|0

+ (−)G(A)+G(B)+10|F [ϕA ]S 2 F [ϕB ]δBRS SF [ϕC ] |0

= (−)G(A)+1 F [ϕA ]S 2 F [ϕQB ]SF [ϕC ]

+ (−)G(A)+G(B)+1 F [ϕA ]S 2 F [ϕB ]SF [ϕQC ] , (2.46)
which turns out to be (−)1 2 3|(Q(2) + Q(3) )|A1 |B2 |C3 . Thus we obtain the BRST
invariance (2.44) of the 3-vertex.
The oscillator representation of the 3-vertex has the form

1 2 3 =
(1) (1) (2) (2) (3) (3)
1 k1 |c−1 c0 ⊗ 2 k2 |c−1 c0 ⊗ 3 k3 |c−1 c0
3
i ki =0
∞
1
3
ij (j )
× exp (−)n+m Nnm αnµ(i) αmµ
2
i,j =1 n,m=0
µ
∞

3 ij (j )
+ (−) n+m+1
Nc nm bn(i)cm , (2.47)
i,j =1 n0,m1
ij ij
where Nnm and Nc nm are the Neumann coefficients given by the Fourier components
of the two-points functions of X and (b, c) in a suitable coordinate. Their explicit forms
can be found in [13,19]. The oscillator representation clearly shows that the 3-vertex is a
grassmann-odd vector and has the ghost number equal to three. In particular, it is symmetric
under the cyclic permutation of open-string indices,

1 2 3 = 2 3 1 . (2.48)
This actually means 1 2 3|A1 |B2 |C3 = 2 3 1|A1 |B2 |C3 for any three states. The
RHS of this equation is identified with (−)G(A)(G(B)+G(C)) 2 3 1|B2 |C3 |A1 . It is equal
to (−)G(A)(G(B)+G(C))1 2 3|B1 |C2 |A3 . Thus Eq. (2.48) means

1 2 3 |A1 |B2 |C3 − (−)G(A)(G(B)+G(C))|B1 |C2 |A3 = 0 (2.49)
for any three states.
2.6. Witten’s Φ 3 -action of open-string field theory
Open-string field Φ is a vector of the open-string Hilbert space H. It is grassmann-

odd and has the ghost number G(Φ) = 1. The cubic action of open-string field theory is
defined by
Definition 2.1 (Cubic action of open-string field theory).

1 1
S cubic [Φ] = ω12 Φ1 Q(2) |Φ2 + 1 2 3 Φ1 |Φ2 |Φ3 . (2.50)
2 3
Selection rules by the ghost number conservation imply that each term in the action
does not vanish for a generic Φ.
Guiding principle for the construction [1] of the above action was open-string gauge
algebra (Q, D). Here Q is the BRST charge. D is a non-commutative associative algebra on
H,

|A D B1 ≡ 0 2 3 S01 |A2 |B3 , (2.51)
where D is understood as a map from H(2) × H(3) to H(1) . The ghost number of the
RHS of Eq. (2.51) is equal to G(A) + G(B). Hence D preserves the ghost number,
G(A D B) = G(A) + G(B). Non-commutativity, A D B = B D A, clearly has its origin
in the use of the 3-vertex. Associativity of the algebra,
(A D B) D C = A D (B D C), (2.52)
follows from the GGRT theorem [13,20]. It states in this particular case that 4-vertex
1 2 3 4|, which is introduced by 1 2 3 4| ≡ 1 2 a|a 3 4|Sa a , becomes symmetric
under the cyclic rotation (1, 2, 3, 4) → (4, 1, 2, 3). Actually one can evaluate the RHS of
Eq. (2.52) as follows:

|A D (B D C)5 = 1 2 3 S15 |A2 |B D C3

= 1 2 3 S15 |A2 4 6 7 S43 |B6 |C7

= 1 2 3 4 6 7 S43 S15 |A2 |B6 |C7

= 1 2 6 7 S15 |A2 |B6 |C7

= 1 2 3 4 S15 |A2 |B3 |C4 .
A similar computation shows |(A D B) D C5 = 4 1 2 3|S15 |A2 |B3 |C4 . Due to the cyclic
symmetry of the above 4-vertex these two become equal. The BRST charge is another
important constituent of the gauge algebra. It is nilpotent and acts as the derivation on the
D-algebra,
Q(A D B) = QA D B + (−)G(A) A D QB. (2.53)
This can be seen from the BRST invariance of the 3-vertex as follows:

Q(1) |A D B1 = Q(1) 0 2 3 S01 |A2 |B3

= 0 2 3 (−)Q(1) S01 |A2 |B3

= 0 2 3 Q(0) S01 |A2 |B3

= 0 2 3 (−) Q(2) + Q(3) S01 |A2 |B3

= 0 2 3 S01 Q(2) |A2 |B3 + (−)G(A) 0 2 3 S01 |A2 Q(3) |B3
= |QA D B1 + (−)G(A) |A D QB1 . (2.54)
The action (2.50) was obtained [1] originally in a form by which role of the open-string
gauge algebra becomes manifest. Let us rewrite the cubic term in the action as follows:

1 2 3 Φ1 |Φ2 |Φ3 = 1 2 3 1 Pa |Φa |Φ2 |Φ3

= 1 2 3 ωab Sb1 |Φa |Φ2 |Φ3

= ωab Φa 1 2 3 S1b |Φ2 |Φ3

= ω12 Φ1 |Φ D Φ2 . (2.55)
By using this expression of the cubic term we can write down the action (2.50) into the
following form:

1 2
S cubic
[Φ] = ω12 Φ1 Q |Φ2 + |Φ D Φ2 .
(2)
(2.56)
2 3
Analogy with the Chern–Simons gauge theory is very suggestive. The action enjoys the
following gauge symmetry:
δρ Φ = Qρ + Φ D ρ − ρ D Φ, (2.57)
where ρ is arbitrary vector of H with G(ρ) = 0. The invariance of the action can be shown
by using the remarks noted in the previous paragraph.
3. Low energy theory vertices
Let us consider a second-quantization of open-string field theory. Schematically it is

performed by the path-integral,

DΦ exp −S cubic [Φ] . (3.1)
The path-integral may be computed by treating the cubic term as a perturbation. All
connected vacuum Feynman graphs contribute to the path-integral. Vacuum Feynman
graph consists of the trivalent vertices connected by the open-string propagators without
any external strings. The propagator is determined from the quadratic part of the action
(2.50). In the Siegel gauge b0 Φ = 0 it formally turns out to be b0 L10 . This causes the short-
and long-distance divergences as world-sheet theory. In this paper we hope to present a
renormalization group analysis for the short-distance divergence.
It is convenient to use the Schwinger representation of the propagator,
∞
1
b0 = dτ b0 exp −τ L0 . (3.2)
L0
0
One can interpret e−τ L0 as the evolution operator for open-string. It is realized by a strip
of length τ (Fig. 5). Open-string diagrams are metrized trivalent ribbon graphs. Metric of
the graph is given by the (imaginary time) parameters τ . To control the short-distance
divergence we introduce the cut-off scale parameter ζ (> 0) and use the regularized
Fig. 5. Open-string strip. τ (> 0) is the Schwinger imaginary time.

The width of open-string is set to π .
propagator given by
∞
1 reg
b0 = dτ b0 exp −τ L0 . (3.3)
L0
2ζ
In presence of the cut-off scale ζ , any open-string diagram which has at least one
internal strip of length less than 2ζ cannot be reproduced from the trivalent vertices by
using the regularized propagator. This is because τ in Eq. (3.3) is greater than 2ζ . Any
internal strip, which connects the trivalent vertices, must have length greater than 2ζ . On
the other hand, when all the internal strips have length more than 2ζ , it can be reproduced
from the trivalent vertices in this cut-off theory (Fig. 6). If we wish to take account of all
the diagrams even in presence of the cut-off scale ζ , we need to introduce an action which
contains higher interactions beyond the cubic term. These interactions are obtained from
graphs which are one-particle irreducible with respect to the regularized propagator (3.3).
They are those graphs all internal strips of which have length less than 2ζ . If we take
the perspective of renormalization group a la Wilson [10] or Polchinski [11], the action
is nothing but a low energy action obtained by integrating out all the contributions from
ζ
length scale less than ζ . We denote the low energy action at the cut-off scale ζ by Seff [Φ].
It becomes the microscopic action S cubic [Φ] as ζ goes to 0, while, as ζ goes to ∞, it
provides a macroscopic description.
Contribution from open-string loops might play an important role in the low energy
description, particularly related with duality between open- and closed-strings [21].
Nevertheless, at the present stage, we do not have a systematic machinery enough to handle
ζ
with it. In this paper we study the classical part of Seff [Φ] which does not include the loop
effect.
3.1. Low energy theory vertices; examples
Since our consideration is limited to the classical case, we call connected open-string
diagrams which are tree simply as open-string diagrams. External open-strings of the
(a) (b)
Fig. 6. An open-string diagram which contains (a) cannot be reproduced from the trivalent vertices at the cut-off
scale ζ . On the other hand, when all internal strips are (b) the diagram can be reproduced.
diagram are clockwise-ordered. Open-string diagram is a metrized connected trivalent tree

ribbon graph with clockwise-ordered external ribbons. Open-string n-diagram is a metrized
connected trivalent tree ribbon graph with clockwise-ordered n external ribbons.
Let M∂n be the set of clockwise-ordered n different points (z1 , . . . , zn ) on the boundary
of two-disk D (or the upper half-plane) divided by the standard action2 of SL2 (R) (n 3).
The dimension of M∂n is n − 3. The set of metrized connected trivalent tree ribbon graphs
with clockwise-ordered n external ribbons is identified with M∂n . Each ribbon graph has
n − 3 internal strips. Their length play the role of local coordinates of M∂n . Infinities of
M∂n are the configurations in which some of zi coincide with one another on ∂D. Stable
compactification of M∂n is given, under the above identification, by adding the trivalent
tree ribbon graphs at least one internal strip of which has infinite length. We denote this
compactification by CM∂n . Topologically CM∂n becomes a (n − 3)-dimensional ball Bn−3 .
We fix orientation of CM∂n by the standard orientation of Bn−3 .
If we have an open-string n-diagram, we can obtain another one by permuting the
labels of external open-strings from (1, 2, . . . , n − 1, n) to (2, 3, . . . , n, 1). In general, it
is different from the original diagram. This permutation of open-string indices gives an
automorphism of CM∂n . We denote this automorphism by s. Clearly, s generates Zn (or its
subgroup) action on CM∂n . As we will see in the subsequent discussions, the automorphism
s is related with asymmetry of open-string vertex under the cyclic-permutation of open-
string indices.

2 (z , . . . , z ) → az1 +b , . . . , azn +b , where a b ∈ SL (R).
1 n cz1 +d czn +d cd 2
Fig. 7. CM∂4 = [−∞, +∞]. The configurations of four points on ∂D at x = 0, ±∞ are depicted in the second
line. Open-string diagram at x ∈ [−∞, +∞] is on the third line. The s, t-channels are, respectively, x > 0 and
x < 0.
3.1.1. Open-string 4-vertex 1 2 3 4 : ζ |

The compactification CM∂4 is [−∞, ∞]. We identify the s-channel of four open-string
interaction with [0, ∞] and the t-channel with [−∞, 0] (Fig. 7). Length of the internal
strip is x in the s-channel while it is |x| in the t-channel.
We can construct a state x| ∈ (H⊗4 )∗ for each x ∈ CM∂4 by applying the Feynman rule
to the corresponding diagram. Explicitly we define x|(1234) as follows:

 1 2 a a 3 4 b0 e−xL0 Sa a for x > 0,
x|(1234) = 0 for x = 0, (3.4)

2 3 a a 4 1 (−b0 )exL0 Sa a for x < 0.
We need not specify insertions of b0 and L0 on the inverse reflector |Sa a because either
way gives the same by the BPZ conjugation. The reason why we attach the superscript
(1234) to the state x| will become clear soon. The state x|(1234) is grassmann-even and
has the ghost number four. We have a (H⊗4 )∗ -valued one-form Ω|(1234) on CM∂4 . It is
defined by
Ω(x)(1234) = dx x|(1234). (3.5)
Let us consider an effect of the cyclic permutation of open-string indices. Another state
can be obtained if we permute the open-string indices from (1, 2, 3, 4) to (2, 3, 4, 1) in
the RHS of Eq. (3.4). We denote this new state by x|(2341). We also let Ω|(2341) be the
one-form dx x|(2341). Two states x|(1234) and x|(2341) are related with each other by
−x|(1234) = −x|(2341). This follows from the anti-symmetry of the inverse reflector and
the cyclic symmetry of the trivalent vertex besides its odd-grassmannity. The permutation
exchanges the s- and t-channels. Thus the automorphism s generates a Z2 -action on CM∂4 .
It is given by s(x) = −x. Therefore, Ω|(2341) is the pull-back of Ω|(1234) by s:
Ω (2341) = s ∗ Ω (1234). (3.6)
We now examine the low energy description at the cut-off scale ζ . We put V4 (ζ ) ≡
(−2ζ, 2ζ ). Open-string diagrams which belong to V4 (ζ ) cannot be reproduced from the
ζ
trivalent vertices at this length scale. Any state x| for x ∈ V4 (ζ ) must appear in Seff [Φ] as
2ζ
a part of an interaction vertex. It will be a form such as 14 −2ζ Ω|(1234)|Φ1 |Φ2 |Φ3 |Φ4 .
To specify the correct form, we first consider the states ±2ζ |. They are expected to
be obtained simply by joining two 3-vertices at one open-string, since x = ±2ζ are the
boundaries of V4 (ζ ). For this to be possible we need to modify slightly the trivalent vertex.
Definition 3.1 (Open-string 3-vertex at ζ ). Open-string 3-vertex at the cut-off scale ζ is

defined by
−ζ L(i)
1 2 3 : ζ = 1 2 3 e 0 . (3.7)
i=1,2,3
The modified 3-vertex (Fig. 8) is a grassmann-odd vector which satisfies the BRST
invariance and the cyclic symmetry in the same manner as the original one. In (3.7) we
attach the propagator of length ζ to each (not a specific) external open-string in order to
keep the cyclic symmetry of 3-vertex. States obtained by joining two modified 3-vertices
(i)
are not ±2ζ | but ±2ζ | 4i=1 e−ζ L0 . Appearance of the propagators of length ζ is a
common phenomenon in the low energy description. Let Vn (ζ ) be the set of open-string
n-diagrams all internal strips of which have length less than 2ζ . They are by no means
obtained from lower diagrams at the cut-off scale ζ . Open-string diagram at the boundary
of Vn (ζ ) has at least one internal strip of length equal to 2ζ . Let us take any two open-
string diagrams which belong, respectively, to Vn1 +1 (ζ ) and Vn2 +1 (ζ ), and join them at
one external string. As the result we obtain a (n1 + n2 )-diagram. It must be at the boundary
of Vn1 +n2 (ζ ). For this to be possible, each external strip needs to have length ζ . This shows
that we must insert e−ζ L0 to each external open-string in order to obtain the vertices at the
cut-off scale ζ .
Fig. 8. Open-string trivalent vertex at the cut-off scale ζ .

defined by an integration of the (H⊗4 )∗ -valued one-form Ω| (3.5) over V4 (ζ ) multiplied
by the external propagators of length ζ ,
2ζ
4
(i)
1 2 3 4 : ζ = Ω|(1234) e−ζ L0 . (3.8)
−2ζ i=1
Indices in the 4-vertex are understood to label the Hilbert spaces attached to clockwise-
ordered four open-strings on ∂D. Thereby the vertex is regarded as a vector of (H(1) ⊗
H(2) ⊗ H(3) ⊗ H(4))∗ .
Open-string 4-vertex (3.8) is grassmann-even and has the ghost number equal to four.
When we use the one-form Ω (2341) instead of Ω (1234) in the above definition, we obtain
another 4-vertex 2 3 4 1 : ζ |. But, owing to the relation (3.6), these two are shown to be
identical:
2ζ 2ζ
4
(i)
4
(i)
2 3 4 1 : ζ = Ω| (2341)
e −ζ L0
= s ∗ Ω|(1234) e−ζ L0
−2ζ i=1 −2ζ i=1
2ζ
4
(i)
=− Ω|(1234) e−ζ L0 = − 1 2 3 4 : ζ . (3.9)
−2ζ i=1
We say this as asymmetry of the open-string 4-vertex under the cyclic permutation. The
factor (−) in (3.9) originates in the fact that the automorphism s does not preserve the
orientation of V4 (ζ ).
The 4-vertex is not invariant under the action of the BRST charge. Instead we have:
Proposition 3.1 (Q-action on 4-vertex). Action of the BRST charge on the 4-vertex (3.8)
becomes as follows:
4

1234:ζ Q (i)
= 1 2 a : ζ a 3 4 : ζ Sa a − 2 3 a : ζ a 4 1 : ζ Sa a .
i=1
(3.10)
Eq. (3.10) has the following geometrical interpretation. Orientation of V4 (ζ ) is given, as

used in Eq. (3.8), by the orientation of CM∂4 = [−∞, ∞]. Therefore, we have
∂V4 (ζ ) = {2ζ } − {−2ζ }, (3.11)

∂{±2ζ } = 0, (3.12)
where ∂ is the boundary operator. Similarity between Eqs. (3.10) and (3.11) including
their signatures should be emphasized. Each term in the RHS of Eq. (3.10) is BRST-
closed as follows from the BRST invariances of the 3-vertex and the inverse reflector.
This corresponds to Eq. (3.12). The action of the BRST charge on the 4-vertex becomes a
representation of the boundary operator ∂.
The 4-vertex depends on the cut-off scale parameter ζ . It vanishes at ζ = 0. The scale
(i)
dependence comes from V4 (ζ ) and e−ζ L0 in Eq. (3.8).
Proposition 3.2 (Scale dependence of 4-vertex). Scale dependence of the 4-vertex (3.8) is
described by
1 d
1 2 3 4 : ζ = 1 2 a : ζ a 3 4 : ζ b0 Sa a − 2 3 a : ζ a 4 1 : ζ b0 Sa a
2 dζ
4
1
− 1 2 3 4 : ζ
(i)
L0 . (3.13)
2
i=1
Now we prove Propositions 3.1 and 3.2. The following lemma becomes useful in the
proof of Proposition 3.1.
Lemma 3.1. The BRST charge acts on the state x|(1234) as

4
∂x 1 2 a a 3 4 e−xL0 Sa a for x > 0,
x| (1234)
Q(i)
= (3.14)
i=1 ∂x 2 3 a a 4 1 exL0 Sa a for x < 0.
Proof of Lemma 3.1. We show Eq. (3.14) for the case of x > 0. We rewrite the LHS of
Eq. (3.14) as follows:

x|(1234) Q(i)
i

(i) −xL
= 1 2 a a 3 4 Q b0 e 0 S
aa
i

= − 1 2 a Q(1) + Q(2) a 3 4 + 1 2 a a 3 4 Q(3) + Q(4)

× b0 e−xL0 Sa a . (3.15)
Eq. (3.15) can be further computed by using the BRST invariance of the 3-vertices:

Eq. (3.15) = 1 2 a Q(a) a 3 4 − 1 2 a a 3 4 Q(a ) b0 e−xL0 Sa a

= − 1 2 a a 3 4 Q(a) + Q(a ) b0 e−xL0 Sa a . (3.16)
Then, by using the BRST invariance of the inverse reflector and the anti-commutation
relation, {Q, b0 } = L0 , Eq. (3.16) becomes as follows:

Eq. (3.16) = − 1 2 a a 3 4 L0 e−xL0 Sa a

= ∂x 1 2 a a 3 4 e−xL0 Sa a . (3.17)
This is the RHS of Eq. (3.14) for the case of x 0.
Proof of Proposition 3.1. The proof is done by a bunch of simple calculations. We

compute the LHS of Eq. (3.10) by using Lemma 3.1.

(i)
1 2 3 4 : ζ Q
i
2ζ 0
(i) (i)
−ζ L0
= Ω| (1234)
Q(j )
e + Ω|(1234) Q(j ) e−ζ L0
0 j j −2ζ j j
−ζ L(i)
= 1 2 a a 3 4 e−2ζ L0 Sa a − 1 2 a a 3 4 Sa a e 0
i
−ζ L(i)
+ 2 3 a a 4 1 Sa a − 2 3 a a 4 1 e−2ζ L0 Sa a e 0 . (3.18)
i
Then we arrange Eq. (3.18) as follows:

Eq. (3.18) = 1 2 a : ζ a 3 4 : ζ Sa a − 2 3 a : ζ a 4 1 : ζ Sa a
−ζ L(i)
− 1 2 a a 3 4 Sa a − 2 3 a a 4 1 Sa a e 0
i

= 1 2 a : ζ a 3 4 : ζ Sa a − 2 3 a : ζ a 4 1 : ζ Sa a
−ζ L(i)
− 1 2 3 4 − 2 3 4 1 e 0 . (3.19)
i
The last term vanishes identically because of the cyclic symmetry of 1 2 3 4| [13]. (We
remind the reader that 1 2 3 4| and our 4-vertex 1 2 3 4 : ζ | are different.) Therefore, we
obtain the RHS of Eq. (3.10).
Proof of Proposition 3.2. The first-order variation of the 4-vertex with respect to the cut-
off scale ζ can be computed as follows:
! "

4
(i)
δ 1 2 3 4 : ζ = δ Ω| (1234)
e −ζ L 0
V4 (ζ ) i=1
! " ! "

4
(i)
4
(i)
−ζ L0 −ζ L0
=δ Ω| (1234)
e + Ω| (1234)
δ e
V4 (ζ ) i=1 V4 (ζ ) i=1

4
(i)
= 2δζ 2ζ | (1234)
+ −2ζ | (1234)
e−ζ L0
i=1
4
1
− 1234:ζ (i)
L0 . (3.20)
2
i=1
It is easy to see that Eq. (3.20) precisely gives the RHS of Eq. (3.13).
3.1.2. Open-string 5-vertex 1 2 3 4 5 : ζ |

The compactification CM∂5 is a two-disk. We fix an identification of CM∂5 with the
set of open-string 5-diagrams as follows. Let#Ui be the fan yi |xi | 0 on the two-plane
(xi , yi ) for 1 i 5. CM∂5 is understood as 5i=1 Ui . See Fig. 9. For each (xi , yi ) ∈ Ui we
associate an open-string diagram as given in Fig. 10. Open-string indices in the diagrams,
say 2i, are understood modulo 5. Length of the two internal strips are given by xi (or
|xi |) and yi . Diagrams at the boundary xi = yi of Ui coincide with those at the boundary
xi+1 = −yi+1 of Ui+1 , while diagrams at the other boundary xi = −yi coincide with
those at the boundary xi−1 = yi−1 of Ui−1 . As the result we can joint these Ui together
in the clockwise order along their boundaries so that they are consistent with the open-
string diagrams. Hence this gives an identification of CM∂5 with the set of open-string
5-diagrams.
We have a state I| ∈ (H⊗5 )∗ for each I ∈ CM∂5 . It is given by a patch-wise
∈ (H⊗5 )∗ for (xi , yi ) ∈ Ui by applying
(12345)
construction. We introduce a state (xi , yi )|i
the Feynman rule to the corresponding open-string diagram:
(12345)
(xi , yi )|i

 2i − 1 2i a b 2i + 1 2i + 2 a b 2i + 3

 −y L −x L

 i 0 i 0 S
 × b0 e Sa a b0 e
 bb for yi xi > 0,

= 2i − 1 2i a a 2i + 1 b b 2i + 2 2i + 3 (3.21)





 × b0 e−yi L0 Sa a (−)b0 exi L0 Sb b for yi −xi > 0,

0 for xi = 0, yi 0.
This state is grassmann-odd and has the ghost number equal to five. We then de-
fine a (H⊗5 )∗ -valued two-form Ω|(12345). We put it equal to a two-form dxi ∧
(a) (b)
Fig. 9. (a) Ui is given by the shaded region on the (xi , yi )-plane. (b) The compactification CM∂5 . It is understood
#
as a union 5i=1 Ui . The arrow denotes the orientation of CM∂5 .
(a) (b)
Fig. 10. Open-string diagram at (xi , yi ) ∈ Ui .
dyi (xi , yi )|(12345)

i on each Ui ,
(12345)
Ω (12345) U = dxi ∧ dyi (xi , yi ) i . (3.22)
i
We examine an effect of the cyclic permutation of open-string indices. We can obtain
another state by permuting the open-string indices from (1, 2, 3, 4, 5) to (2, 3, 4, 5, 1)
(a) (b)
#5
Fig. 11. (a) Ui (ζ ) is the shaded region of Ui . (b) V5 (ζ ) is a union i=1 Ui (ζ ). It becomes the shaded region of
CM∂5 = B2 . The arrow denotes the orientation of V5 (ζ ).
(23451)
in the RHS of Eq. (3.21). We denote this new state by (xi , yi )|i . We call the
(23451)
corresponding two-form Ω| (23451) . As one can easily check, the state (xi , yi )|i turns
(12345)
out to be (xi , yi )|i+3 . This permutation generates a Z5 -action on CM∂5 . It is given by
s(Ui ) = Ui+3 . Therefore, Ω|(23451) is the pull-back of Ω|(12345) by s:
Ω (23451) = s ∗ Ω (12345). (3.23)

We put Ui (ζ ) ≡ {(xi , yi ) ∈ Ui | yi < 2ζ }. Any diagram belonging # to Ui (ζ ) cannot be
reproduced from lower diagrams at the scale ζ . We put V5 (ζ ) ≡ 5i=1 Ui (ζ ) (Fig. 11).
ζ
The state (xi , yi )|(12345)
i must appear in Seff [Φ] as a part of an interaction vertex for any
(xi , yi ) ∈ V5 (ζ ). We then define open-string 5-vertex at the scale ζ as follows:

defined by an integration of the (H⊗5 )∗ -valued two-form Ω| (3.22) over V5 (ζ ) multiplied
by the external propagators of length ζ :

5
(k)
1 2 3 4 5 : ζ = Ω|(12345) e−ζ L0 . (3.24)
V5 (ζ ) k=1
Indices of the vertex are understood to label the Hilbert spaces H(i) attached to clockwise-
ordered five open-strings on ∂D. Thereby the vertex is regarded as a vector of (H(1) ⊗
· · · ⊗ H(5) )∗ .
The 5-vertex is grassmann-odd and has the ghost number equal to five. Another 5-
valent vertex might be obtained by using the two-form Ω (23451) instead of Ω (12345) in the
Fig. 12. A typical configuration of five points on ∂D which belongs to V(k k+1 k+2 a) (ζ ) × V(a k+3 k+4) (ζ )
(⊂ ∂V5 (ζ )). The solid line connecting two points a and a represents open-string strip of length equal to 2ζ .
definition. We denote it by 2 3 4 5 1 : ζ |. But they eventually become the same due to the
relation (3.23):

5
(k)
5
(k)
2 3 4 5 1 : ζ = Ω|(23451) e−ζ L0 = s ∗ Ω|(12345) e−ζ L0
V5 (ζ ) k=1 V5 (ζ ) k=1

5
(k)
= Ω|(12345) e−ζ L0 = 1 2 3 4 5 : ζ . (3.25)
s(V5 (ζ ))=V5 (ζ ) k=1
We say this as symmetry of the open-string 5-vertex under the cyclic permutation.
The 5-vertex is not invariant under the action of the BRST charge. Similarly to the case
of the 4-vertex the BRST transform can be written down using the lower vertices.
Proposition 3.3 (Q-action on 5-vertex). Action of the BRST charge on the 5-vertex (3.24)
becomes as follows:
5
5

12345:ζ Q (i)
=− k k + 1 k + 2 a : ζ a k + 3 k + 4 : ζ Sa a ,
i=1 k=1
(3.26)
where open-string indices in the RHS are understood modulo 5.
The origin of Eq. (3.26) is also in the geometry. The boundary of V5 (ζ ) is a circle. As
a set, it is a sum of V(k k+1 k+2 a) (ζ ) × V(a k+3 k+4) (ζ ). Here we write V4 (ζ ) and V3 (ζ ) (a
single point) as V(a k k+1 k+2) (ζ ) and V(a k+3 k+4) (ζ ) in order to show that external open-
strings participating in the diagrams are labeled in the clockwise order, respectively, by
(k, k + 1, k + 2, a) and (a , k + 3, k + 4) (Fig. 12).
Orientation of V5 (ζ ) is given by CM∂5 (the standard orientation of two-disk). Thereby
the circle is oriented. On the other hand, V4 (ζ ) × V3 (ζ ) has the orientation determined
from CM∂4 × CM∂3 . Comparing these orientations we obtain

5
∂V5 (ζ ) = − V(k k+1 k+2 a) (ζ ) × V(a k+3 k+4) (ζ ). (3.27)
k=1
Looking at Eq. (3.26) and Eq. (3.27) we find a coincidence between the BRST
transformation and the boundary operation. Actually the action of the BRST charge on
the 5-vertex becomes a representation of the boundary operator.
The 5-vertex depends on the cut-off scale. It vanishes at ζ = 0. The dependence comes
(i)
from the integration region V5 (ζ ) besides the propagators i e−ζ L0 in Eq. (3.24).
Proposition 3.4 (Scale dependence of 5-vertex). Scale-dependence of the 5-vertex (3.24)

is described by
1 d 5

1 2 3 4 5 : ζ = − k k + 1 k + 2 a : ζ a k + 3 k + 4 : ζ b0 Sa a
2 dζ
k=1
5
1
− 1 2 3 4 5 : ζ
(k)
L0 . (3.28)
2
k=1
We give proofs of Propositions 3.3 and 3.4. The next lemma is useful in the proof of
Proposition 3.3.
(12345)
Lemma 3.2. The BRST charge acts on the state (xi , yi )|i as follows:

(12345) (i)

(xi , yi ) i Q
i


 2i − 1 2i a b 2i + 1 2i + 2 a b 2i + 3




 × b0 e−yi L0 Sa a ∂x∂ i e−xi L0 Sb b





 + ∂y∂ i e−yi L0 Sa a b0 e−xi L0 Sb b
 for xi > 0,
= (3.29)

 2i − 1 2i a a 2i + 1 b b 2i + 2 2i + 3






 × b0 e−yi L0 Sa a ∂x∂ i exi L0 Sb b




 + ∂y∂ i e−yi L0 Sa a (−)b0 exi L0 Sb b for xi < 0.
This lemma can be shown in the same manner as Lemma 3.1. We omit the proof.
Proof of Proposition 3.3. We put

5
(12345)
Ii ≡ dxi ∧ dyi (xi , yi ) i Q(k)
.
Ui (ζ ) k=1

The LHS of Eq. (3.26) is just 5i=1 Ii . Due to Lemma 3.2 the integral Ii reduces to an
integral on ∂Ui (ζ ). The boundary ∂Ui (ζ ) has three components. Ii becomes a sum of the
three boundary integrals. The boundary integral along xi = yi turns out to equal (−1)×

(the boundary integral of Ii+1 along −xi+1 = yi+1 ). Thus their contributions to i Ii
cancel each other. (These two boundaries are identified in V5 (ζ ).) Net contribution of Ii to
the sum becomes the boundary integral along yi = 2ζ . Therefore, we have
5 5

1 2 3 4 5 : ζ Q(i) = Ii
i=1 i=1
2ζ

5

= dxi 2i − 1 2i a b 2i + 1 2i + 2 a b 2i + 3
i=1 0

× e−2ζ L0 Sa a b0 e−xi L0 Sb b
0

+ dxi 2i − 1 2i a a 2i + 1 b b 2i + 2 2i + 3
−2ζ
5
(k)
× e−2ζ L0 Sa a (−)b0exi L0 Sb b e−ζ L0
k=1
2ζ

5

=− 2i − 1 2i a dx a 2i + 1 b b 2i + 2 2i + 3 b0 e−xL0 Sb b
i=1 0
0 5
(k)
+ dx 2i + 1 2i + 2 b b 2i + 3 a (−)b0 e 0 Sb b
xL
e−ζ L0
−2ζ k=1

× e−2ζ L0 Sa a . (3.30)
We rewrite Eq. (3.30) in terms of the vertices at the scale ζ . It turns out to be

5

Eq. (3.30) = − 2i − 1 2i a : ζ a 2i + 1 2i + 2 2i + 3 : ζ Sa a
i=1

5

=− k k + 1 k + 2 a : ζ a k + 3 k + 4 : ζ Sa a . (3.31)
k=1
This is precisely the RHS of Eq. (3.26).
Proof of Proposition 3.4. We compute the first-order variation of the 5-vertex with respect
to the cut-off scale parameter ζ as follows:

δ 1 2 3 4 5 : ζ
! "

5
(k)
−ζ L0
=δ Ω| (12345)
e
V5 (ζ ) k=1
! " ! "

5
(k)
5
(k)
−ζ L0 −ζ L0
=δ Ω| (12345)
e + Ω| (12345)
δ e
V5 (ζ ) k=1 V5 (ζ ) k=1
5

5
(12345)
5
(k)
dxi ∧ dyi (xi , yi ) i −ζ 0 − δζ 1 2 3 4 5 : ζ
(k)
= e L
L0 .
i=1 δ U (ζ ) k=1 k=1
i
(3.32)
The first term of Eq. (3.32) is further evaluated as follows:
(12345)
5
(k)
dxi ∧ dyi (xi , yi ) i e−ζ L0
i δ U (ζ ) k=1
i
2ζ

= 2δζ dxi 2i − 1 2i a b 2i + 1 2i + 2 a b 2i + 3
i 0

× b0 e−2ζ L0 Sa a b0 e−xi L0 Sb b
0

+ dxi 2i − 1 2i a a 2i + 1 b b 2i + 2 2i + 3
−2ζ
5
−2ζ L (k)
× b0 e 0 Sa a (−)b0 e x i L 0 Sb b e−ζ L0
k=1

= −2δζ 2i − 1 2i a
i
2ζ

× dx a 2i + 1 b b 2i + 2 2i + 3 b0 e−xL0 Sb b
0
0

+ dx 2i + 1 2i + 2 b b 2i + 3 a (−)b0 e 0 Sb b
xL
−2ζ

5
(k)
× e−ζ L0 b0 e−2ζ L0 Sa a . (3.33)
k=1
We write Eq. (3.33) in terms of the vertices at the scale ζ . It turns out to be

Eq. (3.33) = −2δζ 2i − 1 2i a : ζ a 2i + 1 2i + 2 2i + 3 : ζ b0 Sa a (3.34)
i

= −2δζ k k + 1 k + 2 a : ζ a k + 3 k + 4 : ζ b0 Sa a . (3.35)
k
Hence we obtain the first term of the RHS of Eq. (3.28). The second term in Eq. (3.32)
gives the second term of the RHS of Eq. (3.28).
3.2. Conjecture on higher open-string vertices
We want to obtain low energy open-string n-valent vertices for the cases of n 6. These
vertices are determined in principle from the sum of the graphs which are one-particle
irreducible with respect to the regularized propagator. In the previous subsection explicit
constructions are given for n = 3, 4 and 5 cases. V3 (ζ ), V4 (ζ ) and V5 (ζ ) are the moduli of
the corresponding irreducible graphs. The approach taken there may be generalized.
The compactification CM∂n is Bn−3 . Each point of CM∂n represents an open-string n-
diagram. We can obtain a map CM∂n → (H⊗n )∗ : I → I| by applying the Feynman rule
to the diagrams. Since they are trivalent ribbon graphs, each diagram consists of (n − 2)
trivalent vertices and (n − 3) internal strips. Two trivalent vertices are connected at one
external open-string by inserting there the propagator of the form b0 e−τ L0 |Sa a in the
following way:

1 2 a : ζ a 3 4 : ζ b0 Sa a . (3.36)
The trivalent vertex is a grassmann-odd vector with the ghost number three and the
propagator is a grassmann-even vector with the ghost number two. The state I| acquires
the ghost number n and the grassmannity (−)n . We have several ways to identify points
of CM∂n with the diagrams. Different identifications are related with one another by the
automorphisms generated by s. Let us take one of them and fix it. We call the state I|
given under this identification as I|(1 2 ... n−1 n) . If one takes another identification related
by the automorphism s, the corresponding state is called I|(2 3 ... n 1) .
To have an explicit representation
# of the map it is convenient to take a patch-wise
construction. Let CM∂n ≡ i Ui . Each Ui may be identified with a suitable cone of Rn−3 .
The Euclidean coordinates (x1i , . . . , xn−3 i
) of Ui represent the metric of the underlying
ribbon graph. The map I → I| is introduced in a patch-wise manner. We denote the
i ) by (x i , . . . , x i )| .
state given at (x1i , . . . , xn−3 1 n−3 i
The next step is to obtain a (H⊗n )∗ -valued (n − 3)-form Ω|. It is given on each Ui
by dx1i ∧ · · · ∧ dxn−3
i
(x1i , . . . , xn−3
i
)|i . Recall Vn (ζ ) is the set of open-string n-diagrams
which are one-particle irreducible with respect to the regularized propagator (3.3). In other
word it is the set of open-string n-diagrams all internal strips of which have length less
than 2ζ . Vn (ζ ) also becomes a (n − 3)-ball. Open-string n-vertex at the cut-off scale ζ is
defined by an integration of Ω over the ball Vn (ζ ).

n
(i)
1 2 ... n − 1 n : ζ = Ω|(1 2 ... n−1 n) e−ζ L0 .
Vn (ζ ) i=1
Indices in the vertex are understood to label clockwise-ordered n open-strings on ∂D. The
vertex is regarded as a vector of (H(1) ⊗ · · · ⊗ H(n) )∗ . It has the ghost number n and the
grassmannity (−)n . The automorphisms of CM∂n will provide symmetry of the vertex.
Depending on choice of the identifications, we have several Ω. They are expected to be
related with one another by the automorphisms:
s ∗ Ω (1 2 ... n−1 n) = Ω (2 3 ... n 1) .

Fig. 13. A typical configuration of n points on ∂D which belongs to V(k ... k+l a) (ζ ) × V(a k+l+1 ... k+n−1) (ζ )
(⊂ ∂Vn (ζ )). The solid line connecting two points a and a represents open-string strip of length equal to 2ζ .
This implies that the n-vertex becomes symmetric or anti-symmetric under the cyclic
permutation when s preserves the orientation of CM∂n or not.
The boundary ∂Vn (ζ ) is identified, as a set, with a suitable sum of Vl+2 (ζ ) × Vn−l (ζ ).
Vn (ζ ) is oriented by the standard orientation of Bn−3 . Thereby ∂Vn (ζ ) is oriented. On the
other hand, Vl+2 (ζ ) × Vn−l (ζ ) is also oriented by Bl−1 × Bn−l−3 . Taking account of their
orientations we have
1
n n−3
∂Vn (ζ ) = (±)V(k k+1 ... k+l a) (ζ ) × V(a k+l+1 ... k+n−1) (ζ ).
2
k=1 l=1
The factor 1/2 is needed to avoid the double counting of the components. We write Vl+2 (ζ )
and Vn−l (ζ ) as V(k k+1 ... k+l a) (ζ ) and V(a k+l+1 ... k+n−1) (ζ ), in order to show that open-
strings participating in the diagrams are labeled in the clockwise order, respectively, by
(k, k + 1, . . . , k + l, a) and (a , k + l + 1, . . . , k + n − 1) (Fig. 13). The signature (±) must
be determined by a comparison of the orientations of ∂Vn (ζ ) and Vl+2 (ζ ) × Vn−l (ζ ). As
we have observed in the previous examples, action of the BRST charge on the open-string
vertices is expected to provide a representation of the boundary operator ∂. If so, we will
have
n

1 ... n : ζ Q (i)
i=1
1
n
n−3

= (±) k% k + 1 .&'. . k + l a( : ζ a% k + l + 1 &'
. . . k + n − 1( : ζ Sa a .
2
k=1 l=1 l+2 n−l
The n-vertex vanishes at ζ = 0 since Vn (0) becomes a point. The scale dependence of
open-string vertices follows from the above speculation. The dependence comes from the
(i)
ball Vn (ζ ) and the multiplier i e−ζ L0 . Variation of the n-vertex with respect to ζ is a
sum of the variations of these two. Therefore, we have
d
1 2 ... n : ζ
dζ

n
n−3

= (±) k% k + 1 .&'. . k + l a( : ζ a% k + l + 1 &'
. . . k + n − 1( : ζ b0 Sa a
k=1 l=1 l+2 n−l
n

− 1 2 ... n : ζ (i)
L0 .
i=1
To carry out this program and examine the conjectural properties of open-string vertices,
we need a systematic description of CM∂n . In particular, it is required so that the two
possible orientations of each component appearing in ∂Vn (ζ ) must be compared with each
other in a systematic way. Unfortunately we do not know such a description. We leave it
as a future problem. Instead we would like to propose a possible solution in the following
conjecture.
Conjecture. Open-string n-vertex 1 . . . n : ζ | ∈ (H⊗n )∗ (n 3) can be taken so that it

enjoys the following action of the BRST charge:
n
1
n n−3

1 ... n : ζ Q(i)
=− (−)(n+1)(k+l+1) k k + 1 . . . k + l a : ζ
2
i=1 k=1 l=1

× a k + l + 1 . . . k + n − 1 : ζ Sa a ,
(3.37)
besides the following cyclic asymmetry with respect to the open-string indices:

1 2 . . . n − 1 n : ζ = (−)n+1 2 3 . . . n 1 : ζ . (3.38)
Indices in the RHS of Eq. (3.37), say k + l, are understood modulo n.
Open-string indices of the n-vertex are understood to label the Hilbert spaces H(i) attached
to clockwise-ordered n open-strings on ∂D. We regard the n-vertex 1 . . . n : ζ | as a vector
of (H(1) ⊗ · · · ⊗ H(n) )∗ . It has the ghost number equal to n and grassmannity (−)n . We give
a few comments on the above conjecture. First, the asymmetry in Eq. (3.38) is consistent
with the action of the BRST charge (3.37). Secondly, it can be checked that for the cases
of n = 3, 4 and 5, Eqs. (3.37) and (3.38) reproduce correctly the corresponding results
obtained in the previous subsection.
As we referred in the program at the beginning of this subsection, action of the BRST
charge on the vertices is expected to provide a representation of the boundary operator ∂.
For this to be realized, the conjectural action of the BRST charge must be nilpotent, at
least. To show the nilpotency is a non-trivial test of the conjecture. If we accept Eqs. (3.37)
and (3.38), we have the following proposition.
Proposition 3.5. Action of the BRST charge given in (3.37) is nilpotent:

n n

1 2 ... n : ζ Q(i) Q(i) = 0. (3.39)
i=1 i=1
To show this proposition, some complicated calculations are required. We provide the
proof in Appendix A. The cut-off scale dependence of the vertices can be read from the
conjecture.
Proposition 3.6 (Scale dependence of higher vertices). Scale-dependence of the open-

string n-vertex (n 3) is described by
d
1 2 ... n : ζ
dζ

n
n−3

=− (−)(n+1)(k+l+1) k k + 1 . . . k + l a : ζ
k=1 l=1

× a k + l + 1 . . . k + n − 1 : ζ b0 Sa a
n

− 1 2 ... n : ζ (i)
L0 . (3.40)
i=1
We note that Eq. (3.40) reproduces the previous results for the cases of n = 3, 4 and 5.
We proceed on the rest of this paper by assuming the conjecture.
4. A∞ -algebra in low energy theory
To develop low energy description of classical open-string field theory, we introduce the
notion of homotopy associative algebra (A∞ -algebra). A∞ -algebra was introduced in [4]
and further investigated in [5,6] including deformation theory. 3 In string field theory, A∞
algebra was discussed in [22,23].
Let C be a Z-graded module. A graded module ΠC is defined by shifting the degree,
(ΠC)m ≡) C m+1 . We write the shifted degree by K. We also define Bk ΠC ≡ (ΠC)⊗k and
BΠC ≡ k Bk ΠC. We consider a family of maps mk : Bk ΠC → ΠC of degree K = 1.
Each mk induces a map dk : BΠC → BΠC by
dk (x1 , . . . , xn )

n−k p
≡ (−) j=1 K(xj ) x1 ⊗ · · · ⊗ xp ⊗ mk (xp+1 . . . xp+k ) ⊗ xp+k+1 ⊗ · · · ⊗ xn .
p=0
(4.1)

Then we put d̂ ≡ k dk .
Definition 4.1 (A∞ -algebra). (C, mk ) for k = 1, 2, . . . is said to be a A∞ -algebra if d̂ d̂ = 0.

(C, mk ) for k = 0, 1, 2, . . . is said to be a weak A∞ -algebra if d̂ d̂ = 0.
3 We follow the convention used in [5,6].

Let (C, mk ) be a A∞ -algebra. The condition d̂ d̂ = 0 becomes equivalent to the

constraints, dk d1 + · · · + d1 dk = 0 for k 1. These can be written down in terms of mk as

k
k−l p
(−) i=1 K(xi ) mk+1−l x1 , . . . , xp , ml (xp+1 , . . . , xp+l ), xp+l+1 , . . . , xk = 0.
l=1 p=0
(4.2)
It may be instructive to comment on a first few series of these relations:

m1 m1 (x) = 0, (4.3)

m1 m2 (x, y) + m2 m1 (x), y + (−) m2 x, m1 (y) = 0,
K(x)
(4.4)

m2 m2 (x, y), z + (−) m2 x, m2 (y, z) + m1 m3 (x, y, z)
K(x)

+ m3 m1 (x), y, z + (−)K(x)m3 x, m1 (y), z + (−)K(x)+K(y)m3 x, y, m1 (z) = 0.
(4.5)
The first equation implies that m1 is a boundary operator and the second equation shows
that m1 is a derivation with respect to m2 . The third equation is related with an associativity
relation. When m3 vanishes, m2 defines an associative algebra on C by putting x · y ≡
(−)K(x)m2 (x, y). When m3 does not vanish, the algebra “·” is not associative. But it induces
an associative algebra on the cohomology H ∗ (ΠC : m1 ) ≡ Ker m1 / Im m1 .
Apart from general theory of A∞ -algebra, we concentrate on classical open-string field
theory. The open-string Hilbert space H is naturally Z-graded by the ghost number G. We
introduce a (−1)-shifted ghost number operator K by K(A) ≡ G(A) − 1. By using the low
energy open-string vertices we define a family of maps mk : Bk ΠH → ΠH (k 1) as
follows:
Definition 4.2 (mk at the cut-off scale ζ ). For the case of k = 1 let m1 (A) be Q|A. For
the case of k 2 we set 4
mk (A1 , A2 , . . . , Ak : ζ )
[k/2]
= (−) i=1 K(Ak+1−2i )
a 1 2 . . . k : ζ Sa a |A1 1 |A2 2 · · · |Ak k . (4.6)
Our notation in Eq. (4.6) emphasizes that the maps mk for k 2 depend on the cut-off
scale ζ through the open-string vertices. Each map has the (−1)-shifted ghost number
equal to one. This can be seen as follows. Since the ghost numbers of the (k + 1)-vertex
and inverse reflector are, respectively,
equal to k + 1 and three,
the ghost number of
mk (A1 , . . . , Ak : ζ ) becomes 2 + ki=1 K(Ai ) (= (k + 1) + 3 + ki=1 G(Ai ) − 2(k + 1)).

Hence, K(mk (A1 , . . . , Ak : ζ )) = 1 + ki=1 K(Ai ). This means K(mk ) = 1. The following is
the main result of this paper:
Theorem 4.1 (A∞ -algebra in classical open-string field theory). (H, mk ) is a A∞ -algebra.
Namely, the maps mk (k 1) given by Definition 4.2 satisfy the infinite set of algebraic
relations (4.2).
4 [x] is the maximum integer not greater than x.

This theorem tells that the open-string vertices have the structure of A∞ -algebra which
is independent of the cut-off scale. In the microscopic description at ζ = 0, all the maps mk
for k 3 vanish. The corresponding A∞ -algebra is the open-string gauge algebra (Q, D).
This is a non-commutative associative algebra. On the other hand, in the macroscopic
description the higher maps do not vanish and the gauge algebra becomes non-associative.
The scale dependence of the A∞ -algebra (H, mk ) becomes as follows:
Proposition 4.1 (Scale dependence of the A∞ -algebra). Scale-dependence of the maps mk

(4.6) for k 2 are described by
∂mk
A1 , . . . , Ak : ζ
∂ζ

k−1
k−l

= −2 mk+1−l A1 , . . . , Ap , b0 ml (Ap+1 , . . . , Ap+l : ζ ), Ap+l+1, . . . , Ak : ζ
l=2 p=0
k−1

− L0 mk A1 , . . . , Ak : ζ − mk A1 , . . . , Ap , L0 Ap+1 , Ap+2 , . . . , Ak : ζ .
p=0
(4.7)
In the deformation theory of A∞ -algebra there exists a concept of equivalence relation

between two A∞ -algebras. This is called homotopy-equivalence [5,6]. It is interesting to
see whether the A∞ -algebra at ζ = 0 is homotopy-equivalent to the gauge algebra (Q, D)
or not. If it is affirmative, integral of Eq. (4.7) will be the A∞ -map [5,6] which realizes the
homotopy-equivalence. To answer this question is important on the physical ground since
Eq. (4.7) plays an important role in our description of renormalization group of open-
string field theory. Resolution of the question may shed light on geometry of the space of
boundary field theories in two dimensions.
Proof of Theorem 4.1. Let us abbreviate the vector a 1 2 . . . k : ζ |Sa a |A1 1 |A2 2 · · ·
|Ak k by m
*k (A1 , A2 , . . . , Ak : ζ ). We first rewrite Q m
*k (k 2) as follows:

mk A 1 , A 2 , . . . , A k : ζ
Q*

= (−)k+1 a 1 2 . . . k : ζ Q(a) Sa a |A1 1 |A2 2 · · · |Ak k
k

k
= (−) a 1 2 . . . k : ζ Sa a Q(j )
|A1 1 |A2 2 · · · |Ak k (4.8)
j =1

k

(a ) Sa a |A1 1 |A2 2 · · · |Ak k ,
+ (−) k
a 1 2 ... k : ζ Q + Q(j )
j =1
(4.9)
where the BRST invariance of the inverse reflector besides its odd-grassmannity are used
to show the second equality. We treat two terms (4.8) and (4.9) separately. The first term is
*k :
written down easily to a form expressed in terms of m

k−1 p
Eq. (4.8) = (−)k (−) i=1 G(Ai ) a 1 . . . k : ζ Sa a |A1 1 · · · |Ap p
p=0

× Q(p+1) |Ap+1 p+1 |Ap+2 p+2 · · · |Ak k

k−1 p
= (−)k (−)p+ i=1 K(Ai ) *k (A1 , . . . , Ap , QAp+1 , Ap+2 , . . . , Ak ).
m
p=0
(4.10)
On the other hand, we need some care to evaluate the second term. We compute the second
term using the conjecture. By Eq. (3.37) the second term becomes:
k

Eq. (4.9) = (−) 0 1 2 . . . k : ζ
k Q (j ) S0a |A1 1 |A2 2 · · · |Ak k
j =0
k−1 k−l

= (−)k+1 (−)kp p + l + 1 . . . k 0 1 . . . p b : ζ
% &' (
l=2 p=0 k+2−l

× b p + 1 . . . p + l : ζ Sb b S0a |A1 1 · · · |Ak k .
% &' (
l+1
(4.11)
We then permute open-string indices of the first vertices according to Eq. (3.38) as follows:
k−1 k−l

Eq. (4.11) = (−)k+1 (−)kp+(k−l+1)p 0 1 . . . p b p + l + 1 . . . k : ζ
% &' (
l=2 p=0 k+2−l

× b p + 1 . . . p + l : ζ Sb b S0a |A1 1 · · · |Ak k .
% &' (
l+1
(4.12)
Finally, we arrange Eq. (4.12), taking account of the grassmannities, so that it is
*k :
expressed in terms of m
Eq. (4.12)

k−1
k−l p
= (−)k+1 (−)p+l 1+ i=1 K(Ai )
l=2 p=0

× 0 1 . . . p b p + l + 1 . . . k : ζ S0a |A1 1 · · · |Ap p
% &' (
k+2−l

× b p + 1 . . . p + l : ζ Sb b |Ap+1 p+1 · · · |Ap+l p+l

% &' (
l+1
× |Ap+l+1 p+l+1 · · · |Ak k

k−1
k−l p
= (−)k+1 (−)p+l 1+ i=1 K(Ai )
l=2 p=0

×m *l (Ap+1 , . . . , Ap+l ), Ap+l+1 , . . . , Ak .
*k+1−l A1 , . . . , Ap , m
(4.13)
The sum of Eqs. (4.8) and (4.9) is Q*mk (A1 , . . . , Ak ). Using the expressions (4.10) and
*k :
(4.13) we obtain the following relation of m

k−1 p
*k (A1 , A2 , . . . , Ak ) +
Qm (−)k+p+1+ i=1 K(Ai ) *k (A1 , . . . , QAp+1 , . . . , Ak )
m
p=0

k−1
k−l p
+ (−)k+p+l 1+ i=1 K(Ai )
*k+1−l
m
l=2 p=0

× A1 , . . . , Ap , m
*l (Ap+1 , . . . , Ap+l ), Ap+l+1 , . . . , Ak = 0. (4.14)
This relation can be rewritten in terms of mk if one recalls the correspondence
[k/2]
*k (A1 , A2 , . . . , Ak : ζ ) = (−)
m i=1 K(Ak+1−2i )
mk (A1 , A2 , . . . , Ak : ζ ).
After straightforward calculations we finally find out that Eq. (4.14) is rephrased to

k
k−l p
(−) i=1 K(Ai ) mk+1−l A1 , . . . , Ap , ml (Ap+1 , . . . , Ap+l : ζ ),
l=1 p=0
Ap+l+1 , . . . , Ak : ζ = 0.
This is nothing but the relation (4.2). We omit the last part of the proof since the
calculations, though they are straightforward, are needed to be done case-by-case and
require some spaces.
Proof of Proposition 4.1. (We use the same notation as in the proof of Theorem 4.1.) We
*k /∂ζ by using Proposition 3.6.
first compute ∂ m
∂m*k
(A1 , A2 , . . . , Ak : Λ)
∂ζ

d
=
0 1 2 . . . k : ζ S0a |A1 1 |A2 2 · · · |Ak k
dζ
k−1 k−l

= −2 (−)kp p + l + 1 . . . k 0 1 . . . p b : ζ
% &' (
l=2 p=0 k+2−l

× b p + 1 . . . p + l : ζ b0 Sb b S0a |A1 1 · · · |Ak k
(4.15)
% &' (
l+1
k
(i)
− 0 1 ... k : ζ L0 S0a |A1 1 · · · |Ak k . (4.16)
i=0
*k . For this purpose we

We arrange the first term (4.15) so that it is expressed in terms of m
permute open-string indices of the first vertices by using the asymmetry (3.38) as follows:
k−1 k−l

Eq. (4.15) = −2 (−)kp+(k−l+1)p 0 1 . . . p b p + l + 1 . . . k : ζ
% &' (
l=2 p=0 k+2−l

× b p + 1 . . . p + l : ζ b0 Sb b
% &' (
l+1

× S0a |A1 1 · · · |Ak k . (4.17)
Then we rewrite Eq. (4.17), taking account of the grassmannities of |Ai , into the following
form:
Eq. (4.17)

k−1
k−l p
= −2 (−)(l+1) i=1 K(Ai ) 0 1 . . . p b p + l + 1 . . . k : ζ S0a
% &' (
l=2 p=0 k+2−l
× |A1 1 · · · |Ap p
+ ,
× b0(b) b p + 1 . . . p + l : ζ Sb b |Ap+1 p+1 · · · |Ap+l p+l
% &' (
l+1
× |Ap+l+1 p+l+1 · · · |Ak k

k−1
k−l p
= −2 (−)(l+1) i=1 K(Ai )
l=2 p=0

×m *l (Ap+1 , . . . , Ap+l ), Ap+l+1 , . . . , Ak .
*k+1−l A1 , . . . , Ap , b0 m
(4.18)
*k as follows:
While this, we can easily write down the second term (4.16) by means of m

k
Eq. (4.16) = −L0 m
*k (A1 , A2 , . . . , Ak ) − *k (A1 , . . . , L0 Ai , . . . , Ak ).
m (4.19)
i=1
*k /∂ζ in terms of m
Eqs. (4.18) and (4.19) give the following expression of ∂ m *k :
∂m*k
(A1 , A2 , . . . , Ak )
∂ζ

k−1
k−l p
= −2 (−)(l+1) i=1 K(Ai )
l=2 p=0

×m
*k+1−l A1 , . . . , Ap , b0 m
*l (Ap+1 , . . . , Ap+l ), Ap+l+1 , . . . , Ak

k
− L0 m
*k (A1 , A2 , . . . , Ak ) − *k (A1 , . . . , L0 Ai , . . . , Ak ).
m (4.20)
i=1
This describes the scale dependence of m *k . By rewriting this equation in terms of mk , we
find out, after some calculations, that it becomes Eq. (4.7).
The maps mk are introduced by the open-string vertices. Conversely, the open-string
vertices can be written in terms of mk as follows:
Proposition 4.2. Let k 3. We have the following identities:

1 . . . k : ζ A1 1 · · · |Ak k
[k/2]
= (−) i=1 K(Ak+1−2i ) ωab A1 a mk−1 (A2 , . . . , Ak : ζ ) b (4.21)
[k/2]
= (−) i=1 K(Ak−2i ) ωab mk−1 (A1 , . . . , Ak−1 : ζ ) a |Ak b . (4.22)
Remark 4.1. This proposition becomes useful in the subsequent discussions.
Proof of Proposition 4.2. We give a proof only for the case when k is odd. The other case
can be shown in the same manner. We rewrite ωab |A1 a |m2p (A2 , . . . , A2p+1 )b in terms
of the open-string vertices as follows:

ωab A1 a m2p (A2 , . . . , A2p+1 : ζ ) b
p
= ωab A1 1 (−) i=1 K(A2i ) 0 2 . . . 2p + 1 : ζ S0b |A2 2 · · · |A2p+12p+1
p
= (−)1+ i=1 K(A2i ) 0 2 . . . 2p + 1 : ζ ωab S0b |A1 a |A2 2 · · · |A2p+12p+1 ,
(4.23)
where we use odd grassmannities of the (2p + 1)-vertex and the (inverse) reflector to show
the second equality. Since we know ω12 |S23 = 3P1 , Eq. (4.23) can be evaluated into
p
Eq. (4.23) = (−)1+ i=1 K(A2i ) 0 2 . . . 2p + 1 : ζ (−0 Pa )|A1 a |A2 2 · · · |A2p+12p+1
p
= (−) i=1 K(A2i ) 1 2 . . . 2p + 1 : ζ A1 1 |A2 2 · · · |A2p+1 2p+1 . (4.24)
Hence we obtain Eq. (4.21) with k = 2p + 1. Similarly, by rewriting ωab |m2p (A1 , . . . ,
A2p )a |A2p+1 2p+1 in terms of the open-string vertices, we obtain Eq. (4.22) with k =
2p + 1.
5. Low energy description of open-string field theory
We use the Batalin–Vilkovisky (BV) formalism [7,24] for a low energy description
of open-string field theory. This formalism was elegantly used in quantizing open-string
field theory [8,9]. Although the BV formalism was originally intended to quantize gauge
invariant field theories which we already knew, it was used in closed-string field theory [2]
to find the unknown theory.
5.1. Odd symplectic structure of open-string field theory
In order to develop the BV formalism of open-string field theory we need to introduce

an odd symplectic structure ω on the open-string Hilbert space H. We define ω as follows.
Definition 5.1 (Odd symplectic structure of H). Let ω be an element of (H⊗2 )∗ which is
given by

ω(A, B) = (−)K(A) ω12 A1 |B2 , (5.1)
where K(A) is the (−1)-shifted ghost number of A.
The above bilinear form is non-degenerate on H because of the non-degeneracy of the

BPZ pairing. The symmetry of the reflector implies ω(A, B) = −(−)K(A)K(B) ω(B, A). The
selection rule of the BPZ pairing gives ω(A, B) = 0 ⇒ K(A) + K(B) = 1.
Let {φa }a be bases of H. We take the conjugate bases {φ a }a such that they satisfy
ω(φa , φ b ) = δab . The (−1)-shifted ghost number of φ a is K(φ a ) = 1 − K(φa ). We put
ωab ≡ ω(φa , φb ) and ωab ≡ ω(φ a , φ b ). These matrix elements enjoy

ωac ωcb = ωbc ωca = −δba . (5.2)
c c
This may
need a proof. It can be seen that
these two bases are related with each other by
φa = b φ ωb
ba or, equivalently, φ =
a − b φ ba
b ω . By using these relations we obtain
φa = b φb ( c (−)ω ωca ) and φ = b φ ( c (−)ωbc ωca ). These imply Eq. (5.2).
bc a b
Open-string field Φ is a vector of H with K(Φ) = 0. Hence Φ is restricted on a subspace

of H. In the BV formalism this restriction is removed [8,9] by introduction of fields and
anti-fields. Let us expand open-string field by the bases φa :

Φ= t a φa . (5.3)
a
Each coefficient t a is required to have the ghost number G(t a ) = −K(φa ) and the
a
grassmannity (−)G(t ) so that Φ is still grassmann-odd with K(Φ) = 0. We can regard
these coefficients t a as coordinates of a super-manifold (infinite-dimensional super-vector
space). This manifold is endowed with the odd-symplectic structure ω if one identifies the
open-string Hilbert space H with its tangent space. Let F (Φ) be a functional of open-string
field Φ. We introduce hamiltonian vector |∂F /∂Φ ∈ H as follows.
Definition 5.2 (Hamiltonian vector). Hamiltonian vector |∂F /∂Φ of a functional F (Φ) is
a vector of H. It is defined by the following first-order variation of F with respect to Φ:

∂F
δF (Φ) = ω δΦ, , (5.4)
∂Φ
where δΦ is arbitrary vector with K(δΦ) = 0.
The ghost number of hamiltonian vector becomes G(|∂F /∂Φ) = G(F ) + 2. Expansion
of the vector in terms of φ a turns out to have the forms

∂F
= |φ ∂
a L
F = F ∂ |φ
R a
. (5.5)
∂Φ a a
a a
ta
We regard F as a function of in the RHS of the above equations. We also introduce left-
and right-differentials ∂aL ≡ ∂ L /∂t a and ∂aR ≡ ∂ R /∂t a . ∂aL F and F ∂aR are defined by the
first-order variation of F with respect to t a :
R a
δF = δt a ∂aL F = F ∂a δt . (5.6)
a
Let us derive Eq. (5.5). We put δΦ = a δt φa and write down Eq. (5.4) as follows:
∂F
∂F
δF = δt a ω φa , =− ω , φa δt a . (5.7)
a
∂Φ a
∂Φ
These expressions mean

∂F ∂F
∂aL F = ω φa , , F ∂aR = −ω φa . (5.8)
∂Φ ∂Φ
While this, any vector A ∈ H can be expanded in the following forms:

|A = |φ ω(φa , A)
a
=− ω(A, φa )|φ .
a
(5.9)
a a
Let A be |∂F /∂Φ and compare Eq. (5.9) with Eq. (5.8). Then we obtain Eq. (5.5).
Definition 5.3 (Anti-bracket { , }). Let F and K be functionals of open-string field Φ.

Their anti-bracket {F, K} is defined by

∂F ∂K
{F, K} = ω , . (5.10)
∂Φ ∂Φ
Remark 5.1. By Eq. (5.5) the anti-bracket (5.10) has the following expression:

{F, K} = F ∂aR ωab ∂bL K. (5.11)
a,b
Proposition 5.1. The anti-bracket { , } defined by Eq. (5.10) enjoys the following
properties:
G({F, K}) = G(F ) + G(K) + 1, (5.12)
{F, K} = −(−) K(F )K(K)
{K, F }, (5.13)
{F, {K, M}} = {F, K}M + (−) K(F )(K(K)+1)
K{F, M}, (5.14)
0 = (−) K(F )K(M)
{{F, K}, M} + cyclic permutations. (5.15)
Eq. (5.14) means that the anti-bracket is a derivation with respect to itself. Eq. (5.15) is the
Jacobi identity. Thus the anti-bracket defines a super Lie algebra. This proposition can be
shown by standard calculations and we omit the proof.
5.2. Low energy action S[Φ : ζ ]
ζ
Low energy action Seff [Φ] of open-string field Φ can be obtained by integrating out all
the contributions from length scale less than ζ . It includes higher open-string interactions.
These interactions are obtained from graphs which are one-particle irreducible with respect
ζ
to the regularized propagator. We denote the classical part of Seff [Φ] by S[Φ : ζ ]. This
gives a low energy description of classical open string field theory. All the open-string
vertices investigated previously become the elementary interactions at the cut-off scale ζ
and contribute to S[Φ : ζ ]. Thus we arrive at the following definition of S[Φ : ζ ]:
Definition 5.4 (Low energy action of classical open-string field theory). Action of classical
open-string field at the cut-off scale ζ is given by
1 1
S[Φ : ζ ] = ω12 Φ1 Q(2) |Φ2 + 1 2 . . . k : ζ Φ1 |Φ2 · · · |Φk . (5.16)
2 k
k3
This action reduces to the microscopic action (2.50) as ζ → 0. In this subsection we

examine the low energy description from the perspective of the BV formalism. For this
purpose it is convenient to provide another form of the action by which the underlying BV
structure becomes manifest. We rewrite the interactions in (5.16) by using Proposition 4.2.
1 1
1 2 . . . k : ζ Φ1 |Φ2 · · · |Φk = ω12 Φ1 |mk (Φ k : ζ )2
k k+1
k3 k2
1
= ω Φ, mk (Φ k : ζ ) , (5.17)
k+1
k2
where mk (Φ k : ζ ) is the abbreviation for mk (Φ, . . . , Φ : ζ ). We will use similar

% &' (
k
abbreviation frequently. For instance, mk1 +k2 +1 (Φ k1 , A, Φ k2 : ζ ) stands for
mk1 +k2 +1 (Φ, . . . , Φ , A, Φ, . . . , Φ : ζ ).

% &' ( % &' (
k1 k2
The quadratic part of the action is just 12 ω(Φ, m1 (Φ)). Thus we obtain another expression
for S[Φ : ζ ]:

1
S[Φ : ζ ] = ω Φ, mk (Φ k : ζ ) . (5.18)
k+1
k1
Proposition 5.2 (Equation of motion for Φ). First-order variation of the action (5.16) with
respect to Φ becomes as follows:

δS[Φ : ζ ] = ω δΦ, mk (Φ : ζ ) .
k
(5.19)
k1
Therefore, the equation of motion δS = 0 is given by

mk (Φ k : ζ ) = 0. (5.20)
k1
We give a few comments on some relation with deformation theory [5,6] of A∞ -algebra.
Let (C, mk ) be a A∞ -algebra. Let b ∈ (ΠC)0 . Define a family of maps mbk (k = 0, 1, . . .)
by
mbk (x1 , . . . , xk )

= mk+q1 +···+qk+1 b, . . . , b, x1 , b, . . . , b, . . . , b, . . . , b, xk , b, . . . , b .
% &' ( % &' ( % &' ( % &' (
q1 ,...,qk+1 0 q1 q2 qk qk+1
(5.21)
It was shown in [5,6] that (C, mbk ) for k = 0, 1, . . . becomes a weak A∞ -algebra. If mb0 (1)
vanishes, it becomes a A∞ -algebra. So mb0 (1) is an obstruction in the deformation theory.
Let us consider the particular case of open-string field theory. For any Φ ∈ (ΠH)0 we
obtain a weak A∞ -algebra (H, mΦ k ). The obstruction is m0 (1 : ζ ). Now we become
Φ
aware that the equation of motion (5.20) states vanishing the obstruction. For any classical
Φ
solution ΦO , even apart from the trivial one, we still have a A∞ -algebra H, mk O . In the
BV formalism variational formula (5.19) determines hamiltonian vector of the action S. We
also recognize that the hamiltonian vector is nothing but the obstruction of deformation.

∂S[Φ : ζ ]
= mΦ
∂Φ 0 (1 : ζ ). (5.22)
Remark 5.2. The variational formula (5.19) is mod O((δΦ)2 ). By calculations similar to
the proof of Eq. (5.19) (which we give below), the exact result can be found out to be
1
S[Φ + δΦ : ζ ] = S[Φ : ζ ] + ω δΦ, mΦ (δΦ : ζ ) . (5.23)
k+1 k
k0
Proof of Proposition 5.2. The variation can be evaluated as follows:

δS[Φ : ζ ]
1
= ω12 |δΦ1 Q(2) |Φ2 + |Φ1 Q(2) |δΦ2
2
k
1
+ 1 ... k : ζ |Φ1 · · · |δΦi · · · |Φk
k
k3 i=1

= ω12 δΦ1 Q(2) |Φ2
k
1
+ i i + 1 . . . i + k − 1 : ζ δΦi |Φi+1 · · · |Φi+k−1
k
k3 i=1

= ω12 δΦ1 Q(2) |Φ2 + 1 2 . . . k : ζ δΦ1 |Φ2 · · · |Φk
k3

= ω12 δΦ1 mk (Φ k : ζ ) 2 = ω δΦ, mk (Φ k : ζ ) . (5.24)
k1 k1
We use the asymmetry (3.38) of open-string vertices and the symmetry of the reflector to
show the second equality in the above computation. We also use Proposition 4.2 to show
the fourth equality.
In the previous section, in order to emphasize the perspective of renormalization group,

open-string vertices are constructed by using the propagator in the Siegel gauge. These
vertices are used in the definition of S[Φ : ζ ]. Nevertheless it turns out that the action is a
covariant classical action of open-string field.
Definition 5.5 (Gauge transformation of Φ). For any vector ρ ∈ H with K(ρ) = −1 we
define an infinitesimal transformation of Φ by5
δ ρ Φ = mΦ
1 (ρ : ζ ). (5.25)
When ζ = 0, this transformation reduces to δρ Φ = Qρ +Φ Dρ −ρ DΦ. This is precisely the

gauge symmetry of the microscopic action (2.50). In the case of ζ = 0, it is unclear whether
the transformation (5.25) is still gauge symmetry of S[Φ : ζ ]. We have the following
proposition:
Proposition 5.3 (Gauge invariance of the action). The action (5.19) is invariant under the
infinitesimal transformation (5.25) of Φ.
Proof of Proposition 5.3. By using the variational formula (5.19) we rewrite δρ S to the
form

δρ S[Φ : ζ ] = ω δρ Φ, mΦ 0 (1 : ζ )

= ω mΦ1 (ρ : ζ ), m0 (1 : ζ ) .
Φ
(5.26)
We further rewrite Eq. (5.26) by using Proposition 4.2 as follows:
Eq. (5.26)

= ω mq1 +q2 +1 (Φ q1 , ρ, Φ q2 ), mΦ
0 (1)
q1 ,q2

= ωab mq1 +q2 +1 (Φ q1 , ρ, Φ q2 ) a mΦ
0 (1) b
q1 ,q2

= (−)q2 1 . . . q1 + q2 + 2 : ζ Φ1 · · · |Φq1 |ρq1 +1
q1 ,q2
% &' (
q1

× |Φq1 +2 · · · |Φq1 +q2 +1 mΦ (1) q +q +2 . (5.27)
% &' ( 0 1 2
q2
5 We follow the convention of the deformation theory of A -algebra.

∞
We replace the open-string fields by taking account of the asymmetry (3.38) of the open-
string vertices and then rewrite Eq. (5.27), again by using Proposition 4.2 as follows:
Eq. (5.27)

=− 1 . . . q1 + q2 + 2 : ζ Φ1 · · · |Φq2 mΦ
0 (1) q2 +1
q ,q
% &' (
1 2
q2
× |Φq2 +2 · · · |Φq1 +q2 +1 |ρq1 +q2 +2

% &' (
q1

=− ω mq1 +q2 +1 (Φ q2 , mΦ q1
0 (1), Φ ), ρ
q1 ,q2
Φ
= −ω mΦ
1 m0 (1) , ρ . (5.28)
Therefore, δρ S turns out to be
Φ
δρ S[Φ : ζ ] = ω ρ, mΦ1 m0 (1 : ζ ) : ζ . (5.29)
Recall (H, mΦk ) is a weak A∞ -algebra. Especially it follows from Definition 4.1 that
mΦ (m Φ (1)) = 0. Thus δ S = 0.
1 0 ρ
In the BV formalism [7,24] quantum master equation is the criterion for consistency of
a quantum gauge theory. Its classical limit is called classical master equation. It is simply
given by {S, S} = 0, where S is the classical BV action of the gauge theory. Classical
master equation itself is not sufficient to ensure consistency of the quantum theory, but it
must be satisfied at the classical level in order to obtain a consistent quantum gauge theory.
Proposition 5.4 (Classical master equation). The action (5.16) satisfies classical master
equation:

S[Φ : ζ ], S[Φ : ζ ] = 0. (5.30)
The classical master equation (5.30) eventually reduces to the A∞ -algebra (H, mk )
as we will find in the proof of this proposition. Physical significance of classical master
equation is easy to see if we take another form of this equation. The anti-bracket in
Eq. (5.30) is equal to ω(mΦ 0 (1 : ζ ), m0 (1 : ζ )), where we identify the hamiltonian vector
Φ
with the obstruction of the deformation theory.
Definition 5.6 (BRST transformation of Φ). BRST transformation δBRS of open-string

field Φ is defined by the hamiltonian vector (5.22).
δBRS Φ = mΦ
0 (1 : ζ ). (5.31)
δBRS Φ has the ghost number two. Hence, we can attach the ghost number one to δBRS .
We can also see δBRS is grassmann-odd. Physical significance of classical master equation
may be observed in the following propositions:
Proposition 5.5 (Nilpotency of BRST transformation).

δBRS · δBRS = 0. (5.32)
Proposition 5.6 (BRST invariance of the action). The variational formula (5.19) can be
written as
δS[Φ : ζ ] = ω(δΦ, δBRS Φ). (5.33)
In particular, if one takes δBRS Φ as δΦ, we obtain
δBRS S[Φ : ζ ] = 0. (5.34)
Eq. (5.33) is a simple reinterpretation of Eq. (5.19). The BRST invariance of the action
follows from Proposition 5.4 by using the formula (5.33).
Proof of Proposition 5.4. We apply the strategy taken in [2] for a proof of the quantum
master equation of closed-string field theory. We first write down the anti-bracket
ω(mΦ Φ
0 (1), m0 (1)) as follows:
k−1
Φ
ω m0 (1), m0 (1) =
Φ k−l l
ω mk−l (Φ ), ml (Φ )
k2 l=1
k−1
k−l
= k−l
ω mk−l (Φ ), ml (Φ ) l
k
k2 l=1
k−1
l
+ ω mk−l (Φ k−l ), ml (Φ l )
k
k2 l=1
k−1
2
= (k − l)ω mk−l (Φ ), ml (Φ ) .
k−l l
(5.35)
k
k2 l=1
To proceed on the computation we need a suitable identity: We rewrite ω(mk−l (Φ k−l ),

ml (Φ l )), which appears in Eq. (5.35), into the following form by using Proposition 4.2:

ω mk−l (Φ k−l ), ml (Φ l ) = − ωab mk−l (Φ k−l ) a ml (Φ l ) b

= − 1 . . . k − l + 1 : ζ Φ1 · · · |Φk−l ml (Φ l ) k−l+1 . (5.36)
% &' (
k−l
We permute (k − l) open-string fields in this equation by taking account of the asymmetry
of the vertex and then rewrite the equation in terms of mk by using Proposition 4.2 as
follows:

Eq. (5.36) = (−)i i . . . k − l + 1 1 . . . i − 1 : ζ Φi · · · |Φk−l ml (Φ l ) k−l+1
% &' (
k−l−i+1
× |Φ1 · · · |Φi−1
% &' (
i−1

= ωab mk−l Φ k−l−i+1 , ml (Φ l ), Φ i−2 a |Φb

= ω mk−l Φ k−l−i+1 , ml (Φ l ), Φ i−2 , Φ . (5.37)
Hence, we obtain the following identity independent of i:

ω mk−l Φ k−l−i , ml (Φ l ), Φ i−1 , Φ = ω mk−l (Φ k−l ), ml (Φ l ) . (5.38)
Now we proceed on the computation of Eq. (5.35). By using the above identity it
becomes as follows:
k−l
2 k−1
k−l−i
Eq. (5.35) = ω mk−l Φ , ml (Φ l ), Φ i−1 , Φ
k
k2 l=1 i=1
k k−l
2 i
= ω l
mk+1−l Φ , ml (Φ ), Φ k−l−i
,Φ . (5.39)
k+1
k1 l=1 i=0
This vanishes identically due to the relation (4.2) of the A∞ -algebra (H, mk ). Thus we
obtain ω(mΦ0 (1), m0 (1)) = 0.
Φ
Proof of Proposition 5.5. We show δBRS · δBRS Φ = 0. We first rewrite δBRS · δBRS Φ as
follows:

δBRS · δBRS Φ = δBRS mΦ
0 (1)

= δBRS mk (Φ k )
k1

= δBRS m1 (Φ) + δBRS 0 1 . . . k : ζ S0a |Φ1 · · · |Φk
k2

= −m1 (δBRS Φ) + (−)k 0 1 . . . k : ζ S0a δBRS |Φ1 · · · |Φk .
k2
(5.40)

The first term of Eq. (5.40) is equal to − k1 m1 (mk (Φ k )). As for the second term we
further compute by using Proposition 4.2 as follows:

(−)k 0 1 . . . k : ζ S0a δBRS |Φ1 · · · |Φk
k2
k−1

= (−)k+i 0 1 . . . k : ζ S0a |Φ1 · · · |Φi |δBRS Φi+1 |Φi+2 · · · |Φk
k2 i=0
k−1

= (−) k+i
0 1 . . . k : ζ S0a |Φ1 · · · |Φi ml (Φ ) i+1 |Φi+2 · · · |Φk
l
l1 k2 i=0

k

=− mk Φ i , ml (Φ l ), Φ k−1−i
l1 k2 i=1
k−1 k−l

=− i
mk+1−l Φ , ml (Φ ), Φ l k−l−i
. (5.41)
k1 l=1 i=0
By putting these two terms together, we eventually obtain the following expression of
δBRS · δBRS Φ:
k−1 k−l
i
δBRS · δBRS Φ = − m1 mk (Φ ) −
k l
mk+1−l Φ , ml (Φ ), Φ k−l−i
k1 k1 l=1 i=0

k k−l
i
=− mk+1−l Φ , ml (Φ l ), Φ k−l−i . (5.42)
k1 l=1 i=0
This actually vanishes by the A∞ -relation.
6. Renormalization group of open-string field theory
So far, open-string field Φ is simply a vector of the open-string Hilbert space H. In

the presence of the short-distance cut-off scale parameter ζ , classical dynamics of open-
string field is governed by the action S[Φ : ζ ]. As concerns the classical dynamics, any
consideration on off-shell open-string field is irrelevant. In particular, we cannot find any
relation between off-shell open-string fields at different scales. On the other hand, if one
takes the renormalization group perspective, namely, if one regards S[Φ : ζ ] as the classical
ζ
part of Seff [Φ] which could be obtained integrated out all the contributions from length
scale less than ζ , off-shell open-string fields at different scales should be related with one
another by renormalization group flow. In this section we study renormalization group of
classical open-string field theory. Our formulation of renormalization group equation (RG
equation) is based on Proposition 4.1.
Now open-string field Φ is allowed to depend on the cut-off scale parameter ζ .
Equivalently, the super-coordinates t a which appear in the expansion by the bases φa ,
depend on ζ .

Φ(ζ ) = t a (ζ )φa . (6.1)
a
RG equation determines the scale dependence of open-string field Φ or equivalently a flow
t a (ζ ) on the super-manifold.
Definition 6.1 (RG equation of classical open-string field). Let S[Φ : ζ ] be the action
(5.16). RG equation of open-string field Φ is defined by
d
S[Φ(ζ ) : ζ ] = 0. (6.2)
dζ
The total derivation in the RG equation (6.2) can be factorized into a sum of two terms.
One is a derivation through the coordinates t a (ζ ). Using the variational formula (5.19) it is
given by ω(dΦ/dζ, |∂S/∂Φ). The other is a derivation through the open-string vertices
in the action. It is given by ∂S/∂ζ . The factorized form of the RG equation becomes as
follows:

dΦ ∂S[Φ : ζ ] ∂S
ω , = − [Φ : ζ ]. (6.3)
dζ ∂Φ ∂ζ
The RHS of this equation can be calculated using Proposition 4.1. Let Sint [Φ : ζ ] be the
interaction part of the action (5.16):

1
Sint [Φ : ζ ] ≡ ω Φ, mk (Φ : ζ ) .
k
(6.4)
k+1
k2
The corresponding hamiltonian vector becomes

∂Sint [Φ : ζ ]
= mk (Φ k : ζ ). (6.5)
∂Φ
k2
With this notation the scale dependence of the action is described as follows:
Proposition 6.1. Derivation of the action (5.16) with respect to the cut-off scale parameter
ζ has the following form:

∂S ∂Sint [Φ : ζ ] ∂Sint [Φ : ζ ]

[Φ : ζ ] = −ω b0
+ L0 Φ, . (6.6)
∂ζ ∂Φ ∂Φ
Due to this proposition the RG equation (6.3) becomes

dΦ ∂S[Φ : ζ ] ∂Sint [Φ : ζ ]

∂Sint [Φ : ζ ]

ω , = ω b0 + L0 Φ, . (6.7)
dζ ∂Φ ∂Φ ∂Φ
Eq. (6.7) is a classical analogue of the Polchinski’s RG equation [10,11]. RG equation
a la Wilson or Polchinski is dZ(t (ζ ) : ζ )/dζ = 0, where Z is a partition function of a
quantum field theory regularized by the short distance cut-off scale ζ . Roughly speaking,
correspondence with the RG equation (6.2) can be seen if one puts Z = e−S . The
precise relation may be established by using the Legendre transformation as considered
in [25]. There appears a quantum correction in the original equation. It has the form
∼ ∂ 2 Sint /∂φ∂φ. In our case quantum correction appears as the higher loop contribution
of open-string. Description of RG equation of string-field along the line presented in (6.7)
was first given for closed-string field theory [12]. Similarity with the BV master equation
was emphasized there. Resemblance between the two was further investigated in [26].
As we stated earlier, the RG equation (6.2) is introduced to determine the ζ -evolution of
open-string field Φ. Let us solve the RG equation (6.7) in such a form by which it becomes
manifest. For this purpose we first rewrite the RHS of Eq. (6.7) as follows:

∂Sint ∂Sint ∂S ∂Sint
ω b0
+ L0 Φ, = ω b0 + Qb0 Φ,
∂Φ ∂Φ ∂Φ ∂Φ

∂S ∂Sint ∂S
= ω b0 , + ω Qb0 Φ,
∂Φ ∂Φ ∂Φ

∂Sint ∂S ∂S
= ω b0 , + ω Qb0 Φ,
∂Φ ∂Φ ∂Φ

∂Sint ∂S
= ω b0 + Qb0 Φ, , (6.8)
∂Φ ∂Φ
where the anti-commutation relation, {Q, b0 } = L0 , is used to show the first equality. The
second equality follows from Q2 = 0 and the BRST invariance of the reflector. To show
the third equality we use the asymmetry of ω besides Eq. (2.18). The RG equation (6.7)
acquires the following expression by Eq. (6.8):

dΦ ∂Sint [Φ : ζ ] ∂S[Φ : ζ ]
ω − b0
+ Qb0 Φ , = 0. (6.9)
dζ ∂Φ ∂Φ
Thus we obtain
Proposition 6.2. Open-string field Φ(ζ ) obeying the following first-order differential
equation is a solution of the RG equation (6.2):

dΦ ∂Sint [Φ : ζ ]

= b0 + Qb0 Φ. (6.10)
dζ ∂Φ
We do not know whether Eq. (6.10) is equivalent to the RG equation (6.2) or not since
equivalence between Eqs. (6.9) and (6.10) is obscure. But this does not cause any problem.
Skeptical reader can adopt Eq. (6.10) as a definition of RG equation of open-string field and
keep Eq. (6.2) as a proposition. Eq. (6.10) will be also called RG equation of open-string
field.
Proof of Proposition 6.1. We first write down the partial derivation ∂S[Φ : ζ ]/∂ζ by using
Proposition 4.1:
1
∂S ∂mk k
[Φ : ζ ] = ω Φ, (Φ : ζ )
∂ζ k+1 ∂ζ
k2

1
k−1
=− ω Φ, L0 mk (Φ ) +
k p
mk (Φ , L0 Φ, Φ k−1−p
) (6.11)
k+1
k2 p=0
k−1 k−l
1 p
−2 ω Φ, mk+1−l Φ , b0 ml (Φ l ), Φ k−l−p .
k+1
k2 l=2 p=0
(6.12)
We treat two terms (6.11) and (6.12) separately. We rewrite the first term by using
Proposition 4.2 as follows:

k−1
ω Φ, L0 mk (Φ ) +k p
mk (Φ , L0 Φ, Φ k−1−p
)
p=0
k−1

= ωab Φa L(b)
0 |m k (Φ k
)b + ωab Φa mk (Φ p , L0 Φ, Φ k−1−p ) b
p=0
(b)
= ωab Φa L0 |mk (Φ)b

k−1
(p+2)
+ 1 . . . k + 1 : ζ Φ1 · · · |Φp+1 L0 |Φp+2 |Φp+3 · · · |Φk+1 .
% &' ( % &' (
p=0
p+1 k−p−1
(6.13)
(1)
Due to the property (2.18) of the reflector we have ω12 |(L0 |A1 )|B2 = ω12 |A1 ×
(L(2)
0 |B2 ). We permute open-string fields in the second term of Eq. (6.13) by taking
account of the asymmetry (3.38) of the vertices, and then rewrite Eq. (6.13) again in terms
of mk by using Proposition 4.2 as follows:

Eq. (6.13) = ωab L(a)
0 |Φa mk (Φ ) b
k

k−1
(p+1)
+ p + 1 p + 2 . . . p : ζ L0 |Φp+1 |Φp+2 · · · |Φp
% &' (
p=0
k

= ωab L(a)
0 |Φa mk (Φ ) b
k
(1)
+ k 1 2 . . . k + 1 : ζ L0 |Φ1 |Φ2 · · · |Φk+1
% &' (
k

= (k + 1)ω L0 Φ, mk (Φ k ) . (6.14)
As the result we obtain the following expression of Eq. (6.11).

∂Sint
Eq. (6.11) = − ω L0 Φ, mk (Φ ) = −ω L0 Φ,
k . (6.15)
∂Φ
k2
Next, we examine the second term. By using Proposition 4.2 we rewrite Eq. (6.12) as
follows:
k−1 k−l
p
ω Φ, mk+1−l Φ , b0 ml (Φ l ), Φ k−l−p
l=2 p=0

k−1
k−l

= ωab Φa mk+1−l Φ p , b0 ml (Φ l ), Φ k−l−p b
l=2 p=0

k−1
k−l

= 1 . . . k + 2 − l : ζ Φ1 · · · |Φp+1 b0 |ml (Φ l )p+2 |Φp+3 · · · |Φk+2−l
% &' ( % &' (
l=2 p=0
p+1 k−l−p

k−1

= (k + 1 − l) 1 . . . k + 2 − l : ζ b0 |ml (Φ l )1 |Φ2 · · · |Φk+2−l
% &' (
l=2 k+1−l

k−1

= (k + 1 − l)ω b0 ml (Φ l ), mk+1−l (Φ k+1−l ) . (6.16)
l=2
In this computation, in addition to the use of Proposition 4.2, particularly to show the
third equality we use the asymmetry (3.38) of the vertex. We can rewrite Eq. (6.16) in a
convenient form. We notice the equality

ω b0 ml (Φ l ), mk+1−l (Φ k+1−l ) = ω b0 mk+1−l (Φ k+1−l ), ml (Φ l ) . (6.17)
By using this equality we write down Eq. (6.16) to the following form:

k−1

(k + 1 − l)ω b0 ml (Φ l ), mk+1−l (Φ k+1−l )
l=2

k−1
k+1−l
= ω b0 ml (Φ l ), mk+1−l (Φ k+1−l )
2
l=2

k−1
l
+ ω b0 mk+1−l (Φ k+1−l ), ml (Φ l )
2
l=2

k−1
(k + 1 − l) + l
= ω b0 ml (Φ l ), mk+1−l (Φ k+1−l )
2
l=2
k+1
k−1

= ω b0 ml (Φ l ), mk+1−l (Φ k+1−l ) . (6.18)
2
l=2
Hence, the second term is found out to be:

k−1

Eq. (6.12) = − ω b0 ml (Φ l ), mk+1−l (Φ k+1−l )
k2 l=2

∂Sint ∂Sint
= −ω b0 , . (6.19)
∂Φ ∂Φ
Sum of (6.11) and (6.12) is equal to ∂S/∂ζ . Expressions (6.15) and (6.19) show that it
is precisely Eq. (6.6).
7. Analysis of RG flow in Siegel gauge
If we impose the Siegel gauge condition b0 Φ = 0 on open-string field, the RG equation

(6.10) obtains a simple form


= b0 . (7.1)
dζ ∂Φ
Clearly, the evolution of Φ governed by Eq. (7.1) preserves the gauge condition. In this
section we investigate the RG equation (7.1). Throughout this section the Siegel gauge
condition is imposed on open-string field.
Let HS be subspace of the open-string Hilbert space H consisting of states which satisfy
the Siegel gauge condition. Any vector φ ∈ HS does not contain c0 . The odd symplectic
structure ω becomes null on HS . Instead of ω we introduce another bilinear form g:
Definition 7.1 (Bilinear form g on HS ). For any vectors A, B ∈ HS , we define g(A, B) by

g(A, B) = ω12 c0(1) |A1 |B2 . (7.2)
The bilinear form g is non-degenerate on HS . ω12 | = ω21 | implies g(A, B) =

(−)K(A)K(B)+K(A)+K(B) g(B, A). Selection rule (2.16) of the BPZ pairing gives g(A, B) = 0
⇒ K(A) + K(B) = 0.
Let {φi }i be bases of HS . We can take the conjugate bases {φ i }i such that they
satisfy g(φ i , φj ) = δji . The (−1)-shifted ghost number of φ i is K(φ i ) = −K(φi ). We put
gij ≡ g(φi , φj ) and g ij ≡ (−)K(φj ) g(φ i , φ j ). It can be
seen that two bases are related with
each other by φ = j g ij φj or equivalently by φi = j gij φ j . Hence, g ij is the inverse
i
of gij :

g ik gkj = gj k g ki = δji . (7.3)
k k

In the Siegel gauge, open-string field Φ has an expansion of the form Φ = i t i φi .
i
Each coefficient t i has the ghost number G(t i ) = −K(φi ) and the grassmannity (−)G(t ) .
They are coordinates of the super-submanifold constrained by the gauge condition. We
write the ζ -evolution of Φ as

Φ(ζ ) = t i (ζ )φi . (7.4)
i
RG flow t i (ζ )
on the super-manifold is determined from Eq. (7.1) by using this expansion.
We choose φi to be the eigenvectors of L0 :
L0 |φi = ∆i |φi . (7.5)
7.1. Classical solution and RG equation
Classical solutions of open-string field theory are stationary configurations of classical

open-string field action. In our description of open-string field theory, we have different
classical actions depending on the cut-off scale one considers. The classical solutions or
equivalently solutions of the equation of motion (5.20) are expected to depend on the
cut-off scale parameter ζ . It becomes unclear whether there exists any relation between
classical solutions at different scales. We start our discussion by resolving this question.
Let ΦO be a solution of the equation motion at a given scale ζ0 ,

mΦ0 (1 : ζ0 ) Φ=Φ = 0. O
(7.6)
We examine the RG flow of this classical solution. Let ΦO (ζ ) be the solution of the
RG equation (7.1) which satisfies the initial value condition ΦO (ζ0 ) = ΦO . We obtain the
following proposition:
Proposition 7.1. Suppose we have a classical solution ΦO at a given scale ζ0 . The RG flow
of ΦO generates a family of classical solutions {ΦO (ζ )}ζ . Namely, at any value of ζ , ΦO (ζ )
Φ (ζ )
satisfies the equation of motion, m0 O (1 : ζ ) = 0.
To explain an implication of this proposition, we rewrite the RG equation (7.1), by using

the anti-commutation relation {Q, b0 } = L0 , to the form
dΦ(ζ ) Φ(ζ )
= −L0 Φ(ζ ) + b0 m0 (1 : ζ ). (7.7)
dζ
Then we recognize that the RG flow of a classical solution is described by the simple
equation
dΦO (ζ )
= −L0 ΦO (ζ ), (7.8)
dζ
because ΦO (ζ ) satisfies the equation of motion as stated in the above proposition. This
evolution equation is easily integrated and provides the following relation:
ΦO (ζ ) = e−ζ L0 ΦO (0). (7.9)
This means that any classical solution in the microscopic description (ζ = 0) still continues
to be a classical solution in the low energy description (ζ > 0) modulo (7.9). The classical
action of ΦO (ζ ) is independent of ζ . We have S[ΦO (ζ ) : ζ ] = S cubic [ΦO (0)].
In terms of the super-coordinates t i we can write Eq. (7.9) as
tOi (ζ )eζ ∆i = tOi (0). (7.10)
Hence any classical solution in the microscopic description gives rise to a trivial flow on the
super-manifold. It may sound unpleasant since we hope to interpret it as a fixed point of the
i
RG flow. We can save this situation by introducing rescaled coordinates instead iof−ζt ∆. Let
Φ be open-string field at the cut-off scale ζ . We expand it in the form, Φ = i T e iφ .
i
i
The coefficients T define the rescaled coordinates. The RG flow Φ(ζ ) determines a flow
T i (ζ ). It is related with t i (ζ ) by
T i (ζ ) = t i (ζ )eζ ∆i . (7.11)
Eq. (7.10) becomes TOi (ζ ) = TOi (0).
Classical solutions are fixed points of the flow of T i .
Now let us show Proposition 7.1. The following lemma becomes useful in the proof.
0 (1 : ζ ) is described
Lemma 7.1. Let Φ be a open-string field. The scale dependence of mΦ
by
∂mΦ
0
(1 : ζ ) = mΦ
1 (L0 Φ : ζ ) − 2m1 b0 m0 (1 : ζ ) : ζ + (L0 − 2b0 Q)m0 (1 : ζ ).
Φ Φ Φ
∂ζ
(7.12)
Proof of Lemma 7.1. We first compute ∂mΦ

0 /∂ζ by using Proposition 4.1.
∂mΦ ∂mk
0
(1 : ζ ) = (Φ k : ζ )
∂ζ ∂ζ
k2

k−1
=− L0 mk (Φ : ζ ) +
k p
mk (Φ , L0 Φ, Φ k−1−p
: ζ) (7.13)
k2 p=0

k−1
k−l

−2 mk+1−l Φ p , b0 ml (Φ l : ζ ), Φ k−l−p : ζ . (7.14)
k2 l=2 p=0
We treat two terms (7.13) and (7.14) separately. We write down the first term into the
following form:

Eq. (7.13) = − L0 mk (Φ k ) − QΦ
k1

k−1
+ p
mk (Φ , L0 Φ, Φ k−p−1
) − QL0 Φ
k1 p=0
Φ
0 (1) − QΦ − m1 (L0 Φ) − QL0 Φ
= −L0 mΦ
0 (1) − m1 (L0 Φ) + 2L0 QΦ.
= −L0 mΦ Φ
(7.15)
As for the second term, we arrange it as follows:

k−1
k−l

Eq. (7.14) = −2 mk+1−l Φ p , b0 ml (Φ l ), Φ k−l−p
k2 l=1 p=0

k−1

+2 mk Φ p , b0 QΦ, Φ k−1−p
k2 p=0
Φ
1 b0 m0 (1) − Qb0 m0 (1) + 2 m1 (b0 QΦ) − Qb0 QΦ
= −2 mΦ Φ Φ

1 b0 m0 (1) + 2m1 (L0 Φ) + 2Qb0 m0 (1) − 2L0 QΦ.
= −2mΦ Φ Φ Φ
(7.16)
Using expressions (7.15) and (7.16), sum of two terms (7.13) and (7.14) becomes the RHS
of Eq. (7.12).
Proof of Proposition 7.1. Let Φ(ζ ) be a solution of the RG equation (7.1). We compute
Φ(ζ )
dm0 /dζ as follows:
d Φ(ζ ) d
m0 (1 : ζ ) = mk (Φ(ζ )k : ζ )
dζ dζ
k1
dΦ k ∂mΦ
= mk dζ :ζ + ∂ζ (1 : ζ )
0
k1
Φ(ζ ) dΦ ∂mΦ
= m1 dζ :ζ + ∂ζ (1 : ζ ).
0
(7.17)
We further evaluate Eq. (7.17) by using Lemma 7.1 as follows:
Φ(ζ ) dΦ Φ(ζ )
Eq. (7.17) = m1 dζ : ζ + m1 (L0 Φ : ζ )
Φ(ζ ) Φ(ζ ) Φ(ζ )
− 2m1 b0 m0 (1 : ζ ) : ζ + (L0 − 2b0 Q)m0 (1 : ζ )
Φ(ζ ) Φ(ζ ) Φ(ζ )
= (L0 − 2b0 Q)m0 (1 : ζ ) − m1 b0 m0 (1 : ζ ) : ζ , (7.18)
where we use the RG equation (7.7) to show the last equality. Thus we find out
d Φ(ζ ) Φ(ζ ) Φ(ζ ) Φ(ζ )
m0 (1 : ζ ) = (L0 − 2b0 Q)m0 (1 : ζ ) − m1 b0 m0 (1 : ζ ) : ζ . (7.19)
dζ
Φ (ζ )
Now we consider the particular case of ΦO (ζ ). At ζ0 it satisfies m0 O 0 (1 : ζ0 ) = 0.
Φ(ζ )
The RHS of Eq. (7.19) vanishes there. This means dm0 (1 : ζ )/dζ |ζ =ζ0 = 0. Thus, for
Φ(ζ )
any ζ sufficiently close to ζ0 , we have m0 (1 : ζ ) = 0. For such ζ , again by Eq. (7.19),
Φ(ζ ) Φ(ζ )
m0 (1 : ζ ) vanishes. Repeating this argument, we finally obtain m0 (1 : ζ ) = 0 for
arbitrary ζ .
7.2. Beta functions
Let us express the RG equation (7.1) as a flow equation of the super-coordinates T i . We

first introduce the free energy F as a function of T i and ζ .
Definition 7.2 (Free energy).

F (T , ζ ) ≡ S[Φ : ζ ] Φ= T i e−ζ ∆i φi
. (7.20)
i
Proposition 7.2 (Beta function). The RG equation (7.1) is equivalent to the following
evolution equations of T i :
dT i
= β i (T , ζ ), (7.21)
dζ
where beta functions β i are given by
ij ∂ L F (T , ζ )
β i (T , ζ ) = gren (ζ ) . (7.22)
∂T j
j
ij
Here gren (ζ ) in Eq. (7.22) is the inverse of the renormalized bilinear form, gijren (ζ ) ≡
e−ζ(∆i +∆j ) gij .
Fixed points of the evolution equations (7.21) are zeros of the beta functions. According
to Eq. (7.22) these zeros are solutions of ∂ L F /∂T i = 0 since g ij is non-degenerate on HS .
The free energy F is given by Eq. (7.20). Therefore, zeros of the beta functions are nothing
but classical solutions of open-string field theory.
Perturbative expansions (expansions at T = 0) of the beta functions can be read from
the perturbative expansion of F (T , ζ ). They have the following forms (cf. proof of
Proposition 7.2).
ij
β i (T , ζ ) = ∆i T i + kl (ζ )T T + O(T ).
gren (ζ )Cjren k l 3
(7.23)
j,k,l
Here Cijrenk (ζ ) are the renormalized three points functions, Cijrenk (ζ ) ≡ e−2ζ(∆i +∆j +∆k ) Cij k ,
i )+G(T j )+G(T j )G(T k )
where we put Cij k = (−)G(T 1 2 3|φi 1 |φj 2 |φk 3 . Contrary to our
naive expectation, the beta functions given by Eq. (7.22) depend on the cut-off scale
parameter ζ explicitly. In a quantum field theory, a standard argument to show that beta
functions of the theory have no explicit dependence on the cut-off scale is based on a
simple dimensional analysis. In this argument it is assumed that there is at most only one
dimensionful parameter whatever regularization scheme one chooses. The regularization
we choose for open-string field theory does not satisfy this property. We formulate
open-string field theory from the perspective of two-dimensional chiral CFT. But the
regularization we choose is simply to put a restriction on length of open-string evolution.
It is a regularization of one-dimension. Actually we have two length scales. The missing
scale is length of open-string itself. If we restore string length scale ls in the argument,
there appear two dimensionful parameters ζ ls and ls in our regularization.
Remark 7.1. Dependence of the beta functions on the cut-off scale parameter can be seen
as follows: We put Φ(T , ζ ) ≡ i T i e−ζ ∆i φi . By using the RG equation (7.1) we obtain
the following expression for the beta functions:
Φ(T ,ζ )
β i φi = e ζ L 0 b 0 m 0 (1 : ζ ). (7.24)
i
We partial-differentiate Eq. (7.24) with respect to ζ . By using Lemma 7.1 it turns out to be
∂β i
k1 −ζ L0
φ = −2e b0
i ζ L0
mk1 +k2 +1 Φ , e β φj , Φ : ζ . (7.25)
j k2
∂ζ
i k1 +k2 1 j
This remark or Proposition 7.1 shows that zeros of the beta functions (7.22) are
independent of ζ . Let us suppose that 0 and Tc = (Tci )i (= 0) be zeros of the beta functions
at ζ = 0. Namely, β i (T , ζ = 0)|T =0,Tc = 0. Since they are the zeros even at ζ > 0, the beta
functions must have the following forms:

β i (T , ζ ) = i
βN,M (ζ )T N (T − Tc )M , (7.26)
N,M
where N = (ni )i and M = (mi )i are multi-indices taking values in Z1 . This means that
i
ζ -dependence of the beta functions is absorbed into the coefficients βN,M .
The RG equation of open-string field is originally introduced by Eq. (6.2) in
Definition 6.1. In the Siegel gauge we can write down this equation as the Gell-Mann–
Low equation for the free energy:
∂F (T , ζ ) i ∂ L F (T , ζ )
+ β = 0. (7.27)
∂ζ ∂T i
i
This is an equation which determines the evolution of T i . The RG flow T (ζ ) is a solution

of this equation. By using Eq. (7.22) we can rewrite the Gell-Mann–Low equation (7.27)
as follows:
∂F (T , ζ )
=− β i gijren β j . (7.28)
∂ζ
i,j
By a slight look of this expression besides (7.22) of the beta functions, one may feel a
similarity between the free energy F (7.20) and Zamolodchikov’s c-function. But it is
merely a formal resemblance.6
In a conventional approach to open-string field theory, presumably assuming a suitable
decoupling of the ghost sector, one uses mostly the matter Hilbert space Hmatter rather
than the open-string Hilbert space. We identify Hmatter with a subspace of HS by the map
i : Hmatter → HS , where i(O) ≡ c1O. To present a physical application of the proposition,
we follow the conventional approach although the restriction on Hmatter might cause a
problem on the RG equation. Practically we assume that the RG flow or the ζ -evolution of
Φ is closed on Hmatter .7
The bilinear form g, when restricted on Hmatter , turns out to have the form:

g(c1O, c1O ) = I [ϕO ](∞)ϕO (0) , (7.29)
where the RHS is the standard BPZ pairing of the matter CFT. Eq. (7.29) can be shown
if one computes the LHS by using the oscillator representation (2.23) of the reflector. Let
{OI }I be bases of Hmatter . OI are chosen as the eigenvectors of Lmatter0 : Lmatter
0 OI =
∆I OI . We put gI J ≡ g(c1 OI , c1 OJ ). These define a positive-definite symmetric bilinear
form on Hmatter . Open-string field is now supposed to be a vector of Hmatter . We put
Φ = I T I e−ζ(∆I −1) c1 OI . If one
takes the σ -model viewpoint of the matter theory, Φ
describes a generic perturbation I T I ϕOI of the σ -model and β I are the beta functions
associated with this perturbation. Under this circumstance the free energy F can be thought
as a generating function of all correlation functions of the matter theory. Eq. (7.22) can be
written down as follows:
∂F (T , ζ ) ren I
= gI J β . (7.30)
∂T I
I
Closed-string version of Eq. (7.30) can be found in [28], where it was stated as a conjecture.

Proof of Proposition 7.2. Let us put Φ = i T i e−ζ ∆i φi . We can express the coefficients
T i by T i = eζ ∆i g(φ i , Φ). Then dT i /dζ acquires the following form:

dT i i dΦ
=e ζ ∆i
g φ, + ∆i T i . (7.31)
dζ dζ
We evaluate the RHS of Eq. (7.31) by using the RG Eq. (7.1) as follows:

dT i ∂Sint
=e ζ ∆i i
g φ , b0 + ∆i T i . (7.32)
dζ ∂Φ
6 Nevertheless, possibility of taking classical open-string field action as an analogue of c-function, which one
calls g-function, was discussed in [17]. The argument there was based on the assumption that the free energy
does not depend on ζ explicitly. A discrepancy between the two is also discussed recently in [27].
7 This might be shown by a detailed analysis of the symmetric 3-vertex.
We treat two terms of Eq. (7.32) separately. The bilinear form in the first term can be
expressed as a pairing by ω as follows:

∂Sint (1) (2) ∂Sint
g φ i , b0 = ω12 c0 |φ i 1 b0
∂Φ ∂Φ 2

∂Sint ∂Sint
= (−)K(φ ) ω12 b0(1)c0(1)|φ i 1 = ω φ i ,
i
, (7.33)
∂Φ 2 ∂Φ
where the last equality follows from the gauge condition, b0 φ i = 0 besides the anti-
commutation relation, {c0 , b0 } = 1. We remark that φ i is expressed in terms of Φ as
follows:
∂ LΦ
φi = eζ ∆j g ij . (7.34)
∂T j
j
By using this expression for φ i in Eq. (7.33), it follows from Definition 5.2 that the first
term of Eq. (7.32) becomes

∂Sint ∂L
eζ ∆i g φ i , b0 = eζ(∆i +∆j ) g ij j
Sint [Φ : Λ] Φ= T k e−ζ ∆k φ .
∂Φ ∂T k k
j
(7.35)
Next, we examine the second term of Eq. (7.32). In the Siegel gauge, the BRST charge
Q contributes to ω(Φ, QΦ) as an operator c0 L0 . Thus we have the following identity:
ω(Φ, QΦ) = g(Φ, L0 Φ). (7.36)
This means that the quadratic term of the open-string field action has the following
expression in terms of T i :
1 i −ζ(∆i +∆j )
ω Φ, 12 QΦ Φ= T k e−ζ ∆k φ = T e gij ∆j T j . (7.37)
k k 2
i,j
Then we write down the second term of Eq. (7.32) as a differentiation of ω(Φ, QΦ) by T :
∂L 1
∆i T i = eζ(∆i +∆j ) g ij j
ω Φ, 2 QΦ Φ= T i e−ζ ∆k φ . (7.38)
∂T k k
j
By using the expressions (7.35) and (7.38) we proceed on the computation of Eq. (7.32)
as follows:
dT i ζ(∆i +∆j ) ij ∂ L - .
= e g j
Sint [Φ : ζ ] + ω Φ, 12 QΦ Φ= T i e−ζ ∆k φ
dζ ∂T k k
j
∂L -
i −ζ ∆
.
= eζ(∆i +∆j ) g ij S[Φ : ζ ] Φ= k T e k φk
∂T j
j
∂ L F (T , ζ )
= eζ(∆i +∆j ) g ij . (7.39)
∂T j
j
Thus we obtain Eq. (7.22).
(a) (b)
Fig. 14. (a) Trivalent open-string diagram. (b) The image of trivalent diagram on the upper half-plane.
7.3. Perspective of world-sheet boundary theory
To close this section we hint an interpretation of the previous results in terms of world-
sheet boundary theories. In particular we clarify the role of the cut-off scale parameter ζ .
In the next section we further develop this perspective.
(Tree) open-string diagrams can be mapped holomorphically to the upper half-plane
Im z > 0. These maps are called Mandelstam maps. External open-strings in n-diagram
are mapped to n different points on the real line. The inverse of Mandelstam maps are
given by using quadratic differentials on CP1 . There exists one-to-one correspondence [29]
between the sets of open-string tree diagrams and quadratic differentials on CP1 which
satisfy suitable conditions.
Let us explain the use of quadratic differentials by a simple example. Trivalent open-
string diagram (Fig. 14(a)) can be mapped to the upper half-plane as depicted in Fig. 14(b).
Adding the mirror image we obtain Fig. 2. There CP1 is exactly covered by fi (Vi )
(i = 1, 2, 3). Recall Vi is the unit disk |vi | 1 and fi is the holomorphic map (2.34).
The quadratic differential associated with Fig. 14(b) (or Fig. 2) is
9(1 + z2 )
ϕ= dz⊗2 . (7.40)
z2 (z2 − 3)2
√
It has second-order poles at 0 and ± 3 and first-order zeros at ±i. Also it has the following
expansion in the neighborhood of each pole:

1
ϕ= + · · · dz⊗2 , (7.41)
(z − zP )2
(a) (b)
Fig. 15. (a) The s-channel diagram which appears in ∂V4 (ζ ). (b) The s-channel diagram is mapped
holomorphically to the upper half-plane. The mid-points of the trivalent vertices are mapped to ib and i/b. The
external open-strings 1, 2, 3 and 4 are mapped, respectively, to ±a and ±1/a on the real line.
√ √
where P denotes the pole. Let ϕ (i) be a differential on fi (Vi ) which satisfies ( ϕ (i) )⊗2 =
√ (i)
ϕ. Particularly, we can choose ϕ such that
√ dvi
fi∗ ϕ (i) = . (7.42)
vi
√
We then introduce an analytic function ρi on fi (Vi ) by an integral of ϕ (i) in the following
manner:
z
√ (1)
ρ1 (z) − ρ1 (0) = ϕ . (7.43)
0
Due to the property (7.42) we can regard eρi as a holomorphic map from fi (Vi ) to Vi . It is
the inverse of fi :
eρi ◦ fi = 1. (7.44)
The ith strip in Fig. 14(a) is parametrized by ρi = τi + iσi .
Consider the s-channel diagram which appears in ∂V4 (ζ ). See Fig. 15(a). It can be
mapped to the upper half-plane. The image is depicted in Fig. 15(b). The four open-strings
are mapped to ±a and ±1/a on the real line. The mid-points of the trivalent vertices are
mapped to ib and i/b (0 < a, b < 1). a and b are suitable functions of ζ . These functions
are determined from a consideration on a quadratic differential associated with Fig. 15(b).
The quadratic differential needs to have second-order poles at ±a and ±1/a and
first-order zeros at ±ib and ±i/b. In the neighborhood of each pole it must have the
expansion (7.41). These conditions restrict the quadratic differential to the following form:
4a 2 (a 4 − 1)2 (z2 + b2)(1 + b2 z2 )

ϕ= dz⊗2 . (7.45)
(a 2 + b )(1 + a b ) (z2 − a 2 )2 (1 − a 2 z2 )2
2 2 2
√
Let ϕ be a differential defined locally (in each image of the open-string strip and its
mirror image). We further need to impose the condition which fixes width of the internal
strip to π :

√
ϕ = 2πi. (7.46)
|z|=1
This allows us to solve b in terms of a. Thus obtained quadratic differentials describe open-
string 4-diagrams. Finally, we need to relate a with ζ so that ϕ describes Fig. 15(a). It can
be achieved by imposing the condition
i
√
ϕ = ζ, (7.47)
ib
where the integral is taken along the imaginary axis.

The asymptotic forms of a(ζ ) and b(ζ ) can be written down explicitly. They are
expected to approach to zero as ζ goes to ∞. When a is nearly zero we can solve the
condition (7.46) in the following form:
√
b(a) = 3 a + O(a 2 ). (7.48)
z2 +3a 2
Therefore we approximate the quadratic differential (7.45) by (z2 −a 2 )2
dz⊗2 . The condition
(7.47) will be replaced by the following one:
i √ 2
z + 3a 2
dz = ζ. (7.49)
√
z2 − a 2
i 3a
This integral is easily evaluated and we obtain

2
a(ζ ) = √ e−ζ + O(e−3ζ ). (7.50)
3 3
The approximation used above becomes consistent when a is sufficiently small. It is when
ζ is sufficiently large.
Each strip of an open-string diagram has a Kähler metric dρ d ρ̄ (ρ = τ + iσ is a
holomorphic coordinate of strip). They induce a Kähler metric on the upper half-plane.
We call it bulk metric d 2 sbulk . If one extends the bulk metric to the real line, it becomes
singular at the points where the open-strings are inserted. From the perspective of world-
sheet boundary theories we should rescale the bulk metric near the real line such that the
rescaled metric provides a smooth metric on the real line. World-sheet boundary theory
will be constructed on the real line by using such a smooth metric d 2 sboundary. A convenient
choice of a smooth metric on the boundary is
d 2 sboundary = dx dx. (7.51)

If we use the metric (7.51) on the real line, the regularization employed so far in open-
string field theory corresponds to a point-splitting regularization of short-distance behavior
of a boundary theory. The estimation (7.50) shows that the point-splitting is prescribed by
the boundary length scale δl with the order
δl ∼ e−ζ . (7.52)
The low energy action S[Φ : ζ ] can be understood as an analogue of a generating
function of all correlation functions of a world-sheet boundary theory regularized by the
point-splitting (7.52). ζ needs to be sufficiently large. In the next section we identify it with
the so-called boundary open-string field theory.
8. Boundary open-string field theory
In this section we discuss a relation between two different formulations of classical

open-string field theory. One is the formulation given in the previous sections of this
paper. It is based on the Wilson renormalization group of world-sheet theory in which the
microscopic action is identified with the cubic action (2.50) and the cut-off scale parameter
is introduced by length of open-string strip (which appears in the Schwinger representation
of the open-string propagator). The other is the so-called boundary open-string field theory.
It is proposed in [18] intended for the background independent description of open-string
field theory. To discuss their relation we need to introduce boundary states. These are states
of the closed-string Hilbert space. String vertices which describe interactions of open-
strings with a single closed-string are required. We start this section by a consideration
of the moduli problem relevant to constructions of these vertices. Our discussion in this
section is not rigorous although it is physically motivated. We believe that perspectives
presented in this section become useful in our understanding of duality of open- and closed-
strings or holography in string theory.
Let M/1n be the set of clockwise-ordered n different points (z1 , . . . , zn ) on the boundary
of one-punctured two-disk. Let z0 be the puncture.

/ zi ∈ ∂D (1 i n), z0 ∈ D,
Mn = (z0 ; z1 , . . . , zn )
1
. (8.1)
zi = zj if i = j
We also let M/0n be the set of clockwise-ordered n different points (z1 , . . . , zn ) on the
boundary of two-disk:

/ zi ∈ ∂D (1 i n),
Mn = (z1 , . . . , zn )
0
. (8.2)
zi = zj if i = j
/1n /SL2 (R) for n 1,
SL2 (R) acts on the both sets in the standard manner. We put M1n ≡ M
0 /
and Mn ≡ Mn /SL2 (R) for n 3. In the previous sections we denote M0n by M∂n . We can
0
Fig. 16. A typical configuration at the infinities M1(I,s) × M0(s,J ) of M1n .
always put the puncture at the origin of two-disk by using the SL2 (R). Thereby M1n is
identified with M/0n /U (1), where U (1) is the rotation group of ∂D.
To describe infinities of M1n , we introduce a pair of multi-indices (I, J ), where the
multi-indices I = (i1 , . . . , ip ) (p 0) and J = (j1 , . . . , jq ) (q 2) satisfy the conditions:
I ∩ J = ∅ and (i1 , . . . , ip , j1 , . . . , jq ) = (1, . . . , n) mod cyclic permutations. At infinities
of M1n there appear M1(I,s) × M0(s,J ) for all the pairs (I, J ). “s” denotes the singular point
in the configuration at the infinities. See Fig. 16. Stable compactification of M1n , which we
call CM1n , is defined inductively by adding CM1(I,s) × CM0(s,J ) to the infinities of M1n .8
0
CMn = Mn ∪
1 1
CM(I,s) × CM(s,J ) ,
1 0
(8.3)
(I,J )
where CM11 ≡ M11is a point. Topologically CM1n becomes a (n − 1)-dimensional ball

/n0 /U (1). We fix orientation
Bn−1 . This can be seen from the identification of M1n with M
of CMn by the standard orientation of Bn−1 . Taking account of the orientations we obtain
1

∂CM1n = (±)CM1(I,s) × CM0(s,J ). (8.4)
(I,J )
The signature (±) must be determined by a comparison of the orientations of ∂CM1n and
CM1(I,s) × CM0(s,J ). Later we will discuss about this problem.
In the previous sections we introduced Vn0 (ζ ) ≡ Vn (ζ ), which is the subset of CM0n ,
in order to obtain open-string vertices at the cut-off scale ζ . Analogous consideration
is also possible at the present situation: we identify M1n with M /n0 /U (1). Consider the
configurations near the infinities. In these configurations some of n points (on ∂D) are
getting “close” to one another. We can measure how close they are by length of open-
string strips. Length of open-string strips provide a local coordinates at least near the
boundary of CM1n . If we put a restriction on their length by the scale parameter ζ , the
set of forbidden configurations provides a neighborhood of the infinities. Let us denote this
8 CM0 ∂
(s,J ) is CM(s,J ) in the previous notation.
neighborhood by Pn1 (ζ ). We put Vn1 (ζ ) ≡ CM1n \ Pn1 (ζ ). This plays the same role as Vn0 (ζ )
in the construction of string vertex. Vn1 (ζ ) is topologically Bn−1 . Thus obtained Vn1 (ζ ) for
ζ > 0 satisfies

∂Vn1 (ζ ) = 1
(±)V(I,s) (ζ ) × V(s,J
0
) (ζ ). (8.5)
(I,J )
This equation plays an important role of our understanding of boundary states.

Boundary states are introduced as states of closed-string. For their description we
prepare some terminology of closed-string field theory. We follow the convention used
in [2]. Let Hmatter
c and Hghost
c be respectively the matter and ghost Hilbert spaces of
critical bosonic string. We put Haux ≡ Hmatter c ⊗ Hghost
c . The SL2 -invariant vacuum |0
of the auxiliary Hilbert space Haux is set to have no ghost number. The SL2 -invariant
vacuum 0| of the dual Hilbert space Haux ∗ is also set to have no ghost number. Both
states are grassmann-even. Dual pairing between Haux and Haux ∗ is prescribed based on
0|c−1 c̄−1 c0 c̄0 c1 c̄1 |0 = 1, where c±1,0 and c̄±1,0 are, respectively, the ghost zero modes
of chiral and anti-chiral parts on CP1 . Closed-string Hilbert space Hc consists of vectors
ψ of Haux which satisfy the conditions, b0− ψ = L− 0 ψ = 0. Closed-string field Ψ is a
9
grassmann-even vector of Hc and has the ghost number G(Ψ ) = 2.

Closed-string Hilbert space has an odd symplectic structure ωc . We write it in the form,
ω (A, B) = ω12
c c |A |B for A, B ∈ H . ωc | is an element of (H⊗2 )∗ . It is defined
1 2 c 12 c
by the restriction of R12 c |c −(2) on H . Here R c | is the reflector of H
0 c 12 aux . It is given
by using the BPZ pairing as follows. R12 |A1 |B2 ≡ A |B for any A, B ∈ Haux . The
c T
reflector is a grassmann-even vector and has the ghost number equal to six. Hence the odd
symplectic form ω12 c | is a grassmann-odd vector and has the ghost number equal to seven.
The selection rule of ωc becomes: ωc (A, B) = 0 ⇒ G(A) + G(B) = 5. The inverse of the
odd symplectic form is called sewing ket. We denote it by |S12 c
. It is an element of Hc⊗2
and satisfies ω12 |S23 =3 P1 . The sewing ket is grassmann-odd and has the ghost number
c c
equal to five.
We want to gain ultimately a field theory which describes interactions of open- and
closed-strings. If such a field theory consistently exists, we need to have all the string
vertices for these interactions. Recent attempts in this direction can be found in [23,30].
Nevertheless the construction is still beyond completion. One of the reasons is that we
still do not have a suitable physical perspective of closed-strings by open-strings or open-
strings by closed-strings. We also wish to find such a physical perspective in the following
discussion. Our discussion is based on several assumptions. They are physically reasonable
and acceptable if a field theory of interacting open- and closed-strings exists consistently.
We assume an existence of string vertices 0c ; 1 . . . n : ζ | for n 1. Here the index 0c
represents a closed-string on D and the indices i (1 i n) label clockwise-ordered n
open-strings on ∂D. The vertices describe interactions between a single closed-string and
n open-strings at the cut-off scale ζ . If one takes the closed-string picture, an incoming
closed-string is reflected at ∂D and at the same time decaying into open-strings. The
interaction term 0c ; 1 . . . n : ζ |Ψ 0c |Φ1 · · · |Φn should not vanish. Recalling the ghost
9 We put L± ≡ L ± L̄ , b± ≡ b ± b̄ and c± ≡ (c ± c̄ )/2.

0 0 0 0 0 0 0 0 0
numbers G(Ψ ) = 2 and G(Φ) = 1, this requires that the ghost number of the vertex
0c ; 1 . . . n : ζ | is equal to n + 4. Also for the non-vanishing, odd-grassmannity of Φ
requires the following cyclic asymmetry with respect to the open-string indices:

0c ; 1 2 . . . n − 1 n : ζ = (−)n+1 0c ; 2 3 . . . n 1 : ζ . (8.6)
Explicit construction of the vertices was examined in [31] for the cases of n = 1, 2
by a slightly different formulation. In our formulation it will be generalized based on a
conjectural map (or a section of the Hilbert bundle on CM1n ),
∗
CM1n −→ Hc × Ho⊗n
. (8.7)
I → I|
This map gives rise to a (Hc × Ho⊗n )∗ -valued (n − 1)-form Ω| on CM1n . The vertex will
be defined as an integration of thus obtained Ω| on Vn1 (ζ ):

n
(i)
0c ; 1 . . . n : ζ = Ω|(1 ... n) e−ζ L0 . (8.8)
i=1
Vn1 (ζ )
As is the case of open-string field theory, action of the BRST charge on the vertices is
expected to be a representation of the boundary operator ∂. Since it is an interacting theory
of closed- and open-strings, the BRST charge in question should be a sum of the closed-
string BRST charge Qc and the open-string BRST charge Qo . Hence we arrive at the
following conjectural action of the BRS charge:

n
0c ; 1 . . . n : ζ Q c +
c
(0 )
Q(i)
o = (±) 0c ; I, a : ζ a , J : ζ S o ,
aa (8.9)
i=1 (I,J )
where we denote the open-string inverse reflector by |Sao a .
8.1. Boundary states
So far, our consideration of the string vertices in this section excludes the case of n = 0.
In this particular case we have the state 0c | ∈ Hc∗ . (It is different from the SL2 -invariant
vacuum 0|.) This is closed-string vertex which describes a simple reflection of a single
closed-string at ∂D. It is a BRST invariant state and has the ghost number four. If we take
the dual by using the sewing ket of closed-string, it acquires a familiar form [32]:
c
B ≡ 0 c S
0 c ∗c . (8.10)
This is a boundary state of closed-string and is called the Ishibashi state. It satisfies

Qc B = 0, G B = 3. (8.11)
Due to the correct ghost number we have the non-vanishing coupling ωc (Ψ, |B) with
closed-string field. Strictly speaking, the Ishibashi state is not a state of Hc because its
norm turns out to be divergent. Without any contradiction there is no local field operator
in 2D CFT which corresponds to this state. For a suitable class of 2D CFT, physically
acceptable bases of solutions of Qc |B = 0 are found in [33] and called the Cardy bases.
Roughly speaking, they are characterized by the reflection conditions of closed-string at
∂D. Of course it can be rephrased as the boundary conditions of open-strings when they
interact with the closed-string via the vertex (8.8). For simplicity we impose the Neumann
condition on the reflection of closed-string.
Let us consider the following boundary state in the presence of an open-string field Φ:

+(∗ ) 1 c
B[Φ : ζ ] ≡ e−ζ L0 c 0c S c + 0 ; 1 . . . k : ζ S |Φ · · · |Φk .
0 c ∗c c 0 c ∗c 1
k
k1
(8.12)
The ghost number of this state becomes three. This means that we have a non-vanishing
coupling with closed-string field Ψ of the form, ωc (Ψ, |B[Φ : ζ ]). If one takes the closed-
string picture, it describes an interaction of a closed-string with a reservoir of open-strings.
Due to this interaction or a possible decay to open-strings the boundary state (8.12) is not
invariant under the action of the closed-string BRST charge Qc . Our first investigation is
about an interpretation of this action from the open-string viewpoint.
8.2. Role of Qc in open-string field theory
Let us examine the boundary state (8.12) from the open-string picture. We regard the
boundary state as a Hc -valued function of Φ. Then the corresponding hamiltonian vector
|∂B[Φ]/∂Φ is an element of Hc × Ho . It is given by the variational formula (5.4),

∂B[Φ : ζ ]
δ B[Φ : ζ ] = ωo δΦ, , (8.13)
∂Φ
where we denote the open-string odd symplectic structure by ωo . After a little calculation
we find out the following expression of the hamiltonian vector:

∂B[Φ : ζ ] +(∗c ) o
= e−ζ L0 0c ; 1 . . . k − 1 k : ζ S0cc ∗c |Φ1 · · · |Φk−1 Sk∗ .
∂Φ o
k1
(8.14)
We now compute action of the closed-string BRST charge on the boundary state (8.12).
By a simple evaluation it can be rewritten in the following form:

∂B[Φ : ζ ]
Qc B[Φ : ζ ] = ωo Qo Φ,
∂Φ
+(∗c ) (−)k+1
+ e−ζ L0 0c ; 1 . . . k : ζ
k
k1

k

(i) c
× Q (0c )
c+ Q oS |Φ1 · · · |Φk .
0 c ∗c (8.15)
i=1
To proceed computation of the second term of Eq. (8.15) we need to use the conjectural
form (8.9). The result is a suitable weighted sum of the quantities,
+(f∗c )
ωo mp (Φ p : ζ ), e−ζ L0 0c ; 1 . . . q 0o : ζ S0cc ∗c |Φ1 · · · |Φq S0oo ∗o , (8.16)
for p 2 and q 0. Their weights are determined by the signatures in Eq. (8.9) or
equivalently in Eq. (8.5). So our computation may stop here. What we can learn from
Eq. (8.16) is the appearance of the A∞ -algebra. If one recalls that hamiltonian vector of the
classical open-string field action (5.16) is the sum of mk , one may infer that the weighted
sum eventually gives rise to the following result:

∂S[Φ : ζ ] ∂B[Φ : ζ ]
Qc B[Φ : ζ ] = ωo , , (8.17)
∂Φ ∂Φ
where |∂S[Φ : ζ ]/∂Φ is the hamiltonian vector given by (5.22).
Eq. (8.17) must be tested. First of all, the above Qc -action needs to be nilpotent,
Qc (Qc |B[Φ]) = 0. Let us show this: by using the open-string anti-bracket we rewrite
the RHS of Eq. (8.17) as {S[Φ], |B[Φ]}. We then compute Qc (Qc |B[Φ]) as follows:

Qc Qc |B[Φ] = (−)Qc |B[Φ], S[Φ] = (−) Qc |B[Φ], S[Φ]

= |B[Φ], S[Φ] , S[Φ]
1
=− S[Φ], S[Φ] , |B[Φ] , (8.18)
2
where we use the Jacobi identity (5.15) to show the last equality. Since we have the classical
master equation {S, S} = 0 by Proposition 5.4, Eq. (8.18) vanishes identically. Another
consistency may be obtained if we can find the signatures in Eq. (8.9) so that they provide
a representation of the boundary operator ∂ and also give rise to Eq. (8.17). As such a
solution we finally find out the following ones:

n
0c ; 1 . . . n : ζ Qc
(0 c )
+ Q
o
(i)
i=1

n−2
n

= + l − 1 a( : ζ
(−)k(n+1)+(l+1)n 0c ; k% . . . k &'
l=0 k=1 l+1

. k + n − 1( : ζ Sao a .
× %a k + l . .&' (8.19)
n+1−l
These are consistent with the asymmetry (8.6).
Eq. (8.17) itself was alluded previously from a different perspective in [23,31]. These
authors intended to construct new symmetries of open-string field theory. Their argument
is as follows. Let ψ ∈ Hc be a BRST-closed state (Qc ψ = 0) with G(ψ) = 2. Associated
with such a state we can introduce a non-vanishing function Fψ (Φ) ≡ ωc (ψ, |B(Φ)).
Then, since Qc ψ vanishes, Eq. (8.17) implies {S[Φ], Fψ (Φ)} = 0. This means that Fψ
generates a symmetry of the theory. As a trivial example we can consider the massless
U (1) gauge field aµ (of open-string) in presence of the massless anti-symmetric tensor
field bµν (of closed-string). In this simple case infinitesimal transformations, δaµ = ηµ
and δbµν = ∂µ ην − ∂ν ηµ , correspond to a symmetry in question. ηµ is identified with a
suitable component of ψ = Qc η.
Significance of the Qc -action (8.17) should be stressed. The BRST transformation δBRS
of open-string field Φ is defined by Eq. (5.31). It is just the hamiltonian vector of S[Φ : ζ ].
By using the variational formula (5.4) we can rewrite the Qc -action into

Qc B[Φ : ζ ] = δBRS B[Φ : ζ ] . (8.20)
Thus the closed-string BRST charge induces the BRST transformation of open-string field.
Eq. (8.20) becomes a key in the construction of boundary open-string field theory.
8.3. Boundary open-string field theory
A framework for background independent open-string field theory was proposed in

[18] based on the BV formalism. It is conventionally called boundary open-string field
theory. This formulation was further investigated in [34–36]. Construction based on the
BV formalism requires a triple ((X , ω), V ). X is a super-manifold. ω is an odd symplectic
structure of X and V is a nilpotent fermionic vector field on X . Having such a triple, the
BV action, if it exists, is obtained as the hamiltonian function of V . The nilpotency of
V ensures the classical master equation. And the BRST transformation of the theory is
given by V . In order to construct boundary open-string field theory, a hypothetical “space
of all open-string world-sheet theories” is taken [18] as X . ω is determined by correlation
functions of world-sheet theories. V is practically identified with the closed-string BRST
charge Qc .
We want to interpret the macroscopic open-string field theory presented in this paper as
a boundary open-string field theory.
Our formulation is based on the Wilson renormalization group. We start at the trivial
classical solution Φ = 0. This solution describes a flat Minkowski space R(1,25) and is a
conformal point of the above X . We regards it as a UV fixed point. The open-string Hilbert
space H used in this paper is the tangent of X at Φ = 0. A local coordinate patch of X
centered at this point will be given by the tangent space via the exponential map. Our naive
expectation is that, excluding the critical points, theory at Φ = 0 on this patch is described
by the classical action S[Φ : ζ ] (or the fluctuation given in Remark 5.2). To pursue this
naive picture we need to interpret the scale parameter ζ correctly.
The regularization employed in this paper corresponds to a point-splitting regularization
of short-distance on the boundary when ζ is sufficiently large. The point-splitting is
prescribed by the boundary length scale a ∼ e−ζ measured by the boundary metric dx dx.
Consider a space of all open-string world-sheet cut-off theories which cut-off length scale
is a. The tangent space at Φ = 0, which we denote by Y(a), is identified with the open-
string Hilbert space H. The RG flow in the previous sections defines a map

Rζ : Y(e−ζ ) −→ Y(e−ζ ),
ζ
(8.21)
where ζ > ζ . The previous construction of the RG flow is the reverse of the above map.
This is due to the different interpretation of the scale parameter ζ . On the boundary e−ζ

plays the role of the short distance cut-off. Both spaces Y(e−ζ ) and Y(e−ζ ) are identified
with H. The RG flow becomes the flow on H generated by

= − b0 + Qb0 Φ . (8.22)
dζ ∂Φ
The classical action S[Φ : ζ ] defines a boundary open-string field theory constructed
on Y(e−ζ ). By Proposition 5.6, the action S[Φ : ζ ] satisfies δS[Φ : ζ ] = ω(δΦ, δBRS Φ).
Thus it is the hamiltonian function of δBRS . We now regards δBRS as a nilpotent fermionic
vector field on Y(e−ζ ). As we find in Eq. (8.20), it can be identified with the closed-string
BRST charge Qc . Therefore, geometrical ingredients of both theories become same.
In order to define a “space of all open-string world-sheet theories”, we first need to take a
continuum limit by letting ζ → ∞. The existence of infinitely many irrelevant operators10
causes serious problems. The hypothetical space X is supposed to be as follows in the
original description [18]. Let I0 be a two-dimensional Lagrangian (on D) which describes
a fixed bulk closed-string background. One considers two-dimensional Lagrangians of the
form
I = I0 + I∂D , (8.23)
where I∂D is a suitable boundary term describing the coupling to external open-strings,

I∂D = dθ V(X, b, c), (8.24)
∂D
where dθ is the standard line element of S 1 . The space X is roughly introduced as the
space of Lagrangians I with I0 fixed and I∂D allowed to vary. In order to obtain X as such,
one needs to define I∂D correctly by a suitable regularization and then take a continuum
limit. In [36,37] such a prescription was partially examined. It is amusing to reconsider the
results of [36,37] from the perspective of our formulation.
Acknowledgements
The interpretation as a boundary string field theory in the last section benefitted
from discussions with A. Tsuchiya, A. Fujii, K. Higashijima and K. Murakami. The
interpretation is modified to some extent including the addition of Subsection 7.3 in the
revised version. It benefitted from discussions with H. Kawai. We also thank T. Kugo and
T. Eguchi. We are grateful to T. Kubota and N. Yokoi for useful discussions at the early
stage of this work, and to T. Kimura for a helpful instruction to draw figures in the text.
Finally, we thank the hospitality of the Lawrence Berkely National Laboratory, where some
of this work was done.
Appendix A
In this appendix we give a proof of Proposition 3.5 presented in the text. We first
compute the LHS of Eq. (3.39) by using Eq. (3.37) as follows:
n n

1 2 ... n : ζ Q(i)
Q (i)
i=1 i=1
10 States φ of HS which satisfy ∆ > 0.

i i
1
n n−3

=− k + l a( : ζ
(−)(n+1)(k+l+1) k% . . . &'
2
k=1 l=1 l+2
n

. . . k + n − 1( : ζ Sa a
× a% k + l + 1 &' Q(j )
n−l j =1

1
n n−3 k+l

= (−)(n+1)(k+l+1)+n−l k ... k + l a : ζ Q +
(a)
Q(j )
2
k=1 l=1 j =k

× a k + l + 1 . . . k + n − 1 : ζ Sa a (A.1)
1
n n−3

+ (−)(n+1)(k+l+1) k . . . k + l a : ζ
2
k=1 l=1

k+n−1

× a k + l + 1 . . . k + n − 1 : ζ Q(a ) + Q (j ) Sa a ,
j =k+l+1
(A.2)
where we use the BRST invariance (2.28) of the inverse reflector to show the last
equality. We rewrite (A.1) by replacing two vertices in the equation. Taking account of
the asymmetries (3.38) and the grassmannities, it becomes as follows:
1
n n−3

(A.1) = . . k + n − 1( : ζ
(−)(n+1)k k% + l + 1 .&'
2
k=1 l=1 n−l

)
k+l

× a
k . . . k + l : ζ Q +
(a
Q(j ) Sa a . (A.3)
% &' (
l+2 j =k
After a slight change of the open-string indices Eq. (A.3) turns out to be the second
term (A.2). Thus we obtain the following expression for the LHS of (3.39).
n n

1 2 ... n : ζ Q(i) Q(i)
i=1 i=1

n
n−3

= k + l a( : ζ
(−)(n+1)(k+l+1) k% . . . &'
k=1 l=1 l+2

k+n−1

. . . k + n − 1( : ζ Q(a ) +
× a% k + l + 1 &' Q(j ) Sa a .
n−l j =k+l+1
(A.4)
In order to compute the action of the BRST charge in the RHS of Eq. (A.4), the
following lemma becomes useful:
Lemma 8.1. The action (3.37) of the BRST charge can be written as follows:
n

n−3 n−l−1

1 ... n : ζ Q(i)
=− (−)(n+1)(k+1) k% + l + 2 . . .&'n 1 2 . . . k a( : ζ
i=1 l=1 k=1 n−l

. . k + l + 1( : ζ Sa a .
× %a k + 1 . &' (A.5)
l+2
We omit the proof of this lemma. By using this lemma our computation goes as follows:

n
n−3

Eq. (A.4) = k + l a( : ζ
(−)(n+1)(k+l+1) k% . . . &'
k=1 l=1 l+2
n−l−3
−1
n−l−l
× (−)(n−l+1)(p+1)+1
l =1 p=1

× k + l + l + p + 1 . . . a . . . k + l + p − 1 b : ζ
% &' (
n−l−l

× b k + l + p . . . k + l + l + p : ζ Sb b Sa a

% &' (
l +2

n n−l−l −1

n−3 n−l−3

= k + l a( : ζ
(−)(n+1)(k+p+l)+l(l+p)+1 k% . . . &'
k=1 l=1 l =1 p=1 l+2

× k + l + l + p + 1 . . . a . . . k + l + p − 1 b : ζ

% &' (
n−l−l

× b k + l + p . . . k + l + l + p : ζ Sb b Sa a .

(A.6)
% &' (
l +2
Terms appearing in Eq. (A.6) are conveniently depicted in Fig. 17(a).

We want to show that Eq. (A.6) vanishes identically. For this purpose we exchange
the first and third vertices in Eq. (A.6) taking account of the grassmannities, and then
rewrite Eq. (A.6) by using the asymmetries (3.38) of the vertices. We obtain the following
expression for Eq. (A.6):

n
n−3 n−l−3 −1
n−l−l

Eq. (A.6) = (−)(n+1)(k+l )+l (l+l +p)
k=1 l=1 l =1 p=1

× k + p + l . . . k + p + l + l a : ζ k% + l + 1 . . . a&'

. . . k + n − 1 b( : ζ
% &' (
l +2 n−l−l

. . k + (l : ζ Sb b Sa a .
× %b k . &' (A.7)
l+2
(a)
(b)
Fig. 17. (a) Open-string diagram which appears in Eq. (A.6). (b) Open-string diagram which appears in Eq. (A.7).
Terms appearing in Eq. (A.7) are conveniently depicted in Fig. 17(b).

To compare two expressions (A.6) and (A.7) of the LHS of Eq. (3.39), we replace
the open-string indices in Eq. (A.7) so that they match with those in Eq. (A.6). We
first exchange l ↔ l and then put k ≡ k + p + l and p ≡ n − p − l − l . With these
replacements Eq. (A.7) becomes as follows:

n
n−3 n−l−3 −1
n−l−l

Eq. (A.7) = (−)(n+1)(k+l)+l(l+l +p)
k=1 l=1 l =1 p=1

× k + p + l . . . k + p + l + l a : ζ %k + l + 1 . . . a&'

. . . k + n − 1 b( : ζ
% &' (
l+2 n−l−l

. k + l( : ζ Sb b Sa a
× %b k . .&'
l +2

n
n−3 n−l−3 −1
n−l−l

= k + l a( : ζ
(−)(n+1)(k +p +l)+l(l+p ) k% . . . &'
k =1 l=1 l =1 p =1 l+2

× k + l + l + p + 1 . . . a . . . k + l + p − 1 b : ζ
% &' (
n−l−l

× b k + l + p . . . k + l + l + p : ζ Sb b Sa a . (A.8)
% &' (
l +2
If one compares this expression with Eq. (A.6), one finds that it is equal to (−1)×
Eq. (A.6). This means that Eq. (A.6)= (−1)× Eq. (A.6). Therefore Eq. (A.6) must vanish.
We complete the proof of Proposition 3.5.
References
[1] E. Witten, Non-commutative geometry and string field theory, Nucl. Phys. B 268 (1986) 253.
[2] B. Zwiebach, Closed string field theory: quantum action and the B-V master equation, Nucl. Phys. B 390
(1993) 333, hep-th/9206084.
[3] T. Kugo, H. Kunitomo, K. Suehiro, Non-polynomial closed string field theory, Phys. Lett. B 226 (1989) 48.
[4] J. Stasheff, Homotopy associativity of H spaces I, Trans. Am. Math. Soc. 108 (1963) 275;
J. Stasheff, Homotopy associativity of H spaces II, Trans. Am. Math. Soc. 108 (1963) 293.
[5] K. Fukaya, Floer homology and mirror symmetry I, preprint Kyoto-Math 99-15;
K. Fukaya, Symplectic Geometry, Iwanami-shoten, Tokyo, 2000.
[6] K. Fukaya, Y.-C. Oh, H. Ohta, K. Ono, Lagrangian intersection floer theory: anomaly and obstruction,
preprint Kyoto-Math 00-17.
[7] I.A. Batalin, G.A. Vilkovisky, Quantization of gauge theories with linearly dependent generators, Phys. Rev.
D 28 (1983) 2567;
M. Henneaux, C. Teitelboim, Quantization of Gauge Systems, Princeton Univ. Press, New Jersey, 1992.
[8] C. Thorn, Perturbation theory for quantized string fields, Nucl. Phys. B 287 (1987) 61;
C. Thorn, String field theory, Phys. Rep. 174 (1989) 1.
[9] M. Bochicchio, Gauge fixing for the field theory of the bosonic string, Phys. Lett. B 193 (1987) 31.
[10] K.G. Wilson, J. Kogut, The renormalization group and the epsilon expansion, Phys. Rep. 12 (1974) 75,
Section 11.
[11] J. Polchinski, Nucl. Phys. B 231 (1984) 269.
[12] R. Brustein, S.P. De Alwis, Renormalization group equation and non-perturbative effects in string-field teory,
Nucl. Phys. B 352 (1991) 451;
R. Brustein, K. Roland, Space–time versus world-sheet renormalization group equation in string theory,
Nucl. Phys. B 372 (1992) 201.
[13] A. LeClair, M.E. Peskin, C.R. Preitschopf, String field theory on the conformal plane (I). Kinematical
principles, Nucl. Phys. B 317 (1989) 411.
[14] A. Sen, Descent relations among bosonic D-branes, Int. J. Mod. Phys. A 14 (1999) 4061, hep-th/9902105.
[15] A. Sen, B. Zwiebach, Tachyon condensation in string field theory, JHEP 0003 (2000) 002, hep-th/9912249.
[16] J.A. Harvey, D. Kutasov, E. Martinec, hep-th/0003101.
[17] D. Kutasov, M. Mariño, G. Moore, Some exact results on tachyon condensation in string field theory, hep-
th/0009148;
A.A. Gerasimov, S.L. Shatashvili, JHEP 0010 (2000) 034, hep-th/009103.
[18] E. Witten, On background-independent open-string field theory, Phys. Rev. D 46 (1992) 5467.
[19] D.J. Gross, A. Jevicki, Nucl. Phys. B 283 (1987) 1;
D.J. Gross, A. Jevicki, Nucl. Phys. B 287 (1987) 225;
K. Suehiro, Operator expression of Witten’s superstring vertex, Nucl. Phys. B 296 (1988) 333.
[20] T. Asakawa, T. Kugo, T. Takahashi, On the generalized gluing and resmoothing theorem, Prog. Theor.
Phys. 100 (1998) 437, hep-th/9805119.
[21] J. Khoury, H. Verlinde, On open/closed string duality, hep-th/0001056.
[22] E. Witten, Chern–Simons gauge theory as a string theory, in: The Floer Memorial Volume, Birkhäuser, 1995.
[23] B. Zwiebach, Oriented open–closed string theory revisited, hep-th/9705241.
[24] E. Witten, A note on the antibracket formalism, Mod. Phys. Lett. A 5 (1990) 487;
A. Schwarz, Geometry of Batalin–Vilkovisky quantization, hep-th/9210115.
[25] T. Kubota, G. Veneziano, Off-shell effective actions in string theory, Phys. Lett. B 207 (1988) 419.
[26] H. Hata, B. Zwiebach, Developing the covariant Batalin–Vilkovisky approach to string theory, Ann.
Phys. 229 (1994) 177, hep-th/9301097.
[27] A. Fujii, H. Itoyama, Some computation on g function and disc partition function and boundary string field
theory, hep-th/0105247.
[28] A.M. Polyakov, Gauge Fields and Strings, Harwood Academic, 1987.
[29] K. Streble, Quadratic Differentials, Springer, Berlin, 1984.
[30] T. Kugo, T. Takahashi, Unoriented open-closed string field theory, hep-th/9711100;
T. Asakawa, T. Kugo, T. Takahashi, BRS invariance of unoriented open-closed string field theory, hep-
th/9807066.
[31] H. Hata, M.M. Nojiri, New symmetry in covariant open-string field theory, Phys. Rev. D 36 (1987) 1193.
[32] K. Hashimoto, H. Hata, D-brane and gauge invariance in closed string field theory, Phys. Rev. D 56 (1997)
5179, hep-th/9704125.
[33] J. Cardy, Boundary conditions, fusion rules and the Verlinde formula, Nucl. Phys. B 324 (1989) 581.
[34] E. Witten, Some computations in background-independent off-shell string theory, Phys. Rev. D 47 (1993)
3405, hep-th/9210065.
[35] S. Shatashvili, On the problems with background independence in string theory, hep-th/9311177.
[36] K. Li, E. Witten, Role of short distance behavior in off-shell open-string field theory, Phys. Rev. D 48 (1993)
853, hep-th/9303067.
[37] W. Kummer, D.V. Vassilevich, Renormalizability of the open-string sigma model and emergence of D-
branes, hep-th/0006108.
Rotating deformations of AdS3 × S 3 , the orbifold

CFT and strings in the pp-wave limit
Oleg Lunin, Samir D. Mathur
Department of Physics, The Ohio State University, Columbus, OH 43210, USA
Received 19 July 2002; accepted 5 August 2002
Abstract
We construct an exact metric which at short distances is the metric of massless particles in 5+1
spacetime (moving along a diameter of the sphere) and is AdS3 × S 3 at infinity. We also consider
a set of a conical defect spacetimes which are locally AdS3 × S 3 and have the masses and charges
of a special set of chiral primaries of the dual orbifold CFT. We find that excitation energies for a
scalar field in the latter geometries agree exactly with the excitations in the corresponding CFT state
created by twist operators: redshift in the geometry reproduces ‘long circle’ physics in the CFT. We
propose a map of string states in AdS3 × S 3 × T 4 to states in the orbifold CFT, analogous to the
recently discovered map for AdS5 × S 5 . The vibrations of the string can be pictured as oscillations
of a Fermi sea in the CFT.
1. Introduction
String theory on AdS5 × S 5 is believed to be dual to N = 4 Super-Yang–Mills gauge

theory on the boundary of AdS5 [1]. Weakly coupled gauge theory is dual to a string
background that is not well described by perturbative supergravity. Nevertheless it is
possible to demonstrate several relations between the two dual theories.
String theory on AdS3 × S 3 × M4 is expected to be dual to a 2-dimensional CFT
on the boundary of AdS3 . It has been conjectured that this CFT can be represented as
a deformation of a 2-dimensional sigma model with the orbifold target space M4N /SN
[2]. (Here SN is the permutation group for N variables.) The analogue of free Yang–
Mills would be the ‘orbifold point’ where we have the orbifold with no deformation. This
E-mail address: lunin@pacific.mps.ohio-state.edu (O. Lunin).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 7 7 - 6
92 O. Lunin, S.D. Mathur / Nuclear Physics B 642 (2002) 91–113
orbifold model has given several exact agreements with the dual theory for both BPS and
near-BPS quantities.
An essential feature of the orbifold theory is the existence of ‘twist operators’ σn . The
CFT is described by N copies of a c = 6 CFT. The twist σn links together n of these copies
to yield a c = 6 CFT living on a circle that is n times longer than the circle on which the
initial CFT was defined. The excitations of the CFT on this long circle can have very low
energy if n is large.
In this paper we study two related questions that arise in this duality map. A chiral
primary in the CFT is described by a twist σn which also carries a charge that makes
h = j . We call the resulting chiral primary σn−− . Consider the state where all the N copies
of the CFT are twisted so as to make the state (σn−− )N/n |0NS . Since all copies of the CFT
now live on ‘long circles’ we must have low energy excitations of the CFT, and we must
find that in the dual gravity theory the supergravity fields must have corresponding low
energy eigenmodes.
For the CFT in the Ramond (R) sector this issue was studied in [3], where supergravity
duals were constructed for all the R ground states—these states are in one-to-one
correspondence with the chiral primaries of the NS sector which in turn are of the form
σn−−
1
σn−−
2
· · · σn−−
k
. It was found that the length of the CFT circles was described in the
supergravity dual by the depth of the ‘throat’ of the near horizon geometry. The simplest
geometries were dual to configurations of the type (σn−− )N/n , and for these geometries
the travel time around the CFT ‘long circle’ agreed exactly with the travel time down the
throat of the gravity solution.
In our present study of the NS sector we, therefore, look for the gravity duals of the
CFT states (σn−− )N/n |0NS . We solve the wave-equation in the geometries, and find energy
levels that agree exactly with the energy levels for the CFT with ‘long circles’.
The other question relates to the recent proposal by Berenstein, Maldacena and Nastase
[4] that strings moving fast on the S 5 in AdS5 × S 5 can be described easily in the dual gauge
theory. A supergravity quantum moving fast on the S 3 in AdS3 × S 3 × M4 is described by
a chiral primary σn−− (or its SUSY descendent) with a large value of n. We note that σn−−
can be written in the form (σ2−− )n , and use the ‘bits’ σ2−− , J0− σ2−− , J¯0− σ2−− as analogs of
the Z, Xi of [4] to construct the low lying string excitations of the quantum.
In more detail we do the following:
(a) The chiral primary σn−− in the CFT is described by a supergravity quantum having
angular momentum n on the S 3 . Thus the state (σn−− )N/n |0NS should be described by a
collection of such quanta. If n 1 the quanta will be confined to a narrow width around
the circle of rotation, and if we also have N/n 1 then the number of quanta will be large
and the resulting geometry should be well described by a supergravity solution.
In flat space the metric produced by a massless particle is the Aichelburg–Sexl metric
[5], and we can readily extend this to describe a set of particles uniformly distributed along
the line of motion. We construct an exact solution that goes over to the analogue of the
Aichelburg–Sexl solution in 5 + 1 spacetime near the moving particles, and to AdS3 × S 3
at infinity. Just as in the Aichelburg–Sexl case, the solution at the linearized level turns out
to be exact.
(b) We do not however find any obvious way to separate the scalar wave equation in the
above metric, in contrast to the separability found in the R sector solutions [6,7]. Further
O. Lunin, S.D. Mathur / Nuclear Physics B 642 (2002) 91–113 93
the travel time for quanta in this geometry from infinity to the center does not agree with the
expected time for travel around the ‘long circles’ of the CFT state (σn−− )N/n |0NS . We thus
look for a different metric that could be dual to these chiral primaries. We consider a set
of metrics with a conical defect singularity along a circle; such conical defect singularities
arose in the R sector solutions and were studied in [8–10]. For this class of metrics we find
that the scalar wave equation separates, and the energy levels for the l = 0 harmonic of the
scalar match exactly the corresponding energy levels in the dual CFT on the ‘long circle’.
The travel time for a quantum from infinity to the center and back also agrees exactly with
the travel time around the long circle of the CFT.
These results indicate that for the massless quanta describing the chiral primary
(σn−− )N/n |0NS the lowest energy solution is not the one with an Aichelburg–Sexl type
singularity but rather the one with the conical defect singularity.
(c) We next consider a single quantum with high angular momentum in the geometry
AdS3 × S 3 × M4 . The quantum is described by a chiral primary σn−− , but the remaining
N − n copies of the c = 6 CFT are not twisted into ‘long circles’. The chiral primary
σn−− can be written as (σ2−− )N , where σ2−− is a chiral primary that can be regarded as a
building black to make all higher chiral primaries. We argue that σ2−− plays the role of the
field Z in [4], and that J0− σ2−− , J¯0 σ2−− are the analogs of the variables Xi that describe
stringy excitations around the supergravity quantum. We find that the CFT description of
these string excitations can be interpreted as low energy vibrations of a ‘Fermi fluid’ which
arises from the fermionic excitations that give σn−− its charge.
While we were working on this paper there appeared the paper [11] which overlaps
partly with our discussion in Section 5.
2. The Aichelburg–Sexl type solution in AdS3 × S 3
We begin by recalling the metric produced by a massless point particle moving along
the z direction in 3 + 1 dimensions [5].

ds 2 = −dt 2 + dx 2 + dy 2 + dz2 + 8pδ(t − z) log x 2 + y 2 (dt − dz)2 . (2.1)
If we consider instead a set of such particles distributed uniformly along the z-axis then we
get a time-independent metric:

ds 2 = −dt 2 + dx 2 + dy 2 + dz2 + q log x 2 + y 2 (dt − dz)2 . (2.2)
Note that the function log(x 2 + y 2 ) appearing in the metric is a harmonic function of the
transverse coordinates x, y with δ-function source at the singular line x = y = 0. In 5 + 1
dimensions the corresponding metric will be (xi = x1 , . . . , x4 )
q
ds 2 = −dt 2 + dz2 + dxi dxi + (dt − dz)2 . (2.3)
(xi xi )
We wish to construct solutions that describe a set of massless particles rotating along
a diameter of S 3 in the space AdS3 × S 3 . This metric arises in IIB string theory after
compactification on a 4-manifold M4 which can be T 4 or K3. We will take the case of T 4
for concreteness in this paper, though there is no fundamental difference if we were to take
NS−NS ; we will just call

K3 instead. The curvature is produced by the 2-form gauge field Bµν
this field Bµν hereafter. For unperturbed AdS3 × S × T the 10-d Einstein metric is1
3 4

r2 dr 2
ds = − 1 + 2 dt 2 +
2
2
+ r 2 dχ 2
L 1 + Lr 2
4
+ L2 dθ 2 + cos2 θ dψ 2 + sin2 θ dφ 2 + dzi dzi (2.4)
i=1
and the gauge field is2

r2
B = L2 cos2 θ dφ ∧ dψ + dt ∧ dχ. (2.5)
L
We consider a set of massless particles rotating along the diameter θ = 0 of the S 3 at
the location r = 0 in the AdS3 . (We will ignore the T 4 in what follows—one may assume
that the particle wavefunctions are uniformly smeared along the T 4 .) We thus look for a
solution of the IIB field equations that behave as (2.3) near θ = 0, r = 0, and go over to
AdS3 × S 3 at large r.
We first construct the linear solution corresponding to small strength of the source at
θ = 0, r = 0. It turns out that just as is the case for the Aichelburg–Sexl metric in 3 + 1
dimensions, the linear solution is exact. The full 10d metric is

r2 dr 2
ds 2 = − 1 + 2 dt 2 + 2
+ r 2 dχ 2 + L2 dθ 2 + cos2 θ dψ 2 + sin2 θ dφ 2
L 1 + Lr 2

4
q
+ dzi dzi +
i=1
r 2 + L2 sin2 θ

2 2
2
r2 2 r
× 1 + 2 dt + L cos θ dψ − L
2
dχ − sin θ dφ
2
(2.6)
L L2
and the NS–NS two-form field is
r2 q 1
B = L2 cos2 θ dφ ∧ dψ + dt ∧ dχ −
L L r 2 + L2 sin2 θ

r2
× 1 + 2 dt + L cos θ dψ ∧ r 2 dχ − L2 sin2 θ dφ .
2
(2.7)
L
3. Metrics with conical defects
In [8,9] a set of metrics was considered which could be obtained as special cases of
metrics for rotating charged holes found in [12]. The metrics describe the D1–D5 bound
√ 1 H 2 ] and H
1 The action for the relevant fields is −g [R − 12 µνλ = ∂µ Bνλ + ∂ν Bλµ + ∂λ Bµν .
φψ = L cos θ etc.
2 Thus B 2 2
state carrying angular momentum, and go over to flat space at spatial infinity. In the near
horizon region (the ‘AdS region’), the solutions have the form
2
r dr 2
ds 2 = − 2 + γ 2 dt 2 + r 2 dχ 2 + 2
L r
L2
+γ2

dt 2
+ L dθ + cos θ (dψ − γ dχ) + sin θ dφ − γ
2 2 2 2 2
, (3.1)
L

r2 dt
B = dt ∧ dχ + cos2 θ (dψ − γ dχ) ∧ dφ − γ . (3.2)
L L
It was argued in [8,9] that these solutions describe ground states of the D1–D5 system
in the Ramond sector. It was found in [3] that only a special class of Ramond ground
states are described by these solutions. The solution for the general ground state was then
constructed, and it was found that the singularity had the shape of a complicated curve for
a generic metric.
On the other hand we can write down a larger class of metrics similar to the simple form
(3.1)3:
2
r dr 2
ds = − 2 + γ dt 2 + r 2 dχ 2 + 2
2 2
L r
L2
+γ2
2 2
dt dt
+ L2 dθ 2 + cos2 θ dψ − β̃ − α dχ + sin2 θ dφ − α̃ − β dχ ,
L L
(3.3)
2
r dt dt
B = dt ∧ dχ + cos2 θ dψ − β̃ − αdχ ∧ dφ − α̃ − βdχ . (3.4)
L L L
All these metrics possess a conical defect at θ = 0, r = 0, and are locally AdS3 × S 3
elsewhere. Thus they are quite different from the metric found in the previous section.
Due to the presence of the conical singularity we do not know a priori that all these
metrics are solutions of Type IIB string theory. We will however try to identify a subclass
of such metrics with the dual of chiral primaries of the NS sector, by studying the values
of conserved charges and the spectrum of low energy excitations around these solutions.
Let us briefly discuss the charges associated with metric (3.3). The 2 + 1 gravity theory
obtained by reduction of IIB on S 3 × M4 has a gauge field Aαµ , where α is an index on S 3
and µ is an index in the AdS3 . The field equation for A is
d ∗ dA + dA = ∗j. (3.5)
The conserved charge is thus given by the line integral over the circle t = t0 , r = r0 of
the quantity ∗dA + A. The first part is the usual flux of electrodynamics; it vanishes at
infinity for the metrics we consider. The second part is a Wilson line, and gives the angular
momentum of the solution. The angular momenta corresponding to translations in φ and
3 Such a class of metrics was studied in [10].

ψ directions are:
βL αL
j(ψ) = , j(φ) = , (3.6)
4G3 4G3
where G3 is three-dimensional Newton’s constant. The mass of the solution is
γ2
M =− . (3.7)
8G3
The quantities in the gravity theory can be related to quantities in the boundary CFT by
the following relations [8,9,13]
3L
c= . (3.8)
2G3
1 1
j0 = (j(φ) − j(ψ) ), j¯0 = − (j(φ) + j(ψ) ). (3.9)
2 2
Let l0 + l¯0 give the energy and l0 − l¯0 give the AdS3 momentum of a configuration. Then
to map to CFT levels is achieved by addition of a Sugawara term and a constant shift:
c 6(j0 )2
L0 ≡ l0 + + ,
24 c
c 6(j¯0 )2
L̄0 ≡ l¯0 + + . (3.10)
24 c
For the solutions (3.3)
LM Lγ 2 c c
l0 = l¯0 = =− , j0 = (α − β), j¯0 = − (α + β) (3.11)
2 16G3 12 12
and
c 2
L0 = −γ + 1 + (α − β)2 , (3.12)
24
c 2
L̄0 = −γ + 1 + (α + β)2 . (3.13)
24
The Ramond vacuum of the CFT is dual to the geometry (3.1) with4
α = γ, α̃ = γ , β = β̃ = 0. (3.14)
From the charge and dimensions of the above solutions one infers that the NS vacuum
is given by
α = α̃ = 0, β = β̃ = 0, γ = 1. (3.15)
Chiral primaries in the NS sector have L0 = j0 , L̄0 = J¯0 which corresponds to
α = α̃ = 0, β = β̃ = γ − 1. (3.16)
4 The values of α̃, β̃ are determined by the spectral flow parameters of the CFT [8,9].
For these chiral primaries the geometry (3.3) becomes:

2
r dr 2
ds = − 2 + γ dt 2 + r 2 dχ 2 + 2
2 2
L r
L2
+γ2

dt 2
+ L2 dθ 2 + cos2 θ dψ − β + sin2 θ (dφ − β dχ)2 , (3.17)
L

r2 dt
B = dt ∧ dχ + cos2 θ dψ − β ∧ (dφ − β dχ), γ = β + 1. (3.18)
L L
For these solutions the parameter γ is related to j0 in the following way by Eq. (3.11).
12 2j0
γ =1− j0 = 1 − . (3.19)
c N
Let us see which chiral primaries in the CFT would be dual to these geometries. We
review the structure of chiral primaries of the N = 4 orbifold CFT in Appendix B. Consider
the chiral primary σn−− . This has h = j = n−12 . The states we are interested in are generated
by a set of such chiral primaries:
N n−1
[σn−− ]N/n : L0 = j0 = L̄0 = j¯0 = . (3.20)
n 2
We then find from (3.19) the value of the parameter γ in the metrics (3.17) dual to the state
[σn−− ]N/n |0NS :
1
γ= . (3.21)
n
4. Energy levels and travel time for scalar fields
The state in the CFT created by the chiral primary σn−− |0NS is described in the dual
theory by a supergravity particle in AdS3 × S 3 × M4 [14]. Thus the state (σn−− )N/n |0NS
should be described by a collection of N/n such particles. If n 1 the particles have a
high angular momentum, and their wavefunctions will be confined to a very narrow width
around the diameter on the S 3 on which they rotate. Thus they would be pointlike particles
for supergravity. If we also have N/n 1 then we expect to describe the resulting physics
by a classical supergravity solution.
The configuration constructed in the above way will be naturally smeared uniformly
along the diameter of rotation. We assume that all wavefunctions are constant over the M4 ,
and so can restrict attention to AdS3 × S 3 . We thus have a line of massless particles in
(5 + 1)-dimensional spacetime.
A first guess would, therefore, be that near to this line of particles the solution behaves
like (2.3), the Aichelburg–Sexl solution in flat 5 + 1 spacetime. We therefore consider the
exact solution (2.6).5
5 Note that the B field (2.7) has a divergent quartic invariant H νλσ H τρµ near the singular line,
µνλ H σ τρ H
though the divergence of the invariant is much weaker than would be naively expected from the magnitude of the
H field near the singularity.
At this point we are led to compare this solution with the metrics found in [8,9,15] for
the states of the CFT in the R sector. We expect to have somewhat similar properties for
the R and NS cases, since one is just a spectral flow of the other. In the R sector the wave
equation for a scalar separated between the angular and radial variables. Such a separation
is not obvious for the geometry (2.6). Further, in [3] the travel time around the ‘long circles’
in the CFT agreed exactly with the travel time in the throat of the dual supergravity solution.
We do not find such an agreement for the solution (2.6). We are, therefore, led to consider
other possible solutions that could describe the state (σn )N/n |0NS .
The solutions (3.17), (3.18) have conical defects that are similar to those found for
the R sector metrics in [8,9]. If we assume the relations (3.8)–(3.10) postulated in [8,
13] between quantities in the gravity theory and quantities in the CFT, then we find that
L0 = j0 , L̄0 = j¯0 for these solutions, so they have the right structure to be identified as the
states of the string theory dual to the states (σn )N/n |0NS created by chiral primaries in the
CFT.
We will now study the energy levels of some supergravity excitations around the metrics
(3.17) and see that they agree with the excitation levels expected from the dual CFT.
4.1. Energy levels for l = 0 scalar excitations
There are several scalars in the (5 + 1)-dimensional theory obtained by reduction on

M4 . For concreteness we take M4 = T 4 . Let the T 4 be described by coordinates x1 , . . . , x4 .
Then the graviton h12 is a minimally coupled scalar in the remaining 5 + 1 dimensions
4 ≡ h12 , ✷4 = 0. (4.1)
In Appendix A we solve this wave equation in the metric (3.17). The wave equation is
separable, and we have solutions of the form
4(t, r, χ, θ, ψ, φ) = exp(−iωt + ipψ + iqφ + iλχ)H (r)Θ(θ ). (4.2)
We get normalizable solutions for the frequencies (A.15)

βp 2γ l λ + βq
ωk = ± k+1+ + , k = 0, 1, 2, . . . . (4.3)
L L 2 L
Consider the case l = 0, λ = 0. The spectrum is
2
ωk = (k + 1), k = 0, 1, 2, . . . , (4.4)
nL
where we have used the relation γ = 1/n for chiral primaries (see (3.20)).
Let us relate the energy of such quanta with the expected changes in quantities in the
CFT. For δj0 = δ j¯0 = 0 the relations (3.10) give
2(k + 1)
δ(L0 + L̄0 ) = δ(l0 + l¯0 ) = Lωk = . (4.5)
n
Now let us consider the CFT itself. The graviton h12 is known to be described by a pair
of excitations [16]
1 ¯ 2 + ∂X2 ∂X

¯ 1 .
h12 → √ ∂X1 ∂X (4.6)
2
There is one left mover and one right mover on the ‘long circle’ of length 2πn. We can
have fractional levels for L0 , L̄0 but L0 − L̄0 must be an integer [17,18]. The energy levels
for L0 − L̄0 = 0 are given by
k 2k
δL0 = δ L̄0 = , δ(L0 + L̄0 ) = , k = 1, 2, . . . (4.7)
n n
so that we get exact agreement with (4.5).
4.2. Travel time
In [3] it was found that the time 8tCFT for excitations to travel around the ‘long circle’
of the CFT was exactly equal to the time 8tSUGRA taken for a supergravity quantum to
travel once up and down the ‘throat’ of the dual geometry. In the present case the geometry
at infinity is AdS rather than flat space.
We observe from (4.3) and the fact that γ = n1 , β = n1 − 1 that the frequencies ωk have
the form

1 m1
ω= + m2 , m1 , m2 integral. (4.8)
L n
Consider any wavefunction for the scalar particle made by an arbitrary superposition of
such frequencies. Then the wavefunction returns to its initial form after a time
8tSUGRA = n2πL. (4.9)
Now consider the dual CFT. The left and right movers are massless excitations, and
travel at the speed of light around the spatial circle of the cylinder on which the CFT lives.
This gives Lδχ = δt for the right movers and Lδχ = −δt for the left movers. But the twist
σn makes the circle on which the CFT excitations live a ‘long circle’ of length 2πn times
the initial circle size of the CFT. The state (σn−− )N/n |0NS that we have chosen has all
copies of the CFT arranged into long circles of this length. Thus any excitation of this CFT
state will return to its initial form after a time
8tCFT = 2πnL (4.10)
which is in exact agreement with (4.9).

To summarize, we have seen that the CFT state (σn−− )N/n |0NS has all copies of the
c = 6 CFT joined into ‘long circles’ that wind n times the spatial direction before closing.
The low energy excitations in the CFT are thus of energy ∼ 1/n. In the dual supergravity
solution we do not see any ‘multiwinding’ around the χ direction. What we have instead
is a deep ‘throat’ like structure near r = 0 that leads to a large redshift between r = ∞ and
r ≈ 0. Particle wavefunctions trapped near r = 0 have energies that are low (∼ 1/n) due
to this redshift. This relation ‘multiwinding in CFT → redshift in gravity’ is expected to
have general validity and was found also in the R sector computations of [3].
5. CFT states representing a single string
5.1. String states in AdS5 × S 5
Let us review the idea proposed in [4]. We consider a supergravity particle rotating on
the diameter of S 5 in AdS5 × S 5 , at the origin of AdS5 . The S 5 is described by coordinates
X1 , . . . , X6 , and the rotation is in the plane X5 , X6 . We write Z = X5 + iX6 , and let the
other four X coordinates be called Xi , i = 1, . . . , 4. It was noted in [4] that (for appropriate
choice of supergravity quantum) the state in the dual CFT is created by the operator tr[Z J ],
where J is the angular momentum of the quantum. It was then argued that for large angular
momentum J we can easily describe the excitations of the quantum that change it to a
stringy state. There are 8 bosonic excitations possible, and in a light-cone gauge 4 are
along the directions Xi while 4 are directions in the AdS5 . Stringy excitations in the Xi
directions are described in the dual theory by operators of the form
tr[ZZ · · · ZZXi1 ZZ · · · ZZXi2 ZZ · · · ZZ]. (5.1)

Excitations in the AdS5 directions are given by derivatives ∂µ , µ = 1, . . . , 4.
Each of these insertions Xi , ∂µ have ∆ − J = 1. We do not put insertions of Z, which
has ∆ − J = 2. While the OPE of Xi with Z is regular in the free CFT, the OPE of Z with
Z gives
1 )Z(x2 ) ∼ 1
Z(x . (5.2)
(x1 − x2 )2
The expectation is that the operators with Z insertions would be renormalized to high
dimensions when we go to the values of coupling appropriate to the supergravity
description, and so these would not be visible in the string vibration spectrum.
5.2. String spectrum and energy scales for AdS3 × S 3 × T 4
Let us now consider the analogue of this problem for AdS3 × S 3 × T 4 . We take a
supergravity quantum rotating on the S 3 while staying at the origin in AdS3 . The spectrum
of string states was found in [4,19,20]
∞

∆=J + ω̃m am am + ω̃−m am
(+)† (+)
am + ω̃m ãm
(−)† (−)
ãm + ω̃−m ãm
(+)† (+) (−)† (−)
ãm
m=−∞
4 2
∞
L 2m (i)† (i)
+
α ∆ + J cm cm + ferm. (5.3)
m=−∞ i=1
Here a give oscillators in the AdS directions, ã give oscillations in the sphere directions,
and ci give oscillations in the T 4 . Further,

2
L2 2m 2 L Ts m
ω̃m = sin α̂ + cos α̂ +
2
≈ 1 + cos α̂ , (5.4)
α ∆+J 2π J
where L is a radius of the AdS3 , TS = 2πα1

is the tension of the elementary string and the
approximation in the last step is for small mode numbers m J .
At leading order in m/J we have the frequencies ω̃m ≈ 1 for the oscillators in the S 3
and AdS3 directions, which implies that these excitations have
∆ − J ≈ 1. (5.5)
The oscillators in the T 4 direction have ω̃ ≈ 0 which implies
∆ − J ≈ 0. (5.6)
m
The correction of order in (5.4) depends on the value of α̂. But the sigma model CFT
J
at the orbifold point is equally far in moduli space from the D1–D5 background (α̂ = π/2)
and the NS1–NS5 background (α̂ = 0). Any effects involving α̂ can only be obtained by
studying deformations of the orbifold. We will study the CFT only at the orbifold point,
and it is important to note that in such a study we should try to reproduce only the values
(5.5), (5.6) and not the subleading corrections.
5.3. CFT operators dual to string oscillators
Consider a supergravity quantum rotating on the S 3 while staying at the origin in AdS3 .
Letting this quantum be made from a suitable combination of hab , Bab (a, b are indices on
S 3 ) we obtain a state with h = j = n−1 ¯ n−1
2 , h̄ = j = 2 , which is a created in the dual CFT
by a chiral primary
σn−− |0NS . (5.7)
Let us write explicitly the permutation involved in the twist operator σn . We must finally
sum over all choices of the indices involved in the permutation [21–24], but to do this we
must first start by choosing definite values of the permuted indices. Thus let the operator
be
−−
σn−− = σ(12,...,n) , (5.8)
where (12, . . . , n) is the permutation in the twist created by σn−− . Then we can write
−− −− −− −−
σ(12,...,n) = σ(12) σ(23) · · · σ(n−1,n) . (5.9)
So the chiral primary σ2−− which has a twist of order 2 acts like a building block for other
chiral primaries, and is thus similar to the operator Z in the CFT dual to AdS5 × S 5 .
in the AdS5 × S 5 case were SO(6) rotates of the operator Z. In the
The operators Xi , Z
present case the corresponding rotation group is SO(4) ≈ SU(2)L × SU(2)R . There are two
vibrations on the sphere, which suggests that the string oscillators in the sphere directions
map to the CFT operators
J0− σ2−− , J¯0− σ2−− . (5.10)
is the operator with SO(4) spin opposite to the spin of σ −− , which is
The analogue of Z 2
J0− J¯0− σ2−− . (5.11)

Note that the operators (5.10) have ∆ − J = 1 while the operator (5.11) has ∆ − J = 2;
respectively.
these are similar to the ∆ − J values of the Xi and Z,
For the identification between string oscillators and CFT operators to make sense we
need to check in addition that the OPE of the operators (5.10) with σ2−− is regular, while
the OPE of (5.11) with σ2−− is singular.
In Appendix B we review the structure of twist operators. Let us note some of the
structure of chiral primaries and their excitations. A twist operator σn just cyclically
permutes n copies of the c = 6 CFT into each other. To make this into a chiral primary
with h = j we must add charge, which can be done by adding fractional modes of currents
+
J−k/n .
Now consider the OPE of J0− σ2−− with σ2−− . Working independently in the holomor-
phic and antiholomorphic sectors we find the apparent singularity
1 − −−
[J0− σ2−− ](z)σ2−− (w) ∼ J σ (w) + · · · . (5.12)
(z − w)1/3 13 3
The operator [J 1− σ3−− ](w) has h − h̄ nonintegral. But recall that while fractional modes
3
are allowed for the left and right sectors, the allowed operators and states in the CFT are
only those with h − h̄ integral6 [17,18]. Thus the singular operator on the RHS of (5.12)
does not exist in the CFT, and the OPE is in fact regular
[J0− σ2−− ](z)σ2−− (w) ∼ [J0− σ3−− ](w) + · · · . (5.13)

Thus J0− σ2−− , J¯0− σ2−− are good candidates for the ‘defects’ that can be included in the
chain (5.9) to represent string oscillations:
· · · σ2−− · · · σ2−− [J0− σ2 ]σ2−− · · · σ2−− [J¯0− σ2−− ]σ2−− · · · . (5.14)
Now consider the OPE of J0− J¯0− σ2−− with σ2−− . This time we get the leading singularity
1 − − −−
J¯0− J0− σ2−− (z)σ2−− (w) ∼ J 1 J¯1 σ3 (w) + · · · . (5.15)
|z − w| 2/3 3 3
The operator on the RHS has h − h̄ = 0 and is an allowed operator in the theory. Thus
we should not include J0− J¯0− σ2−− in the chain (5.14), which is analogous to not allowing
Z̄ in the case of AdS5 × S 5 . The combination Z Z can be contracted away to the identity;
− ¯− −− −−
similarly (J0 J0 σ2 )(σ2 ) can be contracted to the identity and thus removed from the
chain.
In Appendix C we discuss other OPEs that help support the picture developed here.
5.4. Mapping string oscillators to CFT operators
The string vibrations in the sphere directions were mapped above to J0− σ2−− , J¯0− σ2−− .
Following the ideas of [4] the oscillators in the AdS3 directions should be mapped to ∂z , ∂z̄ ,
6 For fermionic states h − h̄ is half integral. (h − j ) − (h̄ − j¯) is integral for all states.
which upon acting on σ2−− can be written as

L−1 σ2−− , L̄−1 σ2−− (5.16)
These operators also have ∆ − J = 1.7
There are 4 fermionic vibrations that have ∆ − J = 1, and following the above pattern
it is natural to identify these as
G−,1 −−
1 σ2 , G−,2 −−
1 σ2 , −,1
G −−
1 σ2 , −,2
G −−
1 σ2 . (5.17)
−2 −2 −2 −2
We can place the above ‘modified σ2 ’ operators (5.10), (5.16), (5.17) at various points
along the chain (5.9) and consider the linear combinations of these possibilities that
correspond to definite Fourier harmonics along the chain [4]. This gives the map from
string oscillators to CFT operators:

J
(+)†
ãm |J → e2πimk/J (σ2−− )k−1 [J0− σ2−− ](σ2−− )J −k−1 ,
k=1

J

(−)†
ãm |J → e2πimk/J (σ2−− )k−1 J¯0− σ2−− (σ2−− )J −k−1 ,
k=1

J

(+)i†
bm |J → e2πimk/J (σ2−− )k−1 G−,i1 σ2−− (σ2−− )J −k−1 , i = 1, 2,
−2
k=1

J
−,i −− −− J −k−1
(−)i†
bm |J → 1σ
e2πimk/J (σ2−− )k−1 G (σ2 ) , i = 1, 2,
2 −2
k=1

J
(+)†
am |J → e2πimk/J (σ2−− )k−1 [L−1 σ2−− ](σ2−− )J −k−1 ,
k=1

J

(−)†
am |J → e2πimk/J (σ2−− )k−1 L̄−1 σ2−− (σ2−− )J −k−1 . (5.18)
k=1
After constructing the above state of the twist operators we must symmetrize by
summing over permutations of the N copies of the CFT. The cyclic permutations of
the indices (1, . . . , n) involved in the above operators yield the same state, and so these
permutations must be added with uniform weight. This forces the total ‘momentum’ of the
Fourier modes m to add up to zero in the CFT, which agrees with the similar condition
that arises for the string oscillators from the fact that the total momentum along the string
world sheet must be zero.
5.5. CFT excitations as oscillations of a ‘Fermi fluid’
We can obtain some understanding of the CFT states created by the operators (5.18) by
taking the specific case M4 = T 4 and writing the currents J a in terms of the fermions of
7 See [25] for an alternative interpretation of the AdS oscillators in AdS × S 5 .

5
the CFT. We have 4 real fermions in the holomorphic sector, and 4 real fermions in the
antiholomorphic sector. The holomorphic fermions are grouped into a doublet of complex
fermions (ψ + , ψ − ) which forms a spinor of SU(2)L . The currents are
J + = ψ + (ψ − )† ,
J − = (ψ + )† ψ − ,
1 + + †
J3 = ψ (ψ ) + (ψ − )† ψ − . (5.19)
2
Consider for simplicity the chiral primary σn−− for n odd. The charge that must be added
to the twist operator σn to make it a chiral primary is carried by fermions with fractional
moding:

σn−− = ψ + 1 ψ + 3 · · · ψ + 1 (ψ − )† (ψ − )† · · · (ψ − )†
− 2n − 2n − 2 + n1 1
− 2n 3
− 2n − 12 + n1
− †
× ψ̄ + 1 ψ̄ + 3 · · · ψ̄ + 1 (ψ̄ ) (ψ̄ − )† · · · (ψ̄ − )† σn . (5.20)
− 2n − 2n − 2 + n1 − 2n
1
− 2n
3
− 12 + n1
(Note that ψ + and (ψ − )† both carry the positive SU(2)L spin.)

Now consider a state created by some Fourier mode m of the insertion J0− σ2 in the
manner shown in (5.18). The Fourier function is approximately constant over a large
number of σ2 operators in the chain, since we seek to describe low energy oscillations
of the string by these CFT operators. But if we sum over with uniform weight the states
obtained by insertion of J0− σ2−− over a certain part of the chain of length k − 1 then we
get

σ2−− · · · [J0− σ2−− ] · · · σ2−− ∼ J0− σk−− . (5.21)
The operator J0− acting on σn−− takes a fermion ψ + to a fermion ψ − , at the same level,
but does not change the number of fermions. The operator in (5.21) rotates the spin of the
fermions from one direction of the S 3 to another direction over the segment of the chain
of length ∼ n. We thus picture the modes (5.18) as follows. The fermions given in (5.20)
form a ‘Fermi sea’ over the long circle of length 2πn which results when the twist σn joins
n copies of the CFT into one copy. The Fourier modes (5.18) give low energy oscillations
of this Fermi sea by rotating the sea in different directions in different regions of the long
circle. This situation is reminiscent of the c = 1 matrix model where low energy excitations
of spacetime were given by oscillations of a nonrelativistic Fermi fluid [26].
The operators L−1 behave as L−n on the c = 6 CFT on the long circle as L−n . Their
effect can be seen as the effect of a diffeomorphism in the CFT, which affects both bosonic
and fermionic variables. In this case it is more convenient to bosonise the fermions, so that
we have 6 effective bosons. The diffeomorphism creates a ‘Bogolyubov transformation’

on the bosonic vacuum. (The bosons arising from the fermions have a charge ei e·φ and this
operator is also shifted in the obvious way under the diffeomorphism.)
5.6. T 4 vibrations and the orbifold point
The string vibrations on the torus have ∆ − J ≈ 0. For small ∆ − J we have the
following bosonic and fermionic modes in the CFT
i
∂X− k, ψ +1 , (ψ + )†1 , (ψ − )† , ψ− ,
n − 2 − nk k
2−n − 12 − nk 1 k
2−n
¯ i k,
∂X ψ̄ + 1 , (ψ̄ + )†1 , (ψ̄ − )† , ψ̄ − . (5.22)
− n − 2 − nk k
2−n − 12 − nk 1k
2−n
The modes (ψ + )†1 , ψ− and their antiholomorphic counterparts act as annihilation

2−n 2−n
k 1 k
operators for k < and they become creation operators for k n2 . (Note that since we
n
2,
have restricted attention to k n we will not reach this point in the excitation spectrum
under consideration.)
The energy levels implicit in the mode number − nk in (5.22) are valid only for the CFT
at the orbifold point. In [2] it was argued that the orbifold point has n5 = 1, n1 = N .
√ at the string frequencies (5.3) we note that if we set n5 = 1 and also α̂ = 0, then
Looking
L = α n5 and for the T 4 oscillators we get
2m m
ω̃n = ≈ (5.23)
8+J J
which agrees with the energy levels in (5.22) in the CFT. Similarly, the string modes in the
AdS3 and S 3 directions have energies
m
ω̃n ≈ 1 + . (5.24)
J
In the CFT at the orbifold point we find from the behavior of free fermion energy levels
that
n
e2πimk/n J0− σ(k,k+1)
−−
→ J−−m σn−− . (5.25)
n
k=1
More generally the Fourier modes in (5.18) give operators J −k , G−,i1 , L−1− k and
−n − 2 − nk n
their antiholomorphic partners, so that we get an exact agreement of energy levels. This
agreement of energies cannot be taken too seriously though since at the point n5 = 1, n1 =
N it is unlikely that the string is well described by its free oscillation frequencies (5.4).
In general it appears to be more useful to think of the CFT excitations as generated by
Fourier modes of local insertions as in (5.18), rather than as modes like J −k applied to
−n
the twist σn−− . As we deform the CFT away from the orbifold point we expect the picture
in terms of collective vibrations to be a robust one, since it just uses the fact that locally
along the chain we can apply a symmetry transformation to the spin of the chain. The
exact energy of the state generated by such oscillations can on the other hand depend on
the moduli of the CFT, and so we should use the orbifold point to reproduce just the leading
values (∆ − J ≈ 1, ∆ − J ≈ 0) for the various oscillators of the string.8
8 In [11] the energy levels of the CFT at the orbifold point were compared to the string levels at α̂ = 0, and it
was noted that only those levels in the CFT were found in the string spectrum which were multiples of n5 .
6. Discussion
Let us collect together the above results (and results in [3]) to get a picture of the relation
between rotation in AdS3 × S 3 and twist operators in the dual CFT.
The geometry AdS3 × S 3 is dual to the NS vacuum |0NS of the CFT. In this state the
CFT (at the orbifold point) has N copies of a c = 6 CFT with each copy living on a spatial
circle on length 2π . The operator σn−− joins n of these copies to give a c = 6 CFT living
on a long circle of length 2πn, and adds charge to give h = j, h̄ = j¯. The dual of the state
σn−− |0NS is expected to be a massless particle at the center of AdS3 rotating on the S 3
with SU(2) × SU(2) angular momentum (j, j) ¯ [14].
−−
But in the CFT state σn |0NS we have circles of length 2π and also circles of length
2πn. In [3] dual geometries were constructed for the Ramond sector states which arise
from spectral flows of general chiral primary states σn±± 1
σn±±
2
· · · σn±±
k
|0NS ; these states in
general have circles of several different lengths. Each metric had a singular curve with a
shape depending on the set {ki } and the choices of signs ±.
In our present study of the NS sector we have two cases where simplifications occur:
(a) We take the CFT state (σn−− )N/n |0NS where all circles are of the same length. Since
we have a large number N/n of particles in the dual geometry we have to consider their
back-reaction, and find the appropriate new geometry.
(b) We take the CFT state σn−− |0NS where n N . In this case since we have just
one light particle in the geometry we can ignore its back-reaction and keep the geometry
AdS3 × S 3 . In this approximation we can look at the stringy excitations of this quantum,
following the ideas of [4].
We started by looking for the geometries required in (a). For n 1 the massless
particles are pointlike. We first considered the assumption that the metric will take the
Aichelburg–Sexl form (2.3) near the singular line of massless particles—this metric is
isotropic to leading order near the singularity. It is interesting that the solution (2.6), (2.7)
with this limiting behavior can be found in closed form, and that the linear order solution
turns out to be exact.
But the solution (2.6) does not appear to have the right properties to be the dual of the
CFT state (σn−− )N/n |0NS . Following the analysis of [3] we note that in the CFT all copies
of the CFT live on circles of length 2πn, and so low lying excitation energies should come
in multiples of n2 . Further, since all excitations in the CFT travel at the speed of light around
the spatial circle we expect that any excitation in the dual geometry will be periodic with
period 2πnL. But in the geometry (2.6) the scalar wave equation did not separate so that
the energy levels had no simple form; further the time for a particle to start from infinity
and return to infinity was not order ∼ 2πnL.
We note that the space AdS3 × S 3 is not invariant under rotations in all 5 spatial
directions, and so it is possible that the lowest energy configuration for a given angular
momentum might not be isotropic even near the singular line of particles. In the R
sector study of [3] it was found that the metrics dual to the spectral flow of the states
(σn−− )N/n |0NS had the simple form found in [8,9]: the geometries are locally AdS3 × S 3
and the singular curve is a circle. We thus consider the solution (3.17), (3.18) which is a
simple generalization of the solutions of [8,9], and has the quantum numbers to be dual to
the state (σn−− )N/n |0NS .9 The conical defect makes the solution non-isotropic around the
singular circle.
For the geometries (3.17) we find that the scalar wave equation separates, and we
found the energy eigenvalues (4.3). For the simple case l = 0 (particle with no angular
momentum) we found that the energy levels (4.4) were multiples of n2 L1 . This agreed
exactly with the CFT expectation, where the scalar considered is described by one left
mover and one right mover, and each has energy nk . We see here the phenomenon that was
also seen in [3]. Low energies and long travel times arise in the CFT from long circles. In
the dual geometries we do not have long circles but instead deep throats which create a
large redshift between infinity and r ∼ 0 so that we again get very low energy modes.
We now consider problem (b). In [4] stringy excitations in AdS5 × S 5 were described
in the dual CFT by adding ‘defects’ Xi in a chain of operators Z. We have found a similar
map in the case AdS3 × S 3 , where the basic member of the chain is the chiral primary σ2−−
and the ‘defects’ are given by symmetry operators like J0− acting on σ2−− .
The excitation energies for a string with high J are to leading order ω ≈ 1 in units of
the AdS scale (in the AdS3 and S 3 directions). The corrections to these frequencies depend
on the tension of the string, and thus on the point in moduli space that we choose for the
geometry. The orbifold point can describe at best only one point in moduli space, so we do
not expect to reproduce these corrections by working at the orbifold point.
The chiral primary σn−− has a large number of fermions that carry its charge, and these
form a ‘Fermi sea’ in the c = 6 CFT living on the long circle of length 2πk. The operator
J0− rotates the spin direction of these fermions, and thus Fourier modes of J0− along the
chain generate low energy oscillations of the Fermi sea. When we deform the theory away
from the orbifold point we expect that this concept of collective oscillations will persist
and yield the stringy oscillations of the chiral primary.
We can also combine problems (a) and (b) to consider stringy excitations of a quantum
moving in the background dual to the state (σn−− )N/n |0NS . We can give the string a small
center of mass energy which takes it away from the singular circle; this lets its low energy
excitations be those of a string in AdS3 × S 3 without the singularity. We see that the redshift
between infinity and r ∼ 0 lowers all string excitation energies by a factor n, when these
energies are measured from the boundary of AdS3 . In the dual CFT we start with circles of
length 2πn instead of n, and thus find energies that are also lower by a factor n.
Acknowledgements
We are grateful to Sergey Frolov, Camillo Imbimbo and Arkady Tseytlin for useful
discussions. This work was supported in part by DOE grant DE-FG02-91ER40690.
9 We have not proved directly that this geometry with its conical singularity is a solution of IIB string theory.
By contrast in [3] we did prove that all solutions (which also had singular curves) were true solutions in string
theory: we started with a string having momentum and winding, found its geometry, and then dualized to obtain
D1–D5 metrics with a singular curve.
Appendix A. Energy levels for scalar excitations
Here we solve the wave equation for a massless scalar. We look for the solution of the
Klein–Gordon equation
✷4 = 0 (A.1)
in the metric (3.17).
We write
4(t, r, χ, θ, ψ, φ) = exp(−iωt + ipψ + iqφ + iλχ)H (r)Θ(θ ). (A.2)

Then we get the following equations for the two functions H and Θ (see [7] for details):
2
1 d r 2 dH (ω − βp/L)2 (λ + βq)2
r + γ + − H − ΛH = 0, (A.3)
r dr L2 dr r2
2 + γ 2 r2
L
2
1 d dΘ q p2
sin 2θ − + Θ = −ΛΘ. (A.4)
sin 2θ dθ dθ sin2 θ cos2 θ
The angular equation involves the usual Laplacian on S 3 , so we get Λ = l(l + 2) with
l = 0, 1, 2, . . . . Assuming that λ+βq 0, we get the solution of the radial equation regular
at r = 0 [7]:

(ωL−βp)/2γ x
H (x) = x (λ+βq)/2γ x + γ 2 F a, a + l + 1; c; − 2 , (A.5)
γ
where
l λ + βq + ωL − βp λ + βq
a=− + , c=1+ . (A.6)
2 2γ γ
For large z we have
F (a, a + m; c; z)
∞
B(c)(−z)−a−m (a)n+m (1 − c + a)m+n −n
≈ z (log(−z) + hn )
B(a + m)B(c − a) n!(m + n)!
n=0
B(c)(−z)−a
m−1
(a)n B(m − n) −n
+ z . (A.7)
B(a + m) n!B(c − a − n)
n=0
We will not need an explicit form of the coefficients hn , they can be found, for example,
in [27]. We then find that normalizability at infinity requires that the finite sum disappears
from the last expression, which happens if either one of the conditions:
c − a = −k or a = −l − k − 1, where k = 0, 1, 2, . . . (A.8)
is satisfied. This gives

ωL − βp l λ + βq
=± +1+ +k . (A.9)
2γ 2 2γ
We thus get the frequencies:

βp 2γ l λ + βq
ωk = ± (k + 1 + ) + , k = 0, 1, 2, . . . . (A.10)
L L 2 L
For the discrete set of frequencies with positive sign in (A.10) we can rewrite the
solutions of the wave equation in more convenient form:

x
H (x) = x (x + γ )
c−1 2 k+l/2+1
F c + k, c + k + l + 1; c; − 2
γ

2 −c−k−l/2 2 c+2k+l+1 x
= x (x + γ )
c−1
(γ ) F −k, −(k + l + 1); c; − 2 . (A.11)
γ
The hypergeometric function in the above expression becomes a finite polynomial of
degree k and H (x) ∼ x −1−l/2 as x → ∞. For the frequencies with negative sign in (A.10)
the hypergeometric function in (A.5) is also a finite polynomial.
If λ + βq 0, then instead of (A.5) we get:

(ωL−βp)/2γ x
H (x) = x −(λ+βq)/2γ x + γ 2 F a , a + l + 1; c; − 2 , (A.12)
γ
l λ + βq ωL − βp λ + βq
a = − − + , c = 1 − (A.13)
2 2γ 2γ γ
and we get the frequencies

βp 2γ l λ + βq
ωk = ± k+1+ − , k = 0, 1, 2, . . . . (A.14)
L L 2 L
Combining this with (A.10), we can write the spectrum for a general case as

βp 2γ l λ + βq
ωk = ± k+1+ + , k = 0, 1, 2, . . . . (A.15)
L L 2 L
Appendix B. Chiral primaries in the orbifold CFT
We briefly summarize the nature of chiral primaries in the orbifold CFT and introduce
the notation that we use in this paper. Details can be found in [23,24].
We consider the bound states of n1 D1 branes and n5 D5 branes in IIB string theory. We
set
N = n1 n5 . (B.1)
The D5 branes are wrapped on a 4-manifold M, and thus appear as effective strings in the
remaining 6 spacetime dimensions. M can be T 4 or K3. The D1 branes and the effective
strings from the D5 branes extend along a common spatial direction x5 ≡ y, and y is
compactified on a circle of length 2πR. The low energy dynamics of this system is a
N = (4, 4) supersymmetric (1 + 1)-dimensional conformal field theory (CFT). The CFT
has an internal R-symmetry SU(2)L × SU(2)R ≈ SO(4). This symmetry arises from the
rotational symmetry of the brane configuration in the noncompact spatial directions x1 , x2 ,
x3 , x4 . The group SU(2)L is carried by the left movers in the CFT and the group SU(2)R
is carried by the right movers.
Consider this CFT at the ‘orbifold point’ [2]. Then the CFT is a (1 + 1)-dimensional
sigma model where the target space is the orbifold M N /SN , the symmetric product of N
copies of the 4-manifold M.
The M N /S N orbifold CFT and its states can be understood in the following way. We
take N copies of the supersymmetric c = 6 CFT which arises from the sigma model with
target space M. The vacuum of the theory is just the product of the vacuum in each copy of
the CFT. In the orbifold theory we find twist operators σn [21,22]. The copies 1, 2, . . . , n of
the CFT permute cyclically into each other 1 → 2 → · · · → n → 1 as we circle the point
of insertion of σn . (The other copies are not touched, and we ignore them for the moment.)
In this given twist sector there are operators with various values of j3 .
For odd n we start with σn , which is just a twist operator that permutes the copies of M
around its insertion point. The dimension of σn is

c 1 6 1 1 1
hn = h̄n = n− = n− = n− . (B.2)
24 n 24 n 4 n
To raise the charge of the operator with minimum increase in dimension consider the
application of J +1 . The charge goes up by one unit, while the dimension increases by
−n
only 1
n. The next cheapest operator is J +3 , and continuing to apply the cheapest possible
−n
operator at each stage we construct the chiral operator in this twist sector with lowest
dimension and charge. We will call it σn−− [24]:
σn−− ≡ J +n−2 J +n−4 · · · J +1 J¯+n−2 J¯+n−4 · · · J¯+1 σn (B.3)
− n − n −n − n − n −n
and it has
n−1 n−1
h = j3 = , h̄ = j¯3 = . (B.4)
2 2
In the case of even n, the lightest operator σn in a given twisted sector has dimension
hn = h̄n = n4 (see [24] for details), and the chiral primary with lowest dimension is
σn−− ≡ J +n−2 J +n−4 · · · J0+ J¯+n−2 J¯+n−4 · · · J¯0+ σn . (B.5)

− n − n − n − n
Its dimension and charge are given by (B.4).

Each copy of the CFT has the SU(2) currents J (i)a , J¯(i)a , where the index i labels the
copies. Define

n
n
J = a
J (i)a
, J¯a = J¯(i)a . (B.6)
i=1 i=1
Then we can make three additional chiral primaries from σ −− :

n+1 n−1
+ −−
σn+− = J−1 σn , h = j3 = , h̄ = j¯3 = ,
2 2
n−1 n+1
σn−+ = J¯−1
+ −−
σn , h = j3 = , h̄ = j¯3 = ,
2 2
+ ¯+ −− n+1 n+1
σn++ = J−1 J−1 σn , h = j3 = , h̄ = j¯3 = . (B.7)
2 2
The chiral primaries σn−− , σn+− , σn−+ , σn++ correspond respectively to the (0, 0), (2, 0),
(0, 2), (2, 2) forms from the cohomology of M. Both T 4 and K3 have one form of each of
these degrees.
The operator σ1−− is just the identity operator in one copy of the c = 6 CFT. Thus for
the complete CFT made from N copies we can write the above chiral operators as
σn±± [σ1−− ]N−n . (B.8)

It is understood here that we must symmetrize the above expression among all permutations
of the N copies of the CFT; we will not explicitly mention this symmetrization in what
follows.
More generally we can make the chiral operators

k
s ,s̄ mi
k
σnii i , ni mi = N, (B.9)
i=1 i=1
where si , s̄i can be +, −. This gives the complete set of chiral primaries that result if we
restrict ourselves to the above mentioned cohomology of M.
Appendix C. Some computations in the CFT
In [4] strings in AdS5 × S 5 were argued to be dual to traces like (5.1). We have argued
that strings in AdS3 × S 3 × T 4 are described by similar ‘chains’ of σ2−− operators with
sparsely placed ‘defects’ (Eq. (5.14)); these defects correspond to vibrating the string in
the 2 transverse sphere directions and 2 transverse AdS directions.
For this identification to be valid we must ensure that there are no other ‘defects’ that
can be included in the chain, since every type of defect corresponds to a string oscillator in
the gravity theory. Note that we allow only those operators to appear as defects that have
a regular OPE with the basic member of the chain σ2−− . In computing this OPE we have
to be careful that only operators with h − h̄ integral are allowed operators in the CFT, so
we discard apparently singular terms in the OPE which result in operators with fractional
h − h̄.
We have allowed J0− , G−,i1 , L−1 (and their right moving analogs) to act on σ2−− to
−2
produce the allowed defects. We start by noting that higher modes of these operators
produce singular OPEs with σ2−− and so do not give allowed defects:
− −− 1
[J−n σ2 ](z)σ2−− (w) = − σ −− (z)[J0− σ2−− ](w) + · · · , n > 0, (C.1)
(w − z)n 2
−,i 1
G 1 σ2−− (z)σ2−− (w) = − σ2−− (z) G−,i1 σ2−− (w) + · · · , n > 0,
−( 2 +n) (w − z) n −2
(C.2)
−− −− 1 −−
L−(n+1) σ2 (z)σ2 (w) = − ∂w σ3 (w) + · · · , n > 0. (C.3)
(w − z)n
−
Note that these operators J−n , G−,i1 , L−(n+1) (with n > 0) all have ∆ − J > 1.
− 2 +n
The operators that give allowed defects have ∆ − J = 1. We also have some other
operators with ∆ − J = 1 but which do not give regular OPEs with σ2−− and which
3 :
therefore do not give new allowed defects. An example is J−1
1
3
J−1 σ2−− (z)σ2−− (w) ∼ σ −− (w) + · · · . (C.4)
z−w 3
Next we consider the issue of zero modes. In the string theory the chiral primary is
represented by a particular member of the multiplet of massless particles. By application
of fermionic zero modes d0i† to the string we can change this particle to one of the other
massless particles. To see these zero modes in the dual CFT take the case M4 = T 4 for
concreteness. There are 4 fermionic modes in the c = 6 CFT with ∆ = J = 0: ψ + 1 , ψ̄ + 1 ,
−2 −2
(ψ − 1 )† , (ψ̄ − 1 )† and we can insert these in a chain of σ2−− operators. But
−2 −2

(σ2−− )k ψ + 1 σ2−− (σ2−− )J −k ∼ ψ + 1 σJ−−
+1 . (C.5)
− 2 −2
So these operators can be applied to the chain only in the zeroth Fourier harmonic: they
do not give additional types of local defects but only give a global change to the entire
operator. Thus the string zero modes map to the CFT zero modes (we write only the bosonic
operators)
|J → σJ−− ,
i† j † +β −−
d0 d0 |J → σJ+−
−1 , σJ−+
−1 , ψ +α1 ψ̄ σ ,
−2 − 12 J −1
d01† d02† d03† d04† |J → σJ++

−2 . (C.6)
Taking supersymmetry descendents of the above set gives the other massless particles of
the gravity multiplet.
Finally we note that we also cannot get new ‘defects’ by applying the above zero mode
operators to the allowed defects. Note that ψ + 1 (ψ − 1 )† = J−1
+
. As an example we can
−2 −2
check that J0− J−1
+ −−
σ2 has a singular OPE with σ2−−
1
J0− J−1
+ −−
σ2 (z)σ2−− (w) ∼ J − σ +− (w) + · · · . (C.7)
z−w 0 3
References
[1] J. Maldacena, Adv. Theor. Math. Phys. 2 (1998) 231, hep-th/9711200;

S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 428 (1998) 105, hep-th/9802109;
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9802150.
[2] A. Strominger, C. Vafa, Phys. Lett. B 379 (1996) 99, hep-th/9601029;
J. de Boer, Nucl. Phys. B 548 (1999) 139, hep-th/9806104;
N. Seiberg, E. Witten, JHEP 9904 (1999) 017, hep-th/9903224;
F. Larsen, E.J. Martinec, JHEP 9906 (1999) 019, hep-th/9905064.
[3] O. Lunin, S.D. Mathur, Nucl. Phys. B 623 (2002) 342, hep-th/0109154.
[4] D. Berenstein, J.M. Maldacena, H. Nastase, JHEP 0204 (2002) 013, hep-th/0202021.
[5] P.C. Aichelburg, R.U. Sexl, Gen. Rel. Grav. 2 (1971) 303.
[6] M. Cvetič, F. Larsen, Phys. Rev. D 56 (1997) 4994, hep-th/9705192.
[8] V. Balasubramanian, J. de Boer, E. Keski-Vakkuri, S.F. Ross, Phys. Rev. D 64 (2001) 064011, hep-
th/0011217.
[9] J. Maldacena, L. Maoz, hep-th/0012025.
[10] S.D. Mathur, hep-th/0101118.
[11] Y. Hikida, Y. Sugawara, hep-th/0205200.
[12] M. Cvetič, D. Youm, Nucl. Phys. B 476 (1996) 118, hep-th/9603100.
[13] M. Henneaux, L. Maoz, A. Schwimmer, Ann. Phys. 282 (2000) 31, hep-th/9910013.
[14] J.M. Maldacena, A. Strominger, JHEP 9812 (1998) 005, hep-th/9804085.
[16] S.R. Das, S.D. Mathur, Nucl. Phys. B 478 (1996) 561, hep-th/9606185;
S.R. Das, S.D. Mathur, Nucl. Phys. B 482 (1996) 482, hep-th/9607149.
[17] S.R. Das, S.D. Mathur, Phys. Lett. B 375 (1996) 103, hep-th/9601152.
[18] J.M. Maldacena, L. Susskind, Nucl. Phys. B 475 (1996) 679, hep-th/9604042.
[19] R.R. Metsaev, Nucl. Phys. B 625 (2002) 70, hep-th/0112044.
[20] J.G. Russo, A.A. Tseytlin, JHEP 0204 (2002) 021, hep-th/0202179.
[21] R. Dijkgraaf, C. Vafa, E. Verlinde, H. Verlinde, Commun. Math. Phys. 123 (1989) 485.
[22] A. Jevicki, M. Mihailescu, S. Ramgoolam, Nucl. Phys. B 577 (2000) 47, hep-th/9907144.
[23] O. Lunin, S.D. Mathur, Commun. Math. Phys. 219 (2001) 399, hep-th/0006196.
[24] O. Lunin, S.D. Mathur, Commun. Math. Phys. 227 (2002) 385, hep-th/0103169.
[25] S.R. Das, C. Gomez, S.J. Rey, hep-th/0203164;
R.G. Leigh, K. Okuyama, M. Rozali, hep-th/0204026.
[26] S.R. Das, A. Jevicki, Mod. Phys. Lett. A 5 (1990) 1639.
[27] H.S. Bateman, Higher Transcendental Functions, McGraw–Hill, New York, 1953.
Monodromy of solutions of the

Knizhnik–Zamolodchikov equation:
SL(2, C)/SU(2) WZNW model
Bénédicte Ponsot
Max-Planck-Institut für Gravitationsphysik, Albert Einstein Institut, Am Mühlenberg 1,
D-14476 Golm, Germany
Received 22 April 2002; accepted 15 July 2002
Abstract
Three explicit and equivalent representations for the monodromy of the conformal blocks in the
SL(2, C)/SU(2) WZNW model are proposed in terms of the same quantity computed in Liouville
field theory. We show that there are two possible fusion matrices in this model. This is due to the
fact that the conformal blocks, being solutions to the Knizhnik–Zamolodchikov equation, have a
singularity when the SL(2, C) isospin coordinate x equals the worldsheet variable z. We study the
asymptotic behaviour of the conformal block when x goes to z. The obtained relation inserted into a
four point correlation function in the SL(2, C)/SU(2) WZNW model gives some expression in terms
of two correlation functions in Liouville field theory.
Introduction
The SL(2, C)/SU(2) (or H3+ ) WZNW model is the second simplest non-compact
Conformal Field Theory (i.e., with a continuous spectrum of primary fields) besides
Liouville field theory. This model plays a role in condensed matter physics, as it is
believed to describe the plateau transitions in the Integer Quantum Hall effect [1]. It is also
intensively studied in the context of string theory on AdS3 (see for example [2] for a non-
exhaustive list of references). As its characteristics are closely related to those of Liouville
field theory, one may wonder to what extent it is possible to apply to the H3+ WZNW
model the techniques developed for Liouville field theory (computation of the bulk three
E-mail address: bponsot@aei-potsdam.mpg.de (B. Ponsot).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 5 7 6 - X
B. Ponsot / Nuclear Physics B 642 (2002) 114–138 115
point function: [3,4], bulk one point function in presence of a boundary and boundary two
point function [5], Liouville field theory on the pseudosphere [6], bulk-boundary function
[7], boundary three point function [8], fusion coefficients and proof for crossing symmetry
[9,10]). It turns out in the former model that the construction for the structure constants
[2,11] can be achieved mutatis mutandis along the same lines. Less straightforward is a
proof for crossing symmetry, as well as the construction of the fusion matrix. In Liouville
field theory, the fusion matrix was constructed as Racah–Wigner coefficients for some well
chosen continuous representations of the quantum group Uq (sl(2, R)) [9,10]; in this case
the proof for crossing symmetry boiled down to an orthogonality relation for these Racah–
Wigner coefficients. In the H3+ WZNW model, it was proposed in [12] as a quantum group
structure the “pair” Uq (sl(2)), Uq (osp(1|2)). We will not follow this approach here, as it
turns out that there is a way to avoid quantum group methods to construct the fusion matrix:
as it was noticed in [1] and [13], it is possible to adapt to the H3+ WZNW model some
observation previously made by Zamolodchikov and Fateev in [14] in the context of the
compact SU(2) WZNW model: these authors observed that there exists a 5 point conformal
block in the (k + 2, 1) Minimal Model that satisfies the same Knizhnik–Zamolodchikov
equation [15] as the 4 point conformal block in the SU(2) WZNW model. The adaptation
of this relation to the non-compact case is straighforward and allowed to prove crossing
symmetry in the H3+ WZNW model [13], as a consequence from a similar property of a
5 point function in Liouville field theory. Actually, this method also permits1 to construct
the monodromy of the conformal blocks in the H3+ WZNW model (i.e., the fusion matrix,
also called fusion coefficients) from the knowledge of the monodromy for a special 5 point
function in Liouville field theory. Let us emphasize on the fact that the fusion matrix
is the most important quantity of a CFT on genus zero, as it encodes the information
on degenerate representations, fusion rules, intermediate states appearing in a four point
function; it also permits the computation of boundary structure constants and enters the
consistency relations satisfied by structure constants [16,17].
In the work2 [18] is covered the case where one of the external spins corresponds to a
ˆ
degenerate sl(2) representation, the other spins being generic. The fusion matrix is derived
explicitly thanks to the explicit construction of the conformal blocks in this case (see also
[19]), enlarging the method of screening integrals developed in [20]. These conformal
ˆ
blocks reproduced the fusion rules for sl(2) given in [21].
In the present article, we will consider the problem in its full generality when all the
external spins are generic: there is no formula known for the conformal blocks in general
and the monodromy is infinite dimensional. The paper is organized as follows: in the first
section we will recall some useful information about Liouville field theory and the H3+
WZNW model. In section two we will propose three explicit formulas for the monodromy
of the conformal blocks in H3+ WZNW, constructed in terms of two Liouville fusion
matrices. Let us mention that the case we have here is somewhat different to all other
known cases so far (i.e., minimal models, SU(2) WZNW model, Liouville field theory),
as there are two possible fusion matrices for this model (this was first noticed in [18,19]):
1 This was first observed by J. Teschner.

2 I thank V.B. Petkova for pointing out this reference, as well as the work [19].
116 B. Ponsot / Nuclear Physics B 642 (2002) 114–138
this comes from the fact that the conformal blocks, in addition to the usual singularities at
z = 0, 1, ∞, have also a singularity at z = x (this plays an important role in [22]), where
z is the worldsheet variable, and x the isospin variable (or spacetime coordinate). When
deriving the formula for the fusion matrix, there are two cases to consider, depending
whether Im(z − x) is positive or negative. This accounts for a phase factor in the formulas,
which is different according to the sign of Im(z − x). In Section 3, we will study the
asymptotic behavior of the conformal blocks in the limit where x goes to z, and we will
see that they can be expressed as a sum of two Liouville conformal blocks. The first term
is regular in this limit, whereas the second term contains some singularity. This limit is
interpreted as quantum hamiltonian reduction in [23] (with the external spins satisfying
charge conservation conditions). The obtained relation permits us to rewrite a four point
correlation function in the limit x goes to z as a sum over two Liouville correlation
functions. We finish by some concluding remarks; Appendix A contains some particular
cases of the Liouville fusion coefficients needed in the main text to derive our main result,
in Appendix B we recover some well known special cases from our H3+ WZNW fusion
matrix. Finally, in Appendix C we present a way to find degenerate representations and
ˆ
fusion rules for sl(2) from degenerate representations and fusion rules for the Virasoro
algebra.
1. Requisites
1.1. Liouville field theory
Let the Vα (z, z̄) be the primary fields with conformal weight ∆L α = α(Q − α) where
Q = b + b−1; b is the coupling constant in Liouville field theory that we shall call for short
LFT. The central charge of the Virasoro algebra is c = 1 + 6Q2 .
Let

Vα4 (∞)Vα3 (1)Vα2 (z, z̄)Vα1 (0) ≡ Vα4 ,α3 ,α2 ,α1 (z, z̄)
denote a four point correlation function in LFT and
Fαs 21 (α1 , α2 , α3 , α4 |z) (1)
be the corresponding conformal block in the s-channel. The conformal block is completely
determined by the conformal symmetry (although there is no known closed form for it in
general), and is normalized such that
α −∆α −∆α
∆L L L
Fαs 21 (α1 , α2 , α3 , α4 |z) ∼z→0 z 21 1 2 (1 + O(z)). (2)
Let us note that the Liouville conformal block depends on conformal weights only.
There exist [24] invertible fusion transformations between s- and t-channel conformal
blocks, defining the Liouville fusion matrix:
Fαs 21 (α1 , α2 , α3 , α4 |z)

α α
= dα32 FαL21 α32 3 2 Fαt 32 (α1 , α2 , α3 , α4 |1 − z). (3)
α4 α1
Q +
2 +iR
The explicit expression for the LFT fusion matrix was given in [9], in terms of a b-
deformed 4 F3 hypergeometric function in the Barnes representation:

L α3 α2
Fα21 α32
α4 α1
Γb (2Q − α3 − α2 − α32 )Γb (α3 + α32 − α2 )
=
Γb (2Q − α1 − α2 − α21 )Γb (α1 + α21 − α2 )
Γb (Q − α2 − α32 + α3 )Γb (Q − α3 − α2 + α32 )
×
Γb (Q − α2 − α21 + α1 )Γb (Q − α2 − α1 + α21 )
Γb (Q − α32 − α1 + α4 )Γb (α32 + α1 + α4 − Q)
×
Γb (Q − α21 − α3 + α4 )Γb (α21 + α3 + α4 − Q)
Γb (α1 + α4 − α32 )Γb (α4 + α32 − α1 )
×
Γb (α3 + α4 − α21 )Γb (α21 + α4 − α3 )
Γb (2Q − 2α21 )Γb (2α21 ) 1
×
Γb (Q − 2α32 )Γb (2α32 − Q) i
i∞
Sb (U1 + s)Sb (U2 + s)Sb (U3 + s)Sb (U4 + s)
× ds (4)
Sb (V1 + s)Sb (V2 + s)Sb (V3 + s)Sb (Q + s)
−i∞
with
U1 = α21 + α1 − α2 , V1 = Q + α21 − α32 − α2 + α4 ,

U2 = Q + α21 − α2 − α1 , V2 = α21 + α32 + α4 − α2 ,
U3 = α21 + α3 + α4 − Q, V3 = 2α21 ,
U4 = α21 − α3 + α4 .
−1
The special function entering the formula above is Γb (x) ≡ ΓΓ(Q/2|b,b2 (x|b,b )
−1 ) , where
2
Γ2 (x|ω1 , ω2 ) is the Double Gamma function introduced by Barnes [25], which definition
is
∞

∂ −t
log Γ2 (x|ω1 , ω2 ) = (x + n1 ω1 + n2 ω2 ) .
∂t
n1 ,n2 =0 t =0
The function Γb (x) such defined is such that Γb (x) ≡ Γb−1 (x), and satisfies the following
functional relation
√ 1
2π bbx− 2
Γb (x + b) = Γb (x) (5)
Γ (bx)
Γb (x) is a meromorphic function of x, which poles are located at x = −nb − mb−1 ,
n, m ∈ N. The function Sb (x) is defined as Sb (x) ≡ ΓbΓ(Q−x)
b (x)
.
Let us quote the following properties:
– The Liouville fusion matrix is holomorphic in the following range [9]

| Re(α2 + α3 + α32 − Q)| < Q, | Re(α4 + α1 + α32 − Q)| < Q,
| Re(α2 + α3 − α32 )| < Q, | Re(α4 + α1 − α32 )| < Q,

| Re(α3 + α32 − α2 )| < Q, | Re(α4 + α32 − α1 )| < Q,
| Re(α2 + α32 − α3 )| < Q, | Re(α1 + α32 − α4 )| < Q.
– It satisfies the symmetry properties

α α α α α α
FαL21 ,α32 3 2 = FαL21 ,α32 4 1 = FαL21 ,α32 2 3 . (6)
α4 α1 α3 α2 α1 α4
– As the conformal blocks depends on conformal weights only, so does the Liouville
fusion matrix, i.e., is invariant when one of the αi is substituted by Q − αi .
The LFT fusion matrix was built in terms of the Racah–Wigner coefficients for an
appropriate continuous series of representations of the quantum group Uq (sl(2, R)) with
2
deformation parameter q = eiπb [9,10]. This construction ensures that the fusion matrix
satisfies the set of Moore–Seiberg equations (or polynomial equations) [26].
Let us quote the pentagonal equation:

L α3 α2 L α4 δ1 L α4 α3
dδ1 Fβ1 δ1 F F
β2 α1 β2 γ2 α5 α1 δ1 γ1 γ2 α2
Q +
2 +iR

α4 α3 γ α
= FβL2 γ1 FβL1 γ2 1 2 (7)
α5 β1 α5 α1
and the two hexagonal equations:

L α4 α2 iπ$(∆Lα1 +∆Lα2 +∆Lα3 +∆Lα4 −∆Lα21 −∆Lβ )
Fα21 ,β e
α3 α1

α3 α2 α2 α4 iπ$∆α32
= L
dα32 Fα21 ,α32 F L
e , (8)
α4 α1 α32 ,β α3 α1
Q +
2 +iR
where $ = ±.
We will also need the Liouville three point function; an explicit formula for it was
proposed in [3,4]
C L (α3 , α2 , α1 )

2
Q−α1 −α2 −α3
= πµγ b2 b 2−2b b
Υ0 Υb (2α1 )Υb (2α2 )Υb (2α3 )

× ,
Υb (α1 + α2 + α3 − Q)Υb (α1 + α2 − α3 )Υb (α1 + α3 − α2 )Υb (α2 + α3 − α1 )
(9)
where Υb (x)−1 = Γb (x)Γb (Q − x), γ (x) = Γ (x)
Γ (1−x) , Υ0 = resx=0 dΥdx
b (x)
, and µ is the
cosmological constant.
A four point function in LFT is written (if the Re(αi ), i = 1, . . . , 4 are close enough
to Q/2)
Vα4 ,α3 ,α2 ,α1 (z, z̄)

1 2
= dα21 C L (α4 , α3 , α21 )C L (Q − α21 , α2 , α1 ) Fαs 21 (α1 , α2 , α3 , α4 |z) .
2
Q
2 +iR
(10)
1.2. SL(2, C)/SU(2) WZNW model
Let us denote k the level of the current algebra; k is formally related to the coupling
constant b in LFT by k ≡ b−2 + 2. The central charge of the theory is c = k−2
3k
. The primary
fields Φ (x|z) have conformal weight ∆j = −b j (j + 1).
j 2
The action of the SL(2, C) currents on the primary fields is given by

Dja
J a (z)Φ j (x|w) ∼ Φ j (x|w), a = ±, 3
z−w
a
D
J¯a (z)Φ j (x|w) ∼
j
Φ j (x|w), (11)
z̄ − w̄
where Dja are differential operators representing the sl(2) algebra
∂ ∂ ∂
Dj+ = , Dj3 = x + j, Dj− = x 2 + 2j x, (12)
∂x ∂x ∂x
a their complex conjugates.
the Dj
Let
j
Φ 4 (∞|∞)Φ j3 (1|1)Φ j2 (x|z)Φ j1 (0|0) ≡ Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)
be a four point correlation function in the SL(2, C)/SU(2) WZNW model and Gjs21 (j1 , j2 ,
j3 , j4 |x, z) be the corresponding s-channel conformal block. It is uniquely defined as
(2)
the solution of the Knizhnik–Zamolodchikov equation (z(z − 1)∂z + b2 Dx )G s (x|z) = 0,
where

Dx(2) = x(x − 1)(x − z)∂x2 − (κ − 1)(x 2 − 2zx + z) + 2j1 x(z − 1) + 2j2x(x − 1)

+ 2j3 z(x − 1) ∂x + 2j2 κ(x − z) + 2j1 j2 (z − 1) + 2j2j3 z, (13)
where κ = j1 + j2 + j3 − j4 , and the normalization prescription

G s (x|z) ∼ z∆j21 −∆j2 −∆j1 x j1 +j2 −j21 1 + O(x) + O(z) (14)
in the limit of taking first z → 0, then x → 0. Let us note that the solutions of the KZ
equation have four singular points, located at z = 0, 1, x, ∞, and that there is no closed
form known for the conformal blocks in general.
The monodromy of the conformal blocks (or fusion matrix of the H3+ WZNW model)
is defined as
Gjs21 (j1 , j2 , j3 , j4 |x, z)

− 12 +i∞

H+ j3 j2 t
= dj32 Fj213j32 G (j , j , j , j |1 − x, 1 − z). (15)
j4 j1 j32 1 2 3 4
− 12 −i∞
It is our aim to compute this quantity.

We will also need the expression for the three point function [11]
+
C H3 (j3 , j2 , j1 )

2 1+j1 +j2 +j3 C0 (b)Υb (−2bj1 )
= ν(b)b−2b
Υb (−bj1 − bj2 − bj3 − b)Υb (−bj1 − bj2 + bj3 )
Υb (−2bj2 )Υb (−2bj3)
× . (16)
Υb (−bj1 − bj3 + bj2 )Υb (−bj2 − bj3 + bj1 )
It is known [11] that the expression for a four point correlation function is (if the
Re(ji ), i = 1, . . . , 4 are close enough to −1/2).
Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)

+ + 2
= dj21 B(j21 )C H3 (j4 , j3 , j21 )C H3 (j21, j2 , j1 ) Gjs21 (j1 , j2 , j3 , j4 |x, z) ,
− 12 +iR
(17)
where

B(j ) = (ν(b))−2j −1 γ 1 + b2 (2j + 1) .
2. Monodromy of solutions of the KZ equation
2.1. Method
It follows from a remarkable observation of [14] straightforwardly adapted to the non-

compact case that the conformal block of a special 5 point function in LFT satisfies the
same KZ equation as the conformal block in the H3+ WZNW model (for an explanation
of this relation, see [27]); hence, knowing the precise correspondence between the two
quantities allows us to compute the monodromy of the conformal block in the H3+ WZNW
model in terms of the monodromy of this 5 point LFT conformal block.
Let Fαs 21 (α1 , α2 , − 2b
1
, α3 , α4 |x, z) be the 5 point conformal block corresponding to the
LFT correlation function:

Vα4 (∞)Vα3 (1)V− 1 (x, x̄)Vα2 (z, z̄)Vα1 (0)
2b
it was shown in [14] that

1
Fα21 α1 , α2 , − , α3 , α4 |x, z
s
2b
−1 α −1 α −1 α 1 1
= xb 1 (1 − x)b 3 (x − z)b 2 z 2 γ12 (1 − z) 2 γ13 Gjs21 (j1 , j2 , j3 , j4 |x, z), (18)
where the parameters of the two theories are related by

2α1 = −b j1 + j2 − j3 − j4 − b−2 − 1 ,

2α3 = −b −j1 + j2 + j3 − j4 − b−2 − 1 ,
2α2 = −b(j1 + j2 + j3 + j4 + 1),

2α4 = −b −j1 + j2 − j3 + j4 − b−2 − 1 ,
1
α21 = −bj21 + , α32 = −bj32 + 2b1
,
2b
γ12 = 4b j1 j2 − 4α1 α2 ,
2 γ23 = 4b j2 j3 − 4α2 α3 .
2
One reads off from (18) the following relation between the two monodromies:

H3+ j3 j2 1
iπ$b−1 α2 − 2b α3 α2
Fj21 ,j32 =e Mα21 ,α32 , (19)
j4 j1 $ α4 α1 $
where $ = ±1 depends whether Im(x − z) is negative or positive (this was first seen in [18,
19]). This leads to an ambiguity in the definition of the H3+ WZNW fusion matrix, for this
reason we shall add to the fusion matrix the subscript $. Nevertheless, this ambiguity does
not appear in the bootstrap, where one considers both the holomorphic conformal block
and its antiholomorphic counterpart.
2.2. Monodromy of the Liouville 5 point conformal block
The monodromy of this special 5 point conformal block decomposes in a succession of

elementary braiding and fusing transformations.
Let us remember that the braiding is related to the fusion the following way [26]:

α4 α2 α21 +∆α32 −∆α3 −∆α4 ) L
iπ$(∆L L L L α3 α2
Bα−$21 ,α32 α α
= e Fα21 ,α32 α α . (20)
3 1 4 1
It is then straightforward to obtain the monodromy for this 5 point conformal block thanks
to the picture:

− 2b
1
α2
= B −$
s=±
α21 ,α1 −s 2b
1
α21 − 2b
1
α1
×

α3 α2
= dα32 FαL − 1 ,α
21 2b 32 α4 α1 − s 2b
1

α32 − 2b
1
= FαL −s
2b ,α32 − 2b
1 1
1 α4 α1
where the first picture (top left) represents the s-channel 5 point conformal block, and the
last one (bottom right) the t-channel 5 point conformal block.
Collecting the terms together, we get:

−1 α α
Mα212b,α32 3 2
α4 α1 $

iπ$b−1 (−α21 +α1 ) L α1 − 2b 1
=e F−+
α2 α21 − 2b
1

α32 − 2b 1 α3 α2
× F++ L
FαL − 1 ,α
21 2b 32 α4 α1 −
1
α4 α1 2b

−1 α1 − 2b 1
+ eiπ$b (−α21 −α1 +Q) F−− L
α2 α21 − 2b1

α32 − 2b 1 α3 α2
× F−+ L
FαL − 1 ,α 1 . (21)
α4 α1 21 2b 32 α4 α1 +
2b
2.3. Fusion matrix of the H3+ WZNW model
The formula for the H3+ WZNW fusion matrix follows (see Appendix A for this special
value of the Liouville fusion coefficient when one of the αi , i = 1, . . . , 4 is equal to
−b−1 /2):

H3+ j3 j2 iπ$b−1 α2 − 2b
1
α3 α2
Fj21 ,j32 =e Mα21 ,α32 . (22)
j4 j1 $ α4 α1 $
We find

H+ j3 j2 πΓ (2 + b−2 + 2j21)Γ (1 + 2j32)
Fj213,j32 =
j4 j1 $ sin π(j1 + j2 − j3 − j4 )

eiπ$(j21 −j1 −j2 )
×
Γ (2 + b−2 + j21 + j1 + j2 )Γ (j21 − j3 − j4 )
1
×
Γ (1 + j32 − j1 + j4 )Γ (1 + j3 − j2 + j32 )

−bj3 + 1/2b −bj2
× F−bj21 ,−bj32 +1/2b
L
−bj4 + 1/2b −bj1
eiπ$(j21 −j3 −j4 )
−
Γ (2 + b−2 + j21 + j3 + j4 )Γ (j21 − j1 − j2 )
1
×
Γ (1 + j32 + j1 − j4 )Γ (1 − j3 + j2 + j32 )

−bj3 −bj2 + 1/2b
× F−bj
L
21 ,−bj32 +1/2b −bj −bj + 1/2b
. (23)
4 1
It is possible to obtain an alternative representation for the H3+ fusion matrix: if we

consider the pentagonal Eq. (7) applied to the LFT fusion matrix in the special case
where α2 = − 2b 1
, and if we use the fusion rules of [28], then the LFT fusion matrices
in (7) which contain α2 = − 2b 1
are some residues of the general fusion coefficients (4).
Their expressions are given in Appendix A. We then obtain an equation that permits us to
reexpress each LFT fusion matrix in (21)

α3 α2 α3 α2
FαL − 1 ,α , F L
21 2b 32 α4 α1 − ,α32 α4 α1 + 1
1 1
α21 − 2b
2b 2b
in terms of

α3 α2 − 2b
1
α3 α2 + 2b
1
FαL − 1 ,α − 1 and FαL − 1 ,α − 1 .
21 2b 32 2b α4 α1 21 2b 32 2b α4 α1
Rearranging the terms together, one gets the following representation:

H3+ j3 j2
Fj21 ,j32
j4 j1 $
πΓ (2 + b−2 + 2j21 )Γ (−2j32 − b−2 − 1)
=
sin π(j1 + j2 + j3 + j4 + 2 + b−2 )
−1
Γ (2 + b−2 + j21 + j1 + j2 )Γ −1 (2 + b−2 + j21 + j3 + j4 )
×
Γ (−1 − b−2 − j32 − j2 − j3 )Γ (−1 − b−2 − j32 − j1 − j4 )

−bj3 −bj2
× F−bj
L
21 ,−bj32 −bj −bj
4 1
−2
e−iπ$(j1 +j2 +j3 +j4 +2+b )
−
Γ (j21 − j1 − j2 )Γ (j21 − j3 − j4 )Γ (1 − j32 + j2 + j3 )Γ (1 − j32 + j1 + j4 )

−bj3 + 2b
1
−bj2 + 2b1
× F−bj21 ,−bj32
L
. (24)
−bj4 + 2b
1
−bj1 + 2b1
We can use the same trick again by setting this time α3 = − 2b 1

in (7), so that we can
reexpress

α3 α2 α3 α2
FαL − 1 ,α , F L
21 2b 32 α4 α1 − ,α32 α4 α1 + 1
1 1
α21 − 2b
2b 2b
in terms of

α3 α2 α3 α2
FαL21 ,α32 and FαL21 ,α32 .
α4 − 2b
1
α1 α4 + 2b
1
α1
This gives us a third representation for the H3+ fusion matrix:

H3+ j3 j2
Fj21 ,j32
j4 j1 $
πΓ (−2j21)Γ (1 + 2j32)
=
sin π(−j1 + j2 − j3 + j4 )

eiπ$(j21 +j32 −j1 −j3 )
×
Γ (−j21 − j1 + j2 )Γ (−j21 − j3 + j4 )
1
×
Γ (1 + j32 − j2 + j3 )Γ (1 + j32 + j1 − j4 )

−bj3 + 2b 1
−bj2
× F−bj
L
1 1
21 + 2b ,−bj32 + 2b −bj4 −bj1 + 2b
1
eiπ$(j21 +j32 −j2 −j4 )

−
Γ (−j21 + j1 − j2 )Γ (j21 + j3 − j4 )
1
×
Γ (1 + j32 + j2 − j3 )Γ (1 + j32 − j1 + j4 )

−bj3 −bj2 + 2b
1
× F−bj
L
. (25)
21 + 2b ,−bj32 + 2b
1 1
−bj4 + 2b 1
−bj1
Consistency checks and remarks
1. One might wonder why on the first picture we did not also consider the case where
the fusion between α21 and − 2b 1
gives α21 + 2b1
(this comment also applies for the last
picture replacing α21 by α32 ). It is not difficult to see that if we do the same reasoning
starting with α21 + 2b
1
, we would simply have to replace j21 by −j21 − 1 in the expression
for the fusion matrix. As we consider j21 being of the form − 12 + is with s any real
number, it is enough to consider only the case where the result of the fusion is α21 − 2b 1
(resp. α32 − 2b ).
1
2. The case when one of the external spins corresponds to a sl(2) ˆ degenerate
representation, the other external spins being generic can be found in [18]. The fusion
coefficients appear as residues of the general fusion coefficients, and the monodromy
becomes finite dimensional. We checked in this case that it is indeed possible to find a
basis for the conformal blocks in which the fusion transformation takes a block diagonal
form, reproducing thus the results of [18].
The simplest cases j2 = 1/2, j2 = 1/2b2 and j2 = −k/2 can be found in the Appendix B.
3. The SU(2) WZNW model is obtained by substituting b by ib and by giving the spins
j half integer or integer values in the range 0 j k, the level k being an integer in this
case. One also has to substitute to Liouville theory the (k + 2, 1) Minimal Model with
central charge
6(p − q)2
c=1− , p = k + 2 ≡ b−2 , q = 1.
pq
The second term of the sum in (24) always vanishes as the fusion rules in the SU(2) WZNW
model are such that
1 1
= =0
Γ (j21 − j1 − j2 ) Γ (−n)
with n some positive integer. Only the first term of the sum remains. Then one uses the
fusion rules j21 = j1 + j2 − n and j32 = j3 + j2 − m to rewrite the prefactor in front of the
Liouville fusion matrix. This gives the ($ independent) result:

SU(2) j3 j2
Fj21 ,j32
j4 j1
Γ (2 − b−2 + 2j21 )Γ (2 − b−2 + j32 + j3 + j2 )Γ (2 − b−2 + j32 + j1 + j4 )
=
Γ (2 − b−2 + 2j32 )Γ (2 − b−2 + j21 + j1 + j2 )Γ (2 − b−2 + j21 + j3 + j4 )

−bj3 −bj2
× F−bj
MM
21 ,−bj32 −bj −bj
. (26)
4 1
4. It is straightforward to check the following symmetry properties

H3+ j3 j2 H3+ j2 j3 H3+ j4 j1
Fj21 ,j32 = Fj21 ,j32 = Fj21 ,j32 (27)
j4 j1 $ j1 j4 $ j3 j2 $
using the fact that these properties hold for the Liouville fusion matrix.
5. Using the invariance of the LFT fusion matrix w.r.t. conformal weights, one can
notice the following additional symmetry:

H3+ j3 j2 −iπ$(j1 +j2 +j3 +j4 +2+b−2 ) H3
+
j˜3 j˜2
Fj21 ,j32 =e Fj21 ,j32 ˜ ˜ , (28)
j4 j1 $ j4 j1 $
where j˜ ≡ −j − k2 ≡ −j − 1 − 2b12 , as well as:

H3+ j3 j2 iπ$(j21 −j1 −j2 ) H3
+
j˜3 j2
Fj21 ,j32 =e F ˜ , (29)
j4 j1 $ j21 ,j32 j˜4 j1 −$

H+ j j H+ j3 j2
Fj213,j32 3 2 = eiπ$(j32 −j2 −j3 ) F ˜ 3 . (30)
j4 j1 $ j21 ,j32 j˜4 j˜1 −$
These relations are known to exist in the SU(2) WZWN model [29] (see also [30]). The
difference is that in the SU(2) case, eiπ$(...) is replaced by (−1)(...) : the reason for this is
that there is no $ dependence in the SU(2) case, as we saw above.
6. If Re(ji ), i = 1, . . . , 4 are close enough to −1/2, one is still in the range where both
the Liouville fusion matrix that appear in the formulae (23)–(25) remain holomorphic.
7. Hexagonal equation:
We believe that the Moore–Seiberg equations continue to hold even in non-compact
conformal field theories, for coherency of the operator algebra (there is no proof for this
statement in general). In the special case of LFT, it was proven in [9,10]; this allows us to
derive the following equation for the monodromy of a 5 point conformal block in LFT:
iπ$(∆L +∆L +∆L +∆L −∆L −∆L
1
− 2b α4 α2 α1 α2 α3 α4
α21 − 1 β)
Mα21 ,β e 2b
α3 α1 $
Q
2 +i∞

−1 α3 α2 − 2b
1
α2 α4 iπ$∆L
α32 − 1
= dα32 Mα212b,α32 M e 2b . (31)
α4 α1 1
α32 − 2b ,β α3 α1
$ $
Q
2 −i∞
As the monodromy for a 5 point function in LFT depends on braiding (i.e., depends on $), it
fixes the $ appearing in the exponential, which should be the same as the one parametrizing
the LFT monodromy.
It is possible to derive the relation

− 2b
1
α2 α4 − 2b
1
α4 α2 −1
M = Mα32 ,β eiπ$b β (32)
1
α32 − 2b ,β α3 α1 α α
1 3 $
$
H+
this leads us to the hexagonal equation for Fj213,q jj34 jj21 $

H+ j j
Fj213,q 4 2 eiπ$(∆j1 +∆j2 +∆j3 +∆j4 −∆j21 −∆q −j1 −j2 −j3 −j4 +j21 +q)
j3 j1 $
− 12 +i∞

H+ j3 j2 H+ j2 j4
= dj32 Fj213,j32 Fj323,q eiπ$(∆j32 −j32 ) , (33)
j4 j1 $
j3 j1 $
− 12 −i∞
where β and q are related by β = −bq + 2b 1

.
Let us note that the situation here seems to be different from the already known cases
(rational CFT’s and LFT): apparently, the fusion matrix, for $ given, satisfies one hexagonal
equation only, and not two. But as $ can take two values +1 and −1, we indeed have two
hexagonal equations in this model.
8. Pentagonal equation:
I do not have much to say about it as I don’t know how to prove it.
9. Let us introduce for short δ = j1 + j2 + j3 + j4 + 2 + b −2 , and let us consider for
example (28). This equation is a consequence of the following equality between conformal
blocks:
Gjs21 (j1 , j2 , j3 , j4 |x, z)

δ 2(j +j +1+ 1 ) 2(j3 +j2 +1+ 12 )
= xz−1 − 1 x 1 2 2b2 (1 − x) 2b
−(j1 +j2 +1+ 1

−(j3 +j2 +1+ 1
Gjs21 (j˜1 , j˜2 , j˜3 , j˜4 |x, z).
) )
×z 2b2 (1 − z) 2b2 (34)
We now insert this expression into a four point function

Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)

+ +
= dj21 B(j21 )C H3 (j4 , j3 , j21 )C H3 (j21, j2 , j1 )
− 12 +iR
2
× Gjs21 (j1 , j2 , j3 , j4 |x, z) . (35)
We can rewrite this equation thanks to the following property of the three point function
+ −2 +
C H3 (j3 , j˜2 , j˜1 ) = (ν(b))−b B(j2 )B(j1 )C H3 (j3 , j2 , j1 ), (36)
where B(j ) is the two point function.
It follows
Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)

−1 −2(j1 +j2 +1+ 1 ) −2(j3 +j2 +1+ 12 )
= B(j4 )B(j3 )B(j2 )B(j1 ) |z| 2b2 |1 − z| 2b
−1 2δ 4(j1 +j2 +1+ 1 ) 4(j3 +j2 +1+ 2 )1 −2

× xz − 1 |x| 2b2 |1 − x| 2b (ν(b))2b

+ +
× dj21 B(j21 )C H3 (j˜4 , j˜3 , j21 )C H3 (j21 , j˜2 , j˜1 )
− 12 +iR
2
× Gjs21 (j˜1 , j˜2 , j˜3 , j˜4 |x, z) (37)
from which we conclude the following relation holding at the level of correlation functions
between Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄) and Φj˜4 ,j˜3 ,j˜2 ,j˜1 (x, x̄|z, z̄):
Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)

−1
= B(j4 )B(j3 )B(j2 )B(j1 ) Φj˜4 ,j˜3 ,j˜2 ,j˜1 (x, x̄|z, z̄)
2δ 4(j +j +1+ 1 ) 4(j3 +j2 +1+ 12 ) −2
× xz−1 − 1 |x| 1 2 2b2 |1 − x| 2b (ν(b))2b
−2(j1 +j2 +1+ 1

) −2(j3 +j2 +1+ 1
)
× |z| 2b2 |1 − z| 2b2 . (38)
3. Study of the singular behavior x → z
Let us denote ψαα12,α21 (z) the Liouville chiral vertex operators. They satisfy the operator
product expansion:
− 2b
1
ψαα12,α21 (z)ψ (x)
1
α21 ,α21 − 2b

b−1 α2 α2 − 2b 1 1
α2 − 2b
∼x→z (x − z) FαL ,α − 1 1 ψα ,α − 1 (z)
α1 α21 − 2b
21 2 2b 1 21 2b

1 α2 − 2b 1 α2 + 2b1
+ (x − z) b (Q−α2 ) FαL ,α + 1 1 ψα ,α − 1 (z). (39)
21 2 2b α1 α21 − 2b 1 21 2b
If we insert this relation into (18), we find that the asymptotic behavior when x → z of the
4 point conformal block in the H3+ WZNW is related to the 4 point conformal block in
LFT:
Gjs21 (j1 , j2 , j3 , j4 |x, z)

−1 −1
∼x→z z− 2 γ12 −b α1 (1 − z)− 2 γ13 −b α3
1 1

α2 − 2b 1 1
× Fα ,α − 1
L
1 Fα21 − 1 α1 , α2 −
s
, α3 , α4 |z
21 2 2b α1 α21 − 2b 2b 2b

α2 − 2b 1
+ (x − z) b (−2α2 +Q) FαL ,α + 1
1
21 2 2b α1 α21 − 2b1

1
× Fαs − 1 α1 , α2 + , α3 , α4 |z . (40)
21 2b 2b
The first term of the sum is regular at x = z, whereas the second one is singular.
It is easy to see that the Liouville conformal block F s 1 (α1 , α2 − 2b , α3 , α4 |z) has
1
α21 − 2b
the same monodromy as F−bj s
21
(−bj1 , −bj2 , −bj3 , −bj4 |z); if we then study the behavior
when z → 0 of the first Liouville conformal block multiplied by the spatial factor in front
of the bracket, we then see that we have indeed the equality
F−bj
s
(−bj1 , −bj2 , −bj3 , −bj4 |z)
21

− 12 γ12 −b−1 α1 − 12 γ13 −b−1 α3 1
=z (1 − z) Fαs − 1 α1 , α2 − , α3 , α4 |z . (41)
21 2b 2b
A similar study for the second Liouville conformal block of the sum allows us to
rewrite (40) as
Gjs21 (j1 , j2 , j3 , j4 |x, z)

Γ (2 + b−2 + 2j21 )Γ (2 + b−2 + j1 + j2 + j3 + j4 )
∼x→z
Γ (2 + b−2 + j21 + j1 + j2 )Γ (2 + b−2 + j21 + j3 + j4 )
× F−bj21 (−bj1 , −bj2 , −bj3 , −bj4 |z)
s
Γ (2 + b−2 + 2j21)Γ (−2 − b−2 − j1 − j2 − j3 − j4 )

+
Γ (j21 − j1 − j2 )Γ (j21 − j3 − j4 )
× F−bj21 (−b j˜1, −b j˜2 , −bj˜3, −b j˜4|z)
s

−δ+j1 +j2 +1+ 12 j3 +j2 +1+ 12
× (x − z)δ z 2b (1 − z) 2b . (42)
Remarks
1. It is straightforward to check the property
Gjs21 (j˜1 , j˜2 , j˜3 , j˜4 |x, z)

δ−(j1 +j2 +1+ 1
−(j2 +j3 +1+ 1
∼x→z (x − z)−δ z
) )
2b2 (1 − z) 2b2
× Gjs21 (j1 , j2 , j3 , j4 |x, z). (43)

We recover here a straightforward consequence of Eq. (34).
2. Let us consider again the case of the SU(2) WZNW model: for the same reason we
mentioned previously, the factor
1
Γ (j21 − j1 − j2 )
makes the singular term of (42) vanish. Then one sees that up to a normalization of
the chiral vertex operators preserving the polynomial equations, G(z) is nothing but the
conformal block
F−bj
s
21
(−bj1 , −bj2 , −bj3 , −bj4 |z)
of the (k + 2, 1) minimal model.
3. Degenerate representations and fusion rules for sl(2)ˆ are well known [21]; in
Appendix C we show how this relation permits us to recover them.
4. It is instructive to use relation (42) to express the behavior when x → z of a
correlation function in H3+ in terms of two correlation functions in Liouville field theory.
We consider
Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)

+ +
= dj21 B(j21 )C H3 (j4 , j3 , j21 )C H3 (j21 , j2 , j1 )
D
2
× Gjs21 (j1 , j2 , j3 , j4 |x, z) , (44)
the external spins are such that Re(ji ), i = 1, . . . , 4 are close enough to −1/2, and
D = − 12 + iR. Inserting (42), one then sees that the prefactor in front the conformal
block |F−bj
s
21
(−bj1 , −bj2 , −bj3 , −bj4 |z)|2 multiplied by the H3+ three point function
recombines to give the Liouville three point function
C L (−bj4 , −bj3 , Q + bj21)C L (−bj21 , −bj2 , −bj1 ).

Similarly the prefactor of the conformal block |F−bj s
21
(−b j˜1 , −b j˜2, −b j˜3, −b j˜4 |z)|2
(using the property that the Liouville conformal blocks depend on the Liouville conformal
weights) multiplied by the H3+ three point function recombines to give the Liouville three
point function

−1
B(j1 )B(j2 )B(j3 )B(j4 ) C L (−b j˜4, −b j˜3 , Q + bj21 )C L (−bj21, −b j˜2 , −b j˜1).
We have
Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)

∼x→z a dj21 C L (−bj4 , −bj3 , Q + bj21 )C L (−bj21 , −bj2 , −bj1 )
D
2
× F s−bj21 (z)

−2δ+2(j +j +1+ 1
) 2(j +j +1+ 1
)
+ |x − z|2δ |z| 1 2
2b2 |1 − z|
3 2
2b2

−1
× B(j1 )B(j2 )B(j3 )B(j4 )

× b dj21 C L (−b j˜4 , −bj˜3, Q + bj21 )C L (−bj21 , −bj˜2, −b j˜1)
D
s 2
× F

−bj21 (z)
(45)
with a = γ (j1 + j2 + j3 + j4 + 2 + b −2), b = γ (−j1 − j2 − j3 − j4 − 2 − b −2). A LFT four

point correlation function should factorize over the domain D = − 12 − 2b12 + iR; so we
have to deform the contour of integration from D = − 12 + iR to D = − 12 − 2b12 + iR
in order to rewrite the expression in terms of correlation functions in LFT. While we
deform the contour, we pick up a finite number of those poles −bjp which are in the range
b Q
2 < Re(−bjp ) < 2 , that come from the Liouville three point functions of the regular term
(there are no poles coming from the Liouville three point functions of the singular term in
this case). It would be of course possible to give whatever values we like for the external
spins (for example we could consider them to be real), then the residues that would be
picked up depend on a case by case study, as the poles jp depend on the values of the
external spins j1 , . . . , j4 .
Hence we can rewrite the behavior of the 4 point function in the H3+ WZNW model in
terms of 4 point functions in LFT:
Φj4 ,j3 ,j2 ,j1 (x, x̄|z, z̄)

∼x→z aV−bj4 ,−bj3 ,−bj2 ,−bj1 (z, z̄) + bV−bj˜4 ,−bj˜3 ,−bj˜2 ,−bj˜1 (z, z̄)

−1
× B(j1 )B(j2 )B(j3 )B(j4 )
−2δ+2(j1 +j2 +1+ 1
) 2(j3 +j2 +1+ 1
)
× |x − z|2δ |z| 2b2 |1 − z| 2b2
+ (Residues). (46)
5. If we make the same analysis as above studying this time the behavior when x → 1
of the conformal block, we find (this is a consequence of a relation already noticed in [31]):

z
s −j2 −j2 s ˜ ˜
Gj21 (j1 , j2 , j3 , j4 |x, z) ∼x→1 z (1 − z) Gj˜ j1 , j2 , j3 , j4 , z . (47)
21 x
It is of course possible to make an analysis similar to the one made above to get some
relation at the level of correlation functions.
Concluding remarks
There are several points that deserve further study:
• It remains of course to understand precisely the physical meaning of the singularity of

the conformal blocks when x ∼ z.
• It is straightforward to construct in the case of the spherical branes [2] the boundary
three point function along the lines of [8]: the normalization for the boundary operators
was computed in [2], and the fusion matrix in this case is like the one of Eq. (26) (by
changing b−2 into −b −2 , and the (k + 2, 1) minimal model replaced by LFT). As for
the AdS2 branes, it seems to be a problem to construct a cyclic invariant boundary three
point function, as well as to recover the boundary two point function of [2].
• It is worth trying to generalize this method to the supersymmetric case: this could
maybe lead to a proof at the level of correlation functions for the duality between
N = 2 Liouville and the superconformal SL(2)/U (1) model [32]; maybe the results
presented here work can also help at proving rigorously the duality conjectured by [33]
between the sin-Liouville theory and SL(2)/U (1) WZNW model.
• It would be very interesting to build the fusion matrix as a 6j symbol of a quantum
group. The proposal of [12] is the “pair” Uq (sl(2)), Uq (osp(1|2)). In particular, such
a construction would ensure the validity of the pentagonal equation. We hope to be
able to say more about this problem in the future.
Acknowledgements
It is a pleasure to thank V.A. Fateev, I. Kostov, G. Mennessier, V. Schomerus,

Ch. Schweigert, Al.B. Zamolodchikov and J.-B. Zuber for discussions and interest in this
work.
The author would like to thank V.B. Petkova for clarifying correspondence and for
pointing out useful references after the first version of this paper had appeared.
Work supported by EU under contract HPRN-CT-2000-00122.
Appendix A. Some residues of the Liouville fusion matrix
It is well known that in the case where one of α1 , . . . , α4 , say αi equals − n2 b − m2 b−1
where n, m ∈ N and where a triple (∆α4 , ∆α3 , ∆α21 ), (∆α21 , ∆α2 , ∆α1 ) which contains ∆αi
satisfies the fusion rules of [28], one will find that the fusion coefficients that multiply the
conformal blocks are residues of the general fusion coefficient.
In the case where α2 = − 12 b, the fusion rules are:

α21 = α1 − s b2 ,
α32 = α3 − s b2 ,
where s, s = ±.
There are four entries for the fusion matrix in this special case

α −b/2
FαL1 −sb/2,α3−s b/2 3 ≡ Fs,s
L

α4 α1
which expressions are well known to be:
Γ (b(2α1 − b))Γ (b(b − 2α3 ) + 1)
F++ = ,
Γ (b(α1 − α3 − α4 + b/2) + 1)Γ (b(α1 − α3 + α4 − b/2))
Γ (b(2α1 − b))Γ (b(2α3 − b) − 1)

F+− = ,
Γ (b(α1 + α3 + α4 − 3b/2) − 1)Γ (b(α1 + α3 − α4 − b/2))
Γ (2 − b(2α1 − b))Γ (b(b − 2α3 ) + 1)
F−+ = ,
Γ (2 − b(α1 + α3 + α4 − 3b/2))Γ (1 − b(α1 + α3 − α4 − b/2))
Γ (2 − b(2α1 − b))Γ (b(2α3 − b) − 1)
F−− = . (48)
Γ (b(−α1 + α3 + α4 − b/2))Γ (b(−α1 + α3 − α4 + b/2) + 1)
The dual case where α2 = −b −1 /2 is obtained by substituting b by b−1 .
Appendix B. Special cases of the H3+ fusion matrix
ˆ
Degenerate representations and fusion rules are well known for sl(2) [21]. We will study
three easy cases.
1. j2 = 1/2
This case was first derived in [14].
The fusion rules are j21 = j1 + 1/2 (+), j21 = j1 − 1/2 (−). We are in the case where
the second term of the sum (24) always vanishes. It is straightforward to use the results of
Appendix A to get
H+ Γ (b2 (−2j1 − 1))Γ (b2 (1 + 2j3 ) + 1)

F++3 = ,
Γ (b2 (−j1 + j3 + j4 + 1/2) + 1)Γ (b2 (−j1 + j3 − j4 − 1/2))
H+ Γ (b 2(−2j1 − 1))Γ (b2 (−2j3 − 1))
F+−3 = ,
Γ (−b2 (j 1 + j3 + j4 + 3/2))Γ (b (−j1 − j3 + j4 − 1/2)
2
H+ Γ (1 + b2 (1 + 2j1))Γ (1 + b2 (1 + 2j3))
F−+3 = ,
Γ (1 + b2 (j1 + j3 + j4 + 3/2)))Γ (1 + b2(j1 + j3 − j4 + 1/2))
H+ Γ (1 + b2 (1 + 2j1 ))Γ (−b2 (2j3 + 1))
F−−3 = − .
Γ (b2 (j 1 − j3 − j4 − 1/2))Γ (b (j1 − j3 + j4 + 1/2) + 1)
2
2. j2 = 1/2b2
This case can be found in [18], but as we use the same normalisation as [11], we shall
compare directly with this last paper.
The fusion rules are
1 1
j21 = j1 + (+), j21 = j1 − (−), j21 = −j1 − k/2 (x).
2b2 2b2
We find it more convenient to work directly with the representation (24). The second LFT
fusion matrix of the sum is equal to 1 in the cases:
1 1
j21 = j1 − , or j21 = −j1 − 1 − ,
2b2 2b2
1 1
j32 = j3 − 2 , or j32 = −j3 − 1 − 2 ,
2b 2b
and equal to zero otherwise. It is then straightforward to compute
H+ Γ (−2j1 − b−2 )Γ (2j3 + b−2 + 1)

F++3 = ,
Γ (−j1 + j3 + j4 + b−2 /2 + 1)Γ (−j1 + j3 − j4 − b−2 /2)
H+ Γ (−2j1 − b−2 )Γ (−2j3 − 1)
F+−3 = ,
Γ (−j1 − j3 − j4 − 1 − b−2 /2)Γ (−j1 − j3 + j4 − b−2 /2)
H+ Γ (−2j1 − b−2 )Γ (2j3 + 1)Γ (−2j3 − 1 − b−2 )
F+x3 = ,
Γ (−b−2 )Γ (−j1 + j3 − j4 − b−2 /2)Γ (−j1 − j3 + j4 − b−2 /2)
H+ Γ (2j1 + 2)Γ (2j3 + b−2 + 1)
F−+3 = ,
Γ (j1 + j3 + j4 + b−2 /2 + 2)Γ (j1 + j3 − j4 + 1 + b−2 /2)
H+
Fx+3
Γ (−2j1 )Γ (2j1 + 2 + b−2 )Γ (2j3 + b−2 + 1)
= .
Γ (1 + b−2 )Γ (−j1 + j3 + j4 + b−2 /2 + 1)Γ (j1 + j3 − j4 + 1 + b−2 /2)
(49)
For other cases the second LFT fusion matrix contributes; we find:
−2
H+ e−iπ$b Γ (2j1 + 2)Γ (−2j3 − 1)
F−−3 = ,
Γ (j1 − j3 − j4 − b−2 /2)Γ (j1 − j3 + j4 + 1 + b−2 /2)
−2
H+ e−iπ$(2j3 +b ) Γ (2j1 + 2)Γ (2j3 + 1)Γ (−2j3 − 1 − b−2 )
F−x3 = − ,
Γ (−b−2 )Γ (j1 + j3 + j4 + 2 + b−2 /2)Γ (j1 − j3 − j4 − b−2 /2)
−2
H+ e−iπ$(2j1 +b ) Γ (−2j1)
Fx−3 = −
Γ (1 + b )Γ (−j1 − j3 − j4 − 1 − b−2 /2)
−2
Γ (2j1 + 2 + b−2 )Γ (−2j3 − 1)

× ,
Γ (j1 − j3 + j4 + 1 + b−2 /2)
H+
Fxx3

= Γ (−2j1 )Γ (2j3 + 1)Γ 2j1 + 2 + b−2 Γ −2j3 − 1 − b−2
sin π(−b−2) sin π(j1 − j3 − j4 − b−2 /2) sin π(−j1 + j3 − j4 − b−2 /2)
×
π 2 sin π(j1 + j3 + j4 + 2 + 3b−2 /2)
−2
e−iπ$(j1 +j3 +j4 +2+3b /2) Γ (−2j1 )Γ (2j3 + 1)
−
Γ (−2j1 − 1 − b−2 )Γ (2j3 + 2 + b−2 )
sin π(j1 + j3 + j4 + 2 + b−2 /2)
× .
sin π(j1 + j3 + j4 + 2 + 3b−2 /2)
H+
These coefficients are in agreement with [11] for the choice $ = 1 , excepted for Fxx3 ,
where there seems to be a slight discrepancy in some of the arguments of the gamma
functions. (We did not redo the computations of [11], where the fusion matrix was obtained
thanks to an explicit integral representation for the conformal blocks.)
3. j2 = −k/2
This elementary case was explicitly mentionned in [14]. The fusion rule is j21 = −j1 −
k/2. The first term of the sum vanishes and the second LFT fusion matrix is equal to one.
It remains to evaluate the product of gamma functions; we find:
+ −iπ$(j1 +j3 +j4 +1+ 1
)
F H3 = e b2 .
ˆ
Appendix C. Fusion rules for sl(2) algebra at generic level
ˆ
We present here a way to find degenerate representations and fusion rules for sl(2) from
the knowledge of the degenerate representations and fusion rules of the Virasoro algebra.
We derived in the main text the following asymptotic behavior when x → z for the
conformal blocks of the SL(2, C)/SU(2) WZNW, relating them to two Liouville conformal
blocks:
Gjs21 (j1 , j2 , j3 , j4 |x, z)

Γ (2 + b−2 + 2j21 )Γ (2 + b−2 + j1 + j2 + j3 + j4 )
∼x→z
Γ (2 + b−2 + j21 + j1 + j2 )Γ (2 + b−2 + j21 + j3 + j4 )
× F−bjs
21
(−bj1 , −bj2 , −bj3 , −bj4 |z)
Γ (2 + b−2 + 2j21)Γ (−2 − b−2 − j1 − j2 − j3 − j4 )
+
Γ (j21 − j1 − j2 )Γ (j21 − j3 − j4 )
× F−bj
s
21
(−bJ1 , −bJ2 , −bJ3 , −bJ4 |z)

−2 −j3 −j4 −1− 1
j3 +j2 +1+ 1
× (x − z)j1 +j2 +j3 +j4 +2+b z 2b2 (1 − z) 2b2
with Ji = ji − 2b12 .
Although there is no closed form known for the Liouville conformal blocks, they are
completely determined by the conformal symmetry. They depend on conformal weights
only, i.e., are invariant when −bji (resp. −bJi ) is changed into Q + bji (resp. Q + bJi ).
1. Degenerate representations of the Virasoro algebra.

It is well known that the case where ji (resp. Ji ) equals n2 + m2 b−2 with n, m non-negative
integers, corresponds to a degenerate Virasoro representation V−bji (resp. V−bJi ). The
conformal block F−bj
s
21
then only exists for a finite number of values of −bj21 [28]:
−bj21 = −bj1 + bj2 − ub − vb−1 ,

where u, v are integers such that 0 u n, 0 v m.
It would be equivalent to write Q + bj21 instead of −bj21 in Eq. (50) since the Virasoro
representations V−bj and VQ+bj are equivalent. We will see that this fact will play an
important role in the determination of the degenerate representations and fusion rules
ˆ
for sl(2).
ˆ
2. Degenerate representations and fusion rules for sl(2).
Let us call for short F 1 the first Liouville conformal block of the sum (50) and the second
one F 2 .
ˆ
Claim 1. The spin jm,n that labels the degenerate representation of sl(2) Pjn,m are also
labels for the degenerate Virasoro representation V−bjn,m or V−bJn,m .
ˆ
In other words, it means that Pjn,m is a degenerate representation of sl(2) iff:
n m −2
(a) jn,m = + b or
2 2
n m 1 −2
(b) jn,m = − +1 − + b (50)
2 2 2
with n, m non-negative integers.
ˆ
We now provide two rules that will allow us to determine the sl(2) fusion rules:
Rule 1. If j2 is such that both V−bj2 and V−bJ2 correspond to degenerate Virasoro
ˆ
representations, then the admissible sl(2) fusion rules consist of the set of common fusions
rules plus j21 = j1 + j2 if j2 is of the form (a), and j21 = −j1 − j2 − 2 − b−2 if j2 is of the
form (b).
Rule 2. If j2 is such that V−bj2 is a degenerate Virasoro representation and not V−bJ2 (or
ˆ
converse), then the sl(2) fusion rules are such that the factor Γ −1 (j21 − j1 − j2 ) in front
of F should be equal to zero (resp. Γ −1 (2 + b −2 + j21 + j1 + j2 ) in front of F 1 ). Note
2
that this case happens only if m=0.
We shall start by three easy examples as the generalization is straightforward.

(a) Examples:
(i) j2 = 12
In this case we have −bj2 = − b2 so the Virasoro representation V− b is degenerate. The
2
fusion rules are:
b b
−bj21 = −bj1 − , −bj21 = −bj1 + , or (51)
2 2
b b
Q + bj21 = −bj1 − , Q + bj21 = −bj1 + . (52)
2 2
Let us turn to F 2 : −bJ2 = − b2 + 2b 1
does not correspond to any degenerate Virasoro
representation. We use rule 2 to select the admissible set of fusion rules, and find
j21 = j1 + j2 , j21 = j1 − j2 . (53)
(ii) j2 = 1
2b2
In this case we have −bj2 = − 2b
1
so the Virasoro representation V− 1 is degenerate. The
2b
allowed values for j21 are:
1 1
−bj21 = −bj1 − , −bj21 = −bj1 + , or
2b 2b
1 1
Q + bj21 = −bj1 − , Q + bj21 = −bj1 + . (54)
2b 2b
As for the second term, we have −bJ2 = 0, which corresponds to the identity representa-
tion. The fusion rules are −bj21 = −bj1 + 2b
1
or Q + bj21 = −bj1 + 2b
1
. The common set
of fusion rules consists of
1 1
−bj21 = −bj1 + , −bj21 = bj1 − + Q. (55)
2b 2b
According to the Rule 1, we should also include −bj21 = −bj1 − 2b 1
.
As a conclusion, we are left with the following three possibilities:
1 1
−bj21 = −bj1 − , −bj21 = −bj1 + ,
2b 2b
1
−bj21 = bj1 − + Q. (56)
2b
(iii) j2 = − k2
In this case −bj2 = b + 2b 1
does not correspond to any degenerate Virasoro representation,
and −bJ2 = Q. As VQ and V0 are equivalent Virasoro representations, we see that
−bJ2 = Q actually corresponds to the identity representation. Hence, the fusion rules are
−bj21 = −bj1 + 2b 1
or −bj21 = bj1 + b + 2b 1
. The latter rule is the only acceptable one, as
−1 −2
it makes the term Γ (2 + b + j21 + j1 + j2 ) in front of F 1 vanish.
ˆ
Let us note that the sl(2) representation P− k plays a role very similar to the identity,
2
as the decomposition of its tensor product with an arbitrary representation Pj gives the
representation P−j − k only.
2
(b) General case:
(i) j2 = n2 + m2 b−2 , with n ∈ N, m ∈ N.
The allowed values for j21 are either
j21 = j1 − j2 + u + vb−2 or
j21 = j2 − j1 − (u + 1) − (v + 1)b−2,

(57)
where 0 u n, 0 v m, 0 u n, 0 v m − 1.
(ii) j2 = −( n2 + 1) − ( m2 + 12 )b−2 , with n ∈ N, m ∈ N.
The allowed values for j21 are either
j21 = j1 − j2 − (U + 1) − (V + 1)b−2 or
−2
j21 = j2 − j1 + U + V b , (58)
where 0 U n, 0 V m − 1, n, 0 U 0V m.
These results are in agreement with [21].
References
[1] M.J. Bhaseen, I.I. Kogan, O.A. Solovev, N. Taniguchi, A.M. Tsvelik, Towards a field theory of the plateau
transitions in the integer quantum Hall effect, Nucl. Phys. B 580 (2000) 688, cond-mat/9912060.
[2] B. Ponsot, V. Schomerus, J. Teschner, Branes in the euclidean AdS3 , JHEP 0202 (2002) 016, hep-
th/0112198.
[3] H. Dorn, H.-J. Otto, Two and three point functions in Liouville theory, Nucl. Phys. B 429 (1994) 375,
hep-th/9403141.
[4] A.B. Zamolodchikov, Al.B. Zamolodchikov, Structure constants and conformal bootstrap in Liouville field
theory, Nucl. Phys. B 477 (1996) 577, hep-th/9506136.
[5] V.A. Fateev, A.B. Zamolodchikov, Al.B. Zamolodchikov, Boundary Liouville field theory I, hep-
th/0001012.
[6] A.B. Zamolodchikov, Al.B. Zamolodchikov, Liouville field theory on a pseudosphere, hep-th/0101152.
[7] K. Hosomichi, Bulk-boundary propagator in Liouville theory on a disc, JHEP 0111 (2001) 044, hep-
th/0108093;
Al.B. Zamolodchikov, Conference on Liouville field theory, Montpellier, January 1998, unpublished.
[8] B. Ponsot, J. Teschner, Boundary Liouville field theory: boundary three point function, Nucl. Phys. B 622
(2002) 309, hep-th/0110244.
[9] B. Ponsot, J. Teschner, Liouville bootstrap via harmonic analysis on a noncompact quantum group, hep-
th/9911110.
[10] B. Ponsot, J. Teschner, Clebsch–Gordan and Racah–Wigner coefficients for a continuous series of
representations of Uq (sl(2, R)), Commun. Math. Phys. 224 (2001) 3, math-QA/0007097.
[11] J. Teschner, On structure constants and fusion rules in the SL(2, C)/SU(2) WZNW model, Nucl. Phys. B 546
(1999) 390, hep-th/9712256.
ˆ
[12] B. Feigin, F. Malikov, Modular functor and representation theory of sl(2) at a rational level, q-alg/9511011.
[13] J. Teschner, Crossing symmetry in the H3+ WZNW model, Phys. Lett. B 521 (2001) 127, hep-th/0108121.
[14] A.B. Zamolodchikov, V.A. Fateev, Operator algebra and correlation functions in the two-dimensional
SU(2) × SU(2) chiral Wess–Zumino model, Sov. J. Nucl. Phys. 43 (4) (1986) 657.
[15] V.G. Knizhnik, A.B. Zamolodchikov, Current algebra and Wess–Zumino model in two dimensions, Nucl.
Phys. B 247 (1984) 83.
[16] R.E. Behrend, P.A. Pearce, V.B. Petkova, J.-B. Zuber, Boundary conditions in rational conformal field
theories, Nucl. Phys. B 579 (2000) 707, hep-th/9908036.
[17] I. Runkel, Boundary structure constant for the A-series Virasoro minimal models, Nucl. Phys. B 549 (1999)
563, hep-th/9811178.
(1)
[18] P. Furlan, A.Ch. Ganchev, V.B. Petkova, A1 admissible representations-fusion transformations and local
correlators, Nucl. Phys. B 491 (1997) 635, hep-th/9608018.
[19] J.L. Peterssen, J. Rasmussen, M. Yu, Fusion, crossing and monodromy in conformal field theory base on
SL(2) current algebra with fractional level, Nucl. Phys. B 481 (1996) 577, hep-th/9607129.
[20] V.S. Dotsenko, V.A. Fateev, Conformal algebra and multipoint correlation functions in two-dimensional
statistical models, Nucl. Phys. B 240 (1984) 312;
V.S. Dotsenko, V.A. Fateev, Four point correlations functions and the operator algebra in the two-
dimensional conformal invariant theories with the central charge c < 1, Nucl. Phys. B 251 (1985) 691.
ˆ
[21] H. Awata, Y. Yamada, Fusion rules for the fractional level sl(2) algebra, Mod. Phys. Lett. A 7 (13) (1992)
1185–1195.
[22] J. Maldacena, H. Ooguri, Strings in AdS3 and the SL(2, R) WZW model. Part 3: correlation functions,
hep-th/0111180.
[23] P. Furlan, A.Ch. Ganchev, R. Paunov, V.B. Petkova, Reduction of the rational spin SL(2, C) WZNW
conformal theory, Phys. Lett. B 267 (1991) 63;
P. Furlan, A.C. Ganchev, R. Paunov, V.B. Petkova, Solutions of the Knizhnik–Zamolodchikov equations
with rational isospins and reduction to the minimal models, Nucl. Phys. B 394 (1993) 665, hep-th/9201080.
[24] J. Teschner, Liouville theory revisited, Class. Quantum Grav. 18 (2001) R153–R222, hep-th/0104158.
[25] E.W. Barnes, The theory of the double gamma function, Philos. Trans. R. Soc. London Ser. A 196 (1901)
265.
[26] G. Moore, N. Seiberg, Classical and quantum conformal field theory, Commun. Math. Phys. 123 (1989)
177.
[27] A. Stoyanovsky, A relation between the Knizhnik–Zamolodchikov and Belavin–Polyakov–Zamolodchikov
systems of partial differential equations, math-ph/0112013.
[28] B.L. Feigin, D.B. Fuchs, Representation of the Virasoro algebra, in: A.M. Vershik, D.P. Zhelobenko (Eds.),
Representations of Lie Groups and Related Topics, Gordon and Breach, London, 1990.
[29] P. Furlan, A.Ch. Ganchev, V.B. Petkova, Fusion matrices and c < 1 (quasi) local conformal theories, Int. J.
Mod. Phys. A 5 (1990) 2721.
[30] J.K. Singerland, F.A. Bais, Quantum groups and non-abelian braiding in quantum Hall systems, Nucl.
Phys. B 612 (2001) 229–290, cond-mat/0104035.
[31] A. Parnachev, D.A. Sahakyan, Some remarks on D-branes in AdS3 , JHEP 0110 (2001) 022, hep-th/0109150.
[32] A. Giveon, D. Kutasov, O. Pelc, Holography for non-critical superstrings, JHEP 9910 (1999) 035, hep-
th/9907178;
A. Giveon, D. Kutasov, Little string theory in a double scaling limit, JHEP 9910 (1999) 034, hep-th/9909110;
A. Giveon, D. Kutasov, Comments on double scaled little string theory, JHEP 0001 (2000) 023, hep-
th/9911039.
[33] V.A. Fateev, A.B. Zamolodchikov, Al.B. Zamolodchikov, unpublished.
A three-family standard-like orientifold model:

Yukawa couplings and hierarchy
Mirjam Cvetič a,b,e , Paul Langacker a,b,c,d,∗ , Gary Shiu a,b
a Department of Physics and Astronomy, University of Pennsylvania, Philadelphia, PA 19104, USA
b Institute for Theoretical Physics, University of California, Santa Barbara, CA 93106, USA
c School of Natural Sciences, Institute for Advanced Study, Princeton, NJ 08540, USA
d Department of Physics, University of Wisconsin, Madison, WI 53706, USA
e Isaac Newton Institute for Mathematical Sciences, University of Cambridge, Cambridge, UK
Received 18 June 2002; received in revised form 31 July 2002; accepted 5 August 2002
Abstract
We discuss the hierarchy of Yukawa couplings in a supersymmetric three family standard-like
string model. The model is constructed by compactifying type IIA string theory on a Z2 × Z2
orientifold in which the standard model matter fields arise from intersecting D6-branes. When lifted
to M-theory, the model amounts to compactification of M-theory on a G2 manifold. While the actual
fermion masses depend on the vacuum expectation values of the multiple Higgs fields in the model,
we calculate the leading worldsheet instanton contributions to the Yukawa couplings and examine
the implications of the Yukawa hierarchy.
PACS: 11.25.-w
1. Introduction
The basic premise of string phenomenology is to explore the constructions and the
particle physics implications of four-dimensional string solutions with phenomenologically
viable features (i.e., solutions which give rise to an effective theory containing the
standard model). The moduli space of different compactifications of string theory is
highly degenerate at the perturbative level, and so we are faced with the poorly
understood question of how the string vacuum describing the observable world is selected.
* Corresponding author.
E-mail address: pgl@electroweak.hep.upenn.edu (P. Langacker).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 8 4 - 3
140 M. Cvetič et al. / Nuclear Physics B 642 (2002) 139–156
Nevertheless, by exploring models with quasi-realistic features from various corners of

M-theory, one may deduce some generic physical implications of string derived models.
Prior to the second string revolution, the focus of string phenomenology was on the
construction of such solutions within the framework of the weakly coupled heterotic string.
Over the years, many semi-realistic models have been constructed in this framework, and
the resulting phenomenology has been subsequently analyzed [1]. The richness of semi-
realistic heterotic string models is also in sharp contrast to the apparent no-go theorem
in other formulations of string theory [2]. More recently, the techniques of conformal
field theory in describing D-branes and orientifold planes allow for the construction of
quasi-realistic string models in another calculable regime of M-theory, as illustrated by
the various four-dimensional N = 1 supersymmetric type II orientifolds [3–14]. In these
models, chiral fermions appear on the worldvolume of the D-branes since they are located
at orbifold singularities in the internal space.
Another promising direction to obtain chiral fermions, which has only recently been
exploited in model building, is to consider branes at angles. The spectrum of open strings
stretched between branes at angles may contain chiral fermions which are localized at
the intersection of branes [15]. This fact (or its T-dual version, i.e., branes with flux)
was employed in [16–24] in constructing semi-realistic brane world models. However,
the semi-realistic models considered in this context are typically non-supersymmetric, and
the stability of non-supersymmetric models (and the dynamics involved in restabilization)
is not fully understood. This was one of the motivations of [25–27] in constructing
chiral supersymmetric orientifold models with branes at angles. The constraints on
supersymmetric four-dimensional models are rather restrictive. Despite the remarkable
progress in developing techniques of orientifold constructions, there is only one orientifold
model [25–27] that has been constructed so far with the ingredients of the MSSM:1 N = 1
supersymmetry, the Standard Model gauge group as a part of the gauge structure, and
candidate fields for the three generations of quarks and leptons as well as the electroweak
Higgs doublets.
The general class of supersymmetric orientifold models considered in [25–27] corre-
sponds (in the strong coupling limit) to M-theory compactification on purely geometrical
backgrounds admitting a G2 metric, providing the first explicit realization of M-theory
compactification on compact G2 holonomy spaces that yields non-Abelian gauge groups
and chiral fermions as well as other quasi-realistic features of the Standard and GUT mod-
els. This work also sheds light on the recent results of obtaining four-dimensional chi-
ral fermions from G2 compactifications of M-theory [25,26,28–30], as further elaborated
in [27].
In this paper, we further explore the basic properties of the models, in particular the
three-family standard-like model in [25,26]. The construction, the chiral spectrum and
some of the basic features of the model were described in the original work [25,26].
The details of the chiral and non-chiral spectra, the explicit evaluation of the gauge
couplings, the properties of the two extra U (1) symmetries, and further phenomenological
1 Models with features of the Grand Unified Theories (GUTs) were also constructed in [25–27].
M. Cvetič et al. / Nuclear Physics B 642 (2002) 139–156 141
implications associated with charge confinement in the strongly coupled quasi-hidden

sector were discussed in [31].
The model is not fully realistic. In addition to the Standard Model group, there are
two additional U (1) symmetries, one of which has family non-universal and therefore
flavor changing couplings, and a quasi-hidden non-Abelian sector which becomes strongly
coupled above the electroweak scale. The perturbative spectrum contains a fourth family
of exotic (SU(2)-singlet) quarks and leptons, in which, however, the left-chiral states have
unphysical electric charges. In [31] it is argued that these could decouple from the low
energy spectrum due to hidden sector charge confinement, and that anomaly matching
requires the physical left-chiral states to be composites. The model has multiple Higgs
doublets and additional exotic states. The low energy predictions for the gauge couplings
depend on the choice moduli parameters. The study in [31] reveals that αstrong can be fitted
to the experimental value, while sin2 θW and αEM are off by about a factor of 2 and 3,
respectively.
The purpose of this paper is to carry out further the analysis of the couplings
in the model. In particular, we focus on the calculation of Yukawa couplings and
study their physical implications. The Yukawa couplings among chiral matter are due
to the world-sheet instanton contributions associated with the action of string world-
sheet stretching among intersections where the corresponding chiral matter fields are
located. The leading contribution to the Yukawa couplings is therefore proportional to
exp(−A/(2πα )) where A is the smallest area of the string world-sheet stretching among
the brane intersection points. The complete calculation of the Yukawa couplings involves
techniques of calculating correlation functions involving twisted fields in the conformal
field theory of open strings. The origin of the Yukawa couplings, i.e., their world-sheet
instanton origin and the consequences of the exponential hierarchies within intersecting
brane constructions, was first discussed and analyzed in [17]. (For related applications to
the fermion mass hierarchy within GUT intersecting brane constructions, see [32].)
The purpose of our work is to systematically evaluate the Leading order contributions
to the Yukawa couplings for the supersymmetric three family standard-like model. Even
though, we will approach the study only in the leading order of world-sheet instanton
contributions, we shall elucidate these features explicitly and discuss the consequences of
the resulting hierarchies of the Yukawa couplings. The method can also be further applied
to other constructions involving intersecting branes.
The structure of the paper is as follows. In Section 2 we briefly describe the features of
the model and the chiral spectrum. In Section 3 we focus on the calculation of the Yukawa
couplings both in the quark and lepton sectors of the model. In Section 4 we discuss some
physical implications of the hierarchical structure of these couplings and other possible
low energy implications. The conclusions are given in Section 5.
2. Brief description of the model
The model is an orientifold of type IIA on T6 /(Z2 × Z2 ). The orbifold actions

have generators θ , ω acting as θ : (z1 , z2 , z3 ) → (−z1 , −z2 , z3 ), and ω : (z1 , z2 , z3 ) →
(z1 , −z2 , −z3 ) on the complex coordinates zi of T6 , which is assumed to be factor-
Table 1
3a = m3a + 12 n3a
D6-brane configuration for the three-family model. Here, m
Type Na (n1a , m1a ) × (n2a , m2a ) × (n3a , m
3a ) Group
A1 8 (0, 1) × (0, −1) × (2, 0̃) Q8,8
A2 2 (1, 0) × (1, 0) × (2, 0̃) Sp(2)A
B1 4
(1, 0) × (1, −1) × (1, 3/2) SU(2)
B2 2 (1, 0) × (0, 1) × (0, −1̃) Sp(2)B
C1 6+2
(1, −1) × (1, 0) × (1, 1/2) SU(3), Q3 , Q1
C2 4 (0, 1) × (1, 0) × (0, −1̃) Sp(4)
izable. The orientifold action is ΩR, where Ω is world-sheet parity, and R acts by
R : (z1 , z2 , z3 ) → (z̄1 , z̄2 , z̄3 ). The model contains four kinds of O6-planes, associated with
the actions of ΩR, ΩRθ , ΩRω, ΩRθ ω. The cancellation of the RR crosscap tadpoles
requires an introduction of K stacks of Na D6-branes (a = 1, . . . , K) wrapped on three-
cycles (taken to be the product of 1-cycles (nia , mia ) in the ith two-torus), and their images
under ΩR, wrapped on cycles (nia , −mia ). In the case where D6-branes are chosen parallel
to the O6-planes, the resulting model is related by T-duality to the orientifold in [4], and is
non-chiral. Chirality is however achieved using D6-branes at non-trivial angles.
The cancellation of untwisted tadpoles imposes constraints on the number of D6-branes
and the types of 3-cycles that they wrap around. The cancellation of twisted tadpoles
determines the orbifold actions on the Chan–Paton indices of the branes (the explicit
form of the orbifold actions are given in [25,26]). The condition that the system of branes
preserves N = 1 supersymmetry requires [15] that each stack of D6-branes is related to the
O6-planes by a rotation in SU(3): denoting by θi the angles the D6-brane forms with the
horizontal direction in the ith two-torus, supersymmetry preserving configurations must
satisfy θ1 + θ2 + θ3 = 0. This in turn imposes a constraint on the wrapping numbers and
(i) (i) (i)
the complex structure moduli χi = R2 /R1 , where R1,2 are the respective sizes of the ith
two-torus.
An example leading to a three-family standard-like model massless spectrum corre-
sponds to the following case. The D6-brane configuration is provided in Table 1, and
satisfies the tadpole cancellation conditions. The configuration is supersymmetric for
χ1 : χ2 : χ3 = 1 : 3 : 2.
The rules to compute the spectrum are analogous to those in [18]. Here, we summarize
the resulting chiral spectrum in Table 2, found in [25,26], where

Iab = n1a m1b − m1a n1b n2a m2b − m2a n2b n3a m3b − m3a n3b , (1)
is the intersection number of D6a - and D6b -branes [16,17].
The chiral spectrum is given in Table 3 (see [25]). Here, we list also the chiral matter
from the aa sectors. The charges of the matter fields under various U (1) gauge fields of
the model are tabulated. The generators Q3 , Q1 and Q2 refer to the U (1) factor within the
corresponding U (n), while Q8 , Q8 are the U (1)’s arising from Higgsing the U Sp(8). The
last column provides the charges under a particular anomaly-free U (1) gauge field:
1 1 1
QY = Q3 − Q1 + Q8 + Q8 . (2)
6 2 2
Table 2
General spectrum on D6-branes at generic angles (namely, not parallel to any O6-plane in all three-tori). The
spectrum is valid for tilted tori. The models may contain additional non-chiral pieces in the aa sector and in ab,
ab sectors with zero intersection, if the relevant branes overlap
Sector Representation
aa U (Na /2) vector multiplet
3 Adj. chiral multiplets
ab + ba Iab chiral multiplets in ( a , b ) representation
ab + b a Iab chiral multiplets in ( a , b ) representation

aa + a a − 12 Iaa − 4k Ia,O6 chiral multiplets in representation
2
− 12 Iaa + 4k Ia,O6 chiral multiplets in representation
2
Table 3
The chiral spectrum of the open string sector in the three-family model. To be complete, we also list (in the bottom
part of the table, below the horizontal line) the chiral states from the aa sectors, which are not localized at the
intersections
Sector SU(3) × SU(2) × Sp(2)B Q3 Q1 Q2 Q8 Q8 QY Q8 − Q8 Field
×Sp(2)A × Sp(4)
A1 B1 3 × 2 × (1, 2̄, 1, 1, 1) 0 0 −1 ±1 0 ± 12 ±1 HU , HD
3 × 2 × (1, 2̄, 1, 1, 1) 0 0 −1 0 ±1 ± 12 ∓1 HU , HD
A1 C1 2 × (3̄, 1, 1, 1, 1) −1 0 0 ±1 0 1, −2 1, −1 U
D,
3 3
2 × (3̄, 1, 1, 1, 1) −1 0 0 0 ±1 1, −2 −1, 1 U
D,
3 3
2 × (1, 1, 1, 1, 1) 0 −1 0 ±1 0 1, 0 1, −1 N
E,
2 × (1, 1, 1, 1, 1) 0 −1 0 0 ±1 1, 0 −1, 1 N
E,
B1 C1 (3, 2̄, 1, 1, 1) 1 0 −1 0 0 1 0 QL
6
(1, 2̄, 1, 1, 1) 0 1 −1 0 0 − 12 0 L
B1 C2 (1, 2, 1, 1, 4) 0 0 1 0 0 0 0
B2 C1 (3, 1, 2, 1, 1) 1 0 0 0 0 1 0
6
(1, 1, 2, 1, 1) 0 1 0 0 0 − 12 0
B1 C1 2 × (3, 2, 1, 1, 1) 1 0 1 0 0 1
6 0 QL
2 × (1, 2, 1, 1, 1) 0 1 1 0 0 − 12 0 L
B1 B1 2 × (1, 1, 1, 1, 1) 0 0 −2 0 0 0 0
2 × (1, 3, 1, 1, 1) 0 0 2 0 0 0 0
A1 A1 3 × 8 × (1, 1, 1, 1, 1) 0 0 0 0 0 0 0
3 × 4 × (1, 1, 1, 1, 1) 0 0 0 ±1 ±1 ±1 0
3 × 4 × (1, 1, 1, 1, 1) 0 0 0 ±1 ∓1 0 ±2
3 × (1, 1, 1, 1, 1) 0 0 0 ±2 0 ±1 ±2
3 × (1, 1, 1, 1, 1) 0 0 0 0 ±2 ±1 ∓2
A2 A2 3 × (1, 1, 1, 1, 1) 0 0 0 0 0 0 0
B1 B1 3 × (1, 3, 1, 1, 1) 0 0 0 0 0 0 0
3 × (1, 1, 1, 1, 1) 0 0 0 0 0 0 0
B2 B2 3 × (1, 1, 1, 1, 1) 0 0 0 0 0 0 0
C1 C1 3 × (8, 1, 1, 1, 1) 0 0 0 0 0 0 0
3 × (1, 1, 1, 1, 1) 0 0 0 0 0 0 0
C2 C2 3 × (1, 1, 1, 1, 5 + 1) 0 0 0 0 0 0 0
This linear combination QY plays the role of hypercharge. There are two additional
non-anomalous U (1) symmetries, i.e., 13 Q3 − Q1 and Q8 − Q8 . The spectrum of chiral
multiplets corresponds to three quark–lepton generations, a number of vector-like Higgs
doublets, and an anomaly-free set of chiral matter. It includes states corresponding to the
right-handed SU(2)-singlet fields of a fourth family. However, their natural left-handed
partners, from the B2 C1 sector, have the wrong hypercharge. It is argued in [31] that these
disappear from the low energy spectrum due to the strong coupling of the first Sp(2) group,
to be replaced by composites with the appropriate quantum numbers to be the partners of
the extra family of right-handed fields.
The gauge couplings of the various gauge fields in the model (determined by the volume
of the 3-cycles that the corresponding D6-brane wraps around [25]) were calculated
explicitly in [31]. In this paper we shall focus on the Yukawa couplings.
3. Yukawa couplings
In this section we calculate the leading contribution to the Yukawa couplings among the
chiral matter fields in the model. The string theory calculation of the Yukawa couplings
requires the techniques of computing string amplitudes that involve twisted fields of
the conformal field theory describing the open strings states at each intersection. In
particular, the quantization of the open string sector associated with the string states at
the interection of two D-branes at a general angle θ involves states with the boundary
conditions that are a linear combination of Dirichlet and Neumann boundary conditions.
Thus the mode expansion is in terms of αn−k modes and the non-integer powers of the
αn−ki
world-sheet coordinate z ≡ exp(τ + iσ ), i.e., Xi ∼ n n−k z−(n−k) , where n are integers,
and k = 2π θ
. Therefore states in this sector are created by acting with αn−k (n − k < 0)
on the “twisted” vacuum σk |0, where σk is the conformal field that ensures the correct
boundary conditions on the open string states. As a consequence the string amplitude
for the three states, each of them at an intersection of two D-branes where the three
intersections form the edges of a triangle, involves
a calculation of a correlator of the
type 0|∂z Xi1 ∂z Xi2 ∂z Xi3 σk1 σk2 σk3 |0 (with 3i=1 ki = 1). (The fermionic sector of the
correlator can be determined in a straightforward way by employing the bosonisation
procedure of the world-sheet fermionic degrees of freedom.) Since each state is localized
at the intersection of the D-branes, this amplitude involves the contribution of the world-
sheet instantons, and it is thus exponentially suppressed by the area of the corresponding
intersection triangle. The results of the calculation should be analogous to Yukawa coupling
calculations for the twisted closed string states of orbifolds [33]. However, the subtleties of
the open-string sector calculations (such as the so-called “doubling trick”, that allows one
to express the open string modes in terms of the holomorphic world-sheet coordinate z,
only; z is now defined on the whole complex plane, along with the boundary conditions
for states specified on the real line) require further study.
The Yukawa couplings can therefore be expressed as a sum over the world-sheet
instantons associated with the action of the string world-sheet stretching among the
intersection points where the corresponding chiral matter fields are located. The couplings

are schematically of the form: ∞
n=1 Zn exp[−(ncn A)/(2πα )]. Here A is the smallest area
of the triangle associated with the corresponding brane intersections and α is the string
tension, related to the string scale by Ms = (α )−1/2 . (The factor 1/(2π) in the exponents
is due to the normalization of the string Nambu–Goto action with the pre-factor 1/(2πα ).)
The pre-factors Zn and the coefficients cn in the exponents are of O(1). (The coefficients
cn should in principle include the multiplicity factors due to the orbifold and orientifold
symmetries.) The leading contribution to the Yukawa couplings is therefore proportional
to Z1 exp[−(c1 A)/(2πα )]. The world-sheet instanton origin of Yukawa couplings and the
implications for hierarchies within interesecting D-branes was originally studied in [17].
At this stage we shall approach the study systematically by studying the leading order
contributions, only. Within this context we shall evaluate the intersection areas A explicitly
in terms of the moduli of internal tori. Indeed, even in the leading order in the determination
of the Yukawa couplings there remains an uncertainty, since Z1 and c1 , which are
coefficients of O(1), can only be determined by an explicit string calculation. (Note also
that the physical values of the Yukawa couplings also depend on the normalization of the
kinetic energy terms for the corresponding matter field, which we will not address here
either.)
In particular, we shall explore the basic building blocks for the calculation of A, by first
positioning the branes very close to the symmetric positions in the six-torus. As the next
step we shall then explore the consequences for the Yukawa coupling hierarchy when the
branes are moved from the symmetric positions.
There are couplings between the A1 B1 , B1 C1 and C1 A1 sectors. Since the C1 A1
sector is the same as A1 C1 , and furthermore, A1 = A1 (because the A1 brane is the
same as its own orientifold image), in principle there are non-zero couplings of the form
(A1 B1 )(B1 C1 )(A1 C1 ), which could give rise to the Yukawa couplings of the two families
of quarks and leptons from the B1 C1 sector (see Table 3). The third family has no Yukawa
couplings, since the left-handed quarks and leptons in this family arise from the B1 C1
sector instead, and hence the three-point couplings are not gauge invariant.
The basic ingredients for calculating the intersection areas are given in Fig. 1.
This is an initial symmetric configuration of the A, B, C sectors of branes, associated
with the U (1)8,8 , U (2)L and {U (3)C , U (1)1 } sectors, respectively. The set of A branes,
associated with U (1)8 and U (1)8 , are positioned very close to the corresponding
orientifold plane. (Had they all been positioned exactly on top of the orientifold plane, the
gauge group would have been enhanced to U Sp(8).) Thus the couplings associated with
the pairs of states that are charged under U (1)8 and U (1)8 , respectively, are approximately
degenerate. We denote the two sets of Higgs fields with U (1)8 charges as H{U,D},{I,Iα
I}
where α = {1, 2, 3}. Here, α labels the intersection points of the A1 and B1 branes (where
the Higgs fields are located). The pairs of states denoted by {I, I I } indices correspond to
the two sets of fields appearing at the same intersections. Analogous notation is used for
the corresponding right-handed quark and lepton sector. The set of fields associated with
U (1)8 charges are denoted by H → H and {U , D}
→ {U , D }.
We have also positioned branes associated with U (3)C and U (1)1 nearby, which ensures
γ
at this stage the near degeneracy of the couplings associated with the quark doublets QL
γ
and leptons LL (γ = (i, ii)) as well as that of the Up- and Down-sector. Due to this large
degeneracy, we shall only describe the couplings for the Up-quark sector.
Fig. 1. The initial symmetric configuration of the A, B, C sectors of branes, associated with the U (1)8,8 ,
U (2)L and {U (3)C , U (1)1 } sectors, are denoted by dashed, dotted, and solid lines, respectively. The intersections
denoted by α = (1, 2, 3) and γ = (i, ii) correspond to the appearance of Higgs and left-handed families,
respectively.
From Fig. 1, which depicts the location of the intersections of the A, B, C branes,
it is evident that there are different Yukawa couplings associated with the location of the
intersections of the two types of left-handed quarks γ = (i, ii) and the location of the three
types of Higgs fields HUα {I,I I } where α = 1, 2, 3 (and H → H sectors).
While the Higgs fields (and the right-handed quarks) associated with index I and I I
formally appear at the same intersection, the orientifold and orbifold projection in the
construction of these states ensure that only pairs of the Higgs and right-handed quarks
with the same I or I I index couple to each other.
Thus, in this degenerate case, the Yukawa interactions take the form:
γ
HYukawa = hα,γ QL U I I HUα I I + U
I HU I + U
α , HU → U , HU , (3)
α,γ
where hα,γ ∼ exp(−Aα,γ /(2πα )).

The area of the triangle associated with the three intersection points (in the six-
dimensional internal space) can be calculated in terms of the products of the vectors a ,
b and c, specifying the respective locations of the three intersections points:
1

Area = a − b × a − c
2

1
2
2

2
= a − b a − c − a − b · a − c . (4)
2
After these preliminaries we are set to calculate the minimal intersection areas Aα,γ . These
can be easily determined from Fig. 1, which depicts the position of the building block
Fig. 2. Intersection areas of the branes in the third torus. The thin solid lines denote the lattice, the thick solid
lines U (3)C branes, the dotted lines U (2)L branes and the dashed ones U (1)8,8 branes. Again α = (1, 2, 3) and
γ = (i, ii) denote the location of the three Higgs fields and the two left-handed quark families, respectively.
branes in the fundamental domain of the toroidal lattice, and Fig. 2 which depicts the
relevant intersection areas for the third toroidal lattice.
Employing Eq. (4) we obtain the straightforward results for the intersection areas:
1
A1,i = 0, A2,i = A3,i = R1(3) R2(3) ,
3
1 (3) (3) 3 (3) (3)
A2,ii = A3,ii = R1 R2 , A1,ii = R1 R2 , (5)
12 4
(i)
where R1,2 refer to the two sizes of the ith torus, along the xi and yi axis, respectively.2
(i) (i)
Note also that R1 R2 corresponds to the area of the ith two-torus. Due to the symmetry
of the configuration there is no contribution from the area arising from the first two two-
tori. We can therefore encounter a sizable hierarchy among different Yukawa couplings. In
particular the sub-leading terms h1,ii for α = 1 are smaller than the couplings for α = 2, 3,
as can be seen in (5).
(i)
There are phenomenological constraints on the possible values of R1,2 . The Planck scale
and various Yang–Mills couplings are related to the string coupling gs by
2 Ms8 V6
MP(4d) = , (6)
(2π)7 gs2
2 This notation differs slightly from [31], in which R (i) represented radii, i.e., R (i) in this paper corresponds
1,2 1,2
i in [31].
to 2πR1,2
and
1 Ms3 V3
2
= , (7)
gYM (2π)4 gs
where V6 is the volume of the six-dimensional orbifold and V3 is the volume of the three-
cycle that a specific set of D6-brane wraps. (These volume factors have been explicitly
(i)
calculated in [31] in terms of the wrapping numbers (ni , mi ) and R1,2 .)
Using (6) and (7) one can eliminate gs and obtain the relationship between gYM , MP4d
and Ms :
√
(4d)
√ V6
2
gYM MP = 2π Ms . (8)
V3
(4d) √
For a fixed value of Ms /MP , the gYM depend only on the ratios V6 /V3 , which are
(i) (i)
functions of the complex structure moduli χi = R2 /R1 only, and have been explicitly
evaluated in [31].
Since each gauge group factor of the Standard Model arises from a separate set of branes
wrapping a specific three-cycle, there is no internal direction transverse to all the branes.
It therefore follows from (6)–(7) that the large Planck scale MP4d cannot be generated by
taking any of the internal directions much larger than the inverse of the string scale Ms ,
since for perturbative values of the string coupling gs that would make (at least one of)
the gauge couplings unrealistic. Thus a large Planck scale is generated from a large string
scale and not from a large volume, which is then also compatible with the gauge coupling
constraints (7), (8). (Note also that experimental bounds on the Kaluza–Klein modes of
the Standard Model gauge bosons imply that the extra dimensions cannot be larger than
O(TeV−1 ), but this is a much weaker bound than the one obtained by the arguments above.)
Finally, the R1,2 ’s cannot be much smaller than the string scale Ms−1 as this would again
(i)
make the Planck scale and gauge couplings unrealistic. One should however point out that
(i)
there still remains some flexibility in adjusting the sizes R1,2 ’s by an order of magnitude
or so away from O(Ms−1 ).
(i)
The above constraints that limit generic values of the sizes R1,2 ’s to be close to the
inverse of the string scale Ms−1 (and Planck scale MP4d close to Ms ) have implications for
the hierarchy of the Yukawa couplings. Had one had R1,2 Ms−1 the couplings would
(i)
have been exponentially suppressed. However, since R1,2 (i)

= O(Ms−1 ), the range dictated
from constraints on the Planck scale and gauge couplings, the hierarchy among Yukawa
couplings is non-degenerate and may potentially have interesting phenomenological
(i)
implications. For definiteness, we will require Ms R1,2 2π .
3.1. Lepton–quark splitting
The eight C1 -branes are split in sets of six and two, thus ensuring the breakdown of
U (4) (Pati–Salam type) symmetry down to U (3)C and U (1)1 . We chose to split them in the
first two-torus, keeping U (3)C along the Z2 symmetric position and moving U (1)1 branes
(1)
relative to U (3)C ones by a distance ηR1,2 away in the x- and y-direction, respectively.
Fig. 3. The brane configurations in the first torus, depicting the breaking on U (4) Pati–Salam symmetry down
to U (3)C × U (1)1 . The U (1)1 branes (denoted by dash-dotted line) are positioned in a Z2 symmetric way
(1)
relative to U (3)C branes (denoted by a solid line). The separation between them is ηR1,2 in the respective
x- and y-directions. The relevant interesection area in the first torus, contributing to the lepton Yukawa coupling
is denoted by a shaded area. (U (2)L and U (1)8,8 branes are denoted by a dotted and a dashed line, respectively.)
Table 4
The values of Aα,γ and Bα,γ for the respective U (1)8,8 and U (2)L sides of the intersection triangles in the
third torus for various α and γ
(α, γ ) Aα,γ
2 Bα,γ
2
(1, i) 0 0
4 (3) 2 1 (3) 2 1 (3) 2
(2, i), (3, i) R
3 1 R
3 1 + 2 R2
2 (3) 2 1 (3) 2 1 (3) 2
(2, ii), (3, ii) 3 R1 6 R1 + 4 R2
(3) 2 1 (3) 2 3 (3) 2
(1, ii) 2R1 R
2 1 + 4 R2
(See Fig. 3.) It now becomes a straightforward exercise to determine the new areas
associated with the lepton Yukawa couplings. The areas associated with the lepton Yukawa
couplings can be expressed in terms of the areas for the quark Yukawa couplings by

1 (1) 2 (1) 2 (1) 2 2 (1) 2
+ ηR1 Aα,γ + ηR2 Bα,γ
lepton 2 + 4Aquark 2 ,
Aα,γ = ηR1 ηR2 α,γ (9)
2
where Aα,γ and Bα,γ specify the vectors for the respective U (1)8,8 and U (2)L sides of
the triangles for the corresponding (α, γ ) intersections (in the third toroidal direction).
The areas (9) for lepton Yukawa couplings are always larger than those of the quark
couplings. This formula is valid as long as η is less or ∼1/2. The values of Aα,γ and Bα,γ
quark
are give in Table 4, while Aα,γ are listed in Eq. (5).
3.2. Up–Down Yukawa coupling splitting
The degeneracy of the Yukawa couplings that are associated with states charged under
U (1)8 and U (1)8 can be removed by splitting the branes associated with the first and
Fig. 4. The splitting of A-type branes (associated with U (1)8,8 and denoted by the dashed lines) from the
orientifold planes. For simplicity the figure shows only the fundamental domain of each of the three two-tori.
The solid and dotted lines denote the U (3)C and U (2)L branes, respectively.
second Abelian factors from the orientifold plane by a distance 4i and 4i in the ith torus
(i = 1, 2, 3) (see Fig. 4). This in turn provides a mechanism for Up–Down sector splitting.
One can show that the basic ingredients for determining the Up-type (Down-type)
Yukawa couplings is to study the intersection of the A-type branes (associated with
(i) (i)
U (1)8,8 ) moved by a distance +4i R1 [+4i R1 ] away from the (vertical) orientifold
(3) (3)
planes in the first two (i = (1, 2)) two-tori, and a distance +43 R2 [−43 R2 ] from the
(horizontal) orientifold plane in the third torus. Fig. 4 depicts these basic displacements
of the A-type branes in the fundamental domain of each of the two-tori for the Up- and
Down-sectors, respectively. In addition, Figs. 5 and 6 depict the new (α, γ ) intersection
areas in the third toroidal direction for the Up- and Down-sectors, respectively. One
can now explicitly calculate the new areas by essentially employing the magnitude
of vectors (Ai )2 = (4i R1(i) )2 and (Bi )2 = (4i R2(i) )2 associated with the sides of the
intersection triangles in the first two two-tori for U (1)8,8 and U (2)L branes, as well as
the corresponding vectors Aα,γ and Bα,γ in the respective U (1)8,8 and U (2)L sides of the
intersection triangles of the third two-torus. For the sake of simplicity we set 42 = 0, since
this significantly simplifies the analytic expression for the intersection area, although the
complete formula is straightforward to obtain. The intersection area is:

1 2 2 2
Aα,γ =
U/D
41 R1(1) 41 R2(1) + 41 R1(1) Aα,γ 2 + 4 R (1) 2 B
1 2 α,γ
2 + 4 ÃU/D 2 ,
α,γ
2
(10)
U/D
where Ãα,γ refers to the corresponding intersection area in the third toroidal plane. The
formula is valid as long as 41,3 are less or ∼1/2. For U (1)8 displacements the analogous
area formulae are valid with the replacement 4i → 4i .
Fig. 5. Relevant intersection areas in the third toroidal lattice for the Up-sector.
Due to the orbifold and the orientifold symmetries, it is evident from Figs. 5 and 6 that
a number of Up–Down Yukawa couplings remain degenerate. In particular the following
relations hold:
1,i = A1,i ,
AU D
3,ii = A2,ii > A2,ii = A3,ii ,

AU D U D
2,i = A3,i > A3,i = A2,i ,

AU D U D
AU D
1,ii < A1,ii . (11)
Except for the most suppressed Yukawa couplings between the α = 1 Higgs fields and
γ = ii quarks, the areas associated with the remaining Up- and Down-Yukawa couplings
pair-up.
The explicit values for the areas and the vectors Aα,γ and Bα,γ are given in Table 5 for
the Up-sector. In the Up-sector the area for (3, i) is obtained from (2, i) (and (2, ii) from
(3, ii)) by changing 43 → −43 . Similarly the Down-sector area for (1, ii) is obtained from
the Up-sector area for (1, ii) by changing 43 → −43 .
Fig. 6. Relevant intersection areas in the third toroidal lattice for the Down-sector.
Table 5
The values of A2α,γ , Bα,γ
2 and Ã
α,γ for various α and γ . The results reduce to those in Table 4 for 43 = 0
(α, γ ) Aα,γ
2 Bα,γ
2 Ãα,γ
8 (3) 2 2 (3) 2
(3) 2 4 2 (3) (3)
(1, i) 3 43 R1 3 43 R1 + 43 R2 3 43 R1 R2
4 (3) 2 1 (3) 2 1 (3) 2 1 (1 + 24 )2 R (3) R (3)
(2, i) 3 (1 + 243 )R1 3 (1 + 243 )R1 + 2 (1 + 243 )R2 3 3 1 2
2 (3) 2 1 (3) 2 1 (3) 2 1 (1 + 44 )2 R (3) R (3)
(3, ii) 3 (1 + 443 )R1 6 (1 + 443 )R1 + 4 (1 + 443 )R2 12 3 1 2
(3) 2 1
(3) 2 (3) 2 3 1 − 4 4 2 R (3) R (3)
(1, ii) 2 1 − 43 43 R1 4
2 1 − 3 43 R1 + 34 1 − 43 43 R2 4 3 3 1 2
4. Implications of the Yukawa coupling hierarchy
The basic results for the Yukawa couplings are given in Eqs. (3), (5), (9) and (10). It is
difficult to discuss the implications for the fermion masses without a detailed knowledge
of the Higgs vacuum expectation values (VEVs), which in turn depend on the details
of the soft supersymmetry breaking, the effective µ terms for the Higgs fields, and the
normalization factors of the kinetic energy terms, which have not been determined. As
was discussed in [31], the large number of Higgs doublets and the lack of a compelling
mechanism to generate effective µ terms, at least at the perturbative level, are significant
drawbacks of the construction. Also, the construction contains a strongly coupled quasi-
hidden section, which is a candidate for dynamical supersymmetry breaking, and the
detailed study of these phenomena is in progress [34].
Nevertheless, we can make a few general comments about the implications of the
Yukawa couplings, emphasizing the simplest case in which the D6-branes are positioned
very close to the symmetric positions in the six-torus, as in (5). In this case, there are only
four independent Yukawa couplings, h1,i , h2,ii = h3,ii , h2,i = h3,i and h1,ii . As discussed
in Section 3 there are theoretical uncertainties concerning the prefactors and numerical
factors in the exponents. For definiteness, we will assume that hα,γ ∼ exp(−Aα,γ /(2πα ))
is a good approximation at least for the ratios of Yukawa couplings. We will also
(3)
assume that Ms R1,2 2π . In that case, h1,i ∼ 1, h2,ii = h3,ii 0.59, h2,i = h3,i 0.12,
(3)
and h1,ii 0.009, with all but h1,i being extremely small for R1,2 much larger than
(3)
the minimum value of 2π/Ms . Intermediate values for the R1,2 will yield non-trivial
hierarchies for the Yukawas.
It is convenient to rewrite (3) as
4 √
2
HU K + HU3 K
HYukawa = QL i
h1,i HU K + 2 h2,i
1
√ U K
K=1
2
4 √
2
HU K + HU3 K
+ QL ii
h1,ii HU K + 2 h2,ii
1
√ K ,
U (12)
K=1
2
where the index K represents the four terms (I, I I , and the primed terms) in (3). When
some of the Higgs fields acquire VEVs this will yield a 2 × 4 mass matrix for the two U
quarks and four antiquarks. However, in the special case that the two rows are proportional
(i.e., that they are aligned in the K direction), there will only be a single non-zero mass
eigenvalue. Let us first consider the case of large sizes, so that all of the couplings are small
except h1,i . Then, there will only be one significant mass term, corresponding to QiL and a
linear combination of the U K , with coefficients depending on the VEVs of the H 1 . The
UK
other mass eigenvalue will be exponentially small. In the special case of radiative symmetry
breaking, usually associated with supergravity mediated supersymmetry breaking but also
occurring for gauge mediation, the second mass would be exactly zero. That is because
only the HU1 K ’s have the large Yukawa couplings needed to drive their (presumably
positive) mass-squares at the string scale to negative values at low energies, and the VEVs
of the other Higgs doublets would vanish. (The small h1,ii would lead to a tiny mixing
between Q1L and Q2L , but not generate a second non-zero mass because the two terms
(3) √
would be aligned in K.) On the other hand, for small R1,2 both h1,i and 2 h2,ii (the
√
Yukawa coupling for the relevant state (HU2 K + HU3 K )/ 2 ), could be significant, leading
to two non-zero mass eigenstates provided that the terms are not aligned in K. For radiative
breaking, the two large Yukawas could drive both relevant mass-squares negative, and
alignment would not be expected except for very specific values for the mass-squares at
the string scale and the effective µ parameters and kinetic terms. In this case, the hierarchy
mt mc , mu = 0 could be achieved by a hierarchy in the VEVs of the HU1 K ’s relative to
√
the (HU2 K + HU3 K )/ 2 ), which could be achieved by modest differences in the relevant
soft supersymmetry breaking and other terms.
Thus, it is possible to achieve a hierarchy of Up-mass eigenvalues, associated with the
hierarchy of Yukawa couplings or of VEVs or both. As discussed after (11), even after
moving the branes from their symmetric positions the Up and Down Yukawas are the
same up to relabelling except for the smallest coupling h1,ii . Thus, the hierarchy mt mb
would have to come about because the HU1 K VEVs are much larger than those for the
1 , analogous to the large tan β region of the MSSM. This can easily occur for moderate
HDK
differences in the soft mass-squares, especially if the effective A parameters are small. The
full hierarchy of mt , mb , mc and ms (with md = mu = 0) could most likely be achieved for
appropriate soft and effective µ parameters and kinetic energy terms, but we do not pursue
this in detail since these have not been calculated. Similarly, non-trivial quark mixing could
be generated by different K dependence of the VEVs in the HU and HD sectors.
In the symmetric case the charged lepton Yukawas are the same as for the Up- and
Down-quarks, and the charged leptons couple to the same Higgs doublets as the Down-
quarks. This is analogous to the t–b–τ Yukawa universality of the simplest version of
SO(10) grand unification, which is successful for large tan β. (In addition to the b–τ
Yukawa relation at the string scale, one also has s–µ unification. Of course, the quark
Yukawas are enhanced by QCD and other effects in the running down from the string
scale, leading to a successful mb /mτ prediction, but a rather large value for ms /mµ .) The
corrections to the lepton Yukawas from moving the U (1)1 -branes in (9) decrease the lepton
Yukawas relative to the symmetric quark couplings, increasing the mb /mτ prediction. Such
a shift is acceptable as long as is small. The U (1)8,8 shifts in (10) have the same effects
on the leptons as the quarks.
One expects Dirac neutrino masses comparable to the quark and charged lepton masses
close to the symmetric points. The possibility of a neutrino seesaw was commented
on in [31]. In particular, Majorana masses for the right-handed neutrinos N cannot be
significantly larger than the scales at which the two additional U (1) factors of the model
are broken. It was shown that when the charge confinement and anomaly conditions
associated with the strongly coupled quasi-hidden sector are taken into account, then there
would be scalar fields with the appropriate quantum numbers to break both U (1) s at a
high scale at which the interactions become strongly coupled. This could be 1015 GeV or
higher, which could lead to acceptable seesaw mass scales for the neutrinos. However, the
actual potential for those fields and their couplings to the N states (needed to estimate the
actual masses and mixings) would be non-perturbative effects, beyond the scope of this
investigation.
5. Conclusions
We have considered the Yukawa couplings in a supersymmetric three family standard-

like string model. In particular, we have calculated the leading order contributions to
the world-sheet instantons associated with the action of the string world-sheet stretching
among the intersection points corresponding to the chiral matter fields. We considered both
the case in which the branes are located very close to symmetric positions in the six-torus,
which leads to a high degeneracy of Yukawa couplings, and the consequences of moving
some of the branes away from the symmetric positions. In general there is a large hierarchy
of Yukawa couplings, which increases exponentially as the sizes of the tori are increased.
The actual fermion masses depend on the vacuum expectation values of the Higgs fields,
which in turn depend on the supersymmetry breaking and on the effective µ parameters.
There are typically either two or one massive generations of fermions.
Acknowledgements
We are especially grateful to Angel Uranga for many discussions and collaboration
at an early stage. We would also like to thank Jing Wang for useful discussions and
collaborations on related work. This work was supported by the DOE grants EY-76-02-
3071 and DE-FG02-95ER40896; by the National Science Foundation Grant No. PHY99-
07949; by the University of Pennsylvania School of Arts and Sciences Dean’s fund (M.C.
and G.S.) and Class of 1965 Endowed Term Chair (M.C.); by the University of Wisconsin
at Madison (P.L.); by the W.M. Keck Foundation as a Keck Distinguished Visiting
Professor at the Institute for Advanced Study (P.L.). We would also like to thank ITP,
Santa Barbara, during the “Brane World” workshop (M.C., P.L. and G.S.), and the Isaac
Newton Institute for Mathematical Sciences, Cambridge, during the M-theory workshop
(M.C.), for hospitality during the course of the work.
References
[1] For reviews, see, e.g., B.R. Greene, in: Lectures at Trieste Summer School on High Energy Physics and
Cosmology, 1990;
F. Quevedo, hep-th/9603074;
A.E. Faraggi, hep-ph/9707311;
Z. Kakushadze, G. Shiu, S.H. Tye, Y. Vtorov-Karevsky, Int. J. Mod. Phys. A 13 (1998) 2551;
G. Cleaver, M. Cvetič, J.R. Espinosa, L.L. Everett, P. Langacker, J. Wang, Phys. Rev. D 59 (1999) 055005,
and references therein.
[2] L.J. Dixon, V. Kaplunovsky, C. Vafa, Nucl. Phys. B 294 (1987) 43;
For generalization of this theorem to less than 4 dimensions, see, J. Erler, G. Shiu, Phys. Lett. B 521 (2001)
114.
[3] C. Angelantonj, M. Bianchi, G. Pradisi, A. Sagnotti, Ya.S. Stanev, Phys. Lett. B 385 (1996) 96.
[4] M. Berkooz, R.G. Leigh, Nucl. Phys. B 483 (1997) 187.
[5] Z. Kakushadze, G. Shiu, Phys. Rev. D 56 (1997) 3686;
Z. Kakushadze, G. Shiu, Nucl. Phys. B 520 (1998) 75;
Z. Kakushadze, Nucl. Phys. B 512 (1998) 221.
[6] G. Zwart, Nucl. Phys. B 526 (1998) 378;
D. O’Driscoll, hep-th/9801114.
[7] G. Shiu, S.H. Tye, Phys. Rev. D 58 (1998) 106007.
[8] Z. Kakushadze, G. Shiu, S.H. Tye, Nucl. Phys. B 533 (1998) 25;
Z. Kakushadze, G. Shiu, S.H. Tye, Phys. Rev. D 58 (1998) 086001.
[9] G. Aldazabal, A. Font, L.E. Ibáñez, G. Violero, Nucl. Phys. B 536 (1999) 29.
[10] Z. Kakushadze, Phys. Lett. B 434 (1998) 269;
Z. Kakushadze, Phys. Rev. D 58 (1998) 101901;
Z. Kakushadze, Nucl. Phys. B 535 (1998) 311.
[11] M. Cvetič, M. Plümacher, J. Wang, JHEP 0004 (2000) 004.

[12] M. Klein, R. Rabadán, JHEP 0010 (2000) 049.
[13] G. Aldazabal, L.E. Ibáñez, F. Quevedo, A.M. Uranga, JHEP 0008 (2000) 002.
[14] M. Cvetič, A.M. Uranga, J. Wang, Nucl. Phys. B 595 (2001) 63.
[15] M. Berkooz, M.R. Douglas, R.G. Leigh, Nucl. Phys. B 480 (1996) 265.
[16] R. Blumenhagen, L. Görlich, B. Körs, D. Lüst, JHEP 0010 (2000) 006.
[17] G. Aldazabal, S. Franco, L.E. Ibáñez, R. Rabadán, A.M. Uranga, J. Mod. Phys. 42 (2001) 3103;
G. Aldazabal, S. Franco, L.E. Ibáñez, R. Rabadán, A.M. Uranga, JHEP 0102 (2001) 047.
[18] R. Blumenhagen, B. Körs, D. Lüst, JHEP 0102 (2001) 030.
[19] C. Angelantonj, I. Antoniadis, E. Dudas, A. Sagnotti, Phys. Lett. B 489 (2000) 223.
[20] L.E. Ibáñez, F. Marchesano, R. Rabadán, JHEP 0111 (2001) 002;
For generalization to models with an extra U (1) at the string scale, see, C. Kokorelis, hep-th/0205147.
[21] S. Förste, G. Honecker, R. Schreyer, Nucl. Phys. B 593 (2001) 127;
S. Förste, G. Honecker, R. Schreyer, JHEP 0106 (2001) 004.
[22] R. Blumenhagen, B. Körs, D. Lüst, T. Ott, Nucl. Phys. B 616 (2001) 3.
[23] D. Bailin, G.V. Kraniotis, A. Love, Phys. Lett. B 530 (2002) 202.
[24] C. Kokorelis, hep-th/0207234.
[25] M. Cvetič, G. Shiu, A.M. Uranga, Phys. Rev. Lett. 87 (2001) 201801.
[26] M. Cvetič, G. Shiu, A.M. Uranga, Nucl. Phys. B 615 (2001) 3.
[27] M. Cvetič, G. Shiu, A.M. Uranga, hep-th/0111179.
[28] M. Atiyah, E. Witten, hep-th/0107177.
[29] E. Witten, hep-th/0108165.
[30] B. Acharya, E. Witten, hep-th/0109152.
[31] M. Cvetič, P. Langacker, G. Shiu, UPR-990-T, hep-ph/0205252.
[32] C. Kokorelis, hep-th/0203187.
[33] L.J. Dixon, D. Friedan, E.J. Martinec, S.H. Shenker, Nucl. Phys. B 282 (1987) 13;
S. Hamidi, C. Vafa, Nucl. Phys. B 279 (1987) 465.
[34] M. Cvetič, P. Langacker, J. Wang, in preparation.
Muon decay to one loop order in the left–right

symmetric model
M. Czakon a , J. Gluza b , J. Hejczyk b
a Institut für Theoretische Physik, Universität Karlsruhe, D-76128 Karlsruhe, Germany
b Department of Field Theory and Particle Physics, Institute of Physics, University of Silesia, Uniwersytecka 4,
PL-40-007 Katowice, Poland
Received 28 May 2002; accepted 31 July 2002
Abstract
One loop corrections to the muon decay are studied in a popular and self-consistent version of
the left–right symmetric model. It is shown quantitatively, that the corrections do not split into those
that come from the Standard Model sector, and some decoupling terms. For a heavy Spontaneous
Symmetry Breaking (SSB) scale of the order of at least 1 TeV, the contributions from the top quark
have a logarithmic behaviour and there is a strong quadratic dependence on the heavy Higgs scalar
masses. The dependence on the light Higgs boson mass is small. The heavy neutrinos are shown to
play an important role, although secondary in comparison with the heavy scalar particles as long as
the heavy neutrinos’ Majorana Yukawa coupling matrix hM obeys unitarity bounds.
1. Introduction
Embedding the Standard Model (SM) into a larger gauge group increases the number
of degrees of freedom. For the Left–Right Symmetric Models (LRSM) based on the
SU(2)L ⊗ SU(2)R ⊗ U (1)B−L gauge group [1,2] these are connected with new fields and
interactions. The model is complex with extra particles of different types. New neutral
leptons, charged and neutral gauge bosons, neutral and charged Higgs scalars appear.
There are many different versions of the LR models with equal or different left and right
gauge couplings gL,R , and specific Higgs sector representations. This robust structure is
a challenge and a good theoretical laboratory for testing many phenomena beyond the SM.
E-mail address: czakon@particle.uni-karlsruhe.de (M. Czakon).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 3 8 - 7
158 M. Czakon et al. / Nuclear Physics B 642 (2002) 157–172
The purpose of the present work is to study numerically one loop corrections to muon
decay which come from the extended gauge sector of the LRSM. Since the history of the
LRSM is already quite long, there have been some interesting attempts to study radiative
corrections within its framework [3–5]. To our knowledge however, there has never been
a complete calculation performed. We start our systematic one loop level study of the
model from a low-energy muon decay calculation. The subject has already been explored
qualitatively in [6]. The main result of this paper was to show that the quadratic top mass
dependence of the oblique corrections to r is lost. In the SM these corrections come from
constraints imposed by the SSB sector on the Weinberg angle counter-term. Here similar
constraints connect the counter-term with the heavy SSB scale. As a result the top quark is
effectively massless. By the same the SM one loop corrections do not constitute a subset
of the full contributions. Therefore in general, it is not true that one can properly fit New
Physics Models (NPM) by taking one loop SM corrections modified with tree level NPM
couplings. These issues are further explored in [7].
The question which we wish to answer is the following: can we or can we not
accommodate the present experimental life-time of the muon within a model that has
a minimal Higgs sector structure supported by phenomenology and the smallest possible
number of unknown free parameters? It is common wisdom that when there are many
free parameters any data can be fitted. We show however, that this induces a strong
correlation between the heavy parameters. In fact a full decoupling is not observed, and
if the additional masses tend to infinity independently, a huge correction results, which is
incompatible with data.
We first discuss assumptions on model’s structure and parameters. The renormalization
scheme is then introduced and the corrections to muon decay enumerated. Numerical esti-
mates follow with a study of the dependence on the heavy masses and the heavy symmetry
breaking scale. Conclusions together with an outlook close the paper. Appendix A gathers
our notational conventions and main components of the model.
2. Structure and parameters of the model
As noticed in the Introduction, there are many versions of LRSM. A complete analysis
at the one loop level requires the model to be fixed. We choose to explore the most popular
version of the model with a Higgs representation with a bidoublet Φ and two (left and right)
triplets ∆L,R [8]. We also assume that the VEV of the left-handed triplet ∆L vanishes,
∆L = 0 and the CP symmetry can be violated by complex phases in the quark and
lepton mixing matrices. Left and right gauge couplings are chosen to be equal, gL = gR .
We call this model, the Minimal Left–Right Symmetric Model (MLRSM). The necessary
definitions can be found in Appendix A (for details, see [6,8–10]).
We also take advantage of several approximations which come from phenomenological
studies.
1. Mixing of fermions. As usual in one loop analyses, we neglect quark mixings.

The case of neutrino mixings deserves however additional comments. The effective
light neutrino mass matrix M = −MD MR−1 MD T (Eq. (A.22)) yields three light Majorana
M. Czakon et al. / Nuclear Physics B 642 (2002) 157–172 159
neutrinos which are predominantly composed of the usual “active” neutrinos νL with a
very small O(1/mN ) admixture of “sterile” neutrinos νLc = νR . The diagonalization of
MR (Eq. (A.24)) produces 3 heavy Majorana neutrinos which are mainly composed of νR .
In order to get light neutrino masses at the eV scale (as concluded from experimental
data [11,12]) and without extra relations between MD and MR matrices, mN must be
large mN > 109 (1013) GeV for the lepton (quark) see-saw mechanism. However, we
would also like to explore lower scales with vR of the order of TeV. A crude light–
heavy (LH) mixing estimate O(1/mN ) would give in this case larger couplings. However,
they would lead to a problem with obtaining the light neutrino spectrum, namely, from
Eq. (A.22) their masses would be much above the eV scale. A fine-tuning of MD and/or
MR parameters or additional discrete symmetries must be applied for the full neutrino
mass matrix to get the proper light neutrino spectrum. Therefore it has been argued in [13]
that it is not natural to obtain large LH mixings for heavy neutrinos at the TeV scale. In
accordance with these arguments we assume here that the light and heavy neutrino sectors
are disconnected (negligible mixings). In this way, W1 couples only to light neutrinos,
while W2 couples to the heavy ones. Z1 and Z2 turn out to couple to both of them [9,10].
This generates automatically an extended flavour symmetry, where transitions outside of a
family composed from a lepton, a light and a heavy neutrino are forbidden.
As lower limit on the heavy neutrino masses we use the direct experimental limit from the
lack of Z → νN decay, which is mN MZ .
2. Mixing of charged gauge bosons. In principle the model allows for mixing of charged
gauge bosons. However, experimental data analyses give the following conservative upper
bound on the mixing angle [14,15]
|ξ | 0.013 rad. (1)

The tree level contribution to r coming from the mixing is proportional to
2
MW
sin2 ξ 2
1
. (2)
MW 2
Even if the second charged gauge boson had a mass of the order of the SM W , this number
would be negligible compared to the experimental value which is of the order of 3%.
We therefore put ξ = 0, which also means that κ2 = 0. There are several advantages of
this approximation. First there is no need to renormalize the mixing of the gauge bosons.
It turns also out, that together with the previous approximation on lepton sector mixings,
there is no need to renormalize the a priori possible mixing between the light and the
heavy neutrino within a family. At last, the QED contributions to the process form a self-
contained class as in the SM (see Section 4).
This model has the nice feature that all the constraints on the right handed sector come
uniquely from one loop corrections. The tree level W1 exchange diagram is not sensitive
to the additional gauge structure anymore [16].
3. Yukawa couplings to charged Higgs scalars. The approximations from the preceding
two points, leave still a possibility for muon decay through one of the charged Higgs scalars
H1+ . It turns out, that the experimental data on polarized muon decay asymmetries are
compatible even with a decay through scalar currents only. However, inverse muon decay
bounds these contributions to be at most one order of magnitude smaller than the SM left-
handed current decay. We assume in this work that these diagrams are either negligible or
require only to be included at the tree level in which case, the space left for r in Eq. (18)
would be respectively smaller.
3. Renormalization in one loop order
As the basic set of input parameters we choose the electromagnetic coupling constant
and the masses of the four gauge bosons, Higgs scalars and fermions. It turns out that as
long as no corrections need to be included to tree level Higgs scalar exchange diagrams, the
on-shell conditions of gauge bosons only suffice to fix all of the necessary counter-terms.
Moreover, decoupling effects should be automatically included.
In the present approximation, where we neglect charged gauge boson mixing, only the
Weinberg angle requires renormalization. We recall here its counter-term [6]
(δMZ2 2 + δMZ2 1 ) − (δMW
2 + δM 2 )
W1
2
δsW = 2cW
2 2
(MZ2 2 + MZ2 1 ) − (MW
2 + M2 )
2 W1
1 (MW2 + MW1 )(δMZ2 + δMZ1 ) + (MZ2 + MZ1 )(δMW2 + δMW1 )

2 2 2 2 2 2 2 2
+
2 ((MZ2 + MZ2 ) − (MW
2
2 + M 2 ))2
1 W 2 1
1 (2MZ1 + MZ2 )δMZ1 + (2MZ2 + MZ1 )δMZ2

2 2 2 2 2 2
− . (3)
2 ((MZ2 2 + MZ2 1 ) − (MW
2 + M 2 ))2
2 W1
As discussed in Section 2, no fermion mixing renormalization is needed, and the
hard corrections (factorized weak contributions) are properly included through the simple
fermion counter-terms
l,ν
δZL,R = Σγl,νL,R , (4)
where we used the following decomposition of fermion self-energy
Σ = p̂PL Σγ L + p̂PR Σγ R + PL ΣL + PR ΣR . (5)

An interesting problem is connected to charge universality and renormalization of the
electromagnetic coupling. We wish here to show that charge universality follows simply
from Ward identities and the constructive proof gives also the correct counter-term. To
one loop the potentially problematic contributions come from diagrams involving a heavy
neutrino and the traditional approach gives a result independent of these masses only after
summation of vertex and external line contributions [6].
Let us start from the following relation, which comes from charge assignments within
a fermion doublet and the definition of the physical fields
 0 0 µ 
  lL lL + νL0 νL0 Z1 amp
0
 
 0  = U0T −1  l 0 l 0 + ν 0 ν 0 Z µ 
0 0  LL L L 2 amp  , (6)
lL lL + νL0 νL0 B 0µ amp 0 0 µ
lL lL A amp
where U0 is the bare neutral sector mixing matrix Eq. (A.12) multiplied by the
renormalization constants of the physical fields
 1 1 1 
 0 c0 0 s0 0  Z2 ZZ2 1 Z2 ZZ2 1 γ 
cW cW sW  Z1 Z1
  
U0 =  0 0 
 1 1 1 
 −sW

sM c − cM
0 0 0 0 0
s −sW sM s + cM
0 0 0 0 0
c cW sM 
 ZZ2 Z ZZ2 2 Z2 ZZ2 2 γ 

 2 1 
−sW
0 c0 c0 + s 0 s 0 −sW
0 c0 s 0 − s 0 c0 0 c0 
cW 1 1 1

M M M M M
Zγ2 Z1 2
Z γ Z2 2
Zγ γ
(7)
and · · ·amp is a shorthand for amputated Green functions. From this we obtain
0 0 µ
lL lL A amp = U0T 33 lL0 lL0 + νL0 νL0 B 0µ amp . (8)
After taking the divergence of the current, we can use the U (1) Ward identity for the
B field and the on-shell renormalization conditions on the fermion propagators and the
electromagnetic vertex, which leads to the following identity
e0 T
e= U0 33 , (9)
cos 2ΘW0
which can be put into the following form

e0 1
0
sin φ 0 tan ΘW − cos φ 0 cos 2ΘW0 0
tan ΘW 1
Zγ2 γ + ZZ2 1 γ
e 0
cos 2ΘW

cos φ 0 tan ΘW 0 + sin φ 0 cos 2Θ 0 tan Θ 0
W W 12
− ZZ2 γ = 1. (10)
cos 2ΘW0
Since none of the above renormalization constants depends on the initial fermion species,
we have obtained the required charge universality. At the same time we can expand this
relation to first order to yield the electromagnetic coupling renormalization counter-term
√
δe 1 sin φ tan ΘW − cos φ cos 2ΘW tan ΘW 12
= − δZγ γ − √ Z Z1 γ
e 2 cos 2ΘW
√
cos φ tan ΘW + sin φ cos 2ΘW tan ΘW 12
+ √ Z Z2 γ . (11)
cos 2ΘW
We have checked by explicit calculation that the above formula gives the same value as
the usual approach. We would like to stress that to our knowledge, such a formula for LR
models has never been derived, although similar methods have been used in SM analyses
[17].
4. Structure of corrections to muon decay
The muon life-time is parametrized through the Fermi coupling constant, the mass of
the muon and the QED corrections to the four-fermion interaction q, which are presently
known up to second order in the fine structure constant [18]
1 G2 m5µ
= F 3 (1 + q). (12)
τµ 192π
The Fermi constant on the other hand is related to the tree level SM coupling of the charged
W boson to fermions through
GF e2
√ = 2 s2
(1 + r), (13)
2 8MW W
where r are higher order corrections to which we already made reference.
There are two problems with these formulas, when moving from the SM to other
interactions. Let us first consider Eq. (12). It is based on the assumption, that the basic
process is described by the four-fermion interaction, which in the charge conserving form
is of a pure V − A type. In fact, as long as the interaction has only an admixture of vector
and axial currents, the QED corrections are finite and gauge invariant, hence meaningful.
Notice however, that if the process is induced also by right-handed currents, then after
moving to the charge conserving form of the interaction (Fierz transformation), there
appear also scalar and tensor interactions, which are known not to have a finite QED
correction. The same problem occurs if we add charged scalar particles to the list. There
are two possibilities to remedy the situation, the first being of course calculating by any
means the process in the full model and resign from the separation of QED corrections.
The second possibility is somewhat simpler. If the tree level corrections from right-handed
and/or scalar interactions are of the same size as the one loop corrections to the basic
diagrams, we can simply ignore one loop contributions to these additional currents and
consider Eq. (12) as approximate and valid to one loop order only.
The second problem we have to face is the fact that the tree level coupling to the light
charged gauge boson can be different from the SM one. This concerns mainly the sine of
the Weinberg angle sW . In fact this happens to be the case of the considered model, where
due to constraints if we fix the mass of the two light gauge bosons, then sW is given by
a function of the heavy SSB scale vR . This dependence is depicted in Fig. 1. For small
values of vR , the difference from the SM value is large. We choose here to include the
change of sW from the SM to the LRSM in r.
r can now be obtained from the formula

(s 2 )SM −ΠW T (0) − δM 2
δe δsW 2
r = 2 W 2
W
+ 2 − 2
+ δ V + δ B
(sW )LRSM MW e sW
LRSM − (sW )SM
2 )
(sW 2
− 2 )
, (14)
(sW LRSM
where δV denotes the vertex corrections, which consist of the proper one loop vertex
diagrams and the incomplete counter-term made only of the fermion wave function
renormalization constants
√
2 sW 1 µ ν
δV = (Λeνe W + Λµνµ W ) + δZLe + δZLνe + δZL + δZLµ , (15)
e 2
Fig. 1. sin2 ΘW as function of vR with MW1 and MZ1 as in Eq. (17).
with Λ being the coefficient in front of the operator γ µ PL , and δB represents the box
contributions. The last term comes of course from the “renormalization” of the Weinberg
2
MW
SM = 1 −
2 )
angle between the two models with (sW 1 2 )
and (sW
MZ2 LRSM as obtained by
1
solving Eqs. (A.9), (A.10).
The strong dependence on the light fermion masses in δe is avoided as usual by a shift
up to the Z1 mass, and insertion of the running of the fine structure constant, for which we
take [20]
α(MZ1 ) = 0.059394 ± 0.000395. (16)

The factorization of the QED corrections is obtained with the Sirlin’s method [19], which
amounts to rejecting the infrared divergent box diagram and replacing the photon vanishing
mass by the W1 mass in the infrared divergent lepton wave function renormalization
constants.
5. Quantitative results
The evaluation of one loop corrections within the LRSM is a task of moderate size as
far as the number of diagrams is concerned. In fact approximately 600 had to be calculated
already after our simplifying assumptions. It would not be possible to perform this work
without using an automated system. For the generation of diagrams we used the C++
library DiaGen [21], which currently contains a topology generator, with several tools
to analyse the properties of the created objects, and a diagram generator with support for
Fig. 2. Contribution of box diagrams to r. With increasing vR heavy particles decouple and the lines aim at
the SM contribution. The (a) line is for (three heavy neutrinos) mN = 100 GeV; (b) is for mN = 500 GeV; (c) is
for mN = 2 TeV. Higgs particle masses obey Eqs. (A.3), (A.4). The gray area shows the experimentally allowed
values of r.
Majorana fields. The output has then been algebraically simplified with FORM [22], and at
last numerically evaluated with the help of the FF [23] based package LoopTools [24].
As discussed in the previous section we parametrized the muon lifetime corrections
coming from the LRSM through r, which is defined analogously as in the SM. With the
present values of the coupling constants and masses [15]
GF = 1.16639(1) × 10−5 GeV−2 , 1/α = 137.0359976 ± 0.00000050,
MW1 = 80.451 ± 0.033 GeV, MZ1 = 91.1875 ± 0.0021 GeV, (17)
the value of r with error is
r = 0.032 ± 0.004. (18)
We depicted this experimentally allowed range by a shaded region on the relevant figures.
As noted already in a previous work [6], we should not expect decoupling in the sense
that for large vR and large masses of the additional particles the SM result for r would be
obtained. Some type of decoupling is however observable. For example if we take the box
diagrams in the ’t Hooft–Feynman gauge, then the result tends to the SM one as depicted
in Fig. 2. It is worth noting however the effect of taking heavy neutrinos with a low vR
as for the (c) curve, where the contribution blows up. This is simply a consequence of
the fact, that the ratio of a neutrino mass and the heavy SSB scale is proportional to the
Yukawa coupling hM (Eq. (A.17)) and the respective diagrams are proportional to at least
the square of these couplings. Obviously, if the Yukawa couplings start to be larger than
one then the perturbative expansion must break down.
Fig. 3. r as function of vR for different heavy neutrino masses. Higgs masses are chosen according to Eqs. (A.3),
(A.4). The (a) line is for (three heavy neutrinos) mN = 100 GeV; (b) is for mN = 500 GeV; (c) is for mN = 2 TeV.
Line (d) shows the results when heavy neutrino masses follow from hM = 1 (see Eq. (19)). The gray area shows
the experimentally allowed values of r.
An interesting effect is obtained, if we take the masses of the Higgses to follow some
simple pattern as in Appendix A Eqs. (A.3) and (A.4). The respective r is shown for
several heavy neutrino masses in Fig. 3. With growing vR the value grows strongly away
from the allowed range and these parameters must be rejected. Although heavy neutrinos
lower down r, we cannot obtain a reasonable value even if their masses are at the edge of
the perturbatively range. The line (d) realizes this situation with the largest possible heavy
neutrino mass as a function of vR (Eq. (A.17))
√
mN = 2 vR . (19)
If we now assume for simplicity that all of the Higgs scalar masses are equal, apart
from the SM Higgs boson, then we obtain the strong dependence as depicted in Fig. 4.
If all the scalars are approximately two times heavier than vR (for large Higgs masses),
the experimental value for the muon decay life-time can be accommodated. Let us note at
this point that large Higgs masses, at least of the order of a few TeV are needed because
of FCNC [4]. It is obvious from Fig. 4 that Higgs scalars, heavy neutrinos and additional
gauge boson masses are very much fine-tuned to be within the SM gray area, e.g., the line
(a ) with mH = 1 TeV and mN = 100 GeV gives vR 800 GeV, which fixes MW2 and
MZ2 to the values as in Fig. 5.
It is interesting, that for a larger SSB breaking scale, the variation of r with the SM
Higgs scalar mass is negligible. This is shown in Fig. 6 for vR = 2390 GeV and neutrino
and heavy scalar masses chosen to fit the experimental value.
Fig. 4. r as function of vR . Sets with and without primes show results for three heavy neutrino masses with
mN = 100 GeV and mN = 2 TeV, respectively. The lines describe different values of Higgs scalar masses: (a) is
for all Higgs masses MH = 1 TeV; (b) is for MH = 5 TeV; (c) is for MH = 10 TeV. The gray area shows the
experimentally allowed values of r.
Fig. 5. Masses of additional gauge bosons and two sets of Higgs scalar particles Eqs. (A.3), (A.4) as function of
the vR scale.
Fig. 6. r as function of the lightest Higgs scalar mass MH 0 . The gray area shows the experimentally
0
allowed values of r and the heavy particle spectrum is chosen to fit approximately to this region, namely,
vR = 2390 GeV, mN = 2 TeV, mH = 5 TeV.
Fig. 7. The contribution of the third quark family to r as function of vR for different top quark masses.
At last let us comment on the dependence on the top quark mass. We show the
contributions of the third quark family in Fig. 7. The values of the top mass spread over
a vary large range to show the behaviour of the correction. As already foreseen in [6], for
low values of vR the variation is described by a negative quadratic function. However for
a vR as low as 1 TeV, only a positive logarithmic contribution is visible. Notice also, that
even in the low vR range, the top mass squared enters with a smaller coefficient than in the
SM.
6. Conclusions
In this paper we have studied the full one loop corrections to the muon decay in a self
consistent left–right symmetric model. We have shown quantitatively that the contributions
have a different structure from the SM ones and that they cannot be separated into these
and some corrections that would vanish with vR . Moreover, we have shown that the muon
decay alone already puts some stringent restrictions on the different heavy particle masses
with respect to the heavy SSB breaking scale.
Our analysis should be extended to cover also other low-energy experiments [14]. This
should elucidate the question of the contribution of the H1+ boson to the muon decay. It
will then also be possible to derive bounds on the extra boson masses.
At last let us note that several of the assumptions that we here took could be raised,
but it is doubtful that this would change qualitatively the numerical results. On the other
hand it would certainly make the analysis more involved, starting from the necessity to
renormalize the ξ angle, LH neutrino mixing and ending with problems with the QED
contributions.
Acknowledgement
M.C. would like to thank the Alexander von Humboldt foundation for fellowship. This
work was partly supported by the Polish Committee for Scientific Research under Grants
No. 2P03B04919 and 2P03B05418.
Appendix A
In this appendix we gather our definitions and give a short account of the particle content
of the model.
A.1. Higgs sector
The Higgs sector contains one bidoublet and two triplets

0 + √ ++
φ1 φ1+ δL,R / 2 δL,R
Φ= , ∆L,R = √ , (A.1)
φ2− φ20 0
δL,R +
−δL,R / 2
with the allowed Vacuum Expectation Values (VEV)

1 κ1 0 1 0 0
Φ = √ , ∆L,R = √ , (A.2)
2 0 κ2 2 vL,R 0
of which κ2 and vL are assumed to vanish.

The full potential has been studied in [8–10]. Here we only recall the physical spectrum
of the particles, which consists of
(i) four neutral scalars with J P C = 0++ (Hi0 , i = 0, 1, 2, 3);

(ii) two neutral pseudoscalars with J P C = 0+− (A0i , i = 1, 2);
(iii) two singly charged bosons (Hi± , i = 1, 2), and
(iv) two doubly charged Higgs particles (δL±± , δR±± ).
If vR κ1 and all the parameters of the potential which enter the Higgs masses are
taken to be 1 (these are combinations of µ1,2 , λ1,...,6 , ρ1,...,4 defined in [8]), then neglecting
terms proportional to the VEV of the SM, the masses satisfy the relations
√
MHa ≡ MH 0 = MH 0 = MA0 = MA0 = MH + = MH + = Mδ ++ = vR / 2, (A.3)
√
1 3 1 2 1 2 L
MHb ≡ MH 0 = Mδ ++ = 2 vR , (A.4)
√ 2
R
MH 0 = 2 κ1 . (A.5)
0
A.2. Gauge boson sector
Gauge boson masses are generated by the following mass terms (with the assumption
of equal couplings for the two SU(2) groups)
−
+µ +µ 2 WLµ
LM = WL , WR MCharged − + h.c.
WRµ

1 µ 2 W3Lµ
µ
+ W3L , W3R , B MNeutral W3Rµ ,
µ
(A.6)
2 Bµ
with

g2 κ+2 −2κ1 κ2
2
MCharged = , (A.7)
4 −2κ1 κ2 κ+ 2 + 2v 2
R
and
 g2 2 2 
2 κ+ − g2 κ+
2 0
1  g2 2 g2 2 
2
MNeutral =  − κ+ −2gg vR2  ,
2 κ+ + 4vR
2 (A.8)
2 2
0 −2gg vR2 2g 2 vR2

where κ+ = κ12 + κ22 . The masses of the physical gauge bosons are then given by

g2 2
2
MW 1,2
= κ+ + vR2 ∓ vR4 + 4κ12 κ22 , (A.9)
4
1 2 2
MZ2 1,2 = g κ+ + 2vR2 g 2 + g 2
4

∓ g 2 κ+ 2 + 2v 2 g 2 + g 2 2 − 4g 2 g 2 + 2g 2 κ 2 v 2 . (A.10)
R + R
The symmetric mass matrices are diagonalized by the orthogonal transformations

± ±
WL cos ξ sin ξ W1
± = − sin ξ cos ξ = , (A.11)
WR W2±
and

W3L cW c cW s sW Z1
W3R = −sW sM c − cM s −sW sM s + cM c cW sM Z2 , (A.12)
B −sW cM c + sM s −sW cM s − sM c cW cM A
where
e e
g= , g = √ , cW = cos ΘW , sW = sin ΘW ,
sin ΘW cos 2ΘW
√
cos 2ΘW
cM = , sM = tan ΘW , s = sin φ, c = cos φ.
cos ΘW
The mixing angles are given by
√
2κ1κ2 g 2 κ+
2 cos 2Θ
W
tan 2ξ = − , sin 2φ = − . (A.13)
vR2 2 cos2 ΘW (MZ2 2 − MZ2 1 )
Fig. 5 sums up the mass dependence of the additional gauge bosons and two sets of
Higgs scalar particles Eqs. (A.3), (A.4) on the vR scale.
A.3. Neutrinos
The MLRSM naturally contains left as well as right handed neutrino states. We chose
the following basis for these fields
c
ν ν
nR = R , nL = Lc , c
νL,R = Cν TL,R , (A.14)
νR νL
where both nL(R) form 6-dimensional vectors. The neutrino mass matrix takes the form

0 MD
Mν = T , (A.15)
MD MR
1 †
MD = √ (hl κ1 + h̃l κ2 ) = MD , (A.16)
2
√
MR = 2 hM vR = MRT . (A.17)
hl , h̃l , hM are the respective Yukawa coupling matrices.
Mν can be diagonalized with the following unitary transformation
T
KL
U= , (A.18)
KR†
where the KL,R matrices have dimension 6 × 3. The LEP neutrino counting results
show that there must be three light active neutrino states. This means that we must have
MD MR , which after requiring “natural” couplings (order one), turns into κ1,2 vR .
Let us now introduce 3 × 3 matrices ULl(h) , URl(h) [9]:

†
ULl
KL = †
, (A.19)
ULh
†
URl
KR = †
. (A.20)
URh
The diagonalization equation assumes the form (mdiag and Mdiag correspond to light and
heavy neutrino mass matrices, respectively):
† T ∗ ∗
ULl URl 0 MD ULl ULh mdiag 0
† T
= . (A.21)
ULh T
URh MD MR URl URh 0 Mdiag
This can be written as:
† ∗
ULl −MD MR−1 MD T
ULl mdiag , (A.22)
URl −MR−1 MDT ∗
ULl , (A.23)
URh MR URh Mdiag,
T
(A.24)
∗ ∗
∗ −1
ULh MD MR URh , (A.25)
where the unitarity of U , and the large scale difference Mdiag mdiag , have been used.
Two important conclusions can be drawn from it:
• the matrices ULl and URh are approximately unitary;

mD
• the elements of the non-diagonal submatrices URl and ULh are small, of order MR
O(1 GeV)
mN , where mN = Mdiag .
The symbol · · · denotes the relevant scale of mass matrices.

Altogether we can write
T
KL O(1) O(1/mN )
U= = . (A.26)
K† R
O(1/mN ) O(1)
References
[1] J.C. Pati, A. Salam, Phys. Rev. D 10 (1974) 275;

R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 2558;
R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 566;
G. Senjanovic, R.N. Mohapatra, Phys. Rev. D 12 (1975) 1502;
R.N. Mohapatra, P.B. Pal, Phys. Rev. D 38 (1998) 2226;
G. Senjanovic, Nucl. Phys. B 153 (1979) 334.
[2] H. Georgi, S.L. Glashow, Phys. Rev. Lett. 32 (1974) 438.
[3] G. Senjanovic, A. Sokorac, Phys. Rev. D 18 (1978) 2708;
G. Senjanovic, A. Sokorac, Phys. Lett. B 76 (1978) 610;
G. Beall, M. Bander, A. Soni, Phys. Rev. Lett. 48 (1982) 848;
J. Basecq, L.F. Li, P.B. Pal, Phys. Rev. D 32 (1985) 175;
A. Pilaftsis, Phys. Rev. D 52 (1995) 459;

Z. Gagyi-Palffy, A. Pilaftsis, K. Schilcher, Nucl. Phys. B 513 (1998) 517;
A.V. Gulov, V.V. Skalozub, Eur. Phys. J. C 17 (2000) 685.
[4] P. Ball, J.M. Frere, J. Matias, Nucl. Phys. B 572 (2000) 3;
M.E. Pospelov, Phys. Rev. D 56 (1997) 259;
K. Kiers, J. Kolb, J. Lee, A. Soni, G.H. Wu, hep-ph/0205082.
[5] T.G. Rizzo, Phys. Rev. D 50 (1994) 3303;
P. Cho, M. Misiak, Phys. Rev. D 49 (1994) 5894;
Barenboim, Bernabeu, Prades, M. Raidal, Phys. Rev. D 55 (1997) 4213.
[6] M. Czakon, J. Gluza, M. Zrałek, Nucl. Phys. B 573 (2000) 57.
[7] M. Czakon, J. Gluza, F. Jegerlehner, M. Zrałek, Eur. Phys. J. C 13 (2000) 275.
[8] J.F. Gunion, J. Grifols, A. Mendez, B. Kayser, F.I. Olness, Phys. Rev. D 40 (1989) 1546;
N.G. Deshpande, J.F. Gunion, B. Kayser, F.I. Olness, Phys. Rev. D 44 (1991) 837.
[9] J. Gluza, M. Zralek, Phys. Rev. D 48 (1993) 5093;
J. Gluza, M. Zralek, Phys. Rev. D 51 (1995) 4695.
[10] P. Duka, J. Gluza, M. Zrałek, Ann. Phys. 280 (2000) 336.
[11] V. Barger, T.J. Weiler, K. Whisnant, Phys. Lett. B 442 (1998) 255.
[12] M. Czakon, J. Gluza, M. Zralek, Phys. Lett. B 465 (1999) 211.
[13] J. Gluza, hep-ph/0201002.
[14] M. Czakon, J. Gluza, M. Zralek, Phys. Lett. B 458 (1999) 355.
[15] D.E. Groom, et al., Particle Data Group Collaboration, Eur. Phys. J. C 15 (2000) 1.
[16] M. Doi, T. Kotani, E. Takasugi, Prog. Theor. Phys. 71 (1984) 1440.
[17] W. Hollik, Fortschr. Phys. 38 (1990) 3, 165.
[18] T. van Ritbergen, R.G. Stuart, Nucl. Phys. B 564 (2000) 343.
[19] A. Sirlin, Phys. Rev. D 22 (1980) 971.
[20] F. Jegerlehner, hep-ph/0105283.
[21] M. Czakon, in preparation.
[22] J.A. Vermaseren, math-ph/0010025.
[23] G.J. van Oldenborgh, J.A.M. Vermaseren, Z. Phys. C 46 (1990) 425.
[24] T. Hahn, M. Perez-Victoria, Comput. Phys. Commun. 118 (1999) 153–165.
New G2 metric, D6-branes and lattice universe ✩

H. Lü
Michigan Center for Theoretical Physics, University of Michigan, Ann Arbor, MI 48109, USA
Received 21 May 2002; received in revised form 4 July 2002; accepted 25 July 2002
Abstract
We construct a new (singular) cohomogeneity-three metric of G2 holonomy. The solution can
be viewed as a triple intersection of smeared Taub–NUTs. The metric comprises three non-compact
radial-type coordinates, with the principal orbits being a T 3 bundle over S 1 . We consider an M-theory
vacuum (Minkowski)4 × M7 where M7 is the G2 manifold. Upon reduction on a circle in the T 3 ,
we obtain the intersection of a D6-brane, a Taub–NUT and a 6-brane with RR 2-form flux. Reducing
the solution instead on the base space S 1 , we obtain three intersecting 6-branes all carrying RR 2-
form flux. These two configurations can be viewed as a classical flop in the type IIA string theory.
After reducing on the full principal orbits and the spatial world-volume, we obtain a four-dimensional
metric describing a lattice universe, in which the three non-compact coordinates of the G2 manifold
are identified with the spatial coordinates of our universe.
1. Introduction
Seven-dimensional manifolds of G2 holonomy have long been known to exist. The

construction of explicit non-compact G2 metrics began ten years ago, when asymptotically
conical metrics of cohomogeneity one were found [1,2]. The physical interest of G2
manifolds has increasingly significantly with the discovery of M-theory, because they are
the most natural compactifying spaces from the eleven-dimensional point of view. It is
expected that M-theory compactified on a G2 manifold gives rise to an N = 1 super-
Yang–Mills theory in D = 4 [3]. The G2 manifold with principal orbits S 3 × S 3 provides
a geometrical demonstration of the classical flop of the type IIA superstring theory [4].
In [5], M-theory dynamics on a G2 manifold were discussed.
✩
Research is supported in full by DOE grant DE-FG02-95ER40899.
E-mail address: honglu@umich.edu (H. Lü).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 1 9 - 3
174 H. Lü / Nuclear Physics B 642 (2002) 173–180
Recently, a large class of new metrics of G2 holonomy have been obtained [6–17],
following the construction of the first examples of asymptotically locally conical spin(7)
manifolds [18]. These examples have non-Abelian isometry groups. G2 metrics with
nilpotent isometry groups were also constructed in [19], which can be obtained by taking
the Heisenberg or Euclidean limits of the non-Abelian examples. Whilst it is of great
interest to construct regular G2 metrics, physically, it is essential to have an appropriate
singularity structure to give rise to chiral fermions in D = 4 [20,21].
In Section 2, we construct a new non-compact cohomogeneity three metric with G2
holonomy. The metric has three non-compact radial-type coordinates, with the principal
orbits being a T 3 bundle over S 1 . The isometry group of the metric is a four-dimensional
nilpotent Lie group. The metric has either power-law singularities or delta-function
singularities. M-theory compactification on smooth and compact G2 manifolds was studied
in [22]; however, it was shown [3] that appropriate singularity is necessary for giving rise
to the four-dimensional N = 1 super-Yang–Mills theory. The solution can be viewed as
the intersection of three smeared Taub–NUTs. The singularity of the metric has its origin
of the smeared Taub–NUTs. When one of the Taub–NUT charges is set to zero, the metric
describes a product of S 1 with a six-dimensional non-compact Calabi–Yau manifold.
In Section 3, we consider an M-theory vacuum (Minkowski)4 × M7 , where M7 is the
G2 manifold. We show that by dimensionally reducing the solution on one of the circles
in the T 3 , we obtain a type IIA configuration with one D6-brane, one Taub–NUT and one
6-brane with an RR 2-form flux. On the other hand, if we reduce the solution on the base
space S 1 , we obtain an intersection of three 6-branes all carrying RR 2-form flux. These
two configurations can be viewed as the classical flop in type IIA string theory on a non-
compact six-dimensional Kähler manifold with a nilpotent isometry group. The origin of
the flop is that the T 3 bundle over S 1 principal orbits can also be viewed as S 1 bundle
over T 3 .
In Section 4, we perform a Kaluza–Klein reduction on the full principal orbits and the
spatial world-volume. We obtain three perpendicularly intersecting membranes in D = 4,
describing a lattice universe. In this picture, the three non-compact coordinates of the G2
manifold are identified with the spatial coordinates of our universe. We conclude the paper
in Section 5.
2. New G2 metric
The metric ansatz is given by

2
ds72 = H1 dx12 + H2 dx22 + H3 dx32 + H1 H2 H3 dz42 + H3 H1−1 dz1 + H1 z2 dz4
2 2
+ H1 H2−1 dz2 + H2 z3 dz4 + H2 H3−1 dz3 + H3 z1 dz4 , (1)
where H1 , H2 and H3 are functions of x1 , x2 and x3 , respectively. The prime on Hi denotes
a derivative with respect to the argument of Hi ;
H1 = ∂x1 H1 , H2 = ∂x2 H2 , H3 = ∂x3 H3 . (2)

H. Lü / Nuclear Physics B 642 (2002) 173–180 175
The natural vielbein basis is

e0 = H1 H2 H3 dz4 , e1 = H1 dx1 , e2 = H2 dx2 ,

e = H3 dx3 ,
3
e = H3 H1−1 dz1 + H1 z2 dz4 ,
4

e5 = H1 H2−1 dz2 + H2 z3 dz4 , e6 = H2 H3−1 dz1 + H3 z1 dz4 . (3)
The associative 3-form in this basis is given by
Φ = e016 + e024 + e035 + e125 − e134 + e236 − e456, (4)

where eij k = ei ∧ ej ∧ ek .
The metric (1) has G2 holonomy if and only if Φ is closed and
co-closed. We find that the closure and co-closure of Φ implies that
Hi = 0, i = 1, 2, 3, (5)
implying that
H 1 = 1 + m1 x 1 , H 2 = 1 + m2 x 2 , H 3 = 1 + m3 x 3 . (6)
Here the constant 1 is included so that Hi does not vanish when mi = 0. Clearly the metric
has a power-law singularity whenever any of the Hi vanishes. The metric can also be recast
in a “co-moving” frame,
9
ds72 = dr12 + dr22 + dr32 + (m1 m2 m3 r1 r2 r3 )2/3 dz42
4

m3 r3 2/3 m1 r1 2/3
+ (dz1 + m1 z2 dz4 )2 + (dz2 + m2 z3 dz4 )2
m1 r1 m2 r2

m2 r2 2/3
+ (dz3 + m3 z1 dz4 )2 . (7)
m3 r3
The fibration in the z1 , z2 and z3 coordinates implies that the constants mi are quantised,
namely [23]
L1 L2 L3
m1 = n1 , m2 = n2 , m3 = n3 , (8)
L2 L4 L3 L4 L1 L4
where ni are integers and Li are the periods of the zi . For simplicity, we can set Li = p
where p is the Plank length, and then mi = ni /p .
The metric has a power-law singularity when any of the Hi vanishes. This can be
avoided by instead taking

Hi = 1 + mα xi − x α
i i (9)
α
such that Hi is positive definite. However, in doing so, we have introduced delta function
singularities.
When all three of the Hi are non-vanishing, the metric describes three intersecting
Taub–NUTs with three independent non-vanishing smeared charges. The metric has three
non-compact coordinates x1 , x2 and x3 . The principal orbits are T 3 bundle over S 1 ; they
are parameterised by the coordinates (z1 , z2 , z3 ) and z4 , respectively. The metric is of

cohomogeneity three since it depends explicitly on the three non-compact coordinates xi .
Despite the dependence on the (z1 , z2 , z3 ) coordinates, they, together with z4 , parameterise
a four-dimensional nilpotent Lie group G, which is the isometry group of the metric, and
thus the four-dimensional principal orbits are homogeneous.
When two of the Hi vanish, the metric describes a direct product of Euclidean 3-space
and a smeared Taub–NUT. If instead only one of Hi vanishes, in which case the metric
was obtained in [24], it describes a product of an S 1 with a Calabi–Yau 6-manifold. To see
this in detail, let us set H3 = 1. The metric of the Calabi–Yau manifold is then given by
ds62 = H1 dx12 + H2 dx22 + H1 H2 dz42 + H2 dz32
2 2
+ H1−1 dz1 + H1 z2 dz4 + H1 H2−1 dz2 + H2 z3 dz4 (10)
and the Kähler form is given by
J = e0 ∧ e5 − e1 ∧ e4 + e2 ∧ e6 , (11)
where the vielbein is given by (3) with H3 = 1.
3. Intersecting D6-branes
Having obtained the new G2 metric, one may consider an M-theory vacuum solution
given by the direct product of Minkowski 4-spacetime and the G2 manifold, namely,
2
ds11 = −dt 2 + dw12 + dw22 + dw32 + ds72 . (12)
The solution can be viewed as a triple intersection of smeared Taub–NUTs, with the metric
represented by Diagram 1
There is a U (1) isometry for each of the zi coordinates, and so we can reduce the metric
on any zi to obtain a solution in type IIA theory. The zi for i = 1, 2, 3 are equivalent, and
hence the reduction can be discussed using z1 as a representative. The resulting type IIA
solution is given by
1
e 6 φ ds10
2
= −dt 2 + dw12 + dw22 + dw32 + H1 dx12
+ H2 dx22 + H3 dx32 + H1 H2 H3 dz42
2 2
+ H1 H2−1 dz2 + H2 z3 dz4 + H2 H3−1 dz32 + H3 H1 H1 z2 dz42
2
− W −1 H3 H1−1 H1 z2 dz4 − H2 H3−1 H3 z4 dz3 ,
t w1 w2 w3 x1 x2 x3 z1 z2 z3 z4
H1 (x1 ) × × × × − × × ∗ − × −
H2 (x2 ) × × × × × − × × ∗ − −
H3 (x3 ) × × × × × × − − × ∗ −
Diagram 1. Triple intersections of Taub–NUTs. Here ×, − and ∗ denote the

world-volume, transverse space, and fibre coordinates, respectively.
2
W = H3 H1−1 + H2 H3−1 H3 z4 ,
3
eφ = W 4 ,

A(1) = W −1 H3 H1−1 H1 z2 dz4 − H2 H3−1 H3 z4 dz3 . (13)
Note that before performing the Kaluza–Klein reduction, we have made a coordinate
transformation z3 → z3 − H3 z1 z4 in the metric (1). Clearly, the solution describes an
intersection of three objects. The one parameterised by H1 is a smeared D6-brane, and
the one parameterised by H2 is a Taub–NUT. The one associated with H3 is a 6-brane
carrying an RR 2-form flux, but it differs from a standard D6-brane.
We can instead reduce the solution on the z4 coordinate, giving the type IIA solution
1
e 6 φ ds10
2
= −dt 2 + dw12 + dw22 + dw32 + H1 dx12 + H2 dx22 + H3 dx32
+ H3 H1−1 dz12 + H1 H2−1 dz22 + H2 H3−1 dz32
2
− W −1 H3 H1−1 H1 z2 dz1 + H1 H2−1 H2 z3 dz2 + H2 H3−1 H3 z1 dz3 ,
2 2
eφ = W 4 , W = H1 H2 H3 + H3 H1−1 H1 z2 + H1 H2−1 H2 z3
3
2
+ H2 H3−1 H3 z1 ,

A(1) = W −1 H3 H1−1 H1 z2 dz1 + H1 H2−1 H2 z3 dz2 + H2 H3−1 H3 z1 dz3 . (14)
The solution describes three intersecting 6-branes all carrying RR 2-form flux. These
6-branes are different from the usual D6-brane coming from the reduction of the fibre
coordinate of a Taub–NUT in D = 11.
The two configurations arising from the reduction on z1 or z4 can be viewed as a
classical flop in the type IIA string theory on the non-compact Kähler manifold. The flop
in D = 10 can be geometrically explained by the fact that the T 3 bundle over S 1 principal
orbits of the four-dimensional nilpotent Lie group G can also be described as an S 1 bundle
over T 3 . However, the two descriptions are somewhat different. In the latter case, the fibre
is the circle group in G generated by ∂/∂z4 . In the former case, the fibre is not the orbit of
a three-dimensional subgroup of G because ∂/∂zi for i = 1, 2, 3 are not themselves Killing
vectors; we must add a multiple of ∂/∂z4 . In fact the flop involves interchanging the fibre
and base spaces of the U (1) fibration. This is analogous to the flop discussed in [4].
4. Lattice universe
The new G2 metric (1) that we have obtained is in fact inspired by the four-dimensional
intersecting membrane solution that describes the lattice universe [24].1 There has been
experimental evidence suggesting that the network of galaxy superclusters and voids seems
to form a three-dimensional lattice with a spacing of about 120h−1 Mpc (where h−1 is the
Hubble constant in units of 100 km s−1 Mpc−1 ) [26–28]. In [24], an M-theory solution was
constructed to describe such a lattice structure, which can be realised by considering non-
standard brane intersections of two M5-branes and one Taub–NUT, or two Taub–NUTs
and one M5-brane. In the latter case, turning off the M5-brane charge causes the solution to
1 In [25], a triply quasi-periodic Gibbons–Hawking metric was obtained.

reduce to the product of 5-dimensional Minkowski spacetime and the non-compact Calabi–
Yau manifold given in (10).
In Section 2, we obtained the new G2 metric (1) by adding an extra fibration on
the seventh coordinate. This procedure follows the general prescription of obtaining G2
manifolds from six-dimensional Kähler manifolds, described in detail in [29].
If we reduce the M-theory solution on the world-volume spatial coordinates wi and also
on the T 3 bundle over S 1 principal orbits, we obtain three perpendicularly intersecting
membranes in D = 4, with the metric
1
ds42 = (H1 H2 H3 ) 2 −dt 2 + H1 dx12 + H2 dx22 + H3 dx32 . (15)
This metric was first obtained in [24], although it was supported by very different field
strength. The functions of Hi in this case are given by (9) describing periodic arrays of
intersecting membranes.
In this static cosmological model the spatial world-volume and the T 3 bundle over S 1
principal orbits are viewed as an internal space, whilst the three non-compact coordinates
x1 , x2 and x3 of the G2 manifold are identified with the spatial coordinates of our universe.
This is rather natural since the principal orbits are clearly compact, and the spatial world-
volume can be wrapped on a compact space such as T 3 . Although it is not likely that the
metric (15) describes our actual universe, since it preserves N = 1 supersymmetry; it is
nevertheless rather suggestive that the lattice structure should emerge from a metric with
G2 holonomy.
5. Conclusions
In this paper, we constructed a new cohomogeneity-three metric with G2 holonomy. It

has three radial-type coordinates, with the principle orbits being a T 3 bundle over S 1 .
The solution can be viewed as three intersecting Taub–NUTs. We performed Kaluza–
Klein reduction on the S 1 and instead on a circle in the T 3 . The two resulting type IIA
configurations can be viewed as a classical flop in type IIA string theory on a non-compact
Kähler six-manifold. Although the type IIA solutions do not describe the triply intersecting
D6-branes advocated in [21,30] for the realisation of chiral fermions, it is nevertheless of
interest to investigate further if chirality could arise from the singularities of our G2 metric.
The metric provides a concrete example for studying such an issue, since it can be viewed
as the lifting of a D6-brane configurations in D = 10.
If we perform a Kaluza–Klein reduction on the entire four-dimensional full principal
orbits and the spatial world-volume, we obtain triply intersecting membranes in D = 4,
describing a lattice universe. The construction takes full advantage of the non-compact
nature of the manifold, in that the three non-compact coordinates are precisely identified
with the spatial coordinates of our universe. It is of great interest to investigate further the
significance of such a configuration arising from a G2 manifold.
Acknowledgement
We are grateful to Mirjam Cvetič for extensive discussions on triply intersecting D6-
branes and chiral fermions, to Gary Gibbons for pointing out the isometry group of the
metric, and to Chris Pope and Justin Vázquez-Poritz for discussions.
References
[1] R.L. Bryant, S. Salamon, On the construction of some complete metrics with exceptional holonomy, Duke
Math. J. 58 (1989) 829.
[2] G.W. Gibbons, D.N. Page, C.N. Pope, Einstein metrics on S 3 , R3 and R4 bundles, Commun. Math.
Phys. 127 (1990) 529.
[3] B.S. Acharya, On realising N = 1 super Yang–Mills in M-theory, hep-th/0011089.
[4] M. Atiyah, J. Maldacena, C. Vafa, An M-theory flop as a large N duality, J. Math. Phys. 42 (2001) 3209,
hep-th/0011256.
[5] M. Atiyah, E. Witten, M-theory dynamics on a manifold of G2 holonomy, hep-th/0107177.
[6] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, Supersymmetry M3-branes and G2 manifolds, Nucl. Phys.
B 620 (2002) 3, hep-th/0106026.
[7] A. Brandhuber, J. Gomis, S.S. Gubser, S. Gukov, Gauge theory at large N and new G(2) holonomy metrics,
Nucl. Phys. B 611 (2001) 179, hep-th/0106034.
[8] N. Hitchin, Stable forms and special metrics, math.DG/0107101.
[9] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, Cohomogeneity one manifolds of Spin(7) and G2 holonomy,
hep-th/0108245.
[10] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, Orientifolds and slumps in G2 and Spin(7) metrics, hep-
th/0111096.
[11] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, M-theory conifolds, Phys. Rev. Lett. 88 (2002) 121602, hep-
th/0112098.
[12] A. Brandhuber, G2 holonomy spaces from invariant three-forms, hep-th/0112113.
[13] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, A G2 unification of the deformed and resolved conifolds,
hep-th/0112138, to appear in Phys. Lett. B.
[14] R. Hernandez, K. Sfetsos, An eight-dimensional approach to G2 manifolds, hep-th/0202135.
[15] S. Gukov, S.-T. Yau, E. Zaslow, Duality and fibrations on G2 manifolds, hep-th/0203217.
[16] K. Behrndt, Singular 7-manifolds with G2 holonomy and intersecting 6-branes, hep-th/0204061.
[17] Z.W. Chong, M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, P. Wagner, General metrics of G2 holonomy and
contraction limits, hep-th/0204064.
[18] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, New complete non-compact Spin(7) manifolds, hep-
th/0103155;
M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, New cohomogeneity one metrics with Spin(7) holonomy,
math.DG/0105119.
[19] G.W. Gibbons, H. Lü, C.N. Pope, K.S. Stelle, Supersymmetric domain walls from metrics of special
holonomy, Nucl. Phys. B 623 (2002) 3, hep-th/0108191.
[20] B. Acharya, E. Witten, Chiral fermions from manifolds of G2 holonomy, hep-th/0109152.
[21] M. Cvetič, G. Shiu, A.M. Uranga, Chiral type II orientifold constructions as M-theory on G2 holonomy
spaces, hep-th/0111179.
[22] G. Papadopoulos, P.K. Townsend, Compactification of D = 11 supergravity on spaces of exceptional
holonomy, Phys. Lett. B 357 (1995) 300, hep-th/9506150.
[23] I.V. Lavrinenko, H. Lü, C.N. Pope, Fibre bundles and generalised-dimensional reductions, Class. Quantum
Grav. 15 (1998) 2239, hep-th/9710243.
[24] M.J. Duff, P. Hoxha, H. Lü, R.R. Martinez-Acosta, C.N. Pope, A lattice universe from M-theory, Phys. Lett.
B 451 (1999) 38, astro-ph/9712301.
[25] S. Nergiz, C. Saclioglu, A quasiperiodic Gibbons–Hawking metric and space–time foam, Phys. Rev. D 53
(1996) 2240, hep-th/9505141.
[26] J. Einasto, M. Einasto, S. Gottlober, V. Muller, V. Saar, A.A. Starobinsky, E. Tago, D. Tucker, H. Andernach,
P. Frisch, A 120 Mpc periodicity in the three-dimensional distributions of galaxy superclusters, Nature 385
(1997) 139.
[27] R. Kirshner, The universe as a lattice, Nature 385 (1997).
[28] J. Einasto, Has the universe a honeycomb structure, astro-ph/9711320.
[29] M. Cvetič, G.W. Gibbons, H. Lü, C.N. Pope, Almost special holonomy in type IIA and M-theory, hep-
th/0203060.
[30] E. Witten, Deconstruction, G2 holonomy, and doublet-triplet splitting, hep-ph/0201018.
The phase diagram of four flavor SU(2) lattice gauge

theory at nonzero chemical potential and
temperature
John B. Kogut a , Dominique Toublan a , D.K. Sinclair b
a Physics Department, University of Illinois at Urbana-Champaign, Urbana, IL 61801-3080, USA
b HEP Division, Argonne National Laboratory, 9700 South Cass Avenue, Argonne, IL 60439, USA
Received 21 May 2002; received in revised form 11 July 2002; accepted 5 August 2002
Abstract
SU(2) lattice gauge theory with four flavors of quarks is simulated at nonzero chemical potential
µ and temperature T and the results are compared to the predictions of effective Lagrangians.
Simulations on 164 lattices indicate that at zero T the theory experiences a second order phase
transition to a diquark condensate state. Several methods of analysis, including equation of state fits
suggested by Chiral Perturbation Theory, suggest that mean-field scaling describes this critical point.
Nonzero T and µ are studied on 123 × 6 lattices. For low T , increasing µ takes the system through
a line of second order phase transitions to a diquark condensed phase. Increasing T at high µ, the
system passes through a line of first order transitions from the diquark phase to the quark–gluon
plasma phase. Metastability is found near the first order line. Presumably, there is a tricritical point
along this line of transitions. We estimate its position to be consistent with theoretical predictions.
PACS: 12.38.Mh; 12.38.Gc; 11.15.Ha
1. Introduction
Recently there has been a resurgence of interest in QCD at nonzero chemical potential
for quark number. Arguments based on instantons [1] and phenomenological gap equations
[2] support the old idea that diquark condensation and a color superconductivity phase
transition [3,4] will occur at a critical chemical potential somewhat greater than that at
E-mail address: j-kogut@uiuc.edu (J.B. Kogut).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 7 8 - 8
182 J.B. Kogut et al. / Nuclear Physics B 642 (2002) 181–209
which nuclear matter forms. Nuclear matter is expected to be the ground state of QCD
above a chemical potential µN , which is slightly less than one-third of the proton’s
mass. These arguments neglect the forces of confinement and may only be reliable
at asymptotically large chemical potentials. Since diquarks carry color in real QCD,
a quantitative understanding of confinement and screening is needed to estimate the
energies of the states which control the phases of the system. Unfortunately, brute force
lattice simulation methods, which are so useful at T = 0 and µ = 0, do not yet provide
a reliable simulation algorithm for these environments in which the fermion determinant
becomes complex.
Given these problems, theorists have turned to simpler models which can be analyzed
and simulated by current methods. One of the more interesting is the color SU(2) version
of QCD which addresses some of the issues of interest [1,2,5]. In this model diquarks
do not carry color, so their condensation does not break color symmetry dynamically.
Instead, the diquark condensed phase resembles a superfluid rather than a superconductor.
The critical chemical potential is also expected to be one-half the mass of the lightest
meson, the pion, because quarks and anti-quarks reside in equivalent representations of
the SU(2) color group. Hence, hadron flavor multiplets include both mesons and diquarks.
Chiral Lagrangians can be used to study the diquark condensation transition in this model
because the critical chemical potential vanishes in the chiral limit, and the model has a
Goldstone realization of the spontaneously broken quark-number symmetry [6–9]. The
problem has also been studied within a Random Matrix Model at nonzero µ and T [10].
Lattice simulations of the model are also possible because the fermion determinant is
real and non-negative for all chemical potentials. One hopes that these developments will
uncover generic phenomena that will also apply to QCD at nonzero chemical potential.
Furthermore, most of the theories that aim at a description of the phase diagram of 3-
color QCD at nonzero baryon chemical potential and temperature can be implemented
to describe the phase diagram of 2-color QCD at nonzero baryon chemical potential and
temperature. Therefore, an important motivation to conduct Lattice simulations is to check
the validity of these theories in the 2-color case. If a theory is not correct in the two-color
case, it is very doubtful that the predictions of the corresponding 3-color theory can be
trusted.
Preliminary lattice simulations of the SU(2) model with four species of quarks,
simulation data and an effective Lagrangian analysis of aspects of the T –µ phase diagram
were recently published [11,12]. It is the purpose of this article to present the continuation
of that work on larger lattices, closer to the theory’s continuum limit with a focus on the
system’s phase diagram at nonzero µ and T . Very early work on this model at finite T
and µ was performed by [13]. A simulation study of the spectroscopy of the light bosonic
modes will be presented elsewhere [14].
In our exploratory study [12] of this model’s phase diagram, we found an unanticipated
result. At fixed but small quark mass, which insures us that the pion has a nonzero mass
and chiral symmetry is explicitly broken, we found that at low T and µ there is a line
of second order transitions separating a hadronic from a superfluid phase, but at high T
and µ it becomes a line of first order transitions separating a diquark condensate from a
quark–gluon plasma phase. Along this line there presumably is a tricritical point, labeled
2 in Fig. 1, where the transition switches from being second order (and well described by
J.B. Kogut et al. / Nuclear Physics B 642 (2002) 181–209 183
Fig. 1. Schematic phase diagram of diquark condensation in the T –µ plane. The thin (thick) line consists of
second (first) order transitions. The dashed line denotes a crossover. Point 1 labels a critical point and point 2
labels a tricritical point. The existence and position of point 1 is a matter of conjecture in the two color model.
mean field theory) at relatively low µ to a first order transition at an intermediate µ value.
We have not confirmed the existence of this point explicitly. We will, however, present
evidence for metastability along the line of high µ and T . We have argued elsewhere [12]
that this simulation result, the existence of a tricritical point 2, has a natural explanation in
the context of chiral Lagrangians. Following the formalism of [6] we argued that trilinear
couplings among the low lying boson fields of the Lagrangian become more significant as
µ and T increase and they can cause the transition to become first order at a µ value in the
vicinity of the results found in the simulation. Using hindsight, this behavior should not
have come as a surprise. µ plays the role of a second ‘temperature’ in this theory in that it
is a parameter that controls diquark condensation, but does not explicitly break the (quark-
number) symmetry. The existence of two competing ‘temperatures’ is a characteristic
of systems which exhibit tricritical behavior. A tricritical point is also found in Chiral
Perturbation Theory, as well as in a new Random Matrix Theory model [15,16].
The conjectured phase diagram of Fig. 1 also has a critical point labeled 1 which
connects the line of first order transitions to the dashed line of crossover phenomena
extending to the µ = 0 axis where we expect a finite T chiral transition between
conventional hadronic matter and the quark–gluon plasma. For sufficiently light quarks
in the four flavor SU(2) model, the µ = 0 transition is known to be first order both from
theoretical arguments [17,18] and simulations [19]. For quark mass m = 0.05 we find that
the transition is smoothed to a crossover at µ = 0 and remains a crossover even as µ
is increased and the dynamics cause diquark condensation. Clearly, the character of the
transition between the hadronic and the quark–gluon plasma phase will depend on the
precise value of the quark mass m. Intensive simulations on a range of lattice sizes at
µ = 0.0 support the idea that at intermediate values of the quark mass there is only a
crossover between hadronic matter and the quark–gluon plasma at nonzero T [20]. At
m = 0.05 we find that the transition at µ = 0.10 is as smooth as that at µ = 0.0, while
the transition at µ = 0.20 is noticeably sharper. This result suggests that if m were chosen
somewhat smaller than m = 0.05, the scenario with the critical point 1 at the end of a
first order line to the left of the diquark phase might emerge. However, this critical point
1 might be absent so the crossover line would intersect the curve separating the diquark
condensation phase from the normal phase. We will be pursuing this suggestion in separate
simulations. The simulations done here are not accurate enough to probe the details of the
region of the phase diagram Fig. 1 where the two first order lines split apart. Simulations
at lower quark masses on larger lattices might be needed.
In QCD at nonzero baryon chemical potential Fodor and Katz [21] have estimated the
position of the critical point 1 using a modified, multivariable Glasgow algorithm [22]. It
will be interesting to find point 1 in the four flavor SU(2) model at smaller m values using
our conventional simulation methods and then checking whether the Fodor–Katz method
can find it with competitive accuracy. If point 1 is absent, the Fodor–Katz method should
be able to find the intersection of the crossover line with the curve delimiting the diquark
phase.
We also present extensive and accurate simulation results on a 164 lattice, essentially
at vanishing T . The critical indices of the transition at nonzero µ are measured and favor
mean-field scaling in agreement with Chiral Perturbation Theory, analyzed through one
loop corrections, and simulations of 2-color QCD in the strong coupling limit [23]. The
critical indices βmag and δ are emphasized. Equation of state fits, suggested by Chiral
Perturbation Theory, also support these conclusions. As a third method to determine the
critical indices, we use the static scaling hypothesis. This analysis is also compatible with
mean field, although this approach is unable to predict the index δ accurately because of
its sensitivity to estimates of the critical chemical potential, µc . Unfortunately, because
simulations at small λ are extremely costly in computing requirements, we have not
been able to approach the critical point closely enough to make definitive measurements,
insensitive to the detailed nature of the critical scaling. In a recent paper in which we study
quenched versions of this and closely related theories, we have indicated how sensitive
measurements of critical scaling are to the form of the scaling variable [24]. This paper
does indicate, however, that there are classes of scaling variables for which there exist
extended regions outside of the true scaling window, where the order parameter scales
with the correct critical exponents.
2. Simulation method and the algorithm
The lattice action of the staggered fermion version of this theory is:
1
Sf = / + m χ + λ χ T τ2 χ + χ̄ τ2 χ̄ T ,
χ̄ D(µ) (1)
2
sites
where the chemical potential µ is introduced by multiplying links in the +t direction by
eµ and those in the −t direction by e−µ [25]. The diquark source term (Majorana mass
term) is added to allow us to observe spontaneous breakdown of quark-number on a finite
lattice. The parameter λ and the usual mass term m control the amount of explicit symmetry
breaking in the lattice action. We will be particularly interested in small values of λ and the
extrapolation λ → 0 for a given m to produce an interesting, realistic physical situation.
This is the system that has been studied analytically using effective Lagrangians at nonzero
chemical potential µ [6,7], and nonzero temperature [15].
Integrating out the fermion fields in Eq. (1) gives:

λτ2 A
pfaffian = det A† A + λ2 , (2)

−AT λτ2
where
A ≡ D(µ)
/ + m. (3)
Note that the pfaffian is strictly positive, so that we can use the hybrid molecular
dynamics [26] method to simulate this theory using “noisy” fermions to take the square
root, giving Nf = 4.
For λ = 0, m = 0, µ = 0 we expect no spontaneous symmetry breaking for small µ. For
µ large enough (µ > mπ /2 according to most approaches including Chiral Perturbation
Theory [6–8]) we expect spontaneous breakdown of quark number and one Goldstone
boson—a scalar diquark. (The reader should consult [5] for a full discussion of the
symmetries of the lattice action, remarks about spectroscopy, Goldstone as well as pseudo-
Goldstone bosons, and for early simulations of the 8 flavor theory at λ = 0.)
3. Simulation results and analysis
We now consider the simulation results for the Nf = 4 theory on 164 , 84 and 123 × 6
lattices, measuring the chiral ( χ̄ χ
) and diquark condensates ( χ T τ2 χ
), the fermion
number density, the Wilson/Polyakov line, etc.
Let us first briefly mention our 84 simulations. These were performed at β = 1.5,
a relatively strong coupling, and m = 0.025. These extend our earlier work with the same
β but m = 0.1. Using a smaller quark mass increases the validity of chiral perturbation
analyses. In addition, we are performing simulations on a 123 × 24 lattice with the same
parameters to examine the spectrum of Goldstone and pseudo-Goldstone bosons of this
theory. For this purpose it is necessary to have the pion mass significantly lighter than that
of non-Goldstone, non-pseudo-Goldstone excitations. We have simulated at λ = 0.0025,
and 0.005. Our measurements of the diquark condensate are given in Fig. 2. Each ‘data’
point represents a 2000 molecular-dynamics time-unit simulation. The fact that a simple
linear extrapolation of our measurements to λ = 0 is very small, not only at µ = 0, where
we know that the condensate vanishes, but also at µ = 0.1 leads us to believe that the
condensate vanishes in this limit, probably up to µ ∼ 0.2. (Note that is known that there
is some small but finite value of µ, below which this condensate vanishes [23].) For µ
much greater than 0.2 but less than the saturation value it is reasonably clear that the
condensate does not vanish in the limit as λ → 0 (on an infinite lattice). Hence there is
a phase transition for some value of µ to a state where quark number is spontaneously
broken by a diquark condensate. The transition occurs over a relatively small range of µ,
but does not appear to be first order. Because of the relatively small scaling window this
implies, we will not be able to analyze critical scaling until we perform measurements at
more µ values within this scaling window.
Fig. 2. χ T τ2 χ
as a function of µ on an 84 lattice at β = 1.5, m = 0.025. The unlabeled squares are a linear
extrapolation to λ = 0.
The chiral condensate remains approximately constant up to µ = µc after which it falls

rapidly approaching zero for large µ. The fermion density rises slowly and approximately
linearly from zero for µ µc . Above µ ≈ 0.6, it increases more rapidly approaching its
saturation value of 2 at µ ≈ 1.0.
We now turn to measurements on a 164 , ‘zero temperature’ lattice. We simulated the
SU(2) model at a relatively weak coupling β = 1.85, within the theory’s apparent scaling
window, in order to attempt to make contact with the theory’s continuum limit. The quark
mass was m = 0.05, as in [12], and a series of simulations were performed at λ = 0.0025,
0.005, and 0.01 so that our results could be extrapolated to vanishing diquark source, λ = 0.
Simple linear and spline extrapolations were used and these procedures appeared to be
sensible away from the transition. None of the conclusions to be drawn here will depend
strongly on the limit. In fact, everything we say could be gathered from our data at a fixed
(small) λ value, 0.005 or 0.0025, say. However, only in the limit of vanishing λ do we
expect real diquark phase transitions (at least when the transitions are second order), so
it is particularly interesting and relevant to investigate the λ → 0 limit. The quantitative
agreement between fits at fixed, small λ values and those using extrapolated data enforce
the prevalent idea that simple extrapolation procedures are reliable except in the immediate
vicinity of the critical point where non-analyticities are inevitable.
Because we are forced by practical considerations to work at λ values large enough that
the condensate is relatively smooth over the transition region, we have to deduce critical
Table 1
Diquark condensates on a 164 lattice at λ = 0.0025, 0.005, and 0.010 in the
second, third, and fourth column, respectively. The first column lists the µ values
µ χ T τ2 χ
χ T τ2 χ
χ T τ2 χ
0.25 0.0617(2) 0.1134(2) 0.1962(4)

0.28 0.1120(5) 0.1732(4) 0.2546(5)
0.29 0.1407(5) 0.1993(5) 0.2782(4)
0.30 0.1772(9) 0.2269(6) 0.3016(5)
0.31 0.2120(6) 0.2544(6) 0.3240(3)
0.32 0.2415(11) 0.2783(7) 0.3470(6)
0.33 0.2711(7) 0.3045(7) 0.3687(5)
0.34 0.2951(11) 0.3253(9) 0.3893(6)
0.35 0.3200(9) 0.3499(7) 0.4082(7)
0.36 0.3437(12) 0.3651(9) 0.4287(6)
0.38 0.3815(13) 0.4044(9) 0.4627(7)
Table 2
Diquark condensates on a 164 lattice at λ = 0.00. The first column lists the µ
values, the second records the linear extrapolations of the diquark condensates
and the third records the quadratic extrapolations
µ χ T τ2 χ
χ T τ2 χ
0.25 0.0100(2) 0.0031(2)

0.28 0.0508(5) 0.0371(5)
0.29 0.0821(5) 0.0693(5)
0.30 0.1275(9) 0.1193(9)
0.31 0.1710(6) 0.1664(6)
0.32 0.2050(11) 0.2031(11)
0.33 0.2380(7) 0.2376(7)
0.34 0.2650(11) 0.2661(11)
0.35 0.2900(9) 0.2896(9)
0.36 0.3220(12) 0.3292(12)
0.38 0.3590(13) 0.3628(12)
scaling from the curvature of the condensate further from µc than we would like. Thus
we run the risk that we are outside the scaling window and the apparent scaling we are
seeing does not represent true critical scaling. However, we saw for quenched theories [24]
at relatively weak couplings, that the form of the natural scaling variable is such that it
has an inflection point when expressed in terms of the simplest scaling variable µ − µc .
This leads to an extended domain where the order parameter scales with the true critical
exponents. It is reasonable to assume that dynamical theories behave similarly.
In Table 1 we present the raw data on the 164 lattice at the three λ values. Table 2
exhibits the results of extrapolating this data to λ = 0. Typically, one thousand time units
of the Hybrid Molecular Dynamics algorithm were accumulated to generate these data
sets. The error bars in the data sets of Table 1 account for the correlations in the raw data
sets. The low λ runs were our most CPU intensive. Table 3 shows the corresponding chiral
condensate data.
Fig. 3. Diquark condensate vs. µ on 164 lattice; β = 1.85; m = 0.05; λ = 0.00 (linear).
Table 3
Quark–antiquark condensates on a 164 lattice at λ = 0.0025, 0.005, and 0.010 in
the second, third, and fourth column, respectively. The first column records the
µ values
µ χ̄ χ
χ̄ χ
χ̄ χ
0.25 0.4123(5) 0.4012(5) 0.3887(5)

0.28 0.4025(8) 0.3850(6) 0.3697(6)
0.29 0.3934(7) 0.3758(5) 0.3616(4)
0.30 0.3816(9) 0.3664(7) 0.3529(5)
0.31 0.3686(6) 0.3541(6) 0.3444(5)
0.32 0.3523(11) 0.3409(8) 0.3335(6)
0.33 0.3381(6) 0.3298(8) 0.3234(5)
0.34 0.3242(11) 0.3157(8) 0.3125(6)
0.35 0.3091(10) 0.3057(7) 0.3026(7)
0.36 0.2987(12) 0.2908(9) 0.2920(7)
0.38 0.2733(12) 0.2722(9) 0.2711(7)
Since the hybrid algorithm suffers from systematic errors proportional to the square
of the discrete time step used in integrating its stochastic differential equations forward
in Monte Carlo time [26], we were forced to a small discrete time step. The simulations
reported here used time steps as small as dt = 0.0025 to control these errors. Even then,
because our statistical errors are so small, these systematic errors might not be negligible.
We will later indicate cases where we used dt values which were larger than optimal and
have seen signs of such errors.
Now let us discuss some results. In Fig. 3 we show the diquark condensate linearly
extrapolated to λ = 0 plotted against the chemical potential µ. The quark mass m was
Fig. 4. Diquark condensate vs. µ on 164 lattice, β = 1.85; m = 0.05; λ = 0.00 (spline).
fixed at 0.05, the coupling was β = 1.85 and the linear extrapolation used the raw data at
λ = 0.0025 and 0.005.
We see evidence to a quark-number violating second order phase transition in this figure.
The dashed line is a power law fit from µ = 0.30 to 0.35 which predicts the critical
chemical potential of µc = 0.2860(2). The power law fit is good, its confidence level is
48 percent, and its critical index is βmag = 0.54(3) which is consistent with the mean
field result βmag = 1/2, predicted by chiral perturbation theory [6,7] including one loop
corrections. Note that the quark mass is fixed at m = 0.05 throughout this simulation, so
chiral symmetry is explicitly broken and the transition of Fig. 2 is due to quark number
breaking alone.
In Fig. 4 we show the diquark data extrapolated to λ = 0 quadratically using the raw
data at λ = 0.010, 0.005 and 0.0025. The fit to the data at µ ranging from 0.30 through
0.35 is also shown. The fit predicts a critical point µc = 0.2870(2), has a confidence level
of 47 percent and has a critical exponent of βmag = 0.54(4), again in agreement with mean
field theory.
Further evidence for βmag = 1/2 can be obtained directly from the raw data without
any extrapolations. For example, in Fig. 5 we show the λ = 0.0025 data and a power law
fit ranging form µ = 0.30 through 0.38. We find µc = 0.2749(2), βmag = 0.54(5) with a
confidence level of 24 percent.
We also considered extrapolations of the form χ T τ2 χ
= ρ where ρ is the solution
of an equation of the form ρ 3 = aρ + bλ with a and b determined as functions of µ by
performing an independent fit at each µ. This form is suggested my mean-field scaling
forms, which would predict that a should be proportional to µ − µc close to the transition.
However, these fits were so poor that this approach was abandoned.
Fig. 5. Diquark condensate vs. µ on 164 lattice, β = 1.85; m = 0.05; λ = 0.0025.
Fig. 6. Fermion number density vs. µ on 164 lattice, β = 1.85; m = 0.05; λ = 0.0025.
In addition, we considered the fermion number density which is predicted to rise linearly
from µc and found good agreement with this expectation, as shown in Fig. 6.
The fit predicts the critical point µc = 0.2765(5), with a critical index 1.17(7) and a
confidence level of 13.1 percent. Plots and fits of the fermion number density at the other
Fig. 7. Diquark condensate vs. µ on 164 lattice, β = 1.85; m = 0.05; λ = 0.005.
λ values were equally good. In fact the fermion number density shows relatively little λ
dependence. This is in agreement with Chiral Perturbation Theory [7].
Finally, we considered the λ = 0.005 simulations and extended the measurements up to
large µ → 1.2, shown in Fig. 7. We see the curious phenomenon observed in the past: as
µ becomes large, the condensate falls to zero. The point is that once µ becomes of order
unity, the discreteness of the lattice comes into play and suppresses the density of states
which forces the condensate to fall toward zero. The quark-number density approaches
2 (per lattice site), the maximum allowed by Fermi statistics. This is a lattice artifact.
Nonetheless, we can try a power law fit to the diquark condensate over the entire region of
µ from 0.30 to 0.60 where the curve is increasing. The result is shown in the figure and
the fit predicts βmag = 0.49(7) and µc = 0.2651(2). Since χ 2 /dof ∼ 5, the fit is poor, but
the deviations from this fit are small enough that this fit should be considered a reasonable
zeroth order approximation.
In summary, βmag measurements are relatively easy and self-consistent. For all the
parameters and procedures we tried, we found good agreement with mean field theory.
It was important, however, to be working at relatively weak coupling, β = 1.85, in the
theory’s scaling window where such an extended scaling window exists. Experience has
shown that at stronger coupling, the true scaling is described by effective Lagrangians
closer to the non-linear sigma model form, where the scaling window in µ − µc is small
and the extended scaling regime exhibits scaling in µ − µc but with an apparent critical
exponent βm which is half the true value [23,24,27].
A second critical index in which we are interested is δ, which controls the response of
the system to symmetry breaking fields at the critical point µc ,
χ T τ2 χ
= Aλ1/δ (4)
Fig. 8. Diquark condensate vs. λ for µc = 0.2855 on 164 lattice, m = 0.05.
in the limit of small λ. Mean field theory predicts δ = 3. This scaling law Eq. (4) is usually
difficult to use in practice because first it requires accurate estimates of µc , and second it
generally suffers from a small scaling window in λ. Not having a precise estimate of µc is
probably our biggest problem, since as we have shown in [24], fits to the extended scaling
regime can underestimate µc by ≈ 8%. In order to obtain some useful data, however, we
simulated the 164 lattice for four estimates of µc each over a range of λ from 0.002 to
0.010. First we note that the chosen dt for the λ = 0.002 simulations was too large and
although plotted on our graphs, these points were excluded from our fits.
We considered 0.2855, 0.2870, 0.2920 and 0.2970 as estimates of µc and tried fits of
the form Aλ1/δ + B. The data and fits are shown in the next four figures, which we discuss
in turn.
The µ = 0.2855 data predicted δ = 4.0(6) with a confidence level of 91 percent, but
with a large, negative constant B = −0.17(7). The fit considers the data from λ ranging
from 0.004 to 0.010 because the data at λ = 0.002 has large dt 2 errors. These results are
shown in Fig. 8. The fact that B is large and negative suggests that this estimate for µc is
too low.
In Fig. 9 we consider the same ideas for the estimate µc = 0.2870 and now find
δ = 2.5(2) and B = −0.02(2), but with a rather poor confidence level of a few percent.
For µc = 0.2920 (Fig. 10), we find δ = 2.1(2), but B is distinctly different from zero, B =
0.05(2). The confidence level is, however, very good, 96 percent. Finally, for µc = 0.2970
we find δ = 2.7(6) and B = 0.02(4) with a confidence level of about one percent. See
Fig. 11.
In summary, δ is in the mean field ‘ballpark’, but the uncertainly in the simulation results
is large. The results give us no reason to doubt the application of mean field theory because
they are all in the vicinity of δ = 3 with some fits higher and some lower. The sensitivity of
the simulation estimates to µc is considerable. We learn from these measurements that only
a focused set of measurements that determine µc very well will produce a useful estimate
of δ.
4. Comparison with chiral perturbation theory
It is interesting to compare the lattice data to the theoretical predictions of Chiral

Perturbation Theory. Two-color QCD at nonzero baryon/quark-number chemical potential
has been thoroughly studied at the tree level and at next-to-leading order within Chiral
Perturbation Theory [6–8]. At next-to-leading order, it was found that the phase transition
between the normal phase and the superfluid diquark phase is second order, that the critical
chemical potential is given by µc = mπ /2 and that the critical exponents are given by
mean-field theory. Furthermore, based on the structure of the loop corrections, it was
conjectured that the critical exponents are given by mean field at any (finite) order in Chiral
Perturbation Theory [8].
Overall, the next-to-leading-order corrections do not dramatically change the leading-
order results [8]. The corrections turn out to be small, as long as µ is close to µc . Therefore,
it should be enough to compare the lattice data to Chiral Perturbation Theory at leading-
order, at least for the lattices sizes, couplings and parameters considered here.
In Chiral Perturbation Theory at leading order, the diquark condensate at nonzero
diquark source is given by
χ T τ2 χ
= χ̄ χ
0 sin α, (5)
where χ̄χ
0 is the chiral condensate at zero chemical potential, zero temperature, and zero
diquark source, and α is given implicitly by
4µ2 cos α sin α = m2π sin(α − φ). (6)
In the equation above tan φ = λ/m, and mπ is the mass of the Goldstone modes at chemical
potential µ = 0, temperature T = 0, and diquark source λ = 0 [7]. Furthermore, the quark–
antiquark (chiral) condensate is given by

χ̄ χ
= χ̄χ
0 cos α. (7)
At nonzero diquark source there is a crossover between the normal phase and the
diquark condensation phase. However, Chiral Perturbation Theory at leading order predicts
that χ̄ χ
2 + χ T τ2 χ
2 does not depend on the chemical potential, the diquark source, or
the quark mass. This relation does not hold at next-to-leading order in Chiral Perturbation
Theory [8]. However, for small chemical potential, quark mass and diquark source,
Chiral Perturbation Theory at leading order should give an accurate description of these
observables. Therefore, we expect that χ̄ χ
2 + χ T τ2 χ
2 depends slightly on the quark
mass and the diquark source, in the same way as the quark–antiquark condensate depends
on the quark mass in three-color QCD at zero temperature and chemical potential [28].
The behavior of χ̄ χ
2 + χ T τ2 χ
2 is presented in Fig. 12. These graphs confirm what was
expected from Chiral Perturbation Theory: χ̄ χ
2 + χ T τ2 χ
2 is close to constant for small
enough µ and λ. Notice that it is impossible to see the critical chemical potential, which
is around 0.29 as determined in the previous section, by looking at χ̄ χ
2 + χ T τ2 χ
2 .
The points at λ = 0.010 show the effects of finite dt 2 errors. The points which lie higher
were generated in simulations with dt = 0.01 while the 4 lower points were generated in
simulations with dt = 0.005.
We could compare the diquark condensate obtained on the lattice with the leading-order
result (5). But we notice that within Chiral Perturbation Theory at leading order we have
that
χ T τ2 χ
= tan α. (8)
χ̄ χ
This observable can also be used as an order parameter of the diquark condensation phase.
In Chiral Perturbation Theory at leading order, for a given quark mass, diquark source
and chemical potential, the ratio (8) only depends on the value of the pion mass mπ
through the minimization equation (6). It is therefore a very suitable observable to use
in the comparison between the lattice results and Chiral Perturbation Theory.
Hence we fit the lattice result for χ T τ2 χ
/ χ̄ χ
with Chiral Perturbation Theory (8),
(6). This is a one-parameter fit. Since we use Chiral Perturbation Theory at leading order
to analyze our results, we can only take into account data corresponding to chemical
potentials and diquark sources that are small enough so that χ̄χ
2 + χ T τ2 χ
2 is constant
for a given diquark source. The previous figure shows that we have to limit our analysis
to data corresponding to 0.28 µ 0.30, where χ̄χ
2 + χ T τ2 χ
2 is close to constant
for a given diquark source. If we use the 11 points that correspond to λ = 0.004, 0.005,
and 0.006, and 0.28 µ 0.30, we get that µc = 0.3027(1). This is an acceptable fit
since χ 2 /dof = 2.3 (remembering that the norm of the condensate is only approximately
constant over this range.) This one-parameter fit and the data we used to perform it are
shown in Fig. 13(a). If we now use the 18 points that correspond to λ = 0.0025, 0.004,
0.005, 0.006, and 0.008, and 0.28 µ 0.30, we get that µc = 0.3028(1), and the fit
has a χ 2 /dof = 3.1. If we use the 25 points that correspond to λ = 0.0025, 0.004, 0.005,
0.006, 0.008, and 0.010, and 0.28 µ 0.30, we get that µc = 0.3026(2). The fit is worse
than the previous ones with a χ 2 /dof = 12. This one-parameter fit and the data we used
(a)
(b)
Fig. 12. χ̄ χ
2 + χ T τ2 χ
2 for different diquark sources and chemical potentials. The pluses are the lattice data
for λ = 0.025, the × for λ = 0.004, the triangles for λ = 0.005, the squares for λ = 0.006, the diamonds for
λ = 0.008, and the dots for λ = 0.010. (a) Shows that χ̄ χ
2 + χ T τ2 χ
2 is almost constant for 0.28 µ 0.30
and for a given diquark source. (b) Shows the transition region on an expanded scale.
to perform it are shown in Fig. 13(a). Again we note that the 3 points at λ = 0.010 which
lie above the curve are those with dt = 0.01 which have larger dt 2 errors, and are largely
responsible for this large χ 2 .
Noting that the new estimates for µc are significantly higher than our previous estimates,
we consider the inclusion of the ‘data’ at µ = 0.31 and µ = 0.32 in our fits. If we include
the measurements from µ = 0.31 to the 25 points that correspond to λ = 0.0025, 0.004,
0.005, 0.006, 0.008 used in the fits of the previous paragraph, the χ 2 /dof increases from
3.1 to 4.9. Including both the µ = 0.31 and µ = 0.32 measurements further increases the
(a)
(b)
Fig. 13. Comparison between Chiral Perturbation Theory and the lattice results for χ T τ2 χ
/ χ̄ χ
. The pluses
are the lattice data for λ = 0.025, the × for λ = 0.004, the triangles for λ = 0.005, the squares for λ = 0.006, the
diamonds for λ = 0.008, and the dots are for λ = 0.010. The results of two different fits with Chiral Perturbation
Theory and the data we used for these fits are depicted in the solid curves for the different diquark sources: (a) for
λ = 0.004, 0.005, 0.006 only; (b) for all λ.
χ 2 /dof to 5.9. In both cases the new µc is entirely consistent with the estimates of the
previous paragraph.
In summary, we find that Chiral Perturbation Theory at leading order describes the data
well, with a critical chemical potential given by µc = 0.3027(3). This is somewhat higher
than our earlier estimates in the previous section, and might explain why our δ estimates
were less than compelling. It is also understandable in terms of what was learned from the
quenched studies performed in [24]. In Chiral Perturbation Theory, the critical exponents
are given by mean-field theory even at next-to-leading order. From the quality of the fits of
the lattice results with Chiral Perturbation Theory that are presented above, we conclude
that the critical exponents measured on the lattice are consistent with mean field.
We notice that Chiral Perturbation Theory at leading order does not describe the data
very well at large chemical potential and diquark source. This is presumably a sign that
higher order corrections have to be taken into account. The introduction of µ breaks the
original symmetry and allows the condensates to vary independently. This can, in part,
be implemented by replacing the effective potential based on non-linear sigma models,
by a phenomenological effective potential suggested by linear sigma models, where the
condensates are less tightly constrained [24].
5. Scaling function
The static scaling hypothesis states that the data for χ T τ2 χ

/(µ − µc )βmag for different
diquark sources and chemical potentials should collapse onto a single curve for µ > µc
when plotted against λ/(µ − µc )∆ [29]. The gap exponent ∆ = βmag δ; ∆ = 3/2 in mean-
field theory. We can fit the lattice results to the scaling function if we assume some form for
the scaling function. We will try several possibilities for the form of the scaling function:
linear, of the form f (x) = a + bx s , of the form f (x) = a + bx + cx 2/3 and the simplest
mean-field form. The number of parameters changes according to the assumption for the
form of the scaling function. We present our results in Table 4 using part of the data
(µ = 0.297, 0.30, 0.31, 0.32, 0.33, and 0.34). Table 5 shows the corresponding chiral
condensate data.
The different forms assumed for the scaling function give similar results for the
critical chemical potential and for the two critical exponents that we studied. The best
Table 4
Diquark condensates on a 164 lattice at four estimates of the critical µ, 0.2855, 0.2870, 0.2920, and 0.2970 in
the second, third, fourth, and fifth column, respectively. The first column records the λ values
λ χ T τ2 χ
χ T τ2 χ
χ T τ2 χ
χ T τ2 χ
0.002 0.1239(6) 0.1315(4) 0.1678(8) 0.2194(14)

0.004 0.1678(4) 0.1732(3) 0.1890(5) 0.2034(7)
0.006 0.2031(4) 0.2073(2) 0.2190(4) 0.2308(5)
0.008 0.2303(4) 0.2338(2) 0.2440(4) 0.2566(4)
0.010 0.2526(3) 0.2579(2) 0.2662(3) 0.2753(4)
Table 5
Quark–antiquark condensates on a 164 lattice at four estimates of the critical µ, 0.2855, 0.2870, 0.2920, and
0.2970 in the second, third, fourth, and fifth column, respectively. The first column records the λ values
λ χ̄ χ
χ̄ χ
χ̄ χ
χ̄ χ
0.002 0.4244(9) 0.4261(6) 0.4362(9) 0.452(2)

0.004 0.3882(6) 0.3875(4) 0.3836(5) 0.3780(9)
0.006 0.3730(4) 0.3726(3) 0.3668(4) 0.3619(5)
0.008 0.3622(4) 0.3611(3) 0.3566(4) 0.3511(4)
0.010 0.3525(4) 0.3528(2) 0.3464(3) 0.3415(4)
(a)
(b)
Fig. 14. Scaling function and 95%-confidence ellipse for the critical exponents βmag and δ from our best fit,
which corresponds to µc = 0.2948, βmag = 0.58, and δ = 2.28. The pluses are the lattice data for λ = 0.025, the
× for λ = 0.004, the triangles for λ = 0.005, the squares for λ = 0.006, the diamonds for λ = 0.008, and the dots
for λ = 0.010.
fit corresponds to a critical chemical potential that is consistent with the result we found
in the previous section, µc = 0.2948(7), a value of βmag = 0.58(4) which is consistent
with mean field, and a value of δ = 2.28(3) which is about three quarters of the mean-field
result. None of these fits are very good: the best fit has a χ 2 /dof = 17. It corresponds to
the scaling function of the form f (x) = a + bx + cx 2/3 , as presented in Table 6. In Fig. 14,
we show the scaling function together with the 95%-confidence ellipse for the parameters
βmag and δ using the best fit.
Table 6
Determination of the critical chemical potential and two critical exponents using the static scaling hypothesis
assuming different forms for the scaling function f (x) and using the data for diquark sources 0.0025 λ 0.010,
and 0.297 µ 0.340, i.e., 19 data points. The numbers of parameters and a measure of the fit quality are given
for each fit
Form of scaling function Number of parameters µc βmag δ χ 2 /dof
a + bx 5 0.278(7) 0.5(1) 1.1(2) 86
a + bx s 6 0.285(7) 0.51(8) 1.6(4) 74
a + bx + cx 2/3 6 0.2948(7) 0.58(4) 2.28(3) 17
The quality of these fits is rather poor. Furthermore, we find that there is a sizeable
correlation between the different fit parameters: µc is strongly anti-correlated with βmag
and mildly with δ, whereas βmag and δ are mildly correlated with each other. Therefore,
the results from the fits using the scaling function are somewhat suspect. This can be
readily seen by using the same data as above plus the four points that correspond to
µ = 0.2920. The best fit we get has a χ 2 /dof = 66. It gives µc = 0.269(8), βmag = 0.7(1),
and δ = 1.4(2). Notice that µc is much lower than before, that the increase in βmag is
sizeable, and that δ has changed by almost 40%. This result illustrates the problems usually
encountered in the use of the scaling function. Since the scaling function is in general
different above and below the critical point, the choice of data is crucial, and it is difficult
to get a clear result from this type of analysis. The comparison with Chiral Perturbation
Theory presented in the previous section does not suffer from these problems.
Finally, if we impose mean-field exponents, using the same part of the data as above
(µ = 0.297, 0.30, 0.31, 0.32, 0.33, and 0.34), the best fit we get has poor quality:
χ 2 /dof = 243. It gives µc = 0.297(8), which is compatible with the results obtained in
the previous section where we compared the data with Chiral Perturbation Theory. If we
try to use scaling functions derived from the mean-field equation of state as in [24], the
quality of the fit is marginally better, with a χ 2 /dof = 205, and it gives µc = 0.308(2).
6. Runs at finite temperature
We now present our 123 × 6 simulations to investigate the interior of the phase diagram
of Fig. 1. We are particularly interested in scanning the diagram starting at µ = 0 and
following phase transitions and lines of crossover phenomena at high T out to large µ.
In particular, the low mass four flavor theory should have a first order transition on the
µ = 0 axis and this transition should penetrate into the phase diagram. In addition, the
T = 0 transition to a diquark condensate at µc should penetrate into the phase diagram and
separate the normal phase from the phase with a diquark condensate as shown in Fig. 1. It
would be interesting to understand if and how these two lines of transitions merge inside the
phase diagram. Finally, we know from past simulations and effective Lagrangian analyses
that there is a line of first order transitions at high µ and high T which separates the diquark
condensate phase from the quark–gluon plasma phase [12]. We want to confirm the first
order character of the high µ part of this line and to find the low µ end of this line where it
changes to second order at a tricritical point, presumably, labeled 2 in Fig. 1. This point has
Fig. 15. Wilson line and condensates vs. β for µ = 0.00 on 123 × 6 lattice; m = 0.05; λ = 0.005.
also been found within chiral perturbation at nonzero temperature and chemical potential
[15].
We are hopeful that some of these features of the phase diagram will also occur in other
models, so it would be useful to confirm them in this relatively simple setting. Perhaps,
QCD at nonzero baryon number chemical potential has a T –µ phase diagram with some
of these features.
We begin by considering the µ = 0 axis. Our simulations are at nonzero fermion
mass
√ m = 0.05 and nonzero diquark source λ = 0.005 (equivalent to running at m =
0.052 + 0.0052 ). m = 0.05 is large enough that we expect the transition between
hadronic matter and the quark–gluon plasma to be weakened and perhaps softened into
a crossover. This is, in fact, what we find. (At lower quark mass we would expect to find
a first order transition.) In Fig. 15 we show data for the chiral condensate and the Wilson
line and confirm a smooth but rapid crossover in the vicinity of β = 1.9.
Measurements at µ = 0.10 gave similar conclusions: there is a clear but smooth
crossover in the vicinity of β = 1.9, as shown in Fig. 16.
It is interesting that a plot of the same quantities at µ = 0.20 displays a noticeably
sharper crossover. Fig. 17 suggests that µ = 0.20 is near a critical point, such as the point 1
in our generic phase diagram, Fig. 1. At this rather large quark mass m = 0.05, this appears
merely to be a more pronounced crossover but, perhaps, if m were chosen smaller than
0.05, an actual line of first order transitions would be found for the region corresponding
to µ > 0.20 in this phase diagram.
Once we increase µ to 0.25, the induced diquark condensate becomes more significant
at low T , as shown in Fig. 18.
At µ = 0.30 the diquark condensate becomes the most interesting and rapidly varying
quantity. The chiral condensate is relatively smooth and changes from 0.493(6) at β = 1.7
to 0.217(1) at β = 1.9 while the diquark condensate has varied from 0.352(5) to 0.0400(3)
over the same range of coupling. The Wilson line varies from 0.057(3) to 0.241(2) over
the same range. As shown in Fig. 19, the diquark condensate experiences a sharp transition
between β = 1.80 and 1.90.
That transition becomes even sharper as we increase µ to 0.35 as shown in Fig. 20.
Fig. 18. Wilson line and condensates vs. β on 123 × 6 lattice; µ = 0.25; m = 0.05; λ = 0.005.
At µ = 0.40, the simulations indicate a first order transition near β = 1.85(3). In Fig. 21
we show both the diquark condensates and the Wilson line and see strong suggestions of
discontinuities.
The really solid evidence for a first order transition comes from the time evolution of the
observables at β = 1.87 which show signs of metastability. For example, in the Fig. 22 we
Fig. 20. Wilson line and condensates vs. β on 123 × 6 lattice; µ = 0.35 m = 0.05; λ = 0.005.
show the time evolution of the diquark condensate and display several tunnelings between
a state having a condensate near 0.15 and another with a condensate near 0.40.
Runs at µ = 0.50 and 0.60 show clear discontinuities in both the diquark condensate
and the Wilson line near β = 1.87. In Fig. 23 we show the results for µ = 0.60. Note
that this is a slightly weaker coupling than that found at µ = 0.40, where βc ≈ 1.85. This
Fig. 22. Diquark condensate vs. computer time; β = 1.87; µ = 0.40; m = 0.05; λ = 0.005.
suggests that increasing the chemical potential µ increases the transition temperature very
slightly. The effect is in the expected direction, but is very small. It could be studied by
more precise simulations on larger lattices for confirmation.
Finally, we scanned the phase diagram at low T and variable µ. We accumulated data
at β = 1.30, relatively strong coupling and near vanishing T , on the 123 × 6 lattice and
Fig. 24. Diquark condensate vs. µ on 123 × 6 lattice; β = 1.3; m = 0.05; λ = 0.005.
took measurements over a wide range of µ, from 0.15 to 0.60. The results for the diquark
condensate are shown in the Fig. 24 which shows a transition near µ = 0.25. Since β = 1.3
is relatively strong coupling, outside the theory’s continuum scaling window, the value of
µc and βmag found here are subject to large lattice discretization effects and should not be
taken as indicative of continuum physical values. The power law fit shown in the figure
produces a critical index βmag = 0.29(4) at a critical chemical potential µc = 0.2453(2).
The quality of the fit, which extends from µ = 0.26 to 0.38, is rather poor, having a
confidence level of only a fraction of one percent. As we have seen in quenched simulations
and in simulations of QCD at finite chemical potential for isospin such small estimates of
the critical index are often an indication that the scaling is best described in terms of the
scaling form for the effective Lagrangian in the non-linear sigma-model class [24,27,30].
The estimate for the critical chemical potential, µc = 0.2453(2), meshes well with our
measurements at high T above, which indicated that the diquark condensate only started
to be numerically significant for µ 0.25.
The continuous phase transition is also seen in the induced fermion-number density
shown in the Fig. 25. The dotted line fit to that data, which extends all the way from µ =
0.26 to 0.56, has a confidence level of 42 percent, a critical index 0.99(2) and an estimate
for the critical chemical potential, µc = 0.2282(2). The scaling law obtained here is in
good agreement with field theory expectations of unity. Note that such linearity for these
relatively strong couplings has a wider range of validity than would be indicated by the
scaling window, as is seen in the explicit scaling forms from effective Lagrangians [6,7].
Since we only have a single λ value, we have not attempted a fit to such a form.
Fig. 25. Fermion-number density vs. µ on 123 × 6 lattice; β = 1.3; m = 0.05; λ = 0.005.
7. Conclusion and outlook
In this article we have presented a study of the phase diagram of 2-color QCD at nonzero
baryon/quark-number chemical potential and nonzero temperature. We have found two
phases: a “normal” phase where quark number is conserved and a phase with a diquark
condensate which spontaneously breaks quark number. At µ = 0, for smaller quark masses,
there should be a first order finite temperature “deconfinement” transition which divides the
hadronic from the quark–gluon plasma phase. Since, even in this case we expect the line of
first order transitions emerging from this point to terminate at small µ, beyond which there
should be a rapid crossover but no transition, the hadronic and quark–gluon plasma phases
are not distinct. This is the phase we call the “normal” phase. The chiral condensate is non-
zero everywhere. However, where it is large the system is more hadronic in nature; where
it is small, the system is more plasma-like. In the “diquark” phase, the chiral condensate
decreases with increasing diquark condensate, approaching zero for large µ (even before
saturation sets in). We have found that the phase transition between the “normal” phase and
the diquark phase for small T is second order and compatible with mean field theory. These
simulations therefore agree with the predictions of Chiral Perturbation Theory through
next-to-leading order [8].
We have also found that the second order phase transition between the “normal” phase
and the diquark phase becomes first order when µ and T are increased. Therefore, it is
plausible that the µ–T phase diagram contains a tricritical point, although we have not
done precise enough simulations to confirm is existence explicitly. We inferred from crude
measurements that the tricritical point should be at a chemical potential near two-thirds
of the pion mass, and a temperature only slightly lower than the transition temperature
between the hadronic phase and quark–gluon plasma phase at µ = 0. A tricritical point
has also been found in Chiral Perturbation Theory at nonzero temperature and chemical
potential [15]. Therefore, this phase diagram is consistent with the one obtained within
Chiral Perturbation Theory [15].
Now that the phase diagram of this theory has been mapped out, we can turn to
more quantitative properties. In particular, we plan on analyzing the gauge configurations
for their topological content. There are interesting predictions [31] based on effective
Lagrangians for the size, density and interactions among the instantons of the model at
large µ which can be investigated by cooling methods which have been very successful
at vanishing µ. In addition, we can simulate the phase diagram at low µ and high T with
lighter quarks and search for the critical point 1. The SU(3) analog of this point is thought
to be accessible to heavy ion collisions planned for BNL’s RHIC.
Furthermore, the spectroscopy of the model, with an emphasis on Goldstone and
pseudo-Goldstone bosons, is also under consideration. Since the theory’s light modes
control its critical behavior and thermodynamics, quantitative results on spectroscopy are
quite important.
There are also interesting effective Lagrangian predictions for QCD with chemical
potentials associated with the light quarks [32–34]. We are studying QCD with a
chemical potential associated with isospin [15,27]. This model’s phases at nonzero µ and
temperature T are expected to be very similar to those discussed here [32]. Although we
cannot attack the SU(3) theory with a large baryon number chemical potential, these other
situations can be studied both analytically and numerically, and interesting new phases of
matter have been found there which should be investigated further.
Acknowledgements
This work was partially supported by NSF under grant NSF-PHY-0102409 and by the
US Department of Energy under contract W-31-109-ENG-38. D.T. is supported in part
by “Holderbank”-Stiftung. The simulations were done at NPACI and NERSC. B. Klein,
K. Splittorff, M.A. Stephanov and J. Verbaarschot are acknowledged for useful discus-
sions.
References
[1] R. Rapp, T. Schafer, E.V. Shuryak, M. Velkovsky, Phys. Rev. Lett. 81 (1998) 53.
[2] M. Alford, K. Rajagopal, F. Wilczek, Phys. Lett. B 422 (1998) 247.
[3] D. Bailin, A. Love, Phys. Rep. 107 (1984) 325.
[4] B.C. Barrois, Nucl. Phys. B 219 (1977) 390.
[5] S. Hands, J.B. Kogut, M.-P. Lombardo, S. Morrison, Nucl. Phys. B 558 (1999) 327.
[6] J.B. Kogut, M.A. Stephanov, D. Toublan, Phys. Lett. B 464 (1999) 183.
[7] J.B. Kogut, M.A. Stephanov, D. Toublan, J.J. Verbaarschot, A. Zhitnitsky, Nucl. Phys. B 582 (2000) 477.
[8] K. Splittorff, D. Toublan, J.J.M. Verbaarschot, Nucl. Phys. B 620 (2002) 290.
[9] K. Splittorff, D.T. Son, M.A. Stephanov, Phys. Rev. D 64 (2001) 016003.
[10] B. Vanderheyden, A.D. Jackson, Phys. Rev. D 64 (2001) 074016.
[11] S. Hands, J.B. Kogut, S.E. Morrison, D.K. Sinclair, Nucl. Phys. Proc. Suppl. 94 (2001) 457;
J.B. Kogut, D.K. Sinclair, S. Hands, S.E. Morrison, Phys. Rev. D 64 (2001) 094505.
[12] J.B. Kogut, D. Toublan, D.K. Sinclair, Phys. Lett. B 514 (2001) 77.
[13] A. Nakamura, Phys. Lett. B 149 (1984) 391;
S. Muroya, A. Nakamura, C. Nonaka, nucl-th/0111082;
Y. Liu et al., hep-lat/0009009;
See also: M.P. Lombardo, hep-lat/9907025.
[14] J.B. Kogut and D.K. Sinclair, in preparation.
[15] K. Splittorff, D. Toublan, J.J.M. Verbaarschot, hep-ph/0204076.
[16] B. Klein, D. Toublan, J.J.M. Verbaarschot, in preparation.
[17] R.D. Pisarski, F. Wilczek, Phys. Rev. D 29 (1984) 338.
[18] J. Wirstam, Phys. Rev. D 62 (2000) 045012.
[19] J.B. Kogut, Nucl. Phys. B 290 [FS20] (1987) 1.
[20] F. Karsch, AIP Conf. Proc. 602 (2001) 323.
[21] Z. Fodor, S.D. Katz, Nucl. Phys. Proc. Suppl. 106 (2002) 441.
[22] I.M. Barbour, A.J. Bell, Nucl. Phys. B 372 (1992) 385.
[23] R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, Phys. Lett. B 493 (2000) 189;
R. Aloisio, V. Azcoiti, G. Di Carlo, A. Galante, A.F. Grillo, Nucl. Phys. B 606 (2001) 322.
[24] J.B. Kogut, D.K. Sinclair, hep-lat/0201017.
[25] J.B. Kogut, H. Matsuoka, S.H. Shenker, J. Shigemitsu, D.K. Sinclair, M. Stone, H.W. Wyld, Nucl. Phys.
B 225[FS9] (1983) 93.
[26] S. Duane, J.B. Kogut, Phys. Rev. Lett. 55 (1985) 2774;
S. Gottlieb, W. Liu, D. Toussaint, R.L. Renken, R.L. Sugar, Phys. Rev. D 35 (1987) 2531.
[27] J.B. Kogut, D.K. Sinclair, hep-lat/0202028.
[28] J. Gasser, H. Leutwyler, Ann. Phys. 158 (1984) 142;
J. Gasser, H. Leutwyler, Nucl. Phys. B 250 (1985) 465.
[29] N. Goldenfeld, Lectures On Phase Transitions And The Renormalization Group, in: Frontiers in Physics,
Vol. 85, Addison–Wesley, Reading, MA, 1992.
[30] J.B. Kogut, M.P. Lombardo, D.K. Sinclair, Phys. Rev. D 51 (1995) 1282;
M.P. Lombardo, J.B. Kogut, D.K. Sinclair, Phys. Rev. D 54 (1996) 2303.
[31] D.T. Son, M.A. Stephanov, A.R. Zhitnitsky, Phys. Lett. B 510 (2001) 167.
[32] D.T. Son, M.A. Stephanov, Phys. Rev. Lett. 86 (2001) 592.
[33] J.B. Kogut, D. Toublan, Phys. Rev. D 64 (2001) 034007.
[34] T. Schafer, D.T. Son, M.A. Stephanov, D. Toublan, J.J.M. Verbaarschot, Phys. Lett. B 522 (2001) 67.
Matrix models in homogeneous spaces

Yoshihisa Kitazawa
Institute of Particle and Nuclear Studies, High Energy Accelerator Research Organization (KEK), Tsukuba,
Ibaraki 305-0801, Japan
Received 30 July 2002; accepted 7 August 2002
Abstract
We investigate noncommutative gauge theories in homogeneous spaces G/H . We construct such
theories by adding cubic terms to IIB matrix model which contain the structure constants of G.
The isometry of a homogeneous space, G must be a subgroup of SO(10) in our construction. We
investigate CP 2 = SU(3)/U (2) case in detail which gives rise to 4-dimensional noncommutative
gauge theory. We show that noncommutative gauge theory on R 4 can be realized in the large-N limit
by letting the action approach IIB matrix model in a definite way. We discuss possible relevances of
these theories to the large-N limit of IIB matrix model.
1. Introduction
Matrix models which are dimensional reductions of 10-dimensional super-Yang–Mills

theory first appeared as low energy effective theories of N coincident D-branes. They may
also be regarded as matrix regularization of light-cone membrane action [1] or Green–
Schwarz superstring action [2]. Our hope is that they may provide us second quantized
string theory in the large-N limit. It is because they may contain arbitrary numbers of
strings represented as block-diagonal matrices.
Noncommutative D-branes such as the following are formal classical solutions of matrix
models in the large-N limit
1 = p̂,
Acl 2 = q̂,
Acl
[p̂, q̂] = −i, (1.1)
E-mail address: kitazawa@post.kek.jp (Y. Kitazawa).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 8 2 - X
Y. Kitazawa / Nuclear Physics B 642 (2002) 210–226 211
since they solve the equation of motion

[Aµ , [Aµ , Aν ]] = 0. (1.2)
Noncommutative (NC) gauge theory is obtained from matrix models around NC space-
time [3–5]. The advantage of matrix model construction of NC gauge theory is that it
maintains the manifest gauge invariance under U (N) transformations
Aµ → U Aµ U † . (1.3)
The gauge invariant observables of NC gauge theory, the Wilson lines were constructed
through matrix models [6,7]. NC gauge theory exhibits UV–IR mixing which is a charac-
teristic feature of string theory [8].
In this paper we investigate NC gauge theory on homogeneous spaces through matrix
models. It is an interesting problem on its own to study NC gauge theories on curved
manifolds. A homogeneous space is realized as G/H where G is a Lie group and H is
a closed subgroup of G. We further assume that they are symmetric spaces which are
invariant under space reflection (parity). At Lie algebra level in symmetric spaces, the
following commutation relations hold
g = h + m,
[h, h] ⊂ h, [h, m] ⊂ m, [m, m] ⊂ h, (1.4)
where g and h are the generators of G and H , respectively.
We construct NC gauge theories on homogeneous spaces by adding cubic couplings to
IIB matrix model with a large but finite N . The isometry of a homegeneous space G/H
is the group G. The group G has to be a subgroup of SO(10) which is the symmetry of
IIB matrix model. We investigate CP 2 = SU(3)/U (2) case in detail which gives rise to
4-dimensional NC gauge theory. We show that NC gauge theory on R 4 can be realized
in the large-N limit by letting the action approach IIB matrix model in a definite way.
Although SUSY is broken in general with finite N , we argue that SUSY is locally
resurrected in such a limit. We hope that our construction is useful to investigate 4-
dimensional NC super-Yang–Mills theory nonperturbatively. Since the strength of the
cubic couplings formally vanish in the large-N limit, our results may be relevant to
investigate the large-N limit of IIB matrix model.
In Section 2, we construct fuzzy homogeneous spaces G/H as the orbit of a state in
a definite representation of dimension N which is invariant under H modulo the overall
U (1) phase. We then construct NC gauge fields as bi-local states. Such a construction
facilitates matrix model realizations since gauge fields are naturally represented by N × N
Hermitian matrices. In Sections 3, we construct NC gauge theories on homogeneous
spaces G/H through matrix models. For this purpose we deform IIB matrix model
by adding cubic terms in Aµ which contain structure constants of G. We investigate
CP 2 = SU(3)/U (2) case in detail which gives rise to 4-dimensional NC gauge theory.
We show that NC gauge theories on R 4 can be realized in the large-N limit by letting
the action approach IIB matrix model in a definite way. We investigate quantum effects of
4-dimensional NC gauge theory and its formal infrared limit. We conclude in Section 4
with discussions. We discuss possible relevances of these theories to the large-N limit of
IIB matrix model.
212 Y. Kitazawa / Nuclear Physics B 642 (2002) 210–226
2. Non-commutative spacetime
In this section, we formulate a generic procedure to construct fuzzy homogeneous

spaces G/H and gauge fields on them. We pick a state |0 in a definite representation
of G which is invariant under H . In this construction, we identify the states which only
differ by their U (1) phases since they are equivalent as quantum states. The set of all
states which can be reached by multiplying elements of G to |0 is called the orbit of |0.
Fuzzy homogeneous spaces G/H are constructed as the orbit of |0. It is represented
by the irreducible representation which is descended from |0. The dimension of the
representation N can be interpreted as the volume of fuzzy G/H in the unit of non-
commutativity scale. We recall that the basic degrees of freedom in NC gauge theory are
bi-local fields which can be interpreted as zeromodes of open strings [9]. We then construct
NC gauge fields by forming the tensor product of the relevant irreducible representation
and its complex conjugate. Such a construction facilitates matrix model realizations since
gauge fields are naturally represented by N × N Hermitian matrices.
We first recall a fuzzy flat manifold R 2 . We have noncommutative coordinates x̂, ŷ
which satisfy the canonical commutation relation
[x̂, ŷ] = i. (2.1)

Obviously we cannot realize such a commutation relation with finite size matrices.1 We
can construct the creation and annihilation operators â, â † out of them:
1 1
a = √ (x̂ + i ŷ), a † = √ (x̂ − i ŷ),
2 2

â, â = 1.
†
(2.2)
A localized state |0 at the origin can be defined as
â|0 = 0. (2.3)
We can construct surrounding states by applying the creation operators to |0 as
2
â † |0, â † |0, . . . . (2.4)
The states which can be constructed from |0 by applying â † finite times are localized
around the origin. They are uniformly distributed with the density 1/2π in accordance
with the semiclassical quantization condition.
The conjugate momentum operators are identified with
p̂x = ŷ, p̂y = −x̂, (2.5)

since they satisfy the canonical commutation relationship with x̂, ŷ. They generate the
translations along fuzzy plane. We subsequently introduce the adjoint operators Pµ =
[p̂µ , ]. Since Px and Py commute, we can simultaneously diagonalize them. The
1 This commutation relation is realized by the guiding center coordinates of electrons in magnetic field. In fact
NC gauge theory may be realized in quantum Hall system [10].
eigenstates can be constructed as

Pµ exp(ikx x̂ + iky ŷ) ≡ p̂µ , exp(ikx x̂ + iky ŷ)
= kµ exp(ikx x̂ + iky ŷ). (2.6)
They are the analogous of the plane waves in noncommutative space. They can also be
interpreted as the bi-local fields or dipoles with the length |k|.
We next review S 2 = SU(2)/U (1) case as the first example of a curved manifold. In
SU(2), we have the Hermitian operators jˆx , jˆy , jˆz which satisfy the commutation relations
of the angular momentum
[jˆx , jˆy ] = i jˆz . (2.7)

Contrary to R 2 case, such a commutation relation can be realized with finite size matrices
since S 2 is compact. The raising and lowering operators jˆ+ , jˆ− can be formed from jˆx , jˆy
which satisfy
1
jˆ± = √ (jˆx ± i jˆy ),
2
+ −
ˆ ˆ
j , j = jˆz ,

jˆz , jˆ± = ±jˆ± . (2.8)
We consider the (N = 2l + 1)-dimensional representation of spin l. Since we can
diagonalize jˆz , we can label the states with the eigenvalue of jˆz as |m where
jˆz |m = m|m. (2.9)

We consider the semiclassical limit where l is assumed to be large. In such a situation we
expect to recover a large smooth sphere. We can further expect that a flat fuzzy plane is
realized in the neighborhood of a particular point such as the north-pole. In fact the notion
of locality can be introduced in the large-l limit as follows.
As a localized states at the north-pole, we can choose |l. We can subsequently construct
other states by using the lowering operators
jˆ− |l, (jˆ− )2 |l, . . . . (2.10)

Since we have assumed that l is large, these states which are constructed from |l with
finite operations of jˆ− are all localized around the north-pole.
For these states, we can approximate jˆz ∼ l. By rescaling the operators, we can obtain
the following commutation relations from (2.8),
1 1 1
ã = √ j + , ã † = √ j − , 1̃ = jˆz ,
l l l
1 1
ã, ã † = 1̃, [1̃, ã] = ã, 1̃, ã † = − ã † . (2.11)
l l
Since we obtain the identical algebra with (2.2), we conclude that flat fuzzy plane is
realized around the north-pole of S 2 in the large-l limit. The density of the states is 1/2πl
in the original coordinate and the area of S 2 is 4πl 2 . Thus the total number of the states is
2l semiclassically which is consistent with the spin l representation of SU(2).
The local coordinates which satisfy the canonical commutation relationship can be
identified as
1 1
x̃ = √ jˆx , ỹ = √ jˆy . (2.12)
l l
The conjugate momentum operators are
1 1
p̃x = √ jˆy , p̃y = − √ jˆx . (2.13)
l l
In the local patch, the eigenfunctions of the adjoint Pα = [p̃α , ] can be constructed just
like the flat space case as exp(ikx x̃ + iky ỹ). We point out here that adjoint operators
of Jµ = [jˆµ , ] also satisfy the commutation relation of SU(2). Using the commutation
relations in (2.11) which are realized around the north-pole, we find that P2 can be related
to the Casimir operator of SU(2) in a local patch as follows

J2 1
exp(ikx x̃ + iky ỹ) = P exp(ikx x̃ + iky ỹ) + O
2
. (2.14)
l l
The homogeneous space S 2 is generated by jˆx and jˆy starting from the state |l. The state
|l only changes its U (1) phase under the rotation around the z axis. As quantum states,
we may identify the states which differ only by their U (1) phases. Therefore fuzzy S 2 can
also be realized as G/H where G = SU(2) and H = U (1).
We can parameterize CP 1 or S 2 by two complex coordinates uα such that
u∗α uα = 1. (2.15)
Fuzzy S2 can be represented by the following states with spin l
uα1 · · · uα2l . (2.16)
We construct gauge fields on fuzzy S 2 as bi-local fields. Hence the bi-local fields are
represented as
uα1 · · · uα2l u∗β1 · · · u∗β2l . (2.17)
They can be decomposed into the irreducible representations with spin n

2l

uα1 · · · uαn u∗β1 · · · u∗βn (2.18)
n=0
which are traceless under the contractions of α and β indices.
The adjoint generators of SU(2) transformations are represented as

1 ∂ ∂
J i = − uβ σαβ i
− u∗α σαβ
i
(2.19)
2 ∂uα ∂u∗β
which satisfy the SU(2) algebra
[Ji , Jj ] = i'ij k Jk . (2.20)
The Casimir operator acts on the bi-local fields as

1 ∂ 2 1 ∂
J =
2
uα + uα + c.c. = n2 + n. (2.21)
2 ∂uα 2 ∂uα
Therefore
n=2l the eigenvalues of the Laplacian on S 2 , J 2 / l are n(n + 1)/ l with n integers.
Since n=0 (2n + 1) = (2l + 1)2 , a group of representations with spins up to 2l form the
complete basis of N × N Hermitian matrices as it is evident from our construction.
After reviewing the well-known example of fuzzy S 2 case, we can formulate a general
procedure to construct fuzzy homogeneous space G/H as follows. We first consider
a representation of G which contains a (highest weight) state which is invariant under H
modulo the over all U (1) phase. We also require that fuzzy R n where n is the dimension of
G/H is realized in the local patch around such a state. We are thus restricted to symplectic
manifolds. It is because the ( product on such a manifold can be reduced to that of a flat
manifold (Moyal product) locally by choosing the Darboux coordinates. Since the Kähler
form serves as the symplectic form, Kähler manifolds such as CP n satisfy this requirement
[11].
We embed the Lie generators of G into N -dimensional Hermitian matrices where N
is the dimension of the representation. The gauge fields on fuzzy G/H are constructed
as bi-local fields. The bi-local fields are the tensor products of the relevant representation
and the complex conjugate of it. Since they are reducible, we can decompose them into
the irreducible representations. They are guaranteed to form the complete basis of N × N
Hermitian matrices by construction. In what follows we discuss several concrete examples
in higher dimensions following this general procedure.
Our main interest in this paper is CP 2 = SU(3)/U (2) [12]. SU(3) is generated by 8
Hermitian operators t a which satisfy
a b
t , t = if abc t c . (2.22)
The structure constants of SU(3) are
f123 = 1,
1
f147 = f246 = f257 = f345 = f516 = f637 = ,
2
√
3
f458 = f678 = . (2.23)
2
The irreducible representations of SU(3) can be classified by the Young tableaux with
a pair of integers (p, q) which specifies the numbers of boxes in the first and the second
low. We consider the representation (p, 0) which can be realized by totally symmetrizing
the following states
uα1 uα2 · · · uαp , (2.24)

3
where α=1 u∗α uα = 1. Let us consider the state uαi = δαi ,3 for all i. Such a state |p is
of SU(3) which are generated by t 1 , t 2 , t 3 and the eigenstate
a singlet of SU(2) subgroup√
8
of t with the eigenvalue p/ 3.
The commutation relations among t 4 · · · t 7 are

√
4 5 3 8 1
t ,t = i t + i t 3,
2 2
√
6 7 3 8 1
t ,t = i t − i t 3. (2.25)
2 2
We may introduce the raising and lowering operators:
1 1
u± = √ t 6 ± it 7 , v ± = √ t 4 ± it 5 . (2.26)
2 2
As S 2 case, we can construct descendent states by applying lowering operators u− , v − to
|p.
|p, u− |p, v − |p, (u− )2 |p, (v − )2 |p, u− v − |p, . . . . (2.27)
In the large-p limit, the states which are obtained from |p with finite actions of the
lowering operators form a √localized√patch.
After rescaling u, v → p/2 ũ, p/2 ṽ we obtain the following commutation relation
which are realized in such a local patch,
+ − √
ũ , ũ = 1 + O(1 p ),
+ − √
ṽ , ṽ = 1 + O(1 p ). (2.28)
We thus conclude that flat fuzzy R 4 is realized in a local patch of CP 2 in the large-p limit.
The density of the states is 1/π 2 p2 in the original coordinate and the volume of CP 2 is
π 2 p4 /2. Thus the total number of the states is p2 /2 semiclassically which is consistent
with the (p, 0) representation of SU(3).
The local coordinates which satisfy the canonical commutation relationship can be
identified as
√ √
2 4 2
x̃ = √ t , ỹ = √ t 5 , [x̃, ỹ] = i,
p p
√ √
2 6 2
w̃ = √ t , z̃ = √ t 7 , [w̃, z̃] = i. (2.29)
p p
We also find
1 i i i 1 i
t , x̃ = z̃, t 1 , ỹ = − w̃, t 1 , w̃ = ỹ, t , z̃ = − x̃,
2 2 2 2
2 i 2 i 2 i 2 i
t , x̃ = w̃, t , ỹ = z̃, t , w̃ = − x̃, t , z̃ = − ỹ,
2 2 2 2
3 i 3 i 3 i 3 i
t , x̃ = ỹ, t , ỹ = − x̃, t , w̃ = − z̃, t , z̃ = w̃. (2.30)
2 2 2 2
We may thus identify the SU(2) subgroup formed by t 1 , t 2 , t 3 as a subgroup of SO(4) as
−i 14 −i 13 −i 12
t1 = L − L23 , t2 = L + L24 , t3 = L − L34 . (2.31)
2 2 2
The conjugate momentum operators are
p̃x = ỹ, p̃y = −x̃,

p̃w = z̃, p̃z = −w̃. (2.32)
In the flat space limit, the eigenfunctions of the adjoint operators Pα = [p̃α , ] can be
constructed as exp(ik · x) ≡ exp(ikx x̃ + iky ỹ + ikw w̃ + ikz z̃). We introduce here the adjoint
operators T a = [t a , ] which also satisfy the commutation relation of SU(3). Using the
commutation relations in (2.28) which are realized around the north-pole, we find that Pα2
can be related to the Casimir operator of SU(3) in a local patch

2 a 2 1
T exp(ik · x) = Pα exp(ik · x) + O
2
. (2.33)
p p
Thus the Laplacian on CP 2 which reduces to that of flat R 4 in a local patch in the
large-N limit is 2(T a )2 /p. The eigenvalues 2(T a )2 /p are 2n(n + 2)/p for the representa-
tion (n, n).
It is because the bi-local states are represented as
uα1 · · · uαp u∗β1 · · · u∗βp . (2.34)

They can be decomposed into the irreducible representations of (n, n) type

p

uα1 · · · uαn u∗β1 · · · u∗βn (2.35)
n=0
which are traceless under the contractions of any pairs of α and β indices. The adjoint
generators of SU(3) transformations are represented as

a ∂ a ∂
T a = − uβ tαβ − u∗α tαβ (2.36)
∂uα ∂u∗β
which satisfy the SU(3) algebra. The Casimir operator acts on the bi-local fields as

1 ∂ 2 ∂
T =2
uα + uα + c.c. = n2 + 2n. (2.37)
2 ∂uα ∂uα
p
The dimension of the representation (n, n) is (n+1)3 . Since n=0 (n+1)3 = (p +1)2 (p +
2)2 /4, a group of the irreducible representations (n, n) with n up to p form a complete
basis of N × N Hermitian matrices where N = (p + 1)(p + 2)/2 is the dimension of the
representation (p, 0).
Our final example in this section is CP 3 = SO(5)/U (2) [13–18]. SO(5) are generated
by 10 antisymmetric matrices tµν which satisfy:
[tµν , tρσ ] = iδµρ tνσ − iδνρ tµσ − iδµσ tνρ + iδνσ tµρ . (2.38)
Our investigation is constrained to the homogeneous spaces which can be realized through
IIB matrix model. G is maximal in this case since the number of its generators cannot
exceed 10 which is the number of bosonic matrices Aµ in IIB matrix model. SO(5) contains
a subgroup SO(4) = SU(2) × SU(2) which is generated by

1 1 1
j1 = (t23 + t14 ), j2 = (t31 + t24 ), j3 = (t12 + t34 ),
2 2 2
1 1 1
k1 = (t23 − t14 ), k2 = (t31 − t24 ), k3 = (t12 − t34 ). (2.39)
2 2 2
We can simultaneously diagonalize t 2 , j 2 , k 2 , jz , kz .
The irreducible representations of SO(5) are represented by the Young tableaux with
a pair of integers (m, n)5 with m n. We consider the representation (p/2, p/2)5 which
decomposes into the representations of SO(4) = SU(2) × SU(2) as follows

p p p 1 p−2 p
, = 0, + , + ···+ ,0 . (2.40)
2 2 5 2 4 2 2 4 2 4
As a localized state |p, we consider such a state in (p/2, 0)4 which is invariant under
U (2) modulo U (1) phases:

p p+2 p
j 2 |p = |p, j3 |p = |p,
2 2 2
k 2 |p = 0,
2
p + 4p
t 2 |p = |p. (2.41)
2
In a local patch around |p, the following commutation relations are realized
p
[j1 , j2 ] = ij3 ∼ i ,
2
p
[t15 , t25 ] = it12 ∼ i ,
2
p
[t35 , t45 ] = it34 ∼ i . (2.42)
2
We thus find that fuzzy R 6 is realized in the large-p limit in the local patch just like the
previous examples.
The bi-local fields are obtained by considering the direct product of
(p/2, p/2) ⊗ (p/2, p/2) (2.43)
which can be decomposed into the irreducible representations of SO(5) as

p
m
(m, n). (2.44)
m=0 n=0
Since the dimension of the (m, n) representation is
D(m, n) = (1 + m − n)(1 + 2n)(2 + m + n)(3 + 2m)/6, (2.45)
we can check that the dimension of the bi-local fields agrees with those of N ×N Hermitian
matrices
p m
D(m, n) = D(p/2, p/2)2 . (2.46)
m=0 n=0
3. Matrix model realization
In this section, we construct NC gauge theories on homogeneous spaces G/H through

matrix models. We propose to deform IIB matrix model action as follows generalizing the
S 2 = SU(2)/U (1) case [19]2
i
SIIB → SIIB + αfµνρ Tr[Aµ , Aν ]Aρ , (3.1)
3
where fµνρ is the structure constant of G. Since there are 10 Hermitian matrices Aµ in IIB
matrix model, the number of the Lie generators of G cannot exceed 10 in this construction.
This action does not preserve SUSY unless G = SU(2). However we show that NC gauge
theory on R 4 can be realized in the large-N limit by letting the action approach IIB matrix
model in a definite way. Although SUSY is broken in general with finite N , we argue
that SUSY is locally resurrected in such a limit. Since this model possesses the translation
invariance
A µ → A µ + cµ (3.2)
and also
ψ → ψ + ', (3.3)
we remove these zero-modes by restricting Aµ and ψ to be traceless.
The equation of motion is
[Aµ , [Aµ , Aν ]] + iαfµρν [Aµ , Aρ ] = 0. (3.4)

The nontrivial classical solution is
a = αt , µ = 0,
a
Acl other Acl (3.5)
where t a ’s satisfy the Lie algebra of G. Although diagonal matrices also solve the equation
of motion, the nontrivial solution (3.5) minimizes the classical action. In SU(2) case, it is
evaluated for the irreducible representation of spin l
α4
− l(l + 1)(2l + 1). (3.6)
6
For a large but fixed N , the irreducible representation of spin l where N = 2l + 1 is selected
by minimizing the classical action [22]. There is no quantum corrections to worry about
thanks to SUSY.
Let us investigate the analogous problem in CP 2 = SU(3)/U (2) case. The classical
action for the irreducible representation (p, q) of SU(3) is evaluated as
α4
− C2 (p, q) dim(p, q), (3.7)
4
2 See also [20,21].

where C2 (p, q) is the Casimir and dim(p, q) is the dimension of the representation
1
C2 (p, q) = p(p + 3) + q(q + 3) + pq ,
3
(p + 1)(q + 1)(p + q + 2)
dim(p, q) = . (3.8)
2
We can see that the classical action is minimized for the (p, 0) type representation with
the fixed dim(p, q) = N . We also note that reducible representations do not minimize the
classical action. We conclude that the desired representation of (p, 0) type which is relevant
to fuzzy CP 2 is selected by minimizing the classical action. Therefore this model realizes
U (1) NC gauge theory on CP 2 with finite N .
In the large-N limit, the fuzzy R 4 is realized locally as it has been shown
√ in the previous
section. We expand the action around the classical solution as Aµ = α p/2 (p̂µ + âµ ). In
this parameterization, the noncommutativity scale is fixed to be 1. After using the Moyal–
Weyl correspondence,
â → a(x),
â b̂ → a(x) ( b(x),
2
1
Tr → d 4 x, (3.9)
2π
we obtain the following NC gauge theory from (3.1)
2 2
4 p 1 1 1 1
−α 4
d x [Dα , Dβ ]2 + [Dα , φi ]2 + [φi , φj ]2
2 2π 4 2 4

1 1
+ ψ̄Γα [Dα , ψ] + ψ̄Γi [φi , ψ] . (3.10)
2 2 (
√
The cubic terms are suppressed by 2/p. In this way, we identify the coupling constant of
NC gauge theory as

4π 2
gNC =
2
. (3.11)
pα 2
The analogous relations hold for 2- and 6-dimensional cases, respectively, as

1 2 2 2
gNC2 = 2π
2
, gNC6 = (2π)
2 3
. (3.12)
lα 2 pα 2
The classical action assumes the following value in the large-N limit
2π 2 p2 4π 2 N
− 2
∼− 2
. (3.13)
3gNC 3gNC
We need to choose gNC ∼ 1 to obtain interacting NC gauge theory. For this purpose,
we may choose α such that gNC ∼ 1 for a fixed N . We therefore generally need to
choose α 2 ∼ 1/p. In 4 dimensions, we find N ∼ p2 /2 for the (p, 0) representation which
minimizes the classical action. If we let N large in this way, we find that α vanishes in the
√
large-N limit as α 2 ∼ O( 1/N). From (3.10), we can see that SUSY is locally recovered
in this limit. We have argued that NC gauge theory on R 4 can be obtained by expanding IIB
matrix model around fuzzy R 4 [4]. Our matrix model construction makes such a statement
more precise. Although we can formulate NC gauge theories nonperturbatively through
unitary matrices [23], our construction may be useful for supersymmetric gauge theories.
We move on to investigate quantum theory. After the gauge fixing, the quadratic action
for âµ is simply given by
1
2
(2π)2 Tr(âν Pµ Pµ âν ). (3.14)
2gNC
As we have described in the preceding section, we may identify the eigenstates of Pµ2 =
2T 2 /p with (n, n) representations of SU(3) with the eigenvalues
2n(n + 2)/p. (3.15)
In a local patch, we can adopt the four-dimensional approximation for p̂µ as in (2.32). In
such an approximation, the eigenvalues Pα2 are 2n2 /p. They are uniformly distributed over
the momentum space with the density p2 /8π 2 .
The one loop correction to the classical action is

1 1 1 1 + Γ11
Tr log P 2 δµν − Tr log P 2 − Tr log P 2 + Fµν Γ µν , (3.16)
2 4 4 2
√
where Fµν = i 2/p fµνρ P ρ . It differs from that of IIB matrix model by
1 1
Tr log P 2 δµν − Tr log P 2 δµν − 2iFµν . (3.17)
2 2
In dimensions higher than 2, we can expand this expression into the power series of Fµν .
The leading correction is found to be
1 p
1
−3 Tr = −3 (n + 1)3 ∼ −3N. (3.18)
T 2 n(n + 2)
n=1
We find that the one loop correction is of the same (negative) sign with the classical action.
It is unlike the CP 1 = SU(2)/U (1) case where the quantum corrections vanish due to
SUSY. Nevertheless the quantum correction in CP 2 case is proportional to N and hence
the volume of the manifold. As a catch phrase, one might say that the quantum correction to
the cosmological constant is finite in this model. The contribution to (3.18) is dominated by
the states with large Casimir eigenvalues. Such states represent nonlocal states connecting
the opposite sides of spacetime. In this sense we believe that this quantum correction is
the signature of cosmic scale SUSY breaking and there is no contradiction to the notion of
local recovery of SUSY.
We investigate formal semiclassical (or infrared) limit of NC gauge theory on CP 2 . In
this case, we fix the size of spacetime to be 1. In a semiclassical approximation, we can
identify
2
1 p √
Tr → 2
d 4 x g,
(2π) 2
1 √
Pµ Pµ → − √ ∂a g g ab ∂b . (3.19)
g
We may further identify
Pµ → −iKµa (x)∂a , (3.20)
where the Killing vectors in homogeneous spaces are related to the inverse metric as

Kµa Kµb = g ab . (3.21)
µ
Since Pµ satisfy the Lie algebra of SU(3), we find

Kµa ∂a Kνb − Kνa ∂a Kµb = −fµνρ Kρb . (3.22)
We can also parameterize
âµ → Kµa (x)ba (x) + Nµi (x)φi (x), (3.23)
where Nµi are orthogonal to the Killing vectors

Kµa Nµi = 0. (3.24)
µ
In order that (3.24) is consistent with (3.22), Nµi must satisfy
Kµa ∂a Nνi − Kνa ∂a Nµi = −fµνρ Nρi . (3.25)

We may define the inverse metric in the ‘transverse’ space from Nµi

Nµi Nµj = g ij . (3.26)
µ
From (3.22) and (3.25), we can show that g ij is invariant under the isometry
Kµa ∂a g ij = 0. (3.27)
We can subsequently conclude that the transverse space is flat, namely g ij = δ ij .
Since
Pµ aν − Pν aµ = −iKµa Kνb (∂a bb − ∂b ba ) + ifµνρ Kρa ba

− i Kµa Nνi − Kνa Nµi ∂a φi + ifµνρ Nρi φi , (3.28)
we find

Fµν = −iKµa Kνb (∂a bb − ∂b ba ) − i Kµa Nνi − Kνa Nµi ∂a φi . (3.29)
The bosonic part of the action (3.1) gives the following result in the semiclassical limit

√ 1 ac bd 1
d 4x g g g (∂a bb − ∂b ba )(∂c bd − ∂d bc ) + g ab ∂a φi ∂b φi
4 2

1
− fµνρ Kµa Kνb Nρi (∂a bb − ∂b ba )φi . (3.30)
2
We thus obtain gauge theory on the four-dimensional curved manifold CP 2 .

The fermionic part is
1
Tr ψ̄Γµ [Aµ , ψ]. (3.31)
2
In the semiclassical limit, we obtain

1 √
d 4 x g −i ψ̄γ a ∂a ψ , (3.32)
2
where γ a = Γµ Kµa and γ i = Γµ Nµa . They satisfy the following commutation relations
a b i j a i
γ , γ = g ab , γ , γ = δ ij , γ , γ = 0. (3.33)
We note that the fermionic kinetic term is invariant under SU(3) transformations

√
Tr ψ̄Γµ [p̂µ , ψ] → d 4 x g −i ψ̄Γµ Kµa ∂a ψ . (3.34)
It is because an SU(3) rotation

δ p̂µ = iαfµνρ p̂ν = [p̂ρ , p̂µ ] (3.35)
can be undone by the SO(10) transformation of ψ as
i
δψ = − αfµνρ Γ µν ψ. (3.36)
4
In symmetric homogeneous spaces, we can locally choose a flat metric for g ab with the
vanishing spin connection. Since our fermionic action is valid locally, it must be valid
globally due to SU(3) invariance.
Before concluding this section, we make a brief comment on NC gauge theory on CP 3
case. The classical action is evaluated for the irreducible representation (m, n) of SO(5) as
α4
− C2 (m, n) dim(m, n), (3.37)
2
where
C2 (m, n) = m2 + n2 + 3m + n,
(m + n + 2)(m − n + 1)(3 + 2m)(1 + 2n)
dim(m, n) = . (3.38)
6
It is not minimized by an (m, m) type representation but rather minimized by an (m, 0) type
representation for the fixed N = dim(m, n). Unfortunately the (p/2, p/2) representation
which is relevant to CP 3 as it is explained in Section 2 is metastable in our construction.
The quantum correction around the (p/2, p/2) state is found to be much larger than the
classical action since
1
p
m
1 3
−3 Tr 2
= −3 dim(m, n) ∼ − log(4) − 1 p4 , (3.39)
T C2 (m, n) 8
m=0 n=0
while N ∼ p3 /6. So the one loop correction to the cosmological constant is infinite in this
model.
4. Conclusions and discussions
We have investigated NC gauge theories on fuzzy homogeneous spaces G/H through

matrix models. We have considered deformed IIB matrix models with finite N . In our
construction, the isometry of a homogeneous space, G must be a subgroup of SO(10)
which is the symmetry of IIB matrix model. A local patch and the coordinates can be
introduced semiclassically when N is large. We require that a fuzzy flat hyper plane is
realized in the local patch. Although the Hermitian matrices Aµ have been interpreted as
coordinates in matrix models, we have interpreted them as Killing vectors of spacetime.
We have investigated 4-dimensional NC gauge theory on fuzzy CP 2 in detail. We have
shown that NC gauge theory on R 4 is realized by letting the cubic coupling vanish in the
large-N limit. We have therefore made the relation between matrix models and NC gauge
theory more precise. Our construction may be useful for nonperturbative investigations
of supersymmetric NC gauge theories. In string theory, NC gauge theory is realized by
introducing constant Bµν field [24,25]. Let us consider a localized state |p on CP 2 as it
is discussed in Section 2. In the local patch around it, {t 4 , t 5 , t 6 , t 7 } can be interpreted as
the local coordinates. f458 and f678 can be interpreted as constant B12 and B34 fields since
the deformation term behaves locally like

(2π)2 2
if458 2 Tr t8 [â4, â5 ] + [â6 , â7 ]
gNC p

i
→ 2 d 4 x (a4 ( a5 − a5 ( a4 + a6 ( a7 − a7 ( a6 ). (4.1)
gNC
This is consistent with the coupling of Bµν field in IIB matrix model [26].
We hope to draw possible implications for the large-N limit of IIB matrix model
based on our results. We recall that the cubic terms we have added to IIB matrix model
formally vanish in the large-N limit. Nevertheless they affect the theory since different NC
gauge theories are realized based on different homogeneous spaces G/H . The situation
is analogous to the magnetic systems where different polarization directions are realized
depending on the directions of a tiny external magnetic field. We should hence clarify
under what conditions unique physics is realized in the large-N limit of IIB matrix model.
Another related issue is the possibility that the cubic terms may be dynamically generated
in the large-N limit of IIB matrix model. We recall here that four-dimensional distributions
for Aµ are found to be favored in the mean field approximation [27,28]. It is interesting to
study which homogeneous space minimizes the free energy of IIB matrix model under the
mean field approximation.
Acknowledgements
I would like to thank N. Ishibashi, S. Iso, Y. Kimura and T. Masuda for discussions. This
work is supported in part by the Grant-in-Aid for Scientific Research from the Ministry of
Education, Science and Culture of Japan.
References
[1] T. Banks, W. Fischler, S.H. Shenker, L. Susskind, M-theory as a matrix model: a conjecture, Phys. Rev. D 55
(1997) 5112, hep-th/9610043.
[2] N. Ishibashi, H. Kawai, Y. Kitazawa, A. Tsuchiya, A large-N reduced model as superstring, Nucl. Phys.
B 498 (1997) 467, hep-th/9612115.
[3] A. Connes, M. Douglas, A. Schwarz, JHEP 9802 (1998) 003, hep-th/9711162.
[4] H. Aoki, N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, T. Tada, Non-commutative Yang–Mills in IIB matrix
model, Nucl. Phys. 565 (2000) 176, hep-th/9908141.
[5] M. Li, Nucl. Phys. B 499 (1997) 149, hep-th/961222.
[6] N. Ishibashi, S. Iso, H. Kawai, Y. Kitazawa, Wilson loops in noncommutative Yang–Mills, Nucl. Phys. B 573
(2000) 573, hep-th/9910004.
[7] D.J. Gross, A. Hashimoto, N. Itzhaki, Observables of noncommutative gauge theories, Adv. Theor. Math.
Phys. 4 (2000) 893, hep-th/0008075.
[8] S. Minwalla, M.V. Raamsdonk, N. Seiberg, Non-commutative perturbative dynamics, JHEP 0002 (2000)
020, hep-th/9912072.
[9] S. Iso, H. Kawai, Y. Kitazawa, Bi-local fields in noncommutative field theory, Nucl. Phys. B 576 (2000)
375, hep-th/0001027.
[10] L. Susskind, The quantum Hall fluid and noncommutative Chern–Simons theory, hep-th/0101029.
[11] S. Aoyama, T. Masuda, The fuzzy Kähler coset space with the Darboux coordinates, Phys. Lett. B 521
(2001) 376, hep-th/0109020.
[12] D. Karabali, V.P. Nair, Quantum Hall effect in higher dimensions, hep-th/0203264.
[13] S.C. Zhang, J.P. Hu, Science 294 (2001) 823;
J.P. Hu, S.C. Zhang, cond-mat/0110572.
[14] C.N. Yang, SU(2) monopole harmonics, J. Math. Phys. 19 (1978) 2622.
[15] P.M. Ho, S. Ramgoolam, Higher-dimensional geometries from matrix brane constructions, Nucl. Phys.
B 627 (2002) 266, hep-th/0111278.
[16] M. Fabinger, Higher-dimensional quantum Hall effect in string theory, JHEP 0205 (2002) 637, hep-
th/0201016.
[17] J. Castelino, S. Lee, W. Taylor, Longitudinal 5-branes as 4-spheres in matrix theory, Nucl. Phys. B 526
(1998) 334, hep-th/9712105.
[18] Y. Kimura, Noncommutative gauge theory on fuzzy four-sphere and matrix model, hep-th/0204256.
[19] S. Iso, Y. Kimura, K. Tanaka, K. Wakatsuki, Non-commutative gauge theory on fuzzy sphere from matrix
model, Nucl. Phys. B 604 (2001) 121, hep-th/0101102.
[20] V.P. Nair, S. Randjbar-Daemi, On brane solutions in M(atrix) theory, Nucl. Phys. B 533 (1998) 333, hep-
th/9802187.
[21] G. Bonelli, Matrix strings in pp-wave backgrounds from deformed super-Yang–Mills theory, hep-
th/0205213.
[22] R.C. Myers, Dielectric-branes, JHEP 9912 (1999) 022, hep-th/9910053.
[23] J. Ambjorn, Y. Makeenko, J. Nishimura, R. Szabo, Finite N matrix models of noncommutative gauge theory,
JHEP 9911 (1999) 029, hep-th/9911041;
J. Ambjorn, Y. Makeenko, J. Nishimura, R. Szabo, Non-perturbative dynamics of noncommutative gauge
theory, Phys. Lett. B 480 (2000) 399, hep-th/0002158;
J. Ambjorn, Y. Makeenko, J. Nishimura, R. Szabo, Lattice gauge fields and discrete noncommutative Yang–
Mills theory, JHEP 0007 (2000) 013, hep-th/0003208.
[24] N. Seiberg, E. Witten, String theory and noncommutative geometry, JHEP 9909 (1999) 032, hep-th/9908142.
[25] N. Ishibashi, p-branes from (p − 2)-branes in the bosonic string theory, Nucl. Phys. B 539 (1999) 107,
hep-th/9804163;
N. Ishibashi, A relation between commutative and noncommutative descriptions of D-branes, hep-
th/9909176.
[26] Y. Kitazawa, Vertex operators in IIB matrix model, JHEP 0204 (2002) 004, hep-th/0201218.
[27] J. Nishimura, F. Sugino, Dynamical generation of four-dimensional space–time in IIB matrix model,
JHEP 0205 (2002) 001, hep-th/0111102.
[28] H. Kawai, S. Kawamoto, T. Kuroki, T. Matsuo, S. Shinohara, Mean field approximation of IIB matrix model
and emergence of four-dimensional space–time, hep-th/0204240.
Two-loop QCD helicity amplitudes for

e+e− → 3 jets
L.W. Garland a , T. Gehrmann b , E.W.N. Glover a , A. Koukoutsakis a ,
E. Remiddi c
a Department of Physics, University of Durham, Durham DH1 3LE, UK
b Theory Division, CERN, CH-1211 Geneva 23, Switzerland
c Dipartimento di Fisica, Università di Bologna and INFN, Sezione di Bologna, I-40126 Bologna, Italy
Received 7 June 2002; received in revised form 25 July 2002; accepted 29 July 2002
Abstract
We compute the two-loop QCD helicity amplitudes for the process e+ e− → q q̄g. The amplitudes
are extracted in a scheme-independent manner from the coefficients appearing in the general tensorial
structure for this process. The tensor coefficients are derived from the Feynman graph amplitudes
by means of projectors, within the conventional dimensional regularization scheme. The actual
calculation of the loop integrals is then performed by reducing all of them to a small set of known
master integrals. The infrared pole structure of the renormalized helicity amplitudes agrees with the
prediction made by Catani using an infrared factorization formula. We use this formula to structure
our results for the finite part into terms arising from the expansion of the pole coefficients and a
genuine finite remainder, which is independent of the scheme used to define the helicity amplitudes.
The analytic result for the finite parts of the amplitudes is expressed in terms of one- and two-
dimensional harmonic polylogarithms.
1. Introduction
The three-jet production rate in electron–positron annihilation [1,2] and related event
shape observables were measured to a very high precision at LEP, where they were used
in particular for the determination of the strong coupling constant αs . At present, the error
E-mail addresses: L.W.Garland@durham.ac.uk (L.W. Garland), Thomas.Gehrmann@cern.ch

(T. Gehrmann), E.W.N.Glover@durham.ac.uk (E.W.N. Glover), Athanasios.Koukoutsakis@durham.ac.uk
(A. Koukoutsakis), Ettore.Remiddi@bo.infn.it (E. Remiddi).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 2 7 - 2
228 L.W. Garland et al. / Nuclear Physics B 642 (2002) 227–262
on the extraction of αs from these data is dominated by the uncertainty inherent in the
theoretical next-to-leading order (NLO) calculation [3–7] of the jet observables (see [8] for
a review). The planned TESLA [9] linear e+ e− collider will allow precision QCD studies
at energies even higher than at LEP. Given the projected luminosity of TESLA, one again
expects the experimental errors to be well below the uncertainty of the NLO calculation.
The calculation of next-to-next-to-leading order (NNLO), i.e., O(αs3 ), corrections to
the three-jet rate in e+ e− annihilation has been considered as a highly important project
for a long time [10]. In terms of matrix elements, it requires the computation of three
contributions: the tree level γ ∗ → 5 partons amplitudes [11–13], the one-loop corrections
to the γ ∗ → 4 partons amplitudes [14–17], and the two-loop (as well as the one-
loop times one-loop) corrections to the γ ∗ → 3 partons matrix elements. In a previous
publication [18], we have derived both the interference of the tree and two-loop matrix
elements and the self-interference of the one-loop amplitudes averaged over all external
helicities. In the present work, we extend this calculation to compute the two-loop helicity
amplitudes for the process e+ e− → q q̄g.
The most precisely measured observables related to e+ e− → 3 jets are the jet
production rate itself and a number of event-shape variables. The calculation of these
phenomenologically most relevant applications, which also dominate the extraction of
αs , at NNLO accuracy requires only the helicity averaged squared matrix element at the
two-loop level derived in [18]. Nevertheless, the helicity amplitudes presented here are
interesting for a number of reasons:
• Oriented event-shape observables, which measure the spatial orientation of the final-
state jets relative to the direction of the incoming beams require, even for unpolarized
beams [19], the calculation of the polarization tensor of the virtual photon mediating
the interaction. This polarization tensor can be recovered from the helicity amplitudes.
• Likewise, to determine the direction of the decay leptons in the crossed process,
V + 1 jet production at unpolarized hadron colliders, it is necessary to compute the
polarization tensor of the vector boson.
• Polarization of the beams is an important option for the future linear e+ e− collider
TESLA [9], thus providing a direct measurement of event-shape observables in
polarized e+ e− annihilation.
• NNLO predictions for (V + 1)-jet production at the RHIC polarized proton–proton
collider and for (2 + 1) jet production at a currently discussed polarized upgrade of the
HERA collider do require the calculation of the two-loop helicity amplitudes. These
observables would then form part of a full NNLO determination of the polarized parton
distribution functions in the proton.
• The study of formal aspects of two-loop matrix elements, such as their collinear limits
or their high energy behaviour can be carried out more elegantly on the basis of the
underlying helicity amplitudes.
Two-loop helicity amplitudes have up to now only been derived for 2 → 2 bosonic
scattering processes with all external legs on-shell: for gg → γ γ [20], γ γ → γ γ [21,22]
and gg → gg [23,24]. The latter calculation also confirmed earlier results for the squared
two-loop gg → gg matrix element [25]. In the above calculations, which were all carried
L.W. Garland et al. / Nuclear Physics B 642 (2002) 227–262 229
out within dimensional regularization [26–28] with d = 4 − 2 spacetime dimensions, two

different methods were used to access the helicity structure of the matrix element: explicit
contraction with the external polarization vectors [20,21,23,24] or projection onto the
individual components of the Lorentz-invariant decomposition of the amplitude [22]. Once
these are applied to expose the helicity structure, one is left with the task of computing
a large number of two-loop integrals. Using integration-by-parts [28–30] and Lorentz-
invariance [31] identities, these can be reduced [32] to a small number of so-called master
integrals, which were derived for massless on-shell two-loop four-point functions in [33–
38]. If an explicit contraction with the external polarization vectors is performed, one
also has to compute two-loop integrals over the (d − 4)-dimensional subspace of loop
momenta, which reduce however to simple vacuum diagrams [24]. For 2 → 2 scattering
processes with external fermions and all external legs on-shell (e+ e− → e+ e− , q q̄ → q q̄ ,
q q̄ → q q̄, q q̄ → gg, q q̄ → gγ and q q̄ → γ γ ), only the squared, helicity-averaged two-
loop matrix elements were computed so far [39–42].
The method employed here to extract the two-loop helicity amplitudes for e+ e− → q q̄g
is similar to the approach of [22] by applying projections on all components of the Lorentz-
invariant decomposition of the amplitude. Using this approach, the corresponding one-
loop helicity amplitudes were derived in [6]. The master integrals relevant in the present
context are massless four-point functions with three legs on-shell and one leg off-shell.
The complete set of these two-loop integrals was computed in [43], while earlier partial
results had been presented in [44,45].1 The master integrals in [43] are expressed in terms
of two-dimensional harmonic polylogarithms (2dHPLs). The 2dHPLs are an extension of
the harmonic polylogarithms (HPLs) of [48]. All HPLs and 2dHPLs that appear in the
divergent parts of the planar master integrals have weight 3 and can be related to the
more commonly known Nielsen generalized polylogarithms [49,50] of suitable arguments.
The functions of weight 4 appearing in the finite parts of the master integrals can all be
represented, by their very definition, as one-dimensional integrals over 2dHPLs of weight
3, hence of Nielsen’s generalized polylogarithms of suitable arguments according to the
above remark. A table with all relations is included in the appendix of [43]. Numerical
routines providing an evaluation of the HPLs [51] and 2dHPLs [52] are available.
After carrying out ultraviolet renormalization of the amplitudes in the MS scheme, one
is left with poles which are purely of infrared origin. The infrared pole structure of the
amplitudes can be predicted using Catani’s infrared factorization formula [53]. We use this
formalism to present the infrared poles and the finite parts of the helicity amplitudes in a
compact form.
This paper is structured as follows: in Section 2, we outline the calculational method
used to derive the helicity amplitudes and discuss the techniques used to extract the
ultraviolet and infrared pole structure. We also elaborate on the relation to previous work.
The two-loop helicity amplitudes are computed (in the Weyl–van der Waerden formalism,
1 Note that an alternative approach avoiding the need to use the integration-by-parts and Lorentz-invariance
identities to reduce the integrals appearing in the Feynman diagrams to a basis set has recently been proposed [46,
47]. This method relies on obtaining analytic expressions for the basic topologies with arbitrary powers of the
propagators and arbitrary dimensions, which can often be found in terms of nested sums involving Γ -functions.
The Γ -functions can be directly expanded in and the nested sums related to multiple polylogarithms.
which is briefly described in Appendix A) in Section 3. Finally, Section 4 contains a

discussion of the results and conclusions.
2. Method
2.1. Notation
We consider the production of a quark–antiquark–gluon system in electron–positron

annihilation,
e+ (p5 ) + e− (p6 ) → γ ∗ (p4 ) −→ q(p1 ) + q̄(p2 ) + g(p3 ). (2.1)

It is convenient to define the invariants
s12 = (p1 + p2 )2 , s13 = (p1 + p3 )2 , s23 = (p2 + p3 )2 , (2.2)

which fulfil
p42 = (p1 + p2 + p3 )2 = s12 + s13 + s23 ≡ s123 , (2.3)

as well as the dimensionless invariants
x = s12 /s123 , y = s13 /s123 , z = s23 /s123 , (2.4)

which satisfy x + y + z = 1.
The renormalized amplitude |M can be written as
|M = V µ Sµ (q; g; q̄), (2.5)

where V µ represents the lepton current and Sµ denotes the hadron current. In a previous
paper [18], we have considered the unpolarized decay process
γ ∗ (p4 ) −→ q(p1 ) + q̄(p2 ) + g(p3 ) (2.6)

for which the amplitude is obtained from Eq. (2.5) by replacing the lepton current by the
µ
polarization vector of the virtual photon 4 . The hadron current may be perturbatively
decomposed as

√ αs
Sµ (q; g; q̄) = 4πα eq 4παs T ij Sµ (q; g; q̄) +
a (0)
S (1) (q; g; q̄)
2π µ
2
αs
+ Sµ(2) (q; g; q̄) + O αs3 , (2.7)
2π
where eq denotes the quark charge, a is the adjoint colour index for the gluon and i and j
are the colour indices for quark and antiquark. αs is the QCD coupling constant at the
renormalization scale µ, and the Sµ(i) are the i-loop contributions to the renormalized
amplitude. Renormalization of ultraviolet divergences is performed in the MS scheme.
2.2. The general tensor
The most general tensor structure for the hadron current Sµ (q; g; q̄) is
Sµ (q; g; q̄) = ū(p1 )/

p3 u(p2 )(A113 .p1 p1µ + A12 3 .p1 p2µ + A13 3 .p1 p3µ )
p3 u(p2 )(A21 3 .p2 p1µ + A22 3 .p2 p2µ + A233 .p2 p3µ )
+ ū(p1 )/
+ ū(p1 )γµ u(p2 )(B1 3 .p1 + B2 3 .p2 )
+ ū(p1 )/3 u(p2 )(C1 p1µ + C2 p2µ + C3 p3µ )
+ D1 ū(p1 )/3p
/ 3 γµ u(p2 )
+ D2 ū(p1 )γµp
/ 3/3 u(p2 ), (2.8)
where the constraint 3 · p3 = 0 has been applied. All coefficients are functions of s13 , s23
and s123 . The above tensor structure is a priori d-dimensional, since the dimensionality of
the external states has not yet been specified. The hadron current is conserved and satisfies
µ
Sµ (q; g; q̄)p4 = 0; (2.9)
it must also obey the QCD Ward identity when the gluon polarization vector 3 is replaced
with the gluon momentum,
Sµ (q; g; q̄)(3 → p3 ) = 0. (2.10)

These constraints yield relations amongst the 13 distinct tensor structures and applying
these identities gives the gauge-invariant form of the tensor,
Sµ (q; g; q̄) = A11 (s13 , s23 , s123 )T11µ + A12 (s13 , s23 , s123 )T12µ
+ A13 (s13 , s23 , s123 )T13µ + A21 (s13 , s23 , s123 )T21µ
+ A22 (s13 , s23 , s123 )T22µ + A23 (s13 , s23 , s123 )T23µ
+ B(s13 , s23 , s123 )Tµ , (2.11)
where Aij and B are gauge-independent functions and the tensor structures TI J µ and Tµ
are given by
s13 sJ 4
T1J µ = ū(p1 )/
p3 u(p2 )3 .p1 pJ µ − ū(p1 )/3 u(p2 )pJ µ + ū(p1 )/3p
/ 3 γµ u(p2 ),
2 4
(2.12)
s23 sJ 4
T2J µ = ū(p1 )/
p3 u(p2 )3 .p2 pJ µ − ū(p1 )/3 u(p2 )pJ µ + ū(p1 )γµp/ 3 /3 u(p2 ),
2 4
(2.13)
1
Tµ = s23 ū(p1 )γµ u(p2 )3 .p1 + ū(p1 )/3p / 3 γµ u(p2 )
2

1
− s13 ū(p1 )γµ u(p2 )3 .p2 + ū(p1 )γµp / 3 /3 u(p2 ) . (2.14)
2
Each of the tensor structures satisfies both current conservation and the QCD Ward identity.
The coefficients are further related by symmetry under the interchange of the quark and
antiquark,
A21 (s13 , s23 , s123 ) = −A12(s23 , s13 , s123 ),
A22 (s13 , s23 , s123 ) = −A11(s23 , s13 , s123 ),
A23 (s13 , s23 , s123 ) = −A13(s23 , s13 , s123 ),
B(s13 , s23 , s123 ) = B(s23 , s13 , s123 ). (2.15)
2.3. Projectors for the tensor coefficients
The coefficients AI J and B may be easily extracted from a Feynman diagram

calculation, using projectors such that
µ
P(X) 4 Sµ (q; g; q̄) = X(s13 , s23 , s123 ). (2.16)
spins
The explicit forms for the seven projectors in d spacetime dimensions are,
(s23 s123 d + s13 s12 (d − 2)) (s13 + s23 )(d − 2) † ∗
P(A11 ) = †
T11 · 4∗ − 2 s 2 (d − 3)s
T12 · 4
3 s 2 (d
2s13 12 − 3)s123 2s13 12 123
((s23 + s12 )d + 2s13 )
− T † · 4∗
3
2s12 s13 (d − 3)s123 13
(s23 s123 (d − 2) + s13 s12 (d − 4)) † ∗
− T21 · 4
13 123 (d − 3)
2 s2 s
2s23 s12
(s13 + s23 )(d − 4) (s23 + s12 )(d − 4)
+ †
T22 · 4∗ + †
T23 · 4∗
2(d − 3)s12 2 s s s
123 13 23 2s s s 2 s
23 12 13 123 (d − 3)
1
− 2 2 T † · 4∗ ,
2s13 s12 (d − 3)
(s13 + s23 )(d − 2) † ∗
P(A12 ) = − 2 2 T · 4
2s13 s12 (d − 3)s123 11
(d − 2)(s23 s12 (d − 4) + s13 s123 (d − 2)) † ∗
+ 2 s 2 s (d − 3)s
T12 · 4
2s13 12 23 123 (d − 4)
(d − 2)(s13 + s12 )
− 2 †
T13 · 4∗
2s13 s12 s23 (d − 3)s123
((d − 6)(d − 2)(s13 + s23 ) − 4s12 ) † ∗
+ T21 · 4
2(d − 4)s12 13 23 123 (d − 3)
2 s s s
(s23 s12 (d − 4) + s13 s123 (d − 2)) † ∗

− 2 s s 2 (d − 3)s
T22 · 4
2s12 13 23 123
(2s23 + (s13 + s12 )(d − 2)) † ∗
+ T23 · 4
123 (d − 3)
2 s
2s12 s13 s23
(d − 2)
− T † · 4∗ ,
2(d − 4)s12 2 s s (d − 3)
13 23
((s23 + s12 )d + 2s13 ) (d − 2)(s13 + s12 )

P(A13 ) = − †
T11 · 4∗ − 2 s s (d − 3)s
†
T12 · 4∗
3 (d
2s12 s13 − 3)s123 2s13 12 23 123
(s13 s23 (d − 2) + s12 s123 d)
+ †
T13 · 4∗
3
2s13 s12 s23 s123 (d − 3)
((s12 + s23 )(d − 2) + 2s13 ) † ∗
+ 2 s s (d − 3)s
T21 · 4
2s13 12 23 123
(s13 + s12 )(d − 4) (s13 + s12 )(s23 + s12 )(d − 4) † ∗
+ †
T22 · 4∗ − T23 · 4
2s12 s13 s23 s123 (d − 3)
2
12 23 123 (d − 3)
2 s s2 s
2s13
1
+ T † · 4∗ ,
2 s (d
2s23 s13 12 − 3)
(s23 s123 (d − 2) + s13 s12 (d − 4)) † ∗
P(A21 ) = − 2 s 2 s (d − 3)s
T11 · 4
2s13 12 23 123
(−4s12 + (s13 + s23 )(d − 6)(d − 2)) † ∗
+ T12 · 4
2(d − 4)s12 13 23 123 (d − 3)
2 s s s
((s23 + s12 )(d − 2) + 2s13 ) † ∗

+ 2 s s (d − 3)s
T13 · 4
2s13 12 23 123
(d − 2)(s23 s123 (d − 2) + s13 s12 (d − 4)) † ∗
+ T21 · 4
13 23 123 (d − 3)(d − 4)
2 s s2 s
2s12
(s13 + s23 )(d − 2) † ∗ (s23 + s12 )(d − 2)
− T22 · 4 − †
T23 · 4∗
2s12 s23 s123 (d − 3)
2 2 2s13 s23 s12 (d − 3)s123
2
(d − 2)
+ T † · 4∗ ,
2(d − 4)s12
2 s s (d − 3)
13 23
(s13 + s23 )(d − 4) (s23 s12 (d − 4) + s13 s123 (d − 2)) † ∗
P(A22 ) = †
T11 · 4∗ − T12 · 4
2s13 s12 s23 (d − 3)s123
2 2 s s 2 (d − 3)s
2s12 13 23 123
(s13 + s12 )(d − 4) (s13 + s23 )(d − 2) † ∗
+ 2 s (d − 3)s
†
T13 · 4∗ − 2 2 T21 · 4
2s13 s23 12 123 2s12 s23 s123 (d − 3)
(s23 s12 (d − 2) + s13 s123 d) (s13 d + s12 d + 2s23 )
+ †
T22 · 4∗ − †
T23 · 4∗
3 2
2s23 s12 (d − 3)s123 3
2s12 s23 s123 (d − 3)
1
+ T † · 4∗ ,
2 s 2 (d
2s23 12 − 3)
(s23 + s12 )(d − 4) (2s23 + (s13 + s12 )(d − 2)) † ∗
P(A23 ) = †
T11 · 4∗ + T12 · 4
2 s
2s23 s12 s13 123 (d − 3) 2 s (d − 3)s
2s13 s23 12 123
(s13 + s12 )(s23 + s12 )(d − 4) † ∗ (s23 + s12 )(d − 2)
− T13 · 4 − †
T21 · 4∗
2s13 s12 s23 s123 (d − 3)
2 2 2 s (d − 3)s
2s13 s23 12 123
((s13 + s12 )d + 2s23 ) (s13 s23 (d − 2) + s12 s123 d)
− †
T22 · 4∗ + †
T23 · 4∗
3
2s12 s23 s123 (d − 3) 3
2s13 s12 s23 s123 (d − 3)
1
− T † · 4∗ ,
2 s s (d
2s23 13 12 − 3)
1 (d − 2)
P(B) = − T † · ∗ −
2 s 2 11 4
T † · ∗
2 s s (d − 3) 12 4
2(d − 3)s12 13 2(d − 4)s12 13 23
1 (d − 2)
+ †
T13 · 4∗ + T † · ∗
2 s s (d − 3) 21 4
2s23 s12 (d − 3)s13
2 2(d − 4)s12 13 23
1 1
+ 2 2 T † · 4∗ − T † · ∗
2 23 4
2s23 s12 (d − 3) 22 2s13 s12 (d − 3)s23
1
+ T † · 4∗ . (2.17)
2(d − 4)s12 2 s s
13 23
2.4. The perturbative expansion of the tensor coefficients
Each of the unrenormalized coefficients AI J and B has a perturbative expansion of the

form
2
√ (0),un αs (1),un αs (2),un 3
AI J = 4πα eq 4παs T ij AI J +
un a
AI J + AI J + O αs ,
2π 2π
2
√ αs αs
B un = 4πα eq 4παs T aij B (0),un + B (1),un + B (2),un + O αs3 ,
2π 2π
(2.18)
where the dependence on (s13 , s23 , s123 ) is implicit. At tree level,
A(0),un
IJ (s13 , s23 , s123 ) = 0, (2.19)
2
B (0),un(s13 , s23 , s123 ) = . (2.20)
s13 s23
The one-loop contributions can be written in terms of the one-loop box integral in d =
6 − 2 dimensions, Box6 (sij , sik , sij k ), and the one-loop bubble, Bub(sij ), as follows:
(1),un
A11 (s13 , s23 , s123 )

(d − 4) (d − 4)
=N − Bub(s123 ) − Bub(s13 ) − Bub(s123 )

2(s13 + s12 )s13 2s12 s13
((d − 2)s23 s12 + (d − 4)s23 s13 + 4s12 (s12 + s13 ))
− Bub(s23 ) − Bub(s123 )
2s12 s13 (s13 + s12 ) 2

(d − 4)(4s12 + (d − 2)s23 )
− Box6 (s13 , s23 , s123 )
4s12 s13

1 (d − 4) d
+ Bub(s123 ) + 2 Bub(s12 ) − Bub(s123 )

N 2(s13 + s12 )s13 2s13
(s12 + s13 )(ds23 + 4s13 ) + 2s23 s13
+ Bub(s23 ) − Bub(s123 )
2(s13 + s12 ) s13
2 2
(d − 4)(d − 6)
+ Box6 (s12 , s13 , s123 )
4s13

(d − 2)(ds23 + 4s13 )
+ 2
6
Box (s12 , s23 , s123 ) , (2.21)
4s13
A(1),un
12 (s13 , s23 , s123 )

(d − 10)
=N − Bub(s13 ) − Bub(s123 )
2s12 (s23 + s12 )
((d − 10)s13 − 4s12 )
− Bub(s23 ) − Bub(s123 )
2s12 s13 (s13 + s12 )

(4(d − 4)s12 − (d − 2)(d − 10)s13 )
+ 6
Box (s13 , s23 , s123 )
4s12 s13

1 (d − 2)
+ Bub(s12 ) − Bub(s123 )
N 2s23 s13
((d − 2)s12 + 2(d − 6)s23 )
+ Bub(s13 ) − Bub(s123 )
2s23 s12 (s23 + s12 )
(d − 6)(s12 + 2s13 )
+ Bub(s23 ) − Bub(s123 )
2s12 s13 (s13 + s12 )
((d − 2)2 s12 s13 + 4(d − 4)s12 s23 + 2(d − 4)(d − 6)s13 s23 )
+
4s12 s13 s23
× Box6 (s12 , s13 , s123 )

(d − 6)((d − 2)s12 + 2(d − 4)s13 )
+ Box6 (s12 , s23 , s123 ) , (2.22)
4s12 s13
A(1),un
13 (s13 , s23 , s123 )

(d − 6)
=N Bub(s23 ) − Bub(s123 )
2(s13 + s12 )s13

(d − 4)(d − 6)
+ 6
Box (s13 , s23 , s123 )
4s13

1 (d − 4) (d − 4)
+ − Bub(s123 ) − Bub(s13 ) − Bub(s123 )

N s13 (s23 + s13 ) 2s23 s13
2 − ds (s + s )2 − 2(d − 2)s s (s + s ))
(4s12 s13 12 13 23 13 23 13 23
+
2(s23 + s13 )2 s13
2 s
23

× Bub(s12 ) − Bub(s123 )
(2(d − 3)s13 + ds12 )
− 2 (s + s )
Bub(s23 ) − Bub(s123 )
2s13 13 12
(d − 4)((d − 2)s12 + 4s23 )
− Box6 (s12 , s13 , s123 )
4s23 s13

(d − 2)(ds12 + 2(d − 4)s13 )
− 2
6
Box (s12 , s23 , s123 ) , (2.23)
4s13
B (1),un(s13 , s23 , s123 )

2
d − 3d + 4
=N Bub(s123 )
4(d − 4)s13 s23
(4(d − 3)s12 (s12 + s23 ) + (d − 4)(d − 7)s23 s13 )
+ Bub(s13 ) − Bub(s123 )
2s12 s23 (s23 + s12 )s13 (d − 4)

(4(d − 3)s12 2 + (d − 2)(d − 7)s s )
13 23
+ Box6 (s13 , s23 , s123 )
8s12 s13 s23

1 (7d − 16 − d 2 ) (16 − 5d)
+ Bub(s123 ) + Bub(s12 ) − Bub(s123 )

N 4(d − 4)s13 s23 4(d − 4)s13 s23
(s12 + (d − 6)s23 )
− Bub(s13 ) − Bub(s123 )
2(s23 + s12 )s23 s12
(4(d − 3)s23 s12 + (d − 4)(d − 6)s13 s23 + (d − 2)s12 s13 )
−
4s12 s13 s23

× Box6 (s12 , s13 , s123 ) + s13 ↔ s23 . (2.24)
Explicit expansions of the one-loop integrals around ∼ 0 in terms of HPLs and 2dHPLs
are listed in Appendix A of [18].
(2),un
Similarly, the unrenormalized two-loop AI J and B (2),un coefficients were obtained
analytically (making extensive use of the computer algebra programs MAPLE [54],
FORM2 [55] and FORM3 [56], where the latter two are particularly well suited for
handling the large-size expressions arising at intermediate stages of the calculation) in
terms of a basis set of two-loop master integrals. This basis set comprises 14 planar
topologies and 5 non-planar topologies. Five of the topologies require more than one
master integral, so that in total 24 master integrals are needed. A more detailed discussion
can be found in Ref. [18]. However, we note that Laurent expansions for each of these
master integrals have been derived in [43] by solving differential equations for the master
integrals (equations that are differential with respect to the momentum scales involved in
the diagram). The -expansions of A(2),un
IJ and B (2),un can therefore be obtained by directly
substituting the -expansions of the individual master integrals.
2.5. Relation to previous work
We have considered the case where the correlations with the lepton current are ignored
in a previous paper [18]. In this instance, the squared amplitude for the process γ ∗ → q q̄g,
summed over spins, colours and quark flavours, was denoted by

M|M = |4 · S(q; g; q̄)|2 = T (x, y, z). (2.25)
The perturbative expansion of T (x, y, z) at renormalization scale µ2 = q 2 = s123 reads:

2 (2) αs (q 2 )
T (x, y, z) = 16π α
2
eq αs q T (x, y, z) +
2
T (4) (x, y, z)
q
2π
2
αs (q 2 ) 3 2
+ T (6)
(x, y, z) + O αs q , (2.26)
2π
where

T (2) (x, y, z) = M(0) M(0)

y z 2(1 − y − z) − 2yz
= 4V (1 − ) (1 − ) + + , (2.27)
z y yz

T (4) (x, y, z) = M(0) M(1) + M(1)M(0) , (2.28)

T (6) (x, y, z) = M(1) M(1) + M(0)M(2) + M(2)M(0) , (2.29)
where V = N 2 − 1, with N the number of colours. T (4) (x, y, z) was first derived in [3,4]
through to O( 0 ) while an explicit expression for it to all orders in was given in [18].
The contribution to T (6) (x, y, z) from the interference of two-loop and tree diagrams

T (6,[2×0]) (x, y, z) = M(0)M(2) + M(2)M(0) , (2.30)
as well as the one-loop self-interference

T (6,[1×1]) (x, y, z) = M(1)M(1) (2.31)
were first derived in [18].

It is straightforward to obtain the interference of the tree and i-loop amplitudes in terms
of the tensor coefficients, AI J and B. We find

M(0)M(i)
V
= 2(1 − ) (s12 s123 + s12 s13 + s13 s23 )
2
(i)
− (s13 + s23 )(s12 + s13 ) A11 (s13 , s23 , s123 )

+ 2(s12 + s23 )2 − 2 s123 s23 + (s12 + s23 )2
(i)
+ 2 2 (s13 + s23 )(s12 + s23 ) A12 (s13 , s23 , s123 )

+ 2 s23 − (s13 + s23 ) s123 − (s13 + s23 ) A(i) 13 (s13 , s23 , s123 )
2
+ 2 s13 + s232
+ 2s12 s123 − 2 s123 2
− s12 s13 − s12 s23 − s13 s23

+ 2 (s13 + s23 )2 B (i) (s13 , s23 , s123 ) + {p1 ↔ p2 } . (2.32)
The above relation holds for the unrenormalized as well as for the renormalized matrix
element, involving the appropriate unrenormalized or renormalized tensor coefficients,
respectively. Similar, but more lengthy, expressions can easily be obtained for the
interference of i- and j -loop amplitudes. We have checked that inserting the expressions
(i)
for AI J and B (i) into Eq. (2.32) reproduces our earlier results [18] at the one- and two-loop
level both at the master integral level and after making an expansion in .
2.6. Ultraviolet renormalization
The renormalization of the matrix element is carried out by replacing the bare coupling
α0 with the renormalized coupling αs ≡ αs (µ2 ), evaluated at the renormalization scale µ2
2 2
β0 αs β0 β1 αs 3
α0 µ2 S
0 = αs µ 2
1 − + − + O αs , (2.33)
2π 2 2 2π
where
S = (4π) e−γ with Euler constant γ = 0.5772 . . .

and µ20 is the mass parameter introduced in dimensional regularization [26–28] to maintain
a dimensionless coupling in the bare QCD Lagrangian density; β0 and β1 are the first two
coefficients of the QCD β-function:
11CA − 4TR NF 17CA2 − 10CA TR NF − 6CF TR NF
β0 = , β1 = , (2.34)
6 6
with the QCD colour factors
N2 − 1 1
CA = N, CF = , TR = . (2.35)
2N 2
(i),un
We denote the i-loop contribution to the unrenormalized coefficients by AI J and
B (i),un ,
using the same normalization as for the decomposition of the renormalized
amplitude (2.7); the dependence on (s13 , s23 , s123 ) is always understood implicitly. The
renormalized coefficients are then obtained as
A(0)
I J = 0,
AI J = S−1 AI J
(1) (1),un
,
−2 (2),un 3β0 −1 (1),un
A(2)
I J = S AI J − S A , (2.36)
2 I J
and
B (0) = B (0),un,
β0 (0),un
B (1) = S−1 B (1),un −
B ,
2

−2 (2),un 3β0 −1 (1),un β1 3β02
B = S B
(2)
− S B − − 2 B (0),un. (2.37)
2 4 8
For the remainder of this paper we will set the renormalization scale µ2 = q 2 . The full
scale dependence of the tensor coefficients is given by

√ αs (µ2 ) (1) αs (µ2 ) 2
AI J = 4πα eq 4παs T ij a
AI J +
2π 2π
2
3β 0 (1) µ
× A(2)
IJ + A ln 2 + O αs3 ,
2 IJ q
2
√ αs (µ2 ) β0 µ
B= 4πα eq 4παs T aij B (0) + B (1) + B (0) ln 2
2π 2 q
2 2
αs (µ ) 3β0 (1) β1 (0)
+ B (2) + B + B
2π 2 2
2 2
µ 3β 2
µ
× ln 2 + 0 B (0) ln2 2 + O αs3 . (2.38)
q 8 q
2.7. Infrared behaviour of the tensor coefficients
After performing ultraviolet renormalization, the amplitudes still contain singularities,

which are of infrared origin and will be analytically cancelled by those occurring in
radiative processes of the same order. Catani [53] has shown how to organize the infrared
pole structure of the one- and two-loop contributions renormalized in the MS scheme in
terms of the tree and renormalized one-loop amplitudes. The same procedure applies to
the tensor coefficients. In particular, the infrared behaviour of the one-loop coefficients is
given by
A(1) (1),finite
I J = AI J ,
B (1) = I (1) ()B (0) + B (1),finite, (2.39)
while the two-loop singularity structure is
A(2) (1)
I J = I ()AI J + AI J
(1) (2),finite
,

1 β Γ (1 − 2) β0
B (2) = − I (1) ()I (1) () − I (1) () + e−γ
0
+ K I (1) (2)
2 Γ (1 − )

+ H (2)() B (0) + I (1) ()B (1) + B (2),finite, (2.40)
where the constant K is

67 π 2 10
K= − CA − TR NF . (2.41)
18 6 9
(i),finite
The finite remainders AI J and B (i),finite remain to be calculated.
For this particular process, there is only one colour structure present at tree level which,
in terms of the gluon colour a and the quark and antiquark colours i and j , is simply T aij .
Adding higher loops does not introduce additional colour structures, and the amplitudes are
therefore vectors in a one-dimensional space. Similarly, the infrared singularity operator
I (1) () is a 1 × 1 matrix in the colour space and is given by

eγ 1 3 β0 1 1 3
I () = −
(1)
N 2+ + (S13 + S23 ) − + S12 ,
2Γ (1 − ) 4 2N N 2 2
(2.42)
where (since we have set µ2 = s123 )

s123
Sij = − . (2.43)
sij
Note that on expanding Sij , imaginary parts are generated, the sign of which is fixed
by the small imaginary part +i0 of sij . The origin of the various terms in Eq. (2.42) is
straightforward. Each parton pair ij in the event forms a radiating antenna of scale sij .
Terms proportional to Sij are cancelled by real radiation emitted from leg i and absorbed
by leg j . The soft singularities O(1/ 2 ) are independent of the identity of the participating
partons and are universal. However, the collinear singularities depend on the identities of
the participating partons.
Finally, the term of Eq. (2.40) that involves H (2) () produces only a single pole in
and is given by
eγ
H (2) () = H (2), (2.44)
4 Γ (1 − )
where the constant H (2) is renormalization-scheme-dependent. As with the single-pole
parts of I (1)(), the process-dependent H (2) can be constructed by counting the number
of radiating partons present in the event. In our case, there is a quark–antiquark pair and a
gluon present in the final state, so that
H (2) = 2Hq(2) + Hg(2), (2.45)
where, in the MS scheme:

1 5 11π 2 5 2 π2 89 NF
Hg =
(2)
ζ3 + + N +
2
NF + − − NNF − , (2.46)
2 12 144 27 72 108 4N

7 409 11π 2 1 41 π2
Hq(2) = ζ3 + − N 2 + − ζ3 − −
4 864 96 4 108 96
2
3 3 π 2 1 π 25 (N 2 − 1)NF
+ − ζ3 − + + − , (2.47)
2 32 8 N2 48 216 N
so that

589 11π 2 1 41 π 2
H (2) = 4ζ3 + − N 2 + − ζ3 − −
432 72 2 54 48
2
3 π 1 19 π 2
+ −3ζ3 − + + − + NNF
16 4 N2 18 36

1 π 2 NF 5
+ − − + NF2 . (2.48)
54 24 N 27
(2) (2)
The factors Hq and Hg are directly related to those found in gluon–gluon scatter-
ing [25], quark–quark scattering [40] and quark–gluon scattering [41] (which each involve
four partons) as well as in the quark form factor [57–60] and gluon form factor [61]. We
also note that (on purely dimensional grounds) one might expect terms of the type S2ij to
be present in H (2) . Of course such terms are 1 + O() and therefore leave the pole part
unchanged, only modifying the finite remainder. At present it is not known how to system-
atically include these effects.
3. Helicity amplitudes
We can extend the results of the previous section to include Z boson exchange,

e+ (p5 ) + e− (p6 ) → Z ∗ , γ ∗ (p4 ) −→ q(p1 ) + q̄(p2 ) + g(p3 ), (3.1)
where the off-shell vector boson now distinguishes between left- and right-handed
fermions by keeping track of the helicity of the final state quarks.2 A convenient method
to evaluate the helicity amplitudes is in terms of Weyl–van der Waerden spinors, which is
described briefly in Appendix A and in detail in [62,63].
It is also straightforward to include the spin-correlations with the initial state by
contracting the hadronic current with the lepton current Vµ for fixed helicities of the initial
state electron (and positron). Using the spinor calculus of Appendix A we can express the
lepton current Vµ in terms of the helicities of the incident e+ and e− (with momenta p5
and p6 respectively). Explicitly,
γ
Lee
Vµγ e+ +, e− − = eσµȦB p6Ȧ p5B ,
s
LZ
VµZ e+ +, e− − = eσµȦB p6Ȧ p5B ee
,
s − MZ2 + iΓZ MZ
γ
Ree
Vµγ e+ −, e− + = eσµȦB p5Ȧ p6B ,
s
Z
Ree
VµZ e+ −, e− + = eσµȦB p5Ȧ p6B . (3.2)
s − MZ2 + iΓZ MZ
The hadronic current Sµ is related to the fixed helicity currents, SȦB , by
√
Sµ (q+; gλ; q−) = RfV1 f2 2 σµȦB SȦB (q+; gλ; q−), (3.3)
√
Sµ (q−; gλ; q+) = LVf1 f2 2 σµȦB SȦB (q−; gλ; q+). (3.4)
As in Eq. (2.7), the gauge boson coupling is extracted from SȦB . As mentioned earlier, the
left- and right-handed currents couple with a different strength when the vector boson is a
Z.
The currents with the quark helicities flipped follow from parity conservation:
∗
SȦB (q−; gλ; q+) = SḂA (q+; g(−λ); q−) . (3.5)
Charge conjugation implies the following relations between currents with different
helicities:
SȦB (qλq ; gλ; qλq ) = (−1)SȦB (qλq ; gλ; qλq ). (3.6)

All helicity amplitudes are therefore related to the amplitudes with λq = + and λq̄ = −.
2 Note that the full matrix element for any process should be summed over both photon and Z-boson exchange.
Explicitly, we find
SȦB (q+; g+; q−)
p1ȦD p2D p2B p pD p2B p pĊ p
= α(y, z) + β(y, z) 3ȦD 2 + γ (y, z) 1ĊB 3 3Ȧ ∗
p1 p3 p3 p2 p1 p3 p3 p2 p1 p3 p3 p2
p1 p3 ∗
+ δ(y, z) (p + p2ȦB + p3ȦB ). (3.7)
p1 p3 p1 p2 ∗ 1ȦB
The other helicity amplitudes are obtained from SȦB (q+; g+; q̄−) by the above parity
and charge conjugation relations, while the coefficients α, β and γ are written in terms of
the tensor coefficients:
s23 s13
α(y, z) = 2B(s13 , s23 , s123 ) + A12(s13 , s23 , s123 ) − A11(s13 , s23 , s123 ) ,
4
s13
β(y, z) = 2s23 B(s13 , s23 , s123 ) + 2(s12 + s13 )A11 (s13 , s23 , s123 )
4

+ s23 A12 (s13 , s23 , s123 ) + A13 (s13 , s23 , s123 ) ,
s13 s23
γ (y, z) = A11 (s13 , s23 , s123 ) − A13 (s13 , s23 , s123 ) ,
4
s12 s13
δ(y, z) = − A11 (s13 , s23 , s123 ). (3.8)
4
µ
When the hadron tensor is contracted with 4 or the lepton current V µ , the final term
of Eq. (3.7) vanishes.3 Furthermore, current conservation implies the following relation
between the four helicity coefficients,
2s123
α(y, z) − β(y, z) − γ (y, z) − δ(y, z) = 0. (3.9)
s12
This relation is fulfilled automatically once the tensor coefficients are inserted and does
therefore not yield a further reduction of the tensor basis.
As with the tensor coefficients, the helicity amplitude coefficients α, β and γ are vectors
in colour space and have perturbative expansions:
2
√ αs αs 3
Ω = 4πα 4παs T ij Ω + a (0)
Ω +
(1)
Ω + O αs ,
(2)
(3.10)
2π 2π
for Ω = α, β, γ . The dependence on (y, z) is again implicit.
The ultraviolet and infrared properties of the helicity coefficients match those of the
tensor coefficients,
Ω (0) = Ω (0),un,
β0 (0),un
Ω (1) = S−1 Ω (1),un −
Ω ,
2

−2 (2),un 3β0 −1 (1),un β1 3β02
Ω = S Ω
(2)
− S Ω − − 2 Ω (0),un, (3.11)
2 4 8
3 And for this reason was omitted in Ref. [6].

and
Ω (1) = I (1) ()Ω (0) + Ω (1),finite,

1 β0 Γ (1 − 2) β0
Ω (2) = − I (1)()I (1) () − I (1) () + e−γ + K I (1)(2)
2 Γ (1 − )

+ H (2)() Ω (0) + I (1) ()Ω (1) + Ω (2),finite, (3.12)
where I (1)() and H (2) () are defined in Eqs. (2.42) and (2.44), respectively.
At leading order
α (0) (y, z) = β (0) (y, z) = 1 and γ (0)(y, z) = 0. (3.13)
The renormalized next-to-leading order helicity amplitude coefficients can be straightfor-
wardly obtained to all orders in from the tensor coefficients using Eqs. (2.21)–(2.24). For
practical purposes, they are needed through to O( 2 ) in evaluating the infrared-divergent
one-loop contribution to the two-loop amplitude, while only the finite piece is needed for
the one-loop self-interference. They can be decomposed according to their colour structure
as follows:
1
Ω (1),finite(y, z) = NaΩ (y, z) + bΩ (y, z) + β0 cΩ (y, z). (3.14)
N
The expansion of the coefficients through to 2 yields HPLs and 2dHPLs up to weight 4
for aΩ , bΩ and up to weight 3 for cΩ . The explicit expressions are of considerable size,
such that we only quote the 0 -terms here (although these have been known already for a
long time [6]). The expressions through to O( 2 ) can be obtained in FORM format from
the authors. An example of the size and structure of those coefficients can be found in [18],
where we explicitly list the helicity-averaged one-loop times one-loop and tree times two-
loop matrix elements. The one-loop coefficients read:
7 π2 3 1 1 3
aα (y, z) = − − + H(0; z) − H(0, z)G(0; y) − H(1, 0; z) − G(0; y)
4 12 8 2 2 8
1 1 1
+ G(1, 0; y) − H(0; z) − 1 + 2H(0; z) + O(),
2 4(1 − z) 2 4(1 − z)
z 2
bα (y, z) = 2 H(0; z)G(1 − z; y) + H(1; z)G(−z; y) − G(−z, 1 − z; y)
2y
z
+ −H(0; z) + 2H(0; z)G(1 − z; y) − H(1; z) + 2H(1; z)G(−z; y)
2y
1 1
+ G(1 − z; y) − 2G(−z, 1 − z; y) + H(0; z) − H(0; z)
2y(1 − z) 2y
1 1 7 3
+ H(0; z) + 1 + 2H(0; z) + − H(0; z)
4(1 − z)2 4(1 − z) 4 4
1 1 3
+ H(0; z)G(1 − z; y) + H(0, 1; z) − H(1; z) + H(1; z)G(−z; y)
2 2 4
1 3 1
− H(1; z)G(0; y) + G(1 − z; y) + G(1 − z, 0; y)
2 4 2
1 1
− G(−z, 1 − z; y) + G(0, 1 − z; y) − G(1, 0; y) + O(),
2 2
1 1 iπ
cα (y, z) = − H(0; z) − G(0; y) + + O(),
4 4 2
3 π2 3 1 1 3
aβ (y, z) = − − + H(0; z) − H(0; z)G(0; y) − H(1, 0; z) − G(0; y)
2 12 8 2 2 8
1 1
+ G(1, 0; y) + H(0; z) + O(),
2 4(1 − z)
z(1 − z)
bβ (y, z) = 2
−H(0; z)G(1 − z; y) − H(1; z)G(−z; y) + G(−z, 1 − z; y)
2y
z
+ −H(0; z) + 2H(0; z)G(1 − z; y) − H(1; z) + 2H(1; z)G(−z; y)
2y
1
+ G(1 − z; y) − 2G(−z, 1 − z; y) + −2H(0; z)G(1 − z; y)
2y

+ H(1; z) − 2H(1; z)G(−z; y) − G(1 − z; y) + 2G(−z, 1 − z; y)
z z 1
+ −H(1; z) + G(1 − z; y) + − H(0; z)
2(y + z)2 2(y + z) 4(1 − z)
1 3 3
+ H(1; z) − G(1 − z; y) + − H(0; z)
2(y + z) 2 4
1 1 3
+ H(0; z)G(1 − z; y) + H(0, 1; z) − H(1; z) + H(1; z)G(−z; y)
2 2 4
1 3 1
− H(1; z)G(0; y) + G(1 − z; y) + G(1 − z, 0; y)
2 4 2
1 1
− G(−z, 1 − z; y) + G(0, 1 − z; y) − G(1, 0; y) + O(),
2 2
1 1 iπ
cβ (y, z) = − H(0; z) − G(0; y) + + O(),
4 4 2
1 1 1
aγ (y, z) = − + H(0; z) + 1 − H(0; z) + O(),
4 4(1 − z)2 4(1 − z)
1 z
bγ (y, z) = + 2 −H(0; z)G(1 − z; y) − H(1; z)G(−z; y) + G(−z, 1 − z; y)
4 2y
1 1
− H(0; z) + H(0; z) + H(1; z) − G(1 − z; y)
2y(1 − z) 2y
z z
+ H(1; z) − G(1 − z; y) −
2(y + z) 2 2(y + z)
1 1
+ −H(1; z) + G(1 − z; y) − H(0; z)
2(y + z) 4(1 − z)2
1
+ −1 + H(0; z) + O(),
4(1 − z)
cγ (y, z) = 0. (3.15)
It should be noted that these finite pieces of the one-loop coefficients can equally well be
written in terms of ordinary logarithms and dilogarithms, see [3,6]. The reason to express
them in terms of HPLs and 2dHPLs here is their usage in the infrared counter-term of
the two-loop coefficients, which cannot be fully expressed in terms of logarithmic and
polylogarithmic functions.
The finite two-loop remainder is obtained by subtracting the predicted infrared structure
(expanded through to O( 0 )) from the renormalized helicity coefficient. We further
decompose the finite remainder according to the colour structure, as follows:
1
Ω (2),finite(y, z) = N 2 AΩ (y, z) + BΩ (y, z) +CΩ (y, z) + NNF DΩ (y, z)
N2

NF 4
+ EΩ (y, z) + NF FΩ (y, z) + NF,V
2
− N GΩ (y, z),
N N
(3.16)
where the last term is generated by graphs where the virtual gauge boson does not couple
directly to the final-state quarks. This contribution is denoted by NF,V and is proportional
to the charge weighted sum of the quark flavours. In the case of purely electromagnetic
interactions we find,

q eq
NF,γ = . (3.17)
eq
Including Z-interactions, the same class of diagrams yields not only a contribution from
the vector component of the Z, which for the right-handed quark amplitude is given by
Z
q Lqq + Rqq
Z
NF,Z = Z
, (3.18)
2Rqq
but also a contribution involving the axial couplings of the Z [64]. This contribution
vanishes if summed over isospin doublets. The large mass splitting of the third quark family
induces a non-vanishing contribution from this class of diagrams, which can however not
be computed within the framework of massless QCD employed here, but can only be
obtained within an effective theory with large top-quark mass. In contrast to the vector
contribution from these diagrams, which is finite, one could expect divergences in the
axial vector contribution, which would be cancelled by the single unresolved limits of
the corresponding axial contributions to four-parton final states [14,15]. Results from the
four-parton final states show that this axial contribution is numerically very small [65].
The helicity coefficients contain HPLs and 2dHPLs up to weight 4 in the A, B, C, G-
terms, up to weight 3 in the D, E-terms (which do moreover contain only a limited
subset of purely planar master integrals) and up to weight 2 in the F -term. The size of
each helicity coefficient is comparable to the size of the helicity-averaged tree times two-
loop matrix element quoted in [18]. We do, therefore, only quote the A- and D-terms of
each coefficient, which form the leading colour contributions, and which turn out to be
numerically dominant, approximating the full expressions to an accuracy of about 20%.
The complete set of coefficients in FORM format can be obtained from the authors.
These leading colour terms are:
Aα (y, z)
1 2
= π − 13H(0; z) + 6H(1, 0; z) + 6G(1, 0; y)

48y(1 − z)
1 2
− π − 13H(0; z) + 6H(1, 0; z) + 6G(1, 0; y)

48y(1 − y − z)
z z z
− G(0; y) − +
16(1 − y)2 16(1 − y) 12(1 − y − z)2

5π 2
× − − 5H(0; z)G(0; y) − 5H(1, 0; z) + 5G(1, 0; y)
6

z 14π 2
+ − 11H(0; z) + 28H(0; z)G(0; y) + 28H(1, 0; z)
16(1 − y − z) 3

z2 11π 2
+ 11G(0; y) − 28G(1, 0; y) + + 11H(0; z)G(0; y)
16(1 − y − z) 2 6

1 1
+ 11H(1, 0; z) − 11G(1, 0; y) + G(0; y) +
3(1 − y) 48(1 − z)2

π2
× − + π 2 (3H(0; z) + 3H(1; z) − G(1 − z; y) + G(0; y)) + 6ζ3
6
355
− H(0; z) − 6H(0; z)G(1 − z, 0; y) + 10H(0; z)G(0; y) + 45H(0, 0; z)
6
+ 12H(0, 0; z)G(0; y) + 18H(0, 1, 0; z) − H(1, 0; z) − 6H(1, 0; z)G(1 − z; y)
+ 6H(1, 0; z)G(0; y) + 12H(1, 0, 0; z) + 18H(1, 1, 0; z) + 6G(1 − z, 1, 0; y)

1
− 6G(0, 1, 0; y) + π 2 (−8 + 9H(0; z) + 9H(1; z) − 3G(1 − z; y)
72(1 − z)
277
+ 3G(0; y)) + 18ζ3 − − 65H(0; z) − 18H(0; z)G(1 − z, 0; y)
4
+ 39H(0; z)G(0; y) + 81H(0, 0; z) + 36H(0, 0; z)G(0; y) + 54H(0, 1, 0; z)
− 48H(1, 0; z) − 18H(1, 0; z)G(1 − z; y) + 18H(1, 0; z)G(0; y)
+ 36H(1, 0, 0; z) + 54H(1, 1, 0; z) + 18G(1 − z, 1, 0; y) + 15G(0; y)

1
− 18G(0, 1, 0; y) − 9G(1, 0; y) + −2π 2 H(1; z)
48(1 − y − z)2

− π 2 G(0; y) + 12ζ3 − 6H(1, 0; z)G(0; y) − 12H(1, 1, 0; z) − 6G(0, 1, 0; y)

1
+ −4π 2 H(1; z) − 2π 2 G(0; y) + 24ζ3 − 13H(0; z)
48(1 − y − z)

− 12H(1, 0; z)G(0; y) − 24H(1, 1, 0; z) − 20G(0; y) − 12G(0, 1, 0; y)

π2 928
+ − − 5H(0; z) + 12H(0; z)G(1 − z; y) + 36H(0; z)G(0; y)
288 3
− 12H(0; z)G(1; y) + 24H(0, 1; z) + 24H(1; z)G(1 − z; y)
− 24H(1; z)G(−z; y) − 12H(1; z)G(1; y) + 24H(1, 0; z) + 12H(1, 1; z)

− 44G(1 − z; y) + 12G(1 − z, 0; y) − 24G(1 − z, 1; y) + 24G(−z, 1 − z; y)
− 24G(0, 1 − z; y) + 49G(0; y) − 24G(0, 1; y) + 12G(1, 1 − z; y)

ζ3
+ 44G(1; y) − 36G(1, 0; y) + 24G(1, 1; y) + 317 − 18H(0; z)
72

11π 4
+ 90H(1; z) − 72G(1 − z; y) − 18G(0; y) − 18G(1; y) +
360
1 89959 2149
+ − + H(0; z) − 66H(0; z)G(1 − z, 0; y)
72 144 12
− 18H(0; z)G(1 − z, 1, 0; y) + 36H(0; z)G(−z, 1 − z, 0; y)
− 36H(0; z)G(0, 1 − z, 0; y) − 66H(0; z)G(0; y) + 126H(0; z)G(0, 0; y)
− 18H(0; z)G(0, 1, 0; y) + 18H(0; z)G(1, 1 − z, 0; y) − 3H(0; z)G(1, 0; y)
23
− 36H(0; z)G(1, 0, 0; y) + H(0, 0; z) + 72H(0, 0; z)G(0; y) + 36H(0, 0; z)
2
× G(0, 0; y) + 72H(0, 0, 1, 0; z) + 3H(0, 1, 0; z) − 18H(0, 1, 0; z)G(1 − z; y)
+ 36H(0, 1, 0; z)G(−z; y) + 18H(0, 1, 0; z)G(0; y) − 18H(0, 1, 0; z)G(1; y)
+ 36H(0, 1, 1, 0; z) − 71H(1, 0; z) − 66H(1, 0; z)G(1 − z; y) + 18H(1, 0; z)
× G(1 − z, 0; y) + 36H(1, 0; z)G(−z, 1 − z; y) − 36H(1, 0; z)G(−z, 0; y)
− 36H(1, 0; z)G(0, 1 − z; y) + 96H(1, 0; z)G(0; y) + 18H(1, 0; z)
× G(1, 1 − z; y) − 18H(1, 0; z)G(1, 0; y) + 72H(1, 0, 0; z) + 36H(1, 0, 0; z)
× G(0; y) + 72H(1, 0, 1, 0; z) + 36H(1, 1, 0; z)G(1 − z; y) − 36H(1, 1, 0; z)
× G(−z; y) − 18H(1, 1, 0; z)G(1; y) + 36H(1, 1, 0, 0; z) + 18H(1, 1, 1, 0; z)
+ 18G(1 − z, 0, 1, 0; y) + 66G(1 − z, 1, 0; y) + 36G(1 − z, 1, 1, 0; y)
− 36G(−z, 1 − z, 1, 0; y) − 36G(−z, 0, 1, 0; y) + 36G(0, 1 − z, 1, 0; y)
49
+ G(0; y) + 160G(0, 0; y) − 36G(0, 0, 1, 0; y) − 30G(0, 1, 0; y)
3
+ 36G(0, 1, 1, 0; y) − 18G(1, 1 − z, 1, 0; y) + 71G(1, 0; y) − 126G(1, 0, 0; y)
+ 54G(1, 0, 1, 0; y) − 66G(1, 1, 0; y) + 36G(1, 1, 0, 0; y)

11 1
− 36G(1, 1, 1, 0; y) + iπ − H(0; z) + [−11
16(1 − z)2 16(1 − z)

1 44π 2 2345
− 22H(0; z)] + 2ζ3 + − − − 11H(0; z)
48 3 18

− 66H(0; z)G(0; y) − 66H(1, 0; z) − 110G(0; y) + 66G(1, 0; y) ,
Dα (y, z)
1 1 z
= H(0; z) − H(0; z) +
12y(1 − z) 12y(1 − y − z) 6(1 − y − z)2
2
π z π2
× + H(0; z)G(0; y) + H(1, 0; z) − G(1, 0; y) + −
6 4(1 − y − z) 3

+ H(0; z) − 2H(0; z)G(0; y) − 2H(1, 0; z) − G(0; y) + 2G(1, 0; y)

z2 π2
+ − − H(0; z)G(0; y) − H(1, 0; z) + G(1, 0; y)
4(1 − y − z)2 6

1 1 3
− G(0; y) + π 2 + 25H(0; z) − H(0; z)G(0; y)
12(1 − y) 72(1 − z)2 2

1 2
− 9H(0, 0; z) + 6H(1, 0; z) + 4π + 38 + 37H(0; z)
144(1 − z)

− 6H(0; z)G(0; y) − 36H(0, 0; z) + 24H(1, 0; z) − 3G(0; y)

1
π 2 395
+ H(0; z) + 2G(0; y) + − H(0; z) + 2G(1 − z; y)
12(1 − y − z) 72 12

19 1 3661
− G(0; y) − 2G(1; y) − ζ3 + − 25H(0; z) + 24H(0; z)
36 144 18
× G(1 − z, 0; y) + 29H(0; z)G(0; y) − 36H(0; z)G(0, 0; y) + 6H(0; z)
× G(1, 0; y) − 28H(0, 0; z) − 36H(0, 0; z)G(0; y) − 6H(0, 1, 0; z)
+ 40H(1, 0; z) + 24H(1, 0; z)G(1 − z; y) − 30H(1, 0; z)G(0; y)
− 36H(1, 0, 0; z) − 24G(1 − z, 1, 0; y) + 53G(0; y) − 82G(0, 0; y)

+ 6G(0, 1, 0; y) − 40G(1, 0; y) + 36G(1, 0, 0; y) + 24G(1, 1, 0; y) + iπ

1 1 1 8π 2 28
× H(0; z) + [1 + 2H(0; z)] + − + 13H(0; z)
8(1 − z)2 8(1 − z) 48 3 3

+ 12H(0; z)G(0; y) + 12H(1, 0; z) + 31G(0; y) − 12G(1, 0; y) ,
Aβ (y, z)

z z z 47π 2 47π 2
=− G(0; y) − + H(1; z) −
16(1 − y)2 16(1 − y) 16(y + z)2 3 3
× G(1 − z; y) − 94H(0; z)G(1 − z, 0; y) − 94H(0, 1, 0; z) − 99H(1, 0; z)
− 94H(1, 0; z)G(1 − z; y) + 94H(1, 0; z)G(0; y) + 94H(1, 1, 0; z)

z 47π 2
+ 94G(1 − z, 1, 0; y) + 94G(0, 1, 0; y) − 99G(1, 0; y) + −
16(y + z) 3
+ 11 + 44H(0; z) − 94H(0; z)G(0; y) − 94H(1, 0; z) − 55G(0; y)
2
z 5π π2 π2
+ 94G(1, 0; y) + − H(1; z) − G(0; y) + 3ζ3
12(1 − y − z)2 6 2 4
3
+ 5H(0; z)G(0; y) + 5H(1, 0; z) − H(1, 0; z)G(0; y) − 3H(1, 1, 0; z)
2

3 z 19π 2
− G(0, 1, 0; y) − 5G(1, 0; y) + − + 5H(0; z)
2 12(1 − y − z) 3

53 z2
− 38H(0; z)G(0; y) − 38H(1, 0; z) − G(0; y) + 38G(1, 0; y) +
4 8(y + z)3

× −11π H(1; z) + 11π G(1 − z; y) + 66H(0; z)G(1 − z, 0; y)
2 2
+ 66H(0, 1, 0; z) + 33H(1, 0; z) + 66H(1, 0; z)G(1 − z; y)

− 66H(1, 0; z)G(0; y) − 66H(1, 1, 0; z) − 66G(1 − z, 1, 0; y)

z2
− 66G(0, 1, 0; y) + 33G(1, 0; y) + +22π 2 − 33H(0; z)
16(y + z) 2

+ 132H(0; z)G(0; y) + 132H(1, 0; z) + 33G(0; y) − 132G(1, 0; y)

z2
+ 11π 2 − 11H(0; z) + 66H(0; z)G(0; y) + 66H(1, 0; z)
16(y + z)

z2 53π 2
+ 11G(0; y) − 66G(1, 0; y) + − − 53H(0; z)G(0; y)
48(1 − y − z) 2 6

z2
− 53H(1, 0; z) + 53G(1, 0; y) + 11π 2 − 11H(0; z)
16(1 − y − z)

+ 66H(0; z)G(0; y) + 66H(1, 0; z) + 11G(0; y) − 66G(1, 0; y)

z3 11π 2 11π 2
+ H(1; z) − G(1 − z; y) − 33H(0; z)G(1 − z, 0; y)
8(y + z)4 2 2
− 33H(0, 1, 0; z) − 33H(1, 0; z)G(1 − z; y) + 33H(1, 0; z)G(0; y)

+ 33H(1, 1, 0; z) + 33G(1 − z, 1, 0; y) + 33G(0, 1, 0; y)

z3 11π 2
+ − − 33H(0; z)G(0; y) − 33H(1, 0; z) + 33G(1, 0; y)
8(y + z)3 2

z3 11π 2
+ − − 33H(0; z)G(0; y) − 33H(1, 0; z) + 33G(1, 0; y)
16(y + z)2 2
3 2
z 11π
+ − − 11H(0; z)G(0; y) − 11H(1, 0; z) + 11G(1, 0; y)
8(y + z) 6
3
z 11π 2
+ + 11H(0; z)G(0; y) + 11H(1, 0; z) − 11G(1, 0; y)
16(1 − y − z)2 6

z3 11π 2
+ − − 11H(0; z)G(0; y) − 11H(1, 0; z) + 11G(1, 0; y)
8(1 − y − z) 6

5 1 1
+ G(0; y) + π2 − 3H(0; z) − 3H(1; z) + G(1 − z; y)
24(1 − y) 48(1 − z) 6

355
− G(0; y) − 6ζ3 + H(0; z) + 6H(0; z)G(1 − z, 0; y) − 10H(0; z)G(0; y)
6
− 45H(0, 0; z) − 12H(0, 0; z)G(0; y) − 18H(0, 1, 0; z) + H(1, 0; z)
+ 6H(1, 0; z)G(1 − z; y) − 6H(1, 0; z)G(0; y) − 12H(1, 0, 0; z)

1 7π 2
− 18H(1, 1, 0; z) − 6G(1 − z, 1, 0; y) + 6G(0, 1, 0; y) + −
8(y + z) 3
7π 2
× H(1; z) + G(1 − z; y) + 14H(0; z)G(1 − z, 0; y) + 14H(0, 1, 0; z)
3
+ 25H(1, 0; z) + 14H(1, 0; z)G(1 − z; y) − 14H(1, 0; z)G(0; y)

− 14H(1, 1, 0; z) − 14G(1 − z, 1, 0; y) − 14G(0, 1, 0; y) + 25G(1, 0; y)
2
1 π π2
+ H(1; z) + G(0; y) − 2ζ3 + H(1, 0; z)G(0; y)
8(1 − y − z) 2 3 6

1 13π 2
+ 2H(1, 1, 0; z) + G(0, 1, 0; y) + + 10H(0; z)G(0; y)
24(1 − y − z) 6

π2 946
+ 13H(1, 0; z) + 10G(0; y) − 7G(1, 0; y) + − − 5H(0; z)
288 3
+ 12H(0; z)G(1 − z; y) + 36H(0; z)G(0; y) − 12H(0; z)G(1; y)
+ 24H(0, 1; z) + 24H(1; z)G(1 − z; y) − 24H(1; z)G(−z; y)
− 12H(1; z)G(1; y) + 24H(1, 0; z) + 12H(1, 1; z) − 44G(1 − z; y)
+ 12G(1 − z, 0; y) − 24G(1 − z, 1; y) + 24G(−z, 1 − z; y)
− 24G(0, 1 − z; y) + 49G(0; y) − 24G(0, 1; y)

+ 12G(1, 1 − z; y) + 44G(1; y) − 36G(1, 0; y) + 24G(1, 1; y)
11π 4 ζ3
+ + 317 − 18H(0; z) + 90H(1; z) − 72G(1 − z; y) − 18G(0; y)
360 72

1 79987 1735
− 18G(1; y) + − + H(0; z) − 132H(0; z)G(1 − z, 0; y)
144 72 6
− 36H(0; z)G(1 − z, 1, 0; y) + 72H(0; z)G(−z, 1 − z, 0; y)
− 72H(0; z)G(0, 1 − z, 0; y) − 150H(0; z)G(0; y) + 252H(0; z)G(0, 0; y)
− 36H(0; z)G(0, 1, 0; y) + 36H(0; z)G(1, 1 − z, 0; y) − 6H(0; z)G(1, 0; y)
− 72H(0; z)G(1, 0, 0; y) + 23H(0, 0; z) + 144H(0, 0; z)G(0; y)
+ 72H(0, 0; z)G(0, 0; y) + 144H(0, 0, 1, 0; z) + 6H(0, 1, 0; z)
− 36H(0, 1, 0; z)G(1 − z; y) + 72H(0, 1, 0; z)G(−z; y)
+ 36H(0, 1, 0; z)G(0; y) − 36H(0, 1, 0; z)G(1; y)
+ 72H(0, 1, 1, 0; z) − 160H(1, 0; z) − 132H(1, 0; z)G(1 − z; y)
+ 36H(1, 0; z)G(1 − z, 0; y) + 72H(1, 0; z)G(−z, 1 − z; y) − 72H(1, 0; z)
× G(−z, 0; y) − 72H(1, 0; z)G(0, 1 − z; y) + 192H(1, 0; z)G(0; y)
+ 36H(1, 0; z)G(1, 1 − z; y) − 36H(1, 0; z)G(1, 0; y) + 144H(1, 0, 0; z)
+ 72H(1, 0, 0; z)G(0; y) + 144H(1, 0, 1, 0; z) + 72H(1, 1, 0; z)G(1 − z; y)
− 72H(1, 1, 0; z)G(−z; y) − 36H(1, 1, 0; z)G(1; y) + 72H(1, 1, 0, 0; z)

+ 36H(1, 1, 1, 0; z) + 36G(1 − z, 0, 1, 0; y) + 132G(1 − z, 1, 0; y)
+ 72G(1 − z, 1, 1, 0; y) − 72G(−z, 1 − z, 1, 0; y) − 72G(−z, 0, 1, 0; y)
8
+ 72G(0, 1 − z, 1, 0; y) + G(0; y) + 320G(0, 0; y) − 72G(0, 0, 1, 0; y)
3
− 60G(0, 1, 0; y) + 72G(0, 1, 1, 0; y) − 36G(1, 1 − z, 1, 0; y) + 160G(1, 0; y)
− 252G(1, 0, 0; y) + 108G(1, 0, 1, 0; y) − 132G(1, 1, 0; y)

11
+ 72G(1, 1, 0, 0; y) − 72G(1, 1, 1, 0; y) + iπ H(0; z) + 2ζ3
16(1 − z)

1 44π 2 1751
+ − − − 11H(0; z) − 66H(0; z)G(0; y) − 66H(1, 0; z)
48 3 18

− 110G(0; y) + 66G(1, 0; y) ,
Dβ (y, z)

z 4π 2 4π 2
= − H(1; z) + G(1 − z; y) + 8H(0; z)G(1 − z, 0; y)
4(y + z)2 3 3
+ 8H(0, 1, 0; z) + 9H(1, 0; z) + 8H(1, 0; z)G(1 − z; y) − 8H(1, 0; z)G(0; y)

− 8H(1, 1, 0; z) − 8G(1 − z, 1, 0; y) − 8G(0, 1, 0; y) + 9G(1, 0; y)
2
z 4π
+ − 1 − 4H(0; z) + 8H(0; z)G(0; y) + 8H(1, 0; z)
4(y + z) 3

z π2
+ 5G(0; y) − 8G(1, 0; y) + − − H(0; z)G(0; y)
6(1 − y − z)2 6
2
z 7π
− H(1, 0; z) + G(1, 0; y) + − 2H(0; z) + 14H(0; z)
12(1 − y − z) 3

z2 2
× G(0; y) + 14H(1, 0; z) + 5G(0; y) − 14G(1, 0; y) + π H(1; z)
2(y + z) 3
− π 2 G(1 − z; y) − 6H(0; z)G(1 − z, 0; y) − 6H(0, 1, 0; z) − 3H(1, 0; z)

− 6H(1, 0; z)G(1 − z; y) + 6H(1, 0; z)G(0; y) + 6H(1, 1, 0; z)

z2
+ 6G(1 − z, 1, 0; y) + 6G(0, 1, 0; y) − 3G(1, 0; y) + −2π 2
4(y + z) 2

+ 3H(0; z) − 12H(0; z)G(0; y) − 12H(1, 0; z) − 3G(0; y) + 12G(1, 0; y)

z2 2
+ −π + H(0; z) − 6H(0; z)G(0; y) − 6H(1, 0; z) − G(0; y)
4(y + z)
2

z2 5π
+ 6G(1, 0; y) + + 5H(0; z)G(0; y) + 5H(1, 0; z)
12(1 − y − z) 2 6

z2 2
− 5G(1, 0; y) + −π + H(0; z) − 6H(0; z)G(0; y)
4(1 − y − z)

z3 π2
− 6H(1, 0; z) − G(0; y) + 6G(1, 0; y) + − H(1; z)
2(y + z)4 2
π2
+ G(1 − z; y) + 3H(0; z)G(1 − z, 0; y) + 3H(0, 1, 0; z)
2
+ 3H(1, 0; z)G(1 − z; y) − 3H(1, 0; z)G(0; y) − 3H(1, 1, 0; z)

− 3G(1 − z, 1, 0; y) − 3G(0, 1, 0; y)
2
z3 π
+ + 3H(0; z)G(0; y) + 3H(1, 0; z) − 3G(1, 0; y)
2(y + z)3 2
2
z3 π
+ + 3H(0; z)G(0; y) + 3H(1, 0; z) − 3G(1, 0; y)
4(y + z)2 2
2
z3 π
+ + H(0; z)G(0; y) + H(1, 0; z) − G(1, 0; y)
2(y + z) 6

z3 π2
+ − − H(0; z)G(0; y) − H(1, 0; z) + G(1, 0; y)
4(1 − y − z)2 6
3 2
z π
+ + H(0; z)G(0; y) + H(1, 0; z) − G(1, 0; y)
2(1 − y − z) 6

1 1 3
− G(0; y) + −π 2 − 25H(0; z) + H(0; z)G(0; y)
12(1 − y) 72(1 − z) 2
2
1 π π2
+ 9H(0, 0; z) − 6H(1, 0; z) + H(1; z) − G(1 − z; y)
2(y + z) 6 6
− H(0; z)G(1 − z, 0; y) − H(0, 1, 0; z) − 2H(1, 0; z) − H(1, 0; z)G(1 − z; y)
+ H(1, 0; z)G(0; y) + H(1, 1, 0; z) + G(1 − z, 1, 0; y) + G(0, 1, 0; y)

1 π2
− 2G(1, 0; y) + − − H(0; z)G(0; y) − H(1, 0; z)
6(1 − y − z) 6

π 2 395
− G(0; y) + G(1, 0; y) + − H(0; z) + 2G(1 − z; y) − G(0; y)
72 12

19 1 2977
− 2G(1; y) − ζ3 + − 10H(0; z) + 24H(0; z)G(1 − z, 0; y)
36 144 18
+ 29H(0; z)G(0; y) − 36H(0; z)G(0, 0; y) + 6H(0; z)G(1, 0; y)
− 28H(0, 0; z) − 36H(0, 0; z)G(0; y) − 6H(0, 1, 0; z) + 40H(1, 0; z)
+ 24H(1, 0; z)G(1 − z; y) − 30H(1, 0; z)G(0; y) − 36H(1, 0, 0; z)
− 24G(1 − z, 1, 0; y) + 56G(0; y) − 82G(0, 0; y) + 6G(0, 1, 0; y)

1
− 40G(1, 0; y) + 36G(1, 0, 0; y) + 24G(1, 1, 0; y) + iπ − H(0; z)
8(1 − z)

1 8π 2 46
+ − + 13H(0; z) + 12H(0; z)G(0, y) + 12H(1, 0; z)
48 3 3

+ 31G(0; y) − 12G(1, 0; y) ,
Aγ (y, z)
1 2
=+ −π + 13H(0; z) − 6H(1, 0; z) − 6G(1, 0; y)

48y(1 − z)
1 2
+ π − 13H(0; z) + 6H(1, 0; z) + 6G(1, 0; y)

48y(1 − y − z)

z 11π 2 11π 2
+ − H(1; z) + G(1 − z; y) + 22H(0; z)G(1 − z, 0; y)
16(y + z)2 3 3
+ 22H(0, 1, 0; z) + 55H(1, 0; z) + 22H(1, 0; z)G(1 − z; y)
− 22H(1, 0; z)G(0; y) − 22H(1, 1, 0; z) − 22G(1 − z, 1, 0; y)

z 11π 2
− 22G(0, 1, 0; y) + 55G(1, 0; y) + − 11 − 22H(0; z)
16(y + z) 3

+ 22H(0; z)G(0; y) + 22H(1, 0; z) + 33G(0; y) − 22G(1, 0; y)
2
z π π2
+ H(1; z) + G(0; y) − 2ζ3 + H(1, 0; z)G(0; y)
8(1 − y − z)2 3 6

z 10π 2
+ 2H(1, 1, 0; z) + G(0, 1, 0; y) + + 13H(0; z)
48(1 − y − z) 3

+ 20H(0; z)G(0; y) + 20H(1, 0; z) + 20G(0; y) − 20G(1, 0; y)

z2 22π 2 22π 2
+ H(1; z) − G(1 − z; y) − 44H(0; z)G(1 − z, 0; y)
8(y + z)3 3 3
−44H(0, 1, 0; z) − 33H(1, 0; z) − 44H(1, 0; z)G(1 − z; y) + 44H(1, 0; z)
× G(0; y) + 44H(1, 1, 0; z) + 44G(1 − z, 1, 0; y) + 44G(0, 1, 0; y)

z2 44π 2
− 33G(1, 0; y) + − + 33H(0; z) − 88H(0; z)G(0; y)
16(y + z) 2 3

z2 22π 2
− 88H(1, 0; z) − 33G(0; y) + 88G(1, 0; y) + −
16(y + z) 3

+ 11H(0; z) − 44H(0; z)G(0; y) − 44H(1, 0; z) − 11G(0; y) + 44G(1, 0; y)
2
z2 5π
+ + 5H(0; z)G(0; y) + 5H(1, 0; z) − 5G(1, 0; y)
12(1 − y − z)2 6

z2 22π 2
+ − + 11H(0; z) − 44H(0; z)G(0; y) − 44H(1, 0; z)
16(1 − y − z) 3

z3 11π 2
− 11G(0; y) + 44G(1, 0; y) + − H(1; z)
8(y + z)4 2
11π 2
+ G(1 − z; y) + 33H(0; z)G(1 − z, 0; y) + 33H(0, 1, 0; z)
2
+ 33H(1, 0; z)G(1 − z; y) − 33H(1, 0; z)G(0; y) − 33H(1, 1, 0; z)

− 33G(1 − z, 1, 0; y) − 33G(0, 1, 0; y)

z3 11π 2
+ + 33H(0; z)G(0; y) + 33H(1, 0; z) − 33G(1, 0; y)
8(y + z)3 2
3
z 11π 2
+ + 33H(0; z)G(0; y) + 33H(1, 0; z) − 33G(1, 0; y)
16(y + z)2 2
3 2
z 11π
+ + 11H(0; z)G(0; y) + 11H(1, 0; z) − 11G(1, 0; y)
8(y + z) 6

z3 11π 2
+ − − 11H(0; z)G(0; y) − 11H(1, 0; z)
16(1 − y − z) 2 6

z3 11π 2
+ 11G(1, 0; y) + + 11H(0; z)G(0; y) + 11H(1, 0; z)
8(1 − y − z) 6

1 1
− 11G(1, 0; y) + π2 − 3H(0; z) − 3H(1; z) + G(1 − z; y)
48(1 − z) 2 6

355
− G(0; y) − 6ζ3 + H(0; z) + 6H(0; z)G(1 − z, 0; y) − 10H(0; z)G(0; y)
6
− 45H(0, 0; z) − 12H(0, 0; z)G(0; y) − 18H(0, 1, 0; z) + H(1, 0; z)
+ 6H(1, 0; z)G(1 − z; y) − 6H(1, 0; z)G(0; y) − 12H(1, 0, 0; z)

1 7
− 18H(1, 1, 0; z) − 6G(1 − z, 1, 0; y) + 6G(0, 1, 0; y) + π2 −
48(1 − z) 6

277 493
+ 3H(0; z) + 3H(1; z) − G(1 − z; y) + G(0; y) + 6ζ3 + − H(0; z)
6 6
− 6H(0; z)G(1 − z, 0; y) + 4H(0; z)G(0; y) + 45H(0, 0; z) + 12H(0, 0; z)
× G(0; y) + 18H(0, 1, 0; z) − 7H(1, 0; z) − 6H(1, 0; z)G(1 − z; y)
+ 6H(1, 0; z)G(0; y) + 12H(1, 0, 0; z) + 18H(1, 1, 0; z) + 6G(1 − z, 1, 0; y)

1 2
− 10G(0; y) − 6G(0, 1, 0; y) + 6G(1, 0; y) + −π
48(1 − y − z)

1 277
+ 13H(0; z) − 6H(1, 0; z) − 6G(1, 0; y) + π2 − + 23H(0; z)
48 6

+ 6H(0; z)G(0; y) + 6H(1, 0; z) + 10G(0; y) − 6G(1, 0; y)

11 11
11
+ iπ H(0; z) + 1 − H(0; z) − ,
16(1 − z)2 16(1 − z) 16
Dγ (y; z)
2
1 1 z π
=− H(0; z) + H(0; z) + H(1; z)
12y(1 − z) 12y(1 − y − z) 4(y + z)2 3
π2
− G(1 − z; y) − 2H(0; z)G(1 − z, 0; y) − 2H(0, 1, 0; z) − 5H(1, 0; z)
3
− 2H(1, 0; z)G(1 − z; y) + 2H(1, 0; z)G(0; y) + 2H(1, 1, 0; z)

z π2
+ 2G(1 − z, 1, 0; y) + 2G(0, 1, 0; y) − 5G(1, 0; y) + − +1
4(y + z) 3

+ 2H(0; z) − 2H(0; z)G(0; y) − 2H(1, 0; z) − 3G(0; y) + 2G(1, 0; y)

z π2
+ − − H(0; z) − 2H(0; z)G(0; y) − 2H(1, 0; z)
12(1 − y − z) 3

z2 2π 2 2π 2
− 2G(0; y) + 2G(1, 0; y) + − H(1; z) + G(1 − z; y)
2(y + z)3 3 3
+ 4H(0; z)G(1 − z, 0; y) + 4H(0, 1, 0; z) + 3H(1, 0; z)
+ 4H(1, 0; z)G(1 − z; y) − 4H(1, 0; z)G(0; y) − 4H(1, 1, 0; z)

− 4G(1 − z, 1, 0; y) − 4G(0, 1, 0; y) + 3G(1, 0; y)
2
z2 4π
+ − 3H(0; z) + 8H(0; z)G(0; y) + 8H(1, 0; z)
4(y + z) 2 3
2
z2 2π
+ 3G(0; y) − 8G(1, 0; y) + − H(0; z) + 4H(0; z)G(0; y)
4(y + z) 3

z2 π2
+ 4H(1, 0; z) + G(0; y) − 4G(1, 0; y) + − − H(0; z)
6(1 − y − z)2 6
2
z2 2π
× G(0; y) − H(1, 0; z) + G(1, 0; y) + − H(0; z)
4(1 − y − z) 3

z3
+ 4H(0; z)G(0; y) + 4H(1, 0; z) + G(0; y) − 4G(1, 0; y) +
2(y + z)4
2
π π2
× H(1; z) − G(1 − z; y) − 3H(0; z)G(1 − z, 0; y) − 3H(0, 1, 0; z)
2 2
− 3H(1, 0; z)G(1 − z; y) + 3H(1, 0; z)G(0; y) + 3H(1, 1, 0; z)

z3 π2
+ 3G(1 − z, 1, 0; y) + 3G(0, 1, 0; y) + − − 3H(0; z)G(0; y)
2(y + z) 3 2

z3 π2
− 3H(1, 0; z) + 3G(1, 0; y) + − − 3H(0; z)G(0; y)
4(y + z)2 2

z3 π2
− 3H(1, 0; z) + 3G(1, 0; y) + − − H(0; z)G(0; y)
2(y + z) 6
2
z3 π
− H(1, 0; z) + G(1, 0; y) + + H(0; z)G(0; y)
4(1 − y − z)2 6

z3 π2
+ H(1, 0; z) − G(1, 0; y) + − − H(0; z)G(0; y)
2(1 − y − z) 6

1
− H(1, 0; z) + G(1, 0; y) + −π 2 − 25H(0; z)
72(1 − z)2

3 1
+ H(0; z)G(0; y) + 9H(0, 0; z) − 6H(1, 0; z) + 2π 2 − 38
2 144(1 − z)

+ 65H(0; z) − 3H(0; z)G(0; y) − 18H(0, 0; z) + 12H(1, 0; z) + 3G(0; y)
1 19 5 1
− H(0; z) + − H(0; z) − G(0; y)
12(1 − y − z) 72 48 48

1 1
1
+ iπ − H(0; z) + −1 + H(0; z) + . (3.19)
8(1 − z)2 8(1 − z) 8
From the Ω (1),finite and Ω (2),finite, it is possible to recover the finite pieces of the
helicity-averaged tree times two-loop and one-loop squared matrix elements by squaring
(3.7):
Finite(2×0) (x, y, z)

(1 − y)(1 − y − z) (2),finite 1 − y (2),finite
= 8V R α (y, z) + β (y, z)
yz y

− γ (2),finite(y, z) + (y ↔ z) ,
Finite(1×1) (x, y, z)

(1 − y − z) 1
= 4V R (1 − y − z) + |α (1),finite(y, z)|2
yz 2

1−y−z z 1−y−z y
+ + |β (1),finite(y, z)|2 + + |γ (1),finite(y, z)|2
2 y 2 z

2 − 2z (1),finite
+ −3 + y + z + α (y, z)β ∗(1),finite(y, z)
y
− (1 − y − z)α (1),finite(y, z)γ ∗(1),finite(y, z)

− (1 + y + z)β (1),finite(y, z)γ ∗(1),finite(y, z) + (y ↔ z) . (3.20)
It is important to notice that (3.20) corresponds, by the very nature of the Weyl–van
der Waerden helicity formalism, to a scheme with external momenta and polarization
vectors in four dimensions (internal states are always taken to be d-dimensional), which is
sometimes called the ’t Hooft–Veltman scheme [28]. This scheme is different from the
conventional dimensional regularization used in [18], where all external momenta and
polarization vectors are d-dimensional. Nevertheless, one obtains from (3.20) the same
Finite(2×0)(x, y, z) as in Eq. (4.17) and Finite(1×1) (x, y, z) as in Eq. (4.25) of [18], since all
scheme-dependent terms are correctly accounted for by the finite contributions arising from
expanding the tree level and one-loop contributions in the renormalization and infrared
factorization formulae.
It should also be noted that only the O( 0 ) terms of Ω (1),finite contribute to
Finite(1×1)(x, y, z), terms subleading in are not required, since no term is multiplied
with a divergent factor. Comparing the size of these O( 0 ) terms (3.15) with the size of
Finite(1×1)(x, y, z) in [18], it becomes clear that the squared one-loop amplitude can be
evaluated much more elegantly by squaring the finite remainders of the helicity amplitudes
than by computing the squared matrix element.
4. Conclusions and outlook
In this paper, we have presented analytic formulae for the one- and two-loop virtual
helicity amplitudes to the process γ ∗ → q q̄g. These amplitudes have been derived by
defining projectors, which isolate the coefficients of the most general tensorial structure of
the matrix element at any order in perturbation theory. Once the general tensor is known,
the helicity amplitudes follow in a straightforward manner—they are linear combinations
of the tensor coefficients. We applied the projectors directly to the Feynman diagrams and
used the conventional approach of relating the ensuing tensor integrals to a basis set of
master integrals. This latter step is identical to that employed to evaluate the interference
of tree- and two-loop graphs in Ref. [18], apart from the fact that the projector is no longer
the tree-level amplitude. As anticipated, the finite remainder from the interference of tree-
and two-loop amplitudes can be reconstructed from the appropriate helicity amplitudes,
with the difference between treating the external states in d dimensions or four dimensions
being isolated in the infrared-singular terms.
The results presented here therefore complement the earlier calculation of the inter-
ference of tree- and two-loop graphs in Ref. [18]. Knowledge of the helicity amplitudes
allows additional information on the scattering process. In particular, observables that re-
quire knowledge of the polarization tensor of the virtual photon, such as oriented event
shapes in unpolarized e+ e− scattering or event shapes in polarized e+ e− scattering, can be
described at two-loop order.
Similar results can in principle be obtained for (2 + 1)-jet production in deep inelastic
ep scattering or (V + 1)-jet production in hadron–hadron collisions. However, the rather
different domains of convergence of the HPLs and 2dHPLs makes this a non-trivial task,
which is discussed in a separate paper [66]. Nevertheless, the helicity approach will provide
information on the direction of the decay leptons in (V + 1)-jet production (with or
without polarized protons). Determination of the polarized parton distribution functions in
polarized electron–proton scattering will also benefit from the knowledge of the two-loop
helicity amplitudes in the appropriate kinematic region.
Even though the evaluation of two-loop QCD matrix elements is becoming well
established, the virtual corrections form only part of a full NNLO calculation. They must
be combined with the one-loop corrections to γ ∗ → 4 partons [14–17], where one of the
partons becomes collinear or soft, as well as tree-level processes γ ∗ → 5 partons [11–13]
with two soft or collinear partons in a way that allows all of the infrared singularities to
cancel one another. This task is far from trivial, even though the factorization properties
of both the one-loop, one-unresolved-parton contribution [67–72] and the tree-level, two-
unresolved-parton contributions [73–76] have been studied. Early studies for the case of
photon-plus-one-jet final states in electron–positron annihilation in [77,78], which involves
both double radiation and single radiation from one-loop graphs, indicate the feasibility of
developing a numerical NNLO program implementing the experimental definition of jet
observables and event-shape variables, and significant progress is anticipated in the near
future.
Note added
After this paper was first released, part of its results were confirmed in an independent
calculation using the methods described in [46,47]. In hep-ph/0207043, Moch, Uwer and
Weinzierl obtain results for the full one-loop amplitude (3.14) and for the contributions to
the two-loop amplitude (3.16) which are proportional to NF (i.e. the terms DΩ and EΩ ),
all in agreement with the results presented here.
Acknowledgements
EWNG thanks Adrian Signer for useful discussions. This work was 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 (DG 12-MIHT).
Appendix A. Weyl–van der Waerden spinor calculus
The basic quantity is the two-spinor ψA or ψ A and its complex conjugate ψȦ or ψ Ȧ .
Raising and lowering of indices is done with the antisymmetric tensor ε,

0 1
εAB = εAB = εȦḂ = εȦḂ = . (A.1)
−1 0
We define a antisymmetric spinorial “inner product”:
ψ1 ψ2 = ψ1A εBA ψ2B = ψ1A ψ2A = −ψ1A ψ2A = −ψ2 ψ1 , (A.2)
and
ψ1 ψ2 ∗ = ψ1Ȧ ψ2Ȧ . (A.3)

Any momentum vector kµ gets a bispinor representation by contraction with σ µ :

µ k0 + k3 k1 + ik2
kȦB = σȦB kµ = , (A.4)
k1 − ik2 k0 − k3
where σ 0 is the unit matrix and σi are the Pauli matrices. Since
µ
σȦB σ ν ȦB = 2g µν , (A.5)
we have
kȦB pȦB = 2k · p. (A.6)
For light-like vectors one can show that
kȦB = kȦ kB , (A.7)
where
√
(k1 −√
ik2 )/ k0 − k3
kA = , (A.8)
k0 − k3
so that for light-like vectors we have
2k · p = kpkp∗ = |kp|2 . (A.9)
The following relation is often useful:
µ
σȦB σµĊD = 2δȦ Ċ δB D . (A.10)
For massless spin- 21 particles the four-spinors can be expressed in two-spinors as

follows:

pB
u+ (p) = v− (p) = ,
0

0
u− (p) = v+ (p) = ,
pḂ
ū+ (q) = v̄− (q) = (0, −iqȦ ),
ū− (q) = v̄+ (q) = (iq A , 0). (A.11)
The Dirac γ matrices now become
µ
0 −iσḂA
γ =
µ
, (A.12)
iσ µȦB 0
so that, for example:
ū+ (q)γ µ v− (p) = qȦ σ µȦB pB . (A.13)

The general electroweak vertex for vector boson V coupling to two fermions is denoted
Vf f
by ieδij Γµ 1 2 , where i and j are the colour labels associated with the fermions f1 and f2
respectively. The vertex contains left- and right-handed couplings,

1 − γ5 1 + γ5
ΓµV ,f1 f2 = LVf1 f2 γµ + RfV1 f2 γµ , (A.14)
2 2
where for a photon,
γ γ
Lf1 f2 = Rf1 f2 = −ef1 δf1 f2 , (A.15)
and for a Z-boson,

f
I3 1 − sin2 θW ef1 − sin θW ef1
f1 f2 =
LZ δ f1 f2 , RfZ1 f2 = δ f1 f2 . (A.16)
sin θW cos θW cos θW
f
Here, ef represents the fractional electric charge, I3 the weak isospin and θW the weak
V ,f f
mixing angle. In the Weyl–van der Waerden notation, the vertex Γµ 1 2 becomes,

0 −iLVf1 f2 σµḂA
ΓµV ,f1 f2 = . (A.17)
iRfV1 f2 σµȦB 0
For the polarization vectors of outgoing gluons and photons we use the spinorial
quantities
+
√ kȦ bB
eȦB (k) = 2 , (A.18)
bk
−
√ b kB
eȦB (k) = 2 Ȧ ∗ . (A.19)
bk
The gauge spinor b is arbitrary and can be chosen differently in each gauge-invariant
expression. A suitable choice can often simplify the calculation.
References
[1] TASSO Collaboration D.P. Barber et al., Phys. Rev. Lett. 43 (1979) 830;
P. Söding, B. Wiik, G. Wolf, S.L. Wu, Talks given at Award Ceremony of the 1995 EPS High Energy and
Particle Physics Prize, Proceedings of the EPS High Energy Physics Conference, Brussels, 1995, World
Scientific, Singapore, 1996, p. 3.
[2] J. Ellis, M.K. Gaillard, G.G. Ross, Nucl. Phys. B 111 (1976) 253;
J. Ellis, M.K. Gaillard, G.G. Ross, Nucl. Phys. B 130 (1977) 516(E).
[3] R.K. Ellis, D.A. Ross, A.E. Terrano, Nucl. Phys. B 178 (1981) 421.
[4] K. Fabricius, I. Schmitt, G. Kramer, G. Schierholz, Z. Phys. C 11 (1981) 315.
[5] Z. Kunszt, P. Nason, in: Z Physics at LEP 1, in: CERN Yellow Report 89-08, Vol. 1, 1989, p. 373.
[6] W.T. Giele, E.W.N. Glover, Phys. Rev. D 46 (1992) 1980.
[7] S. Catani, M.H. Seymour, Nucl. Phys. B 485 (1997) 291;
S. Catani, M.H. Seymour, Nucl. Phys. B 510 (1997) 503(E), hep-ph/9605323.
[8] S. Bethke, J. Phys. G 26 (2000) R27, hep-ex/0004021.
[9] R.D. Heuer, D.J. Miller, F. Richard, P.M. Zerwas (Eds.), TESLA Technical Design Report Part III: Physics
at an e+ e− Linear Collider, DESY-report 2001-011, hep-ph/0106315.
[10] Z. Kunszt (Ed.), Proceedings of the Workshop on New Techniques for Calculating Higher Order QCD
Corrections, Zürich, 1992, ETH-TH/93-01.
[11] K. Hagiwara, D. Zeppenfeld, Nucl. Phys. B 313 (1989) 560.
[12] F.A. Berends, W.T. Giele, H. Kuijf, Nucl. Phys. B 321 (1989) 39.
[13] N.K. Falck, D. Graudenz, G. Kramer, Nucl. Phys. B 328 (1989) 317.
[14] Z. Bern, L.J. Dixon, D.A. Kosower, S. Weinzierl, Nucl. Phys. B 489 (1997) 3, hep-ph/9610370.
[15] Z. Bern, L.J. Dixon, D.A. Kosower, Nucl. Phys. B 513 (1998) 3, hep-ph/9708239.
[16] E.W.N. Glover, D.J. Miller, Phys. Lett. B 396 (1997) 257, hep-ph/9609474.
[17] J.M. Campbell, E.W.N. Glover, D.J. Miller, Phys. Lett. B 409 (1997) 503, hep-ph/9706297.
[18] L.W. Garland, T. Gehrmann, E.W.N. Glover, A. Koukoutsakis, E. Remiddi, Nucl. Phys. B 627 (2002) 107,
hep-ph/0112081.
[19] P.N. Burrows, P. Osland, Phys. Lett. B 400 (1997) 385, hep-ph/9701424.
[20] Z. Bern, A. De Freitas, L.J. Dixon, JHEP 0109 (2001) 037, hep-ph/0109078.
[21] Z. Bern, A. De Freitas, L.J. Dixon, A. Ghinculov, H.L. Wong, JHEP 0111 (2001) 031, hep-ph/0109079.
[22] T. Binoth, E.W.N. Glover, P. Marquard, J.J. van der Bij, JHEP 0205 (2002) 060, hep-ph/0202266.
[23] Z. Bern, L.J. Dixon, D.A. Kosower, JHEP 0001 (2000) 027, hep-ph/0001001.
[24] Z. Bern, A. De Freitas, L. Dixon, JHEP 0203 (2002) 018, hep-ph/0201161.
[25] E.W.N. Glover, C. Oleari, M.E. Tejeda-Yeomans, Nucl. Phys. B 605 (2001) 467, hep-ph/0102201.
[26] C.G. Bollini, J.J. Giambiagi, Nuovo Cimento B 12 (1972) 20.
[27] G.M. Cicuta, E. Montaldi, Lett. Nuovo Cimento 4 (1972) 309.
[28] G. ’t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.
[29] F.V. Tkachov, Phys. Lett. B 100 (1981) 65.
[30] K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
[31] T. Gehrmann, E. Remiddi, Nucl. Phys. B 580 (2000) 485, hep-ph/9912329.
[32] S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033.
[33] V.A. Smirnov, Phys. Lett. B 460 (1999) 397, hep-ph/9905323.
[34] J.B. Tausk, Phys. Lett. B 469 (1999) 225, hep-ph/9909506.
[35] V.A. Smirnov, O.L. Veretin, Nucl. Phys. B 566 (2000) 469, hep-ph/9907385.
[36] C. Anastasiou, T. Gehrmann, C. Oleari, E. Remiddi, J.B. Tausk, Nucl. Phys. B 580 (2000) 577, hep-
ph/0003261.
[37] T. Gehrmann, E. Remiddi, Nucl. Phys. B (Proc. Suppl.) 89 (2000) 561, hep-ph/0005232.
[38] C. Anastasiou, J.B. Tausk, M.E. Tejeda-Yeomans, Nucl. Phys. B (Proc. Suppl.) 89 (2000) 262, hep-
ph/0005328.
[39] Z. Bern, L. Dixon, A. Ghinculov, Phys. Rev. D 63 (2001) 053007, hep-ph/0010075.
[40] C. Anastasiou, E.W.N. Glover, C. Oleari, M.E. Tejeda-Yeomans, Nucl. Phys. B 601 (2001) 318, hep-
ph/0010212;
C. Anastasiou, E.W.N. Glover, C. Oleari, M.E. Tejeda-Yeomans, Nucl. Phys. B 601 (2001) 347, hep-
ph/0011094.
[41] C. Anastasiou, E.W.N. Glover, C. Oleari, M.E. Tejeda-Yeomans, Nucl. Phys. B 605 (2001) 486, hep-
ph/0101304.
[42] C. Anastasiou, E.W.N. Glover, M.E. Tejeda-Yeomans, Nucl. Phys. B 629 (2002) 255, hep-ph/0201274.
[43] T. Gehrmann, E. Remiddi, Nucl. Phys. B 601 (2001) 248, hep-ph/0008287;
T. Gehrmann, E. Remiddi, Nucl. Phys. B 601 (2001) 287, hep-ph/0101124.
[44] T. Binoth, G. Heinrich, Nucl. Phys. B 585 (2000) 741, hep-ph/0004013.
[45] V.A. Smirnov, Phys. Lett. B 491 (2000) 130, hep-ph/007032;
V.A. Smirnov, Phys. Lett. B 500 (2001) 330, hep-ph/0011056.
[46] S. Moch, P. Uwer, S. Weinzierl, J. Math. Phys. 43 (2002) 3363, hep-ph/0110083.
[47] S. Weinzierl, Comput. Phys. Commun. 145 (2002) 357, math-ph/0201011.
[48] E. Remiddi, J.A.M. Vermaseren, Int. J. Mod. Phys. A 15 (2000) 725, hep-ph/9905237.
[49] N. Nielsen, Nova Acta Leopoldiana (Halle) 90 (1909) 123.
[50] K.S. Kölbig, J.A. Mignaco, E. Remiddi, BIT 10 (1970) 38.
[51] T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 141 (2001) 296, hep-ph/0107173.
[52] T. Gehrmann, E. Remiddi, Comput. Phys. Commun. 144 (2002) 200, hep-ph/0111255.
[53] S. Catani, Phys. Lett. B 427 (1998) 161, hep-ph/9802439.
[54] MAPLE V Release 7, Copyright 2001 by Waterloo Maple Inc.
[55] J.A.M. Vermaseren, Symbolic Manipulation with FORM, Version 2, CAN, Amsterdam, 1991.
[56] J.A.M. Vermaseren, math-ph/0010025.
[57] R.J. Gonsalves, Phys. Rev. D 28 (1983) 1542.
[58] G. Kramer, B. Lampe, Z. Phys. C 34 (1987) 497;
G. Kramer, B. Lampe, Z. Phys. C 42 (1989) 504(E).
[59] T. Matsuura, W.L. van Neerven, Z. Phys. C 38 (1988) 623.
[60] T. Matsuura, S.C. van der Maarck, W.L. van Neerven, Nucl. Phys. B 319 (1989) 570.
[61] R.V. Harlander, Phys. Lett. B 492 (2000) 74, hep-ph/0007289.
[62] H. Weyl, Gruppentheorie und Quantenmechanik, Leipzig, 1928;
B.L. van der Waerden, Göttinger Nachrichten 100, 1929.
[63] F.A. Berends, W. Giele, Nucl. Phys. B 294 (1987) 700.

[64] B.A. Kniehl, J.H. Kühn, Phys. Lett. B 224 (1989) 229.
[65] L.J. Dixon, A. Signer, Phys. Rev. Lett. 78 (1997) 811, hep-ph/9609460;
L.J. Dixon, A. Signer, Phys. Rev. D 56 (1997) 4031, hep-ph/9706285.
[66] T. Gehrmann, E. Remiddi, Nucl. Phys. B 640 (2002) 379, hep-ph/0207020.
[67] Z. Bern, L.J. Dixon, D.C. Dunbar, D.A. Kosower, Nucl. Phys. B 425 (1994) 217, hep-ph/9403226.
[68] D.A. Kosower, Nucl. Phys. B 522 (1999) 319, hep-ph/9901201.
[69] D.A. Kosower, P. Uwer, Nucl. Phys. B 563 (1999) 477, hep-ph/9903515.
[70] Z. Bern, V. Del Duca, C.R. Schmidt, Phys. Lett. B 445 (1998) 168, hep-ph/9810409.
[71] Z. Bern, V. Del Duca, W.B. Kilgore, C.R. Schmidt, Phys. Rev. D 60 (1999) 116001, hep-ph/9903516.
[72] S. Catani, M. Grazzini, Nucl. Phys. B 591 (2000) 435, hep-ph/0007142.
[73] J.M. Campbell, E.W.N. Glover, Nucl. Phys. B 527 (1998) 264, hep-ph/9710255.
[74] S. Catani, M. Grazzini, Phys. Lett. B 446 (1999) 143, hep-ph/9810389;
S. Catani, M. Grazzini, Nucl. Phys. B 570 (2000) 287, hep-ph/9908523.
[75] F.A. Berends, W.T. Giele, Nucl. Phys. B 313 (1989) 595.
[76] S. Catani, in [10].
[77] A. Gehrmann-De Ridder, T. Gehrmann, E.W.N. Glover, Phys. Lett. B 414 (1997) 354, hep-ph/9705305.
[78] A. Gehrmann-De Ridder, E.W.N. Glover, Nucl. Phys. B 517 (1998) 269, hep-ph/9707224.
The systematic study of B → π form factors in

pQCD approach and its reliability
Zheng-Tao Wei a,b , Mao-Zhi Yang a,c
a CCAST (World Laboratory), P.O. Box 8730, Beijing 100080, PR China
b Institute of Theoretical Physics, P.O. Box 2735, Beijing 100080, PR China
c Institute of High Energy Physics, P.O. Box 918(4), Beijing 100039, PR China
Received 7 February 2002; received in revised form 26 June 2002; accepted 26 July 2002
Abstract
The study of exclusive B decays in perturbative QCD are complicated by the endpoint problem.
In order to perform the perturbative calculation, the Sudakov effects are introduced to regulate
the endpoint singularity. We provide a systematic analysis with leading and next-to-leading twist
corrections for B → π form factors in pQCD approach. The intrinsic transverse momentum
dependence of hadronic wave function and threshold resummation effects are included in pQCD
approach. There are two leading twist B meson distribution amplitudes (or generally wave functions)
in general. The QCD equations of motion provide important constraints on B meson wave functions.
The reliability of pQCD approach in B → π form factors is discussed. 70% of the result comes from
the region αs (t)/π < 0.2 and 38% comes from the region where the momentum transfer t 1 GeV.
The conceptual problems of pQCD approach are discussed in brief. Our conclusion is that pQCD
approach in the present form cannot provide a precise prediction for B → π transition form factors.
PACS: 13.25.Hw; 12.38.Bx
1. Introduction
The exclusive B decays provide an important test of the standard model of particle
physics. With the running of BaBar and Belle B-factories and the proceeding of future B
physics projects (BTeV and LHCb, etc.), a large amount of B mesons will be accumulated
to explore the origin of CP violation and determine the CKM parameters, such as the
E-mail addresses: weizt@itp.ac.cn (Z.-T. Wei), yangmz@mail.ihep.ac.cn (M.-Z. Yang).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 2 3 - 5
264 Z.-T. Wei, M.-Z. Yang / Nuclear Physics B 642 (2002) 263–289
angles α, β and γ in unitary triangle. In most cases, the limitation of our theoretical
ability prevents the precise prediction for exclusive B decays so we have to refer to some
phenomenological approaches.
The large B meson mass mB establishes a large scale so that perturbative QCD
(pQCD) may be applicable in exclusive B decays. Recently, two different approaches
based on perturbative QCD were proposed to calculate the exclusive B decays. One
approach, usually say QCD factorization approach, states that two body hadronic B
decays can factorize and the relevant amplitude can be written as the convolutions of non-
perturbative quantities (the light-cone distribution amplitudes of mesons and semi-leptonic
form factors) and perturbatively calculable hard scattering kernels [1]. The other approach
is the modified pQCD, or say pQCD approach for simplicity. In pQCD approach, the semi-
leptonic form factors for B → P transition (where P represents light meson) are claimed
to be perturbatively calculable [2].
The central difference between these two approaches is whether B → P transition
form factors are perturbatively calculable, or specifically, whether Sudakov effects can
cure endpoint singularity. In B → P transition form factors, the endpoint singularity
is generated even at leading twist. In pQCD approach, the transverse momentum are
retained to regulate the endpoint singularity and Sudakov double logarithm corrections are
included to suppress the long distance contributions from configuration of large transverse
separation. In [3], the authors investigated the reliability of Sudakov effects in B → π form
factors. Their conclusion is that Sudakov suppression are so weak that they cannot be
applied in B decays. If their criticism is right, it would bring the entire thought about
Sudakov effects into crisis.
In our previous study [4], we investigated the Sudakov effects in QCD factorization
approach. Our conclusion is contrary to the above criticism. Sudakov effects play an
important role in exclusive B decays. It is well known that the standard approach or say
BLER approach [5] is questionable at experimentally accessible energy scales, typically
a few GeV region [6]. This question also occurs in the hard spectator scattering in
B → P1 P2 decays. The QCD factorization approach gives a finite result at leading twist.
Carefully study shows that there are large contributions coming from the region where
the momentum transfers are small. The introduction of Sudakov suppression enlarges
the application range of pQCD and makes it self-consistent [7]. Although the relevant
numerical result is power suppressed, the Sudakov effects is potentially important. We
disagree with the comment that the main purpose of introduction of Sudakov suppression
is to improve the computation accuracy [3]. At twist-3 level, the chirally enhanced
power correction is logarithmically divergent, so the QCD factorization approach cannot
apply. A phenomenological parametrization method is designed to regulate the endpoint
singularity [8]. This method is not self-consistent and it introduces an arbitrary infrared
cutoff. Sudakov suppression establish a natural infrared cutoff. Obviously, pQCD approach
has the advantage of the parametrization method.
The disagreement of conclusions in [2,4] and [3] motivates us to reconsider the
reliability of pQCD approach in B → π form factors. In this paper, we present a systematic
study of B → π form factors in pQCD approach and examine the reliability of pQCD
calculation. Compare with the recent pQCD analysis of B → π form factors [9], the
present analysis contains three new theoretical ingredients:
Z.-T. Wei, M.-Z. Yang / Nuclear Physics B 642 (2002) 263–289 265
• The intrinsic transverse momentum dependence of wave functions for B and pion
mesons are included. The importance of intrinsic transverse momentum dependence
is first pointed out in pion form factor in [10]. Because Sudakov suppression is not
severely strong for real mB , the effect of intrinsic transverse momentum dependence
of wave function cannot be neglected.
• B meson contains two leading wave functions. The assumption of single B meson
wave function in the previous pQCD analysis is not valid. Equations of motion in
HQET provide important constraint on the choice of B meson wave functions.
• The threshold resummation effect for B meson is taken into account. The perturbative
analysis depends on the endpoint behavior of B meson distribution amplitudes. The jet
function obtained from threshold resummation suppress the endpoint contribution, so
it modifies the power behavior of endpoint contribution. We expect that it can improve
the perturbative result since one of the two B meson distribution amplitudes does not
vanish at endpoint.
In addition, the chirally-enhanced power corrections (twist-3 contribution) are included

in our analysis. It is noted that the contribution of chirally-enhanced power corrections is
comparable with or even larger than leading twist contribution [9]. Our study confirm it.
The numerical results of B → π form factors at large recoil region in pQCD approach are
consistent with derivations of QCD sum rules [11].
The remainder of the paper is organized as follows. Section 2 introduces the theoretical
ingredients of pQCD approach. In Section 3, we present the formulas for B → π form
factors in pQCD approach, perform the numerical analysis, and examine the reliability of
pQCD approach. Finally, in Section 4, the conclusions and discussions are presented.
2. The B → π form factors in pQCD approach
B → π form factors are the basic non-perturbative parameters in semi-leptonic B →

πlν decays and exclusive, nonleptonic decays such as B → ππ, πK decays in QCD
factorization approach. While in pQCD approach, it can be perturbatively calculated from
the universal hadronic distribution amplitudes or wave functions. In this section, we will
give a general discussion of B → π form factors and introduce the main ingredients of
pQCD approach.
The form factors of B → π are defined by the following Lorentz decompositions of
biquark current matrix elements:

π(Pπ )ūγµ b B(PB )

m2 − m2 m2 − m2
= PB + Pπ − B 2 π q µ F+Bπ q 2 + B 2 π qµ F0Bπ q 2 , (1)
q q
where q = PB − Pπ is the momentum carried by lepton pair in semi-leptonic B → πlν
decays. At large recoil limit, q 2 = 0, the two form factors F+Bπ , F0Bπ reduce to one
parameter, F+Bπ (0) = F0Bπ (0).
First, we give our conventions on kinematics. We work in the rest frame of B meson.
The mass difference of b quark and B meson is negligible in the heavy quark limit and
Fig. 1. Lowest order hard-scattering kernel for Bπ form factor.
we take approximation mb = mB in our calculation. The masses of light quarks u, d and

π meson are also neglected. For discussion, the momentum is described in terms of light-
k+ √ −
cone variables. We take k = ( √ , k , k ⊥ ) with k ± = k 0 ± k 3 and k ⊥ = (k 1 , k 2 ). The scalar
2 2
+ − −B+
product of two arbitrary vectors A and B is A · B = Aµ B µ = A B +A 2 − A ⊥ · B ⊥ .
The momentum of pion is chosen to be in the – direction. Under these conventions,
⊥ ), Pπ = (0, η m ⊥ ), P̄π = (η m ⊥ ) with η̄ = 1 − η = q 2 . We
2
PB = ( m√B , m
√B , 0 √B , 0 √B , 0, 0
2 2 2 2 m
√ √ B
define two light-like vectors n+ ≡ ( 2, 0, 0 ⊥ ) and n+ ≡ (0, 2, 0 ⊥ ).
We denote ξ as the momentum fraction of spectator antiquark in B meson, and x as the
momentum fraction of the antiquark in pion. As plotted in Fig. 1.
2.1. The physical picture of factorization in B → π form factors
The most important ingredient of pQCD is factorization, i.e., the separation of the long-
distance dynamics from the short distance dynamics which is pertubatively calculable.
Although rigorous proof of the factorization theorem is technically intricate, the physical
picture is simple and intuitive.
First, we illustrate the factorization of pion electromagnetic form factor in order to
introduce the basic idea of standard approach and the modified approach. A highly virtual
photon collides on a quark in initial pion and makes it change its direction. In exclusive
process γ ∗ π → π , the simple power counting shows that at large momentum transfers, the
valence quarks dominate the process and the higher Fock states and the intrinsic transverse
momentum is power suppressed. The large momentum Q means the high resolution at the
distance of 1/Q. In a small distance region, a parton only sees the other parton relative
to it and the initial and final pion can be considered as a small-size color-singlet object
during the hard interaction. What is important for small-size color-singlet pion is that
soft gluon corrections cancel at the leading twist. This is the well-known phenomena,
“color transparency”. The long-distance interactions only occur before and after the hard
interaction, so their effects can be factorized into the initial and final pions respectively.
Since the hard interaction is restricted in the short distance, we only need the probability
for the q q̄ pair in pion to be within the transverse distance of 1/Q, in other words, the
distribution amplitude of pion φ(x, Q2 ). The scale Q acts as factorization scale as well as
renormalization scale. The above discussions can be grouped into the standard factorization
formula for electromagnetic form factor of pion

Fπ = dx dy φ x, Q2 T (x, y, Q)φ y, Q2 . (2)
The standard factorization is proved successfully in the asymptotic limit. The applica-
tion of this standard approach at the experimentally accessible energy scales, typically a
few GeV region, is criticized in [6]. The authors pointed out that the perturbative calcu-
lations for electromagnetic form factor have a large amount of contributions coming from
soft endpoint region (x, y → 0) where the perturbative analysis is invalid. This is the well-
known “endpoint” problem even there is no endpoint singularity in the convolution of total
amplitude. The Sudakov effects are introduced to modify the endpoint behavior and make
pQCD applicable in few GeV region [7]. The basic idea of this modified approach, or say
pQCD approach is using the mechanism of Sudakov suppression to suppress long-distance
contributions of large transverse separations. Sudakov suppression establishes a factor-
ization scale 1/b in addition to the momentum scale Q. The pion with small transverse
separation has small color dipole moment so that QCD factorization is revised. In pQCD
approach, Sudakov suppression plays a crucial role.
Second, we discuss the factorization in B → π form factors in the large mB limit.
B meson is a heavy-light system. It has large size in longitudinal as well as transverse
directions. If the soft B meson wave functions overlap with the wave functions of final
meson, the separation of long-distance and short distance dynamics is impossible. From
this point, we conclude that B → D (∗) from factors in heavy quark limit are soft dominant
and cannot be calculated by pQCD. For B → π form factors, the situation is different.
The pion carries the energy of m2B when q 2 = 0. When the fractional momentum of the
antiquark in pion is far away from the endpoint region, i.e., x 0, the pion is highly
restricted in longitudinal direction because of Lorentz contraction. The soft gluons that
attach to small size color-singlet pion decouple in the large mB limit. So factorization
is applicable. Simply to say, the soft spectator antiquark in B meson must undergo hard
strong interaction to change into the fast moving parton in pion.
The above perturbative picture is destroyed by the endpoint contribution where
antiquark in pion carries the momentum of order ΛQCD . In the standard approach, B →
π form factor is logarithmically divergent at leading twist (twist-2 for our case). The
divergence becomes more serious for next-to-leading twist (twist-3) contribution and the
linear divergence occurs. The origin of this divergence is that the distribution amplitude
does not provide enough suppression at endpoint. In order to apply the perturbative
picture, the soft endpoint contribution must be suppressed. As we have discussed for pion
electromagnetic form factor, Sudakov effects can provide such suppression. With the help
of Sudakov suppression, pQCD factorization can be applicable. In pQCD approach, the
B → π form factors can be generally written as convolutions of wave functions and hard
scattering kernels

F+,0 = dξ dx d 2 b B d 2 b π ΨB (ξ, bB , µ)Ψπ (x, bπ , µ)T+,0 (ξ, x, bB , bπ , µ).
Bπ
(3)
2.2. Pion wave functions
In the standard approach, the transverse momentum is assumed power suppressed and
neglected. When the two valence quarks in pion carry large longitudinal momentum,
i.e., the momentum fractions x, x̄ 0, the transverse momentum is power suppressed
compared to longitudinal momentum. In the endpoint region, the transverse momentum is
important and cannot be neglected. In pQCD approach, transverse momentum is retained
in hard scattering kernels and the non-perturbative parameters are hadronic wave functions
in general.
The two valence quarks in pion have transverse momentum as well as the longitudinal
momentum. Compared to the collinear limit, the momentum of the quarks in pion (with
momentum Pπ ) changes to
k1 = xpπ + k⊥ , k2 = x̄pπ − k⊥ , (4)
where x and x̄ denote the longitudinal momentum fractions of quark and antiquark,
respectively. For our case, the meson is on-shell and the partons are slightly off-shell.
2 which is power suppressed compared
The off-shellness of the parton is proportional to k⊥
2
to mB .
As we have discussed above, the intrinsic transverse momentum dependence of wave
functions of pion and B meson cannot be neglected because Sudakov suppression is
not strong in the real mB energy. The distribution amplitude can be obtained from the
integration of wave function over the transverse momentum

φ(x) = d 2 k⊥ Ψ (x, k⊥ ), (5)
|k⊥ |<µ
where µ is the ultraviolet cutoff.
Similar to the definition of distribution amplitudes with leading and next-to-leading
twist [12], the light-cone wave functions of pion are defined in terms of bilocal operator
matrix element
π(p)|q̄β (z)qα |0
1
ifπ
= dx d 2 k ⊥ ei(xp·z−k⊥· z⊥ )
4
0

Ψσ (x, k⊥ )
× p
/ γ5 Ψπ (x, k⊥ ) − µπ γ5 Ψp (x, k⊥ ) − σµν pµ zν , (6)
6 αβ
where fπ is the decay constant of pion. The parameter µπ = m2π /(mu + md ) for charged
pion. For neutral pion, we use the same parameter µπ as the charged pion. Ψπ , Ψp and
Ψσ are the twist-2 and twist-3 wave functions, respectively. The twist-3 wave functions
contributes power corrections. But at mB energy scale, the chirally enhanced parameter
µπ
rπ = m B
∼ O(1) is not small. So the twist-3 contribution should be considered in B decays.
In order to perform the analysis in pQCD approach, we need to transform the parameters
in terms of coordinate variable in Eq. (6) into the momentum space configuration. We use
the momentum projection given in [13]:

ifπ pµ p̄ν Ψσ Ψσ ∂

Mαβπ
= p/ γ5 Ψπ − µP γ5 Ψp − iσµν + iσµν pµ , (7)
4 p · p̄ 6 6 ∂k⊥ν αβ
where Ψσ = ∂Ψσ ∂x(x,k⊥ )
. The wave functions Ψπ , Ψp , Ψσ may have different transverse
momentum dependence, this will make the calculation difficult. In order to simplify
discussions, we assume the same transverse momentum dependence for these wave
functions.
In pQCD approach, the convolutions of wave functions and hard scattering kernel are
presented in transverse configuration b-space. We need to define wave function in b-space
through Fourier transformation

Ψ (x, b) = d 2 k ⊥ e−i k⊥ ·b Ψ (x, k⊥ ). (8)
From Eqs. (5) and (8), we can obtain

φ(x) = Ψ (x, b = 0). (9)
The impact parameter b represents the transverse separation of quark and antiquark in
pion. Eq. (9) shows that the distribution amplitude is equal to the wave function at zero
transverse separation.
The intrinsic transverse momentum dependence of wave function of pion is unknown
from the first principle in QCD. We take a simple model in which the dependence of the
wave function on the longitudinal and transverse momentum can be separated into two
parts:
Ψ (x, k⊥ ) = φ(x) × Σ(k⊥ ), (10)
where φ(x) is the pion distribution amplitude. φ(x) and Σ(k⊥ ) satisfy the normalization
conditions
1
φ(x) = 1, d 2 k ⊥ Σ(k⊥ ) = 1. (11)
0
The k⊥ dependence of the wave function is contained in Σ(k⊥ ). In [10], Σ(k⊥ ) is assumed
to be a Gaussian distribution,
β2
Σ(k⊥ ) = exp −β 2 k⊥
2
. (12)
π
Transforming it into the transverse configuration b-space, we obtain

b2
Σ(b) = d 2 k ⊥ e−i k⊥ ·b Σ(k⊥ ) = exp − 2 . (13)
4β
The oscillating parameter β is fixed by requiring the root mean square transverse
momentum (r.m.s.), k⊥
2 1/2 is the order of Λ
QCD . Their relation can be obtain from
1 2
2 dx d k ⊥ k⊥ 2 |Ψ (x, k )|2
⊥ 1
k⊥ = 0 1 = 2. (14)
0dx d 2 k ⊥ |Ψ (x, k⊥ )|2 2β
Thus, k⊥
2 1/2 = √1 . If the root mean square transverse momentum k⊥
2 1/2 = 0.35 GeV,
2β
−2
then β 2 = 4 GeV . The detailed discussions about choice of the oscillation parameter β
can be found in [10].
2.3. B meson wave functions
The intrinsic dynamic of B meson is different from the case of light pion meson. The
momentum components of the spectator quark l are of order ΛQCD . The most convenient
tool to describe B meson is heavy quark effective theory (HQET). Since we have chosen
Pπ in the – direction, the hard scattering amplitude does not dependent on l− , the l−
dependence of the wave functions can be integrated out. In HQET, the B meson wave
functions are defined by the general Lorentz decomposition of the light-cone matrix
element [3,14]
B )
0|q̄β (z)bα (0)|B(p
+ (z2 , t)
− (z2 , t) − Ψ
ifB p / B + mB
+ 2 Ψ
=− 2 ΨB z , t + B B
/z γ5 , (15)
4 2 t αβ
pB
with v = m B
and t = v · z. A path-ordered exponential is implicitly present in the gauge-
independent matrix element.
The wave functions in momentum space can be obtained through the Fourier
transformation:
2
1 d z ⊥ dz− i(l+ z− /2−l ⊥ · z⊥ ) ± 2
ΨB± (l+ , l⊥ ) = 2
e − z⊥ , z− /2 ,
Ψ B (16)
2 (2π) 2π
√ √
where l = (l+ / 2, 0, l⊥ ) and z = (0, z− / 2, z ⊥ ), so that z2 = − z⊥2 and t = z− /2.
A momentum projection for the matrix element of B meson is needed to simplify
the calculation. In [13], the authors obtain a projection which is valid for distribution
amplitudes. We extend their analysis to include the transverse momentum dependence
of wave functions. We observe that their formulas need no modification except the
replacement of distribution amplitudes by wave functions. The proof of this point is given
in Appendix A. So, the momentum-space projection operator for B meson is:

ifB p / B + mB + − µ ∂
Mαβ = −
B
/ + + ΨB n
ΨB n / − − ∆(l+ , l⊥ )γ µ γ5 , (17)
4 2 ∂l⊥ αβ
l+
where ∆(l+ , l⊥ ) = 0 dl(ΨB− (l, l⊥ ) − ΨB+ (l, l⊥ )).
The projection operator can be represented in the form which is helpful to compare with
the results in the previous pQCD analysis [9]

ifB /+ − n
n /− 1 µ ∂
Mαβ = −
B
p B + m B ) ΨB +
(/ ΨB − ∆(l+ , l⊥ )γ µ γ5 , (18)
4 2 2 ∂l⊥ αβ
B are defined by
where ΨB , Ψ
ΨB+ + ΨB− Ψ + − ΨB−
ΨB = , ΨB = B . (19)
2 2
In [1], the authors consider a different projection operator for B meson
ifB
B
Mαβ =− pB + mB )γ5 [ΨB1 + n
(/ / − ΨB2 ] αβ , (20)
4
where ΨB1 , ΨB2 are defined by
ΨB+ − ΨB−
ΨB1 = ΨB+ , ΨB2 = . (21)
2
The above three projection operators are equivalent up to leading power in 1/mb . The
projection given in Eq. (20) neglects the ∆ term which is proportional to l + /mb and thus
power suppressed.
The B meson distribution amplitudes are obtained from transverse momentum integral
of the relevant wave functions or from wave functions at zero transverse separation. φB±
and φB , φ̄B are distribution amplitudes relative to wave functions of ΨB± and ΨB , Ψ B ,
respectively, and they are related by
φB+ + φB− φ + − φB−
φB = , φ̄B = B . (22)
2 2
The equations of motion impose constraint on the wave functions. Using the equation
of motion for the light quark of B meson and neglecting the effects of three-parton and
higher Fock states, we can obtain [3,13]

∂Ψ − Ψ − − Ψ +
B
B
+ B
2 = 0, (23)
∂t t z =0

∂Ψ + 1 ∂ 2 Ψ −
B
B
+ = 0. (24)
∂z2 4 ∂t 2 z2 =0
Similar to the discussion for pion wave function, we consider a model in which the
dependence on the longitudinal and transverse momenta of B meson is factorized:
ΨB± (ξ, l⊥ ) = φB± (ξ ) × ΣB± (l⊥ ), (25)
where ξ = ml+B is the longitudinal momentum fraction. The φB± are the two distribution
amplitudes of B mesons. The normalization conditions are

dξ φB± (ξ ) = 1, d 2 l ⊥ ΣB± (l⊥ ) = 1. (26)
Under the above assumptions, the two constraints become
dφB−
φB+ (ξ ) = −ξ (ξ ), (27)
dξ
ξ 2 m2B φB− (ξ ) = ωB2 φB+ (ξ ), (28)

where ωB2 = d 2 l⊥ l⊥2 ΣB+ (l⊥ ).
The distribution amplitudes of φB± can be solved analytically and the corresponding
solution is:
2 2 2 2
− 2 mB ξ mB + 2 ξ 2 m3B ξ mB
φB (ξ ) = exp − 2
, φ B (ξ ) = 3
exp − . (29)
π ωB 2ωB π ωB 2ωB2
The parameter ωB is fixed by the value of root mean square transverse momentum
of B meson. For the transverse momentum dependent function, we choose the Gaussian
distribution:
2
1 l⊥
ΣB+ (l⊥ ) = exp − . (30)
2πωB 2 2ωB 2
√
The parameters ωB and ωB are not independent, they are related by ωB = 2 ωB . The
root mean square transverse momentum l⊥ ωB
2 1/2 = √ = ωB . In transverse configuration
2
b-space, the transverse momentum dependent function is

+ ωB 2 bB2
ΣB (bB ) = exp − , (31)
2
where bB is the conjugated variable of l⊥ . For ΣB− (bB ) function, we assume it is the same
as ΣB+ (bB ).
A more rigorous study of B meson wave functions in HQET is given recently in [15].
The authors find an analytic solution in the heavy quark limit using the QCD equations of
motion. The transverse momentum distribution can also be derived without the assumption
of Gauss behavior. The obtained B meson wave functions is model-independent.
2.4. Sudakov form factor
There are two types of resummation: Sudakov resummation (or say b-space resumma-
tion) and threshold resummation. These two resummation effects lead to suppression in
different space: the region with large transverse separations b for Sudakov resummation
and the small longitudinal fractional momentum x region for threshold resummation.
First, we discuss the Sudakov resummation in brief. At αs order, the overlap of soft
and collinear divergences produce double logarithms −c ln2 Qb. The transverse impact
parameter b is used to regulate the infrared divergence. In transverse configuration b space,
the Sudakov double logarithms are resumed up to next-to-leading-log approximation.
A exponential factor e−s(x,b,Q) will be obtained from b-space resummation. we present
the explicit expression of the exponent s(x, b, Q) appearing in Sudakov form factor.
Define the variables,

√
q̂ ≡ ln xQ/( 2 ΛQCD ) , b̂ ≡ ln(1/bΛQCD). (32)
According to [16], the exponent s(x, b, Q) is presented up to next-to-leading-log
approximation
s(x, b, Q)

A(1) q̂ A(1) A(2) q̂
= q̂ ln − (q̂ − b̂) + −1
2β1 b̂ 2β1 4β12 b̂
(2) 2γE −1
A A(1) e q̂
− 2
− ln ln
4β1 4β 1 2 b̂

A β2 ln(2q̂) + 1 ln(2b̂) + 1
(1) A(1) β2 2
+ 3
q̂ − + 3
ln (2q̂) − ln2 (2b̂)
4β1 q̂ b̂ 8β1
2γE −1
(1)
A β2 e ln(2q̂) + 1 ln(2b̂) + 1
+ ln −
8β13 2 q̂ b̂

A β2 2 ln(2q̂) + 3 2 ln(2b̂) + 3
(2)
− −
16β14 q̂ b̂
A(2) β2 q̂ − b̂ A(2)β22 q̂ − b̂ 2
− 4
2 ln(2 b̂) + 1 + 6
9 ln (2b̂) + 6 ln(2b̂) + 2
16β1 b̂ 2 432β1 b̂ 3

A β2 18 ln (2q̂) + 30 ln(2q̂) + 19 18 ln2 (2b̂) + 30 ln(2b̂) + 19
(2) 2 2
+ − ,
1728β16 q̂ 2 b̂2
(33)
where the coefficients βi and A(i) are
33 − 2nf 153 − 19nf
β1 = , β2 = ,
12 24
γE
4 67 π 2 10 8 e
A = ,
(1)
A =
(2)
− − nf + β1 ln , (34)
3 9 3 27 3 2
with γE the Euler constant.
The running coupling constant αs up to next-to-leading-log is written as
αs (µ) 1 β2 ln ln(µ2 /Λ2 )
= − . (35)
π β1 ln(µ2 /Λ2 ) β13 ln2 (µ2 /Λ2 )
√
The exponent s(x, b, Q) is obtained under the condition that xQ/ 2 > 1/b, i.e., the
longitudinal momentum should be larger than the transverse degree. So s(x, b, Q) is
defined for q̂ b̂, and set to zero for q̂ < b̂. The previous formulas [17] about the exponent
s(x, b, Q) picks up the most important first six terms in the first and second lines of the
expression of s. Note that the sign of the fifth and sixth terms are different from those
in [17].
The Sudakov form factor e−s(x,b,Q) falls off quickly in large b region and vanishes
as b > 1/ΛQCD . Therefore, it suppresses the long-distance contribution, which is called
Sudakov suppression. In axial-gauge, the Sudakov form factor is included in each hadronic
wave function. So we can define the pion and B hadronic wave functions with Sudakov
corrections as
Ψπ (x, bπ ; t) = exp(−Sπ )Ψπ0 (x, bπ , t),

ΨB (ξ, bB ; t) = exp(−SB )ΨB0 (ξ, bB , t), (36)
where t is the factorization scale in the hard scattering kernel. The Ψπ0 (x, bπ , t) and
ΨB0 (ξ, bB , t) are wave functions without Sudakov corrections.
Combining with the evolution of wave functions and hard scattering kernel, a complete
factor e−Sπ,B for pion and B mesons can be given as
1 ln(t/ΛQCD )
Sπ = s(x, bπ , mB ) + s(x̄, bπ , mB ) −
ln ,
β1 ln(1/(bπ ΛQCD ))
1 ln(t/ΛQCD )
SB = s(ξ, bB , mB ) − ln . (37)
β1 ln(1/(bB ΛQCD ))
About the Sudakov form factor e−s(x,b,mB ) for B and pion meson, some comments are
in order:
• The Sudakov factor for B meson is only associated with the light quark since
there is no collinear divergence associated with the heavy b quark. Due to the
fact that the momentum of light spectator quark is concentrated at the order of
ΛQCD , we may expect the Sudakov effects is small because of the suppression of
B meson wave function. Our numerical analysis shows that its effect is at one percent
level.
• Notice that the effect of the evolution term in the last of Eq. (37) gives slight en-
hancement. If Sudakov suppression takes place, the enhancement will be masked (for
example, the case of pion). On the other hand, if Sudakov suppression is not effective
(for example, the case of B meson), this enhancement will emerge.
• In the endpoint region, the exchanged gluon carries small longitudinal momentum.
The transverse degree cannot be neglected. The mechanism of Sudakov suppression
begin to take into effect, and the long-distance dynamics from large transverse
separations are suppressed by Sudakov effects. If the momentum fraction of one
quark x is small, the Sudakov form factor associated with it is 1, no suppression
takes place. However, the Sudakov form factor associated with the other quark will
provide strong suppression because the momentum fraction of this quark x̄ is large
now.
2.5. Threshold resummation
Now, we discuss the threshold resummation effects. Double logarithms αs ln2 x can
be produced by the higher order loop corrections. It diverges at the endpoint. If the
endpoint contribution is important, the double logarithms αs ln2 x needs to be resumed
to all orders. We review the basic idea of threshold resummation. Our discussion
underlies on the analysis given in [9,18]. The vertex corrections at αs order produce
αs
the double logarithms − 4π CF ln2 x where CF = 4/3 is the color factor. This collinear
divergences can be factorized into a quark jet function St (x). In order to resume the
double logarithms to all orders, it is necessary to introduce the moment (N) space. In
N space, the Sudakov factor has the exponential form up to the accuracy of leading-log
approximation (LL),

(LL) 1 (LL) 2
St (N) = exp − γK ln N , (38)
4
where the anomalous dimension γK(LL) = αs CF /π .
The jet function St (x) can be obtained from St (N) through the transformation

a+i∞
dN
St (x) = (1 − x)−N St (N)St(0) (N), (39)
2πi
a−i∞
where a is an arbitrary real constant larger than all the real parts of poles involved
in the integrand. The St(0) (N) comes from Mellin transformation of the initial condi-
tion St(0) (x) = 1,
1
(0) (0) 1
St (N) = dx (1 − x)N−1 St (x) = . (40)
N
0
The upper index “0” means that there is no QCD corrections.

The contour integral in Eq. (39) can be transformed to

(LL) π
St (x) = − exp αs CF
4
∞
dt 1 αs
× (1 − x)exp(t ) sin αs CF t exp − CF t 2 . (41)
π 2 4π
−∞
The above analysis can be directly extended to next-to-leading logarithms. The jet
1
function St (x) satisfies the normalization condition 0 St (x) = 1. It vanishes at the end
points x → 0 and x → 1. The most important property of jet function St (x) is that it
damps faster than any power of x.
Since the resumed factor St (x) suppresses small x contribution, it may play crucial
role in B decays. The B meson distribution amplitude φB− (x) and the twist-3 distribution
amplitudes φP (x) do not vanish at x = 0 in general. Although the transverse momentum
can regulate the endpoint singularity, there is still a substantial contribution coming from
endpoint region. The factor St (x) can suppress endpoint contribution and make pQCD
more applicable.
In [19], the authors discuss another resummation whose formula is similar to Sudakov
resummation. The obtained Sudakov form factor suppresses small x contribution more
rapidly than the factor St (x). It shows the importance of double-log corrections from
another point.
The factor St (x) presented in Eq. (41) involves one parameter integration. In or-
der to simplify the numerical calculation, a simple parametrization for St (x) is pro-
posed [9]
21+2c Γ (3/2 + c) c
St (x) = √ x(1 − x) , (42)
π Γ (1 + c)
where the parameter c is determined around 0.3. The factor St (x) with the above simple
parameterization form vanishes at x = 0, 1. But it does not damp faster than any power
of x. Thus the above parametrization is proposed only for phenomenological application.
The rigorous treatment should retain the integral.
3. Calculations
3.1. The formulas of B → π form factors in pQCD approach
We have defined B → π matrix element in terms of form factors F+,0

Bπ
in Eq. (1). The
B → π matrix element can also be expressed in another form

π(Pπ )ūγµ b B(PB ) = f1 q 2 PBµ + f2 q 2 Pπµ , (43)
Bπ from f
From Eqs. (1) and (43), we can obtain F+,0 1,2
1 1 1
F+Bπ = (f1 + f2 ), F0Bπ = (f1 + f2 ) + η̄(f1 − f2 ) (44)
2 2 2
q2
with η̄ = .
m2B
In the large recoil region, the B → π transition is dominated by the single gluon
exchange in the lowest order as depicted in Fig. 1. The formulas for the amplitude of
Figs. 1(a) and (b) are
πCF
A= fπ fB m2B
Nc

1
× dξ dx d l⊥ d 2 k ⊥ αs
2
(xηm2B + k⊥ )(ξ xηm2B + (l ⊥ − k ⊥ )2 )
2

∆(ξ, l⊥ ) k ⊥ · (l ⊥ − k ⊥ )
× 2 Ψπ (xη + 1)ΨB + (xη − 1)ΨB + Pπµ
mB ξ xηm2B + (l ⊥ − k ⊥ )2

µπ 1
+ Ψp (ΨB − Ψ B )PBµ + 2 Ψ B − xΨB Pπµ
mB η

µπ Ψσ 1
− (ΨB − Ψ B )PBµ − 2 − x ΨB Pπµ
mB 6 η

µπ Ψ σ k ⊥
2
k ⊥ · (l ⊥ − k ⊥ )
−2 2 −1 + − ΨB
mB 6 xηm2B + k ⊥ 2 ξ xηm2B + (l ⊥ − k ⊥ )2

k ⊥ · (l ⊥ − k ⊥ ) 2(l ⊥ − k ⊥ )2
+ −1 − +
xηm2B + k ⊥ 2 ξ xηm2B + (l ⊥ − k ⊥ )2

mB ∆(ξ, l⊥ )
× Pπµ , (45)
ξ xηm2B + (l ⊥ − k ⊥ )2
and
πCF
B= fπ fB m2B
Nc

1
× dξ dx d 2 l⊥ d 2 k ⊥ αs
(ξ ηm2B + l ⊥2 )(ξ xηm2B + (l ⊥ − k ⊥ )2 )

B )PBµ − ξ(ΨB + Ψ
× 2 Ψπ ξ η(ΨB + Ψ B )Pπµ
l 2
∆(ξ, l⊥ ) ⊥ (l ⊥ − k ⊥ ) · l⊥
+ 1− − Pπµ
mB ξ ηm2B + l ⊥2 ξ xηm2B + (l ⊥ − k ⊥ )2

µπ
+2 Ψp ξ(Ψ B − ΨB )
mB

∆(ξ, l⊥ ) l ⊥2 (l ⊥ − k ⊥ ) · l ⊥
+ 1− − PBµ
mB ξ ηm2B + l⊥2 ξ xηm2B + (l ⊥ − k ⊥ )2

2
+
1 − ξ ΨB + ΨB Pπµ . (46)
η
In hard scattering kernels, transverse momentum k⊥ in the denominators are retained to

regulate the endpoint singularity. The k⊥2 in the numerator are power suppressed compared
2
to mB and they must be dropped in order to ensure the gauge invariance required by
factorization theorem. Transform the formulas from the momentum space into transverse
configuration b-space and include Sudakov factors, we obtain the final formulas for F+Bπ
and F0Bπ in pQCD approach

πCF
F+Bπ = fπ fB m2B dξ dx bB dbB bπ dbπ αs (t)
Nc

× B (ξ, bB )
Ψπ (x, bπ ) (xη + 1)ΨB (ξ, bB ) + (xη − 1)Ψ

µπ 2
+ Ψp (x, bπ ) (1 − 2x)ΨB (ξ, bB ) + −1 Ψ B (ξ, bB )
mB η

µπ Ψσ (x, bπ ) 2
− 1 + 2x − ΨB (ξ, bB ) − ΨB (ξ, bB )
mB 6 η

µπ Ψσ (x, bπ )
+4 ΨB (ξ, bB ) h1
mB 6
µπ Ψσ (x, bπ ) bB
−2 −mB ∆(ξ, bB ) √ h3
mB 6 2 ξ xη mB

∆(ξ, bB )
+ Ψπ (x, bπ ) −ξ η̄ ΨB (ξ, bB ) + Ψ B (ξ, bB ) +
mB

µπ 2ξ
+2 Ψp (x, bπ ) (1 − ξ )ΨB (ξ, bB ) + 1 + ξ − ΨB (ξ, bB )
mB η

∆(ξ, bB )
+ h2 , (47)
mB
and

πCF
F0Bπ = 2
fπ fB mB dξ dx bB dbB bπ dbπ αs (t)
Nc

B (ξ, bB )
× Ψπ (x, bπ )η (xη + 1)ΨB (ξ, bB ) + (xη − 1)Ψ
µπ
+ Ψp (x, bπ ) (2 − η − 2xη)ΨB (ξ, bB ) + ηΨ B (ξ, bB )
mB
µπ Ψσ (x, bπ ) B (ξ, bB )

− η(2x − 1)ΨB (ξ, bB ) − (2 − η)Ψ
mB 6

µπ Ψσ (x, bπ )
+4 η ΨB (ξ, bB ) h1
mB 6
µπ Ψσ (x, bπ ) bB
−2 −mB ∆(ξ, bB ) √ h3
mB 6 2 ξ xη mB

∆(ξ, bB )

+ Ψπ (x, bπ )η ξ η̄ ΨB (ξ, bB ) + ΨB (ξ, bB ) +
mB

µπ
+2 Ψp (x, bπ ) (η − 2ξ + ξ η)ΨB (ξ, bB ) + η(1 − ξ )Ψ B (ξ, bB )
mB

∆(ξ, bB )
+ (2 − η) h2 , (48)
mB
where
√ √
h1 = K0 ξ xη mB bB θ (bB − bπ )I0 xη mB bπ K0 xη mB bB
√ √
+ θ (bπ − bB )I0 xη mB bB K0 xη mB bπ , (49)

h2 = K0 ξ xη mB bπ θ (bB − bπ )I0 ξ η mB bπ K0 ξ η mB bB

+ θ (bπ − bB )I0 ξ η mB bB K0 ξ η mB bπ , (50)
√ √
h3 = K−1 ξ xη mB bB θ (bB − bπ )I0 xη mB bπ K0 xη mB bB
√ √
+ θ (bπ − bB )I0 xη mB bB K0 xη mB bπ . (51)
The function Ki and Ii are modified Bessel functions with i their orders.
The physical quantities should not depend on the choice of the scale parameter t ≡ µ if
the calculation can be performed up to infinite orders. Therefore, the scale parameter can
be chosen as any value in principle. However, in practice the calculation can only be made
perturbatively. To make the perturbative expansion meaningful, the scale parameter should
be chosen in such
√ a way that can make the higher order corrections small. The natural
choice is t = ξ xη mB in the standard approach. If ξ, x → 0, αs (t) will√be divergent
at t ΛQCD . When including the transverse degree of freedom, αs (µ) ln ξ xη mB /µ,
αs (µ) ln b√
B µ and αs (µ) ln bπ µ will appear in higher order corrections. Therefore we take
t = max( ξ xη mB , 1/bB , 1/bπ ). The scale t ≡ µ must be larger than ΛQCD thus avoids
the divergence of coupling constant.
The wave functions include Sudakov corrections coming from Sudakov and threshold
resummations
Ψπ (x, bπ ) = St (x) exp(−Sπ )φπ (x)Σπ (bπ ),

ΨB (ξ, bB ) = St (ξ ) exp(−SB )φB (ξ )ΣB (bB ). (52)
B . The jet function St comes from
The similar expressions are given for Ψp , Ψσ , Ψσ , Ψ
threshold resummation. The simplified parametrization form in Eq. (42) is taken to estimate
threshold resummation effects in this paper. The complete Sudakov factors Sπ,B are given
in Eq. (37). Σπ,B are intrinsic transverse momentum dependence of pion and B meson
wave functions.
We compare our formulas with the results in [3,9]. In [3], only the leading twist of pion
is discussed. Set the twist-3 terms to zero, the two formulas of ours and [3] are the same
except the definition of h1 . The difference comes from the Fourier transform of hard part.
In [9], the single B meson wave function ΨB is assumed and the terms of Ψ B and ∆ are
neglected. The twist-3 power correction is included. The momentum projector in [9] for
pion meson is slightly different from our projector in Eq. (7). Except for these differences,
the formulas in [9] are consistent with ours.
3.2. The assumption of single distribution amplitude
Before we perform analysis of B → π form factors in pQCD approach, we discuss an

assumption of using single B meson distribution amplitude at first. To our knowledge, the
use of B meson distribution amplitude in B decays firstly appeared in [20]. The authors
suggest the simplest momentum projection for B meson which contains two terms
ifB
MB = − p/ B + mB g(ξ ) γ5 φB (ξ ). (53)
4
The function g(ξ ) is assumed to be at the order of 1. Setting g(ξ ) = 1, the B meson
momentum projection reduces to
ifB
MB = − [/
pB + mB ]γ5 φB (ξ ). (54)
4
This is the widely used formula in previous pQCD analysis which contains single B meson
distribution amplitude. Here, we discuss the most familiar model used in the previous
pQCD analysis
2 2
ξ mB
φB (ξ ) = NB ξ (1 − ξ ) exp −
2 2
, (55)
2ωB2
1
where NB is the renormalization constant makes 0 φB (ξ ) dξ = 1. In this model φB (ξ ) has
a peak at ξ = Λ̄/mB where Λ̄ = mB − mb . This model of B meson distribution amplitude
has been used to fit different channels of B decays.
B meson distribution amplitudes are important ingredients in pQCD approach. We
should be careful about the choice of B meson distribution amplitude. In HQET, the
definition of B meson distribution amplitude contains two terms φB+ , φB− . Compare
Eq. (55) and Eq. (29), the difference lies in 1/mB effect and can be neglected. The
choice of single distribution amplitude amounts to taking φB = φB1 and neglects φB2 .
So, the validity of the assumption of single distribution amplitude depends on whether the
contribution of φB2 is small or not. To test this assumption, we make approximation that
φB ≈ φB1 = φB+ where φB+ is given in Eq. (29). This approximation is reasonable because
the calculated form factor F Bπ (0) is: 0.239 using φB in Eq. (55) and 0.227 using φB+
in Eq. (29).
Table 1 shows the numerical result for the contributions from φB1 , φB2 and ∆B . If
considering only φB1 contribution, F Bπ (0) is 0.227. This result is consistent with the
Table 1
The B → π form factors F Bπ = F+,0 Bπ (0) with φ , φ
B1 B2 and ∆B . The column “φB1 ” represents φB1
contribution only. The meanings of columns “φB2 ” and “∆B ” are similar. The column “sum” represents the
summation of all these contributions
φB1 φB2 ∆B Sum

Twist-2 0.042 0.016 0.014 0.072
Twist-3 0.185 0 0.048 0.233
Total 0.227 0.016 0.062 0.305
one using QCD sum rule within the theoretical errors. The φB1 contribution is dominant.
For φB2 term, its contribution vanish at twist-3 but cannot be neglected at twist-2. The
power suppressed term of ∆B is about 20% in the total numerical result. From a general
point, there should have two leading B meson distribution amplitudes (or generally wave
functions) in pQCD framework. But the single B meson distribution amplitude φB1 is
the dominant contribution. It should be noted that φB2 contribution vanishes in the hard
spectator scattering and weak annihilation diagrams in B → ππ decays.
3.3. The B → π form factors for different models of distribution amplitudes
The distribution amplitudes for pion depends on the renormalization scale. This
dependence is controlled by the evolution equation. For the scale related to our discussion,
evolution effect should be important but precise estimate of this effect depends on the
unknown input parameters. In this paper, what we concern most is B meson wave functions
and the reliability of pQCD framework. We will not consider the evolution effects of pion
distribution amplitudes. Thus, the pion distribution amplitudes for both the twist-2 and
twist-3 are taken as their asymptotic form for simplicity, i.e.,
φπ = 6x x̄, φp = 1, φσ = 6x x̄. (56)
For B meson distribution amplitudes (wave functions), three models exist in literatures:
• Model I
2 2 2 2
2 mB ξ mB 2 ξ 2 m3B ξ mB
φB− (ξ ) = exp − , φB+ (ξ ) = exp − .
π ωB 2ωB2 π ωB 3 2ωB2
(57)
• Model II

mB ξ mB ξ m2B ξ mB
φB− (ξ ) = exp − , φB+ (ξ ) = exp − , (58)
ω0 ω0 ω02 ω0
where ω0 = 23 Λ̄ and Λ̄ = mB − mb in this model.

• Model III
2ξ̄ − ξ 2
ΨB− (ξ, k⊥ ) = θ (2ξ̄ − ξ )δ k⊥ − m2B ξ(2ξ̄ − ξ ) ,
2π ξ̄ 2
ξ 2
ΨB+ (ξ, k⊥ ) = θ (2 ξ̄ − ξ )δ k ⊥ − m 2
B ξ(2 ξ̄ − ξ ) , (59)
2π ξ̄ 2
with ξ̄ = Λ̄
mB .
Model I is proposed in [3]. It uses the equations of motion for light spectator quark
in HQET. The ωB is at the order of ΛQCD . The possible range is 0.2–0.5 GeV. Model
II is based on a QCD Sum rule inspired analysis [14]. The above two models are both
Gaussian type and there is a peak near Λ̄/mB . Model III uses the equations of motion for
both light spectator quark and heavy b quark in HQET [15]. The distribution amplitudes
Table 2
Bπ (0) with three models for B meson distribution amplitudes
The numerical results of B → π form factors F+,0
Model φB φ̄B ∆B Total
I 0.564 −0.321 0.062 0.305
II 0.641 −0.338 0.060 0.363
III 0.540 −0.335 0.055 0.260
Table 3
Bπ (0) with different ω
The twist-2 and twist-3 contributions of pion in B → π form factor F Bπ = F+,0 B
ωB (GeV) 0.25 0.3 0.35 0.4 0.45 0.5

Twist-2 0.086 0.078 0.072 0.066 0.061 0.056
Twist-3 0.345 0.280 0.233 0.198 0.171 0.150
Total 0.431 0.358 0.305 0.264 0.232 0.206
have a cutoff at ξ = 2ξ̄ because it takes the approximation mB = ∞.1 The distributions
are linear functions of ξ . For model I, the transverse momentum dependent function
ΣB (bB ) is given in Eq. (31). For model II, the transverse momentum dependence part
is unknown, we take the same transverse momentum dependence function ΣB (bB ) as
model I.
The other input parameters are as follows: decay constants for pion and B meson
fπ = 0.13 GeV, fB = 0.19 GeV; ΛQCD = 0.25 GeV; Λ̄ = mB − mb = 0.5 GeV; the
oscillator parameter in the transverse momentum distribution of pion wave function
β 2 = 4 GeV−2 [10]; parameter in B meson wave function ωB = 0.35 GeV; pion twist-
3 coefficient µπ = 2.2 GeV [8].
Bπ
We present the result of form factors F+,0 at large recoil q 2 = 0. Using B and pion
wave functions and the input parameters presented above, predictions for the F+,0Bπ (0) with
the three models for B meson distribution amplitudes are listed in Table 2.
The predicted B → π form factors F+,0Bπ
(0) in model I, II and model III are around 0.3
which is favored by experiment and consistent with the prediction by other methods, such
as QCD sum rule [11], BSW model [21]. The advantage of model III is that it only uses
the QCD equations of motion. In reality, it is not a model. There is only one universal
phenomenological parameter Λ̄ in wave functions.
At the end of this subsection, we would like to discuss the power corrections. Table 3
shows that the chirally enhanced twist-3 contribution is numerically larger than leading
twist contribution. The large twist-3 contribution is not consistent with the assumption of
twist expansion. The study of power corrections should be studied in a more careful and
more systematic way which is beyond the scope of this paper.
1 We thank J. Kodaira for the discussions about this topic.

3.4. The reliability of pQCD approach in B → π form factors
Now, we examine the reliability of pQCD analysis in B → π form factors. We choose

model I of the B meson distribution amplitudes for illustration. The conclusions of model II
and III are similar.
The basic idea of pQCD approach is to use Sudakov effects to suppress the long-
distance contribution with large transverse separations. A reliable pQCD analysis of B →
π form factors should satisfy that most of the result comes from the region where impact
parameters bπ , bB are both small. In order to study the impact parameter b dependence
of B → π form factors, we introduce a cut-off bc in impact parameters bπ and bB in the
bc bc
integrals of Eqs. (47) and (48) by 0 B dbB 0 π dbπ . In Fig. 2, we show the dependence of
B → π form factors F Bπ = F+,0 Bπ (0) on b c and b c . Fig. 2(a) plots the b c dependence with
π B π
ωB = 0.35 GeV and µπ varied. Fig. 2(b) plots the bπc dependence with µπ = 2.2 GeV
and ωB varied. One can see that varying these two important input parameters µπ and ωB
does not change the behavior of the F Bπ dependence on bc . Similarly, the bBc dependence
of F Bπ is depicted in Figs. 2(c) and (d). From Fig. 2, 50% of F Bπ comes from the
bπc < 1.5 GeV−1 for impact parameter bπ and bBc < 2.0 GeV−1 for impact parameter
bB . The contributions from regions with large impact parameters bπ , bB > 2 GeV−1 are
substantial: 31% for impact parameter bπ and 50% for bB . The calculation of pQCD
approach may still include large long-distance contributions. Therefore, the reliability of
leading order calculations in αs expansion should be checked carefully. The more direct
criteria is to check the distribution of the coupling constant αs (t).
The standard to judge the reliability of perturbative analysis is that most of the
contribution comes from the region where the coupling constant αs (t) is small. Fig. 3
plots the form factors F Bπ coming from the region where a1 αs (t)/π a2 . The last bar
is the contribution of αs (t)/π > 0.9. In our calculation, 70% of the result comes from the
region where αs (t)/π < 0.2 and the contribution for t 1 GeV is 38%. If we consider
the energy 1 GeV is the point that the perturbation theory is broken, the non-perturbative
contribution is about 60%. If we consider a weaker criterion that αs (t)/π = 0.2 is the
broken point, one can see the non-perturbative contribution is 30%. No matter which
criterion is chosen, the non-perturbative contribution is comparable to the perturbative
part. So, the prediction of B → π form factor in pQCD approach cannot be precise.
The non-perturbative contribution constitutes the intrinsic systematic error in pQCD
approach.
One may also wonder if the integration over bB may not converge because the form
factor F+Bπ (0), as shown in Fig. 2(c) and (d), does not saturate at any value of bBc .
Therefore, it is necessary to study the property of the dependence of F+Bπ (0) on bB in the
region bB > 1/ΛQCD . However, since the complete factor e−SB in B meson wave function
is divergent at ΛQCD , we drop this term and perform the calculations again. We find that
the form factors saturate at bB ∼ 1/ΛQCD as shown in Fig. 4. That is to say, even without
any Sudakov suppression, the region of large bB does not contribute.
Fig. 2. bπ and bB dependence of F Bπ .
Fig. 3. The B → π form factors F Bπ vs. the coupling constant αs (t).

Fig. 4. The property of F Bπ vs. bB c in B meson. The dotted curve is the result with Sudakov and evolution effects
in B meson, the dashed one is for the case without Sudakov effect, i.e., without the contribution of s(ξ, bB , mB )
in SB , and the solid curve is the result without both Sudakov and evolution effects in B meson, i.e., without the
total contribution of SB .
3.5. Comparison with other works
Whether B → π form factors are hard dominant or soft dominant is a highly

controversial topic in B physics. There are many theoretical investigations using various
approaches. In this paper, we have studied the perturbative method. The comparison of our
result with other recent works in pQCD approach is helpful to interpret the perturbative
method.
In [9], the authors calculate the B → π form factors in pQCD approach which includes
pion twist-3 power corrections. The intrinsic transverse momentum dependence of B
meson wave function is considered and of pion is neglected. Neglecting the intrinsic
transverse momentum effect in pion is unjustified. The physical reason is that Sudakov
suppression is not so strong for realistic B decays and the large b contribution is very
sensitive to the non-perturbative dynamics in hadrons. We have discussed that the model
of B meson wave function chosen in [9] is nearly same as the φB+ in model I. So, we find an
explanation of this model in HQET. The reliability of B → π form factor in pQCD is given
in [2] by Y. Keum et al. They observed that 97% of the contribution comes from the region
with αs (t)/π < 0.3. This criterion of αs (t)/π < 0.3 is not strong enough to guarantee the
reliability of leading order result in αs expansion series.
The authors in [3] address some critical questions about Sudakov effects in B → π form
factors. We will make our comments on their questions below.
The first question concerns theoretical issue on Sudakov form factor. For pion meson,
it is pointed out that Sudakov form factor with next-to-leading-log (NLL) approximation
is gauge dependent. This dependence can be cancelled by a soft function U (b) which
resummed the single soft logarithms [22]. A complete analysis of B → π form factors
should include the next-to-leading-order (NLO) calculation in order to ensure gauge
invariance. In this paper, we take the NLL Sudakov form factor and hard kernel to leading
order, so the gauge dependence exists in our analysis. Whether this gauge dependence
will lead to large numerical difference is not clear at present. The NLO calculation in
pQCD approach must include the transverse degrees of freedom, therefore the calculation
is difficult in technique.
The second question is about numerical dependence on the uncertainties in meson’s
distribution amplitude. This is an important and inevitable problem in perturbative method.
We find that the three models of B meson wave functions which satisfy the QCD equations
of motion give the consistent numerical results. More importantly, the B meson wave
functions derived in [15] is model-independent. So, the theoretical errors caused by the
B meson wave functions can be reduced largely.
The third question is about the reliability of pQCD approach in B → π form factors.
The conclusion given in [3] is that Sudakov suppression is so weak that pQCD approach
cannot be applied in B physics. After including the intrinsic transverse momentum effects
and threshold resummation, the perturbation behavior of the numerical result in [3] is
similar to ours. Whether the perturbative calculation is reliable or not should be checked by
higher order calculations in αs expansion series. Our present analysis shows that one cannot
guarantee higher order terms will be small. More works are needed before a completely
reliable method is set up to control the higher order contributions.
4. Conclusions and discussions
In this paper, we have presented a systematic analysis of B → π form factors in

pQCD approach. The intrinsic transverse momentum effects and threshold resummation
are important ingredients in the pQCD framework in B decays. One important finding in
this paper is that all the three models of B meson wave functions predict F Bπ about 0.3
in pQCD approach. This result should not be accidental. We find that the numerical results
are not so sensitive to the choice of B meson wave functions in HQET. The B meson wave
functions which satisfy the QCD equations of motion can reduce the theoretical errors
largely. The consistent numerical results give us confidence that our knowledge about the
B meson wave functions is not so poor as expected before. HQET provides a reasonable
framework to interpret the intrinsic dynamics in B meson.
Although the prediction of B → π form factors in pQCD approach seems favorable
in experiment, we cannot avoid many serious conceptual problems of pQCD approach.
The calculation in our paper is at the leading order (LO). There is no strong evidence
that NLO result is small especially in our case where most of contribution comes from
the momentum region of t < 1 GeV. A critical problem in NLO calculation is how to
deal with the dependence on the choice of renormalization scale µ. In this paper, the
scale µ is taken as the largest momentum carried by the exchanged gluon. Whether this
choice is best or not depends on the NLO calculation. However, as we have discussed, the
NLO calculation in B → π form factors is technically difficult. Another problem comes
from power corrections. We only discuss the chirally enhanced twist-3 correction. There
are many other power corrections, such as twist-4, higher Fock states etc. Although one
can argue that they are power suppressed in the heavy quark limit, we do not know their
numerical effects in the realistic case. In principle, all the above issues should be treated in a
more systematic way than the present form. The seeming correspondence of the prediction
of pQCD approach in B → π form factors with other theoretical expectations is based on
many assumptions and one must be careful to obtain a conclusion. To some extent, we think
it is difficult to estimate the theoretical errors in the present pQCD approach in B decays.
Although there are many crucial problems in pQCD approach, completely rejecting
it should be with caution. There are at least 30% hard contribution coming from the
momentum t > 1 GeV in B → π form factors. This contribution cannot be calculated
reliably in any other present theoretical approaches. We cannot throw one method before
we get the final answers.
There is no universal, satisfactory criteria of QCD because we do not know how
to quantitatively calculate the non-perturbative dynamics especially in exclusive, non-
leptonic B-meson decays at present. The complicated strong interactions in exclusive B
decays where both perturbative and non-perturbative dynamics contribute is a challenging
task for the present non-perturbative methods, such as QCD sum rules, lattice QCD etc.
In conclusion, B → π form factors provide an interesting places to study the
perturbative and non-perturbative strong interactions. The success of B meson wave
functions in HQET is helpful to understand the exclusive B decays. The perturbative
method provides a simple phenomenological method to estimate the perturbative aspects
of strong dynamics in B decays.
Acknowledgement
We would like to thank H. Li, S. Brodsky, A. Ali and V. Braun for many valuable
discussions and their comments. Z. Wei also thanks M. Beneke and C.T. Sachrajda for the
discussions in Regensburg, 20–22 July 2001. M. Yang thanks the partial support of the
Research Fund for Returned Overseas Chinese Scholars. This work is supported in part by
National Natural Science Foundation of China.
Appendix A
In this appendix we will derive the projector for B meson in the momentum space with
including transverse momentum. We begin with the generalized Lorentz decomposition of
the light-cone matrix element [3,14]
B )
M(z)αβ ≡ 0|q̄β (z)bα (0)|B(p
+ (z2 , t)
− (z2 , t) − Ψ
ifB p / B + mB Ψ
=− + z2 , t + B
2Ψ B
/z γ5 , (A.1)
B
4 2 t αβ
pB z+ √
with v = m and t = v · z. In light-cone coordinate z = ( √ , z− , z⊥ ), with z± = z0 ± z3 .
B 2 2
To obtain the projector in the momentum space we consider the amplitude of one process,

d 4z
T= M(z)A(z), (A.2)
(2π)4
where A(z) is the hard scattering kernel of the relevant process. Then

d 4z
T= M(z) d 4 l e−il·z A(l)
(2π)4

1 dz+ dz− d 2 z⊥ l+ z− +l− z+
= 4
M(z) dl+ dl− d 2 l⊥ e−i( 2 −l⊥ · z⊥ )
A l+ , l− , l⊥ .
4 (2π)
(A.3)

For the case that A(l) is independent of l− , we have A(l) = A(l+ , l⊥ ). Then the integration
over l− in the above equation can be accomplished to get a delta function δ(z+ /2).
Therefore we get

1 dz− d 2 z⊥ l+ z−
T= M(z)|z+ =0 dl+ d 2 l⊥ e−i( 2 −l⊥ · z⊥ ) A l+ , l⊥ . (A.4)
2 (2π)3
So we just need to consider the case z+ = 0. For convenience the above equation can be
written in the form

1 dz− d 2 z⊥ −i( l+ z− −l ⊥ · z⊥ )
T = dl+ d l⊥ 2
e 2 M(z)|z+ =0 A l+ , l⊥ . (A.5)
2 (2π)3
Introduce Fourier transformation:

± 1 dz− d 2 z⊥ −i( l+ z− −l ⊥ · z⊥ ) ± 2
ΨB (l+ , l⊥ ) = e 2 ΨB − z⊥ , z− /2 , (A.6)
2 (2π)3
and substitute Eq. (A.1) into Eq. (A.5), we can obtain

−ifB
T = dl+ d 2 l⊥
4

/ B + mB
p + 1 dz− d 2 z⊥ −i( l+ z− −l ⊥ · z⊥ )
× 2ΨB (l+ , l⊥ ) + e 2
2 2 (2π)3
− (− z2 , z− /2) − Ψ
Ψ + (− z2 , z− /2)

× B ⊥ B ⊥
/z γ5 A l+ , l⊥ .
z− /2
(A.7)
l+ − +
By defining ∆(l+ , l⊥ ) ≡ 0 dl(ΨB (l, l⊥ ) − ΨB (l, l⊥ )) and making the integrations by
part, we can finally get

−ifB
T = dl+ d 2 l⊥
4

/ B + mB
p ∂
× 2ΨB+ (l+ , l⊥ ) − ∆(l+ , l⊥ )γ µ µ γ5 A l+ , l⊥ . (A.8)
2 ∂l
To deal with ∂/∂l µ , we need to re-express l µ as
l+ µ l− µ µ
lµ = n + n− + l⊥ ,
2 + 2
therefore we can get
∂ µ ∂ µ ∂ ∂
µ
= n− + n+ + µ. (A.9)
∂l ∂l+ ∂l− ∂l⊥
Substitute Eq. (A.9) into Eq. (A.8) and do the integration by part and drop the surface term,
we get

−ifB p / B + mB
T = dl+ d l⊥ 2
ΨB+ (l+ , l⊥ )/
n+ + ΨB− (l+ , l⊥ )/
n−
4 2

∂
− ∆(l+ , l⊥ )γ µ µ γ5 A l+ , l⊥ . (A.10)
∂l⊥
So the projector for B meson in the momentum space is

ifB p / B + mB
Mαβ = − ΨB+ (l+ , l⊥ )/
n+ + ΨB− (l+ , l⊥ )/
n−
4 2

∂
− ∆(l+ , l⊥ )γ µ µ γ5 . (A.11)
∂l⊥ αβ
References
[1] M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Phys. Rev. Lett. 83 (1999) 1914–1917;
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 591 (2000) 313–418;
M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 606 (2001) 245–321.
[2] H. Li, H. Yu, Phys. Rev. Lett. 74 (1995) 4388–4391;
Y. Keum, H. Li, A.I. Sanda, Phys. Lett. B 504 (2001) 6;
Y. Keum, H. Li, A.I. Sanda, Phys. Rev. D 63 (2001) 054008;
C. Lü, K. Ukai, M. Yang, Phys. Rev. D 63 (2001) 074009;
C. Lü, M. Yang, hep-ph/0011238, Eur. Phys. J. C 23 (2002) 275.
[3] S. Descotes, C.T. Sachrajda, Nucl. Phys. B 625 (2002) 239–278.
[4] D. Du, C. Huang, Z. Wei, M. Yang, Phys. Lett. B 520 (2001) 50–58.
[5] S. Brodsky, G. Lepage, Phys. Rev. Lett. 43 (1979) 545;
S. Brodsky, G. Lepage, Phys. Lett. B 87 (1979) 359;
S. Brodsky, G. Lepage, Phys. Rev. D 22 (1980) 2157;
A. Efremov, A. Radyushkin, Phys. Lett. B 94 (1980) 245.
[6] N. Isgur, C. Smith, Nucl. Phys. B 317 (1989) 526–572.
[7] H. Li, G. Sterman, Nucl. Phys. B 381 (1992) 129–140.
[8] M. Beneke, G. Buchalla, M. Neubert, C.T. Sachrajda, Nucl. Phys. B 606 (2001) 245–321.
[9] T. Kurimoto, H. Li, A.I. Sanda, Phys. Rev. D 65 (2002) 014007.
[10] R. Jakob, P. Kroll, Phys. Lett. B 315 (1993) 463–470.
[11] E. Braun, P. Ball, V.M. Barun, Phys. Lett. B 417 (1998) 154–162;
A. Khodjamirian, R. Rückl, S. Weinzierl, O. Yakovlev, Phys. Lett. B 410 (1997) 275–284.
[12] V.M. Braun, I.E. Filyanov, Z. Phys. C 44 (1989) 157;
V.M. Braun, I.E. Filyanov, Z. Phys. C 48 (1990) 239.
[13] M. Beneke, T. Feldmann, Nucl. Phys. B 592 (2001) 3–34.
[14] A.G. Grozin, M. Neubert, Phys. Rev. D 55 (1997) 272–290.
[15] H. Kawamura, J. Kodaira, C. Qiao, K. Tanaka, Phys. Lett. B 523 (2001) 111, hep-ph/0112174.
[16] H. Li, Phys. Rev. D 52 (1995) 3958–3965.
[17] J. Botts, G. Sterman, Nucl. Phys. B 325 (1989) 62–100.
[18] H. Li, hep-ph/0102013.
[19] R. Akhoury, G. Sterman, Y. Yao, Phys. Rev. D 50 (1994) 358.
[20] A. Szczepaniak, E.M. Henley, S. Brodsky, Phys. Lett. B 243 (1990) 287.
[21] M. Wirbel, B. Stech, M. Bauer, Z. Phys. C 29 (1985) 637.
[22] H. Li, Phys. Rev. D 55 (1997) 105.
The QCD phase diagram for small densities

from imaginary chemical potential
Philippe de Forcrand a,b , Owe Philipsen c
a Institut für Theoretische Physik, ETH Zürich, CH-8093 Zürich, Switzerland
b Theory Division, CERN, CH-1211 Geneva 23, Switzerland
c Center for Theoretical Physics, Massachusetts Institute of Technology, Cambridge, MA 02139-4307, USA
Received 28 May 2002; received in revised form 23 July 2002; accepted 29 July 2002
Abstract
We present results on the QCD phase diagram for µB πT . Our simulations are performed with
an imaginary chemical potential µ = iµI for which the fermion determinant is positive. On an 83 × 4
lattice with 2 flavors of staggered quarks, we map out the phase diagram and identify the (pseudo-)
critical temperature Tc (µI ). For µI /T π/3, this is an analytic function, whose Taylor expansion is
found to converge rapidly, with truncation errors far smaller than statistical ones. The truncated series
may then be continued to real µ, yielding the corresponding phase diagram for µB 500 MeV. This
approach provides control over systematics and avoids reweighting. We compare it with other recent
work.
1. Introduction
A prime goal of heavy-ion collision experiments at SPS, LHC (CERN) and RHIC
(Brookhaven) is to probe the transition1 from hadronic matter to a quark–gluon plasma
at high temperatures and small baryon density. Because QCD is strongly interacting, the
only first principles method to predict the phase diagram is by means of lattice simulations.
On the other hand, at finite densities the fermion determinant is complex, thus prohibiting
E-mail address: philipse@lns.mit.edu (O. Philipsen).

1 In this paper the term “transition” is used to label a change in dynamics, whether it proceeds by a smooth
crossover or by a first or second order phase transition. For the last two, we use explicitly the term “phase
transition”.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 2 6 - 0
P. de Forcrand, O. Philipsen / Nuclear Physics B 642 (2002) 290–306 291
standard Monte Carlo sampling of the path integral in what is known as the ‘sign problem’.
Overviews with references to various studies of this problem may be found in [1].
Recently, a first (µ, T ) phase diagram for (2 + 1)-flavor QCD has been presented [2]
by using a two-dimensional generalization [3] of the Glasgow reweighting method [4]. For
such methods cancellations in the reweighting factor occur generically at large volumes
and/or chemical potentials, and it remains to be seen how well this approach allows for
infinite volume and continuum extrapolations. Another work [5] attempts to improve on
this aspect by expanding the product of reweighting factor and observable in powers
of (µ/T ), and computing the coefficients through second order. The calculation is then
equivalent to computing susceptibilities at µ = 0 [6], which is possible on any volume. The
µ/T -range of applicability is limited by either the µ/T -value of the expected critical point
marking the onset of a phase transition, where susceptibilities become non-analytic, or the
radius of convergence of the expansion, whichever is lower. However, a priori nothing is
known about the convergence of the expansion or the error introduced by its truncation.
Moreover, a problem common to all reweighting approaches is that so far it has been
impossible to assess the overlap of reweighted ensembles with the original. Hence, the
results obtained in [2,5] rest on as yet uncontrolled approximations.
In this paper we attempt to develop a controllable alternative approach, avoiding
reweighting in µ altogether. This is achieved by simulating with imaginary chemical
potential, where there is no sign problem and hence no need for reweighting. We employ
the idea [7] that in this case one may fit the non-perturbative data for an observable by
truncated Taylor series in µ/T , thus keeping full control over the associated systematic
error. In the parameter range where this is possible, and in the absence of non-analyticities,
the series may be analytically continued to real values of µ. Non-perturbative evidence for
the viability of this approach has been given for screening masses in the deconfined phase
[8], which can be simulated successfully for real and imaginary µ [9] in the framework of
dimensionally reduced QCD [10]. Here we extend this approach to the critical line Tc (µ).
This is possible by noting that, while giving the location of singularities in thermodynamic
limit, the critical line itself is an analytic function. After continuing it from imaginary to
real µ for various volumes, a finite volume scaling analysis of its location should also
reveal the nature of the transition. Since there is no sign problem, there are no limitations
on the volumes that can be simulated either. In particular, it should then be possible to
identify the location of the critical endpoint of the phase transition which is expected on
general theoretical grounds [11], and for which reweighted numerical results have been
presented in [2].
Our method is limited by a maximal imaginary part of the chemical potential, µcI =
πT /3, above which the system tunnels into an unphysical Z(3) sector. Along the transition
line, this corresponds to an accessible range of real µB 500 MeV, which well covers
densities in heavy-ion experiments [12]. We present numerical results for the location
of the critical line, and find that the truncation error of the Taylor series is far smaller
than statistical errors in the whole accessible range. A finite volume scaling analysis is
beyond the scope of our current computational resources and postponed to later work.
Nevertheless, our results indicate that the part of the phase diagram relevant for heavy-ion
collisions can be reliably obtained from simulations of imaginary chemical potentials.
292 P. de Forcrand, O. Philipsen / Nuclear Physics B 642 (2002) 290–306
In Section 2 we recall some general features of QCD at imaginary chemical potential,

which will be useful in the sequel. The analyticity of the critical line is derived in Section 3.
Numerical results for two flavor QCD with Wilson gauge action and Kogut–Susskind
fermions are presented in Section 4, while Section 5 discusses our approach in comparison
with previous ones. In Section 6 we draw our conclusions.
2. QCD at real and imaginary chemical potential
While this section contains no new results, we would like to discuss some qualitative
features in detail as they play an important role for the following. We denote by µ the
chemical potential for quark number Q, and by µB the chemical potential for baryon
number B = Q/3, i.e., µ = µB /3. The grand canonical partition function is compactly
written as

Z(V , µ, T ) = Tr e−(H −µQ)/T . (1)
An expansion for small chemical potentials proceeds in terms of the dimensionless
parameter µ̄ = µ/T , and hence our interest is in Z(V , µ̄, T ).
The presence of a chemical potential term in the action breaks the separate invariance
under Euclidean time reflection and charge conjugation. In particular, a time reflection can
be compensated by µ → −µ and vice versa, so that
Z(µ̄) = Z(−µ̄). (2)

As a consequence, observables symmetric under time reflection are even functions of µ̄.
Let µR , µI ∈ R denote the real and imaginary parts of µ:
µ = µR + iµI , (3)
and similarly for µ̄. The qualitative features of SU (N) QCD at imaginary chemical
potential, µR = 0, were studied a long time ago [13]. With dynamical fermions the Z(N)
symmetry of the pure gauge theory is explicitly broken. However, in the presence of a
complex chemical potential, a Z(N) transformation of the fermion fields is equivalent to
a shift in µI , thus leading to a new symmetry: the partition function is periodic in µI with
period 2πT /N [13],
Z(µ̄R , µ̄I ) = Z(µ̄R , µ̄I + 2π/N). (4)

The different Z(N) sectors are characterized by the phase ϕ of the Polyakov loop,

P (x) = P (x) eiϕ . (5)
At high temperatures one may use the perturbative effective potential V (ϕ) to obtain
the qualitative behaviour of the theory. In pure gauge theory P (x) = 0 for T < Tc
and P (x)
= 0 for T > Tc . In the deconfined phase there are N degenerate minima at
ϕ = 2πk/N, k = 1, . . . , N , separated by potential barriers, and the transition between them
is of first order [14]. For Nf > 0, µ = 0, explicit Z(N)-breaking results in an expectation
value for the Polyakov loop even in the confinement phase. The expectation value is
Fig. 1. Location of Z(3) transitions, Eq. (6). Analytic continuation is limited to the region within the arc.
real, ϕ = 0, corresponding to the true vacuum of the theory, while the other sectors are
metastable [15]. This is true both for T > Tc and T < Tc . When a chemical potential is
switched on [13,16,17], its real part stabilizes this situation by raising the energy of the
other Z(N) vacua, while the imaginary part has the opposite effect of lowering one of
them relative to the ϕ = 0 vacuum. Hence, once µI exceeds some critical value µcI , a phase
transition to a non-trivial Z(N) sector occurs. Non-perturbatively the same statement
follows from the aforementioned equivalence of certain µI -shifts and Z(N) rotations. The
combined symmetries Eqs. (2), (4) imply that this transition is periodically repeated for
larger values of µI . Moreover, its exact critical values are dictated to be [13]

2π 1
µ̄cI = k+ . (6)
N 2
On the other hand, for purely real µ this transition never occurs. This situation is sketched
qualitatively for SU (3) in Fig. 1. The order of the Z(3) transitions depends on temperature.
A perturbative calculation predicts it to be of first order in the deconfined phase, whereas
at low temperature a strong coupling analysis suggests it to be a smooth crossover [13]. We
shall present numerical results in support of this picture in Section 4.
Note that the first of these Z(3) transitions at µ̄cI = π/3 limits the useful information to
be obtained from imaginary chemical potentials: Eqs. (2), (4) imply
Z(µ̄I = π/3 + µ̄I ) = Z(µ̄I = π/3 − µ̄I ) (7)

so that all observables are symmetric about the transition at µ̄cI . In other words, all
expectation values for µ̄I > µ̄cI are simply copies of those for µ̄I < µ̄cI , determined by
reflection symmetry and periodicity. Moreover, here we are interested in analytically
continuing results obtained on the imaginary axis, and this is only possible for |µ̄| π/3
[8], represented by the arc of circle in Fig. 1. Furthermore, the observable to be continued
has to be free of additional singularities in this region. In the next section we will show
that this is the case for the critical coupling for deconfinement, and hence the associated
temperature Tc (µ). Along the deconfinement transition the latter is in the range T ∼ 160–
170 MeV, which means that our approach is restricted to baryon chemical potentials
µB 500 MeV.
3. Analyticity of the (pseudo-)critical line
On a L3 × Nt lattice, the rescaled chemical potential takes the form µ̄ = Nt aµ, which
up to the constant factor Nt is equivalent to the chemical potential in lattice units. On the
lattice one then explores the phase diagram in the (β, µ̄)-plane, which in the continuum
limit has to be converted to a (T , µ)-diagram.
The location and nature of a phase transition may be determined from the finite volume
scaling of susceptibilities of dimensionless operators (V = L3 ),
2 1
χ = V Nt O − O , O= O(x). (8)
V Nt x,t
In practice we shall use as O(x) the plaquette, the chiral condensate and the modulus
of the Polyakov loop. In a transition region, fluctuations are strong and susceptibilities
display a peak, χmax = χ(µ̄c , βc ), whose location determines the critical parameters. In
a finite volume, the susceptibility is always an analytic function of the parameters of the
theory, even in the presence of a phase transition. The latter reveals itself by a divergence of
χmax in the infinite volume limit, whereas χmax stays finite in the case of a crossover. The
order of the transition is determined by the rate of divergence: for a first order transition
χmax ∝ V , whereas for a second order phase transition, χmax ∝ V ρ with a critical exponent
ρ = γ /dν < 1 [18]. Alternatively, one may consider the finite size scaling of the critical
coupling βc (V ) itself, which attains its infinite volume limit as

βc (V ) − βc (∞) ∼ V −σ , (9)
where σ = 1 for a first order phase transition, σ = 1/dν < 1 for a second order phase
transition, and σ = 0 for a crossover. (For notation and numerical application to SU (3)
pure gauge theory, see, e.g., [19] and references therein).
At zero density, the deconfinement transition is signalled by a peak χmax = χ(0, βc ),
defining the critical coupling through the equations

∂χ ∂ 2 χ
= 0, < 0. (10)
∂β βc ∂β 2 βc
When the chemical potential is switched on, this peak extends as a “mountain ridge” into
the (β, µ̄)-plane. Starting at βc (0), the ridge is defined by maximizing χ while moving
into the (β, µ̄)-plane, i.e., by moving along the direction of the gradient ∇χ(µ̄, β) (or
the direction with the largest second derivative in the case of vanishing gradient). It has a
vanishing first and negative second derivative in the direction perpendicular to the gradient.
For the numerical analysis, however, it is more convenient to consider maxima in the
directions of β and µ̄, respectively. The two pairs of conditions

∂β χ = 0, ∂β2 χ < 0 , ∂µ̄ χ = 0, ∂µ̄2 χ < 0 (11)
Fig. 2. Schematic location of the lines defined by Eq. (11). In the infinite volume limit (right), the lines merge for
a phase transition, but stay separate for a crossover, the bifurcation marking the end point of the phase transition.
then define two separate lines bounding the ridge from below and above, as indicated in
Fig. 2 (left), where the ordering of the lines depends on the sign of the gradient. For large
enough volumes one expects a positive gradient while increasing µ towards a critical point,
and the first condition denotes the lower line. When the volume is taken to infinity, the
mountain ridge changes its shape: where there is a phase transition, it becomes singular and
infinitely narrow, i.e., the two lines merge into one. Where there is a crossover, the ridge
saturates at finite height with finite curvature, and the two lines remain separate (as in Fig. 2
(right) at small µ). In this case there is no phase transition and no uniquely defined pseudo-
critical line. Nevertheless, the physical properties change rapidly in the region of the line
defined by ∂β χ = 0, as is well known from the zero density case. Moreover, for small µ̄ this
line is the one closer to the ridge. Thus, for every µ̄ we may locate the thermal transition
by the conditions specified in Eq. (10), which then represent our implicit definition of the
critical line βc (µ̄).
Let us now consider a finite volume. Expanding the susceptibility around βc (0) and
using χ(µ̄, β) = χ(−µ̄, β), it takes the form
n
χ(µ̄, β) = cnm β − βc (0) µ̄2m . (12)
n,m=0
Because χ(µ̄, β) is analytic, its derivatives are as well. In the neighbourhood of the critical
line we furthermore have ∂β2 χ
= 0. The implicit function theorem2 then tells us that the
implicitly defined βc (µ̄) is analytic in µ̄. Furthermore, its first derivative is obtained from
the chain rule for partial differentiation,

∂βc ∂ 2 χ ∂ 2 χ −1
=− . (13)
∂ µ̄ ∂ µ̄∂β ∂β 2
2 See, e.g., [20]. We thank F. Wilczek for reminding us of this theorem in this context.
Since χ is even in µ̄, the same follows from the last equation for βc , which therefore has a
Taylor expansion

βc (µ̄) = an µ̄2n = cn (aµ)2n , (14)
n=0 n=0
where we have absorbed the temporal lattice length into the coefficients in the second
equation, cn = an Nt2n .
All of these considerations are unchanged if we consider a purely imaginary potential,
µ̄ = iµ̄I . In this case we may define a corresponding real function χ̃(µ̄I , β) with a Taylor
expansion in β, µ̄2I and coefficients c̃nm . Clearly, this is just the analytic continuation
χ̃(µ̄I , β) = χ(iµ̄I , β), with c̃nm = cnm (−1)m . Its maximum in the β-direction is again
defining a function βc (µ̄I ), which by the implicit function theorem is analytic. The
analogue of Eq. (13) then reads

∂βc ∂ 2 χ̃ ∂ 2 χ̃ −1 ∂ 2 χ ∂ 2 χ −1 ∂βc
=− = −i =i . (15)
∂ µ̄I ∂ µ̄I ∂β ∂β 2 ∂ µ̄∂β ∂β 2 ∂ µ̄
Since this equation is true for every µ̄I < µ̄cI we have thus established βc (µ̄I ) = βc (µ̄ =
iµ̄I ), i.e., the critical line of the deconfinement transition at imaginary chemical potential
is simply the analytic continuation of the critical line at real chemical potential.
Next, we need to discuss the infinite volume limit. As remarked above, for every µ̄
the critical coupling will approach its thermodynamic limit in a fashion characteristic of
the order of the transition. However, βc (∞, µ̄) is only shifted from βc (V , µ̄), without
developing any singularities in the thermodynamic limit. Hence, the critical line defined
by Eq. (10) marks the singularities in the partition function, which it smoothly connects to
a pseudo-critical line in the crossover region. It thus remains itself an analytic function of
µ̄ in the thermodynamic limit for all µ̄.
Our strategy is now to first compute the critical line for imaginary chemical potential
and check the convergence of its Taylor expansion. As we shall see, convergence is
surprisingly fast for the whole range of chemical potentials accessible to this method.
Analytic continuation then reduces to simply flipping the sign of the appropriate terms.
Finally, the infinite volume limit has to be taken from the continued results at real µ̄. The
order of the transition can then be determined in a (V , µ̄)-range where the truncation error
of the series is smaller than finite size scaling effects.
4. Numerical results for two light flavors
In order to test the feasibility of our approach, we consider QCD with two flavors of
staggered fermions with bare mass am = 0.025 and µR = 0 on a 83 × 4 lattice. We use
the R-algorithm [21] with a step size δτ = 0.02, sufficiently small for the systematic errors
O(δτ 2 ) to be negligible compared to our statistical errors. For 8 values of β spanning the
relevant temperature regime and 6 values of aµI ∈ {0, π/12}, we have accumulated 2000–
6000 unit-length trajectories each, measuring the gauge action, the Polyakov loop and the
chiral condensate after each trajectory. For one parameter set we have performed additional
simulations on 63 × 4 lattice. To determine the pseudo-critical value βc (aµI ), we use the

Ferrenberg–Swendsen reweighting method [22].
4.1. The Z(3) transition for imaginary µ
First, it is instructive to investigate the Z(3) transition discussed in Section 2. The

predicted critical value for the chemical potential is at µ̄cI = π/3, which on our Nt = 4
lattice corresponds to aµcI = π/12 = 0.262. For this value Fig. 3 shows histograms of the
phase ϕ of the Wilson loop, obtained for various lattice couplings (temperatures). We find
the system tunneling between the sectors with ϕ = 0 and ϕ = −2π/3.
In accord with the predictions made in [13], we observe a smooth distribution consistent
with a crossover at low temperatures, and a pronounced two-state signal indicating a first
order phase transition at high temperatures. The critical coupling for the onset of the Z(3)
transition is visibly between 5.313 < βc < 5.34.
For low temperatures, where the Z(3) transition is continuous, there is frequent
tunneling between the Z(3) sectors. It then follows that also for aµI < (aµI )c the ensemble
average consists of a mixture of the two sectors, where the ϕ = −2π/3 admixture grows
with aµI . This is clearly visible in Fig. 4, where the average phase of the Polyakov
loop gradually moves from ϕ(aµI = 0) = 0 to ϕ((aµI )c ) = −π/3, the latter value
corresponding to the average between the two phases ϕ = 0, −2π/3. For aµI > (aµI )c ,
the phase gets dominated by the vacuum with ϕ = −2π/3.
At high temperatures, on the other hand, the Z(3) sectors are separated by a diverging
potential barrier, and tunneling is more suppressed the higher the temperature. As a result
the Monte Carlo ensemble stays longer in one sector, and its equilibrium is dominated by
the lowest vacuum. Thus, as shown in Fig. 4, for aµI < (aµI )c the system always settles
Fig. 3. Probability distribution of the phase of the Polyakov loop for the critical value aµcI = π/12.
(a)
(b)
Fig. 4. The β- and µI -dependence of the average phase of the Polyakov loop.
in the trivial vacuum with zero phase, whereas for aµ > (aµI )c it settles in the vacuum
sector with ϕ = −2π/3. The dividing line between those situations is ϕ((aµI )c ) = −π/3,
which is independent of temperature by symmetry. In Fig. 4, one observes that indeed
ϕ((aµI )c ) = −π/3 for all β’s, although errors become large at large β. This reflects the
difficulty of tunneling at high temperatures, and implies longer and longer Monte Carlo
time for the system to equilibrate. Hence, at high temperatures great attention must be paid
to Z(3) ergodicity and its possible violation.
From the susceptibilities χ(β) at (aµI )c = π/12, only one peak is apparent. This means
that, within our accuracy, at (aµI )c the critical temperature at which the Z(3) first-order
Fig. 5. Schematic phase diagram for imaginary chemical potential: the vertical line marks the Z(3) transition
(this section), the curved lines the deconfinement transition (next section). The solid line indicates a first order
transition, while the nature of the dotted lines is not yet determined. The diagram is periodically repeated for
larger values of µI .
transition line starts is the same as the critical temperature for the deconfinement transition
determined below. Schematically the (T , µI ) phase diagram then looks as in Fig. 5, with a
special point where three critical lines meet. This point might be tricritical if deconfinement
corresponds to a genuine phase transition, or critical otherwise. We take a first look at this
issue below.
Once ergodicity is ensured, the effects of the Z(3) transition are also visible in other
observables like, e.g., the chiral condensate, as shown in Fig. 6. Note the symmetry of the
observable around aµcI according to Eq. (7). For low β (viz. temperatures) the observable
is continuous at the critical chemical potential, while a cusp indicating a discontinuous
transition develops for large β.
Let us now try to determine the endpoint (βc , aµI = π/12) of the first-order transition
line with better accuracy. We can do so by monitoring the plaquette susceptibility as a
function of β. This procedure gives us βc = 5.325(5). For improved accuracy, and to check
our control of the delicate ergodicity problems mentioned above, we performed additional
simulations on a 63 × 4 lattice at aµI = π/12. Since ϕ = −π/3, one can form a ratio
of cumulants ϕ̂ 4 /ϕ̂ 2 2 , where ϕ̂ = ϕ + π3 , and estimate the critical point βc from the
crossing of the cumulant ratios for the two available volumes. This procedure, illustrated
in Fig. 7, yields a consistent value βc = 5.321(7). Note also that the crossing occurs at a
value of the cumulant ratio 1.8(2) consistent with that of the 3d Ising universality class
1.604(1) [23]. This indicates as the simplest interpretation of the phase diagram: the first-
order Z(3) transition line ends with a second-order critical point in the Ising class. This
picture will likely depend on the number of quark flavors and on their masses.
Fig. 6. The chiral condensate ψ̄ψ as a function of aµI .
Fig. 7. Ratio of cumulants ϕ̂ 4 /ϕ̂ 2 2 , for aµI = π/12.
4.2. The deconfinement transition for imaginary µ
In order to map out the deconfinement transition line in the (β, µ̄I )-plane, we have
measured the susceptibility of the plaquette, of the modulus of the Polyakov loop, and the
disconnected part of a stochastic estimator for the chiral condensate susceptibility. All three
observables yield consistent results for βc (aµI ), as shown in Fig. 8. We have chosen to
display only three aµI -values for clarity of the plots. Note that the critical lattice coupling
is growing with increasing aµI , and thus the critical temperature is an increasing function
(a)
(b)
(c)
Fig. 8. Susceptibilities for the plaquette, the chiral condensate and the absolute value of the Polyakov loop.
Fig. 9. Location of the pseudo-critical deconfinement line as determined from plaquette susceptibilities. The lines
show the fits given in Table 1.
Table 1
The first coefficients of the Taylor expansion of the critical coupling, Eq. (14). Data and fits for βc from χplaq are
shown in Fig. 9
c0 c1 c2 χ 2 /dof Q

βc from χplaq O aµ2I 5.2865(18) 0.596(40) – 0.60 0.66

O (aµI )4 5.2857(23) 0.68(15) −1.2 (2.0) 0.67 0.57

βc from χ|P | O (aµI )2 5.2874(17) 0.640(38) – 0.29 0.89

O (aµI )4 5.2873(21) 0.65(14) −0.2 (2.0) 0.38 0.77
of µI as well. This is in contrast to the situation at real µ, where Tc (µ) is a decreasing

function. This qualitative behaviour is in accord with the one found for screening lengths
in units of inverse temperature, which are increasing (decreasing) functions of µI (µR ) [8].
(An increasing screening length means more energy is required to separate charges before
they are screened, and hence an increasing deconfinement temperature).
Since the plaquette susceptibility is the one that is most accurately determined, it is our
choice for determining the critical line. The values of βc corresponding to six values of aµI
are shown in Fig. 9. Our next task is to determine the reliability of the Taylor expansion
for the critical line, Eq. (14). Results of fitting the data to a polynomial of degree one and
two in (aµI )2 , respectively, are shown in Table 1 and Fig. 9. Excellent fits are obtained
in both cases with compatible coefficients c0,1 . The quartic correction appears negligible
even near aµI = (aµI )c , leaving c2 to be consistent with zero within this range. This is
again in accord with the analogous finding for screening masses, which are equally well
described by the leading (aµI )2 term in that range [8]. Statistically consistent results are
obtained when βc is extracted from Polyakov loop instead of plaquette susceptibilites,

as also shown in Table 1. We then conclude that the (pseudo-)critical line for imaginary
chemical potential µ̄ = iµ̄I , µ̄I π/3, is well described by
βc (aµI ) = c0 + c1 (aµI )2 , (16)
with c0 , c1 as per Table 1 (top line).
4.3. The deconfinement transition in physical units for real µ
With the results of the previous section, it is now trivial to obtain βc (µ̄) for real µ̄,
by continuing µ̄I → −iµ̄I . As a comment on systematics, let us add that the difference
between the linear and quadratic fits in (aµI )2 from Table 1 gets slightly amplified by
analytic continuation (only the sign of one term is flipped). This reflects the fact that
convergence properties in the real and imaginary directions may be different in general.
Such an effect will be large for an observable with pronounced structures, and small on a
smooth observable like ours. In any case, the effect can be monitored, and the error band
of the first fit is even after continuation completely contained within the error band of the
second fit, which can hence be dropped as statistically insignificant.
It is instructive to translate our result into physical units, in order to illustrate the
phenomenological relevance of our approach. However, we caution that at this stage such
a conversion is merely illustrative, with quantitative numbers still afflicted by various
systematic errors: while we have monitored that the truncation error of the Taylor series is
negligible, there might still be sizeable corrections from finite volume and lattice spacing
effects, and we have not investigated the quark mass dependence of our results. In view
of this we content ourselves with the perturbative two-loop expression for the lattice QCD
Fig. 10. Location of the deconfinement transition corresponding to the first fit in Table 1. The error bar gives the
uncertainty in Tc (0) used to set the scale, the dotted lines reflect the error on c1 from Table 1.
scale ΛL in the presence of two massless flavors,

−b1 /2b2
6b0 0
a(β)ΛL = exp(−β/12b0),
β
2
1 2 1 38
b0 = 11 − Nf , b1 = 102 − Nf ,
16π 2 3 16π 2 3
Tc (µ) a(βc (µ))ΛL
= (17)
Tc (0) a(βc (0))ΛL
while the scale is set by the critical temperature Tc (µ = 0) = 173(8) MeV in the chiral limit
for staggered fermions [24] (which is expected to still contain finite lattice spacing errors).
This leads to the result for the (pseudo-)critical line as shown in Fig. 10. As remarked
before, a detailed finite volume study is beyond the scope of the present work but would
reveal the order of the transition along the line.
5. Comparison with other methods
Before concluding, we would like to make some comments comparing our method with
previous studies of the finite µ deconfinement transition [2,5]. The main disadvantage of
our approach is its limitation to the range |µ|/T < π/3, as discussed in Section 2. On
the other hand, within this range we have full control over the necessary approximation,
i.e., the truncation of the Taylor series, which we find to pose no limitation at all within the
accessible range.
We would like to illustrate the importance of this aspect by a comparison with the
reweighting method [2,3]. In Ref. [3], the overlap of the reweighted with the full ensemble
is tested for imaginary µ. The configurations used for reweighting from the µ = 0
ensemble are all in the real Z(3) sector, so that reweighting is by construction insensitive
to the Z(3) transition. This is clearly seen in Fig. 1 of Ref. [3], where the reweighted
chiral condensate is found to be a monotonically increasing function of (aµI ) in the range
0 aµI 0.32. Such a result is inconsistent with the symmetries of the imaginary-µ
theory. Ref. [3] works at the same lattice spacing as we do, and therefore the critical
chemical potential for Z(3) tunneling is at (aµI )c = π/12. According to Eq. (7), the chiral
condensate has to be symmetric about this point and decrease for larger aµI , as it does in
our data, Fig. 6. Not surprisingly, reweighting completely misses this qualitative behavior.3
In the case of real µ, there is no Z(3) transition. However, the relevance of complex Z(3)
sectors was studied in [17]. It was argued there that constraining the Polyakov loop to
be real is a reasonable approximation below and above the deconfinement transition, but a
poor one near it. In any case, the discussion illustrates a potential danger of any reweighting
3 Fig. 1 of Ref. [3] also shows the chiral condensate rising monotonically in the full ensemble. A possible
reason for this strange behavior might be insufficient thermalization, cf. the discussion in Section 4.1. Indeed, at
yet larger µI , the authors of Ref. [3] find, as expected, clear disagreement between reweighted and full ensemble
results [25].
method: it may completely miss a qualitative change in the physics without announcing its
failure, which only shows up by comparison with results obtained without reweighting.
Ref. [2] studied Nf = 2 + 1 flavors, with light quark mass values and lattice spacing as
ours, whereas the bare strange quark mass was chosen eight times heavier than the light
ones. Keeping this difference in mind, we find the location of our continuum transition line
consistent with theirs over the whole range we have studied.
As mentioned in the introduction, another recent work [5] employs a Taylor expansion
of the reweighted path integral. However, in this case only susceptibilities are measured, so
that one has no control over the higher order terms. Let us recall that the Taylor expansion
is in µ/T , which becomes larger than one already at µ ∼ 170 MeV. A priori there is no
way of knowing how large the convergence radius of such an expansion is. Simulating
at imaginary µ provides a framework for obtaining quantitative information about the
convergence. As it turns out, our results non-perturbatively endorse the use of the leading
Taylor coefficient in Tc (µ) up to µ ∼ 170 MeV. Ref. [5] works at quark masses four to
ten times larger than ours, but finds only very weak quark mass dependence. Keeping this
difference in mind, we find our result for the slope c1 of βc ((aµ)2 ) to be consistent with
theirs.
6. Conclusions
The past year has seen progress in the numerical study of finite density QCD. However,
this progress is essentially based on adopting a more pragmatic viewpoint. The net quark
density in heavy-ion collisions is small. Therefore, instead of solving the fundamental
difficulty posed by the sign problem, it is very useful already to explore the regime of small
chemical potential, where the sign problem is moderate and various approaches appear
possible.
We have studied QCD with two light flavors of staggered fermions in the presence of
an imaginary chemical potential µ = iµI . In this case the partition function is periodic in
µI and, in addition to the deconfinement transition, Z(3) transitions occur at (µI /T )c =
2π(k + 1/2)/3. We mapped out the phase diagram in the (T , µI )-plane and found that
the Z(3) transitions are of first order in the deconfined phase and continuous in the
confined phase. For (µI /T ) < π/3, the critical temperature for deconfinement is an
analytic function, irrespective of the order of the transition. Within this range its Taylor
expansion is found to rapidly converge and to be well described by the leading term.
Other recent work uses reweighting of simulations performed at µ = 0, and at one or
more temperatures. In contrast, we vary two parameters in our simulations: the temperature
as usual, and also the imaginary chemical potential µI . Having at our disposal two-
dimensional information allows us a reasonable control over systematic errors.
We have used this property to determine the transition line Tc (µ) for real µ through
analytic continuation. While we cannot, on a single volume, find out whether the transition
is a crossover or a genuine phase transition at each value of µ, a finite-size scaling study
should elucidate this issue, provided that finite size effects are larger than the truncation
error. Our central result is shown in Fig. 10. It is consistent with Refs. [2,5]. The transition
line is well represented by the equation

Tc (µB ) µB 2
= 1 − 0.00563(38) , (18)
Tc (µB = 0) T
while the next-order term O((µB /T )4 ) is statistically insignificant up to µB ∼ 500 MeV.
Acknowledgements
We thank V. Azcoiti, O. Bär, Z. Fodor, M. García Pérez, M. Laine, M.-P. Lombardo,

K. Rajagopal and F. Wilczek for discussions and comments.
References
[1] S. Hands, Nucl. Phys. (Proc. Suppl.) 106 (2002) 142, hep-lat/0109034;
O. Philipsen, Nucl. Phys. (Proc. Suppl.) 94 (2001) 49, hep-lat/0011019;
O. Philipsen, hep-ph/0110051.
[2] Z. Fodor, S.D. Katz, JHEP 0203 (2002) 014, hep-lat/0106002.
[3] Z. Fodor, S.D. Katz, Phys. Lett. B 534 (2002) 87, hep-lat/0104001.
[4] I.M. Barbour, S.E. Morrison, E.G. Klepfish, J.B. Kogut, M.P. Lombardo, Nucl. Phys. (Proc. Suppl.) A 60
(1998) 220, hep-lat/9705042.
[5] C.R. Allton, et al., hep-lat/0204010.
[6] S. Gottlieb, et al., Phys. Rev. D 38 (1988) 2888;
QCD-TARO Collaboration, P. de Forcrand, et al., hep-lat/0011013;
S. Choe, et al., Phys. Rev. D 65 (2002) 054501;
R.V. Gavai, S. Gupta, P. Majumdar, Phys. Rev. D 65 (2002) 054506, hep-lat/0110032.
[7] M.P. Lombardo, Nucl. Phys. (Proc. Suppl.) 83 (2000) 375, hep-lat/9908006.
[8] A. Hart, M. Laine, O. Philipsen, Phys. Lett. B 505 (2001) 141, hep-lat/0010008.
[9] A. Hart, M. Laine, O. Philipsen, Nucl. Phys. B 586 (2000) 443, hep-ph/0004060.
[10] P. Ginsparg, Nucl. Phys. B 170 (1980) 388;
T. Appelquist, R.D. Pisarski, Phys. Rev. D 23 (1981) 2305.
[11] K. Rajagopal, Nucl. Phys. A 661 (1999) 150, hep-ph/9908360;
J. Berges, K. Rajagopal, Nucl. Phys. B 538 (1999) 215, hep-ph/9804233;
M.A. Stephanov, K. Rajagopal, E.V. Shuryak, Phys. Rev. Lett. 81 (1998) 4816, hep-ph/9806219.
[12] J. Cleymans, K. Redlich, Phys. Rev. C 60 (1999) 054908, nucl-th/9903063, and references therein.
[13] A. Roberge, N. Weiss, Nucl. Phys. B 275 (1986) 734.
[14] D.J. Gross, R.D. Pisarski, L.G. Yaffe, Rev. Mod. Phys. 53 (1981) 43;
N. Weiss, Phys. Rev. D 24 (1981) 475.
[15] V. Dixit, M.C. Ogilvie, Phys. Lett. B 269 (1991) 353;
J. Ignatius, K. Kajantie, K. Rummukainen, Phys. Rev. Lett. 68 (1992) 737.
[16] C.P. Korthals Altes, R.D. Pisarski, A. Sinkovics, Phys. Rev. D 61 (2000) 056007, hep-ph/9904305.
[17] P. de Forcrand, V. Laliena, Phys. Rev. D 61 (2000) 034502, hep-lat/9907004.
[18] M.N. Barber, in: C. Domb, J.L. Lebowitz (Eds.), Phase Transitions and Critical Phenomena, Vol. 8,
Academic Press, New York, 1983.
[19] M. Fukugita, M. Okawa, A. Ukawa, Nucl. Phys. B 337 (1990) 181;
Y. Iwasaki, et al., Phys. Rev. D 46 (1992) 4657.
[20] See, e.g., P. Henrici, in: Applied and Computational Complex Analysis, Wiley, 1974, p. 101.
[21] S. Gottlieb, W. Liu, D. Toussaint, R.L. Renken, R.L. Sugar, Phys. Rev. D 35 (1987) 2531.
[22] A.M. Ferrenberg, R.H. Swendsen, Phys. Rev. Lett. 63 (1989) 1195.
[23] H.W.J. Blöte, E. Luijten, J.R. Heringa, J. Phys. A 28 (1995) 6289, cond-mat/9509016.
[24] F. Karsch, E. Laermann, A. Peikert, Nucl. Phys. B 605 (2001) 579, hep-lat/0012023.
[25] Z. Fodor, private communication.
L USIFER : a LUcid approach to SIx-FERmion

production
Stefan Dittmaier a,1 , Markus Roth b
a Deutsches Elektronen-Synchrotron DESY, D-22603 Hamburg, Germany
b Institut für Theoretische Physik, Universität Karlsruhe, D-76131 Karlsruhe, Germany
Received 10 June 2002; accepted 31 July 2002
Abstract
L USIFER is a Monte Carlo event generator for all processes e+ e− → 6 fermions, which is based on
the multi-channel Monte Carlo integration technique and employs the full set of tree-level diagrams.
External fermions are taken to be massless, but can be arbitrarily polarized. The calculation of
the helicity amplitudes and of the squared matrix elements is presented in a compact way. Initial-
state radiation is included at the leading logarithmic level using the structure-function approach. The
discussion of numerical results contains a comprehensive list of cross sections relevant for a 500 GeV
collider, including a tuned comparison to results obtained with the combination of the W HIZARD
and M ADGRAPH packages as far as possible. Moreover, for off-shell top-quark pair production
and the production of a Higgs boson in the intermediate mass range we additionally discuss some
phenomenologically interesting distributions. Finally, we numerically analyze the effects of gauge-
invariance violation by comparing various ways of introducing decay widths of intermediate top
quarks, gauge and Higgs bosons.
1. Introduction
At future e+ e− colliders, such as TESLA [1], some of the most interesting elementary
particle reactions lead to final states involving six fermions. Typically these multi-particle
final states represent the final decay stage of unstable particles that were produced as
resonances in subprocesses. Because of the high precision of future colliders, predictions
that are entirely based on the narrow-width approximation (with possible improvements
E-mail address: stefan.dittmaier@desy.de (S. Dittmaier).

1 Heisenberg fellow of the Deutsche Forschungsgemeinschaft.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 4 0 - 5
308 S. Dittmaier, M. Roth / Nuclear Physics B 642 (2002) 307–343
by spin correlations between production and decays or by resonance expansions) are not
sufficient in the most cases. In practice, this means that the full set of Feynman diagrams
(not only the resonant ones) has to be considered in perturbative calculations, at least in
lowest order. The situation is very similar to four-fermion production, e+ e− → 4f , and
the related radiative processes e+ e− → 4f + γ , at LEP [2]. Although the complexity of
the corresponding calculations increases when turning from four-fermion to six-fermion
production, e+ e− → 6f , most of the existing results are obtained by applying and further
developing the methods that had been worked out for four-fermion production. We briefly
summarize these results according to the subprocesses of interest:
(i) Top-quark pair production
Since top quarks decay via the cascade t → bW+ → bf f¯ into three fermions, the
production of tt̄ pairs corresponds to a particular class of e+ e− → 6f processes: e+ e− →
bb̄f1 f¯1 f2 f¯2 . Here fi f¯i denote two weak isospin doublets. Some processes are already
discussed in the literature. The specific process e+ e− → bbu ¯ d̄µ− ν̄µ was studied in Ref. [3]
with the GRACE package [4]. In Refs. [5–7] the cases of one and two hadronically
decaying W bosons were discussed in more detail, respectively. The former results are
based on a generalization of the program PHACT [8], the latter on the ALPHA algorithm
[9] for matrix elements. Recently various resonance approximations for the top quarks were
compared with a calculation based on full sets of e+ e− → 6f diagrams in Ref. [10]; these
results underline the importance of calculations based on full sets of Feynman diagrams.
However, to our knowledge, results have not yet been presented for all final states, for
instance, not yet for e+ e− → bb̄e− ν̄e νe e+ .
(ii) Vector-boson scattering and quartic gauge-boson couplings
One of the most promising windows to electroweak symmetry breaking is provided
by investigating the scattering of massive vector bosons, V1 V2 → V3 V4 . This subprocess
is initiated by the emission of the vector bosons V1,2 from the incoming e+ e− system
and leads to six fermions in the final state: two remnants from the initial state and four
fermions from the decays of the vector bosons V3,4 . Thus, the corresponding reactions
are of the form e+ e− → e+ e− /e+ νe /ν̄e e− /ν̄e νe + 4f . Many studies of vector-boson
scattering were presented in the literature (see, e.g., Ref. [1] and references therein), but
with very few exceptions they were based on approximations with respect to the kinematics
of the incoming and/or the outgoing vector bosons. In Refs. [11,12] the sensitivity of
the processes e+ e− → νe ν̄e + 4 quarks to possible anomalous quartic gauge-boson
couplings was investigated making use of full e+ e− → 6f matrix elements. The matrix
elements used in Ref. [11] were obtained with ALPHA, while in Ref. [12] the package
O’M EGA [13] delivered the amplitudes and the phase-space generation was performed
with W HIZARD [14].
(iii) Higgs production for intermediate Higgs-boson masses
If the Higgs boson of the Standard Model has an intermediate mass of MH
150 GeV, it predominantly decays via H → WW → 4f . Since the Higgs boson is
either produced by Higgs-strahlung off Z bosons, e+ e− → ZH, or vector-boson fusion,
e+ e− → e+ e− H/ν̄e νe H, the search for the Higgs boson in the intermediate mass range
also proceeds via e+ e− → 6f processes. In Refs. [6,15–17] the Higgs-strahlung signal
S. Dittmaier, M. Roth / Nuclear Physics B 642 (2002) 307–343 309
was studied for various 6f final states by employing full matrix elements from PHACT
and ALPHA. Ref. [17] also contains some results for the vector-boson fusion channel.
(iv) Three-gauge-boson production
Last but not least, all 6f final states in e+ e− → 6f contribute to the signal
of resonant three-gauge-boson production, such as e+ e− → WWZ/ZZZ, from which
valuable information on the quartic gauge-boson couplings can be deduced. In Refs. [5,6]
WWZ production was investigated for some interesting final states using the full 6f matrix
elements from PHACT. However, more detailed studies including more final states are
certainly wanted.
It should be mentioned that the combination of the PHEGAS and HELAC packages
[18] is also able to deal with six-fermion production processes. However, no detailed results
of these programs for e+ e− → 6f have been presented yet in the literature.
In this paper we present the Monte Carlo event generator L USIFER, which is, in its
first version, designed for all Standard Model processes e+ e− → 6 fermions in lowest
order; gluon-exchange diagrams can be optionally included for final states with two leptons
and four quarks (not yet for six-quark final states).2 Technically the approach closely
follows the structure of E XCALIBUR [20] and the lowest-order part of R ACOON WW [21,
22] for the processes e+ e− → 4f (+γ ). This means the matrix elements are evaluated
within the Weyl–van der Waerden (WvdW) spinor technique as described in Ref. [23]
(see also references therein); the external fermions are taken to be massless, but can be
arbitrarily polarized. The phase-space integration is performed with the multi-channel
Monte Carlo integration technique [24], improved by adaptive weight optimization [25].
The lowest-order predictions obtained this way are dressed by initial-state radiation (ISR)
in the leading logarithmic approximation following the structure-function approach [26],
as summarized in Appendix of Ref. [27].
Although the above list of topics shows that several studies of six-fermion production
processes have already been presented in the literature, no detailed tuned comparison
between results from different approaches is available yet. We make a first step to fill
this gap by giving a full list of cross sections for 6f states with up to three neutrinos and
up to four quarks for a centre-of-mass (CM) energy of 500 GeV. Moreover, we compare
these cross sections with results obtained by W HIZARD [14] and M ADGRAPH [28] as far
as the combination of these packages is applicable. In general, we find good numerical
agreement, but for several channels the limitation of these multi-purpose programs
becomes visible. We continue the tuned comparison by comparing some distributions for
topics of phenomenological interest, such as invariant-mass and angular distributions for
top-quark and Higgs-boson production. We conclude the discussion of numerical results by
comparing various schemes for introducing the finite decay widths of unstable particles in
the amplitudes. Already for CM energies in the TeV range the gauge-invariance-breaking
effects in some cases are clearly visible, underlining the importance of this issue. Within
L USIFER several width schemes are implemented, comprising also the complex-mass
2 A full treatment of the production of four or more jets in e+ e− → 6 jet reactions additionally requires the
inclusion of gluon jets, as for instance done in Ref. [19].
scheme, which was introduced in Ref. [21] for tree-level predictions and maintains gauge
invariance. Hence, gauge-violating artefacts can be controlled by comparing a given width
scheme with the complex-mass scheme.
The paper is organized as follows: in Section 2 the calculation of the helicity amplitudes
and of the squared matrix elements is described in detail, followed by a description of
the multi-channel phase-space integration in Section 3. The treatment of ISR is described
in Section 4. Section 5 contains a classification of the processes e+ e− → 6f according
to the underlying resonance subprocesses. In Section 6 we present numerical results,
including a comprehensive list of cross sections, their tuned comparison with W HIZARD
and M ADGRAPH results, some distributions relevant for top-quark and Higgs-boson
production, and the discussion of gauge-invariance-breaking effects. A summary is given
in Section 7.
2. Matrix-element calculation
2.1. Generic amplitudes for eight external fermions
The Feynman diagrams contributing to a process with 8 external fermions can

be classified into different categories according to the number of external fermion–
antifermion pairs that directly fuse to a boson. At tree level there are at least two such pairs,
called “fermion currents” in the following, so that we have three categories: diagrams with
4, 3, or 2 fermion currents. Assuming massless external fermions, the generic diagrams of
these classes are shown in Figs. 1–3. The vector bosons V... represent all gauge-boson fields
γ , Z, W± , g that are allowed by the quantum numbers of the external fermions. Whenever
a gauge boson has to be electrically charged, it is already denoted by W; if a top quark
is present, the charge flow is automatically fixed in the diagram. All purely electroweak
diagrams are supported for arbitrary six-fermion final states. L USIFER optionally includes
also gluon-exchange diagrams, but in its first version such diagrams are only included for
up to four quarks in the final state. This, in particular, implies that there are no gluonic
diagrams of the type shown in Fig. 1. The scalar boson S stands both for the Higgs-boson
field H and for the would-be Goldstone partners χ and φ ± of the gauge bosons Z and W± .
The amplitudes given below are all evaluated within the ’t Hooft–Feynman gauge. Among
Fig. 1. Generic diagrams with 4 fermion currents.

the internal fermions F... the top quark plays a special role, since it is the only fermion
that receives a mass, mt . Note that the first three diagrams in each class appear already in
a theory with only massless fermions, while diagrams (3d), (3e), and (2d) are proportional
to a power of the top-quark mass due to the appearance of the top-quark Yukawa coupling
in the amplitude.
In the generic diagrams the external fermions fa,...,h , which are simply denoted as
a, . . . , h, carry incoming momenta pa,...,h and helicities σa,...,h , respectively. The helicity
amplitudes of these diagrams are calculated within the Weyl–van der Waerden (WvdW)
formalism following the conventions of Ref. [23]. The amplitudes are expressed in terms
of WvdW spinor products,

√ −iφp θp θq −iφq θq θp
pq = pA qB = 2 p0 q0 e
AB
cos sin − e cos sin , (2.1)
2 2 2 2
where pA , qA are the associated momentum spinors for the massless momenta
pµ = p0 (1, sin θp cos φp , sin θp sin φp , cos θp ),

q µ = q0 (1, sin θq cos φq , sin θq sin φq , cos θq ). (2.2)
Fermions are assumed to be incoming and, if necessary, turned into outgoing ones
by crossing, which is performed by inverting the corresponding fermion momenta and
helicities. If spinor products appear with negative momenta −p, −q as arguments, it is
understood that only the complex conjugate spinor products get the corresponding sign
change. We illustrate this by simple examples:
A(p, q) = pq = A(p, −q) = A(−p, q) = A(−p, −q),

B(p, q) = pq∗ = −B(p, −q) = −B(−p, q) = B(−p, −q). (2.3)
The denominator parts of the propagators for vector bosons V and fermions F are
abbreviated by
1
PV (p) = ,
p2 − MV2
+ iMV ΓV (p2 )
1
PF (p) = , (2.4)
p − mF + imF ΓF (p2 )
2 2
where the fermion mass mF is only non-zero for the top quark. The introduction of finite
decay widths ΓV (p2 ) and ΓF (p2 ) in the propagators is described below. Moreover, we
introduce abbreviations for sums of external momenta,
pab = pa + pb , pabc = pa + pb + pc , pabcd = pa + pb + pc + pd . (2.5)

For the Feynman rules and coupling factors we follow the conventions of Ref. [29]. For the
gauge-boson self-interactions we define the following constants,
CWW∓ V = CWV W± = ±gV WW , V = γ , Z,

CWWV1 V2 = −gV1 WW gV2 WW , V1,2 = γ , Z,
1
CWWWW = 2 , (2.6)
sw
with the abbreviations

cw
gγ WW = 1, gZWW = − . (2.7)
sw
The sine and cosine of the weak mixing angle are defined by the masses of the Z and W
bosons as follows:
2
MW
2
cw = 1 − sw2 = . (2.8)
MZ2
Here and in the following the fields denoted in the subscripts are assumed to be incoming.
If the charge flow does not matter, we simply write W and φ instead of W± and φ ± ; the
corresponding coupling is, of course, understood as zero if charge conservation would be
violated. For the couplings of scalar to gauge bosons we introduce
MW
CHZZ = 2s
,
cw w
MW
CHWW = ,
sw
Cφγ W = CφWγ = −MW ,
sw MW
CφZW = CφWZ = − , (2.9)
cw
and for the couplings of the electroweak gauge bosons to fermions
3
Iw,i
sw
gγσ f¯ f = −Qi , gZσf¯ f = − Qi + δσ,− ,
i i i i cw cw sw
1
σ
gW f¯i fi
=√ δσ,− , (2.10)
2sw
where Qi and Iw,i 3
= ±1/2 denote the relative charge and the weak isospin of the fermion
fi , respectively. The field fi corresponds to the weak-isospin partner of fi . For the gluon
coupling to quarks q we introduce
gs
ggσq̄q = , (2.11)
e
√ √
where gs = 4παs is the strong gauge coupling and e = 4πα is the electromagnetic
coupling. The factor e in the denominator is introduced, because we will separate the global
factor e6 from the e+ e− → 6f helicity amplitudes. By convention, the colour operator as
well as the colour indices of external quarks are split off from the generic amplitudes given
below; the whole colour structure will be reinserted when squaring the amplitudes. Finally,
we need the Yukawa coupling of the top quark, for which we define
mt mt
gφσt̄b = √ δσ,+ , gφσtb¯ = √ δσ,− . (2.12)
2sw MW 2sw MW
σ appears in the following with subscripts not listed here, it
If any of the constants C... or g...
is understood to vanish.
The amplitudes for the generic graphs with 4 fermion currents (see Fig. 1) are given by
Mσ(4a)
a ,...,σh
(pa , . . . , ph ) = 4e6δσa ,+ δσb ,− δσc ,+ δσd ,− δσe ,−σf δσg ,−σh
−
g−
σ σ
× gW g f g h C
f¯ f Wf¯ f V f¯ f V f¯ f WWV1 V2
a b c d 1 e f 2 g h
× PW (pab )PW (pcd )PV1 (pef )PV2 (pgh )

σ ,σ ,σe ,σg
× A(4a)
a c
(pa , . . . , ph ), (2.13)
Mσ(4b)
a ,...,σh
(pa , . . . , ph ) = 4e6δσa ,+ δσb ,− δσc ,−σd δσe ,+ δσf ,− δσg ,−σh
−
× gW g σd g− g σh C C
f¯a fb V1 f¯c fd Wf¯e ff V2 f¯g fh WV1 V3 WV2 V3
× PW (pab )PV1 (pcd )PW (pef )PV2 (pgh )PV3 (pabcd )
σ ,σ ,σe ,σg
× A(4b)
a c
(pa , . . . , ph ), (2.14)
σa ,...,σh
M(4c) (pa , . . . , ph ) = −4e6δσa ,−σb δσc ,−σd δσe ,−σf δσg ,−σh
σ σ σ σ
× gVb f¯ f gVd f¯ f gVf f¯ f gVhf¯ f CSV1 V2 CSV3 V4
1 a b 2 c d 3 e f 4 g h
× PV1 (pab )PV2 (pcd )PV3 (pef )PV4 (pgh )PS (pabcd )
σ ,σ ,σe ,σg
× A(4c)
a c
(pa , . . . , ph ), (2.15)
where the auxiliary functions A...
... contain the WvdW spinor products. Explicit results for
the auxiliary functions read
A++++
(4a) (pa , . . . , ph )
= 2pb pd ∗ pf ph ∗ pa pc pe pg − pb pf ∗ pd ph ∗ pa pe pc pg
− pb ph ∗ pd pf ∗ pa pg pc pe , (2.16)
A++++
(4b) (pa , . . . , ph )
= pb pd ∗ pf ph ∗ pa pc pe pg (pab − pcd ) · (pef − pgh )

+ 2pb pd ∗ pa pc pf Pab pe ph Pcd pg − pf Pcd pe ph Pab pg

+ 2pf ph ∗ pe pg pb Pef pa pd Pgh pc − pb Pgh pa pd Pef pc

+ 2pb Pcd pa pd ph ∗ pc pg pf Pgh pe − pd pf ∗ pc pe ph Pef pg

− 2pd Pab pc pb ph ∗ pa pg pf Pgh pe − pb pf ∗ pa pe ph Pef pg ,
(2.17)
A++++ ∗ ∗
(4c) (pa , . . . , ph ) = pb pd pf ph pa pc pe pg . (2.18)
Here the dot in the first line of Eq. (2.17) indicates the usual Lorentz product of four-
vectors, and the extended brackets are shorthands for the expressions
pa Pbc pd = pb pa ∗ pb pd + pc pa ∗ pc pd . (2.19)

The remaining helicity configurations of the auxiliary functions follow from discrete
symmetries. All non-vanishing A...
... are obtained from Eqs. (2.16)–(2.18) by the simple
substitutions
−σ ,σc ,σe ,σg
a
A(4...) (pa , pb , pc , pd , pe , pf , pg , ph )
σa ,σc ,σe ,σg
= A(4...) (pb , pa , pc , pd , pe , pf , pg , ph ),
σa ,−σc ,σe ,σg
σa ,σc ,σe ,σg
= A(4...) (pa , pb , pd , pc , pe , pf , pg , ph ),
σ ,σ ,−σe ,σg
a c
σa ,σc ,σe ,σg
= A(4...) (pa , pb , pc , pd , pf , pe , pg , ph ),
σa ,σc ,σe ,−σg
σa ,σc ,σe ,σg
= A(4...) (pa , pb , pc , pd , pe , pf , ph , pg ). (2.20)
Another useful relation is provided by the fact that taking the complex conjugate of A......
reverses all helicities,
−σa ,−σc ,−σe ,−σg σa ,σc ,σe ,σg ∗
A(4...) (pa , . . . , ph ) = A(4...) (pa , . . . , ph ) , (2.21)
which is useful for checking the amplitudes. For the class of diagrams with 4 fermion
currents it is obvious that we have helicity conservation for the respective (massless)
fermion–antifermion pairs, i.e., each diagram only contributes if
σa = −σb , σc = −σd , σe = −σf , σg = −σh . (2.22)

Mσ(3a)
a ,...,σh
(pa , . . . , ph ) = 8e6δσa ,−σb δσc ,−σd δσe ,−σf δσg ,−σh
× gV−σfa¯ F gVb F f gVd f¯ f gVf f¯ f gVh f¯ f
σ σ σ σ
1 a 1 3 2 b 1 c d 2 e f 3 g h
× PF1 (pacd )PF2 (pbgh )PV1 (pcd )PV2 (pef )PV3 (pgh )
σa ,σc ,σe ,σg
× gVτ F F A(3a),τ (pa , . . . , ph ), (2.23)
2 1 2
τ =±
σa ,...,σh
M(3b) (pa , . . . , ph ) = −4e6δσa ,−σb δσc ,−σd δσe ,+ δσf ,− δσg ,−σh
× gV−σfa¯ F gVb Ff gVd f¯ f gW
−
σ σ σ
g h C
f¯ f V f¯ f WV2 V3
1 a 2 b 1 c d e f 3 g h
× PF (pacd )PV1 (pcd )PV2 (pefgh )PW (pef )PV3 (pgh )

σ ,σ ,σe ,σg
× A(3b)
a c
(pa , . . . , ph ), (2.24)
Mσ(3c)
a ,...,σh
(pa , . . . , ph ) = −4e6δσa ,−σb δσc ,−σd δσe ,+ δσf ,− δσg ,−σh
× gV−σfa¯ F gVσb Ff gVσd f¯ f gW
−
g σh C
f¯e ff V3 f¯g fh WV2 V3
2 a 1 b 1 c d
× PF (pbcd )PV1 (pcd )PV2 (pefgh )PW (pef )PV3 (pgh )

σ ,σ ,σe ,σg
× A(3c)
a c
(pa , . . . , ph ), (2.25)
Mσ(3d)
a ,...,σh
(pa , . . . , ph ) = 4e6δσa ,+ δσb ,− δσc ,+ δσd ,− δσe ,+ δσf ,− δσg ,−σh
−
g− g− g−
σ
× gW g h C
f¯ t φ t̄f Wf¯ f Wf¯ f V f¯ f φWV
a b c d e f g h
× Pt (pacd )PW (pcd )PW (pefgh )PW (pef )PV (pgh )

σ ,σ ,σe ,σg
× A(3d)
a c
(pa , . . . , ph ), (2.26)
σ ,...,σh
M(3e)
a
(pa , . . . , ph ) = 4e6δσa ,+ δσb ,− δσc ,+ δσd ,− δσe ,+ δσf ,− δσg ,−σh
× gφ+f¯ t gW
−
g− g− g σh C
t̄f Wf¯ f Wf¯ f V f¯ f φWV
a b c d e f g h
× Pt (pbcd )PW (pcd )PW (pefgh )PW (pef )PV (pgh )

σ ,σ ,σe ,σg
× A(3e)
a c
(pa , . . . , ph ). (2.27)
At first sight it seems that the occurrence of a top quark in the spinor chain between fa
and fb can violate helicity conservation in this line. However, this is not true since the
top-quark propagator always appears between two chirality projectors, restoring helicity
conservation in the whole fermion line. Thus, as in the case of 4 fermion currents all graphs
with 3 fermion currents vanish if Eq. (2.22) is not fulfilled. We have already made use of
Eq. (2.22) in the expressions for the matrix elements Eqs. (2.23)–(2.27). The auxiliary
functions A...
... read
A++++ ∗ ∗
(3a),+ (pa , . . . , ph ) = −mF1 mF2 pb ph pd pf pa pc pe pg ,
A++++ ∗
(3a),− (pa , . . . , ph ) = −pb ph pa pc pd Pac pe pf Pbh pg , (2.28)

A++++ ∗ ∗
(3b) (pa , . . . , ph ) = pa pc pf ph pe pg pa pd pb Pef pa − pb Pgh pa

+ pc pd ∗ pb Pef pc − pb Pgh pc
+ 2pb ph ∗ pd Pac pg pf Pgh pe

− 2pb pf ∗ pd Pac pe ph Pef pg , (2.29)

A++++ ∗ ∗
(3c) (pa , . . . , ph ) = pb pd pf ph pe pg −pb pc

× pb Pef pa − pb Pgh pa

+ pc pd pd Pef pa − pd Pgh pa
− 2pa pg ph Pbd pc pf Pgh pe

+ 2pa pe pf Pbd pc ph Pef pg , (2.30)
A++++ ++++
(3d) (pa , . . . , ph ) = A(3e) (pa , . . . , ph )
= mt pb pd ∗ pf ph ∗ pa pc pe pg , (2.31)
where the expressions for the remaining polarizations are again obtained by simple
substitutions,
σa ,−σc ,σe ,σg
A(3...),τ (pa , pb , pc , pd , pe , pf , pg , ph )
σa ,σc ,σe ,σg
= A(3...),τ (pa , pb , pd , pc , pe , pf , pg , ph ),
σa ,σc ,−σe ,σg
σa ,σc ,σe ,σg
= A(3...),τ (pa , pb , pc , pd , pf , pe , pg , ph ),
σ ,σ ,σ ,−σg
a c e
σa ,σc ,σe ,σg
= A(3...),τ (pa , pb , pc , pd , pe , pf , ph , pg ),
−σa ,−σc ,−σe ,−σg σa ,σc ,σe ,σg ∗
A(3...),−τ (pa , . . . , ph ) = A(3...),τ (pa , . . . , ph ) |m∗t →mt , (2.32)
...
where the index τ has to be ignored for all functions other than A(3a),τ and the substitution
m∗t → mt ensures that the top-quark mass (if taken complex) remains unaffected by the
complex conjugation.
Mσ(2a)
a ,...,σh
× gV−σfa¯ F gVσb F f gV−σfc¯ F gVσd F f gVf f¯ f gVσh f¯ f
σ
1 a 1 2 1 b 2 c 2 3 2 d 1 e f 3 g h
× PF1 (paef )PF2 (pdgh )PV1 (pef )PV2 (pabef )PV3 (pgh )
σ ,σ ,σe ,σg
× A(2a)
a c
(pa , . . . , ph ), (2.33)
Mσ(2b)
a ,...,σh
σ
1 a 1 2 1 b 3 c 2 2 2 d 1 e f 3 g h
× PF1 (paef )PF2 (pcgh )PV1 (pef )PV2 (pabef )PV3 (pgh )
σ ,σ ,σe ,σg
× A(2b)
a c
(pa , . . . , ph ), (2.34)
Mσ(2c)
a ,...,σh
σ
2 a 1 1 1 b 2 c 2 3 2 d 1 e f 3 g h
× PF1 (pbef )PF2 (pdgh )PV1 (pef )PV2 (pabef )PV3 (pgh )
σ ,σ ,σe ,σg
× A(2c)
a c
(pa , . . . , ph ), (2.35)
Mσ(2d)
a ,...,σh
(pa , . . . , ph ) = 4e6δσa ,+ δσb ,− δσc ,+ δσd ,− δσe ,+ δσf ,− δσg ,+ δσh ,−
−
× gW g− g+ g− g− g−
f¯a t φ t̄fb φ f¯c t Wt̄fd Wf¯e ff Wf¯g fh
× Pt (paef )Pt (pdgh )PW (pef )PW (pabef )PW (pgh )
σ ,σ ,σe ,σg
× A(2d)
a c
(pa , . . . , ph ). (2.36)
Again we exploited the fact that helicity conservation in each spinor chain containing top
quarks is restored by chirality projectors, i.e., that Eq. (2.22) is necessary for non-vanishing
contributions. The functions A...... read
A++++ ∗
(2a) (pa , . . . , ph ) = pd ph pa pe pb Pdh pg pf Pae pc ,

A−+++ ∗ ∗
(2a) (pa , . . . , ph ) = pa pf pd ph pb pc pa pe pa Pdh pg

− pe pf pf Pdh pg , (2.37)

A++++ ∗ ∗
(2b) (pa , . . . , ph ) = − pb pd pa pe pc pg pa pf ph Pcg pa

+ pe pf ∗ ph Pcg pe ,
A−+++ ∗
(2b) (pa , . . . , ph ) = −pa pf pc pg pd Paf pe ph Pcg pb , (2.38)

A++++ ∗ ∗
(2c) (pa , . . . , ph ) = − pb pf pd ph pa pc pb pe pb Pdh pg

− pe pf pf Pdh pg ,
A−+++ ∗
(2c) (pa , . . . , ph ) = −pd ph pb pe pa Pdh pg pf Pbe pc , (2.39)
A++++ ∗ ∗
(2d) (pa , . . . , ph ) = −mt pb pf pd ph pa pe pc pg ,
2
(2.40)
with the following substitutions for the remaining polarizations,
σ ,σ ,−σe ,σg
a c
σa ,σc ,σe ,σg
= A(2...) (pa , pb , pc , pd , pf , pe , pg , ph ),
σa ,σc ,σe ,−σg
σa ,σc ,σe ,σg
= A(2...) (pa , pb , pc , pd , pe , pf , ph , pg ),
−σ ,−σc ,−σe ,−σg σa ,σc ,σe ,σg ∗
a
A(2...) (pa , . . . , ph ) = A(2...) (pa , . . . , ph )
m∗ →mt . (2.41)
t
2.2. Squared matrix elements from generic amplitudes
Having defined all amplitudes involving 8 external massless fermions in the previous
section, we now turn to the evaluation of the squared matrix elements for a given process
2f → 6f . Since the adopted approach is based on helicity eigenstates of the external
fermions, in a first step we calculate the squares of all helicity matrix elements and take
the incoherent sum over the relevant helicity configurations at the end. We widely suppress
the helicity labels in the following.
The fermion lines with an outgoing arrow are enumerated by i = 1, 3, 5, 7; they
correspond to incoming antifermions f¯i or outgoing fermions fi . The fermion lines with an
incoming arrow are enumerated by i = 2, 4, 6, 8; they correspond to incoming fermions fi
or outgoing antifermions f¯i . The respective incoming momenta and helicities are denoted
by pi and σi , i.e., these quantities receive a minus sign by crossing if they belong to final-
state particles. For instance, for a scattering reaction of the incoming f¯7 f8 pair, our notation
is
f¯7 (p7 , σ7 ) + f8 (p8 , σ8 ) → f1 (−p1 , −σ1 ) + f¯2 (−p2 , −σ2 ) + f3 (−p3 , −σ3 )
+ f¯4 (−p4 , −σ4 ) + f5 (−p5 , −σ5 )
+ f¯6 (−p6 , −σ6 ), (2.42)
where the momenta and helicities within parentheses correspond to incoming fields,
i.e., they are identified with appropriate permutations of the momenta pa , . . . , ph and
σa ,...,σh
helicities σa , . . . , σh in the generic amplitudes M(...) (pa , . . . , ph ) of the previous
section. Finally, all relevant Feynman graphs have to be summed and squared taking into
account colour correlations.
In detail the construction proceeds in three steps:
(i) Fermion permutations
First we write down all permutations of (1, 3, 5, 7) and (2, 4, 6, 8) for (a, c, e, g) and
(b, d, f, h), respectively, resulting in (4!)2 = 576 different combinations (a, . . . , h). The
generic amplitudes vanish if at least one of the pairs f¯a fb , f¯c fd , f¯e ff , and f¯g fh does not
appear in a vertex V f¯f with a gauge boson V . Therefore, we can omit such combinations
(a, . . . , h) from the beginning. This excludes, for instance, combinations where at least
one of the pairs consists of a lepton and a quark. However, there are also channels, such
as e+ e− → e+ e− e+ e− e+ e− , where all 576 combinations contribute. In the following we
enumerate the relevant combinations by n = 1, . . . , N (N 576) and denote them by
(an , . . . , hn ). Next we determine the relative sign factors ηn between diagrams that differ
in their fermion-number flow by an interchange of external lines. Obviously

+1, for even permutations (an , . . . , hn ) of (1, . . . , 8),
ηn = an ,...,hn = −1, for odd permutations (an , . . . , hn ) of (1, . . . , 8), (2.43)
0, otherwise,
yields the correct signs.
(ii) Amplitude evaluation
In the next step we evaluate the generic amplitudes Mσ(...) a ,...,σh
(pa , . . . , ph ) with the
identifications (a, . . . , h) = (an , . . . , hn ), i.e., we calculate all graphs shown in Figs. 1–3
with all possible internal bosons. More precisely, in order to avoid double counting we
require appropriate conditions for (a, . . . , h) and the internal bosons to ensure that each
Feynman diagram is calculated only once. These conditions are listed in Table 1, where
Qab = Qb − Qa is the total incoming charge of the pair f¯a fb . An entry “none” means that
all insertions allowed by charge conservation are possible. For example, the second line
in Table 1 for diagram “4b” means that the vector bosons V1 and V2 each can be either
a photon or Z boson, V3 is a W boson with a negative charge −e flowing through the
diagram from the f¯a fb pair to the f¯e ff pair, while no restriction on the numbers a, . . . , h
is imposed.
Table 1
Restrictions on the field insertions and on the charge flow in the generic amplitudes. (The lists for the field
insertions are ordered.)
Amplitude Bosons Fermions Charge flow
4a V{1,2} = {W, W} a < c, e < g Qab = Qcd = −1
V{1,2} = {γ /Z, γ /Z} e<g Qab = −1
4b V{1,2,3} = {W, W, γ /Z} a <g Qab = Qgh = −1
V{1,2,3} = {γ /Z, γ /Z, W} none Qab = −1
4c V{1,2,3,4} = {W, W, W, W}, S = H a <g Qab = Qgh = −1
V{1,2,3,4} = {W, W, Z, Z}, S = H e<g Qab = −1
V{1,2,3,4} = {Z, Z, Z, Z}, S = H a = 1, e < g none
V{1,2,3,4} = {W, γ /Z, W, γ /Z}, S = φ none Qab = −1
3b, 3c V{1,2,3} = {W, W, γ /Z} none none
V{1,2,3} = {W, γ /Z, W} none Qef = −1
V{1,2,3} = {γ /Z/g, W, γ /Z} none none
V{1,2,3} = {γ /Z/g, γ /Z, W} none Qef = −1
2b, 2c none a <c none
3a, 3d, 3e, 2a, 2d none none none
At this stage, we can already sum up diagrams that belong to the same fermion
permutation and have the same colour structure. In this way we obtain the sums M(ew) n
(1g)
and Mn which include all purely electroweak and one-gluon exchange diagrams for
(an , . . . , hn ), respectively. Here one should realize that the number of diagrams can be
rather large, ranging from typically ∼ 102 –103 up to 13896 for e+ e− → e+ e− e+ e− e+ e− .
(iii) Squared matrix element

If no external quarks are involved the squared matrix element |Mlept|2 is simply
obtained by squaring the sum over all contributions ηn M(ew)
n ,

N
2

(ew)
|Mlept| =
2
ηn Mn
. (2.44)

n=1
For two external quarks the situation is as simple as in the purely leptonic case. The quark
line always closes to the same loop in the squared “interference” diagrams, leading to
a global factor 3 from the colour trace,

N
2

(ew)
|M2q | = 3
2
ηn Mn
. (2.45)

colour n=1
For four external quarks non-trivial colour interferences occur. Denoting the four quarks by
q̄1 q2 q̄3 q4 , there are two types of diagrams: one type in which the pairs q̄1 q2 and q̄3 q4 define
the two quark lines, and another one in which q̄1 q4 and q̄3 q2 define the lines. According
to this criterion we divide the set of N permutations (an , . . . , hn ) into two subsets of N/2
permutations for each of the two diagram types and label the permutations by the pair
of indices (n, α) with n = 1, . . . , N/2, α = 1, 2, instead of taking n = 1, . . . , N . Hence,
(1g)
M(ew)n,α and Mn,α denote the sums of purely electroweak and one-gluon diagrams for
(an,α , . . . , hn,α ), respectively, where the index α determines whether q̄1 q2 and q̄3 q4 or
q̄1 q4 and q̄3 q2 correspond to the two quark lines. The squared matrix element is obtained
from terms of the form Mn,α (Mm,β )∗ . If α = β, diagrammatically such a term represents
(X) (Y )
interference graphs in which the two quark lines close separately; for purely electroweak
diagrams (X = Y = ew) each closed quark yields a colour factor 3, leading to a colour
factor 32 = 9 for the whole interference term. If α = β, the two quark lines close to a single
loop, leading to a colour factor 3 for purely electroweak diagrams. If at least one of the
terms Mn,α and (Mm,β )∗ stands for one-gluon diagrams, the colour factors result from
(X) (Y )
some simple traces over Gell–Mann matrices. We summarize the colour factors in terms of
(X,Y )
2 × 2 matrices C4q :

(ew,ew) 3 9 (1g,1g) 2 − 23
C4q = , C4q = ,
9 3 − 23 2

(ew,1g) 0 4
C4q = . (2.46)
4 0
Using these matrices, the squared matrix element reads

N/2 N/2 ∗
2 (ew,ew)
(ew)
|M4q | =
2
ηn,α Mn,α C4q,αβ
(ew)
ηm,β Mm,β
colour α,β=1 n=1 m=1
N/2 N/2 ∗

2 (1g) (1g,1g)
(1g)
+ ηn,α Mn,α C4q,αβ ηm,β Mm,β
α,β=1 n=1 m=1
N/2 N/2 ∗

2 (ew,1g)
(1g)
+ 2 Re ηn,α M(ew)
n,α C4q,αβ ηm,β Mm,β .
α,β=1 n=1 m=1
(2.47)
2.3. Introduction of finite decay widths
For the finite decay widths introduced in the propagator functions (2.4) we provide three
possibilities as options:

Γ, fixed width,
2  Γ × θ p2 , step width,
Γ p = (2.48)
 p2 2
Γ × M 2 θ p , running width,
where M and Γ denote the mass and on-shell decay width of the propagating particle,
respectively, and θ (x) is the usual step function. In the fixed-width scheme, the constant
width Γ is introduced in each propagator, s-channel- or t-channel-like, i.e., for time-
like and space-like momentum transfer, respectively. In the step-width scheme and in
the running-width scheme only s-channel propagators receive a finite width, as actually
demanded from field-theoretical principles, and the factor p2 /M 2 in the running width
reproduces the correct momentum dependence of the imaginary part of a one-loop self-
energy for a particle that decays into massless decay products.
Note that none of these schemes preserves gauge invariance, as, for instance, explained
in Refs. [2,27,30–33]. Nevertheless for the processes e+ e− → 4f (+γ ) the gauge-
invariance-violating effects in the fixed-width scheme turned out to be sufficiently
suppressed [21,31], rendering this simple scheme very useful. Moreover, it was pointed
out in Ref. [21] that gauge invariance is restored in the fixed-width approach if complex
gauge-boson masses are used in all Feynman rules,3 i.e., in particular the weak mixing
angle is derived from the complex gauge-boson masses,
2 − iM Γ
MW W W
2
cw = 1 − sw2 = . (2.49)
MZ2 − iMZ ΓZ
This complex-mass scheme is included as a fourth option in L USIFER.
The so-called fermion-loop scheme [31], which introduces gauge-boson widths by
a gauge-invariant fermion-loop resummation, is not sufficient for six-fermion production in
3 It is not clear whether this scheme, where masses and couplings are complex, is consistent also in higher
perturbative orders.
general and thus not considered in the following, since it does not provide an introduction
of the decay widths of the top quark and the Higgs boson, because these particles do not
(or not entirely) decay into fermion–antifermion pairs. On the other hand, the effective
Lagrangian approach of Ref. [33] or appropriate expansions [30] about resonance poles
could also be used for e+ e− → 6f processes, but this task goes beyond the scope of this
paper.
2.4. Cross-checks
Apart from performing internal checks, we have compared the squared matrix elements
for some phase-space points with the results obtained with M ADGRAPH [28]. Since
M ADGRAPH calculates amplitudes within the unitary gauge without providing a gauge-
invariant introduction of decay widths, we performed the comparison for vanishing decay
widths consistently. We find very good numerical agreement for non-exceptional phase-
space points between our results and the ones of M ADGRAPH, whenever the latter program
delivers an amplitude. For a recent test version of M ADGRAPH, directly obtained from
the authors [34], there are only two e+ e− → 6f channels that are not covered, e+ e− →
e+ e− e+ e− e+ e− and e+ e− → e+ e− e+ e− νe ν̄e . We note that the amplitude check was done
with and without (coherent) inclusion of gluon background diagrams.
3. Multi-channel phase-space integration
Monte Carlo generators are widely used for experimental analyses since they provide
realistic event samples that can be compared with experimental observations after detector
simulation. Another advantage of a Monte Carlo generator is its flexibility; all possible
observables can be studied easily by choosing suitable separation cuts. However, for
a proper use of the Monte Carlo program a basic knowledge of the numerical integration
techniques is helpful. Therefore, the numerical integration of L USIFER is briefly outlined
in the following. The general method is very similar to the one used in R ACOON WW, as
described in Refs. [21,35], but generalized to six-fermion processes.
For six-fermion production, e+ e− → 6f , an integration over a 14-dimensional phase
space has to be performed. Denoting the incoming e± momenta p± and the outgoing
fermion momenta ki (i = 1, . . . , 6), the phase-space integral reads

|M(p+ , p− , k1 , . . . , k6 )|2
dσBorn = dΦ 2→6 2
,
8(2π)14 ECM

6
d3 ki (4) 6
dΦ 2→6
= δ p+ + p− − k j
, (3.1)
2ki0
0 2 2
i=1 j =1 ki = ki +mi
where the fermion masses mi are zero in our case (mi = 0) and ECM is the total CM energy.
The numerical integration (3.1) is rather complicated owing to the rich peaking structure of
the integrand. The amplitude involves a huge number of propagators that become resonant
or are enhanced in various phase-space regions. The problem is even more serious if the
separation cuts, which are required to exclude infrared and collinear singularities from the
physical integration domain, are small.
To obtain reliable numerical results, events have to be sampled more frequently in the
integration domains where the integrand is large. This redistribution of events is called
importance sampling. In practice, this means that the mapping of the pseudo-random
numbers ri , 0 ri 1, into the space of final-state momenta has to be chosen such that
the corresponding Jacobian 1/g(r) cancels the propagators of the differential cross section
at least partially:
1
dΦ 2→6 = d14 r . (3.2)
g(r)
3.1. Propagator mappings
In order to smooth a single propagator with momentum transfer q in the square of

a matrix element,
1
|M|2 ∝ , (3.3)
(q 2 − M 2 )2 + M 2 Γ 2
the phase space is parametrized in such a way that the virtuality q 2 of the propagator is
chosen as an integration variable. The integral over q 2 is transformed into an integral over
the random number r:
2
qmax 1
1
dq =
2
dr . (3.4)
gprop (q 2 (r))
2
qmin 0
The dependence of q 2 on the pseudo-random number r has to be chosen such that the
Jacobian 1/gprop smoothes the propagator (3.3). In L USIFER two types of mappings are
used according to the mass M of the propagating particle. Since all occurring massive
particles possess a finite decay width Γ , a suitable mapping is of Breit–Wigner type
(M = 0, Γ = 0):
q 2 (r) = MΓ tan[y1 + (y2 − y1 )r] + M 2 ,

MΓ
gprop q 2 = (3.5)
(y2 − y1 )[(q 2 − M 2 )2 + M 2 Γ 2 ]
with
2
qmin,max − M 2
y1,2 = arctan . (3.6)
MΓ
For massless particles the following mappings are appropriate (M = Γ = 0):
2 1−ν 1
q 2 (r) = r qmax + (1 − r)(qmin
2
)1−ν 1−ν ,
1−ν
gprop q 2 = , (3.7)
(q ) [(qmax )1−ν − (qmin
2 ν 2 2 )1−ν ]
2 r 2 1−r
q 2 (r) = qmax qmin ,
1
gprop q 2 = . (3.8)
2 /q 2 )
q 2 ln(qmax min
The mapping (3.7) is only valid for ν = 1, while Eq. (3.8) is a substitute of Eq. (3.7) for
ν = 1. The naive expectation ν = 2 from the squared matrix element is not necessarily the
best choice because the propagator poles are partially cancelled in the collinear limit. It
turns out that a proper value is ν 1.
3.2. Multi-channel approach
Obviously, importance sampling of all propagators appearing in the amplitude is

not possible by a single phase-space mapping. Therefore, we apply the multi-channel
approach [20,24] where N phase-space parametrizations with appropriate mappings,
called channels, are used simultaneously. To this end, the phase-space integral is rewritten
into the form
1
N 1
1
dΦ 2→6
= dr θ (r − βi−1 )θ (βi − r) d14 r , (3.9)
gtot (r)
0 i=1 0
i
where the βi define a partition of unity, i.e., βi = j =1 αj , β0 = 0, βN = N j =1 αj = 1.
In Eq. (3.9) an additional random number r is introduced in order to select a channel i,
randomly, with probability αi 0. A channel is defined by its mapping from the pseudo-
random numbers into the space of final-state momenta. The total density gtot is composed
of the single densities gi of the different channels weighted by the a priori weights αi :

N
gtot = αi gi . (3.10)
i=1
In this way, it is possible to simultaneously include different phase-space mappings which

are suitable for different parts of the integrand. To be specific, for each diagram a channel
exists, so that all propagators of the squared diagram are smoothed by the corresponding
local density gi . No special channels are provided for interference contributions.
The a priori weights αi are adapted in the early phase of the Monte Carlo run several
times to optimize the convergence behaviour of the numerical integration. This adaptive
weight optimization [25] increases the a priori weights αi for such channels that correspond
to important diagrams of the process, i.e., to diagrams that give large contributions to the
total cross section.
The actual calculation is performed in two steps: first, a channel is chosen and the
momenta of the final-state particles are calculated with the corresponding phase-space
generator. Secondly, the total density is determined. Besides the evaluation of the matrix
elements, the second step is the most time-consuming part since the local densities of all
channels have to be calculated.
Fig. 4. Topological diagram for the generic phase-space generator.
3.3. Generic construction of phase-space generators
Since six-fermion production processes involve of the order of 102 –104 diagrams, it
is very important to handle the different phase-space mappings in a generic way. All
different channels can be obtained from one generic phase-space generator by choosing
the topological structure of the corresponding diagram, as illustrated in Fig. 4, the order of
the incoming and outgoing particles, and the masses and widths of the internal particles.
In contrast to the calculation of the matrix elements as discussed in Section 2, there
is a fundamental difference between the incoming and outgoing particles, and s- and
t-channel propagators.
For each class of diagrams with the same propagator structure we adopt an own channel
for the numerical integration. Diagrams with four-particle vertices also fit into the generic
diagram of Fig. 4 if one of the propagators is contracted and the corresponding invariant is
sampled uniformly. The same is true for diagrams with exclusively s-channel propagators,
since the s-channel propagator resulting from a fusion of the two incoming lines is
constant.
The actual calculation of the event kinematics is decomposed into three steps:
the calculation of time-like invariants, of 2 → 2 scattering processes with t-channel
propagators, and of 1 → 2 particle decays. The phase-space integration reads
4

si,max
τ

5−τ
dΦ 2→6
= dsi 2→2
dΦj dΦk1→2 , (3.11)
i=1si,min j =1 k=1
where the phase space of the 2 → 2 scattering processes and the 1 → 2 decays are denoted
by Φj2→2 and Φk1→2 , respectively. Note that the number τ of 2 → 2 scattering processes,
which is the number of t-channel lines in Fig. 4, can range from 0 to 5. The four invariants
si (i = 1, . . . , 4) result from the phase-space factorization into scattering processes and
decays. More details on phase-space parametrizations can be found in Refs. [35,36]. If
several t-channel propagators are present, i.e., for τ 2, some of the si (i = 1, . . . , 4) do
not correspond to virtualities of propagators in the diagram; for such variables no mappings
are introduced, they are generated uniformly.
In detail we proceed as follows:
(i) Calculation of time-like invariants

(j )
First of all, the time-like invariants si are determined. The invariants corresponding to
s-channel propagators are calculated from Eqs. (3.5)–(3.8). All other time-like invariants
are sampled uniformly, i.e., calculated from Eq. (3.7) with ν = 0. The determination of
(j ) (j )
the invariants si is ordered in such a way that the si nearest to final-state particles are
calculated first, followed by the next-to-nearest s-channel propagators, and so on.
(j )
The lower and upper limits si,min / max , in general, are functions of the already
determined invariants. To enhance the efficiency of the numerical integration, separation
cuts have to be taken into account. We include invariant-mass cuts of final-state particles
(j )
in the evaluation of si,min / max whenever possible.
(ii) 2 → 2 scattering processes with t-channel propagators
The subpart of the diagram involving t-channel propagators is decomposed into several
2 → 2 scattering processes. The phase-space integration reads
2π tmax
1 ∗
dΦ 2→2
(p1 , p2 ; q1, q2 ) = dφ dt, (3.12)
8 (p1 p2 )2 − p12 p22 0 tmin
where p1,2 are the incoming and q1,2 are the outgoing momenta, and φ ∗ is the azimuthal
angle defined by p1 and q1 in the CM frame of the subprocess, where p1 + p2 = 0. The
argument t = (p1 − q1)2 of the t-channel propagator is calculated according to Eqs. (3.5)–
(3.8). The azimuthal angle φ ∗ is sampled uniformly. Since the corresponding polar angle
cos θ ∗ of this scattering process depends on the invariant t linearly,

p12 q22 + p22 q12 − 2(p1 p2 )(q1 q2 ) + 2 (p1 p2 )2 − p12 p22 (q1 q2 )2 − q12 q22 cos θ ∗
t= ,
(p1 + p2 )2
(3.13)
the calculation of the momenta and the Lorentz transformation into the laboratory (LAB)
frame is simple and, thus, skipped here. Explicit formulas can, for instance, be found in
Ref. [35]. Subprocesses that are nearest to the incoming particles of the whole e+ e− → 6f
scattering process are calculated first, followed by the next-to-nearest, and so on.
For the 2 → 2 scattering processes that are attached to the original incoming particles,
we already take into account possible cuts on angles between outgoing particles and the
beams that effectively reduce the integration range of t, i.e., we include these cuts in the
determination of tmin / max .
(iii) 1 → 2 particle decays
It remains to perform the decays of the s-channel particles. The phase-space integration
reads

(q1 q2 )2 − q12 q22
2π 1
∗
dΦ 1→2
(q1 , q2 ) = dφ d cos θ ∗ , (3.14)
4(q1 + q2 )2
0 −1
where φ∗ and θ∗ are the azimuthal and polar angles in the rest frame of the decaying
particle, respectively. The variables φ ∗ and cos θ ∗ are sampled uniformly. The momenta of
Fig. 5. Topological diagram for a specific phase-space generator.
the outgoing particles are q1 and q2 . As in the former step, the calculation of the momenta
and the Lorentz transformation into the LAB frame is straightforward (see, e.g., Ref. [35]).
We illustrate the general strategy by considering the specific topology shown in Fig. 5,
where a particular choice for the incoming and outgoing momenta p± and ki is made. For
this example the phase-space parametrization (3.11) reads
4

si,max
dΦ 2→6
|Fig.5 = dsi dΦ 2→2 (p+ , p− ; k4, k12356)
i=1 si,min

× dΦ 2→2 (p− , p+ − k4 ; k1 , k2356)

× dΦ 2→2 (p+ − k4 , p− − k1 ; k256, k3 )

× dΦ 1→2
(k2 , k56 ) dΦ 1→2 (k5 , k6 ), (3.15)
where sums of outgoing momenta are abbreviated by kij = ki + kj , kij k = ki + kj + kk ,

etc. Note that only the time-like invariants s1 = k56
2 and s = k 2 correspond to virtualities
2 256
of propagators in the diagram, while s3 = k2356 and s4 = k12356
2 2 correspond to invariant
masses of fictitious final-state particles within the first two 2 → 2 scattering processes.
For an efficient and fast numerical integration L USIFER evaluates the different subparts
of the phase-space mappings for all channels only once for a single event. The local
densities of individual channels are then easily obtained from the Jacobians of these
subparts. This speeds up the numerical integration considerably.
3.4. Cross checks
The Monte Carlo part of L USIFER can, in principle, be applied to arbitrary processes. In
order to check the phase-space integration, we implemented the tree-level matrix elements
of R ACOON WW, as given in Ref. [21], in L USIFER and reproduced the results given
in Table 1 of Ref. [21] for four-fermion production, e+ e− → 4f , and for the radiative
processes, e+ e− → 4f γ , in the fixed width scheme. We found good agreement between
the 4f and 4f γ results integrated with L USIFER and R ACOON WW.
4. Higher-order initial-state radiation at the leading logarithmic level
A first step to improve tree-level predictions by higher-order radiative corrections

consists in including universal corrections such as the leading logarithms of initial-state
radiation. In L USIFER this is done via structure functions [26,27]. The lowest-order
differential cross section is convoluted in the energy fractions x± of the incoming electron
and positron:
1 1

dσBorn+ISR = dx+ ΓeeLL x+ , Q2 dx− ΓeeLL x− , Q2
0 0

× dσBorn (x+ p+ , x− p− ), (4.1)
where the splitting scale Q is not fixed at the leading logarithmic level and quantifies part of
the missing radiative corrections. In L USIFER, Q is set to the CM energy ECM by default.4
The Born cross section in Eq. (4.1) is calculated in the same way as in the case without
ISR but with the momenta x± p± for the incoming e± . The precise form of the structure
function ΓeeLL (x, Q2 ) used in L USIFER is the same as in Eqs. (5.2)–(5.4) of the second
paper in Ref. [22].
5. Classification of 6f final states in e+ e− collisions
The derivation of squared matrix elements described in Section 2 is valid for all
processes involving eight external fermions and restricted for gluonic diagrams to
processes involving up to four external quarks. Thus, these results can be applied to various
six-fermion processes in lepton–lepton, lepton–hadron, and hadron–hadron scattering. In
this paper, we focus on e+ e− → 6f processes in the following, representing an important
class of reactions at future linear colliders such as TESLA [1]. We divide the set of 6f final
states into three (overlapping) subsets corresponding to different subprocesses of interest:
(i) Processes involving top quarks
The most interesting processes involving top quarks correspond to top-pair production
with the subsequent decay of the top quarks into six fermions which proceeds via
W bosons, e+ e− → tt̄ → bW+ b̄W − → 6f . The corresponding diagram is shown in Fig. 6.
All 6f final states that are relevant for tt̄ production are of the form e+ e− → bb̄f1 f¯1 f2 f¯2 ,
where fi f¯i (i = 1, 2) are two weak isospin doublets.
There is a second, but less interesting class of 6f final states that involve top-quark
diagrams but do not contain tt̄ production as a subprocess. These processes are of the form
e+ e− → νe ν̄e bb̄f f¯, where f is any fermion other than the electron.
4 For scattering processes that are dominated by small splitting scales the use of purely s-dependent structure
functions is not a good approximation for treating ISR effects. This is, in particular, the case if e± are scattered
in the very forward direction. Possible improvements for such situations have been, for instance, described in the
context of single-W production in e+ e− → 4f (see, e.g., Refs. [2,37]).
Fig. 6. Diagram for tt̄ production: e+ e− → tt̄ → bW+ b̄W − → 6f .
Fig. 7. Diagram structures for vector-boson scattering: e+ e− → 2f + (V1 V2 → V3 V4 ) → 6f .
(ii) Vector-boson scattering and s-channel Higgs production

One of the most interesting class of subprocesses in e+ e− → 6f is the scattering of
weak gauge bosons, e+ e− → 2f + (V1 V2 → V3 V4 ) → 6f . According to the charges of
the “incoming” vector-boson pair V1 V2 , there are three different types of vector-boson
scattering channels, as shown in Fig. 7: (a) neutral–neutral, (b) charged–charged, and (c)
mixed. Among all 6f final states these channels are distinguished by the appearance of (a)
e+ e− , (b) νe ν̄e , and (c) e+ νe /e− ν̄e in the final state.
ZZ and WW fusion processes, which are included in types (a) and (b), are also
important in Higgs physics. If the Higgs-boson mass MH is large enough that the decay
channel H → WW opens, resonant s-channel Higgs production dominates the vector-
boson scattering cross section, WW, ZZ → H → WW/ZZ, which is one of the two main
production mechanisms for the Higgs boson in e+ e− annihilation. The relevant Feynman
diagram is shown in Fig. 7(d).
Fig. 8. Diagram structures for three-boson production: e+ e− → V1 V2 V3 → 6f .
(iii) Three-boson production and Higgs-strahlung off Z bosons

All 6f final states have in common that they contribute to production processes of
three electroweak gauge bosons, e+ e− → V1 V2 V3 → 6f . As illustrated in Fig. 8, there are
two different types of reactions, distinguished by the charges of the produced bosons: (a)
charged–charged–neutral and (b) neutral–neutral–neutral.
For a sufficiently large Higgs-boson mass, i.e., when H → WW becomes possible, not
only s-channel Higgs production leads to relatively large contributions to 6f production
but also Higgs-strahlung off Z bosons, e+ e− → ZH → ZWW/ZZZ → 6f , which is the
second of the two main Higgs production mechanisms at future e+ e− colliders. The
corresponding Feynman diagram is shown in Fig. 8(c).
6. Numerical results
6.1. Input parameters
For the numerical evaluation we use the following set of Standard Model parameters
[38],
Gµ = 1.16639 × 10−5 GeV−2 , α(0) = 1/137.0359895,

αs (MZ ) = 0.1181,
MW = 80.419 GeV, ΓW = 2.12 GeV,
MZ = 91.1882 GeV, ΓZ = 2.4952 GeV,
mt = 174.3 GeV, Γt = 1.6 GeV,
me = 0.51099907 MeV, (6.1)
where the top-quark width Γt is, of course, only a reasonable estimate. In order to absorb
parts of the renormalization effects (i.e., some universal radiative corrections) into the
electroweak couplings, such as the running of α(Q2 ) from Q2 = 0 to a high-energy scale
and some universal effects related to the ρ parameter, we evaluate amplitudes in the so-
called Gµ scheme, i.e., we derive the electromagnetic coupling α = e2 /(4π) from the
Fermi constant Gµ according to
√ 2 s2
2 Gµ MW w
αGµ = . (6.2)
π
Table 2
Higgs-boson masses and decay widths as provided by HDECAY
MH [GeV] 170 190 230
ΓH [GeV] 0.3835 1.038 2.813
In the structure functions we use α(0) as coupling parameter, which is the correct effective
coupling for real photon emission. For the Higgs-boson mass we take the set of sample
values listed in Table 2, including also the corresponding Higgs-boson widths, which are
calculated with the program HDECAY [39]. If not stated otherwise, we take the value
MH = 170 GeV by default.
With the only exception of Section 6.5, where the interplay between finite-width
schemes and gauge invariance is discussed, we make use of the fixed-width approach,
as described in Section 2.3.
Finally, we specify the following separation cuts:
θ (l, beam) > 5◦ , θ (q, beam) > 5◦ , θ (l, l ) > 5◦ , θ (l, q) > 5◦ ,
El > 10 GeV, Eq > 10 GeV, m(q, q ) > 10 GeV, (6.3)
where θ (i, j ) is the angle between the particles i and j in the LAB system, and l, q, and
“beam” denote charged final-state leptons, quarks, and the beam electrons or positrons,
respectively. The invariant mass of a quark pair qq is denoted by m(q, q ).
6.2. Survey of cross sections
Only very few results presented in the literature are found to be appropriate for a tuned
comparison to our results, because in most cases only plots but no numbers are given,
fermion mass singularities are not always excluded by phase-space cuts, or beamstrahlung
effects are included. Therefore, we decided to adopt a public version of a multi-purpose
generator available from the www and to perform a tuned comparison based on the
setup described in the previous section as far as possible. Specifically, we have used the
generator W HIZARD [14], version 1.21, together with the implemented M ADGRAPH [28]
amplitudes. We did not include beamstrahlung effects in the following results, since they
depend on the details of the inspected collider and would spoil the usefulness of our results
as reference for future studies. Beamstrahlung can, however, be included in L USIFER in
a straightforward way. If not stated otherwise the cross section numbers in this section are
based on 107 weighted Monte Carlo events. In the following tables, the numbers within
parentheses correspond to the statistical errors of the Monte Carlo integrations.
In Table 3 we show the results for all processes e+ e− → 6 leptons with up to two
neutrinos in the final state (three neutrinos are not possible) for a CM energy of 500 GeV.
The first set of final states comprises the reactions that receive contributions from resonant
WWZ production and various other mechanisms. The remaining reactions entirely proceed
via neutral-current interactions, except for the last two processes that receive contributions
from WW → ZZ scattering. Comparing the two sets of cross sections, the ones for
WWZ production are larger. ISR affects the cross sections at the level of a few per cent,
ranging up to ∼ 8%, and this correction is negative in almost all cases. The comparison
Table 3
√
Cross sections for e+ e− → 6f channels with purely leptonic final states for s = 500 GeV
L USIFER W HIZARD & M ADGRAPH
e+ e− → σBorn [fb] σBorn+ISR [fb] σBorn [fb] σBorn+ISR [fb]
µ− µ+ µ− µ+ νµ ν̄µ 0.041382(87) 0.040883(81) 0.04130(19) 0.04077(27)
µ− µ+ µ− ν̄µ ντ τ + 0.040320(80) 0.04002(12) 0.04044(19) 0.03961(15)
µ− µ+ νµ ν̄µ τ − τ + 0.042297(84) 0.041966(89) 0.04212(12) 0.04165(33)
e− ν̄e µ− µ+ νµ µ+ 0.04729(10) 0.04640(12) 0.04713(25) 0.04549(20)
e− ν̄e µ− µ+ ντ τ + 0.04708(14) 0.04629(14) 0.04705(16) 0.04702(62)
e− e+ νe ν̄e µ− µ+ 0.06309(20) 0.06026(18) 0.06139(26) 0.05980(65)
e− e+ µ− µ+ νµ ν̄µ 0.18300(36) 0.16838(32) 0.18302(32) 0.16976(36)
e− e+ µ− ν̄µ ντ τ + 0.18023(21) 0.16555(30) 0.18101(31) 0.16661(43)
e− e+ e− ν̄e νµ µ+ 0.18603(25) 0.17091(41) 0.18624(39) 0.17118(48)
e− e+ e− e+ νe ν̄e 0.19702(36) 0.18065(35) – –
µ− µ+ µ− µ+ µ− µ+ 0.00026849(54) 0.00027067(57) – –
µ− µ+ µ− µ+ τ − τ + 0.0008088(15) 0.0008182(16) 0.0007925(69) 0.0007689(79)
e− e+ µ− µ+ µ− µ+ 0.002514(16) 0.002437(16) 0.00211(18) 0.001467(78)
e− e+ µ− µ+ τ − τ + 0.005102(31) 0.005005(37) 0.00445(16) 0.00464(23)
e− e+ e− e+ µ− µ+ 0.004158(31) 0.004017(39) – –
e− e+ e− e+ e− e+ 0.001803(16) 0.001758(15) – –
µ− µ+ µ− µ+ ντ ν̄τ 0.0010312(16) 0.0010262(16) 0.0010295(36) 0.0010257(43)
e− e+ µ− µ+ ντ ν̄τ 0.0033141(69) 0.0031918(72) 0.003296(14) 0.003133(23)
e− e+ e− e+ νµ ν̄µ 0.0022162(60) 0.0020908(62) – –
νe ν̄e µ− µ+ µ− µ+ 0.003730(14) 0.0034166(76) 0.003613(25) 0.003374(34)
νe ν̄e µ− µ+ τ − τ + 0.007366(17) 0.006785(16) 0.007301(33) 0.006503(67)
with W HIZARD and M ADGRAPH reveals agreement within 1–3σ in general with a few
exceptions, where larger differences occur. These differences are accompanied with larger
errors in the W HIZARD results and are assumed [40] to be due to the fact that the photon
propagators in the splittings γ → f f¯ are not analytically smoothed by mappings in
W HIZARD. In most cases the statistical error given by L USIFER is smaller than the one
of W HIZARD. The missing entries correspond to those final states that are not supported
by M ADGRAPH.5
Table 4 shows a tuned comparison for 6f final states containing two quarks and at
most three neutrinos for a CM energy of 500 GeV. The table is divided into two parts: the
first part comprises all final states that receive contributions from a hadronically decaying
W boson; final states corresponding to hadronically decaying Z bosons form the second
part, where contributions from uū and dd̄ pairs are added, in order to keep the table more
compact. In comparison with the purely leptonic final states the cross sections are larger,
which is mainly due to the colour factor 3 resulting from the two quarks in the final states.
Otherwise the various channels show similar features as their leptonic counterparts. WWZ
production channels have larger cross sections than processes that proceed via neutral-
5 In contrast to the M ADGRAPH version used for checking the amplitudes, as described in Section 2.4,
the M ADGRAPH version within W HIZARD is neither able to add electroweak and gluon-exchange diagrams
coherently, nor to deal with more than 999 Feynman graphs for a given process.
Table 4
√
Cross sections for e+ e− → 6f channels with four leptons and two quarks in the final state for s = 500 GeV
e+ e− → σBorn [fb] σBorn+ISR [fb] σBorn [fb] σBorn+ISR [fb]
µ− µ+ µ− ν̄µ ud̄ 0.11835(22) 0.11714(22) 0.11797(26) 0.11732(71)
µ− µ+ τ − ν̄τ ud̄ 0.11861(42) 0.11695(22) 0.11791(21) 0.11664(29)
e− ν̄e µ− µ+ ud̄ 0.13832(30) 0.13516(34) 0.13785(36) 0.13510(62)
e− e+ µ− ν̄µ ud̄ 0.53352(70) 0.48897(69) 0.53493(61) 0.49378(73)
e− e+ e− ν̄e ud̄ 0.55089(74) 0.50464(72) 0.5514(13) 0.5061(10)
µ− ν̄µ νµ ν̄µ ud̄ 0.18399(11) 0.18182(11) 0.18396(13) 0.18204(14)
µ− ν̄µ ντ ν̄τ ud̄ 0.18406(10) 0.18201(11) 0.18430(13) 0.18197(14)
e− ν̄e νµ ν̄µ ud̄ 0.20272(14) 0.19847(14) 0.20288(14) 0.19843(16)
νe ν̄e µ− ν̄µ ud̄ 1.6326(12) 1.4743(12) 1.6313(13) 1.4746(13)
e− ν̄e νe ν̄e ud̄ 1.6500(17) 1.4906(15) 1.6482(15) 1.4914(14)
µ− µ+ νµ ν̄µ (uū+dd̄) 0.26632(36) 0.26266(33) 0.26647(19) 0.26242(21)

µ− ν̄µ ντ τ + (uū+dd̄) 0.25408(18) 0.25068(19) 0.25427(17) 0.25067(17)
e− ν̄e νµ µ+ (uū+dd̄) 0.28161(24) 0.27514(24) 0.28189(21) 0.27471(21)
e− e+ νe ν̄e (uū+dd̄) 0.35788(62) 0.34361(73) 0.35917(62) 0.34366(36)
µ− µ+ µ− µ+ (uū+dd̄) 0.0043727(73) 0.0043774(77) 0.004368(18) 0.004303(18)
µ− µ+ τ − τ + (uū+dd̄) 0.008731(14) 0.008736(15) 0.008652(26) 0.008585(41)
e− e+ µ− µ+ (uū+dd̄) 0.015466(59) 0.014886(49) 0.01523(11) 0.01396(27)
e− e+ e− e+ (uū + dd̄) 0.010730(58) 0.010207(45) – –
µ− µ+ ντ ν̄τ (uū+dd̄) 0.012057(17) 0.012021(17) 0.012042(12) 0.011964(19)
e− e+ νµ ν̄µ (uū+dd̄) 0.016967(30) 0.016484(32) 0.017006(23) 0.016417(25)
νe ν̄e µ− µ+ (uū+dd̄) 0.044940(85) 0.041135(86) 0.044976(64) 0.04083(11)
current interactions only, and ISR affects cross sections at the level of a few per cent,
mainly in negative direction. It should be mentioned that in the second set of final states the
L USIFER results have again been obtained from 107 events, while the W HIZARD runs have
been performed with 107 events for uū and dd̄ each, so that the final numbers effectively
result from 2 × 107 events. In view of this difference the statistical error of L USIFER is
somewhat smaller. The agreement between the two programs is again within 1–3σ with
a few exceptions for the same reason as explained above.
A tuned comparison of results for final states containing two leptons and four quarks
is presented in Table 5, again for a CM energy of 500 GeV. The notation “4q” stands
for the sum over all possible four-quark configurations of u, d, c, s quarks. In the table
we separately give results that include or exclude the contributions of gluon-exchange
diagrams in the amplitude calculation, as indicated in the “QCD” column. Note that the
large impact of these gluon-exchange diagrams sensitively depends on the separation cuts,
in particular on the minimal invariant mass of quark pairs. The impact of ISR is similar
to the cases discussed previously. The largest cross section is observed for νe ν̄e + 4q
production, which is the only channel that receives contributions from WW → WW/ZZ
scattering and from W fusion to a Higgs boson. The comparison to W HIZARD shows the
same features as in the other cases. However, it should be mentioned that all 4q final states
have been integrated individually with W HIZARD, but in a single run with L USIFER. This
explains the smaller integration errors of the W HIZARD results. Since the M ADGRAPH
version included in W HIZARD is not able to coherently add gluon-exchange and purely
Table 5
√
Cross sections for e+ e− → 6f channels with two leptons and four quarks in the final state for s = 500 GeV
e+ e− → QCD? σBorn [fb] σBorn+ISR [fb] σBorn [fb] σBorn+ISR [fb]
µ− ν̄µ + 4q yes 3.8170(84) 3.7917(86) – –
no 2.9603(28) 2.9172(27) 2.9616(13) 2.9195(13)
e− ν̄e + 4q yes 4.605(13) 4.529(14) – –
no 3.2798(38) 3.1930(40) 3.2708(16) 3.1916(17)
µ− µ+ + 4q yes 1.5150(30) 1.5017(31) – –

no 1.4406(40) 1.4227(29) 1.4343(18) 1.4192(19)
νµ ν̄µ + 4q yes 2.3405(29) 2.3147(28) – –
no 2.2466(33) 2.2138(26) 2.2440(12) 2.2153(13)
e− e+ + 4q yes 6.570(17) 6.053(16) – –
no 6.1439(95) 5.679(10) 6.2001(50) 5.7115(60)
νe ν̄e + 4q yes 19.260(40) 17.441(37) – –
no 18.850(32) 17.038(29) 18.879(11) 17.076(11)
Table 6
√
Cross sections for e+ e− → 6f channels involving intermediate top quarks for s = 500 GeV
e+ e− → QCD? σBorn [fb] σBorn+ISR [fb] σBorn [fb] σBorn+ISR [fb]
µ− ν̄µ νµ µ+ bb̄ – 5.8091(49) 5.5887(36) 5.8102(26) 5.5978(30)
µ− ν̄µ ντ τ + bb̄ – 5.7998(36) 5.5840(40) 5.7962(26) 5.5893(29)
e− ν̄e νµ µ+ bb̄ – 5.8188(45) 5.6042(38) 5.8266(27) 5.6071(30)
e− ν̄e νe e+ bb̄ – 5.8530(68) 5.6465(70) 5.8751(30) 5.6508(36)
µ− ν̄µ ud̄bb̄ yes 17.171(24) 16.561(24) – –
no 17.095(11) 16.4538(98) 17.1025(80) 16.4627(87)
e− ν̄e ud̄bb̄ yes 17.276(45) 16.577(21) – –
no 17.187(21) 16.511(12) 17.1480(82) 16.5288(92)
νe ν̄e µ− µ+ bb̄ 0.024550(45) 0.022472(45) 0.024619(43) 0.022398(41)

νe ν̄e uūbb̄ yes 0.12625(32) 0.11703(32) – –
no 0.06984(15) 0.06369(14) 0.069781(70) 0.063635(86)
νe ν̄e dd̄bb̄ yes 0.13709(41) 0.12636(37) – –
no 0.08648(20) 0.07871(18) 0.086351(83) 0.078533(96)
νe ν̄e bb̄bb̄ yes 0.06741(18) 0.06226(18) – –
no 0.04352(10) 0.03974(12) 0.043473(49) 0.039721(68)
electroweak diagrams, a comparison of cross sections based on the full amplitudes has not
been carried out.
Finally, in Table 6 we collect our results on cross sections for final states that involve
amplitudes with intermediate top quarks. The first set of channels comprises all final
states that are relevant for top-quark pair production, e+ e− → tt̄ → 6f . In the second
set of processes top quarks appear only on non-resonant lines in diagrams. The difference
between the tt̄ production cross sections for two and four quarks in the final states roughly
reflects the factor 3 between leptonically and hadronically decaying W bosons that have
been produced in t → bW+ . Since these channels are strongly dominated by the diagram
Fig. 9. Total cross section of e+ e− → µ− ν̄µ udb¯ b̄ (without gluon-exchange diagrams) as function of the CM
energy with and without ISR on the l.h.s. and in comparison with e+ e− → e− ν̄e ud̄bb̄ (without ISR) on the r.h.s.
shown in Fig. 6, the impact of gluon-exchange diagrams is small. Concerning the ISR
effects and the comparison of the two programs, the same features are observed as in the
other cases discussed above.
Finally, we mention that L USIFER runs faster than the combination of the W HIZARD
and M ADGRAPH packages. The factor in speed varies with the 6f final state from roughly
a factor of 2 up to an order of magnitude, where the superiority of L USIFER becomes more
apparent if a large number of diagrams is involved.
6.3. Results on top-quark pair production
Fig. 9 illustrates the (well-known) energy dependence of the top-quark pair production
cross section for final states where one of the produced W bosons decays hadronically and
the other leptonically. The results for the total cross section are obtained with 5 ×106 events
per CM energy. Gluon-exchange diagrams are not taken into account. The cross section
steeply rises at the tt̄ threshold, reaches its maximum between 400 GeV and 500 GeV, and
then decreases with increasing energy. The l.h.s. of the figure shows that ISR reduces the
cross section for energies below its maximum and enhances it above, thereby shifting the
maximum to a higher energy. This behaviour is simply due to the radiative energy loss
induced by ISR. Near a CM energy of 250 GeV the onset of WWZ production can be
observed. Note that this contribution is entirely furnished by background diagrams, i.e., by
diagrams that do not have a resonant top-quark pair. Another type of background diagrams
exists if electrons or positrons are present in the final state, since an incoming e± line can
then go through to the final state. The impact of such diagrams is illustrated on the r.h.s.
of Fig. 9, where the final states are equal up to the change of µ− ν̄µ to e− ν̄e . For energies
around 500 GeV the difference is of the order of a per cent, but increasing with energy. It
should, however, be noted that this difference strongly depends on the separation of final-
state e± from the beams. For smaller cut angles, or for particular regions in distributions,
Fig. 10. Invariant-mass distribution of the ud̄b quark triple in e+ e− → µ− ν̄µ ud̄bb̄ (without gluon-exchange
diagrams): absolute prediction with and without ISR (l.h.s.) and comparison between L USIFER and W HIZARD
(r.h.s.).
¯ quark triple in e+ e− → µ− ν̄µ ud̄bb̄ (without gluon-exchange diagrams):

Fig. 11. Angular distribution of the udb
absolute prediction with and without ISR (l.h.s.) and comparison between L USIFER and W HIZARD (r.h.s.).
the impact of such background diagrams will be much larger.6 The problem of working out
an optimal strategy to define a clear tt̄ signal, i.e., to systematically suppress background
contributions, obviously goes beyond this study.
In Figs. 10 and 11 we consider two examples of distributions that are interesting for
tt̄ production, again focusing on the channel e+ e− → µ− ν̄µ ud̄bb̄. Fig. 10 shows the
¯ quark triple that results from the top-quark decay. As
invariant-mass distribution of the udb
expected, ISR does not distort the resonance shape but merely rescales the Breit–Wigner-
6 These features are well-known from 4f physics, where forward-scattered e± , in particular, are used to define
the so-called single-W production signal (see, e.g., Refs. [2,37]).
Fig. 12. Invariant-mass distribution of the ud̄sc̄ quark system in e+ e− → νµ ν̄µ ud̄sc̄ (without gluon-exchange
diagrams): absolute prediction with and without ISR (l.h.s.) and comparison between L USIFER and W HIZARD
(r.h.s.).
Fig. 13. Angular distribution of the ud̄sc̄ quark system in e+ e− → νµ ν̄µ ud̄sc̄ (without gluon-exchange diagrams):
absolute prediction with and without ISR (l.h.s.) and comparison between L USIFER and W HIZARD (r.h.s.).
like distribution. More interestingly, the r.h.s. of the figure demonstrates that L USIFER
and W HIZARD yield predictions that are fully compatible within statistical accuracy, both
with and without the inclusion of ISR corrections. Fig. 11 shows the production angular
distribution of the ud̄b quark triple, which is (for resonant top quarks) equal to the top-
quark production angle. ISR tends to flatten the distribution, which is again due to the
impact of effectively reduced scattering energies where the distribution is less angular
dependent. The r.h.s. of the figure reveals agreement between the two programs within
statistical errors.
Fig. 14. Invariant-mass and angular distributions of the four-quark system in e+ e− → νµ ν̄µ + 4q (without ISR
√
and gluon-exchange diagrams) for various Higgs masses and s = 500 GeV.
6.4. Results on Higgs-boson production
In this section we discuss some distributions that are relevant for Higgs-boson
production in the intermediate MH range. Figs. 12 and 13 show the four-quark invariant-
mass distribution and the related production angular distribution of the individual channel
e+ e− → νµ ν̄µ ud̄sc̄ for a CM energy of 500 GeV. Gluon-exchange diagrams are not
included in these evaluations. The reaction is dominated by two mechanisms: ZH
production with the subsequent decays H → WW → 4q and Z → νµ ν̄µ , and “continuous”
WWZ production. In the invariant-mass distribution the narrow Higgs resonance shows up
at Mud̄sc̄ = MH = 170 GeV√ over a continuous background from WWZ production in the
range 2MW Mud̄sc̄ s − MZ . The l.h.s. of Fig. 12 shows that ISR does not influence
the resonance structure strongly; the largest ISR effect is observed at the upper edge of
the spectrum, where the effective CM energy loss by ISR reduces the rate. The r.h.s. of
the figure illustrates the agreement between the L USIFER and W HIZARD results within
statistical accuracy. Fig. 13 shows that ISR significantly distorts the four-quark angular
distribution at intermediate angles, where ZH production dominates; in the very forward
and backward regions the spectrum is mainly due to contributions from the subprocess
e+ e− → (γ ∗ /Z∗ )Z → WWZ, where the total momentum of the ud̄sc̄ quark system
correspond to the off-shell particles γ ∗ /Z∗ . The r.h.s. of the figure again demonstrates
the good agreement between the two different Monte Carlo programs.
In Fig. 14 we show the analogous distributions for three different Higgs-boson masses,
but now summed over all four-quark configurations of the first two generations. Fig. 15
shows the same distributions after replacing the νµ ν̄µ pair in the final state by νe ν̄e . ISR
and gluon-exchange diagrams are not included in these evaluations. The crucial difference
between the νµ ν̄µ and νe ν̄e channels lies in the Higgs production mechanisms: while the
former receives only contributions from ZH production, the latter additionally involves
Fig. 15. Invariant-mass and angular distributions of the four-quark system in e+ e− → νe ν̄e + 4q (without ISR
√
and gluon-exchange diagrams) for various Higgs masses and s = 500 GeV.
W fusion, WW → H → WW, which dominates the cross section. Therefore, the cross
section of νe ν̄e + 4q is an order of magnitude larger than the one of νµ ν̄µ + 4q. The
invariant-mass distributions of the two channels look similar, and for MH = 170 GeV
resemble the shape already observed for the single channel e+ e− → νµ ν̄µ ud̄sc̄ in Fig. 12.
Note that for MH = 190 GeV and 230 GeV the high-energy tails of the distributions show
some Higgs mass dependence. This is due to the subprocess of ZH production where the
Higgs decays via H → ZZ → (νµ ν̄µ /νe ν̄e ) + 2q, which is not yet possible for the smaller
Higgs mass MH = 170 GeV. The corresponding boundary in M4q , which is clearly seen in
the plots for MH = 190 GeV, is determined by the two extreme situations where the decay
H → ZZ proceeds along the ZH production axis. For MH = 230 GeV this boundary is
hidden by the Higgs peak and the upper kinematical limit in the M4q spectrum. In contrast
to the invariant-mass distributions, the shape of the four-quark angular distributions of the
νµ ν̄µ and νe ν̄e channels look very different. For νµ ν̄µ , i.e., for ZH production, intermediate
production angles dominate, and this dominance is more pronounced for smaller Higgs-
boson masses, where more phase space exists. For νe ν̄e , i.e., W-boson fusion, forward
and backward production of Higgs bosons is preferred, and the MH dependence is mainly
visible in the overall scale of the distribution, but not in the shape itself.
6.5. Finite-width decay widths and gauge-invariance violation
We conclude our discussion of numerical results by considering the behaviour of

various cross sections in the high-energy limit, using the different schemes for introducing
finite decay widths as described in Section 2.3. Table 7 shows the results for the three
reactions e+ e− → µ− ν̄µ ud̄bb̄, e+ e− → µ− ν̄µ + 4q, and e+ e− → νe ν̄e µ− ν̄µ ud̄, which are
typical representatives for top-quark pair production, WWZ production, and WW → WW
scattering. All three examples confirm the expectation from 4f (+γ ) studies that the fixed-
Table 7
Born cross sections (without ISR and gluon-exchange diagrams) for e+ e− → µ− ν̄µ ud̄bb̄, e+ e− → µ− ν̄µ + 4q
and e+ e− → νe ν̄e µ− ν̄µ ud̄ for various CM energies and schemes for introducing decay widths
σ (e+ e− → µ− ν̄µ ud̄bb̄) [fb]
√
s [GeV] 500 800 1000 2000 10000
L USIFER fixed width/ 17.095(11) 8.6795(83) 6.0263(76) 1.8631(31) 0.08783(28)
step width
running width 17.106(10) 8.6988(85) 6.0700(73) 2.3858(31) 212.61(28)
complex mass 17.085(10) 8.6773(84) 6.0249(76) 1.8627(31) 0.08800(32)
W.&M. step width 17.1025(80) 8.6823(44) 6.0183(31) 1.8657(12) 0.08837(20)

σ (e+ e− → µ− ν̄µ + 4q) [fb]
√
s [GeV] 500 800 1000 2000 10000
L USIFER fixed width/ 2.9603(28) 2.5949(26) 2.3573(25) 1.4055(20) 0.22593(63)
step width
running width 2.9845(25) 2.7354(25) 2.6829(27) 5.2921(66) 1623.0(30)
complex mass 2.9600(25) 2.5948(26) 2.3559(25) 1.4048(20) 0.22532(63)
W.&M. step width 2.9616(13) 2.5932(13) 2.3611(13) 1.4082(11) 0.2224(12)

σ (e+ e− → νe ν̄e µ− ν̄µ ud̄) [fb]
√
s [GeV] 500 800 1000 2000 10000
L USIFER fixed width 1.6326(12) 4.1046(35) 5.6795(61) 11.736(16) 26.380(55)
step width 1.6333(12) 4.1044(37) 5.6720(56) 11.734(15) 26.380(55)
running width 1.6398(12) 4.1324(39) 5.7206(54) 12.881(14) 12965(12)
complex mass 1.6330(12) 4.1037(34) 5.6705(54) 11.730(14) 26.387(57)
W.&M. step width 1.6313(13) 4.1053(35) 5.6695(49) 11.741(12) 26.565(83)
width scheme, in spite of violating gauge invariance, practically yields the same results as
the complex-mass scheme that maintains gauge invariance. For the first two processes the
step width and the fixed width lead to the same results, since no propagators of unstable
particles with space-like momenta (t-channel propagators) contribute, i.e., the widths in
the propagators are never switched off by the step function in Eq. (2.48). For the last
example, the difference between fixed and step widths is also marginal. In this context, the
comparison with the W HIZARD and M ADGRAPH results is particularly interesting, since
M ADGRAPH employs the step width within the unitary gauge, in contrast to L USIFER,
where the ’t Hooft–Feynman gauge is used. Thus, there is a difference between the
M ADGRAPH and L USIFER results for the step width, since gauge invariance is broken
in this approach. However, this difference is not yet numerically significant in the shown
numbers. Finally, all examples of Table 7 show that the running-width scheme breaks
gauge invariance so badly that deviations from the complex-mass scheme are already
visible below 1 TeV. Above 1 TeV these deviations grow rapidly, and the high-energy
limit of the prediction is totally wrong.
From these results we can draw similar conclusions as known from 4f (+γ ) production.
If finite decay widths are introduced on cost of gauge invariance, the result is only reliable
if it has been compared to a gauge-invariant calculation, as it is for instance provided by the
complex-mass scheme. Moreover, our numerical studies show that the fixed-width scheme
is in fact a good candidate for reliable results also in six-fermion production, although
it does not respect gauge invariance. Whether this observation generalizes to all 6f final
states (or even further) is, however, not clear.
7. Summary and outlook
The investigation of six-fermion production is one of the most important tasks at a future
high-energy e+ e− collider owing to a variety of interesting subprocesses leading to such
final states. These, in particular, comprise top-quark pair production, massive vector-boson
scattering, triple gauge-boson production, and Higgs-boson production for intermediate
Higgs masses.
In this paper, the Monte Carlo event generator L USIFER has been introduced, which in
its first version deals with all processes e+ e− → 6 fermions at tree level in the Standard
Model. In the predictions all Feynman diagrams are included, the number of which is
typically of the order of 102 –104 . Fermions other than top quarks, which are not allowed as
external fermions, are taken to be massless, and polarization is fully supported. The helicity
amplitudes are generically calculated with spinor methods and are presented explicitly. The
phase-space integration is based on the multi-channel Monte Carlo integration technique.
More precisely, channels and appropriate mappings are provided for each individual
diagram in a generic way. Owing to the potentially large number of Feynman diagrams per
final state, an efficient generic approach has been crucial, in order to gain an acceptable
speed and stability of the program. Initial-state radiation is included at the leading
logarithmic level employing the structure-function approach.
The performance of L USIFER has been demonstrated in detail. In particular, a compre-
hensive survey of cross section results is presented, including all final states with up to
three neutrinos and up to four quarks. Moreover, these cross sections are confronted with
results obtained with the multi-purpose packages W HIZARD and M ADGRAPH in a tuned
comparison, as far as these programs were applicable. Apart from a few cases, where the
limitations of W HIZARD and M ADGRAPH becomes visible, we find good numerical agree-
ment.
We have supplemented the numerical results on cross sections by presenting some
distributions that are phenomenologically interesting for top-quark pair and Higgs-boson
production. A comparison to W HIZARD and M ADGRAPH results shows also in this case
good agreement within statistical errors.
Finally, we have numerically investigated possible effects from gauge-invariance
violation due to the introduction of the finite decay widths of unstable particles in the
amplitudes. Similarly to the known results in four-fermion production, it turns out that the
use of running decay widths in general leads to a totally wrong high-energy behaviour of
the cross section, since gauge cancellations are disturbed. Although an approach based on
fixed gauge-boson widths does not maintain gauge invariance either, the gauge-breaking
effects are found to be sufficiently suppressed in the considered examples. This conclusion
is based on a comparison with the gauge-invariant result obtained in the so-called complex-
mass scheme, where gauge invariance is restored by introducing appropriate complex
couplings that are derived from complex mass parameters.
Apart from emphasizing the issue of speed and stability, a third motivation in the
construction of L USIFER lies in the simplicity and flexibility of both the underlying concept
and the actual computer code. This simplicity considerably facilitates the task of going
a step further in theoretical sophistication, i.e., beyond a tree-level Monte Carlo program
improved by universal corrections. The high accuracy of future e+ e− colliders requires
the inclusion of non-universal radiative corrections to subprocesses such as top-quark pair
and Higgs-boson production, or vector-boson scattering. To improve L USIFER accordingly
will be the subject of future work.
Acknowledgements
We thank Wolfgang Kilian for his aid in the W HIZARD installation and Tim Stelzer and
David Rainwater for their help in the application of M ADGRAPH to six-fermion production
processes. Moreover, Ansgar Denner is gratefully acknowledged for discussions and for
carefully reading the manuscript. This work was partially supported by the European
Commission 5th framework contract HPRN-CT-2000-00149.
References
[1] E. Accomando, et al., ECFA/DESY LC Physics Working Group Collaboration, Phys. Rep. 299 (1998) 1,
hep-ph/9705442;
J.A. Aguilar-Saavedra, et al., TESLA technical design report part III: physics at an e+ e− linear collider,
hep-ph/0106315.
[2] M.W. Grünewald, et al., in: S. Jadach, G. Passarino, R. Pittau (Eds.), Reports of the Working Groups on
Precision Calculations for LEP2 Physics, CERN, Geneva, 2000, p. 1, CERN 2000-009, hep-ph/0005309.
[3] F. Yuasa, Y. Kurihara, S. Kawabata, Phys. Lett. B 414 (1997) 178, hep-ph/9706225.
[4] T. Ishikawa, T. Kaneko, K. Kato, S. Kawabata, Y. Shimizu, H. Tanaka, MINAMI-TATEYA Group
Collaboration, KEK-92-19.
[5] E. Accomando, A. Ballestrero, M. Pizzio, Nucl. Phys. B 512 (1998) 19, hep-ph/9706201.
[6] E. Accomando, A. Ballestrero, M. Pizzio, hep-ph/9709277.
[7] F. Gangemi, G. Montagna, M. Moretti, O. Nicrosini, F. Piccinini, Nucl. Phys. B 559 (1999) 3, hep-
ph/9905271.
[8] A. Ballestrero, E. Maina, Phys. Lett. B 350 (1995) 225, hep-ph/9403244;
A. Ballestrero, hep-ph/9911318.
[9] F. Caravaglios, M. Moretti, Phys. Lett. B 358 (1995) 332, hep-ph/9507237.
[10] K. Kolodziej, hep-ph/0110063.
[11] F. Gangemi, hep-ph/0002142.
[12] R. Chierici, S. Rosati, M. Kobel, LC-PHSM-2001-038, 2nd ECFA/DESY Study 1998–2001, p. 1906.
[13] M. Moretti, T. Ohl, J. Reuter, LC-TOOL-2001-040, 2nd ECFA/DESY Study 1998–2001, p. 1981. hep-
ph/0102195.
[14] W. Kilian, LC-TOOL-2001-039, 2nd ECFA/DESY Study 1998–2001, p. 1924.
[15] G. Montagna, M. Moretti, O. Nicrosini, F. Piccinini, Eur. Phys. J. C 2 (1998) 483, hep-ph/9705333.
[16] E. Accomando, A. Ballestrero, M. Pizzio, Nucl. Phys. B 547 (1999) 81, hep-ph/9807515.
[17] F. Gangemi, G. Montagna, M. Moretti, O. Nicrosini, F. Piccinini, Eur. Phys. J. C 9 (1999) 31, hep-
ph/9811437.
[18] C.G. Papadopoulos, Comput. Phys. Commun. 137 (2001) 247, hep-ph/0007335;
A. Kanaki, C.G. Papadopoulos, Comput. Phys. Commun. 132 (2000) 306, hep-ph/0002082;
A. Kanaki, C.G. Papadopoulos, hep-ph/0012004.
[19] S. Moretti, Phys. Lett. B 420 (1998) 367, hep-ph/9711518;

S. Moretti, Nucl. Phys. B 544 (1999) 289, hep-ph/9808430;
S. Moretti, Eur. Phys. J. C 9 (1999) 229, hep-ph/9901438.
[20] F.A. Berends, R. Pittau, R. Kleiss, Nucl. Phys. B 424 (1994) 308, hep-ph/9404313;
F.A. Berends, R. Pittau, R. Kleiss, Comput. Phys. Commun. 85 (1995) 437, hep-ph/9409326.
[21] A. Denner, S. Dittmaier, M. Roth, D. Wackeroth, Nucl. Phys. B 560 (1999) 33, hep-ph/9904472.
[22] A. Denner, S. Dittmaier, M. Roth, D. Wackeroth, Phys. Lett. B 475 (2000) 127, hep-ph/9912261;
A. Denner, S. Dittmaier, M. Roth, D. Wackeroth, Nucl. Phys. B 587 (2000) 67, hep-ph/0006307;
A. Denner, S. Dittmaier, M. Roth, D. Wackeroth, Eur. Phys. J. C 20 (2001) 201, hep-ph/0104057.
[23] S. Dittmaier, Phys. Rev. D 59 (1999) 016007, hep-ph/9805445.
[24] F.A. Berends, P.H. Daverveldt, R. Kleiss, Nucl. Phys. B 253 (1985) 441;
J. Hilgart, R. Kleiss, F. Le Diberder, Comput. Phys. Commun. 75 (1993) 191.
[25] R. Kleiss, R. Pittau, Comput. Phys. Commun. 83 (1994) 141, hep-ph/9405257.
[26] E.A. Kuraev, V.S. Fadin, Yad. Fiz. 41 (1985) 753, Sov. J. Nucl. Phys. 41 (1985) 466;
G. Altarelli, G. Martinelli, in: J. Ellis, R. Peccei (Eds.), Physics at LEP, Vol. 1, CERN, Geneva, 1986, p. 47,
CERN 86-02;
O. Nicrosini, L. Trentadue, Phys. Lett. B 196 (1987) 551;
O. Nicrosini, L. Trentadue, Z. Phys. C 39 (1988) 479;
F.A. Berends, W.L. van Neerven, G.J. Burgers, Nucl. Phys. B 297 (1988) 429;
F.A. Berends, W.L. van Neerven, G.J. Burgers, Nucl. Phys. B 304 (1988) 921, Erratum.
[27] W. Beenakker, et al., in: G. Altarelli, T. Sjöstrand, F. Zwirner (Eds.), Physics at LEP2, Vol. 1, Geneva, 1996,
p. 79, CERN 96-01, hep-ph/9602351.
[28] T. Stelzer, W.F. Long, Comput. Phys. Commun. 81 (1994) 357, hep-ph/9401258;
H. Murayama, I. Watanabe, K. Hagiwara, KEK-91-11.
[29] M. Böhm, H. Spiesberger, W. Hollik, Fortsch. Phys. 34 (1986) 687;
A. Denner, Fortsch. Phys. 41 (1993) 307;
A. Denner, S. Dittmaier, G. Weiglein, Nucl. Phys. B 440 (1995) 95, hep-ph/9410338.
[30] R.G. Stuart, Phys. Lett. B 262 (1991) 113;
H.G. Veltman, Z. Phys. C 62 (1994) 35;
A. Aeppli, F. Cuypers, G.J. van Oldenborgh, Phys. Lett. B 314 (1993) 413, hep-ph/9303236.
[31] E.N. Argyres, et al., Phys. Lett. B 358 (1995) 339, hep-ph/9507216;
W. Beenakker, et al., Nucl. Phys. B 500 (1997) 255, hep-ph/9612260.
[32] S. Dittmaier, hep-ph/9710542.
[33] W. Beenakker, F.A. Berends, A.P. Chapovsky, Nucl. Phys. B 573 (2000) 503, hep-ph/9909472.
[34] T. Stelzer, private communications.
[35] M. Roth, Doctoral thesis, hep-ph/0008033.
[36] E. Byckling, K. Kajantie, Particle Kinematics, Wiley, London, 1973, p. 158ff.
[37] Y. Kurihara, et al., Prog. Theor. Phys. 103 (2000) 1199, hep-ph/9912520;
Y. Kurihara, et al., Eur. Phys. J. C 20 (2001) 253, hep-ph/0011276;
G. Montagna, M. Moretti, O. Nicrosini, A. Pallavicini, F. Piccinini, Eur. Phys. J. C 20 (2001) 217, hep-
ph/0005121;
G. Passarino, hep-ph/0101139;
G. Passarino, Nucl. Phys. B 619 (2001) 313, hep-ph/0108255.
[38] D.E. Groom, et al., Particle Data Group Collaboration, Eur. Phys. J. C 15 (2000) 1.
[39] A. Djouadi, J. Kalinowski, M. Spira, Comput. Phys. Commun. 108 (1998) 56, hep-ph/9704448.
[40] W. Kilian, private communications.
Initial-state interactions and single-spin asymmetries

in Drell–Yan processes ✩
Stanley J. Brodsky a , Dae Sung Hwang a,b , Ivan Schmidt c
a Stanford Linear Accelerator Center, Stanford University, Stanford, CA 94309, USA
b Department of Physics, Sejong University, Seoul 143-747, South Korea
c Departamento de Física, Universidad Técnica Federico Santa María, Casilla 110-V, Valparaíso, Chile
Received 27 June 2002; accepted 23 July 2002
Abstract
We show that the initial-state interactions from gluon exchange between the incoming quark
and the target spectator system lead to leading-twist single-spin asymmetries in the Drell–Yan
↑
process H1 H2 → + − X. The QCD initial-state interactions produce a T -odd spin-correlation
SH2 · PH1 × Q
between the target spin and the virtual photon production plane which is not power-law
suppressed in the Drell–Yan scaling limit at large photon virtuality Q2 at fixed xF . The single-spin
asymmetry which arises from the initial-state interactions is not related to the target or projectile
transversity distribution δqH (x, Q). The origin of the single-spin asymmetry in πp↑ → + − X is a
phase difference between two amplitudes coupling the proton target with Jpz = ± 12 to the same final-
state, the same amplitudes which are necessary to produce a nonzero proton anomalous magnetic
moment. The calculation requires the overlap of target light-front wavefunctions with different orbital
angular momentum: Lz = 1; thus the SSA in the Drell–Yan reaction provides a direct measure of
orbital angular momentum in the QCD bound state. The single-spin asymmetry predicted for the
Drell–Yan process πp↑ → + − X is similar to the single-spin asymmetries in deep inelastic semi-
inclusive leptoproduction p↑ → πX which arises from the final-state rescattering of the outgoing
quark. The Bjorken-scaling single-spin asymmetries predicted for the Drell–Yan and leptoproduction
processes highlight the importance of initial- and final-state interactions for QCD observables.
 2002 Published by Elsevier Science B.V.
PACS: 12.38.-t; 12.38.Bx; 13.85.-t; 13.88.+e
✩
Work partially supported by the Department of Energy, contract DE-AC03-76SF00515, by the LG Yonam
Foundation, by Fondecyt (Chile) under grant 8000017 and by MECESUP (Chile) program FSM9901.
E-mail addresses: sjbth@slac.stanford.edu (S.J. Brodsky), dshwang@sejong.ac.kr (D.S. Hwang),
ischmidt@fis.utfsm.cl (I. Schmidt).
0550-3213/02/$ – see front matter  2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 1 7 - X
S.J. Brodsky et al. / Nuclear Physics B 642 (2002) 344–356 345
1. Introduction
Single-spin asymmetries in hadronic reactions provide a remarkable window to QCD

mechanisms at the amplitude level. In general, single-spin asymmetries measure the
correlation of the spin projection of a hadron with a production or scattering plane [1].
Such correlations are odd under time reversal, and thus they can arise in a time-reversal
invariant theory only when there is a phase difference between different spin amplitudes.
Specifically, a nonzero correlation of the proton spin normal to a production plane measures
the phase difference between two amplitudes coupling the proton target with Jpz = ± 12 to
the same final-state. The calculation requires the overlap of target light-front wavefunctions
with different orbital angular momentum: Lz = 1; thus a single-spin asymmetry (SSA)
provides a direct measure of orbital angular momentum in the QCD bound state.
Consider the SSA produced in semi-inclusive deep inelastic scattering p↑ → πX.
In the target rest frame, such a single target spin correlation corresponds to the T -odd
triple product Sp · pπ × q . (The covariant form of this correlation is µνσ τ Sp pν q σ pπτ .)
µ
Significant asymmetries AU L and AU T of this type have in fact been observed for targets
polarized parallel to or transverse to the lepton beam direction [2,3].
In a recent paper [4] we have shown that the QCD final-state interactions (gluon
exchange) between the struck quark and the proton spectator system in semi-inclusive
deep inelastic lepton scattering can produce single-spin asymmetries which survive in the
Bjorken limit. Such effects are proportional to the matrix element of a higher-twist quark–
quark–gluon correlator in the target hadron, and thus it has been assumed on dimensional
grounds that any SSA arising from this source must be suppressed by a power of the
momentum transfer Q in the Bjorken limit. However, another momentum scale enters into
the semi-inclusive process—the transverse momentum r⊥ = pπ ⊥ − q⊥ of the emitted pion
relative to the photon direction, and we have shown that the power-law suppression due
to the higher-twist quark–quark–gluon correlator takes the form of an inverse power of r⊥
rather Q. As shown in the appendix, r⊥ 2 can be written in terms of the invariant momentum
transfer squared t from the proton to the spectator system and the Bjorken variable.
Corrections from spin-one gluon exchange in the initial- or final-state of QCD processes
are not suppressed at high energies because the coupling is vector-like. Therefore, as
a consequence of the gauge coupling of QCD, single-spin asymmetries in semi-inclusive
deep inelastic scattering survive in the Bjorken limit of large Q2 at fixed xbj and fixed r⊥ .
Recently it has been shown [5,6] that the same type of final-state interaction is the origin
of the leading-twist diffractive component in deep inelastic scattering, implying that the
pomeron is not a universal property of the target proton’s wavefunction, and that it depends
in detail on the deep inelastic scattering (DIS) process itself. Diffractive processes in DIS
in turn lead to nuclear shadowing in the case of nuclear targets, showing that shadowing is
not an intrinsic property of nuclear wavefunctions.
The final-state phases which we compute are analogous to the “Coulomb” phases to the
hard subprocess which arises from gauge interactions between outgoing charge particles
in QED [7]. More specifically, we require the difference between the gauge interaction
phases for the Jpz = ± 12 amplitudes. The phases depend on the spin because the outgoing
particles interact at different impact separation corresponding to their different relative
orbital angular momentum.
346 S.J. Brodsky et al. / Nuclear Physics B 642 (2002) 344–356
Fig. 1. The initial-state interaction in the Drell–Yan process.
In our previous paper [4], we explicitly evaluated the SSA for electroproduction for
a specific model of a spin-1/2 proton of mass M with charged spin-1/2 and spin-0
constituents of mass m and λ, respectively, as in the QCD-motivated quark–diquark model
of a nucleon. The basic leptoproduction reaction is then γ ∗ p → q(qq)0. Our analysis
predicts a nonzero SSA for the target spin normal to the photon to quark-jet Sp · pq × q
which can be determined by using a jet variable such as thrust to determine the current
quark direction; i.e., we predict a SSA even without final-state jet hadronization. Our
mechanism is thus distinct from a description of SSA based on transversity and phased
fragmentation functions.
Recently Collins [8] has pointed out some important consequences of these results
for SSA in deep inelastic scattering. In his treatment the final-state interactions of the
struck quark are incorporated into Wilson line path-ordered exponentials which aug-
ment the light-cone wavefunctions. (See also [9].) Since the final-state interactions ap-
pear at short light-cone times x + = O(1/ν) after the virtual photon acts, they can
be distinguished from hadronization processes which occur over long times. Collins has
stressed the fact that single-spin asymmetries probe the partonic structure associated with
chiral-symmetry breaking. Furthermore, these results show that time-reversal-odd par-
ton densities are allowed, opening up a whole range of phenomenological applications
[10–12]. In particular, as noted by Collins, initial-state interactions between the annihilat-
ing antiquark and the spectator system of the target can produce single-spin asymmetries
in the Drell–Yan process.
In this paper we shall extend our analysis to initial- and final-state QCD effects to predict
single-spin asymmetries in hadron-induced hard QCD processes. Specifically, we shall
consider the Drell–Yan (DY) type reactions [13] such as πp↑ → + − X. Here the target
particle is polarized normal to the pion-to-lepton-pair production plane. The target spin
asymmetry can be produced due to the initial-state gluon-exchange interactions between
the interacting antiquark coming from one hadronic system and the spectator system of the
other. This is shown in the diagram of Fig. 1. The importance of initial-state interactions in
the theory of massive lepton pair production, Q⊥ broadening, and energy loss in a nuclear
target has been discussed in Refs. [14–16].
The orientation of the target spin Sz = ±1/2 corresponds to amplitudes differing by
relative orbital angular momentum Lz = 1. The initial-state interaction from a gluon
exchanged between the annihilating antiquark and target spectator system depends in detail
on this relative orbital angular momentum. In contrast, the initial or final-state interactions
due to the exchange of gauge particles between partons not participating in the hard
subprocess do not contribute to the SSA. Such spectator-spectator interactions occur at
large impact separation and are not sensitive to just one unit difference Lz = 1 of the
orbital angular momentum of the target wavefunction.
Our mechanism thus depends on the interference of different amplitudes arising from
the target hadron’s wavefunction and is distinct from probabilistic measures of the target
such as transversity. It is also important to note that the target spin asymmetries which
we compute in the DY and DIS processes require the same overlap of wavefunctions
which enters the computation of the target nucleon’s magnetic moment. In addition, by
selecting different initial mesons in the DY process, we can isolate the flavor of the
annihilating quark and antiquark. The flavor dependence of single-spin asymmetries thus
has the potential to provide detailed information of the spin and flavor content of nucleons
at the amplitude level.
As in our analysis of semi-inclusive DIS, we shall calculate the single-spin asymmetry
in the Drell–Yan process induced by initial-state interactions by adopting an effective
theory of a spin-1/2 proton of mass M with charged spin-1/2 and spin-0 constituents of
mass m and λ, respectively, as in a quark–diquark model. We will take the initial particle
to be just an antiquark. The result for specific meson projectiles such as Mp↑ → + − X is
then obtained by convolution with the antiquark distribution of the incoming meson. One
can also incorporate target nucleon wavefunctions with a quark-vector diquark structure.
In a more complete study, one should allow for a many-parton light-front Fock state
wavefunction representation of the target. The results, however, can be normalized to the
quark contribution to the proton anomalous moment, and thus are in large part model-
independent.
2. Crossing
There is a simple diagrammatic connection between the amplitude describing the initial-
state interaction of the annihilating antiquark, which gives a single-spin asymmetry for
the Drell–Yan process πp↑ → + − X, and the final-state rescattering amplitude of the
struck quark, which gives the single-spin asymmetries in semi-inclusive deep inelastic
leptoproduction p↑ → πX. The crossing of the Feynman amplitude for γ ∗ (q̃)p(P ) →
(q̃ + r)(P − r) in DIS gives (−q̃ − r)p(P ) → γ ∗ (−q̃)(P − r) for DY by reversing the
four-vectors of the photon and quark lines. The outgoing quark with momentum q̃ + r in
DIS becomes the incoming antiquark with momentum −q̃ − r in DY. We can use crossing
of the Lorentz invariant amplitudes for DIS as a guide for obtaining the amplitudes for DY
amplitude [17]. (In the next section it will be convenient to label q̃ = −q with q + > 0.)
In general, one cannot use crossing to relate imaginary parts of amplitudes to each other,
since under crossing, real and imaginary parts become connected. However, in our case, the
relevant one-gluon exchange diagrams in DIS and DY are both purely imaginary at high
energy, so their magnitudes are related by crossing. Thus a crucial test of our mechanism
is an exact relation between the magnitude and flavor dependence of the SSA in the Drell–
Yan reaction and the SSA in deep inelastic scattering. We thus predict the DY SSA of the
proton spin with the normal to the antiquark to virtual photon plane: Sp · pq × q̃. It is

identical—up to a sign—to the SSA computed in DIS for Sp · pq × q .
The phase arising from the initial- and final-state interactions in QCD is analogous to the
Coulomb phase of Abelian QED amplitudes. The Coulomb phase depends on the product
of charges and relative velocity of each ingoing and outgoing charged pair [7]. Thus the
sign of the phase in DY and DIS are opposite because of the different color charge of the
ingoing 3C antiquark in DY and the outgoing 3C quark in DIS. In order to check the sign,
we will carry out the DY calculation explicitly in the next section.
The asymmetry in the Drell–Yan process is thus the same as that obtained in DIS, with
the appropriate identification of variables, but with the opposite sign. This has been stressed
recently by Collins [8]. Therefore, the single-spin asymmetry transverse to the production
plane in the Drell–Yan process can be obtained from the results of our recent paper [4]:

e1 e2 2(M + m)r 1 m2 λ2
Py = − r + (1 − ) −M +
2 2
+
8π [(M + m)2 + r⊥2 ] ⊥ 1−
2 2
1 r⊥2 + (1 − )(−M 2 + + 1− )
m λ
× ln . (1)
r⊥2 2 λ2
(1 − )(−M 2 + m + 1− )
2 2
Here = 2Pq ·q = 2Mν
q
where ν is the energy of the lepton pair in the target rest frame. The
kinematics are given in detail in Appendix A.
3. Calculation
We can check the results of the previous section obtained using crossing by performing
a direct calculation of the qp↑ → γ ∗ (qq)0 amplitude where we take a spin-zero diquark
for the proton spectator system. The kinematics are (q − r)p(P ) → γ ∗ (q)(P − r), as in
Fig. 1. The J z = + 12 two-particle Fock state is given by [18,19]
↑ +
Ψ
two particle P , P⊥ = 0⊥

d 2 k⊥ dx ↑
= √ ψ 1 x, k⊥ + 12 ; xP + , k⊥
x(1 − x) 16π 3 +2
↑
+ ψ 1 x, k⊥ − 1 ; xP + , k⊥ , (2)
−2 2
where
 ↑
 ψ 1 x, k⊥ = M + m x ϕ,
+2
(3)
 ψ ↑ 1 x, k⊥ = − (+k +ik ) ϕ.
1 2
−2 x
The scalar part of the wavefunction ϕ depends on the dynamics. In the perturbative theory
it is simply
√
e/ 1 − x
ϕ = ϕ x, k⊥ = . (4)
k2 +m2 k 2 +λ2
M 2 − ⊥ x − ⊥1−x
In general one normalizes the Fock state to unit probability.

Similarly, the J z = − 12 two-particle Fock state has two components

 ψ ↓ 1 x, k⊥ = (+k x−ik ) ϕ,
1 2
+2
(5)
 ψ ↓ 1 x, k⊥ = M + m ϕ.
−2 x
The spin-flip amplitudes in (3) and (5) have orbital angular momentum projection l z = +1
and −1, respectively. The numerator structure of the wavefunctions is characteristic of the
orbital angular momentum, and holds for both perturbative and non-perturbative couplings.
We require the interference between the tree amplitude of Fig. 1(a) and the one-loop
amplitude of Fig. 1(b). The contributing amplitudes have the following structure through
one-loop order:

m e1 e2
A(⇑→↓) = M + C h+i g1 , (6)
8π
1
+r − ir 2 e1 e2
A(⇓→↓) = C h+i g2 , (7)
8π
1
−r − ir 2 e1 e2
A(⇑→↑) = C h+i g2 , (8)
8π

m e1 e2
A(⇓→↑) = M + C h+i g1 , (9)
8π
where

C = −ge1 P + β − 2(1 − ) (10)
1
h= 2
. (11)
λ2
r⊥ + (1 − )(−M 2 + m +
2
1− )
The label ⇑ / ⇓ corresponds to Jpz = ± 12 of the proton spin. The second label ↑/↓ gives
the spin projection Jqz = ± 12 of the interacting spin-1/2 constituent antiquark of the other
proton. Here e1 and e2 are the electric charges of the proton’s constituents q and (qq)0 ,
respectively, and g is the coupling constant of the effective proton-q-(qq)0 vertex. The first
term in (6) to (9) is the Born contribution of the tree graph. The crucial result will be the
fact that the contributions g1 and g2 from the one-loop diagram Fig. 1(b) are different, and
that their difference is infrared finite. A gauge boson mass λg will be used as an infrared
regulator in the calculation of g1 and g2 . The final result for g1 −g2 is infrared finite, and λg
can be set to zero. The calculation will be done using light-cone time-ordered perturbation
theory, or equivalently, by integrating Feynman loop-diagrams over dk − .
We take q to lie in the ẑ − x̂ plane, q = (q x , q y , q z ) = (q 1 , 0, q 3 ). We denote q + as
q + = βP + . (12)
From energy conservation, we get
q⊥2
q− = +
. (13)
P (β − )
Then we have the relation

q 2 = q + q − − q⊥2 = q 2 , (14)
β − ⊥
where
β . (15)
Further details on the kinematics are given in Appendix A.
The covariant expression for the four one-loop amplitudes of diagram Fig. 1(b) is:
Aone-loop (I )

d 4k
= ige1 (−e1 e2 )
(2π)4
N (I )
×
(k 2 − m2 + i)((−k + q)2 − m2 + i)((k − r)2 − λ2g + i)
1
×
((k − P )2
− λ2 + i)

d 2 k⊥ N (I )
= −ige1 (−e1 e2 ) P + dx +4
2(2π)4 P x(β − x)(x − )(1 − x)

1
× dk − 2 )−i
2 +k
q⊥ )2 )−i
k⊥ +
(−k − + q − ) − (m +(−
(m 2
k− − xP +
⊥
(β−x)P +
1
× 2 )−i
, (16)
(λ2g +(k⊥ −r⊥ )2 )−i (λ2 +k⊥
(k − − r − ) − (x−)P +
(k − − P − ) + (1−x)P +
where we used k + = xP + . The numerators N (I ) are given by

m
N (⇑→↓) = N M + , (17)
x
1
+k − ik 2
N (⇓→↓) = N , (18)
x
1
−k − ik 2
N (⇑→↑) = N , (19)
x

m
N (⇓→↑) = N M + , (20)
x
where

N = 2P + β − xq − −P + (1 − x) + (1 − ) , (21)
q⊥2
and q − = P + (β−) as given in (13). For the [current]–[gauge boson propagator]–[current]
factor, in Feynman gauge only the −g +− term of the gauge boson propagator −g µν
contributes in the Bjorken limit, and it provides a factor proportional to q − in the numerator
which cancels the q − in the denominator which provides the imaginary part. Therefore, the
result scales in the Bjorken limit.
The integration over k − in (16) does not give zero only if 0 < x < 1. We first consider
the region < x < 1.
Aone-loop (I )

d 2 k⊥ N (I )
= −ige1 (−e1 e2 ) × (2πi) P + dx
2(2π)4 P +4 x(β − x)(x − )(1 − x)
1
× 2 )−i
(λ2 +k⊥
2 )−i (m2 +k⊥
P− − (1−x)P +
− xP +
1
×
(λ2 +k⊥ q⊥ )2 )−i
(m2 +(−k⊥ +
2 )−i
−P − + (1−x)P +
+ q− − (β−x)P +
1
× , (22)
(λ2 +k⊥
2 )−i (λ2g +(k⊥ −r⊥ )2 )−i
P− − (1−x)P +
− r− − (x−)P +
where we used k + = xP + . The result is identical to that obtained from light-cone time-
ordered perturbation theory.
The phases χi needed for single-spin asymmetries come from the imaginary part of (22),
which arises from the potentially real intermediate state allowed before the rescattering.
The imaginary part of the propagator (light-cone energy denominator) gives

(λ2 + k⊥
2)
(m2 + (−k⊥ + q⊥ )2 )
−iπδ −P − + + q −
−
(1 − x)P + (β − x)P +
1 (β − )2
= −iπ + δ x −−δ , (23)
P q⊥
2
where
q⊥ · (k⊥ − r⊥ )
δ = 2(β − ) . (24)
q⊥2
Since the exchanged momentum δP + is small, the light-cone energy denominator

g (k⊥ −r⊥ )2 +λ2
corresponding to the gauge boson propagator is dominated by the (x−) term. This

gets multiplied by (x −), so only (k⊥ − r⊥ ) + λg appears in the propagator, independent
2 2
of whether the photon is absorbed or emitted. The contribution from the region 0 < x <
thus compliments the contribution from the region < x < 1.
We can integrate (22) over the transverse momentum using a Feynman parametrization
to obtain the one-loop terms in (6) to (9).
1
1
g1 = dα , (25)
m2 λ2
α(1 − α)r⊥2 + αλ2g + (1 − α)(1 − )(−M 2 + + 1− )
0
1
α
g2 = dα . (26)
m2 λ2
α(1 − α)r⊥
2 + αλ2g + (1 − α)(1 − )(−M 2 + + 1− )
0
We define:
2 2 2 2
Pz = C −1 A(⇑→↑) − A(⇓→↑) + A(⇑→↓) − A(⇓→↓) , (27)

Px = C −1 A(⇑→↑)∗ A(⇓→↑) + A(⇑→↑)A(⇓→↑)∗

+ A(⇑→↓)∗ A(⇓→↓) + A(⇑→↓)A(⇓→↓)∗ , (28)

Py = C −1 i A(⇑→↑)∗ A(⇓→↑) − A(⇑→↑)A(⇓→↑)∗

+ i A(⇑→↓)∗ A(⇓→↓) − A(⇑→↓)A(⇓→↓)∗ , (29)
where the normalization from the unpolarized cross section is

2 2 2 2
C = A(⇑→↑) + A(⇓→↑) + A(⇑→↓) + A(⇓→↓) . (30)
We can assume for convenience that the initial-state interactions generate a phase
when exponentiated, as in the Coulomb phase analysis of QED. The rescattering phases
eiχi (i = 1, 2) with χi = tan−1 ( e8π
1 e2 gi
h ) are thus distinct for the spin-parallel and spin-
antiparallel amplitudes. The difference in phase arises from the orbital angular momentum
k⊥ factor in the spin-flip amplitude, which after integration gives the extra factor of the
Feynman parameter α in the numerator of g2 compared to g1 , as we can see in (25)
and (26). Notice that the phases χi are each infrared divergent for zero gauge boson
mass λg → 0, as is characteristic of Coulomb phases. However, the difference χ1 − χ2
which contributes to the single-spin asymmetry is infrared finite. We have verified that the
Feynman gauge result is also obtained in the light-cone gauge using the principal value
prescription. The small numerator coupling of the light-cone gauge boson is compensated
by the small value for the exchanged l + = δP + momentum.
The virtual photon and produced hadron define the production plane which we will
take as the ẑ − x̂ plane. From Eqs. (6)–(9) and (29), the azimuthal single-spin asymmetry
transverse to the production plane is given by
e1 e2 2(M + m)r 1
Py = −
8π [(M + m)2 + r⊥
2]

m2 λ2
× r⊥2 + (1 − ) −M 2 + +
1−
2 2
1 r⊥2 + (1 − )(−M 2 + + 1− )
m λ
× 2 ln , (31)
r⊥ 2
(1 − )(−M 2 + m + 1−
λ 2
)
in agreement with the crossing properties described in Section 2. The linear factor of
r 1 = r x reflects the fact that the single-spin asymmetry is proportional to Sp · q × r since
q ∼ −ẑ|q | and Sp = ±ŷ. The kinematics are given in more detail in Appendix A. The
prediction for Py as a function of and r⊥ is identical but with opposite sign to that
illustrated in Fig. 4 of Ref. [4].
Our analysis can be generalized to the corresponding calculation in QCD. The initial-
state interaction from gluon exchange has the strength e4π 1 e2
→ CF αs (µ2 ). The scale of
αs in the MS scheme can be identified with the momentum transfer carried by the gluon
µ2 = e−5/3 (k⊥ − r⊥ )2 [20]. The matrix elements coupling the proton to its constituents
will have the same numerator structure as the perturbative model since they are determined
by orbital angular momentum constraints. The strengths of the proton matrix elements
can be normalized quark by quark according to their contributions to the target nucleon’s
anomalous magnetic moment weighted by the quark charge squared. In QCD, r⊥ is the
magnitude of the momentum of the current quark jet relative to the virtual photon direction.
αs (r⊥2 )x Mr ln r 2
⊥
Notice that for large r⊥ , Py decreases as bj
2
⊥
. The physical proton mass M
r⊥
appears since it is present in the ratio of the L = 1 and L = 0 matrix elements. This form
z z
is expected to be essentially universal.
4. Summary
We have shown that the same physical mechanism which produces a leading-twist
single-spin asymmetry in semi-inclusive DIS also leads to a leading twist single-spin
asymmetry in the Drell–Yan process. The initial-state interaction between the annihilating
antiquark with the spectator of the target produces the required phase correlation. The
equality in magnitude, but opposite sign, of the single-spin asymmetries in semi-inclusive
DIS and the corresponding Drell–Yan processes is an important check of our mechanism.
It has been conventional to assume that the effects of initial- and final-state interactions
are always power-law suppressed for hard processes in QCD. In fact, this is not in general
correct, as can be seen from our analyses of leading-twist single-spin asymmetries in
the Drell–Yan process and semi-inclusive deep inelastic scattering. The initial- and final-
state interactions which survive in the scaling limit occur in light-cone time τ = O(1/ν)
immediately before or after the hard subprocess. Other initial- and final-state interactions,
such as those between the spectator of the incident hadron and the spectator of the
target hadron in the DY process, take place over long time scales, and they only provide
inconsequential unitary phase corrections to the process. This is in accord with our intuition
that interactions which occur at distant times cannot affect the primary reaction.
A natural framework for the wavefunctions which appear in the SSA calculations is the
light-front Fock expansion [21,22]. In principle, the light-front wavefunctions for hadrons
can be obtained by solving for the eigen-solutions of the light-front QCD Hamiltonian.
Such wavefunctions are real and include all interactions up to a given light-front time.
The final-state gluon-exchange corrections which provide the SSA for semi-inclusive DIS
occurs immediately after the virtual photon strikes the active quark. Such interactions are
not included in the light-front wavefunctions, just as Coulomb final-state interactions are
not included in the Schrödinger bound state wavefunctions in QED. Collins [8] has argued
that since the relevant rescattering interactions of the struck quark occur very close in
light-cone time to the hard interaction, one can augment the light-front wavefunctions by
Fig. 2. An example of a final-state interaction which can cause a single asymmetry in pp↑ → π X.
a Wilson line factor which incorporates the effects of the final-state interactions in semi-
inclusive DIS. However, such augmented wavefunctions are not universal and process
independent; for example, in the case of the DY process, an incoming Wilson line of
opposite phase must be used.
Our formalism can be adopted to single-spin asymmetries in more general hard inclusive
reactions, such as pp↑ → πX, where the pion is detected at high transverse momentum
[23,24]. In such cases one must identify the hard quark–gluon subprocess and analyze a set
of gluon exchange corrections which connect the spectators of the polarized hadron with
the active quarks and gluons of the hard subprocess. An example of a final-state interaction
which can cause a single asymmetry in pp↑ → πX is shown in Fig. 2. However, this type
of final-state interaction cannot be readily identified as an augmented target wavefunction.
It is also clear from our analyses that there are potentially important corrections to the hard
quark propagator in hard exclusive subprocesses such as deeply virtual Compton scattering
or exclusive meson electroproduction. These rescattering interactions of the propagating
quark can provide new single-spin observables and will correct analyses based on the
handbag approximation.
It should be emphasized that the same overlap of light-front wavefunctions with
Lz = 1 which gives single-spin asymmetries also yields the Pauli form factor F2 (t) and
the generalized parton distribution E(x, ζ, t) entering deeply virtual Compton scattering
[25–29]. Each quark of the target wavefunction appears additively, weighted linearly by
the quark charge in the case of the Pauli form factor and weighted quadratically in the case
of deep inelastic scattering, the Drell–Yan reaction and deeply virtual Compton scattering.
The empirical study of single-spin asymmetries in hard inclusive and exclusive
processes thus provides a new window to the investigation of hadron spin, angular
momentum, and the flavor structure of hadrons.
Acknowledgements
We thank Harut Avakian, John Collins, Paul Hoyer, Xiang-Dong Ji, and Stephane
Peigne for helpful conversations.
Appendix A. Kinematics
From Fig. 1 we have [a = (a + , a − , a⊥ )]

2
+ M
P = P , + , 0⊥ , (A.1)
P

λ2 + r⊥2
P = (1 − )P + , , −r ⊥ , (A.2)
(1 − )P +

+ q +q ⊥
2 2
q = βP , , q⊥ , (A.3)
βP +

m2 + ( q⊥ − r⊥ )2
Pq = (β − )P + , ,
q ⊥ −
r⊥ , (A.4)
(β − )P +
where
m2 + (
q⊥ − r⊥ )2
s − m2 − M 2 = 2Pq · P = + M 2 (β − ). (A.5)
(β − )
Energy conservation gives
m2 + (
q⊥ − r⊥ )2 q 2 + q⊥2 λ2 + r⊥
2
+ M2 = + . (A.6)
(β − ) β (1 − )
Here s and q 2 are large. It is convenient to work in a frame where β − = O(1) and q⊥2
q⊥2
is large so that s β− . In such a frame, q ∼ −ẑ|
q |. Then Eq. (A.6) gives
β − 2
q⊥2 = q . (A.7)

From (A.1), (A.3) and (A.7) we have
q⊥2 q2
2P · q = + βM 2 , (A.8)
β −
and thus to leading twist
q2
= , (A.9)
2P · q
where P · q/M = ν is the energy of the lepton pair in the target rest frame.
From (A.1) and (A.2) we have
λ2 + r⊥2
2P · P = + (1 − )M 2 . (A.10)
1−
Since t = (P − P )2 , we have 2P · P = −t + M 2 + λ2 . Therefore, (A.10) gives

r⊥2 = (1 − ) −t + M 2 + λ2 − (1 − )M 2 − λ2 , (A.11)
which relates r⊥2 to invariants.
References
[1] D.W. Sivers, Phys. Rev. D 43 (1991) 261.

[2] HERMES Collaboration, A. Airapetian, et al., Phys. Rev. Lett. 84 (2000) 4047;
HERMES Collaboration, A. Airapetian, et al., Phys. Rev. D 64 (2001) 097101.
[3] A. Bravar, SMC Collaboration, Nucl. Phys. B (Proc. Suppl.) 79 (1999) 520.
[4] S.J. Brodsky, D.S. Hwang, I. Schmidt, Phys. Lett. B 530 (2002) 99.
[5] S.J. Brodsky, P. Hoyer, N. Marchal, S. Peigne, F. Sannino, Phys. Rev. D 65 (2002) 114025.
[6] S. Peigne, hep-ph/0206138.
[7] S. Weinberg, Phys. Rev. 140 (1965) B 516.
[8] J.C. Collins, Phys. Lett. B 536 (2002) 43.
[9] X.D. Ji, F. Yuan, hep-ph/0206057.
[10] D. Boer, P.J. Mulders, Phys. Rev. D 57 (1998) 5780.
[11] M. Anselmino, F. Murgia, Phys. Lett. B 442 (1998) 470.
[12] D. Boer, Phys. Rev. D 60 (1999) 014012.
[13] S.D. Drell, T.M. Yan, Phys. Rev. Lett. 25 (1970) 316;
S.D. Drell, T.M. Yan, Phys. Rev. Lett. 25 (1970) 902, Erratum.
[14] G.T. Bodwin, S.J. Brodsky, G.P. Lepage, Phys. Rev. Lett. 47 (1981) 1799.
[15] G.T. Bodwin, S.J. Brodsky, G.P. Lepage, Phys. Rev. D 39 (1989) 3287.
[16] S.J. Brodsky, A. Hebecker, E. Quack, Phys. Rev. D 55 (1997) 2584.
[17] A. Brandenburg, V.V. Khoze, D. Muller, Phys. Lett. B 347 (1995) 413.
[18] S.J. Brodsky, S.D. Drell, Phys. Rev. D 22 (1980) 2236.
[19] S.J. Brodsky, D.S. Hwang, B.Q. Ma, I. Schmidt, Nucl. Phys. B 593 (2001) 311.
[20] S.J. Brodsky, A.H. Hoang, J.H. Kühn, T. Teubner, Phys. Lett. B 359 (1995) 355.
[21] G.P. Lepage, S.J. Brodsky, Phys. Rev. D 22 (1980) 2157.
[22] S.J. Brodsky, G.P. Lepage, in: A.H. Mueller (Ed.), Perturbative Quantum Chromodynamics, World
Scientific, Singapore, 1989.
[23] E704 Collaboration, A. Bravar, et al., Phys. Rev. Lett. 77 (1996) 2626.
[24] K. Heller, in: C.W. de Jager, T.J. Ketel, P. Mulders (Eds.), Proceedings of Spin 96, World Scientific,
Singapore, 1997.
[25] D. Muller, D. Robaschik, B. Geyer, F.M. Dittes, J. Horejsi, Fortsch. Phys. 42 (1994) 101.
[26] X.D. Ji, Phys. Rev. Lett. 78 (1997) 610.
[27] A.V. Radyushkin, Phys. Rev. D 56 (1997) 5524.
[28] M. Diehl, T. Feldmann, R. Jakob, P. Kroll, Nucl. Phys. B 596 (2001) 33;
M. Diehl, T. Feldmann, R. Jakob, P. Kroll, Nucl. Phys. B 605 (2001) 647, Erratum.
[29] S.J. Brodsky, M. Diehl, D.S. Hwang, Nucl. Phys. B 596 (2001) 99.
Chiral fermions on the lattice

Oliver Jahn a,b,∗ , Jan M. Pawlowski c
a Dublin Institute for Advanced Studies, 10 Burlington Road, Dublin 4, Ireland
b Institut für Theoretische Physik, ETH Zürich, CH-8093 Zürich, Switzerland 1
c Institut für Theoretische Physik III, Universität Erlangen, Staudtstraße 7, D-91054 Erlangen, Germany

Dedicated to the memory of L. O’Raifeartaigh
Abstract
We discuss topological obstructions to putting chiral fermions on an even-dimensional lattice. The
setting includes Ginsparg–Wilson fermions, but is more general. We prove a theorem which relates
the total chirality to the difference of generalised winding numbers of chiral projection operators. For
an odd number of Weyl fermions this implies that particles and anti-particles live in topologically
different spaces.
1. Introduction
The formulation of lattice theories with chiral fermions has been a long-standing
problem [1,2] which is closely related to the fermion doubling problem. It has been proven
early on, that it is impossible to maintain chiral symmetry or define a chiral gauge theory
on the lattice, if some basic assumptions are met. This is the celebrated Nielsen–Ninomiya
(NN) no-go theorem [3–7]. It states that each left-handed fermion on a lattice must be
accompanied by a right-handed fermion with the same quantum numbers. Thus, a lattice
theory for a single Weyl fermion seems to be ruled out, but also a lattice realisation of
chiral symmetry since the latter would assign opposite charges to left- and right-handed
particles.
* Corresponding author.
E-mail addresses: jahn@itp.phys.ethz.ch (O. Jahn), jmp@theorie3.physik.uni-erlangen.de
(J.M. Pawlowski).
1 Present address.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 3 9 - 9
358 O. Jahn, J.M. Pawlowski / Nuclear Physics B 642 (2002) 357–371
In turn, the Ginsparg–Wilson (GW) relation describes how close one can get to the naïve
chiral symmetry in a lattice theory [8]. The derivation of the GW relation also highlights
how the no-go theorem could be circumvented. The GW relation is derived from a block
spin transformation of a lattice action which enjoys the naïve chiral symmetry. In this
sense it defines the ‘natural’ quantum chiral symmetry on the lattice. For a long time,
investigations of lattice theories with GW fermions have been hampered by the fact that
no explicit example was known of a Dirac operator (in an interacting theory) obeying the
GW relation. This gap was filled in [9] (and, less explicitly, in [10]). It also turned out that
the action of GW fermions is invariant under a generalised chiral transformation of the
fields [11,12].
Let us now focus on how the no-go theorem is avoided: a chiral symmetry based on
the GW relation fails to meet the key assumption of the NN theorem, namely that the
symmetries act on the fermions ψ and ψ̄ as if they were (Dirac-) conjugates of each
other. While this is the case in Minkowski space–time, ψ and ψ̄ have to be treated as
independent fields in the Euclidean space–time used for the formulation of lattice theories.
Consequently, ψ and ψ̄ can transform independently under symmetry transformations.
In the present contribution we study the properties of general chiral transformations on
the lattice for the case of free fermions. We derive a statement about generalised chiral
projections including those which are related to GW Dirac operators. To that end we
consider Dirac operators D which allow the definition of local chiral projections: Pψ and
Pψ̄ with DPψ = Pψ̄ D and Pψ/ 2
ψ̄
= Pψ/ψ̄ . Apart from this, some technical conditions have
to be met in order to ensure the vanishing of lattice artifacts in the continuum limit. Then
the following statement can be proven:
χ = n[Pψ ] − n[Pψ̄ ],
where n[Pψ/ψ̄ ] is a (generalised) winding number of Pψ/ψ̄ defined in (7) and χ is the total
chirality of all fermion species emerging in the continuum limit. It is also shown that the
windings n[Pψ/ψ̄ ] are integers. As a corollary this implies a version of the no-go theorem:
for total chirality ±1 it is impossible to find chiral projections Pψ/ψ̄ with Pψ̄ = 1 −Pψ . The
theorem proven here applies to even-dimensional Euclidean lattices; in particular it covers
the case of 4-dimensional Euclidean lattices instead of the 3-dimensional spatial lattices
considered in [4,5,7]. Ref. [6] deals with 4-dimensional lattices but restricts the form of the
action rather strongly and does not allow for momentum dependent chiral projections. The
latter, in particular, has important consequences. A first account of the present work was
given in [13].
The paper is organised as follows. In Section 2 we state the theorem and some
corollaries. We also discuss the necessity and implications of certain properties imposed
on the Dirac operator and the projections. In Section 3 the winding numbers n[Pψ/ψ̄ ]
are evaluated and the theorem is proven. We close with a brief discussion of our
findings. Some technical details are deferred to the appendices together with an example
which highlights the difference between 3-dimensional (Hamiltonian) and 4-dimensional
(Euclidean) lattices.
O. Jahn, J.M. Pawlowski / Nuclear Physics B 642 (2002) 357–371 359
2. Theorem
We state the theorem in Section 2.1 and discuss its implications in Section 2.2.
2.1. Setting and theorem
We consider free, massless fermions on an infinite d-dimensional hyper-cubic lattice

Λ with lattice spacing a, Λ = {nµ a | nµ ∈ Z}, where d = 2l is even. At each lattice site,
two M-component spinors ψ(x), ψ̄(x) ∈ CM are defined. The action is given in terms of
a translationally invariant Dirac operator D(x − y) ∈ CM×M ,

S= ψ̄(x)D(x − y)ψ(y). (1)
x,y∈Λ
The Fourier transform of D,

D(k) ≡ e−ik·x D(x), (2)
x∈Λ
is periodic with periods 2π/a on R2l and can therefore be considered as a function from
the 2l-torus T 2l to CM×M . The spinors ψ, ψ̄ live in the spaces defined by the constraints
Pψ (k)ψ(k) = ψ(k) and ψ̄(k)Pψ̄ (k) = ψ̄(k), (3)
with translationally invariant Hermitean projection operators Pψ/ψ̄ = Pψ/ 2 . This includes
ψ̄
the trivial case Pψ̄ = Pψ = 1. For a theory with chiral fermions, the projections Pψ/ψ̄ are
necessarily non-trivial, as one has to remove part of the degrees of freedom. We would also
like to emphasise that Pψ and Pψ̄ can be different since in a Euclidean theory ψ and ψ̄ are
independent fields. We will prove the following:
Theorem. Given a lattice theory in 2l dimensions with the action (1) and projections Pψ̄ ,
Pψ as in (3) , where the Dirac operator D and the projection operators Pψ/ψ̄ have the
properties (i)–(iii):
(i) Locality: |Dij (x)|, |Pψ̄ij (x)|, |Pψij (x)| < ce−|x|/λ for some real constants c, λ.
This implies that D(k) and Pψ/ψ̄ (k) are analytic in kµ in a strip around the real
axis;
(ii) Spin- 21 zeros: the real poles k (i) of the propagator D(k)−1 have the form
(i)
(kµ − kµ ) (i)†
D(k)−1 = Σ + finite, (4a)
|k − k (i) |2 µ
Σµ(i)† Σν(i) + Σν(i)† Σµ(i) = 2δµν Πψ(i) , with Πψ(i) = Πψ(i) 2 = Πψ(i)† ,
Σµ(i) Σν(i)† + Σν(i) Σµ(i)† = 2δµν Πψ̄(i) , with Πψ̄(i) = Πψ̄(i) 2 = Πψ̄(i)† ; (4b)
(iii) Compatibility of chiral projections: the chiral projection operators and the Dirac
operator satisfy
DPψ = Pψ̄ D. (5)
Then the total chirality of all fermion species in the continuum limit is given by
χ = n[Pψ ] − n[Pψ̄ ], (6)
where

1 i l
n[P ] ≡ tr P (dP )2l ∈ Z. (7)
l! 2π
T 2l
Moreover n[P ] is a topological invariant with integer values for local, translationally
invariant, Hermitean projection operators P on the Fourier space T 2l .
The theorem implies that
(1) A non-zero total chirality is only possible with non-trivial projections (Pψ = 1 or
Pψ̄ = 1 or both);
(2) An odd total chirality is only possible if the spaces onto which Pψ̄ and Pψ project
are neither identical nor orthogonal complements, i.e., Pψ̄ = Pψ and Pψ̄ = 1 − Pψ ,
because otherwise n[Pψ̄ ] = n[Pψ ] or n[Pψ̄ ] = −n[Pψ ]. The spaces onto which they
project (and their orthogonal complements) are in fact inequivalent fibre bundles over
T 2l .
The theorem does not exclude even non-zero chirality χ = 2n[Pψ ] if Pψ̄ = 1 − Pψ . This
is in contrast to the situation for 3-dimensional spatial lattices, where only zero chirality
is possible [4–7]. An example with χ = 4 in 2 dimensions (and χ = 16 in 4 dimensions)
presented in Appendix A shows, that this is realised.
GW fermions [8] are included in (5) as they are defined with
{γ2l+1 , D} = aDγ2l+1D, where γ2l+1 = il γ1 · · · γ2l , with γ2l+1
2
= 1. (8)
Thus, they admit the (non-unique) definition of Pψ/ψ̄ with Pψ = 12 (1 − γ2l+1 ) and
Pψ̄ = 12 (1 + γ2l+1 (1 − aD)). One can prove with (8) that Pψ/ψ̄ satisfy (5). Moreover,
Pψ2 = Pψ and Pψ̄2 = Pψ̄ . A similar analysis applies to Dirac operators satisfying a recently
discussed generalisation of the GW relation [14].
2.2. Necessity and implications of the properties (i)–(iii)
Before proving the theorem in the next section, we first would like to elaborate a bit on
the properties (i)–(iii).
Locality (i) of D and the chiral projections guarantees that the continuum limit of the
lattice theory does not depend on the details of the discretisation. In a local theory, only the
behaviour at the zeros of D matters in the continuum limit.
The structure of the zeros (ii) is determined by the requirement that the fields in the
continuum limit carry spin- 21 representations of the Euclidean group. A pole of D −1 of the
form (4) gives rise to a continuum action density ψ̄k · Σψ (we suppress the superscript (i)
on Σ). Πψ and Πψ̄ project onto the right and left eigenspaces with vanishing eigenvalues
of D, i.e., onto those components of ψ and ψ̄ that survive the continuum limit. Like
the γ -matrices, the Σµ define spin- 21 representations of the rotation group SO(2l). The
fermions ψ̄, ψ live in the two different representations generated by
1 ψ 1 1 ψ̄ 1
Σµν ≡ Σµ† Σν − Σν† Σµ , Σµν ≡ Σµ Σν† − Σν Σµ† . (9)
2 4 2 4
The matrix Σµ couples the representations Σ ψ̄ and Σ ψ to a vector. Therefore, the
continuum action ψ̄k · Σψ is rotationally invariant. In turn, fermions in the spin
representations (9) and the requirement of the correct continuum limit lead to a pole
structure of D −1 as in (4).
It follows from (4b) that we can write Σ more explicitly as
n−

Σµ(i) =U (i)
diag σµ , . . . , σµ , σµ† , . . . , σµ† V (i)† , (10)

n+
where the σµ and σµ† form right- and left-handed 2l−1 -dimensional irreducible represen-
tations of (4b) without projections. These are unique up to bi-unitary transformations.
Here, right-handed means il σ1† σ2 · · · σ2l−1†
σ2l = +1, which implies that σ † is left-handed,
il σ1 σ2† · · · σ2l−1 σ2l† = −1. For 2l = 4, one can choose σ4 = 1 and σi = iτi where τi are the
Pauli matrices. We also have
(i) (i)
Πψ̄ = U (i) U (i)† , Πψ = V (i) V (i)† , with
V (i)†
V (i)
=U (i)†
U (i)
= 12l−1 (n+ +n− ) . (11)
Thus, a pole of the form (4) gives rise to n+ right- and n− left-handed fermions in the
(i)
continuum limit, where n+ and n− are the number of σµ and σµ† in Σµ , respectively.
The corresponding components of ψ and ψ̄ , respectively, are obtained as eigenspaces for
eigenvalues ±1 of the chirality operators
Γψ(i) ≡ il Σ1(i)† Σ2(i) · · · Σ2l−1
(i)† (i)
Σ2l , Γψ̄(i) ≡ il Σ1(i) Σ2(i)† · · · Σ2l−1
(i) (i)†
Σ2l . (12)
(i)
The Γψ/ψ̄ are Hermitean and have eigenvalues ±1 and 0. Taking into account the
projection (3), the total chirality of all fermion species in the continuum limit is given
by
1 (i) (i)
χ = l−1 tr Pψ k Γψ . (13)
2
i
The possibility of vector-like (Dirac) zeros is contained in (ii): if the Σµ are Hermitean,
(4b) turns into the standard anti-commutation relations for γ -matrices in the image of
Πψ = Πψ̄ .
The compatibility condition (iii) for D and the chiral projections is, for local Pψ/ψ̄ ,
equivalent to the following compatibility condition for Pψ and the spin representations
Σ ψ(i) at the zeros k (i) ,
ψ(i)
Σµν , Pψ k (i) = 0, (14)
the form of Pψ̄ then follows from that of Pψ via (5). Hence the property (5) is, at its
root, only a constraint at the points k (i) . It is for this reason, that the proof of the theorem
boils down to the calculation of windings at these points. The relation (14) ensures that
the projected spinor Pψ ψ also transforms under SO(4); it contains complete irreducible
components of the representation Σ ψ(i) only. It also follows that

Πψ , Pψ k (i) = 0. (15)
To prove that (5) implies (14) and (15), first note that it implies Pψ D −1 = D −1 Pψ̄ . Since
Pψ/ψ̄ are analytic, we can expand in powers of q ≡ k − k (i) to get

Pψ k (i) Σν† = Σν† Pψ̄ k (i) , (16)
for all ν, where we have used that we are free to choose qµ = δµν |q| (we have suppressed
the superscript (i) on Σ again). Using (16) and its Hermitean conjugate one concludes that
Σµ† Σν Pψ (k (i) ) = Pψ (k (i) )Σµ† Σν . With (4b) and (9) this leads to (14) and (15).
Conversely, for any local Pψ satisfying Eq. (14), we can define Pψ̄ ≡ DPψ D −1 . Except
for the zeros of D, analyticity of Pψ̄ follows in the set where Pψ and D have this property.
Eqs. (14) and (4b) imply Pψ (k (i) )Σµ† = Pψ (k (i) )Σµ† Σν Σν† = Σµ† Σν Pψ (k (i) )Σν† for ν = µ
(no sum). As DΣµ = O(k), the poles of D −1 drop out and Pψ̄ is finite and analytic there
as well. It goes without saying that similar statements and relations like (14)–(16) follow
ψ̄
for Σµν , Πψ̄ , Pψ̄ (k (i) ).
Local projection operators Pψ/ψ̄ satisfying (5) are also relevant in non-chiral theories
where the constraints (3) are not needed: they can be used to define charges Qψ ≡ 1 − 2Pψ ,
Qψ̄ ≡ 2Pψ̄ − 1 and a ‘chiral’ symmetry
ψ → eiαQψ ψ, ψ̄ → ψ̄eiαQψ̄ . (17)

The existence of such a symmetry with local charges, however, implies the existence of
local projections only if their eigenvalues are non-zero and non-degenerate in the entire
Brillouin zone, for instance, if the charges are integer-valued. The theorem presented in
this paper applies to such a symmetry as well. It implies that a symmetry of this kind that
goes over to the standard chiral symmetry in the continuum limit necessarily acts on ψ and
ψ̄ in an asymmetric way (not as if ψ and ψ̄ were Dirac conjugates of each other). Note,
however, that symmetries with charges whose eigenvalues vanish somewhere can be useful
in non-chiral theories, e.g., [11,14]. The theorem does not make a statement about these
symmetries. Indeed, the symmetries used in [11,14], essentially, lead to projections that
satisfy Pψ̄ = 1 − Pψ as well as (5) although the image of Pψ contains only a single left-
handed fermion. These projections are discontinuous at some points in momentum space,
so they are not local.
(i)
The chirality of a fermion in the continuum limit is determined by Γψ/ψ̄ in (12) rather
than Pψ or Pψ̄ . The images of the latter may well contain fermions of different chirality.
Hence, our setting includes theories with any number of left- and right-handed fermions
(in particular vector-like theories for which we can set Pψ = Pψ̄ = 1). If only one chirality
(i)
is desired, one has to require that Γψ/ ψ̄
has a definite sign in the image of Pψ (k (i) ). This
is not necessary for our theorem, so we do not make this requirement. However, we still
use the term ‘chiral’ projections for Pψ/ψ̄ since they treat left- and right-handed fermions
differently in general.
(i)
Accordingly, also the charges Qψ/ψ̄ need not coincide with the chirality operator Γψ/ψ̄
(i) (i)
at the zeros of D. Eq. (14) only guarantees that Qψ (k ) and Γψ are simultaneously
(i)
diagonalisable, as are Qψ̄ (k (i) ) and Γψ̄ . Eq. (17) goes over to the standard chiral
(i)
symmetry in the continuum limit only if Γψ/ψ̄ has only the eigenvalues 1 and 0 in the
image of Pψ/ψ̄ (k (i) ).
We close the section with an explicit—and relevant—example, the overlap Dirac
operator [9] in four dimensions. It is a GW Dirac operator, see (8), and, for vanishing
gauge field, it is given by
aD = 1 − cos θ + iγ · p̂ sin θ, (18)
where p̂ = p/|p|, pµ (k) and θ (k) are periodic functions in momentum space and γµ a
(fixed) set of Dirac matrices. Local chiral projections can be defined as Pψ̄ = 12 (1 + γ5 )
and Pψ = 12 (1 − Qψ ) where [12]
Qψ = γ5 (1 − aD) = γ5 cos θ − iγ5 γ · p̂ sin θ. (19)
Regarding −iγ5 γµ and γ5 as basis vectors in a 5-dimensional space, Qψ takes values on

the unit sphere in this space. It follows from Eq. (7) that n[Pψ ] measures the degree, or
winding number, of the map Qψ : T 4 → S 4 . The degree can be expressed as the number
of times a fixed point on the target is taken, weighted by the orientation (the sign of the
Jacobian), provided the Jacobian does not vanish at the chosen point. We may choose the
point Qψ = γ5 , i.e., θ = 0. This corresponds to the zeros of D. If D has only a single zero,
Qψ has unit winding number. For the overlap operator this can be explicitly verified by
studying the functions θ and pµ . So we find n[Pψ ] = −1 and n[Pψ̄ ] = 0 and verify the
theorem (6). For general functions θ (k) and pµ (k), the orientation of a zero is given by
(minus) the winding number of p̂ : S 3 → S 3 around the zero. Since this coincides with the
definition of chirality in (13), the theorem holds true.
3. Proof of the theorem
First, in Section 3.1, we prove that the winding number n[P ] is an integer. Then, in
Section 3.2, we show that n[Pψ ] − n[Pψ̄ ] in (6) can be transformed into the formula for
the total chirality as given on the r.h.s. of (13).
3.1. The winding number n[P ]
The power of the theorem depends crucially on the fact that n[P ] is an integer for local
projection operators P (k). Hence, before tackling the proof of the relation (6) we argue that
n[P ] ∈ Z. This discussion will also shed some light on the interpretation of the invariant
n[P ]. To that end we express P in terms of an orthonormal basis Ψ = (ψ1 , . . . , ψN ) ∈
CM×N (where N = tr P is the rank of P ),
P ≡ Ψ Ψ †, with Ψ † Ψ = 1N . (20)
Then, a general basis is given by Ψ v ≡ Ψ v with v ∈ U (N). Since P is periodic in kµ , Ψ
in general satisfies the boundary conditions
2π
Ψ k + âµ = Ψ (k)uµ (k), with uµ (k) ∈ U (N) and âµ ν = δµν . (21)
a
The transition functions uµ defined in this way satisfy the cocycle conditions

uµ (k)uν k + âµ = uν (k)uµ k + âν . (22)
They define a U (N) fibre bundle over T 2l . If this fibre bundle is non-trivial, the eigenspace
of P does not admit a globally smooth basis. We will see that n[P ] measures (part of ) this
non-triviality.
To characterise the fibre bundle, we define the U (N) gauge potential and field strength
A = Ψ † dΨ, F = dA + A ∧ A. (23)
They obey the expected boundary conditions

A k + âµ = u†µ (k) A(k) + d uµ (k), (24)

F k + âµ = u†µ (k)F (k)uµ (k). (25)
It follows from (20) and (23) that, for even dimension d = 2l, the winding number n[P ] as
defined in (7) is given by the integral of the lth Chern character (cf. Appendix C):

n[P ] = chl (F ). (26)
In general, this is not an integer but only a multiple of 1/l!. For U (N) bundles over T 2l ,
however, it is an integer. This follows directly from the Atiyah–Singer index theorem.
There, the lth Chern character for a torus U (N)-bundle is shown to be identical to the
index of the related Dirac operator, which is an integer.
In the light of the discussion above there is a natural interpretation of (5) as a map
between inequivalent U (N)-bundles over the torus T 2l . For highlighting this fact and for
later use in Section 3.2 let us discuss this in more detail. On Tr2l ≡ T 2l \ {k (i) } we can write
Eq. (5) as
−1/2
Pψ̄ = ε(D)Pψ ε† (D), with ε(D) ≡ D D † D , (27)
ε(D) is unitary. In (27) we have used that [Pψ , (D † D)−1/2 ] = 0, which follows directly
from (5) and its Hermitean conjugate. Two orthonormal bases Ψψ̄ of Pψ̄ and Ψψ of Pψ are
therefore related by
Ψψ̄ = ε(D)Ψψ g, with g = Ψψ† ε† (D)Ψψ̄ ∈ U (N). (28)
The function g is continuous except at the zeros k (i) of D where ε(D) is ill-defined. It is
not periodic but satisfies the boundary conditions

g k + âµ = uψ† ψ̄
µ (k)g(k)uµ (k), (29)
ψ ψ̄
where uµ and uµ are the transition functions of Ψψ and Ψψ̄ . These are thus related by

µ (k) = g (k)uµ (k)g k + âµ .
uψ̄ † ψ
(30)
Hence, if g (restricted to the boundary) carries non-trivial topology, the two sets of
transition functions uψ/ψ̄ define inequivalent U (N)-bundles. However, since g is smooth
outside of the zeros of D, its non-trivial content can be extracted from its windings at k (i) ,
which are introduced by ε(D).
3.2. Proof of Eq. (6)
The discussion in the previous section already suggests that the difference n[Pψ̄ ] −
n[Pψ ] is directly related to a homotopy class of the map ε(D) around the zeros of D.
Indeed, we shall see that it is given by

n[Pψ̄ ] − n[Pψ ] = νi [g], (31)
i
where

2l−1
νi [g] = lim b2l−1 tr g −1 dg , with
ε→0
k−k (i) =ε
l
(l − 1)! i
b2l−1 = (−1)l−1 , (32)
(2l − 1)! 2π
is the π2l−1 winding of g ∈ GL(N) about k (i) . The gauge function g is given in terms of
ε(D) by (28).
From (31) only a few technical steps have to be invoked in order to prove (6). For the
proof of (31) we resort to the fact that the difference of winding numbers is given by the
integral of the difference of Chern character chl , see (26):

n[Pψ̄ ] − n[Pψ ] = chl [FL ] − chl [FR ], (33)
where FL , FR are the field strengths of Aψ̄ = A[Pψ̄ ] and Aψ = A[Pψ ], respectively. The
integral over a Chern character is a topological invariant of the underlying fibre bundle. It
can therefore be calculated with any gauge field obeying the same boundary conditions,
in particular we can replace Aψ̄ by Ãψ̄ = Aψ ≡ g −1 (A + d)g. Now we use that chl
g
can (locally) be related to its Chern–Simons form cs2l−1 by chl [F ] = d cs2l−1 [A, F ].
Furthermore

cs2l−1 Ag , F g − cs2l−1 [A, F ] = cs2l−1 g −1 dg, 0 + dα2l−2 A, F, dg g −1 . (34)
Upon differentiation, the second term drops out. With cs2l−1 [A, 0] = b2l−1 tr A2l−1 (see,
e.g., [15, p. 400]) and integrating over the Brillouin zone I 2l ≡ [−π/a, π/a]4, we find
n[Pψ̄ ] − n[Pψ ] = ν[g], (35)

where

2l−1
ν[g] = b2l−1 tr g −1 dg , (36)
∂I 2l
is the total π2l−1 winding on ∂I 2l . Thus, (31) follows with the remark, that g is continuous
except for the zeros of D and the density of ν is closed. Then, ν[g] equals the sum of the
winding numbers of g: ν[g] = i νi [g].
It is left to relate νi [g] to the chirality χ . First note that in g = Ψψ† ε† (D)Ψψ̄ one can
replace Ψψ and Ψψ̄ by their values at k (i) and ε† (D) by D −1 ,

νi [g] = νi [gi ], with gi = Ψψ† k (i) D −1 Ψψ̄ k (i) . (37)
The first replacement is allowed because the bases Ψψ/ψ̄ can be chosen with finite
derivatives (at least locally) and higher-order terms drop out in Eq. (32) in the limit
ε→ √ 0; the second because the set of positive definite Hermitean matrices is contractible,
so D † D can be deformed to 1.
Now we employ the explicit form of D about its zeros k (i) . Using (5) and the cyclicity
of the trace we arrive at
2l−1 2l−1
tr gi−1 dgi = − tr Pψ k (i) J , where J ≡ D −1 dD. (38)
Intuitively one expects that only those parts of D and D −1 can contribute to νi [gi ] that
carry the Dirac structure q · Σ (i) and |q|−2 q · Σ (i)† , respectively (q ≡ k − k (i) ). Indeed we
find

νi [gi ] = − lim b2l−1 tr Pψ k (i) J0 2l−1 , (39)
ε→0
k−k (i) =ε
where
1
J0 ≡ q · Σ (i)† dq · Σ (i) . (40)
|q|2
As the derivation of (39) is a bit technical we defer it to Appendix B. For performing the
integration in (39) we use the symmetry properties of the integral. We infer from (40) that
1 l−1
J02l−1 = 2l (−1)l−1q · Σ (i)† dq · Σ (i) dq · Σ (i)† dq · Σ (i) + O(q · dq) ,
|q|
(41)
where the algebra (4b) has been used and O(q · dq) denotes terms containing q · dq as a
factor. These do not contribute when integrated over 3-spheres centered at q = 0. The first
term yields

il 1
lim b2l−1 J02l−1 = l−1 εµ ···µ Σ (i)† Σ (i) · · · Σµ(i)† Σ (i)
ε→0 2 (2l)! 1 2l µ1 µ2 2l−1 µ2l
k−k (i) =ε

1
= Γψ(i) . (42)
2l−1
By inserting (42) into (39) and the latter into (37), we obtain
1 (i) (i)
n[Pψ̄ ] − n[Pψ ] = − l−1 tr Pψ k Γψ = −χ. (43)
2
i
Thus n[Pψ ] − n[Pψ̄ ] is given by the total chirality of all fermion species appearing in the
eigenspace of Pψ . ✷
For the proof we have employed some cohomology theory, many readers might be
unfamiliar with. It is possible to avoid the use of (34) at the expense of tedious calculations.
The integrand of the difference n[Pψ̄ ] − n[Pψ ] is a total derivative dω2l−1 , which
can be explicitly calculated by using Pψ̄ = D −1 Pψ D on Tr2l = T 2l \ {k (i) }, following

from (5). Then n[Pψ̄ ] − n[Pψ ] = ∂T 2l ω2l−1 . In four dimensions (l = 2) it follows after a
r
straightforward, but rather lengthy calculation, that on Tr2l , the integrand in n[Pψ̄ ] − n[Pψ ]
can be written as
4
dω3 = tr DPψ D −1 d DPψ D −1 − tr Pψ (dPψ )4 , (44)
with

2
ω3 ≡ − tr J (J Pψ ) − (J Pψ ) − 2(J Pψ ) dPψ + J Pψ J dPψ − 2J Pψ (dPψ ) .
2 3 2 2
3
(45)

Then, one proceeds by exploiting the properties of J on ∂Tr = i {k | k − k = ε}
2l (i)
as discussed in Appendix B. Since Pψ is smooth and J has only linear singularities,

only the first two terms on the r.h.s. of (45) can contribute. On ∂Tr2l it follows that
ω3 |∂Tr2l = − 13 tr Pψ (k)J03 + O(q −2 ). Then one proceeds as from (39). We add that even
deriving (45) is quite tedious and it gets increasingly complicated when going to higher
dimensions. Moreover, within this approach the underlying topological structure gets
obscured.
4. Conclusions
We have investigated the topological obstructions to implementing chiral symmetry on

Euclidean lattices in general even dimensions. Our findings are summarised in Section 2
in terms of a theorem. Its setting allows for general local chiral projections and Dirac
operators, which, as a specific case, includes Ginsparg–Wilson fermions [8]. Within this
setting the total chirality χ is given by the difference of winding number n[Pψ ] and n[Pψ̄ ]:
χ = n[Pψ ] − n[Pψ̄ ].
Here, Pψ/ψ̄ are the projection operators defining the spaces of fermions: Pψ ψ =
ψ, ψ̄Pψ̄ = ψ̄ , see (3). The invariants n[P ] were shown to be integers.
This constitutes a generalisation of the Nielsen–Ninomiya no-go theorem. Let us briefly
recapitulate the setting and the consequences of the original theorem and its variants.
In [3–5,7] Hamiltonian (spatial) lattices and general symmetric projections are considered,
whereas [6] deals with Euclidean lattices and constant projections. Then, in both settings,
the no-go theorem states that a non-vanishing total chirality χ is excluded.
In turn, the present formulation of the theorem implies that the projections Pψ̄ and
1 − Pψ have to be topologically inequivalent for odd chirality. Hence, in this case
symmetric projections Pψ̄ = 1 − Pψ cannot be used. In particular this rules out symmetric
projections for one Weyl fermion: χ = 1.
Evidently, the realisation of an even number of only left handed fermions with
symmetric projections Pψ̄ = 1 − Pψ is not excluded by χ = n[Pψ ] − n[Pψ̄ ]. However, if
such a theory could be realised, the projections cannot be constant. For constant projections
both winding numbers n[Pψ ] and n[Pψ̄ ] vanish and the total chirality is zero, in accordance
with the no-go theorem of [6]. Consequently, this additional option, can only be realised for
momentum-dependent chiral projections. In Appendix A we present an example for such
a case. The example indicates that, as in the case of doubling modes, χ = 2d is needed if
one requires invariance under the discrete Euclidean group.
Acknowledgements
We thank H. Banerjee, C. Nash and V.V. Sreedhar for discussions. Most of this work
was done in collaboration with L. O’Raifeartaigh, who sadly passed away in November
2000. His enthusiasm, deep insight and more generally his humanity are greatly missed.
J.M.P. thanks the Dublin Institute for Advanced Studies for hospitality and financial
support. The work of O.J. has been supported by the Alexander von Humboldt Foundation
through a Feodor Lynen Research Fellowship (grant number V-3-FLF-1068701).
Appendix A. Example for χ = 0
In this appendix, we provide an example of a lattice theory with an even number of

fermions of the same chirality and Pψ̄ = 1 − Pψ . According to Ref. [7], this is not possible
on 3-dimensional, spatial lattices. It is not ruled out by (6) for even dimensional space–
time lattices, however, and the example shows that it can indeed occur. To keep expressions
simple, we actually present a 2-dimensional example and indicate how it can be generalised
to 4 dimensions. Our theorem also holds in 2 dimensions, where the winding number (7)
takes the form

i
n[P ] ≡ tr P (dP )2 , (A.1)
2π
T2
(i) (i) (i)† (i)

and the total chirality is given by χ = i tr[Pψ (k (i) )Γψ ] with Γψ = iΣ1 Σ2 . Now
consider the naïve Dirac operator
D = γµ sin kµ (A.2)
(we have set the lattice spacing a to 1) and define chiral projections Pψ ≡ 2 (1 − Qψ )
1
and
Pψ̄ ≡ 1 − Pψ with
γ3 cos k1 cos k2 + γ1 sin k2 − γ2 sin k1
Qψ ≡ , (A.3)
1 + sin2 k1 sin2 k2
with γ3 ≡ iγ1 γ2 . It easy to see that Q2ψ = 1 and Q†ψ = Qψ , so that Pψ/ψ̄ are projections,
and that DQψ = −Qψ D, so that Eq. (5) holds with Pψ̄ = 1 − Pψ . Furthermore Qψ is
analytic in a strip around real ki , so Pψ/ψ̄ are as well.
(i)
Near the zeros of D, kµ = iµ π with the multi-index iµ = 0, 1,

D = (−1)i1 γ1 q1 + (−1)i2 γ2 q2 + O q 2 , (A.4)
(i)
where qµ = kµ − kµ . We read off Σµ = (−1)iµ γµ and find
Γψ(i) = (−1)i1 +i2 γ3 . (A.5)

On the other hand, Eq. (A.3) gives

Qψ k (i) = (−1)i1 +i2 γ3 . (A.6)
So Pψ projects onto the left-handed component of ψ at all zeros of D. The total chirality
after chiral projection is χ = −4, the theory contains 4 species of left-handed fermions.
Eq. (A.3) can be generalised to 4 dimensions as follows: we put cos θ = cos k1 cos k2 ×
cos k3 cos k4 , so that sin θ still cancels the poles at sin kµ = 0. The fraction is the scalar
product of a unit vector tµ with γµ . In order for Qψ to anticommute with D, tµ
has to be orthogonal to the vector pµ = sin kµ . This can be achieved with the choice
t = (p2 , −p1 , p4 , −p3 )/|p|. The resulting projected theory contains 16 species of right-
handed fermions. Note that this construction is not possible in 3 dimensions: since t
depends only on p/|p| and is orthogonal to p, it can be considered as a vector field of
unit length on S 3 ; on S 2 , however, all vector fields vanish somewhere, so such a t cannot
exist. This argument does not exclude a different construction, of course.
The above example does not contradict Ref. [6]. There, the chiral charge Qψ is assumed
to be momentum independent.
Appendix B. Derivation of (39)
Here we present the technical details of the derivation of (39). In view of (32) and (37),
Eq. (39) is equivalent to

(i) 2l−1
lim tr Pψ k J = lim tr Pψ k (i) J02l−1 , (B.1)
ε→0 ε→0
k−k (i) =ε k−k (i) =ε
where J0 = |q|−2 q · Σ † dq · Σ as defined in (40). We have dropped the index (i) since we
focus on one zero of D. Only the singular pieces of J can contribute to the integral on the
l.h.s. of (B.1). For tracking them down we write D and D −1 in an expansion about k (i) as
given in (4a):
qµ †
D(k) = qµ Σµ + M(k), D(k)−1 = Σ + M(k), (B.2)
|q|2 µ
where we have put q ≡ k − k (i) and the matrices M and M have analytic entries. We are
only interested in the divergent term in J = |q|2 q · Σ dD + O(|q|0 ). With (4b) it follows
1 †
that Σ † = Πψ Σ † Πψ̄ . Moreover the projection operators Πψ/ψ̄ commute with Pψ (k (i) )
(see (15)). Hence for calculating the l.h.s. of (B.1) only the singular piece of J Πψ is
required. Consequently, we are only interested in the non-vanishing part of Πψ̄ dDΠψ . To
obtain it, consider

q · Σ = q · ΣD −1 DΠψ = Πψ̄ DΠψ + q · Σ MΠ ψ +q ·Σ , (B.3)
and
ψ + O(q).
q · Σ = DD −1 q · Σ = MΠ (B.4)
These imply Πψ̄ DΠψ = q · Σ + O(q 2 ), and we find
1
J Πψ = J0 + O |q|0 , with J0 ≡ q · Σ † dq · Σ. (B.5)
|q|2
Furthermore it follows from (16) that [J0 , Pψ (k (i) )] = 0. Thus, (B.1) follows.
Appendix C. Chern characters in terms of P
We show that the winding number n[P ] is given by the integrated Chern character of
the fibre bundle associated with P , see Eq. (26). To this end, we use
F = dΨ † ∧ dΨ + Ψ † dΨ ∧ Ψ † dΨ, (C.1)
and
2
(dP )2 Ψ = dΨ Ψ † + Ψ dΨ † Ψ = Ψ dΨ † ∧ 1 − Ψ Ψ † dΨ = Ψ F, (C.2)
to find

tr F l = tr Ψ † Ψ F l = tr Ψ † (dP )2l Ψ = tr P (dP )2l . (C.3)
The Chern characters can now be expressed as

1 iF l 1 i l
chl (F ) ≡ tr = tr P (dP )2l , (C.4)
l! 2π l! 2π
which coincides with the integrand in the definition (7) of n[P ].
References
[1] P. Hasenfratz, Nucl. Phys. Proc. Suppl. 106 (2002) 159, hep-lat/0111023.
[2] M. Creutz, Rev. Mod. Phys. 73 (2001) 119, hep-lat/0007032.
[3] L.H. Karsten, J. Smit, Nucl. Phys. B 183 (1981) 103.
[4] H.B. Nielsen, M. Ninomiya, Nucl. Phys. B 185 (1981) 20;
H.B. Nielsen, M. Ninomiya, Nucl. Phys. B 195 (1981) 541, Erratum.
[5] H.B. Nielsen, M. Ninomiya, Nucl. Phys. B 193 (1981) 173.
[6] L.H. Karsten, Phys. Lett. B 104 (1981) 315.
[7] D. Friedan, Commun. Math. Phys. 85 (1982) 481.
[8] P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
[9] H. Neuberger, Phys. Lett. B 417 (1998) 141, hep-lat/9707022;
H. Neuberger, Phys. Lett. B 427 (1998) 353, hep-lat/9801031.
[10] P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125, hep-lat/9801021.
[11] M. Lüscher, Phys. Lett. B 428 (1998) 342, hep-lat/9802011.
[12] M. Lüscher, Nucl. Phys. B 549 (1999) 295, hep-lat/9811032.
[13] L. O’Raifeartaigh, Nucl. Phys. Proc. Suppl. 101 (2001) 107.
[14] K. Fujikawa, Nucl. Phys. B 589 (2000) 487, hep-lat/0004012;
K. Fujikawa, M. Ishibashi, H. Suzuki, hep-lat/0202017.
[15] M. Nakahara, Geometry, Topology and Physics, Institute of Physics, 1990.
Worldline formalism in a gravitational background

Fiorenzo Bastianelli, Andrea Zirotti
Dipartimento di Fisica, Università di Bologna and INFN, Sezione di Bologna via Irnerio 46,
I-40126 Bologna, Italy
Received 11 June 2002; accepted 7 August 2002
Abstract
We analyze the worldline formalism in the presence of a gravitational background. In the worldline
formalism a path integral is used to quantize the worldline coordinates of the particles. Contrary to the
simpler cases of scalar and vector backgrounds, external gravity requires a precise definition of the
ultraviolet regularization of the path integral. Taking into account the UV regularization, we describe
the first quantized representation of the one-loop effective action for a scalar particle. We compute
explicitly the contribution to the graviton tadpole and self-energy to test the validity of the method.
The results obtained by usual field theoretical Feynman diagrams are reproduced in an efficient way.
Finally, we comment on the technical problems related to the factorization of the zero mode from the
path integral on the circle.
1. Introduction
The worldline path integral formulation of quantum field theory provides an alternative
and efficient method for computing Feynman diagrams. This method has quite a long
history [1]. More recently, it has been developed further by viewing it as the particle limit
of string theory [2] and discussed directly as the first quantization of point particles [3,4]
(see [5] for a review and a list of references). In all these developments a major problem was
the inclusion of gravity, even as a background. In fact, gravity generically decouples in a
naive particle limit of string theory, while in a direct worldline formulation the gravitational
background leads to a path integral which necessarily requires a detailed discussion of
ultraviolet regularizations. Nevertheless, much progress has been made in [6] where string
inspired rules were developed.
E-mail addresses: bastianelli@bo.infn.it (F. Bastianelli), zirotti@bo.infn.it (A. Zirotti).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 8 3 - 1
F. Bastianelli, A. Zirotti / Nuclear Physics B 642 (2002) 372–388 373
In this paper we address the use of the worldline formalism in the presence of
background gravity starting directly from the first quantization of point particles. We
consider for simplicity the case of the one-loop effective action for a scalar particle in a
gravitational background. We shall show that by properly taking into account an ultraviolet
regularization the gravitational background can successfully be included into the worldline
formalism.
The Euclidean one-loop effective action Γ [g] that we shall consider is the one obtained
by quantizing a Klein–Gordon field φ coupled to gravity

√ 1
S[φ, g] = d D x g g µν ∂µ φ ∂ν φ + m2 φ 2 + ξ Rφ 2 (1)
2

and formally reads (e−Γ [g] = Dφ e−S[φ,g] )
1
Γ [g] = Tr log −✷ + m2 + ξ R (2)
2
with ξ describing an additional non-minimal coupling.
This effective action can also be obtained by considering the first quantization of a scalar
point particle with coordinates x µ and action [7]
1
1 −1
S e, x µ = dτ e gµν (x)ẋ µ ẋ ν + e m2 + ξ R(x) , (3)
2
0
where e is an auxiliary einbein and ẋ µ = ∂τ x µ , and requiring that the worldline is a closed
loop (i.e., imposing periodic boundary conditions for all fields). By a standard gauge fixing
procedure [3] one can eliminate the einbein by the gauge condition e(τ ) = 2T (the factor
2 is conventional and it is used here to obtain standard QFT formulas), thus leaving an
integration over the proper time parameter T (the ghosts decouple and a factor T1 is due to
the presence of an isometry on the circle)
∞
1 dT
Dx e−Sgf [x ] ,
µ
Γ [g] = − (4)
2 T
0
where
1
1 2
Sgf x µ
= dτ gµν (x)ẋ ẋ + T m + ξ R(x) .
µ ν
(5)
4T
0
Now one is left with a path integration over the coordinates x µ . This is non-trivial as
the non-linear sigma model in (5) needs a regularization. Such non-linear sigma models
have been used previously to evaluate trace anomalies in 2, 4 and 6 dimensions [8–10],
and in that context three different regularizations have been analyzed: mode regularization
(MR) [8,11], time slicing (TS) [12], and dimensional regularization (DR) [13]. The DR
regularization was developed after the results of [14] which dealt with non-linear sigma
model in the infinite propagation time limit. All these regularizations require different
counterterms to produce the same physical results. The optimal choice for perturbative
374 F. Bastianelli, A. Zirotti / Nuclear Physics B 642 (2002) 372–388
calculations is the DR scheme which requires a coordinate invariant counterterm
1
∆SDR = dτ 2T VDR (6)
0
to be added to (5) with VDR = − 18 R. The MR and TS schemes on the contrary need
each non-covariant counterterms, VMR and VTS , that compensate the non-covariance of
the regularization procedure.
A second technical issue concerns the ways one treats a constant zero mode for the
path integral on the circle. One option was already used in trace anomalies calculations. It
µ
consists in first considering loops with a fixed base-point x0 in target space, and then
integrating over the position of that base-point. The coordinates x µ (τ ) have Dirichlet
µ
boundary conditions (DBC) x µ (0) = x µ (1) = x0 , so that the quantum fields y µ (τ ) =
µ µ
x (τ ) − x0 describe fluctuations around the background position x0 which must vanish at
µ
τ = 0, 1. These quantum fields have a kinetic term without zero modes and the propagators
can be derived immediately. This way of casting the path integral computation delivers a
covariant effective lagrangian density. A second option, sometimes called “string inspired”,
µ 1
consists in directly separating out the constant zero mode x0 = 0 dτ x µ (τ ) of the
µ
differential operator ∂τ2 on the circle. The fields y µ (τ ) = x µ (τ )−x0 are now defined on the
circle and thus satisfy periodic boundary conditions (PBC). The corresponding propagators
are periodic and translationally invariant. This set up is simpler than the first one since in
actual computations one can use translational invariance on the circle. However, it has the
disadvantage that it produces an effective lagrangian density with certain total derivative
terms which are non-covariant. This non-covariance invalidates any advantage of using
Riemann normal coordinates [15]. These two options are summarized in Figs. 1 and 2.
Other choices for treating the zero mode can be found in [5,16]. The total derivative terms
of the “string inspired” method are present not only in the gravitational case. They exist
also for standard field theories in flat space, including gauge theories, but in that case
they are not bothersome, since they do not violate gauge invariance. In fact, they are even
beneficial since their addition leads to a more compact form of the effective action [17].
In the following we shall consider the one-loop effective action as the generator of 1PI
graphs and we shall evaluate directly the terms obtained after functional differentiation.
In such a situation total derivative terms which may be present in the effective lagrangian
density are harmless. Thus the simplest way to set up the computations in the worldline
scheme is to use the “string inspired” method for separating out the zero mode together
with worldline dimensional regularization. This way we will obtain the correct contribution
Fig. 1. Dirichlet boundary con- Fig. 2. Periodic boundary

ditions at x0 (DBC). conditions without zero mode
(PBC).
to the tadpole and self-energy of the graviton, showing the correctness and efficiency of the
worldline formalism in this context.
As already mentioned, previous results in the string inspired framework for describing
theories coupled to gravity were presented in [6], and more recently in [18] where
perturbative formulations of certain gravitational theories have been systematically carried
out by using relations between open and closed string amplitudes. In the present paper we
only consider external gravity, but it would be quite interesting to extend our results by
learning how to include other particles as well as gravity itself in the loops, and possibly
making contact with the rules developed in [6,18].
The paper is organized as follows. In Section 2 we describe the worldline formalism
with background gravity. In Section 3 we present explicit computations of the one- and
two-point functions, which give the scalar particle contributions to the cosmological
constant and graviton self-energy, respectively. In Section 4 we test the correctness of
our results by first showing that the expected Ward identities are satisfied, and then by
comparing with a standard Feynman graph calculation. In Section 5 we discuss the total
derivative terms that one finds in computing directly the effective lagrangian density with
the PBC (“string inspired”) propagators. Finally we present our conclusion in Section 6
and put in an Appendix A our conventions and some useful formulae.
2. The worldline formalism
The worldline formalism for a scalar particle in a gravitational background leads us to

consider the following representation of the one-loop effective action
∞
1 dT
Dx e−Sgf [x ] ,
µ
Γ [g] = − (7)
2 T
0
where
1
1
Sgf x µ
= dτ gµν (x)ẋ µ ẋ ν + T m2 + ξ R(x) (8)
4T
0
and with the fields x µ (τ ) satisfying periodic boundary conditions at τ = 0, 1. Because

of derivative interactions present in this non-linear sigma model divergences may arise in
the quantum-mechanical loop corrections. Thus one needs a regularization. However, one
does not need infinite renormalization: the covariant path integral measure produces other
infinities that cancel the original ones [19]. This measure is of the form

Dx = Dx det gµν (x(τ )), (9)

0τ <1

where Dx = τ d D x(τ ) denotes the standard translationally invariant measure. It can be
represented more conveniently by introducing bosonic a µ and fermionic bµ , cµ ghosts
which satisfy the same periodic boundary conditions of the coordinates x µ [8,9]

Dx = Dx det gµν (x(τ )) = Dx Da Db Dc e−Sgh [x,a,b,c] , (10)
0τ <1
where
1
1
Sgh [x, a, b, c] = dτ gµν (x) a µ a ν + bµ cν . (11)
4T
0
After having selected a regularization scheme one may explicitly check that all divergences
cancel and only certain spurious finite terms are left over. The latter are compensated by a
finite counterterm VCT associated with the chosen regularization scheme.
Here we will adopt dimensional regularization which was shown in [13] to require the
covariant counterterm
1
∆SDR [x] = dτ 2T VDR (x) (12)
0
with VDR = − 18 R. In dimensional regularization, as developed in [13], one may proceed

as follows: (i) one extends the original compact space I = [0, 1] by adding d infinite
dimensions, (ii) uses partial integration in the regulated d + 1 dimensions to cast in simpler
forms the integrals arising in perturbation theory, (iii) computes those simpler forms by first
removing the regularization (i.e., sending d → 0) in case that no ambiguities are left over
at d = 0 [10,13]. This procedure frees one from the need of computing tricky integrals
at arbitrary complex d + 1 dimensions, as done instead in the usual QFT dimensional
regularization. In all the cases analyzed so far this recipe has been enough to compute the
required integrals.
Collecting all terms, the formula for the effective action in the worldline DR scheme is
given by
∞
1 dT
Γ [g] = − Dx Da Db Dc e−S (13)
2 T
0
with
1
1
S= dτ gµν ẋ µ ẋ ν + a µ a ν + bµ cν + T m2 + ξ̄ R , (14)
4T
0
where ξ̄ = ξ − 14 takes into account the DR counterterm.

This effective action can be used immediately to obtain (1PI) correlation functions by
varying the metric gµν and then setting gµν = δµν . Alternatively, one can obtain correlation
functions directly in momentum space. One considers the effective action as a power series
in hµν = gµν − δµν , substitutes the hN term with plane waves of definite polarizations

N
(i) ipi ·x
hµν (x) = "µν e (15)
i=1
(i)
and then picks us the terms linear in each "µν : this gives the contribution to the N -graviton
amplitude in momentum space Γ (p11 ,...,pNN ) (see notation in Eq. (39)).
" ,...,"
This way one is left with quantum mechanical correlation functions on the circle of the
form
µ1 ν1 µ ν µ ν
ẋ1 ẋ1 + a1 1 a1ν1 + b1 1 c1ν1 eip1 ·x1 · · · ẋNN ẋNN + aNN aNN + bNN cNN eipN ·xN ,
µ µ µ ν
(16)
where the fields x1 , a1 stand for x(τ1 ), a(τ1 ) and so on (this formula is exact for ξ̄ = 0, the
general case has additional contact terms due to vertices with multiple graviton legs arising
from the expansion of the ξ̄ R term).
On the circle the free kinetic term of the coordinates x µ has a zero mode. One option is
to split
µ

x µ (τ ) = x0 + y µ (τ ), y µ (τ ) = ynµ e2πinτ , (17)
n=0
µ
where x0 is the constant zero mode of the differential operator ∂τ2 and y µ (τ ) are the
quantum fluctuations. After inclusion of the quantum fluctuations one must integrate over
all possible zero modes, i.e., all possible positions of the particle loop in target space. Thus
the path integration is split as
1
Dx = D
d D x0 Dy. (18)
(4πT ) 2
The kinetic term for the quantum fields y µ is now invertible and the corresponding path
integral is normalized to unity
1
Dy e− 0 dτ 4T ẏ = 1.
1 2
(19)
The propagators are translationally invariant and read

µ
y (τ )y ν (σ ) = −2T δ µν ∆(τ − σ ),
µ
a (τ )a ν (σ ) = 2T δ µν ∆gh (τ − σ ),
µ
b (τ )cν (σ ) = −4T δ µν ∆gh (τ − σ ), (20)
where ∆ and ∆gh are given by
1 1 1 1
∆(τ − σ ) = − 2 2
e2πin(τ −σ ) = |τ − σ | − (τ − σ )2 − ,
4π n 2 2 12
n=0
∞
∆gh (τ − σ ) = e2πin(τ −σ ) = δ(τ − σ ). (21)
n=−∞
With these propagators one can compute the averages in (16) using the Wick theorem.
Note that integration over the zero mode d D x0 in (18) produces through the exponentials
in (16) a delta function
(2π)D δ D (p1 + · · · + pN ) (22)

enforcing momentum conservation. With momentum conservation the constant part of the
propagator ∆ drops out from (16) and may be set to zero. Thus instead of ∆ one may use
the effective propagator
1 1
∆0 (τ − σ ) = |τ − σ | − (τ − σ )2 (23)
2 2
which satisfy ∆0 (0) = 0.
We will apply and test this set up to compute one- and two-point functions in the next
section.
3. One- and two-point functions
We begin with the rather simple one-point function which gives the scalar particle
contribution to the cosmological constant in Fig. 3. Taking from the effective action the
term linear in hµν and substituting for hµν the expression (15) with just one plane wave
produces
∞
1 dT −m2 T 1 1
Γ (p)
"
= e D
d D x0 "µν
2 T (4πT ) 2 4T
0
1

× dτ ẏ µ ẏ ν + a µ a ν + bµ cν eip·(x0+y) . (24)
0
The integration over d D x0 gives the momentum delta function (2π)D δ D (p), which for
simplicity we factorize together with the polarization tensor "µν . Momentum conservation
eliminates the exponential and the remaining Wick contractions leave us with1
∞ 1
δ µν dT −m2 T 1
dτ •∆• (τ − τ ) + ∆gh (τ − τ ) .
µν
Γ(0) =− e D
(25)
4 T (4πT ) 2
0 0
The integrals of the propagators can be treated in DR, but it is immediately clear from
Eq. (21) that only the term from the ghost zero mode contributes
• •
∆ (τ − τ ) + ∆gh (τ − τ ) = 1. (26)
1 Here we consider ∆(τ, σ ) = ∆(τ − σ ) as function of two variables, and dots on the left/right denote
derivatives with respect to τ /σ . Later on we will also denote by ∆|τ evaluation at coinciding points σ = τ .
Fig. 3. Graviton tadpole. Fig. 4. Graviton self-energy.
Here we see explicitly the effect of the ghosts that eliminate potential divergences in
quantum mechanics. Now the integral over the proper time leads directly to a gamma
function if the target space dimension D is turned into a complex number (this is
dimensional regularization in target space and regulates the QFT ultraviolet divergences).
So we are left with the result
D
µν δ µν (m2 ) 2 D
Γ(0) = − Γ − . (27)
4 (4π) D2 2
Note that the terms linear in hµν and coming from the expansion of the scalar curvature
in (14) vanish at zero momentum. This tadpole diagram diverges at even dimensions D
and must be renormalized. Of course, one may keep D fixed and use instead a cut-off in
the proper time as an alternative regularization.
We now discuss the more interesting two-point function. Let us consider first the simpler
case with ξ̄ = 0 (i.e., ξ = 14 ). This is special since only vertices with one graviton are
present, see Fig. 4. The quadratic part in the metric fluctuations hµν are
∞
1 dT −m2 T 1
Γ (p
"1 ,"2
1 ,p2 )
=− e D
2 T (4πT ) 2
0
1 2
µ ν
1 1
× D
d x0 dτ hµν ẏ ẏ + a a + b c
µ ν µ ν
, (28)
2 4T
0 lin "1 ,"2
where
(1) ip1 ·x (2) ip2 ·x
hµν = "µν e + "µν e . (29)
As before the zero mode integration gives a delta function for momentum conservation,
which we factorize again for notational simplicity. Then a straightforward application of
the Wick theorem produces
∞
1 1 dT
e−m
"1 "2 2T
Γ(p,−p) =−
8 (4π) D2 T 1+ D
2
0

× r1 I1 + r2 I2 − 2Tp2 (r3 I3 − r4 I4 ) + 4T 2 p4 r5 I5 , (30)
(1) µναβ (2)
where p = p1 = −p2 and ri = "µν Ri "αβ with
µναβ
R1 = δ µν δ αβ ,
µναβ
R2 = δ µα δ νβ + δ µβ δ να ,
µναβ 1 µα ν β
R3 = 2
δ p p + δ να pµ pβ + δ µβ pν pα + δ νβ pµ pα ,
p
µναβ 1
R4 = 2 δ µν pα pβ + δ αβ pµ pν ,
p
µναβ 1
R5 = 4 pµ pν pα pβ (31)
p
while the integrals coming from the quantum mechanical correlation functions are given
by
1 1
dσ ( •∆• + ∆gh )|τ ( •∆• + ∆gh )|σ e−2Tp
2∆
I1 = dτ 0 ,
0 0
1 1

dσ •∆• 2 − ∆2gh e−2Tp ∆0 ,
2
I2 = dτ
0 0
1 1
dσ •∆ •∆• ∆• e−2Tp
2∆
I3 = dτ 0 ,
0 0
1 1
dσ ( •∆• + ∆gh )|τ (∆• )2 e−2Tp
2∆
I4 = dτ 0 ,
0 0
1 1
dσ (•∆)2 (∆• )2 e−2Tp
2∆
I5 = dτ 0 . (32)
0 0
Translational invariance can be used at once by fixing σ = 0. Then one obtains the
following results by using dimensional regularization when necessary
1
dτ e−Tp
2 (τ −τ 2 )
I1 = ,
0
1
I2 = Tp2 − 2 + I1 ,
4
1 1
I3 = − (1 − I1 ),
8 2Tp2
1
I4 = (1 − I1 ),
2Tp2
1 3
I5 = 2
− (1 − I1 ). (33)
8Tp 4T 2 p4
At this stage the proper time integral can be carried out at complex D and yields
Fig. 5. Additional graph for graviton self-energy.

D 1 D 2 D2
(4π) 2 Γ(p,−p) = − Γ − P (R1 + R2 − R3 − R4 + 3R5 )
8 2
D
− m2 2 (2R2 − R3 − R4 + 3R5 )

1 D 2 2 D2 −1
− Γ 1− p m (R2 − R3 + 2R5 ), (34)
32 2
(i)
where in our shorthand notation we have factorized the polarization tensors "µν , suppressed
tensor indices, and used the expression
1

2 x
x
P = dτ m2 + p2 τ − τ 2 . (35)
0
The additional terms ∆Γ(p,−p) present for the case ξ̄ = 0 correspond to Fig. 5 and can
be quickly derived by using the expansion of the scalar curvature reported in Eq. (A.3) of
Appendix A. They read

D ξ̄ D 2 2 D2 −1
(4π) 2 ∆Γ(p,−p) = − Γ 1 − p m (2R1 + R2 − R3 − 2R4 + 4R5 )
8 2
D −1
− 4 P 2 2 (R1 − R4 + R5 )

ξ̄ 2 D 4 2 D2 −2
− Γ 2− p P (R1 − R4 + R5 ). (36)
2 2
4. Ward identities and standard Feynman graphs
Ward identities follow from general coordinate invariance and can be used to test our
previous results. General coordinate invariance implies conservation of the induced stress
tensor
1 δΓ [g]
∇µ(x) √ = 0. (37)
g(x) δgµν (x)
We use the notation

δ n Γ [g]
µ ν1 ,...,µn νn
≡ Γ(x11,...,x (38)
δgµ1 ν1 (x1 ) · · · δgµn νn (xn ) gµν =δµν n)
and Fourier transform to momentum space by

Γ (p1 ,...,pn ) = (2π)D δ(p1 + · · · + pn )Γ(p1 ,...,pn )

= dx1 · · · dxn eip1 x1 +···+ipn xn Γ(x1 ,...,xn ) . (39)
Thus from (37) we get the following Ward identity relating the one- and two-point
functions
µν,αβ 1 µα µβ 1 αβ
pµ Γ(p,−p) + pµ δ νβ Γ(0) + δ να Γ(0) − pν Γ(0) = 0. (40)
2 2
It is immediate to verify from Eqs. (27), (34), (36) that this identity is satisfied for any
value of ξ .
One may note that the two-point function can be written in a more compact form which
makes it easier to check the Ward identity. Defining the tensors
S1 = R1 − R4 + R5 ,
S2 = R2 − R3 + 2R5 (41)
µναβ µναβ
which satisfy pµ S1 = pµ S2
= 0 allows one to write the full two-point function as

D 1 D 2 D2 D D
full
(4π) 2 Γ(p,−p) =− Γ − m (R1 − R2 ) + P 2 2 − m2 2 (S1 + S2 )
8 2

1 D 2 2 D2 −1
− Γ 1− p m S2
32 2

ξ̄ D 2 2 D2 −1 D −1
− Γ 1− p m (2S1 + S2 ) − 4 P 2 2 S1
8 2

ξ̄ 2 D 4 2 D2 −2
− Γ 2− p P S1 . (42)
2 2
The value ξ = 0 (ξ̄ = − 14 ) describes the result for a scalar with a minimal coupling.
(D−2)
A conformally coupled scalar needs instead the value ξ = 4(D−1) (i.e., ξ̄ = 4(1−D)
1
)
together with m2 = 0. Finally, the value ξ = 14 (ξ̄ = 0) allows for the simplest computation
in the worldline formalism as all possible vertices contain one graviton only: this is going
to be quite useful for the worldline description of spin 1/2 fermions.
To dispel any further doubt we have repeated the above calculations using the standard
Feynman rules obtained from the action in Eq. (1) and found the expected agreement.2
5. On the factorization of zero modes
One can repeat the previous worldline computation using instead the propagators
with Dirichlet boundary conditions for both coordinates and ghosts. The propagators are
different and, in particular, are not translationally invariant. As a consequence the integrals
arising from Wick contractions are more complicated and rather laborious to evaluate.
2 It is likely that the explicit result for the two-point function due to a scalar loop is present somewhere in the
literature, however, we have not found it.
These propagators are again given by

µ
y (τ )y ν (σ ) = −2T δ µν ∆(τ, σ ),
µ
a (τ )a ν (σ ) = 2T δ µν ∆gh (τ, σ ),
µ
b (τ )cν (σ ) = −4T δ µν ∆gh (τ, σ ) (43)
but with Green functions ∆ and ∆gh satisfying vanishing DBC
∞

2
∆(τ, σ ) = − 2 2 sin(πmτ ) sin(πmσ )
π m
m=1
= (τ − 1)σ θ (τ − σ ) + (σ − 1)τ θ (σ − τ ),
∞

∆gh (τ, σ ) = 2 sin(πmτ ) sin(πmσ ) = δ(τ, σ ), (44)
m=1
where θ (τ − σ ) is the standard step function and δ(τ, σ ) is the Dirac’s delta function
vanishing at the boundaries. Note again that these functions are not translationally
invariant. Moreover the values of ∆ and •∆ at coinciding points (which we denote by
∆|τ and •∆|τ ) are non-vanishing and, in fact, not even constant. One must keep track of
them in the integrals obtained after Wick contractions. For example, the complete form of
the integrals written in Eq. (32) for the two-point function in the PBC method are given in
general by
1 1
dσ ( •∆• + ∆gh )|τ ( •∆• + ∆gh )|σ eTp
2 (∆| +∆| −2∆)
I1 = dτ τ σ
,
0 0
1 1
2
I2 = dτ dσ •∆• 2 − ∆2gh eTp (∆|τ +∆|σ −2∆) ,
0 0
1 1
dσ (•∆ −• ∆|τ ) •∆• (∆• − ∆• |σ )eTp
2 (∆| +∆| −2∆)
I3 = dτ τ σ
,
0 0
1 1
dσ ( •∆• + ∆gh )|τ (∆• − ∆• |σ )2 eTp
2 (∆| +∆| −2∆)
I4 = dτ τ σ
,
0 0
1 1
dσ (•∆ −• ∆|τ )2 (∆• − ∆• |σ )2 eTp
2 (∆| +∆| −2∆)
I5 = dτ τ σ
. (45)
0 0
Note that the expressions in (32) are immediately recovered when using the properties
of the PBC propagators. Dimensional regularization can still be used to compute these
integrals, but their values with the DBC propagators differs from the ones reported in (33).
This is correct, as there are additional terms that must be included. In fact, the ghosts have
vanishing boundary conditions at τ = 0, 1 and so they cannot create the covariant measure
√ µ
g(x0 ) for the integration over the base-point x0 . This factor must appear directly in the
path integral measure
1
Dx = D
d D x0 g(x0 ) Dy (46)
(4πT ) 2
and generates additional terms (compare with √ Eq. (18)). Suffice here to mention that the
extra contributions coming
√ from the factor g(x0 ) (there
√ are quadratic terms in hµν from
the direct expansion of g(x0 ) and a cross term from g(x0 ) and the action) are essential
to recover the final answer reported in Eq. (42). It is quite simple to √ verify this for the
tadpole. The full contribution comes from the expansion of the factor g(x0 ), while the
remaining term from the path integral vanishes since
1 1
• •
dτ ( ∆ + ∆gh )|τ = dτ ∂τ (∆• |τ ) = 0 (47)
0 0
as one verifies from Eq. (44) using dimensional regularization3 (compare this result with
Eq. (26)). As a consequence we conclude that the string inspired approach is more efficient
for computing correlation functions.
On the other hand, if one is interested in computing directly the effective action the
opposite is true. In fact, non-covariant total derivative terms which seem to arise in the
string inspired approach make the choice of Riemann normal coordinates not useful at
all [15]. To test these expectations we now compute the leading terms of the effective
action (in the proper time expansion) using both schemes, and exhibit the total derivative
term arising at the leading order. The computation is structurally the same as the one carried
out in the heat kernel approach by DeWitt [20]. The difference is that instead of solving
the heat equation with an ansatz to find recursive relations for the so-called Seeley–DeWitt
coefficients, we compute those coefficients directly with a worldline path integral.
We start from the effective action written as in Eq. (13). Using arbitrary coordinates we
get the following expansion
∞
dT e−m T
2
1
Γ [g] = − Z(T ), (48)
2 T (4πT ) D2
0
where Z(T ) is given by

1 2
Z(T ) = d x0 g(x0 ) 1 − S3 − S4 + S3 + · · ·
D
(49)
2
with the vertices
1
1 1
S3 = dτ ∂α gµν (x0 )y ẏ ẏ + a a + b c ,
α µ ν µ ν µ ν
T 4
0
3 This identity is valid also in mode regularization, but not in time slicing. It is easy to check that also these
other regularizations produce the correct value of the tadpole.
1
1 1
S4 = dτ ∂α ∂β gµν (x0 )y y ẏ ẏ + a a + b c + T ξ̄ R(x0 ) .
α β µ ν µ ν µ ν 2
(50)
T 8
0
Notice that we expand around the fixed point x0 (the “base-point” in the DBC method
and the “zero-mode point” in the PBC method) but keep the metric arbitrary. Thus all
propagators carry the inverse metric evaluated at that point
µ
y (τ )y ν (σ ) = −2T g µν (x0 )∆(τ, σ ) (51)
and similarly for the ghosts.
We use a short hand notation for the various tensor structures appearing after the Wick
contractions (see appendix). We find that S3 vanishes as it contains an odd number of
fields and
T
−S4 = − A1 g µν ∂ 2 gµν + 2A2 ∂ µ ∂ ν gµν − T ξ̄ R,
2
1 2 T 2
S3 = − B1 g µν ∂α gµν + 4B2 g µν ∂ α gµν ∂ β gβα + 2B3 (∂α gµν )2
2 4
2
+ 4B4 ∂ µ g να (∂ν gµα ) + 4B5 ∂ µ gµν (52)
with all tensor structures evaluated at x0 , and with

1
A1 = dτ ∆|τ ( •∆• + ∆gh )|τ ,
0
1
A2 = dτ •∆2 |τ ,
0
1 1
B1 = dτ dσ ( •∆• + ∆gh )|τ ∆( •∆• + ∆gh )|σ ,
0 0
1 1
B2 = dτ dσ ( •∆• + ∆gh )|τ ∆• ∆• |σ ,
0 0
1 1

B3 = dτ dσ ∆ •∆• 2 − ∆2gh ,
0 0
1 1
B4 = dτ dσ ∆• •∆ •∆• ,
0 0
1 1
B5 = dτ dσ ∆• |τ •∆• ∆• |σ . (53)
0 0
Using the DBC propagators in Eq. (44) together with dimensional regularization gives
the following values
1 1 1 1
A1 = − , A2 = , B1 = − , B2 = ,
6 12 12 12
1 1 1
B3 = , B4 = − , B5 = − (54)
8 24 12
so that with the help of formula (A.2) in the Appendix A one can cast the final result as

√ 1
ZDBC (T ) = d D x g 1 − T + ξ̄ R + O T 2 . (55)
12
On the other hand, using the PBC propagators in Eq. (21) gives
1
A1 = − , A2 = 0, B1 = 0, B2 = 0,
12
1 1
B3 = , B4 = , B5 = 0 (56)
24 24
and inserting these values into (52) produces

√ √ 1 T √
ZPBC (T ) = d D x g − T g + ξ̄ R + ∂µ gg αβ Γαβ µ + O T 2 .
12 12
(57)
We see explicitly the total derivative term appearing at this perturbative order in the PBC
method. It is manifestly non-covariant and reads

T √
∆Z(T ) = d D x ∂µ g g αβ Γαβ µ + O T 2 . (58)
12
It implies that Riemann normal coordinates would not be useful to simplify the calculations
in the PBC (“string inspired”) method: one would not know how to reconstruct the final
expression in arbitrary coordinates. We interpret this total derivative term as an infrared
1
effect due to the non-local constraint 0 dτ y µ (τ ) = 0 imposed on the quantum fields. Of
course, the same total derivative term is found using mode regularization or time slicing.4
6. Conclusions
We have discussed the worldline formalism for a scalar particle coupled to a

gravitational background. Using an ultraviolet regularization we have shown how the
results expected from QFT follow unambiguously. We have seen that the easiest way
to proceed for calculating correlation functions at one-loop is to use: (i) periodic
boundary conditions with factorization of the zero mode and (ii) worldline dimensional
regularization. With these prescriptions we have computed the scalar particle contribution
4 In fact one of us (FB) originally computed this term in a discussion with Christian Schubert using only MR
and TS regularizations.
to the graviton tadpole and self-energy. We have also seen explicitly that the PBC method
suffers from non-covariant total derivative terms arising when one wants to compute the
effective action directly. One may hope that future improvements will show a way of using
the PBC method for effective action calculations, but at present the secure path is to use
the DBC method if one wants to employ the simplifying properties of Riemann normal
coordinates.
We expect that the worldline formalism with gravity can be extended to include other
types of particles in the loop. More ambitiously one would like to establish a connection
with the string inspired rules of Refs. [6,18].
Acknowledgement
We thank Christian Schubert for discussions and comments.
Appendix A
We use the following conventions for the curvature tensors

[∇µ , ∇ν ]V λ = Rµν λ ρ V ρ , Rµν = Rλµ λ ν , R = R µ µ > 0 on spheres.
(A.1)
In a self-evident, short-hand notation the scalar curvature is given by the sum of the
following 7 terms
3 1 1 2
R = ∂ µ ∂ ν gµν − g µν ∂ 2 gµν + (∂α gµν )2 − ∂ µ g να (∂ν gµα ) − g µν ∂α gµν
4 2 4
2
+ g µν ∂ α gµν ∂ β gβα − ∂ µ gµν . (A.2)
The linear and quadratic terms in the expansion around flat space read as
R = R (1) + R (2) + · · · ,
R (1) = ∂ µ ∂ ν hµν − ✷h,
2
1 1 3
R (2) = hµν ✷hµν − 2hµν ∂µ ∂ α hαν − ∂ν h − ∂ α hαµ − ∂µ h + (∂α hµν )2
2 2 4
1
− (∂µ hνα ) ∂ ν hµα , (A.3)
2
where hµν = gµν − δµν and h = δ µν hµν .
References
[1] R.P. Feynman, Phys. Rev. 80 (1950) 440;

J.S. Schwinger, Phys. Rev. 82 (1951) 664;
Y. Nambu, Prog. Theor. Phys. 5 (1950) 82;
V. Fock, Sov. Phys. J. 12 (1937) 404.
[2] Z. Bern, D.A. Kosower, Phys. Rev. Lett. 66 (1991) 1669;

Z. Bern, D.A. Kosower, Nucl. Phys. B 379 (1992) 451;
Z. Bern, L.J. Dixon, D.A. Kosower, Phys. Rev. Lett. 70 (1993) 2677, hep-ph/9302280.
[3] A.M. Polyakov, Gauge Fields and Strings, Harwood Academic, Chur, 1987.
[4] M.J. Strassler, Nucl. Phys. B 385 (1992) 145, hep-th/9205205;
M.G. Schmidt, C. Schubert, Phys. Lett. B 318 (1993) 438, hep-th/9309055;
M.G. Schmidt, C. Schubert, Phys. Lett. B 331 (1994) 69, hep-th/9403158;
E. D’Hoker, D.G. Gagne, Nucl. Phys. B 467 (1996) 272, hep-th/9508131;
E. D’Hoker, D.G. Gagne, Nucl. Phys. B 467 (1996) 297, hep-th/9512080.
[5] C. Schubert, Phys. Rep. 355 (2001) 73, hep-th/0101036.
[6] Z. Bern, D.C. Dunbar, T. Shimada, Phys. Lett. B 312 (1993) 277, hep-th/9307001;
D.C. Dunbar, P.S. Norridge, Nucl. Phys. B 433 (1995) 181, hep-th/9408014.
[7] L. Brink, S. Deser, B. Zumino, P. Di Vecchia, P. Howe, Phys. Lett. B 64 (1976) 435.
[8] F. Bastianelli, Nucl. Phys. B 376 (1992) 113, hep-th/9112035.
[9] F. Bastianelli, P. van Nieuwenhuizen, Nucl. Phys. B 389 (1993) 53, hep-th/9208059.
[10] F. Bastianelli, O. Corradini, Phys. Rev. D 63 (2001) 065005, hep-th/0010118;
F. Bastianelli, N.D. Hari Dass, Phys. Rev. D 64 (2001) 047701, hep-th/0104234.
[11] F. Bastianelli, K. Schalm, P. van Nieuwenhuizen, Phys. Rev. D 58 (1998) 044002, hep-th/9801105;
F. Bastianelli, O. Corradini, Phys. Rev. D 60 (1999) 044014, hep-th/9810119.
[12] J. de Boer, B. Peeters, K. Skenderis, P. van Nieuwenhuizen, Nucl. Phys. B 446 (1995) 211, hep-th/9504097;
J. de Boer, B. Peeters, K. Skenderis, P. van Nieuwenhuizen, Nucl. Phys. B 459 (1996) 631, hep-th/9509158.
[13] F. Bastianelli, O. Corradini, P. van Nieuwenhuizen, Phys. Lett. B 494 (2000) 161, hep-th/0008045.
[14] H. Kleinert, A. Chervyakov, Phys. Lett. B 464 (1999) 257, hep-th/9906156;
F. Bastianelli, O. Corradini, P. van Nieuwenhuizen, Phys. Lett. B 490 (2000) 154, hep-th/0007105.
[15] K. Schalm, P. van Nieuwenhuizen, Phys. Lett. B 446 (1999) 247, hep-th/9810115.
[16] E.S. Fradkin, A.A. Tseytlin, Phys. Lett. B 158 (1985) 316;
E.S. Fradkin, A.A. Tseytlin, Nucl. Phys. B 261 (1985) 1.
[17] D. Fliegner, P. Haberl, M.G. Schmidt, C. Schubert, Ann. Phys. 264 (1998) 51, hep-th/9707189.
[18] Z. Bern, L.J. Dixon, D.C. Dunbar, M. Perelstein, J.S. Rozowsky, Nucl. Phys. B 530 (1998) 401, hep-
th/9802162;
Z. Bern, in: H.E. Haber (Ed.), Proc. of the 5th International Symposium on Radiative Corrections (RADCOR
2000), hep-th/0102186.
[19] T.D. Lee, C.N. Yang, Phys. Rev. 128 (1962) 885;
E.S. Abers, B.W. Lee, Phys. Rep. 9 (1973) 1.
[20] B.S. DeWitt, Relativity, Groups and Topology, in: B. DeWitt, C. DeWitt (Eds.), Lectures at Les Houches
1963, Gordon & Breach, New York, 1964;
B.S. DeWitt, Relativity, Groups and Topology II, in: B. DeWitt, R. Stora (Eds.), Lectures at Les Houches
1983, North Holland, Amsterdam, 1984.
pp-wave/Yang–Mills correspondence:
an explicit check
Youngjai Kiem a , Yoonbai Kim b , Sangmin Lee b , Jaemo Park c
a Department of Physics, KAIST, Taejon 305-701, South Korea
b BK21 Physics Research Division and Institute of Basic Science Sungkyunkwan University,
Suwon 440-746, South Korea
c Department of Physics, POSTECH, Pohang 790-784, South Korea
Abstract
We present an explicit evidence that shows the correspondence between the type IIB supergravity
in the pp-wave background and its dual supersymmetric Yang–Mills theory at the interaction level.
We first construct the cubic term of the light-cone interaction Hamiltonian for the dilaton-axion sector
of the supergravity. Our result agrees with the corresponding part of the light-cone string field theory
(SFT) and furthermore fixes its previously undetermined p+ -dependent normalization. Adopting
thus fixed light-cone SFT, we compute the matrix elements of light-cone Hamiltonian involving
three chiral primary states and find an agreement with a prediction from the dual Yang–Mills theory.
1. Introduction
While the duality between type IIB string theory in a pp-wave background [1–4] and a
sector of N = 4 supersymmetric Yang–Mills theory with large R charge [5] is in a sense a
part of the AdS5 /CFT 4 correspondence, its many novel features make one to approach it
from a different angle. The authors of Ref. [5] succeeded in reproducing the string spectrum
[6] from perturbative Yang–Mills theory, thereby putting the duality on a firm ground at
the free theory level. Subsequent papers made some progress on the important question of
E-mail addresses: ykiem@muon.kaist.ac.kr (Y. Kiem), yoonbai@skku.ac.kr (Y. Kim),

sangmin@newton.skku.ac.kr (S. Lee), jaemo@physics.postech.ac.kr (J. Park).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 7 5 - 2
390 Y. Kiem et al. / Nuclear Physics B 642 (2002) 389–404
string interactions both on the Yang–Mills theory side [7–10] and on the string theory side
[11].1
String spectrum in the pp-wave background can be obtained from that of the full AdS, at
least in principle, by simply taking a limit. However, the dictionary between AdS and CFT
at the interaction level is seemingly lost. In the AdS/CFT correspondence, the correlators of
the CFT are recovered from string theory (supergravity in practice) by using boundary-to-
bulk propagators and bulk supergravity interaction vertices. The pp-wave limit magnifies
the neighborhood of a null geodesic and pushes away the original boundary, so that after
taking the limit the notion of the bulk-to-boundary propagator becomes unclear.
Without clear understanding of how holography is realized for the pp-wave case (see
however Refs. [8,26,33,35]), one should first specify what physical quantities can be
computed and compared between the two theories. The string theory in the pp-wave
background takes the simplest form in the light-cone gauge. As shown in Ref. [5], the light-
cone Hamiltonian is related to the AdS variables as Hl.c. = µ(∆ − J ) where ∆ is the AdS
energy and J is a component of angular momentum along S 5 . In the language of Yang–
Mills defined on S 3 × R, ∆ is again energy and J is a U (1) R charge. Since J does not
change due to perturbation, the interaction Hamiltonians of the two theories are expected to
be the same once appropriate observables are identified. That is, if matrix elements among
well-defined states can be computed reliably in both theories, that will provide a non-trivial
check of the duality at the interaction level. Note that scattering amplitudes are not well-
defined in a pp-wave since the strings are confined by an effective gravitational potential;
we are solving a quantum mechanics for ‘particles in a box’. Note also that the quantum
mechanical approach proved useful already in recovering the string spectrum from Yang–
Mills theory [5].
In Ref. [11] the cubic term of the interaction Hamiltonian of light-cone string field
theory in a pp-wave background was determined up to an overall function of light-cone
momenta p+ . Once one fixes the normalization, one can in principle compute arbitrary
matrix elements among three single particle states. On the other hand, in the course of
computing the second order correction to the energy spectrum in the Yang–Mills theory,
the authors of Ref. [10] made a concrete proposal for a set of matrix elements.
In this paper, we make an attempt to build a bridge between the two computations.
Specifically, we fix the normalization of the cubic Hamiltonian of Ref. [11] by comparing
it with a supergravity calculation. We then use it to compute the matrix elements of three
chiral primary states and find that the result agrees with the prediction of Ref. [10]. Strictly
speaking, the supergravity analysis in the present paper is valid for µp+ α 1 while the
Yang–Mills analysis of Ref. [10] is valid for µp+ α 1. However, our experience in
AdS/CFT [54] tells us that the cubic interactions involving chiral primaries are strongly
protected by supersymmetry. It is plausible that the nearly chiral string states of Ref. [5]
show only a small deviation from their chiral cousins. In any case, confirmation of the
proposal of Ref. [10] for chiral primaries should be a first consistency check as a prelude to
a full-fledged comparison involving the string states of Ref. [5]. Verification of the proposal
for them will be reported elsewhere.
1 See Refs. [12–53] for related recent developments.

Y. Kiem et al. / Nuclear Physics B 642 (2002) 389–404 391
This paper is organized as follows. In Section 2, we carefully perform the light-cone

quantization of the dilaton-axion system of IIB supergravity in a pp-wave background. The
fact that the cubic Hamiltonian is nearly the same as the one in the flat spacetime enables
us to determine it uniquely using an argument partially based on Lorentz symmetry. In
Section 3, we compare the cubic Hamiltonian H3 of Ref. [11] with that of Section 2 and
fix the normalization of the latter. We then compute the matrix elements of H3 for three
chiral primary states, and compare them with the proposal of Ref. [10]. In Appendix A, we
present a computation of the same matrix elements in the full AdS space. We again find an
agreement, but the derivation is not fully faithful.
2. Light-cone quantization of IIB supergravity in a pp-wave background
One of the most efficient ways of constructing the light-cone Hamiltonian of superstring
theory is to solve the constraints from (super)symmetries, which uniquely fixes the
Hamiltonian in the flat background [55,56]. In pp-wave backgrounds whose symmetry
generators are inherited from those of AdS space [18,32], however, the counterparts
of the flat space generators J +− and J −I are not present. This absence leaves the
overall p+ -dependent factor of the interaction Hamiltonian undetermined, barring us from
performing the comparison to the corresponding proposals based on Yang–Mills theory.
It thus appears necessary to directly construct the light-cone Hamiltonian of the pp-wave
supergravity from the known covariant action and to compare it with the zero mode part of
the string interaction Hamiltonian.
We begin with the determination of the cubic interaction Hamiltonian of the IIB
supergravity in a pp-wave background. The final answer that we get should be the same as
the one obtained from the light-cone string field theory computations after fixing its overall
p+ -dependent normalization. For this purpose, it is enough to work out the simplest non-
trivial case, namely, the dilaton-axion system. In the Einstein frame, the bosonic action of
type IIB supergravity with manifest SL(2, R) invariance is given as follows:

1 10 √ ∂µ τ ∂ µ τ̄ Mij i j
SIIB = 2 d x −g R − 2
− F3 · F3
2κ 2(Im τ ) 2

1 2 ij j
− |F5 | − C4 ∧ F3i ∧ F3 , (2.1)
4 4
where

−φ 1 |τ |2 − Re τ H3
τ = χ + ie , Mij = , F3i = , (2.2)
Im τ − Re τ 1 F3
and
5 = F5 − 1 C2 ∧ H3 + 1 B2 ∧ F3 ,
F (2.3)
2 2
and the coupling constant is given by 2κ 2 = (2π)7 gs2 α 4 . As is well-known, the action
given in the above is incomplete, for the self-duality condition
5 = F
∗F 5
should be separately imposed. While this procedure is easy to perform on the classical
solutions, it is difficult to implement at the quantum level. Concentrating on the dilaton-
axion sector allows us to bypass this subtlety. Note that there is no coupling in (2.1)
between the dilaton-axion and F 5 .
Expanding (2.1) up to cubic orders for the dilaton-axion sector around τ = τ + i = i
(φ = 0, χ = 0) and setting other perturbations to be zero, we have

1 √ 1 i
S = 2 dx 10 −g − ∇µ τ ∇ µ τ̄ − (τ − τ̄ )∇µ τ ∇ µ τ̄ , (2.4)
2κ 2 2
√ √
where
√ we suppressed the prime in τ . Rescaling τ → 2 κτ and τ̄ → 2 κ τ̄ shows that
2 κ is the cubic coupling constant. The background metric g is set to that of the pp-waves
in IIB supergravity
2 2 √
ds 2 = gµν dx µ dx ν = −4 dx + dx − − µ2 x I x I dx + + dx I , −g = 2, (2.5)
where x + is the light-cone ‘time’ and I = 1, . . . , 8. Written explicitly, the action is

+ − I 1 1 1
S = dx dx dx ∂+ τ ∂− τ̄ + ∂− τ ∂+ τ̄ − µ2 x I x I ∂− τ ∂− τ̄ − ∂I τ ∂I τ̄
2 2 4

√ 1 1 1
+ i 2 κ(τ − τ̄ ) ∂+ τ ∂− τ̄ + ∂− τ ∂+ τ̄ − µ2 x I x I ∂− τ ∂− τ̄ − ∂I τ ∂I τ̄ .
2 2 4
(2.6)
√
One might try to construct the quantizable light-cone Hamiltonian from (2.6) treating 2 κ
as a perturbation expansion parameter. This approach, however, has a subtle problem. The
light-cone canonical momenta are computed to be
√
2Πτ = ∂− τ̄ + i 2 κ(τ − τ̄ )∂− τ̄ , (2.7)
√
2Πτ̄ = ∂− τ + i 2 κ(τ − τ̄ )∂− τ. (2.8)
The cubic interaction terms of supergravity action involve two derivatives. Due to them,
the canonical momentum Πτ gets the second, interaction-dependent term. This behavior
greatly complicates the imposition of the standard canonical commutation relations when
we try to quantize (2.6). Specifically, the cubic part of the commutation relation yields
nonlinear constraints between the τ operator and the τ̄ operator. In the case of nonabelian
gauge theories having single derivative cubic interactions, this difficulty can be overcome
as follows; the interaction dependent part of the canonical momentum (proportional to
∂− AI − ∂I A− + gYM [A− , AI ]) vanishes upon choosing the light-cone gauge A− = 0.
In our case, a simple cure is to introduce a nonlocal field redefinition (and its complex
conjugated version)
√ i 1
τ →τ − 2κ (τ − τ̄ )∂− τ , (2.9)
2 ∂−
√ i 1
τ̄ → τ̄ − 2 κ (τ − τ̄ )∂− τ̄ (2.10)
2 ∂−
that deletes the part of the cubic action that contributes to Πτ . We will then try to quantize
the resulting system. In (2.9), we understand
x −
1
f (x − ) = dx − f (x − ) (2.11)
∂−
or its momentum space version
i 1 1
→ += , (2.12)
∂− 2p α
where α = 2p+ throughout the rest of this paper. Strictly speaking, the field redefinition
(2.9) should be applied to nonzero modes with p+ = 0. One can show that a slightly
modified version of (2.9) for the zero mode p+ = 0 part may be used to delete the cubic
terms involving ∂+ -derivative and zero modes.
Upon performing the field redefinition, the action (2.6) changes to:

+ − I 1 1 1
S = dx dx dx ∂+ τ ∂− τ̄ + ∂− τ ∂+ τ̄ − µ2 x I x I ∂− τ ∂− τ̄ − ∂I τ ∂I τ̄
2 2 4

√ 1 1
+ i 2 κ −(τ − τ̄ )∂I τ ∂I τ̄ + ∂I (τ − τ̄ )∂− τ ∂I τ̄
2 ∂−

1 1
+ ∂I (τ − τ̄ )∂− τ̄ ∂I τ + quartic and higher terms , (2.13)
2 ∂−
where we have formally used integration by parts. The cubic interaction terms in (2.13)
do not involve any ∂+ -derivatives. An important point is that they do not involve any
µ-dependent parts either and, in fact, they are identical to the flat space supergravity results.
The field redefinition that removes interaction terms having ∂+ -derivatives also removes
the µ-dependent interaction terms (without using integration by parts). We propose to
take (2.13) as the starting point for further analysis.
The main subtlety of light-cone Hamiltonian construction is the careful treatment of
p+ = 0 zero modes and the implementation of the extra global constraint from them. In the
flat spacetime case, Lorentz invariance emerges only after properly incorporating the zero
mode effect, as reported, for example, in [57]. What makes our problem solvable is that the
zero mode part of the action (2.13) is µ-independent. The only µ-dependent term in (2.13)
involves ∂− -derivative and vanishes for the zero modes. Taken together, these mean that
the extra µ-independent contribution to the light-cone Hamiltonian coming from the zero
modes should make the total Hamiltonian compatible with Lorentz invariance when µ = 0.
Fortunately, for the theory at hand (2.13), the zero-mode contribution to the light-cone
Hamiltonian which ensures Lorentz invariance can be inferred from an earlier work by
Goroff and Schwarz [58]. The light-cone Hamiltonian thus obtained from (2.13) is given
as follows:
H = H2 + H3 , (2.14)
where

1 2 I I
H2 = dx − dx I µ x x ∂− τ ∂− τ̄ + ∂I τ ∂I τ̄ , (2.15)
4
and

√ 1 1
H3 = dx − dx I τ ∂I τ ∂I τ̄ − ∂I
2 κi (τ ∂− τ ) ∂I τ̄
2 ∂−

1 1
− ∂I (τ ∂− τ̄ ) ∂I τ + c.c. + HGS . (2.16)
2 ∂−
The extra term HGS guarantees the Lorentz invariance up to cubic terms when µ = 0 and
is given by
√
2κ ∂I ∂I
HGS = τ ∂I τ ∂I τ̄ − τ ∂I τ ∂− τ̄ − τ ∂I τ̄ ∂− τ
4i ∂− ∂−

∂I ∂I
+ 2
τ ∂− τ ∂− τ̄ + c.c. (2.17)
∂−
In the above equations c.c. represents the complex conjugate. We note that the relative
normalization between HGS and the other terms in H3 is uniquely fixed by requiring the
eventual Lorentz invariance when µ = 0.
It is worthwhile to give a brief review of the work by Goroff and Schwarz [58]. They
considered the construction of light-cone Hamiltonian for a nonlinear sigma model with
SL(D − 2, R) invariance to mimic the D-dimensional ‘gravity’ theory:

1 ∂j ij 1 ∂i ∂j ij
S = d D x K+− − γ ij Kij + γ Ki− − 2
γ K−− . (2.18)
2 ∂− 2 ∂−
The field γ ij is a traceless symmetric ‘metric’ with i = 1, . . . , D − 2, and
Kµν = ∂µ γij ∂ν γ ij , (2.19)
where µ = +, −, 1, . . . , D − 2. Since our dilaton-axion system is a nonlinear sigma model
with SL(2, R) invariance, the model (2.18) when D = 4 should be similar to our system.
This becomes clear when we consider the following two points. First, we introduce
two fields C and C with opposite helicities for two physical modes of four-dimensional
gravitons. The first term of (2.18) then produces −∂+ C∂− C − ∂+ C∂
− C and nothing else,
up to cubic terms, analogous to our action (2.13). While this term alone is not J i+ invariant,
adding three extra terms in (2.18) makes it invariant. Secondly, the action of J i+ on fields
in flat spacetime is given entirely by orbital parts with no extra spin contributions; the index
structure of γij is irrelevant as far as the action of J i+ is concerned, making it easier to
apply to the dilaton-axion system. The extra term HGS can be read off from (2.18) and it
can be shown that HGS makes the action (2.13) Lorentz invariant when µ = 0.
The cubic Hamiltonian H3 is consistent with the light-cone superstring field theory
construction in the pp-wave background of Ref. [11]; as a functional of classical fields,
H3 of supergravity light-cone Hamiltonian is identical to the flat space (µ = 0) result.
Of course, this does not mean that the matrix elements of the quantum operator H3 are
µ-independent. As we have emphasized in the introduction, quantum mechanics of free
particles (µ = 0) and that of bounded particles (µ = 0) are quite different. For nonzero µ,
the Hilbert space consists of harmonic oscillator modes whereas the µ = 0 Hilbert space
is a collection of free particles. As we will see in the next section, the matrix elements
of H3 actually have some explicit µ dependence due to their dependence on the frequency
of harmonic oscillators.
To further compare with the analysis of [11], we write down the classical Hamiltonian
in momentum space,

3
dαr
H3 = δ(Σαr )h3 (αr )τ1 τ2 τ̄3 + c.c. (2.20)
2π
r=1
There are two contributions to h3 :

√
(0) 2κ 1
h3 = (α1 p2 − α2 p1 )2 (2.21)
i 4α1 α2
from H3 other than HGS and
√
(GS) 2κ 1 1 1
h3 = · + (α1 p2 − α2 p1 )2 (2.22)
i 8 α12 α22
from HGS , which sum up to yield
√
2 κ 1 α32
h3 = · (α1 p2 − α2 p1 )2 . (2.23)
i 8 α12 α22
Here, pt are the transverse momenta of the rth particle.
3. Supergravity sector of light-cone string field theory in a pp-wave background
Spradlin and Volovich [11] adopted the light-cone string field theory of [56] to
study string interactions in the pp-wave background. They showed that the prefactor
of the cubic Hamiltonian H3 is the same as the one in the flat space up to an overall
function f (α1 , α2 , α3 ). We compute the dilaton-axion sector of H3 and compare it to our
supergravity result in the previous section. It will be shown that the function f is indeed a
constant, and the precise value of the normalization constant will also be determined. We
then proceed to compute the matrix elements of H3 for three chiral primary fields.
3.1. Dilaton-axion system and normalization of H3
We begin with a brief review of IIB supergravity in the light-cone gauge [55] following
the notation of [56]. A single superfield Φ contains the 256 physical degrees of freedom of
IIB supergravity in the light-cone gauge:
1 4 1
Φ(α, p, λ) = α 2 A(α, p) + 2
A∗ (α, p) a···h λa · · · λh + Aabcd (α, p)λa λb λc λd
4 8!α 4!
1 ab 1 ab∗
+ αA (α, p)λa λb − A (α, p) a···h
λc · · · λh
4 6!α
+ (fermions), (3.1)
where λa is an SO(8) spinor with positive chirality. Only the fields on the first line will
concern us in this paper. The quadratic Hamiltonian in the pp-wave background is given
by [6]

1 dα d 8 p 8
H2 = d λ Φ(−α, −p, −λ)(∆B + ∆F )Φ(α, p, λ), (3.2)
2 2π (2π)8
1 ∂2 ∂
∆ B = p 2 − α 2 µ2 2 , ∆F = αµλΠ , (3.3)
4 ∂p ∂λ
where Π = γ 1 γ 2 γ 3 γ 4 is the SO(4) chirality operator. Expansion of H2 in component
fields is straightforward. The SO(8) scalars A, A∗ give

dα ∗
H2 = A (−α)∆B A(α). (3.4)
2π
This is exactly the same as the quadratic Hamiltonian of τ field in (2.15) upon identifying
τ (α) = A(α). For the reasons explained in the previous section and in Ref. [11], we expect
that ‘classically’ H3 is essentially the same as in the flat spacetime. In terms of the light-
cone superfields, H3 is given by
√
H3 = N 2 κ dµ3 v I J (Λ)PI PJ Φ(1)Φ(2)Φ(3), (3.5)
3

dαr d 8 pr
dµ3 = d λr δ(Σαr )δ 8 (Σpr )δ 8 (Σλr ),
8
(3.6)
2π (2π)8
r=1
P = α1 p2I − α2 p1I ,
I
(3.7)
Λa = α1 λa2 − α2 λa1 , (3.8)
where we have introduced a normalization constant N . Furthermore, the prefactor vI J is
given by
1 IK JK a b c d 16 I J
vI J = δI J + 2
γab γcd Λ Λ Λ Λ + δ a···h Λa · · · Λh
6α 8!α 4
i IJ a b 4i I J
− γab Λ Λ − γ abc···h Λc · · · Λh , (3.9)
α 6!α 3 ab
where α = α1 α2 α3 . The terms on the second line are irrelevant for supergravity but become
important for string states.
The dilaton-axion part of the interaction Hamiltonian is obtained from H3 in (3.5) by
performing the fermionic integrals and collecting terms involving A and A∗ . The Λ0 and
the Λ8 terms of (3.9) give nonvanishing answer (the Λ8 term producing the result shown
below and the Λ0 term producing its complex conjugate):
2
√ α3 2
H3 = 12N 2 κ d µ̄3 2 2 P τ1 τ2 τ̄3 + c.c., (3.10)
α1 α2
where d µ̄3 is the bosonic part of dµ3 defined in (3.6). We have H3 from (2.23), which
was obtained from the construction of light-cone Hamiltonian from the covariant action.
After the 90-degree rotation of (2.23) on the complex plane τ , under which τ → iτ and
τ̄ → −i τ̄ (H2 remains invariant), we have

√ 2
2κ α
H3 = d µ̄3 2 3 2 P2 τ1 τ2 τ̄3 + c.c. (3.11)
8 α1 α2
Comparing (3.10) and (3.11), we find that the normalization constant should be
1
N= . (3.12)
3 · 25
In conclusion, the possible function f (α1 , α2 , α3 ) undetermined from the symmetry
consideration should be set to a constant.
3.2. Matrix elements of H3 for chiral primaries
3.2.1. Interaction Hamiltonian

It is convenient to divide the SO(8) spinor λa into two groups depending on their SO(4)
chirality. For example, we may choose a basis of gamma matrices in which Π = γ 1 γ 2 γ 3 γ 4
is diagonal, so that λa s (a = 1, 2, 3, 4) have positive chirality and λā s (ā = 5, 6, 7, 8) have
negative chirality. In such a basis, the chiral primary field is related to the components of
the superfield as
1 ∗ 1
s(α) = √ A5678(α), s(−α) = s(α) = √ A1234(−α) (α > 0). (3.13)
2 2
√
The factor of 2 was introduced to normalize the quadratic Hamiltonian in the standard
way:

H2 = d µ̄3 s(−α)(∆B − 4µ|α|)s(α). (3.14)
In order to obtain the cubic Hamiltonian H3 for s(α), one has to expan (3.5) and collect s 3
terms. It is clear that only the middle component
1
γ I K γ J K Λa Λb Λc Λd (3.15)
6(α1 α2 α3 )2 [ab cd]
of v I J will contribute to H3 for chiral primaries since other terms simply do not have
the right number of λs to saturate the λ integral in (3.5). Moreover, the relation (3.13)
implies that it is sufficient to pick out the terms with all Λa s having the same chirality, i.e.,
Λ1 Λ2 Λ3 Λ4 ≡ Λ1234 and Λ5 Λ6 Λ7 Λ8 ≡ Λ5678 . When the spinor indices are restricted to
the same SO(4) chirality subspace, say a = 1, 2, 3, 4 space, it can be shown that
jK j K j K
iK
γ[ab γcd] = δ ij abcd , iK
γ[ab γcd] = −δ i j abcd , iK
γ[ab γcd] = 0. (3.16)
I K vanishes if I and J do not belong to the same SO(4). This fact can
First, note that the γab
be verified by using the aforementioned basis of gamma matrices where the matrix Π is
diagonal. Second, since there are only four components of a spinor with the same chirality,
the LHS of (3.16) must be proportional to abcd . With a suitable choice of the basis for γ
JK
matrices, one can use the self-dual property of γab
1 i k 1 i k
ik
γab = abcd γcd
ik
, γab = − abcd γcd (3.17)
2 2
to show that the RHS of (3.16) is correctly normalized and that there is a relative minus
sign between the two SO(4)s.
Without loss of generality, one may assume that α (1) , α (2) > 0 and α (3) < 0. In such a
case, one encounters the following expression,

Λ1234 λ(1) + λ(2) 1234, (3.18)
(1) (2)
where Λ = α1 λ(2) − α2 λ(1) . Only the terms proportional to λ1234 λ1234 saturate the
λ-integral. Collecting all such terms, one finds
1
(α1 + α2 )4 λ(1) (2)
1234 λ1234 = (|α1 | + |α2 | + |α3 |)4 λ(1) (2)
1234 λ1234 . (3.19)
16
All in all, we find that H3 for chiral primaries is given by
√
23/2 2 κ (|α1 | + |α2 | + |α3 |)4 2
H3 = d µ̄ 3 P − P 2
s s s
⊥ 1 2 3 , (3.20)
3 · 27 α12 α22 α32
where and ⊥ denote the four transverse directions coming from AdS5 and the other four
from S 5 , respectively.
3.2.2. Quantization
Consider a massless scalar in the pp-wave background in the light-cone gauge2

10 √ 1
S = d x −g − (∇φ) 2
2

+ − 8 1 2 2
= dx dx d x ∂+ φ∂− φ − µ x (∂− φ) − (∂I φ) .
2 2
(3.21)
4
The Dirac quantization procedure for a constrained system gives the quantum commutation
relation
i
φ(x − , x ), ∂− φ(y − , y ) = δ(x − − y − )δ 8 (
x − y ). (3.22)
2
The normalizable on-shell wavefunctions are labeled by α, n
+ −iαx −
φ(x + , x − , x ) = e−iEx fn (
x ). (3.23)
Here, fn is the n -excited state wave function of an eight-dimensional harmonic
oscillator
with ω = µ|α|/2 and the energy is given by E = µ(|| n|| ≡ 8I =1 nI .
n|| + 4) where ||
We find it convenient to focus on the x − (or α in the momentum space) dependence,
suppressing the transverse directions as long as no confusion arises. Going to the
2 For simplicity, we consider a massless scalar in place of the chiral primary field s. The difference between
the two fields will not be important in what follows.
momentum space only in the x − direction via Fourier-transform

dα −
φ(x − ) = φ(α)e−iαx , (3.24)
2π
the commutation relation becomes
1
φ(α1 ), φ(α2 ) = 2πδ(α1 + α2 ), (3.25)
2α1
which is solved by
√
2π
φ(α) = √ an (α)fn (
x ) (α > 0), (3.26)
2 α n
where the oscillators satisfy

am (α1 ), an† (α2 ) = δm,
n α1 δ(α1 − α2 ). (3.27)
The corresponding single particle states are normalized accordingly,
α1 , m|α
2 , n = δm,
n α1 δ(α1 − α2 ). (3.28)
In terms of the oscillators the quadratic Hamiltonian is expressed as
dα †
H2 = µ n|| + 4).
a (α)an (α)(|| (3.29)
2πα n
n
3.2.3. Matrix elements of H3

Suppose we have a cubic Hamiltonian of the form

dα1 dα2 dα3
H3 = (2π)δ(Σαr )h3 (αr )s(α1 )s(α2 )s(α3 ). (3.30)
(2π)3
It is easy to show that the matrix element is given by
2π · 3!
3|H3 |1, 2 = h3 (αr )δ(Σαr ) × (
nr |E(br )|0). (3.31)
23/2(2π)3/2
√
Note that we have included the symmetry factor 3! and factors of (2π)s and 2 s
that we introduced when writing fields in terms of creation/annihilation operators. The
dimensionless factor (nr |E(br )|0) arises from the overlap integral of the wave function in
the transverse directions. In the language of quantum mechanics of a harmonic oscillator,
we have the following expressions:
x ) = (
fn ( n|
x ) = ( x )|0),
n|O( (3.32)
2

x ) = µα exp − 1 a † · a † + µ|α| x · a † − 1 µ|α|
O( x2 . (3.33)
2π 2 4
Integration over the eight transverse directions becomes a simple Gaussian integral which
produces
24 µ2 (α1 α2 α3 )2
× (
nr |E(br )|0), (3.34)
(2π) (|α1 | + |α2 | + |α3 |)4
2
where the operator E(br ) defined by

 
3 1 − b12 −b1 b2 −b1 b3
1
E ≡ exp †
a(r) †
M rs a(s) , M ≡  −b1 b2
rs
1 − b22 −b2 b3  , (3.35)
2
r,s=1 −b1 b3 −b2 b3 1 − b32
√
and br ≡ 2|αr |/(|α1 | + |α2 | + |α3 |) is precisely the same as the operator Ea0 defined in
Eqs. (4.10), (4.11) of [11]. Before using h3 for the chiral primaries in (3.20), it is useful to
realize that within the momentum integral one can write
⊥

P2 − P2⊥ = α1 α2 α3 E123 − E123 , (3.36)

⊥ ≡ E ⊥ −E ⊥ −E ⊥ are the contributions to the energy
where E123 ≡ E3 −E1 −E2 and E123 3 1 2
difference from the two SO(4) directions. This identity [11] follows from the definition of
the free part of the light-cone Hamiltonian for a chiral primary field,

1 2 1
H= p + (µα) x − 4µ,
2 2
(3.37)
α 4
and the fact that the sum of three x terms vanishes due to momentum conservation
(α1 + α2 + α3 = 0). The constant term in (3.37) originates from the ∆F term in (2.15)
and cancels the zero point energy of the bosonic harmonic oscillator part. Since one has
only the difference between the two SO(4) directions, the zero point energy cancels out
automatically and the term 4µ is irrelevant. Note also that in (3.31), α1 and α2 are positive
while α3 = −(α1 + α2 ) is negative. Since the eigenvalue of Hamiltonian (3.37) has the
same sign as α, if one defines Er to be the absolute value of the energy of the rth state, one
⊥ . Taking everything into account, we insert
gets P2 = α1 α2 α3 E123 and P2⊥ = α1 α2 α3 E123
the h3 of (3.20) into (3.31) to find
π ⊥
3|H3 |1, 2 = gα 2 µ2 α1 α2 α3 E123 − E123 δ(Σαr ) × ( nr |E(br )|0). (3.38)
2
The authors of Ref. [10] made a proposal for the matrix elements that is supposed to be
valid when the energy difference between the in-state and the out-state is nearly zero.3 In
the Yang–Mills variables, the proposal reads
3|H3 |12 = µ(∆3 − ∆2 − ∆1 )C123 δJ1 +J2 ,J3 . (3.39)

The value of C123 depends on the states {|1, |2, |3}. To compare with the results of
Ref. [10], we introduce the following operators in Yang–Mills and their counterparts in
supergravity,
1
AJ = √ Tr Z J ←→ |α, 0, (3.40)
J NJ
3 For the supergravity modes, the energy difference is always an integer (times µ), so one may think that we
are comparing a zero with another zero in the rest of the section. A related subtlety that arises in the computation
of extremal correlators in AdS [54] was circumvented by using analytic continuation. In the same spirit, we use
analytic continuation to give a meaning to the coefficient multiplying the ‘zero’.
1
BJ = √ Tr φZ J ←→
aφ† |α, 0, (3.41)
N J +1
1 J

CJ = √ Tr φZ l ψZ J −l ←→
aψ† aφ† |α, 0. (3.42)
J N J +2 l=0
There are four processes with E123 ≈ 0. The value of C123 for each case can be read off
from Section 3.2 of Ref. [10]:
√
J1 J2 J3 (0)
AA → A : C123 = ≡ C123 , (3.43)
N

(0) J2
AB → B : C123 = C123 × , (3.44)
J3
J2
(0)
AC → C : C123 = C123 × , (3.45)
J3
√
(0) J1 J2
Bφ Bψ → C : C123 = C123 × . (3.46)
J3
Note that since J is proportional to α, the relative factor matches precisely with
nr |E(br )|0). Therefore, it suffices to check the correspondence for the case AA → A.
(
Following Ref. [10], we switch from the unit normalization i|j = δij to a normaliza-
tion suitable for the continuum limit, i|j = Ji δij = αi δ(αi − αj ), and use J = µR 2 α/2
and R 4 = 4πgs α 2 N to find
3|H3 |1, 2 = πgα 2 µ2 α1 α2 α3 E123 δ(Σαr ) (3.47)
which, indeed, agrees with the supergravity result (3.38) with nr = 0 (up to a numerical
factor of two). In fact, the authors of Ref. [10] considered only the processes with strictly
⊥ . Our result (3.38) suggests that the Yang–
vanishing E123 and slightly nonzero E123
Mills computation should also distinguish the two SO(4) directions and E123 in (3.47)
be replaced by E123⊥ − E .4 It would be interesting to verify this expectation explicitly
123
on the Yang–Mills side.
Acknowledgements
We are grateful to Jin-Ho Cho, Shiraz Minwalla, Jongwon Park and Hyeonjoon Shin
for useful discussions, and Peter Lee for pointing out an error in the original version of
this paper. We also thank the organizers of the KIAS Workshop on Strings and Branes for
hospitality and for an opportunity to present this work. Y. Kiem is especially grateful to
Seungjoon Hyun for pointing out a subtlety in Section 2 regarding the zero mode treatment.
The work of Y. Kiem was supported by Korea Science and Engineering Foundation
(KOSEF) Grant R02-2002-000-00146-0. The work of Y. Kim was supported by Korea
4 After submitting this paper, we were informed that the relative minus sign between the two SO(4) directions
and its implications have been noticed independently in Refs. [10,59].
Research Foundation Grant KRF-2001-015-DP0082. The work of S. Lee was the result of
research activities (CNNC) supported by KOSEF. The work of J. Park was supported by
the POSTECH BSRI research fund 1RB0210601.
Appendix A. AdS computation
In this appendix, we compute the matrix elements for three chiral primary states
(∆ = J ). The AdS supergravity action for the fields corresponding to the chiral primaries
is known to be [54]

√ 1
S = d 5 x −g − ∇s I ∇ s̄ I − m2I |s I |2 − GI J K s I s J s̄ K + c.c. . (A.1)
2
When all three s fields have ∆ − J = 0 , the coupling constant is

√ 21/2 J1 J2 J3 (J32 − 1)(J3 + 2)(J3 − 2)∆123
G123 = ( 2 κ) × F123 , (A.2)
(J12 − 1)(J22 − 1)(J1 + 2)(J2 + 2)
√
1 (J1 + 1)(J1 + 2)(J2 + 1)(J2 + 2)
F123 = √ √ . (A.3)
2π 3 (J3 + 1)(J3 + 2)
The coefficient FI J K comes from the overlap integral of spherical harmonics on S 5 . This
factor gets modified when one considers fields with different values of ∆ − J , but as we
discussed in Section 3.2.3, the change again matches ( nr |E(br )|0).
It is straightforward to compute the matrix element of the cubic Hamiltonian from the
action (A.1). Each on-shell wavefunction to the linearized equation of motion corresponds
to a single particle state in the quantum theory. For simplicity, we restrict ourselves to the
ground state (E = ∆) for which the normalized wave-packet is given by
√
∆(∆ − 1)
s= , (A.4)
π(cosh ρ)∆
where ρ is the radial variable of the AdS global coordinates. Following the standard recipe
of quantum field theory, we find
√
1 (∆1 − 1)(∆2 − 1)(∆3 − 1)
3|H3 |12 = 3/2 × G123. (A.5)
2 π (∆3 − 1)(∆3 − 2)
Using the G123 written above, we find the following remarkably simple result:
√
∆1 ∆2 ∆3
3|H3 |12 = (∆3 − ∆1 − ∆2 ) , (A.6)
N
which is in agreement with the proposal of [10].
All of the above appear sensible, but we have to admit that there are a few reasons to
doubt the validity of this computation. First, in the process of obtaining the cubic term in
the action (A.1), one made a nonlinear field redefinition in which the on-shell condition
∇ 2 s = m2 s was used. In particular, on-shell condition removed cubic couplings with time
derivatives, which may cause trouble in quantization. Second, the method used in this
appendix gives answers that are finite in the pp-wave limit only when ∆123 = 0. If the cubic
Hamiltonian we used were the correct one, the matching between the AdS Hamiltonian and
the pp-wave Hamiltonian would be valid for arbitrary values of ∆123 .
References
[1] J. Figueroa-O’Farrill, G. Papadopoulos, Homogeneous fluxes, branes and a maximally supersymmetric

solution of M-theory, JHEP 0108 (2001) 036, hep-th/0105308.
[2] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, A new maximally supersymmetric background of
IIB superstring theory, JHEP 0201 (2001) 047, hep-th/0110242.
[3] R.R. Metsaev, Plane wave Ramond–Ramond background, hep-th/0112044.
[4] M. Blau, J. Figueroa-O’Farrill, C. Hull, G. Papadopoulos, Penrose limits and maximal supersymmetry, hep-
th/0201081.
[5] D. Berenstein, J. Maldacena, H. Nastase, Strings in flat space and pp-waves from N = 4 super Yang–Mills,
JHEP 0204 (2002) 013, hep-th/0202021.
[6] R.R. Metsaev, A.A. Tseytlin, Exactly solvable model of superstring in plane wave Ramond–Ramond
background, hep-th/0202109.
[7] C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, A new double-scaling limit of N = 4 super Yang–
Mills theory and pp-wave strings, hep-th/0205033.
[8] D. Berenstein, H. Nastase, On lightcone string field theory from super Yang–Mills and holography, hep-
th/0205048.
[9] D.J. Gross, A. Mikhailov, R. Roiban, Operators with large R charge in N = 4 Yang–Mills theory, hep-
th/0205066.
[10] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, PP-wave string
interactions from perturbative Yang–Mills theory, hep-th/0205089.
[11] M. Spradlin, A. Volovich, Superstring interactions in a pp-wave background, hep-th/0204146.
[12] M. Blau, J. Figueroa-O’Farrill, G. Papadopoulos, Penrose limits, supergravity and brane dynamics, hep-
th/0202111.
[13] D. Berenstein, C.P. Herzog, I.R. Klebanov, Baryon spectra and AdS/CFT correspondence, hep-th/0202150.
[14] N. Itzhaki, I.R. Klebanov, S. Mukhi, pp-wave limit and enhanced supersymmetry in gauge theories,
JHEP 0203 (2002) 048, hep-th/0202153.
[15] J. Gomis, H. Ooguri, Penrose limit of N = 1 gauge theories, hep-th/0202157.
[16] J.G. Russo, A.A. Tseytlin, On solvable models of type IIB superstring in NS–NS and R–R plane wave
backgrounds, JHEP 0204 (2002) 021, hep-th/0202179.
[17] L.A. Pando Zayas, J. Sonnenschein, On Penrose limits and gauge theories, hep-th/0202186.
[18] M. Hatsuda, K. Kamimura, M. Sakaguchi, From super-AdS5 × S 5 algebra to super-pp-wave algebra, hep-
th/0202190.
[19] M. Alishahiha, M.M. Sheikh-Jabbari, The pp-wave limits of orbifolded AdS5 × S 5 , hep-th/0203018.
[20] M. Billo’, I. Pesando, Boundary states for GS superstrings in an Hpp-wave background, hep-th/0203028.
[21] N. Kim, A. Pankiewicz, S.-J. Rey, S. Theisen, Superstring on pp-wave orbifold from large-N quiver gauge
theory, hep-th/0203080.
[22] M. Cvetic, H. Lu, C.N. Pope, Penrose limits, pp-waves and deformed M2-branes, hep-th/0203082.
[23] T. Takayanagi, S. Terashima, Strings on orbifolded pp-waves, hep-th/0203093.
[24] U. Gursoy, C. Nunez, M. Schvellinger, RG flows from Spin(7), CY 4-fold and HK manifolds to AdS,
Penrose limits and pp waves, hep-th/0203124.
[25] J. Michelson, (Twisted) toroidal compactification of pp-waves, hep-th/0203140.
[26] S.R. Das, C. Gomez, S.-J. Rey, Penrose limit, spontaneous symmetry breaking and holography in pp-wave
background, hep-th/0203164.
[27] C.-S. Chu, P.-M. Ho, Noncommutative D-brane and open string in pp-wave background with B-field, hep-
th/0203186.
[28] A. Dabholkar, S. Parvizi, Dp branes in pp-wave background, hep-th/0203231.

[29] D. Berenstein, E. Gava, J. Maldacena, K.S. Narain, H. Nastase, Open strings on plane waves and their
Yang–Mills duals, hep-th/0203249.
[30] J.P. Gauntlett, C.M. Hull, pp-waves in 11 dimensions with extra supersymmetry, hep-th/0203255.
[31] P. Lee, J. Park, Open strings in pp-wave background from defect conformal field theory, hep-th/0203257.
[32] M. Hatsuda, K. Kamimura, M. Sakaguchi, Super-pp-wave algebra from super-AdS × S algebras in eleven
dimensions, hep-th/0204002.
[33] E. Kiritsis, B. Pioline, Strings in homogeneous gravitational waves and null holography, hep-th/0204004.
[34] A. Kumar, R.R. Nayak, Sanjay, D-brane solutions in pp-wave background, hep-th/0204025.
[35] R.G. Leigh, K. Okuyama, M. Rozali, pp-waves and holography, hep-th/0204026.
[36] D. Bak, Supersymmetric branes in pp-wave background, hep-th/0204033.
[37] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, A semi-classical limit of the gauge/string correspondence, hep-
th/0204051.
[38] K. Skenderis, M. Taylor-Robinson, Branes in AdS and pp-wave spacetimes, hep-th/0204054.
[39] S. Mukhi, M. Rangamani, E. Verlinde, Strings from quivers, membranes from moose, hep-th/0204147.
[40] M. Alishahiha, M.M. Sheikh-Jabbari, Strings in pp-waves and worldsheet deconstruction, hep-th/0204174.
[41] V. Balasubramanian, M. Huang, T.S. Levi, A. Naqvi, Open strings from N = 4 super Yang–Mills, hep-
th/0204196.
[42] S. Frolov, A.A. Tseytlin, Semiclassical quantization of rotating superstring in AdS5 × S 5 , hep-th/0204226.
[43] H. Takayanagi, T. Takayanagi, Open strings in exactly solvable model of curved spacetime and pp-wave
limit, hep-th/0204234.
[44] I. Bakas, K. Sfetsos, pp-waves and logarithmic conformal field theories, hep-th/0205006.
[45] C. Ahn, More on Penrose limit of AdS4 × Q1,1,1 , hep-th/0205008.
[46] A. Parnachev, D.A. Sahakyan, Penrose limit and string quantization in AdS3 × S 3 , hep-th/0205015.
[47] M. Li, Correspondence principle in a pp-wave background, hep-th/0205043.
[48] K. Oh, R. Tatar, Orbifolds, Penrose limits and supersymmetry enhancement, hep-th/0205067.
[49] S.D. Mathur, A. Saxena, Y.K. Srivastava, Scalar propagator in the pp-wave geometry obtained from
AdS5 × S 5 , hep-th/0205136.
[50] R. Gopakumar, String interactions in pp-waves, hep-th/0205174.
[51] O. Bergman, M.R. Gaberdiel, M.B. Green, D-brane interactions in type IIB plane-wave background, hep-
th/0205183.
[52] K. Dasgupta, M.M. Sheikh-Jabbari, M. van Raamsdonk, Matrix perturbation theory for M-theory on a pp-
wave, hep-th/0205185.
[53] G. Bonelli, Matrix strings in pp-wave backgrounds from deformed super Yang–Mills theory, hep-
th/0205213.
[54] S. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three-point functions of chiral operators in D = 4, N = 4
SYM at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hep-th/9806074.
[55] M.B. Green, J.H. Schwarz, Extended supergravity in ten dimensions, Phys. Lett. B 122 (1983) 143.
[56] M.B. Green, J.H. Schwarz, L. Brink, Superfield theory of type (II) superstrings, Nucl. Phys. B 219 (1983)
437.
[57] T. Maskawa, K. Yamawaki, The problem of P + = 0 mode in the null-plane field theory and Dirac’s method
of quantization, Prog. Theor. Phys. 56 (1976) 270.
[58] M. Goroff, J.H. Schwarz, D-dimensional gravity in the light-cone gauge, Phys. Lett. B 127 (1983) 61.
[59] P. Lee, S. Moriyama, J. Park, Cubic interactions in pp-wave light cone string field theory, hep-th/0206065.
Nuclear Physics B 642 [FS] (2002) 407–432
Spin and superconducting instabilities near

a Van Hove singularity
J. González
Instituto de Estructura de la Materia, Consejo Superior de Investigaciones Científicas,
Serrano 123, 28006 Madrid, Spain
Abstract
We apply a wilsonian renormalization group approach to the system of electrons in a two-
dimensional square lattice interacting near the saddle-points of the band, when the correlations
at momentum Q ≡ (π, π) prevail in the system. The detailed consideration of the spin degrees
of freedom allows to discern the way in which the SU(2) spin invariance is preserved in the
renormalization process. Regarding the spin correlations, we find two different universality classes
which correspond, in the context of the extended Hubbard model, to having the bare on-site
interaction U repulsive or attractive. The first class is characterized by a spin instability which
develops through the condensation of particle–hole pairs with momentum Q, with the disappearance
of the Fermi line in the neighborhood of the saddle-points. Within that class, the attractive or
repulsive character of the nearest-neighbor interaction V dictates whether there is or not a d-wave
superconducting instability in the system. For the Hubbard model with just on-site interaction, we
show that some of the irrelevant operators are able to trigger the superconducting instability. The
naturalness of the competing instabilities is guaranteed by the existence of a range of doping levels in
which the chemical potential of the open system is renormalized to the level of the saddle-points. We
incorporate this effect to obtain the phase diagram as a function of the bare chemical potential, which
displays a point of optimal doping separating the regions of superconductivity and spin instability.
PACS: 71.10.Hf; 74.20.Mn; 71.10.-w
E-mail address: imtjg64@pinar2.csic.es (J. González).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 3 7 - 5
408 J. González / Nuclear Physics B 642 [FS] (2002) 407–432
1. Introduction
During the last years there has been much effort devoted to the study of strongly
correlated electron systems. The interest has been maintained by the behavior displayed
by the high-Tc copper-oxide compounds since the discovery of their superconductivity 15
years ago [1]. There are a number of features exhibited by these materials that do not fit
into the conventional theoretical frameworks. The normal state of the cuprates shows, for
instance, unusual transport properties and, more strikingly, a pseudogap phase in which
part of the density of states is lost at the Fermi level while the system remains conducting.
It seems that a new paradigm is needed to describe these materials, in the same way as the
Fermi liquid picture accounts for the behavior of conventional metals.
From the theoretical point of view, progress has been made during the past decade
in understanding the foundations of Landau’s Fermi liquid theory and, consequently, the
possible deviations that may open the way to a new kind of metallic behavior [2–4]. The
most powerful method used in this task has been the renormalization group (RG) approach
developed for interacting fermion systems [2]. We have learned from it that the Fermi liquid
picture is a very robust description of the metallic state. There are only a few perturbations
that may destabilize the Fermi liquid, favoring the formation of states with different types
of symmetry breaking. The Fermi liquid represents itself a universality class in which any
electron system falls at dimension D 2, unless the interaction is sufficiently long-ranged
[5–11] or the Fermi surface develops singular points [12].
Soon after the discovery of the high-Tc superconductivity, it was proposed that
the presence of nonlinear dispersion near the Fermi line of the copper-oxide layers
could be at the origin of the unconventional behavior [13,14]. The fermion systems
in a two-dimensional (2D) square lattice have necessarily saddle-points in their band
dispersion, which give rise to Van Hove singularities where the density of states diverges
logarithmically. In the most common instances, the two inequivalent saddle-points lie
at the boundary of the Brillouin Zone, and their hybridization has been proposed to
explain the existence of a d-wave order parameter in the superconducting phase [15–
19], as observed experimentally. Further investigations have shown that the unconventional
transport properties in the normal state may be accounted for by the proximity of the Fermi
level to the Van Hove singularity (VHS) in the copper-oxide layers [20–24].
A careful examination of the kinematics near the saddle-points has shown indeed that
a superconducting instability with d-wave order parameter arises in the t − t Hubbard
model with bare repulsive interaction [25,26]. The mechanism at work is of the same kind
described by Kohn and Luttinger as giving rise to a p-wave pairing instability in the three-
dimensional Fermi liquid [27,28], but adapted now to the 2D model with saddle-points
near the Fermi line. Other studies have considered in detail the influence of the entire
Fermi line in the development of the instabilities of the system [29,30]. They have given
further support to the picture of a competition between a spin-density-wave instability and a
pairing instability with d-wave order parameter in the t − t Hubbard model with the Fermi
level at the VHS. More recently, a refined renormalization program has been implemented
in Ref. [31] by trying to handle the momentum dependence of the vertex functions in the
scaling procedure, what has confirmed the appearance of different phases with symmetry
breaking in the spin and the charge sector.
J. González / Nuclear Physics B 642 [FS] (2002) 407–432 409
Despite all the results obtained in the system of electrons near the VHS, there are
still important obstacles precluding a precise description of the effective theory at low
energies. From a technical point of view, the source of the problem is the appearance of
infrared singularities in the RG approach after accomplishing the renormalization of the
leading logarithm. Some vertex functions, like the four-point interaction with vanishing
total incoming momentum at the one-loop level or the electron self-energy at the two-loop
level, get log2 (Λ) corrections in terms of the energy cutoff Λ. After applying the standard
RG program, the renormalized quantities still contain factors of the form log(Λ). This fact
questions the predictability of the theory since the argument of the logarithm has a hidden
energy scale, which sets the strength of the corrections. From a formal point of view, the
theory becomes nonrenormalizable in the standard RG approach, since the energy cutoff is
not the only dimensionful variable that appears in the scaling process.
The problem of the renormalizability of the theory can be best handled by adopting a
wilsonian RG approach, in which only the high-energy modes that live at the cutoff Λ are
integrated out at each RG step. In the present paper we follow Shankar’s RG program for
interacting fermion systems [2], which has the advantage of decoupling the renormalization
of the BCS channel (with vanishing momentum of the colliding particles) from that of the
rest of the channels at the one-loop level.
Moreover, the important feature of the wilsonian approach is that it allows to set free
the chemical potential, so that it can readjust itself at each step of integration. The issue of
the renormalization of the chemical potential has been discussed in Ref. [2] in the context
of Fermi liquid theory, and it reaches great significance when considering the system of
electrons near the VHS. The chemical potential cannot be fixed at the singularity from
the start, since it is actually the scale needed to regularize the infrared singularities that
appear in the standard RG procedure. On the other hand, the final location of the chemical
potential relative to the VHS is not arbitrary, since it is a dynamical quantity that scales in
a predictable way upon renormalization.
We remark that the renormalization devised in the paper assumes a constant value of
the bare chemical potential, instead of a constant particle number of the system. That
is, we describe a situation appropriate for an electron system in contact with a charge
reservoir, which sets the nominal value µ0 of the ensemble. The renormalization accounts
for the reduction suffered by the effective chemical potential inside the electron system
due to the repulsive interaction. This description of the electron system at constant nominal
chemical potential is most appropriate when dealing with the Cu–O layers of the cuprate
superconductors, since it provides a realization of the contact of the 2D layers with the
charge reservoir. The conclusion is that a variation in the external chemical potential does
not have always a linear correspondence with the variation of the final renormalized value
of µ, which is identified with the Fermi energy of the electron system.
The renormalization of the chemical potential makes possible to address the question
of the naturalness of the picture in which the Fermi level is fine-tuned to the VHS. The
strength of the predicted instabilities depends crucially on the proximity of the Fermi
energy to the singularity. This has been the main criticism to the proposals claiming that
the features of the copper-oxide materials could be related to the properties of electrons
interacting near a VHS. We will show that the chemical potential is renormalized towards
the VHS in a certain range of filling levels, in such a way that it may become pinned to
the singularity in the low-energy theory. This fact was already anticipated in Refs. [16,32,
33], and it has been used to cure the infrared singularities of the electron self-energy in
Ref. [21]. In the present paper, we will take into account such an effect to determine in a
predictable way the strength of the pairing instability in the system, as a function of the
different values of the bare chemical potential.
In the next section we describe the system to which our analysis applies. In Section 3
we classify the different renormalized vertices that arise by explicit consideration of the
spin degrees of freedom. The universality classes of the system are obtained in Section 4,
where we also show the way in which the SU(2) spin invariance is preserved along the RG
flow. Section 5 is devoted to establish the properties of the spin instability of the system,
while Section 6 analyzes the renormalization of the chemical potential to determine the
region of the phase diagram in which the superconducting instability prevails. Finally, the
last section is devoted to draw the main conclusions of this work.
2. The model
We take as starting point of our analysis a system of interacting electrons in the 2D

square lattice with nearest-neighbor hopping t and next-to-nearest-neighbor hopping t .
The band dispersion of the model is given by

ε(k) = −2t cos(kx ) + cos(ky ) + 4t cos(kx ) cos(ky ), (1)
where we have set the lattice spacing equal to one. Some of the energy contour lines are
shown in Fig. 1. The dispersion has two inequivalent saddle-points A and B at the boundary
of the Brillouin Zone. In their neighborhood, the energy of the one-particle states can be
approximated by the quadratic form
εA,B (k) ≈ ∓(t ∓ 2t )kx2 ± (t ± 2t )ky2 , (2)

where the momenta kx and ky measure now small deviations from A and B.
Fig. 1. Contour energy map for the t − t model about the Van Hove filling.
As a consequence of the nonlinear character of the dispersion, the density of states n(ε)
diverges logarithmically at the level of the saddle-points

c log(t/|ε|)
n(ε) ≈ with c ≡ 1/ 1 − 4(t /t)2 . (3)
4π 2 t
This implies that, when the Fermi level is close to the VHS, most part of the low-
energy states are concentrated in the neighborhood of the two saddle-points. In order to
apply the RG approach, we may take two patches where the quadratic approximation
(2) holds around the saddle-points. Higher-order corrections to the expression (2) are
irrelevant under the scaling that makes the action of the model a fixed-point of the RG
transformations, as we see in what follows.
We consider then a model whose action at the classical level is

S= dt d 2 p iΨ + +
aσ (p)∂t Ψ aσ (p) − (εa (p) − µ0 )Ψ aσ (p)Ψ aσ (p)
a

+
+ dt d 2 p1 d 2 p2 d 2 p3 d 2 p4 U (p1 , p2 , p3 , p4 )Ψ +
aσ (p1 )Ψ bσ (p2 )
a,b,c,d
× Ψ cσ (p4 )Ψ dσ (p3 )δ(p1 + p2 − p3 − p4 ), (4)
where the indices a, b, c, d run over the two patches around A and B.
The scaling transformation that leaves invariant the kinetic term of the action is
∂t → s∂t , (5)
p→s 1/2
p, (6)
−1/2
Ψ aσ (p) → s Ψ aσ (p). (7)
It is easily checked that, with the transformation (5)–(7), the interaction term in the
action (4) is also scale-invariant for a constant value of the potential U (p1 , p2 , p3 , p4 ). If
this is not constant, provided that it is a regular function of the arguments we can resort to
an expansion in powers of the momenta. Only the constant term is significant, since the rest
of higher-order terms fade away upon scaling to the low-energy limit s → 0. This means
that we meet the first requirement to apply the RG program, that is to have a model which
converges to a fixed-point under RG transformations at the classical level.
In the above scaling, we already find the first deviation in the RG program with respect
to the analysis of Fermi liquid theory. In the case of a model with circular Fermi line,
the interaction term is scale invariant only for very special kinematics of the scattering
processes [2]. In our model, we have seen that no constraint is needed on the four momenta
involved in the interaction at the classical level. It is only after taking into account virtual
processes that the interactions will start to grow large under scaling for some particular
choices of the kinematics. This will single out a number of so-called marginally relevant
channels among all the scattering processes, recovering then the similitude with the
analysis of Fermi liquid theory at the quantum level.
The two-patch RG analysis of the t − t Hubbard model has proven to give the dominant
instabilities of the system with the Fermi level at the VHS. For t > 0.276 t, a ferromagnetic
phase has been found below a certain critical frequency [25,26,34,35], in agreement with
the results obtained from Monte Carlo calculations [36]. In this paper we will be interested
in the regime with t < 0.276 t, where the competition between a spin instability and a
pairing instability arises, making the model more appropriate for the comparison with the
phenomenology of the cuprates.
3. Wilsonian renormalization group
In what follows we apply a wilsonian RG approach to obtain the low-energy effective

theory of the system. We proceed by progressive integration of the modes in two thin shells
of width dΛ at distance Λ in energy below and above the Fermi level, as depicted in Fig. 2.
For the time being, we will assume that the Fermi level is located precisely at the VHS,
unless otherwise stated. This is crucial to obtain a significant renormalization in any of the
interaction channels, and later on we will comment on the naturalness of this situation.
The vertex functions may become relevant, that is increasingly large at low energies,
only for very definite choices of the kinematics. Focusing on the four-point interaction
vertex, this is renormalized by a quantity of order dΛ at each RG step only when the
momentum transfer along a pair of external lines is either 0 or Q ≡ (π, π), or when the total
momentum of the incoming modes vanishes (BCS channel). In the present work we deal
with the latter two instances, since the first corresponds to the case of forward-scattering
interactions, which are subdominant in the range t < 0.276 t that we are considering.
In this regime, the divergences at vanishing momentum-transfer are related to charge
instabilities of the system, which have been treated in detail elsewhere [37]. We will see
that divergences in the channel with momentum transfer Q give rise to a spin instability,
which competes with the superconducting instability in the BCS channel in the model with
a bare on-site repulsive interaction.
Fig. 2. Picture of the density of states n(ε) and of the renormalization of the chemical potential µ by integration
of states at the energy cutoff Λ.
(a) (b)
Fig. 3. BCS vertices that undergo renormalization by particle–particle diagrams. The solid and dashed lines stand
for modes in the neighborhood of the two different saddle points.
(a) (b) (c) (d)
Fig. 4. Direct and exchange vertices that undergo renormalization by particle–hole diagrams.
(a) (b) (c) (d)
Fig. 5. Umklapp vertices that undergo renormalization by particle–hole diagrams.
The different kinematics which may appear in the BCS channel are listed in Fig. 3. We
allow for the possibility of Umklapp processes in which the incoming modes scatter from
one of the saddle points to the other.
The different kinematical possibilities that arise in the channel with momentum transfer
Q are classified in Figs. 4 and 5. The first includes the interactions in which the incoming
modes are at different saddle points, while the latter contains the Umklapp processes. The
other important distinction is between direct (D) and exchange (E) interactions. Direct
processes are those in which the momentum transfer Q is taken by the same scattered
fermion line, while in a exchange process the momentum transfer takes place between two
different fermion lines connected only by the interaction.
(a) (b) (c) (d)
Fig. 6. Particle–particle diagrams renormalizing the BCS vertices at the one-loop level.
The interaction vertices depicted in Figs. 3–5 are all renormalized upon reduction of the
cutoff Λ. This can be traced back to the divergent behavior of the different susceptibilities
of the model. By integration of the high-energy modes in the shells of width dΛ, the
particle–hole susceptibility at momentum Q gets a contribution [38]

c dΛ 1 + 1 − 4(t /t)2
dχph (Q) = , where c ≡ log . (8)
4π 2 t Λ 2t /t
In the same fashion, the contribution to the particle–particle susceptibility at zero total
momentum is
c dΛ
dχpp (0) = 2
log(Λ) . (9)
4π t Λ
In the latter case, the result of the differential integration diverges logarithmically in the
limit Λ → 0. This has been a source of problems in the usual RG analyses of the model.
The definition of the argument in the logarithm needs an additional scale, while a proper
RG scaling requires that the energy is the only dimensionful variable in the problem. It has
to be realized that the coefficient at the right-hand side of Eq. (9) represents actually the
density of states. This has to be born in mind for the correct implementation of the RG
approach, as we will discuss later.
Let us deal first with the renormalization of the vertices with BCS kinematics in Fig. 3.
At the one-loop level, the vertices VI and VU get corrections of order dΛ/Λ from the
diagrams shown in Fig. 6. It is important to realize that these are the only diagrams to
be taken into account to first order in dΛ. There are also corrections from particle–hole
diagrams but, as long as the momentum that goes into the particle–hole loop is not precisely
zero or Q, these terms are of order (dΛ)2 and therefore irrelevant in the low-energy limit,
as shown graphically in Fig. 7.
The BCS vertices mix between themselves alone at the one-loop level, and the situation
is similar in that respect to the general analysis of the 2D Fermi liquid [2]. The degree of
renormalization depends on the density of states n(ε) at the shells integrated out. For later
use, we consider at this point the most general case in which the chemical potential µ does
not coincide from the start with the level of the VHS. The differential RG equations take
Fig. 7. Picture of the high-energy shells of width dΛ at a given saddle-point. The dark regions represent the
contribution to a particle–hole diagram when q is the total incoming momentum.
(a) (b) (c) (d)
Fig. 8. Particle–hole diagrams renormalizing the vertices EQ⊥ and EU ⊥ at the one-loop level.
then the form

∂VI
Λ = cn(µ − Λ) VI2 + VU2 , (10)
∂Λ
∂VU
Λ = 2cn(µ − Λ)VI VU . (11)
∂Λ
These equations were considered in Ref. [25], and they also appear as the leading order in
the RG approach of Ref. [31].
We consider next the renormalization of the vertices EQ⊥ and EU ⊥ , which have also
the property that they mix only between themselves in the one-loop corrections linear in
dΛ. These have been represented in Fig. 8. It can be checked that any other diagrams
give irrelevant contributions of order (dΛ)2 , because they involve either a particle–hole
susceptibility at momentum different from Q or a particle–particle susceptibility with total
momentum different from zero. In the latter case, for instance, it is shown in Fig. 9 that the
number of intermediate states produced by integration of high-energy modes is quadratic,
instead of linear in dΛ.
Fig. 9. Same scheme as in Fig. 7. The dark regions represent the contribution to a particle–particle diagram when
q is the total incoming momentum.
The differential RG equations for the pair of vertices read

2 + E2
EQ⊥
∂EQ⊥ U⊥
Λ = −c , (12)
∂Λ 4π 2 t
∂EU ⊥ EQ⊥ EU ⊥
Λ = −c . (13)
∂Λ 2π 2 t
These equations were obtained in Ref. [25], where the names Uinter and Uumk were used
instead of EQ⊥ and EU ⊥ introduced in the present paper. The same equations also arise at
the dominant level in the functional renormalization of Ref. [31].
We now turn to the rest of the vertices, DQ , DQ⊥ , EQ , DU , DU ⊥ , EU , which
renormalize among themselves at the one-loop level. It is clear that the vertices DQ and
EQ cannot be distinguished from each other just by looking at the external legs. The
same applies to DU and EU . At the one-loop level, one can still discern whether the
momentum transfer Q takes place along the same scattered fermion line or not. However,
the different corrections have to organize so that the above pairs of vertices enter in the
combinations DQ − EQ and DU − EU , which are the quantities that make physical
sense. In that respect, the situation is similar to what happens with the couplings g1 and
g2 in the one-dimensional electron systems [39].
The one-loop renormalization of the vertices provides an explicit proof of the above
statement. The vertex DQ gets linear corrections in dΛ from the diagrams shown in
Fig. 10, while EQ is renormalized by the diagrams shown in Fig. 11. Their RG equations
read then
2 + D 2 + D 2 + D 2 − 2D E
∂DQ DQ U U⊥ Q Q − 2DU EU
= c
Q⊥
Λ 2
, (14)
∂Λ 4π t
2 + E2
EQ
∂EQ U
Λ = −c . (15)
∂Λ 4π 2 t
These two equations can be combined to be written in terms of the physical vertex,
∂(DQ − EQ ) (DQ − EQ )2 + (DU − EU )2 + DQ⊥
2 + D2
U⊥
Λ = c . (16)
∂Λ 4π 2 t
(a) (b)
(c) (d)
Fig. 10. Particle–hole diagrams renormalizing the vertex DQ at the one-loop level.
(a) (b)
Fig. 11. Particle–hole diagrams renormalizing the vertex EQ at the one-loop level.
The RG equations for the remaining vertices also depend on the combinations DQ −
EQ and DU − EU . In the case of DQ⊥ , we have
∂DQ⊥ (DQ − EQ )DQ⊥ + (DU − EU )DU ⊥

Λ = c . (17)
∂Λ 2π 2 t
Finally, the RG equations for DU , DU ⊥ and EU take the form
∂DU DQ DU − DQ EU − DU EQ + DQ⊥ DU ⊥
Λ = c , (18)
∂Λ 2π 2 t
∂EU ⊥ EQ EU ⊥
Λ = −c , (19)
∂Λ 2π 2 t
∂DU ⊥ (DQ − EQ )DU ⊥ + (DU − EU )DQ⊥
Λ = c . (20)
∂Λ 2π 2 t
As a final check, the equation for DU − EU turns out to depend on the physical
combination of couplings
∂(DU − EU ) (DQ − EQ )(DU − EU ) + DQ⊥ DU ⊥
Λ = c . (21)
∂Λ 2π 2 t
4. Universality classes
We discuss now the universality classes in which the system may fall regarding the spin
correlations. We will focus on the analysis of bare repulsive interactions, that is where the
competition between spin and superconducting instabilities arises. We will see that our RG
scheme is able to preserve the spin-rotational invariance of models whose bare interactions
have such a symmetry. This provides another nontrivial check of our RG approach, as our
framework offers the possibility to analyze the scaling of interactions with and without the
SU(2) spin symmetry.
The interactions of physical interest have the property that DQ − EQ = DU − EU
and DQ⊥ = DU ⊥ . These conditions are maintained along the RG flow if they are satisfied
by the bare couplings. Thus, it is useful to work with the set of couplings
D ± ≡ DQ − EQ ± DU − EU , (22)
±
D⊥ ≡ DQ⊥ ± DU ⊥ . (23)
From the results of the preceding section, these new couplings satisfy the equations
± 2 ± 2
∂D ±
D + D⊥
Λ =c , (24)
∂Λ 4π 2 t
∂D⊥± D± D±
⊥
Λ =c . (25)
∂Λ 2π 2 t
The universality classes of the system can be obtained from the integrals of Eqs. (24)
and (25). We stick to the case in which D − = D⊥ −
= 0. The flow for the couplings D +
+
and D⊥ is represented in Fig. 12. Focusing on interactions that are repulsive at the initial
stage of the RG, that is D + > 0 and D⊥ +
> 0, we observe two possible behaviors of the
renormalized couplings. In the case in which the bare couplings satisfy D + D⊥ +
, the flow
is bounded and it converges monotonically to the origin of the space of couplings. If we
start otherwise from a point with D + < D⊥ +
, the flow becomes unstable and it approaches
+ +
a regime in which D → −∞ and D⊥ → +∞.
The regions with stable and unstable flow correspond to respective universality classes,
which imply quite different physical properties. Let us focus, for instance, on the extended
Hubbard model with on-site interaction U and interaction V between nearest-neighbor
sites. The appropriate bare values for the couplings in Figs. 4 and 5 are
DQ = DQ⊥ = U − 4V , (26)
EQ = EQ⊥ = U + αV , (27)
DU = DU ⊥ = U − 4V , (28)
Fig. 12. Flow of the renormalized couplings in the (D + , D⊥

+
) plane.
EU = EU ⊥ = U − βV , (29)
with 0 < α, β < 4. We have for the initial values of the flow D + = −(8 + α − β)V and
+
D⊥ = 2U − 8V . With the physically sensible choice α = β, we see that the attractive or
repulsive character of the on-site interaction dictates whether the RG flow is bounded or
not in the upper half-plane of Fig. 12.
The fact that the flow is not bounded for U > 0 points to the development of some
instability in the system. The divergence of the renormalized couplings represents the
failure to describe the model in terms of the original fermion variables. The underlying
physical effect is the condensation of boson degrees of freedom, as we will show in the
next section. The preservation of the spin-rotational invariance at each step of the RG
process helps to clarify the physical interpretation of the instability and to discern the issue
of the spontaneous breakdown of the symmetry.
We pay attention then to the way in which the SU(2) spin symmetry is preserved in
our RG framework. This can be analyzed by looking at the response functions for the
different components of the spin operator. Since the renormalized interactions grow large
at momentum transfer Q = (π, π), we focus on the correlations of the operator

σσ
Sj (Q) = Ψ+σ (k + Q)σj Ψ σ (k), j = x, y, z. (30)
k
The scaling properties of the response functions can be studied in the same fashion as
for the interacting one-dimensional fermion systems [40]. The response function Rz (ω) for
the Sz (Q) operator, for instance, is renormalized by the diagrams shown in Fig. 13. After
taking the derivative with respect to the cutoff and imposing the self-consistency of the
diagrammatic expansion, we obtain
∂Rz 2c 1 c 1
=− 2 + 2 DQ − EQ + DU − EU − DQ⊥ − DU ⊥ Rz . (31)
∂Λ π tΛ π t Λ
(a) (b)
(c) (d)
(e) (f)
Fig. 13. First-order contributions to the correlator of the Sz operator.
(a) (b)
Fig. 14. First-order contributions to the correlators of the Sx and Sy operators.
The response functions Rx (ω) and Ry (ω) for the other two components of the spin
operator are both renormalized by the diagrams shown in Fig. 14. Following the same
procedure as for Rz (ω), we obtain
∂Rx 2c 1 c 1
=− 2 − 2 (EQ⊥ + EU ⊥ ) Rx , (32)
∂Λ π tΛ π t Λ
and a completely similar equation for Ry (ω).
The response functions Rx (ω), Ry (ω) and Rz (ω) can be made exactly equal if the
equation
DQ − EQ + DU − EU − DQ⊥ − DU ⊥ = −EQ⊥ − EU ⊥ , (33)
is satisfied all along the flow. From Eqs. (12), (13), (24), and (25), we observe that this is
automatically fulfilled when the condition is imposed for the initial values of the couplings.
In the case of the extended Hubbard model, we have indeed for the bare couplings in
Eqs. (26)–(29)
DQ⊥ + DU ⊥ − DQ + EQ − DU + EU = EQ⊥ + EU ⊥ = 2U + (α − β)V .

(34)
The condition is actually satisfied by the couplings of any Hamiltonian that is invariant
under rotations. We show in this way that the SU(2) spin symmetry can be preserved at
each point of the RG flow of the couplings, so that the low-energy effective action keeps
the invariance of the bare Hamiltonian.
5. Spin instability
We proceed to determine the physical properties of the universality class corresponding

to the unstable flow in the upper half-plane of Fig. 12. The divergence of the renormalized
+
couplings D⊥ − D + and EQ⊥ + EU ⊥ results in the divergence of the response functions
Rx , Ry and Rz at a certain value of their argument. This points at the development of an
instability in the spin sector at the corresponding value of the energy measured from the
Fermi level.
The divergence of the response functions implies the existence of a pole at a given
frequency ωc . From the solution to Eqs. (12), (13), (24), and (25), the value of the pole is
given by
+
1 − D⊥ (Λ0 ) − D + (Λ0 ) χph (Q, ωc ) = 0, (35)
+
where D⊥ (Λ0 ) and D + (Λ0 ) are the initial values of the couplings. As long as the
susceptibility χph at momentum Q diverges logarithmically in the low-frequency limit,
it is clear that the above condition is satisfied no matter how small the initial value of the
+
coupling D⊥ − D + may be.
It is important to bear in mind that the susceptibility χph at momentum Q has a finite
imaginary part, which is essential to discern the nature of the ground state of the system.
The imaginary part is computed in Appendix A, and it turns out to be c /(8πt). Eq. (35)
can be written then in the form
+ c
1 − D⊥ (Λ0 ) − D + (Λ0 ) log(iΛ0 /ωc ) = 0, (36)
4π 2 t
which shows that the pole occurs for a pure imaginary value ωc = i|ωc |.
The appearance of a pole in the correlator of a boson operator for a pure imaginary
frequency corresponds to a phenomenon of condensation, in the same fashion as it happens
in the case of a pairing instability [41]. In the present instance, the boson-like object is the
spin operator at momentum Q defined in Eq. (30). The fact that the pole arises at a value
i|ωc | means that the instability pertains actually to the theory posed at finite temperature,
and that there is a transition to a condensed phase at a temperature of the order of magnitude
given by |ωc |.
(a)
(b)
Fig. 15. Self-consistent equations for the dressed propagators in the particle–hole condensate, in terms of the
undressed propagators at the two inequivalent saddle-points.
In our case, the boson operator that acquires a nonvanishing mean value due to the spin
instability is the vector

σσ
d 2 k dω Ψ +
σ (k)σ Ψ σ (k + Q).
This has important consequences, since the diagrammatic approach has to be rebuilt below
the point of the transition, in the same way as in the case of a pairing instability [42].
Let us focus on the Hubbard model, i.e., on a model with interaction between currents
with opposite spin projections. To fix ideas, suppose that the vector S gets the nonzero
mean value pointing in the x direction. Then, there are two different kinds of one-particle
propagators, since the presence of the condensate leads to the consideration of correlators
of the type Ψ + A↑ (k, ω)Ψ B↓ (k, ω), as well as of the usual propagators for well-defined
spin projection near each of the saddle-points. To include all the different possibilities, we
define the propagator Gaσ,bσ (k, ω), with indices a, b labelling the saddle-points and σ, σ
labelling the spin projections:

Gaα,bα (k, ω) = i Ψ +
aσ (k, ω)Ψ bσ (k, ω) . (37)
The Schwinger–Dyson equations for the one-particle propagators take the form shown
graphically in Fig. 15, where the insertion of the wavy line represents the factor

U d 2 k dω Ψ + A↑ (k)Ψ B↓ (k + Q) ≡ ∆. (38)
We have, for instance, the closed set of equations
GA↑,A↑ = G(0) (0)

A↑,A↑ + GA↑,A↑ ∆GB↓,A↑ , (39)
∗
GB↓,A↑ = G(0)
B↓,B↓ ∆ GA↑,A↑ , (40)
where the superindex 0 denotes the corresponding propagator before the introduction of
the condensate. Eqs. (39) and (40) can be combined to give an equation for GA↑,A↑ , which
reads
∗
GA↑,A↑ = G(0) (0) (0)
A↑,A↑ + GA↑,A↑ ∆GB↓,B↓ ∆ GA↑,A↑ . (41)
The solution to Eq. (41) takes the form
G(0)
A↑,A↑ (k, ω)
GA↑,A↑ (k, ω) = (0) (0)
(42)
1 − GA↑,A↑ (k, ω)|∆|2 GB↓,B↓ (k, ω)
in terms of the propagators at the two different saddle-points
1
G(0)
A↑,A↑ (k, ω) = , (43)
ω − εA (k) + i. sgn(ω)
1
G(0)
B↓,B↓ (k, ω) = . (44)
ω − εB (k) + i. sgn(ω)
The important point is to determine the pole structure of the propagator (42). Its frequency
dependence can be expressed in the form
ω − εB (k)
GA↑,A↑ (k, ω) = (45)
(ω − εA (k) + i. sgn(ω))(ω − εB (k) + i. sgn(ω)) − |∆|2
u(k)2 v(k)2
= + (46)
ω − εu (k) + i. sgn(ω) ω − εv (k) + i. sgn(ω)
with appropriate weights u(k)2 , v(k)2 , and εu (k), εv (k) being the roots of the denominator
in Eq. (45)

1
εu,v (k) = εA (k) + εB (k) ± (εA (k) − εB (k))2 + 4|∆|2 . (47)
2
From the physical point of view, the most important feature is the appearance of a gap
in the quasiparticle spectrum near the saddle-points. This can be checked by determining
the shape of the Fermi line, which is given by setting either εu (k) = 0 or εv (k) = 0. Both
conditions lead to the equation
εA (k)εB (k) − |∆|2 = 0. (48)
By recalling that εA (k) = −t− kx2 + t+ ky2 and εB (k) = t+ kx2 − t− ky2 , we end up with the
equation satisfied by the points of the Fermi line
2
t− kx − t+ ky2 t+ kx2 − t− ky2 + |∆|2 = 0. (49)
Solving Eq. (49) for the variable ky2 , for instance, we find that there is a solution only
for values of kx2 such that
2
2 2 4
t+ − t− kx − 4|∆|2t+ t− 0. (50)
Reminding that t± ≈ t ± 2t , this condition implies that, for small values of t , there is a
gap in the spectrum of quasiparticles in the range
2t kx2 |∆|. (51)
We see, therefore, that the gap opens up in the neighborhood

of the saddle-points. The size
of the part of the Fermi line destroyed is bounded by |∆|/t , in units of the inverse lattice
spacing.
The formation of the quasiparticle gap has its origin in the hybridization of modes
at different saddle-points, as a consequence of the enhanced scattering with momentum
transfer exactly equal to Q. Quite remarkably, this is an effect that can be studied in the
weak coupling regime of the model, and the gap appears for arbitrarily small strength
U of the interaction. From the technical point of view, the discussion carried out in this
section parallels the treatment of the one-particle Green functions in the usual description
of the superconducting instability [42]. However, it is clear that the physical setting is
quite different. In the present situation, the condensate is made of particle–hole pairs with
a nonvanishing average projection of the spin. The fact that a macroscopic number of these
pairs has been formed is what forces the quasiparticles to live out of the range already
excited by the condensate.
An important issue concerns the spontaneous breakdown of the spin-rotational symme-
try in the condensate. Let us consider the model at zero temperature regarding this matter.
It is clear that the nonvanishing average spin cannot have in principle any preferred direc-
tion in space. Recalling our definition in Eq. (38), a real value of ∆ implies that the spin of
the condensate points in the x direction, since

d 2 k dω Ψ +A↑ (k)Ψ B↓ (k + Q)

2(∆ + ∆∗ )
+ d 2 k dω Ψ + B↓ (k + Q)Ψ A↑ (k) + A ←→ B = . (52)
U
A purely imaginary value of ∆ implies otherwise that the spin of the condensate lies in the
y direction. Finally, it may also be that the nonvanishing mean value is realized for the z
component of the spin

d 2 k dω Ψ +
A↑ (k)Ψ B↑ (k + Q)

− d 2 k dω Ψ + A↓ (k)Ψ B↓ (k + Q) + A ←→ B = 0. (53)
In the ground state of the model at zero temperature, the spin of the condensate has to
point in a definite direction and the SU(2) rotational symmetry is spontaneously broken.
As a consequence, two Goldstone bosons arise in the spectrum, which correspond to the
spin waves that propagate on top of the particle–hole condensate. These are the gapless
excitations of the model, together with the quasiparticle excitations that exist sufficiently
far away from the saddle-points.
6. Superconducting instability
We now turn to the instability that arises from the divergent flow of Eqs. (10) and (11).
The integral of these equations depends on the position of the chemical potential with
respect to the VHS. For this reason, it is crucial to know how µ depends on the cutoff Λ
as this is progressively lowered.
The issue of the renormalization of the chemical potential has to be treated necessarily
in the framework of the wilsonian RG approach. As the high-energy modes are integrated
out at the scale Λ, µ shifts its position by a quantity proportional to dΛ. At the same time,
it is the chemical potential which sets the level to measure the energy cutoff, as shown
graphically in Fig. 2. The outcome is that µ adjusts itself at each step of the RG process,
until the point in which the cutoff Λ is lowered down to the final chemical potential.
At the computational level, the shift of µ is given by the frequency and momentum-
independent part of the electron self-energy, with intermediate states taken from the high-
energy modes being integrated. The renormalization is proportional to the charge of the
occupied states in the lower slice of width dΛ, which couples through the forward-
scattering vertex F in the usual Hartree and exchange diagrams. The RG equation for
the chemical potential reads
dµ
= F (µ − Λ) n(µ − Λ). (54)
dΛ
The perturbative approach is further improved by incorporating the renormalization of the
F vertex, which bears a well-known dependence on the energy scale measured from the
VHS [26,37]
F0
F (ε) ≈ . (55)
1 − F0 log |ε|/(4π 2 t)
When the density of states n(ε) is a smooth function of the energy, the integration of
high-energy modes produces a steady downward flow of µ. The physical interpretation
of this effect corresponds to the upward displacement of the one-particle levels due to
the repulsive electronic interaction. In the neighborhood of the VHS, the dynamics of µ
becomes highly nonlinear given the singular behavior of the density of states in Eq. (54).
It turns out that, in a certain range of initial values, the chemical potential is renormalized
down to the VHS and precisely pinned to it in the low-energy regime. As stated in the
Introduction, this result pertains to a statistical description in terms of the grand canonical
ensemble. The physical picture is appropriate then for an open system in contact with a
charge reservoir, which sets the bare value µ0 of the chemical potential.
In order to evaluate the influence of the VHS on the renormalization of the chemical
potential, we have solved Eq. (54) with the approximate density of states
c log(t/|ε|)
n(ε) = for |ε| 0.5 t, (56)
4π 2 t
= const for |ε| > 0.5 t. (57)
This expression has the correct normalization for the logarithmic singularity in the 2D
square lattice. The behavior of the integrals of Eq. (54) with such a density of states is
shown in Fig. 16. It is manifest that, for initial values of the chemical potential µ0 t
above the singularity, the final renormalized value of µ lies very close to the VHS. These
results are important to assure that the enhancement of the instabilities due to the divergent
density of states does not rely on fine-tuning the Fermi level to the VHS, as the chemical
potential tends to pin itself in a natural way to the singularity.
Fig. 16. Scaling of the chemical potential as a function of the high-energy cutoff. The results correspond to the
Hubbard coupling U = 4t.
Fig. 17. Flow of the renormalized BCS couplings in the (VI , VU ) plane.
The integrals of Eq. (54) can be used now to find the solutions of Eqs. (10) and
(11) displaying the superconducting instability. The form of the flow in the coupling
constant space is shown in Fig. 17. In the case of bare repulsive interactions, either the
BCS couplings scale to zero for VI > VU , or there is an unstable flow giving rise to the
superconducting instability when VI < VU . The latter instance is realized in lattice models
which have a nearest-neighbor attractive interaction V besides the on-site U repulsive
interaction. When V < 0, the bare coupling VI = U + 4V is obviously smaller than the
bare coupling VU = U − 4V . We are however more interested in the case of the pure
(a) (b)
Fig. 18. Particle–hole corrections to the BCS vertices in the Hubbard model.
Hubbard model, in which the bare couplings lie in the diagonal of the first quadrant in
Fig. 17.
The couplings read directly from the hamiltonian of the Hubbard model correspond to
the boundary between the regions of stable and unstable flow. This means that the slightest
perturbation may drive the system to either of the two sides, which stresses the role played
by the irrelevant operators under these conditions. There are actually perturbations that
fade away when the theory is scaled to low energies, but that may be important because
they may destabilize the flow in the BCS channel.
In the particular case of the Hubbard model, such irrelevant perturbations are given
by the iteration of particle–hole diagrams of the type shown in Fig. 18. Apart from the
particle–particle diagrams, these are the only corrections that arise from the bare couplings
of the model, and they are not enhanced at low energies since the particle–hole bubbles do
not have the appropriate kinematics to be of order ∼ dΛ in the wilsonian approach [2].
The iteration of the bubbles in Fig. 18 gives rise to antiscreening diagrams, i.e., to
corrections that add to the bare repulsive interaction. We recall that the particle–hole bubble
with total momentum about Q is enhanced with the factor c given after Eq. (8), while that
with momentum about the origin is proportional to the factor c given after Eq. (3). As long
as in the present paper we remain in the range t < 0.276 t, we have that c is greater than
c, and we face the instance in which the irrelevant perturbations make VU slightly larger
than VI at the beginning of the RG flow.
We have solved the RG equations (10) and (11) taking as initial values for VI and VU the
result of adding the ladder series built from the diagrams in Fig. 18, with a bare Hubbard
coupling U = 4t. Moreover, in the resolution we have introduced the dependence of µ on
Λ that arises from Eq. (54). This is one of the main accomplishments of our RG procedure,
since the knowledge of how the VHS is approached is essential to regularize the effect of
the divergent density of states.
The results can be synthesized in the determination of the line at which the transition to
the superconducting state takes place in the model. That is characterized by the energy
at which the BCS couplings grow large or, more conveniently, by the point at which
these couplings have a singularity. This depends on the initial position µ0 of the chemical
potential, and it has been represented as a function of this variable in Fig. 19.
We find that the BCS couplings diverge only for values of µ0 in the range of attraction
to the VHS, that is when the renormalized chemical potential is pinned to the singularity.
There is an optimal value of µ0 for which the scale of the transition reaches a maximum,
Fig. 19. Plot of the energy scale of the superconducting instability (thin line) and of the transition to the spin
instability phase (shaded region).
as the chemical potential stays closer to the VHS during a greater part of the RG flow. For
lower values of µ0 , the scale of the instability decreases, as a consequence of the fact that
the renormalized chemical potential is not precisely pinned then to the VHS.
We have also represented in Fig. 19 the energy at which the spin instability opens up,
according to the estimate of Section 5. We realize that this scale is always above the energy
at which the singularity develops in the BCS channel, whenever the spin instability exists
in the system. This happens for values of µ0 higher than the optimal one. For lower values,
the renormalized chemical potential deviates from the VHS by an amount even larger than
the gap that would be due to the spin instability, so that this does not find the conditions to
develop. We have then a picture in which the pairing instability exists alone for µ0 below
the optimal doping, but it is actually precluded above that level since the spin instability
sets in before with the formation of a gap in the quasiparticle spectrum.
We comment finally on the symmetry of the condensate wavefunction. The fact that the
Umklapp interaction VU becomes increasingly repulsive when approaching the instability
implies that the wavefunction must have opposite signs in the saddle-points A and B. As
long as in the unstable flow we approach the asymptotic regime VI = −VU , the response
function for the d-wave operator
Ψ+ + + +
A↑ (k)Ψ A↓ (−k) − Ψ B↑ (k)Ψ B↓ (−k) + h.c. (58)
develops a singularity at the frequency where the coupling VI − VU blows up. By the same
token, it is easily seen that the response function for the s-wave operator does not display
any divergence at low energies. Without the need of knowing precisely the shape of the
gap, we may assure then that the symmetry of the order parameter is of d-wave type, with
nodal lines at the bisectors of the four quadrants. This is in agreement with the results
of more general analyses, which show that the symmetry of the order parameter can be
ascertained from the topology of the Fermi line alone [12].
7. Conclusions
In this paper we have presented a study of the different phases of the system of electrons
interacting near a Van Hove singularity, when the correlations at momentum Q ≡ (π, π)
prevail over those at zero momentum. In the context of a model with nearest-neighbor
and next-to-nearest-neighbor hopping, this happens for 0 < t < 0.276t, according to the
comparison of the prefactors c and c that appear in Eqs. (3) and (8), respectively. We have
applied a wilsonian RG approach following the same lines developed by Shankar in Ref.
[2] for the analysis of Fermi liquid theory. We have paid attention to the spin degrees of
freedom when considering the different interactions, what has allowed us to discern the
universality classes of the system.
We have seen that, regarding the spin correlations, there is a universality class
characterized by a spin instability in the low-energy theory, in opposition to the regime
of couplings with smooth behavior of the correlators for the spin operators. In the case of
the extended Hubbard model with on-site interaction U and nearest-neighbor interaction
V , the spin instability arises for U > 0, irrespective of the value of V , and it is absent for
U < 0.
Several authors have previously considered the competition between the spin and the
superconducting instabilites in the universality class corresponding to the divergent flow
in the upper half-plane of Fig. 12 [15,17,25,26,29–31]. Our analysis has shed light into
a number of features of the spin instability. We have seen that this takes place through
the condensation of particle–hole pairs with momentum Q. The fact that a macroscopic
number of these pairs has been formed is what forces the quasiparticles to live out of the
range ∆ excited by the condensate in the neighborhood of the saddle-points. The Fermi
line is destroyed in a region whose size is bounded by |∆|/t , in units of the inverse
lattice spacing.
This effect provides a paradigm for the disappearance of the Fermi surface of an
electron system which differs from the understanding of such a phenomenon in Mott–
Hubbard insulators. Those systems are supposed to be in a strong-coupling regime, in
which the double occupancy of each lattice site is highly suppressed. In our case, we need
otherwise to constrain the Fermi level near the VHS, departing sensibly from half-filling
as t is increased. Most remarkably, the instability takes place no matter how small the
bare couplings may be in the above picture. This is what ultimately allows to discern the
symmetry breaking in the ground state within our RG approach.
We have seen that two different behaviors arise also in the space of couplings for
the BCS channel, starting from bare repulsive interactions VI and VU . The d-wave
superconducting instability develops in models corresponding to the region with unstable
flow in the upper half-plane of Fig. 17. This is the case of the extended Hubbard model
with U > 0 and attractive interaction V . When U > 0 and the nearest-neighbor interaction
is repulsive, the couplings in the BCS channel scale down to zero. The Hubbard model with
just on-site interaction is placed at first sight on the boundary between the regions with
stable and unstable behavior. We have shown that the model has irrelevant perturbations
that drive the system towards the side with divergent RG flow. Since the departure from
the limit behavior is weak, the superconducting instability is overshadowed by the spin
instability, up to a point of optimal doping beyond which the latter is absent.
The use of the wilsonian RG approach provides some advantages over other RG
methods, the most important being the possibility of studying the renormalization of the
chemical potential. Given the divergent behavior of the density of states at the VHS, it
is clear that all the positions of the Fermi level cannot be equally stable. The scaling of
the chemical potential can be obtained by letting it free to evolve and computing the shift
from the integration of high-energy modes near the cutoff at each RG step. Following
this procedure, we have seen that there is a range of attraction near the VHS where the
chemical potential is renormalized down to the singularity. This guarantees the naturalness
of the different instabilities since, rather than relying on the fine-tuning of the Fermi level,
they arise from its precise pinning to the VHS in the low-energy effective theory.
Appendix A
In this appendix we compute the imaginary part of some of the susceptibilities of the
model. It turns out that the particle–hole susceptibility χph (Q, ω) and the particle–particle
susceptibility χpp (0, ω) have a nontrivial imaginary part, while this vanishes for χph (0, ω)
and χpp (Q, ω) at any finite frequency.
In our model, the susceptibility χph (Q, ω) is given by

dωq d 2q 1
χph (Q, ω) = i
2π (2π)2 ω + ωq − εA (q) + i. sgn(ω + ωq )
1
× , (A.1)
ωq − εB (q) + i. sgn(ωq )
where the energy cutoff is implicit in the integration over the momenta. According to the
standard prescription, the imaginary part of (A.1) is given by

dωq d 2q
Im χph (Q, ω) = −2π 2 sgn(ω + ωq ) sgn(ωq )
2π (2π)2

× δ ω + ωq − εA (q) δ ωq − εB (q) . (A.2)
In the limit of small t , Eq. (A.2) leads to a quantity which does not depend on the
frequency. Taking ω > 0, we have

1
Im χph (Q, ω) = d 2 q δ ω + 2t qx2 − qy2
4π
q0
1 1 ω(t − 2t )
= dqx , where q0 = . (A.3)
8πt qx2 + ω/(2t) 8tt
−q0
After a little of algebra, we obtain

1 t t2 c
Im χph (Q, ω) = log
+ 2
−1 = . (A.4)
8πt 2t 4t 8πt
We see therefore that the imaginary part is equal to π/2 times the prefactor of log(Λ) in
the real part of the susceptibility. It can be checked that the same relation holds between
the real and the imaginary part of the particle–particle susceptibility χpp (0, ω).
Turning now to the susceptibility χph (0, ω), we have

dωq d 2q 1
Im χph (0, ω) = Re
2π (2π)2 ω + ωq − εA (q) + i. sgn(ω + ωq )
1
×
ωq − εA (q) + i. sgn(ωq )

dωq d 2q
= −2π 2 sgn(ω + ωq ) sgn(ωq )
2π (2π)2

× δ ω + ωq − εA (q) δ ωq − εA (q)

1
= − δ(ω) d 2 q. (A.5)
4π
We see then that the imaginary part of the susceptibility is zero for any finite value of the
frequency.
A result similar to (A.5) is obtained for the imaginary part of the susceptibility
χpp (Q, ω). In this channel, the pole that arises after summing up the ladder series
corresponds to the appearance of excited states in the spectrum. We conclude therefore
that the breakdown of symmetry through a mechanism of condensation can only take
place in the particle–particle channel at zero momentum and in the particle–hole channel
at momentum Q, as stated in the text.
References
[1] See, for instance, E. Dagotto, Rev. Mod. Phys. 66 (1994) 763;
P.W. Anderson, The Theory of Superconductivity in the High-Tc Cuprates, Princeton Univ. Press, Princeton,
1997.
[2] R. Shankar, Rev. Mod. Phys. 66 (1994) 129.
[3] J. Polchinski, in: J. Harvey, J. Polchinski (Eds.), Proceedings of the 1992 TASI in Elementary Particle
Physics, World Scientific, Singapore, 1992.
[4] W. Metzner, C. Castellani, C. di Castro, Adv. Phys. 47 (1998) 3.
[5] P.-A. Bares, X.-G. Wen, Phys. Rev. B 48 (1993) 8636.
[6] J. Gan, E. Wong, Phys. Rev. Lett. 71 (1993) 4226.
[7] A. Houghton, H.-J. Kwon, J.B. Marston, R. Shankar, J. Phys. Condens. Matter 6 (1994) 4909.
[8] C. Castellani, C. Di Castro, Physica C 235–240 (1994) 99;
C. Castellani, C. Di Castro, A. Maccarone, Phys. Rev. B 55 (1997) 2676;
C. Castellani, S. Caprara, C. Di Castro, A. Maccarone, Nucl. Phys. B 594 (2001) 747.
[9] C. Nayak, F. Wilczek, Nucl. Phys. B 417 (1994) 359.
[10] J. González, F. Guinea, M.A.H. Vozmediano, Nucl. Phys. B 424 (1994) 595.
[11] S. Chakravarty, R.E. Norton, O.F. Syljuasen, Phys. Rev. Lett. 74 (1995) 1423.
[12] J. González, F. Guinea, M.A.H. Vozmediano, Phys. Rev. Lett. 79 (1997) 3514.
[13] J. Labbé, J. Bok, Europhys. Lett. 3 (1987) 1225;
J. Friedel, J. Phys. (Paris) 48 (1987) 1787;
J. Friedel, J. Phys. (Paris) 49 (1988) 1435;
R.S. Markiewicz, B.G. Giessen, Physica C 160 (1989) 497;
C.C. Tsuei, et al., Phys. Rev. Lett. 65 (1990) 2724;
D.M. Newns, et al., Phys. Rev. Lett. 69 (1992) 1264.
[14] A review of the Van Hove scenario for high-Tc superconductivity has been made by R.S. Markiewicz,
J. Phys. Chem. Solids 58 (1997) 1179.
[15] H.J. Schulz, Europhys. Lett. 4 (1987) 609;
P. Lederer, G. Montambaux, D. Poilblanc, J. Phys. (Paris) 48 (1987) 1613;
J.E. Dzyaloshinskii, Pis’ma Zh. Eksp. Teor. Fiz. 46 (1987) 97, JETP Lett. 46 (1987) 118.
[16] J. González, F. Guinea, M.A.H. Vozmediano, Europhys. Lett. 34 (1996) 711, cond-mat/9502095.
[17] L.B. Ioffe, A.J. Millis, Phys. Rev. B 54 (1996) 3645.
[18] D. Zanchi, H.J. Schulz, Phys. Rev. B 54 (1996) 9509.
[19] D.Z. Liu, K. Levin, Physica C 275 (1997) 81.
[20] P.C. Pattnaik, et al., Phys. Rev. B 45 (1992) 5714.
[21] J. González, F. Guinea, M.A.H. Vozmediano, Nucl. Phys. B 485 (1997) 694.
[22] D. Menashe, B. Laikhtman, Phys. Rev. B 59 (1999) 13592.
[23] G. Kastrinakis, Physica C 340 (2000) 119.
[24] V.Yu. Irkhin, A.A. Katanin, Phys. Rev. B 64 (2001) 205105.
[25] J.V. Alvarez, J. González, F. Guinea, M.A.H. Vozmediano, J. Phys. Soc. Jpn. 67 (1998) 1868.
[26] J. González, F. Guinea, M.A.H. Vozmediano, Phys. Rev. Lett. 84 (2000) 4930.
[27] W. Kohn, J.M. Luttinger, Phys. Rev. Lett. 15 (1965) 524.
[28] For a review, see M.A. Baranov, A.V. Chubukov, M.Yu. Kagan, Int. J. Mod. Phys. B 6 (1992) 2471.
[29] C.J. Halboth, W. Metzner, Phys. Rev. B 61 (2000) 7364;
C.J. Halboth, W. Metzner, Phys. Rev. Lett. 85 (2000) 5162.
[30] C. Honerkamp, M. Salmhofer, N. Furukawa, T.M. Rice, Phys. Rev. B 63 (2001) 35109.
[31] B. Binz, D. Baeriswyl, B. Douot, Eur. Phys. J. B 25 (2002) 69.
[32] R.S. Markiewicz, J. Phys. Condens. Matter 2 (1990) 665.
[33] D.M. Newns, P.C. Pattnaik, C.C. Tsuei, Phys. Rev. B 43 (1991) 3075.
[34] V.Yu. Irkhin, A.A. Katanin, M.I. Katsnelson, Phys. Rev. B 64 (2001) 165107.
[35] C. Honerkamp, M. Salmhofer, Phys. Rev. Lett. 87 (2001) 187004.
[36] S. Sorella, R. Hlubina, F. Guinea, Phys. Rev. Lett. 78 (1997) 1343.
[37] J. González, Phys. Rev. B 63 (2001) 45114.
[38] H.Q. Lin, J.E. Hirsch, Phys. Rev. B 35 (1987) 3359.
[39] J. Sólyom, Adv. Phys. 28 (1979) 201.
[40] H.J. Schulz, in: V.J. Emery (Ed.), Correlated Electron Systems, Vol. 9, World Scientific, Singapore, 1993.
[41] A.A. Abrikosov, L.P. Gorkov, I.E. Dzyaloshinski, Methods of Quantum Field Theory in Statistical Physics,
Dover, New York, 1975, Chapter 7.
[42] E.M. Lifshitz, L.P. Pitaevskii, Statistical Physics, Part 2, Pergamon Press, Oxford, 1980, Chapter 5.
Correlation functions of the XXZ spin- 12 Heisenberg

chain at the free fermion point from their multiple
integral representations
N. Kitanine a,1 , J.M. Maillet b , N.A. Slavnov c , V. Terras d,2
a Graduate School of Mathematical Sciences, University of Tokyo, Japan
b Laboratoire de Physique, UMR 5672 du CNRS, ENS Lyon, France
c Steklov Mathematical Institute, Moscow, Russia
d NHETC, Department of Physics and Astronomy, Rutgers University, USA
Received 3 May 2002; accepted 7 August 2002
Abstract
Using multiple integral representations, we derive exact expressions for the correlation functions
of the spin- 12 Heisenberg chain at the free fermion point ∆ = 0.
PACS: 71.45.G; 75.10.Jm; 11.30.Na; 03.65.Fd
1. Introduction
In the article [1] new multiple integral representations for the correlation functions of
the XXZ spin- 21 Heisenberg chain have been obtained. In the present article, we apply the
results of [1] to compute the correlation functions of the spin- 21 Heisenberg chain at the
free fermion point ∆ = 0.
E-mail addresses: kitanine@ms.u-tokyo.ac.jp (N. Kitanine), maillet@ens-lyon.fr (J.M. Maillet),

nslavnov@mi.ras.ru (N.A. Slavnov), vterras@physics.rutgers.edu (V. Terras).
1 On leave of absence from Steklov Institute at St. Petersburg, Russia.
2 On leave of absence from LPMT, UMR 5825 du CNRS, Montpellier, France.
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 8 0 - 6
434 N. Kitanine et al. / Nuclear Physics B 642 [FS] (2002) 433–455
Generically, the Hamiltonian of the finite cyclic XXZ chain with M sites (where M is
supposed to be even) has the form
M

y y z h
HXXZ = σmx σm+1
x
+ σm σm+1 + ∆ σmz σm+1 − 1 − σmz . (1.1)
2
m=1
x,y,z
Here, σm denote the local spin operators (Pauli matrices) associated with the mth site
of the chain, ∆ is the anisotropy parameter, and h an external classical magnetic field. The
specialization of this model to the case ∆ = 0 is known as the XX chain (isotropic XY
model [2]):
M

y y h z
HXX = σm σm+1 + σm σm+1 − σm .
x x
(1.2)
2
m=1
In spite of the fact that the XX spin- 21 chain is equivalent to a model of free fermions,
its correlation functions are quite non-trivial. They had been studied for a long period by
numerous authors [3–8] and the key results in this field are presently known. It is worth
mentioning however that the methods used in the works listed above rely essentially on
the free fermion features of the XX model. Therefore, they cannot be applied to the more
general XXZ case, at least without significant modifications. On the contrary, we expect
our present approach, which relies on multiple integral representations of correlation
functions, to be instructive for the study of the general case as well.
In 1992 [9], multiple integral representations for the correlation functions of the XXZ
chain at zero temperature, ∆ > 1 and h = 0 have been obtained from the q-vertex operator
approach. Later, in 1996 [10] (see also [11]), similar answers were formulated for the case
|∆| 1. A proof of these formulas, together with their extension to non-zero magnetic
field, has been obtained in 1999 [12,13] for both regimes using algebraic Bethe ansatz
and the actual resolution of the quantum inverse scattering problem [13,14]. It results from
these articles that, starting from the so-called elementary blocks, one can in principle obtain
a multiple integral representation for any n-point correlation function of the XXZ chain.
,
More precisely, if |ψg denotes the ground state, and Emm m the elementary operators
,
acting on the quantum space Hm at site m as the 2 × 2 matrices Elk = δl, δk, , the
elementary blocks for correlation functions are defined as
m j ,j

ψg | j =1 Ej |ψg
Fm {j , j } = . (1.3)

ψg |ψg
The methods developed in [9–12] enable one to obtain the quantity (1.3) in terms of an
integral with m integrations. It is clear that an arbitrary correlation function in the ground
state can be expressed as a linear combination of the elementary blocks (1.3) and, hence, as
a linear combination of such multiple integrals. It should be stressed however that, although
these formulas are quite explicit, the actual analytic computation of these multiple integrals
has been missing up to now. Moreover, the evaluation of correlations of physical relevance,
like spin–spin correlation functions, is a priori quite involved. Indeed, if we consider, for
N. Kitanine et al. / Nuclear Physics B 642 [FS] (2002) 433–455 435
example, the correlation function

σ1z σmz , the identity

m−1 11

ψg |σ1z σmz |ψg ≡
ψg | E111 − E122 Ej11 + Ej22 Em − Em
22
|ψg (1.4)
j =2
shows that the corresponding linear combination of elementary blocks is actually given as
a sum of 2m terms. This means that the number of terms to sum up grows exponentially
with m, which in particular makes it extremely difficult to solve the problem of asymptotic
behavior at large distance.
Thus, up to recently, the situation in this field was as follows. On one hand, the free
fermion limit (∆ = 0) of the XXZ model was well studied, but the extension of these
results to the general case came up against serious problems. On the other hand, the
multiple integrals approach formally provided the possibility to compute the correlation
functions for arbitrary ∆; however, because of the technical reasons mentioned above, no
result has up to now been reproduced via this method, even for the simplest case of free
fermions.3
The goal of this paper is to study the correlation functions of the XX chain using the
new multiple integral representations obtained in [1]. In fact, these new representations are
nothing but re-summations of the multiple integral expressions for the elementary blocks.
In particular, they enable us to present the spin–spin correlation functions of the type (1.4)
as a sum of only m terms instead of 2m . We would like to point out that in this paper we do
not obtain new results, but only reproduce the known answers via a new method. Moreover,
we consider here the XX model only as a test for the relevance of the formulas obtained in
[1]. We hope that some of the methods developed in the present publication can be applied
(perhaps after certain modifications) to the general XXZ chain as well.
This article is organized as follows. In the next section, we introduce some useful
notations and give the list of formulas obtained in [1] for the correlation functions of the
XXZ model. In Section 3, we compute the correlation function of the third components
of spin. The emptiness formation probability is considered in Section 4. In Section 5,
we obtain a Fredholm determinant representation for the correlation function
σ1+ σm+1 −
.
The asymptotic analysis of this Fredholm determinant is performed in Section 6. Some
perspectives are discussed in the conclusion.
2. Correlation functions of the XXZ chain
For the reader’s convenience, we gather in this section the list of results obtained in
[1] for the correlation functions of the XXZ model. Since eventually we study the limit
∆ = 0, we hereafter restrict ourselves to the case |∆| < 1.
3 Recently, in [15], the probability to find in the ground state a string of particles with spin down (the emptiness
formation probability) was computed at ∆ = 0 by the method of multiple integrals. Let us stress that, contrary to
the spin–spin correlation functions, this quantity can be expressed as a single elementary block. We consider the
emptiness formation probability in Section 4 of the present article.
Let us first of all recall that the Hamiltonian (1.1) possesses the symmetries

M/2
U HXXZ (∆, h)U −1 = −HXXZ (−∆, h), U= z
σ2m ,
m=1

M
V HXXZ (∆, h)V −1 = HXXZ (∆, −h), V= σmx . (2.1)
m=1
Due to this freedom, there is no common definition of the XXZ model. It means that the
expressions of correlation functions obtained in different publications may coincide up to
a common sign and/or the sign of ∆ and h.
The standard parameterization of the anisotropy parameter is ∆ = cosh η. In the regime
|∆| < 1 the parameter η is imaginary, and we set η = −iζ , ζ > 0. The free fermion point
∆ = 0 corresponds to ζ = π/2.
The general structure of the expressions obtained in [1] for the spin–spin correlation
functions is the following:

m−1

β
σ1α σm+1 = d n+1 z d nλ d 2 µ f m ({λ, z})Γnαβ ({λ, µ, z})Sh ({λ, z}). (2.2)
n=0 C Cλ Cµ
z
αβ
Here α, β = x, y, z, and the functions Γn ({λ, µ, z}) and f ({λ, z}) are purely algebraic
quantities, which in particular do not depend on the regime nor on the magnetic field.
Their explicit forms for specific correlation functions are given below.
The integration contour Cz surrounds the point zj = −iζ /2 (zj = −iπ/4 for ∆ = 0),
where the function f ({λ, z}) has a pole. All other singularities of the integrand (2.2) are
outside the contour Cz .
The contours Cλ and Cµ depend on ∆ and h. In all the examples considered below
we have Cλ = Cµ . For |∆| < 1 and h 0, the contour Cλ is an interval [−Λ, Λ] of the
real axis, where the value of Λ is uniquely defined by ∆ and h, although in the general
case the dependency Λ = Λ(∆, h) is rather implicit. At ∆ = 0, however, the integration
domain can be found explicitly from the fact that cosh 2Λ = 4/ h. Note that if h → 0, one
has Λ → ∞. On the other hand, if h approaches its critical value hc = 4, then Λ → 0 and
all the correlation functions become trivial, which comes physically from the fact that the
ground state of the Hamiltonian (1.2) is then purely ferromagnetic. Therefore we consider
below only the case 0 h hc . Due to the symmetry (2.1), we also do not need to consider
negative magnetic field, although all our results remain valid for h < 0 as well.4
Finally, the integrand (2.2) contains a function Sh ({λ, z}), which also depends on the
value of the magnetic field. This function is equal to the determinant of a matrix of elements
ρ(λj , zk ), where ρ(λ, z) is the so-called ‘inhomogeneous density’, solution of the integral
4 Actually the restriction h 0 is not necessary. In the case h < 0, the integration contour C becomes
λ
[−Λ + iπ iπ iπ iπ
2 , −∞ + 2 ] ∪ R ∪ [+∞ + 2 , Λ + 2 ], where Λ is the real positive solution of the equation
cosh 2Λ = 4/|h|.
equation
Λ
−2πiρ(λ, z) + K(λ − µ)ρ(µ, z) dµ = t (λ, z), (2.3)
−Λ
with
i sin 2ζ −i sin ζ
K(λ) = , t (λ, z) = .
sinh(λ + iζ ) sinh(λ − iζ ) sinh(λ − z) sinh(λ − z − iζ )
(2.4)
Note that, at z = −iζ /2, the function ρ(λ, z) coincides with the spectral density of the
ground state. In the free fermion limit ∆ = 0 (ζ = π/2), one has K(λ) = 0, and thus
i i
ρ(λ, z) = t (λ, z) = . (2.5)
2π π sinh 2(λ − z)
After this general setting, let us now be more specific and present the explicit formulas
for some of the correlation functions of the XXZ chain in the domain |∆| < 1. We consider
below essentially three different cases:
zz =
σ z σ z
(a) the correlation function gm 1 m+1 in a magnetic field;
(b) the emptiness formation probability τ (m) in a magnetic field;
(c) the correlation functions gm+− =
σ + σ − in zero magnetic field.
1 m+1
The reader can find the derivation of the corresponding multiple integral representations
in [1].
(a) The correlation function of the third components of spin canz be evaluated from the
generating functional
exp(βQ1,m ), where Q1,m = 12 m k=1 (1 − σk ), as

z z 2 ∂
2 ∂
σ1 σm+1 = 2Dm 2 − 4Dm + 1
exp(βQ1,m )
∂β ∂β β=0
2
= 2Dm Q1,m − 4Dm
Q1,m + 1.
2
(2.6)
Here, the symbols Dm and Dm
2 denote, respectively, the first and the second lattice
derivative,
Dm f (m) ≡ f (m + 1) − f (m),
Dm
2
f (m) ≡ f (m + 1) + f (m − 1) − 2f (m). (2.7)
The expectation value of the functional
exp(βQ1,m ) is given by (5.8) of [1]:

m

n Λ n
iζ iζ m
1 dzj sinh(za − 2 ) sinh(λa + 2)

exp(βQ1,m ) = n
d λ iζ iζ
(n!)2 2πi sinh(za + −
n=0 Cz j =1 −Λ a=1 2 ) sinh(λa 2)
j k ({λ}|{z})]detn [ρ(λj , zk )].

× Wn ({λ}|{z})detn [M (2.8)
Here and further we use the notation detn for the determinants of n × n matrices. The
function Wn is defined by

n
n
sinh(λa − zb − iζ ) sinh(zb − λa − iζ )
Wn ({λ}, {z}) = , (2.9)
sinh(λa − λb − iζ ) sinh(za − zb − iζ )
a=1 b=1
j k are
and the entries of the matrix M
n
sinh(λa − λj − iζ ) sinh(λj − za − iζ )
j k ({λ}|{z}) = t (zk , λj ) + eβ t (λj , zk )
M ,
sinh(λj − λa − iζ ) sinh(za − λj − iζ )
a=1
(2.10)
where the functions t (λ, z) and ρ(λ, z) are defined in (2.4) and (2.3), respectively.
For h = 0, it is more convenient to derive the correlation function
σ1z σm+1
z
from the
generating functional
exp(βQ1,m )σm+1 :
z

z z ∂
σ1 σm+1 = −2Dm−1 exp(βQ1,m )σm+1
z
, h = 0. (2.11)
∂β β=0
z
The expectation value of the functional
exp(βQ1,m )σm+1 is given by (6.7) of [1]:
z
exp(βQ1,m )σm+1
m
n n
1 dzj sinh(za − iζ2 ) sinh(λa + iζ2 ) m
= d n+1
λ iζ iζ
(n!)2
a=1 sinh(za + 2 ) sinh(λa − 2 )
2πi
n=0 Cz j =1 R
n
sinh(λn+1 − za ) sinh(λn+1 − za − iζ )
n n
sinh(λa + iζ2 )
× −
iζ sinh(λn+1 − λa ) sinh(λn+1 − λa − iζ )
a=1 sinh(za + 2 ) a=1 a=1

× Wn ({λ}|{z})detn M j k ({λ}|{z})

× detn+1 ρ(λj , z1 ), . . . , ρ(λj , zn ), ρ λj , − iζ2 . (2.12)
(b) The emptiness formation probability (the probability to find in the ground state a
string of particles with spin down in the first m sites) is defined as
m
1 − σz
τ (m) = k
. (2.13)
2
k=1
The multiple integral representation of this correlation function can be easily obtained from
the generating functional
exp(βQ1,m ) (see Eq. (C.9) of [1]):
Λ
1 Zm ({λ}, {ξ })
τ (m) = lim m detm ρ(λj , ξk ) d m λ, (2.14)
ξ1 ,...,ξm →− iζ2 m! a<b sinh(ξa − ξb )
−Λ
where
m m
sinh(λa − ξb ) sinh(λa − ξb − iζ ) detm t (λj , ξk )
Zm ({λ}, {ξ }) = t m .
sinh(λa − λb − iζ ) a>b sinh(ξa − ξb )
a=1 b=1
(2.15)
(c) Let us consider finally the correlation functions gm +− =
σ + σ − and g −+ =
1 m+1 m
σ1− σm+1
+
. +− −+
It is easy to see that gm (h) = gm (−h). For zero magnetic field, these two
correlation functions coincide. In this case their multiple integral representation is given
by (6.13) of [1]:
+ −
σ1 σm+1

m−1
n+1
dzj
1
= d n+2 λ
n!(n + 1)! 2πi
n=0 Cz j =1 R

n+1
sinh(za − iζ m
n
sinh(λa + iζ m
2) 2)
× iζ iζ
a=1 sinh(za + 2) a=1 sinh(λa − 2)

n+1
1 sinh(λn+1 − z a − iζ ) sinh(λn+2 − z a )
× na=1
sinh(λn+1 − λn+2 ) a=1 sinh(λn+1 − λa − iζ ) sinh(λn+2 − λa )

×W j k detn+2 ρ(λj , z1 ), . . . , ρ(λj , zn+1 ), ρ λj , − iζ .
n ({λ}, {z})detn+1 M
2
(2.16)
Here,
n+1
n
n ({λ}, {z}) = a=1 b=1 sinh(λa − zb − iζ ) sinh(zb − λa − iζ )
W n n n+1 n+1 ,
a=1 b=1 sinh(λa − λb − iζ ) a=1 b=1 sinh(za − zb − iζ )
(2.17)
are
and the entries of the (n + 1) × (n + 1) matrix M

n
sinh(λa − λj − iζ ) sinh(λj − zb − iζ )
n+1
j k = t (zk , λj ) − t (λj , zk )
M ,
sinh(λj − λa − iζ ) sinh(zb − λj − iζ )
a=1 b=1
j n,

iζ
Mn+1,k = t zk , − , j = n + 1. (2.18)
2
To conclude this section, let us recall once more that all the multiple integral
representations given above hold for arbitrary −1 < ∆ < 1. In order to specialize these
expressions to the case of the XX model, one has to set ζ = π/2, Λ = arccosh(4/ h)/2,
and to use the expression (2.5) for the inhomogeneous density ρ(λ, z).
3. Correlation function of the third components of spin
Let us begin our calculations with the correlation function

σ1z σm+1z
, which is the
simplest spin–spin correlation function of the XX model.
j k
Observe that, at ζ = π/2, Eq. (2.8) simplifies drastically. First of all, the matrix M
(2.10) becomes proportional to the Cauchy matrix
j k ({λ}|{z}) = 2(eβ − 1) π
M , ζ= , (3.1)
sinh 2(λj − zk ) 2
and, hence, to compute its determinant one can use the identity
n
1 j >k sinh(xj − xk ) sinh(yk − yj )
detn = n . (3.2)
sinh(xj − yk ) j,k=1 sinh(xj − yk )
However, it is more important to notice that detn M j k is proportional to (eβ − 1)n : this
means that, if one takes the first (respectively, the second) derivative with respect to β and
sets β = 0 as in (2.6), only the terms n 1 (respectively, n 2) in the sum (2.8) do not
vanish. Thus, after some simple computation, one obtains

Λ
1 dλ

Q1,m = dz ϕ m (z)ϕ −m (λ) , (3.3)
4π 2 sinh2 (λ − z)
Cz −Λ

Λ
2
1
Q21,m =
Q1,m + 2
d z d 2λ ϕ m (za )ϕ −m (λa )
32π 4
Cz −Λ a=1
2
1
× det2 , (3.4)
sinh(λj − zk )
where we have introduced the notation
sinh(z − iπ
4 )
ϕ(z) = . (3.5)
sinh(z + iπ
4 )
Let us first consider the integral (3.3). As the contour Cz surrounds only the singularity
z = −iπ/4, where ϕ(z) admits a pole of order m, the value of the z-integral in (3.3)
is given by the corresponding residue at z = −iπ/4. However, this way to compute the
z-integral is not very convenient, especially for large m. Instead, we suggest to deform the
original contour Cz into an infinite horizontal strip of boundary Γ given by z = z0 − π
and z = z0 , where 0 < z0 < 3π/4. Then, obviously,

ϕ m (z) dz ϕ m (z) dz ϕ m (z)
= − 2πi Res . (3.6)
sinh2 (λ − z) sinh2 (λ − z) sinh2 (λ − z) z=λ
Cz Γ
Since the integrand is a periodic function with period iπ , and since it vanishes at z → ±∞,
it is clear that the integral with respect to the new contour Γ is equal to zero. Thus, to
compute the z-integral in (3.3), it is enough to take the residue in the second order pole at
z = λ. The remaining integral with respect to λ is then trivially computable, and we obtain
m

Q1,m = arctan(sinh 2Λ). (3.7)
π
In the next sections, we shall deal with the change of integration variables cosh 2λ =
(cos p)−1 . Therefore, let us at this stage introduce p0 such that cosh 2Λ = (cos p0 )−1 .
Then (3.7) takes the form
mp0

Q1,m = , (3.8)
π
where p0 = arccos(h/4).
The integral (3.4) can be taken in the same manner. The integration with respect to z1
and z2 leads to

2 mp0 mp0 2
Q1,m = +
π π
m
Λ m m m
1 ϕ 2 (λ1 )ϕ − 2 (λ2 ) − ϕ − 2 (λ1 )ϕ 2 (λ2 ) 2
+ dλ1 dλ2 . (3.9)
4π 2 sinh(λ1 − λ2 )
−Λ
In fact, we do not need to compute

Q21,m itself, but only its second lattice derivative.
Differentiating (3.9) with respect to m, we immediately arrive at
2
2 p0 1
Dm 2
Q1,m = 2 − 2 2 (1 − cos 2mp0 ). (3.10)
π π m
Combining (3.8) and (3.10), we finally obtain
2
z z 2p0 2
σ1 σm+1 = − 1 − 2 2 (1 − cos 2mp0 ). (3.11)
π π m
This result was obtained in [16]. It is worth mentioning that, in spite of the fact that
we have formally restricted ourselves to the case h 0, the result (3.11) remains valid
for h < 0 as well. For zero magnetic field, p0 = π/2, and the constant contribution to
the correlation function disappears. In order to get rid of this constant term from the
very beginning, it is more convenient to derive
σ1z σm+1
z
from the generating functional
z
exp(βQ1,m )σm+1 (see (2.12)). Then, for ∆ = 0, the corresponding sum reduces to the
single term n = 1, which gives

z z 2i dλ1 dλ2 ϕ m (z)ϕ −m (λ1 )
σ1 σm+1 = 3 dz . (3.12)
π cosh 2λ2 cosh 2λ1 sinh 2(λ2 − z)
Cz R
Using the method of calculations described above, we find

2
σ1z σm+1
z
= (−1)m − 1 , at h = 0. (3.13)
π 2 m2
4. Emptiness formation probability
The emptiness formation probability (2.13) constitute one of the simplest examples of
correlation functions. In particular, unlike the spin–spin correlation functions studied in
Sections 3, 5 and 6, it can be directly expressed as a single elementary block of the form
(1.3), therefore, as a single (multiple) integral which can be written in the symmetric form
(2.14). In this section, we explain how to compute this integral in the case ∆ = 0, and how
to analyze its asymptotic behavior in the limit m → ∞. Note that for ∆ = 1/2, we have
also been able to compute this quantity explicitly for any values of m [17].
Setting ζ = π/2 in Eqs. (2.14), (2.15), we have

Λ
1 Zm ({λ}, {ξ }) i
τ (m) = lim m detm d m λ,
ξ1 ,...,ξm →− iπ
4
m! a<b sinh(ξa − ξb ) π sinh 2(λj − ξk )
−Λ (4.1)
and
m m
sinh 2(λa − ξb )
Zm ({λ}, {ξ }) = 2−m
2
cosh(λa − λb )
a=1 b=1

m
−1 2
× sinh (ξa − ξb )detm . (4.2)
sinh 2(λj − ξk )
a>b
Once again, we have to deal with determinants of Cauchy matrices. Computing them via
(3.2) and setting ξj = −iπ/4, we obtain
2 Λ m
a>b sinh (λa − λb ) m
2
2m
τ (m) = m m d λ. (4.3)
m!(2π)m a=1 cosh (2λa )
−Λ
The representation (4.3) can be reduced to a Toeplitz determinant. Indeed, the change of
variables cosh 2λj = (cos pj )−1 leads to
p0
m p0
2m −m
2
(pa − pb ) m 1
τ (m) = sin 2
d p= ∆ e−ip ∆ eip d m p,
m!(2π)m 2 m!(2π)m
−p0 a>b −p0
(4.4)
where ∆(e±ip )
denote Van der Monde determinants of variables Due to the e±ipj .
symmetry of the integrand with respect to all pj , one can replace one of these Van der
Monde determinants with the product of its diagonal elements multiplied by m!, which
gives us
p0
m
1
τ (m) = e−i(k−1)pk detm ei(j −1)pk d m p
(2π)m
−p0 k=1
p0
1 i(j −k)p
= detm e dp . (4.5)
2π
−p0
The representation (4.5) of τ (m) as a Toeplitz determinant has already been obtained in
[15] from the multiple integral representation given in [12].
Thus, (4.5) provides an explicit expression of the emptiness formation probability, at
least if m is small enough. However, it is more important to be able to extract the asymptotic
behavior of τ (m) at m → ∞. There exist several ways to do this. Firstly, one can analyze
the determinant (4.5) as in [15]. Secondly, the determinant (4.5) can be transformed to a
Fredholm determinant of a linear integral operator [7], the asymptotic behavior of which
can be evaluated from the matrix Riemann–Hilbert problem [18]. Here we propose a third
approach, based on the application of the saddle point method directly to (4.3). It is possible
that this method can be used also for the general XXZ model.
Let us rewrite (4.3) in the following way:
2
2m 2
τ (m) = m
em S({λ}) dλ, (4.6)
(2π)
D
where
1 1
m m
S({λ}) = log sinh 2
(λa − λb ) − log cosh 2λa , (4.7)
m2 m
a>b a=1
and the domain D is defined by −Λ λ1 · · · λm Λ. The integrand in (4.6) is
positive within the domain D and vanishes on its boundary. Moreover, it is not difficult to
check that the matrix of the second derivatives ∂ 2 S/∂λj ∂λk is negatively defined. Hence,
the integrand has a unique maximum in D, which is given by the system
2
m
∂S
m = coth(λj − λa ) − 2 tanh 2λj = 0. (4.8)
∂λj m a=1
a=j
Following the standard arguments of the saddle point method, we assume that, in the limit
m → ∞, the solutions of the system (4.8) are distributed on the interval [−Λ, Λ] according
to a certain density ρ0 (λ). Then, in this limit, (4.8) becomes an integral equation for this
density:
Λ
tanh 2λ = P .V . coth(λ − µ)ρ0 (µ) dµ. (4.9)
−Λ
In its turn, the integral (4.6) can be approximated by the value of the integrand in the saddle
point:
2 2S
τ (m) → 2m em 0 , m → ∞, (4.10)
where
Λ Λ
1
S0 = log sinh (λ − µ)ρ0 (λ)ρ0 (µ) dλ dµ −
2
log cosh 2λρ0 (λ) dλ. (4.11)
2
−Λ −Λ
The analytic expression of the function ρ0 (λ) can be determined as follows. Setting
x = e2λ , we transform (4.9) into
b
x dy
= P .V . ρ̂0 (y), (4.12)
x2 + 1 x −y
a
1 −2λ
where ρ̂0 (x) = 2e ρ0 (λ)and a = e−2Λ , b = e2Λ . The solution of the singular integral
equation (4.12) can be obtained in a standard way via the scalar Riemann–Hilbert problem.
Let us define
b
dy
f± (x) = ρ̂0 (y). (4.13)
x − y ± i0
a
Then we have
b
2πi ρ̂0 (x) = f− (x) − f+ (x), ρ̂0 (x) dx = 1. (4.14)
a
At the same time, f (x) satisfies the relation
2x
f− (x) + f+ (x) = , (4.15)
x2 + 1
and, hence,
b
1 dy y
f (x) = f (x) C +
0
, (4.16)
πi y − x (y 2 + 1)f+0 (y)
a
where f 0 (x) = ((x − a)(x − b))−1/2 and C is a constant. Substituting this expression into
(4.14), we eventually obtain

1 x+1 a+b
ρ̂0 (x) = , (4.17)
π x2 + 1 2(x − a)(b − x)
which results into

1 cosh λ cosh 2Λ
ρ0 (λ) = . (4.18)
π cosh 2λ sinh(Λ − λ) sinh(Λ + λ)
This enables us to obtain the analytic expression of S0 in terms of Λ:

√
cosh 2Λ
S0 = − log 2 . (4.19)
sinh Λ
Taking into account that cosh 2Λ = 4/ h, we finally obtain the following asymptotic
equivalent of τ (m) in terms of the magnetic field h:
m2
4−h 2
τ (m) → , m → ∞. (4.20)
8
Thus, this method provides an alternative derivation of the asymptotic behavior of the
emptiness formation probability [18].
5. Correlation function σ1+ σm+1

− as Fredholm determinant
Unlike the correlations of the third components of spin (see Section 3), the correlation
function
σ1+ σm+1
−
remains non-trivial even at the free fermion point ∆ = 0. In this
section, we show how to compute it from (2.16) for zero magnetic field.
Let us first consider the part of the integrand (2.16) which depends on λn+1 and λn+2 :

n+1
detn+2 ρ(λj , zk ) a=1 sinh(λn+1 − za − iζ ) sinh(λn+2 − za )
n ,
sinh(λn+1 − λn+2 ) a=1 sinh(λn+1 − λa − iζ ) sinh(λn+2 − λa )
where one should set zn+2 ≡ −iζ /2 in the last column of the determinant of densities. We
can shift the integration contour for λn+2 by −iζ , and then replace λn+2 by λn+2 − iζ in
the integrand. This changes the sign of the density function, and we obtain

n+1
detn+2 ρ(λj , zk ) sinh(λn+1 − za − iζ ) sinh(λn+2 − za − iζ )
− na=1 .
sinh(λn+1 − λn+2 + iζ ) a=1 sinh(λn+1 − λa − iζ ) sinh(λn+2 − λa − iζ )
We see that for ζ = π/2 the integrand becomes an antisymmetric function of λn+2 and
λn+1 and, hence, the corresponding integral vanishes. It remains then to take into account
the contribution of the poles which have been crossed during the shift of the λn+2 -contour.
It is easy to see that we have crossed only one singularity, which corresponds to the pole
of the function ρ(λn+2 , − iζ2 ) at λn+2 = − iζ2 . The residue in this point gives

σ1+ σm+1
−

m−1
n+1
dzj
1
= d n+1 λ
n!(n + 1)! 2πi
n=0 Cz j =1 R

n+1
sinh(za − iζ m n
+ iζ m
2) sinh(λa 2)
× iζ iζ
a=1 sinh(za + 2) a=1 sinh(λa − 2)
n+1 iζ

1 a=1 sinh(λn+1 − za − iζ ) sinh(za + 2)
× iζ n iζ
sinh(λn+1 + 2) a=1 sinh(λn+1 − λa − iζ ) sinh(λa + 2 )
n ({λ}, {z})detn+1 M
×W j k detn+1 [ρ(λj , zk )]. (5.1)
and (ρ(λj , zk )) become Cauchy matrices at
At this stage, we can again use the fact that M
ζ = π/2. To compute their determinants, it is convenient to use the following modification
of (3.2):
n
1 j >k cosh(xj − xk ) cosh(yk − yj ) 1
detn = detn . (5.2)
sinh 2(xj − yk ) 2n nj,k=1 cosh(xj − yk ) sinh(xj − yk )
j k and detn+1 ρ(λj , zk ) into (5.1),

Substituting the corresponding expressions of detn+1 M
we arrive at

σ1+ σm+1
−
n+1
1
m−1 n
1 i 1
= d n+1 λ ϕ −m+1 (λa )
2i n!(n + 1)! π sinh(λn+1 + iπ
4 )
n=0 R a=1
 1 
sinh(zk −λ1 )

n+1  
dzj n+1  .. 
1  . 
× ϕ m−1
(za )detn+1 detn+1  .
2πi sinh(λj − zk )  1 
Cz j =1 a=1  sinh(z k −λ n ) 
1
sinh(zk + iπ
4 )
(5.3)
As we have seen above, the correlation function
σ1z σm+1
z

and the emptiness formation
probability τ (m) at ∆ = 0 are completely described by only one or two terms at ∆ = 0. The
peculiarity of the correlation function
σ1+ σm+1
−
is that, even in the limit of free fermions,
all the terms of the corresponding series survive. Nevertheless, it is possible to express
(5.3) into a more compact form.
The contour integrals with respect to zk in (5.3) can be easily computed. Due to the
symmetry of the integrand with respect to all the variables zk , 1 k n + 1, we can make
the replacement

n+1
1 1
detn+1 −→ (n + 1)! .
sinh(λj − zk ) sinh(λa − za )
a=1
Then, inserting for each zk the factors sinh−1 (λk − zk ) and ϕ m−1 (zk ) into the kth column of
the remaining determinant, we can integrate separately each of these columns with respect
to zk , using the method described in Section 3:

1 1 i n+1
m−1 n
− detn+1 Uj k
σ1+ σm+1 = d n+1 λ ϕ −m+1 (λa ) , (5.4)
2i n! π
n=0
sinh(λn+1 + iπ 4 )
a=1
R
where
 #

 ϕ m−1 (zk ) dzk ϕ m−1 (λj )−ϕ m−1 (λk )
 2πi Cz sinh(zk −λj ) sinh(λk −zk ) = , j n, j = k,
1

 sinh(λj −λk )
# ϕ m−1 (zk ) dzk ϕ m−1 (λj )
Uj k = − 2πi1
Cz sinh2 (zk −λj ) = 2i(m − 1) cosh(2λj ) , j n, j = k, (5.5)

 #

 m−1 m−1
 2πi
1 ϕ (zk ) dzk
= ϕ (λkiπ) , j = n + 1.
Cz sinh(zk +
iπ
4 ) sinh(λk −zk ) sinh(λk + 4 )
The sum (5.4) is very similar to the expansion of a Fredholm determinant of a linear
integral operator. To make it more clear, let us introduce
m−1 m−1
(ϕ(λ)/ϕ(µ)) 2 − (ϕ(µ)/ϕ(λ)) 2
V (λ, µ) = i , (5.6)
π sinh(λ − µ)
where V (λ, λ) is defined by continuity from (5.6). Then Eq. (5.4) can be written as

m−1
− 1
σ1+ σm+1 = j k ,
d n+1 λ detn+1 U (5.7)
n!
n=0 R
with

 V (λj , λk ), j, k n,




m−1
 V (λj , λn+1 )

ϕ 2 (λn+1 )
, k = n + 1, j n,
 2i sinh(λn+1 + iπ
4 )
j k =
U m−1 (5.8)


iϕ 2 (λk )
, j = n + 1, k n,

 π sinh(λk + iπ
4 )



 ϕ m−1 (λn+1 )
, j, k = n + 1.
2π sinh2 (λn+1 + iπ
4 )
It is shown in Appendix A that (5.7) results into the derivative of a Fredholm determinant:

+ − ∂ α
σ1 σm+1 = det I + V (λ, µ) + R(λ, µ) , (5.9)
∂α 2π
where I denotes the identity operator and
m−1
(ϕ(λ)ϕ(µ)) 2
R(λ, µ) = . (5.10)
sinh(λ + 4 ) sinh(µ + 4 )
iπ iπ
The operator I + V + 2π α
R acts on the real axis, and α is an auxiliary parameter. Since
R(λ, µ) is a one-dimensional projector, the determinant in (5.9) is a linear function of α.
Hence, the derivative of the determinant does not depend on α. To compare (5.9) with the
result obtained in [7] one can make the standard change of variables cosh 2λ = (cos p)−1 ,
cosh 2µ = (cos q)−1 . Then, the kernel V and the projector R become
2 (p − q)
sin m−1 im
V (p, q) = − , R(p, q) = (−1)m e 2 (p+q) , (5.11)
2 (p − q)
1
π sin
where the integral operator acts on the interval [−π/2, π/2]. Finally, replacing p with −p
and q with −q, we arrive at

+ − ∂ 2 (p − q)
sin m−1 α − im (p+q)
σ1 σm+1 = (−1)m det I − + e 2 . (5.12)
∂α π sin 12 (p − q) 2π
This formula coincides with the result of [7] up to the factor (−1)m . The existence of this
factor is due to the fact that we use a different definition for the Hamiltonian, as it was
mentioned in the beginning of Section 2.
6. Long-distance asymptotics of σ1+ σm+1

−
The leading asymptotic behavior of the correlation function

σ1+ σm+1
−
was computed
in [3,4]. Later, in [7], a Fredholm determinant representation of the dynamic temperature
correlation function was obtained for an arbitrary value of the external magnetic field. The
determinant (5.12) appears to be a particular case of this result. To compute its asymptotic
behavior at large m, one can use the methods of the matrix Riemann–Hilbert problem
which were developed in [8] to study the dynamic temperature correlations. However,
these methods allow to find the asymptotics of the determinant only up to a multiplicative
constant, whereas the determinant (5.12) can be computed explicitly as a finite product of
Γ -functions. In this section, we present the corresponding derivation and reproduce the
results of the papers [3,4].
Observe first that the kernel V (p, q) (5.11) is degenerated:
1 i(p−q)(k− m )
m−1
2 (p − q)
sin m−1
V (p, q) = − =− e 2 . (6.1)
π sin 12 (p − q) π
k=1
Thus, the corresponding Fredholm determinant can be reduced to thedeterminant of a
matrix of finite size. Indeed, if a kernel K(p, q) has the form K(p, q) = m
k=1 fk (p)gk (q),
then

det I + K(p, q) = detm (δj k + Mj k ), (6.2)
with

Mj k = fj (p)gk (p) dp. (6.3)
C
Here C is the contour where the operator I + K acts. In our case, C = [−π/2, π/2], and
1 m
fk (p) = − eip(k− 2 ) , k = 1, . . . , m − 1,
π
α −ip m
fm (p) = e 2 ,
2π
gk (q) = e−iq(k− 2 ) , k = 1, . . . , m.
m
(6.4)
Thus, for j < m,
$
2 0, for j − k even,
Mj k = −δj k − (−1)(j −k−1)/2t j −k
1
, for j − k odd, (6.5)
π
whereas the elements of the last line of the matrix M are given by
$
α 0, for k even,
Mmk = (k−1)/2 1 , for k odd. (6.6)
π (−1) k
Note that the parameter α enters only the last line. Hence, taking the derivative of the
determinant with respect to α, we need to differentiate only the elements of this line, and
we obtain
+ − 2m−1
σ1 σm+1 = − m detm (aj k ), (6.7)
π
where
$
0, for j − k even,
aj k = (−1)(j −k−1)/2 1 , for j − k odd, (6.8)
j −k
for j < m, and

$
0, for k even,
amk = (6.9)
(−1)(k−1)/2 k1 , for k odd,
for j = m.
Our goal is now to compute detm (aj k ). To do this, let us first reorder the columns and
the lines of the matrix aj k such that it becomes a 2 × 2 block-matrix. One has to move:

1. The columns with the number 2k to the position k for k = 1, . . . , m2 ;

2. The lines with the number 2j − 1 to the position j for j = 1, . . . , m2 ;
m
3. The last line with the number m to the position 2 + 1.
m
Here, denotes the integer part of m. After these transformations, we arrive at
2

a2j −1,2k 0
detm (aj k ) = (−1) m−1
det , (6.10)
0 a2j −2,2k−1
with a0,2k−1 ≡ am,2k−1 .
m+1
Hereby the sizes of the blocks are m2 × m2 for a2j −1,2k and m+1 2 × 2 for
a2j −2,2k−1 . Observe now that
(−1)j −k−1
a2j −1,2k = a2j −2,2k−1 = . (6.11)
2j − 2k − 1
Hence, we obtain

1 1
detm (aj k ) = −det m2 det m+1 . (6.12)
2j − 2k − 1 2 2j − 2k − 1
It remains to use the analog of (3.2) for rational functions
n
1 j >k (xj − xk )(yk − yj )
detn = n , (6.13)
xj − yk j,k=1 (xj − yk )
which gives
m m m+1 m+1

2
2
j −k
2
2
j −k
detm (aj k ) = (−1)m−1 . (6.14)
k=1 j=1
j −k− 1
2 k=1 j=1
j −k− 1
2
j=k j=k
After the computation of the products with respect to j , we substitute the result into (6.7)
and eventually obtain
m m+1
(−1)m 2
Γ 2 (k) 2
Γ 2 (k)
σ1+ σm+1
−
= . (6.15)
2
k=1
Γ (k − 2 )Γ (k + 2 ) k=1 Γ (k − 12 )Γ (k + 12 )
1 1
Thus, we have computed the Fredholm determinant (5.12) as a finite product of Γ -

functions. This form enables one to evaluate the large m asymptotic behavior of the
correlation function
σ1+ σm+1
−
in a rather simple way (see Appendix B). The result reads
∞ % &
(−1)m 1 dt −4t 1 (−1)m −4
σ1+ σm+1
−
= √ exp e − 1− +O m .
2m 2 t cosh2 t 8m2
0 (6.16)
7. Conclusion
To conclude this article, we would like to discuss the significance of our results and
the perspectives they open concerning the computation of correlation functions in a more
general case. Actually, the purpose of the work presented here was double-folded.
Our first goal was to demonstrate that it was really possible in the free fermion limit to
compute the spin–spin correlation functions using their multiple integral representations.
Indeed, the ability of this method to provide effective results, even in the simplest case
∆ = 0, has for a long time been seriously under question (see however [19] and [20,21]
for the Ising case). Thanks to the new formulas obtained in [1], which correspond in fact
to certain re-summations of the elementary blocks [9–12], we were able here to solve this
problem.
The second goal of this paper concerns the possible application of these new integral
representations to the general XXZ model. We hope that some of the technical methods
presented here are not specific to the free fermion point, and that they can also be
successfully adapted to study a more general case.
In particular, the method we used here to compute the z-integrals might be quite efficient
for the evaluation of the long distance asymptotics of the spin–spin correlation functions.
Recall that, in this paper, we have deformed the original contour Cz into a horizontal strip
Γ of width iπ . At ∆ = 0, the contribution coming from Γ vanishes due to the periodicity
of the integrand. In the general case this property is no longer valid since the density
function ρ(λ, z) is no longer iπ -periodical (except at ζ = 2n π
, where n is a positive integer).
Nevertheless, it seems that one can control the order of the contribution coming from Γ
as m → ∞. Indeed, it is possible to choose the strip such that |ϕ(z)| < 1 (see (3.5)) on its
boundaries. Then the factor ϕ m (z) becomes exponentially small uniformly on an arbitrary
finite interval of the new integration contour for z. Preliminary estimates show that the
integrals over Γ decrease as some negative powers of m as m → ∞. Thus, it is very
possible that the leading long distance asymptotics of the spin–spin correlation functions
are given by the residues of the integrand within the strip Γ .
Finally, we would also like to draw the reader’s attention to the method used for the
evaluation of the long-distance asymptotics of the emptiness formation probability. In fact,
the saddle point approach can be applied directly to the representation (2.14) in the general
case as well. One can thus expect that the emptiness formation probability decreases as
exp{−c(∆, h)m2 } as m → ∞. The main obstacle to finding the coefficient c(∆, h) is
related to the asymptotic analysis of detm t (λj , ξk ) (recall that for ∆ = 0 this determinant
is explicitly computable). The problem we mention here coincides one to one with the
problem of the computation of the partition function of the six-vertex model with domain
wall boundary conditions [22,23], which was solved for the homogeneous case in [24,25].
For our purpose, it would be desirable to extend these results to the inhomogeneous case
as well, which is still an open problem. It seems nevertheless that the emptiness formation
probability admits a Gaussian behavior.
Acknowledgements
N.K. would like to thank the University of York, the SPhT in Saclay and JSPS for
financial support. N.S. is supported by the grants INTAS-99-1782, Leading Scientific
Schools 00-15-96046, the Program Nonlinear Dynamics and Solitons and by CNRS.
J.M.M. is supported by CNRS. V.T. is supported by the DOE grant DE-FG02-96ER40959
and by CNRS. N.K., N.S. and V.T. would like to thank the Theoretical Physics group of the
Laboratory of Physics at ENS Lyon for hospitality, which made this collaboration possible.
Appendix A. Fredholm determinant
Let us first recall the definition of a Fredholm determinant: if an operator I + K acts on

an interval C as

[(I + K)φ](λ) = φ(λ) + K(λ, µ)φ(µ) dµ, (A.1)
C
then its Fredholm determinant is
∞
1
det(I + K) = dλ1 · · · dλn detn K(λj , λk ). (A.2)
n!
n=0 C
The series (5.7) has almost the same form. To make it exactly the same, we observe that
j k m − 1, and hence detn+1 U
rank U j k = 0 as soon as n > m − 1. Therefore, the sum in
(5.7) can be extended to infinity. We can then use the identity

∂
detn+1 Wj k = Wn+1,n+1 − detn (Wj k + αWj,n+1 Wn+1,k ) (A.3)
∂α α=0
which holds for any arbitrary matrix W . The sum (5.7) can thus be written as
+ −
σ1 σm+1
∞

1 ϕ m−1 (λn+1 ) dλn+1 ∂
= −
n=0
n! 2π sinh2 (λn+1 + iπ4 )
∂α
R

× d n λ detn V (λj , λk )
R

m−1 m−1
α ϕ 2 (λn+1 )ϕ 2
(λk ) dλn+1
+ V (λj , λn+1 ) . (A.4)
2π sinh(λn+1 + iπ ) sinh(λk + iπ
)
4 4 α=0
R
The series (A.4) is now exactly of the form (A.2), which means that the correlation function
σ1+ σm+1
−
can be represented as the following Fredholm determinant:

σ1+ σm+1
−

ϕ m−1 (ν) dν ∂
= −
2π sinh (ν + 4 )
2 iπ ∂α
R

m−1 m−1
α ϕ 2 (ν)ϕ 2 (µ) dν
× det I + V (λ, µ) + V (λ, ν) . (A.5)
2π sinh(ν + 4 ) sinh(µ + 4 ) α=0
iπ iπ
R
To reduce this determinant to the form (5.9), we use the following lemma:
Lemma A.1. Let an integral operator

I + K(λ, µ) + α K(λ, ν)y(ν)x(µ) dν
C
acts on an interval C. Hereby the kernel K(λ, µ) and the functions x(λ), y(λ) are such
that the Fredholm determinant of this operator exists. Then

∂
x(ν)y(ν) dν − det I + K(λ, µ) + α K(λ, ν)y(ν)x(µ) dν
∂α
C C α=0
∂
= det I + K(λ, µ) + αy(λ)x(µ) . (A.6)
∂α
Proof. Assume first that det(I + K) = 0, i.e., that there exists the inverse operator
I − G = (I + K)−1 . Then one can extract the determinant of the operator I + K:

∂
∂α
C C α=0

∂
= det(I + K) x(ν)y(ν) dν − det I + α G(λ, ν)y(ν)x(µ) dν .
∂α
C C α=0
(A.7)
'
The operator C G(λ, ν)y(ν)x(µ) dν is a one-dimensional projector, hence,

det I + α G(λ, ν)y(ν)x(µ) dν = 1 + α G(µ, ν)y(ν)x(µ) dν dµ. (A.8)
C C
Substituting this into (A.7), we obtain


∂
∂α
C C α=0

= det(I + K) δ(ν − µ) − G(µ, ν) y(ν)x(µ) dν dµ. (A.9)
C
In its turn the last equality is equivalent to

∂
∂α
C C α=0

∂
= det(I + K) det I + α (I − G)(λ, ν)y(ν)x(µ) dν dµ . (A.10)
∂α
C
Finally, transforming the product of the two determinants into a single one, we arrive at
(A.6).
If det(I + K) = 0, we can consider the modified operator I + γ K, where γ is some
complex. The Fredholm determinant det(I + γ K) is an entire function of γ . Hence, it has
a finite number of isolated zeros in any closed domain of the complex plane. Thus, we can
choose γ such that det(I + γ K) = 0, repeat all the transformations described above and
continue the result to the point γ = 1. Thus, the lemma is proved. ✷
It remains to observe that the structure of the determinant (A.5) coincides with the one
of (A.6) if we set K(λ, µ) = V (λ, µ) and
m−1
ϕ 2 (λ)
x(λ) = y(λ) = √ . (A.11)
2π sinh(λ + iπ
4 )
Thus, we arrive at (5.9).
Appendix B. Asymptotic study of the product of Γ -functions
We need to compute the asymptotic behavior of the quantity

N
Γ 2 (k)
e−φN = , (B.1)
k=1
Γ (k − 12 )Γ (k + 12 )
where N → ∞. Expanding φN into a Taylor series, we obtain
∞
N 2n
1 1
φN = 2 ψ (2n−1) (k), (B.2)
(2n)! 2
k=1 n=1
where
d 2n
ψ (2n−1) (z) = log Γ (z). (B.3)
dz2n
The sum with respect to k in (B.2) can be computed using

N

ψ (s) (k) = Nψ (s) (N + 1) + s ψ (s−1) (N + 1) − ψ (s−1) (1) . (B.4)
k=1
Substituting this into (B.2), we obtain
∞ 2n
1 1
φN = 2 Nψ (2n−1) (N + 1) + (2n − 1)
(2n)! 2
n=1

× ψ (2n−2) (N + 1) − ψ (2n−2) (1) . (B.5)
This series is absolutely convergent, and we can make an asymptotic estimate of each term
separately. To do this, let us present φN in the form
1
φN = log N + S1 + S2 , (B.6)
4
where
∞
1
2n − 1 1 2n (2n−2)
S1 = Nψ (N + 1) + ψ(N + 1) + C − log N − 2 ψ (1).
4 (2n)! 2
n=2
(B.7)
Here C = −ψ(1) is Euler’s constant. The last term in (B.6) is
∞
2n
1 1
S2 = 2 Nψ (2n−1) (N + 1) + (2n − 1)ψ (2n−2) (N + 1) . (B.8)
(2n)! 2
n=2
Recall also the asymptotic expansion for the logarithm of Γ -function
1 1
n−1
B2k
log Γ (z) = z log z − z − log z + log(2π) +
2 2 2k(2k − 1)z2k−1
k=1

+ O z1−2n , |z| → ∞, (B.9)
where B2k are Bernoulli numbers. Using (B.9) one can easily see that at N → ∞ the term
S1 has a finite limit, while S2 vanishes.
To compute the limiting value of S1 , one can use, for example, the integral representa-
tion for the derivatives of ψ(z):
∞
e−t z t 2n−1
ψ 2n−1
(z) = dt, n 1, (z) > 0. (B.10)
1 − e−t
0
Then the series in (B.7) becomes the expansion of the exponent, and we obtain
∞ % &
1 dt −4t 1
S1 = − e − , N → ∞. (B.11)
4 t cosh2 t
0
This gives us the constant contribution to φN .
In order to find the corrections to this quantity, we need first to take into account the
higher order corrections to ψ (N + 1) and ψ(N + 1) in S1 and then to take several terms
from the series S2 . In particular, we find
∞ % &
1 1 dt −4t 1 1
φN = log N − e − 2
+ 2
+ O N −4 , N → ∞. (B.12)
4 4 t cosh t 64N
0
This formula can be directly applied for the computation of the asymptotic behavior of
(6.15). For another related expansion, see also [26].
References
[1] N. Kitanine, J.M. Maillet, N. Slavnov, V. Terras, Spin–spin correlation functions of the XXZ- 12 Heisenberg
chain in a magnetic field, hep-th/0201045, Nucl. Phys. B (2002), in press.
[2] E. Lieb, T. Shultz, D. Mattis, Ann. Phys. 16 (1961) 407.
[3] T.T. Wu, Phys. Rev. 149 (1966) 380.
[4] B.M. McCoy, Phys. Rev. 173 (1968) 531.
[5] H.G. Vaidya, C.A. Tracy, Phys. Lett. A 68 (1978) 378.
[6] B.M. McCoy, J.H.H. Perk, R.E. Schrock, Nucl. Phys. B [FS8] 220 (1983) 35.
[7] F. Colomo, A.G. Izergin, V.E. Korepin, V. Tognetti, Theor. Math. Phys. 94 (1993) 19.
[8] A.R. Its, A.G. Izergin, V.E. Korepin, N.A. Slavnov, Phys. Rev. Lett. 70 (1993) 1704.
[9] M. Jimbo, K. Miki, T. Miwa, A. Nakayashiki, Phys. Lett. A 168 (1992) 256.
[10] M. Jimbo, T. Miwa, J. Phys. A: Math. Gen. 29 (1996) 2923.
[11] M. Jimbo, T. Miwa, Algebraic Analysis of Solvable Lattice Models, American Mathematical Society, 1995.
[12] N. Kitanine, J.M. Maillet, V. Terras, Nucl. Phys. B 567 [FS] (2000) 554, math-ph/9907019.
[13] N. Kitanine, J.M. Maillet, V. Terras, Nucl. Phys. B 554 [FS] (1999) 647, math-ph/9807020.
[14] J.M. Maillet, V. Terras, Nucl. Phys. B 575 (2000) 627, hep-th/9911030.
[15] M. Shiroishi, M. Takahashi, Y. Nishiyama, Emptiness formation probability for the one-dimensional
isotropic XY model, cond-mat/0106062.
[16] B.M. McCoy, E. Barouch, Phys. Rev. A 3 (1971) 789.
[17] N. Kitanine, J.M. Maillet, N.A. Slavnov, V. Terras, hep-th/0201134, J. Phys. A: Math. Gen. (2002), in press.
[18] P.A. Deift, A.R. Its, X. Zhou, Ann. Math. 146 (1997) 149.
[19] H. Au-Yang, J.H.H. Perk, Physica A 144 (1987) 44.
[20] J.R. Reyes Martínez, Phys. Lett. A 227 (1997) 203.
[21] J.R. Reyes Martínez, Physica A 256 (1998) 463.
[22] A.G. Izergin, V.E. Korepin, Commun. Math. Phys. 99 (1985) 271.
[23] A.G. Izergin, Sov. Phys. Dokl. 32 (1987) 878.
[24] V.E. Korepin, P. Zinn-Justin, Thermodynamic limit of the six-vertex model with domain wall boundary
conditions, cond-mat/0004250.
[25] P. Zinn-Justin, Six-vertex model with domain wall boundary conditions and one-matrix model, math-
ph/0005008.
[26] H. Au-Yang, J.H.H. Perk, Phys. Lett. A 104 (1984) 131.
RSOS revisited
G. Takács a , G.M.T. Watts b
a Institute for Theoretical Physics, Eötvös University, Pázmány Péter sétány 1/A, Budapest H-1117, Hungary
b Department of Mathematics, King’s College London Strand, London WC2R 2LS, UK
Received 21 May 2002; received in revised form 30 July 2002; accepted 5 August 2002
Abstract
We investigate the issues of unitarity and reality of the spectrum for the imaginary coupled affine
(1) (2)
Toda field theories based on a1 and a2 and the perturbed minimal models that arise from their
(1)
various RSOS restrictions. We show that while all theories based on a1 have real spectra in finite
(2)
volume, the spectra of a2 models is in general complex, with some exceptions. We also correct the
S matrices conjectured earlier for the φ15 perturbations of minimal models and give evidence for
a conjecture that the RSOS spectra can be obtained as suitable projections of the folded ATFTs in
finite volume.
PACS: 11.55.Ds; 11.30.Na; 11.10.Kk
Keywords: Unitarity; Finite size effects; Minimal models; Integrable perturbations; Kinks
1. Introduction
Unitarity plays an important role in quantum field theory (QFT). In many application of
QFT, the framework of quantum theory requires a positive definite conserved probability,
which is guaranteed by the Hermiticity of the Hamiltonian and the unitarity of the S matrix.
Hermiticity also guarantees that the spectrum of the Hamiltonian is real, an important
condition if the Hamiltonian is to be interpreted as the energy of the physical system.
However, there are applications of the QFT formalism when this is not a necessary
physical requirement: non-unitary QFTs (even those with complex spectra) appear to play
an increasing role in the investigation of statistical mechanical systems (e.g., disordered
E-mail addresses: takacs@ludens.elte.hu (G. Takács), gmtw@mth.kcl.ac.uk (G.M.T. Watts).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 7 6 - 4
G. Takács, G.M.T. Watts / Nuclear Physics B 642 [FS] (2002) 456–482 457
systems [1]). There has been some interest in the thermodynamics of such theories
recently [2]. In the study of disordered systems, sigma models based on supergroups play a
prominent role in recent developments aimed at classifying the possible universality classes
[3] and their scattering theories have also been investigated in [4]. In the latter paper it
was shown that the scattering amplitudes of the OSP(1|2n) σ -model [4] are related to the
S-matrix of the imaginary coupled Toda field theory based on the twisted affine algebra
(2)
a2n , due to a relation between the algebras of the non-local conserved charges in the two
theories.
In this paper we intend to study the simplest cases: the finite-size spectra of perturbed
(1) (2)
minimal models related to imaginary coupled Toda theories a1 (sine-Gordon) and a2
(ZMS), with periodic boundary conditions. Our main aim is to establish conditions under
which these theories have real spectra, continuing our earlier work started in [5]. Reality
of the spectrum could be interesting for several reasons: (i) it affects the large distance
asymptotics of correlation functions (in case of a complex spectrum, these asymptotics can
show oscillating behaviour), (ii) it could allow a redefinition of the theory that renders it
unitary (although, as we discuss, physical requirements may prevent that, and it is not at
all clear whether this can be performed while maintaining the interpretation of the given
model as a local QFT). Finite size spectra are also interesting in their own right: periodic
boundary conditions can also be thought of as realization of finite temperature.
We start by recalling the general issues involved, and the relationships between the three
distinct concepts of probability conservation, unitarity and the reality of the spectrum.
For any imaginary coupled affine Toda field theory (ATFT), there are three different
classes of models which can be considered, namely, the original (unfolded, unrestricted),
the folded and the restricted (RSOS) models. We define these in Section 3 and discuss the
(1)
reality of the spectrum in these three classes for the theories related to a1 (sine-Gordon)
(2)
and a2 (ZMS) in turn.
In the case of a1(1) , the finite-size spectrum of the RSOS models turns out to be a subset
of the spectrum of a suitable folded model—i.e., the state space of the RSOS model can
be thought of as a projection of the space of the folded model. Using this relation, we
(1)
can conclude that the spectra of all a1 related models are real. In this paper we also give
evidence that a similar relationship exists between the RSOS and folded models based
(2)
on a2 . However, this is not sufficient to establish the reality of these models. We then turn
to transfer matrix arguments and numerical analysis to investigate the spectrum.
In all cases in which our studies find no violation of reality, there are several possibilities
remaining:
1. Apart from cases in which the S-matrix can be shown to be Hermitian analytic (e.g.,
perturbations of unitary minimal models), we only have numerical (and sometimes
matrix perturbation theory) results. Since these numerical tests require ‘scanning’ over
the full range of rapidity of each particle in a multi-particle state, there is always a
chance that such methods miss some region of rapidities in which reality is violated.
In addition, numerical diagonalisation always introduces some numerical errors, and
so some threshold condition must be defined to separate real cases from non-real cases.
It remains a question whether reality violations below the threshold are genuine or not.
458 G. Takács, G.M.T. Watts / Nuclear Physics B 642 [FS] (2002) 456–482
2. We could only examine transfer matrices containing up to five particles, as the

numerical calculations become progressively more and more difficult as we increase
the number of particles involved. Therefore, it is possible that higher particle transfer
matrices would introduce further constraints on the reality of the spectrum.
(2)
3. In the case of a2 , our methods only allow us to determine the large volume
asymptotics of the spectrum which have a power-like decay as a function of the
volume. It is possible that further contributions (decaying exponentially with the
volume, e.g., those related to vacuum polarisation) would spoil reality.
4. For a2(2) , the bootstrap is closed only in certain regimes of the parameter space and we
do not know the S-matrices of all possible particles in the spectrum. It is possible that
reality of the spectrum is only violated in sectors which contain such ‘extra’ particles.
As a result, we can only have definite results in the cases when the spectrum is complex:
if we find that some multi-particle transfer matrix has non-phase eigenvalues then we can
conclude that the Hamiltonian must have complex eigenvalues. Whenever our examination
finds that the spectrum is real we cannot the possibility exclude that further study would
find a complex spectrum.
However, the theories for which we cannot find any violation of reality show some very
striking patterns, and therefore we believe it is possible that several or possibly all theories
falling into these patterns have real spectra. Clearly, further understanding is necessary: the
most promising approach would be the development of some exact method to describe the
(2)
finite size spectra (e.g., an extension of the a2 NLIE [6] to describe excited states) and
a deeper understanding of the relation between the folded and the RSOS models in finite
volume.
As a side result, we also correct some minor mistakes in our previous paper [5] and
the φ15 S-matrices conjectured earlier by one of us in [7]. We summarise our results in
Section 5.
2. Unitarity and related concepts in QFT
In this section we discuss the relation between the concepts of probability conservation,
unitarity and reality of the spectrum in quantum field theory which are often confused.
We also recall the definitions of Hermitian analyticity and R-matrix unitarity and their
relationship with unitarity of the S-matrix.
2.1. Unitarity, reality and probabilistic interpretation
For all the theories we consider there exists a non-degenerate sesquilinear form (which
we call an inner product for short, even if it is not positive definite) on their space of states
which is conserved under the time evolution described by the Hamiltonian, i.e., the time
evolution operator and the S-matrix preserves this inner product (and the Hamiltonian is
Hermitian with respect to the conjugation defined by it). For perturbed conformal field
theories such an inner product is inherited from the standard inner product used in CFT,
even away from the critical point.
In the cases when this inner product is positive definite (and defines a Hilbert space
structure on the space of states) this implies the usual Hermiticity/unitarity, with the
consequence that the S-matrix has phase eigenvalues and the energy spectrum is real.
This inner product makes possible the usual probabilistic interpretation in quantum theory.
These theories are called unitary QFTs.
On the other hand, in many systems the inner product is indefinite. It is still possible for
the spectrum to be real and the S-matrix eigenvalues to be phases in which case we call
the theory a real QFTs. However, generally when the inner product is indefinite, then the
spectrum is complex and the eigenvalues of the S-matrix are not phases, in which case we
call the QFT non-real.1
One can see that the three notions of unitarity, reality and probability conservation are
in fact different. If we want the theory to have a positive definite conserved probability,
then unitarity is necessary, which is a stronger notion than reality. However, often this is
not required; moreover, there exist exotic probability theories which allow for negative
or complex probabilities (which in that case of course loses the interpretation in terms of
frequency of events).
An important point is whether the S-matrix has any relevance in the case when the
theory is non-unitary. It is easier to accept this when the theory is still real, as in the
case of the scaling Lee–Yang model (i.e., Virasoro minimal model (2, 5) perturbed by
its single non-trivial relevant operator φ13 )2 or, for indeed, of any φ13 perturbations of
Virasoro minimal models. However, there are examples of non-real theories (e.g., the
Virasoro minimal model (3, 14) perturbed by operator φ15 [9]) for which the finite size
spectrum extracted using thermodynamic Bethe Ansatz (TBA) for the vacuum energy
and the Bethe–Yang equations for multi-particle states matches perfectly the spectrum of
the Hamiltonian obtained numerically using truncated conformal space approach (TCSA),
even for the complex part of the spectrum.
To summarise, we believe that for a large class of theories (unitary, real and non-real),
and in particular for the RSOS models considered in this paper, the spectrum is determined
by the S-matrix even when the spectrum is complex.
2.2. Unitarity and Hermitian analyticity
In analytic S-matrix theory the property of unitarity is closely linked with that of
Hermitian analyticity [10,11]. Without entering into details, we intend to recall these
concepts and their relations here as they are going to play an important role later.
Unitarity is simply the statement that using a Hermitian conjugation † with respect to a
positive definite inner product, the S operator that maps out-states into the in-states has the
property
SS † = S † S = I. (2.1)
1 There does not seem to be any agreed convention for the naming of what we call real and non-real theories.
We simply use these names in this paper for convenience.
2 For an early discussion of this issue, cf. [8].
Given a theory that is non-unitary but real, since the S operator has phase eigenvalues
it is obviously possible to define a new positive definite inner product with respect to
which S is unitary. However, this inner product may be inconsistent with the rules of
analytic S matrix theory and/or may render the Hamiltonian non-Hermitian (while the
Hamiltonian was Hermitian with respect to the original, indefinite inner product). The
scaling Lee–Yang model is an example of this situation. Its spectrum contains a single
scalar particle which appears as a bound state in the two-particle scattering. S-matrix theory
then relates the norm of one-particle states to two-particle states through the residue of the
corresponding pole in the two-particle S-matrix. The natural inner product in the perturbed
CFT formalism (defined through the Hermiticity of Virasoro generators) is indefinite, with
n-particle states having the signature (−1)n (or −(−1)n ), depending on choice, in which
case the sign of the residue is consistent with S-matrix theory. However, if one attempts
to use a positive definite inner product, then the residue of the two-particle S-matrix at the
bound state pole has the “wrong sign”, which is how it is often stated in the literature. As
a result, the natural inner product of the scaling Lee–Yang model is indefinite, and it is
an example of a real but non-unitary theory. Therefore, the existence of a positive definite
redefinition of the inner product does not mean the theory can be made unitary because it
may conflict with some other physically motivated requirements. For a further discussion
of this issue see [8]. The property that the residues of the bound state poles have the “right
sign” is sometimes called “one-particle unitarity” [12].
In integrable theories the whole S operator is encoded in the set of two-particle
S-matrices SAB (θ ), where A and B denote the particles (or multiplets if there are internal
quantum numbers) and θ is their relative rapidity. It is a simple matter to prove that unitarity
of all possible two-particle S-matrices

kl ∗
ij SAB (θ )mn = δim δj n
SAB (θ )kl (2.2)
k,l
(where we explicitly wrote the multiplet indices) implies the unitarity of multi-particle
transfer matrices (defined in Appendix C), and also means that the spectrum is real. In
writing Eq. (2.2) we assumed that we had chosen an orthonormal basis in the internal
multiplet space.
If, in addition, the inner product on the space of states is positive definite then the theory
in question is a unitary QFT. We shall abbreviate the property (2.2) as TU (two-particle
unitarity).
Hermitian analyticity (HA) tells us something about the behaviour of the S-matrix
elements under complex conjugation. It states that
∗
= SBA (−θ ∗ )lk .
ji
SAB (θ )kl
ij (2.3)
On the other hand, the S-matrices we investigate here are derived from quantum group
R-matrices. R-matrices also satisfy a relation known as “unitarity” in quantum group
theory which we call here R-matrix unitarity (RU) and takes the form

ij SBA (−θ )lk = δim δj n .
SAB (θ )kl nm
(2.4)
k,l
We see that RU and HA together imply TU and thus reality of the spectrum (note that
Eq. (2.2) is meant only for physical, i.e., real values of θ ). Therefore for the models we
consider Hermitian analyticity implies a real spectrum since RU is automatically satisfied
due to the quantum group symmetry.
These notions can be appropriately generalised to RSOS theories, where the multi-
particle polarisation spaces are not simply tensor products of one-particle ones. Rather,
the multiplet structure is labelled by so-called RSOS sequences, which denote ‘the vacua’
between which the particles mediate and are constrained according to so-called ‘adjacency
rules’. The two-particle S-matrix, therefore, carries four vacuum indices a, b, c, d and it is
specified in the following way (we omit particle species labels for simplicity):
d
a c
←→ Sab
cd
(θ − φ). (2.5)
b
θ φ
For RSOS theories TU, HA and RU take the form:

TU : cd
Sab cd
(θ )Sae (θ )∗ = δbe , (2.6)
d
∗
HA : cd
Sae (θ )∗ = Sad
ce
−θ , (2.7)

RU : cd
Sab ce
(θ )Sad (−θ ) = δbe . (2.8)
d
3. The models
The models we study can be classified in the following way:
1. The original (unfolded, unrestricted) models. These have an infinite set of degenerate
(1)
‘vacua’ and their spectra are built from a fundamental soliton doublet (a1 ) or triplet
(2)
(a2 ) by closing the S-matrix bootstrap.
2. Folded models. Using the periodicity of the field theoretic potential, one can choose
to identify the ground states after k periods (see [13] for the sine-Gordon case), i.e.,
in a k-folded model one has k ground states, between which the solitons mediate. The
spectrum and the scattering theory for this case can be straightforwardly written down
using the well-known S-matrices of the original model.
3. Restricted (RSOS) models. At certain ‘rational’ values of the coupling it is possible
to define a space of ‘RSOS states’ as a quotient of the full space of states on which
the action of the S operator is well defined. These states can be labelled by sequences
of ‘vacua’ and the S-matrix factories into 2-particle ‘RSOS type’ S-matrices. These
RSOS S-matrices describe the scattering theory of the φ13 or φ12 /φ21 /φ15 perturbations
(1) (2)
of Virasoro minimal models, respectively for a1 and a2 . These restrictions were
discussed in the following papers: the φ13 case in [14], φ12 /φ21 in [15] and φ15 in [7].
The above construction of the RSOS states relies on the action of the quantum group
and has only been defined for the theory on the full line, i.e., in infinite spatial volume.
However, given the two-particle S-matrix derived in this way one can easily define the
theory in finite volume.
All of the models enumerated above are integrable and therefore there are numerous
results for their finite volume spectra. We shall mainly use results obtained using two
methods.
Firstly, the so-called nonlinear integral equation (NLIE) approach, which gives exact
results for the spectrum and was pioneered by Klümper et al. [16] on the one hand and
Destri and de Vega [17] on the other. The original developments concerned mainly the
spectra of the a1(1) case (for the excited states see also [18–20]).
Secondly, we use the related method of the Bethe–Yang equations (cf. [21] and
references therein), which describe the large volume asymptotics of the finite size
spectrum. If the NLIE for a given theory is known, then the asymptotics can be derived
independently and for a1(1) it was checked that they agree with the results from the Bethe–
Yang equations [19,20]. In general when the NLIE is not known this is the only known
analytic way to obtain information about the finite size spectrum (in some cases, TBA
equations are known for the excited states [22], but these cases are covered by the NLIE as
well).
In finite volume, since the field theory potential is periodic, it is possible to introduce
twisted sectors of the original model [23]. These are labelled by a twist parameter α,
defined mod 2π . The twisted sectors have different finite size spectra which are given by
a modification of the original NLIE in which α appears explicitly [23].
The finite size spectrum of the k-folded model is the union of the spectra of twisted
sectors where the twist runs through the values [13]
2πm
α= , m = 0, . . . , k − 1. (3.1)
k
When we turn to specific models we shall discuss a relation between the finite size
spectra of folded and RSOS models, namely that the finite-size spectrum of an RSOS
model is exactly a subset of the spectrum of an appropriate folded model. Such a relation
(1)
first emerged for the case of a1 [24], and we shall give evidence that a similar relationship
holds for a2(2) .
(1)
To be precise, for a1 it is known that both the exact spectra (described by the NLIE)
and, as a consequence, their asymptotics (described by the Bethe–Yang equation) of folded
and RSOS models are related. For a2(2) it is known that the exact vacuum energies in finite
volume are related (this was shown using the NLIE framework in [6]), but at present the
NLIE for excited states is not known. We shall present evidence that the asymptotics of
the excited state energies given by the Bethe–Yang eqns are similarly related. We believe
similar relations will hold for the exact spectra of every ATFT and the corresponding RSOS
models.
There exists some evidence in support of this claim. Al.B. Zamolodchikov has shown
that within the perturbed conformal field theory framework, the perturbative expansion for
the vacuum energy of the sine-Gordon model with a suitable value of the twist parameter
could be reinterpreted as a perturbative expansion in the RSOS model [23], to all orders
in perturbation theory. There is also a quantum group argument for the agreement of the
vacuum energies [25], which relies on a formulation of the partition function in terms of
the system in infinite volume. In addition, using the a1(1) NLIE it is possible to calculate
(a) the exact ultraviolet conformal weights and (b) the energies of the ground state and
excited states in the twisted sectors of sine-Gordon model to very high numerical accuracy.
Comparison of these data to (a) the known CFT data and (b) numerical finite volume
spectra extracted using TCSA, respectively, shows an excellent agreement [24].
It should be possible to find a proof using quantum group arguments directly for the
theory in finite volume, by finding an appropriate projection from the folded model to the
RSOS space which commutes with the transfer matrices, but we are unaware of such an
argument. The knowledge of this projection would also be useful because it could be used
to select systematically the RSOS states in the NLIE approach.
(1) (2)
We now proceed to give our conventions for the models based on a1 and a2 and give
the RSOS—folded relations that we shall check in Section 4.
3.1. Conventions for a1(1)
(1)
The a1 theory is simply the sine-Gordon model, for which we take the action to be

1 m20
AsG = d x ∂ν Φ∂ Φ + 2 cos βΦ .
2 ν
(3.2)
2 β
As is well known, introducing the parameter q = exp(8πi/β 2), one can show that the
(1)
model is invariant with respect to Uq (a1 ). The spectrum consists of a doublet of solitons
and a collection of scalar particles (breathers). As
described by [14], the RSOS theory
Mp,p + φ13 is obtained as a restriction at β = 8πp/p , that is at q = exp(πip /p).
Putting together the results of [20] and [13], it is straightforward to see that the finite
volume spectrum of this RSOS model is a subset of that of the 2p-folded model.
(2)
3.2. Conventions for a2
(2)
We take the action of the a2 model to be

1 m2 √ √
AZMS = d2 x ∂ν Φ∂ ν Φ + 0 exp(2i γ Φ) + 2 exp(−i γ Φ) . (3.3)
2 γ
Introducing the parameter
2
iπ
q = exp (3.4)
γ
one can show that the model has a symmetry under the affine quantum group Uq (a2(2)) and
as a result its S-matrix can be explicitly constructed [15].
The spectrum of the original model consists of a fundamental soliton transforming as a

triplet under Uq (a1(1)), having an S-matrix of the form
2π γ
0
S00 (x, γ )R(x, q), where x = exp(2πθ/ξ ) and ξ = , (3.5)
3 (2π − γ )
where S00 0 is a scalar function which is always a pure phase for real x and γ , and R(x, q)
(2)
is the R-matrix of Uq (a2 ) in the triplet representation.
Depending on γ , there may be other particles in the spectrum. For ξ < π there are
breathers B0n , n = 1, . . . , [π/ξ ] formed as bound states of K0 transforming in the singlet
representation, and for ξ < 2π/3 there are higher kinks Ki , n = 1, . . . , [2π/(3ξ )], also
formed as bound states of K0 and transforming in the triplet representation, and their
associated breathers Bin .
In addition, it is known that there are values of γ for which the bootstrap does not
close on this particle content, and for which there are more particles in the spectrum.
Since we are only able to investigate the transfer matrices for the kinks Ki and breathers
Bin , it is only possible to prove that any particular model has a complex spectrum. If we
find that the spectrum for these particles is real, that does not imply the reality of the full
spectrum simply because the S-matrix bootstrap could still generate S-matrices with non-
phase eigenvalues.
The S-matrices of the higher kinks have the form

SKm ,Kn (θ ) = Sm,n
0
(x, γ )R (−1)m+n x, q (3.6)
where Smn0 is a scalar factor.
0 are phases and there are only two different matrix structures
Since the scalar factors Sm,n
for the two-particle S-matrices SKm ,Kn , i.e., R(x , q) if m + n is even and R(−x, q) if m + n
is odd, for the purposes of finding out if the eigenvalues of the transfer matrices (defined
in Appendix C) it is hence to sufficient to consider
• For π/γ < 1: transfer matrices containing only K0 .

• For π/γ > 1: transfer matrices containing only K0 and/or K1 .
3.2.1. RSOS models related to a2(2)

The a2(2) model has two inequivalent RSOS restrictions [7,15]. From the conjectured
(2)
ground state NLIE of the a2 -related models [6] one can determine the minimal value
of the folding number for which the RSOS ground state is in the spectrum of the folded
model. Together with the results in [7,15], it leads to the following conjectures
1. Mp,p + φ12 is a projection of the p-folded ZMS model at γ = πp/p .

The RSOS S-matrices are given in Appendix A where we have specified all the
formulae including the 6j symbols since the ones we found in the literature all had
typos.3
3 We are grateful to Giuseppe Mussardo (SISSA) for providing us with a correct set of formulae.
As for the unrestricted case, the matrix part of the S-matrices depend only on q so they
depend on p only modulo 2p. In fact, we only need study the values 1 p < p since
the substitution p → −p changes the S-matrix elements to their complex conjugate
(if we choose the square root branches in the 6j symbols appropriately) and thus
simply complex conjugates the transfer matrices and their eigenvalues too as well.
(n.b. the models Mp,p + φ21 are contained in this class through the identification
Mp,p + φ21 ≡ Mp ,p + φ12 .)
2. Mp,p + φ15 is a projection of the 2p-folded ZMS model at γ = 4πp/p .
The RSOS restriction leading to the S-matrices was performed in [7], however, certain
amplitudes had wrong prefactors. The reason was that in the scattering amplitudes of
two (oppositely) charged solitons into two neutral ones there is a Clebsh–Gordan factor
which was not taken properly into account and therefore the amplitudes of [7] do not
satisfy the Yang–Baxter equation and R-matrix unitarity. The corrected amplitudes are
listed in Appendix B.
These perturbations lead to renormalisable field theories if the conformal weight of the
perturbing field is less than or equal to two, which is equivalent to the condition π/γ 1/2
for both cases, which is also the condition for the unrestricted model to be well-defined.
4. Results
The leading approximation (e.g., in the NLIE formalism) to finite size effects in large
volume is given by the Bethe–Yang equations which are summarised in Appendix C. In
particular it can be seen that the spectra of the models can only be real if the transfer
matrix eigenvalues (denoted in Appendix A by λ(s) (ϑ|ϑ1 , . . . , ϑN )) are all phases for real
values of the particle rapidities ϑi .
We check for which values of the coupling the various transfer matrices have phase
eigenvalues. There are some simplifications we can make.
1. In all our calculation we omit the scalar prefactors of the S-matrices since these are
irrelevant for determining whether the eigenvalues are pure phases or not. We call
the transfer matrices obtained from the S-matrices without the scalar factor ‘reduced’
transfer matrices.
2. The R-matrices of all models we consider have some discrete symmetries (which we
describe here for real values of the rapidity).
(a) R(x, q ∗ ) = R(x, q)∗ , which means that we only need to consider 0 arg(q) π .
(b) R(x, −1/q) = U R(x, q)U −1 where U is a diagonal matrix whose non-zero
entries are ±1.
The reduced transfer matrices are still in general very complicated, even for the original
models, and, therefore, in our study we diagonalised them numerically for a large set of
values for the rapidities and other parameters. In all our plots we show only the eigenvalues
of the reduced transfer matrices.
(1)
4.1. a1 related models
(1)
In the case of a1 related models, since the folded and original models are all unitary,
their finite size spectra are real and, therefore, all φ13 perturbations have real spectra
as well, regardless of whether they are perturbations of unitary or non-unitary minimal
models. In addition, the NLIE formalism [24] yields manifestly real spectra as well.
In other words:
⇒ RSOS real,
original unitary ⇒ folded unitary
RSOS unitary.
One might think that, since the RSOS spectrum is simply a subset of the spectrum of
the unitary unrestricted model, the RSOS model would simply inherit a positive definite
inner product and also be unitary. However, the fact that the spectrum of the RSOS and
unrestricted models are different, means that the constraints of analytic S-matrix theory
may enforce different inner products. As an example, consider the value 8π/β 2 = 2/5. The
particle content of the unrestricted model is a soliton doublet s, s̄ and a single breather B.
The RSOS model, on the contrary, is the Lee–Yang model with a single particle B (the
solitons are removed from the spectrum by the RSOS restriction). There is a first order pole
in SBB (θ ) at θ = 2πi/3 which must have an explanation in terms of on-shell diagrams.
In the unrestricted sine-Gordon model, it is explained by the famous Coleman–Thun
mechanism [26] as the sum of singular contributions from diagrams with internal soliton
lines. In the RSOS model there are no solitons so this pole must be explained by a single
diagram in which B occurs as a bound state of two B particles. As we discussed before,
the sign of the residue of this pole forces the inner product to be indefinite.
By numerical diagonalisation of the reduced transfer matrices we found that the transfer
matrix eigenvalues for the RSOS models Mp,p + φ13 are a subset of those for the

2p-folded sine-Gordon model with β = 8πp /p. We also observed that the numerically
computed transfer matrix eigenvalues are all phases, as expected from the general argument
above.
In Fig. 1, we plot the arguments of the eigenvalues of the reduced two-particle transfer
matrices of the RSOS models (M7,m + φ13 ) and the 14-folded sine-Gordon model. The
RSOS models are only defined for m integer, i.e., for arg(q)/π taking values m/7. For
these values the eigenvalues of the transfer matrices (shown as blobs) are a subset of the
eigenvalues of the matrices of the folded model shown as lines.
The two-particle space of the 14-folded model has dimension 28, and the spectrum
consists of 12 eigenvalues each of multiplicity 2 and 4 of multiplicity 1. The RSOS model
excludes the eigenvalues of multiplicity 1, 2 of the other eigenvalues, and each of the
remaining 10 eigenvalues appears with multiplicity 1 only.
There are RSOS models Mp,p + φ13 for which the S-matrix is Hermitian analytic
(e.g., the unitary case p = p + 1 or the series p = 2, see [14]) which then implies that the
spectrum is real. Our arguments above, however, extend to all φ13 perturbations.
Fig. 1. Comparison of the eigenvalues of the matrix part of the two-particle transfer matrices for the RSOS model
(1)
(M7,m + φ13 ) and the 14-folded a1 ATFT. The horizontal axis is arg(q)/π restricted to the fundamental
domain between 0 and 1, and the vertical axis is the argument of the eigenvalues. The folded models are shown
in green/solid lines, and the RSOS models (which are only defined for discrete values of q) are shown as blobs.
The relative rapidity of the particles was chosen θ = 1/9.
(2)
4.2. a2 related models
(2)
In this section, we investigate models related to a2 . We start with the 1-folded model
where we correct a mistake in [5]. We then consider multi-particle transfer matrices and
show the emergence of a pattern which seems to carry over to the folded case and it plays
an important role in the RSOS case as well.
For all three classes of model, we must consider separately the case π/γ > 1 for which
there are higher kinks, and π/γ < 1 for which there are no higher kinks and consequently
fewer constraints on the values of γ allowed by the reality of the spectrum.
We recall that due to the symmetries of the R-matrix, the spectrum of models with
−π < arg(q) < 0 is obtained by conjugating the spectrum of the model with 1/q, and so
we can restrict our attention to models with 0 arg(q) π .
(2)
4.2.1. The 1-folded models based on a2
We shall consider first the two-particle transfer matrix, and then the higher particle
number transfer matrices in turn. As we have mentioned, for our investigation of the
reality of the spectrum, we can ignore the scalar prefactors in the S-matrices and need only
consider the ‘reduced’ transfer matrices constructed out of the appropriate R-matrices. The
two-particle ‘reduced transfer matrices’ are simply the R-matrices themselves: if the kinks
being scattered are Km and Kn , then the appropriate R-matrix is R((−1)m+n x, q).
As calculated in [5], the eigenvalues of R(x, q) are three pairs of doubly degenerate
eigenvalues,
√ √
1 − q2 x 1 + q2 x
1, √ , √ (4.1)
q2 − x q2 + x
and three eigenvalues λ satisfying

x − q 4 x + q 6 λ3 + q 6 2 + q 2 λ2 + x 2 λ + x q 2 − 1 1 − 3q 4 + q 8 λ2 + λ

− q 2 1 + 2q 2 x 2 λ2 + λ + 1 − q 4 x 1 + q 6 x = 0. (4.2)
As stated in [5], for x negative, there are obviously eigenvalues which are not phases,
except when q 4 = 1. However, in [5] we stated incorrectly that all the eigenvalues were
phases for x non-negative and q a phase, and instead the correct result is that there are
non-phase eigenvalues for 0 < | arg(q)/π| < 1/4. This is due to branch cuts that appear in
the solutions of the cubic Eq. (4.2) which we overlooked before.
We were not able to diagonalise any higher particle number transfer matrices exactly.
Therefore, we resorted to numerical methods for the cases of three, four, five and six
particles, combined with matrix perturbation theory in the case of three and four particles.
The 1-folded models with π/γ > 1

If π/γ > 1 then the model contains both K0 and K1 . Since the matrix part of SK0 ,K1
is R(−x, q), the eigenvalues are just given by (4.1) and (4.2) with x replaced by −x,
and from looking at (4.1) we see that the two-particle transfer matrix has non-phase
eigenvalues unless q 4 = 1. This means that, unless q 4 = 1, the theory is non-real. We
have not investigated the case q 4 = 1 any further.
The 1-folded models with π/γ < 1

If π/γ < 1 then the only particles in the spectrum are the fundamental kinks (and the
first breather for π/γ > 2/3). In particular, the model contains no higher kinks, and the
only reduced transfer matrices to be considered are those constructed out of R(x, q) with
x positive.
The eigenvalues of the two-particle case have already been presented, and show that the
theory is non-real for 0 < | arg(q)/π| < 1/4 and for 3/4 < | arg(q)/π| < 1.
Numerical investigation of the three-particle transfer matrices show that the theory is
also non-real if 1/4 < | arg(q)/π| < 1/3 or 2/3 < | arg(q)/π| < 3/4. It turns out that
the strongest violation of reality happens for small x and y, and one can expand the
eigenvalues in a perturbation series around x = y = 0. Some care must be taken as
the x = y = 0 transfer matrices have a non-trivial Jordan form, so the Lidskii–Vishik–
Lyusternik generalised perturbation theory [27] must be used for the expansion. Without
entering into technicalities we only wish to mention that one obtains exactly the same
pattern as described above.
Consideration of the four-particle transfer matrix adds further regions of non-reality,
1/3 < | arg(q)/π| < 3/8 and 5/8 < | arg(q)/π| < 2/3, and of the five-particle transfer
matrix additional regions of 3/8 < | arg(q)/π| < 2/5 and 3/5 < | arg(q)/π| < 5/8.
Assuming this pattern to continue for higher particle number transfer matrices, we are
led to the conjecture that the eigenvalues of the n particle transfer matrices are always
phases for

1 1 arg(q) 1
− + 1 (4.3)
2 2n π 2 2n
and at the isolated points

arg(q) 1 1

π = 2 ± 2m , 1mn (4.4)
and for every other value of q there are non-phase eigenvalues for some value of the
rapidities.
This would lead to the result that the spectrum of the 1-folded model is always complex
except for the isolated points

arg(q) 1 1
=
π 2 1± n . (4.5)
(2)
4.2.2. Higher folded models based on a2
We investigated the folded transfer matrices using numerical diagonalisation and found
that the regions of real and complex spectrum are identical in every case to those of the
1-folded models, independent of the folding number.
It is obvious that the regions of complex spectrum contain those of the 1-folded model,
since the spectrum of the folded model contains that of the 1-folded model, but it appears
that there are no further constraints arising from the twisted sectors.
4.2.3. RSOS models based on a2(2)

The first observation to make is that since the spectrum of the RSOS models is a subset
of the spectrum of the appropriate folded model, if the folded model has a real spectrum
then so does the RSOS model.
Next, since the RSOS spectrum is a subset of the folded spectrum, it is possible that the
RSOS spectrum is real while that of the folded is complex. This is well known to be the
case for the perturbations of the unitary minimal models Mp,p±1 + φ12 (note that φ15 is
never a relevant perturbation of a unitary minimal model), and we believe that this property
is shared by many other models.
As evidence for our assertion that the spectra of the RSOS models are subsets of those
of the folded models, in Figs. 2 and 3 we plot the eigenvalues of the reduced two-particle
transfer matrices for K0 –K0 scattering of the 10-folded model and of the associated RSOS
models (We performed similar checks for the K0 –K1 and K0 –K0 –K0 transfer matrices
with equally convincing results).
These figures also show that for this particular choice of rapidity difference, there are
regions of arg(q)/π for which the folded transfer matrix has non-phase eigenvalues which
are included in the regions for which we believe that the folded transfer matrices have
non-phase eigenvalues for some values of rapidity. We also see that some of the RSOS
restrictions do manage to omit these non-phase eigenvalues, while others do not. We report
more fully on our findings in the later sections.
4.2.4. The φ12 perturbations with π/γ > 1

These theories have at least one higher kink K1 and we find that the most stringent
restrictions already arise from the two-particle transfer matrix involving K0 and K1 .
The theories with p = ±1 mod p are of the ‘unitary type’: their S-matrices (and
transfer matrices) are proportional to those of the unitary models Mp,p+1 + φ12 and
Mp,p+1 + φ21 , and since the scattering itself is manifestly unitary the eigenvalues are
all phases. Indeed, these S-matrices do satisfy Hermitian analyticity which together with
R-matrix unitarity implies the S-matrix unitarity equation for the two-particle S-matrices.
(2)
(M10,m + φ12 ) and the 10-folded a2 ATFT. The horizontal axis is arg(q)/π restricted to the fundamental
domain between 0 and 1, and the vertical axis is the argument of the eigenvalues. The phase eigenvalues of the
folded models are shown in green/thin solid lines, the non-phase eigenvalues in pink/thick solid lines, and the
RSOS models (which are only defined for discrete values of q and happen in this case all to be pure phases) are
shown as blobs. The relative rapidity of the particles was chosen θ = 5.
(2)
(M5,m + φ15 ) and the 10-folded a2 ATFT. The graph is labelled as for Fig. 2, except that there are now also
non-phase eigenvalues of the RSOS model which are shown as black blobs. The relative rapidity of the particles
was chosen θ = 5.
Our numerical calculations show that the transfer matrices of the theories Mp,p + φ12
where p = (p ± 1)/2 mod p also have phase eigenvalues only, although we do not know
any explanation for this fact yet.
Every other theory has non-phase eigenvalues in the K0 –K1 transfer matrix; conversely,
the models described above still have phase eigenvalues when we consider the three-
particle transfer matrices for all combinations of K0 and K1 .
We have not tested these models beyond the three-particle transfer matrix. In addition,
it is possible that particles arising through the S-matrix bootstrap will also introduce non-
Fig. 4. Numerical results for the eigenvalues of the transfer matrices of Mrs + φ12 . The heavy (black) points
are models for which the multi-particle spectrum appears to be real up to 5 particles, and the light (red) points
those which are non-real, with [π/γ ] = [s/r] plotted vertically against r = (folding number). See text for further
details.
phase eigenvalues in the transfer matrices; since the bootstrap has not been completed for
the majority of these models, we cannot say anything more about them.
4.2.5. The φ12 perturbations with 1/2 < π/γ < 1

These theories are more usually known as the ‘φ21 ’ perturbations. They have no higher
kinks in their spectra and at most one breather, and the S-matrix bootstrap closes on these
particles. The only possible violation of reality of the spectrum could be introduced by the
fundamental kink S-matrix, since the kink-breather and breather-breather S-matrices are
pure phases.
We show a selection of our results (up to folding number 50 and particle number 5) in
Fig. 4.
The small (red) points are values which we find definitely to be ‘non-real’. The
remaining large (black) points are values which appear to have real spectra up to and
including five-particle states. The solid (red, blue and purple) lines are the first three pairs
of real series discussed in point 3 below and the dashed (green) lines are models for which
a TBA equation for the ground state is known (see below).
We find a complicated pattern of results, with increasingly more models being ruled out
by higher and higher particle number transfer matrices. Since we have only investigated
transfer matrices involving at most 5 particles, our results are somewhat sketchy but they
show the following patterns:
1. All theories for which the folded models have real spectra are obviously real after
RSOS restriction, that is those with | arg(q)/π| = (1 ± 1/n)/2. For the range of γ
allowed, this gives only the models M4m,2m+1 + φ12 ≡ M2m+1,4m + φ21 and the
models M2m+1,m+1 + φ12 ≡ Mm+1,2m+1 + φ21 .
2. The perturbations of the unitary minimal models have manifestly unitary S-matrices
and real spectra, corresponding to the models Mp+1,p + φ12 ≡ Mp,p+1 + φ21 .
3. It appears that several other infinite series of models may also have real spectra, at
least this appears to be the case from our numerical tests of the transfer matrices up to
5 particles. These models form sequences which tend to the ‘real’ theories described
in 1 above. In fact the first example of such a sequence are the unitary models which
tend towards the value π/γ = 1, which is the first ‘real’ model with n = 1.
The next examples are the series M4k±1,3k±1 + φ12 tending towards arg(q)/π = 3/4,
M3k±1,2k±1 + φ12 tending towards arg(q)/π = 2/3, M8k±3,5k±2 + φ12 tending
towards arg(q)/π = 5/8, and so on.
These three pairs of series and the unitary models are shown on Fig. 4 as solid lines.
Some of these models have been considered before in various contexts. The first of
these theories, M3,5 + φ21 , was considered by G. Mussardo in [28], where he noted that
its spectrum was real despite the fact that the S-matrix was ‘non-unitary’.
Since then, (at least) three series of models have been conjectured to have a ground state
described by real TBA equations,4 these being Mm+1,2m+1 + φ21 [29,30], M2m+1,4m +
φ21 [31], and M2m+1,4m−2 + φ21 [32]. These are massive counterparts of the |I |=1, 2
and 4 massless flows considered in [32] with π/γ = 1/2 + |I |/(2r) and shown in Fig. 4
as dashed (green) lines. However, as is well known, simply knowing the TBA equations
for the ground state does not determine the excited state spectrum uniquely (a particular
example of this effect being the ‘type II’ ambiguity noted in [9]) and the scaling function
can describe the ground state of a genuine unitary model and also a ‘non-real’ non-unitary
model. So, while TBA equations are a useful alternative way to study the spectrum of a
theory, it is not easy to deduce reality properties from the TBA equation for a ground state.
4.2.6. The φ15 perturbations with π/γ > 1

These theories again have at least one higher kink K1 and again we find that the
most stringent restrictions already arise from the two-particle transfer matrix involving K0
and K1 .
Our numerical analysis shows that only two series of models have the possibility to be
real. These are the theories with p = 2p ± 1 mod 4p for which the S-matrix is Hermitian
analytic which guarantees the reality of the spectrum at the level of the Bethe–Yang
equations, and the theories with p = ±1 mod 4p for which the reality of the spectrum
is mysterious.
4.2.7. The φ15 perturbations with π/γ < 1

Again, these theories have no higher kinks in their spectra, and at most one breather, and
the S-matrix bootstrap closes on these particles. The only possible violation of reality of the
spectrum could be introduced by the fundamental kink S-matrix, since the kink-breather
and breather-breather S-matrices are pure phases.
4 We thank R. Tateo for pointing this out to us.

We find an even more complicated pattern of results than for the φ12 perturbations, once
again with increasingly more models being ruled out by higher and higher particle number
transfer matrices. In this case the size of the transfer matrices is much larger, and we have
only investigated transfer matrices involving at most 4 particles. Our results are even more
sketchy but they show the following patterns:
1. All theories for which the folded models have real spectra are still obviously real after
RSOS restriction, that is those with | arg(q)/π| = (1 ± 1/n)/2. For the range of γ
allowed, this gives only the models Mp,2p+1 + φ15 and M2p+1,4p+4 + φ15 .
2. Again, it appears that several other infinite series of models may also have real spectra,
at least this appears to be the case from our numerical tests of the transfer matrices up
to 4 particles. These models form sequences which tend to the ‘real’ theories described
in 1 above.
The first example are the theories, Mk,4k−1 + φ15 tending towards arg(q)/π = 1 and
the series Mk,3k±1 + φ15 tending towards arg(q)/π = 3/4.
At the next level, we find that up to four particles, there are four infinite series tending
to arg(q)/π = 2/3, namely, M3k±1,8k±3 + φ15 and M6k±1,16k±2 + φ15 . Since the
existence of this second series is in some sense a new phenomenon, we checked that
it survives our numerical tests at the five particle transfer matrix level, but we have no
opinion whether or not it will survive at all higher particle numbers.
Again, three infinite series of these models have been considered before in various
contexts [32]. These series are exactly those which are related to the infinite series of φ21
perturbations by interchanging the two terms in the ZMS potential as described in, e.g., [9]:
M2m+1,m+1 + φ12 ↔ M2m+1,4m+4 + φ15 ,

M4m,2m+1 + φ12 ↔ Mm,2m+1 + φ15 ,
M4m−2,2m+1 + φ12 ↔ M2m−1,4m+2 + φ15 . (4.6)
These series are shown as dashed (green) lines in Fig. 5. The final pair of theories is
particularly interesting as being an example of a ‘type II’ pair in that these are different
theories but share exactly the same ground state scaling function and have a common sub-
sector of multi-particle states with identical finite size energy corrections.
5. Conclusions
(1) (2)
We have investigated the finite volume spectra of theories based on a1 and a2
imaginary coupled affine Toda field theories. Our main results can be summarised as
follows:
(1)
1. All theories (original, folded, RSOS) based on a1 have real spectra. To show this,
we used the fact that the finite volume spectrum of the RSOS theories can be obtained
as suitable projections of folded theories, which are in turn manifestly unitary QFTs
Fig. 5. Numerical results for the eigenvalues of the transfer matrices of Mrs + φ15 . The heavy (black) points
are models for which the multi-particle spectrum appears to be real up to 4 particles, and the light (red) points
those which are non-real, with π/γ = s/4r plotted vertically against r = (folding number)/2. See text for further
details.
related to sine-Gordon theory. We also presented new evidence for this (previously
known) correspondence between the folded and RSOS models based on the transfer
matrix method.
2. We conjectured that a similar correspondence exists between folded a2(2) models and
RSOS models based on a2(2) , and supported this by numerical diagonalisation of their
transfer matrices.
3. We presented substantial evidence that unrestricted (both the folded and the original)
a2(2) models have complex finite volume spectra in general, perhaps with the exception
of a few special values of the coupling constant.
4. Similarly, it seems that RSOS theories based on a2(2) have complex spectra in general,
with the exception of some special sequences asymptotically approaching the special
values of the coupling for which the unrestricted models may have real spectrum.
These sequences in particular include the perturbations of the unitary minimal models,
for which we know that the spectrum is real.
There are quite a few open questions remaining. First of all, the NLIE description must
be extended to excited states of a2(2) related models; this could provide us conclusive
(2)
evidence on whether or not the spectra of the unrestricted a2 theories are real for the
special values of the coupling constants for which the transfer matrix calculations found
no violation of reality. On the other hand, our results that show that the spectrum is complex
in general must be reproduced by such an extension of the NLIE.
Second, the projection of the folded spectrum to the RSOS spectrum must be explicitly
realized. Together with the extension of the NLIE to excited states, this would open the
way to (a) a systematic description of the spectra of the RSOS models; (b) determining
whether the sequences for which we found no violation of reality with our methods really
have real spectra.
It seems appropriate to mention that another interesting problem is the relation of the
(2)
a2 model and ‘φ21 ’ perturbations to the corresponding q-state Potts models. See [33] for
TBA equations for the ground states of these models and for a discussion of the particle
spectra in the q-state Potts models.
(2)
Third, as we mentioned in the text, the full spectrum of a2 related theories is not
yet closed in full generality. An example is M3,5 + φ12 , for which the closed spectrum
is not known up to date. One could attempt to close the S matrix bootstrap (we know of
some attempts that failed, and we ourselves tried unsuccessfully); but equally well, it could
(2)
possibly be found by extending the NLIE to describe the excited state spectrum of a2
related theories.
Acknowledgements
The authors would like to thank P. Dorey, H. Saleur and R. Tateo for valuable comments
and discussions. We also thank G. Mussardo for providing a correct set of q-6j symbols for
the φ12 perturbations. This work was supported by a Royal Society joint project grant. G.T.
is supported by a Zoltán Magyary Fellowship from the Hungarian Ministry of Education,
and is partly supported from Hungarian funds OTKA T029802/99 and FKFP-0043/2001.
G.T. is also grateful to the Department of Mathematics, King’s College, London for their
hospitality.
Appendix A. Scattering amplitudes for Mp,p + Φ1,2
The scattering amplitudes for the fundamental kinks in Mp,p +Φ1,2 were written down
by Smirnov in [15]. Here we briefly recall his result and give explicit expressions for the
necessary q-6j symbols as it seems there are none available in the literature that are free
of misprints and errors.
The allowed vacuum sequences {j1 , . . . , jn } have the following adjacency rules:
jk+1 = 1, if jk = 0,

jk+1 ∈ max(0, jk − 1) · · · min(jk + 1, p − 3 − jk ) , if jk = 0 (A.1)
and are of two types: either all jk are integers or all jk ∈ Z + 1/2. The maximum value
for jk is jmax = (p − 2)/2. The adjacency rules for these two sequences are shown on
Fig. 6 for the case when jmax is integer; for the case jmax ∈ Z + 1/2 two very similar
graphs result. Let us introduce the notation

iπp 2πϑ 2 πp
q = exp , x = exp , where ξ = , (A.2)
p ξ 3 2p − p
Fig. 6. Adjacency rules for kinks in Φ12 perturbations for jmax ∈ Z.
and the q-numbers

q n − q −n
[n] = . (A.3)
q − q −1
Then the 2-kink scattering amplitudes take the form
cd
Sab (ϑ) = cd
Sab (x, q)S0 (ϑ),

(q 2 − 1)(q 3 + 1) 1 − 2 (Cb +Cd −Ca −Cc ) 1 c
3 1 d
cd
Sab (x, q) = 5
δ bd + − 1 q 2
q2 x 1 a b q

1 c d
+ (1 − x)q − 2 + 2 (Cb +Cd −Ca −Cc )
3 1
, (A.4)
1 a b q
where
−1
1 π π 2πi
S0 (ϑ) = ± sinh (ϑ − πi) sinh ϑ−
4i ξ ξ 3
∞ ξ

sin kϑ sinh πk cosh( π
− )k
× exp −2i 3 6
ξk
2
dk . (A.5)
k cosh πk2 sinh 2
0
Ca = a(a + 1) (A.6)
and the q-6j symbols take the form

1 a−2 a−1 1 a+2 a+1
= = 1,
1 a a−1 q 1 a a+1 q

√
1 a+1 a+1 1 a a [2a + 4][2a]
= = ,
1 a a q
1 a + 1 a + 1 q [2a + 2]

1 a a+1 [2]
= ,
1 a a + 1 q [2a + 1][2a + 2]

1 a a−1 [2]
= ,
1 a a − 1 q [2a][2a + 1]

√
1 a+1 a 1 a a+1 [2a + 4][2a]
= = ,
1 a a+1 q 1 a+1 a q [2a + 2]

√
1 a a−1 1 a a+1 [2a − 1][2a + 3]
= = ,
1 a a+1 q 1 a a−1 q [2a + 1]

1 a a 1 a a+1 [2] [2a + 3]
= =− ,
1 a a+1 q 1 a a q [2a + 2] [2a + 1]

1 a a 1 a+1 a [2]
= = ,
1 a+1 a q 1 a a q [2a + 2]

1 a a 1 a a−1 [2] [2a − 1]
= = ,
1 a a−1 q 1 a a q [2a] [2a + 1]

1 a a 1 a−1 a [2]
= =− ,
1 a−1 a q 1 a a q [2a]

1 a a [2a − 1][2a + 3] − 1
= . (A.7)
1 a a q [2a][2a + 2]
We remark that the square roots in (A.7) carry a sign ambiguity which must be resolved
by an appropriate choice of the branches of the square root function. One must consider
the square roots of each q-numbers separately and fix a sign for the expressions

[n], n = 1, 2, . . . (A.8)
and then use the selected representative consistently in all formulas. Such a choice of
branch is necessary in order for the amplitudes to satisfy the Yang–Baxter equation,
R-matrix unitarity and appropriate crossing relations.
Appendix B. Scattering amplitudes for Mp,p + Φ1,5
In this appendix we present a corrected version of the S-matrix of Φ1,5 perturbations

written down in the paper [7] by one of the authors. The original formulas have misprints
and some of them have the wrong normalisation factors. The correct ones can be obtained
by imposing crossing symmetry and RU on the amplitudes. The choice of the normalisation
factors amount to defining the scalar product of the multi-kink states correctly (i.e.,
satisfying the constraints imposed by the quantum group symmetry).
In Φ1,5 perturbations, the allowed vacuum sequences are composed of highest weights
of the group Uq 4 (sl(2)) where

iπp
q = exp . (B.1)
4p
They are labelled by jk = 0, 1/2, . . . , jmax with jmax = (p − 2)/2. The adjacency rules for
a sequence {j1 , . . . , jn } are |jk+1 − jk | = 0 or 1/2 (Fig. 7). The zero difference corresponds
Fig. 7. Adjacency rules for kinks in Φ15 perturbations.
to a neutral kink, while the non-zero to charged ones. Let us introduce the notation

π 4 πp
y = exp ϑ , ξ= ,
ξ 3 p − 2p
q 4z − q −4z
[z]4 = . (B.2)
q 4 − q −4
The scattering amplitudes for charged kinks are (with a sign corrected with respect to [7])
2
y q 1 [2b + 1]4 [2d + 1]4 1/2
cd
Sab = − − 2 − + q δac
q y q [2a + 1]4 [2c + 1]4
2 5
y q 1
+ − − + q δbd S0 (ϑ), (B.3)
q5 y2 q
where S0 (ϑ) is the function defined in (A.5). For the sign ambiguities occurring as a result
of square roots of q-numbers see the remark at the end of Appendix A.
The remaining amplitudes include neutral kinks. The neutral kink-neutral kink scatter-
ing, neutral kink-charged kink forward scattering and neutral kink-charged kink reflection
were correct in the original paper [7] and are the following
q 6 y 2 + y 2 q 8 − q 8 − q 4 y 2 + y 2 − q 10 y 2 + y 4 q 2 − y 2 q 2
aa
Saa = S0 (ϑ),
y 2q 5
(y 2 + q 6 )(y 2 − 1)
ba
Sab = S0 (ϑ),
y 2q 3
(q 4 − 1)(y 2 + q 6 )
ba
Saa =− S0 (ϑ). (B.4)
yq 5
The amplitudes describing two charged kinks turning into two neutral ones were
incorrectly normalised (they include a non-trivial Clebsh–Gordan in the unrestricted
theory [7]). The correct form is
caa (q 4 − 1)(y 2 − 1)
aa
Sab =i . (B.5)
cba q 2y
For the reverse process the amplitude is
cab (q 4 − 1)(y 2 − 1)
ab
Saa =i , (B.6)
caa q 2y
where the normalisation factors cab are

α (−1) 2a [2b+1]4 , a = b ± 1 ,
cab = 1 [2a+1]4 2 (B.7)
α2 , a = b.
α1,2 are constants which are left free by the constraints of RU and crossing. The cab appear
in the crossing relations, which take the form
cad bc
cd
Sab (ϑ) = S (iπ − ϑ). (B.8)
cbc da
Appendix C. Transfer matrices and the Bethe–Yang equations
C.1. RSOS transfer matrices
When we put the theory on a cylindrical space-time with spatial volume L, the allowed
sequences of vacua (see Appendices A and B for φ12 and φ15 perturbations, respectively,
and [14] for the φ13 case) are further restricted by the condition a1 = aN+1 , where N
is the number of particles. We take all particles to be point-like and ignore all vacuum
polarisation contributions. Let us define the following (generalised) transfer matrix
b1 b2 bN b1
b b ...b
Ta11a22...aNN (ϑ|ϑ1 , ϑ2 , . . . , ϑN ) = ··· ϑ
a1 a2 aN a1
ϑ1 ϑ2 ϑN
which translates to

N
a bj+1
...aN =
T (ϑ|ϑ1 , ϑ2 , . . . , ϑN )ba11 ba22 ...b Sbjj+1 (ϑ − ϑj )
N
aj (C.1)
j =1
with the identification aN+1 ≡ a1 , bN+1 ≡ b1 . In the following we shall not always write
down the matrix indices explicitly. From the Yang–Baxter equation, it is straightforward to
prove that these transfer matrices form a commuting family

T (ϑ|ϑ1 , . . . , ϑN ), T (ϑ |ϑ1 , . . . , ϑN ) = 0. (C.2)
We define the following specialised transfer matrices
...aN = T (ϑk |ϑ1 , ϑ2 , . . . , ϑN )a1 a2 ...aN

Tk (ϑ1 , ϑ2 , . . . , ϑN )ba11 ba22 ...b N b1 b2 ...bN
b
a bj+1
= (−1)δ δakk+1 Sbjj+1aj (ϑk − ϑj ). (C.3)
j =k
Apart from a phase coming from a plane wave factor of rapidity ϑk in the wave function,
this gives the monodromy corresponding to taking the kth kink around the spatial circle
multiplied by a factor (−1)δ . The total phase must equal (−1)F depending on the statistics
of the particle k (F = 1 for fermions F = 0 for bosons). Thus we obtain the so-called
Bethe–Yang equations [21]:
exp(imk R sinh ϑk )Tk (ϑ1 , ϑ2 , . . . , ϑN )ba11 ba22 ...b N

...aN Ψ
a1 a2 ...aN
= (−1)F +δ Ψ b1 b2 ...bN , (C.4)
where Ψ a1 a2 ...aN is the wave function amplitude, defined by the decomposition of the state
|Ψ as follows

|Ψ = Ψ a1 a2 ···aN |Ka1 a2 (ϑ1 ) · · · KaN a1 (ϑN ). (C.5)
a1 ,...,aN
The energy and the momentum of the state (relative to the vacuum) are given by

N
N
E= mk cosh ϑk , P= mk sinh ϑk . (C.6)
k=1 k=1
Due to the commutation relation (C.2), Eq. (C.4) for the vector Ψ are compatible and can
be reduced to scalar equations by simultaneously diagonalising the transfer matrices. Let us
denote the eigenvalues of T (ϑ|ϑ1 , . . . , ϑN ) by λ(s)(ϑ|ϑ1 , . . . , ϑN ) with the corresponding
eigenvectors ψ (s) (ϑ1 , . . . , ϑN ) (s is just an index enumerating the eigenvalues and the
eigenvectors can be chosen independent of ϑ due to the commutativity (C.2)). Then the
solutions of the Bethe–Yang equations (C.4) are given by
Ψ a1 a2 ...aN = ψ (s) (ϑ1 , . . . , ϑN )a1 a2 ...aN , (C.7)

where the rapidities solve the scalar Bethe Ansatz equations
exp(imk R sinh ϑk )λ(s) (ϑk |ϑ1 , . . . , ϑN ) = (−1)F +δ . (C.8)
C.2. Folded transfer matrices
For folded models the vacua are labelled by an integer a modulo the folding number k.
The allowed sequences satisfy
ai+1 = ai + Q mod k, (C.9)

where Q are the possible topological charges of the solitons

±1, sine-Gordon,
Q= (C.10)
+1, 0, −1, ZMS.
In finite volume we require a1 = aN+1 mod k for periodic boundary conditions. Then
the transfer matrix has the same form as in Eq. (C.1) except that now the matrix S is
constructed from the scattering matrix
S of the sine-Gordon/ZMS model in the following
way:
(ϑ) =
cd ad Q ,Q
dc
Sab SQab ,Qbc (ϑ), (C.11)
where Qab denotes the charge of the soliton connecting the vacua a and b (see Fig. 8).
Fig. 8. Vacuum labels and topological charges for two-particle S-matrices in folded models.
References
[1] C. Mudry, C. Chamon, X.-G. Wen, Nucl. Phys. B 466 (1996) 383–443, cond-mat/9509054;
N. Read, H. Saleur, Nucl. Phys. B 613 (2001) 409–444, hep-th/0106124;
M.J. Bhaseen, J.-S. Caux, I.I. Kogan, A.M. Tsvelik, Nucl. Phys. B 618 (2001) 465–499, cond-mat/0012240.
[2] H. Saleur, B. Wehefritz-Kaufmann, Phys. Lett. B 481 (2000) 419–426, hep-th/0003217.
[3] S. Guruswamy, A. LeClair, A.W.W. Ludwig, Nucl. Phys. B 583 (2000) 475–512, cond-mat/9909143;
N. Read, H. Saleur, Nucl. Phys. B 613 (2001) 409, hep-th/0106124.
[4] H. Saleur, B. Wehefritz-Kaufmann, Nucl. Phys. B 628 (2002) 407–441, hep-th/0112095.
[5] G. Takács, G. Watts, Nucl. Phys. B 547 (1999) 538–568, hep-th/9810006.
[6] P. Dorey, R. Tateo, Nucl. Phys. B 571 (2000) 583–606, hep-th/9910102.
[7] G. Takács, Nucl. Phys. B 489 (1997) 532–556, hep-th/9604098.
[8] J.L. Cardy, G. Mussardo, Phys. Lett. B 225 (1989) 275–278.
[9] H.G. Kausch, G. Takács, G.M.T. Watts, Nucl. Phys. B 489 (1997) 557–579, hep-th/9605104.
[10] D.I. Olive, Nuovo Cimento 26 (1962) 73.
[11] J.L. Miramontes, Phys. Lett. B 455 (1999) 231–238, hep-th/9901145.
[12] T.R. Klassen, E. Melzer, Nucl. Phys. B 338 (1990) 485–528.
[13] Z. Bajnok, L. Palla, G. Takács, F. Wágner, Nucl. Phys. B 587 (2000) 585–618, hep-th/0004181.
[14] N. Reshetikhin, F.A. Smirnov, Commun. Math. Phys. 131 (1990) 157–178.
[15] F.A. Smirnov, Int. J. Mod. Phys. A 6 (1991) 1407–1428.
[16] A. Klümper, P.A. Pearce, J. Stat. Phys. 64 (1991) 13;
A. Klümper, M. Batchelor, P.A. Pearce, J. Phys. A 24 (1991) 3111.
[17] C. Destri, H.J. de Vega, Phys. Rev. Lett. 69 (1992) 2313;
C. Destri, H.J. de Vega, Nucl. Phys. B 438 (1995) 413–454, hep-th/9407117.
[18] D. Fioravanti, A. Mariottini, E. Quattrini, F. Ravanini, Phys. Lett. B 390 (1997) 243–251, hep-th/9608091.
[19] C. Destri, H.J. de Vega, Nucl. Phys. B 504 (1997) 621–664, hep-th/9701107.
[20] G. Feverati, F. Ravanini, G. Takács, Phys. Lett. B 430 (1998) 264–273, hep-th/9803104;
G. Feverati, F. Ravanini, G. Takács, Nucl. Phys. B 540 (1999) 543–586, hep-th/9805117;
G. Feverati, F. Ravanini, G. Takács, Phys. Lett. B 444 (1998) 442–450, hep-th/9807160.
[21] T.R. Klassen, E. Melzer, Nucl. Phys. B 382 (1992) 441–485, hep-th/9202034.
[22] P.E. Dorey, R. Tateo, Nucl. Phys. B 482 (1996) 639–659, hep-th/9706140;
P.E. Dorey, R. Tateo, Nucl. Phys. B 515 (1998) 575–623, hep-th/9607167.
[23] Al.B. Zamolodchikov, Nucl. Phys. B 432 (1994) 427–456;
Al.B. Zamolodchikov, Phys. Lett. B 335 (1994) 436–443.
[24] G. Feverati, F. Ravanini, G. Takács, Nucl. Phys. B 570 (2000) 615–643, hep-th/9909031.
[25] P. Zinn-Justin, J. Phys. A 31 (1998) 6747–6770, hep-th/9712222.
[26] S. Coleman, H.J. Thun, Commun. Math. Phys. 61 (1978) 31.
[27] V.B. Lidskii, USSR Comput. Math. Math. Phys. 6 (1966) 73–85;
J. Moro, J.V. Burke, M.L. Overton, SIAM J. Matrix Anal. Appl. 18 (1997) 793–817.
[28] G. Mussardo, Int. J. Mod. Phys. A 7 (1992) 5027–5044.
[29] F. Ravanini, M. Stanishkov, R. Tateo, Int. J. Mod. Phys A 11 (1996) 677–698, hep-th/9411085.
[30] M.J. Martins, Phys. Rev. Lett. 69 (1992) 2461–2464, hep-th/9205024;
M.J. Martins, Nucl. Phys. B 394 (1993) 339–355.
[31] E. Melzer, Supersymmetric analogs of the Gordon-Andrews identities, and related TBA systems, hep-
th/9412154.
[32] P.E. Dorey, C. Dunning, R. Tateo, Nucl. Phys. B 578 (2000) 699–727, hep-th/0001185.
[33] P.E. Dorey, A. Pocklington, R. Tateo, Integrable aspects of the scaling q-state Potts models I: bound states
and bootstrap closure and Integrable aspects of the scaling q-state Potts models II: finite size effects, in
preparation.
Non-commutative Chern–Simons for the quantum

Hall system and duality
Eduardo Fradkin, Vishnu Jejjala, Robert G. Leigh
Department of Physics, University of Illinois at Urbana-Champaign, 1110 W. Green Street,
Urbana, IL 61801, USA
Abstract
The quantum Hall system is known to have two mutually dual Chern–Simons descriptions, one
associated with the hydrodynamics of the electron fluid, and another associated with the statistics.
Recently, Susskind has made the claim that the hydrodynamical Chern–Simons theory should be
considered to have a non-commutative gauge symmetry. The statistical Chern–Simons theory has a
perturbative momentum expansion. In this paper, we study this perturbation theory and show that
the effective action, although commutative at leading order, is non-commutative. This conclusion is
arrived at through a careful study of the three-point function of Chern–Simons gauge fields. The non-
commutative gauge symmetry of this system is thus a quantum symmetry, which we show can only
be fully realized through the inclusion of all orders in perturbation theory. We discuss the duality
between the two non-commutative descriptions.
1. Introduction
Fractional quantum Hall (FQH) states are topological quantum fluid phases of two-
dimensional electron gases in large magnetic fields. These incompressible fluid states have
universal wave functions, the Laughlin wave functions [1] and their hierarchical extensions
[2–4], and their associated spectrum of low-energy excitations have universal quantum
numbers, such as their charge and statistics. Smooth, short distance, changes in the precise
form of the electron–electron interactions do not change the universal properties of these
ground states and of their low energy spectra. On closed surfaces, the ground states of FQH
E-mail addresses: efradkin@uiuc.edu (E. Fradkin), jejjala@uiuc.edu (V. Jejjala), rgleigh@uiuc.edu

(R.G. Leigh).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 6 1 6 - 8
484 E. Fradkin et al. / Nuclear Physics B 642 [FS] (2002) 483–500
systems exhibit degeneracies which are a direct manifestation of the topological nature of
these fluid states [5,6].
The universal properties of FQH states can also be described in the form of an effective
quantum field theory for the degrees of freedom with energies low compared to the
cyclotron energy h̄ωc and/or the Coulomb energy e2 /() √ (whichever is smallest), and
at distances long compared to the magnetic length = eh̄/cB (and/or any other short
distance scale related to the interaction). The low energy theory of the FQH states is a
Chern–Simons gauge theory. There are two equivalent, mutually dual, descriptions of this
effective theory. In the first, the Chern–Simons gauge theory is derived directly from a
theory of interacting electrons [7–9], and the gauge field is the statistical gauge field. The
second description is based on a hydrodynamic picture [10–15]. In this picture the gauge
fields embody the conserved charge densities and currents of the electron fluid. Naturally,
both descriptions, which are dual to each other, yield an effective action for slowly varying
electromagnetic gauge fields which also has a Chern–Simons form, a result required by
gauge invariance and broken time-reversal symmetry. Thus Chern–Simons gauge field
theories appear as the natural field-theoretic description of the low energy physics of FQH
states.
In Ref. [16], Susskind argued that quantum Hall systems are inherently non-commu-
tative, and that their effective field theory should be a non-commutative extension of the
Chern–Simons gauge theory. From a heuristic point of view, this is expected, because of the
non-commutative magnetic algebra of the quantum mechanics of electrons in a magnetic
field [17]. Thus, when acting on the Hilbert space of excitations of an incompressible
(gapped) state, physical observables such as the charge and current densities, obey a local
extension of the magnetic algebra [17], a magnetic current algebra related to the W∞
symmetry of area preserving diffeomorphisms [18–20]. The physical observables of this
magnetic current algebra obey Moyal products controlled by the magnetic length .
The non-commutative gauge theory that Susskind discusses is an effective hydrody-
namic theory of the fractional quantum Hall states. Thus, as such it describes only the low
energy physics of these incompressible fluid states. It is an extension of the conventional
effective Chern–Simons descriptions which have been used quiet successfully to describe
interesting phenomena such as edge states. Susskind’s non-commutative hydrodynamics
is a natural extension of these theories. However, Susskind’s proposal poses a number of
questions. For instance, the non-commutativity of Susskind’s theory, at the level of the ef-
fective action of Ref. [16], is controlled not by the magnetic length , but instead by the par-
ticle density ρ (or rather, by the mean separation between the electrons). It is thus natural to
inquire what is the connection between Susskind’s non-commutative Chern–Simons gauge
theory and the magnetic current algebra as they superficially appear to be controlled by
different (although obviously related) non-commutativity parameters. On the other hand,
since the non-commutativity happens on such a short-distance scale, it is also necessary
to explain in what sense this theory is robust, i.e., insensitive to other non-universal short
distance physical phenomena not considered in this theory. For instance, at short distances
the particular form of the interaction matters since it is what determines how big the energy
gaps are and how fast do the excitations move. We will see below that what is special about
the non-commutativity is that it is the only short distance effect associated with the broken
time-reversal symmetry and parity of electrons in magnetic fields, and that in this sense
E. Fradkin et al. / Nuclear Physics B 642 [FS] (2002) 483–500 485
this structure is robust. Nevertheless, the effects of non-commutativity show up primarily

in non-uniform states and in practice both the time-reversal even and odd processes are
important. For example, it is well known from the classic work of Girvin et al. [21] that the
spectrum of collective modes with a finite length scale, such as the roton modes, is due to
a combination of the magnetic algebra and electron–electron interaction effects.
In this paper, we provide a complementary derivation of Susskind’s result, based directly
on the field theory of interacting electrons coupled to gauge fields. Our calculation is
an extension of the standard methods of Ref. [8], which were used before to derive the
conventional Chern–Simons effective gauge theory and the universal physics they contain:
the values of the quantum Hall conductivity for which the FQH state is stable, the ground
state degeneracy, and the quantum numbers of the excitations. Here we will show that a
non-commutative Chern–Simons gauge theory emerges naturally within this framework.
We will find that to lowest order of approximation, the non-commutativity parameter does
not agree with the one found by Susskind. However, we will show that Susskind’s result is
required by the constraints imposed by the magnetic current algebra of the charge densities
and currents [22]. Hence, higher order quantum corrections lead to finite renormalizations
of the effective non-commutativity parameter. In contrast the level of the Chern–Simons
theory is not renormalized beyond its Gaussian value, due to the constraints of topological
and gauge invariance, and incompressibility. We will show that these results are universal
in the sense that these are the only effects at this length scale which are odd under time-
reversal (and parity).
Throughout the paper, we shall use the following notation:
aµ : statistical CS gauge field,

bµ : hydrodynamical CS gauge field,
Aµ : electromagnetic gauge field.
The paper is organized as follows. In Section 2, we briefly review Susskind’s fluid
dynamics derivation of non-commutative Chern–Simons theory for the Laughlin FQH
states. In Section 3, we review the derivation of commutative Abelian Chern–Simons
theory for the FQH states in the Jain sequences, based on the notion of flux attachment,
as the effective field theory of FQH states. In Section 4, we present the results of our
calculation of the effective gauge field action, obtained by integrating out fermions. The
bulk of this calculation, as well as notational details are relegated to Appendix A. In
Section 5, we discuss the non-commutative version of Chern–Simons duality. Here we
discuss the connection between non-commutative Chern–Simons theory, the underlying
magnetic algebra and the W∞ algebra of the currents. Finally, in Section 6 we present our
conclusions.
2. Hydrodynamic Chern–Simons theory and non-commutativity
Susskind’s discussion concerns the hydrodynamic Chern–Simons field. We briefly

review that derivation here, largely to establish notation. In the hydrodynamic limit, we
have a charged fluid of number density ρ in a magnetic field B. A fluid element is labeled
by coordinates x a (y), the y i being co-moving coordinates. The Lagrangian is taken to be

M 2 eB
L = d y ρ0
2
ẋ + ab x ẋ − V (ρ) .
a b
(2.1)
2 a 2
The gauge generators are of the form

1
G= d 2 y πa δx a
M
a
∂ ωc b ∂x
= ρ0 d y Λ(y) i ẋa + ab x
2 ij
, (2.2)
∂y 2 ∂y j
where ζ i (y) = ij ∂j Λ is the transformation of y i . This symmetry, when restricted to the
lowest Landau level, reduces to area preserving diffeomorphisms (APD) in space. The
conservation of G implies that
e
VΣ − !Σ = constant, (2.3)
M
where VΣ is the vorticity and ΦΣ is the magnetic flux through a region Σ.
Assuming that the fluid is near static equilibrium at ρ ρ0 , we may expand
x a = y a + θ ab bb (y), (2.4)
where θ ≡ 1/2πρ0 . The Lagrangian becomes

Mθ
L= d 2 y ωc ij ḃi bj + ḃi2 − cs2 (∇ × b)2 (2.5)
4π
while Eq. (2.3) becomes

e ωc
Φ0 + ωc θ b − θ 2 dbi ḃi + ij bj = constant. (2.6)
M 2
In the vacua of interest, the vorticity vanishes, and we find the constraint

∂ ωc ∂bk
ij j ωc bi − θ ḃk + kl bl = 0. (2.7)
∂y 2 ∂y i
This may be enforced by introducing a Lagrange multiplier b0 , and consequently, we find
that the action is (at leading order in powers of B)

h̄θ θ ij
S= d 3
y µνρ
b µ ∂ν b ρ + ∂i b µ ∂j b ν b ρ . (2.8)
4π2 3
Now the supposition is that this action should be interpreted as the leading term in a
momentum expansion of

k 2i
SNCCS = h̄ 3
d y µνλ
bµ , ∂ν bλ − bµ , bν , bλ , (2.9)
4π 3
where the ,-product is

i ij ∂ ∂
f (x) , g(x) = exp θ f (y)g(z) . (2.10)
2 ∂y i ∂zj y=z=x
Thus the field bµ is a (hydrodynamic) non-commutative Abelian Chern–Simons gauge

field. The non-commutativity is controlled by the parameter θ = 1/2πρ0 = 2 /ν. If we
couple this to the electromagnetic potential and compute the Hall conductance, we find
that the filling fraction is given by ν = 1/k. Hence, this theory describes the quantum
hydrodynamics of fractional quantum Hall states in the Laughlin sequence ν = 1/k, with
k an odd integer.
3. Duality invariant formalism: commutative limit
Now let us review the dual Chern–Simons pictures of the FQH states following the
arguments of Refs. [23,24]. Let us describe the problem of electrons in a large magnetic
field in a Feynman path-integral picture. We will denote the histories of the electrons in
(2 + 1)-dimensional space–time by a set of conserved 3-currents jµ (with µ = 0, 1, 2). Let
us write the (fermionic) path-integral in the form of a sum over current configurations (i.e.,
a sum over the histories of the particles)
iS[j ]/h̄−i e d 3 x A j µ +iφ[j ]
Z= e h̄c µ
, (3.1)
[j ]
where φ[j ] is a phase which accounts for the statistics of the particles. Here, S[j ] is the
action for N interacting fermions, Aµ is the external magnetic field.
Next, we note that for every configuration of currents [j ] it is possible to compute
the linking number νL ([j ]) of the configuration, which is a topological invariant and it
is an integer. (Technically this requires the assumption that the currents do not cross or
equivalently that there is a “hard core” condition.) The expression of νL [j ] in terms of j is
known as the Gauss invariant, which is non-local:

νL [j ] = j ∧ ∂ −1 · j. (3.2)
We will express the fact that the currents jµ are conserved, i.e., ∂ · j = 0 by introducing
[11] a hydrodynamic field b, related to the particle current by
1 µνλ
jµ = ∂ν bλ , (3.3)
2π
which we will denote in short hand form as j = 2π 1
∂ ∧ b.
When written in terms of b, the link invariant is local and it has a Chern–Simons form
[25,26]:

1
νL [j ] = d 3 x b · ∂ ∧ b. (3.4)
4π 2
Hence, for any arbitrary integer n, we can shift the action S[j ] → S[j ] + 2πinh̄νL [j ]
without changing any quantum mechanical amplitude.
To enforce current conservation, we insert bµ into the partition function by making use
of the identity
3
1
∂ ∧ b = [Db][Da]ei d x a· j − 2π ∂∧b .
1
1 = [Db]δ j − (3.5)
2π
To represent the δ-function, we have introduced another vector field a. Thus, we may write
the full fermionic path integral in the equivalent form
iS[j ]/h̄+iφ[j ]+i d 3 x (a− e A)·j − 1 a·∂∧b+ 2n b·∂∧b
Z = [Db][Da] e h̄c 2π 4π . (3.6)
[j ]
Hence, a system of interacting fermions in an external electromagnetic field is equivalent

to another system with two gauge fields, aµ and bµ , with the fermions minimally coupled
to aµ − (e/h̄c)Aµ . The gauge field aµ is the statistical gauge field of Refs. [7,8,27], and it
corresponds to 2n fluxes being attached to each fermion. Conversely, the gauge field bµ is
the hydrodynamic field used by Wen [5,11,12]. As noted in Ref. [24], the effective action
of Eq. (3.6) obeys the Chern–Simons level quantization condition [28,29], and thus it is a
consistent form of flux attachment, on both open and closed manifolds.
4. The non-commutative effective action for the statistical Chern–Simons field
We will now use this description of interacting electrons in magnetic fields to give a
derivation of Susskind’s non-commutative hydrodynamics. It was shown in Refs. [8,24],
that for systems with filling factors ν in the Jain sequences [4], where ν −1 = 2n + p−1 and
n and p are integers, at a mean-field level the statistical field screens the external magnetic
field B down to the effective field Beff = B/(2np + 1). At this level of approximation the
fermions fill up an integer number p of the Landau levels of Beff . Naturally, this is just
Jain’s picture of composite fermions filling up effective Landau levels [4]. Fluctuations
about this mean-field state play a key role.
Now, if we integrate out the fermions, i.e., we sum over all current configurations
including the effects of the fermion phase factor as well as interaction effects, we will
clearly obtain an effective action for ã and b̃, the fluctuating pieces of the statistical gauge
field a and the hydrodynamic gauge field b. (Formally, this is done by expanding the
fermion determinant in powers of the gauge fields):

h̄2 2
Sfermion = −i h̄ tr ln i h̄D0 + µc − D + Sint (b̃), (4.1)
2Mc
where Sint (b̃) is the term of the action due to electron–electron interactions; here we have
used the constraint relating the current to the hydrodynamic gauge field. The details of this
calculation are given explicitly by Lopez and Fradkin in Ref. [8,24,33]. They computed the
leading quadratic term in ã, which amounts to a calculation of the polarization tensor of
electrons filling up p effective Landau levels. The resulting expression is rather complex
and depends explicitly on the form of the interactions. However, to lowest order in the
momentum expansion, i.e., at distances long compared to the effective magnetic length
eff , it is simply given by (see Fig. 1)

h̄p
Seff [ã] = d 3 x µνλ ãµ ∂ν ãλ + · · · . (4.2)
4π
Fig. 1. Two point function for ã.
Other effects, such as the contributions from the electron–electron interaction, appear at
higher orders in derivatives (and in the parity even sector only). Thus, to leading order we
find the following effective Lagrangian [24]
h̄p µνλ h̄ µνλ 2nh̄ µνλ

Leff ã, b̃, Ã = ãµ ∂ν ãλ + ãµ ∂ν b̃λ − b̃µ ∂ν b̃λ
4π 2π 4π
1 e µνλ
− Ãµ ∂ν b̃λ , (4.3)
2π c
where Ã is a small fluctuation of the external electromagnetic field. Note that if we integrate
out ã, at leading order we get (up to boundary conditions)

1 h̄ µνλ
Leff [b̃] = − 2n + b̃µ ∂ν b̃λ + · · · . (4.4)
p 4π
The coefficient of Leff [b̃] is ν −1 = 2n + 1/p. Thus p = 1 corresponds to the Laughlin

series. In particular, also for p = 1, this is the form of the action that follows from
hydrodynamic arguments [6,11–14]. It is straightforward to show [8] that this theory leads
to the correct value of the Hall conductivity
ν e2
σxy = , (4.5)
2π h̄c
where
1 1
= 2n + . (4.6)
ν p
This result is exact because it saturates a Ward identity, i.e., the f -sum rule, as discussed
in Refs. [8,9].
The coefficient of the Chern–Simons action is known as the Chern–Simons level.
General topological consistency arguments [28,29] show that if the theory is quantized
on a closed manifold, the level of a Chern–Simons action must be quantized for the theory
to be invariant under large gauge transformations. This requirement is met by the effective
action of Eq. (4.3), which contains both the statistical and the hydrodynamic gauge fields,
but not by either the effective action for the statistical field or the hydrodynamic field alone
except for the Laughlin sequence, p = 1. We will return to this issue in the context of the
non-commutative theory in Section 5.
4.1. Trilinear Chern–Simons couplings
Hence, to quadratic order in fluctuations and at long distances, one finds a commutative
Abelian Chern–Simons gauge theory. We will investigate now the trilinear terms in this
expansion.
We have seen that to quadratic order in ã, and to lowest order in a gradient expansion, the
effective action for a Laughlin FQH state and its Abelian generalizations is a (commutative)
Abelian Chern–Simons theory. We will show now that the contributions to the effective
action trilinear in ã make the effective theory non-commutative. The next correction,
(3)
Seff [ã], is trilinear in the fields ã, and it is shown in Fig. 2. The details are given in
Appendix A.
In momentum space, we evaluate

(3) 1
Seff [ã] = d 3 P d 3 Q d 3 R Π µνλ (P , Q, R)ãµ (P )ãν (Q)ãλ (R). (4.7)
3!
It is difficult to evaluate the resulting sums and integrals entering in Π µνλ (P , Q, R) in
closed form. They are given in the form of a series with frequency poles at integer multiples
of the cyclotron frequency and with residues dictated by gauge invariance multiplied by
factors involving Laguerre polynomials of the variable P 2 /2Beff . The odd-parity terms of
this tensor contain a phase factor of the form

exp −i2eff P ∧ Q/2 (4.8)
which we recognize as a Moyal phase. (To be more precise it is the Fourier transform of
Eq. (2.10) with θ → 2eff .) Notice that here 2eff = h̄c/eBeff is the effective magnetic length,
which is related to the magnetic length by
p 2
2eff = (2np + 1)2 = . (4.9)
ν
To leading order in an expansion in momenta there is one component Π 012 which survives.
We find, up to an energy–momentum conserving δ-function
h̄p −i2 P ∧Q/2 2 2
Π 0[12] (P , Q, R) = e eff f P ,Q , (4.10)
2π
where f (P 2 , Q2 ) is a regular function of the momenta P and Q and f (0, 0) = 1.
We are being selective in keeping the momentum dependent phase. To be fully
consistent we should expand the exponential and group the momenta into higher derivative
Fig. 2. Diagrams contributing to the three point function for ã.

corrections. However, we claim that this phase is a Moyal phase indicative of the
non-commutativity of this theory. This phase automatically accompanies all Feynman
diagrams, and at least at one loop is the only odd parity contribution.
Thus, to the order of approximation we have kept, the effective Lagrangian for the
statistical gauge fields ã is a non-commutative Chern–Simons gauge theory, of the form of
Eq. (2.9), at level p:

h̄p µνλ 2i
Leff [ã] = ãµ ∂ν ãλ + ãµ , ãν , ãν + · · · , (4.11)
4π 3
where , denotes a Moyal product of Eq. (4.8) with non-commutativity parameter θa =
1/2eff . Hence, to the present order of approximation, the trilinear terms just calculated
turn the terms associated with the statistical gauge field ã in the effective action Seff [a, b]
of Eq. (4.3) into a non-commutative Chern–Simons gauge theory. The resulting effective
action still satisfies the quantization condition of the level of its Chern–Simons terms.
However, there are a number of features of this result that seem to be wrong. Although
this effective theory is non-commutative, the non-commutativity is associated with the
statistical gauge field ã instead of the hydrodynamic gauge field b̃ as in Susskind’s action.
In addition, the level of the non-commutative gauge theory is p instead of the denominator
of the filling factor as suggested by Ref. [16]. Likewise, the length scale of the non-
commutativity is neither the average particle separation as in Susskind’s action nor the
magnetic length as required by the magnetic current algebra, but instead the effective
magnetic length eff . We will show below, using the exact symmetries of the physical
system, that at higher orders there must be corrections which conspire to yield the correct
behavior for physical observables, such as the actual charge density and current.
These corrections are not hard to find. In fact the expansion of the fermion determinant
leads to terms in the effective action for ã with more than three external legs. In particular,
the terms with an odd number of external legs have parity-odd pieces involving Moyal
products as well as more derivatives. These terms cannot be neglected in the computation
of the effective action for the hydrodynamic field b̃.
5. The non-commutative duality
Let us now make contact with Susskind’s hydrodynamic effective theory. We will do
that by integrating out the statistical gauge field a, and calculating an effective action
for the hydrodynamic field b. Since the theory is now non-linear in ã, we will do this
in perturbation theory. Thus, we go back and consider the full non-commutative effective
Lagrangian of Eq. (4.3):

h̄p µνλ 2i h̄ µνλ

Leff [ã, b] = ãµ ∂ν ãλ − ãµ , ãν , ãλ + · · · + ãµ ∂ν bλ
4π 3 2π
2nh̄ µνλ e µνλ
− bµ ∂ν bλ − Ãµ ∂ν bλ + · · · (5.1)
4π 2πc
and integrate out ã perturbatively. Here · · · denotes terms with more that three factors of ã.
This expansion leads to an effective Lagrangian for the hydrodynamic field b, the dual
theory, of the form

h̄ µνλ 2i e µνλ
Leff [b] = − bµ ∂ν bλ − bµ , bν , bλ + · · · − Ãµ ∂ν bλ . (5.2)
4πν 3 2πc
In order to write the effective action in this form, we have rescaled the ,-product. At the
present level of approximation, the effective non-commutativity parameter appearing in
Eq. (5.2) is
1 2
θb = + ··· (5.3)
2np + 1 ν
and the terms denoted by · · · are contributions to the trilinear term in b originating from
diagrams in the expansion on ã of the type shown in Fig. 3. Such contributions, among
other things, lead to finite renormalizations of the non-commutativity parameter.
For the particular case of the Laughlin sequence where p = 1 and ν = 1/(2n + 1), we
obtain an effective hydrodynamic theory of Susskind’s form at level k = 2n + 1 but with an
effective non-commutativity parameter θ = 2 /νk. In contrast, in Susskind’s theory [16]
the non-commutativity parameter is θ = 2 /ν = 1/(2πρ), i.e., the length scale of non-
commutativity is the mean distance between electrons.
The effective hydrodynamic theory of Eq. (5.2) is dual to the theory described in
terms of the statistical gauge field a. Under this transformation particles and fluxes
are interchanged. This is just the particle-flux duality which underlies the physics of
the fractional quantum Hall states. Hence, the effective theory of Eq. (5.2) is the non-
commutative extension of the duality picture familiar from the commutative Chern–Simons
theories of the fractional quantum Hall effect [30].
The effective action Seff [Ã] for an electromagnetic perturbation Ã can be computed by
integrating out the hydrodynamic field b. To lowest order we find

νe2 2i e
Seff = d 3
x µνλ
Ã µ ∂ν Ã λ − Ã µ , Ã ν , Ã λ + · · · (5.4)
4π h̄c2 3 h̄c
with a non-commutativity parameter which is now
2
θA = + ···. (5.5)
2np + 1
This result implies that the Hall conductivity is σxy = νe2 / h, which is correct, but it
predicts a non-commutativity parameter θ which is not equal to 2 . Below we will present
Fig. 3. Two loop diagrams contributing to Π µνλ . The internal wavy lines are propagators of the statistical gauge
field ã.
an argument, essentially a Ward identity, which implies that for a translationally invariant
system in a magnetic field the non-commutativity parameter should be exactly equal to 2
without any renormalization. This result also implies that the non-commutativity parameter
of the hydrodynamic theory must be exactly equal to 2 /ν = 1/(2πρ) as in Susskind’s
theory. Hence, the non-linear terms in the effective action for the statistical field ã must
lead to perturbative contributions in the hydrodynamic theory whose total net effect is a
finite renormalization of its non-commutativity parameter to the correct value 2 /ν. In
contrast the level of the Chern–Simons action cannot be renormalized as it is protected by
topological invariance.
5.1. Magnetic Ward identity
In the previous sections we showed that the Moyal phase appearing in Eq. (4.10)
indicates that the effective action for ã should be considered non-commutative. However,
this result led to a number of inconsistencies which we will discuss here. As we noted
above, the Moyal phase appearing in Eq. (4.10) is controlled by Beff , which sets the length
scale relevant to the perturbation theory about the mean field state. However, as we noted
above, Beff is not a physical scale.
The mean field theory that we used here, usually called the average field approximation,
is known to break a number of symmetries of the physical electron gas which must be
preserved exactly. Let us reexamine what we have done. We first attached 2n flux quanta
to each electron. This procedure does not change anything and it does not break any
symmetry. However, at the mean field level, there are significant changes in the Hilbert
space: the system behaves like a system of composite fermions filling up an integer number
p of Landau levels of the partially screened magnetic field Beff . In particular, these effective
levels are mixtures of the original Landau levels. This is the physics of the (unprojected)
Jain wave functions [4]. If taken literally, this mean field theory makes a number of wrong
predictions. For instance it implies that the bosonic bound state with the leading residue,
i.e., a residue which vanishes like ∼ q 2 as q → 0 as required by gauge invariance, has an
energy gap at the effective cyclotron frequency ωeff = ω/(2np + 1). As it is well known,
this result violates Kohn’s theorem [31], which states that in a translationally invariant
system this gap must be at the exact cyclotron frequency ωc = eB/mc, without any mass
renormalization. Kohn’s theorem follows from the fact that the system as a whole must
obey the magnetic algebra of the full magnetic field B. (A similar approximation also leads
to a spurious zero-field Hall effect in theories of anyon superfluidity [32].) Similarly, the
excitation spectrum predicted by this mean field theory is wrong, since it predicts that the
excitations of this state are composite fermions instead of particles with fractional charge
and fractional statistics. The problem with the length scale in the Moyal phase has the same
origin.
The solution to these problems is well known. At the level of wave functions, the
correct physical behavior is recovered only after the states are projected onto the lowest
Landau level [4]. In the context of the Chern–Simons theories, which deal with electrons
as point particles and are not projected onto the lowest Landau level, the solution to
these problems is also well understood. It has been shown in Refs. [8,9,33], that unlike
other mean field theories, for fractional quantum Hall states, quantum fluctuations even
at the Gaussian level, restore some of the symmetries broken by the mean field state.
Physically this happens since, due to the Gauss Law constraint of the Chern–Simons theory,
a local density fluctuation is equivalent to a local flux fluctuation. In particular, the leading
quantum corrections, described by the Gaussian commutative effective action, yield upon
integrating out the gauge fields a and b, the correct electromagnetic polarization tensor
which saturates the f -sum rule in the q → 0 regime and thus it is exact in this limit [8,24,
33]. Hence, in the uniform limit this theory predicts, for the Jain states with filling factor
ν = p/(2np + 1), a Hall conductivity equal to
ν e2
σxy = ,
2π h̄
a cyclotron mode with frequency ωc and residue ∼ q 2 , a spectrum of quasiholes carrying
charge e/(2np + 1) and fractional statistics π/(2np + 1), and, for a surface of genus g,
a ground state degeneracy of (2np + 1)g . Thus the (Gaussian) commutative Chern–Simons
effective action of Eq. (4.3) yields a complete description of the universal long-distance
data of fractional quantum Hall states (for a recent discussion see Ref. [24]). (This is
no longer the case when the gap collapses as in the limiting states ν = 1/2n, which are
compressible states of the 2DEG [27].)
Let us see what this line of analysis implies for the non-commutative effective theory
of Eqs. (4.3) and (4.11). We noted before that in the q → 0 limit the Gaussian theory is
exact. However, in this regime the theory is commutative. Although at shorter distances,
the physics of the 2DEG is strongly non-universal, we showed above that in the odd-
parity sector, the theory becomes non-commutative. However, the length scale of the
non-commutativity is eff which is not a physical length scale. This inconsistency has
exactly the same origin as the violation of the global magnetic symmetry. As before, this
inconsistency is also resolved by quantum corrections to the mean field state.
To understand this problem, let us look first at the actual electromagnetic currents Jµ in
the system. (As usual, this is a 3-vector formed by the physical charge density fluctuation
and the two components of the charge current.) These currents are obtained as from
the effective action for the external electromagnetic gauge field Seff [Ã] = − log Z[Ãµ ],
where Z[Ã] is the partition function of the full 2DEG and Ã is a weak electromagnetic
perturbation. In 1992, Iso, Karabali and Sakita [18], and independently Martinez and
Stone [19], showed that the electromagnetic currents and charge densities for a 2DEG in a
large magnetic field B, in an incompressible state with filling factor ν, obey a W∞ algebra,
which is a local extension of the global magnetic algebra of finite translation operators in a
magnetic field. Let ρ̂(x, y) be the charge density operator projected on the Lowest Landau
level, and ρ̂(p, q) be its Fourier transform

ρ̂(p, q) = dx dy ρ̂(x, y)ei(px+qy)/h̄. (5.6)
These authors [17–19] showed that the projected density operators obey the algebra
2

ρ̂(p, q), ρ̂ p , q = −2i sin pq − p q e (pp +qq )/2 ρ̂ p + p , q + q .
2
2
(5.7)
In the holomorphic basis

ρ̂mn = dx dy ρ̂(x, y)zm z̄n (5.8)
the algebra becomes

min(n,k)
(−1)j n!k!
[ρ̂mn , ρ̂kl ] = ρ̂m+k−j,n+l−j − (n ↔ l)(m ↔ k).
j ! (n − j )!(k − j )!
j =1
(5.9)
Hence, the algebra of the charge density operators [17,18], as well as of the current density
operators as shown in Ref. [19], contains explicitly a Moyal product controlled by the
magnetic length scale of the full magnetic field B. This is a very general result which
follows from the incompressibility of the system and from the properties of the quantum
states in a magnetic field.
This result poses important constraints, which can be regarded as Ward identities, on the
values of the parameters of the effective action Seff [Ã] of an external electromagnetic field
in an incompressible quantum Hall state. In addition to the condition of gauge invariance,
which follows from current conservation, these results require that for a fractional quantum
Hall state at filling factor ν, the effective action Seff [Ã] must have the form of a non-
commutative Chern–Simons action whose level is ν (in units of e2 /h̄) and whose non-
commutativity parameter θ = 2 = h̄c/eB, where B is the full magnetic field. As we
saw above, this requirement in turn implies that the non-commutativity scale for the
hydrodynamic field is 1/(2πρ), where ρ is the average areal density. Therefore, we
conclude that the higher orders in perturbation theory, among other things, must necessarily
renormalize the non-commutativity parameter of the hydrodynamic field to this exact
value.
6. Conclusions
In this paper we have used the notion of flux attachment to derive Susskind’s non-
commutative hydrodynamic action for the fractional quantum Hall effect. We have
presented two mutually dual effective theories, which are non-commutative extensions of
the familiar (commutative) Chern–Simons descriptions of the FQHE. We have used this
duality to show that while the hydrodynamic effective action must have a non-commutative
parameter controlled by the average particle distance (as in Susskind’s action), the effective
action of the electromagnetic field must have a non-commutative parameter controlled only
by the magnetic length. We have also shown that while mean-field-like approximations
do lead to effective actions with the correct form of a non-commutative Chern–Simons
theory, the non-commutativity parameter is not correct at lowest order and must acquire
corrections in order to satisfy the constraints imposed by the global magnetic algebra, and
its local (Moyal) extension. This is in marked contrast with the Hall conductivity which is
already correct at the level of the Gaussian theory.
We would like to comment on the validity and usefulness of these non-commutative
effective theories. Clearly the commutative abelian Chern–Simons theory gives a faithful
description of the universal data encoded in fractional quantum Hall fluids: the Hall
conductivity, the fractional charge and statistics of the excitations and the global
topological degeneracy of their ground states. The non-commutative theory describes the
local extension of the global magnetic algebra, i.e., the W∞ algebra of the local currents
and densities of incompressible charged two-dimensional fluids in large magnetic fields.
In fact, as argued by Susskind, non-commutative Chern–Simons theory is a description
of the area-preserving diffeomorphisms of these incompressible fluids. Hence, in a sense,
except for the universal quantum numbers they encode, the information contained in these
effective theories is essentially kinematic in nature. Indeed, there is much dynamics taking
place at the scale of the magnetic (and Coulomb) lengths, such as the energetics of the
excitations, which is not described by these effective theories. Although in a formal sense
this is an inconsistency, what matters here is that the non-commutative terms included
are the only contributions at these length scales which are odd under parity and time-
reversal. Obviously a full description of the physics requires the (strongly non-universal)
microscopic physics which controls the energetics of these states.
These considerations are particularly important for applications of these ideas to non-
uniform ground states such as Wigner crystals, quantum Hall smectic and nematic states
(for recent work see Refs. [34–40]). Recent work on these interesting phases of the two-
dimensional electron gas have revealed that magnetic symmetry plays a crucial role in
low-energy dynamics of these states. Interestingly, recent work on non-commutative field
theories have suggested that their phase diagrams involve analogs of these phases [41].
It remains an important and open problem to understand the implications and restrictions
imposed by the non-commutative structure on the phase transitions between these states
[42,43].
Acknowledgements
Discussions with Moshe Rozali and Michael Stone are gratefully acknowledged. Work
supported in part by US Department of Energy, grant DE-FG02-91ER40677 (R.L.), and
by the National Science Foundation through the grant DMR-01-32990 (E.F.). V.J. has been
supported by a GAANN fellowship from the US Department of Education, grant 1533616.
Appendix A
In this appendix, we give details of the perturbative computations. For a detailed

discussion of the physics behind this theory, see Ref. [8]. We borrow their notation,
and pick up the discussion at an appropriate point. Within this appendix, we have set
e = h̄ = c = 1, and thus Beff = 1/2eff . The Feynman rules are obtained from the Lagrangian
density for non-relativistic fermions in a magnetic field:
i 1 2 †
L = ψ † Lψ − a0 ψ † ψ + aj Dj† ψ † ψ − ψ † Dj ψ + a ψ ψ, (A.1)
2M 2M
where
D = ∂ + iA, D † = ∂ − iA, (A.2)
L = i∂0 − A0 + µ + (1/2m)D 2. (A.3)
The propagator is written

ψ(x)ψ † (y) = iG(x, y). (A.4)
√
√ to rescale spatial coordinates u i = Beff xi and
It is convenient for this calculation
similarly rescale momenta Pi → Pi / Beff . We then have

dk ∗
iG(u, v) = S 0 0 ϕm,k u ϕm,k v , (A.5)
2π x ,y ,m
where for brevity we have written the summation symbol
∞ p−1
0 0 −iω (x 0 −y 0 )
Sx 0 ,y 0 ,m = θ x − y 0
−θ y − x 0
e m (A.6)
m=p m=0
and

ϕm,k u = Nm eikux e−(uy −k) /2 Hm (uy − k)
2
(A.7)
√
with normalization Nm = (2m m! π )−1/2 . We choose a gauge such that
(x)
iDx ∂
=α i + uy ≡ dx(u) , (A.8)
M ∂ux
(x)
iDy ∂
= iα ≡ dy(u) , (A.9)
M ∂uy
√
where α = Beff /M. These derivatives have the effect of shifting the indices of the Landau
functions:

dk ∗
da(u1 ) iG(u1 , u2 ) = α Sx 0 ,y 0 ,m (J )
cm,a ϕm+J,k u 1 ϕm,k u2 ,
2π
J

dk (J˜) ∗
da iG(u3 , u1 ) = α
(u1 )†
Sx 0 ,y 0 ,m ϕm,k u3 c̃m,a ϕm+J˜,k u1 ,
2π
J˜
where we have introducted constants c and c̃ whose non-zero components are:

m+1 m+1
cm,x =
(1)
, cm,y = −i
(1)
, (A.10)
2 2

m m
(−1)
cm,x = , (−1)
cm,y = +i , (A.11)
2 2

m+1 m+1
c̃m,x =
(1)
, c̃m,y = +i
(1)
, (A.12)
2 2

m m
(−1)
c̃m,x = , (−1)
c̃m,y = −i . (A.13)
2 2
In the three-point Feynman diagrams (Fig. 2), for each spatial component of a gauge field
that appears at an external leg, there is a spatial derivative of the form given above acting
on the fermion line. If we sum over all such possibilities, the net effect on a vertex is to
replace its Landau functions1 by a factor

∗
CJ˜1 ,J2 (m1 , m2 ; µ)ϕm +J˜ ,k
u2 ϕm2 +J2 ,k2 u 2 . (A.14)
1 1 1
J˜1 ,J2
It is not difficult to show that
CJ˜1 ,J2 (m1 , m2 ; 0) = δJ˜1 ,0 δJ2 ,0 , (A.15)

α (J2 ) (J˜1 )

CJ˜1 ,J2 (m1 , m2 ; a) = cm δ
2 ,a J˜1 ,0
+ c̃m δ
1 ,a J2 ,0
, a = x, y. (A.16)
2
C is essentially the “Fourier transformed” three-point vertex of ψ † aµ ψ.
We write the correction to the effective action as

1
δSeff = d 3 x1 d 3 x2 d 3 x3 aµ1 (x1 )aµ2 (x2 )aµ3 (x3 )Π µ1 µ2 µ3 (x1 , x2 , x3 ), (A.17)
3!
where Π is
3
dkj
Sx 0 ,x 0 ,mj CJ˜j ,Jj+1 (mj , mj +1 ; µj +1 )ϕmj +Jj ,kj uj
2π j j+1
j =1
J,J˜

∗

× ϕm u j . (A.18)
j−1 +J˜j−1,kj−1
(Note: actually, this is just one of the two Feynman diagrams. The other can contribute
only to Π aaµ .) To proceed, we Fourier transform the CS fields

aµ (x) = dP 0 d 2 P aµ (P )eiP x e−i P ·u .
0 0
(A.19)
Note that P here is the rescaled momentum.

It is then a simple matter to perform the integrals over the time coordinates; the net
effect is that the sums and θ -functions are replaced by 2πδ(P0 + Q0 + R0 ) times
∞ ∞ p−1 p−1 p−1 ∞
m1 =p m2 =p m3 =0 − m1 =0 m2 =0 m3 =p
T= + cyclic, (A.20)
(ωm3 − ωm1 + P0 )(ωm3 − ωm2 − R0 )
where we cycle simultaneously on P0 , Q0 , R0 and m1 , m2 , m3 .
Next, we can do the spatial integrals, using the orthogonality for the ϕ’s

∗ −i P ·u
d 2 u ϕm,k u ϕm e
,k u
2
= 2πδ k − k − Px e−P /4 e−iPy (k+k )/2 Lm,m P , (A.21)
1 The Landau functions come from the fermion propagator, but at each vertex there are two such functions.
where

m !

Lm,m P = θ m − m
(−P)m−m Lm−m |P|2
m! m

m! ∗ m −m m −m 2
+θ m −m P L |P| , (A.22)
m ! m
√
where we have defined P = (Px + iPy )/ 2.
Thus we find
3
dkj −Pj2 /4 −iPj,y (kj +kj−1 )/2
2πδ(kj − kj −1 − Pj,x )e e
2π
j =1

3

×T C˜ (mj −1 , mj ; µj )L
Jj−1 ,Jj ˜ Pj .
mj +Jj ,mj−1 +Jj−1 (A.23)
j =1
J,J˜
Performing the kj -integrations, we find

1 −i(P1,x P2,y −P1,y P2,x )/2
e
2π

3
−Pj2 /4
×T e CJ˜j−1 ,Jj (mj −1 , mj ; µj )Lmj +Jj ,mj−1 +J˜j−1 Pj . (A.24)
j =1
J,J˜
(We have dropped an overall energy–momentum conserving δ-function.) We note that

this result carries an overall Moyal phase, times a complicated function of energies and
momenta.
To proceed further, we need to evaluate the rather complicated sums in Eq. (A.24). We
will do so by expanding in energies and momenta (keeping the overall Moyal phase intact).
We find that there is a non-zero contribution to the partity odd polarization
p −i2 P1 ∧P2 /2 2 2
Π 0[12] (P1 , P2 , P3 ) = e eff f P1 , P2 , (A.25)
2π
where f (P 2 , Q2 ) is a regular function of the momenta P and Q and f (0, 0) = 1.
Contributions to other polarizations of Π are also calculable, but are of no relevance to
the discussions of this paper.
References
[1] R.B. Laughlin, Phys. Rev. Lett. 50 (1983) 1395.

[2] F.D.M. Haldane, Phys. Rev. Lett. 51 (1983) 605.
[3] B.I. Halperin, Phys. Rev. Lett. 52 (1984) 1583.
[4] J.K. Jain, Phys. Rev. Lett. 63 (1989) 199;
J.K. Jain, Phys. Rev. B 40 (1989) 8079;
J.K. Jain, Adv. Phys. 41 (1992) 105.
[5] X.G. Wen, Q. Niu, Phys. Rev. B 41 (1990) 9377.
[6] X.G. Wen, Phys. Rev. B 40 (1989) 7387.

[7] S.C. Zhang, T.H. Hansson, S. Kivelson, Phys. Rev. Lett. 62 (1989) 82.
[8] A. Lopez, E. Fradkin, Phys. Rev. B 44 (1991) 5246.
[9] S.C. Zhang, Int. J. Mod. Phys. B 6 (1992) 25.
[10] B. Blok, X.G. Wen, Phys. Rev. B 43 (1991) 8337.
[11] X.G. Wen, Adv. Phys. 44 (1995) 405.
[12] X.G. Wen, A. Zee, Phys. Rev. B 46 (1992) 2290.
[13] J. Fröhlich, A. Zee, Nucl. Phys. B 364 (1991) 517.
[14] J. Fröhlich, T. Kerler, Nucl. Phys. B 354 (1991) 369.
[15] S. Bahcall, L. Susskind, Int. J. Mod. Phys. B 5 (1991) 2735.
[16] L. Susskind, The quantum Hall fluid and non-commutative Chern–Simons theory, hep-th/0101029.
[17] S.M. Girvin, T. Jach, Phys. Rev. B 29 (1984) 5617.
[18] S. Iso, D. Karabali, B. Sakita, Phys. Lett. B 296 (1992) 143.
[19] J. Martinez, M. Stone, Int. J. Mod. Phys. B 7 (1993) 4389.
[20] A. Capelli, C.A. Trugenberger, G.R. Zemba, Nucl. Phys. B 396 (1993) 465.
[21] S.M. Girvin, A.H. MacDonald, P. Platzman, Phys. Rev. B 33 (1986) 2481.
[22] C.D. Fosco, A. López, Aspects of non-commutative descriptions of planar systems in high magnetic fields,
hep-th/0106136.
[23] E. Fradkin, C. Nayak, A. Tsvelik, F. Wilczek, Nucl. Phys. B 516 [FS] (1998) 704.
[25] F. Wilczek, A. Zee, Phys. Rev. Lett. 51 (1983) 2250.
[26] Y.S. Wu, A. Zee, Phys. Lett. 174B (1984) 325.
[27] B.I. Halperin, P.A. Lee, N. Read, Phys. Rev. B 47 (1993) 7312.
[28] E. Witten, Commun. Math. Phys. 121 (1989) 351.
[29] Y. Hosotani, Phys. Rev. Lett. 62 (1989) 2785.
[30] The concept of duality used here is essentially electromagnetic duality. See D.-H. Lee, M.P.A. Fisher, Phys.
Rev. Lett. 63 (1989) 903;
S.A. Kivelson, D.-H. Lee, S.-C. Zhang, Phys. Rev. B 46 (1992) 2223;
C.A. Lütken, G.G. Ross, Phys. Rev. B 48 (1993) 2500;
E. Fradkin, S.A. Kivelson, Nucl. Phys. B 474 (1996) 543.
[31] W. Kohn, Phys. Rev. 123 (1961) 1242.
[32] Y.H. Chen, F. Wilczek, E. Witten, B.I. Halperin, Int. J. Mod. Phys. B 3 (1989) 1001.
[34] A.A. Koulakov, M.M. Fogler, B.I. Shklovskii, Phys. Rev. Lett. 76 (1996) 499.
[35] R. Moessner, T.J. Chalker, Phys. Rev. B 54 (1996) 5006.
[36] E. Fradkin, S.A. Kivelson, Phys. Rev. B 59 (1999) 8065.
[37] A. Lopatnikova, S.H. Simon, B.I. Halperin, X.-G. Wen, Phys. Rev. B 64 (2001) 155301, cond-mat/0105079.
[38] D. Barci, E. Fradkin, S.A. Kivelson, V. Oganesyan, to be published in Phys. Rev. B cond-mat/0105448.
[39] M. Fogler, cond-mat/0107306.
[40] L. Radzihovsky, A. Dorsey, Phys. Rev. Lett. 88 (2002) 216802, cond-mat/0110083.
[41] S. Gubser, S.L. Sondhi, Nucl. Phys. B 605 (2001) 395–424.
[42] J. Sinova, V. Meden, S.M. Girvin, Phys. Rev. B 62 (2000) 2008.
[43] J. Moore, A. Zee, J. Sinova, Phys. Rev. Lett. 87 (2001), 046801/1-4 cond-mat/0012341.
Integrability and exact solution for coupled BCS

systems associated with the su(4) Lie algebra
X.-W. Guan a , A. Foerster a,b , J. Links c , H.-Q. Zhou c
a Instituto de Fisica da UFRGS, Av. Bento Goncalves, 9500, Porto Alegre, 91501-970, Brazil
b Institut für Theoretische Physik, der FU-Berlin, Arnimallee 14, Berlin, Germany
c Centre for Mathematical Physics, School of Physical Sciences, The University of Queensland, 4072, Australia
Received 13 May 2002; received in revised form 9 August 2002; accepted 26 August 2002
Abstract
We introduce an integrable model for two coupled BCS systems through a solution of the Yang–
Baxter equation associated with the Lie algebra su(4). By employing the algebraic Bethe ansatz,
we determine the exact solution for the energy spectrum. An asymptotic analysis is conducted to
determine the leading terms in the ground state energy, the gap and some one point correlation
functions at zero temperature.
PACS: 71.24.+q; 74.20.Fg
Keywords: BCS model; Cooper pair; Quantum R-matrix; Ground state
1. Introduction
The role of the Yang–Baxter equation in the study of quantum mechanical models
has a long and distinguished history. Notable examples are the XY Z chain [1], the t–J
model at supersymmetric coupling [2] and the Hubbard model [3], each of which is both
integrable and exactly solvable as a result of the formulation for each model through the
Quantum Inverse Scattering Method (QISM). The key concept of the QISM is the notion
of mutually commuting transfer matrices, the existence of which is a result of the Yang–
Baxter equation. In each of the above examples the Hamiltonian of the model is defined as
the logarithmic derivative of the transfer matrix, and by the nature of the construction this
yields a model defined on a one-dimensional lattice with nearest neighbour interactions,
E-mail address: guan@if.ufrgs.br (X.-W. Guan).

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 7 6 - 9
502 X.-W. Guan et al. / Nuclear Physics B 642 [FS] (2002) 501–514
where it is possible for both integrability and solvability to hold for a variety of boundary
conditions.
The application of the QISM however can be applied on a much more general level.
In the context of the present work, it is appropriate to mention, for example, the work
of Gaudin [4] in relation to the construction of systems with long range interactions.
Very closely related to Gaudin’s Hamiltonians is the BCS model, the exact solution of
which was quite remarkably found in 1963 by Richardson [5], while integrability was
established much later by Cambiaggio et al. in 1997 [6]. The exact solution of the BCS
model has come under close scrutiny in recent years due to its application in the theory
of metallic nanograins [7]. Specifically, the experiments of Ralph, Black and Tinkham
[8] have shown that it is not valid to apply the BCS mean field theory for systems of
nanoscale size. As a result, one has to turn to the exact solution of Richardson in order to
conduct a reliable analysis. However, the approaches adopted in [5,6] make no reference
to the Yang–Baxter equation or the QISM (indeed QISM was not developed until many
years after Richardson’s work), and since historically the two facets of integrability and
solvability have been intimately linked in the QISM framework, it was a natural question
to ask whether the BCS model could be recast through this technique. An affirmative
answer was given in [9,10] with the surprising result that the R-matrix solution of the
Yang–Baxter equation which is needed in the construction of the BCS Hamiltonian is
one of the simplest known examples, being that associated with the su(2) algebra. Given
that a great volume of literature exists devoted to solutions of the Yang–Baxter equation
associated with representations of simple Lie algebras, there is a vast opportunity to
investigate generalized models. An important step towards this has already been achieved
in [11] where a connection has been established between Chern–Simons theory and
integrability of models associated with an arbitrary Lie algebra, which is achieved through
the Knizhnik–Zamolodchikov–Bernard equations.
Such generalized models can be interpreted as coupled BCS systems, at least in the
sense that every simple Lie algebra can be generated by a system of simple roots which
each form an su(2) subalgebra. An example of this was given in [12] where the Lie algebra
employed was so(5). In this instance, the model constructed reproduces the one studied by
Richardson in 1966 [13] describing proton–proton and neutron–neutron pairing as well as
a coupling term for the scattering of proton–neutron pairs. Here we shall introduce a model
based on the su(4) Lie algebra symmetry which can also be interpreted as a nuclear system
where there are now different types of pairing interactions. The Hamiltonian takes the
form of two BCS systems which individually describe pairing interactions for the protons
and neutrons and the scattering of bound proton pairs–neutron pairs, which is in contrast
to the proton–neutron pairs of [12]. This interpretation is possible because the number
operator for each system provides a good quantum number; i.e., the number operators are
conserved. Therefore we can identify each BCS system with a particular distinguishable
particle, which in this case are the protons and neutrons. It is worth remarking that this
situation is inherently different to the pairing models described in [14] based on higher spin
representations of the su(2) algebra. In these cases, the only good quantum number is the
total number of particles in the combined system. There, individual particle numbers are
not conserved and thus the models can be interpreted as describing a Josephson tunneling
phenomena.
X.-W. Guan et al. / Nuclear Physics B 642 [FS] (2002) 501–514 503
The paper is organized as follows. In Section 2 we present the construction of the model
through the QISM. In Section 3 the exact solution of the model is given by means of
the algebraic Bethe ansatz. An analysis of the asymptotic solutions of the Bethe ansatz
equations is presented in Section 4, where the ground state energy, the gap in the spectrum
of elementary excitations, as well as the derivation of some correlation functions in this
asymptotic regime, are presented. A summary of the main results can be found in Section 5.
2. Coupled pairing Hamiltonian and integrability
Let us begin by introducing the following Hamiltonian

Ω
H = BCS(1) + BCS(2) − g bj+ (1)bj+ (2)bk (2)bk (1)
j,k

Ω
2
+g bj+ (1)bk (1) nj (2) − nk (2)
j,k

Ω
2
+g bj+ (2)bk (2) nj (1) − nk (1) , (2.1)
j,k
where

Ω
Ω
BCS(a) = 2j nj (a) − g bj+ (a)bk (a). (2.2)
j =1 j,k
Above the operators bj (a), bj+ (a) are the annihilation and creation operators for the hard-
core bosons (or Cooper pairs) in system a, and j refers to the single particle energy level
with energy j . We will assume that the values j are distinct. Further, g is a coupling
strength constant for the scattering of Cooper pairs and nj (a) = bj+ (a)bj (a), is the Cooper
pair number operator. As in the case of the usual BCS system there is a blocking effect (e.g.,
see [7]), as there is no scattering of any unpaired states. For each level j there are actually
sixteen local states, but the nature of the Hamiltonian means that only on a subspace
spanned by four of these states, where there are no unpaired states, is the scattering non-
trivial (see (2.7)). Hereafter we will restrict our analysis to this subspace.
On this restricted subspace the operators bj+ (a) = cj†↑ (a)cj†↓ (a), bj (a) = cj ↓ (a)cj ↑ (a),
where cj σ , cj†σ , σ = ↑, ↓, are the familiar fermion operators, satisfy the hard-core boson
relations
+ 2
bj (a) = 0, bj (a), bk+ (b) = δab δj k 1 − 2bj+ (a)bj (a) ,

bj (a), bk (b) = bj+ (a), bk+(b) = 0, for k = j.
We can see from the Hamiltonian expression that the exchange interaction of Cooper
pairs in one system depends on the number of Cooper pairs in the other system. For
example, if in system (1), the level j is empty and the level k is occupied by one Cooper
Fig. 1.
Fig. 2.
pair, just for certain configurations of system (2) it is possible that this Cooper pair in (1)
scatters from level k to j . This means that the Hamiltonian (2.1) presents naturally some
“selection rules” for the scattering of states. We illustrate these configurations to indicate
the possible pair scatterings in Fig. 1.
In addition, the double-pair scattering terms of the form (as shown in Fig. 2) are
also present. What the above indicates is that besides the number of Cooper pairs being
conserved in each system, the number of double pairs (to be more precise, the number of
energy levels which are completely filled) is also conserved. This can be seen in each of
the scattering processes depicted graphically above. In each case the scattering does not
overall change the number of completely filled levels. There are further symmetries in the
Hamiltonian. For example, there is a reflection symmetry which interchanges the labels 1
and 2 for the two BCS systems. This arises as a result of a global so(3) ⊕ u(1) symmetry
that the model possesses, which will be made more clear later. In that which follows we
shall first discuss the integrability of the Hamiltonian (2.1) in the context of the QISM.
In order to built up a mechanism to construct an integrable su(4) pairing model, let us
first recall the quantum R-matrix associated with the Lie algebra su(4), which acts in the
tensor product of two 4-dimensional spaces V ⊗ V and can be written as
(λ.I ⊗ I + ηP )
R(λ) = . (2.3)
(λ + η)
Above λ is the usual spectral parameter, P is the permutation operator with matrix elements
Pαβ,γ δ = δαδ δβγ , α, β, γ , δ = 1, 2, 3, 4 and η is the quasiclassical limit parameter; i.e.,
lim R(λ) = I ⊗ I.
η→0
It is known that this R-matrix satisfies the Yang–Baxter equation (YBE)
R12 (λ − µ)R13 (λ)R23 (µ) = R23 (µ)R13 (λ)R12 (λ − µ). (2.4)

The R-matrix may be viewed as the structural constants for the Yang–Baxter algebra
generated by the monodromy matrix T (λ), namely,
1 2 2 1
R12 (λ − µ) T (λ) T (µ) = T (µ) T (λ)R12 (λ − µ). (2.5)
Consequently, the R-matrix (2.3) allows us to construct a realization of the monodromy
matrix through
T (λ) = G0 R0Ω (λ − Ω ) · · · G0 R01 (λ − 1 ). (2.6)

Here the subscript 0 denotes the auxiliary space and G satisfying
[R, G ⊗ G] = 0,
is a class of c-valued solutions of the YBE (2.4). As a consequence of the Yang–Baxter
algebra (2.5), the transfer matrices t (λ) = tr0 T (λ) mutually commute for different values
of the spectral parameter λ. This transfer matrix is the starting point in the construction of a
su(4)-type Gaudin Hamiltonian, from which we can obtain the su(4) pairing Hamiltonian,
as will be shown below. For this purpose we make the following identification for the basis
states
|1 = |0 = ,
+ +
|2 = b (1)b (2)|0 = ,
|3 = b+ (1)|0 = ,
|4 = b+ (2)|0 = , (2.7)
and choose the G-matrix to be given by

 2η 
exp( Ωg ) 0 0 0

2η(1 − n(1) − n(2))  exp( −2η 0 0
G ≡ exp =

0 Ωg ) , (2.8)
Ωg 0 0 1 0
0 0 0 1
to construct the transfer matrix t (λ). It can be verified that

t (j ) = tr0 G0 R0Ω (j − Ω ) · · · G0 P0j · · · G0 R01 (j − 1 )
= Gj Rj,j −1 (j − j −1 ) · · · Gj Rj 1 (j − 1 )
× Gj Rj,Ω (j − Ω ) · · · Gj Rj,j +1 (j − j +1 )Gj {tr0 P0j }
= Gj Rj,j −1 (j − j −1 ) · · · Gj Rj 1 (j − 1 )
× Gj Rj,Ω (j − Ω) · · · Gj Rj,j +1 (j − j +1 )Gj . (2.9)
Above tr0 P0j = 1. Next, taking the quasiclassical limit, we find

Rj,k (λ)|η→0 = I ⊗ I + ηrj,k (λ) + O η2 , (2.10)
2η
Gj |η→0 = I + 1 − nj (1) − nj (2) + O η2 , (2.11)
Ωg
Pj,k −1
where rj,k (λ) = λ . Thus it follows that

Ω
1
t (j )|η→0 = 1 + η τj − + ··· , (2.12)
j − k
k=1
k=j
where
4 αβ βα
2 Ω
α,β Ej Ek
τj = 1 − nj (1) − nj (2) + . (2.13)
g j − k
k=1
k=j
Here E αβ = |αβ|, α, β = 1, . . . , 4 are the Hubbard operators. An immediate conse-

quence from the Yang–Baxter algebra (2.5) is that [τj , τk ] = 0. In addition, as a result
of the so(3) ⊕ u(1) symmetry mentioned earlier, it can be shown that there are extra con-
served operators K and χ such that
[τj , K] = [τj , χ] = [K, χ] = 0.

Above, K is the Casimir operator of an so(3) subalgebra acting on the Ω-fold tensor
product

Ω
+ −
K= Lj Lk + L− +
j Lk + 2 Lj Lk ,
1 0 0
(2.14)
j,k
where (L0 , L+ , L− ) are the basis elements of this canonical so(3) subalgebra
L+ = E 34 = b+ (1)b(2),
L− = E 43 = b+ (2)b(1),
L0 = E 33 − E 44 = n(1) − n(2). (2.15)
The u(1) operator χ explicitly reads

Ω
Ω
2
χ= Ej33 + Ej44 = nj (1) − nj (2) . (2.16)
j =1 j =1
Any Hamiltonian which is defined in terms of the mutually commuting set of operators
{τj , K, χ} (2.17)
will necessarily be integrable where the operators in (2.17) represent the constants of the
motion. By making the following choice

Ω
g3
Ω
3g 2
Ω
g
H = −g j τj + τj τk + τj + K
16 4 2
j =1 j,k=1 j =1
g
Ω
gΩ 2
+ χ(χ − Ω) + 2 j + − 2gΩ, (2.18)
2 4
j
we produce the Hamiltonian (2.1). In order to determine the energy spectrum of this model,
we will need to determine the eigenvalues of the conserved operators, to which we turn
next.
3. Bethe ansatz solutions
Besides proving the integrability of the model, we can also obtain its exact solution
from the algebraic Bethe ansatz for the standard su(4) vertex model constructed from
the R-matrix (2.3). Employing the nested algebraic Bethe ansatz [15] we can obtain the
eigenvalue of the transfer matrix (2.6) as
2η v − v − η
N
i
Λ v, {λj }{ui } = e g
v − vi
i=1
− 2η

Ω
v − i − η N
v − vi + η v − ul −
M η
+e g 2
η
2
η
v − i + 2 i=1 v − vi v − ul + 2
i=1 l=1

Ω
v − i − η M
v − ul + 3η Q
v − wj
+ 2
η
2
v − i + v − ul + η2 v − wj + η
i=1 2 l=1 j =1

Ω
v − i − η Q
v − wj + 2η
+ 2
η . (3.1)
v − i + 2 j =1 v − wj + η
i=1
Above the parameters vj , um and wk satisfy the Bethe ansatz equations
− 4η

Ω
vj − i − η N
vj − vl + η vj − ul −
M η
e g 2
η
2
η = 1,
vj − i + 2 l=1 vj − vl − η vj − ul + 2
i=1 l=1
l=j
− 2η

N
um − vi + η
M
um − ui + η um − wl − 2
Q η
e g 2
η = ,
um − vi − 2 um − ui − η um − wl + η2
i=1 i=1 l=1
i=m

M
wk − ul + η
Q
wk − wl + η
2
η = ,
wk − ul − 2 wk − wl − η
l=1 l=1
l=k
j = 1, . . . , N, m = 1, . . . , M, k = 1, . . . , Q.

Defining N(a) = Ω j =1 nj (a), we can readily determine that the quantum numbers N, M
and Q are given by
N = N(1) + N(2) − N(1)N(2),

M = N(1) + N(2) − 2N(1)N(2),
Q = N(2) − N(1)N(2). (3.2)
The eigenvalues of the integrals of motion τj (2.13) can be obtained from the expansion of
the eigenvalue of the transfer matrix (3.1) in the parameter η. Explicitly, the eigenvalues of
τj are given by
2 1 1
N Ω
Λj = + + , (3.3)
g vl − j j − k
l=1 k=1
k=j
where the parameters satisfy the following equations
4 1 1 1
Ω M N
+ + =2 ,
g vj − i vj − ul vj − vl
i=1 l=1 l=1
l=j
2
N M Q
1 1 1
− −2 = ,
g um − vi ul − um um − wl
i=1 l=1 l=1
l=m

M
1 Q
1
=2 ,
wk − ul wk − wl
l=1 l=1
l=k
j = 1, . . . , N, m = 1, . . . , M, k = 1, . . . , Q. (3.4)
We will also need the eigenvalues of the operators K and χ . Through use of (2.15), (2.16),
(3.2) we find that χ has eigenvalue M while the eigenvalues of K are
1
(M − 2Q)(M − 2Q + 2).
2
Finally, utilizing (3.3) and noting the following identities which can be derived from
(3.4):

Q
M
1 1
Q Q
=2 = 0,
wk − ul wk − wl
k=1 l=1 k=1 l=1

M
N
1 2M
= ,
um − vi g
m=1 i=1

N
Ω
1 2(2N − M)
=− ,
vj − i g
j =1 i=1

Q
M
ul wk Q M
− = MQ,
ul − wk ul − wk
k=1 l=1 k=1 l=1

M
N
vi um M N
− = MN,
vi − um vi − um
m=1 i=1 m=1 i=1

N
Ω
i
N
Ω
vj
− = NΩ,
i − vj i − vj
j =1 i=1 j =1 i=1

Q
M
wk
= Q(Q − 1),
wk − ul
k=1 l=1
2
M M
N
M
Q
um um
− um + + = M(M − 1),
g um − vi um − wl
m=1 m=1 i=1 m=1 l=1
4
N
N
Ω
vj
N
M
vj
vj − + = N(N − 1),
g i − vj vj − ul
j =1 j =1 i=1 j =1 l=1
we can present from the relation (2.18) the eigenvalue of the Hamiltonian (2.1) as

N
M
E=4 vi − 2 um − g(2N − 3M). (3.5)
i=1 m=1
Let us make some small remarks about the degeneracies of the spectrum. Though the
eigenstates of the Hamiltonian have not been made explicit here, it can be deduced by
the standard arguments (e.g., [16]) that each is a highest weight state with respect to the
so(3) symmetry algebra (2.15). In particular, the highest weight which is given by the
eigenvalue of the operator L0 is M − 2Q, so we can conclude that the multiplet generated
by (2.15) acting on this highest weight state has dimension M − 2Q + 1. Therefore for each
solution of (3.4) with given N, M and Q, the corresponding energy level has degeneracy
M − 2Q + 1.
4. Asymptotic solutions
As the Bethe ansatz equations (3.4) take the form of coupled non-linear equations it
is unlikely to find analytic solutions, and one tends to resort to numerical analysis. It
is however possible to conduct an asymptotic analysis for small values of the coupling
parameter g. Below we undertake this for the ground state of the system and some
elementary excitations.
Fig. 3.
For the ground state, we consider first the case g = 0. Letting N1 and N2 denote the
number of Cooper pairs in each system, then it is clear that the ground state corresponds
to filling the Fermi sea, which is illustrated in Fig. 3, where without loss of generality we
assume that N1 > N2 . For small g = 0 we see that the ground state will be described by a
solution of (3.4) with N = N1 , M = N1 − N2 , Q = 0.
Thus the Bethe equations (3.4) reduce to two levels for the parameters vj and um . For a
small g > 0 it is appropriate to consider the asymptotic solution
vj = j + gδj + g 2 σj , j = 1, . . . , N1 , (4.1)
um = N1 −m+1 + gαm + g 2 βm , m = 1, . . . , N1 − N2 . (4.2)
Substituting these into the Bethe equations (3.4) with the configuration N = N1 , M =
N1 − N2 , Q = 0, one can find that

g g2
Ω
1
N2
1
vj ≈ j − + − , j N2 , (4.3)
4 16 j − i j − l
i=N1 +1 l=1
l=j

g g2
Ω
1
N1
1
vj ≈ j − + − , j > N2 , (4.4)
2 4 j − i j − l
i=N1 +1 l=N2 +1
l=j

g2
Ω
1
N2
1
um ≈ N1 −m+1 + +
4 N1 −m+1 − i N1 −m+1 − i
i=N1 +1 i=1
i=N1 −m+1

N1
2
− , m = 1, . . . , N1 − N2 . (4.5)
N1 −m+1 − l
l=N2 +1
l=N1 −m+1
The asymptotic ground state energy is deduced to be given by


N1
N1
E0 ≈ 4 j − 2 l − (N1 + 2N2 )g
j =1 l=N2 +1
N Ω
g2
2 Ω N1 N2
1 2 2
+ + − .
4 j − i j − i j − i
j =1 i=N1 +1 j =N2 +1 i=N1 +1 i=1
It is important to point out that from the above ground state energy we can infer
some results about the asymptotic behaviour of zero temperature correlation functions.
Specifically, by employing the Hellmann–Feynman theorem we have that
1 ∂E0
ni (1) + ni (2) = .
2 ∂i
The result obtained is

ni (1) + ni (2)
N
g2
2 N1
1 2
≈ + , for i > N1 ,
8 (j − i )2 (j − i )2
j =1 j =N2 +1

ni (1) + ni (2)
Ω
g2 2
N2
2
≈1− − , for N1 i > N2 ,
8 (j − i )2 (j − i )2
j =N1 +1 j =1

ni (1) + ni (2)
Ω
g2 1
N1
2
≈2− + , for i N2 .
8 (j − i )2 (j − i )2
j =N1 +1 j =N2 +1
Next let us consider a possible excitation which can be obtained by breaking one Cooper
pair in BCS(2). In the g = 0 case, the excited state is depicted in Fig. 4.
For non-zero g, we choose N = N1 − 2, M = N1 − N2 − 1 and block the levels with
energy N2 , N2 +1 . From the asymptotic solutions (4.3)–(4.5), we obtain the excitation
energy

N1
N1
E1 ≈ 4 j − 2 l − N2 − N2 +1 − (N1 + 2N2 − 4)g
j =1 l=N2 +2
N −1 Ω

N1
Ω 2 −1
N
g2 2
1 2 2
+ + − .
4 j − i j − i j − i
j =1 i=N1 +1 j =N2 +2 i=N1 +1 i=1
Therefore the gap obtained through the breaking a Cooper pair in BCS(2) is given by
Fig. 4.
Fig. 5.

g2
Ω
1
Ω
2
∆1 ≈ N2 +1 − N2 + 4g + +
4 i − N2 i − N2 +1
i=N1 +1 i=N1 +1

N1
2
N2
2
+ + .
j − N2 N2 +1 − i
j =N2 +2 i=1
Another possibility for an excitation at g = 0 is to break the Cooper pair at level
N1 in BCS(1) (see Fig. 5). For g = 0, the configuration should be accommodated as
N = N1 − 1, M = N1 − N2 − 1 with the levels N1 , N1 +1 blocked.
We find the gap obtained in this case to be given by
N
g2
2 N2
1 2
∆2 ≈ N1 +1 − N1 + g + +
4 N1 +1 − j N1 − i
j =1 i=1

1 −1
N Ω
2 2
+ + .
N1 +1 − j i − N1
j =N2 +1 i=N1 +1
For the above asymptotic analysis to be valid, we require that the coefficients of the
terms in g 2 must be much smaller than g −1 , imposing a stringent constraint on g. In each
case this will depend explicitly on Ω, N1 , N2 and the distribution of the single particle
energy levels i . For the case of the usual BCS model such a constraint was studied in [18].
5. Conclusion
To summarize, we have constructed an integrable pairing Hamiltonian based on the

su(4) Lie algebra. This model can be interpreted as describing two coupled BCS systems
of different types, such as for protons and neutrons in a nuclear system. The Bethe
ansatz equations and the energies of the model have been calculated. For small values of
the coupling parameter g, we asymptotically analyzed the ground state and elementary
excitations, and the expectation values for the occupation numbers. An open problem
that we will address in the future is the exact calculation of form factors and correlation
functions using the techniques developed by Babujian et al. in [17].
Acknowledgements
X.-W.G. gratefully acknowledges the Centre for Mathematical Physics, the University
of the Queensland for kind hospitality and FAPERGS (Fundação de Amparo ã Pesquisa
do Estado do Rio Grande do Sul) for financial support. A.F. would like to acknowledge
M. Karowski and H. Babujian for discussions and the group of Theoretical Physics of
the FU-Berlin for their kind hospitality. She also thanks DAAD (Deutsche Akademischer
Austauschdienst) and FAPERGS for financial support. H.-Q.Z. and J.L. thank the
Australian Research Council.
References
[1] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, San Diego, 1982.
[2] F.H.L. Essler, V.E. Korepin, Phys. Rev. B 46 (1992) 9147;
A. Foerster, M. Karowski, Phys. Rev. B 46 (1992) 9234;
A. Foerster, M. Karowski, Nucl. Phys. B 396 (1993) 611.
[3] B.S. Shastry, Phys. Rev. Lett. 56 (1986) 1529;
B.S. Shastry, Phys. Rev. Lett. 56 (1986) 2453;
M.J. Martins, P.B. Ramos, Nucl. Phys. B 522 (1998) 413.
[4] M. Gaudin, J. Physique 37 (1976) 1087;
H. Babujian, J. Phys. A 26 (1993) 6981.
[5] R.W. Richardson, Phys. Lett. 3 (1963) 277;
R.W. Richardson, Phys. Lett. 5 (1963) 82;
R.W. Richardson, N. Sherman, Nucl. Phys. 52 (1964) 221;
R.W. Richardson, N. Sherman, Nucl. Phys. 52 (1964) 253.
[6] M.C. Cambiaggio, A.M.F. Rivas, M. Saraceno, Nucl. Phys. A 624 (1997) 157.
[7] J. von Delft, D.C. Ralph, Phys. Rep. 345 (2001) 61;
J. von Delft, Ann. Phys. (Leipzig) 10 (2001) 219.
[8] D.C. Ralph, C.T. Black, M. Tinkham, Phys. Rev. Lett. 76 (1996) 688;
D.C. Ralph, C.T. Black, M. Tinkham, Phys. Rev. Lett. 78 (1997) 4087.
[9] H.-Q. Zhou, J. Links, R.H. McKenzie, M.D. Gould, Phys. Rev. B 65 (2002) 060502(R).
[10] J. von Delft, R. Poghossian, cond-mat/0106405.
[11] M. Asorey, F. Falceto, G. Sierra, Nucl. Phys. B 622 (2002) 593.
[12] J. Links, H.-Q. Zhou, M.D. Gould, R.H. McKenzie, J. Phys. A 35 (2002) 6459.
[13] R.W. Richardson, Phys. Rev. 144 (1966) 874;
R.W. Richardson, Phys. Rev. 154 (1967) 1007.
[14] J. Links, H.-Q. Zhou, R.H. McKenzie, M.D. Gould, cond-mat/0110105, Int. J. Mod. Phys. B, in press;
J. Links, K.E. Hibberd, Int. J. Mod. Phys. B 16 (2002) 2009.
[15] B. Sutherland, Phys. Rev. B 12 (1975) 3795;
O. Babelon, H.J. de Vega, C.M. Viallet, Nucl. Phys. B 200 (1982) 266.
[16] A.N. Kirillov, J. Sov. Math. 36 (1987) 115.
[17] H. Babujian, A. Fring, M. Karowski, Form factors of the SU(N ) model, in preparation.
[18] M. Schechter, Y. Imry, Y. Levinson, J. von Delft, Phys. Rev. B 63 (2001) 214518.
Elusive physical electron propagator in QED-like

effective theories
D.V. Khveshchenko
Department of Physics and Astronomy, University of North Carolina, Chapel Hill, NC 27599, USA
Abstract
We study the previously conjectured form of the physical electron propagator and its allegedly
Luttinger type of behavior in the theory of the pseudogap phase of high-temperature copper-oxide
superconductors and other effective QED-like models. We demonstrate that, among a whole family
of seemingly gauge-invariant functions, the conjectured “stringy ansatz” for the electron propagator
is the only one that is truly invariant. However, contrary to the results of the earlier works, it appears
to have a negative anomalous dimension, which makes it a rather poor candidate to the role of the
physical electron propagator. Instead, we argue that the latter may, in fact, feature a “super-Luttinger”
behavior characterized by a faster than any power-law decay Gphys (x) ∝ exp(−const · ln2 |x|).
1. Introduction
In a generic (1 + 1)-dimensional (hereafter 2D) many-fermion system, an arbitrarily

weak short-range repulsive interaction is known to completely destroy the conventional
Fermi liquid, thereby giving rise to the so-called Luttinger behavior. As one of the
hallmarks of the Luttinger regime, the electron propagator exhibits an algebraic decay with
distance which is controlled by a non-universal exponent and appears to be faster than in
the non-interacting case.
In higher dimensions (3D and 4D) the Fermi liquid is believed to be more robust,
although it is not expected to remain absolutely stable. In the absence of a clear-cut
evidence, however, the possibility of a non-Fermi liquid behavior in D > 2 has recently
become the subject of an intense debate.
E-mail address: khvesh@physics.unc.edu (D.V. Khveshchenko).
PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 7 9 3 - 9
516 D.V. Khveshchenko / Nuclear Physics B 642 [FS] (2002) 515–529
While in the case of short-ranged repulsive interactions any departures from the Fermi
liquid are likely to be limited to the infinitely strong coupling regime, the long-ranged
forces are considered to be capable of destroying the Fermi liquid even at finite coupling.
One of the most extensively studied such examples, the finite density system of non-
relativistic massive fermions which are minimally coupled to an Abelian gauge field, was
indeed shown to manifest a distinctly non-Fermi liquid behavior, although the latter turned
out to be rather different from the Luttinger one [1].
More recently, there has been a renewed interest in the 3D relativistic counterpart of this
model governed by the conventional QED3 action
N
1 2
S ψ, ψ̄ , A = dz ψ̄f i γ̂ ∂µ + γ̂ Aµ − m ψf + 2 ∂µ Aν − ∂ν Aµ
µ µ
,
4g
f =1
(1)
for the N -flavored Dirac fermions which are described by a reducible (four-component)
representation corresponding to the choice of the γ -matrices γ̂µ = σ̂µ ⊗ σ̂3 constructed
from the triplet σµ of Pauli matrices. The use of the parity-even representation guarantees
that radiation corrections do not generate any parity-breaking Chern–Simons terms.
Among the previously discussed examples of the 3D condensed matter systems that
support the Dirac-like low-energy excitations and allow for such an effective description
are the so-called flux phase in doped Mott insulators [2,3] and d-wave superconductors
with strong phase fluctuations [4–7] proposed as an explanation of the pseudogap and
insulating (spin and/or charge density wave) phases of the high-Tc cuprates. Also, the non-
Lorentz-invariant version of QED3 was shown to provide a convenient description of the
normal semimetalic state of highly oriented pyrolytic graphite [8–10].
The number N of the fermion flavors depends on the problem in question, although it
is not necessarily equal to the number of different conical Dirac points in the bare electron
spectrum. In the above mentioned condensed matter-inspired QED-like models [2–10],
N = 2 is a number of the electron spin components, while the number of the conical points
turns out to be either four [4–7] or two [8–10].
In some of the strongly correlated electron systems described by the effective QED-
like theories, such as the problem of graphite, the theory is formulated in terms of
the electrons themselves which enables one to directly compute various experimentally
relevant observables [9,10].
However, in other systems the effective gauge fields serve as a convenient representation
of such bosonic collective modes as spin or pairing fluctuations, while the Dirac fermions
correspond to some auxiliary fermionic excitations, such as spinons [2–4], “topological”
fermions [5–7], and so forth.
In the latter case, a generic quantum mechanical amplitude written in terms of the
Lagrangian fermions turns out to be gauge-dependent, while all the physical observables,
which experimental probes can only couple to, must be manifestly gauge-invariant. In
light of that, it is absolutely imperative to establish the correct form of the physical
electron propagator in terms of the Lagrangian fermions. Until this task is accomplished,
no theoretical prediction obtained in the framework of the QED-like effective models can
be put under a decisive test against experimental data which will be the ultimate means of
ascertaining the validity of these scenarios.
D.V. Khveshchenko / Nuclear Physics B 642 [FS] (2002) 515–529 517
2. Exact versus limited gauge invariance of fermion amplitudes
In the recent theory of the pseudogap phase of the high-Tc superconductors [4–6], the
authors advocated the use of the “stringy ansatz” with the inserted Wilson-like exponent of
the line integral

GΓ (x − y) =
0|ψ(x) exp +i Aµ (z) dz ψ̄(y)|0,
µ
(2)
Γ
as a viable candidate for the gauge-invariant physical electron propagator (in spite of its
being gauge-independent, the function GΓ explicitly depends on the choice of the contour
Γ which connects the end points x and y).
Moreover, the authors of Refs. [4–6] argued that the amplitude (2) exhibits the so-called
Luttinger-type behavior
x̂
GΓ (x) ∝ , (3)
|x|d+η
characterized by a positive (albeit varying from one reference to another) value of the
anomalous exponent η, provided that the contour Γ is chosen as the straight-line segment
between the end points.
Although the idea of using Eq. (2) for elucidating the behavior of the physical electron
propagator has been previously entertained, one must be cautioned by the fact that, apart
from its relative simplicity, it never received any firm justification. A potential problem
with both the simple form of the line integral and the particular choice of the contour
Γ in Eq. (2) stem from the fact that in the theory of Refs. [4–6] the phase φ(x) of the
(singular) gauge transformation Ψ (x) = eiφ(x)ψ(x) which relates the physical electron
operator Ψ (x) to the Lagrangian fermion field ψ(x) is not uniquely defined.
In what follows, we address some of the misconceptions pertaining to the use of Eq. (2)
in the earlier works and ascertain the status of the previously obtained results. To this end,
we start out by casting GΓ (x − y) in the form of a functional integral taken over both the
gauge and fermion Lagrangian variables
GΓ (x − y)

= D[A] D[ψ̄] D[ψ] ψ(x)ψ̄(y) exp i Aµ (z) dz exp iS[ψ, ψ̄, A] .
µ
(4)
Γ
Upon integrating the fermions out, the amplitude (4) turns into a functional average over
different gauge field configurations

GΓ (x − y) = D[A] G(x, y|A) exp i Aµ (z) dzµ exp iSeff [A] , (5)
Γ
applied to the inverse Dirac operator G(x, y|A) =

x|[i ∂ˆ + Â(z) − m]−1 |y computed for
a given gauge field configuration Aµ (z).
The weight of the average in Eq. (5) is controlled by the effective gauge field action

1
Seff [A] = 2 dx (∂µ Aν − ∂ν Aµ )2
4g

1
+ dx dy Aµ (x)Πµν (x − y)Aν (y) + · · · , (6)
2
where the dots stand for the higher-order (non-Gaussian) terms produced by fermion
polarization (processes of “light–light scattering” and alike) which we hereafter neglect,
as in all of the previous works on the subject.
With the transverse fermion polarization Πµν (q) = Π(q)(δµν − qµ qν /q 2 ) taken in to
account, the gauge field propagator computed in the covariant λ-gauge assumes the form

g2 qµ qν
Dµν (q) = 2 δµν + (λ − 1) 2 . (7)
q + g 2 Π(q) q
In order to obtain Eq. (7) one
has to complement (6) with a (generally, non-local) gauge
fixing term %Seff [A] = dx dy ∂µ Aµ (x)
x|[1 − g 2 Π(i∂)/∂ 2 ]|y∂ν Aν (y)/2g 2 λ.
In the case of the conventional (that is weakly coupled, g 2 1) QED4 , the fermion
polarization behaves as Π(q) ∝ q 2 , and the above gauge fixing term becomes local in the
coordinate space. As a result, the ultraviolet (UV) divergencies occur at fermion momenta
of order the upper momentum cut-off.
In contrast, QED3 develops its highly non-trivial behavior at intermediate fermion
momenta m p Λ = Ng 2 controlled by the (this time, dimensionful) coupling
constant g [2]. However, adhering to the customary terminology, we will refer to this
regime as the UV one, for this is where all the logarithmic corrections originate from,
whereas above the cut-off Λ no terms ∝ ln(Λ/p) can possibly occur.
In this regime, the gauge propagator (7) is totally dominated by the fermion polarization
which, due to the parity conserving structure of the reducible four-fermion representation
produces no Chern–Simons terms and, in the leading 1/N approximation, reads as

N
Π(q) = q 2. (8)
8
In this way, the number of fermion flavors N becomes the actual parameter controlling
perturbation expansion, regardless of the strength of the bare coupling g.
Without making any additional approximations, the function G(x, y|A) can be ex-
pressed in the form of a quantum mechanical path integral [11] (see also [12] where this
representation was used in a non-perturbative calculation of the Dirac fermion propagator
in a static spatially random gauge field which is pertinent to such problems as the effect
of vortex disorder on the quasiparticle properties of d-wave superconductors or that of
dislocations in layered graphite)
∞ )=x
x(τ τ
i
S0 [x,p] dzµ
G(x, y|A) = dτ Dx Dp e exp i dτ Aµ (z) , (9)
dτ
0 x(0)=y 0
where xµ (τ ) is a fermion trajectory parametrized by the proper time τ . The spinor structure
of the fermion propagator is fully accounted for by integrating over the additional variable
pµ (τ ) with the free (matrix-valued) fermion action

τ
dx µ

S0 x(τ ), p(τ ) =
dτ pµ − γ̂ − m ,
µ
(10)
dτ
0
thus providing a systematic improvement of the celebrated Bloch–Nordsieck model. In the
latter, all the spin-related effects are ignored which makes this model exactly soluble but
restricts its applicability to the infrared (IR) regime |p2 − m2 | m2 near the fermion mass
shell. In the IR regime, the relevant fermion trajectories contributing to (9) deviate only
slightly from the straight-path contour Γ (which coincides with the fermion world line in
the case of a light cone-like separation between the end points, (x − y)2 = 0).
One must recognize that the IR regime can only exist if the fermions are massive, while
in the case of m = 0 the entire region below the upper cut-off Λ falls into the opposite,
UV, regime. In the UV regime, the fermion trajectories which dominate the amplitude (9)
may strongly depart from the straight-path contour Γ , for there is no mass term to suppress
such deviations.
Combining Eqs. (5) and (9) we represent the amplitude (2) in the form manifesting its
gauge invariance
∞ )=x
x(τ

GΓ (x − y) = dτ Dx Dp ei S0 [x,p] exp i ∂µ Aν dΣ µν , (11)
0 x(0)=y
where the brackets stand for the (normalized) functional average over the gauge field with
the weight exp(iSeff [A]) which is to be taken for each trajectory xµ (τ ) before the sum over
all the different trajectories is executed.
In order to obtain the area integral in the exponent we applied Stokes’ theorem to the
line integral taken along a closed contour which is composed of a trajectory xµ (τ ) and the
straight-line segment Γ , parametrized as xµ(0) (τ ) = τ (x − y)µ /τ and traced backwards.
The averaging procedure generates a sum of all the multi-loop diagrams with no
couplings between the fermion polarization insertions into the gauge field propagators and
the open fermion line which describes the fermion propagating from y to x.
Elucidating the structure of the gauge invariant amplitude (2) helps one to appreciate
the difference between (11) and all the other (ξ = 0) members of the family of functions
labeled by a continuous parameter ξ (not to be confused with the covariant gauge
parameter λ!)

0|ψ(x) exp(i(ξ − 1) Γ Aµ (z) dzµ )ψ̄(y)|0
Gξ (x − y) =

0| exp(iξ Γ Aµ (z) dzµ )|0
∞ )=x
x(τ
iS0 [x,p]
exp(iξ Γ Aµ dz + i ∂µ Aν dΣ )
µ µν
= dτ Dx Dp e .

exp(iξ Γ Aµ dzµ )
0 x(0)=y
(12)
Amongst others, Eq. (12) also includes the function

0|ψ(x)ψ̄(y)|0
G1 (x − y) = , (13)

0| exp(i Γ Aµ (z) dzµ )|0
which was invoked by the authors of Ref. [5] who asserted that G1 (x) given by Eq. (13) is
identical to the amplitude (3) and, therefore, can be used to compute GΓ (x) = G0 (x).
The UV anomalous dimension of Eq. (13) can be deduced rather straightforwardly by
computing the ratio between the wave function renormalization factor determining the
anomalous dimension of the ordinary (gauge variant) fermion propagator [14]
2 N)(2−3λ)

0|ψ(x)ψ̄(y)|0 ∝ G(x − y)(Λ|x − y|)(4/3π (14)
(hereafter G(x) stands for the free fermion propagator which also yields the bare value of
Gξ (x) for any ξ ) and the Gaussian average of the Wilson line

µ
exp i Aµ dz
Γ

1 µ 2
= exp − dz1 dz2 Dµν (z1 − z2 ) ∝ (Λ|x − y|)(4/π N)(2−λ) ,
ν
(15)
2
Γ Γ
thus indeed resulting in the anticipated Luttinger-type behavior of Eq. (13) which is
characterized by the overall positive anomalous dimension
16
η13d = . (16)
3π 2 N
Notably, this result is free of the gauge parameter λ, thus creating the impression that
Eq. (13) represents a truly gauge invariant function.
As one immediate objection to the assertion of Ref. [5], one might recall that the very
nature of the relationship between the electrons and the auxiliary Lagrangian fermions im-
plemented via the singular gauge transformation
0|Ψ (x)Ψ (y)|0 =
0|ψ(x) exp[i(φ(x) −
φ(y)]ψ̄(y)|0 implies that the physical electron propagator can only be given by a single
average over the gauge field, rather than a ratio of such.
However, except for the case of the original amplitude G0 (x), Eq. (12) is represented
by a ratio of the two averages and, therefore, it cannot be truly gauge invariant. Indeed, the
gauge fields, over which one averages in both the numerator and denominator of Eq. (12),
may transform totally independently of one another (Aµ 1,2 → A1,2 + ∂ f 1,2 ), thus resulting
µ µ
in the overall phase factor exp[iξ(f (x) − f (x) − f (y) + f 2 (y))] which only vanishes
1 2 1
for f 1 (x) = f 2 (x), as long as ξ = 0.

Nevertheless, as shown below, Gξ (x) turns out to be independent of the gauge parameter
λ within the class of the covariant gauges for any ξ . Presumably, it was this limited gauge
independence that led the authors of Ref. [5] to their conclusion that the (seemingly gauge-
invariant) amplitude G1 (x) must be identical to the (truly invariant) amplitude G0 (x), for
they apparently coincide
with each other in the axial gauge (x − y)µ Aµ (z) = 0, in which
the line integral Γ Aµ (z) dzµ vanishes.
However, should this happen to be valid, the same argument would also apply to an
arbitrary Gξ (x) which is seemingly gauge invariant to absolutely the same extent as
Eq. (13) is (that is, provided that one uses the identical gauge transformation in both
numerator and denominator), and all these functions coincide with each other when
computed in the axial gauge.
Nonetheless, one can readily see that Gξ (x) does bear a non-trivial ξ -dependence
by simply evaluating the integrand in Eq. (12) for certain chosen trajectories. As an
example, we consider the case of a square-shaped closed contour consisting of four straight
segments, each of length |x − y|. Having all the adjoint sides of the square at right angles
to each other guarantees one from a possible occurrence of the additional (“cusp”) UV
singularities [11].
Applying the method of dimensional regularization near d = 3 we obtain

exp(i ∂µ Aν dΣµν + iξ Γ Aµ dzµ )

exp(iξ Γ Aµ dzµ

4(Λ|x − y|)3−d d −1
∝ exp − , (2 − ξ )
π (d+1)/2N 2

× I1 1 − (λ − 1)(d − 2) − 2I2 (λ − 1)(d − 1)
2N
∝ (Λ|x − y|)8(2−ξ )/π , (17)
where the terms in the exponent proportional to
1 1
dα dβ 2
I1 = = (18)
|α − β|d−1 (3 − d)(2 − d)
0 0
and
1 1
αβ dα dβ 2 (1−d)/2
I2 = = 2 −1 , (19)
(α 2 + β 2 )(1+d)/2 (3 − d)(1 − d)
0 0
correspond to the line integrals taken along the same and two adjoint sides of the square,
respectively (pairs of the opposite sides produce no terms ∝ ln(Λ|x − y|) which is what we
are after). Despite the fact that the gauge parameter λ cancels out, as expected, the values
of the exponent given by Eq. (17) are markedly different for, say, ξ = 0 and ξ = 1.
In fact, Eq. (17) can either vanish or grow with increasing |x − y|, depending merely on
whether ξ is greater or smaller than 2. This observation suggests that for different values of
ξ the functions Gξ (x) are indeed different (it would be rather fortuitous if the sum over all
the fermion trajectories were ξ -independent, despite the difference between contributions
from each individual trajectory).
It is worth mentioning that in the case of massive fermions the different functions (12)
do become approximately identical in the vicinity of the mass shell (|p2 − m2 | m2 )
where the leading functional dependence
on the gauge field originates from the eikonal
phase factor G(x, y|A) ∝ exp(i Γ Aµ dzµ ). Therefore, the entire IR divergence of the
ordinary (gauge variant) propagator can be obtained by simply averaging this exponential
factor, while any Gξ (x) appears to be totally IR divergence-free.
This factorization of the exponent of the line integral in the IR regime which had
long been known in the case of QED4 [13] was extended into the 3D case in Ref. [9],
where it was shown that for any ξ the function (12) possesses the same simple pole in
the momentum representation Gξ (p) ∝ (p̂ − m)/(p2 − m2 ), which corresponds to the

universal (ξ -independent) behavior in the coordinate space Gξ (x) ∝ e−m|x| .
In order to avoid confusion we reiterate that the universal IR behavior by no means
prevents the different functions (12) from having different UV anomalous dimensions, as
manifested by Eq. (17). Incidentally, in the massless case it turns out to be the UV exponent
ηξ which controls the power-law decay (3) extending to arbitrarily long distances.
3. Perturbative calculation of fermion amplitudes
Having demonstrated the possibility of a non-trivial ξ -dependence non-perturbatively,

we now present a direct perturbative calculation of Gξ (x). Expanding (12) to second order
in Aµ (z) we find three different kinds of correction terms
Gξ (x − y) = G(x − y)

− dz1 dz2 G(x − z1 )Â(z1 )G(z1 − z2 )Â(z2 )G(z2 − y)

1 µ µ
+ ξ− G(x − y) Aµ (z1 ) dz1 Aν (z2 ) dz2
2
Γ Γ

µ
+ (1 − ξ ) dz1 G(x − z1 )Â(z1 )G(z1 − y) Aµ (z2 ) dz2 . (20)
Γ
The first term in the r.h.s. of (20) corresponds to the first-order self-energy correction to
the ordinary propagator which is given by the leading term of 1/N expansion of Eq. (14)

δ1 Gξ (x − y) = − dz1 dz2 G(x − z1 )γ µ Dµν (z1 − z2 )γ ν G(z2 − y)
4
= G(x − y) (2 − 3λ) ln(Λ|x − y|). (21)
3π 2 N
The second term in (20) originates from 1/N expansion of the Wilson line (15) inserted
into the numerator and denominator of Eq. (12) with the prefactors (ξ − 1) and ξ ,
respectively,

1 µ
δ2 Gξ (x − y) = ξ − G(x − y) dz1 dz2ν Dµν (z1 − z2 )
2
Γ Γ
4
= G(x − y) 2 (1 − 2ξ )(2 − λ) ln(Λ|x − y|). (22)
π N
The third term in (20) stems from expanding both G(x, y|A) and the Wilson line in the
numerator of Eq. (12) to first order in Aµ (z)

δ3 Gξ (x − y) = (1 − ξ ) dz1 G(x − z1 )γ µ G(z1 − y) dz2ν Dµν (z1 − z2 )
Γ

≈ (1 − ξ ) dz1 G(x − y)γ G(z1 − y)
µ

+ G(x − z1 )γ µ G(x − y) dz2ν Dµν (z1 − z2 )
Γ
2
= 3 (1 − ξ )G(x − y)γ µ
π N

(ẑ1 − ŷ) 1
× dz2ν dz1
|z1 − y| |z1 − z2 |2
3
Γ
µ µ
(z − z2 )(z1ν − z2ν )
× λδµν + 2(1 − λ) 1
|z1 − z2 |2
8
= G(x − y) (1 − ξ )λ ln(Λ|x − y|). (23)
π 2N
This result was obtained with logarithmic accuracy, and in the course of the calculation we
used the coordinate space representation of the 3D propagators

1 x̂ 4 xµxν
G(x) = , D µν (x) = λδ µν + 2(1 − λ) (24)
4π x 3 π 2 N|x|2 |x|2
and the following d-dimensional integrals

xα 2π d/2 y α
dx d = (25)
|x| |x − y|d−1 ,(d/2) |y|d−1
and

x α (x β − y β )(x γ − y γ )
dx
|x|d |x − y|d+1
α βγ
2π d/2 2y δ − y β δ αγ − y γ δ αβ yαyβ yγ
= + (d − 1) . (26)
3(d − 1),(d/2) |y|d−1 |y|d+1
dependence in Eqs. (21)–(23) results from taking the d → 3 limit of the

The logarithmic
line integral Γ dzµ (z − y)µ /|z − y|d−1 = (Λ|x − y|)3−d /(3 − d) → ln(Λ|x − y|).
Combining Eqs. (21)–(23) together we read off the overall UV anomalous dimension
16
ηξ3d = (3ξ − 2). (27)
3π 2 N
In particular, for ξ = 1 we reproduce Eq. (16), while for ξ = 0 one obtains
32
η03d = − . (28)
3π 2 N
Thus, despite the fact that the gauge parameter λ cancels out, as expected, the functions
Gξ (x) are indeed different for different values of the parameter ξ .
At first sight, the non-trivial dependence of Eq. (27) on the parameter ξ (which is linear
to first order in 1/N ) may seem to be in conflict with the fact that all the functions Gξ (x)
coincide with each other when computed in the axial gauge.
In fact, there would only be a contradiction, if Eq. (12) were truly gauge invariant.
Instead, albeit invariant within the class of covariant gauges, the function Gξ (x) may take
a different value in the axial gauge, for the latter does not belong to this class, for any
ξ = 0.
On the other hand, the unique property of the function G0 (x − y) which is given by a
single average instead of a ratio of the two does guarantee its true gauge invariance and
the possibility to obtain Eq. (28) in arbitrary gauge, including both the covariant and the
axial ones. In fact, it was shown in Ref. [15] that these two calculations do agree with each
other, and yet another independent calculation employing the so-called “radial” (Fock–
Schwinger) gauge (x − y)µ Aµ (x) = 0 [16] provides further support for this conclusion.
However, the finding that the UV exponent η03d assumes the negative value suffices to
disqualify the “stringy ansatz” (2) from being a sound candidate for the physical electron
propagator, since, in all of the QED-like models discussed so far, repulsive electron–
electron interactions are expected to result in a suppression, rather than enhancement, of
any amplitude describing propagation of physical electrons.
The above analysis can be easily extended to the case of the conventional weak coupling
QED4 which demonstrate that the situation in 3D is not at all exceptional. Instead of
Eqs. (21)–(23) we now get
g2
δ1 Gξ (x − y) = −G(x − y) λ ln(Λ|x − y|), (29)
8π 2
g2
δ2 Gξ (x − y) = G(x − y) 2 (1 − 2ξ )(3 − λ) ln(Λ|x − y|), (30)
8π
g2
δ3 Gξ (x − y) = G(x − y) 2 (1 − ξ )λ ln(Λ|x − y|) (31)
4π
and, instead of Eqs. (24)–(26), we use the formulas

1 x̂ g2 1+λ x µx ν
G(x) = , Dµν (x) = δµν + (1 − λ) , (32)
2π 2 x 4 4π 2 |x|2 2 |x|2

xα π d/2 yα
dx d = , (33)
|x| |x − y| d−2 ,(d/2) |y|d−2

x α (x β − y β )(x γ − y γ )
dx
|x|d |x − y|d
α βγ
π d/2 y δ − y β δ αγ − y γ δ αβ yαyβ yγ
= + (d − 2) . (34)
2(d − 2),(d/2) |y|d−2 |y|d
Combining Eqs. (29)–(31) together we obtain the UV anomalous dimension of Gξ (x) in
the 4D case
3g 2
ηξ4d = (2ξ − 1). (35)
8π 2
Interestingly enough, for ξ = 0 and ξ = 1 the values of Eq. (35) differ only in their sign,
and the anomalous dimension η04d = −3g 2 /8π 2 is again negative.
4. Alternate forms of physical electron propagator
The apparent problem with the unphysical decay of the conjectured form of the physical
electron propagator (2) which appears to be slower than in the case of non-interacting
electrons compels one to explore alternate proposals. To this end, one may be able
to benefit from the previous studies of the conventional QED4 where the problem of
constructing gauge invariant asymptotical quantum states with quantum numbers of an
electron (“dressed charges”) was long recognized, and attempts to explicitly solve it have
long been under way.
A (non-local) composite operator creating such a state can be sought in the form

Ψ (x) = exp i dy f µ (x − y)Aµ (y) ψ(x), (36)
where the vector function f µ (x) is subject to the condition ∂µ f µ (x) = δ(x) imposed by
the requirement of gauge invariance.
In the original proposal which was put forward by Dirac in his pioneering work of the
fifties, the physical electron was identified with the operator

∇i Ai
ΨD (x) = exp i 2 ψ(x). (37)
∇
Indeed, the expectation value
0|ΨD (x)Aµ (y)ΨD† (x)|0 = δµ0 /4π| x − y| describes
the Coulombic electrostatic field, thus suggesting that Eq. (37) provides a plausible
representation of a static charge.
Moreover, it was found that the Fourier transform of the gauge-invariant amplitude
0|ΨD (x)Ψ on the mass shell [17].

D (y)|0 is IR finite at the single point pµ = (m, 0)
It was also argued in Ref. [17] that, once created, the coherent state ΨD† (x)|0 remains
t
stable, unlike that produced by the operator ΨT (x) = exp(i −∞ A0 (t ) dt )ψ(x) whose
propagator
0|ΨT (x)Ψ T (y)|0 is given by Eq. (2) where (x − y)µ ∝ δµ0 .
Furthermore, the amplitude
0|ΨD (x)Ψ D (y)|0 was found to undergo multiplicative UV
renormalization, featuring a positive UV exponent
4d
ηD = g 2 /8π 2 . (38)
The authors of Ref. [17] were also able to generalize the construction (37) onto the case of
a moving charge

δµν − (u + v)µ (u − v)ν

Ψv (x) = exp i 2 ∂ A
µ ν ψ(x), (39)
∂µ − (uµ ∂µ )2 + (vν ∂ν )2
is the time-like unit vector and vµ = (0, v) is the velocity of the charge.
where uµ = (1, 0)
The formula (39) was obtained by requiring that, in accord with the anticipated physical
interpretation of Eq. (39), the electromagnetic field
0|Ψv (x)Aµ (y)Ψv† (x)|0 associated
with the coherent state created by this composite operator reproduces the classical Lienard–
Wiechert potentials.
It was also found that the corresponding propagator
µν
δ − (uµ + v µ )(uν − v ν )
Gv (x − y) =
0|ψ(x) exp i 2 ∂µ A ν ψ̄(y)|0, (40)
∂α − (uα ∂α )2 + (vα ∂α )2
√
is both IR-finite at pµ = m(1, v)/ 1 − v 2 and multiplicatively UV-renormalizable, its
anomalous dimension being explicitly v-dependent [17]

g2 1 1−v
ηv = − 2 3 + 2 ln
4d
. (41)
8π v 1+v
The combination of all the before mentioned properties makes Eq. (39) the best (up-to-
date) available candidate for the physical electron operator in the conventional (massive)
QED4 .
It is worth mentioning that Eq. (41) coincides with the UV dimension of the 4D
“stringy” amplitude (2) only in the unphysical limit v → ∞, while its v = 0 value yields
Eq. (38).
By construction, Eq. (41) cannot be applied directly to the massless, let alone the
3D massless, case which corresponds to the limit v → 1 where the UV dimension (41)
increases towards infinitely high positive values.
Albeit not being immediately applicable to the case of m = 0, the calculation of Gv (x)
carried out in Ref. [17] suggests that the logarithmic growth of (41) gets effectively cut-off
at max(|1 − v|, 1/Λ|x|). Therefore, it is not inconceivable that the massless counterpart of
Eq. (40) may exhibit a faster than a power-law decay

Gphys(x) ∝ exp −const · ln2 (Λ|x|) , (42)
where the constant is proportional to either 1/N (3D) or g 2 (4D), thus placing the
effective QED-like theories into the class of “super-Luttinger” models, alongside the
one-dimensional metals with unscreened Coulombic interactions where Gphys(x) ∝
exp(−const · ln3/2 (Λ|x|)) [18].
In light of this possibility, the previous claims of discovering the Luttinger-like behavior
in the framework of the QED3 -theory of the pseudogap phase [4–6] may need to be
prepared to handle a potentially much stronger suppression of the physical amplitudes in
order to reconcile such a behavior with the available photoemission, tunneling, and other
experimental data.
5. Discussion
Evidently, one could escape the whole problem with the singular massless limit, if the
Dirac fermions acquired a finite mass via the mechanism of chiral symmetry breaking [19].
In the context of the QED3 theory of the pseudogap phase, such an intrinsic instability
is expected to result in the onset of a spin (SDW) and/or charge (CDW) density wave
[7,20], provided that such a transition does occur for the physical number of fermion flavors
(N = 2).
Thus far, there have been only preliminary conclusions drawn on the role of the strong
spatial anisotropy of the quasiparticle dispersion (3 = v12 q12 + v22 q22 with v1 /v2 > 10) in
the high-Tc cuprates.
Although, in accord with general physical expectations, this anisotropy was found to
decrease upon IR renormalization, [20,21], there has been no consensus reached, regarding
universality (or a lack thereof) of the critical number of flavors Ncr , much less regarding
the possibility that the spatial anisotropy can drive this critical number to the values in
excess of 2.
To this end, it may be helpful to complement the renormalization group approach
adopted in Refs. [20,21] by a direct analysis of the strong coupling regime where the
gauge field propagator (7) is dominated by the fermion polarization Π(q). In the case
of anisotropic fermion dispersion, the latter becomes a function of the spatial components
of the momentum q multiplied by the corresponding velocities vi qi , thus allowing one to
scale such factors out in the Schwinger–Dyson equation for the fermion mass [19] and
suggesting that Ncr may even remain the same as in the isotropic case.
In any event, the effect of the spatial anisotropy of the fermion dispersion may be rather
different from that of the Lorentz non-invariant (e.g., pure Coulombic) couplings which can
cause Ncr to decrease [9,10], as compared to the value obtained in the Lorentz-invariant
case.
One should, however, be alerted by the fact that even in the perfectly Lorentz-invariant
case the status of the analyzes based on the Schwinger–Dyson equation (which, in this
particular case, suggests Ncr > 3) remains somewhat unclear. In this regard, it was pointed
out in Ref. [24] that the Schwinger–Dyson equation may systematically overestimate the
actual critical number of flavors which might, in fact, be as low as 3/2.
It is, therefore, quite remarkable that, as an alternate route, the intrinsic SDW instability
of the pseudogap phase of the cuprates can instead manifest itself through a divergent
staggered spin susceptibility [22,23]
1−64/(3π 2N)
χQ (q) ∝ q 2 , (43)
where the momentum q corresponds to the deviation from the SDW ordering vector
Q = (π, π).
In order for this behavior to occur, the presence of massless neutral fermions is rather
necessary than problematic. Curiously enough, the susceptibility (43) displays the negative
anomalous dimension which is exactly twice the value (28).
Although this observation does not necessarily imply that the staggered spin sus-
ceptibility given by the average χQ (x − y) =
G(x, y|A)G(y, x|A) is directly related
to the product of the two averages |G 0 (x − y)| 2 ∝
G(x, y|A) exp(−i
Γ A ν dz ν ) ×
G(y, x|A) exp(i Γ Aµ dz ), it may help one to elucidate the real physical meaning of
µ
the negative anomalous dimensions exhibited by the gauge invariant amplitude (2).
It is possible that, albeit being unfit for the role of the physical electron propagator, the
amplitude (2) may still bear some important information about the vertex functions which
also determine the behavior of the gauge-invariant susceptibilities.
As far as the experimental status of the QED3 theory of the cuprates is concerned,
it is presently unclear whether it can accommodate the phenomenon of time reversal
symmetry breaking, as suggested by the recent experiment [25], without generating the
Chern–Simons term in the fermion polarization.
In contrast, broken time reversal symmetry seems to be an intrinsic feature of some
alternate approaches to the pseudogap phase which, instead of the phase fluctuations of the
local parent dx 2 −y 2 -wave order parameter [4–7], focus on other bosonic collective modes,
such as emergent d-symmetrical CDW [26] or incipient secondary pairing (dx 2 −y 2 →
dx 2 −y 2 + idxy ) [27,28].
To this end, it is worth mentioning that the recent experimental reports of a genuine
quantum-critical behavior in the Ca-doped cuprates [29] are in quantitative agreement with
the predictions [28] based on the theory of Ref. [27]. Notably, in this alternate (Higgs–
Yukawa-type) theory of the cuprates the anomalous dimension of the electron propagator
was found to be positive [28,30].
We conclude by stressing that, regardless of the status of the QED3 theory of the
cuprates itself, the problem of constructing the gauge invariant fermion propagator in
the effective massless QED-like theories, thus far, has received a lesser attention than it,
arguably, deserves.
This problem needs to be settled before one can start drawing solid, rather than wishful,
conclusions about the true behavior in these as well as other gauge field models, including
non-Abelian and discrete symmetry (say, Zn ) ones.
In the present paper, we undertook an attempt to clarify this issue which may have
already resulted in a widespread confusion, in particular, as far as the QED3 theory of
the pseudogap phase of the cuprates is concerned. We demonstrated that the previously
proposed ansatz exibits a clearly unphysical behavior and, therefore, needs to be modified.
In the course of this analysis we established an interesting property of “partial gauge
invariance” of the family of functions (12) with ξ = 0 which, we believe, might have
been partly responsible for the erroneous conclusions drawn in [5]. We also conjectured
an alternate form of the physical electron propagator which decays with distance faster
than any power-law, thus challenging the interpretation of the experimental data based on
the idea of the Luttinger-like behavior in the cuprates proposed in Refs.[4–6].
Acknowledgements
The author acknowledges valuable communications with V.P. Gusynin. This research
was supported by the NSF under Grant No. DMR-0071362.
References
[1] P.A. Lee, N. Nagaosa, P.A. Lee, Phys. Rev. Lett. 64 (1990) 2450;
J. Gan, E. Wong, Phys. Rev. Lett. 71 (1993) 4226;
D.V. Khveshchenko, P.C.E. Stamp, Phys. Rev. Lett. 71 (1993) 2118;
D.V. Khveshchenko, P.C.E. Stamp, Phys. Rev. B 49 (1994) 5842;
C. Nayak, F. Wilczek, Nucl. Phys. B 417 (1994) 359;
C. Nayak, F. Wilczek, Nucl. Phys. B 430 (1994) 534;

L.B. Ioffe, D. Lidsky, B.L. Altshuler, Phys. Rev. Lett. 73 (1994) 472;
B.L. Altshuler, L.B. Ioffe, A. Millis, Phys. Rev. B 50 (1994) 14048;
Y.B. Kim, et al., Phys. Rev. B 50 (1994) 17917;
S. Chakravarty, R.E. Norton, O. Syljuasen, Phys. Rev. Lett. 74 (1995) 1423.
[2] N. Dorey, N.E. Mavromatos, Nucl. Phys. B 386 (1992) 614;
I.J.R. Aitchison, N.E. Mavromatos, Phys. Rev. B 53 (1996) 9321.
[3] D.H. Kim, P.A. Lee, Ann. Phys. 272 (1999) 133.
[4] W. Rantner, X.-G. Wen, Phys. Rev. Lett. 86 (2001) 3871, cond-mat/0105540.
[5] Z. Tesanovic, M. Franz, Phys. Rev. Lett. 87 (2001) 257003;
M. Franz, Z. Tesanovic, O. Vafek, cond-mat/0203047;
M. Franz, Z. Tesanovic, O. Vafek, cond-mat/0203333;
M. Franz, Z. Tesanovic, O. Vafek, cond-mat/0204536.
[6] J. Ye, Phys. Rev. Lett. 87 (2001) 227003;
J. Ye, Phys. Rev. B 65 (2002) 214505.
[7] I.F. Herbut, Phys. Rev. Lett. 88 (2002) 047006, cond-mat/0202491.
[8] J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Phys. Rev. Lett. 69 (1992) 172;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Nucl. Phys. 406 (1993) 771;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Nucl. Phys. 424 (1994) 595;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Phys. Rev. Lett. 77 (1996) 3589;
J. Gonzalez, F. Guinea, M.A.H. Vozmediano, Phys. Rev. B 59 (1999) 2474.
[9] D.V. Khveshchenko, Phys. Rev. Lett. 87 (2001) 206401;
D.V. Khveshchenko, Phys. Rev. Lett. 87 (2001) 246802.
[10] E.V. Gorbar, V.P. Gusynin, V.A. Miransky, I.A. Shovkovy, cond-mat/0202422.
[11] N.S. Craigie, H. Dorn, Nucl. Phys. B 185 (1981) 204;
N.G. Stefanis, Nuovo Cimento A 83 (1984) 205;
A.I. Karanikas, C.N. Ktorides, Phys. Rev. D 52 (1995) 5883;
A.I. Karanikas, C.N. Ktorides, N.G. Stefanis, Phys. Rev. D 52 (1995) 5898.
[12] D.V. Khveshchenko, A.G. Yashenkin, cond-mat/0202173;
D.V. Khveshchenko, A.G. Yashenkin, cond-mat/0204215.
[13] L.S. Brown, Quantum Field Theory, Cambridge Univ. Press, Cambridge, 1992, pp. 521–524.
[14] D. Nash, Phys. Rev. Lett. 62 (1989) 3024.
[15] D.V. Khveshchenko, Phys. Rev. B 63 (2002) 235111, cond-mat/0112202.
[16] V.P. Gusynin, D.V. Khveshchenko, M. Reenders, cond-mat/0207372.
[17] M. Lavelle, D. McMullan, Phys. Rep. 279 (1997) 1;
E. Bagan, M. Lavelle, D. McMullan, Phys. Rev. D 56 (1997) 3732;
E. Bagan, M. Lavelle, D. McMullan, Ann. Phys. 282 (2000) 471, 503.
[18] H.J. Schulz, Phys. Rev. Lett. 71 (1993) 1864.
[19] T. Appelquist, D. Nash, L.C.R. Wijewardhana, Phys. Rev. Lett. 60 (1988) 2575;
T. Appelquist, J. Terning, L.C.R. Wijewardhana, Phys. Rev. Lett. 75 (1995) 2081.
[20] Z. Tesanovic, O. Vafek, M. Franz, Phys. Rev. B 65 (2002) 180511;
M. Franz, D.E. Sheehy, Z. Tesanovic, cond-mat/0203219.
[21] D. Lee, I. Herbut, cond-mat/0201088.
[22] W. Rantner, X.-G. Wen, cond-mat/0201521.
[23] V. Gusynin, A. Hams, M. Reenders, Phys. Rev. D 63 (2001) 045025.
[24] T. Appelquist, A.G. Cohen, M. Schmaltz, Phys. Rev. D 60 (1999) 045003.
[25] A. Kaminski, et al., cond-mat/0203133.
[26] S. Chakravarty, R.B. Laughlin, D.K. Morr, C. Nayak, Phys. Rev. B 63 (2001) 094503.
[27] M. Vojta, Y. Zhang, S. Sachdev, Phys. Rev. B 62 (2000) 6721;
M. Vojta, Y. Zhang, S. Sachdev, Phys. Rev. Lett. 85 (2000) 4940.
[28] D.V. Khveshchenko, J. Paaske, Phys. Rev. Lett. 86 (2001) 4672.
[29] Y. Dagan, G. Deutscher, Phys. Rev. Lett. 87 (2001) 177004;
R. Connelli, et al., Eur. Phys. J. 22 (2001) 411.
[30] M. Reenders, cond-mat/0110168.
CUMULATIVE AUTHOR INDEX B641–B642
Bagnoud, M. B641 (2002) 61 Hoffmann, L. B641 (2002) 188

Bastianelli, F. B642 (2002) 372 Hwang, D.S. B642 (2002) 344
Berglund, P. B641 (2002) 351 Hwang, S. B641 (2002) 376
Bilal, A. B641 (2002) 61
Blumenhagen, R. B641 (2002) 235 Isidro, J.M. B641 (2002) 111
Bouttier, J. B641 (2002) 519
Jahn, O. B642 (2002) 357
Brandhuber, A. B641 (2002) 351
Jegerlehner, F. B641 (2002) 285
Brodsky, S.J. B642 (2002) 344
Jejjala, V. B642 (2002) 483
Carlevaro, L. B641 (2002) 61
Kalmykov, M.Yu. B641 (2002) 285
Corley, S. B641 (2002) 131 Karabali, D. B641 (2002) 533
Cristofano, G. B641 (2002) 547 Khveshchenko, D.V. B642 (2002) 515
Cvetič, M. B642 (2002) 139 Kiem, Y. B641 (2002) 256
Czakon, M. B642 (2002) 157 Kiem, Y. B642 (2002) 389
Kim, S.-S. B641 (2002) 256
Dabholkar, A. B641 (2002) 223
Kim, Y. B642 (2002) 389
De Forcrand, P. B642 (2002) 290 Kitanine, N. B641 (2002) 487
Dermíšek, R. B641 (2002) 327 Kitanine, N. B642 (2002) 433
Di Francesco, P. B641 (2002) 519 Kitazawa, Y. B642 (2002) 210
Dittmaier, S. B642 (2002) 307 Kogut, J.B. B642 (2002) 181
Körs, B. B641 (2002) 235
Faraggi, A.E. B641 (2002) 93
Koukoutsakis, A. B642 (2002) 227
Faraggi, A.E. B641 (2002) 111
Fjelstad, J. B641 (2002) 376 Langacker, P. B642 (2002) 139
Foerster, A. B642 (2002) 501 Lee, S. B642 (2002) 389
Fradkin, E. B642 (2002) 483 Leigh, R.G. B642 (2002) 483
Levman, G. B642 (2002) 3
Garavuso, R. B641 (2002) 111 Links, J. B642 (2002) 501
Garland, L.W. B642 (2002) 227 Lü, H. B642 (2002) 173
Gehrmann, T. B642 (2002) 227 Lunin, O. B642 (2002) 91
Ghilencea, D.M. B641 (2002) 35 Lüst, D. B641 (2002) 235
Glover, E.W.N. B642 (2002) 227
Gluza, J. B642 (2002) 157 Maiella, G. B641 (2002) 547
González, J. B642 (2002) 407 Maillet, J.M. B641 (2002) 487
Groot Nibbelink, S. B641 (2002) 35 Maillet, J.M. B642 (2002) 433
Guan, X.-W. B642 (2002) 501 Månsson, T. B641 (2002) 376
Guitter, E. B641 (2002) 519 Marotta, V. B641 (2002) 547
Mathur, S.D. B642 (2002) 91
Hejczyk, J. B642 (2002) 157 Mesref, L. B641 (2002) 188
Henkel, M. B641 (2002) 405 Meziane, A. B641 (2002) 188
0550-3213/2002 Published by Elsevier Science B.V.

PII: S 0 5 5 0 - 3 2 1 3 ( 0 2 ) 0 0 8 4 9 - 0
532 Nuclear Physics B 642 (2002) 531–532
Nair, V.P. B641 (2002) 533 Shiu, G. B642 (2002) 139

Nakatsu, T. B642 (2002) 13 Sinclair, D.K. B642 (2002) 181
Nandi, S. B641 (2002) 327 Slavnov, N.A. B641 (2002) 487
Niccoli, G. B641 (2002) 547 Slavnov, N.A. B642 (2002) 433
Suyama, T. B641 (2002) 341
Okuda, T. B641 (2002) 393
Ooguri, H. B641 (2002) 3
Takács, G. B642 (2002) 456
Ott, T. B641 (2002) 235
Terras, V. B641 (2002) 487
Park, J. B642 (2002) 389 Terras, V. B642 (2002) 433
Parvizi, S. B641 (2002) 223 Toublan, D. B642 (2002) 181
Pawlowski, J.M. B642 (2002) 357
Philipsen, O. B642 (2002) 290 Vafa, C. B641 (2002) 3
Ponsot, B. B642 (2002) 114 Veretin, O. B641 (2002) 285
Vives, O. B641 (2002) 93
Raby, S. B641 (2002) 327
Ramgoolam, S. B641 (2002) 131 Watts, G.M.T. B642 (2002) 456
Remiddi, E. B642 (2002) 227 Wei, Z.-T. B642 (2002) 263
Rey, S.-J. B641 (2002) 256
Roth, M. B642 (2002) 307
Rühl, W. B641 (2002) 188 Yang, M.-Z. B642 (2002) 263
Sato, H.-T. B641 (2002) 256 Zhou, H.-Q. B642 (2002) 501
Schmidt, I. B642 (2002) 344 Zirotti, A. B642 (2002) 372

Nucl - Phys.B v.642 PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Nucl - Phys.B v.642 PDF

Caricato da

Copyright:

Formati disponibili

Nuclear Physics B 642 (2002) 3–10

The structure of the pion and nucleon, and leading

PACS: 12.40.Nn; 13.60.Hb; 13.60.Rj

Keywords: Nucleon; Pion; Meson exchange; Structure

E-mail address: levman@physics.utoronto.ca (G. Levman).

4 G. Levman / Nuclear Physics B 642 (2002) 3–10

2. Meson exchange picture of leading baryon production

3. Leading neutron production and pion exchange

In hadro-production experiments (projectile = p, π ) leading neutron production is

• an absorptive factor a(t);

G. Levman / Nuclear Physics B 642 (2002) 3–10 5

• ρ, a2 exchange and ∆ production increase neutron production, while absorptive effects

4. The effective flux

where deconvolution is written as division. By construction and definition feff corrects

It is important to emphasize: experiments find that a convenient parameterization of feff is

6 G. Levman / Nuclear Physics B 642 (2002) 3–10

5. The structure of the pion and the effective flux

• W is the single relevant leptonic variable (Q2 = 0);

In contrast to photo-production, the situation in electro-production is less well

G. Levman / Nuclear Physics B 642 (2002) 3–10 7

5.2. F2π at low x

8 G. Levman / Nuclear Physics B 642 (2002) 3–10

• one-pion-exchange is not assumed. It is only assumed that pion exchange dominates

If absorption approximately compensates for background exchanges in hadro- and photo-

G. Levman / Nuclear Physics B 642 (2002) 3–10 9

absorptive effects is approximately 20–30%. Background processes must contribute at

10 G. Levman / Nuclear Physics B 642 (2002) 3–10

[1] ZEUS Collaboration, S. Chekanov et al., Nucl. Phys. B 637 (2002) 3.

Classical open-string field theory: A∞ -algebra,

E-mail address: nakatsu@het.phys.sci.osaka-u.ac.jp (T. Nakatsu).

for n = 0, 1, 2, . . . . Here I, J are the multi-indices satisfying I ∪ J = {1, . . . , n}.

[Note. To study the renormalizability of φ 4 -theory in d-dimensions (i.e., to search an

n-valent vertex at Λ = Λ0 e−ζ 4-valent vertex at Λ0

The evolution of the coupling constants can be read as

We start Section 7 with a further investigation on the renormalization group flow

2. Microscopic open-string field theory

2.1. Open-string Hilbert space

imposing the conditions:

2.2. BPZ conjugation

The Belavin–Polyakov–Zamolodchikov (BPZ) conjugation is a general operation of

ϕ (0) (z0 ) (dz0 )∆ = ϕ (∞) (z∞ )(dz∞ )∆ , (2.8)

It can be generalized to any product of operators by letting (ϕn(∞)

(ϕnl ) . This generalization induces a linear map T : H(∞) → (H(0) )∗ by

2.3. Reflector ω12 |

Reflector ω12 | is a vector of (H⊗2 )∗ and is determined by the condition

|A1 |B2 = (−)G(A)G(B)|B2 |A1 . (2.24)

2.4. Inverse reflector |S12 

√ of fi (Vi ) on the z-plane. They exactly

2.5. Symmetric 3-vertex 1 2 3|

Let Vi for i = 1, 2, 3 be unit disks |vi | 1 on the vi -plane. Let fi for i = 1, 2, 3 be

where F and S are meromorphic functions given by

ui -plane) to the w-plane, F and S in (2.35) are given by 1

ϕ U (z)(dz)∆ = ϕ V (v)(dv)∆ . (2.38)

2.6. Witten’s Φ 3 -action of open-string field theory

Open-string field Φ is a vector of the open-string Hilbert space H. It is grassmann-

Definition 2.1 (Cubic action of open-string field theory).

3. Low energy theory vertices

Let us consider a second-quantization of open-string field theory. Schematically it is

Fig. 5. Open-string strip. τ (> 0) is the Schwinger imaginary time.

3.1. Low energy theory vertices; examples

diagram are clockwise-ordered. Open-string diagram is a metrized connected trivalent tree

3.1.1. Open-string 4-vertex 1 2 3 4 : ζ |

Definition 3.1 (Open-string 3-vertex at ζ ). Open-string 3-vertex at the cut-off scale ζ is

2.3. Reflector ω12 |

Reflector ω12 | is a vector of (H⊗2 )∗ and is determined by the condition

|A1 |B2 = (−)G(A)G(B)|B2 |A1 . (2.24)

2.4. Inverse reflector |S12

2.5. Symmetric 3-vertex 1 2 3|

3.1.1. Open-string 4-vertex 1 2 3 4 : ζ |

Lemma 3.1. The BRST charge acts on the state x|(1234) as

3.1.2. Open-string 5-vertex 1 2 3 4 5 : ζ |

dyi (xi , yi )|(12345)

Conjecture. Open-string n-vertex 1 . . . n : ζ | ∈ (H⊗n )∗ (n 3) can be taken so that it

× |Φq2 +2 · · · |Φq1 +q2 +1 |ρq1 +q2 +2

The bilinear form g is non-degenerate on HS . ω12 | = ω21 | implies g(A, B) =