Nuclear Physics B

NUCLEAR PHYSICS B
Journal devoted to the experimental and theoretical study of the fundamental

constituents of matter and their interactions
Editorial Board
Supervisory Editors: G. ALTARELLI, Geneva, Switzerland; W. BARTEL, Hamburg, Germany; C. BECCHI, Genova,
Italy; R.H. DIJKGRAAF, Amsterdam, The Netherlands; D. KUTASOV, Chicago, IL, USA; L. MAIANI, Rome, Italy;
H. MURAYAMA, Berkeley, CA, USA; T. NAKADA, Geneva, Switzerland; N. NEKRASOV, Bures-sur-Yvette, France;
H. OOGURI, CalTech, Pasadena, CA, USA; H. SALEUR, USC, Los Angeles, CA, USA; A. SCHWIMMER, Rehovot,
Israel
Associate Editors for Particle Physics: S.J. Brodsky, SLAC; M. Dine, UC, Santa Cruz; G. Dvali, New York; G.W.
Gibbons, Cambridge, UK; D. Gross, St. Barbara; S.S. Gubser, Princeton; L. Ibanez, Madrid; R.L. Jaffe, MIT; K. Kajantie,
Helsinki; M. Luescher, CERN; E. Rabinovici, Jerusalem; L. Randall, Princeton; A. Sen, Bombay; G. Steigman, Ohio
State; G. t Hooft, Utrecht; Y. Totsuka, KEK; P. van Baal, Leiden; S. Weinberg, Texas; J. Zinn-Justin, Saclay
Associate Editors for Field Theory and Statistical Systems: B. Duplantier, Saclay; A.W.W. Ludwig, Santa Barbara;
S. Lukyanov, Rutgers, Piscataway; T. Miwa, Kyoto; G. Mussardo, Trieste; M. Oshikawa, Tokyo; G. Parisi, Frascati;
A.M. Polyakov, Princeton; R. Sasaki, Kyoto; M. Zirnbauer, Cologne
Aims and Scope
Nuclear Physics B focuses on the domain of high energy physics and quantum field theory, and includes three main sections, i.e. particle physics, field theory and statistical systems and physical mathematics. The emphasis is on original
research papers. Conference Proceedings on the topics covered by Nuclear Physics B are published in the (separately
available) journal Nuclear Physics B-Proceedings Supplements.
Abstracted/Indexed in:
Chemical Abstracts/Current Contents: Physical, Chemical & Earth Sciences. Also covered in the abstract and citation
database SCOPUS . Full text available on ScienceDirect .
Subscription Information 2008
Volumes 788805 are scheduled for publication in 54 issues. Publication frequency: three times a month. Subscription
prices are available upon request from the Publisher or from the Regional Sales Office nearest you or from this journals
website (http://www.elsevier.com/locate/nuclphysb). A combined subscription to Nuclear Physics A ISSN 0375-9474
(volumes 798814), Nuclear Physics B ISSN 0550-3213 (volumes 788805) and Nuclear Physics B Proceedings Supplements ISSN 0920-5632 (volumes 174185) is available at a reduced rate. Subscriptions are accepted on a prepaid
basis only and are entered on a calendar year basis. Issues are sent by standard mail (surface within Europe, air delivery
outside Europe). Priority rates are available upon request. Claims for missing issues should be made within six months
of the date of dispatch.
Orders, claims, and journal enquiries: please contact the Customer Service Department at the Regional Sales Office
nearest you:
Orlando: Elsevier, Customer Service Department, 6277 Sea Harbor Drive, Orlando, FL 32887-4800, USA; phone: (+1)
(877) 8397126 [toll free number for US customers], or (+1) (407) 3454020 [customers outside US]; fax: (+1) (407)
3631354; e-mail: usjcs@elsevier.com
Amsterdam: Elsevier, Customer Service Department, PO Box 211, 1000 AE Amsterdam, The Netherlands; phone:
(+31) (20) 4853757; fax: (+31) (20) 4853432; e-mail: nlinfo-f@elsevier.com
Tokyo: Elsevier, Customer Service Department, 4F Higashi-Azabu, 1-Chome Bldg, 1-9-15 Higashi-Azabu, Minato-ku,
Tokyo 106-0044, Japan; phone: (+81) (3) 5561 5037; fax: (+81) (3) 5561 5047; e-mail: jp.info@elsevier.com
Singapore: Elsevier, Customer Service Department, 3 Killiney Road, #08-01 Winsland House I, Singapore 239519;
phone: (+65) 63490222; fax: (+65) 67331510; e-mail: asiainfo@elsevier.com
Advertising information
Europe, USA, Canada and ROW: Advertising orders and enquiries can be sent to: Miss Katrina Barton, phone: (+44)
(0) 20 7611 4117; fax: (+44) (0) 20 7611 4463; e-mail: commercialsales@elsevier.com.
South America: Advertising orders and enquiries can be sent to: Mr Tino DeCarlo, The Advertising Department,
Elsevier Inc., 360 Park Avenue South, New York, NY 10010-1710, USA; phone: (+1) (212) 633 3815; fax: (+1) (212)
633 3820; e-mail: t.decarlo@elsevier.com.
US mailing notice Nuclear Physics B (ISSN 0550-3213) is published weekly by Elsevier B.V., P.O. Box 211, 1000
AE Amsterdam, The Netherlands. The annual institutional subscription price in the USA is US$ 13612 (valid in North,
Central and South America), including air speed delivery. Periodical postage paid at Rahway NJ and additional mailing
offices. USA Postmaster: Send change of address: Nuclear Physics B, Elsevier, 6277 Sea Harbor Drive, Orlando, FL
32887-4800. Airfreight and mailing in the USA by Mercury International Limited, 365, Blair Road, Avenel, NJ 07001.
Nuclear Physics B 802 (2008) 139

www.elsevier.com/locate/nuclphysb
Algebra of transfer-matrices and YangBaxter equations

on the string worldsheet in AdS5 S 5
Andrei Mikhailov a,,1 , Sakura Schfer-Nameki b
a California Institute of Technology, 1200 E California Blvd., Pasadena, CA 91125, USA
b Isaac Newton Institute for Mathematical Sciences, 20 Clarkson Road, Cambridge, CB3 0EH, UK
Received 18 January 2008; accepted 28 April 2008

Available online 14 May 2008
Abstract
Integrability of the string worldsheet theory in AdS5 S 5 is related to the existence of a flat connection
depending on the spectral parameter. The transfer matrix is the open-ended Wilson line of this flat connection. We study the product of transfer matrices in the near-flat space expansion of the AdS5 S 5 string
theory in the pure spinor formalism. The natural operations on Wilson lines with insertions are described
in terms of r- and s-matrices satisfying a generalized classical YangBaxter equation. The formalism is
especially transparent for infinite or closed Wilson lines with simple gauge invariant insertions.
2008 Elsevier B.V. All rights reserved.
1. Introduction
Integrability of superstring theory in AdS5 S 5 has been a vital input for recent progress in
understanding the AdS/CFT correspondence. However quantum integrability of the string worldsheet sigma-model is far from having been established. The notion of quantum integrability is
well developed for relativistic massive quantum field theories, which describe scattering of particles in two spacetime dimensions. But the string worldsheet theory is a very special type of
a quantum field theory, and certainly not a relativistic massive theory. It may not be the most
natural way to think of the string worldsheet theory as describing a system of particles. It may
be better to think of it as describing certain operators, or rather equivalence classes of operators.
* Corresponding author.
E-mail addresses: andrei@theory.caltech.edu (A. Mikhailov), ss299@theory.caltech.edu (S. Schfer-Nameki).

1 On leave from Institute for Theoretical and Experimental Physics, 117259, Bol. Cheremushkinskaya, 25, Moscow,
Russia.
0550-3213/$ see front matter 2008 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2008.04.029
A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139
What does integrability mean in this case? Progress in this direction could be key to understanding the exact quantum spectrum, which goes beyond the infinite volume spectrum that is obtained
from the asymptotic Bethe ansatz [1,2].
The transfer matrix usually plays an important role in integrable models, in particular in
conformal ones [3]. The renormalization group usually acts nontrivially on the transfer matrix
[4,5]. But the string worldsheet theory is special. The transfer matrix on the string worldsheet is
BRST-invariant, and there is a conjecture that it is not renormalized. This was demonstrated in a
one-loop calculation in [6].
In this paper we will revisit the problem of calculating the Poisson brackets of the worldsheet
transfer matrices [712]. The transfer matrix is a monodromy of a certain flat connection on the
worldsheet, which exists because of classical integrability. One can
think of it as a kind of Wilson
line: given an open contour C, we calculate T [C] = P exp( C J ). Instead of calculating the
Poisson bracket we consider the product of two transfer matrices for two different contours, and
considering the limit when one contour is on top of another:
At first order of perturbation theory studying this limit is equivalent to calculating the Poisson
bracketswe will explain this point in detail. We find that the typical object appearing in this calculation is a dynamical (= field-dependent) R-matrix suggested by Maillet [1315]. The Maillet
approach was discussed recently for the superstring in AdS5 S 5 in [10,12,16].
The transfer matrix is a parallel-transport type of object. Given two points x and y on the string
worldsheet, we can consider the tangent spaces to the target at these two points, Tx (AdS5 S 5 )
and Ty (AdS5 S 5 ). The transfer matrix allows us to transport various vectors, tensors and spinors
between Tx (AdS5 S 5 ) and Ty (AdS5 S 5 ). This allows to construct operators on the worldsheet
by inserting the tangent space objects (for example + x) at the endpoints of the Wilson line:
or inside the Wilson line:
We study the products of the simplest objects of this type at the first order of perturbation theory.
The results are summarised in Section 2. The subsequent sections contain derivations, the main
points are in Sections 4.5 and 6. In Section 8 we discuss the consistency conditions (generalized
YangBaxter equations).
2. Summary of results
This section contains a summary of our results, and in the subsequent sections we will describe
the derivation.
2.1. Definitions
2.1.1. The definition of the transfer matrix
Two-dimensional integrable systems are characterized by the existence of certain currents J a ,
which have the property that the transfer matrix

a
T [C] = P exp J ea ,
(2.1)
C
is independent of the choice of the contour. In this definition ea are generators of some algebra.
The algebra usually has many different representations, so the transfer matrix is labelled by a
representation. We will write T [C] where the generators ea act in the representation .
For the string in AdS5 S 5 the algebra is the twisted loop algebra Lpsu(2, 2|4) and the
coupling of the currents to the generators is the following:
[]
[] 0
2
[] 4
1
3
J+ = J0+ N0+ e[]
(2.2)
+ J3+
e + J2+ e
+ J1+
e + N0+ e[]
,
[]
[] 0
[]
1
2
3
4
e + J2 e
+ J3
e + N0 e[]
.
J = J0 N0 e[] + J1
(2.3)
Here eam are the generators of the twisted loop algebra. We will use the evaluation representation
of the loop algebra. In the evaluation representation eam are related to the generators of some
representation of the finite dimensional algebra psu(2, 2|4) in the following way:
e3 = z3 t1 ,
2
e
= z2 t2 ,
e1 = zt1 ,
etc.,
(2.4)
where z is a complex number, which is called spectral parameter. Further details on the conventions can be found in Section 3.1 and in [6].
2.1.2. Setup: Expansion around flat space and expansion in powers of fields
The gauge group g0 psu(2, 2|4) acts on the currents in the following way:
0 J1 = [0 , J1 ],
0 J2 = [0 , J2 ],
0 J0 = d0 + [0 , J0 ],
where 0 g0 .
0 J3 = [0 , J3 ],
(2.5)
In terms of the coordinates of the coset space

J = dgg 1 ,
g P SU(2, 2|4).
(2.6)
The gauge invariance (2.5) acts on g as follows:

g hg,
h = e ,
g0 .
(2.7)
given by Eq. (2.1) and the other is g 1 T g.
There are two versions of the transfer matrix. One is T

Notice that g 1 T g is gauge invariant, while T is not. We should think of T [C] as a map from
the (supersymmetric) tangent space T (AdS5 S 5 ) at the starting point of C to T (AdS5 S 5 ) at
the endpoint of C.
The choice of a point in AdS5 S 5 leads to the special gauge, which we will use in this paper:
g = eR
1 ( + )
L
R
eR
1 x
(2.8)
Here R is the radius of AdS space, and it is introduced in (2.8) for convenience. The action has
a piece quadratic in x, and interactions which we can expand in powers of x, . There are also
pure spinor ghosts , w. All the operators can be expanded2 in powers of x, , , w. We will refer
to this expansion as expansion in powers of elementary fields, or expansion in powers of x.
Every power of elementary field carries a factor R 1 . The overall power of R 1 is equal to twice
the number of propagators plus the number of uncontracted elementary fields. A propagator is a
contraction of two elementary fields.
The currents are invariant under the global symmetries, up to gauge transformations. For
example the global shift

1
x, [x, ] + ,
3R 2
results in the gauge transformation of the currents with the parameter

1
h , x; e = exp 2 [x, ] + .
2R
Sg 0 x = x + +
(2.9)
(2.10)
To have the action invariant we should also transform the pure spinors with the same parameter:

1
1
[x, ], ,
w+ =
[x, ], w+
=
(2.11)
2R 2
2R 2
and same rules for w , .

2.2. Fusion and exchange of transfer matrices
2.2.1. The product of two transfer matrices
Consider the transfer matrix in the tensor product of two representations 1 2 . There are
two ways of defining this object. One way is to take the usual definition of the Wilson line

a
P exp J (z)ea ,
(2.12)
and use for ea the usual definition of the tensor product of generators of a Lie superalgebra:
1 (ea ) 1 + ()F a 2 (ea ),
(2.13)
where a is 0 if ea is an even element of the superalgebra, and 1 if ea is an odd element of the

superalgebra.
Another possibility is to consider two Wilson lines T1 and T2 and put them on top of each
other. In other words, consider the product T2 T1 . In the classical theory these two definitions
of the composite Wilson line are equivalent, because of this identity:
e e = e1+1 .
(2.14)
2 The expansion in powers of elementary fields is especially transparent in the classical theory where it can be explained
in the spirit of [17]. We write
x=
N

a=1
ka w +
a eika w+i
ka +kb )w + .
Gab (ka , kb )a b ei(ka +kb )w+i(
ab
where a , a = 1, 2, . . . , N are nilpotents: a2 = 0 for every a. The nilpotency of a implies that the powers of x higher
than x N automatically drop out.
But at the first order in h there is a difference. The difference is related to the singularities in the
operator product of two currents.
Consider the example when the product of the currents has the following form:
1 ab c
A J + ,
(2.15)
w c +
where dots denote regular terms. Take two contours C1 and C2 and calculate the product
J+a (w)J+b (0) =
T2 [C2 ]T1 [C1 ],
(2.16)
where the indices 1 and 2 indicate that we are calculating the monodromies in the representations 1 and 2 , respectively. For example, suppose that the contour C1 is the line = 0 (and
runs from to +), and the contour C2 is at = y (and [, +]). Suppose that we
bring the contour of 2 on top of the contour of 1 , in other words y 0. Let us expand both
T2 [C2 ] and T1 [C1 ] in powers of R 2 , and
think of them as series of multiple integrals of J .
Consider for example a term in which one J comes from T2 [C2 ] and another J comes from
T1 [C1 ]. We get

d1 d2 J+a (y, 2 )(ea 1)J+b (0, 1 )(1 eb )

1
(2.17)
Aab J c (ea 1)(1 eb ).
=
d1 d2
2 1 + iy c +
The pole 2 11 +iy leads to the difference between limy0 T2 [C + y]T1 [C] and T2 1 [C].
Indeed, the natural definition of the double integral when y = 0 would be that when 1 collides
with 2 we take a principle value:

V.P.
d1 d2 J+a (0, 2 )(ea 1)J+b (0, 1 )(1 eb ).
(2.18)
Here V.P. means that we treat the integral as the principal value when 1 collides with 2 . Modulo
the linear divergences, which we neglect, the integral (2.18) is finite. This is because ea 1
commutes with 1 eb . But such a V.P. integral is different from what we would get in the limit
y 0, by a finite piece. Indeed:

dw J+a (w + i)J+b (0) = V.P. dw J+a (w)J+b (0)
(2.19)
c
+ iAab
c J+ (0).
(2.20)
The second row is the difference between the V.P. prescription and the limy0 prescription. The
c
additional piece iAab
c J+ (0) could also be interpreted as the deformation of the generator to
c
which J+ couples in the definition of the transfer matrix:

F b
J+c ec 1 + ()F c ec J+c ec 1 + ()F c ec + iAab
(2.21)
eb .
c ea ()
We have two different definitions of the transfer matrix in the tensor product of two representations. Is it true that these two definitions actually give the same object? There are several logical
possibilities:
1. There are several ways to define the transfer matrix, and they all give essentially different
Wilson line-like operators.
2. We should interpret Eq. (2.21) as defining the deformed coproduct on the algebra of generators. The algebra of generators is in our case a twisted loop algebra of psu(2, 2|4). There are
at least three possibilities:
(a) The proper definition of the transfer matrix actually requires the deformation of the algebra of generators ea , and the deformed algebra has deformed coproduct.
(b) The algebra of generators is the usual loop algebra, but it has a nonstandard coproduct;
limy0 T2 [C + y]T1 [C] is different from T1 2 [C], the difference being the use of a
nonstandard coproduct. We are not aware of a mathematical theorem which forbids such
a nontrivial coproduct.
(c) The coproduct defined by Eq. (2.21) is equivalent to the standard one, in a sense that it
is obtained from the standard coproduct by a conjugation:

0 ec = ec 1 + ()F c ec ,
(2.22)
c
F c
ab
F c
e = ec 1 + () ec + iAc ea () eb
i
i
= e 2 r ec 1 + ()F c ec e 2 r .
(2.23)
We will argue that what actually happens (at the tree level) is a generalization of 2(c). The deformation (2.23) is almost enough to account for the difference between limy0 T2 [C + y]T1 [C]
and T1 2 [C], but in addition to (2.23) one has to do a field-dependent generalized gauge transformation.3 The correct statement is:
for a contour C going from the point A to the point B
lim T2 [C + y]T1 [C] = e
y0
i
2 r (A)
T1 2 [C]e 2 r (B) ,
(2.24)
where r is field dependent (dynamical). In fact r is of the order h.

This paper is all about the
tree level. Therefore all we are saying is:
lim T2 [C + y]T1 [C]
y0

i
(2.25)
r (A)T1 2 [C] T1 2 [C]r (B) + ,
2
where dots stand for loop effects. The hat over the letter r shows that this is a field-dependent
object. We will also use a field-independent r-matrix which will be denoted r without a hat; r
is the leading term in the near-flat-space expansion of r , which is the expansion in powers of
elementary fields explained in Section 2.1.2:
= T1 2 [C] +

i 2
z1 z12 t 2 t 2 , x t 2 , x z22 z22 t 2
2

i 3
z1 z1 t 1 t 3 , L t 3 , L z23 z2 t 1
2

i 1
z1 z13 t 3 t 1 , R t 1 , R z21 z23 t 3 + .
r = r
(2.26)
3 Generalized gauge transformation is J f (d + J )f 1 . If f exp g then this is a usual (or proper gauge trans0
formation as defined in Section 2.1.2). If we relax this condition we get the generalized gauge transformation see
Section 5.
Fig. 1. Fusion of transfer matrices.
Fig. 2. Exchange of transfer matrices.
Here r is given by Eq. (2.33) and dots stand for the terms of quadratic and higher orders in
x and . The pure spinor ghosts do not enter into the expression for r , only the matter fields x
and .
The special thing about the constant term r is that it is a rational function of the spectral
parameter with the first order pole at zu = zd . The coefficients of the x, -dependent terms are
all polynomials in zu , zd , zu1 , zd1 . The field dependence of the r matrix in this example is related
to the fact that the pair of Wilson lines with loose ends is not a gauge invariant object.4
Eq. (2.24) is schematically illustrated in Fig. 1. A consequence of (2.24) is the equivalence
relation for the exchange of the order of two transfer matrices, see Fig. 2:

lim TCu uzu TCd dzd = exp(i r ) lim TCu uzu TCd dzd exp(i r ).
(2.27)
Cu Cd
Cu
Cd
2.2.2. Relation to Poisson brackets

At the tree level the calculation of the fusion of transfer matrices is equivalent to the calculation of the Poisson brackets. This follows from the definition of the Poisson bracket:

1
{T1 , T2 } = lim
(2.28)
lim T1 [C + y]T2 [C] lim T2 [C + y]T1 [C]
h0
y0+
i h y0+
and the equation:

lim T1 [C + y]T2 [C] + lim T2 [C + y]T1 [C] = 2T1 2 [C] + O h 2 ,
y0+
y0+
which holds to the first order in h . These two equations and Eq. (2.25) imply

T1 [C], T2 [C] = r (A)T1 2 [C] T1 2 [C]r (B)
(2.29)
(2.30)
and therefore the calculation of r is actually equivalent to the calculation of the Poisson brackets.
To derive (2.29) we expand the product T [C + y]T [C] as normal ordered product plus contractions. At the tree level only one contraction is needed; schematically we get
J (w)J (0) = :J (w)J (0): + F (w, w),
4 We use the special gauge (2.8), therefore in our formalism the lack of gauge invariance translates into the lack of
translational invariance.
Fig. 3. An infinite Wilson line with an operator insertion.
where F (w, w)
is w1 or w12 or w1 or w12 times some expression regular at w 0; see Section 3.
Then Eq. (2.29) follows from the relation

1
1
1
+
= 2V.P. n
lim0+
(2.31)
n
n
(w + i)
(w i)
w
applied to the singular part of F (w, w).
The standard calculation of the Poisson bracket of two transfer matrices involves the equal
time Poisson brackets of the currents {J ( ), J ( )}. This is proportional to ( ) or (
). This is equivalent to what we are doing because

1
2i(1)n
1
( ).
=
lim
(2.32)
(w i)n
(n 1)!
0+ (w + i)n
We conclude that the difference between our approach based on the notion of fusion and the
standard approach to calculating the Poisson brackets is a matter of notations. (But we believe
that our notations are more appropriate for calculating beyond the tree level.)
2.2.3. r- and s-matrices and generalized classical YBE
The open ended contours like the ones shown in Figs. 1 and 2 are strictly speaking not gauge
invariant. In our approach we fix the gauge (2.8) and therefore it is meaningful to consider these
operators as operators in the gauge fixed theory. Nevertheless we feel that these are probably not
the most natural objects to study, at least from the point of view of the differential geometry of
the worldsheet.
The natural objects to consider are infinite (or periodic) Wilson lines with various operator
insertions, see Fig. 3. How to describe the algebra formed by such operators? What is the relation between
and
? We will find that the description of this algebra involves
matrices r and s which have the following form:

(z1 , z2 ) 3 1
(z1 , z2 ) 0
(2.33)
z1 z2 t t 3 + z13 z2 t 3 t 1 + z12 z22 t 2 t 2 + 2 4
t t 0,
4
4
z1 z2
z1 z24

s = z11 z23 z13 z2 t 3 t 1 + z12 z22 z12 z22 t 2 t 2 + z13 z21 z1 z23 t 1 t 3 , (2.34)
r=
where
2
2

(z1 , z2 ) = z12 z12 + z22 z22 ,
(z1 , z2 ) = 1 + z14 z24 z14 z24 .
The notations used in (2.33), (2.34) are explained in Section 3.1. In Section 8 we will study the
consistency conditions for r and s, which generalize the standard classical YangBaxter algebra.
At the tree level we will get a generalization of the classical YangBaxter equations:

(r12 + s12 ), (r13 + s13 ) + (r12 + s12 ), (r23 + s23 )

(2.35)
+ (r13 + s13 ), (r23 s23 ) = t123 ,
where the RHS is essentially a gauge transformation; the explicit expression for t is (8.7). Note
that neither r nor s satisfy the standard classical YBE on their own, and even the combination
r s satisfies an analogue of the cYBE only when acting on gauge invariant quantities. Therefore
we have a generalization of the classical YangBaxter equations with the gauge invariance built
in.
2.3. Infinite Wilson lines with insertions
To explain how r and s enter in the description of the algebra of transfer matrices, we have to
introduce some notations.
2.3.1. General definitions
Consider a Wilson line with an operator insertion, shown in Fig. 3. For this object to be gauge
invariant, we want O to transform under the gauge transformations in the representation
of the gauge group g0 psu(2, 2|4). We will introduce the notation H(1 2 ) for the space of
operators transforming in the representation 1 2 of g0 . With this notation5 :
O H( ).
(2.36)
Here means the representation dual to .

For example, we can take the evaluation representation of the loop algebra corresponding
to the adjoint of psu(2, 2|4), with some spectral parameter z, and take O = J2+ :

J2+ H adz adz .
(2.37)
In other words, consider:

0

+

ad J (z) ad(J2+
ad J (z) .
P exp
)P exp
(2.38)
This is gauge invariant because ad ad ad as a representation of psu(2, 2|4) and therefore

also as a representation of g0 . Of course, we could also pick O = ad(J1+
) or ad(J3+
). These
operators have engineering dimension (1, 0). Geometrically they correspond to + x or + .
We want to study the objects of this type in the situation when two contours come close to each
other. For example, consider a Wilson line in the representation u with some operator O inserted
at the endpoint. Let us take another Wilson line, an infinite one, carrying the representation d ,
and put the Wilson line with the representation u on top of the one carrying d . In the limit
when the separation goes to zero we should have a Wilson line carrying u d at and d
at +.
This defines maps F , see Fig. 4. If O is inserted inside the contour (rather than at the endpoint) we get G . To summarize:
F+ : H(u ) H(u d d ),
(2.39)
F : H(d ) H(u
(2.40)
u ),
5 If is a trivial (zero-dimensional) representation, then the Wilson line terminates:

O H( ).
. In this case
10
Fig. 4. Fusion operations F+ , F and G+ .
G+ : H(u u ) H(u d u d ),
(2.41)
G : H(d d ) H(u d u d ).
(2.42)
2.3.2. Split operators

i into the upper line, and O j into
We also want to be able to insert two operators: Oup
dn
i i
Odn is gauge inthe lower line, such that they are not separately gauge invariant, but i Oup
variant. For example, for a gauge invariant operator O we can insert C t2 {t2 , O} where
C = C (xup , xdn , up , dn ) is some kind of a parallel transport. This will be gauge invariant.
We will use a thin vertical line to denote such a split operator
In the tensor product notations, for example when we write C t2 {t2 , O}, we assume that the
first tensor generator in the tensor product (in this case t2 ) acts on the upper Wilson line, and the
second (in this case {t2 , O}) on the lower line. We will need such operators in the limit where
the upper contour approaches the lower contour. Strictly speaking the split operator will depend
on which parallel transport is used even in the limit of coinciding contours, by the mechanism
similar to what we described in Section 2.2.1. We will not discuss this dependence in this paper,
because it is not important at the tree level.
The exchange map R acts as follows:

R : Hsplit 1out 1in , 2out 2in
Hsplit 2out 2in , 1out 1in .
The pictorial representation of R is
(2.43)
11
2.3.3. Switch operators

Given a representation of psu(2, 2|4) we denote the evaluation representation z . Consider
in
out
u = zu , u = zu and d = zd , where zuin , zuout and zd are three different complex numbers.
in
out
Take O = 1. This is gauge invariant because zu and zu are equivalent as representations of
the gauge group g0 . We can think of such O as the operator changing the spectral parameter,
or the switch operator.
in
out
For abbreviation we write uin = zu and uout = zu . Let us first consider the operation G+
in Fig. 4, with O = 1. In Section 6.1 we will show that G+ (1) is given (at the tree level) by this
formula:

i
(r + s)|uin d (r + s)|uout d + .
G+ (1) = 1 +
(2.44)
2
Here the r matrix appears from the diagrams involving the interaction of currents in the bulk of
the contours. It comes from the deformed coproduct, see Eq. (2.23). The matrix s comes from the
diagrams which are localized near the insertion of O. These are the additional diagrams existing
because we inserted the impurities.
The corresponding exchange relation is
where
in

out

, zdn ir+ zup
, zdn + ,
R(1switch 1)=1 + ir+ zup
r+ = r + s.
(2.45)
Similarly, if we lift the switched contour from the lower position to the upper position, we should
insert R(1 1switch ):
in

out

R(1 1switch ) = 1 + ir zup
, zdn ir zup
, zdn + ,
r = r s.
(2.46)
It is useful to write down explicit formulas for r = r s following from (2.33) and (2.34):

2

r + s
1 2
=
zd zd2 zu zd3 t 1 t 3 + zu2 zd2 t 2 t 2 + zu3 zd t 3 t 1

4
4
2 u d zu zd

(2.47)
+ zu2 zd2 zu2 zu2 zd2 zd2 t 0 t 0 ,

2

1 2
r s
zu zu2 zu zd3 t 1 t 3 + zu2 zd2 t 2 t 2 + zu3 zd t 3 t 1
=

4
4
2 u d zu zd

+ zu2 zd2 zu2 zu2 zd2 zd2 t 0 t 0 .
(2.48)
12
We will use the notation

R+ = R(1switch 1),
(2.49)
R = R(1 1switch ).
(2.50)
2.3.4. Intersecting Wilson lines

In this paper we mostly consider exchange and fusion as relations in the algebra generated
by transfer matrices with insertions. It is also possible to think of these operations as defining
vertices connecting several Wilson lines in different representations. For example the fusion can
be thought of as a triple vertex:
Such vertices will become important if we want to consider networks of Wilson lines. We want to
define this triple vertex so that the diagram is independent of the position of the vertex, just as it is
independent of the shape of the contours. At the tree level we suggest the following prescription:
The subscripts go-around and V.P. require explanation. They indicate different prescriptions
for dealing with the collisions of the currents
coupled to t 1 with the currents coupled to
1 t. Suppose that we consider the integral dw Ja t a 1 and the integration contour has to
pass through several insertions of Jb 1 t b . The prescription is such that to the right of the point
V
the collision as the principal value integral, while to the left of V the contour for
we treat
dw (J a ta ) 1 it goes around the singularity in the upper half-plane:
The insertion of 1 + 2r is necessary to have independence of the position of the vertex V . Notice
that in defining the worldsheet fusion we use r rather than r + s or r s. This is different from
the formula (2.44) for G+ which uses r + s.
13
2.4. Outline of the calculation

2.4.1. Use of flat space limit
We will use the near flat space expansion of T [C + y]T [C], see Section 2.1.2. For our calculation it is important that the transfer matrix is undeformable. The definition given by Eqs. (2.1),
(2.2) and (2.3) cannot be modified in any essential way. More precisely, we will use the following
statement. Suppose that there is another definition of the contour independent Wilson line of the
form

T new = P exp I a ea ,
(2.51)
C
where the new currents I have ghost number zero and coincide with J at the lowest order in the
near flat space expansion. In other words:
1
I1 = R + ,
R
1
1
I3 = L + ,
I2 = x + ,
R
R
I0 = 0 + ,
where dots denote the terms of the order R12 or higher. Let us also require that T new is invariant
(up to conjugation) under the global symmetries including the shifts (2.9). Then
new B

T
(2.52)
= exp (A) T exp (B) ,
A
where (w, w)
is a power series in x and with zero constant term. Eq. (2.52) says that the
transfer matrix is an undeformable object.
2.4.2. Derivation of r
We will start in Section 4 by calculating the couplings of d x and d . These are the standard
couplings of the form R 1 d x (t2 1 + 1 t2 ) plus corrections proportional to R 3 d x arising
as in Section 2.2.1. These couplings are defined up to total derivatives, i.e., up to the couplings of
dx. In particular, a different prescription for the order of integrations would add a total derivative
coupling. It will turn out that with one particular choice of the total derivative terms the coupling
is of the form

i 2
exp
r dx t 1 + 1 t2 + dL t3 1 + 1 t3
2

i
(2.53)
+ dR t1 1 + 1 t1 exp r .
2
where r is the c-number matrix defined in Eq. (2.33). These total derivative terms are important,
because they correspond to the field dependence of r in (2.24). The same prescription for the
total derivatives gives the right couplings for [x, d x] and [, d ] (Sections 5.2, 5.2.2 and 5.3).
The best way to fix the total derivatives in our approach is by looking at the effects of the global
shift symmetry (2.9) near the boundary, as we do in Section 6.2 deriving (2.26).
According to Section 2.4.1 Eq. (2.53) implies that:
lim T2 [C + y]T1 [C]

i
i
r T1 2 [C] exp r exp (A) .
= exp (A) exp
2
2
y0
(2.54)
14

i
The right-hand side is e 2 r (A) T1 2 [C]e 2 r (B) , the difference between r and r is due to the
field dependent gauge transformation with the parameter .
2.4.3. Boundary effects and the matrix s
We then proceed to the study of the boundary effects and derive the exchange relations for the
simplest gauge invariant insertionthe switch operator, see Eqs. (2.45) and (2.46). The matrix s
given by Eq. (2.34) arises from the diagrams localized on the insertion of the switch operator.
2.4.4. Dynamical vs. c-number
The r and s matrices appearing in the description of the exchange relations are generally
speaking field dependent, and in our approach they are power series in x and . These series
depend on which insertions we exchange, although the leading c-number term in r given by
(2.33) should be universal. For the exchange of the switch operator we claim that r and s entering
Eqs. (2.44), (2.45) and (2.46) are exactly c-number matrices given by (2.47) and (2.48). In other
words, all the field dependent terms cancel out. The argument based on the invariance under the
global shift symmetry is given in Section 6.1.
2.4.5. BRST transformation
1
The action of Q on the switch operator is the insertion of ()F ( zout
of this action with the exchange relation is verified in Section 7.
1
).
zin
The consistency
3. Short distance singularities in the product of currents

3.1. Notations for generators and tensor product
Recall that the notations for generators of Lpsu(2, 2|4) is
e3 = z3 t3 ,
2
e
= z2 t2 ,
e1 = zt3 .
(3.1)
The collective notations for the generators of psu(2, 2|4) are

, , [ ] .
tai , i Z4 , a ,
(3.2)
The coproduct for superalgebra involves the operator (1)F , which has the property (1)F t3 =
t3 (1)F , see (2.21). The origin of ()F can be understood from this example

e1 (t1) e2 (t 1) e3 (t
=e
1)
e1 (1t) e2 (1t ) e3 (1t ) |0 |0
1 (t1+()F t) 2 (t 1+()F t ) 3 (t 1+()F t )
|0 |0,
(3.3)
(3.4)
where 1,2,3 are three Grassman variables and t, t , t three generators of some algebra, acting on the representation generated by a vector |0, where ()F |0 = |0, ()F t|0 = t|0,
()F t t|0 = t t|0, etc.
When we write the tensor products we will omit ()F for the purpose of abbreviation. For
example
1 t3 ()F t3 ,
1 t3 1,
1 1 t3 ()F
(3.5)
t3
(3.6)
()
t3 ,
(3.7)
1 t3 1 ()F t3 1,
t3
t3
15
(3.8)
1 1 t3 1 1,
t3 t3 ()F t3 .
(3.9)
(3.10)
j
Generally speaking 1 1 1 ta 1 1 means

j
()j F ()j F ()j F ta 1 1.

With these notations we have
3

t 1 1 t3 = 1 t3 t3 1 = t3 t3 .
(3.11)
(3.12)
We also use the following abbreviations:

2
e1 e
(3.13)
= z1 t3 z2 t2 = zu1 zd2 t3 t2 ,

1
2
2
2
= e1 e
e
e1 ,
e1 e
(3.14)
2

1
e1 e1 = e1 e1 + e1 e1 .
(3.15)
2
When we write Casimir-like combinations of generators, we often omit the Lie algebra index:
1
t 1 t 3 = C
t t3 ,
t 3 t 1 = C t3 t1 ,
0
0
t 0 t 0 = C [][ ] t[]
t[
].
t 2 t 2 = C t2 t2 ,
(3.16)
We will also use this notation:

t i t j t k = fa b c C a a C b b C c c tai tb tck ,
(3.17)
where

fabc = fab c Cc c = Str [ta , tb ]tc .
(3.18)
For example:
0
t 3 t 1 t 0 = f[]
C
C C [][ ] t3 t1 t[
].
Using these notations we can write, for example

i

t t 4i 1, t j 1 t 4j = ()i+j +ij t (i+j ) mod 4 t 4i t 4j .
(3.19)
(3.20)
3.2. Short distance singularities using tensor product notations

Short distance singularities in the products of currents were calculated in [6,18]. Here is the
table in the tensor product notations

1
t 1 t 3 , + L ,
wu wd

w u w d 3 1
2
t 3 t 1 , + R
t t , R ,
J3+ J2+ =
wu wd
(wu wd )2

1
t 1 t 3 , + x ,
J1+ J1+ =
wu wd
J1+ J2+ =
16

2
w u w d 3 1
t 3 t 1 , + x
t t , x ,
2
w u wd
(wu wd )

1/2
1/2
t 0 t 0 , + R
t 0 t 0 , R ,
J0+ J1+ =
2
w u wd
(wu wd )

1/2
1/2
J0+ J3+ =
t 0 t 0 , + L
t 0 t 0 , L ,
2
w u wd
(wu wd )

1
t 1 t 3 , L ,
J1 J2+ =
w u wd

1
t 1 t 3 , L ,
J1+ J2 =
w u wd

1
t 3 t 1 , + R ,
J3 J2+ =
w u w d

1
t 3 t 1 , + R ,
J3+ J2 =
w u w d

1
t 1 t 3 , x ,
J1+ J1 =
w u wd

1
t 3 t 1 , + x .
J3+ J3 =
w u w d
J3+ J3+ =
Such tensor product notations are very useful and widely used in expressing the commutation
relations of transfer matrices. We will list the same formulas in more standard index notations in
Appendix A.3.
4. Calculation of
In this section we will give the details of the calculation which was outlined in Section 2.2.1.
4.1. The order of integrations
As we discussed in [6] the intermediate calculations depend on the choice of the order of
integrations. We will use the symmetric prescription. This means that if we have a multiple
integral, we will average over all possible orders of integration. For example in this picture:
we have three integrations, and therefore we average over 6 possible ways of taking the integrals.
Another prescription would give the same answer (because after regularization the multiple integral is convergent, and does not depend on the order of integrations), but will lead to a different
distribution of the divergences between the bulk and the boundary.
17
4.2. Contribution of triple collisions to

Triple collisions contribute to the comultiplication because of the double pole. Let us for
example consider this triple collision:
Of course this is not really a collision, since only the lower two points collide. But we still call it
a triple collision. This has to be compared to:
where the integrals are understood in the sense of taking the principal value. We have to average over two ways of integrating: (1) first integrating over the position of the zu2 d+ x on the
upper contour, and then zd2 d+ x on the lower contour and (2) first integrating over the position
of zd2 d+ x and then integrating over the position of zu2 d+ x. The first way of doing integrations does not contribute to , and the second does. Indeed, the contraction d+ x d+ x gives
1
z2 z2 t 2 t 2 , and after we integrate over wd we get:
(w w
)2 u d
u

u
Then integration over wu gives the imaginary contribution ( wdw
) = i:
u w
The contribution from the contractions d+ L d+ R is similar, and the result for the contribution
of triple collisions to is:

1
triple eam = i C+ C , 1 eam eam 1 ,
(4.1)
2
where 1/2 is because we average over two different orders of integration, and C is defined as

C+ = z1 t 3 z3 t 1 + z2 t 2 z2 t 2 + z3 t 1 z1 t 3 ,
(4.2)
3 3 1 2 2 2 2 1 3 3
C = z t zt + z t z t + zt z t .
(4.3)
18
The expression (4.1) for trpl should be added to dbl which is generated by the double collisions. We will now calculate dbl and = dbl + trpl .
4.3. Coupling of dx
We have just calculated the contribution of triple collisions; now we will discuss the contribution of double collisions and the issue of total derivatives.
Effect of double collisions:
Collision
J1+ J1+ :
J1 J1 :
J3+ J3 :
J3 J3 :
J3+ J3+ :
J1 J1+ :
J0 J2 :
contributes i times

zu3 zd3 t 1 t 3 , d+ x

+ 2zu zd t 1 t 3 , d x + zu zd t 1 t 3 , d+ x

+ 2zu1 zd3 t 3 t 1 , d+ x

+ zu3 zd3 t 3 t 1 , d x

2zu1 zd1 t 3 t 1 , d+ x zu1 zd1 t 3 t 1 , d x

2zu zd3 t 1 t 3 , d x

+ 32 zd2 zd2 dx, t 2 t 2 .
(4.4)
In the calculation of the contribution of J0 J2 we take an average of first taking an integral

over the position of J0 and then taking an integral over the position of J2 . To summarize:

1 dbl
(dx) = zu3 zd3 + zu zd t 1 t 3 , d+ x
i

+ zu1 zd3 + zu3 zd1 2zu1 zd1 t 3 t 1 , d+ x

+ zu zd3 zu3 zd + 2zu zd t 1 t 3 , d x

+ zu3 zd3 zu1 zd1 t 3 t 1 , d x

3
+ zu2 zu2 t 2 t 2 , dx .
2
Effect of triple collisions:

1 trpl
(dx) = C + C , 1 z2 d+ x + z2 d x
i

= zu3 zd3 zu zd t 1 t 3 , d+ x + zu1 zd5 zu3 zd1 t 3 t 1 , d+ x

+ zu2 zd4 zu2 t 2 t 2 , d+ x

+ zu3 zd zu zd5 t 1 t 3 , d x + zu1 zd1 zu3 zd3 t 3 t 1 , d x

+ zu2 zu2 zd4 t 2 t 2 , d x .
This leads to the following expression for the total :
2
2

1
1 2
(dx) =
zu zu2 + zd2 zd2 zu1 zd1 t 3 t 1 , d+ x
i
2
2
2

1 2
zu zu2 + zd2 zd2 zu zd t 1 t 3 , d x
2

+ zu2 zd4 zu2 t 2 t 2 , d+ x
(4.5)
19

+ zu2 zu2 zd4 t 2 t 2 , d x

3
+ zu2 zu2 t 2 t 2 , dx .
(4.6)
2
The calculations of this section can only fix the coupling of d x up to total derivatives, i.e., terms
proportional to dx = d+ x + d x. Only the terms proportional to dx = d+ x d x are fixed. To
fix the terms proportional to dx, we have to either study the couplings of x dx or look at what
happens at the endpoint of the contour. We will discuss this in Sections 5 and 6. The result it that
the following additional coupling:

1 2
z zu2 t 2 t 2 , dx ,
2 u
should be added to (4.6).
(4.7)
4.4. Coupling of dL
Similar to the dx terms, we can discuss the d coproduct.
Effect of double collisions. Here is the table:
Collision
J1+ J2+
J1 J2
J1 J2+
J1+ J2
J0 J3
contributes i times

2zu3 zd2 t 1 t 3 , d+ L

+ 2zu zd2 t 1 t 3 , d+ L + 4zu zd2 t 1 t 3 , d L

2zu zd2 t 1 t 3 , d L

2zu3 zd2 t 1 t 3 , d L

+ 32 z3 z1 t 3 t 1 , dL .
Contribution of triple collisions:

1 trpl
(dL ) = C + C , 1 z1 d+ L + z3 d L
i

= zu3 zd2 1 zu4 zd4 t 1 t 3 , d+ L + zu2 zd3 1 zu4 zd4 t 2 t 2 , d+ L

+ zu1 zd4 1 zu4 zd4 t 3 t 1 , d+ L

+ zu3 zd2 1 zu4 zd4 t 1 t 3 , d L + zu2 zd 1 zu4 zd4 t 2 t 2 , d L

+ zu1 1 zu4 zd4 t 3 t 1 , d L

= zu3 zd2 + zu2 zd3 1 zu4 zd4 t 1 t 3 , d+ L

+ zu1 zd4 1 zu4 zd4 t 3 t 1 , d+ L

+ zu3 zd2 + zu2 zd 1 zu4 zd4 t 1 t 3 , d L

+ zu1 1 zu4 zd4 t 3 t 1 , d L .
Just as in case of the couplings of dx, we observe that only the couplings proportional to d+ x
d x are fixed by the calculation in this section. In fact the analysis of Section 5 will show that
we have to add the following total derivative coupling:

(1/2) z3 z1 t 3 t 1 , dL .
(4.8)
20
Adding this to dbl + trpl we get

2
2

1
(dL ) = zu zd2 zd2 zd2 + zu2 zu2 t 1 t 3 , d L
i

+ 2zu3 zu1 zd4 zu3 t 3 t 1 , d L

2zu1 zu3 zu1 zd4 t 3 t 1 , d+ L .
(4.9)
4.5. The structure of

At the first order of perturbation theory = 0 + where 0 (t) = t 1 + 1 t is the trivial
coproduct. It follows from Sections 4.3 and 4.4 that is given by the following formula:
=

i
r, 0 ,
2
(4.10)
where
r=

(zu , zd ) 3 1
(zu , zd ) 0
zu zd t t 3 + zu3 zd t 3 t 1 + zu2 zd2 t 2 t 2 + 2
t t 0.
4
zu4 zd
zu4 zd4
(4.11)
We used the notations:

2
2

(zu , zd ) = zu2 zu2 + zd2 zd2 ,
(zu , zd ) = 1 + zu4 zd4 zu4 zd4 .
The following identities are useful in deriving (4.10):
3 1

zu zd t t 3 + zu2 zd2 t 2 t 2 + zu3 zd t 3 t 1 , zu1 t3 1 + 1 zd1 t3 = 2zu3 t 3 t 1 , t3 ,
0

t t 0 , zu1 t3 1 + 1 zd1 t3 = 2 z1 t 3 t 1 , t3 ,
3 1

zu zd t t 3 + zu2 zd2 t 2 t 2 + zu3 zd t 3 t 1 , zu2 t2 1 + 1 zd2 t2

(4.12)
= 2zu1 zd3 t 1 , t2 t 3 + 2zd2 t 2 , t2 t 2 .
Here denotes the symmetric tensor product; it is the opposite of . The minus sign in the last
= C . So in particular 2z3 t 3 {t 1 , t 3 } = (z3 t 3 ) {t 1 , t 3 }
line of (4.12) is because C
u
3
1
3
3
{t , t } (zd t ).
5. Generalized gauge transformations
5.1. Dress code
The coupling of fields to the generators of the algebra is strictly speaking not defined unambiguously, because of the possibility of a generalized gauge transformation
J f (d + J )f 1 ,
(5.1)
where f is a group-valued function of fields, depending on the spectral parameter z. A proper

gauge transformation would not depend on z and would belong to the Lie group of g0 , while
f in (5.1) belongs to the Lie group of g and does depend on z. Therefore it would perhaps be
appropriate to call (5.1) generalized gauge transformation or maybe change of dressing. If
there is some insertion A into the contour, then we should also transform A f Af 1 .
21
One of the reasons to discuss the transformations (5.1) is that different prescriptions for the
order of integrations are related to each other by such a change of dressing. A similar story
for log divergences was discussed in [6]. Different choices of the order of integration lead to
different distribution of the log divergences between the bulk and the boundary.
We agreed in Section 4.1 to use the symmetric prescription for the order of integrations. It
turns out that with this prescription limy0 T2 [C + y]T1 [C] comes out in the wrong dressing
in the sense that the limit cannot be immediately presented in the form

P exp J a (ta ) .
(5.2)
In particular x + x couples to a different algebraic expression than x x , while in (5.2) they
0 ). However, it turns out that it is possible to satisfy the dress code
should both couple to (t[]
(5.2) by the change of dressing of the type (5.1).
We will now stick to the symmetric prescription for the order of integrations and study the
asymmetry between the couplings of x d+ x and x d x, and the asymmetry between the couplings
of d+ and d . Then we will determine the generalized gauge transformation needed to
satisfy (5.2), and this will fix the total derivative couplings discussed in Section 4.3. It turns out
that in the symmetric prescription we will have to do the generalized gauge transformation (5.1)
with the parameter:

i 2
z z2 t 2 t 2 , dx + z1 z3 t 3 t 1 , dL
f =1
2

+ z3 z t 1 t 3 , dR + .
(5.3)
In Sections 5.2 and 5.3 we will show that the gauge transformation with this parameter indeed
removes the asymmetry. In Section 6.2 we will derive (5.3) using the invariance under the shift
symmetries.
5.2. Asymmetry between the coupling of x d+ x and x d x
5.2.1. Coupling proportional to zu4 x dx
The most obvious asymmetry is that there is a term with zu4 x d+ x but no term with zu4 x d x.
The term with zu4 x d+ x comes from this collision:
The result is

1
i z2 d+ x 1, z2 t 2 x, t 2 .
4
(5.4)
This is unwanted, so we want to do the generalized gauge transformation with the parameter
i 2 2 2
z t t ,x .
2
which removes this coupling and adds instead a total derivative coupling to dx

i 2 2 2
z t t , dx .
2
(5.5)
(5.6)
22
We will now argue that the change of dressing with the parameter (5.5) also removes the asymmetry between the coupling of x d+ x and x d x.
Also the coefficient of zu2 zd2 x d+ x is different from the coefficient of zu2 zd2 x d x. Let us
explain this.
5.2.2. Asymmetric couplings of the form zu2 zd2 x dx
There is a contribution from a double collision, and from a triple collision. The double collision is:
and we have to take into account the interaction vertex in the action
S
1
str[x, + x][x, x].
6
(5.7)
The calculation is in Section A.2, and the result is:

1
iC z2 t2 , [x, d x] z2 t2 .
2
(5.8)
There is also a triple collision:
It contributes:

1 2
i z [d+ x, x], t 2 z2 t 2 .
4
(5.9)
The sum of Eqs. (5.8) and (5.9) amounts to the following asymmetry of the form zu2 zd2 :

1
i z2 [d+ x, x], t 2 z2 t 2 .
4
(5.10)
We see that (5.4) + (5.10) is

1
z2 d+ x 1 + 1 z2 d+ x , i z2 t 2 x, t 2 .
2
This is undone with the generalized gauge transformation with the parameter
[x, t 2 ], which adds an additional total derivative coupling:

1 2 2
i z t dx, t 2 .
2
This is the additional coupling of Eq. (4.7).
(5.11)
1
2 2
2 i(z t )
(5.12)
23
5.3. Asymmetry in the couplings of d

The situations with the couplings of d is similar. There are asymmetric couplings of the
form zu4 L d+ R which are removed by the generalized gauge transformation. This generalized
gauge transformation should also remove the asymmetry in the couplings of zu2 zd2 d+ and
zu2 zd2 d , but we did not check this.
Terms of the form zu4 L d+ R come from
z3 d+ R z1 d+ L
1
1
2 [L ,dR ]+ 2 [R ,dL ]
. They are similar to (5.4)

3
1 1 3 1
z t t , L .
i z d+ R 1,
4

(5.13)
This should be removed with the generalized gauge transformation which simultaneously introduces the total derivative coupling:

i 1 3 1
z t t , dL .
2
(5.14)
This is the additional coupling of (4.8).

6. Boundary effects
6.1. The structure of G
6.1.1. Introducing the matrix s
Here we will derive Eq. (2.44) in Section 2.3.3. We inserted the switch operator on the upper
line, which turns zuin into zuout . Naively Eq. (4.10) implies that:
But this is wrong because there is an additional boundary contribution related to the second order
poles in the short distance singularities of the products of currents. Notice that these second order
poles correspond to the terms in the approach of [1315] (see Appendix B). At the first order
in the x-expansion the contributing diagram is this one:
and similar ones. This turns 1

s = C+ C ,
i
2 r
+ into 1
i
2 (r
+ s) + where
(6.1)
24
and C are given by (4.2) and (4.3). Therefore G+ of the switch operator is the following split
operator:

i in
r zu , zd + s zuin , zd
2

i out
r zu , zd + s zuout , zd + .
G+ (1switch ) = 1 +
(6.2)
6.1.2. Cancellation of field dependent terms

Dots in (6.2) denote the contribution of the higher orders of the string worldsheet perturbation
theory. Those are the terms of the order h 2 and higher. The terms with 12 (r + s) are of the order
h . Remember that we are also expanding in powers of elementary fields. It turns out that all the
terms of the order h (i.e., tree level) in G+ (1switch ) are c-number terms written in (6.2), there
are no corrections of the higher powers in x and . This is because such corrections would
contradict the invariance with respect to the global shifts (2.9). Indeed, suppose that R(1switch
1) contained x and . For example, suppose that there was a term linear in x, something like
xt t . Then the variation under the global shift (2.9) will be proportional to t t and there
is nothing to cancel it.6 This implies that R(1switch 1) is a c-number insertion, i.e., no fielddependent corrections to (2.45)(2.48).
6.2. Boundary effects and the global symmetry
We explained in Section 3 of [6] that the global shifts act on the capital currents by the
gauge transformations (normal gauge transformation, not generalized)
S .J = dhh1 + hJ h1
1
h = 1 R 2 [x, ] + .
2
(6.3)
Suppose that the outer contour is open-ended, then this is not invariant under the global shifts
The infinitesimal shift of this is equal to
6 If we inserted some operator O which is not gauge invariant, for example O = t 2 , the variation under the global

shift will give [t2 , [, x]]. This is linear in x, but x will contract with d+ x in (z2 J2+ d + + z2 J2 d ) resulting
in the x-independent expression of the form z2 t 2 [t 2 , ], which will cancel the -variation of the field dependent
terms (2.26).
25
Therefore because of this contraction:
We have the imaginary contribution:
Using the terminology from Section 2.3 we should say that F+ (1) is such that

1
S F+ (1) = i t 2 , zd2 zd2 t 2 .
(6.4)
2
There are similar considerations for the super-shifts. Therefore

1
1
F+ (1) = const + i x, t 2 zd2 zd2 t 2 + i L , t 1 zd1 zd3 t 3
2
2
3

1
3
1 1
+ i R , t zd zd t + .
(6.5)
2
The relation between this formula and the generalized gauge transformation with the parameter (5.5) is the following. Part of (6.5) comes from (5.5), and another part from the following
diagrams:
These two diagrams contribute

1
1
i x, t 2 zd2 zd2 t 2 + i zu2 zu2 t 2 x, t 2 .
(6.6)
4
4
And the generalized dressing transformation with the parameter (5.5) gives the boundary term
i 14 [x, t 2 ] (zd2 zd2 )t 2 i 14 (zu2 zu2 )t 2 [x, t 2 ] which in combination with (6.6) gives

1 2 2
x, t zd zd2 t 2 ,
(6.7)
2
which is in agreement with (6.5). Similar diagrams with fermions give terms with in (6.5).
Notice that the constant terms in (6.5) are essentially the same as in Section 6.1

i
F+ (1) = 1 +
(6.8)
r(zu , zd ) + s(zu , zd ) + .
2
The difference between F+ (1) and G+ (1switch ) is that G+ (1switch ) is a c-number while F+ (1)
is field-dependent. That is because 1switch is invariant under the gauge transformations, because
out
in
uz and uz are the same as representations of the finite dimensional g0 Lpsu(2, 2|4).
i
26
7. BRST transformations
Here we discuss the action of QBRST on the switch operators end verify that it commutes
They are both
with G+ . There are two BRST currents, holomorphic Q and antiholomorphic Q.
= 0. The total BRST operator is their sum
nilpotent and {Q, Q}
QBRST = Q + Q.
(7.1)
Here we will consider the holomorphic BRST operator Q. The action of Q on the currents is

(z)
Q, J (z) = D z1
(7.2)
(see for example Section 2 of [6]). In other words
1
1
Q, TAB (z) = (B)TAB (z) TAB (z) (A).
z
z
The switch operator turns zin into zout . We have

1
1
Q . 1switch = out in .
z
z
(7.3)
(7.4)
According to (2.44) the fusion of the switch operator on the upper contour is the split operator
i
out
in
2 (r+ + r+ ). Therefore
QG+ 1switch =

out
i 1
in
+ r+
zout ( 1) + z1 (1 ) r+
2
out
1

in
r+
zin ( 1) + z1 (1 ) .
+ r+
(7.5)
(7.6)
1
z1in ) is essentially the same
Now we have to calculate G+ Q1switch . The action of G+ on ( zout
as the action on the switch
1
in
i out 1
1
1
( 1) + zout
( 1) zin
( 1) r+
r+ zout ( 1) zin
2
plus the contribution of this diagram
This diagram contributes

1 in 1
1 zd4 t 3 t 1 , ,
zu
i zuout
(7.7)
where we have used the short distance singularity

(wu ) {w+ , }(wd ) =

1
t 3 t 1, .
w u wd
Therefore the condition [Q, G+ ] = 0 can be written as follows:

1
r+ (zuin , zd ) 1
1
r+ (zuout , zd ) 1
, out 1 + 1
, in 1 + 1
0=
2
zu
zd
2
zd
zu
(7.8)
1
zuout
1
zuin

1 3 1
t t , .
zd4
This can be verified using the identity:

r+ (zu , zd )
1
1
(zu ) 1 + (zd ) 1 ,
2

1
= 1 4 zu1 t 3 t 1 , zd3 , t 1 t 3 .
zd
27
(7.9)
(7.10)
Notice that this identity can be used to derive Eq. (2.47) for r+ . Also notice that the second term
on the right-hand side (1 14 )zd3 {, t 1 } t 3 is a gauge transformation and does not contribute
zd
because the switch operator is gauge invariant.

In this case the
As another example let us consider the upper Wilson line terminating on .
second term on the right-hand side of (7.10) does contribute

1 3 1
1 4 zd , t , t 3 .
(7.11)
zd

But this cancels with the contribution of the contraction of with the integral d (1 zd4 )N
arising in the expansion of the lower Wilson line:
Indeed, this contraction gives

t 1 1 zd4 t 3 ,
(7.12)
and therefore there is an additional contribution to [Q, G]:

1 3

t 1 1 zd4
, t , .
zd
(7.13)
Notice that (7.11) cancels with (7.13) because [{, t 1 }, ] t 3 t 1 [, {t 3 , }] = 0.

8. Generalized YBE
The r-matrix (4.11) does not satisfy the classical YangBaxter equation in its usual form,
but the deviation from zero is a polynomial in z1 , z2 , z3 , z11 , z21 , z31 . Using the notation of
Eq. (3.17)
[r12 , r13 ] + [r12 , r23 ] + [r13 , r23 ]

z2 z2
1
= t 0 t 2 t 2 z22 z32 z14 32 22 + 2 2 4
z2 z3 z2 z3 z1

4
1
1
1
3
3
2
3 2 3
+ t t t z1 z3 z2 +
z1 z32 z2 z15 z32 z2 z1 z36 z2 z1 z32 z25
28

z33 z2
1
3 4
+ t t t z2 z3 z1 + 3 +
z2 z3 z23 z3 z14
+ t 1 t 1 t 2 z1 z2 z36 z1 z25 z32 z15 z2 z32 + 4z1 z2 z32

0
+ permutations.
z13 z23 z32

(8.1)
We will now explain why (8.1) is not zero and what replaces the classical YangBaxter equation.
We will also derive a set of generalized YBE which we conjecture to be relevant in the quantum
theory.
The consistency conditions follow from considering the different ways of exchanging the
product of three Wilson lines with insertions. We first consider the case of gauge invariant insertions; in this case the R-matrices are c-numbers. Then we will consider the case of non-gaugeinvariant insertions, namely loose endpoints. In this case the R-matrices are field-dependent, and
the generalized YangBaxter equations are of the dynamical type.
Fig. 5. Generalized YBE 1.
29
Fig. 6. Generalized YBE 2.
8.1. Generalized quantum YBE

To understand the quantum consistency conditions for the R-matrices let us put the Wilson
line with the spectral parameter switch on top of two other Wilson lines, the other two Wilson
lines having no operator insertions. The equations of this section will not change if we put a
constant gauge invariant operator at the point on the upper contour where we switch the spectral
parameter (instead of just 1). For example, C t2 t2 is a constant gauge invariant operator. It is
gauge invariant because commutes with g0 .
The generalized quantum YangBaxter equations (qYBE) are obtained from the exchanges
1
1
illustrated in Fig. 5. The notations are: = R+ , = R+
, = R , = R
. The insertion of the
spectral parameter changing operator is marked by a black bar.
Equating LHS and RHS in Fig. 5 yields
1
1
1
1
1
R23, R13,+ R23,
R23,+ R12,+ R23,+
R23, R13,+
R23,
R23, R13,+ R23,
1
R12,+ .
= R12,+ R13,+ R12,+
(8.2)
30
After cancellations of RR 1
1
1
R23, R13,+ R23,
R23,+ R12,+ R23,+
= R12,+ R13,+ .
(8.3)
Here the sign

= means that the ratio of the left-hand side and the right-hand side commutes
with O
1
1
1
1
R12,+
R23, R13,+ R23,
R23,+ R12,+ R23,+
= T123 ,
R13,+
(8.4)
1
= O1 .
T123 O1 T123
(8.5)
at the first order of perturbation theory the left-hand side of (8.4) is, cf. (2.47) and (2.48),
[r23, , r13,+ ] + [r13,+ , r12,+ ] + [r23,+ , r12,+ ] =
and the right-hand side of (8.4) is

= 4t 0 z22 z22 t 2 z32 z32 t 0 , t 2

4t 0 z2 z23 t 1 z33 z31 t 0 , t 3

4t 0 z23 z21 t 3 z3 z33 t 0 , t 1
(8.6)
(8.7)
0
(the indices of t 0 contract with the indices of another t 0 , so t 0 t 0 stands for C [1 1 ][2 2 ] t[
1 1 ]
0
2
2
1
t[2 2 ] ; similarly the indices of t contract with the indices of another t , and t with t 3 ). Our
O is just the spectral parameter switch, it is a constant gauge invariant operator. In particular,
[t 0 , O] = 0, cf. (8.5). In other words, Eqs. (8.6), (8.7) is the generalized classical YangBaxter
equation modulo gauge transformation.
Similarly, putting the switch operator on the lower contour (see Fig. 5) we get the following
consistency condition:
1
1
R23, R12, R23,
= R12, R13, .
R23,+ R13, R23,+
(8.8)
Finally we turn to the exchange of R+ and R , which is derived in Fig. 6. Equating the LHS
and RHS of this graph, we obtain

1
1
1 1
R23 R13+ R23
R23+ R12 R23+
R23 R13
R23 R23+

1
1
1 1
(8.9)
= R12+ R13 R12+ R12 R23+ R12 R12+ R13
R12+ R12 .
Note that we equate of course only one side of the insertion at a time. Note also, that for R+ = R
this returns to a standard YBE. Another way of writing it
1 1

1
R12 R23+
= adR12+ (R13 )adR12 (R23+ )adR12+ R13
1

.
adR23 (R13+ )adR23+ (R12 )adR23 R13
(8.10)
In [19,20] another generalization of quantum YBE was proposed as the quantum version of a
more restricted set of classical YBE. The main difference to the equations here is that the ones in
[19,20] impose the standard qYBE on R (and thus the standard YBE on the classical r-matrix)
and supplement these by equations of the type RSS = SSR. However the main problem with
this approach is that the case of principal chiral models and strings on AdS5 S 5 do not fall in
the class of models where r satisfies the YBE separately from s.
31
8.2. Some speculations on charges

Strictly speaking our derivation of equations like (8.3) only applies to the terms quadratic in r
(i.e., tree level). Although the derivation outlined in Figs. 5 and 6 seems to apply also at the level
of higher loops, in fact there might be subtleties associated to overlapping diagrams involving all
three lines.
Nevertheless, let us for a moment take the proposed generalized qYBE (8.3) seriously and see
how it could be put to use in order to construct a quadratic algebra of RT T type. The relation
(8.3) can be thought of in the following way. Let us begin with the standard YBE, which reads
R12 R13 R23 = R23 R13 R12 .
(8.11)
This can formally be thought of as R12 and R13 commute up to conjugation by R23 , or explicitly

1
1
R12 R13 = R23 R13 R23
(8.12)
R23 R12 R23
.
The relation (8.3) generalizes this version of the YBE naturally, in that

1
1
R12,+ R13,+ = R23, R13,+ R23,
R23,+ R12,+ R23,+
.
(8.13)
If we interpret this as RT T relations, we obtain

1
1
T12,+ T13,+ = R23, T13,+ R23,
R23,+ T12,+ R23,+
.
(8.14)
Naively one might then conclude that this equation is in fact of the type that has been discussed
in [19], Eq. (14)
A12 T1 B12 T2 = T2 C12 T1 D12 ,
(8.15)
where T1 = T12,+ and T2 = T13,+ and

1
A12 = R23,
,
B12 = 1,
1
C12 = R23,
R23,+ ,
1
D12 = R23,+
.
(8.16)
However, in [19] it is required that the matrices A, B, C, D satisfy a set of equations, in particular
A and D have to separately satisfy the standard YBE, as well as equations of the type ACC =
CCA and DCC = CCD as well as [A12 , C13 ] = 0 and [D12 , C32 ] = 0 have to hold (note that it
is pointed out in [19] that these are only sufficient conditions). We do not require these equations,
but only seem to be imposing Eq. (8.3). This is in fact a much weaker equation, but has the vital
advantage that it gives as a classical limit and the algebra of r-, s-matrices as we require it.
In view of the algebra (8.13) the standard argument of construction of commuting charges
does not go through, namely [tr2 (T12,+ ), tr3 (T13,+ )] is not obviously vanishing, as the conjugation in this case is by R23,+ and R23, , respectively, which do not agree in the present case. At
this point a construction that appears in [19] is useful, despite the fact that their transfer matrix
algebra is different from ours. First let us simplify notation and suppress the physical space index
of the T -matrices, so we consider the exchange relation of T2 and T3 . Ti is an element of End
(a ) a a (at least for finite dimensional representations). Thus we can label them by T(a,a)
,
where a denotes the dual representation. The generalized RTT relations then become
1
1
=: R(3,3)(2,
T(2,2)
T(3,3)
= R23 T(3,3)
R23
R23+
T(3,3)
T(2,2)
,
2)
R23+ T(2,2)
(8.17)
where Rab acts on the a part of T(a,a)

, etc., and we defined
1
1
R(3,3)(2,
= R23 R23
2)
R23+ R23+
.
(8.18)
32
We require that R satisfies the YBE, in order for the exchange algebra of T(a,a)
to be consistent.
At this point the deviation from the construction in [19] is necessary. Our Rab matrices obey
the generalized YBE (8.3) and the complete set of consistency conditions on R should imply
YBE for R. This requires in particular additional relations for R12 and R21 . Once the YBE
for R are established, we define the dual RTT algebra as
T(2,2)
T(3,3)
R(3,3)(2,
= T(3,3)
T(2,2)
.
2)
(8.19)
Consider a matrix representation (scalar matrix) of (8.19) given by (2,2)

and (3,3)
. There is a
natural inner product between the representations and their duals, in particular a a .Ta a . Thus
acting with 22 33 on the generalized YBE in the form (8.17) we obtain
(22 . T22 )(33 . T33 ) = (33 . T33 )(22 . T22 ),
(8.20)
and thus

(22 . T22 ), (33 . T33 ) = 0.
(8.21)
This allows for construction of a family of infinite commuting charges by expanding these expressions in powers of the spectral parameter.
8.3. Contours with loose endpoints
The consistency condition for the exchange of contours with endpoints is more complicated.
Again, we can compare (123) (213) (231) (321) to (123) (132) (312) (321).
When we exchange (123) (213) we get the insertion of the split operator F1 (F+ (1)):
At the first order of perturbation theory

F1 F+ (1) = 1 + r + s + q,
(8.22)
where q are field-dependent (= dynamical) terms. Indeed, the main difference between the exchange of the switch operator and the exchange of the endpoint is that the endpoint is not gauge
invariant and therefore the exchange matrix is field dependent. The expansion of q in powers of
x and starts with:
q=

1 2
x, t t 2 + 3 , t 1 t 3 + 1 , t 3 t 1 + .
2
(8.23)
33
Then, when we exchange (213) (231) (321) we get additional contributions coming from
the contraction of q12 with the currents integrated over line 3, for example
(8.24)
On the other hand, if we look at the field independent (leading) terms, we will get an equation
identical to (8.4), but now T123 does not act as the identity on the endpoint, because the endpoint
is not gauge invariant. But in fact the T on the right-hand side of (8.4) cancels with the terms
arising from the contractions (8.24).
To summarize, we have the following two types of consistency conditions:
1. Consistency conditions for the exchange of gauge invariant operators. In this case the righthand side of (8.4) does not spoil the consistency because of Eq. (8.5), which expresses the
gauge invariance of the inserted operators.
2. Consistency conditions for the insertions which are not gauge invariant. In this case the
right-hand side of (8.4) cancels against the terms arising from the diagrams like (8.24).
9. Conclusions and discussion
We have setup a formalism in which to compute the product of two transfer matrices, using
the operator algebra of the currents. In particular, to leading order in the expansion around flat
spacetime, a structure reminiscent of classical r-matrices appears. This is however modified in
that we require a system of r- and s-matrices, which satisfy a generalized classical YBE. This is
related in the approach of [15] to Poisson brackets being non-ultralocal.
We consider it a first step towards constructing the analog of a quantum R-matrix, which satisfies a generalized quantum YBE. The situation is different from [19,20], because the classical
r-matrix in our case does not satisfy the standard classical YBE (which is one of the assumptions
that goes into the construction in [19,20]) but the combined equation for r and s (8.6).
The most promising direction to extend this work is to construct the quantum conserved
charges from the T -matrices, as outlined in Section 8.2. It would also be interesting to test the
generalized quantum YBE explicitly at higher orders in the 1/R expansion.
It would also be interesting to understand how the r-, s-matrices found here relate to the
classical su(2|2) r-matrices found from the light-cone string theory and super-YangMill dual in
[2123]. The connection, if it exists, would presumably be along the line of our speculations in
Section 8.2.
Acknowledgements
We thank Jean-Michel Maillet for very interesting discussions. The research of A.M. is supported by the Sherman Fairchild Fellowship and in part by the RFBR Grant No. 06-02-17383
and in part by the Russian Grant for the support of the scientific schools NSh-8065.2006.2. The
research of S.S.N. is supported by a John A. McCone Postdoctoral Fellowship of Caltech. We
thank the Isaac Newton Institute, Cambridge, for generous hospitality during the completion of
this work.
34
Appendix A. Calculation of the products of currents

Here we will describe some methods for calculating the singularities in the product of two
currents. We will only discuss two examples. The first example is the collision J3+ J3+ and the
collision J1+ J2+ . The second is the singularities proportional to x dx in the collision J2+ J2+ ,
which we needed in Section 5.2.2.
A.1. Collisions J3+ J3+ and J1+ J2+
Collision J3+ J3+ .

+ L + [R , + x] (wa ) + L + [R , + x] (wb ).
(A.1)
The cubic vertex ([L , L ]+ x) does not contribute to the singularity, but the other cubic
vertex does

1
1
d 2 v str [R , + R ] x .
S
(A.2)
2
After integration by parts the interaction vertex becomes

1
1
(A.3)
d 2 v str [ R , + R ]x + str[R , + R ]x .
2
Integrating by parts + in the second term we get

1
1
2
d v str [ R , + R ]x + str[R , R ]+ x .
(A.4)
2
This implies that (A.1) gives the same singularity as the following collision in the free theory:

1
+ L [+ R , x] + [R , + x] (wa )
2

1
(A.5)
+ L [+ R , x] + [R , + x] (wb ).
2
The singularity is
1

1
t , x(wa ) t 3 + t 3 t 1 , x(wb )
2
(wa wb )
1

1
1
t , + x(wa ) t 3 t 3 t 1 , + x(wb ) .
+
2 w a wb
(A.6)
This is equal to
1

w a w b 1
2
t , + x t 3 +
t , x t 3 .
2
wa wb
(wa wb )
Collision J1+ J2+ .

+ R + [L , + x] + x + 1/2[L , + L ] + .
We have to take into account the interaction vertex in the action

1
1
d 2 v str [L , L ]+ x .
S
(A.7)
(A.8)
(A.9)
35
It is convenient to denote the contracted fields by using prime. For example, this notation

1
(A.10)
str [L , L ]+ x ,
2
means that L is contracted with the + R in J1+ , and + x with + x in J2+ . Therefore L
remains uncontracted. There is another possible contraction

1
str [L , L ]+ x .
2
In the interaction vertex (A.10) let us integrate by parts . We will get
1

1
str [ L , L ]+ x str [L , L ] + x .
2
2
In the second expression let us integrate by parts + . The result is
1
1

1
str [ L , L ]+ x + str [+ L , L ] x + str [L , + L ] x .
2
2
2
The first term coincides with (A.11), and together with (A.11) gives

str [ L , L ]+ x .
(A.11)
(A.12)
(A.13)
(A.14)
This is easy to contract, and precisely cancels the direct hit [L , + x] + x. The second and
third terms combine with the direct hit
+ R 1/2[L , + L ]
(A.15)
to give the same contribution as the collision

+ R [L , + L ],
(A.16)
which gives:
+ L
f
.
wa wb
(A.17)
A.2. Terms x dx in the collision J2+ J2+

Consider this collision:
More explicitly, we are looking at

1
1
+ x + x, [x, + x] (wu ) + x + x, [x, + x] (wd ).

6
6
(A.18)
Couplings to x dx receive contributions from the quartic interaction vertex

1
(A.19)
str[x, + x][x, x].
6
We denote the contracted fields x (wu ) and x (wd ). When x in the interaction vertex gets
contracted with + x in one of the J2+ , this contribution cancels the direct hit + x
S
36
1
6 [x, [x, + x]].
Let us study the diagrams in which x in the interaction vertex remains uncontracted. There are the following possibilities:

1
d 2 v str 2[x , + x ][x, x]
(A.20)
6
+ [x, + x ][x , x]
(A.21)

+ [x, + x ][x , x] .
(A.22)
Here prime and double prime mark the contracted elementary fields; for example in the first term
2[x , + x ][x, x] the elementary field x contracts with + x(wu ) in zu2 J2+ (wu ) and + x
contracts with + x(wd ) in zd2 J2+ (wd ); while [x, x] remains uncontracted. This gives

1
(2)
2
d v str
C t2 , [x, x] t2
2
6
(v wu )(v wd )

(1)
C x, t2 , x t2
(v wu )2 (v wd )

2

(1)
2
C
t
x,
t
,

(v wu )(v wd )2

1
1
C t2 , [x, x] t2
d 2 v str
=
(A.23)
2
2
(v wu )(v wd )

1 w d w u
=
C t2 , [x, x] t2 .
(A.24)
2
2 (wd wu )
(A simple way to get the singularity of this integral is by considering
the current-generator coupling

1
iC z2 t2 , [x, x] z2 t2 .
2
w u .)
This contributes to
(A.25)
Taking into account that C [t2 , t 0 ] t2 = C t2 [t2 , t 0 ] we can rewrite (A.25) using the
-product:

1
iC z2 t2 , [x, x] z2 t2 .
2
A.3. Short distance singularities using index notations
In the main text we gave the expressions for the short distance singularities in the tensor
product notations. Here we list the singularities using more conservative index notations
1 L
f
,
R 3 w 1 w2
1 L
(w1 )J2 (w2 ) = 3

f
,
J1+
R w 1 w2
J1
(w1 )J2+ (w2 ) =
(w1 )J2+ (w2 ) =

J3
(w1 )J2 (w2 ) =

J3+
(A.26)
(A.27)
1 + R
f
,
R 3 w 1 w 2
(A.28)
1 + R
f ,
R 3 w 1 w 2
(A.29)
1 x
f ,
R 3 w a wb
1 + x
(w)J3 (0) = 3
f ,
J3+
R w a w b
J1+
(w)J1 (0) =
(w1 )J2+ (w2 ) =

J1+
1 + L
1
f + O
,
3
w
w
R 1
R4
2
37
(A.30)
(A.31)
(A.32)
2 + R
f
R 3 w 3 w2

1 w 3 w 2
1
+ 3
f
+
O
,
R
2
R (w3 w2 )
R4

1 + x
1
f + O
,
J1+ (wa )J1+ (wb ) = 3
R w a wb
R4
J3+
(w3 )J2+ (w2 ) =
2 + x
f
R 3 w a wb

1 w a w b
1
3
x f + O
,
2
R (wa wb )
R4

1
1
1
(w1 )J3+
(w3 ) = 2
C
+
O
,
J1+
2
R (w1 w3 )
R4

1
1
1
C +O
,
J2+ (wm )J2+ (wn ) = 2
R (wm wn )2
R4
R (w0 )
+ R (w0 )
1
1
[]
[]
(w1 ) = 3
+
+
O
f
,
J0+ (w0 )J1+
2R (w0 w1 )2 (w0 w1 )
R4
L (w0 )
+ L (w0 )
1
1
[]
[]
+
+O
f
,
J0+ (w0 )J3+ (w3 ) = 3
2R (w0 w3 )2 (w0 w3 )
R4
1
+ x (w0 )
1
x (w0 )
[]
[]
(w2 ) = 3
+
+
O
f
.
J0+ (w0 )J2+
2
(w0 w2 )
2R (w0 w2 )
R4
(A.33)
(A.34)
J3+
(wa )J3+ (wb ) =
(A.35)
(A.36)
(A.37)
(A.38)
(A.39)
(A.40)
Appendix B. Very brief summary of the Maillet formalism

Let us briefly review the situation in Maillet et al.s work and how this connects to our present
analysis. In [15] a formalism was developed which generalizes the classical YBE to incorporate
the case of non-ultralocal Poisson brackets. Consider the algebra of L-matrices (spatial component of the Lax operator)

L(1 , z1 ), L(2 , z2 ) = r(1 , z1 , z2 ), 1 L(1 , z1 ) + L(1 , z1 ) 1 (1 2 )

+ s(1 , z1 , z2 ), 1 L(1 , z1 ) L(1 , z1 ) 1 (1 2 )

s(1 , z1 , z2 ) + s(2 , z1 , z2 ) (1 2 ).
(B.1)
The terms proportional to are the so-called non-ultralocal terms. The algebra (B.1) is a deformation of the standard ultralocal one by terms depending on the matrix s, which unlike r is
38
symmetric. Jacobi-identity for {,} yields a generalized, dyamical YBE7

[r12 s12 , r13 + s13 ] + [r12 + s12 , r23 + s23 ] + [r13 + s13 , r23 + s23 ]
(r+s)
(r+s)
+ H1,23
H2,13
= 0.
(B.2)
The dynamicity is due to the terms Hi,j k , which arise if r + s is field dependent and are defined
by

(r+s)
L(1 , z1 ) 1 1, 1 (r + s)23 (2 , z2 , z3 ) = H1,23
(1 , z1 , z2 , z3 )(1 2 ). (B.3)
In the case of s = 0 and r constant (field-independent) the relation (B.2) reduces to the standard
classical YBE. This formulation was applied to the O(n) model [15] and the complex sineGordon model [14] (where in both cases the r-, s-matrices are dynamical), as well as the principal
chiral field [13], in which case the terms Hi,j k vanish. Note that the field-dependence of the r-,
s-matrices seems to be due to the field-dependence of the non-ultralocal term.
References
[1] N. Beisert, R. Hernandez, E. Lopez, A crossing-symmetric phase for AdS5 S 5 strings, JHEP 0611 (2006) 070,
hep-th/0609044.
[2] N. Beisert, B. Eden, M. Staudacher, Transcendentality and crossing, J. Stat. Mech. 0701 (2007) P021, hep-th/
0610251.
[3] V.V. Bazhanov, S.L. Lukyanov, A.B. Zamolodchikov, Integrable structure of conformal field theory. III: The Yang
Baxter relation, Commun. Math. Phys. 200 (1999) 297324, hep-th/9805008.
[4] C. Bachas, M. Gaberdiel, Loop operators and the Kondo problem, JHEP 0411 (2004) 065, hep-th/0411067.
[5] A. Alekseev, S. Monnier, Quantization of Wilson loops in WessZuminoWitten models, hep-th/0702174.
[6] A. Mikhailov, S. Schfer-Nameki, Perturbative study of the transfer matrix on the string worldsheet in AdS5 S 5 ,
arXiv: 0706.1525 [hep-th].
[7] A. Das, J. Maharana, A. Melikyan, M. Sato, The algebra of transition matrices for the AdS5 S 5 superstring,
JHEP 0412 (2004) 055, hep-th/0411200.
[8] A. Das, A. Melikyan, M. Sato, The algebra of flat currents for the string on AdS5 S 5 in the light-cone gauge,
JHEP 0511 (2005) 015, hep-th/0508183.
[9] M. Bianchi, J. Kluson, Current algebra of the pure spinor superstring in AdS5 S 5 , JHEP 0608 (2006) 030, hepth/0606188.
[10] N. Dorey, B. Vicedo, A symplectic structure for string theory on integrable backgrounds, JHEP 0703 (2007) 045,
hep-th/0606287.
[11] A. Mikhailov, Bihamiltonian structure of the classical superstring in AdS5 S 5 , hep-th/0609108.
[12] J. Kluson, Reduced sigma-model on O(N ): Hamiltonian analysis and Poisson bracket of Lax connection,
JHEP 0709 (2007) 100, arXiv: 0707.3264 [hep-th].
[13] J.M. Maillet, Hamiltonian structures for integrable classical theories from graded KacMoody algebras, Phys. Lett.
B 167 (1986) 401.
[14] J.M. Maillet, New integrable canonical structures in two-dimensional models, Nucl. Phys. B 269 (1986) 54.
[15] J.M. Maillet, KacMoody algebra and extended YangBaxter relations in the O(N ) nonlinear sigma model, Phys.
Lett. B 162 (1985) 137.
[16] J. Kluson, Note about classical dynamics of pure spinor string on AdS5 S 5 background, Eur. Phys. J. C 50 (2007)
10191030, hep-th/0603228.
[17] A.A. Rosly, K.G. Selivanov, On amplitudes in self-dual sector of YangMills theory, Phys. Lett. B 399 (1997)
135140, hep-th/9611101.
[18] V.G.M. Puletti, Operator product expansion for pure spinor superstring on AdS5 S 5 , JHEP 0610 (2006) 057,
hep-th/0607076.
[19] L. Freidel, J.M. Maillet, Quadratic algebras and integrable systems, Phys. Lett. B 262 (1991) 278284.
7 Note that the signs are slightly different in [15] from the equations we will be using.
39
[20] L. Freidel, J.M. Maillet, On classical and quantum integrable field theories associated to KacMoody current algebras, Phys. Lett. B 263 (1991) 403410.
[21] A. Torrielli, Classical r-matrix of the su(2|2) SYM spin-chain, Phys. Rev. D 75 (2007) 105020, hep-th/0701281.
[22] S. Moriyama, A. Torrielli, A Yangian double for the AdS/CFT classical r-matrix, JHEP 0706 (2007) 083, arXiv:
0706.0884 [hep-th].
[23] N. Beisert, F. Spill, The classical r-matrix of AdS/CFT and its Lie bialgebra structure, arXiv: 0708.1762 [hep-th].

Logarithmically enhanced corrections to the decay rate

and forwardbackward asymmetry in B Xs +
Tobias Huber a,b , Tobias Hurth c,d,,1 , Enrico Lunghi e
a Institute for Theoretical Physics, University of Zurich, CH-8057, Zurich, Switzerland
b Institut fr Theoretische Physik E, RWTH Aachen, D-52056 Aachen, Germany
c CERN, Department of Physics, Theory Division, CH-1211 Geneva, Switzerland
d SLAC, Stanford University, Stanford, CA 94309, USA
e Fermi National Accelerator Laboratory, PO Box 500, Batavia, IL 60510, USA
Received 21 January 2008; received in revised form 28 February 2008; accepted 28 April 2008
Abstract
We study logarithmically enhanced electromagnetic corrections to the decay rate in the high dilepton
invariant mass region as well as corrections to the forwardbackward asymmetry (FBA) of the inclusive rare
decay B Xs + . As expected, the relative effect of these corrections in the high dilepton mass region
is around 8% for the muonic final state and therefore much larger than in the low dilepton mass region.
We also present a complete phenomenological analysis, to improved NNLO accuracy, of the dilepton mass
spectrum and the FBA integrated in the low dilepton mass region, including a new approach to the zero
of the FBA. The latter represents one of the most precise predictions in flavour physics with a theoretical
uncertainty of order 5%. We find (q02 ) = (3.50 0.12) GeV2 . For the high dilepton invariant mass
7
region, we have B(B Xs )high = (2.40+0.69
0.62 ) 10 . The dominant uncertainty is due to the 1/mb
corrections and can be significantly reduced in the future. For the low dilepton invariant mass region, we
confirm previous results up to small corrections.
Keywords: B physics; Rare decays
* Corresponding author at: CERN, Department of Physics, Theory Division, CH-1211 Geneva, Switzerland.
E-mail address: tobias.hurth@cern.ch (T. Hurth).

1 Heisenberg Fellow.
doi:10.1016/j.nuclphysb.2008.04.028
T. Huber et al. / Nuclear Physics B 802 (2008) 4062
41
1. Introduction
At the beginning of the LHC era, the search for physics beyond the Standard Model (SM)
is the main focus of particle physics. In principle, there are two ways to search for possible
new degrees of freedom. At the high-energy frontier one tries to produce those new degrees of
freedom directly, while at the high-precision frontier one analyses the indirect virtual effects of
such new particles. It is a matter of fact that high-precision measurements allow to analyse new
physics scales presently not accessible at direct collider experiments.
There are theoretical arguments like the hierarchy problem which let us expect new physics
at the electroweak scale. However, the indirect constraints on new physics by the present flavour
data indicate a much higher new physics scale when such new effects are naturally parametrized
by higher-dimensional operators. Thus, if there is new physics at the electroweak scale, then
its flavour structure has to be highly non-trivial and the experimental measurement of flavourviolating couplings is mandatory. This flavour problem has to be solved by any new physics
scenario at the electroweak scale. Moreover, the present electroweak data also indicate a slightly
higher new physics scale (little hierarchy problem).
Among inclusive flavour-changing neutral current (FCNC) processes (for a review see [1,2]),
the inclusive B Xs + decay presents an important test of the SM, complementary to the
inclusive B Xs decay. It is particularly attractive because of kinematic observables such
as the dilepton invariant mass spectrum and the forwardbackward asymmetry (FBA). These
observables are dominated by perturbative contributions if the cc resonances that show up as
large peaks in the dilepton invariant mass spectrum are removed by appropriate kinematic cuts.
In the so-called perturbative q 2 -windows, namely in the low-dilepton-mass region 1 GeV2 <
q 2 = m2 < 6 GeV2 , and also in the high-dilepton-mass region with q 2 > 14.4 GeV2 , theoretical
predictions for the invariant mass spectrum are dominated by the perturbative contributions, and
a theoretical precision of order 10% is in principle possible.
The integrated branching ratio has been measured by both Belle [3] and BaBar [4] based on
a sample of 152 106 and 89 106 B B events respectively. In the low-dilepton invariant mass
region, 1 GeV2 < q 2 < 6 GeV2 , the experimental results read
6 (Belle),
(1.493 0.504+0.411
0.321 ) 10

+
B B Xs low = (1.8 0.7 0.5) 106

(1)
(BaBar),
6
(average).
(1.60 0.50) 10
Measurements for the high-q 2 region, 14.4 GeV2 < q 2 < 25 GeV2 , are also available:
+0.061
6 (Belle),
(0.418 0.1170.068 ) 10

+
+0.08
6
B B Xs high = (0.5 0.250.07 ) 10

(2)
(BaBar),
(0.44 0.12) 106

(average).
By the end of the present B factories an experimental accuracy of 15% is finally expected.
The recently calculated NNLL QCD contributions [514] have significantly improved the
sensitivity of the inclusive B Xs + decay in testing extensions of the SM in the sector of
flavour dynamics, in particular, the value of the dilepton invariant mass q02 for which the differential FBA vanishes is one of the most precise predictions in flavour physics with a theoretical
uncertainty of order 5%. This well corresponds to the expected experimental sensitivity of 46%
at the proposed Super-B factories [1517].
Also non-perturbative corrections scaling with 1/m2b , 1/m3b , or 1/m2c [1824] have to be
taken into account. Moreover, factorizable long-distance contributions away from the resonance
42
peaks are important; here using the KrgerSehgal approach [25] avoids the problem of doublecounting.
In the high-q 2 region, one encounters the breakdown of the heavy-mass expansion at the endpoint; while the partonic contribution vanishes in the end-point, the 1/m2b and 1/m3b corrections
tend towards a non-zero value. In contrast to the endpoint region of the photon energy spectrum
in the B Xs decay, no partial all-order resummation into a shape function is possible here.
2
However, for an integrated
high-q spectrum an effective expansion is found in inverse powers of meff
b = mb (1 smin ) rather than mb [26,27]. The expansion converges less rapidly,
depending on the lower dilepton mass cut smin [11]. Recently it was suggested [24] that the
large theoretical uncertainties can be reduced by normalizing the B Xs + decay rate to the
semileptonic B Xu decay rate with the same q 2 cut.
A hadronic invariant-mass cut is imposed in the present experiments (BaBar: mX < 1.8 GeV,
Belle: mX < 2.0 GeV) in order to eliminate the background such as b c( se+ )e =
b se+ e + missing energy. The high-dilepton mass region is not affected by this cut, but in
the low-dilepton mass region the kinematics with a jet-like Xs and m2X mb QCD implies the
relevance of the shape function. A recent SCET analysis shows that using the universality of
jet and shape functions the 1030% reduction of the dilepton mass spectrum can be accurately
computed using the B Xs shape function. Nevertheless effects of subleading shape functions
lead to an additional uncertainty of 5%. It is also shown that the zero of the FBA is not affected by
the cut [28,29]. We emphasize that, as in all previous phenomenological analyses of the inclusive
B Xs + mode, the predictions in this paper are given under the assumption that there is no
cut in the hadronic mass region.
Finally, as in the B Xs case, there are unknown subleading non-perturbative corrections
of order O(s /mb ) which may be estimated by an additional uncertainty of order 5%.
Recently, further refinements were presented such as the NLO QED two-loop corrections to
the Wilson coefficients whose size is of order 2% [12]. Furthermore, it was shown that in the QED
one-loop corrections to matrix elements large collinear logarithms of the form log(m2b /m2lepton )
survive integration if only a restricted part of the dilepton mass spectrum is considered. This adds
another contribution of order +2% in the low-q 2 region for B(B Xs + ) [30]. For B(B
Xs e+ e ), in the current BaBar and Belle setups, the logarithm of the lepton mass gets replaced
by angular-cut parameters and the integrated branching ratio for the electrons is expected to be
close to that for the muons.
In the present manuscript, we calculate these corrections due to large collinear logarithms also
for the dilepton mass spectrum in the high-q 2 window and for the FBA for the first time. In the
high-q 2 region these corrections are much larger than in the low-q 2 one; we already anticipate
that in the muon channel the former account for 8% of the rate while the latter are only about
+2%. We also update the theoretical prediction of the observables using several improvements
offered recently in the literature among which the two-loop matrix element of the semileptonic
operator P9 and the 1/m3b corrections are the most important to date. We present a new phenomenological analysis of the dilepton mass spectrum in the low- and high-q 2 region and of the
FBA. In particular, a new approach to the zero of the FBA is proposed. Moreover, we explore a
strategy to reduce the uncertainties due to the 1/mb corrections by normalizing with the semileptonic B Xu decay rate with the same q 2 cut, as was recently proposed in the literature [24].
We already anticipate that with this normalisation the dilepton mass spectrum, integrated over the
high-q 2 region, receives only a 3% shift in the central value by the non-perturbative corrections.
Finally, we derive model-independent formulae for non-SM values of the high-scale Wilson
coefficients.
43
The present manuscript is organised as follows. In Section 2 we present the master formulae
for the calculation of the observables; in particular we discuss the normalization of the branching
ratio and forwardbackward asymmetry and explain the details of the perturbative expansion. In
Section 3 we present the novel analytical results on the logarithmically enhanced QED corrections to the forwardbackward asymmetry and to the decay width in the high-q 2 region and also
discuss some non-perturbative subtleties. Furthermore, we analyse the relation of the collinear
logarithms and the experimental angular cuts. In Section 4 we present the SM predictions for the
branching ratio and forwardbackward asymmetry. In Section 5 we draw our conclusions.
2. Master formulae
In this section, we present master formulae for the various observables. We closely follow
here the notations of Ref. [30] in order to keep the paper compact. The new results are explicitly
derived in the next section.
2.1. Dilepton invariant mass spectrum
Following Ref. [30], the master formula for the B Xs + branching ratio reads

Vts Vtb 2 4 (s )
dB(B Xs + )

,
= B(B Xc e )exp
ds
Vcb C u
(3)
where s = q 2 /mb,pole . (s ) and u are defined by

2 5
d (B Xs + ) GF mb,pole 2
Vts Vtb (s ),
=
ds
48 3
G2F m5b,pole
|Vub |2 u ,
(B Xu e ) =
192 3
(4)
(5)
and we have u = 1 + O(s , em , 2 /m2b ). The normalization with the measured semileptonic
decay rate minimizes the uncertainty due to the fifth power of mb ; the factor C,

Vub 2 (B Xc e )

C =
(6)
,
Vcb (B Xu e )
is used to avoid spurious uncertainties due to the b Xc e phase-space factor. This factor can
be determined from a global analysis of the semileptonic data [31]. The quantity (s )/u can
be expressed in terms of the low-scale Wilson coefficients and various functions of s that arise
from the matrix elements. The main formula reads

(s )
eff
=
Re Ci (b )Cjeff (b )Hij (b , s ) ,
(7)
u
ij
where Cieff (b ) = Ci (b ) only for i = 7, 8. The functions Hij (b , s ) are given in Eq. (116) of
Ref. [30] and can be expressed in terms of the coefficients MiA listed in Table 6 of Ref. [30], as
well as the building blocks S99 , S77 , S79 , and S1010 given in Eqs. (112)(115) of Ref. [30].
In the low-q 2 region, there are only a few modifications compared to the analysis in Ref. [30]:
we update the parametric inputs, the influence of the cc resonances are taken into account and the
two-loop matrix element of the operator P9 is given in a more precise way (see Eqs. (48) and (49)
44
of the present paper). The combined effect of these improvements does not change appreciably
the central value for the branching ratio obtained in Ref. [30].
In the high-q 2 region, there are several additional changes necessary besides the ones mentioned above. The electromagnetic contributions to the building blocks S99 , S77 , S79 , and S1010
are different and are presented here for the first time in Eqs. (40)(47). O(1/m3b ) corrections
are numerically relevant and explicitly given in Eq. (6) of Ref. [24]. Finally, the functions F19 ,
F29 , F17 , F27 , F89 , and F87 that appear in the coefficients MiA (listed in Table 6 of Ref. [30]) are
different in the high-q 2 region and are known only numerically. They have been calculated in
Ref. [11].2
2.1.1. New normalization
The authors of Ref. [24] have shown that it is possible to drastically reduce the size of 1/m2b
and 1/m3b power corrections to the integrated decay width, by normalizing it to the semileptonic
b u rate integrated over the same q 2 -interval. This procedure will help reducing the uncertainties induced by the large power corrections to the decay width integrated over the high-s
region. The new observable is defined as follows:
1 d (BX
+
1

s )
s
Vts Vtb 2 s0 ds (s )
s0 d
ds

R(s0 ) =
(8)
= 4
1
1
d (B 0 Xu )
V
ub
d
s
(
s
)
s
u
s
s0 d
0
ds
where (s ) was defined in Eqs. (3) and (5) and the differential u (s ) is given by
2
2 3
du GF |Vub | mb,pole
u (s ).
=
ds
192 3
(9)
Explicit expressions for the O(1, s , 1/m2b , 1/m3b ) contributions to u (s ) can be found in
Eqs. (5), (11) of Ref. [24]. The O(s2 ) correction to u (s ) is given by s2 X2 (s ) with X2 defined in Eq. (60) of Ref. [32]. We have not included electromagnetic corrections of order O()
in the normalization because they are presently not known. However, the impact of those effects in the conventional approach where the fully integrated semileptonic decay rate is used as
normalization is less than 2.5%. We would also like to note that in Eq. (8), contrary to Eq. (3),
corrections of O(s ) to u (s ) can contain residual logarithmically enhanced terms once the
integration over s is restricted to the high-q 2 region. We leave the investigation of terms of O()
and O(s ) to a future analysis.
Note that in the definition of R we used only neutral semileptonic decays in order to reduce the
uncertainties coming from weak annihilation contributions (see further remarks in Section 4.1).
Obviously, for the comparison of the theoretical ratio R with experiment, a separate measurement of B 0 Xu and B Xu with a high-q 2 cut is necessary. These are quantities with
larger experimental uncertainties. However, they are not expected to negate the drastic reduction
of the theoretical uncertainty in the ratio R (see Section 4.2 for numerical results).
2.2. Forwardbackward asymmetry
In complete analogy to the formula for the branching ratio, one can derive also a formula that
expresses the FBA (see [34]) in terms of the low-scale Wilson coefficients and various building
2 Numerical expressions of these functions can be obtained from the authors upon request.
45
blocks. Normalizing to the semileptonic B Xu e decay width, we get

FBA (
Vts Vtb 2 4
s)
dAFB (B Xs + )

,
= B(B Xc e )exp

ds
Vcb
C u
(10)
FBA (
s ) is defined by (z = cos )
where
1
dz sgn(z)
1
2 5
d2 (B Xs + ) GF mb,pole 2 FBA
Vts Vtb (s ),
=
ds dz
48 3
FBA

=
Re Cieff (b )Cjeff (b )Hij (b , s ) .
u
(11)
(12)
ij
Again, the functions Hij (b , s ) depend on the coefficients MiA listed in Table 6 of Ref. [30] and
on the building blocks S710 and S910 :

Re(M A M 10 )SA10 + Hii ,
when i = j,
Hij = A=7,9 A i 10i
(13)
10
A
+ Mi Mj )SA10 + Hij , when i = j,
A=7,9 (Mi Mj
where S710 and S910 are given by

(em)

S710 = 6(1 s )2 1 + 8 s f710 (s ) + u(1) + u(em) + 8 s 710 (s ) + u(em) f710 (s )

81
2
+ 16 s2 u(2) + 4u(1) f710 (s ) 2 s 6 2 1 14s + 9s 2 ,
(14)
mb
mb

(em)

S910 = 3s (1 s )2 1 + 8 s f910 (s ) + u(1) + u(em) + 8 s 910 (s ) + u(em) f910 (s )
41
122
+ 16 s2 u(2) + 4u(1) f910 (s ) 2 s 2 + 2 s 2 (4 3s ).
(15)
mb
mb
(em)
The functions fi10 (s ) (i = 7, 9) can be found in Eqs. (15)(17) of Ref. [9]. The functions ij
represent the electromagnetic corrections to the matrix elements which are calculated for the first
time in the present paper (see Eqs. (32)(38)). The SAB also include non-perturbative O(1/m2b )
corrections from Refs. [19,22]. Contrary to the expression for the branching ratio the quantity 1 ,
which is related to the kinetic energy of the b-quark, does not drop out here. The quantities
Hij = bij + cij + eij
(16)
have the same meaning as in Eq. (117) of Ref. [30]. They need to be included only for i = 1, 2.
The additional ln(m2b /m2 )-enhanced electromagnetic corrections eij for the FBA read
(em)
e210 = 24s (1 s )2 s2 2 210 (s ),

4
e110 = e210 .
3
(em)
(17)
The function 210 (s ) is again new (see Eq. (34)), while the O(2QCD /m2c ) non-perturbative
contributions were calculated in Ref. [21]
46
c210 = + s
2
(1 s )2 (1 + 3s )F (r),
3m2c
1
c110 = c210 ,
6
(18)
where r 1/yc = s/(4m2c ) and the function F (r) is listed for example in Appendix A of
Ref. [30]. The finite bremsstrahlung contributions bij were calculated in Ref. [13]. We do not
present these corrections here but do include them in the numerical analysis.
2.3. Normalization of the FBA
As discussed in Section 2, the normalization of the FBA by the semileptonic or by the
B Xs + decay rate is important in order to cancel the overall factor m5b,pole . Moreover,
the elimination of renormalon ambiguities requires the analytical conversion to a short-distance
mass and a complete expansion in s [35,36].
If we extract the zero of the FBA without using any normalization, the conversion and the
expansion of the overall m5b,pole factor induces a large dependence on the scheme we choose for
the short distance b mass. In fact, we find a 11% (+5%) shift in switching from the 1S scheme
to the MS (pole) one. This large scheme mb dependence was overlooked in previous analyses
of the zero. It can be eliminated by normalizing the FBA by a rate proportional to m5b and fully
expanding in s .
The most natural choice is to normalize the FBA directly by the differential B Xs +
differential decay rate because this normalisation is also used in the experiment. Unfortunately,
this option is not viable because of the poor convergence of the perturbative series for the latter
(the most serious problem is that the lowest term in the expansion is sub-dominant). It is therefore
not surprising that also this procedure leads to a large mb scheme dependence of the zero: we
find a shift of 12% when one compares the 1S to the MS scheme.
Therefore, we propose here a definition of the FBA that, at the same time, removes the m5b
dependence, does not suffer from renormalon ambiguities, has a nicely behaved perturbative
expansion, and can directly be compared with the experimental quantity. The following double
ratio satisfies all the requirements:

FBA

(s )
dAFB (B Xs + ) dB(B Xs + )
(s )
=
,
ds
ds
u
u
(19)
where the quantities are defined in Eqs. (3) and (10). The critical point is that each of the ratios
in brackets gets fully expanded in s , but no overall expansion is done. This is allowed for
two reasons: on the one side, the fully expanded numerator and denominator are renormalon
free, on the other one, they separately correspond to two measurable quantities. As we show in
Section 4.5, also the mb scheme dependence of the zero turns out to be minimal (the difference
between the 1S, MS and pole schemes is about 1%).
This discussion shows that the estimation of the perturbative error of the zero via the comparison of different normalisations as done in [12] is questionable because some normalizations
are affected by renormalon ambiguities and the problem of the large scheme dependence of the
b quark mass should be addressed. As we will discuss in Section 4.5, the standard analysis of the
scale dependence offers a better guidance to the understanding of the perturbative errors in our
predictions.
47
2.4. Perturbative expansion

Some remarks about the perturbative expansion are in order: Large logarithms of the form
s log(mb /MW ) have to be resummed at a given order in perturbation theory using renormalization group techniques. In the case of b s+ , the first large logarithm of the form
log(mb /MW ) arises already without gluons. Moreover, the amplitude of b s+ is proportional to em . So naturally we have an expansion in s and in = em /s while resumming
all powers of s log(mb /MW ). The Leading Order (LO) terms are of order , NLO terms are
proportional to s , NNLO terms are proportional to s2 .
It is well known that this naive s expansion is problematic, since the formally-leading
O(1/s ) term in C9 is accidentally small and much closer in size to an O(1) term. In addition
2 sin2 ). Therefore, also specific higher
the NLO terms are enhanced by a factor of m2t /(MW
W
n
m
order terms in the general s expansion are numerically important.
The b s+ decay amplitude has the following structure (up to an overall factor of GF ):

A = ALO + s ANLO + s2 ANNLO + O s3

3
em
2 em
+ 2 Aem
(20)
+ O 3
LO + s ANLO + s ANNLO + O s
em
with ALO s ANLO and Aem
LO s ANLO . All these terms are included in the numerical analysis in a complete manner, together with the appropriate bremsstrahlung corrections, while also
the term ANNLO is practically complete due to the calculations in Refs. [514]. The only missing parts originate from the unknown two-loop matrix elements of the QCD-penguin operators
whose Wilson coefficients are very small.
Among the contributions to Aem
NNLO , we include only the terms which are either enhanced
2 sin2 ) (with respect to Aem ) [12] or contribute to the
by an additional factor of m2t /(MW
W
NLO
ln(m2b /m2 )-enhanced terms at the decay width level. As discussed above, the latter terms were
calculated in Ref. [30] for the dilepton mass spectrum in the low-q 2 region, while the corresponding terms for the high-q 2 region and the FBA are calculated for the first time in the present
paper.
The perturbative expansion of the ratio (s )/u has a similar structure to that of the squared
amplitude (up to bremsstrahlung corrections and non-perturbative corrections)

A2 = 2 A2LO + s 2ALO ANLO + s2 A2NLO + 2ALO ANNLO

+ s3 2(ANLO ANNLO + ) + O s4

em
em
+ 3 2ALO Aem
LO + s 2 ANLO ALO + ALO ANLO

em
em
+ s2 2 ANLO Aem
NLO + ANNLO ALO + ALO ANNLO

4
em
+ s3 2 ANLO Aem
NNLO + ANNLO ANLO + + O s

+ O 4 .
(21)
We assume that all products in Eq. (7) are expanded in s and . Regarding QCD, a strict NNLO
calculation of (s )/u should only include terms up to order 2 s2 .
In the numerical calculation of (s )/u , however, we include all the terms that are written explicitly in Eq. (21). The term ANLO ANNLO of order 2 s3 is formally an NNNLO term,
but numerically important. Within the electromagnetic corrections the same is true for the
term Aem
NLO ANNLO . We emphasize that those terms beyond the formal NNLO level which are
48
proportional to |C7 |2 and |C8 |2 are scheme independent. The dots in Eq. (21) stand for the unknown terms ALO ANNNLO and Aem
LO ANNNLO and, consequently, can safely be neglected due to
em , and A
ALO s ANLO , Aem
A
s NLO
s NNNLO ANNLO . Thus, one can argue that our calLO
culations pick up the dominant NNNLO QCD corrections. In the following we will therefore call
the accuracy of our calculations improved NNLO.3
Finally let us comment on the mass schemes that we use: The pole mass of the b quark
appears explicitly in the calculation of the matrix elements (s s/m2b,pole ) and in several loop
functions. Unfortunately, the precise determination of the numerical value of m2b,pole is hindered
by renormalon ambiguities that appear in the relation between any short distance mass definition
(e.g. MS, 1S, kinetic schemes) and the pole one. We eliminate these renormalon uncertainties
utilizing the Upsilon expansion described in Refs. [35,36]. Every occurrence of the pole mass
is converted analytically to the 1S-mass before any numerical evaluation of the branching ratio
is performed. In our analysis we use the conversion formula up to order s2 [36]. We follow a
similar approach for the treatment of the charm pole mass that appears in the calculation of some
matrix elements. We adopt the MS charm mass as input and expand the pole mass at order s2
using the formulae presented in Ref. [36]. For what concerns the top mass, we take the pole mass
as input and convert it to the MS scheme at order s3 . The perturbative expansion of the FBA
proceeds along the same lines.
3. Analytical results
In this section, we calculate the electromagnetic corrections to the matrix elements for the
FBA in the low- and high-q 2 region and for the dilepton mass spectrum in the high-q 2 region
and discuss the phenomenological implications of those collinear logarithms in the ee and
final states.
Moreover, we present a more precise fixing of the two-loop matrix elements of the P9 operator.
Finally, we make some remarks on some non-perturbative subtleties.
3.1. Log-enhanced corrections to the FBA, low- and high-q 2 region
First we derive the basic formulae to obtain the expressions for the ln(m2b /m2 )-enhanced corrections to the FBA, where we shall focus here on the un-normalized asymmetry4 corresponding
to the LHS of Eq. (11). We follow closely the notation of Section 5 of Ref. [30].
It was shown in Appendix of Ref. [19] that the angular FBA with respect to , the angle in
the dilepton c.m.s. between the directions of the momenta of the decaying B and the positively
charged lepton, is equivalent to the energy asymmetry between the two leptons in the rest-frame
Events in which cos > 0 in the dilepton c.m.s. correspond to events in which
of the decaying B.
E > E+ measured in the B-meson

rest-frame, where E denotes the -energy. Sticking to the
3 In fact, this scheme is very similar to the one proposed in Ref. [6] and used in many previous analyses of the
b s+ observables. In these works the formally-leading, but accidentally small O(1/s ) term in C9 is treated
as O(1) and absorbed into the NLO coefficient: as consequence the two-loop matrix element of P9 and the three-loop
mixing of the four-quark operators into P9 are left out even though they are formally NNLO terms. This is the main
difference between the NNLO scheme proposed in Ref. [6] and the improved NNLO used in the present manuscript.
4 In this subsection we suppress the normalization factor, thus, the symbol A
FB denotes the un-normalized FBA, see
Eqs. (23), (27), and (31).
49
latter frame and defining the scaled energies

y
2E
,
mb
(22)
we can write the fully differential FBA as

dAFB = PF ds dy+ dy (1 + s y+ y )|A|2 sgn(y y+ ),
(23)
with the pre-factor

PF =
G2F mb |Vtb Vts |2

.
32 3
(24)
We are mainly interested in ln(m2b /m2 )-enhanced electromagnetic corrections. They are derived
by means of the splitting function and we shall adopt the kinematics from Fig. 2 of Ref. [30]. In
the collinear limit the fully differential FBA reads
dA(m)
s dy+ dy (1 + s y+ y )f(m) (x)|A|2 sgn(y y+ ).
FB,coll = PF dx d
(25)
We shall only retain the ln(m2b /m2 )-enhanced part of f(m) , which then becomes independent
of E,

[1 + (1 x)2 ]
mb
(m)
f (x) = 4 e
(26)
ln
.
x
m
The approximation of dropping the non-log-enhanced terms in the splitting function is justified
by the fact that even for high q 2 the logarithmically enhanced part dominates numerically. This
(em)
can be seen, for instance, by the fact that the non-log-enhanced term of (1 s )2 (1 + 2s )99 (s )
(see Eq. (94) of Ref. [30]) vanishes at s = 1.
In the squared amplitude in Eq. (25) we keep only those terms that are relevant for the FBA,
i.e. all terms that do not drop out upon integration over the sign-function,

|A|2 = 2 Re C10 C9 + s C10 (C2 + CF C1 )f2 (s ) P10 tree P9 tree

+ 2 Re C10 C7 P10 tree P7 tree .

(27)
As in the case of the decay width we must consider the difference
dA(m)
FB,coll,2
(m)
dAFB,coll,3
(28)
,
ds
ds
where we stay differential in the double and in the triple invariant, respectively. At this point
more care is required compared to the calculation of the decay width. Due to the emergence
of the sign-function in Eq. (25) we must distinguish between photon-emission from the + and
from the in case of the double invariant. In the former case we have to stay differential in
s = (xp
1 + p2 )2 /m2b , in the latter in s = (p1 + xp
2 )2 /m2b . Also the y change accordingly. We
therefore have
(m)
dAFB,coll,2,
ds
1

= PF
x+
s
x(1+
x)
dy +
dx
0
1s
s /x
1 s
1
s
1
dx
s /x
dy
dx
0
1
dy
x+
s
x(1+
x)
50

(m)
f (x) 2
|A| |s s /x;y
1y +s /x .
x
(29)
The two expressions corresponding to upper and lower sign should be equal due to the antisymmetry of |A|2 in y+ y . The case of the triple invariant is simpler since we stay differential in
s = (p1 + p2 )2 /m2b , namely
ds
s
1+
2
(m)
dAFB,coll,3,
= PF
1
dy
dx
s

dy f(m) (x)|A|2 |y 1y +s ,
(30)
1+s
2
where the expression for the upper and lower sign should again be equal due to the antisymmetry
of |A|2 . We finally have to combine the expressions according to Eq. (28).
The corrections to the un-normalized FBA read
dAFB G2F m5b 2
FBA
V Vtb
(s )
=
ds
48 3 ts
2
(em)

G2 m5
710 (s )
= F 3b Vtb Vts (1 s )2 s 48 s Re C7 C10
48

(em)
(em)
24s Re C9 C10
910 (s ) + s Re (C2 + CF C1 )C10
210 (s ) ,
(31)
with

1 + 3s
s ln s
1 + s
7 16 s + 9s
s
)
+
+
ln(1
ln
,
4(1 s )
1 s
2
(1 s )
m2
(32)
2
mb
5 16 s + 11s
(em)
910 (s ) = ln
+ ln(1 s )
4(1 s )
m2

1 5s
1 + s
(1 3s ) ln s
+
(33)
ln
,
1 s
2
(1 s )
2

mb
7 (s ) + i7I (s )
b
8 (em)
(em)
(
s
)
ln
+
210 (s ) = ln
(34)
9 910
5 GeV
24s (1 s )2
m2

(em)
(s ) = ln
710
m2b

(em)
(em)
The functions 710 (s ) and 910 (s ) are known in the entire q 2 -region, whereas the function
(em)
210
(s ) was obtained by a least squares fit. The function 7 , valid in the low-s region, reads
7 (s ) = 0.259023 28.424s + 205.533s 2 603.219s 3 + 722.031s 4 ,

7I (s ) = 12.20658 215.8208(s a) + 412.1207(s a)2 (s a)2 (s a),
(35)
(36)
with a = (4m2c /m2b )2 . In the high-s region the function 7 reads ( = 1 s )

7 (s ) = 77.0256 2 264.705 3 + 595.814 4 610.1637 5 ,
(37)
7I (s ) = 135.858 2
(38)
618.990 + 1325.040 1277.170 .

3
The polynomials in the high-s region were obtained such as to have a double zero at s = 1.
51
3.2. Log-enhanced corrections to the BR, high-q 2 region

The log-enhanced corrections to the differential decay width given in Eq. (4) read

2
G2 m5
d
(em)
(em)
= F 3b Vtb Vts (1 s )2 s 8(1 + 2s ) |C9 |2 99 (s ) + |C10 |2 1010 (s )
ds
48

(em)
(em)
+ s Re (C2 + CF C1 )C9 29 (s ) + s2 2 (C2 + CF C1 )2 22 (s )
(em)
(em)
(s ) + s2 2 Re (C2 + CF C1 )C7 27
(s )
+ 96 s Re C7 C9 79

8 2 2
(em)
+8 4+
s |C7 |2 77 (s ) .
s
(em)
(em)
(39)
(em)
Exact analytical expressions are known for the functions 99 (s ), 1010 (s ), 77 (s ), and
(em)
79 (s ) (see Eqs. (94) and (100)(102) of Ref. [30]). Therefore, they hold in the entire q 2 region, while the other -functions are obtained by a least-squares fit in the high-q 2 region (for
fixed values of mb and mc ):
2

mb
4 (s ) + i4I (s )
b
16 (em)
(em)
,
29 (s ) = ln
(40)
+ 1010 (s ) ln
9
5 GeV
8(1 s )2 (1 + 2s )
m2

2

mb
4 (s )
b
5 (s )
(em)
+
ln
22 (s ) = ln
5 GeV
8(1 s )2 (1 + 2s ) 9(1 s )2 (1 + 2s )
m2

64 (em)
b
+ 1010
(41)
(s ) ln2
,
81
5 GeV
2

6 (s ) + i6I (s )
mb
b
8 (em)
(em)
(s ) = ln
(
s
)
ln
,
+
27
(42)
9 79
5 GeV
96(1 s )2
m2
with the functions i , valid in the high-s region ( = 1 s ):
4 (s ) = 148.061 2 + 492.539 3 1163.847 4 + 1189.528 5 ,
(43)
4I (s ) = 261.287 2 + 1170.856 3 2546.948 4 + 2540.023 5 ,

5 (s ) = 221.904 2 + 900.822 3 2031.620 4 + 1984.303 5 ,
6 (s ) = 298.730 2 + 828.0675 3 2217.6355 4 + 2241.792 5 ,
6I (s ) = 528.759 2 + 2095.723 3 4681.843 4 + 5036.677 5 .
(44)
(45)
(46)
(47)
The fits are excellent for s > 0.65. Again, the polynomials were obtained such as to have a double
zero at s = 1.
3.3. Phenomenological implications of the collinear logarithms
After inclusion of the NLO QED matrix elements calculated in the previous subsections the
electron and muon channels receive different contributions due to terms involving ln(m2b /m2 ).
We emphasize that this is the only source of the difference between the electron and muon channel.
However, as already pointed out in Ref. [30], the presence of this logarithm is strictly related
to the definition of the dilepton invariant mass. The collinear logarithm ln(m2b /m2 ) disappears
52
if all photons emitted by the final state on-shell leptons are included in the definition of q 2 :
(p1 +p2 )2 (p1 +p2 +p )2 . If none of these photons was included in the definition of q 2
i.e. if a perfect separation of leptons and collinear photons was possible experimentallythen
our expressions containing ln(m2b /m2 ) were directly applicable. This is the case for muons since
the separation of muons and collinear photons is practically perfect [46]. For electrons with the
current BaBar and Belle setups the lepton mass gets replaced by an effective mass parameter
which is found to be of the same order as m . Hence our numerical results in Section 4 for muonic
final states should be applied also to the case of electronic final states in the present BaBar and
Belle setups. All results in Section 4 for the electron channel are given under the assumptions of
perfect separation of electrons and collinear photons. We refer the reader to Section 6 of Ref. [30]
for more details.
3.4. Two-loop matrix element of P9
In Ref. [12], an estimate of the two-loop (NNLO) matrix element of the operator P9 denoted
(2)
99 (s ) in Eq. (74) of Ref. [30]was presented which was neglected in all previous analyses.
The estimate is based on the fact that the QCD corrections to b s+ are identical to those of
b u(t b), in the limit of vanishing strange (bottom) quark mass. However, the specific
two-loop calculation of the b u decay was available as an expansion in (1 s ) [37], while
2 /m2
the two-loop contribution of the top quark decay was only known as an expansion of MW
t
(which translates into an expansion in q 2 /m2b for b s) up to the second order based on Pade
approximation methods [38]. In the meanwhile, the QCD corrections to b u are also known
as an expansion in s [32,39].
(2)
by the in s = expanded results
Thus, in the low-s region we approximate the function 99
(2)
99 (s ) X2 ( = s )/X0 ( = s )
(48)
where X2 () is given in Eq. (60) of Ref. [32] and X0 () in Eq. (2) of Ref. [39].
(2)
by the in (1 s ) =
Analogously, in the high-s region we approximate the function 99
expanded results of X2 ( = 1 s ) and X0 ( = 1 s ) as given in Eqs. (2) and (3) of Ref. [37].
Note that the normalization of the Xi differs by a factor of 2 in Ref. [37] and Refs. [32,39].5
3.5. Some non-perturbative subleties
For a detailed discussion of non-perturbative corrections we refer the reader to Section 5 of
Ref. [11]. Here we add few remarks on some specific issues.
The long-distance corrections related to the cc intermediate states originate from the nonperturbative interactions of the cc pair in the process B Xs cc Xs + . If the dilepton
invariant mass is near the first two J P C = 1 cc resonances ( and ), this effect is very
large and shows up as a peak in the dilepton mass spectrum which can easily be eliminated by
suitable kinematical cuts. More delicate is the estimate of such long-distance effects away from
the resonance peaks in the perturbative windows.
Within the KS-approach [25], one reabsorbs charm rescattering effects into the matrix element
operators is estimated by means of experimental data on (e+ e
of P9 ; the effects of b ccs
cc hadrons) using a dispersion relation. To be more specific, the function h(z, s) appearing in
5 There is a typo in Eq. (4) of Ref. [37]: the powers of the last four s have to be increased by 1 each.
53
Eq. (A.5) of Ref. [11] (which corresponds to the expression g(yc ) + 8/9 log(mb /mc ) 4/9 in
Eq. (72) of Ref. [30]) is replaced by (z = m2c /m2b )
s
h(z, s ) h(z, 0) + P
3

m2
ds
cc (
Rhad
s )
cc
(s ),
+ i Rhad
s (s s )
3
(49)
cc (
s ) = (e+ e cc)/
(e+ e + ). This method is exact only in the limit in

where Rhad
colour-singlet
which the B Xs cc transition can be factorized into the product of s b and cc
currents. It is possible to take into account non-factorizable effects and reproduce the correct
cc by a purely phenomenological factor = 2.6.
hadronic branching ratios by multiplying Rhad
However, a model-independent way to estimate the non-factorizable cc long-distance effects far
from the resonance region exists by means of an expansion in inverse powers of the charm-quark
mass [20,21]. Having included those 1/mc corrections, an inflation of the factorizable KScorrections using the phenomenological factor would lead to a double-counting. The numerical
impact of the KS contribution to the integrated rate in the low- and high-s regions is about +2%
and 10%, respectively.
In Section 5 of Ref. [11], another large non-perturbative error for the high-q 2 region was
located in the linear 1/mb corrections. The physical observables defined in terms of a q 2 -cut
are sensitive to a 1/mb term via the uncertainty on the value of mb,pole or equivalently via the
relation of the pole mass and the hadron mass

b + O 1/m2b .
MB = mb 1 + /m
(50)
Choosing as reference cut the value smin = 0.6, the integrated normalized dilepton mass spectrum
2 can be written as
defined in terms of qmin

2
Rcut qmin
=
dq 2
2
q 2 >qmin

= 1 6.2
d (B Xs + )
(B Xc e)
2
qmin
m2b

0.6 + O

2
qmin
m2b
2
0.6
ds R(s),
(51)
0.6
which implies
mb
Rcut
7.4
.
Rcut
mb
(52)
Using the pole mass scheme with mb = 0.1 GeV, this leads to a 15% error on Rcut . However,
this error gets now significantly reduced in our updated analysis using the kinematical 1S scheme
for the mb mass.
4. Numerical results
The numerical inputs that we adopt are summarized in Table 1.
54
Table 1
Numerical inputs that we use in the phenomenological analysis. Unless explicitly specified, they are taken from PDG
2004 [45]
s (Mz ) = 0.1189 0.0010 [40]
e (Mz ) = 1/127.918
2 sin2 = 0.2312
sW
W
|Vts Vtb /Vcb |2 = 0.962 0.002 [41]
|Vts Vtb /Vub |2 = (1.28 0.12) 102 [41]
BR(B Xc e )exp = 0.1061 0.0017 [43]
MZ = 91.1876 GeV
MW = 80.426 GeV
2
eff
2 = (0.12 0.02) GeV
2
eff
1 = (0.243 0.055) GeV [42]
fu0 fs = (0 0.04) GeV3 [24]
me = 0.51099892 MeV
m = 105.658369 MeV
m = 1.77699 GeV
mc (mc ) = (1.224 0.017 0.054) GeV [42]
m1S
b = (4.68 0.03) GeV [31]
mt,pole = (170.9 1.8) GeV [44]
mB = 5.2794 GeV
C = 0.58 0.01 [31]
1 = (0.06 0.06) GeV3 [31]
fu0 + fs = (0 0.2) GeV3 [24]
fu = (0 0.4) GeV3 [24]
4.1. Branching ratio in the high-q 2 region

For the branching ratio integrated over the region q 2 > 14.4 GeV2 we find:
+0.01
+0.14
+0.006
high
+ 0.06
0.02mt + 0.003
0.05mb
B = 2.40 107 1 + 0.02
0
b
C,mc
+0.0002
+ 0.001 0.002CKM 0.02BRsl 0.052 0.191 0.14fs 0.02fu
s

= 2.40 107 1+0.29
(53)
0.26 ,
high
+0.02
+0.16
+0.005
+ 0.08
0.02mt + 0.0009
0.05mb
Bee = 2.09 107 1 + 0.04
0
b
C,mc
+0.0003
+ 0.002 0.002CKM 0.02BRsl 0.052 0.221 0.16fs 0.02fu
s

= 2.09 107 1+0.32
(54)
0.30 .
The scale uncertainty has been estimated by varying the matching scale 0 and the low-energy
scale b by factors of 2 around their central values (120 and 5 GeV, respectively). In our approach
we normalized the b s decay width to the semileptonic rate averaged over neutral and
charged B mesons; hence, we average fu0 and fu . From the values quoted in Table 1 we obtain:
fu = 0 0.2 and fs = 0 0.1.6
The other parametric uncertainties are obtained by varying the inputs within the errors given
in Table 1. We assume the errors on C and mc to be fully correlated. The total error is obtained by
adding the individual uncertainties in quadrature. We note that here and in the following all errors
are parametric or perturbative uncertainties only and that subleading non-perturbative corrections
of order O(s /mb ) will give an additional uncertainty.
6 Regarding the weak annihilation contributions (f 0 , f + , f 0 , f + ) we follow the analysis in Ref. [24]. Here are some
s
u u
s
additional remarks in order: Using flavor SU(3) one concludes that fu0 fs0 . However, we stress that the additional
s b. Nevertheless, due
assumption fs0 fs+ comes from neglecting iso-singlet effects in the SU(3) decomposition of bs
ub
to a mismatch in the flavor of the spectator quark in the B meson with the quark appearing in the operators (bu
s b), it is not unreasonable to assume fs+ , fu0 < fu+ . This leads to the relation fu0 fs0 fs+ < fu+ [24] that
and bs
we use in the numerics. We note again that in the definition of R we use only neutral semileptonic decays, thus, the
dependence on fu+ disappears in the ratio. The actual numerical inputs we adopt are taken from Refs. [24,33]. Because
all our expressions for the branching ratio and the ratio R0 are linear in these non-perturbative parameters, hence it is
trivial to adjust our results to accommodate for different sets of non-perturbative inputs.
55
(em)
Fig. 1. The log-enhanced part of (1 s )2 (1 + 2s )99 (s ), see Eq. (94) of Ref. [30]. This function integrates to zero,
but gives contributions of opposite sign to the low-s and high-s branching ratio, respectively (shaded areas). The moduli
of the absolute shifts are of comparable size. However, the relative effect is much larger in the high-s compared to the
low-s region due to the steep decrease of the differential decay width at large s .
Log-enhanced QED bremsstrahlung corrections shift the central value by about 8% and
20% for the muonic and the electronic final state, respectively. These relative shifts are much
larger than the corresponding ones in the low-q 2 region. In fact, these corrections vanish when
integrated over the whole dilepton invariant mass spectrum, but are relevant if one is restricted
to certain regions in phase space. One observes that the absolute shifts in the central values
of the branching ratio are of similar size but of opposite sign in the low-q 2 compared to the
high-q 2 region, namely for the muonic final state +3 108 and 2 108 respectively. But
since the differential decay width decreases steeply at large q 2 , the relative effect is much more
pronounced in the high-q 2 compared to the low-q 2 region. See also Fig. 1 for illustration.
4.2. Ratio R(s0 )
The numerical results we obtain for the ratio R(s0 ) discussed in Section 2.1.1 are for s0 =
14.4 GeV2 :
R(s0 ) = 2.29 103 (1 0.04scale 0.02mt 0.01C,mc 0.006mb 0.005s
high
0.09CKM 0.0032 0.051 0.03fu0 +fs 0.05fu0 fs )

= 2.29 103 (1 0.13),
high
R(s0 )ee
(55)
= 1.94 103 (1 0.06scale 0.02mt 0.02C,mc 0.004mb 0.006s

0.09CKM 0.012 0.091 0.05fu0 +fs 0.05fu0 fs )
= 1.94 103 (1 0.16).
(56)
Note that uncertainties from poorly known O(1/m3b ) power corrections are now under control;
the largest source of error is Vub .
56
4.3. Branching ratio in the low-q 2 region

For the branching ratio integrated in the range 1 GeV2 < m2 < 6 GeV2 it was found in
Ref. [30]:
B = [1.59 0.08scale 0.06mt 0.024C,mc 0.015mb 0.02s (MZ )
0.015CKM 0.026BRsl ] 106 = (1.59 0.11) 106 ,
(57)
Bee = [1.64 0.08scale 0.06mt 0.025C,mc 0.015mb 0.02s (MZ )

0.015CKM 0.026BRsl ] 106 = (1.64 0.11) 106 .
(58)
We note that we find for B a +1.8% shift of the central value due to the KS-corrections which
were discussed in Section 3.4 and which were not included in the previous analysis. Moreover,
the update of the input parameters (CKM, mt , s ) leads to a 3.1% shift; thus, we end up to an
overall change of 1.3% within our new phenomenological analysis compared with the previous one in Ref. [30]. The total error, being from parametric and perturbative uncertainties only,
remains unchanged.
4.4. Integrated FBA in low-s region
By the end of the current B factories the fully differential FBA will not be accessible experimentally, contrary to the integral over one or more bins in the low-q 2 region. However,
these integrals can already serve of gain important information on the shape of the FBA and
to constrain the parameter space of new physics models (see for instance [47]). We now give the
results for the integrated FBA based on Eq. (19). We subdivide the low-s region into the two
bins s [1, 3.5] GeV2 and s [3.5, 6] GeV2 , which we will call bin 1 and bin 2, respectively.7
We then integrate the numerator of the double ratio, Eq. (10), over the respective bin, and divide
the result by the denominator of the double ratio, Eq. (3) integrated over the same bin, i.e. we
first integrate FBA and branching ratio separately and subsequently divide the two numbers. Our
results read:
FB
A bin 1 = [9.09 0.83scale 0.03mt 0.24mc ,C 0.20mb 0.18s (MZ ) 0.022 ]%
A FB
ee bin 1
= [9.09 0.91]%,
(59)
= [8.14 0.80scale 0.03mt 0.23mc ,C 0.19mb 0.18s (MZ ) 0.022 ]%
= [8.14 0.87]%,
(60)
FB
A bin 2 = [+7.80 0.55scale 0.02mt 0.31mc ,C 0.34mb 0.16s (MZ ) 0.202 ]%

A FB
ee bin 2
= [+7.80 0.76]%,
(61)
= [+8.27 0.47scale 0.02mt 0.30mc ,C 0.33mb 0.15s (MZ ) 0.192 ]%

= [+8.27 0.69]%.
(62)
The total error is obtained by adding the individual ones in quadrature. For the entire low-s
region, we get
7 Predictions for different bins can be produced upon request.
57
Table 2
Dependence of the zero of the FBA on the b quark mass scheme. The input values are
mb (mb ) = 4.205 GeV [48] and mb,pole = 4.8 GeV

A FB
low
1S
MS
Pole
3.50
3.38
3.47
3.34
3.52
3.41

= [1.50 0.90]%,
A FB
ee low
(63)

= [0.86 0.85]%.
(64)
The relative errors in the respective bins are considerably smaller than for the entire low-s
region since the respective values in each bin are similar in size and of opposite sign.
4.5. Analysis of the zero of the FBA
The basic formula for the extraction of the zero is Eq. (19). We expand the two ratios separately in s , , 1 and 2 , and keep all the terms as specified in Section 2.4. It is understood that
also the conversion of the mass scheme for the bottom, the charm and the top quark is performed
in the way described there. The results for q02 , the zero of the FBA in the low-s region, are
2
q0 = [3.50 0.10scale 0.002mt 0.04mc ,C 0.05mb 0.03s (MZ )
0.0011 0.012 ] GeV2

2
q0
ee
= (3.50 0.12) GeV2 ,
(65)
= [3.38 0.09scale 0.002mt 0.04mc ,C 0.04mb 0.03s (MZ )

0.0021 0.012 ] GeV2
= (3.38 0.11) GeV2 .
(66)
The central values are obtained for the matching scale 0 = 120 GeV and the low-energy scale
b = 5 GeV. The uncertainty from missing higher order perturbative corrections have been estimated by increasing and decreasing the scales 0 and b by factors of 2. Uncertainties induced
by mt , mb , mc , C, s (MZ ), 1 and 2 are obtained by varying the various inputs within the errors
given in Table 1. We assume the errors on C and mc to be fully correlated. The total errors are
again obtained by adding the individual ones in quadrature. In order to show the stability of the
zero under change of the b quark mass scheme we collect in Table 2 the results we obtained in
the 1S, MS and pole schemes.
The total errors are 3.4% and 3.3% respectively and therefore quite small. It is often argued
that especially the small dependence at the zero is an accident and should be increased by
hand. We argue in the following that the small dependence is a reasonable reflection of the
perturbative error.
One test of our estimation of the perturbative error on the zero consists in comparing the
central values and b dependences of lower-order predictions. The comparison between the
NNLO + QED, NNLO and NLO extraction of the zero reads (here we consider only the muon
58
Fig. 2. b -dependence of the forwardbackward asymmetry for the muonic final state. The lines are the NNLO QCD
result; the dashed line corresponds to b = 5 GeV, and the solid lines to b = 2.5, 10 GeV. The shaded area is the region
spanned by the NLO asymmetry for 2.5 GeV < b < 10 GeV.
channel and quote only the scale uncertainty):
(3.50 0.10) GeV2 NNLO + QED,

2
q0 = (3.45 0.11) GeV2 NNLO,
(3.11 0.39) GeV2 NLO.
(67)
In Fig. 2, we plot the b dependence of the FBA in the low-q 2 region and compare NNLO and
NLO QCD results. Here we note that the scale uncertainty is maximal at q 2 1 GeV2 , decreases
smoothly at larger q 2 and almost vanishes at the edge of the low-q 2 region. Moreover, the band
between the two solid lines, representing the scale uncertainty of the NNLO FBA, lies to a large
extent within the shaded area of the NLO scale uncertainty. The same holds true if we compared
NNLO + QED vs. NLO QCD results.
In Fig. 3 we show the entireparametric and perturbativeerror band of the full NNLO +
QED asymmetry over the whole low-s region. The plot shows that the perturbative expansion
converges nicely everywhere in the low-q 2 region.
The numerical results of Eq. (67) and the plots therefore suggest that the variation of the scale
b properly describes the size of the missing perturbative corrections.
As already stressed before, the errors considered here are parametric and perturbative ones
only; and unknown subleading non-perturbative corrections of order O(s /mb ) may give an
additional uncertainty of order 5%.
4.6. New-physics formulae
Here, we present numerical formulae for the various observables for non-SM values of the
high-scale Wilson coefficients of the operators P7 , P8 , P9 and P10 :

high
B = 2.399 0.002576I(R10 ) 0.01277I(R7 ) + 0.0002656I R7 R8

+ 0.0004108I R7 R9 + 0.002027I(R8 ) 0.00003375I R8 R10

+ 0.001676I R8 R9 + 0.1079I(R9 ) + 3.022R(R10 )
59
Fig. 3. The full NNLO + QED asymmetry (dashed line) and the totalparametric and perturbativeerror band (shaded
area).

+ 0.001146R R10 R7 + 0.0001236R R10 R8 0.0173R R10 R9

0.1312R(R7 ) 0.0143R(R8 ) + 0.0008111R R8 R7 + 0.9375R(R9 )

0.0568R R9 R7 0.006099R R9 R8 + 1.558|R10 |2 + 0.003436|R7 |2

+ 0.00004162|R8 |2 + 0.2231|R9 |2 107 ,
(68)

high
Bee = 2.085 0.002576I(R10 ) 0.011I(R7 ) + 0.0002656I R7 R8

+ 0.0004108I R7 R9 + 0.002162I(R8 ) 0.00003375I R8 R10

+ 0.001676I R8 R9 + 0.09845I(R9 ) + 2.73R(R10 )

+ 0.001146R R10 R7 + 0.0001236R R10 R8 0.0173R R10 R9

0.1119R(R7 ) 0.01279R(R8 ) + 0.0006912R R8 R7 + 0.8243R(R9 )

0.04872R R9 R7 0.005478R R9 R8 + 1.411|R10 |2 + 0.002655|R7 |2

+ 0.00003702|R8 |2 + 0.2016|R9 |2 107 ,
(69)
where
(00)eff
1 + R7,8 =
C7,8
(0 )
(00)eff,SM
C7,8
(0 )
(11)
and 1 + R9,10 =
C9,10 (0 )
(11)SM
C9,10
(70)
(0 )
and we refer the reader to the definition of the Wilson coefficients given in Ref. [30].
The analogous formulae for the integrated forwardbackward asymmetry in bin 1 and bin 2
are (we give separately the FBA from Eq. (10), AFB
, and the normalizing branching ratio from
Eq. (3), B ):
FB
A bin 1 = 7.877 0.84I(R10 ) + 0.1901I(R8 ) + 0.1901I R8 R10

+ 0.02323I(R9 ) 8.136R(R10 ) + 0.02323I(R9 )R(R10 )

6.114R R10 R7 0.6183R R10 R8 + 5.048R R10 R9
5.942R(R7 ) 0.6059R(R8 ) + 4.947R(R9 )
60

0.02323I(R10 )R(R9 ) + 0.01569R R9 R7 0.08693|R10 |2

0.0128|R9 |2 108 ,
(71)

(B )bin 1 = 8.653 + 0.07321I(R7 ) + 0.0167I R7 R8 + 0.002753I R7 R9

0.003923I(R8 ) + 0.01471I R8 R9 0.03578I(R9 )

+ 10.62R(R10 ) + 0.008387R R10 R7 + 0.001049R R10 R8

0.05413R R10 R9 + 0.262R(R7 ) + 0.02037R(R8 )

+ 0.05047R R8 R7 + 2.264R(R9 ) 0.4778R R9 R7

0.05645R R9 R8 + 5.463|R10 |2 + 0.2087|R7 |2 + 0.002892|R8 |2

+ 0.7634|R9 |2 107 ,
(72)
FB
A bin 2 = 5.459 0.7026I(R10 ) + 0.1426I(R8 ) + 0.1426I R8 R10

+ 0.03706I(R9 ) + 5.472R(R10 ) + 0.03706I(R9 )R(R10 )

4.48R R10 R7 0.4645R R10 R8 + 7.832R R10 R9
4.352R(R7 ) 0.4645R(R8 ) + 7.622R(R9 )

0.03706I(R10 )R(R9 ) + 0.01172R R9 R7 0.1382|R10 |2

0.02034|R9 |2 108 ,
(73)

(B )bin 2 = 7.052 + 0.02856I(R7 ) + 0.005661I R7 R8 + 0.002082I R7 R9

+ 0.01401I(R8 ) + 0.01112I R8 R9 0.03512I(R9 ) + 9.472R(R10 )

+ 0.006338R R10 R7 0.04942R R10 R9 0.4596R(R7 )

0.05505R(R8 ) + 0.01678R R8 R7 + 2.394R(R9 )

0.3527R R9 R7 0.04016R R9 R8 + 4.879|R10 |2

+ 0.07016|R7 |2 + 0.6891|R9 |2 107 .
(74)
The integrated asymmetries in bin 1, bin 2, and in the whole low-q 2 region are then according to
Eq. (19):
A FB
A FB
A FB
=
bin 1
=
bin 2
=
low
(AFB
)bin 1
(B )bin 1
(AFB
)bin 2
(B )bin 2
(75)
(76)
FB
(AFB
)bin 1 + (A )bin 2
(B )bin 1 + (B )bin 2
(77)
5. Summary and outlook

In this paper we extend the analysis of log-enhanced QED corrections initiated in Ref. [30] to
the B Xs decay width in the high-q 2 region and to the forwardbackward asymmetry.
We give a complete phenomenological analysis of all relevant quantities related to these
observables, including the ratio R(s0 ) recently proposed in Refs. [24,49]. We propose a new
61
approach to the extraction of the zero of the FBA and argue that the scale dependence obtained
by this procedure is a reasonable reflection of the perturbative error.
It is well known that the measurements of those quantities, in addition to the B Xs
branching ratio, will allow to fix magnitude and sign of all relevant Wilson coefficients in the
SM. Since at the end of the current B factories quantities integrated over certain q 2 -bins will be
already accessible with high precision, the following observation is important, namely that the
double differential decay width decomposed according to (z = cos )

3

d2
=
1 + z2 HT q 2 + 2zHA q 2 + 2 1 z2 HL q 2 ,
2
dq dz 8
(78)

d
= HT q 2 + HL q 2 ,
2
dq
(79)
where

dAFB
= 3/4HA q 2 ,
2
dq
includes a third quantity which depends on a different combination of Wilson coefficients [50].
The NNLO + QED analysis of this observable will be published in a forthcoming paper.
Acknowledgements
We would like to thank Martin Beneke, Christoph Greub, Ulrich Haisch, Gino Isidori, Mikoaj
Misiak, Lalit Sehgal, and Peter Uwer for interesting and helpful discussions and suggestions.
This work was supported by Schweizerischer Nationalfonds (SNF) and by Deutsche Forschungsgemeinschaft, SFB/TR 9 Computergesttzte Theoretische Teilchenphysik. Fermilab is operated by Fermi Research Alliance, LLC under Contract No. DE-AC02-07CH11359 with the
United States Department of Energy.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
T. Hurth, Int. J. Mod. Phys. A 22 (2007) 1781, hep-ph/0703226.

T. Hurth, Rev. Mod. Phys. 75 (2003) 1159, hep-ph/0212304.
M. Iwasaki, et al., Belle Collaboration, hep-ex/0503044.
B. Aubert, et al., BaBar Collaboration, Phys. Rev. Lett. 93 (2004) 081802, hep-ex/0404006.
C. Bobeth, M. Misiak, J. Urban, Nucl. Phys. B 574 (2000) 291, hep-ph/9910220.
H.H. Asatryan, H.M. Asatrian, C. Greub, M. Walker, Phys. Rev. D 65 (2002) 074004, hep-ph/0109140.
H.H. Asatryan, H.M. Asatrian, C. Greub, M. Walker, Phys. Rev. D 66 (2002) 034009, hep-ph/0204341.
A. Ghinculov, T. Hurth, G. Isidori, Y.P. Yao, Nucl. Phys. B 648 (2003) 254, hep-ph/0208088.
H.M. Asatrian, K. Bieri, C. Greub, A. Hovhannisyan, Phys. Rev. D 66 (2002) 094013, hep-ph/0209006.
A. Ghinculov, T. Hurth, G. Isidori, Y.P. Yao, Eur. Phys. J. C 33 (2004) S288, hep-ph/0310187.
A. Ghinculov, T. Hurth, G. Isidori, Y.P. Yao, Nucl. Phys. B 685 (2004) 351, hep-ph/0312128.
C. Bobeth, P. Gambino, M. Gorbahn, U. Haisch, JHEP 0404 (2004) 071, hep-ph/0312090.
H.M. Asatrian, H.H. Asatryan, A. Hovhannisyan, V. Poghosyan, Mod. Phys. Lett. A 19 (2004) 603, hep-ph/
0311187.
M. Gorbahn, U. Haisch, Nucl. Phys. B 713 (2005) 291, hep-ph/0411071.
A.G. Akeroyd, et al., hep-ex/0406071.
J. Hewett, et al., hep-ph/0503261.
M. Bona, et al., arXiv: 0709.0451 [hep-ex].
A.F. Falk, M.E. Luke, M.J. Savage, Phys. Rev. D 49 (1994) 3367, hep-ph/9308288.
A. Ali, G. Hiller, L.T. Handoko, T. Morozumi, Phys. Rev. D 55 (1997) 4105, hep-ph/9609449.
J.W. Chen, G. Rupak, M.J. Savage, Phys. Lett. B 410 (1997) 285, hep-ph/9705219.
G. Buchalla, G. Isidori, S.J. Rey, Nucl. Phys. B 511 (1998) 594, hep-ph/9705253.
G. Buchalla, G. Isidori, Nucl. Phys. B 525 (1998) 333, hep-ph/9801456.
62
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
C.W. Bauer, C.N. Burrell, Phys. Rev. D 62 (2000) 114028, hep-ph/9911404.

Z. Ligeti, F.J. Tackmann, Phys. Lett. B 653 (2007) 404, arXiv: 0707.1694.
F. Kruger, L.M. Sehgal, Phys. Lett. B 380 (1996) 199, hep-ph/9603237.
M. Neubert, JHEP 0007 (2000) 022, hep-ph/0006068.
C.W. Bauer, Z. Ligeti, M.E. Luke, Phys. Rev. D 64 (2001) 113004, hep-ph/0107074.
K.S.M. Lee, I.W. Stewart, Phys. Rev. D 74 (2006) 014005, hep-ph/0511334.
K.S.M. Lee, Z. Ligeti, I.W. Stewart, F.J. Tackmann, Phys. Rev. D 74 (2006) 011501, hep-ph/0512191.
T. Huber, E. Lunghi, M. Misiak, D. Wyler, Nucl. Phys. B 740 (2006) 105, hep-ph/0512066.
C.W. Bauer, Z. Ligeti, M. Luke, A.V. Manohar, M. Trott, Phys. Rev. D 70 (2004) 094017, hep-ph/0408002.
I. Blokland, A. Czarnecki, M. Slusarczyk, F. Tkachov, Phys. Rev. D 71 (2005) 054004, hep-ph/0503039.
M.B. Voloshin, Phys. Lett. B 515 (2001) 74, hep-ph/0106040.
A. Ali, T. Mannel, T. Morozumi, Phys. Lett. B 273 (1991) 505.
A.H. Hoang, Z. Ligeti, A.V. Manohar, Phys. Rev. D 59 (1999) 074017, hep-ph/9811239.
A.H. Hoang, hep-ph/0008102.
A. Czarnecki, K. Melnikov, Phys. Rev. Lett. 88 (2002) 131801, hep-ph/0112264.
K.G. Chetyrkin, R. Harlander, T. Seidensticker, M. Steinhauser, Phys. Rev. D 60 (1999) 114015, hep-ph/9906273.
I. Blokland, A. Czarnecki, M. Slusarczyk, F. Tkachov, Phys. Rev. Lett. 93 (2004) 062001, hep-ph/0403221.
S. Bethke, Prog. Part. Nucl. Phys. 58 (2007) 351, hep-ex/0606035.
J. Charles, et al., CKMfitter Group, Eur. Phys. J. C 41 (2005) 1, hep-ph/0406184.
A.H. Hoang, A.V. Manohar, hep-ph/0509195.
P.A. Movilla Fernandez, CDF Collaboration, arXiv: 0705.3910 [hep-ex].
S. Eidelman, et al., Particle Data Group, Phys. Lett. B 592 (2004) 1.
J. Berryhill, A. Ishikawa, private communication.
A. Ali, E. Lunghi, C. Greub, G. Hiller, Phys. Rev. D 66 (2002) 034002, hep-ph/0112300.
R. Boughezal, M. Czakon, T. Schutzmeier, Phys. Rev. D 74 (2006) 074006, hep-ph/0605023.
C.W. Bauer, Z. Ligeti, M.E. Luke, Phys. Lett. B 479 (2000) 395, hep-ph/0002161.
K.S.M. Lee, Z. Ligeti, I.W. Stewart, F.J. Tackmann, Phys. Rev. D 75 (2007) 034016, hep-ph/0612156.

0 B
0 mixing and lepton flavour violation in SUSY
Bd,s
d,s
GUTs: Impact of the first measurements of s
J.K. Parry a, , Hong-Hao Zhang b

a Center for High Energy Physics, Tsinghua University, Beijing 100084, China
b School of Physics & Engineering, Sun Yat-Sen University, Guangzhou 510275, China
Received 1 March 2008; received in revised form 17 April 2008; accepted 5 May 2008
Abstract
0 B 0 mixing and lepton flavour violation in the
In this work we re-examine the correlation between Bd,s
d,s
light of recent experimental measurements in the Bs system. We perform a generic SUSY analysis of the
allowed down squark mass insertion parameter space. In the SUSY GUT scenario this parameter space is
then used to make predictions for LFV branching ratios. We find that the recent measurement for the CP
phase s excludes the lowest rates for and provides a lower bound of 3 109 for tan = 10.
Future experimental improvements in the bound on and the measurement of s will constitute a
strong test of the SUSY GUT scenario.
PACS: 12.10.-g; 12.60.Jv; 13.35.Dx; 14.40.Nd; 13.20.He
1. Introduction
In recent years there have been great efforts made in the search for new physics (NP) in the
B system which has been seen as an extremely promising place to find NP effects as deviations
of flavour-changing neutral-currents (FCNC) from their Standard Model (SM) expectations. The
first measurements of Bs B s mixing have been observed at D [1] and CDF [2] experiments
E-mail addresses: jkparry@tsinghua.edu.cn (J.K. Parry), zhh98@mail.sysu.edu.cn (H.-H. Zhang).

doi:10.1016/j.nuclphysb.2008.05.010
64
J.K. Parry, H.-H. Zhang / Nuclear Physics B 802 (2008) 6376
reporting a mass difference,

17 ps1 < Ms < 21 ps1
(90% C.L. D),

1
Ms = 17.77 0.10 0.07 ps
(CDF).
(1)
(2)
The constraints from Eq. (2) strongly restrict the allowed parameter space for the NP contribution
to Bs B s mixing [311].
The first measurement of the CP phase associated with Bs B s mixing, s , was reported by
the D Collaboration [12]. More recently, flavour-tagged measurements have now been made by
both the CDF Collaboration [13] and the D Collaboration [14]. Combining these latest results
the UTfit group [15] found that the new physics phase deviates from zero by about 3 ,

s = sNP 2s
(3)
with s = 0.018 0.001 and,
sNP = [60.9, 18.58] [156.90, 106.40]
(4)
at 95% C.L. There still remains a two fold ambiguity in the measurement due to the symmetry
in s ( s ).
In the Bd system the measured mass difference given by the Heavy Flavour Averaging
Group (HFAG) [16] is,
Md = 0.507 0.004 ps1
(5)
and the phase, d , has been measured as,

d = 0.757 0.044.
(6)
A very appealing model of NP is the supersymmetric grand unified theory (SUSY GUT). In
such models there is a deep underlying connection between quarks and leptons above the GUT
scale. For example in SU(5) the singlet down quarks are related to the lepton doublet as they live
in the same 5 representation. This deep relation between down quarks and leptons is broken at
the GUT scale but its presence may still be felt all the way down at the electroweak scale. As
a result of this relationship the right-handed down squark and left-handed slepton mixings are
related to each other. In such SUSY GUTs there is hence a deep connection between LFV rates
such as BR( ) and FCNCs such as Bs B s mixings and its associated CP phase [1720].
In this work we re-examine the correlation between FCNCs and LFV rates in SUSY GUTs.
Specifically we are interested in the connection between the allowed NP parameter space of
Bs B s mixing and the rate of . To this end we first perform a generic SUSY analysis,
d . Our analysis in generic in
of the allowed parameter space of down squark mass insertions RR
that no particular SUSY-breaking mechanism shall be assumed. Having determined the allowed
parameter space we then exploit the GUT relationship of squark and slepton mass insertions to
make realistic predictions for the rates of , e and their ratio. Recently a similar
study was undertaken [18] on the correlation between LFV and Bq mixing. We believe that the
NP contribution to Bd mixing was under-estimated in that work leading to over-estimates for the
branching ratio of e . In this work we use an analytic form for the gluino contribution to Bd
mixing derived in Appendix A in the mass insertion approximation. This allows us to correctly
compute the NP contribution to both Bd and Bs mixing. Furthermore the recent measurement of
the CP phase s has prompted a re-examination of this correlation. Crucially we find that these
recent measurements severely restrict the allowed parameter space and provide a lower bound
65
for the rate of . Finally, we explore an example of an SO(10) inspired model and discuss
the tension between the constraints of LFV and FCNCs.
2. Constraints on the SUSY contribution to Bq mixing
The B = 2 transition between Bq and B q mesons is defined as,
0 B=2 0
q
Bq Heff B q = 2MBq M12 ,
(7)
where MBq is the mass of the Bq meson. We can then define the Bq mass eigenstate difference
as,
q
q
q
Mq MH ML = 2M12
(8)
and its associated CP phase,
q
q = arg M12 .
(9)
In the Standard Model M12 is given by,

q,SM
M12
2

2
G2F MW
MBq B fB2q B Bq Vtq Vtb S0 (xt ),
2
12
(10)
where GF is Fermis constant, MW the mass of the W boson, B = 0.551 is a short-distance

QCD correction identical for both the Bs and Bd systems. The bag parameter B Bq and decay
constant fBq are non-perturbative quantities and contain the majority of the theoretical uncertainty. Vtq and Vtb are elements of the CabibboKobayashiMaskawa (CKM) matrix, and
2 ) = 2.32 0.04.
2t /MW
S0 (xt m
The NP contribution to Bq mixing may be parameterized in a model independent way as,
Mq = MqSM |1 + Rq |,
(11)
q = qSM
(12)
+ qNP
= qSM
q,NP

+ arg 1 + rq eiq ,
q,SM
where Rq rq eiq = M12 /M12 denotes the relative size of the NP contribution. The dominant SUSY contribution to Bq mixing comes from the gluino contribution which may be written
as,
d 2
d 2
d d
g
q
q
Rq = a1 (mg , x) q3
(13)
+ a4 (mg , x) q3
+ q3
+ .
RR
LL
LL q3 RR
d
mass insertions as they are expected to be
Here we have ignored terms proportional to RL,LR
heavily suppressed due to constraints from b s [17]. In Appendix A we give details of the
q
q
functions a1 and a4 .
We can now constrain both the magnitude and phase of the NP contribution, rq and q ,
through the comparison of the experimental measurements with SM expectations. From the definition of Eq. (11) we have the constraint,

Mq
=
1 + 2rq cos q + rq2
q
(14)
MqSM
for rq and q . The values for q given by the UTfit analysis [15,21] at the 95% C.L. are,
d = [0.53, 2.05],
(15)
66
s = [0.62, 1.93].
(16)
From Eq. (12) the phase associated with NP can also be written in terms of rq and q ,
rq sin q
sin qNP =
,
1 + 2rq cos q + rq2
cos qNP =
1 + rq cos q
1 + 2rq cos q + rq2
(17)
Here [15,21] gives the 95% C.L. constraints,

dNP = [16.6, 3.2] ,
sNP
(18)
= [60.9, 18.58] [156.90, 106.40] .
(19)
In order to consistently apply the above constraints all input parameters are chosen to match
those used in the analysis of the UTfit group [15,21] with the non-perturbative parameters,
1
fBs B B2s = 262 35 MeV,
(20)
= 1.23 0.06 MeV,
(21)
fBs = 230 30 MeV,
(22)
fBd = 189 27 MeV.
(23)
Now we have a clear picture of the constraints to be imposed on the rq q plane as listed
in Eqs. (14)(19). In the following sections these constraints shall be used in a generic SUSY
analysis of the mass insertion parameter space, followed by a study of the correlation of Bq
mixing and LFV rates in the minimal SUSY GUT.
3. Correlation between Bq mixing and LFV rates in SUSY GUTs
In many ways the prototypical GUT is SU(5) as it is the minimal group which can unify the
gauge group of the Standard Model. In the SU(5) GUT the quarks and leptons are placed into
10 = (Q, uc , ec ) and 5 = (L, d c ) representations. Due to the symmetry of this simple SUSY
GUT there exists relations amongst the slepton and squark soft SUSY breaking masses,
m210 = m2Q = m2U = m2E ,
m25 = m2L = m2D .
(24)
These relations hold for scales at and above the GUT scale. Interestingly this implies that lefthanded slepton mixing and right-handed down squark mixing are related. This relation can still
be felt very strongly at the electroweak scale in the correlation of LFV rates and FCNCs.
Further renormalization group (RG) evolution down to the electroweak scale has very little
effect on the size of these off-diagonal elements [17]. So we may assume that these GUT scale
values are approximately equal to their values at the electroweak scale. Hence it is a fair approximation to assume that (m2 )ij (m2 )ij at the electroweak scale. Then the (ijd )RR (m2 )ij /m2q
D
contributions to FCNC and (ijl )LL (m2 )ij /m2 contributions to LFV are clearly correlated. We
L
l
may explicitly write the rate of li lj in the very suggestive form [18],
4
3 mq d 2 2
BR(li lj ) 2 8 ij RR tan ,
GF MS
where mq is the average squark mass and MS is the typical SUSY scale.
(25)
67
We also need to consider the RGE effects of the CKM mixings in the left-handed down squark
matrices. These effects can be approximated as,
ijd

LL
1 2 (3m20 + A20 ) M
Y V Vtj
ln
.
MW
8 2 t ti
m2q
(26)
Here m0 is the universal scalar mass, A0 is the universal A-term and M is the scale at which the
flavour blind soft SUSY breaking is communicated. These effects may be quite important in the
d d contribution to Eq. (13) due to the typically large value of a 100a . We shall take a
LL
4
1
RR
minimal approach and assume M to be the GUT scale 2 1014 GeV, in which case the mass
d )
d
2
3
insertions are of the order, (23

LL and (13 )LL . Larger values of M will lead to an
d
even more restricted parameter space for (13,23 )RR .
If we assume that the second term of Eq. (13) dominates then we may derive the relation,

2 s 2 d
( ) 2 ( d ) 2
Rs
a4 23 LL 23 RR 2 BR( ) ,
(27)
a d ( d ) ( d )
R
BR( e )
d
4
13 LL
13 RR
here we have made use of the ratio, |a4s |/|a4d | Vtd2 /Vts2 2 , as derived in Appendix A. This
relation differs from that derived in [18] where it was assumed that, a4s /a4d = 1. In Appendix A
we show the full form of the functions a1 and a4 where it is shown that the correct ratio is
d )
|a4s |/|a4d | Vtd2 /Vts2 2 . As a result the allowed parameter space for the mass insertion (13
RR
d
2
is suppressed relative to (23 )RR by a factor as shown above.
4. Numerical results and discussion
In this section we shall show the results of numerical calculations of our generic SUSY anald
ysis of the allowed parameter space for the mass insertions (13,23
)RR . We then make use of this
allowed parameter space to study the correlation of FCNCs and LFV rates.
For the numerical analysis we fixed the values of tan = 10, MS = m0 = M1/2 = mq =
(300, 500) GeV, A0 = 0 and the ratio x m2g /m2q = 1. Taking these values we scan over
d
)RR and require fits to the values of q and qNP as described in the previous section.
(23,13
d )
In Fig. 1 we show the allowed parameter space for the mass insertion (23
RR as dictated by
the constraints imposed from Ms (red/blue) and s (green/pink). The allowed regions form
d )
rings in the complex (23
RR plane with wedges cut into it corresponding to allowed regions
from the measurement of s . It is clear therefore that the recent measurements of s substantially restrict the allowed parameter space and provide important information. In the SUSY-GUT
scenario the same mass insertion also contributes to the LFV decay , the resulting bounds
from which can be seen by the solid black line shown in Fig. 1. The two black lines correspond
d )
to the values m0 = 300, 500 GeV and clearly exclude considerable regions of the (23
RR parameter space. In fact one of the two allowed regions is strongly disfavoured in the SUSY-GUT
scenario.
d )
Fig. 2 shows the respective allowed parameter space of the mass insertion (13
RR constrained
d
by the measurements of Md and d . For (13 )RR we can see that the measurement of d has
d )
reduced the allowed parameter space to a small slice in the complex (13
RR plane. Increasing
the size of mq leads to an increase in the allowed parameter space due to a suppression of the
functions a1,4 1/m2q .
68
d )
Fig. 1. Allowed parameter space for the mass insertion (23
RR , with mq = 300 (blue/pink) and 500 (red/green) GeV
respectively. Red/blue points are constrained by the measurement of Ms while green/pink points have the extra constraint from the measurement of s both at the 95% C.L. Solid black lines show the respective constraints for each value
of m0 from bounds on from Belle [22]. (For interpretation of the references to colour in this figure legend, the
reader is referred to the web version of this article.)
From these allowed regions of parameter space we can plot the corresponding allowed regions
of the rq q parameter space. These are shown in Figs. 3 and 4. Without the restrictions from
the CP phases d,s the allowed regions of parameter space form loops in the rq q plane. The
constraints from the CP phases represent slices as shown in [6,10]. The accurate measurement
of d leaves the allowed region of parameter space as a thin slice shown by the green points in
Fig. 3. On the other hand, the present measurement of s has a two fold ambiguity shown by the
two green wedges in Fig. 4.
Next we would like to discuss the implications of this allowed parameter space on predictions
for the LFV rates of and e . We have shown that in a SUSY GUT the mass
insertions associated with the off-diagonal down squark mass matrix elements and those of the
sleptons are clearly related. This relation leads to a strong correlation between the Bq mixing and
LFV rates. Here we would like to expose this correlation in making realistic predictions for LFV
rates and to extract information about the NP CP phase.
Let us first consider the 23 sector. Fig. 5 shows the predicted rate for plotted against
the NP phase sNP . The possibility of sNP = 0 is excluded at the 3 level [15], which also
excludes the lowest rates for at the centre of the plot where sNP 0. The general
feature of the plot is that larger LFV rates are predicted for larger NP phases. Without the input of
s this plot is symmetric in sNP sNP and extends for all sNP (, ). The inclusion of the
s constraint drastically reduces the allowed parameter space and the resulting prediction space,
as shown in Fig. 5. These predictions are independent of the choice of MS = mq due to the 1/MS2
dependence of both BR(li lj ) and the functions a1,4 shown in Appendix A. There are two
allowed regions, one with large NP phase and one with smaller NP phase. From Fig. 5 it is clear
69
d )
Fig. 2. Allowed parameter space for the mass insertion (13
RR , with mq = 300 (small) and 500 (large) GeV respectively.
Red/blue points are constrained by the measurement of Md while green/pink points have the extra constraint from the
measurement of d both at the 95% C.L. (For interpretation of the references to colour in this figure legend, the reader
is referred to the web version of this article.)
Fig. 3. Allowed region of rd d parameter space for the NP contribution to Bd mixing. Red points agree with the Md
constraint while green points also agree with that of d both at the 95% C.L. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
70
Fig. 4. The allowed region of the rs s parameter space for the NP contribution to Bs mixing. Points in red agree with
the constraint from Ms , while green points also agree with the s measurement both at the 95% C.L. (For interpretation
of the references to colour in this figure legend, the reader is referred to the web version of this article.)
Fig. 5. Predictions for BR( ) from constraints on Bs mixing. The red points conform to the Ms constraint while
green points show the additional constraint from s , both at 95% C.L. The present experimental bound is shown by the
solid black horizontal line. (For interpretation of the references to colour in this figure legend, the reader is referred to
the web version of this article.)
71
Fig. 6. Predictions for BR( e ) from constraints on Bd mixing.
that the upper bound on disfavours the large phase solution. For the second region, there
is a lower bound on the BR( ) 3 109 . This second allowed region is just below the
present experimental bounds for the branching ratio, BR( ) < 6.8 108 BaBar [23,24]
and BR( ) < 4.5 108 Belle [22]. From these plots it is clear that improvements of
the bound on in conjunction with an improved measurement of s will provide a much
stricter test of the SUSY-GUT scenario.
In the 13 sector, we can see from Fig. 2, that the allowed parameter space for the mass inserd )
tion (13
RR is far more restricted. This is due to the smaller mass difference in the Bd system, as
a result the functions a1 and a4 are enhanced by the ratio, Md /Ms Vtd2 /Vts2 2 , see Appendix A. This restriction leads to a much more suppressed prediction for the rates of e .
These rates are plotted in Fig. 6 against the allowed values of the NP phase dNP . Including
the constraints from d the predicted values are in the region of 1014 109 . The present experimental bounds for the branching ratio are BR( e ) < 1.1 107 BaBar [23,24] and
BR( e ) < 1.2 107 Belle [22]. Looking at Fig. 6 we can see that the allowed region of
d )
the (13
RR parameter space predicts rates well below the present experimental bounds. Therefore the present bounds have no impact on the allowed parameter space.
The ratio of the branching fraction of and e is less dependent on the SUSY
parameter space and so it is interesting to consider the allowed size of this ratio. Using the
d )
d
NP
parameter space of (13
RR and (23 )RR allowed by the constraints of d,s and d,s we can
make predictions for the ratio BR( )/ BR( e ). Fig. 7 shows the resulting plot of
BR( )/ BR( e ) plotted against the prediction for the NP CP phase sNP . We can see
that the numerical results show this ratio ranging from 0.01 up to tens of thousands. Applying the
constraints from the CP phases d,s implies that this ratio must lie in the region 10. This large
d )
ratio is again due to the suppression of the (13
RR allowed parameter space due to the smallness
of the mass difference in the Bd system as opposed to the larger mass difference in the Bs system.
72
Fig. 7. BR( )/ BR( e ) from constraints on Bs and Bd mixing. Red points are from the Md,s constraint
only, green points are for both Md,s and d,s constraints. (For interpretation of the references to colour in this figure
legend, the reader is referred to the web version of this article.)
Here our results disagree with those presented in [18] due to their erroneous assumption of the
ratio ais /aid = 1.
Let us now look at a specific example of a simple SUSY GUT. The example we pick is an
SO(10) inspired model as introduced in [19,25]. In this model there are two Higgs representations
one for the up/neutrino and down/charged lepton Yukawa couplings separately. The neutrino
Yukawa coupling is then related to the up quark Yukawa, where two limiting cases are possible:
(i) Y has CKM like mixing, (ii) Y has MNS like mixing. It is assumed that SO(10) breaks to
SU(5) at the scale M10 and further breaks to the SM gauge group at the scale MGUT . Assuming
that the flavour blind SUSY breaking occurs at the scale M = M10 = 1017 GeV, then the mass
insertions can be written as,
3 2
M10
yt Vj 3 Vi3 ln
,
2
MGUT
8
l
3
M10
ij LL 2 yt2 Vj3 Vi3 ln
,
MR
8
ijd
RR
(28)
(29)
where the matrix V is VCKM and UMNS for cases (i) and (ii) respectively. The scale MR
1015 GeV is the scale of the right-handed neutrinos. Case (i) produces relatively small LFV rates
l )
d
with (23
LL 0.008 and (23 )RR 0.003. The present bound on the rate of leads to
the weak bound m0 200 GeV. Alternatively the contribution to Bs mixing is also very small,
for m0 200 GeV we have too small a value |Rs | 0.2. Case (ii) produces large LFV rates
l )
d
with (23
LL 0.25 and (23 )RR 0.08. Here the present bound on the rate of leads
to the strong bound m0 1000 GeV. Here again the contribution to Bs mixing is very small, for
m0 1000 GeV we once again have too small a value |Rs | 0.2.
73
5. Summary
In this work we have re-examined the constraints imposed on the parameter space of NP
contributions to Bs and Bd mixing in the light of recent measurements of Ms and its associated
CP phase s . These constraints were then imposed to make a generic SUSY analysis of the
d )
d
allowed parameter space of the mass insertions (13
RR and (23 )RR .
In SUSY GUTs there is a deep underlying connection between (s)quarks and (s)leptons. In
such a theory we should therefore expect that the flavour mixings observed in the quark and
lepton sectors are correlated. Assuming that such a SUSY GUT exists, we have re-examined the
correlation between LFVs and FCNCs.
From our numerical analysis we have found that the allowed parameter space for the mass
d )
insertion (13
RR is particularly restricted by the present measurements of Md and d . This is
due to the small mass difference in the Bd system compared to the Bs system. As a result, the
predicted branching ratio for e is particularly suppressed. On the other hand, the present
experimental determination of Ms and particularly s are not yet so restrictive on the allowed
d )
values of (23
RR . The larger mass difference in the Bs system also means that the predicted
values of the branching ratio of are much larger and close to the present bounds from the
B factories. The main conclusion of this work is that the recent measurement of s substantially
restricts the allowed parameter space and excludes the region where the lowest rates for
are to be found, leading to a lower bound of 3 109 for tan = 10. Future experimental
improvement of the bound on the decay rate and the measurement of s will lead to a
strong test of the SUSY-GUT scenario. We have also considered a specific example of a SUSY
SO(10) GUT and found there to be significant tension between the constraints from FCNCs and
LFV decay rates.
Acknowledgements
J.K.P. would like to thank E. Kou for useful discussions concerning the gluino contribution
to Bs mixing. H.H.-Z. thanks Prof. Y.P. Kuang and Prof. Q. Wang for their encouragement.
This work is supported by the National Natural Science Foundation of China under Grant
No. 10747165, and Sun Yet-Sen University Science Foundation.
0 B
0 mixing in SUSY models
Appendix A. Bd,s
d,s
To examine the contributions from new physics we may write the B = 2 effective Hamiltonian in the following form,
B=2
Heff
=
Ci ()Qi () + h.c.
(A.1)
i=1
The operators are defined as,
qL bL ,
Q1 = qL bL
(A.2)

qR bL ,
Q2 = qR bL

Q3 = qR bL qR bL ,

qL bR ,
Q4 = qR bL
(A.3)
(A.4)
(A.5)
74

Q5 = qR bL qL bR
.
(A.6)
We shall also need to define the hadronic matrix elements of the above operators such that,
1
Bq |Q1 |B q = MBq fB2q B1 (),
3
5
Bq |Q2 |B q = RBq MBq fB2q B2 (),
24
1
24
1
4
1
12
(A.7)
(A.8)
(A.9)
(A.10)
(A.11)
MB
q
where we have defined RBq = ( mb ()+m
)2 . The values of the bag parameters have been calq ()
culated on the lattice and can be found in [26].
The dominant SUSY contribution is from the gluino, which gives the following Wilson coefficients [2732]
C1 (MS ) =
g
C4 (MS ) =
g
C5 (MS ) =
d 2
s2 (MS )
24xf6 (x) + 66f6 (x) q3
,
LL
2
216mq
(A.12)
d d
s2 (MS )
504xf6 (x) 72f6 (x) q3
,
LL q3 RR
2
216mq
(A.13)
d d
s2 (MS )
24xf6 (x) + 120f6 (x) q3
.
LL q3 RR
2
216mq
(A.14)
Here only the LL and RR terms have been kept as the contributions from RL and LR are
suppressed.
The above Wilson coefficients are evaluated at the scale MS = (Mg + Mq )/2 =
( x + 1)Mq /2.
These Wilson coefficients need to be renormalization group (RG) evolved to the scale of the
bottom quark. Using the magic numbers from [32], we have,

(,)
(,) ai
bi
C (MS )
+ ci
C (b ) =
(A.15)
i
with = s (MS )/s (t ). Let us now write the Bs B s mixing parameter

q,SM

q

g
Mq 2M12 = 2M12 (1 + Rq ) = MqSM 1 + Rq + ,
q,NP
where Rq rq eiq = M12
q,SM
/M12
(A.16)
parameterizes the contribution from new physics. We cong
q,g
q,SM
sider the dominant effect to come from the gluino Rq = M12 /M12 . From the above effective
Hamiltonian, Wilson coefficients, operator matrix elements and RG evolution we can write,
d 2
d 2
d d
g
q
q
+ a4 (mg , x) q3
+ q3
+
Rq = a1 (mg , x) q3
RR
LL
LL q3 RR
(A.17)
75
have we have only kept the LL, RR terms as they will dominate. The complete form of the
functions a1 (mg , x) and a4 (mg , x) are as follows,
(1,1)
s2 (MS ) 1
(1,1) ai
bi
,
A mBs fB2s B1 ()
+ ci
2
3
216mq
i

(4,4)
s2 (MS )
q,SM q
(4,4) ai 1
2
bi
B4 ()
m
f
R
+ ci
M12 a4 (mg , x) =
Bs Bs Bq B
2
4
216 mq
i
(4,5)
(4,5) ai 1
+A
bi
B4 ()
+ ci
4
i
(5,4)
(5,4) ai 1
+B
bi
+ ci
B5 ()
12
i

(5,5)
(5,5) ai 1
bi
+ ci
B5 () ,
+A
12
q,SM q
a1 (mg , x) =
M12
(A.18)
(A.19)
where we have defined, A = 24xf6 (x) + 66f6 (x) and B = 504xf6 (x) 72f6 (x) where,
6(1 + 3x) ln x + x 3 9x 2 9x + 17
,
6(x 1)5
6x(1 + x) ln x x 3 9x 2 + 9x + 1
f6 (x) =
3(x 1)5
f6 (x) =
and x = m2g /m2q .
(A.20)
(A.21)
(1,1)
(4,4)
The dominant contributions to the above functions come from b1

= 0.865, b4
= 2.87,
(4,5)
(5,4)
(5,5)
b4
= 0.961, b4
= 0.09, and b5
= 0.863, with ai = (0.286, 0.692, 0.787, 1.143,
0.143).
So we can write the approximate functions as,
s2 (MS ) 1
(A.22)
mBs fB2s B1 ()A0.8650.286 ,
216m2q 3

1

s2 (MS )
q,SM q
2
RBq mBs fBs B2.871.143 + A0.9611.143 B4 ()
M12 a4 (mg , x)
2
4
216mq

1

B5 () .
+ B0.091.143 + A0.8630.143
(A.23)
12
q,SM q
a1 (mg , x)
M12
We can see that the values of the functions a1,4 are not the same for the Bs and Bd systems. From
Eqs. (A.18), (A.19) we can approximate the ratio of these functions as,
|a1s |
|a1d |
|a4s |
|a4d |
MdSM Vtd2
2 2
MsSM
Vts
(A.24)
so that the value of the functions a1 and a4 are enhanced in the Bd relative to those derived
in the Bs system. This enhancement will lead to a more constrained parameter space of the
13 mass insertion relative to the 23 sector. It is also important to notice that a1 and a4 are complex
d,SM
,
parameters. This is particularly important in the case of the Bd system where the phase of M12
dSM = 2, is large.
76
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
V.M. Abazov, et al., D Collaboration, Phys. Rev. Lett. 97 (2006) 021802, hep-ex/0603029.
A. Abulencia, et al., CDF Collaboration, Phys. Rev. Lett. 97 (2006) 242003, hep-ex/0609040.
B. Dutta, Y. Mimura, Phys. Rev. Lett. 97 (2006) 241802, hep-ph/0607147.
J.K. Parry, Nucl. Phys. B 760 (2007) 38, hep-ph/0510305.
J.K. Parry, hep-ph/0606150.
P. Ball, R. Fleischer, Eur. Phys. J. C 48 (2006) 413, hep-ph/0604249.
J.K. Parry, Mod. Phys. Lett. A 21 (2006) 2853, hep-ph/0608192.
A. Lenz, U. Nierste, JHEP 0706 (2007) 072, hep-ph/0612167.
A. Freitas, E. Gasser, U. Haisch, Phys. Rev. D 76 (2007) 014016, hep-ph/0702267.
P. Ball, hep-ph/0703214.
S. Trine, arXiv: 0710.4955 [hep-ph].
V.M. Abazov, et al., D Collaboration, Phys. Rev. D 76 (2007) 057101, hep-ex/0702030.
T. Aaltonen, et al., CDF Collaboration, arXiv: 0712.2397 [hep-ex].
V.M. Abazov, et al., D Collaboration, arXiv: 0802.2255 [hep-ex].
M. Bona, et al., UTfit Collaboration, arXiv: 0803.0659 [hep-ph].
H.Y. Cheng, Int. J. Mod. Phys. A 21 (2006) 4209, hep-ph/0603003.
M. Ciuchini, A. Masiero, P. Paradisi, L. Silvestrini, S.K. Vempati, O. Vives, Nucl. Phys. B 783 (2007) 112, hepph/0702144.
K. Cheung, S.K. Kang, C.S. Kim, J. Lee, Phys. Lett. B 652 (2007) 319, hep-ph/0702050.
D. Chang, A. Masiero, H. Murayama, Phys. Rev. D 67 (2003) 075013, hep-ph/0205111.
T. Moroi, Phys. Lett. B 493 (2000) 366, hep-ph/0007328.
M. Bona, et al., UTfit Collaboration, JHEP 0803 (2008) 049, arXiv: 0707.0636 [hep-ph].
K. Abe, et al., Belle Collaboration, hep-ex/0609049.
A. Masiero, S.K. Vempati, O. Vives, Nucl. Phys. B 649 (2003) 189, hep-ph/0209303.
W.M. Alberico, A. Drago, C. Ratti, Nucl. Phys. A 706 (2002) 143, hep-ph/0110091.
P. Ball, S. Khalil, E. Kou, Phys. Rev. D 69 (2004) 115011, hep-ph/0311361.
F. Gabbiani, A. Masiero, Nucl. Phys. B 322 (1989) 235.
J.S. Hagelin, S. Kelley, T. Tanaka, Nucl. Phys. B 415 (1994) 293.
E. Gabrielli, A. Masiero, L. Silvestrini, Phys. Lett. B 374 (1996) 80, hep-ph/9509379.
F. Gabbiani, E. Gabrielli, A. Masiero, L. Silvestrini, Nucl. Phys. B 477 (1996) 321, hep-ph/9604387.
D. Becirevic, et al., Nucl. Phys. B 634 (2002) 105, hep-ph/0112303.

CP observables with spinspin correlations

in chargino production
A. Bartl a,b , K. Hohenwarter-Sodek a , T. Kernreiter a, ,
O. Kittel c , M. Terwort d
a Faculty of Physics, University of Vienna, Boltzmanngasse 5, A-1090 Wien, Austria
b AHEP Group, Institut de Fsica CorpuscularC.S.I.C., Universitat de Valncia, Edifici Instituts dInvestigaci,
Apt. 22085, E-46071 Valncia, Spain

c Physikalisches Institut der Universitt Bonn, Nuallee 12, D-53115 Bonn, Germany
d Institut fr Experimentalphysik, Universitt Hamburg, Notkestrae 85, D-22607 Hamburg, Germany
Received 27 February 2008; received in revised form 10 April 2008; accepted 29 April 2008
Abstract
We study the CP-violating terms of the spinspin correlations in chargino production e+ e 1 2 ,
and their subsequent two-body decays into sneutrinos plus leptons. We propose novel CP-sensitive observables with the help of T-odd products of the spinspin terms. These terms depend on the polarizations of
both charginos, with one polarization perpendicular to the production plane. We identify two classes of CPsensitive observables; one requires the reconstruction of the production plane, the other not. Our framework
is the Minimal Supersymmetric Standard Model with complex parameters.
PACS: 11.30.Pb; 11.30.Er; 14.80.Ly
Keywords: Supersymmetry; CP violation; Supersymmetric partners of known particles
1. Introduction
Supersymmetric (SUSY) extensions of the Standard Model (SM), like the Minimal Supersymmetric Standard Model (MSSM) [1], give rise to new sources of CP violation [2]. From a
E-mail address: tkern@hephy.oeaw.ac.at (T. Kernreiter).

doi:10.1016/j.nuclphysb.2008.04.031
78
A. Bartl et al. / Nuclear Physics B 802 (2008) 7791
mathematical point of view, this means that in the SUSY Lagrangian complex parameters enter
whose phases cannot be removed by redefining the fields. The presence of CP phases can drastically alter the phenomenology of the underlying model (for a recent review, see [3]). For instance,
contributions of SUSY CP phases to the electric dipole moments (EDM) of electron, neutron, and
that of the atoms 199 Hg and 205 Tl can be close or beyond the present experimental upper bounds
[4,5], and thus in turn constrain the size of these CP phases [4,5]. These constraints, however, are
strongly model dependent, see e.g. [5]. Thus measurements of CP observables outside the EDM
sector are necessary to independently determine or constrain the CP phases. Furthermore, nonvanishing phases can significantly change masses, cross sections and decay branching ratios of
SUSY particles, compared to the real case, see e.g. [6,7]. Hence, in determining the underlying
model parameters, the effect of their CP phases has to be taken into account. The phases could
be measured once supersymmetric particles are accessible at future colliders. A genuine signal
for CP violation would be the measurement of non-vanishing CP-sensitive observables.
In this paper, we propose CP-sensitive observables in chargino production
e+ e 1 2 ,
(1)
1+ 1 , 2+ 2
the CP-sensitive terms in the amwithin the MSSM. For the processes e+ e
plitude squared vanish at tree level since all coupling factors are real [8,9]. The chargino mass
matrix, in the weak basis, is given by

M2
2mW sin
MC =
(2)
,
2mW cos
with M2 the SU(2) gaugino mass parameter, and tan the ratio of the vacuum expectation values
of the two Higgs fields. M2 and tan can be chosen real and positive, while the higgsino mass
parameter can be complex = ||ei . At tree level, is the only CP-violating phase that
gives rise to CP-sensitive observables in chargino production, while phases from other sectors
can contribute at loop level [10,11]. For example, phases of the gaugino mass parameter M1 ,
and the trilinear coupling parameters At in the stop-sector, lead to rate differences in 1+ 2
production and that of the charge conjugated pair 1 2+ at the percent level [10]. For chargino
decays, rate asymmetries of the partial chargino decay widths can exceed 10%, mainly due to the
phases of M1 and At,b [12] (see also [13]).
Another class of promising CP-sensitive observables are based on so-called T-odd correlations
(or T-odd products), see e.g. [14]. They can give rise to CP-violating effects already at tree-level,
and therefore suffer not from loop suppression as rate asymmetries. Previous studies of CPsensitive observables, based on T-odd products in chargino production and decay, have been
focussing on the spin correlations between production and decay of only one chargino [8,9]. The
corresponding terms in the amplitude squared involve the polarization vector perpendicular to the
production plane of one of the produced charginos. Such a transverse polarization component is
a genuine signal of CP violation. The transverse polarization is then retrieved from asymmetries
in the azimuthal distribution of the decay products, which can be as large as 30%, even for small
of order /10 [9]. In such an analysis of the spin-correlations, the polarization, i.e. the decay,
of only one chargino needs to be considered. However, if the decays of both charginos are taken
into account, one can probe their spinspin correlations [15]. These are terms in the amplitude
squared that include the polarization vectors of both charginos. The angular distributions of the
decay products of the two charginos are correlated to one another due to total angular momentum
conservation. Spinspin correlations in chargino production and decay have been utilized for the
determination of CP-even coupling factors [15,16]. Moreover, spinspin correlations have been
79
used for the definition of CP-sensitive observables in the decays of third generation squarks
[17,18].
In the present paper, we propose novel CP-sensitive observables with the help of T-odd
products in the chargino spinspin correlation terms. We take the decay of both charginos into
account, and consider, for definiteness, their subsequent leptonic two-body decays
i+ + ,
j ,
i, j = 1, 2 (i = j ), , = e, .
(3)
By analyzing the CP-odd parts of the spinspin terms, we find two independent T-odd products.
The first one includes the momenta of the beams and those of the two decay leptons. Thus a
reconstruction of the production plane is not necessary for a measurement of the corresponding CP-sensitive observables. From an experimental perspective, this seems to be advantageous
compared to CP-sensitive observables in chargino production and leptonic decays, where such
a reconstruction is essentially required [8,9]. The second T-odd product which we find involves
the momenta of the charginos, requiring the reconstruction of the production plane. We consider
two sorts of CP-sensitive observables which we obtain from the spinspin correlations. Defining
their theoretical statistical significances, we can make a quantitative comparison of their accessibility.
The paper is organized as follows. In Section 2, we present analytical formulae for the amplitude squared of chargino production and decay e+ e i+ j + . We identify the
T-odd products that are involved in the spinspin terms of the amplitude squared in Section 3,
and define the associated CP-sensitive observables in Section 4. We present numerical results in
Section 5. In Section 6, we give a summary and conclusions.
2. Cross section
Chargino production e+ e i+ j proceeds via , Z exchange in the s-channel, and
exchange in the t-channel, see the Feynman diagrams in Fig. 1. The exchange vanishes for
non-diagonal chargino production e+ e 1 2 . For diagonal chargino production, e+ e
i+ i , the squared amplitude contains no CP-violating terms at tree level [8,9].
We write the differential cross section for chargino production and decay e+ e i+ j
+ , , = e, , generically as
d =
1
|T |2 dLips,
2s
(4)
with the center-of-mass energy s, and the Lorentz invariant phase space element dLips, which
can be found in [9]. The amplitude squared |T |2 was calculated in Ref. [15] in the spin density
Fig. 1. Feynman diagrams for chargino production e+ e i+ j [15].
80
matrix formalism1

3

2
2
a

|T | = 4 ( i ) ( j ) P Di Dj +
Pa D
Dj
i
2
a=1
3

b
Pb D
Di
j
b=1
3

a
b
Pab D
D
i
j

(5)
a,b=1
with the propagators ( i,j ) = 1/[p2 i,j m2i,j + imi,j i,j ] of the decaying charginos. The
amplitude squared has contributions from chargino production (P ) and decay (D). The terms P
and Di , Dj are those parts of the spin density production and decay matrices, respectively, that
a depend
are independent of the polarizations of the charginos. The contributions Pa and D
i
a
on the polarization basis vectors s + (for their definition see Appendix B, Eq. (B.3)) of the
i
b depend on the polarization basis vectors s b of the

decaying chargino i+ , while Pb and D
j
decaying chargino j . We choose a coordinate frame such that a, b = 3 denote the longitudinal
polarizations, a, b = 1 the transversal polarizations in the production plane, and a, b = 2 the
a and
transversal polarizations perpendicular to the production plane. The quantities Di , Dj , D
i
b
a
b
Dj are given in Appendix A. The full expressions for the quantities P , P , P and Pab can
be found in Ref. [15].
The contributions to the amplitude squared which depend on the polarizations of both
charginos are the spinspin correlation terms Pab . The T-odd parts of the spinspin correlation terms are from pure Z exchange (ZZ) and from Z interference (Z e ), and are those
which include one chargino spin vector with a component perpendicular to the production plane,
i.e., those with ab = 12, 21, 23, 32 [15]
2

g 4
(Z) L2e c+ + Re2 c+ Im Oij L Oij R f ab ,
4
cos W

g4
(Z)(e ) Le c+ Im Vi1 Vj 1 Oij R f ab ,
Pab (Z e ) =
2
2 cos W
Pab (ZZ) =
(6)
(7)
where the left and right chiral couplings of the charginos to the Z boson are
1
Oij L = Vi1 Vj1 Vi2 Vj2 + ij sin2 W ,
(8)
2
1
Uj 1 Ui2
Uj 2 + ij sin2 W ,
Oij R = Ui1
(9)
2
with W the weak mixing angle and the unitary 2 2 mixing matrices U and V which diagonalize the chargino mass matrix, Eq. (2), U MC V 1 = diag(m1 , m2 ). In Eqs. (6) and (7), g is
the SU(2) weak coupling constant, Le = 1/2 + sin2 W , Re = sin2 W , (Z) = i/(s m2Z ),
(e ) = i/(t m2 e ), with s = (pe + pe+ )2 , t = (pe p )2 , and m e (mZ ) is the mass of the
j
electron sneutrino (Z boson). The dependence on the beam polarizations is given by the factors
c+ = (1 + P )(1 P+ ),
c+ = (1 P )(1 + P+ ),
1 For a detailed discussion of the spin density matrix formalism, we refer to Ref. [19].
(10)
81
where P and P+ are the degrees of longitudinal polarization of the electron and positron beam,
respectively, with 1 P 1. The kinematical dependence of the spinspin correlation terms,
Eqs. (6) and (7), is given by the function [15]
b,
b, a,
f ab = s sa,
+ p + pe (pe+ p ) s s + pe+ p (p + pe )

a,
+ s + p + pe p pe+
j
i
i
sb
j

b,
+ s p + pe+ p pe
j
i
j

sa + ,
(11)
with 0123 = 1.
Note that the spinspin correlation terms in Eqs. (6) and (7) depend on the imaginary parts of
the products of chargino couplings, Im(Oij L Oij R ) and Im(Vi1 Vj 1 Oij R ), and thus are manifestly
CP-sensitive, i.e., sensitive to the phase of the chargino sector. We also give the spinspin
correlation terms in the laboratory system in Appendix C.
3. Identifying the T-odd products in the spinspin correlation terms
For an identification of the T-odd products in chargino production and decay, we consider the
kinematical dependence of the spinspin correlation terms of the amplitude squared, Eq. (5),
3

a
b
Pab D
D
i
j
a,b=1
3

f ab sa + p+ sb p
a,b=1
= p p+ p + pe (pe+

i
p ) + p p+ p pe+ (pe

j
j
p + )
i

+ p+ p + pe p (pe+ p ) + p p + pe+ p (pe p+ ) ,

j
(12)
a and b , respectively, see

where the scalar products (sa + p+ ) and (sb p ) appear in D
Dj
i
i
Eq. (A.2). We have used the explicit expression for f ab , Eq. (11), and the completeness relation
for the chargino spin vectors [15,19]

c
c,
s
k
sc,
k
= g
p p
k
m2k
(13)
If we now substitute in the center-of-mass system the chargino 3-momenta by the corresponding
lepton 3-momenta p + p + , p p , on the right-hand side of Eq. (12), we find the
i
j
T-odd product
OT = p e (p + + p )p e (p + p ),
(14)
of the unit momentum vectors p = p/|

p|.
If we do not replace the chargino momenta, we find
an additional T-odd product
prod
OT
= (p e p )p e (p p + ) + (p e p + )p e (p p ).
j
prod
(15)
Since the T-odd product OT includes the chargino momentum p , it will require the reconj
struction of the production plane. However, as we will see later, this product gives rise to larger
CP-sensitive observables. Note that in writing all momentum vectors as unit vectors, the T-odd
prod
products OT and OT are dimensionless.
82
4. CP-sensitive observables
In this section, we define our CP-sensitive observables, which depend on the T-odd parts of
we define its
the spinspin correlations for chargino production and decay. For an operator O,
expectation value by

|2 dLips
O|T
1
d
=
O
dLips.
O =
(16)
2
dLips
|T | dLips
We now define two classes of CP observables; one class requires the reconstruction of the
chargino momenta, the other class not.
4.1. CP-sensitive observables without the knowledge of the production plane
Using Eq. (16) and the T-odd product OT , Eq. (14), we define the two CP-sensitive observables

OT and Sgn(OT ) .
(17)
The integration in Eq. (16) with OT and Sgn(OT ) as operators, projects out the CP-sensitive parts
in the spinspin correlation terms of the amplitude squared. The contributions from the terms
a D and b b D ,
of the spin correlations between production and decay in Eq. (5), Pa D
j
P Dj i
i
cancel each other. The observable
Sgn(OT ) represents an up-down asymmetry, which gives
the relative number of events for which the sign of the T-odd product OT in Eq. (14) is positive
(N+ /N ), subtracted from the relative number of events where it is negative (N /N ).
OT , on
the other hand, gives the expectation value of the momentum configuration OT , Eq. (14), for the
event sample.
Note that, since OT does not include the chargino momenta, the CP-sensitive observables in
Eq. (17) can be probed without the knowledge of the production plane. Note further, that for
the production of the charge conjugated pair of charginos, e+ e i j+ + , the
observables
OT and
Sgn(OT ) change sign. In order that the two observables from the two
chargino production processes e+ e 1 2+ and e+ e 2 1+ , do not sum up to zero, one
has to distinguish from which chargino ( 1 or 2 ) the final state leptons originate. This can
be achieved by using the different energy distributions of the leptons, since their kinematical
limits depend on the mass of the decaying chargino. In Section 4.4, we give a numerical example
showing how the leptons can be distinguished by using their different energy distributions.
4.2. CP-sensitive observables which require the knowledge of the production plane
prod
For the T-odd product OT , Eq. (15), we analogously obtain the CP observables
prod

prod
OT
and Sgn OT
.
(18)
The CP-sensitive observables in Eq. (18) not only receive contributions from the spinspin correa D
lation terms, but also from the spin correlation terms between production and decay, Pa D
j
i
b
b
and P Dj Di , see Eq. (5).
prod
Since the T-odd product OT includes the chargino momenta, a measurement of the CPsensitive observables requires the reconstruction of the production plane. The extent to which
such a reconstruction can be accomplished depends on the decay pattern of the produced
83
charginos. For their subsequent two-body decays which we consider here, e+ e 1 2

+ , the chargino momentum three-vector can be reconstructed up to a sign ambiguity
in its second component, if the masses of the involved particles are known [2022]. The two
solutions (true and false) can then be combined using statistical methods, as shown in [21].
4.3. Theoretical statistical significances
We have defined two kinds of CP-sensitive observables,
O and
Sgn(O), based on the
prod
T-odd products O = OT , OT . The observable
O is obtained by matching the complete kinematical (angular) dependence of the spinspin correlation terms in the amplitude squared. In the
literature, this technique is known by the name optimal observables [22,23]. In order to compare
the two kinds of observables, we define their theoretical statistical significances. A comparison
of the numerical values of
O and
Sgn(O) alone cannot be used to decide which observable
is more sensitive to the CP phases. The statistical significances also allow us to compare the obprod
servables which are based on the T-odd product OT , with those which are based on OT , which
includes the chargino momentum.
where O = OT , Oprod or
The theoretical statistical significance of the CP observable
O,
T
prod
O = Sgn(OT ), Sgn(OT ), is defined by [22,24]
=
S[O]
|
O|
N
,
O 2
(19)
with the number of events N = 8 (e+ e 1+ 2 ) BR( 1+ e e+ ) BR( 2

e e ) L, where L denotes the integrated luminosity. The factor 8 appears since there are 4
possibilities to sum over the lepton flavors = e, ,
and two charge assignments for chargino
L is called effective asymmetry in [22].

production (e+ e 1 2 ). The quantity S[O]/
We have obtained the significance S[O], Eq. (19), by imposing that the observable should at
least be larger than its absolute statistical error [24]
|
O|
> 1,
(20)
where the absolute error is approximated by

O]
O]
2 S[
= S[
O 2
O
O 2 .
N
N
(21)
The theoretical statistical significance is thus equal to the number of standard deviations to which
the corresponding CP observable can be determined to be non-zero. For an ideal detector, a significance of, e.g., S = 1 implies that the CP observables can be measured at the statistical 68%
confidence level. We remark that the theoretical statistical significance in Eq. (19) is solely a theoretical definition. Background and detector simulations for particle reconstruction efficiencies
are not included. In order to give realistic values of the statistical significances, a detailed Monte
Carlo analysis would be required, which is however beyond the scope of the present work. In the
following we comment on the major SUSY and SM backgrounds and discuss how they can be
reduced.
84
Table 1
Input parameters M2 , ||, , tan , and m e, . All mass parameters are given in GeV. The other masses are meR =
107 GeV, meL = 125 GeV, m1 = 106 GeV, m2 = 127 GeV, m 0 = 73 GeV, m 0 = 127 GeV, m 0 = 208 GeV,
1
m 0 = 247 GeV
M2
||
tan
m e,
m1
m2
152
200
0.5
103
125
246
Fig. 2. Normalized energy distributions of the signal leptons from chargino pair production e+ e 2+ 1 ;
1 (light blue line) and e+ e 1+ 2 ; 2 (dark blue line), for the parameters as given in Table 1.
Furthermore, we show the energy distributions of the background leptons from the reactions e+ e W + W ;
W (dashed orange line [27]), e+ e 1+ 1 ; 1 (dashed red line), and e+ e 2+ 2 ;
2 (dashed green line). (For interpretation of the references to colour in this figure legend, the reader is referred
to the web version of this paper.)
4.4. Lepton energy distributions and backgrounds

+
+
A main SM background will be from W pair production,
e e W W , where both W s
+
+
decay leptonically W and W . For s =

500 GeV and unpolarized beams
the cross section is (e+ e W + W ) = 7.4 pb [25], and , BR(W + + )BR(W
) = 4.4%, for , = e, [26]. With a beam polarization of (P , P+ ) = (0.9, 0.6), the
cross section for e+ e W + W is about a factor 3 larger [25]. A main SUSY background
would be from the pair production of equal charginos, e+ e 1 1+ and e+ e 2 2+ .
For the scenario as given in Table 1, the cross sections are (e+ e 1 1+ ) = 974 fb, and
(e+ e 2 2+ ) = 145 fb at s = 500 GeV with (P , P+ ) = (0.9, 0.6). The signal cross
section is (e+ e 1+ 2 ) = 527 fb. The chargino branching ratios are BR( 1+ e e+ ) =
33% and BR( 2+ e e+ ) = 8%.
A large part of the background can be cut by using the different energy distributions of the
final state leptons. In Fig. 2, we show the normalized energy distributions of the leptons
stemming from the various reactions e+ e i+ j ; j , where we have neglected the
effect of the longitudinal chargino polarization which may somewhat change the box-shape of the
85
distributions. For the energy distribution of the leptons from e+ e W + W ; W ,

we have fitted the curve in Fig. 3 of Ref. [27].
We can observe from Fig. 2 that the energy distribution of the signal leptons from
e+ e 1 2+ , 1 (light blue line), does only slightly overlap with that from the
charge conjugated process e+ e 1+ 2 , 2 (dark blue line) in this scenario. In
the first place, this is essential for a measurement of the CP observables, since they change
sign for the charge conjugated process. Secondly, the background from equal charginos pair
production e+ e 2 2+ can be totally eliminated, since both leptons have an energy in
the interval E [84, 122] GeV. As discussed above, a signal event has a lepton with energy
E [7, 59] GeV, and an oppositely charged lepton with energy E [54, 189] GeV. On the
other hand, the lepton energy distribution from 1+ 1 production overlaps with that of a signal
lepton (the one from the 2 decay) only in the interval E [54, 74] GeV. The background from
1+ 1 production can be eliminated, if we require one lepton with E > 74 GeV.
The energy distribution of the leptons from W decays overlaps with that from the signal in
the whole range, see Fig. 2. However, the signal involves one lepton with a rather low energy
E [7, 59] GeV, where the background is low. In this interval, the signal is about twice as large
as the W background. Thus, as any CP-even background, which contributes only to the denominator but not to the numerator of the CP observables, it will reduce them and the corresponding
significances only slightly. Moreover, the high energetic background leptons from W decays
can be significantly reduced by the cut E 189 GeV, see Fig. 2.
Additional important SUSY background originates from the following processes [28,29]:
(a) stau pair production e+ e + , with subsequent decays 10 , followed by leptonic
tau decays, (b) selectron and smuon production e+ e + , = e, , followed by 10 ,
and (c) neutralino production e+ e i0 j0 and their decays into leptons via sleptons.
Additional SM background reactions are [28,29]: (i) photon-induced tau pair production
e+ e e+ e + , followed by leptonic tau decays, (ii) photon-induced W pair production
e+ e e+ e W + W , with W , (iii) tau pair production followed by leptonic tau decays, (iv) Z pair production followed by the decays into leptons, and (v) single boson production,
e+ e Z and Z with Z + , .
Recently a detailed NLO study of chargino pair production e+ e 1+ 1 and decay
1 at a linear collider with s = 1 TeV has been performed [28]. The SM and SUSY
backgrounds to the experimental signature e + E
/ have been taken into account. A signal to
background ratio of 0.62 has been obtained after appropriate cuts [28]. In particular, the background from photon-induced pair production, which may exceed the size of the signal cross
section by a factor of 104 , has been reduced by a factor of 106 . Further the authors of Ref. [28]
have shown that SM processes lead to a flat background distribution, which can be easily subtracted, while SUSY backgrounds are more challenging, since their kinematic distributions are
similar to the signal in general. We expect that our proposed discrimination criteria for the lepton
energies, together with cuts as applied in [28,29], will enhance the signal to background ratio
also for our observables.
5. Numerical results
We present numerical results for the CP-sensitive observables defined in Section 4 for
chargino production and decay, e+ e 1 2 + , for , = e, . We study the
dependence of the CP observables on the phase of the higgsino mass parameter = ||ei
86
in the framework of the general MSSM, where restrictions on from the electron and neutron
EDMs are less severe compared to the constrained MSSM [4,5]. For example, the cancellation of
various contributions to the EDMs allow for the possibility of O(1) [5]. Since our analysis
does not include all relevant parameters necessary to predict the actual values of the EDMs, we
do not take the EDMs into account,
and show the full -dependence of the observables.
Our study is for the ILC with s = 500 GeV and longitudinal beam polarizations (P , P+ ) =
(0.9, 0.6). This choice enhances the e exchange contribution, yielding larger cross sections
and CP observables. We also provide numerical results for the chargino cross sections and the
branching ratios. Furthermore, we give the theoretical statistical significances of the CP-sensitive
observables.
For the calculation of the chargino decay widths and branching ratios we consider their twobody decays [30]
1,2
W n0 , e e , , , e eL
,
L , 1,2 ,
2 Z 1 , h 1 .
(22)
We assume the GUT inspired relation |M1 | = 5/3M2 tan2 W , and in the stau sector we fix the
trilinear scalar coupling parameter A = 250 GeV.
Before resuming with the numerical investigation, we address the parameter dependence of
L O R ) and Im(V V O R ) on which our CP-sensitive
the CP-sensitive coupling factors Im(O12
12
11 21 12
observables depend, see Eqs. (6) and (7). When we expand them by using the parametrization of
the chargino mixing matrices U and V , we find

L R 1
R
Im V11
(23)
= 2 Im O12
V21 O12
O12 = sin 21 sin 22 sin(1 2 + 1 2 ),
8
with the chargino mixing angles 1 , 2 , and the phases 1 , 1 , 2 , 2 of the matrices U, V . Their
explicit dependence on the parameters of the chargino system can be found in [7,16]. In particular, the phases are zero when is zero. One finds that the CP-sensitive coupling factors (and
therefore the CP observables) are largest for large gaugino-higgsino mixing, i.e. for M2 ||.
Furthermore, a small value for tan is preferable, as tan results in 1 , 2 and
2 , 1 0, leading to sin(1 2 + 1 2 ) 0. We therefore choose a mixed scenario with
small tan , see the parameters in Table 1.
prod
In Fig. 3(a), we show the -dependence of the CP-sensitive observables
OT and
OT ,
Eq. (17) and (18), respectively. One can clearly see their asymmetric dependence on . At
= 0(mod ), the chargino couplings are real and therefore the CP-sensitive observables vanish. The observables attain their largest values at = 0.75 of about
OT = 6.5 103
prod
and
OT = 2 102 . Fig. 3(b) shows the -dependence of the CP-sensitive observables
prod
Sgn(OT ) and
Sgn(OT ). At = 0.75 the observables reach
Sgn(OT ) = 1.9% and
prod
Sgn(OT ) = 5.6%.
In Fig. 4(a), we show the chargino production cross section 12 (e+ e 1+ 2 ) as a
function of . The production cross section varies between 12 = 618 fb for = 0(mod 2),
and 12 = 227 fb for = . Fig. 4(b) shows the branching ratios BR( 1+ e e+ ) and
BR( 2 e e ), which are about 32% and 7%, respectively. The -dependence of the cross
section (e+ e 1+ 2 e e+ e e ) is given in Fig. 4(c). In contrast to the CP-sensitive
observables, the cross section shows a symmetric dependence on . The phase ambiguity
2 of the higgsino mass parameter can only be resolved by a measurement of
87
prod
Fig. 3. -dependence of the CP-sensitive observables (a)

OT (red solid line) and
OT (blue dashed line),
prod
and (b)
Sgn(OT ) (red solid line) and
Sgn(OT ) (blue dashed line), for chargino production and decay
+
+
e e 1 2 e e+ e e , for the scenario defined in Table 1, at s = 500 GeV with longitudinal beam polarizations (P , P+ ) = (0.9, 0.6). (For interpretation of the references to colour in this figure legend, the reader is referred
to the web version of this paper.)
CP-sensitive observables. In Fig. 4(d), we present the theoretical statistical significances, S[O],
as defined in Eq. (19), for an integrated luminosity L = 500 fb1 . For
Sgn(OT ) and
OT ,
which do not require a reconstruction of the production plane, the theoretical statistical significances reach 4 and 5 standard deviations, respectively. The theoretical statistical significances of
prod
prod
the CP-sensitive observables
Sgn(OT ) and
OT are at the 12- level for the considered
scenario.
; 10 and
We note that also other leptonic chargino decay chains, i.e. 1,2
1,2 W 1 ; W , can be used for measuring the spinspin correlation terms of the
chargino production amplitude. The inclusion of these decay chains can be done in a similar
fashion as in Ref. [18]. However, we expect that this would not lead to larger statistical significances of the CP-sensitive observables. Although the total number of events is increased, the
CP-sensitive observables are reduced, since they have to be weighted with the corresponding
decay branching ratios [18].
Finally, note that the T-odd products, Eqs. (14) and (15), can also be used for the definition
of CP-sensitive observables in neutralino production with subsequent decays. This is possible
for the pair production (e+ e i0 j0 ), with i > j 2. The neutralino polarizations can then
again be obtained from the lepton distributions, e.g., in the decays 0 [31].
i,j
6. Summary and conclusions

We have proposed novel CP-sensitive observables in chargino production e+ e 1 2 .
These CP observables are sensitive to the phase of the higgsino parameter . They arise on treelevel, and rely on T-odd products in the chargino spinspin correlations. These are the terms
of the matrix element, which include the polarizations of both charginos, with one component
perpendicular to the production plane. The chargino polarization can be deduced from the distri
, = e, .
butions of their leptonic decay products 1,2
In order to probe the CP-sensitive spinspin correlation terms, we have identified two different T-odd products. The first one, OT , does not involve the chargino momentum, which has
the advantage that it is not necessary to reconstruct the production plane. We recall that other
88
Fig. 4. -dependence of (a) the chargino production cross section 12 (e+ e 1+ 2 ), (b) the branching ratios BR( 1+ e e+ ) (red solid line) and BR( 2 e e ) (blue dashed line), (c) the cross section
prod
(e+ e + e e+ e e ), and (d) the theoretical statistical significances S[Sgn(O
)] (red solid line),
prod
S[OT ] (blue dashed line), S[OT ] (magenta dotted line), and S[Sgn(OT )] (green dot-dashed line), for the scenario
defined in Table 1, at s = 500 GeV with longitudinal beam polarizations (P , P+ ) = (0.9, 0.6), and for (d), with
an integrated luminosity L = 500 fb1 . (For interpretation of the references to colour in this figure legend, the reader is
referred to the web version of this paper.)
T-odd products proposed in the literature always require such a reconstruction, if only one lepprod
tonic chargino decay is considered. The second T-odd product, OT , in contrast includes the
chargino momentum. Based on these T-odd products, we have defined two sorts of CP-sensitive
observables. One is an up-down asymmetry, giving the difference of events with positive and
negative T-odd products. The other sort of CP-sensitive observables are the expectation values of
the T-odd products for the event sample.
In the numerical study, we have found that the observables are largest in mixed scenarios
with small tan . We have defined theoretical significances to
decide which CP observable is
most sensitive to the CP phase . For a linear collider with s = 500 GeV and longitudinally
polarized beams, (P , P+ ) = (0.9, 0.6), with an integrated luminosity of L = 500 fb1 , the
CP-sensitive observables that are based on the T-odd product OT yield S[OT ] 5. We find larger
prod
prod
significances S[OT ] 12 for the CP-sensitive observables that are based on OT . Thus the
largest CP-violating effects are obtained if the chargino production plane can be reconstructed.
However, only a detailed experimental study with background and detector simulations can show
whether the CP-sensitive observables are accessible.
89
Finally, we remark that our proposed method for analyzing T-odd products in the spinspin
correlations terms can also be used for the definition of CP-sensitive observables in other fermion
pair production processes, such as neutralino productions.
Acknowledgements
We thank K. Desch and P. Wienemann for very helpful discussions on the background from
W pairs. M.T. thanks R. Kgerler for very useful discussions and encouragement. This work
is supported by the Fonds zur Frderung der wissenschaftlichen Forschung (FWF) of Austria,
project No. P18959-N16. The authors acknowledge support from EU under the MTRN-CT-2006035505 and MTRN-CT-2006-503369 network programs. A.B. was supported by the Spanish
grants SAB 2006-0072, FPA 2005-01269 and FPA 2005-25348-E of Ministero de Educacion y
Ciencia. O.K. was supported by the SFB Transregio 33: The Dark Universe.
c in the spin density matrix formalism
Appendix A. Quantities D and D
The coefficients in Eq. (5) of the chargino decay matrices for k+ + and k ,
= e, , are
Dk =
and

g2
|Vk1 |2 m2k m2 ,
2
(A.1)

c
D
= g 2 |Vk1 |2 mk sc p .
k
(A.2)
The positive sign in Eq. (A.2) holds for the decay k , and the negative sign for the
charge conjugated decay k+ + .
Appendix B. Momentum and polarization vectors
We choose a coordinate system with the z-axis along the p e direction in the center-of-mass
system. The 4-momenta of the charginos i and j+ are
p + = q(Ei /q, sin , 0, cos ),
i
p = q(Ej /q, sin , 0, cos ),
(B.1)
with their energies and common momentum

Ei,j
s + m2i,j m2j,i
,
=
2 s
2 (s, m2i , m2j )

q=
,
2 s
(B.2)
respectively, with the scattering angle (p e , p ), and the kinematic function (a, b, c) =
j
a 2 + b2 + c2 2(ab + ac + bc). The azimuthal angle can be set to zero, due to rotational invariance around the beam axis.
The three spin basis vectors of k+ and k are chosen to be
s 2 s 3
k
s1 = 0, 2k
= (0, cos , 0, sin ),
k
| s s 3 |
k
90

s2
k
s3
k
= 0,
p e p
k
= (0, 0, 1, 0),
|p e p |
k

Ek
Ek
1
=
(q/Ek , sin , 0, cos ).
p =
q,
k
mk
q
mk
(B.3)
They fulfill the orthonormality relations sc sd = cd and sc p = 0. The 4-momenta of

k
the leptons in the decays i+ + and j are
p+ = |p + |(1, cos + sin + , sin + sin + , cos + ),
(B.4)
p = |p |(1, cos sin , sin sin , cos ),
(B.5)
respectively, with
|p + | =
m2i m2
2(Ei + q cos + )
|p | =
m2j m2

2(Ej q cos )
(B.6)
and
cos + = sin sin + cos + + cos cos + ,
cos = sin sin cos + cos cos .
(B.7)
Appendix C. Spinspin correlation terms in the laboratory system

The spinspin correlation terms in the laboratory system are

g4
(Z)2 L2 c+ + R 2 c+ Im O L O R E sq sin2 ,
e
e
i
ij
ij
4
2 cos W

g4
R
Ei sq sin2 ,
(Z)(
)
L
c
Im
V
V
O
P12 (Z e ) =
e
e
+
j
1
ij
i1
4 cos2 W

g4
(Z)2 L2 c+ + R 2 c+ Im O L O R m sq sin 2,
P23 (ZZ) =
e
e
j
ij
ij
4
4 cos W

g4
(Z)(e ) Le c+ Im Vi1 Vj 1 Oij R mj sq sin 2,
P23 (Z e ) =
2
8 cos W
P12 (ZZ) =
(C.1)
(C.2)
(C.3)
(C.4)
which we obtain by inserting the momenta and spin vectors in the laboratory system, Eqs. (B.1)
and (B.3), into Eqs. (6) and (7). In order to obtain the terms P21 (ZZ) and P21 (Z e ), one has
to exchange Ei Ej in Eqs. (C.1) and (C.2). In order to obtain the terms P32 (ZZ) and
P32 (Z e ), one has to exchange mj mi in Eqs. (C.3) and (C.4).
References
[1] H.E. Haber, G.L. Kane, Phys. Rep. 117 (1985) 75;
H.P. Nilles, Phys. Rep. 110 (1984) 1.
[2] See, for example, H.E. Haber, Nucl. Phys. B (Proc. Suppl.) 62 (1998) 469, hep-ph/9709450.
[3] T. Ibrahim, P. Nath, arXiv: 0705.2008 [hep-ph].
91
[4] T. Ibrahim, P. Nath, Phys. Rev. D 57 (1998) 478, hep-ph/9708456;

T. Ibrahim, P. Nath, Phys. Rev. D 58 (1998) 019901, Erratum;
M. Brhlik, G.J. Good, G.L. Kane, Phys. Rev. D 59 (1999) 115004, hep-ph/9810457;
A. Bartl, T. Gajdosik, W. Porod, P. Stockinger, H. Stremnitzer, Phys. Rev. D 60 (1999) 073003, hep-ph/9903402;
A. Pilaftsis, Nucl. Phys. B 644 (2002) 263, hep-ph/0207277;
A. Bartl, W. Majerotto, W. Porod, D. Wyler, Phys. Rev. D 68 (2003) 053005, hep-ph/0306050.
[5] V.D. Barger, T. Falk, T. Han, J. Jiang, T. Li, T. Plehn, Phys. Rev. D 64 (2001) 056007, hep-ph/0101106;
S. Yaser Ayazi, Y. Farzan, Phys. Rev. D 74 (2006) 055008, hep-ph/0605272.
[6] M. Brhlik, G.L. Kane, Phys. Lett. B 437 (1998) 331, hep-ph/9803391.
[7] A. Bartl, K. Hidaka, T. Kernreiter, W. Porod, Phys. Rev. D 66 (2002) 115009, hep-ph/0207186.
[8] S.Y. Choi, M. Drees, B. Gaissmaier, Phys. Rev. D 70 (2004) 014010, hep-ph/0403054;
A. Bartl, K. Hohenwarter-Sodek, T. Kernreiter, H. Rud, Eur. Phys. J. C 36 (2004) 515, hep-ph/0403265;
O. Kittel, A. Bartl, H. Fraas, W. Majerotto, Phys. Rev. D 70 (2004) 115005, hep-ph/0410054;
A. Bartl, H. Fraas, S. Hesselbach, K. Hohenwarter-Sodek, T. Kernreiter, G. Moortgat-Pick, Eur. Phys. J. C 51 (2007)
149, hep-ph/0608065.
[9] A. Bartl, H. Fraas, O. Kittel, W. Majerotto, Phys. Lett. B 598 (2004) 76, hep-ph/0406309.
[10] K. Rolbiecki, J. Kalinowski, Phys. Rev. D 76 (2007) 115006, arXiv: 0709.2994 [hep-ph].
[11] P. Osland, A. Vereshagin, Phys. Rev. D 76 (2007) 036001, arXiv: 0704.2165 [hep-ph].
[12] W.M. Yang, D.S. Du, Phys. Rev. D 67 (2003) 055004, hep-ph/0211453;
H. Eberl, T. Gajdosik, W. Majerotto, B. Schrausser, Phys. Lett. B 618 (2005) 171, hep-ph/0502112.
[13] J. Fujimoto, T. Ishikawa, Y. Kurihara, M. Jimbo, T. Kon, M. Kuroda, Phys. Rev. D 75 (2007) 113002;
K. Rolbiecki, arXiv: 0710.1748 [hep-ph].
[14] G. Valencia, hep-ph/9411441, and references therein;
G.C. Branco, L. Lavoura, J.P. Silva, CP Violation, Oxford Univ. Press, New York, 1999.
[15] G.A. Moortgat-Pick, H. Fraas, A. Bartl, W. Majerotto, Eur. Phys. J. C 7 (1999) 113, hep-ph/9804306;
G.A. Moortgat-Pick, Spin effects in chargino/neutralino production and decay, Doctoral thesis, Universitt
Wrzburg, 1999 (in German).
[16] S.Y. Choi, A. Djouadi, H.K. Dreiner, J. Kalinowski, P.M. Zerwas, Eur. Phys. J. C 7 (1999) 123, hep-ph/9806279;
S.Y. Choi, A. Djouadi, M. Guchait, J. Kalinowski, H.S. Song, P.M. Zerwas, Eur. Phys. J. C 14 (2000) 535, hepph/0002033.
[17] A. Bartl, E. Christova, K. Hohenwarter-Sodek, T. Kernreiter, Phys. Rev. D 70 (2004) 095007, hep-ph/0409060;
K. Kiers, A. Szynkman, D. London, Phys. Rev. D 74 (2006) 035004, hep-ph/0605123.
[18] A. Bartl, E. Christova, K. Hohenwarter-Sodek, T. Kernreiter, JHEP 0611 (2006) 076, hep-ph/0610234.
[19] H.E. Haber, hep-ph/9405376.
[20] A. Bartl, H. Fraas, S. Hesselbach, K. Hohenwarter-Sodek, T. Kernreiter, G.A. Moortgat-Pick, JHEP 0601 (2006)
170, hep-ph/0510029.
[21] M.R. Buckley, H. Murayama, W. Klemm, V. Rentala, arXiv: 0711.0364 [hep-ph].
[22] S.Y. Choi, M. Drees, J. Song, JHEP 0609 (2006) 064, hep-ph/0602131.
[23] D. Atwood, A. Soni, Phys. Rev. D 45 (1992) 2405;
M. Diehl, O. Nachtmann, Z. Phys. C 62 (1994) 397;
B. Grzadkowski, J.F. Gunion, Phys. Lett. B 350 (1995) 218, hep-ph/9501339.
[24] A. Bartl, H. Fraas, K. Hohenwarter-Sodek, T. Kernreiter, G. Moortgat-Pick, A. Wagner, Phys. Lett. B 644 (2007)
165, hep-ph/0610431.
[25] W. Beenakker, A. Denner, Int. J. Mod. Phys. A 9 (1994) 4837.
[26] W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.
[27] A. Denner, S. Dittmaier, M. Roth, D. Wackeroth, Eur. Phys. J. Direct C 2 (2000) 4, hep-ph/9912447.
[28] J. Kalinowski, W. Kilian, J. Reuter, T. Robens, K. Rolbiecki, arXiv: 0803.4161 [hep-ph].
[29] Y. Kato, K. Fujii, T. Kamon, V. Khotilovich, M.M. Nojiri, Phys. Lett. B 611 (2005) 223, hep-ph/0411249.
[30] O. Kittel, hep-ph/0504183.
[31] M. Terwort, Diploma thesis, Universitt Wien and Universitt Bielefeld, 2006 (in German).

New solutions to the GinspargWilson equation

Nigel Cundy
Institut fr Theoretische Physik, Universitt Regensburg, D-93040 Regensburg, Germany
Received 15 February 2008; received in revised form 8 April 2008; accepted 8 May 2008
Abstract
The overlap operator is just the simplest of a class of Dirac operators with an exact chiral symmetry.
I demonstrate how a general class of chiral Dirac operators can be constructed, show that they have no
fermion doublers and that they are all exponentially local, and test my conclusions numerically for a few
examples. However, since these operators are more expensive than the overlap operator, it is unlikely that
they will be useful in practical simulations.
PACS: 12.38.Gc; 11.30.Rd
Keywords: Chiral fermions; Lattice QCD
1. Introduction
For many years, it seemed that simulating chiral symmetry on the lattice was impossible,
because of the NielsonNinomiya theorem [1], which, in one form, states that it is impossible
to have a Dirac operator which is local, has the correct continuum limit, hyper-cubic symmetry,
no doublers, and anti-commutes with 5 . A method of avoiding the NielsonNinomiya theorem
was suggested shortly afterwards, in 1982, by Ginsparg and Wilson [2], who proposed that the
smoothest way of breaking chiral symmetry on the lattice in the continuum limit is to use a Dirac
operator which, rather than anti-commuting with 5 , satisfies the GinspargWilson relation,
1
aD5 + a5 D = a 2 D(5 R + R5 )D.
2
I will call R the GinspargWilson function, and it plays a key role in this paper.
E-mail address: nigel.cundy@physik.uni-regensburg.de.
doi:10.1016/j.nuclphysb.2008.05.004
(1)
N. Cundy / Nuclear Physics B 802 (2008) 92105
93
The GinspargWilson relation was soon forgotten as no solutions were known. After this
relation was rediscovered in 1998 [3], Martin Lscher showed that it implied an exact symmetry
of the QCD Lagrangian for any lattice Dirac operator obeying the GinspargWilson relation
with a local R, and that this symmetry reduces to chiral symmetry in the continuum limit [4].
Inspired by the work of Kaplan [5] and using an infinite number of fermion fields to avoid the
NielsonNinomiya theorem, Neuberger and Narayanan had already, a few years earlier, found
a Dirac operator which satisfied the GinspargWilson relation with R = 1, namely the overlap
operator [6,7], which can be written in the form given in Section 2, Eqs. (13) and (14). Non-local
solutions to the GinspargWilson equation have also been found, for example in [8], and the
various fixed point actions, including the classically perfect action [9,10], are known to satisfy
the GinspargWilson relation [3], but in practice a truncated form of the Dirac operator has
to be used, meaning that the chiral symmetry becomes inexact. Fujikawa has constructed the
algebraic solutions to a particular form of a generalised GinspargWilson equation [11,12]. The
construction of these operators requires two steps: firstly constructing an intermediate operator
from the matrix sign function of a Wilson-type operator, then taking a root of that operator.
Thus, these operators effectively require a nested series of roots of a matrix, meaning that they
will be considerably slower than overlap fermions in practice. Werner Kerler has constructed
generalised chiral lattice Dirac operators by solving a particular form of the GinspargWilson
relation [1315], and I discuss the relationship between his work and my own in Section 5. The
overlap operator, Fujikawas and Kerlers solutions are all particular forms of my most general
solution. I shall not discuss the generalisation of the GinspargWilson equation obtained by using
different kernels in the matrix sign function.
Of course, not every possible GinspargWilson function R will lead to a lattice chiral symmetry. An easy, though over simplified, way of seeing this is to note that if, in an expansion in
the lattice spacing, R is inversely proportional to the lattice spacing, a (or worse) then the righthand side of Eq. (1) will remain constant (or diverge) rather than reduce smoothly to zero in the
continuum limit. More strictly, the lattice chiral symmetry requires that {5 , R} is non-vanishing
and local [16]. This places a number of restrictions on the lattice Dirac operator D, such as that if
it satisfies the GinspargWilson symmetry and has no doublers then it cannot be ultra-local [17],
but at best, like the overlap operator [18], exponentially local.
I define DC as the class of possible suitable Dirac operators (local, 5 -Hermitian, with the
correct continuum limit, and no doublers) which have their eigenvalues lying on a curve in the
complex plane. It is interesting that the overlap operator, the generalisations by Fujikawa and
Kerler, and the Dirac operator in the continuum are all within DC . In the continuum, the eigenvalue spectrum is a straight line along the imaginary axis; for overlap fermions, the curve is a
circle. However, for example, Wilson fermions, which have no chiral symmetry, and staggered
fermions, which have doublers, are outside DC . It is an interesting hypothesis whether all suitable lattice Dirac operators within DC satisfy GinspargWilson chiral symmetry. In this paper,
I demonstrate the plausibility of this hypothesis by showing that a class of Dirac operators with
the correct continuum limit, no fermion doublers, and with their eigenvalue spectrum lying on
an arbitrary curve in the complex plane are all suitable chiral fermionsin that they obey a
GinspargWilson relation with local R and are local themselves.
It is also worth considering the reverse hypothesis, whether all possible GinspargWilson
lattice Dirac operators are within the class DC . In Section 2.2, I demonstrate that this is true
for Dirac operators satisfying [D, D ] = 0, a condition obeyed in the continuum. However, in
94
general it does not seem to be the case. Certain fixed point lattice Dirac operators [3]1 offer one
known counterexample, and I discuss others in Section 2.2. However, these Dirac operators are
related to a Dirac operator within DC by a simple chirally invariant transformation. However,
there are distinct homotopy classes within DC which cannot be mapped to each other by this
transformation.
In Section 2, I review the GinspargWilson chiral symmetry, to give the notation and tools
which I will use in Section 3 to construct the new GinspargWilson operators and demonstrate
that, with sufficiently smooth gauge fields, they are exponentially local. I test the locality numerically for a few examples in Section 4 and, after outlining some generalisations to this work in
Section 5, I conclude in Section 6.
2. GinspargWilson chiral symmetry
2.1. Introduction
In this section, I review the demonstration that the GinspargWilson equation implies an exact
lattice chiral symmetry and gives a lattice topological charge satisfying an index theorem. This
section is based on the work of Martin Lscher [4].
Consider the following chiral transformation of the fermion fields:
1
i = e5 (S 2 aRD) i ,
i = i e(S 2 aDR)5 ,
1
which for small enough can be written as

1
i = i + 5 S aRD i ,
2

1
i = i + i S aDR 5 .
2
(2)
(3)
If there is a mass-less fermionic action

Sf =
Nf
i Di ,
(4)
i=1
where D is 5 -Hermitian (D = 5 D5 ), then it is trivial to demonstrate that the action is conserved under the chiral transformation if and only if the general GinspargWilson relation is
fulfilled
1
D5 S + S5 D = a D(R5 + 5 R)D.
(5)
2
We can write R = RA + RC , where RA = 12 (R 5 R5 ) and RC = 12 (R + 5 R5 ) so that
{RA , 5 } = 0 and [RC , 5 ] = 0. Since RA does not contribute to the GinspargWilson relation,
we can simplify without any loss of generalisation by restricting R to those functions which
commute with 5 . Additionally I restrict S to 5 -Hermitian operators. By taking the Hermitian
conjugate of Eq. (5), it is clear that, with these restrictions, R must be Hermitian. It is now
1 I thank Wolfgang Bietenholz for bringing the relevant section of this paper to my attention.
95
possible to simplify Eq. (5) so that it reads

D S + S D = aD RD.
(6)
This can be associated with the continuum chiral symmetry because in the continuum limit (for
suitable S and R, i.e. they are local and with the correct form in the continuum) it reduces to
5 D + D5 = 0. The GinspargWilson relation is usually defined with S = 1, and S can easily
be absorbed into the definition of D and R. Of course, the Dirac operator must also satisfy the
usual criteria for a suitable Dirac operator, namely that it is exponentially local (according to
the definition Dxy e|xy| for positive and ), that the Dirac operator has no fermion
doublers, and that expanding the Fourier transformed operators in the lattice spacing, a, gives

D(p)
= ia p + Ab T b + O a 2 ,
S(p)
= 1 + O(a),
R(p) = O(1),
(7)
where T b are the (Hermitian) generators of the gauge group, and A = Ab T b represents the
gauge fields. It can be shown that a GinspargWilson Dirac operator correctly resolves the U(1)
anomaly by noting that the fermion measure is not invariant under this transformation:

1
1
d d = dete5 (S 2 aRD) dete(S 2 aDR)5 d d ,

(8)
which leads to a definition of a lattice topological index [4,19,20]

1
Qf = Tr 5 S a5 RD .
2
(9)
As an aside, I note that, in the continuum limit, the YangMills action can be written as [21]

1 2
1
(10)
F = cs Tr S aRD + const,
4
2
and the electromagnetic field tensor as [22]

1
F = cf Tr S aRD ,
2
(11)
where cs and cf are normalisation constants. This means that all the elements of QCD and the
electro-weak Lagrangians can be constructed from any GinspargWilson operator.
R can, of course, be trivially constructed algebraically for any lattice Dirac operator (assuming
that its inverse exists),
1
1
(12)
+ S,
D D
but only a few possible Dirac operators will give the local R needed for the lattice chiral symmetry. It is trivial to show that inserting the Wilson Dirac operator gives a non-local R since the
inverse of the Wilson operator is non-local and has additive mass renormalisation [23].
One exact solution to the GinspargWilson equation, with local R, has been known for over
ten years. The mass-less overlap operator [24], given by
R=S
DO = 1 + 5 (5 DW ),
(13)
96
where is the matrix sign function, satisfies the GinspargWilson equation with R = 1. DW
can in principle be any valid lattice Dirac operator with the correct continuum limit, no fermion
doublers, and a suitably chosen mass term (e.g. in Eq. (14), any m satisfying mc < m < 2, where
mc 0 is the Wilson critical mass, will suffice, although in practice m should be tuned to improve
the locality of the operator). For the purposes of this work, I will use the simplest possibility, the
Wilson Dirac operator, which I shall write as

1
DW =
(14)
( + ) m.
2
and are the forward and backward lattice Dirac operators, defined as

(x) = eiaA (x+a 2 ) (x + a )

(x),

iaA (xa 2 )
(x a ).
(x) = (x) e
(15)
Thus, taking the Fourier transform to obtain the momentum representation of DW , I obtain
D W (p) = i H + W m,

i i(p,x) iaA (x+a ) iap

2 e
H =
e
e
eiaA (xa 2 ) eiap ,
2 x
1 i(p,x) i(p,y) iaA (x+a ) iap

2 e
W =
e
e
e
+ eiaA (ya 2 ) eiap
2 ,x,y

1 eiaA (x+a 2 ) eiaA (ya 2 ) .
(16)
Note that in the free theory,

H = sin(ap ),

ap
W=
2 sin2
.
2
(17)
Using Eq. (16), the momentum representation of the overlap operator can be expressed as
1
D O (p) = 1 + ( H + W m)
B(p)

2
B(p)
= (m W ) + H H + i (W m)H H (W m) .
(18)
It is clear that, because the term inside the square root is real and greater than or equal to zero,
for gauge fields which are sufficiently smooth and where DW

DW does not have an exact zero
eigenvalue, the Fourier representation of the overlap operator is an analytic function of the momentum, and that the momentum is bound pa . From the PaleyWeiner theorem [25],
this is enough to ensure exponential locality. Numerical experience has shown that the overlap
DW has a zero eigenvalue [26]. D O (p) is

operator remains exponentially local even when DW
only zero at p = 0, meaning that it has no unwanted doublers. It is also clear that it has the correct
continuum limit. The overlap topological charge is

1
QO = Tr (5 DW ) .
2
(19)
97
2.2. The eigenvalue spectrum of generalised GinspargWilson operators

I now demonstrate that those chiral Dirac operators satisfying both [D, D ] = 0 and the
GinspargWilson equation have an eigenvalue spectrum on a curve in the complex plane. This
condition, along with 5 -Hermiticity, guarantees that the eigenvalue spectrum is symmetric under
reflection in the real axis.
I start with the GinspargWilson equation
D + D = D RD.
Using a spectral decomposition, which is valid if D and
I obtain
(20)
D
share eigenvalues, i.e.
2 Re() = ||2 (, R).
[D, D ] = 0,
(21)
It is clear that Eq. (20) implies that

D, D (1 RD) = D [D, R]D.
(22)
Thus, if [D, D ] = 0 then [D, R] = 0 and is also an eigenvector of R. This enables me to write
2 Re() = ||2 R (),
(23)
R
where
is some real function of the eigenvalue. Thus, in this case, the eigenvalue spectrum is
constrained to a curve in the complex plane.
In the general case, the eigenvalues of a GinspargWilson operator will not lie on a curve. For
example, we can consider a Dirac operator D, defined by [27]
T DT = DGW ,
(24)
where DGW DC is some GinspargWilson operator obeying the relation

DGW 5 + 5 DGW = DGW 5 RGW DGW ,
(25)
and T is some local operator which commutes with 5 , satisfies the correct continuum limit and
whose inverse both exists and is local. In the context of the renormalisation group and the fixed
point action, T is equivalent to a change in the blocking procedure used to modify the lattice
spacing. Given that T is invertible, D will satisfy a GinspargWilson equation
D5 + 5 D = D5 T RGW T D.
(26)
T 1 |
For every vector satisfying DGW | = 0 there will be a vector

which is a zero eigenvector of D. Otherwise there will (in general) be no relation between the eigenvalues and
eigenvectors of D and DGW unless T commutes with DGW or is unitary. Every particular
GinspargWilson operator is related to at least one operator within the class DC by some transformation T . The index of the Dirac operator is unchanged:
1
1
1
QT = Tr RD = Tr T RGW T T DGW (T )1 = QGW .
(27)
2
2
Not every operator within DC can be mapped to every other operator in DC . To move from
one Dirac operator to another which commutes with it, it is necessary to use a transformation T
which commutes with both Dirac operators, and both R functions. In some cases R might have
some zero eigenvalues, and the transformation needed to increase the number of zero eigenvalues
of R is not invertible.
98
3. Additional solutions to the GinspargWilson equation

Consider the Dirac operator

1
5 (5 DW ) + (5 DW )5 5 (5 DW ),
Dr = 1 + r
2
(28)
where r[x] is some real, positive, and analytic function (which, as I shall demonstrate, is enough
to ensure locality), and which satisfies
r[1] = 1.
(29)
The condition given in Eq. (29) ensures that, for a suitable choice of DW , such as the one given
in Eq. (14), the operator has the correct continuum limit and no doublers. It is trivial to show
that because r commutes with both 5 and , Dr is 5 -Hermitian. Furthermore, it satisfies the
GinspargWilson relation with
S = 1,

1
1
1
Rr =
2 + r (5 + 5 ) (5 + 5 )
D
2
D
=
2 + (5 + 5 )r[ 12 (5 + 5 )]

1 + r[ 12 (5 + 5 )]2 + r[ 12 (5 + 5 )](5 + 5 )
(30)
From Eq. (16), I obtain

1
D r (p) = 1 + r[x]( H + W m)
B(p)

1

1
1
H (p) + W (p) m
+
H (p) + W (p) m .
x=
2
B(p)
B(p)
(31)
DW has no zero
It is clear that if the gauge field A and the function r are analytic and DW
eigenvalues, then the Fourier representation of the Dirac operator is analytic. Therefore, using
the same argument as for the overlap operator, it is exponentially local. It is also clear, given
Eq. (29), that these Dirac operators have the correct continuum limit and that the doublers have
infinite mass. In the momentum representation, R is given by
R r (p) =
2 + 2r[x]x
.
1 + (r[x])2 + 2r[x]x
(32)
There is potentially a pole in R r at 1 + r[x]2 + 2r[x]x = 0 or, equivalently, at r = x + x 2 1.

Since r and x are both real, 1 x 1 and r is positive, the pole cannot contribute for any
x = 1. Expanding R r around x = 1 gives
R r =
2(1 r [1])(1 + x) + 12 (1 + x)2 (2r [1] r [1]) + O((1 + x)3 )

2(1 + x) + 12 (1 + x)2 (2r [1]2 + 2r [1] + r [1]) + O((1 + x)3 )
(33)
is always real and positive

Therefore, R r has a smooth limit to x = 1 and does not diverge. B(p)
(again assuming analytic r, A and that DW DW has no exact zero eigenvalues), so R r is analytic,
and Rr will (at worst) fall of exponentially with distance. Thus both the Dirac operator and
99
the GinspargWilson function Rr are exponentially local, and this operator should be a suitable
lattice Dirac operator with chiral symmetry. Given my experience with the overlap operator, I do
not expect locality to break down when DW

DW has an exact zero eigenvalue.
I now need to demonstrate that these operators have a well defined topological charge; which
is most easily done by comparing the eigenvalue spectrum of this operator with that of the overlap
operator. I write the non-zero modes of the Hermitian overlap operator as |+ and | , where
the zero modes are |0 and the unpaired eigenvectors of the overlap operator with eigenvalue
2 are |2 . Since DO and Dr commute, it is easy to show, that these eigenvectors are also
eigenvectors of 5 Dr and Rr . Additionally, it can be shown that
Dr |0 = 0,
2 |Dr |2 = 2 |DO |2 ,
+ |Dr |+ = |Dr | ,
Rr |0 = Rr |2 = 1,
Rr |+ = Rr | .
(34)
Hence,
1
1
Tr 5 Rr Dr = Tr 5 DO = Qf .
(35)
2
2
Therefore these Dirac operators will have the same zero modes and topological charge as the
overlap operator.
I note that writing the operator as r(cos )ei , which I have done in Eq. (28), is only possible
] = 0. It would not be possible to perform a similar decomposition with, for

because [DO , DO
example, the Wilson operator to shift the Wilson eigenvalues to a closed curve on the complex
plane.
4. Numerical tests
To test the locality of the Dirac operator Dr and the GinspargWilson function R, I measured

LD (x, y) = (x)Dr (y) ,

LR (x, y) = (x)Rr (y) ,
(36)
where (x) is a point source. I then plot the mean value of the locality L(|x y| = d) against d,
the distance between x and y in lattice units. The operator with r = 1/|x| mimics the continuum
operator in that the eigenvalue spectrum is on two linesone along the imaginary axis containing
the physical modes, and one at a constant mass (in the continuum limit an infinite mass), containing the doublers. However, because this operator is impractical,2 I shall use an approximation
2 Since H and W do not commute in the non-Abelian theory, I have only been to write the operator in a simplified
form in the free or Abelian theory. To calculate the operator exactly would require calculating eigenvalues of the overlap
operator to a high precision and using some rational or polynomial approximation to |x| to simulate the rest of the
eigenvalue spectrum, which is possible but impractical. It is also not clear if this operator is local: r is not analytic;
however we can construct successively better approximations to r all of which will be local. I also note that because
R1/|x| = 0 for the physical modes (it is not zero for the eigenvalues corresponding to the doublers; so that the Nielson
Ninomiya theorem is still avoided) the left and right chiral projectors are identical for the physical modes. Thus if this
operator is local, it might be possible to use it to circumvent the CP and T violations described in [28]. However, a full
discussion of this interesting topic is beyond the scope of this paper, and deserves a full treatment in a future work.
100
Fig. 1. The eigenvalue distributions of the various Dirac operators used in this study.
to this function. I therefore set r as theChebyshev approximations (over the range 0.1 < x < 1)
of order n 1 of the function f = 1/ x 2 , for n = 1, 5, 9 and 13, normalised so that it satisfies
Eq. (29). This gives me four different Dirac operators, D1 , D5 , D9 and D13 . The (theoretical)
eigenvalue distributions for the operators Dn are shown in Fig. 1. D1 is the overlap operator.
The kernel of the matrix sign function, DW , is as defined by Eq. (14), but with two levels of
stout smearing [29] at parameter 0.1, and with m = 1.5. I tested the locality of these operators
on configurations from a 123 48 dynamical overlap ensemble, with lattice spacing a 0.13 fm.
I apply the Dirac operator Dn or GinspargWilson function Rn to a unit source at x, and calculate
its projection onto another unit source vector at y, averaging over x, y and configurations. The
results are shown in Figs. 2 and 3.
It can be seen that in all cases both the Dirac operator and the GinspargWilson function are
exponentially local, with the same rate of decay as the overlap operator. They are all constrained
so that LD < D e|xy| and LR < R e|xy| . The value of seems to be independent of n,
and (we might conjecture) seems to be general across all possible functions r. The value of
increases with increasing n for the Dirac operator (although there is very small change for the
GinspargWilson function), and does depend on r. It is possible to suspect that other forms of the
GinspargWilson Dirac operator could have improved locality compared to the overlap operator.3
It can be concluded that these Dirac operators and GinspargWilson functions are, as expected,
3 For the same kernel. It is known that using an improved kernel can improve the locality [30], but this is a separate
issue.
101
Fig. 2. The locality function LD on a 123 48 dynamical overlap ensemble.
exponentially local. The various small wiggles in the curves are caused by the geometry of the
lattice and the breaking of rotational symmetry.
5. Generalisation of the GinspargWilson Dirac operator
The proposed operator given in Eq. (28) can be easily generalised further. One possibility is
to multiply by a function q:

1
1
Drq = q (5 + 5 ) 1 + r (5 + 5 ) 5 ,
(37)
2
2
where we again have the constraints that q[x] must be positive, analytic, and q[1] = q[1] (for
the sake of being definite, I shall use q[1] = 1, although the value can easily be absorbed
into the fermion renormalisation constant), and, to give the correct continuum limit, restricted to
q = 1 + O(a). This leads to a GinspargWilson function
Rrq =
1
Rr [x],
q[x]
(38)
and it is clear that, with q as specified, this contains no poles in the Fourier representation, so
Rrq will be local. We can, of course, generalise further, and consider an operator

1
1
1
Drqth = t (5 + 5 ) h q (5 + 5 ) 1 + r (5 + 5 ) 5 ,
(39)
2
2
2
102
Fig. 3. The locality function LR on a 123 48 dynamical overlap ensemble.
where h(z) is analytic, positive except for z = 0 and satisfies h(0) = 0, h(2) > 0 and h(z) =
h(z ); and t obeys the same conditions as q and r. Finally, I can choose S = 1 in Eq. (6), which
will modify the above Dirac operator to give:

1
1
DrqthS =
t (5 + 5 )
S(5 ) 2

1
1
h(q (5 + 5 ) 1 + r (5 + 5 ) 5 ,
(40)
2
2
where S(z) is an analytic function satisfying S(1) = S(1) = 1 and S(z ) = S(z) . Once again,
from the Fourier representation, it can be shown that DrqthS and RrqthS are both exponentially
local and have the correct continuum limit. Fujikawas operators [11,12] are members of this
class, with r = q = S = t = 1, h(z) = z1/(2k+1) , and a particular form for the kernel operator DW
which gives the correct continuum limit.
Kerler [1315] considered, at first, a alternative form of lattice chiral symmetry
5 D + D5 V = 0,
(41)
where V is unitary and 5 -Hermitian. This relates to the standard GinspargWilson relation with
the substitution of variables V = 1 RD = D1 D. He later extended equation (41) to include
some possible S = 1 forms of the GinspargWilson equation (following the transformations outlined in [28]), and, by considering Dirac operators which are functions of V and using a spectral
decomposition, he showed that operators of the form

1/2
V +V
1 1/2
1/2
W
H
V
V
,
DK = i G(V )G(V )
2i
2
103
(42)
where H (z) = H [z] (with the expansion around z = 0 giving the correct continuum limit; for
example using DW as the kernel of the matrix sign function, H [z] = z + O(z2 )), W (1) = 0,
both functions are Hermitian and

G(V ) = (1 sk ) + sk V /N,

= sk + (1 sk )V /N,
G

1
,
N = 1 2sk (1 sk ) 1 V + V
(43)
2
satisfy a lattice chiral symmetry. It is clear that Kerlers solution (with V = 5 ) and my own have
a certain similarity, although they were derived from different approaches: his from attempting
an algebraic solution of the eigenvalue equivalent of the GinspargWilson equation; mine from
desiring to test a seemingly general property of chiral Dirac operators. His solutions are a particular form of my most general DrqthS solution, specifically with r = 1 and with a particular form
of S and t. Kerler did not discuss in detail the locality of the GinspargWilson function, which
is crucial in determining if we truly have a lattice chiral symmetry. It is unclear whether these
generalisations offer any advantage over the operator given in the previous sections.4
Finally, it is worth spending a moment considering whether it is necessary to use 5 to
construct these operators, and not some other unitary operator u. There are several ways in which
we can construct u:
(1) u = C/ C C;
(2) u = C1 C for C not (anti-)Hermitian;
(3) u = eA , for anti-Hermitian A.
u, of course, would have to be local and have the correct continuum limit. I have already discussed the first option. For the second option, forcing u to be local, having the correct continuum
limit and being 5 -Hermitian will place considerable restrictions on the possible choices of C
(one option is C = 1 + DGW , where DGW is another GinspargWilson operator) but if we choose
to use a GinspargWilson operator constructed the matrix sign function this will revert to another
form of Eq. (40). For the third option, it is not obvious how to construct suitable operators which
are both 5 -Hermitian and free of doublers; for example i5 (1ei/||5 DW ) is not 5 -Hermitian,
and 1 eD5 D5 has doublers. Thus using the matrix sign function remains the only currently
known possibility to construct local chiral lattice Dirac operators free from doublers.
One can also, of course, transform the Dirac operators using the method outlined in Section 2.2. By choosing T and T which commute with D, we can map the eigenvalue spectrum to
any closed curve (which passes through the origin to get the correct continuum limit and through
two to remove the doublers). The only restriction with the mapping is caused by the number
of zero modes of R; but since RrqthS has an arbitrary number of zero modes depending on the
4 One possibility might be to absorb a perturbative expansion of the fermion renormalisation into the Dirac operator;
or to otherwise reduce the higher order lattice artifacts.
104
choice of functions, we can map to any possible R and hence D. Thus the eigenvalue spectrum
of any Dirac operator within DC can be mapped to the eigenvalue spectrum of at least one of
the operators DrqthS by a suitable transformation. The eigenvectors of DrqthS are determined
by DW , and can be freely modified (within certain constraints) by adjusting DW . Given that
the Dirac operators can be defined in terms of their eigenvalues and eigenvectors spectrum, it is
plausible, if not yet proven, that all Dirac operators within DC are GinspargWilson operators.
6. Conclusion
I have demonstrated that the overlap operator is just the simplest member of a class of chiral
Dirac operators by constructing additional GinspargWilson operators. I propose that any lattice
Dirac operator which has
(1) The correct continuum limit with no doublers;
(2) Eigenvalues which lie on a closed loop in the complex plane, symmetric under reflection
in the imaginary axis, and single valued with respect to the angle from the center of the
GinspargWilson circle;
will be exponentially local (in a sufficiently large volume and sufficiently smooth gauge fields),
will satisfy a lattice chiral symmetry with exponentially local (or better) GinspargWilson function R with an exact index theorem, will be 5 -Hermitian, and will thus be a suitable lattice
Dirac operator. Of course, these additional Dirac operators are more expensive to simulate than
the overlap operator while it is not clear that they have any benefits over the overlap operator.
Thus it is unlikely that they will have any more than theoretical interest.
Furthermore, I have shown that chiral Dirac operators fall into certain homotopy classes determined by the number of zero modes of the GinspargWilson function R. In the continuum limit,
all these Dirac operators will reduce to the same, universal, operator, and it is most unlikely that
this division has any physical relevance. However, it seems likely that the perfect action (the
lattice action with no discretization errors) will have to fall into the same homotopy class as the
continuum Dirac operator.
Acknowledgements
I am grateful of support from the DFG. Computations were carried out on a Linux cluster
at the Jlich Supercomputing center. I am grateful for useful discussions with Andreas Schfer,
Stephan Durr and Wolfgang Bietenholz, and the referee for pointing out an error in the first draft
of this paper.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
H.B. Nielsen, B. Ninomiya, Nucl. Phys. B 185 (1981) 20.

P.H. Ginsparg, K.G. Wilson, Phys. Rev. D 25 (1982) 2649.
P. Hasenfratz, V. Laliena, F. Niedermayer, Phys. Lett. B 427 (1998) 125, hep-lat/9801021.
M. Lscher, Phys. Lett. B 428 (1998) 342, hep-lat/9802011.
D.B. Kaplan, Phys. Lett. B 288 (1992) 342, hep-lat/9206013.
R. Narayanan, H. Neuberger, Nucl. Phys. B 412 (1994) 574, hep-lat/9307006.
R. Narayanan, H. Neuberger, Phys. Rev. Lett. 71 (1993) 3251, hep-lat/9308011.
P.J. de A. Bicudo, Phys. Lett. B 478 (2000) 379, hep-lat/9909157.
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
P. Hasenfratz, F. Niedermayer, Nucl. Phys. B 414 (1994) 785.

W. Bietenholz, U.J. Wiese, Phys. Lett. B 378 (1996) 222, hep-lat/9503022.
K. Fujikawa, Nucl. Phys. B 589 (2000) 487, hep-lat/0004012.
K. Fujikawa, M. Ishibashi, Nucl. Phys. B (Proc. Suppl.) 106 (2002) 712, hep-lat/0110023.
W. Kerler, Nucl. Phys. B 646 (2002) 201, hep-lat/0204008.
W. Kerler, Int. J. Mod. Phys. A 18 (2003) 2565, hep-lat/0212021.
W. Kerler, Nucl. Phys. B 680 (2004) 51, hep-lat/0307011.
P. Hasenfratz, Nucl. Phys. B 525 (1998) 401, hep-lat/9802007.
I. Horvath, Phys. Rev. D 60 (1999) 034510, hep-lat/9901014.
P. Hernandez, K. Jansen, M. Luscher, Nucl. Phys. B 552 (1999) 363, hep-lat/9808010.
Y. Kikukawa, A. Yamada, Phys. Lett. B 448 (1999) 265, hep-lat/9806013.
D.H. Adams, Ann. Phys. 296 (2002) 131, hep-lat/9812003.
I. Horvath, hep-lat/0607031.
K.F. Liu, A. Alexandru, I. Horvath, hep-lat/0703010.
W. Bietenholz, hep-lat/0001001.
H. Neuberger, Phys. Rev. Lett. 81 (1998) 4060, hep-lat/9806025.
R. Paley, N. Wiener, Fourier Tansform in Complex Domain, Providence, RI, 1934, Theorem XII.
S. Krieg, Internal collaboration report.
T.-W. Chiu, Phys. Lett. B 474 (2000) 89, hep-lat/9910029.
P. Hasenfratz, M. Bissegger, Phys. Lett. B 613 (2005) 57, hep-lat/0501010.
C. Morningstar, M.J. Peardon, Phys. Rev. D 69 (2004) 054501, hep-lat/0311018.
W. Bietenholz, Nucl. Phys. B 644 (2002) 223, hep-lat/0204016.
105

Aspects of multiple membranes

David S. Berman , Laura C. Tadrowski, Daniel C. Thompson
Queen Mary College, University of London, Department of Physics, Mile End road, London, E1 4NS, England, UK
Received 1 May 2008; accepted 13 May 2008
Abstract
This paper examines various aspects of the recently proposed theory of coincident membranes by Bagger
and Lambert. These include the properties of open membranes and the resulting boundary theory with an
interpretation in terms of the fivebrane and marginal supersymmetric deformations of the interactions with
the relation to the holographic dual.
1. Introduction
Two of the most outstanding problems in M-theory are the understanding of the world volume theories of coincident membranes and coincident fivebranes [1]. In a fascinating recent
development Bagger and Lambert [24] have proposed a Lagrangian description of coincident
interacting membranes which relies on the scalar fields of the membrane world volume taking
values in some nonassociative algebra. (A similar approach was also described by Gustavsson in
[5].) Remarkably, the theory could be made to possess the N = 8 supersymmetry required of a
membrane with the introduction of a gauge connection whose kinetic piece is a twisted version
of ChernSimons theory. The goal of this paper is to investigate properties of this theory. The
two aspects will both involve breaking the supersymmetry in interesting ways. The first, through
introduction of a boundary and the second, by directly altering the interaction potential while
preserving some supersymmetry.
First, we will consider the theory of open membranes i.e., the BaggerLambert theory with a
boundary. The motivation for this is to see aspects of fivebrane physics from the point of view
E-mail addresses: d.s.berman@qmul.ac.uk (D.S. Berman), l.c.tadrowski@qmul.ac.uk (L.C. Tadrowski),

d.c.thompson@qmul.ac.uk (D.C. Thompson).
doi:10.1016/j.nuclphysb.2008.05.006
D.S. Berman et al. / Nuclear Physics B 802 (2008) 106120
107
of the membrane boundary. If the BaggerLambert action is a good description of the membrane
then this would be expected since the fivebrane is the M-theory analogue of a D-brane.
The key issue is to examine the twisted ChernSimons theory since this gives rise to dynamical degrees of freedom on the membrane boundary. In fact we will show that the membrane
boundary is described by a sigma model whose target space is six-dimensional. This is consistent
with a self-dual string in the fivebrane world volume.
We will then consider possible marginal deformations of the interactions that preserve some
fraction of supersymmetry. This is the M-theory analogue of the Leigh Strassler deformed N = 4
YangMills theory whose supergravity dual was described in [7]. We will examine the relationship between the proposed supersymmetry preserving deformations of the membrane theory and
the deformed supergravity dual described and investigated in [8].
Whether this is the true description of coincident membranes is still not certain however supersymmetric field theories have proved of immense interest over the years [9] and has been
shown to possess a rich and deep theoretical structure. Therefore the arrival of an entirely new
supersymmetric field theory where the fields are nonassociative demands study.
2. An effective theory of interacting membranes
In [2] Bagger and Lambert proposed a theory with N = 8 supersymmetry to describe multiple
coincident membranes. The novel insight allowing this construction is that the fields take values
in a nonassociative algebra, denoted here by A. This nonassociative algebra, also called a three
algebra, is endowed with a bilinear product and a totally antisymmetric three-bracket instead of
the standard commutator found in Lie algebras. The three bracket or triple product is given by
the antisymmetrised associator. For example the associator of three transverse scalars is
I J K I

X , X , X = X XJ XK XI XJ XK
(1)
and the three bracket is then
I J K
1 [I J K]
X ,X ,X .
X ,X ,X =
12
One can introduce a basis {T a } of A satisfying
a b c
T , T , T = f abc d T d ,
(2)
(3)
where the totally antisymmetric structure constants1 f abcd obey the fundamental identity, akin
to the Jacobi identity of Lie algebras, given by
f efg d f abc g = f ef a g f bcg d + f ef b g f cag d + f ef c g f abg d .
(4)
We remark that at this stage we have not specified the dimension of the algebra which we shall
denote by n.
To make the supersymmetry algebra close [3] it is necessary to introduce non-propagating
fields A b a , which gauge the transformation:
XaI = cd f cdb a XbI b a XbI .
(5)
1 We raise and lower algebraic indices with a positive definite trace form metric which, in this paper, we take to be
simply ab .
108
The gauge field is antisymmetric as a consequence of the antisymmetry of f cda b so the gauge
group G SO(n). In fact since we are only examining the connection, one can only make statements about the algebra but it will be assumed that there is a full group structure.
As a consequence of the transformation law (5) the group G is restricted by insisting that one
may write:
A b a = f cdb a A cd
(6)
for some n n matrix valued A cd with f cdb a satisfying the fundamental identity which implies
f abcd must be an invariant four form of the group.
The Lagrangian for the full N = 8 theory including these gauge fields is given by
1
i
i
L = D X aI D XaI + a D a + b I J a XcI XdJ f abcd
2
4
2

1 abcd
2 cda efgb
V (X) +
Aab Acd + f g f
Aab Acd Aef ,
f
2
3
(7)
with bosonic potential

1 I J K 2
,
Tr X , X , X
12
and supersymmetry transformations
V (X) =
(8)
XaI = i I a ,
(9)
1
a = D XaI I XbI XcJ XdK f bcd a I J K ,
6
A b a = i I XcI d f cdb a ,
(10)
(11)
where the covariant derivative acts as D Xa = Xa A b a Xb . The gauge kinetic term is similar to ChernSimons theory but twisted with the structure constants of the algebra. Obviously,
the gauge fields are non-propagating as is required to give the correct degrees of freedom for
supersymmetry.
In the remains of this paper we shall explore two truncations of this theory. We shall examine
the gauge sector of the theory in closer detail by switching off the scalars and spinors. This
will allow us to investigate the boundary theory of open coincident membranes. We shall then
restrict our attention to the scalar-spinor sector and study marginal deformations of the theory
that preserve N = 2 supersymmetry.
3. Twisted ChernSimons and the self-dual string boundary theory
When open membranes end on a fivebrane the boundary may be described by an N = (4, 4)
self-dual string theory (see for example, [1]). The self-dual string can be regarded from two perspectives; it can be viewed simply as a solitonic solution of the fivebrane world volume equations
of motion [10] or one can think of it as the boundary theory of coincident membranes [11,12].
From the fivebrane perspective one can take a Maldacena style limit and consider the resulting
geometry of the fivebrane to describe the string. One finds that in this near horizon limit the selfdual string is described by a fivebrane with AdS3 S 3 geometry [6].
In this section we shall study the self-dual string from the point of view of the boundary theory
of coincident membranes. We will examine the gauge sector in isolation (setting XI = = 0)
and compare the resulting boundary theory to the fivebrane description.
109
To make concrete progress we need to specify a gauge group of the twisted ChernSimons
theory. This essentially means solving the fundamental identity (4). There is only one finitedimensional known solution (though other more exotic solutions have been discussed in [13]).
When the dimension of the algebra is four it is known that
f abcd abcd
(12)
satisfies the fundamental identity and the associated gauge group is then SO(4). There is some evidence in the literature that this n = 4 algebra, which we denote by A4 , may be the only possible
solution to the fundamental identity [13,14]. This restriction to the dimension four algebra is not
something that we will find problematic, rather we find that this will have a natural interpretation
in what follows. We will thus restrict ourself to the gauge group G = SO(4).
First we carry out the usual decomposition of SO(4)
= SU(2) SU(2) by splitting the so(4)
gauge field into self-dual and anti-self-dual parts using t Hooft matrices [15]. Let us perform
this decomposition in our twisted ChernSimons theory. We write the gauge field as
A = A+ + A ,
(13)
where
A+ = A+ ,
A = A
(14)
and where denotes Hodge star on the matrix indices i.e.

A+

ab
1
= abcd A+
cd .
2
(15)
We also have that 2 = 1 and that for A4 the structure constants when viewed as operators have
the same action on gauge fields as . Using the relation (6) between A and A we find (omitting
indices for convenience) that
A = A+ A = f A = 2 A = A.
Then the twisted ChernSimons action becomes

1 abcd
2 cda efgb
LTCS =
Aab Acd + f g f
Aab Acd Aef
f
4
3

Tr
2
=
Ad A + A A A
4
3

Tr
2 +
2
Tr
+
+
+
+
=
A dA + A A A
A dA + A A A
4
3
4
3
+

= LCS A LCS A .
(16)
(17)
Implicitly we are working at level k = 1 and henceforward we shall drop the overall 4 normalisation of the Lagrangian. We see that the twisted ChernSimons action has decomposed into two
SU(2) ChernSimons theories but note that there is an all important relative minus sign between
the two.2
2 Carrying out the same analysis but starting with the un-tilded A leads to an apparent relative minus sign on the A
cubic term but this is not physically significant since it can be removed by a field redefinition.
110
Let us now consider what happens when our membrane has a boundary. With a boundary,
regular ChernSimons theory gives rise to a (chiral) WessZuminoWitten theory of propagating
boundary degrees of freedom [1618] (see also [19] for a helpful review). The naive first thought
is that the boundary theory is a WZW model with a target space S 3 S 3 . However the relative
minus sign between the two ChernSimons terms means we must be much more careful.
We shall briefly outline the process to arrive at the WZW model and see how this minus sign
plays a role. We start by considering a single SU(2) ChernSimons theory with level k = 1.
One writes the three manifold as M = R with boundary M = R. For simplicity
we consider only the case where is a disc such that the boundary of M is a cylinder with
angular coordinate and R identified with time. Upon doing a gauge variation of the Chern
Simons Lagrangian one picks up a surface term. To eliminate this we choose a suitable boundary
condition, for example that the time component of the gauge field A0 vanishes at the boundary.
Equipped with this boundary condition one is able to recast the ChernSimons theory as
S = d 3 0ij Tr(A0 Fij Ai 0 Aj ),

(18)
so that in the bulk A0 becomes a Lagrange multiplier enforcing Gauss law for the spatial
components of the gauge field. Integrating out this Lagrange multiplier places the remaining
components of the gauge field to be pure gauge Ai = g 1 i g. Writing out the resulting action in
terms of the group element one finds
1
1

1
SWZW = d dt Tr g gg t g +
d 3 Tr g 1 gg 1 gg 1 g .
3
M
(19)
This action is invariant under g L( )gR(t). L() corresponds to transformations which do not
vanish at t = and therefore states lie in representations of this symmetry. R(t) is a gauge
transformation which must fixed. Hence the action is actually a chiral WessZuminoWitten
theory [18].
Let us explicitly calculate the action by parameterising the group element with Euler angles
in the form
g(, t) = eiu(,t)2 eix(,t)3 eiv(,t)2 ,
with

2 =
0
i
i
0

,
3 =
1
0

0
.
1
We find the kinetic term contributes

Tr g 1 gg 1 t g = 2 xt x + ut u + vt v + cos 2x( ut v + vt u) .
(20)
(21)
(22)
The WessZumino term is simply the pull back of the canonical three form of the group which
we calculate as
3
1
= Tr g 1 dg
3
= 4 sin 2x dx du dv
= d(2 cos 2x du dv).
(23)
111
Altogether we find the WZW action is
SWZW = 2 d dt ( xt x + ut u + vt v + 2 cos 2x ut v).
(24)
To see this in a more familiar form we do a field redefinition

1
1
u = ( + ),
v = ( ),
2
2
and find after some trivial trigonometry

SWZW = d dt cos2 x t + sin2 x t + xt x

+ cos 2x( t t ) .
(25)
(26)
We can now read off the metric, the 2-form and its field strength by picking out the symmetric
and antisymmetric terms in this action
ds 2 = cos2 x d 2 + sin2 x d 2 + dx 2 ,
(27)
B = cos 2x d d,
(28)
H = dB = 2 sin 2x dx d d = 4 g dx d d.
(29)
The target space is a three-sphere (written in Hopf coordinates) with unit radius and with one
unit of normalised flux through the sphere.
For the multiple M2 theory we can read off the complete six-dimensional target space metric
and field strength by simply taking into account the extra minus sign entering into (17) to find
ds 2 = cos2 x d 2 + sin2 x d 2 + dx 2 cos2 x d 2 sin2 x d 2 d x 2 ,
H = dB = 2 sin 2x dx d d 2 sin 2x d x d d .
(30)
(31)
The flux is anti-self-dual i.e. H = H and the geometry here can be thought of as a sphere of
unit radius cross a sphere of imaginary (unit) radius. The sphere of imaginary radius is a space of
constant negative curvature which we identify with AdS3 . Indeed, if one analytically continues
x i x so as to absorb the minus signs into the metric of the target space then the result would be
the metric and flux for AdS3 S 3 .3 This is very satisfactory, from examining the boundary theory
we see the target space is six-dimensional with an anti-self-dual three form flux and a geometry
of AdS3 S 3 . Comparing this to the fivebrane we see that this is indeed the description of the
self-dual string in a Maldacena style limit [6]. The restriction to SO(4) now makes sense since
it ensures the dimension of the boundary theory target space is six i.e., the dimension of the
fivebrane.
One may wonder whether one could reinstate k, the level, and introduce a parameter to control
the number of membranes in this way. This has one appealing property that the self-dual string
charge will be given by k. However, there are various difficulties with this approach. First, the
supersymmetry requires that we also multiply the matter sector by k which does not feel appropriate for the interacting membrane theory. Second, the scaling of radius of the target space S 3
with k does not match the scaling found in [6].
3 This appears to be legitimate locally but one should be very careful in how to view the global nature of the space.
112
4. Marginal deformations
We now shift our attention back to closed membranes and begin by describing the Bagger
Lambert theory in N = 2 superspace language. The gauge field is removed and so the N = 8
supersymmetry is no longer manifest and only an SU(4) U (1) SO(8) R-symmetry remains.
The fields are redefined using a notation suited to the SU(4) U (1) symmetry as follows:
X I Z A ZA 4(1) 4(1),
A A 4(1) 4(1),
AB 1(2) 1(2)
6(0).
(32)
The N = 2 supervariation in this form is,

Z A = i A ,

A = 2 Z A + i1 ABCD [ZB , ZC , ZD ] + 3i3 Z A , Z B , ZB .
(33)
One may then formulate the theory in terms of N = 2 chiral superfields, Z A , which satisfy
A = 0,
DZ
(34)
with an expansion
Z A = Z A (y) + A (y) + F A (y),
(35)
where y = x + i . The Clifford algebra has a real basis and conjugation of the complex
spinors and is defined as = T 0 .
The N = 2 supersymmetry algebra closes when the following constraints are imposed:

A B
Z , Z , ZB = 0,
(36)
A B

Z , Z , B = 0,
(37)
A B

A B
, Z , ZB + Z , , ZB = 0.
(38)
These follow from the single superspace constraint
A B

Z , Z , ZB = 0
(39)
and result in the vanishing of the third term in the supervariation of the fermion. Hence, in what
follows, 3 does not appear in our analysis. When the dust settles one is left with the following
Lagrangian written in terms of N = 2 chiral superfields obeying the constraint (39):
A
A

1
4
2
d Tr Z , ZA + d W Z + d 2 W (ZA )
L=
(40)
2
with W a holomorphic function of the nonassociative algebra which describes the interaction:
W =

1
ABCD Tr Z A , Z B , Z C , Z D .
8
In fact, closure of the full N = 8 theory requires 1 = 16 [3].
(41)
113
5. A proposal for a marginal deformation

In [7,8], the supergravity dual description of multiple membranes was studied and possible
deformations preserving the AdS4 structure were investigated. The preservation of the AdS4
structure indicates the possibility of deforming the membrane theory while preserving conformal invariance. The deformation of the S 7 of course indicates that the deformation of the theory
would not preserve the SO(8) R-symmetry and hence the N = 8 supersymmetry. This is the
M-theory analogue of the Leigh Strassler deformation of N = 4 YangMills whose supergravity
dual was determined by Lunin and Maldacena [7].
We will proceed by describing properties of the Leigh Strassler deformation and then describe
the proposed M-theory analogue for the membrane theory. The result will be to allow a marginal
deformation that preserves the N = 2 structure.
The Leigh Strassler deformation works by introducing a deformed product in the field theory
parametrised by (for this reason it is also referred to as deformed YangMills):
f g = ei(Qf Qg Qg Qf ) f g
1
= eiQ[f, Qg] f g,
(42)
where Qif , Qg denote the charges of the fields with respect to two global U (1) symmetries. As
such the deformation picks out two U (1)s as a special subgroup of the SO(6) R-symmetry.
This deformation only produces an effect within the superpotential. Expressed in N = 1 superspace using chiral fields I , I = 1, 2, 3, the superpotential is
W I J K I J K .
After deforming this using (42) the superpotential becomes

1 2 , 3 = 1 ei 2 3 ei 3 2 .
(43)
(44)
Properties of the deformation can be summarised as follows:

(1) The deformed product relies on the use of two U (1)s.
(2) The phase factor is defined in terms of the U (1)1 U (1)2 charges of the fields and involves
a commutator over the field indices.
(3) All products involving two different fields pick up a phase factor but effectively only commutators of independent fields are modified by the deformation.
(4) The star product with any third field produces no new phases, for example

1 2 3 = 1 2 3 .
(45)
One should also note that the deformed product is very reminiscent of the noncommutative
Moyal product and indeed from the D-brane perspective it is as if the transverse space to the
brane has been made noncommutative. This analogy becomes more explicit from the perspective
of the supergravity dual where one begins with the nondeformed brane solutions and then carries
out a series of solution generating transformations to switch on background fields. One can then
make the choice of whether the background fields are on the brane world volume, in which case
the resulting deformation is to make the theory noncommutative, or transverse to the brane world
volume in which case one produces this Leigh Strassler deformation.
114
We will now attempt now to generalise this deformation to the Bagger and Lambert theory by
introducing a deformed triple product with analogous properties to the deformed Lie bracket. We
introduce a phase factor in the triple product which depends on the charges of the fields under
the U (1)1 U (1)2 U (1)3 global symmetry. For chiral fields labelled A, B, C we propose the
following deformed associator
1
A, B, C = eiQ[A, QB, QC] A, B, C,
(46)
where the
denote the charges of the field under the U (1)1 U (1)2 U (1)3 global symmetry.
For all A, B, C this gives the phase factor
Qi
Q1[A, Q2B, Q3C] = Q1A Q2B Q3C + Q1B Q2C Q3A + Q1C Q2A Q3B
Q1C Q2B Q3A Q1B Q2A Q3C Q1A Q2C Q3B ,
Q1[A, Q2B, Q3C]
= Det Q,
where Q is the matrix of charges denoted by

1
QA Q2A Q3A
Q = Q1B Q2B Q3B .
Q1C Q2C Q3C
(47)
(48)
(49)
From this antisymmetrisation it can be seen that cyclic permutations of A, B, C will deform with the same phase structure as will cyclic permutations of C, B, A. Furthermore, the
multiplicative phase factor accompanying the latter will be the reciprocal of the former. This
deformation of products of three fields results in many nice features, which are analogous to
the Leigh Strassler deformation. In particular, the three bracket of the nonassociative algebra
deforms in a manner analogous to the deformation of the commutator, that is:

3
1
2
[ZA , ZB , ZC ] = eiQ[A, QB, QC] ZA , ZB , ZC + cyclic

3
1
2
eiQ[A, QB, QC ] ZC , ZB , ZA + cyclic ,
(50)
for all values of A, B, C. We find that properties of the deformation can be summarised as follows:
(1) The deformed product relies on the use of three U (1)s.
(2) The phase factor is defined in terms of the U (1)1 U (1)2 U (1)3 charges of the fields and
involves a triple product over the field indices.
(3) All products involving three different fields pick up a phase factor but effectively only triple
products of independent fields are modified by the deformation.
(4) The star product with any fourth field produces no new phases.
We find preservation of the algebraic properties of the theory. Furthermore, the reality condition is respected by the deformed triple product. For the purpose of explicit computation, the
chiral fields have been assigned the following charges

1 2 3 4
Z , Z , Z , Z Z 1 , Z 2 , ei1 Z 3 , ei1 Z 4 : U (1)1 ,
(51)
1 i 2 i 3 4
2
2
Z , e Z , Z : U (1)2 ,
Z ,e
(52)
i 1 i 2 3 4
3
3
Z , Z , Z : U (1)3 .
e Z ,e
(53)
115
For all possible choices of A = B = C (1, 2, 3, 4) preserving the ordering in even permutations
of (1, 2, 3, 4) the deformation is given by

A B C
Z , Z , Z = ei Z A , Z B , Z C + cyclic

ei Z C , Z B , Z A + cyclic ,
(54)
provided the sequence (A, B, C, D) in

Tr Z A , Z B , Z C , Z D ,
(55)
can be written as a positive permutation of (1, 2, 3, 4). The cyclicity of the trace and the presence
of a totally antisymmetric tensor in the superpotential ensure this can be done for every term in
the action. This ordering is important for re-expressing the deformation simply. It can then be
seen that the deformation preserves the properties of the associator within the trace

Tr Z A , Z B , Z C , Z D = Tr Z B , Z C , Z D , Z A ,
(56)
for all inequivalent A, B, C, D (1, 2, 3, 4) and that using this property we can then show

Tr Z A , Z B , Z C , Z D = Tr Z A , Z B , Z C , Z D .
(57)
It then follows for the triple product

Tr Z A , Z B , Z C , Z D = Tr Z A , Z B , Z C , Z D .
(58)
Analysis of anti-chiral fields reveals the same results with all exponential factors mapping to
their reciprocal. Also note that, triple products containing triple products obey:

, , [A, B, C] = , , [A, B, C] .
(59)
As for N = 4 super-YangMills, the effects of the deformation on the action are found only
in the superpotential. The N = 2 Lagrangian becomes
1

A

1
4
L =
d Tr Z , ZA ABCD d 2 Tr Z A , Z B , Z C , Z C
2
8
2
+ d W (ZA ).
(60)
The supersymmetry algebra remains unchanged along with the superspace constraint
A B

Z , Z , ZB = 0.
(61)
That the deformed theory can be expressed in N = 2 superspace means that N = 2 supersymmetry is manifest. In fact one may have preserved more but we expect only one quarter of the
supersymmetry to be preserved as in Leigh Strassler.
Note that although we began with the N = 2 version of the multi-membrane theory, we may
consider this as a deformation to the full N = 8 theory that preserves N = 2 supersymmetry.
The route via the N = 2 superspace version of the membrane was chosen because of familiarity
with the -deformation of N = 4 SYM theory and as in that case is much easier to formulate
using the superfield language.
We would like to see the effect of the deformation on the bosonic scalar potential:
V=

1
Tr X I , X J , X K , X I , X J , X K .
2 3!
(62)
116
To start with we will look at the terms arising in the undeformed superpotential that survive the Grassman integration. There are four different types of terms that do this. These are
(Z, [Z, , ]), (, [Z, Z, ]), (F, [Z, Z, Z]) and (Z, [Z, Z, F ]). We neglect the fermions for
the moment. Symmetries of the triple-product and the properties of the trace allow us to group
terms

1
W| = ABCD Tr Z A , Z B , Z C , DEF G [ZE , ZF , ZG ]
8

1
ABCD Tr Z A , Z B , CEF G [ZE , ZF , ZG ], Z D
8

1 ABCD A BEF G
Tr Z ,
[ZE , ZF , ZG ], Z C , Z D

8

1
ABCD Tr AEF G [ZE , ZF , ZG ], Z B , Z C , Z D ,
(63)
8
which becomes

241 A C D
W| =
Tr Z , Z , Z , ZC , ZD , ZA
8

241 A
Tr Z , ZC , Z C , Z D , ZD , ZA
8

241 A
(64)
Tr Z , ZC , ZD , Z C , Z D , ZA .
8
The first and second terms vanish using the constraint

A B
Z , Z , ZB = 0,
(65)
whilst, using the symmetries of the trace and triple-product gives

241 A
Tr Z , ZC , ZD , Z C , Z D , ZA
8

241 A
Tr Z , ZC , ZD , Z C , Z D , ZA + Z C , ZC , ZD , Z D , ZA
=
8

+ Z C , Z D , [ZC , ZD , ZA ] ,

241 A C D
Tr Z , Z , Z , [ZC , ZD , ZA ] ,
=
8

241 A B C
=
(66)
Tr Z , Z , Z , [ZA , ZB , ZC ] .
8
In the undeformed theory, this term is proportional to the term quadratic in the auxiliary field,
which arises from the chiral superfield kinetic term. These combine to produce the sixth order
scalar field term in the potential of the component Lagrangian.
The chiral kinetic term of the N = 2 superfield Lagrangian is invariant under the deformation
but we have already seen that the triple-products, present in the superpotential, are modified
while the constraint equations are not. The field equation for the auxiliary field becomes
W| =
F A = 1 ABCD [ZB , ZC , ZD ] .
(67)
After the deformation, the superpotential can be written, using the deformed symmetries of the
trace and triple product, as

W = 31 Tr Z 1 , Z 2 , Z 3 , Z 4 .
(68)
117
That both the auxiliary field equation and the superpotential can be simply expressed in the
deformed theory by replacing the triple product with a star-triple product is due to the symmetries
of the star product outlined earlier.
As for the undeformed case, the only purely bosonic contribution coming from the superpotential is of the same form as the term coming from the chiral kinetic piece and is given by

W| Tr Z A , Z B , Z C , [ZA , ZB , ZC ] .
(69)
5.1. The -deformation related to the supergravity dual
We now wish to examine other deformations, which may be related to deformations of the
AdS4 S 7 supergravity dual of the membrane. Matching such deformations will provide further
support to the BaggerLambert conjectured multi-membrane theory.
The deformations in [8] preserve the AdS4 (and hence the conformality of the membrane
theory) but deform the S 7 using M-theory solution generating transformations. These solution
generating transformations involve identifying a three torus and then acting with a solution generating transformation on a T 2 T 3 to deform the solution. It was shown that different choices
of the T 2 produced distinct solutions. A whole spectrum of solutions could be generated with a
great diversity in their apparent properties. In one case the effects of the deformation were such
that the entropy of related black-brane was left invariant by the deformation. In another case,
the deformations produced a different entropy involving a simple multiplicative factor with the
deformation parameter. There were also more complicated examples.
Motivated by the deformation of the supergravity dual, of a T 2 , we propose the deformation
of the membrane theory should be with a deformed two product. The deformed associator, given
by replacing the regular product with the star product as given in (42), is
A, B, C = A (B C) (A B) C.
(70)
For star products of the nonassociative theory we have no reason to assume that we will
recover the properties that were found for the star product in N = 4 SYM or those described in
Section 5. Even if these properties were preserved then we could not necessarily expect the two
terms present in the associator to transform in a homogenous manner. It would then be impossible
to express the deformed associator as a product of a phase (containing all the information of the
deformation) and the undeformed associator as we did in (46). Such a factorisation was crucial
in allowing us to express the deformed action simply.
Interestingly, for some choices of T 2 this factorisation property is present. For other choices
we lose this property and star-products with third fields produce additional phase-factors under
the deformation. However, this can occur in such a way as to ensure that the two constituent terms
in any associator transform in an homogenous manner. Again, we find we are able to generate
families of solutions through the deformation with very different properties. All results depend
on the particular choice of fields and charge assignments of the U (1) U (1) subgroup. Let us
see how the associator deforms under the star product. Using this deformation, the product of
three fields always transforms like
1
A (B C) = eiQ[A, QBC] eiQ[B, QC] A (B C),
(71)
where
is the charge of (B C) under the global symmetry U (1)i . Additivity of charge for
this global symmetry gives the phase as
QiBC
ei(Q[A, QB] +Q[A, QC] +Q[B, QC] ) ,

1
(72)
118
where the ordering of the field indices is determined by the ordering of the fields in the product.
We can see now, that given three fields the constituent terms within the associator can be expected to transform in the same way under the deformation. Furthermore, we can conclude that
associators related by complete antisymmetry
A, B, C C, B, A
(73)
will pick up reciprocal phase factors. However we lose the ability to relate phase factors for
cyclic permutations of all associators. This is a result of the replacement of a product of three
field charges, antisymmetrised on the field indices, with three commutators of field charges.
This symmetry is lost for deformations of general associators and triple products under arbitrary
global U (1) U (1)s.
We look now at two particular examples to highlight the behaviour of the nonassociative
BaggerLambert theory under this deformation.
5.2. Example I: [Z 2 , Z 3 , Z 4 ] under U (1)1 U (1)3
In this example, the star-product with a third field produces a new phase factor under the
deformation. This happens in such a way as to ensure each associator transforms into the old
undeformed associator and a multiplicative phase factor. Using the arbitrary charge assignments
from before and deforming under U (1)1 U (1)3 we see

2 3 4
Z , Z , Z = Z 2, Z 3, Z 4 Z 4, Z 3, Z 2

+ Z 3, Z 4, Z 2 Z 2, Z 4, Z 3

+ e2i Z 4 , Z 2 , Z 3 e2i Z 3 , Z 2 Z 4 .
(74)
The triple products do not transform in a similar manner to that of the U (1) U (1) U (1)
deformation defined earlier. However, it may be relatable to one of the less symmetric cases
found for the M2 supergravity dual.
5.3. Example II: [Z 1 , Z 2 , Z 3 ] under U (1)2 U (1)3
In this example, we find that star-products with a third field produce no new deformative
phase factors just as we found in with the original -deformation. The deformation can therefore
be fully expressed in terms of a deformation of the first product taken within any associator.
Furthermore, for this particular choice, each and every bi-linear star-product produces exactly
the correct phase factor to reproduce the effects of the deformation defined using three global
U (1) symmetries. For two deformations with very different origins it is surprising that we find
the same resultant deformation of triple-product terms within the superpotential. We find, under
U (1)2 U (1)3 ,
1 2 3

Z , Z , Z = Z 1, Z 2, Z 3 + Z 2, Z 3, Z 1

+ Z 3, Z 1, Z 2 Z 3, Z 2, Z 1

Z 2, Z 1, Z 3 Z 1, Z 3, Z 2 .
(75)
On further investigation it was found that third and higher order star products were irrelevant in
the deformation
Z 1, Z 2, Z 3

= Z1 Z2 Z3 Z1 Z2 Z3 + Z2 Z3 Z1

Z2 Z3 Z1 + Z3 Z1 Z2 Z3 Z1 Z2

Z3 Z2 Z2 + Z3 Z2 Z1 Z2 Z1 Z3

+ Z 2 Z 1 Z 3 Z 1 Z 3 Z 2 + Z 1 Z 3 Z 2.
119
(76)
This can be written

1 2 3
Z , Z , Z = ei Z 1 , Z 2 , Z 3 + cyclic ei Z 3 , Z 2 , Z 1 + cyclic .
(77)
Furthermore, if we then take the product with Z 4 from

Tr Z 4 , Z 1 , Z 2 , Z 3
(78)
we obtain

Tr Z 4 , Z 1 , Z 2 , Z 3 ,
(79)
which produces exactly the same term that we would generate using the U (1)1 U (1)2 U (1)3
deformation completely antisymmetrised over the three field indices as already described in Section 5. Thus the marginal deformation described in Section 5 will be dual to the deformation
described in [8].
6. Conclusions
We are still some way from a full understanding the interacting membrane theory but there are
significant indications that we are heading in the right direction. One, we have reproduced aspects
of the fivebrane from the open membrane. Two, we have explicit examples of supersymmetry
preserving marginal deformations that may be related to deformations of the supergravity dual.
The key issue that needs to be further explored is to understand how the number of membranes
enters the theory and of course relate this to the number of degrees of freedom and hopefully to
the proposal described in [20].
An immediate technical question is the study of the membrane supersymmetry in the presence
of a boundary and the supersymmetry of the boundary theory. One can also try to understand
quantum properties of the twisted ChernSimons theory. Famously, the partition function of
ChernSimons theory [17] leads to the RaySinger Torsion of the three manifold. The role of the
RaySinger torsion of the membrane and the partition function of the ChernSimons theory at
level-1 (coming from the anti-self-dual sector) are still very much open questions.
Note added
As this paper was being prepared, two interesting preprints appeared discussing aspects of the
Bagger and Lambert theory [21,22].
Acknowledgements
This work was in part supported by the EC Marie Curie Research Training Network, MRTNCT-2004-512194. L.T. and D.T. are supported by an STFC (nee PPARC) grant. We are gratefull
to Tom Brown, Paul Heslop, Sanjaye Ramgoolam and Bill Spence for discussions.
120
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
D.S. Berman, M-theory branes and their interactions, Phys. Rep. 456 (2008) 89, arXiv: 0710.1707 [hep-th].
J. Bagger, N. Lambert, Modeling multiple M2s, Phys. Rev. D 75 (2007) 045020, hep-th/0611108.
J. Bagger, N. Lambert, Gauge symmetry and supersymmetry of multiple M2-branes, arXiv: 0711.0955 [hep-th].
J. Bagger, N. Lambert, Comments on multiple M2-branes, JHEP 0802 (2008) 105, arXiv: 0712.3738 [hep-th].
A. Gustavsson, Algebraic structures on parallel M2-branes, arXiv: 0709.1260 [hep-th].
D.S. Berman, P. Sundell, AdS3 OM theory and the self-dual string or membranes ending on the five-brane, Phys.
Lett. B 529 (2002) 171, hep-th/0105288.
O. Lunin, J.M. Maldacena, Deforming field theories with U (1) U (1) global symmetry and their gravity duals,
JHEP 0505 (2005) 033, hep-th/0502086.
D.S. Berman, L.C. Tadrowski, M-theory brane deformations, Nucl. Phys. B 795 (2008) 201, arXiv: 0709.3059
[hep-th].
D.S. Berman, E. Rabinovici, Supersymmetric gauge theories, hep-th/0210044.
P.S. Howe, N.D. Lambert, P.C. West, The self-dual string soliton, Nucl. Phys. B 515 (1998) 203, hep-th/9709014.
A. Strominger, Open p-branes, Phys. Lett. B 383 (1996) 44, hep-th/9512059.
P.K. Townsend, D-branes from M-branes, Phys. Lett. B 373 (1996) 68, hep-th/9512062.
A. Gustavsson, Selfdual strings and loop space Nahm equations, arXiv: 0802.3456 [hep-th].
Y. Kawamura, Cubic matrix, generalized spin algebra and uncertainty relation, Prog. Theor. Phys. 110 (2003) 579,
hep-th/0304149.
G. t Hooft, Computation of the quantum effects due to a four-dimensional pseudoparticle, Phys. Rev. D 14 (1976)
3432;
G. t Hooft, Phys. Rev. D 18 (1978) 2199, Erratum.
E. Witten, Nonabelian bosonization in two dimensions, Commun. Math. Phys. 92 (1984) 455.
E. Witten, Quantum field theory and the Jones polynomial, Commun. Math. Phys. 121 (1989) 351.
S. Elitzur, G.W. Moore, A. Schwimmer, N. Seiberg, Remarks on the canonical quantization of the ChernSimons
Witten theory, Nucl. Phys. B 326 (1989) 108.
G.V. Dunne, Aspects of ChernSimons theory, hep-th/9902115.
D.S. Berman, N.B. Copland, A note on the M2M5 brane system and fuzzy spheres, Phys. Lett. B 639 (2006) 553,
hep-th/0605086.
S. Mukhi, C. Papageorgakis, M2 to D2, arXiv: 0803.3218 [hep-th].
M.A. Bandres, A.E. Lipstein, J.H. Schwarz, N = 8 superconformal ChernSimons theories, arXiv: 0803.3242 [hepth].

Extended supersymmetric -model based on

the SO(2N + 1) Lie algebra of the fermion operators
Seiya Nishiyama , Joo da Providncia, Constana Providncia,
Flvio Cordeiro
Centro de Fsica Terica, Departamento de Fsica, Universidade de Coimbra, P-3004-516 Coimbra, Portugal
Received 11 December 2007; received in revised form 6 May 2008; accepted 13 May 2008
Dedicated to the memory of Hideo Fukutome.
Abstract
Extended supersymmetric -model is given, standing on the SO(2N +1) Lie algebra of fermion operators
composed of annihilationcreation operators and pair operators. Canonical transformation, the extension of
the SO(2N ) Bogoliubov transformation to the SO(2N +1) group, is introduced. Embedding the SO(2N +1)
group into an SO(2N + 2) group and using SO(2N+2)
U (N+1) coset variables, we investigate a new aspect of the
supersymmetric -model on the Khler manifold of the symmetric space SO(2N+2)
U (N+1) . We construct a Killing
)
potential which is just the extension of the Killing potential in the SO(2N
U (N ) coset space given by van Holten
et al. to that in the SO(2N+2)

U (N+1) coset space. To our great surprise, the Killing potential is equivalent with
the generalized density matrix. Its diagonal-block matrix is related to a reduced scalar potential with a
FayetIlipoulos term. The reduced scalar potential is optimized in order to see the behaviour of the vacuum
expectation value of the -model fields and a proper solution for one of the SO(2N + 1) group parameters
is obtained. We give bosonization of the SO(2N + 2) Lie operators and vacuum functions in terms of the
SO(2N +2)
U (N+1) coset variables, a U (1) phase and the corresponding Khler potential.
A preliminary version of this work has been presented by S. Nishiyama at the YITP workshop YITP-W-07-05 on
String Theory and Quantum Field Theory held at Kinki University, 610 August 2007.
E-mail addresses: seikoaquarius@ybb.ne.jp, nisiyama@teor.fis.uc.pt (S. Nishiyama), providencia@teor.fis.uc.pt
(J. da Providncia), cp@teor.fis.uc.pt (C. Providncia), flaviocordeiro_704@hotmail.com (F. Cordeiro).
doi:10.1016/j.nuclphysb.2008.05.008
122
S. Nishiyama et al. / Nuclear Physics B 802 (2008) 121145
1. Introduction
The supersymmetric extension of nonlinear models was first given by Zumino under the introduction of scalar fields taking values in a complex Khler manifold [1]. The higher-dimensional
nonlinear -models defined on symmetric spaces and on hyper-Khler manifolds have been
intensively studied in various contexts in modern versions of elementary particle physics, superstring theory and supergravity theory [25].
In nuclear physics, HartreeBogoliubov theory (HBT) [6] has been regarded as the standard approximation in the many-body theoretical descriptions of superconducting fermion systems [7]. In the HBT an HB wavefunction (WF) for such systems represents Bose condensate
states of fermion pairs. Standing on the Lie-algebraic viewpoint, the pair operators of fermion
with N indices form the SO(2N ) Lie algebra and accompany the U (N ) Lie algebra as a subalgebra. SO(2N) (= g) and U (N ) (= h) denote the special orthogonal group of 2N dimensions
and the unitary group of N dimensions, respectively. One can give an integral representation
of a state vector on the group g, the exact coherent state representation (CS rep) of a fermion
system [8]. It makes possible global approach to problems of the HBT [7]. The canonical
transformation of the fermion operators induces the well-known generalized Bogoliubov transformation for the fermions.
van Holten et al. have discussed a procedure for consistent coupling of gauge- and matter superfields to supersymmetric -models on the Khler coset spaces. They have presented a way of
constructing the Killing potentials and have applied their method to the explicit construction of
)
supersymmetric -models on the coset spaces SO(2N
U (N ) . They have shown that only a finite number
of the coset models can be consistent when coupled to matter superfields with U (N ) quantum
numbers reflecting spinorial representations of SO(2N ) [2]. Higashijima et al. have given RicciSp(N )
SU(N )
SO(2N )
flat metrics on compact Khler manifolds, [SU(NM)U
(M)] , U (N ) and U (N ) and non-compact
Khler manifolds, applying their technique of the gauge theory formulation of supersymmetric
nonlinear -models on the Hermitian symmetric spaces [3]. Preceding these works, Deldug and
Valent have investigated the Khlerian -models in two-dimensional spacetime at the classical quantum level. They have presented a unified treatment of the models based on irreducible
G
[9]. van Holten has also disHermitian symmetric spaces corresponding to the coset spaces H
cussed the construction of -models on compact and non-compact Grassmannian manifolds,
SU(N+M)
SU(N,M)
[SU(N )U (M)] and [SU(N )U (M)] [10].
For providing the general microscopic means for a unified self-consistent description for Bose
and Fermi type collective excitations in soft nuclei with strong collective correlations, Fukutome,
Yamamura and one of the present authors (S.N.) have proposed a new fermion many-body theory
basing on the SO(2N + 1) Lie algebra of fermion operators [11]. The set of the fermion operators composed of creationannihilation and pair operators forms a larger Lie algebra, the Lie
algebra of the SO(2N + 1) group. A representation of an SO(2N + 1) group has been derived
by a group extension of the SO(2N ) Bogoliubov transformation for fermions to a new canonical
transformation group. The fermion Lie operators, when operated onto the integral representation
of the SO(2N + 1) WF, are mapped into the regular representation of the SO(2N + 1) group
and are represented by boson operators. The boson images of the fermion Lie operators are expressed by closed first order differential forms. The creationannihilation operators themselves
as well as the pair operators are given by the Schwinger-type boson representation as a natural
consequence [12,13].
123
Along the same way as the above, we give an extended supersymmetric -model on Khler
G
coset space H
= SO(2N+2)
U (N+1) , basing on the SO(2N + 1) Lie algebra of the fermion operators.
Embedding the SO(2N + 1) group into an SO(2N + 2) group and using SO(2N+2)
U (N+1) coset variables [14], we investigate a new aspect of the supersymmetric -model on the Khler manifold
of the symmetric space SO(2N+2)
U (N+1) . We construct a Killing potential which is just the extension
of the Killing potential in the
SO(2N +2)
U (N+1)
SO(2N )
U (N )
coset space given by van Holten et al. [2] to that in the
coset space. To our great surprise, the Killing potential is equivalent with the generalized density matrix. Its diagonal-block part is related to a reduced scalar potential with a
FayetIlipoulos term. The reduced scalar potential is optimized in order to see the behaviour
of the vacuum expectation value of the -model fields. We get, however, a too simple solution. The optimization of the reduced scalar potential plays an important role to evaluate the
criteria for supersymmetry-breaking and internal symmetry-breaking. To find a proper solution
for the extended supersymmetric -model, after rescaling Goldstone fields by a mass parameter, minimization of the deformed reduced scalar potential is also made. In the process of the
minimization of the reduced scalar potential with respect to the FayetIlipoulos parameter ,
we meet a surprising occurrence of a zero of the reduced scalar potential in the model with
gauged U (1).
In Section 2, we give a brief recapitulation of the induced representation of the SO(2N + 1)
canonical transformation group and embedding of the SO(2N + 1) group into an SO(2N + 2)
group. We also introduce SO(2N+2)
U (N+1) coset variables. In Section 3, we give an extended supersymmetric -model on the coset space SO(2N+2)
U (N+1) based on the SO(2N + 1) Lie algebra of the fermion
operators and study a new aspect of the extended supersymmetric -model on the Khler manifold, the symmetric space SO(2N+2)
U (N+1) . In Section 4, we express the Killing potential in terms of the
coset variables. We also prove its equivalence with the generalized density matrix and then derive
a reduced scalar potential. The reduced scalar potential is optimized in order to see the behaviour
of the vacuum expectation value of the -model fields. We get, however, a too simple solution.
We meet a surprising occurrence of a zero of the reduced scalar potential with gauged U (1) and
consider the reason why such a surprising consequence occurs. In Section 5, we devote ourselves
to find a proper solution for the extended supersymmetric -model by making a minimization of
the deformed reduced scalar potential. We see again the surprising occurrence of a zero of the reduced scalar potential with gauged U (1). Finally, in Section 6, we discuss on the optimized scalar
potential and make some concluding remarks. In Appendices AC, we give a bosonization of the
SO(2N + 2) Lie operators and vacuum functions in terms of the corresponding Khler potential,
+2)
the SO(2N
U (N+1) coset variables and a U (1) phase. Throughout this paper, we use the summation
convention over repeated indices unless there is the danger of misunderstanding.
2. Bogoliubov transformation generated by SO(2N + 1) Lie algebra of fermion operators
and embedding into SO(2N + 2) group
Let c and c , = 1, . . . , N , be annihilation and creation operators of the fermion satisfying
the canonical anti-commutation relations

c , c = ,

{c , c } = c , c = 0.
(2.1)
124
We introduce the set of fermion operators consisting of the following annihilation and creation
operators and pair operators:
c , c ,

E = c c , E = c c , E = c c ,
(2.2)
E = E , E = E
, E = E (, = 1, . . . , N).
It is well known that the set of fermion operators (2.2) form an SO(2N + 1) Lie algebra. As a
consequence of the anti-commutation relation (2.1), the commutation relations for the fermion
operators (2.2) in the SO(2N + 1) Lie algebra are

U (N ) algebra ,
E , E = E E
(2.3)

E , E = E E ,
(2.4)

[E , E ] = 0,
E , E = E + E E E ,

c , c = 2E ,
[c , c ] = 2E ,
(2.5)

[c , E ] = 0,
c , E = c c .
c , E = c ,
We omit the commutation relations obtained by Hermitian conjugation of (2.4) and (2.5). The
SO(2N + 1) Lie algebra of the fermion operators contains the U (N ) (= {E }) and the SO(2N )
(= {E , E , E }) Lie algebras of the pair operators as sub-algebras.
An SO(2N ) canonical transformation U (g) belongs to the fermion SO(2N ) Lie operators.
The U (g) is the generalized Bogoliubov transformation [6] specified by an SO(2N ) matrix g

U (g) c, c U (g) = c, c g,
(2.6)

def a b
,
g g = gg = 12N ,
det g = 1,
g=
(2.7)
b a

U (g)U (g ) = U (gg ), U g 1 = U 1 (g) = U (g),

U (12N ) = Ig
(unit operator on g).
(2.8)
) and b = (b ) are N N matrices.

The bar denotes the complex conjugation. The HB (SO(2N )) WF |g is generated as |g =
U (g)|0 (|0: the vacuum satisfying c |0 = 0). The matrix g is composed of the matrices a and
b satisfying the ortho-normalization condition. The |g is expressed as

1
q c c |0,
|g = 0|U (g)|0 exp
(2.9)
2

1
1 i
4e 2,
0|U (g)|0 = 00 (g) = det(a) 2 = det 1N + q q
(2.10)

SO(2N
)
coset space ,
variables of the
q = ba 1 = q T
U (N )

det(a )
i
.
= ln
(2.11)
2
det(a)
(c, c ) is a 2N -dimensional row vector ((c
), (c )). a
= (a
The det means determinant and the symbol T denotes the transposition.
The canonical anti-commutation relation gives us not only the above Lie algebras but also
the other three algebras. Let n be the fermion number operator n = c c . The operator (1)n
125
anti-commutes with c and c ;

c , (1)n = c , (1)n = 0.
Let us introduce an operator as
e = Z + X c X c ,
(2.12)
c . Due to the relation
X X + Z 2 = 1,
= c
= , we have
(2.13)
sin , 2 = .
From (2.1), (2.12) and (2.13), we obtain

1
1
n
n
= c , c , (1) GX ,
e c , c , (1) e
2
2
X X
2Z X
X X
def

GX = X X
X X
2ZX .
2ZX
2Z X
2Z 1
Let G be the (2N + 1) (2N + 1) matrix defined by

a b 0
def
G = GX b a 0
0 0 1
X = a Y b Y ,

a XY
b + X Y 2Z X
= b+ XY a
2ZX , Y = X a X b ,
X Y
2ZY 2Z Y 2Z 2 1
Y Y + Z 2 = 1,
Z = cos ,
X =
(2.14)
(2.15)
where X and Y are the column vector and the row vector, respectively. The SO(2N + 1) canonical
transformation U (G) is generated by the fermion SO(2N + 1) Lie operators. The U (G) is an
extension of the generalized Bogoliubov transformation U (g) [6] to a nonlinear transformation
and is specified by the SO(2N + 1) matrix G. We identify this G with the argument G of U (G).
Then U (G) = U (GX )U (g) and U (GX ) = exp().
From (2.6), (2.14) and (2.15) and the commutability of U (g) with (1)n , we obtain
x
A B
2

1
x
1
n
nB
, (2.16)
U (G) c , c , (1) U (G) = c , c , (1) A
2
2
2
y
where
x = 2ZX ,
y = 2ZY ,
x y
,
2(1 + z)
x y
= b + X Y = b +
,
2(1 + z)
A = a X Y = a
B
(2.17)
z = 2Z 2 1.
By using the relation U (G)(c, c , 1 )U (G) = U (G)(c, c , 1 )U (G)(z + )(1)n and the
2
2
third column equation of (2.16), Eq. (2.16) can be written as

1
1
U (G) c, c , U (G) = c, c , (z )G,
(2.18)
2
2
126
def
B
G=
x
2
y
2
G G = GG = 12N+1 ,
det G = 1,
(2.19)
U G1 = U 1 (G) = U (G),
U (G)U (G ) = U (GG ),

U (12N+1 ) = IG .
(2.20)
(c,
) and B = (B ) are
N N matrices. The U (G) is a nonlinear transformation with a q-number gauge factor z
where = x c x c and 2 = x x = z2 1 [11]. The matrix G belongs to the SO(2N + 1)
group. It is transformed to a real (2N + 1)-dimensional orthogonal matrix as
1
1 1N 0
1N
2
2
O = V GV 1 , V = i 1N i 1N 0 .
(2.21)
c , 1 ) is a (2N
2
+ 1)-dimensional row vector ((c ), (c ), 1 ).

2
A = (A
The SO(2N + 1) WF [14,15] |G = U (G)|0 is expressed as

1
1

|G = 0|U (G)|0 1 + r c exp
q c c |0, r =
(x + q x ), (2.22)
2
1+z

1 i
1 + z
4e 2.
det 1N + q q
0| U (G) |0 = 00 (G) =
(2.23)
2
Following Fukutome [15], we define the projection operators P+ and P onto the sub-spaces
of even and odd fermion numbers, respectively, by

1
P2 = P ,
P+ P = 0,
1 (1)n ,
2
and define the following operators with the spurious index 0:
1
def 1
E 0 0 = (1)n = (P P+ ),
2
2
def
def
0
E 0 = c P = P+ c ,
E = c P+ = P c ,
def
def
E 0 = c P+ = P c ,
E 0 = E 0 ,
def
def
E0 = E0 .
E0 = c P = P+ c ,
def
P =
(2.24)
(2.25)
We introduce the indices p, q, . . . running over N + 1 values 0, 1, . . . , N . Then the operators of

(2.2) and (2.25) can be denoted in a unified manner as E p q , Epq and E pq . They satisfy
, Epq = Eqp (p, q = 0, 1, . . . , N),

E p q = E q p , E pq = Eqp
p

r
p
r
E q , E s = qr E s ps E q
U (N + 1) algebra ,
p

(2.26)
E q , Ers = ps Eqr pr Eqs ,
pq

q
p
q
p
E , Ers = ps E r + qr E s pr E s qs E r ,
[Epq , Ers ] = 0.
The above commutation relations in (2.26) are of the same form as (2.3) and (2.4).
127
The SO(2N + 1) group is embedded into an SO(2N + 2) group. The embedding leads us to
a unified formulation of the SO(2N + 1) regular representation in which paired and unpaired
modes are treated in an equal way. Define (N + 1) (N + 1) matrices A and B as

x
A x2
B
2
A = y 1+z , B =
(2.27)
, y = x T a x b.
y
1z
2
2
2
2
Imposing the ortho-normalization of the G, matrices A and B satisfy the ortho-normalization
condition and then form an SO(2N + 2) matrix G represented as [14]

A B
G=
,
G G = GG = 12N +2 ,
(2.28)
det G = 1.
B A
By using (2.17) and (2.15), the matrices A and B can be decomposed as

T 1

T
x
x2
1N xr
1N + xr 2q
a 0
b
2
2
,
B=
A=
(1+z)r T
(1+z)r T q 1
0 1
0
1+z
1z
2
2
2
2
from which we get the inverse of A, A1 , as
1

x
1N
a
0
1+z
A1 =
.
0
1
r T
1

0
,
1
(2.29)
(2.30)
From (2.29) and (2.30), we obtain a SO(2N+2)

U (N+1) coset variable with the (N + 1)th component as

q
r
Q = BA1 =
(2.31)
= QT ,
r T 0
from which the SO(2N + 1) variables q and r are shown to be the independent variables of the
SO(2N +2)
U (N+1) coset space. The paired mode q and unpaired mode r variables in the SO(2N + 1)
algebra are unified as the paired variables in the SO(2N + 2) algebra [14]. We denote the (N + 1)dimension of the matrix Q by the index 0 and use the indices p, q, . . . .
3. -model on the SO(2N + 2)/U (N + 1) coset manifold
Let us introduce a (2N + 2) (N + 1) isometric matrix U by

U T = B T , AT .
(3.1)
If one uses the matrix elements of U and U as the co-ordinates on the manifold SO(2N + 2),
a real line element can be defined by a Hermitian metric tensor on the manifold. Under the
transformation U VU the metric is invariant. Then the metric tensor defined on the manifold
may become singular, due to the fact that one uses too many co-ordinates.
According to Zumino [1], if A is non-singular, we have relations governing U U as

U U = A A + B B = A 1N+1 + BA1 BA1 A = A 1N+1 + Q Q A,
(3.2)

ln det U U = ln det 1N+1 + Q Q + ln det A + ln det A ,
where we have used the SO(2N+2)
U (N+1) coset variable Q (2.31). If we take the matrix elements of
Q and Q as the co-ordinates on the SO(2N+2) coset manifold, the real line element can be well
U (N+1)
128
defined by a Hermitian metric tensor on the coset manifold as

ds 2 = Gpq rs dQpq d Q rs Qpq = Qpq and Q rs = Q rs ; Gpq rs = Grs pq .
(3.3)
We also use the indices r, s, . . . running over 0 and , , . . . . The condition that the manifold
under consideration is a Khler manifold is that its complex structure is covariantly constant
relative to the Riemann connection:
def Gpq rs
def Gpq rs
Gpq rs,tu =
(3.4)
= Gtu rs,pq ,
Gpq rs,tu =
= Gpq tu,rs ,
tu
Q
Q tu
and that it has vanishing torsions. Then, the Hermitian metric tensor Gpq rs can be locally given
through a real scalar function, the Khler potential, which takes the well-known form

K Q , Q = ln det 1N +1 + Q Q ,
(3.5)
and the explicit expression for the components of the metric tensor is given as
Gpq rs =
1
1
2 K(Q , Q)
1N +1 + Q Q
= 1N +1 + QQ
sp
qr
rs
pq
Q Q
(r s) (p q) + (p q, r s).
(3.6)
K(Q , Q)
uniquely since
Notice that the above function does not determine the Khler potential
the metric tensor Gpq rs is invariant under transformations of the Khler potential,
Q).
K Q , Q K Q , Q = K Q , Q + F(Q) + F(
(3.7)
Q)
are analytic functions of Q and Q,
respectively. In the case of the Khler metric
F(Q) and F(
tensor, we have only the components of the metric connections with unmixed indices

def
pq tu rs = G vw tu Gpq vw,rs , pq tu rs = G tu vw Gvw rs,pq , G vw tu = G 1 vw tu ,

(3.8)
and only the components of the curvatures

R pq rs tu vw = Gvw vw pq vw tu,rs = Gpq vw, tu rs Gt u v w pq t u tu rs v w vw ,
R rs pq vw tu = Gtu tu rs tu vw ,pq = R pq rs tu vw .
(3.9)
In two- or four-dimensional spacetime, the simplest representation of N = 1 supersymmetry

is a scalar multiplet = {Q, L , H } where Q and H are complex scalars and L 12 (1 + 5 )
is a left-handed chiral spinor defined through a Majorana spinor. In superspace language the
multiplet is written as a chiral superfield:
= Q + R L + R L H.
(3.10)
The most general theory of the supersymmetric -model can be constructed from the [N ] scalar
multiplets [] = {Q[] , L[] , H [] } ([] = 1, . . . , [N ]). The supersymmetry transformations are
given by

1
[]
/Q R + H [] L ,
H [] = L /L[] ,
(3.11)
2
where is the Majorana spinor parameter.
+2)
Let the Khler manifold be the SO(2N
U (N +1) coset manifold and redenote the complex scalar fields
Qpq as Q[] ([] = 1, . . . , N (N+1) (= [N ])) and suppose that the spinors [] and [] span the
Q[] = R L[] ,
L[] =
129
fibres. Following Zumino [1] and van Holten et al. [10], the Lagrangian of a supersymmetric
-model can be written completely in terms of co-ordinates on the fibre bundle in the following
form:
[]
[]
/ L[] + W;[][] R[] L[] + W ;[][] L R[]
Lchiral = G[][] Q [] Q[] + L D
1
[]
[]
[ ]
G[][] W ;[] W;[] + R [][][ ][] L L[] L L ,
2
(3.12)
[]
and the Khler covariant derivative is defined as D L[] = L[] + [] [] [ ] L Q[ ] . The

curvature tensors R are given by (3.9). This form of the Lagrangian has also been derived by
Higashijima
and Nitta with the use of the Khler normal co-ordinate expansion of the Lagrangian

L = d 4 K( , ) [16]. The Lagrangian Lchiral (3.12) is manifestly a scalar under the general
co-ordinate transformations Q[] Q [] on the manifold, provided that the fermion transforms
as a vector and the superpotential does as a scalar:
def
L [] =
Q [] []
,
Q[] L
W (Q ) = W (Q).
(3.13)
The auxiliary fields H [] are eliminated through their field equations

[] [ ]
H [] = [] [] [ ] R L + G [][] W ,[] .
(3.14)
The right-handed chiral spinor R is defined as R = C LT and C is the charge conjugation. The
comma , [] (, []) denotes a derivative with respect to Q[] (Q [] ), while the semicolon denotes
a covariant derivative using the affine connection defined by the [] [] [ ] :
W;[] = W,[] =
W
,
Q[]
W;[][] =
W;[]
[] [ ] [] W;[ ] .
Q[]
(3.15)
4. Expression for SO(2N + 2)/U (N + 1) Killing potential

Let us consider an SO(2N + 2) infinitesimal left transformation of an SO(2N + 2) matrix G
to G , G = (12N+2 + G)G, by using the first equation of (A.3):

1
A + AA + BB
B + AB + B A
+ A
B
G = N+1
(4.1)
G
=
+ BA A + A A + B B .
B + AB
B
1N+1 + A

1 ) in the G frame. With the aid of (4.1), the
Let us define a SO(2N+2)
U (N+1) coset variable Q (= B A

Q is calculated infinitesimally as
+ BA)(A + AA + BB)
1
Q = B A 1 = (B + AB
Q BQ.
= Q + B QA + AQ
(4.2)
The Khler metrics admit a set of holomorphic isometries, the Killing vectors, Ri[] (Q) and
i[] (Q)
(i = 1, . . . , dim G), which are the solution of the Killing equation
R
i [] (Q),[] = 0,
Ri [] (Q),[] + R
Ri [] (Q) = G[][[] Ri[] (Q).
(4.3)
These isometries define infinitesimal symmetry transformations and are described geometrically
by the above Killing vectors which are the generators of infinitesimal co-ordinate transformations
Q)
such that G (Q, Q)
=
keeping the metric invariant: Q = Q Q = R(Q) and Q = R(
130
The Killing equation (4.3) is the necessary and sufficient condition for an infinitesimal
G(Q, Q).
co-ordinate transformation
[]
Q[] = B AT Q QA + QB Q
= i Ri[] (Q),
i[] (Q),
Q [] = i R
(4.4)
where i are the infinitesimal and global group parameters. Due to the Killing equation, the
i[] (Q)
can be written locally as the gradient of some real scalar
Killing vectors Ri[] (Q) and R
i
such that
function, the Killing potentials M (Q, Q)
i [] (Q)
= iMi ,[] .
R
Ri [] (Q) = iMi ,[] ,
(4.5)
According to van Holten et al. [2] and using the infinitesimal SO(2N + 2) matrix G given by
the first of (A.3), the Killing potential M can be written for the coset SO(2N+2)
U (N+1) as
) = tr AM A + BM B + B M B ,
M A, B, B = Tr(G M

B
A
A + (M
AT )T ,
(4.6)
M
M
M
=
M
M
,
B M
AT
B , M B = M
B ,
M
M B = M
where the trace Tr is taken over the (2N + 2) (2N + 2) matrices, while the trace tr is taken over
the (N + 1) (N + 1) matrices. Let us introduce the (N + 1)-dimensional matrices R(Q; G),
RT (Q; G) and X by

R(Q; G) = B AT Q QA + QB Q,
RT (Q; G) = AT + QB ,
(4.7)
1
= X .
X = 1N+1 + QQ
In (4.4), putting i = 1, we have Q = R(Q; G) which is the Killing vector in the coset space
SO(2N +2)
U (N+1) , and tr of holomorphic matrix-valued function RT (Q; G), tr RT (Q; G) = F(Q) is a
holomorphic Khler transformation. Then the Killing potential M is given as
G) = tr (Q, Q;
G),
iM (Q, Q;
def
(4.8)
(Q, Q; G) = RT (Q; G) R(Q; G)Q X
= QAQ AT BQ + QB X .
From (4.6) and (4.8), we obtain
iM B = X Q,
iM B = Q X ,
iM A = 1N+1 2Q X Q.
(4.9)
, Eq. (4.9), their components are written in the form

Using the expression for M
B = X Q,
B = Q X ,
i M
i M
A = Q X Q,
AT = 1N+1 QX Q = X .
i M
i M
(4.10)
It is easily checked that the result (4.9) satisfies the gradient of the real function M (4.5). Of
course, putting r = 0 in Q (2.31), the Killing potential M in the SO(2N+2)
U (N+1) coset space leads to
)
the Killing potential M in the SO(2N
U (N ) coset space obtained by van Holten et al. [2].
To make clear the meaning of the Killing potential, using the (2N + 2) (N + 1) isometric
matrix U (U U = 1N+1 ), let us introduce the following (2N + 2) (2N + 2) matrix:
R
W = UU =
K

K
,
1N+1 R
R = BB ,
131
(4.11)
K = BA ,
which satisfies the idempotency relation W 2 = W and is Hermitian on the SO(2N + 2) group.
W is a natural extension of the generalized density matrix in the SO(2N ) CS rep to the
SO(2N + 2) CS rep. Since the matrices A and B are represented in terms of Q = (Qpq ) as
1
A = 1N+1 + Q Q 2 U,
B = Q 1N+1 + Q Q 2 U ,
then, we have
1
R = Q 1N+1 + Q Q Q = Q Q = 1N+1 ,
1
= Q.
K = Q 1N+1 + Q Q
U U (N + 1),
(4.12)
(4.13)
is expressed in terms of the subSubstituting (4.13) into (4.10), the Killing potential i M
matrices R and K of the generalized density matrix (4.11) as

K
= R
i M
(4.14)
,
K (1N+1 R)
from which we finally obtain

K
= R
i M
.
K 1N+1 R
(4.15)
To our great surprise, the expression for the Killing potential (4.15) just becomes equivalent with
the generalized density matrix (4.11).
The expression for the Killing potential M is described in terms of the SO(2N+2)
U (N+1) coset
variable Qpq but include an inverse matrix X given by (4.7). The variable Qpq has already been
expressed in terms of the variables q and r through (2.31). To obtain the concrete expression
for the Killing potential in terms of the SO(2N + 1) variables q and r , we must also represent
the inverse matrix in terms of the variables q and r . After some algebraic manipulations, the
inverse matrix X in (4.7) leads to the form

Qqq Qqr
1
X=
(4.16)
= ,
,
= 1N + qq
Qqr Qr r
where each sub-matrix is expressed by the variables q and r as

1+z

rr q r r T q ,
Qqq =
2
1+z
1+z
Qq r =
q r ,
Qr r =
.
2
2
(4.17)
Then, substituting (2.31) and (4.16) into (4.9) and introducing an auxiliary function = rr
q r r T q = , we can get the Killing potential M expressed in terms of only q, r and 1 + z =
2Z 2 as,

q + Z 2 (q + q r r T ) r + Z 2 r
iM B =
(4.18)
,
Z 2 r T q r
Z 2 (r T q q r T )

q Z 2 (q + r r T q ) Z 2 (q q r r )
iM B =
(4.19)
,
Z 2 r q r
r Z 2 r
132
iM A

1 2q q + 2Z 2 (q q + q q r r T + r r T q q r r T )
= N
2q r + 2Z 2 (q r + r r T q r)
2r q + 2Z (r q + r q r r )
1 2r r + 2Z 2 r r

.
(4.20)
In the above expressions for the Killing potential M , each block-matrix is easily verified to
satisfy the following identities and relations:

1 Z2
1 Z2 2
T
r q r = 0,
r q r = 0,
r r =
,
r r =
, (4.21)
Z2
Z2
1 2r r + 2Z 2 r r = 2Z 2 1,
r =
1 Z2
Z2
r,
r =
(4.22)
1 Z2
Z2
r ,
q q = 1N .
(4.23)
Using r = 2Z1 2 (x + q x),

the third identity of (4.21) can be easily proved.
Using (4.21), (4.22) and (4.23), we get compact forms of the Killing potential M as,
Z 2 r
q + Z 2 (rr q + q r r T )
(q Z 2 (q rr + r r T q )) (Z 2 r )
iM B =
(4.24)
,
2 r T
0
Z
( B )

iM A =
(Z 2 r )
1N
2q q
(0)
+ 2Z 2 (q rr q
2Z 2 r q
r r T )

2Z 2 q r
.
2Z 2 1
(4.25)
Let us introduce the gauge covariant derivatives

D Q[] = Q[] gi Ai Ri[] (Q),

[]
[]
D L[] = L[] gi Ai Ri[] ,[] (Q)L + [] [] [ ] L Q[ ] .
(4.26)
Ai are gauge fields corresponding to local symmetries and gi are coupling constants. They are
components of vector multiplets V i = (Ai , i , D i ), with i representing the gauginos and D i
the real auxiliary fields. Introducing the gauge fields in Lagrangian (3.12), via the gauge covariant
derivatives (4.26), the -model is no longer invariant under the supersymmetry transformations.
To restore the supersymmetry, it is necessary to add the terms

[]
i [] (Q) i [] gi tr D i Mi + i , (4.27)
Lchiral = 2G[][] Ri [] (Q) L iR + R
R L
where i are FayetIlipoulos parameters. Then the full Lagrangian for this model consists of the
usual supersymmetry YangMills part and the chiral part

1 i i
1
1
L = tr F
(4.28)
F + iD
/ i D i D i + Lchiral ( D ) + Lchiral .
4
2
2
Eliminating the auxiliary field D i by D i = gi (Mi + i ) (not summed for i), we can get a
scalar potential
2
1
VSC = gi2 tr Mi + i ,
2
(4.29)
133
in which a reduced scalar potential arising from the gauging of SU(N + 1) U (1) including a
FayetIlipoulos term with parameter is of special interest:
VredSC =
2
gU
(1)
2(N + 1)
( iMY )2 +
2
gSU(N+1)
tr(iMt )2 .
(4.30)
The new quantities tr(iMt )2 and iMY in VredSC are defined below and the trace tr, taken
over the N N matrix, is used. Then we have
2
2
2
(iMY ) , iMY = tr(iM A ),

tr(iMt ) = tr(iM A )
N +1
2
tr(iM A ) = N + 2 tr() + 2Z tr rr 4Z tr rr + 2Z 1,
tr(iM A )2 = N 4 tr() + 4 tr() + 12Z 2 tr rr 16Z 2 tr rr
4
4Z r r tr rr + 8Z r r tr rr + 1 4Z r r.
(4.31)
+ q x),
we can approximately calculate the quantities r r and tr(rr ) as

1
1
1 Z2
def 1
x x
, =
,
N + tr q q
r r =
N
4Z 4
Z2
(4.32)
1 1
1 Z 2 1 def
x x
= rr .
tr rr = r r =
4Z 4
Z 2
Using r =
1
(x
2Z 2
In (4.31), approximating tr(), tr(rr ), etc., by , tr(rr ), etc., respectively, and using
(4.32), tr(iM A ) and tr(iM A )2 are computed as

tr(iM A ) = 1 N + 2 2Z 2 1 ,
(4.33)

tr(iM A )2 = 1 + N 4 2Z 2 1 + 4 2Z 4 1 2 .
Substituting (4.33) into (4.30), we obtain the final form of the reduced scalar potential as
VredSC =
2
gU
(1)
2
gSU(N+1)
2

+ 1 N + 2 2Z 2 1 + 2
N 2 2Z 2 1
2(N + 1)
N +1

+ 2(N 1)Z 4 + 4Z 2 (N + 2) 2 ,
(4.34)
which is written in terms of the SO(2N + 1) parameter Z, the mean value of , i.e., , and the
FayetIlipoulos parameter .
In order to see the behaviour of the vacuum expectation value of the -fields, it is very important to analyze the form of the reduced scalar potential. From the variation of the reduced scalar
potential with respect to Z and , we can obtain the following relations:
2

2
2
2
gU
(1) + 1 N + 2 2Z 1 2gSU(N+1) 1 (N 1)Z + 1 = 0, (4.35)
2

2

2
gU
(1) + 1 N + 2 2Z 1 2Z 1
2

2
2gSU(N+1)
(4.36)
2Z 1 2(N 1)Z 4 + 4Z 2 (N + 2) = 0.
Multiplying by 2Z 2 1 for (4.35) and using (4.36), we have a g 2 -independent relation

(N + 1) Z 2 1 = 0,
(4.37)
134
from which, since = 0, we get very simple solutions

Z 2 = 1,
=
2
2
2
1 gU (1) (N 1) + 2gSU(N+1) gU (1)
.
2
2
2
gU
(1) + NgSU(N+1)
(4.38)
)
2
These solutions just correspond to the SO(2N
U (N ) supersymmetric -model since Z = 1. From the
second solution of (4.38) and the definition of in (4.32), we have

tr q q = 2N
2
2
gU
(1) + NgSU(N+1)
2
2
2
gU
(1) (N 1) + 2gSU(N+1) gU (1)
N,
(4.39)
which is a vacuum expectation value of q q, i.e., tr(q q), the invariant norm of the complex
scalar Goldstone fields. Putting the solutions (4.38) into (4.34), the minimization of the reduced
scalar potential with respect to the FayetIlipoulos parameter is realized as follows:

2
2
gU
2
1 N
1
(1) gSU(N+1)
min
VredSC =
+ VredSC ,
2 N (N 1)
+
2
2
2 N + 1 gU
N
+
Ng
(1)
SU(N+1)

2
gSU(N+1)
1
1
1
min
=2
VredSC
N
,
min = ,
min = N 1 2 ,
N
N
+
1
N
N
tr q q min = N (N 1) minimized value of tr q q .
(4.40)
min does
In the second line of (4.40), we can see that the minimized reduced scalar potential VredSC
2
not include the coupling constant gU (1) . We consider the reason why such a surprising consequence occurs. Using the solutions Z 2 = 1 and (4.38), the first term of the reduced scalar
potential in the right-hand side of (4.34) is calculated as
2
gU
2
(1)
+ 1 N + 2 2Z 2 1
2(N + 1)

2
2
2
2
2
gU
gU
(1)
(1) (N 1) + 2gSU(N+1) gU (1)
=
+1N +
2
2
2(N + 1)
gU
(1) + NgSU(N+1)
2

2
4
gU
N2
2
(1) gSU(N+1)
=
N 1
2
2
2
2(N + 1) (gU
N
=min
(1) + NgSU(N+1) )
= 0,
(4.41)
where we have used the min in (4.40). This fact means that a zero of the reduced scalar potential
in the model with gauged U (1) occurs in the process of the minimization of the reduced scalar
potential with respect to the FayetIlipoulos parameter .
5. Proper solution for extended supersymmetric -model
To find a proper solution for the extended supersymmetric -model, after rescaling the Goldstone fields Q by a mass parameter, as van Holten et al. did [2,10], we also introduce the
135
(N + 1)-dimensional matrices Rf (Qf ; G), Rf T (Qf ; G) and Xf in the following forms:
Rf (Qf ; G) = B AT Qf Qf A + f Qf B Qf ,
Rf T (Qf ; G) = A + f Qf B ,

1
(5.1)

r
q
1
f f
,
Xf = 1N+1 + f 2 Qf Qf
= X , Qf =
1 T
f rf
0
def 1
rf =
(x
+
f
q
x),
f
=
.
2
m
2Z
Due to the rescaling, the Killing potential M is deformed as
iMf (Qf , Q f ; G) = trf (Qf , Q f ; G),
def
2
f (Qf , Qf ; G) = Rf T (Qf ; G) Rf (Qf ; G)f Qf Xf

(5.2)
= f Qf AQf AT f BQf + f Qf B Xf ,
from which we obtain an f -deformed Killing potential Mf
iMf B = f Qf Xf ,
iMf B = f Xf Qf ,
iMf A = 1N+1 2f 2 Qf Xf Qf .
(5.3)
After the same algebraic manipulations, the inverse matrix Xf in (5.1) leads to a different form
deformed from the previous one (4.16)

1
Qf qq Qf qr
= f ,
,
f = 1N + f 2 qq
Xf =
(5.4)
Qf qr Qf r r
where each sub-matrix is expressed by the variables q and r and the parameters f and Z as

Qf qq = f Z 2 f rf rf f 2 q rf rfT q f ,
(5.5)
Qf q r = f Z 2 f q rf , Qf r r = Z 2 (see Appendix C).
Substituting (5.1) and (5.4) into (5.3) and introducing an f -deformed auxiliary function f =
rf rf f 2 q rf rfT q = f , we can get the f -deformed Killing potential Mf A as,
iMf A
1N 2q f q
2
+ 2 Z2 (f 2 q f f f q + f 2 q f q rf rfT
f
+ f 2 rf rfT q f q rf rfT )
2
2 f1 rf f q + 2 Zf (rf f f f q
+ rf f q rf rfT )
2
2 f1 q f rf + 2 Zf (q f f f rf
+ rf rfT q f rf )
1
Z
1 2 2 rf f rf + 2 2 rf f f f rf
f
f
(5.6)
in which the identities similar to (4.21) hold even if r, and are changed to rf , f and f .
Using rf = 2Z1 2 (x + f q x),
the following relations also can be easily proved:
12

Z2
1
1
1
r
r
+
2
rf f f f rf = 2 2Z 2 1 + 1 2 ,
f
f
f
2
2
f
f
f
f
(5.7)
136
f f f rf =
q f q =
1 Z2
f rf ,
Z2
rf f f f =
1 Z2
r f ,
Z2 f
1
(1N f ).
f2
(5.8)
Using these, we get a more compact form of the f -deformed Killing potential Mf A as,
1 2q q
N
f
Z2
2
2
q
r
f
f
Z
1
T
iMf A = + 2 f (f q f rf rf f q f f rf rf f )
.
2
2 Zf rf f q
1
(2Z 2
f2
1) + 1
1
f2
(5.9)
Owing to the rescaling, the f -deformed reduced scalar potential is written as follows:
2
2
gU
gSU(N+1)
(1)
2
2
Vf redSC =
( iMf Y ) +
tr(iMf t ) ,
2(N + 1)
2
1
(iMf Y )2 ,
tr(iMf t )2 = tr(iMf A )2
N +1
iMf Y = tr(iMf A ),
(5.10)
in which each f -deformed Killing potential is computed straightforwardly as

1
Z2
1
tr(iMf A ) = 1 2 2 N + 2 2 tr(f ) + 2 2 tr f rf rf
f
f
f
2

Z
1
1
4 2 tr f rf rf f + 2 2Z 2 1 + 1 2 ,
(5.11)
f
f
f

1
1
1
1
tr(iMf A )2 = N 4 2 1 2 N 4 4 tr(f ) + 4 4 tr(f f )
f
f
f
f

1
1
1
1
+ 4 2 1 2 Z 2 tr f rf rf 4 2 1 2 Z 2 tr f rf rf f
f
f
f
f
2
2

Z
+ 12 4 tr f rf rf f 16 4 tr f f rf rf f
f
f

Z4
Z4
4 4 rf f f rf tr f rf rf + 8 4 rf f f rf tr f rf rf f
f
f
+

1
1 2
Z4
1
2
1
+
2
1
+
1
4
r f f rf ,
1
2Z
f4
f2
f2
f2
f4 f
(5.12)
and the quantities rf f f rf and tr(rf rf ) can be approximately calculated in a fashion similar
to (4.32) as

1
1
1 Z2
def 1
2
rf f f rf =
x

,

=
tr
q
q
,
N
+
f
f
f
f
N
4Z 4
Z2
2

def 1 Z 1

1 1
tr rf rf = rf rf =
x f x rf rf , rf rf =
.
4
2
f
4Z
Z
(5.13)
137
In (5.11) and (5.12), approximating tr(f ), tr(f rf rf ), etc., by f , f tr(rf rf ), etc., respectively, and using (5.13), tr(iMf A ) and tr(iMf A )2 are computed as

1
1
tr(iMf A ) = 1 + 1 2 2 N + 2 2 2Z 2 1 f ,
f
f

1
1
tr(iMf A ) =1 + N 4 2 1 2 N
f
f
(5.14)

1
1 1
2
2Z 1 1 2 Z f
4 2
f f2
f

1
4
+ 4 4 2Z 1 f .
f
Substituting (5.14) into (5.10), we obtain the f -deformed reduced scalar potential as
2

1
1
2
+ 1 + 1 2 2 N + 2 2 2Z 1 f
Vf redSC =
2(N + 1)
f
f

2
gSU(N+1) 1 1
1
1
+2
N 1 2 N + 1 + 3 2 Z 2 f
N +1 f2 f2
f
f

1
1
+ 1 2 N + 1 + 2 f
f
f

1
+ 2 2(N 1)Z 4 + 4Z 2 (N + 2) f 2 ,
f
2
gU
(1)
(5.15)
and, from the variation of this with respect to Z and f , we get the following relations:

1
1
2
2
gU
2Z
1

+
1
+
1
2
N
+
2
f
(1)
f2
f2

1
1
1
2
2gSU(N+1)
1 2 N + 1+3 2
4
f
f

1
2
2 (N 1)Z + 1 f = 0,
(5.16)
f

2

1
1
2
2
gU
(1) + 1 + 1 2 2 N + 2 2 2Z 1 f 2Z 1
f
f

1
1
1
2
1 2 N + 1 + 3 2 Z2
2gSU(N+1)
2
f
f

1
1
1
1
1 2 N
1+ 2
2
2
f
f

1
4
2
2 2(N 1)Z + 4Z (N + 2) f = 0.
(5.17)
f
Multiplying by (2Z 2 1) for (5.16) and using (5.17), we have a g 2 -independent relation

(N + 1) 4 Z 2 1 f 1 f 2 = 0.
(5.18)
138
Then, due to 0 Z 2 1 and f > 0, the rescaling parameter f is shown to satisfy f 2 1.

From (5.18), we can finally reach our ultimate goal of proper solutions for Z 2 and f as
2
2
gU
(1) + NgSU(N+1)
2
2
Z =1+ 1f
,
2
2
2
2
gU (1) {(2 f 2 )N 1} gSU(N+1) {(1 f 2 )N 2} gU (1) f 2
f =
2
2
2
2
2
2
1 gU (1) {(2 f )N 1} gSU(N+1) {(1 f )N 2} gU (1) f
.
2
2
2
gU
(1) + NgSU(N+1)
(5.19)
This is the solution for the
SO(2N+2)
U (N+1)
supersymmetric -model. The solution (5.19), if
f2
= 1,
SO(2N+2)
U (N+1)
reduces to a simple solution (4.38). In order to find a proper solution for the extended
supersymmetric -model, minimization of the f -deformed reduced scalar potential has been
made after rescaling Goldstone fields by the inverse of mass parameter f . Then the proper solutions for Z 2 and f (5.19) have been produced.
From the definition of the f in (5.13), the vacuum expectation value of q q, tr(q q), i.e.,
the invariant norm of the complex scalar Goldstone fields is obtained as
2
2
gU
2N
N
(1) + NgSU(N+1)
tr q q = 2 2
2.
2
2
2
2
2
f gU (1) {(2 f )N 1} gSU(N+1) {(1 f )N 2} gU (1) f
f
(5.20)
If
f2
= 1, this reduces to the result (4.39).
6. Discussions and concluding remarks

In the course of producing the proper solutions for Z 2 and f (5.19) in the preceding
section, the FayetIlipoulos term has made a crucial role. Substituting (5.19) into (5.15), the
minimization of the f -deformed reduced scalar potential with respect to the FayetIlipoulos
parameter is realized as follows:

2
2
2
gU
1 N
1
1
1
(1) gSU(N+1)
12 2 N +N +2 2
+
Vf redSC =
2
2
2 N + 1 gU
N
f
f
+
Ng
(1)
SU(N+1)
min
+ Vf redSC ,

(6.1)
1
1 1
min = 1 2 2 N 1 2 2 ,
f
f N

2
2
g
1
1
1
1
1
1
1
SU(N+1)
min
.
N
+
Vf redSC = 2
4
2
2
4
N +1
8
8
f
f
f
f N
Thus we get the minimized potential Vfmin
redSC if we choose the FayetIlipoulos parameter to
be min and have the same surprising consequence as in the previous section, vanishing of the
reduced scalar potential with gauged U (1). Putting the min into (5.19), we have
2
Zmin
=
1
1
,
+
2 2 (1 f 2 )N
f min =

1
1
1f2 ,
N
2
f 2 1,
(6.2)
139
which are independent from the coupling constants gU (1) and gSU(N+1) . We get the vacuum
expectation value of q q, tr(q q)min in a form similar to the corresponding one in (4.40) as

2N (N 1) + N 2 (1 f 2 )
.
tr q q min =
f 2 {2 N (1 f 2 )}
(6.3)
Eqs. (6.1), (6.2) together with (6.3), if f 2 = 1, reduce to (4.40).

In this paper, we have given an extended supersymmetric -model on the Khler coset space
G
H
SO(2N +2)
U (N+1) ,
basing on the SO(2N + 1) Lie algebra of the fermion operators. Embedding the
SO(2N + 1) group into an SO(2N + 2) group and using the SO(2N+2)

U (N+1) coset variables [14], we
have investigated a new aspect of the extended supersymmetric -model which has never been
)
seen in the usual supersymmetric -model on the Khler coset space SO(2N
U (N ) given by van Holten
et al. [2]. We have constructed a Killing potential, the extension of the Killing potential in the
SO(2N )
U (N )
coset space to that in the SO(2N+2)

U (N+1) coset space. To our great surprise, the Killing potential
is equivalent with the generalized density matrix which is a useful tool to study fermion manybody problems. Its diagonal-block part is related to a reduced scalar potential with the Fayet
Ilipoulos term. The reduced and the f -deformed reduced scalar potentials have been optimized
in order to see the behaviour of the vacuum expectation value of the -model fields. We have
)
got, if f 2 = 1, a simple solution Z 2 = 1 corresponding to the SO(2N
U (N ) supersymmetric -model
2
and the proper solutions for Z and f . The FayetIlipoulos term has made an important role
to get such solutions.
)
The above simple solution for the SO(2N
U (N ) supersymmetric -model and proper solution for the
SO(2N +2)
U (N+1)
extended supersymmetric -model, however, are only an extremely little part of whole
solutions which should be obtained from anomaly-free supersymmetric coset models. As already
shown in the work by van Holten et al. [2], if we construct some quantum field theories based
on pure coset models, we meet with a serious problem of anomalies in a holonomy group which
particularly occur in pure supersymmetric coset models due to their chiral fermions [17]. This is
)
SO(2N+2)
also the cases for our orthogonal cosets, coset SO(2N
U (N ) and its extended coset U (N+1) , though
each spinor representation of SO(2N ) group and its extended SO(2N + 2) group is anomaly
free. To construct the consistent supersymmetric coset models, we must embed each coset coordinate in each anomaly-free spinor representation of SO(2N ) group and SO(2N + 2) group and
give a corresponding Khler potential and then a Killing potential for each anomaly-free
SO(2N )
U (N )
model and its extended anomaly-free SO(2N+2)

U (N+1) model based on each positive chiral spinor representation. To achieve such a subject on the case of the SO(2N ) group/algebra, van Holten
et al. have proposed the method of constructing the Khler potential and then the Killing potential [2]. This idea is considered to be very suggestive and useful for our present purpose of
constructing the corresponding Khler potential and then the Killing potential for the case of
the SO(2N + 2) group/algebra. This problem will be worked out in a separate paper based on a
spinor or a vector representation of SO(2N + 2) group/algebra. Finally, we have given bosonization of the SO(2N + 2) Lie operators, vacuum functions and differential forms for their bosons
expressed in terms of the SO(2N+2)
U (N+1) coset variables, a U (1) phase and the corresponding Khler
potential. This provides a powerful tool for describing the Goldstone bosons but accompanying fermionic modes in the present model. The effectiveness of SO(2N+2)
U (N+1) Khler manifold is
expected to open a new field for exploration of low-energy elementary particle physics by the
supersymmetric -model.
140
Acknowledgements
One of the authors (S.N.) would like to express his sincere thanks to Professor Alex H. Blin
for kind and warm hospitality extended to him at the Centro de Fsica Terica, Universidade
de Coimbra, Portugal. This work was supported by the Portuguese Project POCTI/FIS/451/94.
The authors thank the Yukawa Institute for Theoretical Physics at Kyoto University. Discussions
during the YITP workshop YITP-W-07-05 on String Theory and Quantum Field Theory were
useful to complete this work.
Appendix A. Bosonization of SO(2N + 2) Lie operators
Consider a fermion state vector | corresponding to a function (G) in G SO(2N + 2):

| = U (G)|00|U (G)| dG = U (G)|0 (G) dG.

(A.1)
The G is given by (2.27) and (2.28) and the dG is an invariant group integration. When an
infinitesimal operator IG + G and a corresponding infinitesimal unitary operator U (12N+2 + G)
is operated on | , using U 1 (12N +2 + G) = U (12N+2 G), it transforms | as

U (12N+2 G)| = (IG G)|

= U (G)|00|U (12N +2 + G)G | dG

= U (G)|0 (12N +2 + G)G dG

= U (G)|0(12N +2 + G) (G) dG,
(A.2)

1N+1 + A
B
12N+2 + G =
,
B
1N+1 + A
tr A = 0,
B = B ,
A = A,
(A.3)

1
Bpq E qp + Bpq Eqp ,
p
q
qp
G = A q E p + Bpq E + Bpq E qp .
2
Eq. (A.2) shows that the operation of IG G on the | in the fermion space corresponds to
the left multiplication by 12N +2 + G for the variable of the G of the function (G). For a small
parameter , we obtain a representation on the (G) as

e G (G) = e G G = (G + GG) = (G + dG),
(A.4)
G = Ap q E q p +
which leads us to a relation dG = GG. From this, we express it explicitly as,

dA d B
AA + BB
AB + B A
dG =
=

,
dB d A
BA + AB
A A + B B
A
B

dA =
= (AA + BB),
dB =
= (BA + AB).

(A.5)
141
A differential representation of (G), d(G), is given as

p

A q
Bpq
A p q
Bpq
d(G) (G) =
+
+
+
(G).
Ap q
Bpq
A p q
Bpq
(A.6)
Substituting (A.5) into (A.6), we can get explicit forms of the differential representation
d(G) (G) = G (G),
(A.7)
where each operator in G is expressed in a differential form as
Bqr
A q r
+ Ap r
=
E
,
E p q = Bpr
p
Bpr
Aq r
Bqr
A p r
p
q
qp
Bqr
A r
+ Bpr
=E ,
E pq = A r
Ap r
Aq r
Bqr
Bpr
p
p
E pq = E pq ,
E pq = E qp .
E q = E q ,
(A.8)
We define the boson operators Ap q , A p q , etc., from every variable Ap q , A p q , etc., as

def 1
def 1
A = A+
,
,
A = A
A
2
2

1
1
def
def
T
,
A = A
,
A = A+
A
A
2
2

T
A = 1,
A,
A, A = 1,

T
= A, A = 0,
[A, A]
A , A = A , A = 0,
(A.9)
where A is a complex variable. Similar definitions hold for B in order to define the boson oper pq , etc. By noting the relations
ators Bpq , B
A
= A A AT A,
A
A
A
= A B BT A,
A
B
A
B
(A.10)
the differential operators (A.8) can be converted into a boson operator representation

pr Aq r Ap r + ApT r A q r = B Bq r Aq r Ap r ,
E p q = Bpr Bqr BTqr B
p r
E pq = Ap r Bqr BTqr A p r Aq r Bpr + BTpr A q r = Ap r Bq r Aq r Bpr ,
(A.11)
by using the notation ApT r+N = Bpr and BTpr+N = Ap r to use a suffix r running from 0 to N
and from N to 2N . Then we have the boson images of the fermion SO(2N + 1) Lie operators as
E = E = B r B r A r A r ,
E = E = A r B r A r B r ,
(A.12)
c = E 0 E 0 = A r A0 r B0r + A0 r B0r B r

def
= 2 A r Y r + Y r B r , Y r = 1 A0 r B0r
2
and E 0 = 0 and E 00 = 0. The last representation for c in (A.12) involves, in addition to the
original A and B bosons and X bosons, their complex conjugate bosons and the Y r
0
142
bosons. The complex conjugate bosons arise from the use of matrix G as the variables of representation and the Y r bosons arise from extension of algebra from SO(2N ) to SO(2N + 1) and
embedding of the SO(2N + 1) into SO(2N + 2).
Using the relations

det A = A1 p q det A,
Ap q

1 r
A s = A1 s q A1 p r ,
Ap q
we get the relations which are valid when operated on functions on the right coset
1 q
A r
=
,
Bpq r<p
Qpr
=
Qrp A1 s q
A1 p q ,
p
A q
Qrs
2

s<r<p
(A.13)
SO(2N+2)
SU(N+1)
(A.14)
using which, the bosonized operators (A.11) are expressed by the closed first order differential
+2)
form over the SO(2N
U (N+1) coset space in terms of the coset variables Qpq and a phase variable .
Appendix B. Vacuum function for bosons
We show here that the function 00 (G) in G SO(2N + 2) corresponds to the free fermion
vacuum function in the physical fermion space. Then the 00 (G) must satisfy the conditions

1
p
E q + pq 00 (G) = E pq 00 (G) = 0, 00 (12N+2 ) = 1.
(B.1)
2
12 , the proof of
The vacuum function 00 (G) which satisfy (B.1) is given by 00 (G) = [det(A)]
which is made easily as follows:

1
1
p
2
E q + pq det(A)
2

1

1
1
q
p
2
2
= pq det(A) + Bpr
Bqr
A r
+A r
det(A)
q
p
2
B
A
Bqr
A r
pr
r
1

1

1
2 A q r
2
det(A)
= pq det(A)
2
A p r

1
1
1
21
A q r
= pq det(A)
det(A)
1
2
2 [det(A)]
2
A p r

1

1
1
21
= 0,
A A 1 qp det(A)
= pq det(A)
(B.2)
1
2
2 [det(A)]
2

1

1
p
q
2
2 = 0.
E pq det(A) = A r
Bqr
A r
+ Bpr
det(A)
p
q
A r
A r
Bqr
Bpr
(B.3)
The vacuum functions 00 (G) in G SO(2N + 1) and 00 (g) in g SO(2N ) satisfy

1
c 00 (G) = E + 00 (G) = E 00 (G) = 0, 00 (12N+1 ) = 1,
(B.4)
2

1
e + 00 (g) = e 00 (g) = 0,
2
143
00 (12N ) = 1.
(B.5)
By using the SO(2N + 2) Lie operators E pq , the expression (2.22) for the SO(2N + 1) WF
|G is converted to a form quite similar to the SO(2N ) WF|g as

1
pq
|G = 0|U (G)|0 exp
(B.6)
Qpq E
|0,
2
where we have used the nilpotency relation (E 0 )2 = 0. Eq. (B.6) leads to the property
U (G)|0 = U (G)|0. On the other hand, from (2.29) we get

T 1
1+z
1
1z
det A =
det a,
det B =
+
x q x x x det b = 0. (B.7)
2
2
2(1 + z)
Then we obtain the vacuum function 00 (G) expressed in terms of the Khler potential as

1
2 = e 14 K(Q,Q ) ei 2 ,
0|U (G)|0 = 00 (G) = det(A)

1
1 + z
1 + z 1 K(q,q ) i
2
00 (G) = 00 (G) =
=
e 2.
det(a)
e 4
2
2
(B.8)
(B.9)
Appendix C. Derivation of (5.5)

Using the representation for Qf given in (5.1), the inverse of the matrix Xf , Xf1 , is expressed
as
Xf1
f1 + rf rf
f q rf
f rfT q
1 + rf rf
f1 = 1N + f 2 qq ,
(C.1)
from which the inverse matrix Xf (= [1N+1 + f 2 Qf Qf ]1 ) in (5.4) is given as

Xf =
Qf qq
Qf qr
Qf qr
Qf qq =
f1
Qf qr =
Qf r r
+ rf rf
1 + rf rf
Qf r r =
f
1 + rf rf
f2
1 + rf rf
q rf rfT q
Qf qq q rf ,
1
,
(C.2)

1 + f rfT q Qf qr .
Let us introduce an N N matrix Yf defined in the form

Yf = f rf rf
f2
1 + rf rf
f q rf rfT q .
(C.3)
The Qf qq in (C.2) is related to Yf by

Qf qq = (1N + Yf )1 f .
By repeated uses of both the relation and the identity
(C.4)
144
rf f rf =
1 Z2
,
Z2
rfT q f q rf =

1
1
1
+
r
,
f f
f2
Z2
rfT q f rf = rf f q rf = 0,
(C.5)
we can compute Yfn (n 1) as

1 n1
1 Z 2 n1
1
n
n1
f rf rf (1)
1
Yf =
Z2
1 + rf rf Z 2
f2
1 + rf rf
f q rf rfT q .
(C.6)
Using the formula for an infinite summation, an inverse matrix (1N + Yf )1 is calculated as
(1N + Yf )1 = 1N +
(1)n Yfn
n=1

= 1N
n=0
+
n=0
1 Z2
Z2
n
f rf rf
1
1
1 + rf rf Z 2
1
n
f2
1 + rf rf
= 1N Z 2 f rf rf + Z 2 f 2 f q rf rfT q .
f q rf rfT q
(C.7)
The geometric series expansion appearing in the right-hand side of the first, second and third
lines of (C.7) converge only if Z 2 is sufficiently close to 1. However, by direct calculation of
(1N + Yf )(1N + Yf )1 using (C.3), the last equation of (C.7) and (C.5), it is proved to be 1N .
Then one can see that the expression obtained for the matrix (1N + Yf )1 in (C.7) is indeed true
for all 0 Z 2 1. Substituting (C.7) into (C.4) and (C.2), we can get (5.5). If we put f = 1, we
also get (4.17).
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
B. Zumino, Phys. Lett. B 87 (1979) 203.

S. Groot Nibbelink, T.S. Nyawelo, J.W. van Holten, Nucl. Phys. B 594 (2001) 441.
K. Higashijima, T. Kimura, M. Nitta, Nucl. Phys. B 623 (2002) 133.
U. Lindstrm, M. Rocek, Nucl. Phys. B 222 (1983) 285;
N.J. Hitchin, A. Karlhede, U. Lindstrm, M. Rocek, Commun. Math. Phys. 108 (1987) 535.
M. Arai, M. Naganuma, M. Nitta, N. Sakai, Nucl. Phys. B 652 (2003) 35.
N.N. Bogoliubov, Sov. Phys. Usp. 67 (1959) 236.
P. Ring, P. Schuck, The Nuclear Many-Body Problem, Springer, Berlin, 1980.
A. Perelomov, Generalized Coherent States and Their Applications, Springer, 1986;
A. Perelomov, Phys. Rep. 146 (1987) 135, and references therein.
F. Delduc, G. Valent, Nucl. Phys. B 253 (1985) 494.
J.W. van Holten, Z. Phys. C 27 (1985) 57;
J.W. van Holten, Nucl. Phys. B 260 (1985) 125;
Y. Achiman, S. Aoyama, J.W. van Holten, Phys. Lett. B 150 (1985) 153;
S. Aoyama, Z. Phys. C 32 (1986) 113.
H. Fukutome, M. Yamamura, S. Nishiyama, Prog. Theor. Phys. 57 (1977) 1554.
J. Schwinger, in: L.C. Biedenharn, H. vanDam (Eds.), Quantum Theory of Angular Momentum, Academic Press,
New York, 1965, p. 229.
[13] M. Yamamura, S. Nishiyama, Prog. Theor. Phys. 56 (1976) 124.

[14] H. Fukutome, Prog. Theor. Phys. 58 (1977) 1692;
H. Fukutome, S. Nishiyama, Prog. Theor. Phys. 72 (1984) 239.
[15] H. Fukutome, Int. J. Quantum Chem. 29 (1981) 955;
H. Fukutome, Prog. Theor. Phys. 65 (1981) 809.
[16] K. Higashijima, M. Nitta, Prog. Theor. Phys. 105 (2001) 243.
[17] C.L. Ong, Phys. Rev. D 27 (1983) 3044;
G. Moore, P. Nelson, Phys. Rev. Lett. 53 (1984) 1519;
E. Cohen, C. Gomez, Nucl. Phys. B 254 (1985) 235;
S. Aoyama, J.W. van Holten, Nucl. Phys. B 258 (1985) 18.
145

Wilson lines and transverse-momentum dependent

parton distribution functions:
A renormalization-group analysis
I.O. Cherednikov a,b,1 , N.G. Stefanis a,
a Institut fr Theoretische Physik II, Ruhr-Universitt Bochum, D-44780 Bochum, Germany
b Bogoliubov Laboratory of Theoretical Physics, JINR, 141980 Dubna, Russia
Received 25 February 2008; received in revised form 5 May 2008; accepted 14 May 2008
Abstract
The renormalization-group properties of gauge-invariant transverse-momentum dependent (TMD) parton distribution functions (PDF) in QCD are addressed. We perform an analysis of their leading-order
anomalous dimensions, which are local quantities, making use of the renormalization properties of contourdependent composite operators in QCD. We argue that attaching individual gauge links with transverse
segments to quark fields in the light-cone gauge, the associated gauge contours are joined at light-cone
infinity through a cusp-like junction point. We find that the renormalization effect on the junction point
creates an anomalous dimension which has to be compensated in order to recover the results in a covariant
gauge. To this end, we include in the definition of the TMD PDF an additional soft counter term (gauge
link) along that cusped contour. We show that the eikonal factors entering this counter term are peculiar to
the Mandelstam field formalism and are absent when one uses a direct gauge contour.
PACS: 13.60.Hb; 13.85.Hd; 13.87.Fh; 13.88.+e
E-mail addresses: igorch@theor.jinr.ru (I.O. Cherednikov), stefanis@tp2.ruhr-uni-bochum.de (N.G. Stefanis).

1 Alexander von Humboldt Fellow.
doi:10.1016/j.nuclphysb.2008.05.011
I.O. Cherednikov, N.G. Stefanis / Nuclear Physics B 802 (2008) 146179
147
1. Introduction
Theoretical interest in the use of Wilson lines (also termed gauge links or eikonal phases) has
been greatly stimulated recently by both theoretical investigations and experiments on single spin
asymmetries (SSA) [16]see also [7] for a quite recent review and further references. This interest appears in the context of a gauge-invariant formulation of parton distribution functions
(PDF) in terms of hadronic matrix elements. Because these matrix elements involve quark
antiquark field operators at different spacetime points, one has to introduce a path-ordered gauge
link of the form

y
[y, x|C] = P exp ig

(1)
dz Aa (z)ta
x[C ]
that restores gauge invariance, albeit introducing an implicit, i.e., functional dependence on the
contour C adopted. Here, C is, in general, an arbitrary path in Minkowski space and P denotes the
path-ordering instruction that orders the Lie-algebra valued gluon fields with the earliest contour
point x furthest to the right, whereas the final point y is put furthest to the left. [Throughout this
(F)
work, ta stands for the generator Ta of the fundamental representation (labelled F) of color
SU(Nc ). Note that there is an implied sum over the color index a.]
The concept of gauge links is so pervasive throughout YangMills theories because it ensures
local gauge invariance independent of the particular dynamical theory. The concept of using
contour-dependent operators in QCD is an old one and mostly studied in connection with the
renormalization of singularities caused by contour obstructions, like end- or cross-points, and
cuspssee, for instance, in [831] and further references cited therein. The renewed interest in
Wilson lines in SSA and day-to-day applications is due to the potential breakdown of universality
in transverse-momentum dependent (TMD) parton distribution functions (or PDF for short) [32
37] and the ensuing discussion about a process dependence of the gauge link caused by the
color flow of the particular hard-scattering process. As a consequence, it is argued that there is
a change of the overall sign of the single transverse spin asymmetry from the deeply inelastic
scattering (DIS) to the DrellYan (DY) case and, hence, a breakdown of universality [32]. The
content and formulation of this question is coterminous with the properties of Wilson lines within
the purview of specific gauge-fixing prescriptions and the imposition of boundary conditions at
light-cone infinity.
This brings in questions about the role of final (initial) state interactions (FSI) between the
struck quark and the target spectators that may yield important effects like shadowing and SSA
[3840]. The core issue here is the choice of the gauge adopted for the Wilson line that has to
be inserted to ensure gauge invariance. For singular gauges, like the light-cone gauge A+ = 0,
the gauge link vanishes by choice without any restriction put on the gauge potential A at infinity. Then, to avoid light-cone singularities at k + = 0, specific boundary conditions on the gluon
propagator have to be imposed in order to exhaust the remaining gauge freedom and recover
the results obtained in non-singular gauges, say, in the Feynman gauge. It was shown in [41],
and further worked out and detailed in [33], that the effects of FSI in the light-cone gauge can
be properly taken into account by including a gauge link (an eikonalized quark line) which
involves a path in the transverse direction. In covariant, i.e., non-singular, gauges this additional
term does not contribute and, therefore, the modified definition, proposed by Belitsky, Ji, and
Yuan, for the TMD PDF reduces to the correct gauge-invariant one. These important findings not
withstanding, yet a consistent gauge-invariant picture for TMD PDFs in the whole phase space
148
is still incomplete, because the behavior of the gauge contour at infinite transverse distance is
largely arbitrary [41]. The crucial point isas we have recently shown [42]that splitting the
gauge link and allowing the separated contours to stretch out to infinity in the transverse configuration space, induces an additional contribution that cannot be dispensed with by imposing
suitable boundary conditions [33]. Instead, one has to compensate this new contribution by incorporating into the definition of the TMD PDF an eikonal factor which provides a soft counter
term in the sense of Collins and Hautmann [4345].
In the present work, we shall investigate these issues from the point of view of the renormalization group and address parton distribution functionsintegrated and unintegratedaiming
for a more suitable definition of TMD PDFs. Our considerations will employ contour-dependent
operators and we will calculate the leading gluon radiative corrections in the light-cone gauge.
The renormalization of such operators is supremely simple to deal with when stated in terms of
anomalous dimensions because these quantities are local and do not depend on the length of the
gauge contour. Therefore, they provide a powerful tool to access the renormalization properties of
Wilson lines and take into account those contributions originating from geometrical obstructions,
notably, endpoints, or sharp bends in the contour, as first pointed out by Polyakov [8]. In fact, we
will show in more detailed form than in our brief presentation in [42] (see also [46]) that, adopting the light-cone gauge, the leading gluon radiative corrections associated with the transverse
gauge link give rise to an anomalous dimension that exhibits a ln p + behaviorcharacteristic
of a contour with a cusp. We will fathom out the physics underlying this findingin particular,
the renormalization effect on the junction point of two individual gauge contours joined nonsmoothly through a cusp. Moreover, we will present a new definition for the TMD PDF that (i)
reduces to the correct integrated case and (ii) coincides with the result obtained in the Feynman
gauge that is untainted by contour obstructions (the reason being, we reiterate, that in this case
A vanishes at infinity).
The remainder of the paper is organized as follows. In Section 2, we give a summary of the
kinematics used and sketch the spacetime picture of DIS. Section 3 discusses integrated PDFs
and their gauge-invariance and renormalization-group properties. In Section 4, we reinvigorate
our statements on the transverse gauge link, presented briefly in [42], by a formal derivation,
making use of a classical current along the lines of thought described by Jackiw, Kabat, and
Ortiz in [47]. The calculation of the leading one-loop anomalous dimension of the TMD PDF
in the light-cone gauge is outlined in Section 5. Here we also present a generalized factorization
rule for gauge links which takes into account the possibility that the contours may be joined
non-smoothly via a cusp-like junction point. The same section contains the evaluation of the soft
counter term to supplement the definition of the TMD PDF and its interpretation as an intrinsic
Coulomb phase, in analogy to the phase found by Jakob and Stefanis [48] for QED within a
manifestly gauge-invariant formulation in terms of Mandelstam fields [49,50]. In Section 6 we
turn our attention to the real-gluon contributions and the evolution equations, providing a tangible
proof that the integrated PDF obtained from our modified TMD PDF definition coincides with the
correct one with no artefact of the cusped contour left over. Section 7 addresses the application
of our approach to the DrellYan case. Finally, a summary and further discussion of our findings
together with our conclusions is given in Section 8.
2. Kinematics and spacetime picture of (SI)DIS
In what follows, we employ null-plane coordinates with

P = P 0 P 3 / 2,
P = (P + , P , P ),
P 2 = 2P + P P 2 ,
(2)
149
Fig. 1. (a) Schematic Feynman graph for DIS and (b) its spacetime picture at the amplitude level. The thick line denotes
the struck quark, whereas the other solid lines along the x + direction represent spectators, and curly lines mark exchanged
gluons.
where P + > 0 is large and the other components P , P > 0 are small. Moreover, we will
visualize the spacetime structure of the chief hadronic reactions, like DIS, Semi-Inclusive DeepInelastic leptonnucleon Scattering (SIDIS), etc., as a series of snapshots on a plane of equal x +
as depicted in Fig. 1. [For a pedagogical exposition of this graphical method, see, e.g., Refs. [51,
52], and the review in Ref. [53].]
Let us now fix the kinematics relevant for the cases to be considered in our work. We introduce
two light-cone vectors
1
(1, 1, 0 )
2
with the following properties of their plus/minus light-cone components
n = (1, 1, 0 ),
n+ =
n = 0,
2,
n n = 1,
n =
n+ = 0,
(3)
1
n =
,
2
(n )2 = n2 = 0,
(4)
where is an arbitrary parameter having the dimension of mass. Then, the momentum of the
initial hadron reads
M2
P 2 = M 2.
n ,
2
The momentum of the struck quark before being measured [51] by the photon is
+
= xP + = 2x
kin = xP + k kin
P = n +
(5)
(6)
with
xM 2
kin
=
,
2 2
k = (0+ , 0 , k ),
k in = k ,
(7)
whereas the off-shell photon has the momentum

q = x n +
Q2
n
2x
q + = 2x ,
Q2
q =
.
2 2x
(8)
150
After the interaction with the highly virtual photon the struck quark acquires the momentum

xx M 2 + Q2
n .
kout = kin + q = (x x )n +
(9)
2x
Choosing a specific Lorentz frame upon setting = Q/(2x ), corresponds to an almost lightlike
hadron that moves along the plus direction. In the Bjorken limit Q2 , the variables x and
x coincide up to O(M 2 /Q2 ) terms, so that one gets for the photon
Q
q+ = ,
2
Q
q = ,
2
(10)
meaning that the photon moves along the negative x 3 -direction, whereas the struck quark (after
being probed by the photon) moves along the minus direction to infinity:
Q2
Q
k =
= = q .
(11)
2 2x
2
Therefore, in the Bjorken limit, one can estimate the corresponding conjugated spacetime lightcone coordinates to be
k + = 0,
xP 1/P + small,
xk
xP+ 1/P large,
xk+
1/k large,
1/k small.
(12)
(13)
In the next section, the spacetime picture, described above, will be used to motivate the introduction of the extra transverse gauge link.
3. Integrated PDFs: Gauge-invariance and renormalization-group properties
A well-known example for an integrated PDF is provided by the single parton distribution
of a quark of fractional longitudinal momentum x and flavor i in a fast moving hadron which
contains the non-perturbative physics in DIS2 :

d ik +
1
H (P ) i ( , 0 ) + i (0 , 0 )H (P ) ,
e
fi/H (x) =
(14)
2
2
where the quark momentum is defined by k + = xP + with x = Q2 /(2P q).
To give this expression a physical meaning, i.e., elevate it to an observable, one has to ensure that it is gauge invariant.3 To achieve this goal, one usually introduces ad hoc a gauge-link
operator
y
[y , x ] = P exp ig
dz
A+
a (0, z , 0 )ta
(15)
which is path ordered along a lightlike line from the point x , where the quark was removed from
the hadronic state by the annihilation operator i (0 , 0 ), to the point y , where it was recreated
2 Strictly speaking, this is an unrenormalized quantity. The renormalization properties of the operators will be considered shortly in detail.
3 Otherwise, one would have to define a standard gauge to be used in all calculations in order to make them comparable because what is a quark in one gauge is a quark plus gluons in another.
151
by i ( , 0 ). The gauge-link operator represents the struck quark as an eikonal parton [54]
with fixed color charge g, after its interaction with the highly virtual photon (0 < Q2 = q 2
1 GeV2 ) in the DIS process, as it moves with (almost) the speed of light along a lightlike line in
the x direction. The spacetime visualization of this process is illustrated in Fig. 1(b).
It should be clear that, ultimately, expressions (14) and (15) have to be quantized, using, for
instance, functional-derivative techniques. This means that in employing Eq. (15) in Eq. (14),
the gluon potential in the gauge link has to be Wick contracted with corresponding terms in the
interaction Lagrangian, accompanying the Heisenberg fermion (quark) field operators. The consequence of this operation is that all terms in the expectation value in (14), proportional to the
gauge parameter and originating from the gluon propagator in a covariant gauge, will ultimately
cancel. This has been explicitly proved for the quark propagator in leading-order perturbation
theory long ago in [14] (see also [12,13] and [55] for alternative formulations). Indeed, it has
been shown in these works that after renormalization in a MS scheme, the anomalous dimension
associated with the ordinary quark self-energywhich is gauge dependentgets additional contributions, stemming from the two endpoints of the gauge link (coined in [14] the connector),4
that exactly cancel its gauge-parameter term, so that the anomalous dimension of the composite
gauge-independent quark propagator (termed hybrid in [14]) is indeed free of the gauge parameter. [Below, the terminology of [14] is adopted.] This translates into the following sum rule
for the anomalous dimensions
hybrid = 2q + connector ,
(16)
where

s CF a
+ O s2 ,
4

s CF (3 + a)
+ O s2 ,
connector =
4
2q =
(17)
(18)
and
hybrid =

3s CF
+ O s2 ,
4
(19)
with s = g 2 /4 , CF = (Nc2 1)(2Nc ) = 4/3, and a being the gauge parameter. Note that we
use the same conventions for the definition of the anomalous dimension as in [14], i.e., we write
hybrid =
1 dZhybrid
2 Zhybrid d
(20)
with analogous expressions for the other anomalous dimensions.

The above sum rule (16) is nothing but a logarithmic, i.e., additive, version of the Ward
Takahashi/SlavnovTaylor, identities in terms of ratios of the various renormalization constants
of the QCD Lagrangian in the MS scheme [9,1214]:
Zconnector Z1q
Z3 Z 3
Nc s
1
=
=
=1+
(3 + a) ,
Zhybrid
Z1 Z1
8

(21)
4 These contributions are generated by singularities at the endpoints of the line integrals that are dimensionally regularized in D = 4 dimensions and give 1/ poles.
152
where, Zhybrid = Zconnector Z2q , with Z2q , Z3 , Z 3 being, respectively, the renormalization constants of the quark, the gluon, and the ghost field, whereas Z1q , Z1 , Z 1 are the renormalization
constants pertaining to the quarkgluonquark vertex, the three-gluon vertex, and the ghost
gluonghost vertex, respectively.5
Some important comments are here in order: (i) This leading-order result can be formally
proved for smooth contours in every order of QCD perturbation theory [10,11,56]. (ii) It was
shown in [9,12,13,56] that the connector can be renormalized by multiplying it with an appropriate renormalization constant and by replacing in the exponent the strong coupling by its renormalized version. Indeed, the renormalization constants Z2q and Z1q do not depend on the path
chosen in any order of g [10,12,13,56] and the crucial SlavnovTaylor identities are satisfied.
This has been explicitly shown in [9,13] up to the order g 4 and to all orders in g in [10]. (iii) The
straight line is actually enough for the renormalization of the connector for any smooth contour
[23,57,58]. The reason is that what matters are only the singularities induced by the endpoints
of the contour, which are multiplicatively renormalizable using exclusively dimensional regularization [14], whereas the specific path itself (for instance, its length) is irrelevantprovided no
local obstructions like cusps and self-intersections are involved. Such obstructions would induce
additional anomalous dimensions because of discontinuities in the contour slope, as discussed in
detail in [8,12,16], that have to be taken into account in the anomalous-dimensions sum rule.6
We will show below the key role of a cusped gauge contour in the eikonalized TMD PDF with a
transverse gauge link.
The eikonalized quark PDF reads [54] (see also [53,60])

1
d ik +
fi/a (x) =
(22)
P | i ( , 0 ) + [ , 0 ]i (0 , 0 )|P ,
e
2
2
and is a manifestly gauge invariant quantity, but has an anomalous dimension that comprises
contributions stemming from the two endpoints of the lightlike contour C ,0 (recall the remarks on the anomalous dimensions given above). Alternatively, one may be tempted to split the
connector [ , 0 ] into two gauge links that connect the points 0 and through a point at
infinity, the aim being to associate each of them with a quark field operator. Inserting a complete
set of intermediate states, one can then recast Eq. (22) in the following form
split
fi/a (x)

1
=
2 n

1
=
2 n
d ik +
P | i ( , 0 )[ , ] |n +
n|[ , 0 ]i (0 , 0 )|P
e
2
d ik +
P | i ( , 0 |C1 )|n +
n|i (0 , 0 |C2 )|P ,
e
2
(23)
where we have introduced the path-dependent Mandelstam fields [49,50]

x
(x|C2 ) = P exp ig
d A
a ( )ta (x),
(24)
[C2 ]
5 It is worth noting that in the covariant gauge a = 3, the gauge-contour divergences cancel among themselves and
the SlavnovTaylor identity (21) becomes trivially satisfied [12,14].
6 Just recently, Pobylitsa [59] has studied inequalities of a particular class of anomalous dimensions depending on cusp
angles.
153
for the fermion and

(x|C1 ) = (x)P exp ig
x
0
d A
a ( )ta
(25)
[C1 ]
for the antifermion. These field operators represent the struck quark as an eikonal line [53] along
a light-like ray in the minus light-cone direction while interacting with the gluon field of the
hadron, thus mimicking the motion of a struck quark in a real experiment [51].
The above two definitions (14) and (23) are equivalent, because the anomalous dimension of
the direct gauge link [ , 0 ] is preserved when splitting the contour into two distinct contours
through infinity. This means, in particular, that the junction pointwhich is shifted to infinity
is not creating any anomalous-dimension artefact. By virtue of the smooth connection of the
contours C1 and C2 , direct calculation shows that the renormalization of the junction point z
preserves the algebraic identity
[x2 , z | C1 ][z, x1 | C2 ] = [x2 , x1 | C = C1 C2 ]
(26)
in any order of the coupling [13]. This is not trivial because in a local gauge-invariant theory the
factorized gauge link does depend, in general, on the junction point. However, for purely lightlike smooth contours C1 and C2 , the above gauge-link factorization property is satisfied and the
transport of color information by two different routes does not depend on the junction point (the
latter being hidden at infinity in our case), with complete cancellation of the contributions at
infinity. This is a well-established property which has been demonstrated by many authors in the
early literature on studies of path-ordered exponentials. The analogous situation for non-smooth
contours, which are not purely lightlike, and the generalization of Eq. (26) will be discussed in
the next section.
4. Derivation of the transverse gauge link
To substantiate the use of the transverse gauge link and prepare the ground for the gaugeinvariant formulation of unintegrated PDFs, which bear a full transverse-momentum dependence,
let us introduce a Coulomb source in terms of a classical current and write

j (y) = g dy (4) (y y ),
(27)
y = v ,
which corresponds to a charged point-like particle moving with the four-velocity v along the
straight line v t . The gauge field related to such a current has the form

A ( ) = d 4 y D ( y)j (y),
(28)
where D is the gluon Greens function in an arbitrary covariant gauge.
Here and in below, we will use two dimensionless light-cone vectors n
1
n = (1, 0 , 1),
2
(n+ )2 = (n )2 = 0,
(n+ n ) = 1,
(29)
and the metric tensor

g = gT + (n+ ) (n ) + (n+ ) (n )
with gT = .
(30)
154
Fig. 2. Spacetime picture of the DIS process with a gauge connector (double line) (a) and a split-gauge link (two double
lines) at light-cone infinity (b). The infinitely distant parts of the contours, where the behavior of the fields is not specified,
are represented by the symbol for ground (earth) in electricity, introduced in [48]. Else, the same designations as in
Fig. 1 are used.
Appealing to the spacetime structure of this process, illustrated in Fig. 2(a), we recast the
current in the form

j (y) = g n+
0
d (4) (y n+ ) + n

d (4) (y n+ )

+
+
dq eiq y
dq + eiq y
+
= g (2) (y ) n+
(y
)
(y
)
n
,
2 q + i0
2 q + i0
(31)
which makes it clear that the first term in this expression corresponds to a gauge field created by
a source moving from minus infinity to the origin in the plus light-cone direction, before being
struck by the photon, whereas the second term corresponds to a gauge field being created by a
source moving, after the collision, from the origin to plus infinity along a light-cone ray in the
minus light-cone direction. Note that the underlying kinematics are those of Section 2.
To continue, we approximate the gluon Greens function D by the free gluon propagator in
the light-cone gauge A+ = 0 and obtain
d 4 k ikz
e
D (k)
(2)4

d 4 k eikz
k (n ) + k (n )
=
,
g
[k + ]
(2)4 k 2 + i0
(z) =
(32)
where 1/[k + ] denotes the regularization prescription to handle the light-cone (pole) singularity
at k + . Taking into account that the n -part of the current (31) does not contribute in the A+ =
(A n ) = 0 gauge, one is able to obtain the transverse components = i = 1, 2 of the gauge
field; viz.,

d 4 k ik
e
D (k) dy + dy d 2 y eiky (y ) (2) (y )
2(2)4

+
i
1
dk + eik
d 2 k k
= g
eik .
2
2 [k + ]
(2)2 k 2
Ai ( ) = gn+
(33)
155
The second, purely transverse, integral over d 2 k in Eq. (33) gets factorized and finally yields
2
i
i
d k k
eik = i ln | |,
(34)
2
(2)2 k 2
while the first integral over dk + can be performed on account of the preferred regularization
prescription for the pole at k + = 0. Note that in Eq. (34), is an auxiliary infrared (IR) regulator
which ultimately drops out from all physical quantities. The connection between the regularization procedure in the momentum-space representation with the behavior of the gauge field at
light-cone infinity can be anticipated from the following expression
dk + eik
= i
2 k + i0

d
dk + i(k + i0)ik +
= i
e
2
d ( + ),
(35)
where = 1. Then, one readily finds

g
C ln | |,
A ( ; = 1) =
(36)
4
emphasizing that this expression is dependent on the boundary conditions via the parameter C .
Obviously, the longitudinal components A vanish.
Note that the magnitude of A in our expression (36) turns out to be two times smaller
than that obtained in Ref. [47] (see also [61] and references cited therein). The reason for this
difference lies in the fact that in our case the source travels not along the whole plus lightlike
axis, but only along half of it, changing its direction at the origin, as a result of its collision with
the hard photon, and in agreement with the physical picture of the process.
Let us continue by supplying within the same approach the gauge field in a general covariant
gauge, characterized by the gauge parameter a, and using the gluon propagator

1
kk
D (k) = 2
(37)
g (1 a) 2
.
k + i0
k + i0
Starting from Eq. (28), one gets after some simple algebraic manipulations
g
A = 0,
A+ ( ) = ( ) ln | |,
A = 0,
(38)
4
where fields in a covariant gauge are marked by a prime accent in order to distinguish them
from those in the light-cone gauge [cf. Eq. (9) in Ref. [47]]. Notice that the a-dependent terms,
which are proportional to k under the d 4 k-integral, do not contribute by virtue of the deltafunction (k ). Next we give the (singular) gauge transformation which connects these two
representations in the light-cone gauge and in covariant gauges:

ALC
= A + ,
( ) =
d A+ ( ).
(39)
Now we are ready to discuss the origin of the transverse contribution in Mandelstams gaugeinvariant formalism in the light-cone gauge. First, note that the analogous expression to the
Mandelstam field (24) in a covariant gauge reads

cov ( ; n+ ) = P exp ig
dz A+
cov (z , ) cov ( , ),
(40)
156
+
where the gauge field A+
cov differs from the special case, A , given by Eq. (38). Second, performing a regular gauge transformation

x
U (x ) = exp ig dz A+
(41)
in the light-cone gauge on both sides of (40), one can eliminate the Wilson-line integral in the
phase. However, the regular transformation U (x ) does not exhaust the gauge freedom in the
light-cone gauge completely and is, therefore, insufficient to trivialize the interaction of the struck
quark with the gluon field of the spectators. More explicitly, a residual singular transformation
Using ( , ) is still allowed and one realizes that the singular gauge transformation (39) reflects exactly this remaining gauge freedom. Carrying out this additional gauge transformation,
one generates an additional phase that is now associated with the quark field itself; viz.,
( , )LC
= Using ( , )LC ( , )

= 1 ig
dz
A+
source (z , ) + O

2
g LC ( , ),
(42)
which is now completely gauge-fixed, and hence, represents a quark with a fixed color charge.
[This is indicated by a wide hat over the label LC which abbreviates light cone.] Finally, by
taking into account expression (36), one finds that the quark wave function in the light-cone
gauge acquires a phase that may formally be written as

2
( , )LC
LC ( , ).
dz ALC
= 1 + ig
source ( , z ) + O g
(43)
The above arguments make it clear that a complete gauge fixing can be achieved in (42) by
inserting the additional singular gauge transformation Using which contains the cross-talk effects
of the struck parton with the light-cone source. As a result, taking the product of two (local)
quark field operators in the fixed light-cone gauge differs from what one finds in a covariant
gauge. This difference is encapsulated in two phase factors so that one gets

, ) + (0 , 0 )
(
LC

= LC ( , )P exp +ig
+
dz ALC
source ( , z )

P exp ig
dz ALC
source ( , z ) LC (0 , 0 ).
(44)
In the next section, we will use the results obtained above in order to demonstrate the role of the
transverse link in the restoration of the prescription-independence of the anomalous dimension
of the TMD PDF. Specifically, the explicit expression for the transverse gauge field at infinity,
Eq. (33), will be used in the diagrammatic calculations of the gluon radiative corrections pertaining to the anomalous dimension of the TMD PDF.
157
5. Calculation of the one-loop anomalous dimension of the TMD PDF in the light-cone
gauge
We have stressed before the importance of the renormalization effect on the junction point of
the decomposed transverse contours at infinity. In this section we will explain exactly what this
means in mathematical detail. We will prove that the factorization of the gauge link into factors,
each associated with a distinct contour starting (ending) at light-cone infinity, has to be modified
to include an additional phase factor which accounts for the cusp anomalous dimension induced
by the junction point of these decomposed contours.
5.1. Definitions
Our starting point is the operator definition of the (unpolarized) TMD distribution of a quark
with momentum k = (k + , k , k ) in a quark with momentum p = (p + , p , 0 ):
fq/q (x, k ) =
1
2

d d 2 ik + +ik
, )[ , ; , ]
q(p)(
e
e
2
2(2)
[ , ; , ] + [ , ; , 0 ][ , 0 ; 0 , 0 ]

(0 , 0 )q(p) + ,
=0
(45)
where the lightlike and transverse gauge links are defined, respectively, by
[ , z ; z , z ] P exp ig
n
Aa t (z + n )
[ , ; , ] P exp ig
d l Aa t ( + l ) .
a
(46)
Let us emphasize that in the definition of the transverse gauge link the contour is defined in terms
of the two-dimensional vector l which is absolutely arbitrary. We will show explicitly that this
arbitrariness does not affect the local properties of the gauge linkin particular the anomalous
dimension.
Employing the light-cone axial gauge
A+ = (A n ) = 0,
(n )2 = 0,
(47)
the gluon propagator has an additional pole, related to the plus light-cone component of the gluon
momentum, and reads

q n
i
+ q n
LC
D
(48)
.
g
(q) = 2
[q + ]
q 2 + i0
To give this expression a mathematical meaning, we apply the following pole prescription

1
1
1
1
1
1
(49)
,
+
,
=
=
[q + ] Ret/Adv q + i
[q + ] PV 2 q + + i q + i
158
Fig. 3. One-loop gluon contributions (curly lines) to the UV-divergences of the TMD PDF in a general covariant gauge.
Double lines denote gauge links. Diagrams (b) and (c) are absent in the light-cone gauge. The omitted Hermitian conjugate diagrams are symbolically abbreviated by (h.c.).
where has the dimension of mass, to be kept small but finite, and where we used the abbreviations Ret for retarded, Adv for advanced, and PV for the principal value.7 In what follows,
we regularize collinear poles by means of the quark virtuality p 2 < 0, whereas IR singularities
are regularized by an auxiliary gluon mass which is put at the end back to zero. The described
regularization procedure works well in the one-loop order, while at a higher loop order one may
need to apply more sophisticated methods.
In the tree approximation, where the gauge links are equal to unity and the quarkgluon interactions vanish, one trivially gets

d d 2 ik + +ik
1
(0)
, ) + (0 , 0 )|p
fq/q (x, ) =
e
p|(
2
2(2)2
1
+
(50)
u(p),
= (p + xp + ) (2) (k ) u(p)
2
where u(p) denotes a quark spinor and summation over spin indices is tacitly assumed. Moreover, we use the short-hand notation
+ u(p)
u(p)
0 (p),
2p +
(51)
which implies
(0)
fq/q
(x, ) = (1 x) (2) (k )0 (p).
(52)
5.2. One-loop calculation

Our objective here is to discuss the leading-order (LO) g 2 quarkgluon interactions, stemming, on one hand, from the standard QCD vertices, and, on the other hand, from the interactions
of the quarks with the gauge links. Because in the TMD case the distance between the quark
fields is spacelike, i.e., = 2 = 0, the UV-divergent contributions arise only due to virtual gluon corrections. Therefore, in LO (alias, at the one-gluon exchange level), diagrams (a),
(b), and (c) contribute in a covariant gauge (see Fig. 3), while in the light-cone gauge, only diagrams (a) and (d) give non-vanishing contributionswith the latter diagram being associated
with the transverse gauge link.
7 We remark that another possible prescriptionthe so-called MandelstamLeibbrandt pole prescription [6264]
is outside the scope of the present investigation, though we will make some related comments in connection with the
anomalous dimension in Eq. (79).
159
The Hermitian-conjugate (h.c. for short) contributions (omitted in Fig. 3) are generated by the
corresponding mirror diagrams. In what follows, we consider first the left set of diagrams
(in order to show explicitly how the transverse gauge link comes into play) and then we take the
sum left + right which gives the total contribution to the TMD PDF.
These diagrams yield contributions proportional to delta-functions:
1-loop
fq/q
(x, k ; , ) = (1 x) (2) (k )0 (p)(p, s , , ),
where 1-loop results from the diagrams shown in Fig. 3. The quark self-energy diagram (a)
gives (in dimensional regularization with = 4 2 )

(p q)

dq
i p
d (q) 2
(a) (p, s ; , , ) = g 2 CF 2
(53)
(2) (p q)2 (q 2 2 + i0) LC
p
with
dLC (q) = g
q (n )
q (n )
,
+
[q ]
[q + ]
(54)
where the dependence on the auxiliary mass scale is hidden in the pole prescription [q + ] and
a ( a).
The g -proportional term gives a Feynman-like contribution, namely,

(p q)
i p
dq
(a)
2
2
Feynman (p, s , , ) = g CF
(55)
(2) (p q)2 (q 2 2 ) p 2
and generates no extra light-cone singularities. After carrying out the momentum integral, one
gets
(a)
Feynman (p, s , , )
=

1
s
2
2
.
CF ( ) 4 2 (1 ) dx x(1 x) 1 2
4
p
xp
(56)
Performing the remaining integral, one finally finds

s
2
(a)
Feynman (p, s , , ) = CF ( )(1 ) 4 2
4
p

2
2 2 p 2
p 2 2
.
1 + 2 + 2 ln
ln
+
O

p
2
p2
(57)
Note that the mirror diagram gives precisely the same contribution, doubling this result.
5.3. Evaluation of the pole-prescription-dependent contributions
The calculation of the [q + ]-dependent part
(a)
pole (p, s , , ; )

dq
1
q i p
q(
p q)
+ + (p q)
2
2
= g CF
+
(2) (p q)2 (q 2 2 )
[q + ]
[q + ]
p2
(58)
160
is more demanding, owing to the presence of light-cone singularities, and we will consider its
evaluation in detail. After some simple transformations of the numerator, we find

i p
dq
(p q)
1
(a)
2
2
+
+
+
pole = g CF
2
.
(p
+ p)
(2)
(p q)2
(q 2 2 )[q + ] p 2
(59)

dq
One observes that the integral (q 2 2 )[q + ] vanishes (which is true for the considered pole prescriptions but not for the MandelstamLeibbrandt one), while the rest can be recast in the form

i p
(a)
pole = g 2 CF 2 (p
+ + + p)
p 1 (p) + n 2 (p) 2 ,
p
(60)
where
( )
i
1 (p) =
(4)/2 (p 2 )
1
0

(1 x)
2
dx
x(1 x) 1 2
[xp + ]
xp
(61)
bearing in mind that by virtue of + + = (n ) = 0 the term 2 (p) does not contribute. Thus,
one has
(a)
(a)
Feynman + pole (p, s , , , )

2 2 p 2
p 2 2
s
2
ln
CF 4 2 ( ) (1 ) 1 + 2 + 2 ln
=
4
p
p
2
p2

1
2
2 + p
(1 x)
2
(62)
.
dx
1 ln x(1 x) 1 2
+O
p+
[x]
xp
2

In order to evaluate the integral dx(1 x)/[x], one has to use a specific pole prescription
for [x]. Let us consider three possible prescriptions: Advanced, retarded and principal value:

1
1
1
1
1
1
1
1
(63)
=
=
=
,
,
+
[x] Ret x + i
[x] Adv x i
[x] PV 2 x + i x i
using temporarily for convenience the short-hand notation = /p + . In the limit of small ,
we keep only logarithmic terms and omit any powers of .

The UV-divergent part (in the MSscheme) then reads

1
s
2 + p
i
(a)
UV = CF 1 ln 4 + E
(64)
i C ,
1 + ln +
4

p+
p
2
where the numerical constant C depends on the pole prescription according to (cf. [33])
advanced,
0,
C = 1, retarded,
(65)
1 , principal value.
2
161
On the other hand, the finite part of the pole-prescription dependent gluon radiative corrections
is

s
2 2 p 2
p 2 2
2
(a)
finite (p, s , , , ) = CF 1 + ln 2 + 2 ln
ln
4
p
p
2
p2

+
2
2 p
i
iC ln 2
1 + ln +
+
p
p
2
p

1
(1 x)
2
(66)
+ dx
ln x(1 x) 1 2
.
[x]
xp
0
Evaluating this UV-finite integral (66) by setting 2 = 0 (which is justified given that this
integral is IR finite and the magnitude of the IR regulator is irrelevant), we obtain
1
dx
0

(1 x)
1
1
ln x(1 x) = 2 + (1 i) Li2
Li2
.
x i
i
1 i
(67)
As a result, the complete prescription-dependent finite part for 2 = 0 becomes

s
i
2 2 + p
2
(a)
finite (p, s , , , ) = CF 1 + ln 2
iC
1
+
ln
ln
4
p+
p+
2
p
p2

1
1
+ 2 + (1 i) Li2
Li2
.
(68)
i
1 i
A key remark here is that any dependence on the pole prescription in Eqs. (64) and (66) is
cancelled by taking into account analogous contributions originating from the transverse gauge
link as depicted in Fig. 3(d).
5.4. Contribution of the transverse gauge link at light-cone infinity
The issue at stake in this subsection is the proof of the cancellation of all pole-prescription
dependent terms in dealing with the gauge-invariant formulation of the TMD PDF. To be more
specific, we focus on the effects related to the interactions with the gluon field of the transverse
gauge link

d l A ( , 0+ ; l + )
P exp +ig
0
d l A ( , 0 ; l )
P exp ig
(69)
and show that the contribution of the diagram 3(d) cancels out all terms proportional to the
numerical factor C in Eqs. (64) and (66) (or equivalently (68)). Before we proceed, note that,
as it is obvious from Eq. (64), the UV-divergent part depends not only on the pole prescription but
also on the logarithmic p + -term. The effects related to this latter dependence will be considered
subsequently.
162
In Section 4, we have worked out the transverse components of the gluon field in the lightcone gauge and found Eq. (33). This expression can be further evaluated to read
2

d q iq l
dq + iq +
+
e
A (q),
e
A ( , 0 ; l ) =
(70)
2
(2)2
finally assuming the form
d l A ( , 0+ ; l )

=
dq + iq +
e
2
d 2 q
i
l A (q)
.
2
(q l ) + i0
(2)
(71)
Consider now the free gluon propagator resulting from the correlation between longitudinal and
transverse gluons; viz.,

A (q)Ai (q ) =
(q 2
q i n
(i)(2)4 (4) (q + q ),
2 )[q + ]
(72)
and use the relation

+
eiq
= 2iC (q + )
[q + ]
(73)
to find the contribution of the diagram in Fig. 3(d):

dq
+ (p q)
(d)
(q + )
(p, , g; ) = g 2 CF 2 2iC
(2)
(p q)2 (q 2 2 )

1
(1 x)(x)
+ p
2
2
.
1
= iC s CF 4 2 ( ) +
2p
[x(1 x)]
p
xp 2
(74)
One sees explicitly that by virtue of the relation

1
1
1
= lim
= PV i(x),
[x] 0 x i
x
(75)
the transverse-gauge link contribution (74) exactly cancels the dependence on the pole prescrip(a)
(a)
tion in both the UV-divergent part UV and in the finite part finite . To show this explicitly,
we collect all pole-prescription dependent terms of diagram (a) in Fig. 3 and add to them the
contribution from the transverse gauge link, i.e., Eq. (74). Then, we have
1
lim
0
0

1 + C
C
dx(1 x)
i2C (x) ln x(1 x)

x i
x + i
1
= lim
dx
0
0

1x
ln x(1 x)
x i
(76)
which establishes the independence of the result on the parameter C the latter encoding
the adopted pole-prescription. As a result, the complete UV-divergent part of the TMD PDF
163
Fig. 4. Graphical representation of a generic TMD PDF (shaded oval) in coordinate space. The double lines denote the
lightlike and transverse gauge links, connecting the quark field points (0 , 0 ) and ( , ), by a composite contour
through light-cone infinity. The contour obstruction at infinite transverse and lightlike distance ( , ) is symbolized
by a cross, whereas the broken line indicates that this obstruction is hidden.
fq/q (x, k ) is

s 1 1 + p
i
+
iC
CF
iC
1
+
ln
4 2p +
p+
2

s 1
i
+ p
= CF 1
(77)
.
1 + ln +

2p +
p
2
(a+d)
UV
(p, , s ; ) =
Next, taking into account that

+
+ p
=+
2p +
and recalling that the mirror (which we termed before right) counterparts of the evaluated diagrams yield the complex-conjugated contributions, one can conclude that the imaginary terms
above mutually cancel, so that the UV-divergent part of diagrams (a) and (d) contains only contributions due to the p + -dependent term, notably,

s
1 3
(a+d)
UV
(78)
(s , ) = 2 CF
+ ln + E + ln 4 ,
4
p
plus those terms originating from the standard MS renormalization. Hence, there is an extra
anomalous dimension associated with the p + -dependent term which at the one-loop level, considered here, is given by

s
3
LC
+ ln + = smooth .
1-loop = CF
(79)
4
p
The difference between smooth and LC is exactly that term induced by the additional divergence which ultimately has to be compensated by a suitable redefinition of the TMD PDF, if we
want to reproduce the same anomalous dimension as in a covariant gauge.
Here, some comments are in order. It was shown (see, e.g., [65]) that the Mandelstam
Leibbrand (ML) prescription [6264]
1
q
1
=
=
[q + ]ML q + + i0 q q + q + i0
(80)
164
Fig. 5. Renormalization effect on the junction point due to gluon corrections (illustrated by a shaded oval with gluon
lines attached to it) for (a) two smoothly joined gauge contours C1 and C2 at point 3 and (b) the same for two contours
joined by a cusp (indicated by the symbol ) at infinite transverse distance (marked by the earth symbol) off the light
cone. All contours shown are assumed to be arbitrary non-lightlike paths in Minkowski space.
yields a p + -independent anomalous dimension of the quark fields, i.e.,

3 s
(81)
CF + O s2 .
4
Moreover, the ML-prescription, in contrast to the PV/Adv/Ret ones (cf. Eq. (65)), entails additional poles in the complex q 0 -plane which allow for a Wick rotation and, therefore, it does
not break the standard power counting rules. However, on the one hand, it is not clear how the
ML-prescription can be related to any boundary conditions on the gauge field at light-cone infinity, thus making the popular initial/final state interactions interpretation questionable. On the
other hand, the ML-regularization appears to be not sufficient for the calculation of the transverse
gauge field at light-cone infinity (in the form of an expression analogous to, say, Eq. (49)). The
latter issue is potentially crucial for reproducing the results obtained in covariant gauges, while
within the PV/Adv/Ret methods, the similarity between the light-cone and covariant gauges can
be explicitly established. These issues will be further investigated and quantified in a separate
work.
To continue, recall that in a covariant gauge the gluon field vanishes at infinity and, hence, the
only anomalous dimensions ensuing from the gauge link stem from its endpoints that are joined
by a smooth direct contour.8 Actually, p + = (p n ) cosh defines an angle between the
direction of the quark momentum p and the lightlike vector n . Then, in the large limit,
one has ln p + , . Thus, we come to recognize that the defect of the anomalous
dimension, , can be identified with the well-known cusp anomalous dimension [17]
LC
ML
=
s
CF ( coth 1),
d
s
= lim
cusp (s , ) = CF .
d
cusp (s , ) =
d
d ln p +
(82)
This is an important observation that deserves to be discussed in some detail.

As we mentioned earlier, splitting the gauge contour for non-purely lightlike contours through
the light-cone infinity, is not equivalent to the situation with a direct contour between the two field
8 It is worth reiterating that all smooth gauge contours yield the same anomalous dimensions,
smooth , as the straight
line between the endpoints, because only the latter are relevant.
165
points. To understand the deeper reason for this difference, we have to study again the algebraic
identity (26) for decomposing (factorizing) gauge contours (links). The crucial question here is
whether the defect of the anomalous dimension, we calculated, is compatible with this identity
when the junction point is assumed to be at infinite distance in the transverse configuration space.
To answer this question, consult Fig. 5. Panel (a) of this figure shows the renormalization effect
(illustrated by a shaded oval with gluon lines attached to it) on the junction point in the algebraic
identity (26). The contour C1 C2 is smooth and non-self-intersecting owing to the assumption
that C1 and C2 are smoothly connected at 3. Then, both contours the direct one, C, and the
decomposed one, C1 C2 , between the endpoints 1 and 2, cannot be distinguished from each other
by switching on gluon quantum corrections. In particular, no anomalous dimension emerges from
the junction point 3 and thus (symbolically)
C = C1 C2 .
(83)
Now we may ask what changes are induced, if we allow the junction point 3 to be shifted to
infinity in the transverse direction off the light cone. The graphics at right of Fig. 5 helps the eye
catch the key features of the situation involving two non-lightlike contours C1 and C2 . It turns out
that the naive assumption that
C = C1 C2
(84)
is incorrect for contours containing transverse segments. Instead, we found that in this case
C = C1 C2 + cusp .
(85)
Consequently, the validity of the algebraic identity (26) is not conserved and we have to replace
it by the generalized gauge-link factorization rule

[2, 1|C] = 2, |C2 , 1|C1 eicusp ,
(86)
which is valid for arbitrary paths in Minkowski space. In this expression, cusp takes care of the
effect induced by the cusp-like junction point. We will consider an explicit example of such a
phase in the next subsection, where we show how to compensate it by an eikonal factor in the
definition of the TMD PDF. One may associate this phase with final (or initial) state interactions,
as proposed by Ji and Yuan in [41], and also by Belitsky, Ji, and Yuan in [33]. However, these
authors (and also others) did not recognize that the junction point in the split contour (taking a
detour to light-cone infinity) is no more a simple point, but becomes a cusp obstruction ln p+
that entails an anomalous dimension as the result of a non-trivial renormalization effect owing to
gluon radiative corrections.
These arguments make it clear that the naive decomposition of gauge contours that stretch
out to light-cone infinity along the transverse direction is erroneous, simply because the basic
algebraic identity (26), which is tacitly assumed, is inapplicable to such contours and has to be
replaced by Eq. (86). It almost goes without saying that the modified factorization rule, expressed
through Eqs. (85) and (86), is valid when one is composing non-smoothly any gauge contours
with a cusp obstruction at the junction point.9 What marks out a cusped contour from all the
others, however, is that it gives rise to an anomalous dimension proportional to ln p+ , i.e., to a
jump in the four-velocity. This is a salient ingredient in describing correctly a DIS process in
spacetime, because if the two quarks (the struck one and a spectator) are separated also in the
9 A similar factorization rule holds for contours joined through a self-crossing point.
166
transverse coordinate space, the gluons emitted mismatch in rapidity and, hence, the contour
liaising them has to have a sharp bend and cannot be the direct one.
5.5. Compensating the defect of the anomalous dimension by a soft counter term
The defect of the anomalous dimension owing to the gauge-contour cusp at light-cone infinity
represents a distortion of the gauge-invariant formulation of the TMD PDF in the light-cone
gauge. To restore its consistency, we have to dispense with the anomalous-dimension artefact
of the cusp. This can be achieved by supplying the original definition of fq/q (x, k ) by a soft
counter term in the sense of Collins and Hautmann [4345,66]:
R (p + , n |0) (p + , n | ),
(87)
where the eikonal factors are given by

+
(p , n |0) =
0|P exp ig

a
d t
Aa ( )
|0,
Ccusp
(p + , n | ) =
0|P exp ig
(88)
d t a Aa ( + ) |0,
(89)
Ccusp
and evaluate R along the non-smooth (non-lightlike) integration contour Ccusp , defined by

Ccusp : = [p+ s, < s < 0] [n
(90)
s , 0 < s < ] [l , 0 < < ] ,
with n
being the minus light-cone vector, as illustrated in Fig. 6.
Contour (90) is obviously cusped: at the origin, the four-velocity p+ , which is parallel to the
plus light-cone ray, is replacednon-smoothlyby the four-velocity n
, which is parallel to
the minus light-cone ray. This jump in the four-velocity becomes visible in the standard leadingorder term

g

ds
(v1 v2 )
ds
= g2
(v1 s v2 s )2

ds
ds
v12 s
(v1 v2 )
,
2(v1 v2 )ss
+ v22 s
(91)
in which v1 = p + , v2 = n , and the change of the four-velocity at the origin produces an angledependence via (v1 v2 ) = p + . This means that exactly at this point the contour has a cusp
that is characterized by the angle ln p + = ln(p n ), and, therefore, the corresponding
eikonal factor (89) gives rise to a cusp anomalous dimension. Obviously, this is exactly what we
need in order to compensate the extra term in the anomalous dimension found in the preceding
subsection.
Next, we show that the one-loop gluon virtual corrections, contributing to the UV divergences
of R and displayed in Fig. 7, yield an anomalous dimension that neutralizes the cusp artefact
. Note that in the light-cone gauge A+ = (n A) = 0 only the first lightlike ray < s < 0
and also the transverse segment contribute, since the other eikonal line along the minus lightlike
ray depends on the longitudinal component of the gauge field and vanishes due to the gauge
condition.
167
Fig. 6. The integration contour associated with the additional soft counter term.
Calculate first the diagram (a) in Fig. 7. In leading order, the first nontrivial term in Eq. (89)
reads
(1-loop)
(u, )
= ig CF u u
2 2

= ig CF
2 2

d
d q

d

d q eiqu( ) q n + q n
g
(2) q 2 2
[q + ]

1
2u+
u2
+
.
(2) q 2 2
(q u i0)2 (q u i0)[q + ]
(92)
The first term in the square bracket vanishes since u is chosen to point along the p + direction, i.e., u = (p + , 0 , 0 ), u2 = 0, and by recalling that in dimensional regularization
u2 /(u2 ) = 0. Notice that in a covariant gauge this diagram would be tantamount to the selfenergy contribution of the struck quark. However, in the light-cone gauge, we are employing,
a second term in the parenthesis in the second line of Eq. (92) appears which stems from the
gluon propagator in that gauge. This term, being not lightlike, also entails a contribution to the
cusp-dependent part, as we will now show. Indeed, one has

dq
1
(1-loop)
a
(93)
,
(u, ) = ig 2 2 CF 2p +
2
2
(2) (q )(q u i0)[q + ]
an expression which would correspond to a vertex-like contribution of the pure gauge link in a
covariant gaugeas one may appreciate from Eq. (91).
The pole-prescription dependent integral can be evaluated in analogy to our previous calculations in the preceding subsection, so that

(1-loop)
a
(u, ) = g 2 CF 2
42
2
( )
(4)2
1
dx
0
1
x
x 2 [ 2x
]
(94)
where the bracketed term in the denominator is to be evaluated with the aid of Eq. (75). The last
(1-loop)
step in obtaining an explicit expression for a
(u, ) is to carry out the line integral over x
(which enters because of the appearance of the gauge link in the lightlike direction). To do so, we
have to take care of the additional (logarithmic) singularity ln , which cannot be regularized
by the parameter , where is an extra regulator. The origin of this singularity is related to
168
Fig. 7. Virtual gluon contributions to the UV-divergences of the soft counter term, given by Eq. (87). The designations
are as in Fig. 3.
the vector u which defines the lightlike direction. There are, of course, several possibilities
how to regularize this type of integral. For instance, one can get a regular expression in
after the integration over das discussed in Refs. [19,21] having recourse to the fact that the
(1-loop)
(u, )/ is -independent. However, for technical convenience, we apply
derivative a
here a different regularization technique by making use of an auxiliary regulator and absorb
the light-cone singularity ln inside a redefined parameter = 2 , the latter being contained
inside the pole-prescription contributioncf. Eq. (75). This is possible, given that does not
depend on the scale parameter and hence does not contribute to the evolution of the considered
quantity (see Section 6).
Performing all these operations and taking into account that u+ = p + , one gets for the UV
part of diagram (a) in Fig. 7

s 1
(1-loop)
() = CF ln + i iC .
UV
(95)

p
2
Evidently, this expression yields a cusp-dependent contribution, ln p + , as we already mentioned, which would be completely absent in a covariant gauge, thus, underlying their mutual
difference. Besides, there is a dependence on the choice of the pole prescription (via the numerical parameter C ). In order to cancel this latter dependence, one needs to take into account the
contribution of diagram (d) in Fig. 7; viz.,

2
d q (l q )
dq
dq + iq +
e
(d) = g 2 CF 2 p +
(2)
2
(2)2 (q 2 2 )[q + ]
1
1
(96)
(i)(2)4 (4) (q + q )
.

q u + i0 q l + i0
Using Eqs. (70)(73), we find

2
d
q
1
2
(d) = ig 2 CF 2 C
=
C
iC
( )
4
s
F
(2)2 q 2 + 2
2
and extracting the UV-pole and adding it to Eq. (95) we finally arrive at

s 1
(a+d)
UV
() = CF ln + i iC + iC

p
2

s 1
= CF ln + i
.

p
2
(97)
(98)
This result exhibits the independence on the pole prescription of the soft factor R, in close analogy to Eq. (79). Taking into account the corresponding mirror diagram (which doubles the real
part and cancels the imaginary one), we obtain the total UV-divergent part of the soft factor in
169
the one-loop order:

(1-loop)
UV
() =
s 2
CF ln + ,
(99)
making it apparent that there is no dependence on the pole prescription, as now all C -dependent
terms are absent.
To conclude, we have shown at the one-loop level that the soft counter term (soft eikonal
factor) has the following two important properties:
(i) it gives rise to the same cusp anomalous dimension as fq/q (x, k ), but with an opposite sign,
and
(ii) it bears no dependence on the choice of the pole prescription to go around the light-cone
singularity in the light-cone gauge (with corresponding terms cancelling among themselves).
Therefore, it is reasonable to redefine the conventional TMD PDF by including into its definition the soft counter term ab initio. This provides
mod
(x, k ; , )
fq/q

d d 2 ik + +ik
1
, )[ , ; , ]
q(p)(
e
=
2
2
2(2)
[ , ; , ] + [ , ; , 0 ][ , 0 ; 0 , 0 ]

(0 , 0 )q(p) (p + , n |0 , 0 ) (p + , n | , ) ,
(100)
which represents one of the main results of our investigation here and in [42].
Let us finish this section by giving a physical interpretation to the soft counter term, using
Mandelstams formalism [49,50]. To this end, we utilize the exponentiation theorem for nonAbelian path-ordered exponentials [17] and recast the exponential operator (89) in the form

s n (u, n ) ,
(u, n ) = exp
(101)
n=1
where the functions n have, in general, a complicated structure that is, however, irrelevant for
our purposes here. The leading term in this series, 1 , is just a non-Abelian generalization of the
Abelian expression
dx dy (x y)D (x y).
1 (u, n ) = 4CF
(102)
Ccusp
Then, by virtue of the current

b
b
d (4) (z v ),
j (z) = t v
(103)
Ccusp
evaluated along the contour Ccusp (cf. Eq. (90)) and where the velocity v equals either u , n ,
or l (depending on the segment of the contour along which the integration is performed), one
can rewrite (102) as follows

1 (u, n ) = t a 4
(104)
dx d 4 z ab D (x z)jb (z).
Ccusp
170
This expression looks formally very similar to the intrinsic Coulomb phase found by Jakob and
Stefanis (JS) [48] in QED for Mandelstam charged fields involving a gauge contour which is a
timelike straight line. The name intrinsic derives from the fact that this phase is different from
zero even in the absence of external charge distributions. Its origin was ascribed by JS to the
long-range interaction of the charged particle with its oppositely charged counterpart that was
removed behind the moon after their primordial separation.10 This phase is acquired during
the parallel transport of the charged field along a timelike straight line from infinity to the point
of interaction with the photon field and is absent in the local approach, i.e., for local charged
fields joined by a connector. It is different from zero only for Mandelstam fields with their own
gauge contour attached to them and keeps track of its full history since its primordial creation.
Keep in mind that the connector is introduced ad hoc in order to restore gauge invariance and is
not part of the QCD Lagrangian. In contrast, when one associates a distinct contour with each
quark field, one, actually, implies that these Mandelstam field variables should also enter the
QCD Lagrangian (see [48] for more details). However, a consistent formulation of such a theory
for QCD is still lacking and not without complications of its own.
The analogy to our case is the following. First, formally adopting a direct contour for the
gauge-invariant formulation of the TMD PDF in the light-cone gauge (Fig. 8(a) shows an example of the contributing diagrams), the connector gauge link does not contribute any anomalous
dimensionexcept at the endpoints; this anomalous dimension being, however, irrelevant for
the issue at stake. Hence, there is no intrinsic Coulomb phase in that case. Second, splitting the
contour and associating each branch to a quark field, transforms it into a Mandelstam field and,
as a result, adding together all gluon radiative corrections at the one-loop order, a p + -dependent
term survives that gives rise to an additional anomalous dimension. We have shown that this extra anomalous dimension can be viewed as originating from a contour with a discontinuity in the
four-velocity x(
) at light-cone infinitya cusp obstruction.
Classically, it is irrelevant how the two distinct contours C1 and C2 in Fig. 5 are joined, i.e.,
smoothly or by a sharp bend. But switching on gluon quantum corrections, the renormalization
effect on the junction point reveals that the contours are not smoothly connected, but go instead
through a cusp. Here, we have a second analogy to the QED case discussed above. Similarly
to the particle behind the moon, this cusp-like junction point is hidden and manifests itself
only through the path-dependent phase (104). Note in the same context that integrating over
the transverse momentum (see next section), the p + -dependent terms, resulting from virtual
gluon corrections, cancel against their counterparts from real-gluon corrections, so that this cuspinduced phase disappears [see for illustration Fig. 8(b)].
In our previous paper [42], we concentrated on the anomalous dimension of the TMD PDF,
and, therefore, only the UV-divergent parts were studied. In the present work, however, we take
also into account the UV-finite parts and, consequently, the dependence on the transverse momentum appears explicitly, as we discussed above.
6. Real-gluon contributions and evolution equations
In this section, we concentrate our efforts on two subjects: (i) First we discuss in some detail
the evolution behavior of the TMD PDF and establish the connection between our approach and
that of Collins and Soper [67]. (ii) Second, we prove that the integrated PDF, obtained from our
10 The existence of a balancing charge behind the moon was postulated before by several authorssee [48] for related
referencesin an attempt to restore the Lorentz covariance of the charged sector of QED.
171
Fig. 8. (a) Schematic illustration of the direct gauge contour with virtual gluon-line insertions (denoted by curly lines).
(b) Similar illustration for real gluon exchanges over the cut, the latter indicated by a dashed line. In both panels double
lines represent the gauge links.
modified definition, coincides with the standard one with no any artefact of the cusped contour
used in the TMD PDF left over.
(i) Evolution behavior. The modified TMD PDF (100) depends on two arbitrary mass-scale
parameters: the UV scale and the extra regulator . The -dependence is described by the
standard renormalization-group evolution equation (see below) and is controlled by the UVanomalous dimension, which arises as the sum of the anomalous dimensions of all the ingredients
of the TMD PDF (100):
fq/q = 2q +
i
gauge
link + R
i=1

3 s
=
(105)
CF + O s2 .
4
The anomalous dimensions associated with the quark fields and the soft counter term R are
marked by self-explaining labels. We have used for convenience a short-hand notation to denote
the anomalous dimension of each gauge link on the right-hand side of Eq. (100) by a number in
the order the gauge link appears from the left to the right.
Before we proceed, a couple of important remarks are here in order. One realizes that the
anomalous dimension of fq/q coincides with the anomalous dimension of the conventional quark
propagator in the light-cone gauge, but with the opposite sign due to the different Dirac structure.
Up to the sign, this result also coincides with the anomalous dimension of the gauge-invariant
quark propagator in a covariant gauge [14]. The anomalous dimension of R, R , cancels precisely
those contributions in the sum above which contain the p + -dependent terms.
On the other hand, the dependence on is more complicated and is described by an integral
kernel to be determined below. In order to derive the corresponding evolution equation, one needs
to calculate the real-gluon contributions, depicted in Fig. 9. Here, we present this calculation in
the small- limit (that corresponds to the large-rapidity limit within the CollinsSoper
approach [67]). Let us emphasize that in the case of the integrated PDFs, where the dependence
on the regularization parameter appears at the intermediate steps of the calculations (in the
172
Fig. 9. The leading-order real gluon contributions to the TMD PDF are shown. The diagrams (b) and (d) with a transverse
gauge link do not contribute to the TMD PDF in the light-cone gauge. The dashed line marks the cut.
light-cone gauge), it cancels out in the final expression. This will be demonstrated below. In
contrast, in the unintegrated PDFs, this dependence remains and, thus, it should be treated by
means of a corresponding evolution equation. Note that the diagrams in Fig. 9 are UV-finite and
do not contribute to the anomalous dimensions. However, they do depend on the regularization
parameter . The diagram (a) yields

+ (p q)

d 4 q (p q)
(a)
2
real = g CF
Disc D (q)
(2)4
(p q)4
(p + k + q + ) (2) (q k ),
(106)
where the absorptive part of the gluon propagator reads

q n + q n
.
Disc D (q) = 2(q + ) q 2 2 g +
(107)
[q + ]PV
In the last equation we have adopted the PV-prescription, because the real gluon contributions
are prescription-independent and the diagrams with transverse gauge links do not contribute. The
-divergences can be isolated by means of the standard rules given in [68]:
1x
1
= (1 x) ln + +
.
2
+
2
p
(1 x)+
(1 x) + (/p )
After some standard calculations, one gets the Feynman (-independent) part
(a)real
Feynman =
|1 x| k 2 + x2 + x(x 3)p 2
s
C
.
F
p + [k 2 + x2 x(1 x)p 2 ]2
2 2
The -dependence appears through the pole-contributions, i.e.,

s
x
1
(a)real
= 2 CF
(1 x) ln + 2
.
pole
(1 x)+
p k + x2 x(1 x)p 2
(108)
(109)
(110)
On the other hand, the diagram (c) in Fig. 9 yields

d 4 q iqu( ) (2)
e
(q k )
(2)4
0
0

q n + q n
.
2(q + ) q 2 2 g
[q + ]PV
In the small- limit, a straightforward calculation gives

s
1
(c)
CF (1 x) 2
1 ln + .
real =
p
2 2
k + 2
real = ig 2 CF u u (p + xp + )
(c)
(111)
(112)
173
Finally, the (logarithmic) dependence of the modified TMD PDF on is determined in terms of
the equation
d mod
(x, k ; , )
f
d q/q

s
2
2
(2)
= CF (1 x) (k ) ln 2 ln 2

1
1
1
+
0 (p).
k 2 + x2 x(1 x)p 2 k 2 + 2
(113)
We can recast this equation, which governs evolution with respect to , in a form which formally
resembles the standard CollinsSoper evolution equation [67,69] with respect to , namely,

d mod
mod
(x, k ; , ) = K() + G(, ) fq/q
(x, k ; , ).
f
d q/q
(114)
The renormalization-group behavior of the functions K() and G(, ) [70] is determined by
the universal cusp anomalous dimension

s
1 d
1 d
ln K() =
ln G(, ) = cusp = CF + O s2 .
2 d
2 d
(115)
Extracting explicit expressions for K() and G(, ) from our Eq. (113), we can readily show
that they each satisfy Eq. (115) with respect to the cusp anomalous dimension. We emphasize
that the parameter in our approach plays a role akin to the rapidity parameter in the additional
evolution equation of Collins and Soper, with Eq. (113) being the analogue of the CollinsSoper
equation.
Therefore, the dependence on the dimensional regularization scale of the re-defined TMD
PDF (100) is given by the following renormalization-group equation

3 s
1 d
mod
(x, k ; , ) =
ln fq/q
CF + O s2 .
2 d
4
(116)
It is important to appreciate that only the modified TMD PDF, given by Eq. (100), obeys such a
simple UV-evolution; without the soft counter term, non-trivial extra contributions would arise in
the corresponding anomalous dimension on the right-hand side of Eq. (105). Taking logarithmic
mod (x, k ; , ) with respect to both scales and , we get
derivatives of fq/q

d mod
d
(x, k ; , ) = 0,
fq/q
(117)
d d
which establishes the formal analogy between our approach and the CollinsSoper one. This
equation ensures the absence of extra UV-singularities related to artefacts owing to the light-cone
gauge and is equivalent to our initial requirement of the cancellation of undesirable divergences.
(ii) Integrated modified PDF. Consider now the integration of Eq. (100) over k . We collect all -dependent terms from the virtual and the real-gluon contributionscontributing UV
divergencesand perform the k integration. We find that the result is -independent, so that
the DGLAP evolution of this quantity is guaranteed. Below, we demonstrate this cancellation
explicitly. The integration of the UV-divergent term (78) trivially gives

s 1
s 1
2 CF ln (1
(118)
x) d 2 k (2) (k ) = 2 CF ln + (1 x).
174
On the other hand, the integration (in the dimensional regularization) of the -dependent realgluon contribution (110) yields

1
s
2 CF ln + (1 x)2 d 22 k 2
p
k + 2

s
42
= 2 CF ( ) ln + (1 x)
,
(119)
p
2
where 2 = x2 x(1 x)p 2 . After extracting the UV-divergent term
2
s 1
CF ln + (1 x),
(120)
we observe that it exactly cancels the right-hand side of Eq. (118). The same cancellation occurs
between the -dependent terms in the virtual and the real-gluon contributions of the soft factor.
Integration over the transverse momentum (using dimensional regularization) in the modified
TMD PDF (100) yieldsat least formallythe integrated PDF

mod
(x, k ; , ) = fi/a (x, )
d 2 k fi/a
(121)
which bears no -dependence as well.
From the above considerations it becomes apparent that the renormalization-group properties
of this distribution are described by the DGLAP equation
d
fi/a (x, ) =
d
j
1

dz
x
Pij
fj/a (z, ),
z
z
(122)
where the integral kernel reads (in leading order)

s
1 + x2
3
Pij (x) = CF
(1 x) +
+ O s2
2
(1 x)+
(123)
and the ()+ -regularization is defined in the standard manner by

1
0
f (z)
dz
(1 z)+
1
dz
0
f (z) f (0)
.
1z
(124)
Thus, one may conclude that the extra cusp-dependent terms are not present in the integrated case
and, consequently, the UV properties of the standard PDFs (governed by the DGLAP equation)
are not affected by the additional parameter , as expected.
7. DrellYan and universality
The study presented in the preceding sections was performed for the semi-inclusive DIS
(SIDIS). In this section we discuss the DY case within our approach and comment on universality.
We now ask ourselves to what extent our analysis can be applied to other reactions, like the
DY lepton-pair production. It was shown by Collins [32] that the direction of the integration
contours in the lightlike gauge links entering the definition of the TMD PDF should be reversed
175
Fig. 10. (a) Spacetime picture of the DrellYan process at the amplitude level. The thick line denotes the struck quark,
whereas the other solid lines along the x + direction represent spectators, and curly lines mark exchanged gluons attaching
to the gauge links (double lines). (b) Splitting the gauge link in the DY process in analogy to SIDIS in Fig. 2 and using
the same designations as there. Notice that the two gauge links are separated by a transverse distance off the light
cone.
relative to the SIDIS, see Fig. 10:

[ , ; , ] [ , 0 ; , 0 ]SIDIS
[ , ; , ][ , 0 ; 0 , 0 ]DY .
(125)
Later, Belitsky et al. [33] argued that the transverse gauge links, they had introduced to exhaust
the gauge invariance in the light-cone gauge, should be used in the DY case with the reverse
sign, keeping in mind that the lightlike gauge links do not contribute. Hence, one has for DY the
following combination of gauge links
[ , ; , ] [ , ; , 0 ]SIDIS
[ , ; , ][ , ; , 0 ]DY .
(126)
In our approach, the latter replacement should be supplied with a change of sign of the additional regulator (cf. the pole prescription in Eq. (49)), i.e.,
SIDIS DY .
(127)
This change reflects, in fact, the different behavior of the gauge fields at the plus and minus lightcone infinity subject to the proviso of different pole-prescriptions. In the non-polarized case (we
exclusively discuss in the present investigation), this affects only the purely imaginary terms of
the UV-divergent partsconsult Eq. (64). Thus, in the DY case, one has

1
s
2 + p
i
(a)
DY
+
+
iC
1
+
ln
,
UV DY = CF 1 ln 4 + E
(128)
4

p+
p+
2
DY differs from the SIDIS case and is defined as
where the numerical factor C
1, advanced,
DY
retarded,
C = 0,
1,
principal value.
2
(129)
These imaginary terms occur at the intermediate steps of the calculations, but do not contribute
to the final expressions for the unpolarized TMD PDFs. Exactly the same arguments hold for
the additional soft factor (92). This means that the (real-valued) UV anomalous dimension of the
176
unpolarized TMD PDFs is universal as regards the SIDIS and the DY processes:
= fDY
.
fSIDIS
q/q
q/q
(130)
This, however, may not be true for the spin-dependent TMD PDFs, since in that case the imaginary parts play a crucial role and, thus, a sign change (expressed in (128)) might indeed affect
the renormalization-group properties and the corresponding evolution equations. This is an interesting task which will be pursued separately elsewhere.
In this paper we have applied renormalization-group techniques to TMD PDFs, defined in a
gauge-invariant way. We have shown by explicit calculation in the light-cone gauge of the oneloop gluon radiative corrections to the quantity fq/q (x, k ) that a contribution appears, which
is proportional to ln p+ . This contribution gives rise to an anomalous dimension that is formally
equal to the universal cusp anomalous dimension and helps unravel a cusp obstruction in the
composed gauge contour at light-cone infinity. The origin of this anomalous dimension can be
traced to the renormalization effect on the junction point of the split contours, each associated
with a gauge link and attached individually to a quark fieldtransforming it into a Mandelstam
path-dependent field [49,50]. Guided by this finding, we derived a generalized factorization rule
for cusp-connected gauge links which contains an additional phase factor and worked it out. For
gauge links joined along lightlike contours, this factor reduces to unity, while for more convoluted
contours which run off to light-cone infinity in the transverse configuration space, this eikonal
factor contributes an anomalous dimension due to the cusp. In this context, we emphasize that
we verified that integrating over the transverse momenta in fq/q (x, k ) no artefact owing to the
contour cusp remains, thus invigorating the validity of the standard integrated PDF.
In order to eliminate the cusp anomalous dimension and recover the well-known results in
a covariant gauge, where the gauge field vanishes at infinity, we proposed a new definition for
the TMD PDF, which includes a soft counter term in the sense of Collins and Hautmann [43,
44], in order to eliminate the contribution from the cusp anomalous dimension. This counter
term enters in addition to the transverse gauge links, previously introduced by Belitsky et al.
[33], and comprises two eikonal factors caused by a particle-like current flowing along a cusped
contour meandering from (0 , + , 0 ) to ( , + , ) with a sharp bend in the transverse
direction. We have argued that each of these eikonal factors resembles in crucial aspects the
intrinsic Coulomb phase found before by Jakob and Stefanis [48] in a formulation of QED in
terms of Mandelstam fields. In the present case, the cusp-like junction point of the two individual
gauge contours plays a similar role as the so-called particle behind the moon, postulated in
QED in connection with the Lorentz-covariance restoration of its charged sector. Both quantities
share the feature of being hidden at infinity and reveal themselves only in terms of (pathdependent) phases, being independent of external charge distributions (QED case) and unrelated
to boundary conditions to avert light-cone singularities (TMD PDF case in QCD). The origin of
the phenomenon is in both cases the same and peculiar to the inclusion of the individual pathdependent exponential into the field operator supposed to describe the quark as a Mandelstam
field. No such effect appears in cases where the dynamics of the process allow one to use a direct
contour between the two field points. In that case, one has to deal only with the connector which
has well-known renormalization properties [14,67].
The intrinsic Coulomb phase in QED tells us that each charged particle, though primordially
separated from its balancing counterpart, is still in harness with it. In the TMD PDF case, this
177
phase accumulates effects due to the interaction of the struck quark with its target spectators, as
pointed out by Ji and Yuan in [41] and reinforced by Belitsky et al. in [33]. However, the existence
of a cusp at light-cone infinity went unnoticed, because in previous works the UV divergences of
the TMD PDF were not considered. In [35] UV divergences were addressed, but only within the
CollinsSoper approach, which is formulated off-the-light-cone and, hence, the ln p+ term does
not appear there at all.
The appearance of the cusp anomalous dimension in the present context is, in actual fact,
not really surprising. We know from the so-called modified factorization of exclusive reactions
that retaining transverse degrees of freedom amounts to the inclusion of Sudakov factors for
each quark in the hard-scattering subprocess [71]see for a review [72]. The connection of the
Sudakov factors to the cusp anomalous dimension within the modified factorization scheme was
worked out in detail in [73] up to the level of the next-to-leading-order logarithmic accuracy.
Our results may have a wide range of applications. Chief among them:
More precise data analyses of various experimental data on hard-scattering cross sections.
Development of more accurate Monte Carlo event generators (to estimate exclusive components of inclusive cross sections) [66,74].
Better description of polarized TMD PDFs and the phenomenology related to SSA and spin
physics [2,3,32,7577].
Acknowledgements
We would like to thank A.P. Bakulev, A.V. Efremov, A.I. Karanikas, S.V. Mikhailov,
P.V. Pobylitsa, and O.V. Teryaev for stimulating discussions and useful remarks. This investigation was partially supported by the HeisenbergLandau Programme (grants 2007 and 2008),
the Deutsche Forschungsgemeinschaft under contract 436RUS113/881/0, and the Russian Federation Presidents Grant 1450-2003-2.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
A.V. Efremov, O.V. Teryaev, Phys. Lett. B 150 (1985) 383.

D.W. Sivers, Phys. Rev. D 41 (1990) 83.
D.W. Sivers, Phys. Rev. D 43 (1991) 261.
J.-w. Qiu, G. Sterman, Phys. Rev. Lett. 67 (1991) 2264.
J.C. Collins, Nucl. Phys. B 396 (1993) 161.
A. Efremov, V. Korotkiian, O. Teryaev, Phys. Lett. B 348 (1995) 577.
V. Barone, A. Drago, P.G. Ratcliffe, Phys. Rep. 359 (2002) 1.
A.M. Polyakov, Nucl. Phys. B 164 (1979) 171.
V.S. Dotsenko, S.N. Vergeles, Nucl. Phys. B 169 (1980) 527.
I.Y. Arefeva, Phys. Lett. B 93 (1980) 347.
R.A. Brandt, F. Neri, M.-a. Sato, Phys. Rev. D 24 (1981) 879.
N.S. Craigie, H. Dorn, Nucl. Phys. B 185 (1981) 204.
S. Aoyama, Nucl. Phys. B 194 (1982) 513.
N.G. Stefanis, Nuovo Cimento A 83 (1984) 205.
S.V. Ivanov, G.P. Korchemsky, A.V. Radyushkin, Yad. Fiz. 44 (1986) 230.
G.P. Korchemsky, A.V. Radyushkin, Phys. Lett. B 171 (1986) 459.
G.P. Korchemsky, A.V. Radyushkin, Nucl. Phys. B 283 (1987) 342.
G.P. Korchemsky, A.V. Radyushkin, Phys. Lett. B 279 (1992) 359.
I.A. Korchemskaya, G.P. Korchemsky, Phys. Lett. B 287 (1992) 169.
A.I. Karanikas, C.N. Ktorides, N.G. Stefanis, Phys. Lett. B 301 (1993) 397.
178
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
[51]
[52]
[53]
[54]
[55]
[56]
[57]
[58]
[59]
[60]
[61]
[62]
[63]
[64]
[65]
[66]
[67]
[68]
[69]
[70]
A. Bassetto, I.A. Korchemskaya, G.P. Korchemsky, G. Nardelli, Nucl. Phys. B 408 (1993) 62.
I.A. Korchemskaya, G.P. Korchemsky, Nucl. Phys. B 437 (1995) 127.
A.I. Karanikas, C.N. Ktorides, N.G. Stefanis, Phys. Rev. D 52 (1995) 5898.
G.C. Gellas, A.I. Karanikas, C.N. Ktorides, N.G. Stefanis, Phys. Lett. B 412 (1997) 95.
A.I. Karanikas, C.N. Ktorides, N.G. Stefanis, S.M.H. Wong, Phys. Lett. B 455 (1999) 291.
A.I. Karanikas, C.N. Ktorides, N.G. Stefanis, Eur. Phys. J. C 26 (2003) 445.
E.V. Shuryak, I. Zahed, Phys. Rev. D 62 (2000) 085014.
M.A. Nowak, E.V. Shuryak, I. Zahed, Phys. Rev. D 64 (2001) 034008.
A.E. Dorokhov, I.O. Cherednikov, Phys. Rev. D 66 (2002) 074009.
A.E. Dorokhov, I.O. Cherednikov, Phys. Rev. D 67 (2003) 114017.
A.E. Dorokhov, I.O. Cherednikov, Ann. Phys. 314 (2004) 321.
J.C. Collins, Phys. Lett. B 536 (2002) 43.
A.V. Belitsky, X. Ji, F. Yuan, Nucl. Phys. B 656 (2003) 165.
D. Boer, P.J. Mulders, F. Pijlman, Nucl. Phys. B 667 (2003) 201.
X. Ji, J.-P. Ma, F. Yuan, Phys. Rev. D 71 (2005) 034005.
X. Ji, J.-P. Ma, F. Yuan, Phys. Lett. B 597 (2004) 299.
J.C. Collins, A. Metz, Phys. Rev. Lett. 93 (2004) 252001.
S.J. Brodsky, P. Hoyer, N. Marchal, S. Peigne, F. Sannino, Phys. Rev. D 65 (2002) 114025.
S.J. Brodsky, D.S. Hwang, I. Schmidt, Phys. Lett. B 99 (2002) 530.
S.J. Brodsky, D.S. Hwang, I. Schmidt, Nucl. Phys. B 642 (2002) 344.
X. Ji, F. Yuan, Phys. Lett. B 543 (2002) 66.
I.O. Cherednikov, N.G. Stefanis, Phys. Rev. D 77 (2008) 094001.
J.C. Collins, F. Hautmann, Phys. Lett. B 472 (2000) 129.
J.C. Collins, F. Hautmann, JHEP 0103 (2001) 016.
F. Hautmann, Phys. Lett. B 655 (2007) 26.
I.O. Cherednikov, N.G. Stefanis, arXiv: 0711.1278 [hep-ph].
R. Jackiw, D. Kabat, M. Ortiz, Phys. Lett. B 277 (1992) 148.
R. Jakob, N.G. Stefanis, Ann. Phys. 210 (1991) 112.
S. Mandelstam, Ann. Phys. 19 (1962) 1.
S. Mandelstam, Phys. Rev. 175 (1968) 1580.
D.E. Soper, Nucl. Phys. B (Proc. Suppl.) 53 (1997) 69.
D.E. Soper, in: J.L. Rosner (Ed.), Theoretical Advanced Study Institute in Elementary Particle Physics (TASI 2000):
Flavor Physics for the Millennium, 430 June 2000, Boulder, Colorado, 430 June 2000, World Scientific, Singapore, 2001, pp. 267316.
J.C. Collins, D.E. Soper, G. Sterman, Adv. Ser. Direct. High Energy Phys. 5 (1988) 1.
J.C. Collins, D.E. Soper, Nucl. Phys. B 194 (1982) 445.
K. Kanaya, Phys. Rev. D 26 (1982) 1758.
J.-L. Gervais, A. Neveu, Nucl. Phys. B 163 (1980) 189.
N.G. Stefanis, in: D.V. Shirkov, D.I. Kazakov, A.A. Vladimirov (Eds.), 10th International Conference on Problems
of Quantum Field Theory, Alushta, Ukraine, 1317 May 1996, Joint Institute for Nuclear Research, RU-141980
Dubna, Moscow Region, Russia, 1996, pp. 199206, hep-th/9607063.
N.G. Stefanis, in: B.M. Barbashov, G.V. Efimov, A.V. Efremov (Eds.), Problems of Quantum Field Theory, Dubna,
Russia, 1317 July 1998, Joint Institute for Nuclear Research, RU-141980 Dubna, Moscow Region, Russia, 1998,
pp. 449453, hep-ph/9811262.
P.V. Pobylitsa, hep-th/0702123.
R. Brock, et al., CTEQ Collaboration, Rev. Mod. Phys. 67 (1995) 157.
Y.V. Kovchegov, Phys. Rev. D 55 (1997) 5445.
S. Mandelstam, Nucl. Phys. B 213 (1983) 149.
G. Leibbrandt, Phys. Rev. D 29 (1984) 1699.
G. Leibbrandt, Rev. Mod. Phys. 59 (1987) 1067.
A.V. Belitsky, A.V. Radyushkin, Phys. Rep. 418 (2005) 1.
F. Hautmann, H. Jung, arXiv: 0712.0568 [hep-ph].
J.C. Collins, D.E. Soper, Nucl. Phys. B 197 (1982) 446.
G. Curci, W. Furmanski, R. Petronzio, Nucl. Phys. B 175 (1980) 27.
J.C. Collins, Adv. Ser. Direct. High Energy Phys. 5 (1989) 573.
J. Grammer, G.D.R. Yennie, Phys. Rev. D 8 (1973) 4332.
[71]
[72]
[73]
[74]
[75]
[76]
H.-n. Li, G. Sterman, Nucl. Phys. B 381 (1992) 129.

N.G. Stefanis, Eur. Phys. J. direct C 7 (1999) 1, hep-ph/9911375.
N.G. Stefanis, W. Schroers, H.-C. Kim, Eur. Phys. J. C 18 (2000) 137.
J.C. Collins, Acta Phys. Pol. B 34 (2003) 3103.
M. Anselmino, F. Murgia, Phys. Lett. B 442 (1998) 470.
P.J. Mulders, R.D. Tangerman, Nucl. Phys. B 461 (1996) 197;
P.J. Mulders, R.D. Tangerman, Nucl. Phys. B 484 (1997) 538, Erratum.
[77] U. DAlesio, F. Murgia, arXiv: 0712.4328 [hep-ph].
179

N = 3 superparticle model
I.L. Buchbinder a , I.B. Samsonov b,c,
a Department of Theoretical Physics, Tomsk State Pedagogical University, 634041 Tomsk, Russia
b Institut fr Theoretische Physik, Leibniz Universitt Hannover, 30167 Hannover, Germany 1
c Laboratory of Mathematical Physics, Tomsk Polytechnic University, 634050 Tomsk, Russia
Received 12 February 2008; received in revised form 10 April 2008; accepted 16 May 2008
Abstract
We consider the formulation and quantization of the N = 3 superparticle model, both with and without
central charge. Without the central charge the action possesses U (3) invariance and therefore is naturally
quantized in the N = 3 harmonic superspace. The quantization reproduces the N = 3 supergauge strength
multiplets, described by analytic N = 3 superfields and a gravitino multiplet as a constrained N = 3 chiral
superfield. When the central charge is present, it breaks the U (3) R-symmetry of N = 3 superalgebra
down to SU(2) U (1), and the corresponding superparticle model is formulated in the N = 2 harmonic
superspace extended by a pair of extra Grassmann variables. The quantization of such a model leads to the
massive BPS N = 3 vector multiplet. It is shown that upon additional superfield constraints such multiplet
reduces to the massive N = 2 vector multiplet.
1. Introduction
The models of relativistic particles and superparticles have deep relations to string and field
theories. They can be considered not only as toy models which hint how to quantize superstring
theories, but also describe the dynamics of D0-branes, point-like objects which form a part of
the physical content of the type IIA string theory. From the quantum field theory point of view,
the quantization of superparticles results in superfield realizations of supersymmetry multiplets
* Corresponding author at: Laboratory of Mathematical Physics, Tomsk Polytechnic University, 634050 Tomsk,
Russia.
E-mail addresses: joseph@tspu.edu.ru (I.L. Buchbinder), samsonov@mph.phtd.tpu.edu.ru (I.B. Samsonov).
1 Alexander von Humboldt fellow at Leibniz Universitt Hannover.
doi:10.1016/j.nuclphysb.2008.05.014
I.L. Buchbinder, I.B. Samsonov / Nuclear Physics B 802 (2008) 180207
181
with corresponding equations of motion and constraints [16]. This is especially important for
the models with extended supersymmetry since their superfield equations of motion are usually
entangled with superfield constraints [7]. In particular, the unconstrained superfield formulation
of N = 2 supersymmetric models of hypermultiplets and gauge multiplet are given within the
so-called harmonic superspace approach [8,9]. This approach is crucial for the superfield quantization of these models which usually requires the use of unconstrained superfields [10] (see also
[11] for applications to problem of effective action).
We point out that free equations of motion for the N = 2 super-YangMills (SYM) and hypermultiplet models can be naturally derived by quantizing the N = 2 superparticle in harmonic
superspace [12,13].2 However, the unconstrained superfield formulation for N = 4 SYM theory
has not been achieved yet despite the many attempts made in this direction. The N = 3 SYM
theory in harmonic superspace [17], which is known as a maximally supersymmetric field theory with the unconstrained superfield formulation, deserves special attention. Since the N = 3
SYM model is equivalent to the N = 4 one on-shell, it can be considered as an N = 3 superfield
formulation for the N = 4 SYM model. Some quantum aspects of the N = 3 SYM model in
harmonic superspace were studied in [18,19].
Inspired by the success of harmonic superspace formulation for the N = 3 SYM model we
pose the question: which N = 3 supermultiplets, apart from supergauge one, admit the description in terms of the N = 3 superfields? For this purpose, we study the relativistic superparticle
model in the N = 3 d = 4 harmonic superspace and quantize it. A generic discussion of d = 4
superparticles in harmonic superspaces with N 2 and the existence in their quantum spectra
of supergauge and supergravity multiplets was given in [1214].
Note that different models of superparticles can be considered, depending on whether their
actions contain mass and central charge terms. In this paper we consider in detail all such N = 3
superparticles and find the superfield realizations of corresponding supermultiplets. In particular, the GuptaBleuler quantization of the massless N = 3 superparticle without central charge
reproduces the N = 3 SYM supermultiplet realized on superfield strengths. These N = 3 superfields satisfy the Grassmann and harmonic shortness conditions [12,20]. Another interesting
multiplet appearing in this case is the N = 3 gravitino multiplet (with the highest helicity 3/2)
which is described by a chiral N = 3 superfield.
The quantization of the N = 3 superparticle with the central charge term is not so straightforward. One of the features in this case is that the central charge in the N = 3 superalgebra breaks
the group of internal automorphisms U (3) down to SU(2) U (1), and therefore the SU(3) harmonic superspace approach is not applicable here. Regarding the preserved R-symmetry group
one can use SU(2) harmonics instead. Therefore the appropriate formulation of such a superparticle is given in the N = 2 harmonic superspace, which is extended by a pair of additional
Grassmann coordinates. We show that the central charge term in this model coincides with the
central charge of N = 2 harmonic superparticle studied in [12,13]. Hence, the quantization proceeds in the same way as in the N = 2 superparticle model and leads to the supermultiplets of
the N = 3 supersymmetry with central charge, realized on superfields in the N = 3 superspace
with SU(2) harmonic variables.
One of the simplest multiplets appearing in the quantization of the N = 2 superparticle with
a central charge is the massive q + -hypermultiplet described by an unconstrained N = 2 analytic
2 The harmonic superspace approach is also very effective for studying 1d supersymmetric models with extended
supersymmetry (see e.g., the recent works [15] and reviews [16] on the supersymmetric mechanics).
182
superfield in harmonic superspace [12,13]. In our case, the quantization results in a similar q +
superfield, which depends on the extra Grassmann spinor coordinate in a chiral way. Such a
superfield describes the massive N = 3 vector multiplet where the mass is related to the central
charge of the superalgebra by the BPS condition. The on-shell field content of this multiplet is
given by 5 complex scalars, 4 Dirac spinors and 1 complex vector. Thus it has twice as much
components compared to the N = 2 non-BPS massive vector multiplet [21].
Naturally, one is led to question, whether it is possible to impose such extra constraints on
the N = 3 massive vector superfield, which eliminate half of the states and reduce the above
multiplet to the N = 2 massive vector multiplet. We give the positive answer to this question and
show that in N = 3 superspace these constraints look very similar to the equations of motion
in the massive N = 1 WessZumino model. These equations preserve N = 3 supersymmetry,
but violate CPT invariance of the multiplet. The resulting N = 3 superfield describes exactly
the N = 2 massive vector multiplet on-shell. In other words, we show that the massive non-BPS
N = 2 vector supermultiplet can be described by the N = 3 superfield under specific constraints,
which violate the CPT invariance of the N = 3 superalgebra with central charge.
The paper is organized as follows. In Section 2 we introduce the actions for the N = 3 superparticle both with and without the central charge term. The quantization of the massless N = 3
superparticle without central charge term is considered in Section 3, where the N = 3 gravitino
and SYM multiplets are derived. In Section 4 we quantize the N = 3 superparticle with the central charge and, in the simplest case, obtain the N = 3 massive vector supermultiplet. We also
show that this multiplet can be reduced to the massive N = 2 vector supermultiplet by imposing
extra superfield constraints. In Section 5 we summarize the results and discuss some unresolved
problems.
2. N = 3 superparticle action in harmonic superspace
2.1. N = 3 superparticle model without central charge term
...
The N = 3 superspace is parameterized by coordinates Z M = {x m , i , i }, where , ,
are the indices of SL(2, C) and i, j, . . . denote the indices of SU(3). In this coordinate system the
superinvariant Cartan forms M = M d are
m = dx m i di m i + ii m d i ,
M
= i = di ,
(2.1)
i
= d i .
In terms of the Cartan forms (2.1), the massive superparticle action is given by

1
d e1 m m + em2 ,
Ssp =
(2.2)
2
where e( ) is the einbein field, m is the mass of the superparticle and is the worldline parameter.
Action (2.2) is invariant under supertranslations
i = i ,
x m = ii m i + ii m i ,
i = i ,
which correspond to conserved charges (supercharges)

i = 2ie1 m i m .
Q
Qi = 2ie1 m m i ,
(2.3)
(2.4)
183
Together with the particle momenta, they generate the N = 3 superalgebra (3.37) after quantization. The model (2.2) also respects the U (3) R-symmetry of the N = 3 superalgebra.
In general, the arbitrary superfields on the full N = 3 superspace with coordinates Z M (as well
as the arbitrary superfields on any extended superspace) have a large number of components and
do not correspond to irreducible representations of the N = 3 superalgebra. The construction of
the irreducible superfields with fewer number of components is realized within the harmonic superspace formalism [8,9]. Following this approach, in the case of the N = 3 supersymmetry [17],
one extends the superspace Z M with the harmonic variables uIi which are SU(3) matrices,
u u = 133 ,
det u = 1.
(2.5)
Conditions (2.5) can be written in terms of the matrix elements

j
uIi u iJ = JI ,
ij k u1i u2j u3k
uIi
as
uIi u I = i ,
= 1,
j
ij k u i1 u 2 u k3
(2.6)
= 1.
(2.7)
Here the capital Latin indices I, J, . . . are SU(3) ones with the values 1, 2, 3. The invariant Cartan
forms on the SU(3) group [12]
JI = duIi u iJ = uIi d u iJ
(2.8)
satisfy the following identity

11 + 22 + 33 = 0,
(2.9)
which can be derived from (2.7).

The action for a particle on the coset space SU(3)/(U (1) U (1)) can be constructed with the
use of the Cartan forms (2.8),
SSU(3) = S + SWZ + S ,

R2
d e1 21 12 + 32 23 + 31 13 ,
S =
2

1 i

is2

is1
1 i
d u i u 1 ui u 1 +
d u 3i u i3 u3i u i3 ,
SWZ =
2
2
3

ij k 1 2 3

J I i
I
i j k
I ui u J J + ui uj uk + ij k u 1 u 2 u 3 2 .
S = d
(2.10)
(2.11)
(2.12)
(2.13)
I,J =1
Action S is a kinetic term which appears here with the constant R 2 /2, SWZ is the WessZumino
term for harmonic variables with the constants s1 , s2 , and action S takes into account the con The Lagrange
straints (2.6), (2.7) with the help of Lagrange multipliers (IJ ) = (IJ ) and = .
multipliers have ten independent real degrees of freedom and the corresponding constraints thus
single out eight independent components from eighteen components of the arbitrary complex
3 3 matrix uIi . In addition, there is the local U (1) U (1) symmetry which further reduces this
number to six, i.e., the particle moves effectively in the coset space SU(3)/(U (1) U (1)).
The variation of action (2.13) over the Lagrange multipliers gives the following constraints
for harmonic variables
JI
S
= uIi u iJ JI = 0,
JI
(2.14)
184
S
j
(2.15)
= ij k u1i u2j u3k + ij k u i1 u 2 u k3 2 = 0.
Eq. (2.14) shows that the matrix u is unitary while (2.15) means Re det u = 1. These constraints
uniquely imply det u = 1. Therefore both conditions (2.6), (2.7) are satisfied and the matrices uIi ,
u iI belong to the SU(3) group. In particular, for any SU(3) matrix we have Im det u = 0, or,
1
2 = ij k u1i u2j u3k ij k u i1 u 2 u k3 = 0.
(2.16)
Hence, (2.16) appears as a consequence of (2.14), (2.15). This constraint will be used in the next
section for constructing the Dirac bracket.
The unitary matrices uIi rotate the Grassmann variables and supersymmetric Cartan forms
(2.1),
i I = uIi i ,
i I = u iI i ,
i I = uIi i .
i I = u iI i ,
(2.17)
The N = 3 harmonic superspace is parameterized by a set of coordinates ZH = {x m , I , I , u},

where I , I are given by (2.17). Apart from the usual complex conjugation, there is also
conjugation which acts on the Grassmann variables and harmonics as (cf. [9,20])
u1i u i3 ,
1 3 ,
u2i u i2 ,
2 2 ,
u3i u i1 ,
3 1 .
(2.18)
This conjugation is natural in harmonic superspace ZH , since the N = 3 SYM action is known
to be real under (2.18). Applying the conjugation to action (2.10) swaps the constants s1 , s2 in
the WessZumino term. Therefore, action (2.10) is real under (2.18) if s1 = s2 .
The action of the N = 3 superparticle without central charges moving in the harmonic superspace ZH is a sum of (2.2) and (2.10),

Ssp + SSU(3) = Ssp + S + SWZ + S = d L1 ,
(2.19)
where L1 denotes the Lagrangian of the superparticle.
2.2. N = 3 superparticle model with the central charge term
The superparticle action (2.2) admits the extension by the WessZumino term [4],

Sc = d Z ij i j + Zij i j .
(2.20)
Here, Zij and its conjugate Z ij are the constant antisymmetric matrices,
Z ij = Z j i ,
Zij = Zj i ,
(Zij ) = Z ij .
(2.21)
The WessZumino term (2.20) is also invariant under supersymmetry (2.3), but up to a total
derivative. Added to action (2.2), it leads (upon taking into account boundary contributions) to
the conserved Noether supercharges with central charge terms,

i = 2ie1 m i m + 2Zij j ,
Qi = 2ie1 m m i + 2Z ij j ,
(2.22)
Q
which generate the N = 3 superalgebra with central charge (4.32) after quantization. Therefore
we refer to action (2.20) as a central charge term in the superparticle model.
185
Since Zij are constants, they break the U (3) R-symmetry of the N = 3 superalgebra. To understand which symmetry survives, we notice that any 3 3 antisymmetric matrix is degenerate,
det Z = 0,
det Z = 0.
(2.23)
Moreover, performing some rotation with the SU(3) matrix

brought to the normal form [21,22],

0 z 0
ij
ij
i j kl
Z
Z = vk vl Z = z 0 0 ,
0 0 0

0 z 0

k l
Zij Zij = vi vj Zkl = z 0 0 .
0 0 0
vji ,
the matrices Zij ,
Correspondingly, (2.20) takes the following form

Sc = d z 1 2 2 1 + z 1 2 2 1 .
Z ij
can be
(2.24)
(2.25)
(2.26)
Note that action (2.26) is nothing but the WessZumino term of the N = 2 superparticle, which
respects the SU(2) symmetry realized on the coordinates 1 , 2 and 1 , 2 . There is also the
U (1) symmetry which transforms the 3 and 3 variables with a phase factor,
3 ei 3 ,
3 ei 3 .
(2.27)
As a result, we conclude that the WessZumino term (2.20) breaks the internal automorphisms
symmetry U (3) down to the SU(2) U (1).
The Grassmann variables of the N = 3 superspace can be rearranged as

1 2 3

1 , 2 , 3 i , 3 ,
(2.28)
i , 3 ,
, ,
where the underlined indices i, j are the SU(2) ones with the values 1, 2. Action (2.26) can now
be written as

i
Sc = d z ij i j z ij j ,
(2.29)
where ij is the antisymmetric two-dimensional tensor, 12 = 12 = 1. Action (2.29) has manifest SU(2) invariance realized on the indices i, j . Therefore a natural harmonic extension of such
superparticle model is given by the SU(2) harmonic variables u
i ,
u
i SU(2),
ui = ij u
j ,
u+i u
i = 1.
(2.30)
The Cartan forms on the SU(2) group

+i
i++ = u+
i du ,
i
i = du
i u ,
+i
i = u
i du
(2.31)
are used to write down the particle action on a sphere S 2 SU(2)/U (1) [12,13],

i

+i
+i
SSU(2) = 2R 2 d e1 ++ d u
.
u
+i u
n
d u
i
i u
i u
2
(2.32)
186
Here R is the radius of the sphere, n is the electric charge of the particle which couples to a
magnetic field produced by a monopole situated in the center of the sphere (see [12] for details)
and is the Lagrange multiplier.
The harmonics u
i convert SU(2) indices i, j , . . . into U (1) ones , e.g.,
= u
i ,
i
i
i

= ui .
(2.33)
3
Apart from usual complex conjugation, the harmonic superspace {x m , ,
, 3 , , ui } has
also conjugation defined as3
i
u
i =u ,

= ,

i = u ,
u
i
= ,

3

3 = ,
3 = 3 ,

(2.34)
which leaves the action (2.32) invariant.

3
The action of a superparticle in harmonic superspace {x m , ,
, 3 , , ui } is given by the
sum of (2.2), (2.29) and (2.32)

Ssp + Sc + SSU(2) = d L2 ,
(2.35)
where L2 denotes the Lagrangian of the N = 3 harmonic superparticle with central charges.
3. Quantization of the N = 3 superparticle without the central charge term
3.1. Hamiltonian formulation and constraints
We start with the action of the N = 3 harmonic superparticle given by (2.19). The corresponding Lagrangian reads
1
1 m
R2 1 2
m +
+ 32 23 + 31 13 em2
2e
2e 2 1
2

is
is1 1 i
2 3 i
u u u1i u i1 +
u u u3i u i3
2 i 1
2 i 3
3

j
+
JI uIi u iJ JI + ij k u1i u2j u3k + ij k u i1 u 2 u k3 2 .
L1 =
(3.1)
I,J =1
The canonical momenta are defined by the Lagrangian (3.1) in a standard way,
L1
= e1 m ,
x m

L1
i = = ipm m i ,
i

L1
i =
= ipm i m = i ,
i
pm =
(3.2)
(3.3)
(3.4)
3 This conjugation is natural in the N = 2 harmonic superspace with coordinates {x m , , , u } [9] and acts on

i
extra Grassmann variables 3 , 3 as the usual complex conjugation.
187
vi1 =
is1 1
L1 R 2 2 1
=
ui 2 + u3i 31
u ,
i
2e
2 i
u 1
(3.5)
vi2 =

L1 R 2 1 2
=
ui 1 + u3i 32 ,
i
2e
u 1
(3.6)
vi3 =
is2 3
L1 R 2 2 3
=
ui 2 + u1i 13 +
u ,
i
2e
2 i
u 3
(3.7)
v1i =
is1 i
L1
R2 i 2
=
u 2 1 + u i3 13 +
u ,
1
2e
2 1
u i
(3.8)
v2i =

L1
R2 i 1
=
u 1 2 + u i3 23 ,
2
2e
u i
(3.9)
v3i =
L1
u 3i
is2 i
R2 i 2
u 2 3 + u i1 31
u .
2e
2 3
(3.10)
Note that Eqs. (3.3), (3.4) are the constraints since they do not allow to express the Grassmann
velocities through the corresponding momenta.
The standard mass-shell constraint appears from the equation of motion for the einbein,
0=

1
L1
= 2 m m R 2 21 12 + 32 23 + 31 13 e2 m2 .
e
2e
(3.11)
It is convenient to pass from the canonical harmonic momenta viI , vIi to the covariant ones
CJI ,
DJI ,
DJI = uIi vJi u iJ viI ,
CJI = uIi vJi + u iJ viI ,
(3.12)
which can be written manifestly as

DJI = R 2 e1 JI
D11
CJI
= is1 ,
=0
D22
(I = J ),
= 0,
D33
(3.13)
= is2 ,
I, J = 1, 2, 3.
(3.14)
(3.15)
Eqs. (3.14), (3.15) are nothing but the constraints for the harmonic variables. In terms of covariant
momenta DJI , the mass-shell constraint (3.11) reads
p m pm

1 1 3
D3 D1 + D21 D12 + D32 D23 m2 = 0.
2
R
Let us now define the Poisson brackets in a standard way:

n

n
,
x , pm P = m
j

i
j
i , P = i ,
, j P = ji ,
i J
I j
j
u I , vj P = ji IJ .
ui , vJ P = i JI ,
(3.16)
(3.17)
The harmonic covariant momenta (3.13) form su(3) algebra with the Cartan generators S1 =
D11 D22 , S2 = D22 D33 under the Poisson brackets (3.17), e.g.,
188

D21 , D32 P = D31 ,
1 2
D2 , D 1 P = S 1 ,

D21 , D31 P = 0,
2 3
D3 , D 2 P = S 1 ,

D32 , D31 P = 0,

S1 , D21 P = 2D21 ,
(3.18)
etc.
There are the following non-trivial Poisson brackets between the functions
constraints (2.14), (2.16)
I K

2
CJ , L P = 2JK LI ,
D2 , 2 P = 2,
I K
CJ , CL P = JK DLI LI DJK AIJK
L,
CJI ,
D22
and the
(3.19)
which mean that they are second-class. Hence, they can be taken into account by introducing the
Dirac bracket

1
1
[f, g}D = [f, g}P + f, CJI P IJ , g P f, IJ P CJI , g P
2
2

K

1
1
1
L
L , g P ,
+ f, D22 P [2 , g]P [f, 2 ]P D22 , g P f, JI P AJI K
2
2
4
(3.20)
where f and g are arbitrary phase space functions. The applications of the Dirac and Poisson
brackets resemble only the harmonic variables and momenta, while for the other superspace
coordinates one can freely use the Poisson bracket instead of (3.20).
Eqs. (3.14) contain the following first-class constraints
S1 is1 0,
S2 is2 0.
There are also the spinor constraints

Di = i + ipm m i 0,
(3.21)

D i = i ipm i m 0,
which anticommute non-trivially under the Poisson brackets (3.17),

i

D , D j P = 2iji m pm .
(3.22)
(3.23)
In general, the constraints (3.22) are second-class. However, if the dynamics of the superparticle
is constrained to the surface pm p m 0, the matrix in the rhs of (3.23) is degenerate and the
superparticle action is invariant under -symmetry.4 In this case both first-class and second-class
constraints are entangled in (3.22). As follows from (3.16), the condition pm p m 0 is satisfied
only for the massless superparticle, m = 0, with the following additional constraint for harmonic
variables
D21 D12 + D32 D23 + D31 D13 0.
(3.24)
As we will show in the next section, this case corresponds exactly to physical supermultiplets
upon quantization.
Apart from the constraints considered above, it is meaningful to introduce the following extra
harmonic constraints [12,13]
D21 0,
D32 0,
D31 0.
(3.25)
4 The -symmetry was first observed in the model of a massive N = 2, d = 4 superparticle with the central charge [4]
and in the case of a massless N = 1 superparticle in [23]. The group-theoretical and geometrical origin of -symmetry
as a manifestation of local extended supersymmetry of the superparticle worldline was found in [24]. This observation
leads to the development of the superembedding approach to the description of superbranes (see [25] for a review and
references).
189
These constraints are first-class and reduce the mass-shell condition (3.16) to the physical one,
p m pm m2 0.
(3.26)
Eqs. (3.25) freeze the dynamics of harmonic variables leaving only the motion of a particle
in {x m , i , i } superspace. This fact emphasizes the unphysical meaning of the harmonic variables in field theory. Indeed, upon quantization, these constrains will yield the equations which
eliminate an infinite number of auxiliary fields with arbitrary number of SU(3) indices and leave
only physical components.
The canonical Hamiltonian is defined using the momenta (3.2)(3.10) via the Legendre transform,
H1 = x m pm u Ii vIi u iI viI + i i + i i L1 ,
(3.27)
where L1 is given by (3.1). Now we express the velocities from (3.2)(3.10) and substitute them
into the Hamiltonian (3.27),

e 3 1
e m
D1 D3 + D12 D21 + D23 D32
p pm m2 +
2
2
2R
+ 11 (S1 + is1 ) + 33 (S2 is2 ) L ,
H1 =
(3.28)
where L corresponds to the last line in the Lagrangian (3.1) with Lagrange multipliers. This
Lagrangian L is not essential when the harmonic constraints (2.14), (2.15) are taken into account by the Dirac bracket (3.20). Therefore we omit L further, assuming the use of the Dirac
bracket (3.20) in what follows. Note also that the velocities 11 , 33 can not be eliminated from
the Hamiltonian and remain arbitrary functions. They play the role of Lagrange multipliers (as
well as the einbein e) and we denote them as ( ), ( ), respectively. As a result we obtain the
total Hamiltonian which is a linear combination of first-class constraints,

e
1
H1 = p m pm 2 D13 D31 + D12 D21 + D23 D32 m2
2
R
+ (S1 + is1 ) + (S2 is2 ).
(3.29)
The Hamiltonian equation of motion for any phase space coordinate f has the standard form
f = [f, H1 ]D .
(3.30)
Here we do not write down these equations in detail, however they can be easily figured out.
3.2. GuptaBleuler quantization
According to the postulates of canonical quantization, one replaces the canonical momenta
(3.2)(3.10) by the corresponding differential operators,
pm i
viI
,
x m
,
u iI
i i
vIi
.
uIi
,
i
i i
,
i
(3.31)
190
The spinor constraints (3.22) turn into the covariant spinor derivatives,5
Di =

+ i m i m ,

D i = i i i m m ,

(3.32)
and the covariant harmonic momenta (3.12) lead to the covariant harmonic derivatives,
DJI = uIi
u iJ i .
uJi
u I
(3.33)
Further we will use covariant spinor derivatives contracted with harmonics,

DI = uIi Di ,
D I = u iI D i ,
which satisfy the following algebra

I
D , D J = 2iJI m m ,
(3.34)
I J
D , D = 0,
{D I , D J } = 0.
(3.35)
The supercharges (2.4) are promoted to the operators

Qi =

+ i m i m ,

Q i = i i i m m ,
which form the N = 3 superalgebra,

i

i
j
Q , Q j = 2iji m m ,
Q , Q = 0,
i , Q
} = 0.
{Q
j
(3.36)
(3.37)
The operators (3.31) should be realized in some Hilbert space formed by the functions |,

| = x m , i , i , u .
(3.38)
Superfield (3.38) should satisfy some equations of motion and constraints which originate from
the superparticle constraints. The superparticle has both first- and second-class constraints. The
first-class constraints form closed algebra under the Poisson or Dirac bracket. Therefore, they all
should be imposed on states (3.38),
S1 (s1 ,s2 ) = s1 (s1 ,s2 ) ,
S2 (s1 ,s2 ) = s2 (s1 ,s2 ) ,

1
m m + 2 X + m2 (s1 ,s2 ) = 0,
R
(3.39)
(3.40)
where6
X = D12 D21 + D23 D32 + D13 D31 .
(3.41)
(s1 ,s2 )
is a function of harmonic variables with definite

Eqs. (3.39) mean that the superfield
U (1) charges. Note that Eqs. (3.39) covariantly constrain the SU(3) harmonic dynamics to the
one on a coset SU(3)/(U (1) U (1)). Eq. (3.40) is the mass-shell constraint which gives Klein
Gordon-like equations for all physical fields. Note that the zero modes of the operator (3.41)
(s ,s )
(states which are annihilated by this operator, X0 1 2 = 0) satisfy the standard KleinGordon
5 The operators D i , D
i are multiplied here also by i for convenience, so that the operators (3.32) are related to
each other by complex conjugation rater than Hermitian one. The same concerns the derivatives (4.26)(4.28) as well as
the supercharges (3.36), (4.31) and U (1) charges S1 , S2 in (3.39).
6 Here we use a particular ordering of the operators D I although other orderings are also possible.
J
equation without a harmonic term,

(s ,s )
m
m + m2 0 1 2 = 0.
191
(3.42)
The second-class constraints are accounted either by constructing the corresponding Dirac
bracket or by applying the GuptaBleuler method. In our case, the second-class harmonic constraints (3.19) are taken into account by the Dirac bracket (3.20), while the spinorial ones (3.22)
should be considered la GuptaBleuler. It means that they have to be divided into two complex
conjugate subsets with weakly commutative constraints in each subset. There are four different
ways of separation of derivatives (3.32) or (3.34) into such subsets:
1 2 3
D , D , D {D 1 , D 2 , D 3 },
(3.43a)
1

2
3
D , D2 , D3 D1 , D , D ,
(3.43b)

1 2
3
D , D , D 3 D 1 , D 2 , D ,
(3.43c)

1

2
3
D 1 , D , D 3 D , D 2 , D .
(3.43d)
Different choices of subsets (3.43a)(3.43d) lead to different types of quantization of the superparticle. In the following subsections we consider them separately.
3.2.1. N = 3 gravitino multiplet
In this subsection we will show that the separation of fermionic constraints (3.43a) leads to
the N = 3 gravitino multiplet with the highest helicity 3/2. First, we consider the massive case,
m = 0, where the spinor constraints Di 0, D i 0 are second-class since the matrix in the
rhs of (3.23) is invertible. There is no -symmetry in the model and, hence, no extra constraints.
According to (3.43a), the physical state is annihilated only by the derivative D i , while Di kills
the conjugate superfield,
D i (s1 ,s2 ) = 0,
Di (s1 ,s2 ) = 0.
(3.44)
The dynamics of such a field is described by the set of Eqs. (3.39), (3.40), (3.44) which take
into account all the superparticle constraints. Note that Eqs. (3.44) are nothing but the chirality
conditions for the field (s1 ,s2 ) . Therefore we refer to such a quantization as a chiral quantization.
The non-zero modes of the operator X propagate analogously to zero ones. It means that the
superfield (s1 ,s2 ) describes an unphysical multiplet with infinite number of component fields.
To make it physical, we impose the additional harmonic constraints (3.25),
D21 (s1 ,s2 ) = 0,
D32 (s1 ,s2 ) = 0,
D31 (s1 ,s2 ) = 0.
(3.45)
In general, function (s1 ,s2 ) is given by a series in harmonic variables. Eqs. (3.45) reduce this
series to a monomial
is
j1 ...js
(s1 ,s2 ) = u i31 u 3 2 u1j1 u1js
1 s2
j1 ...js
i1 ...is 1
2
s2
(3.46)
where i1 ...is 1 2 is a totally symmetric traceless tensor. Indeed, D21 , D32 , D31 are raising operators
2
in the su(3) algebra which define the highest weight vector (3.46) [18,20]. As a result, we obtain
the chiral superfield with fixed number of SU(3) indices (symmetric and traceless) on mass-shell,
m
j1 ...js s
j1 ...js s
D i i1 ...is 1 2 = 0,
(3.47)
m + m2 i1 ...is 1 2 = 0.
2
192
Let us consider, e.g., the simplest representation without SU(3) indices that corresponds to
the choice of U (1) charges s1 = s2 = 0. The solution of the chirality condition D i = 0 is most
naturally given in chiral coordinates y m = x m + ii m i ,
(y, ) = + i i + i j ij k Fk () + i j d (ij )
+ i j k ikl l + i j k ij k T( )

+ i j k l ikr j lm S(rm) + i j k l ij k Gl()

+ i j k l m ij k lmn n + i j k l m n ij k lmn U,
(3.48)
where all components depend on y m . Note that both bosonic and fermionic components in (3.48)
satisfy the KleinGordon equation owing to (3.47), but there are no Dirac equations for spinors.
Therefore, such a multiplet is unphysical. We assume that the mass should be introduced not
directly but through a central charge, as is shown in the next section. Therefore for the rest of
this section we consider only the massless case, m = 0.
In the massless case the superparticle action (2.2) is well known to respect the -symmetry
since the matrix in the rhs of (3.23) is degenerate. Half of the constraints Di 0, D i 0 turn
into first-class ones. But if we deal with the harmonic superparticle with the action (2.10), the
standard mass-shell constraint (3.42) is replaced by Eq. (3.40) with the harmonic contribution
due to operator X. Therefore, for the states out from the kernel of operator X, the matrix m m
is invertible and the constraints Di 0, D i 0 still belong to the second class. As explained
before, the -symmetry of harmonic superparticle is restored and one half of these second-class
constraints turn into first-class ones, if the dynamics is constrained by (3.24). Upon quantization,
constraint (3.24) is imposed on the states implying the condition X = 0. Namely such states
are interesting from the physical point of view.
In particular, on the surface of constraints (3.25), the condition (3.24) is satisfied and the kinetic part of the action for harmonics (2.11) can be omitted. Therefore the dynamics is described
effectively by the action

1
m m
S=
(3.49)
d
+ SWZ + S ,
2
e
which is invariant under the following transformations of -symmetry

i = ipm i m ,
i = ipm m i ,
x m = i i m i ii m i ,

e = 4 i i + i i ,
(3.50)
i ( ) are anticommuting local parameters. Note that the harmonic terms SWZ ,
where
S , given by (2.12) and (2.13) respectively, do not violate the -symmetry as the harmonics do
not transform, uIi = 0, u iI = 0. Therefore the fields (s1 ,s2 ) with all values of U (1) charges
should obey the constraints originating from the -symmetry.
The transformations (3.50) are generated by the Poisson brackets of coordinates with the
following first-class constraints,
i ( ),
i = ipm m D i 0,
i = ipm m D i 0.
(3.51)
Upon quantization, (3.51) turn into the differential operators,

i = m m D i ,
i = m m D i ,
(3.52)
193
where Di , D i are the covariant spinor derivatives (3.32). These differential operators should
annihilate the physical states,
m m D i (s1 ,s2 ) = 0.
(3.53)
is a chiral N = 3 superfield (3.44) constrained by (3.39), (3.42),

(3.45). First-class constraints (3.42) and (3.53) arise from the following one as the integrability
conditions (cf. [12,13] in N = 2 case)
Recall that the superfield (s1 ,s2 )
D i Dj (s1 ,s2 ) = 0.
(3.54)
It is the constraint (3.54) which leads to the correct component structure of the multiplet and
eliminates all auxiliary fields in the decomposition (3.48), despite the fact that it is stronger than
(3.42) and (3.53). Therefore, we use further (3.54) rather than (3.42), (3.53).7
Let us summarize all the equations for the superfield (s1 ,s2 ) in a single list,
S1 (s1 ,s2 ) = s1 (s1 ,s2 ) ,
S2 (s1 ,s2 ) = s2 (s1 ,s2 ) ,
D21 (s1 ,s2 ) = D32 (s1 ,s2 ) = D31 (s1 ,s2 ) = 0,

D i (s1 ,s2 ) = 0,
D i Dj (s1 ,s2 ) = 0.
(3.55)
The solution of the pure harmonic constraints in the first two lines of (3.55) is given by (3.46).
The other constraints in (3.55) give the chirality and linearity conditions,
j1 ...js s
D k i1 ...is 1 2 = 0,
j1 ...js
D k Dl i1 ...is 1
s2
= 0.
(3.56)
Consider Eqs. (3.56) in the simplest case of the scalar superfield without SU(3) indices,
D i = 0,
D i Dj = 0.
(3.57)
The component structure of a general chiral superfield is given by (3.48). The linearity condition eliminates all unphysical components,
= + i i + i j ij k Fk() + i j k ij k T( ) .
(3.58)
Owing to (3.57), the component fields in (3.58) satisfy the standard dAlembert, Weyl and
Maxwell equations,
= 0
1 complex scalar,
m m i = 0
m m Fk() = 0
m m T( ) = 0
3 Weyl spinors,
3 vectors,
1 gravitino.
(3.59)
Therefore, superfield describes the N = 3 supersymmetric gravitino multiplet.

7 In fact, (3.54) is a consequence of -symmetry constraints (3.51) since on the surface of constraints (3.22), (3.26)
j
the following relations dij i j D i D j , d ij i j D i D hold. It is easy to see that the constraints
j
dij 0, d ij 0 are the generators -transformations (3.50) with the parameters i = k ij j , i = kij , where k ij ,
kij are new bosonic local parameters. Analogously, the constraint (4.44) follows from -symmetry ones (4.41).
194
3.2.2. N = 3 supergauge multiplet

In this subsection we will show that the separations of constraints (3.43b)(3.43d) lead to the
N = 3 supergauge multiplet [12,20]. First, we will analyze the separation (3.43b) in details and
then will comment on the (3.43c) and (3.43d) cases.
Recall that superfield (s1 ,s2 ) satisfies the first-class constraints (3.39), (3.40). Now we impose
also the spinorial constraints from the first subset in (3.43b),
D1 (s1 ,s2 ) = 0,
D 2 (s1 ,s2 ) = 0,
D 3 (s1 ,s2 ) = 0.
(3.60)
(s1 ,s2 )
is analytic, i.e., it is short in the component exConstraints (3.60) show that superfield
pansion. Therefore, we refer to such type of quantization as an analytic quantization. Note that
different types of analytic subspaces in full N = 3 harmonic superspace introduced in [20] correspond to different subsets of Grassmann derivatives (3.43b)(3.43d) annihilating the state.
It is easy to observe that the spinor derivatives in (3.60) do not commute with the operator X
given by (3.41),
1
D , X = D2 D21 D3 D31 ,
[D 2 , X] = D 1 D21 + D23 D 3 ,
[D 3 , X] = D 2 D32 + D 1 D31 .
(3.61)
Therefore, the analytic quantization is consistent only if the state (s1 ,s2 ) satisfies extra harmonic
constraints (3.45). Owing to these constraints, the operators in the rhs of (3.61) vanish on the state
while the constraint (3.40) has no harmonic part and turns into the usual mass-shell constraint
(m = 0),
(s1 ,s2 ) = 0.
(3.62)
Eqs. (3.45) leave only zero modes of the operator X, which are massless. For such modes, the
harmonic variables are not dynamical and the action (3.49) possesses -symmetry (3.50). Let us
project the generators of -symmetry (3.52) with harmonics,
I = m m D I ,
I = m m D I .
(3.63)
(s1 ,s2 )
since the constraints (3.51) are first-class.

Operators (3.63) should annihilate the state
Owing to the analyticity (3.60), it is sufficient to impose three of the six operators (3.63) as the
constraints,
m m D 1 (s1 ,s2 ) = 0,
m m D2 (s1 ,s2 ) = 0,
m m D3 (s1 ,s2 ) = 0.
(3.64)
These constraints (3.64), as well as the mass-shell condition (3.62), follow from the more general
ones
2 2 (s ,s )
3 2 (s ,s )
(D 1 )2 (s1 ,s2 ) = 0,
D 1 2 = 0,
D D 1 2 = 0,
3 2 (s ,s )
D 1 2 = 0.
(3.65)
In spite of constraints (3.65) being stronger than (3.62) and (3.64), they should be also imposed
on the state by the same reasons as constraint (3.54), obtained for the N = 3 gravitino multiplet.
We summarize all the constraints for the superfield (s1 ,s2 ) in a single list,
S1 (s1 ,s2 ) = s1 (s1 ,s2 ) ,
S2 (s1 ,s2 ) = s2 (s1 ,s2 ) ,
D21 (s1 ,s2 ) = D32 (s1 ,s2 ) = D31 (s1 ,s2 ) = 0,
195
D1 (s1 ,s2 ) = D 2 (s1 ,s2 ) = D 3 (s1 ,s2 ) = 0,

2

2
(D 1 )2 (s1 ,s2 ) = D 2 (s1 ,s2 ) = D 3 D 2 (s1 ,s2 ) = D 3 (s1 ,s2 ) = 0.
(3.66)
Further we consider some examples of solutions of these constraints for the lowest values of
U (1) charges.
Let s1 = s2 = 0. The corresponding state is described by the chargeless superfield . It is easy
to show that under constraints (3.66), this superfield is just a constant, = const. Therefore this
case is trivial.
The next case with s1 = 1, s2 = 0 was considered in [12]. We denote the corresponding superfield by (1,0) = W 1 . As a consequence of (3.66), it satisfies the following equations of motion
and constraints
D21 W 1 = D32 W 1 = D31 W 1 = 0,
D1 W 1 = D 2 W 1 = D 3 W 1 = 0,

2
2

(D 1 )2 W 1 = D 2 W 1 = D 3 W 1 = D 2 D 3 W 1 = 0.
(3.67)
Eqs. (3.67) are known to describe the N = 3 superfield strength of the gauge multiplet [12,20].
The component structure of W 1 can be most easily found in the analytic coordinates,

yAm = x m i 1 m 1 3 m 3 2 m 2 ,
(3.68)
in which the spinor derivatives D1 , D 2 , D 3 take the most simple form,
D1 =
,
1
D 2 = 2 ,
D 3 = 3 .
(3.69)
Then we have,
W 1 = 1 + 1 + 2 3 3 2 i2 1 m m 2
m
+ 2 3 F() i 1 2 3 (
m 1) ,
(3.70)
where I = uIi i is a triplet of complex scalars, I = u iI i is a triplet of Weyl spinors,

is also a Weyl spinor, and F() is a Maxwell field strength. All these components satisfy the
corresponding free equations of motion.
Let us consider briefly the separation of constraints (3.43c) leading to other superfield realizations. Consideration for the case when the superfield (s1 ,s2 ) satisfies the following Grassmann
shortness conditions
D1 (s1 ,s2 ) = 0,
D2 (s1 ,s2 ) = 0,
D 3 (s1 ,s2 ) = 0
(3.71)
can be done similarly as in the previous case. The physically interesting representation appears
if the U (1) charges take the values s1 = 0, s2 = 1. We denote such a superfield by (0,1) = W 3 .
It has the following equations of motion and constraints
D21 W 3 = D32 W 3 = D31 W 3 = 0,
D1 W 3 = D2 W 3 = D 3 W 3 = 0,
2
(D 1 )2 W 3 = (D 2 )2 W 3 = (D 1 D 2 )W 3 = D 3 W 3 = 0,
(3.72)
196
which describe the N = 3 Maxwell multiplet as well [12,20]. In particular, in the coordinates
(y m , 1 , 2 , 3 ),
A

yA m = x m i 1 m 1 3 m 3 + 2 m 2 ,
(3.73)
W 3 has the following component field decomposition

W 3 = 3 3 + 1 2 2 1 + i3 1 m m 1 i3 2 m m 2
1 2 m (
3) .
1 2 F( )
+ i 3 m
(3.74)
Here, 3 = u i3 i is a triplet of complex scalars, is a Weyl spinor, I is a triplet of Weil spinors,

and F( )
is a Maxwell field strength. It is easy to see that superfields (3.70) and (3.74) are related
to each other by the conjugation (2.18).
In concluding this subsection, let us briefly comment on the last case of constraints (3.43d) on
the example of a superfield (1,1) W 2 ,
D2 W 2 = 0,
D 1 W 2 = 0,
D 3 W 2 = 0.
It is easy to see that the harmonic derivatives

therefore should also annihilate this superfield,
D31 W 2 = 0,
D12 W 2 = 0,
D31 ,
D12 ,
(3.75)
D32
commute with
D32 W 2 = 0.
The -symmetry leads to the following linearity constraints

1 2 2 1 3 2 3 2 2
D W = D D W = D W = (D 2 )2 W 2 = 0.
D2 ,
D 1 , D 3 and
(3.76)
(3.77)
Such a superfield under constraints (3.75)(3.77) also describes the N = 3 Maxwell multiplet
which is equivalent to (3.70). The component structure of W 2 is similar to (3.70) with the change
of index from 1 to 2.
4. Quantization of the N = 3 superparticle with a central charge term
4.1. Hamiltonian formulation and constraints
Let us consider the N = 3 harmonic superparticle with central charges described by the action
(2.35). The Lagrangian L2 reads

1 m
1
m em2 z ij i j z ij i j
2e
2

i +i
+i
+i
+ 2R 2 e1 ++ u
(4.1)
.
u
i u 1 n ui u
i u
2
Recall that the underlined indices i, j , . . . denote SU(2) ones with the values 1, 2. The consequent quantization of this model is similar to the one for the N = 2 superparticle in harmonic
superspace [12,13]. Therefore we follow the same steps keeping, however, basic details of calculations.
The Lagrangian (4.1) defines the following canonical momenta for superspace coordinates
L2 =
pm =
L2
= e1 m ,
x m
(4.2)

i

L2
= ipm m i + z ij j ,
(4.3)

L2
j
= ipm i m + z ij ,

L
2
3 = = ipm m 3 ,
3

L2
= ipm 3 m ,
3 =
3
i =
v +i =
(4.4)
(4.5)
(4.6)
L2
1
2 1 i ++
inu+
i ,
= 2R e u i
2
u i
(4.7)
L2
1
= 2R 2 e1 u+
+ inu
i i
i .
+i
2
u
Following [12,13], we introduce the covariant harmonic momenta
vi =
+i
= 2iR 2 e1 ++ ,
D ++ = u+
i v
= vi u+i
+i
u
i v
= in,
197
D = vi ui = 2iR 2 e1 ,
2 = vi u+i
+i
+ u
i v
and define the Poisson brackets,

n
n
x , pm P = m
,

i
j
j
i
i , P = i ,
, j P = j ,
3

3
3 , P = ,
, 3 P = ,
+i
+j
j
i
u , vj P = j ,
ui , v P = i .
=0
(4.8)
(4.9)
(4.10)
(4.11)
The full list of constraints is given by

p m pm + R 2 D D ++ m2 0,

i
i
i
D = + ipm m i + z 0,
m
D i = i ipm i z i 0,

D3 = 3 + ipm m 3 0,

D 3 = 3 ipm i m 0,
(4.14)
D in 0,
(4.17)
+i
1 = u
i u
2 = vi u+i
1 0,
+i
+ u
i v
(4.12)
(4.13)
(4.15)
(4.16)
(4.18)
0.
(4.19)
There is also one more extra harmonic constraint

D ++ 0,
(4.20)
which is necessary to freeze the harmonic dynamics and to keep only the physical degrees of
freedom.
198
Constraints (4.12), (4.17), (4.20) are first-class and should be imposed on the state upon quantization. Second-class harmonic constraints (4.18), (4.19) are accounted by the Dirac bracket,
1
1
[f, g}D = [f, g}P + [f, 2 ]P [1 , g]P [f, 1 ]P [2 , g]P .
(4.21)
2
2
It is easy to see that spinor constraints (4.15), (4.16) are second-class, since we consider here the
massive case. They can be taken into account by using the GuptaBleuler method. In general,
if the mass of the superparticle is arbitrary and is not related with the central charges z, z , the
spinor constraints (4.13), (4.14) belong to the second class. The quantization of such a particle
model does not lead to physical supermultiplets. Therefore, we consider further only a special
case, when the central charges are correlated with the mass by BPS condition,
zz = m2 .
(4.22)
In this case, the superparticle model possesses the -symmetry which is realized on the superspace coordinates as follows,

j
i = ipm m i z ij ,

i
= ipm i m + z ij j ,

i
e = 4 i i + i ,
3 = 0.
3 = 0,
(4.23)
The generators of -symmetry (4.23) given by

i
i
= ipm m D i zD 0,
i = ipm m D i + z Di 0,
(4.24)
correspond to the additional first-class constraints.

4.2. GuptaBleuler quantization
Upon quantization, the canonical momenta are replaced by the differential operators,
pm i
,
x m
3 i
,
3
,
i
3 i
i i
,
i
v +i
,
u
i
vi
.
u+i
(4.25)
The spinor constraints (4.13)(4.16), as well as the harmonic momenta (4.9), (4.10), turn into the
covariant spinor and harmonic derivatives,

+ i m m iz ,

D = ui D = i m m + i z
,

D 3 = 3 i 3 m m ,
D3 = + i m 3 m ,
3

D = u
D 0 = u+i +i ui i
D ++ = u+
i
i
,
+,
u
u
ui
ui
D = u
i D =
i
(4.26)
(4.27)
(4.28)
(4.29)
with the following anticommutation relations

+
+
D , D = 2iz ,
D , D = 2i z ,

+ 3
m
D , D = D , D 3 = D , D +
= 2i m ,
+ + + +
= D , D
= 0.
D , D = D , D = D , D
199
(4.30)
Supercharges (2.22) in the harmonic superspace are described now by the operators,

= i m m + i z ,
+ i m m iz ,
Q

Q 3 = 3 i 3 m m ,
Q3 = + i m 3 m ,
(4.31)
3

Q
=
which form the N = 3 superalgebra with a central charge,

+
+
Q , Q = 2iz ,
Q , Q = 2i z ,

+ 3
= Q , Q 3 = 2im m .
Q , Q
(4.32)
Let us introduce the state | = (n) , which is a superfield on harmonic superspace with the
equations of motion and constraints originating from the superparticle constraints (4.12)(4.19).
The first-class constraint (4.17) leads to the following equation
D 0 (n) = n (n) ,
(4.33)
which shows that this superfield has a definite U (1) charge. The other harmonic constraints
(4.18), (4.19) are accounted by the Dirac bracket (4.21). The mass-shell constraint (4.12) is also
first-class, therefore we have

m
m R 2 D D ++ + m2 (n) = 0.
(4.34)
Furthermore, we require that the superfield (n) obeys constraint (4.20),
D ++ (n) = 0,
(4.35)
which removes all unphysical degrees of freedom. Under this additional constraint (4.35) the
mass-shell condition (4.34) simplifies to

m
m + m2 (n) = 0.
(4.36)
Now we have to take into account the spinorial constraints (4.13)(4.16) using GuptaBleuler
method. The covariant spinor derivatives should be divided into two subsets with commuting
constraints in each subset. Clearly, there are two ways of separating these constraints into such
subsets,

+ +
3
D , D , D 3 D , D
(4.37)
, D ,

+ + 3
D , D , D D , D , D 3 .
(4.38)
Both these cases lead to equivalent results. Therefore, we consider only (4.37) in detail. The
corresponding constraints
D+ (n) = 0,
(n)
D +
= 0,

D 3 (n) = 0,
3
show that (n) is analytic with respect to + , +
and is chiral in 3 , variables.
(4.39)
200
Finally, we have to take into account constraints (4.24) originating from the -symmetry
(4.23). Upon quantization, the generators of -transformations (4.24) turn into the differential
operators
m

zD
= m D
.
= m m D + z D ,
(4.40)
(n) = 0 are satisfied

Note that owing to the analyticity (4.39), constraints + (n) = 0, +

automatically, while the projections in (4.40) lead to the equations

(n)
m
m
m D zD
= 0.
m D + z D (n) = 0,
(4.41)

Let us introduce the operator

i
+ z D D ,
zD
D
4
which commutes with covariant spinor derivatives as
Y =
1 +
= ,
D ,Y
z
1 +
=
D ,Y
.
z
As the superfield (n) is analytic, both constraints (4.41) follow from

zD D z D D (n) = 0.
(4.42)
(4.43)
(4.44)
Despite this equation being stronger than the pair (4.41), it should be imposed on the physical
states as well, since the first-class constraint (4.42) is a function of spinorial constraints (4.13),
(4.14) and forms the algebra (4.43) with the generators of -symmetry. As a result, all superparticle constraints are accounted by the corresponding equations for the field (n) .
4.2.1. N = 3 massive vector multiplet
Let us consider a solution of constraints for the superfield (n) on the physically interesting
example of a superfield (1) q + with U (1) charge +1. We will show that such a superfield
describes the N = 3 massive vector multiplet.
To begin with, we list once again all the constraints for the q + superfield,
D0q + = q +,
D+ q + = D + q + = 0,
++ +
(4.45)
(4.46)
D q = 0,
D 3 q + = 0,
2
2
z D q + = 0,
z D

m
m + m2 q + = 0.
(4.47)
(4.48)
(4.49)
(4.50)
Eq. (4.45) is satisfied automatically, while the pair (4.49), (4.50) follows from (4.46), (4.47). To
solve (4.46), we pass from the central coordinates to the analytic ones,

xAm = x m i + m + m + ,
(4.51)
and transform Grassmann and harmonic derivatives as well as the superfield q + such that D+ ,
D +
become short,
D+ D+ = e D+ e =
(4.52)

D D = e D e = + + 2i m m 2iz ,
xA

= + 2i m m + 2i z
D
D = e D e
,
xA

2
2
D ++ D++ = e D ++ e = D ++ + iz + + i z + ,
+
+
D +
=
D = e D e
q + q+ = e q + ,
201
(4.53)
(4.54)
(4.55)
(4.56)
(4.57)
where
= iz + i z + .
(4.58)
In this representation, constraints (4.46) are solved automatically if q+ does not depend on ,
,

3
q+ = q+ xAm , + , +
(4.59)
, , 3 , ui .
++ q+ = 0 which follows from
Next, we expand (4.59) over + , +
and solve the equation D
(4.47). As a result we have
+ 2 i
2
i
+

+ +
F ui i z + F i u
q+ = u+
iz
i F +
i
+ 2i + m + m F i u
i ,
(4.60)
where all the components depend on (xAm , 3 , 3 ). Here F i is a doublet of complex scalars
satisfying KleinGordon equation,
( + zz)F i = 0,

and is a massive Dirac spinor,
(4.61)
i m m iz = 0,
im m + i z = 0.
(4.62)
In what follows, the mass is correlated with the central charges as

m = iz = i z .
Now we recall that
(4.63)
q+
depends also on
3 , 3
variables in a chiral way,
D 3 q+ = 0.
(4.64)
In the chiral coordinates y m = xAm + i3 m 3 , the derivative D 3 is short, D 3 = 3 , and all

components in (4.60) depend on 3 , but not on 3 ,

= y m , 3 ,
= y m , 3 .
F i = F i y m , 3 ,
(4.65)
Let us consider the decomposition of these components in the series over 3 ,
i
F i (y, 3 ) = f i (y) + 3 (y) + (3 )2 g i (y),

(y, 3 ) = (y) + 3 F () (y) + 3 C + (3 )3 (y),
(y, 3 ) = (y) + 3 A (y) + (3 ) (y).
(4.66)
202

i
Owing to (4.61), the fields f i , g i , obey the KleinGordon equation. Unfortunately, there is
i
i
no Dirac equation for the spinors . However, the two Weyl spinors with KleinGordon
equation are equivalent to a pair of Dirac spinors satisfying usual Dirac equation.8

Let us study the consequences of the Dirac equations (4.62). They imply that and
are usual Dirac spinors while the components A , C, F obey
m
m m C + m F zA = 0,
m m A + z C + z F = 0.
(4.67)
(4.68)
Eqs. (4.67), (4.68) have the following solutions

1
C = m A m ,
z
( + zz)A m = 0,
1 mn
Fmn ,
F =
z
(4.69)
(4.70)
mn = 1 ( m n n m ). As a result,
where Fmn = m A n n A m , A m = 12 m A and

4
vector A n corresponds to the complex massive vector field (with Proca equation) plus a complex
scalar C with KleinGordon equation.
Summarizing these results we have the following field content in q+ subject to (4.45)(4.50):
Complex massive vector A m describes 6 bosonic degrees of freedom;

f i , g i , C are complex scalars with 10 bosonic degrees of freedom;

, are massive Dirac spinors with 8 fermionic degrees of freedom;
i
The doublet of spinors does not satisfy Dirac equation since it originates from the scalars
i
F i obeying only the KleinGordon equation (4.62). However, as noted above, correspond
to two Dirac spinors with 8 fermionic degrees of freedom.
This is nothing but the field content of N = 3 massive vector multiplet with BPS mass [21]. Note
that it has double the number of components in comparison with the massive (non-BPS) N = 2
vector multiplet.
4.2.2. Superfield reduction of components in massive vector multiplet
In the previous subsection, we have shown that the quantization of the N = 3 harmonic superparticle with central charges correlated with the mass as in (4.63) leads to the N = 3 massive
vector BPS multiplet. As this multiplet has double the number of states in comparison with the
massive (non-BPS) N = 2 vector multiplet, it is natural to ask whether it is possible to impose
additional constraints on the q + superfield which reduce its component content to the massive
N = 2 supergauge multiplet. As we will show, this is possible if one relaxes the condition of the
CPT invariance of the multiplet.
To begin with, we introduce a conjugation which acts as a standard conjugation (2.34)
in harmonic superspace and changes the signs of the central charges (and mass), z = z, z = z.
8 Let be a function with the spinor index satisfying KleinGordon equation, m + m2 = 0. Factorizing
the box operator we rewrite this equation as im m mi n n m = 0. By denoting = mi m m the

KleinGordon equation can be rewritten as a pair im m m = 0, im + m = 0 that is nothing but the

Dirac equation for the spinor .
For instance, the superfield conjugated to (4.60) is

+ 2 i
+
i

+ +
q + v = u+
F ui
iz
i F
2
+ m +
i
i z + F i u
i + 2i m F ui .
203
(4.71)
Note that q+ depends on 3 , 3 in a chiral way while q + is antichiral with respect to these
variables. It is natural to restrict the dependence on 3 , 3 by imposing equations which are
similar to the ones in the massive WessZumino model,
1
1 3 2 +
(4.72)
(D3 )2 q+ + izq + = 0,
D q i z q+ = 0.
4
4
Eqs. (4.72) are conjugate to each other with respect to conjugation and have the following
consequences:
1 3 2 i
D F + izF i = 0,
4
1 3 2
D iz = 0,
4
1 3 2
D + iz = 0,
4
1 2 i
(D3 ) F i z F i = 0,
4
1 2
(D3 ) + i z = 0,
4
1 2
(D3 ) i z = 0.
4
(4.73)
(4.74)
(4.75)
Clearly, (4.73) are nothing but the usual WessZumino equations for the N = 1 superfields F i ,
F i . Therefore for the components of these superfields we have
( + zz)f i = 0,
im m i
i
i
g = izf ,
i
+ i z
i
( + zz)fi = 0,
= 0,
im m i
g = i z f .
(4.76)
i
+ iz
= 0,
(4.77)
(4.78)

2
2

One can easily construct a Dirac spinor

from the spinors satisfying (4.77). Note
1
satisfies the Dirac equation with opposite sign of the

that the conjugated spinor =
1
mass.
Let us consider the pair of Eqs. (4.74). They lead to the KleinGordon equations for spinors
, , while , are expressed from them,
= iz ,
= iz .
(4.79)
The other components obey the following equations

m m A + z C + z F = 0,
m
m F + m C
+ zA = 0,
(4.80)
(4.81)
which are solved by

1
1 mn
C = m Am ,
F =
Fmn ,
z
z
( + zz)Am = 0.
(4.82)
(4.83)
Considering (4.82) together with (4.69) we conclude that field strength Fmn = m An n Am
is real, Fmn = Fmn , while m Am is imaginary and corresponds to a real scalar B = im Am . As
204
a result, the complex vector An splits into a real vector obeying Proca equations, and a real scalar
with KleinGordon equation. The resulting multiplet exactly corresponds to a massive N = 2
vector multiplet:
Two complex scalars f i and the real one B give 5 real bosonic degrees of freedom;

Dirac spinors , describe 8 fermionic degrees of freedom;
Real massive vector field Am has 3 bosonic components on-shell.
To show that Eqs. (4.72) preserve the N = 3 supersymmetry with a central charge, we note
that the supercharges (4.31) are conjugate to each other with respect to conjugation rather than
to ,

+
Q+
= Q ,
Q
= Q ,
+ = Q+
Q
,
= Q
Q
.
(4.84)
Hence, the supersymmetry variation is real under such a conjugation,

+
+
= .
= + Q
+ Q+
+ 3 Q3 + 3 Q
+ Q
Q
3
q+
(4.85)
q +
and
transform in the same way under supersymmetry, and the
Therefore, both superfields
conjugation in Eqs. (4.72) does not break the N = 3 supersymmetry. However, the resulting
multiplet is not CPT selfconjugated and the CPT symmetry is lost. In terms of superfields it is
obvious since the conjugation involves the change of the sign of the mass and the central

charge. In components it leads to the fact that the spinors , are standard Dirac ones

while their conjugates , satisfy the Dirac equation with opposite sign of the mass.
As a result, Eqs. (4.72) reduce the number of components in the N = 3 vector multiplet
by half resulting in the N = 2 massive (non-BPS) vector multiplet. In other words, the N = 2
massive vector multiplet is equivalent to a half of the N = 3 vector multiplet which respects the
N = 3 supersymmetry with central charge, but is not CPT selfconjugated. Although this fact is
well known [21], we establish this correspondence by superfield considerations. In particular,
this multiplet is realized as a single constrained N = 3 superfield. It would also be very tempting
to find a supersymmetric action for superfields q+ , q + reproducing the corresponding equations
of motion.
5. Conclusion
Let us summarize the results obtained in the quantization of N = 3 superparticle.
(1) The models of the N = 3 superparticles both with and without central charge term, are
considered in the harmonic superspace. Since the N = 3 superalgebra with central charge
possesses SU(2) U (1) R-symmetry rather than U (3), the description of the N = 3 superparticle with central charge is achieved in the N = 2 harmonic superspace (with SU(2) har3
monic variables) extended by a pair of extra Grassmann variables: {x m , ,
, 3 , , ui }.
(2) By quantizing the N = 3 superparticle without central charge, we obtain N = 3 superfield
realizations of the N = 3 supergauge multiplet and the gravitino multiplet (with highest
helicity 3/2). The latter is described by a chiral N = 3 superfield satisfying linearity constraints, while the former is given by the superfield strengths, which are short superfields in
the N = 3 harmonic superspace subject to Grassmann and harmonic shortness conditions.
These superfields were originally introduced in [12] and studied in [20].
205
(3) The N = 3 superparticle with central charge is quantized similarly to the N = 2 superparticle in harmonic superspace [12,13]. The resulting massive N = 3 supergauge multiplet is
given by 5 complex scalars, 4 Dirac spinors and 1 complex vector on-shell. It is embedded into a superfield q + , which is analytic in ,
variables and chiral with respect to
3
3 , . The equation of motion for this superfield is similar to the one for q-hypermultiplet,
D ++ q + = 0.
(4) We notice that the number of states of the massive N = 3 supergauge multiplet is doubled
compared to the massive (non-BPS) N = 2 vector multiplet with 5 real scalars 4 spinors
and 1 real vector. The doubling of states in the representations of the N = 3 superalgebra
with central charge is required for CPT invariance [21]. However, if we abandon the CPT
invariance, the numbers of states in these two multiplets coincide. We have shown how this
can be achieved at a superfield level: by imposing the extra superfield constraints on the N =
3 superfield q + we reduce the number of states by one half, arriving at the massive N = 2
supergauge multiplet realized as a constrained N = 3 superfield. Of course, these constraints
break the CPT invariance manifestly. In components, the loss of CPT invariance means that

the Dirac spinor and its conjugate satisfy Dirac equations with different signs of
the mass.
The results of the quantization of the massless N = 3 superparticle without a central charge
term are rather expected: the superfield realizations of the N = 3 supergauge and gravitino multiplets are achieved. However, the quantization of the N = 3 superparticle with a central charge
leads to a superfield description for the massive N = 3 supergauge multiplet which was previously unknown. It would be interesting to develop the classical and quantum field theory of
this multiplet in a superfield realization. If a Lagrangian superfield formulation of this model
is achievable, the classical and quantum properties would be as interesting as for the massive
N = 2 vector superfield [2628].
We have shown the relations between the N = 3 and the N = 2 massive vector multiplets to
be even deeper. Indeed, the N = 2 massive vector multiplet is described by an N = 3 superfield
under specific superfield constraints which manifestly break CPT symmetry of N = 3 superalgebra with central charge. In other words, the N = 2 massive vector multiplet can be viewed as a
half of the N = 3 massive vector multiplet, which does not have its CPT conjugate. This means
that the free N = 2 massive vector multiplet possesses N = 3 supersymmetry with a central
charge, if we neglect CPT invariance. Finally, the N = 3 superfield under the additional constraints can be considered as an alternative formulation for the massive N = 2 vector multiplet.
It would be interesting to study whether it is possible to include the non-Abelian selfinteraction
of this multiplet in terms of N = 3 superfields, and to build some action directly in the N = 3
superspace.
Another obvious continuation of this research would be the study of the N = 4 superparticle
in a similar way. The case of the massive superparticle with a central charge would be the most
tempting and should lead to a massive N = 4 vector multiplet realized as an N = 4 superfield.
This model would be of high interest both at the classical and at the quantum level.
Acknowledgements
The authors are grateful to O. Lechtenfeld, A.V. Galajinsky, N.G. Pletnev, D. Sorokin,
B.M. Zupnik for stimulating discussions. We would like to thank also J. Brdel for careful reading and improving the text. I.B.S. is very grateful to D. Sorokin for intensive consultations on the
206
aspects of superparticles quantization and for critical reading of the draft and to INFN, Sezione di
Padova & Dipartimento di Fisica Galileo Galilei, Universit degli Studi di Padova for kind hospitality and support during the period when the essential part of this work was done. The present
work is supported particularly by INTAS grant, project No. 05-1000008-7928, by RFBR grants,
projects No. 06-02-16346 and No. 08-02-90490, by DFG grant, project No. 436 RUS/113/669/03 and grant for LRSS, project No. 2553.2008.2. I.B.S. acknowledges the support from INTAS
grant, project No. 06-1000016-6108.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
R. Casalbuoni, The classical mechanics for BoseFermi systems, Nuovo Cimento A 33 (1976) 389.
A.I. Pashnev, D.V. Volkov, Supersymmetric Lagrangian for particles in proper time, Teor. Mat. Fiz. 44 (1980) 321.
L. Brink, J.H. Schwarz, Quantum superspace, Phys. Lett. B 100 (1981) 310.
J.A. de Azcrraga, J. Lukierski, Supersymmetric particles with internal symmetries and central charges, Phys. Lett.
B 113 (1982) 170;
A. Frydryszak, J. Lukierski, N = 2 massive matter multiplet from quantization of extended classical mechanics,
Phys. Lett. B 117 (1982) 51;
J.A. de Azcrraga, J. Lukierski, Supersymmetric particles in N = 2 superspace: Phase-space variables and Hamiltonian dynamics, Phys. Rev. D 28 (1983) 1337.
L. Lusanna, B. Milewski, N = 2 super YangMills and supergravity constraints from coupling to a supersymmetric
particle, Nucl. Phys. B 247 (1984) 396;
J.A. Shapiro, C.C. Taylor, Superspace supergravity from the superstring, Phys. Lett. B 186 (1987) 69.
D.V. Uvarov, A.A. Zheltukhin, N = 2 massive superparticle: The minimality principle and the k symmetry, Phys.
Rev. D 61 (2000) 015004;
D.V. Uvarov, N = 2 supersymmetric YangMills theory and the superparticle: Twistor transform and kappa symmetry, Mod. Phys. Lett. A 18 (2003) 1611.
R. Grimm, M. Sohnius, J. Wess, Extended supersymmetry and gauge theories, Nucl. Phys. B 133 (1978) 275;
M.F. Sohnius, Bianchi identities for supersymmetric gauge theories, Nucl. Phys. B 136 (1978) 461;
M.F. Sohnius, Supersymmetry and central charges, Nucl. Phys. B 138 (1978) 109;
P.S. Howe, K.S. Stelle, P.K. Townsend, The relaxed hypermultiplet: An unconstrained N = 2 superfield theory,
Nucl. Phys. B 214 (1983) 519.
A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic superspace: Key to N = 2 supersymmetric theories,
JETP Lett. 40 (1984) 912;
A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, Unconstrained N = 2 matter, YangMills and
supergravity theories in harmonic superspace, Class. Quantum Grav. 1 (1984) 469.
A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic Superspace, Cambridge Univ. Press, UK, 2001,
306 p.
A. Galperin, E.A. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic supergraphs. Green functions, Class. Quantum
Grav. 2 (1985) 601;
A. Galperin, E.A. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic supergraphs. Feynman rules and examples, Class.
Quantum Grav. 2 (1985) 617.
I.L. Buchbinder, E.I. Buchbinder, S.M. Kuzenko, Non-holomorphic effective potential in N = 4 SU(n) SYM, Phys.
Lett. B 446 (1999) 216, hep-th/9810239;
E.I. Buchbinder, I.L. Buchbinder, E.A. Ivanov, S.M. Kuzenko, B.A. Ovrut, Low-energy effective action in N = 2
supersymmetric field theories, Phys. Part. Nucl. 32 (2001) 641;
I.L. Buchbinder, E.A. Ivanov, Complete N = 4 structure of low-energy effective action in super YangMills theories, Phys. Lett. B 524 (2002) 208, hep-th/0111062;
I.L. Buchbinder, E.A. Ivanov, A.Yu. Petrov, Complete low-energy effective action in N = 4 SYM: A direct N = 2
supergraph calculation, Nucl. Phys. B 653 (2003) 64, hep-th/0210241;
I.L. Buchbinder, N.G. Pletnev, Hypermultiplet dependence of one-loop low-energy effective action in the N = 2
superconformal theories, JHEP 0704 (2007) 096, hep-th/0611145.
V.P. Akulov, D.P. Sorokin, I.A. Bandos, Particle mechanics in harmonic superspace, Mod. Phys. Lett. A 3 (1988)
1633.
V.P. Akulov, I.A. Bandos, D.P. Sorokin, Particle in harmonic N = 2 superspace, Sov. J. Nucl. Phys. 47 (1988) 724,
Yad. Fiz. 47 (1988) 11361146.
207
[14] P.A. Grassi, P. van Nieuwenhuizen, Harmonic superspaces from superstrings, Phys. Lett. B 593 (2004) 271, hep-th/
0402189.
[15] E. Ivanov, O. Lechtenfeld, N = 4 supersymmetric mechanics in harmonic superspace, JHEP 0309 (2003) 073,
hep-th/0307111;
S. Bellucci, E. Ivanov, A. Sutulin, N = 8 mechanics in SU(2) SU(2) harmonic superspace, Nucl. Phys. B 722
(2005) 297, hep-th/0504185;
E. Ivanov, Nonlinear (4, 8, 4) multiplet of N = 8, d = 1 supersymmetry, Phys. Lett. B 639 (2006) 579, hep-th/
0605194;
F. Delduc, E. Ivanov, New model of N = 8 superconformal mechanics, Phys. Lett. B 654 (2007) 200, arXiv:
0706.2472 [hep-th].
[16] S. Bellucci, S.J. Gates Jr., E. Orazi, A journey through garden algebras, in: Lectures Given at Winter School on
Modern Trends in Supersymmetric Mechanics (SSM05), Frascati, Italy, 712 March 2005, in: Lecture Notes in
Physics, vol. 698, 2006, pp. 147, also in: Frascati 2005, Modern Trends in Supersymmetric Mechanics, pp. 147,
hep-th/0602259;
S. Bellucci, S. Krivonos, Supersymmetric mechanics in superspace, in: Lectures Given at Winter School on Modern
Trends in Supersymmetric Mechanics (SSM05), Frascati, Italy, 712 March 2005, in: Lecture Notes in Physics,
vol. 698, 2006, pp. 4996, also in: Frascati 2005, Modern Trends in Supersymmetric Mechanics, pp. 4996, hep-th/
0602199.
[17] A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, N = 3 supersymmetric gauge theory, Phys. Lett.
B 151 (1985) 215;
A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, Unconstrained off-shell N = 3 supersymmetric
YangMills theory, Class. Quantum Grav. 2 (1985) 155.
[18] F. Delduc, J. McCabe, The quantization of N = 3 super-YangMills off-shell in harmonic superspace, Class. Quantum Grav. 6 (1989) 233.
[19] I.L. Buchbinder, E.A. Ivanov, I.B. Samsonov, B.M. Zupnik, Scale invariant low-energy effective action in N = 3
SYM theory, Nucl. Phys. B 689 (2004) 91, hep-th/0403053.
[20] L. Andrianopoli, S. Ferrara, E. Sokatchev, B. Zupnik, Shortening of primary operators in N extended SCFT(4) and
harmonic superspace analyticity, Adv. Theor. Math. Phys. 3 (1999) 1149, hep-th/9912007.
[21] S. Ferrara, C.A. Savoy, B. Zumino, General massive multiplets in extended supersymmetry, Phys. Lett. B 100
(1981) 393.
[22] B. Zumino, Normal forms of complex matrices, J. Math. Phys. 3 (1962) 1055.
[23] W. Siegel, Hidden local supersymmetry in the supersymmetric particle action, Phys. Lett. B 128 (1983) 397.
[24] D.P. Sorokin, V.I. Tkach, D.V. Volkov, Superparticles, twistors and Siegel symmetry, Mod. Phys. Lett. A 4 (1989)
901.
[25] D.P. Sorokin, Superbranes and superembeddings, Phys. Rep. 329 (2000) 1, hep-th/9906142.
[26] G.A. Khelashvili, V.I. Ogievetsky, Nonrenormalizability of the massive N = 2 super YangMills theory, Mod. Phys.
Lett. A 6 (1991) 2143.
[27] S.M. Kuzenko, On massive tensor multiplets, JHEP 0501 (2005) 041, hep-th/0412190.
[28] I.L. Buchbinder, N.G. Pletnev, One-loop effective action in N = 2 supersymmetric massive YangMills model, in
preparation.

N = 4 superparticle and super-YangMills theory

in USp(4) harmonic superspace
I.L. Buchbinder a , O. Lechtenfeld b , I.B. Samsonov b,c,,1
a Department of Theoretical Physics, Tomsk State Pedagogical University, 634041 Tomsk, Russia
b Institut fr Theoretische Physik, Leibniz Universitt Hannover, 30167 Hannover, Germany
c Laboratory of Mathematical Physics, Tomsk Polytechnic University, 634050 Tomsk, Russia
Received 1 May 2008; accepted 20 May 2008

Abstract
We study the N = 4 harmonic superparticle model, both with and without central charge and quantize it.
Since the central charge breaks the U (4) R-symmetry group of the N = 4 superalgebra down to USp(4), we
consider the superparticle dynamics in N = 4 harmonic superspace with USp(4)/(U (1) U (1)) harmonic
variables. We show that the quantization of a massive superparticle with central charge leads to a superfield
realization of the N = 4 massive vector multiplet in N = 4 harmonic superspace. In the massless case
without central charge the superparticle quantization reproduces three different multiplets: the N = 4 SYM
multiplet, the N = 4 gravitino multiplet and N = 4 supergravity multiplet. The SYM multiplet is described
by six analytic superfield strengths with different types of analyticity. We show that these strengths solve
the N = 4 SYM constraints and can be used for the construction of actions in N = 4 harmonic superspace.
1. Introduction
The N = 4 super-YangMills (SYM) field theory, being the maximally extended rigid supersymmetric model, possesses many remarkable properties. The symmetry of this model is so large
that the only freedom in the classical action is the choice of the gauge group, and the quantum
E-mail addresses: joseph@tspu.edu.ru (I.L. Buchbinder), lechtenf@itp.uni-hannover.de (O. Lechtenfeld),

samsonov@mph.phtd.tpu.edu.ru (I.B. Samsonov).
1 Alexander von Humboldt fellow at Leibniz Universitt Hannover.
doi:10.1016/j.nuclphysb.2008.05.015
I.L. Buchbinder et al. / Nuclear Physics B 802 (2008) 208246
209
dynamics is free of divergences. It worth pointing out that this theory has profound relations with
superstring theory, particularly due to the AdS/CFT correspondence (see, e.g., [1]).
The problems of N = 4 SYM theory in the quantum domain are mainly related to the effective action and correlation functions of composite operators. The superfield approaches seem to
be more efficient for these purposes, since they allow one to use the supersymmetries in explicit
form. However, a description of N = 4 SYM theory in terms of unconstrained N = 4 superfields is still missing. For various applications, formulations in terms of N = 1 superfields (see,
e.g., [2]), in terms of N = 2 superfields [3,4], or in terms of N = 3 superfields [5] are used. All
attempts to find an unconstrained N = 4 harmonic superfield formulation for the N = 4 SYM
theory have been futile so far [712], for a number of different types of harmonic variables originating from various cosets of the SU(4) group. However, one may still hope that there exists
some other harmonic superspace, not based on some SU(4) coset, which is better suited for
a superfield realization of N = 4 supergauge theory.2 In other words, we need new superfield
representations of the known irreducible multiplets of the N = 4 superalgebra realized in an
appropriate harmonic superspace.
The main purpose of this paper is to construct N = 4 superparticle models in harmonic superspace, to quantize them and to derive superfield representations of the N = 4 superalgebra as
a result of their quantization. It is well known that the quantization of superparticles is closely
related to superfield formulations of the corresponding field theories. Indeed, the superparticle
models are rich of symmetries such as reparameterization invariance, supersymmetry and, in
particular cases, the kappa-symmetry (see, e.g., [13] for a review). All these symmetries are accompanied by constraints in the Hamiltonian formulation. Upon quantization, these constraints
turn into equations of motion as well as superfield differential constraints, which together define
superfield representations of irreducible multiplets of supersymmetry. For instance, the quantization of the N = 1 superparticle was achieved in [1417], the N = 2 gauge multiplet and
hypermultiplets were obtained in [1721] by quantizing the N = 2 superparticle and, finally,
the N = 3 superparticle was recently studied by two of us [22], where massive and massless
N = 3 vector multiplets as well as the N = 3 gravitino multiplet were derived. A particular case
of massless superparticles with arbitrary N > 2 extended supersymmetry in SU(N ) harmonic
superspace was analyzed in [18], where the corresponding superfield strengths were derived. We
point out the significance of harmonic superparticles with N = 2 and N = 3 extended supersymmetries [18,19,22], since they yield equations of motion for the corresponding field theories in
harmonic superspaces which possess unconstrained superfield descriptions.
In the present paper we study models of the N = 4 harmonic superparticle both in the massive case with central charge and in the massless case without central charge. It is well known
that a central charge breaks the U (4) R-symmetry group of the N = 4 superalgebra down to
USp(4) [23,24]. Therefore, we consider it as crucial to introduce USp(4) harmonic variables
which are employed for the corresponding N = 4 harmonic superspace. The various cosets of
the USp(4) group were introduced and studied in [25], and the corresponding harmonic variables
were further applied in [26,27] to d = 5 and d = 6 N = 4 SYM models. In our work we find
them useful also for d = 4 N = 4 SYM and superparticle models and in [28] to d = 3 N = 6
ChernSimons theory. We start with the formulation and quantization of the N = 4 superparticle
in such a harmonic superspace and find superfield representations of various multiplets of the
2 As is well known, an unconstrained superfield formulation of N = 4 SYM theory is impossible in standard N = 4
superspace.
210
N = 4 superalgebra. In the massive case with nonzero central charge, the quantization leads to
the massive N = 4 vector multiplet, represented by analytic superfields subject to several Grassmann and harmonic shortness conditions. In the massless case with vanishing central charge, this
multiplet reduces to the usual N = 4 SYM multiplet if one also imposes reality conditions. As
a result, the N = 4 SYM multiplet is described by six analytic superfields with different types
of analyticity and harmonic shortness. Apart form the SYM multiplet, we also derive the N = 4
gravitino multiplet (with highest helicity 3/2) and the multiplet of N = 4 supergravity. These
multiplets are represented by N = 4 chiral-analytic and chiral superfields, respectively.
The quantization of the N = 4 superparticle appears to be very fruitful since it not only
hints at a suitable N = 4 harmonic superspace based on USp(4), but it also gives an appropriate formulation of N = 4 superfield strengths in such a harmonic superspace. We show that
exactly these superfields appear in the solution of the N = 4 SYM constraints with the help of
USp(4) harmonic variables. Finally, we employ these superfields for constructing some integral
invariants and sketch certain superfield actions which describe an F 4 term in such a harmonic
superspace. Note that similar superfields were introduced in [811,18] by contracting the N = 4
superfield strengths with harmonics on some coset of SU(4) and have been exploited for constructing N = 4 invariant actions and correlation functions of composite operators in various
works (see, e.g., [12,29]).
The paper is organized as follows. In Section 2 we introduce the N = 4 harmonic superspace
with USp(4) harmonic variables and review the basic constructions in it. In Section 3 we consider
the N = 4 harmonic superparticle model and develop the Lagrangian and Hamiltonian formulations for it. The quantization of the superparticle is given in Section 4, where the superfield
representations for the massive and massless vector multiplets as well as for the gravitino and
supergravity multiplets are found. In Section 5 we show how USp(4) harmonic variables help
solving the SYM constraints and construct various actions in such an N = 4 harmonic superspace. In a summary we discuss the results obtained and some ideas of their further application
to N = 4 SYM theory. Technical details are collected in Appendices AC, where we address
the problem of an unconstrained N = 4 superfield description of the F 2 term in harmonic superspace.
2. N = 4 USp(4) harmonic superspace
2.1. N = 4 superalgebra with central charges
We start with a short review of N = 4 superspace and superalgebra constructions just to fix
our notations. The generators of N = 4 superalgebra with central charges can be represented by
the following differential operators
m
m + iZ ij j ,
Qi = i i
i
x
j
i
Q
(2.1)
+ ii m m + i Z ij ,
=
i
where {x m , i , i } are the superspace coordinates and Z ij = Z j i , Z ij = (Z ij ) are constant

antisymmetric matrices of central charges. The Greek letters , denote the SL(2, C) indices
while the small Latin ones i, j, . . . = 1, 2, 3, 4 correspond to R-symmetry. The operators (2.1)
satisfy the standard anticommutation relations of N = 4 superalgebra with central charges,
i
j
i , Q
} = 2i Z ij ,
{Q
Q , Q = 2i Z ij ,
j

211
j
= 2iji m m .
Qi , Q
(2.2)
x
It is well known that without central charges the N = 4 superalgebra possesses U (4) Rsymmetry group. However, the non-zero central charges break the U (4) R-symmetry group down
to USp(4) [23,24]. Indeed, the relations (2.2) are invariant under those U (4) transformations with
the matrices ui k which leave the antisymmetric constant tensor Z ij invariant, ui k uj l Z kl = Z ij .
Hence, Z ij plays the role of invariant tensor in Sp(4) group. The resulting the R-symmetry group
is given by the intersection of U (4) and Sp(4) groups, that is nothing but USp(4).
By applying the rotations with some unitary matrices ui j U (4) to the central charge tensor
ij
Z one can bring it to the normal form [24,30],
0
0 z1 0
0
0
0
z
Z ij Z ij = ui k uj l Z kl = 1
(2.3)
,
0
0
0 z2
0
0
z2
0
where z1 , z2 are non-negative numbers. We restrict ourself in the further considerations to the
case z1 = z2 z, so we are left with the only central charge. More generally one can consider
complex central charge z by giving arbitrary phase factor to it due to the U (1) rotations in the
U (4) group. This is sufficient for obtaining short representations of N = 4 superalgebra with
central charge, when the masses of multiplets are related with the central charges by the BPS
condition [23,24],
m2 = zz.
(2.4)
The central charge matrix (2.3) can be written as

Z ij = zij ,
Z ij = z ij ,
where
(2.5)
0 1 0 0
0 1 0 0
1 0 0 0
1 0 0 0
ij = (ij )1 =
ij =
,
.
0 0 0 1
0 0 0 1
0 0 1 0
0 0 1 0
The matrix will be considered further as the invariant tensor in the USp(4) group.
The N = 4 superspace possesses the following supercovariant Cartan forms
m
m i
m
m
i
= dx i di + ii d ,
M
= i = di ,
i = d i ,
(2.6)
(2.7)
which will be further used in the construction of the superparticle Lagrangian.

2.2. USp(4) harmonic variables
The harmonic variables on different cosets of the USp(4) group were introduced in [25]. In
the present work we are interested in the harmonics on the USp(4)/(U (1) U (1)) coset which
we review in this subsection in some details.
The USp(4) harmonic variables are 4 4 unitary matrices u = (ui j ) preserving the antisymmetric tensor (2.6),
u USp(4)
uu = 1,
uuT = .
(2.8)
212
Note that it is not necessary to impose the constraint det u = 1 since it follows from (2.8).
Let us denote the elements of complex conjugate matrix as u = (u i j ).3 Then the identities (2.8) can be written for the matrix elements as
ui j u k j = ki ,
ui j j k ul k = il .
(2.9)
As follows from (2.9),

u i j = ik uk l lj ,
(2.10)
the conjugated matrix in the USp(4) group is not independent, but is expressed through the
original one with the help of invariant tensor . In other words, the fundamental and conjugated
representations are equivalent, similarly as for the SU(2) group. Hence, the invariant tensors ij
and ij are used to lower and rise the USp(4) indices, e.g.,
uij = ui k kj = ik u k j ,
)
u ij = ik uk j = u i k kj .
(2.11)
= ij .
Here we assume (ij

Now we introduce the usp(4) algebra as a space spanned on the following differential operators
S1 = D11 D22 ,
S2 = D33 D44 ,
D (++,0) = D21 ,
D (,0) = D12 ,
D (0,++) = D43 ,
D (0,) = D34 ,
D (+,+) = D41 + D23 ,
D (,) = D32 + D14 ,
D (+,) = D31 D24 ,
D (,+) = D42 D13 ,
(2.12)
where
Dji = ui k
.
uj k
(2.13)
The commutation relations of the operators (2.12), given by (A.3), show that D (++,0) , D (0,++) ,
D (+,+) , D (,+) are rising operators, D (,0) , D (0,) , D (,) , D (+,) are lowering ones and
S1 , S2 are Cartan generators in the USp(4) group. The operators S1 , S2 measure the U (1) charges
of the generators of USp(4) group,

S1 , D (s1 ,s2 ) = s1 D (s1 ,s2 ) ,
(2.14)
S2 , D (s1 ,s2 ) = s2 D (s1 ,s2 ) .
It is convenient to label the harmonic variables by their U (1) charges as well,
u1 i = ui(+,0) ,
(,0)
u2 i = ui
(0,+)
u3 i = ui
(0,)
u4 i = ui
(2.15)
The harmonic derivatives (2.12) can now be rewritten in the more useful form for practical calculations with harmonics (2.15),
S1 = ui(+,0)
(+,0)
ui
(,0)
ui
(,0)
ui
(0,+)
S2 = ui
(0,+)
ui
(0,)
ui
(0,)
ui
3 We assume that the complex conjugation flips the position of indices while the transposition changes their order, so
that u = (ui j ), u = (u i j ), uT = (uj i ), u = (u j i ).
(,0)
D (,0) = ui
(,0)
ui
D (,) = ui(,0)
(0,)
+ ui
ui(0,)
(0,)
D (0,) = ui
ui(,0)
(0,)
ui
(,0)
D (,) = ui
213
ui(0,)
,
(0,)
ui
ui(,0)
(2.16)
Using the notations (2.15), the basic relations for harmonics (2.9) can be written as orthogonality
(,0)
u(+,0)i ui
(+,0) (0,+)i
ui
(0,)
= 1,
(+,0) (0,)i
= ui
= u(0,+)i ui
= ui
(2.17)
(0,+) (,0)i
= ui
(0,)
uj
(,0) (0,)i
=0
(2.18)
(0,+) (0,)i
= ji .
(2.19)
and completeness conditions,

(,0)
u(+,0)i uj
(+,0) (,0)i
uj
+ u(0,+)i uj
Apart from the usual complex conjugation there is the following conjugation for harmonics
[25],

(,0)
ui
= u(0,)i ,
(,0)i = u(0,) ,
u
i

(0,)
ui
= u(,0)i ,
(0,)i = u(,0) .
u
i
(2.20)
It is the conjugation (2.20) which allows one to define real objects in harmonic superspace with
USp(4) harmonics.
The invariant Cartan forms on the USp(4) group are given by
(,0)
(0,)
,
(0,) = u(0,)i dui
,
(,0) = u(,0)i dui
1
(0,)
(,0)
,
+ u(0,)i dui
(,) = u(,0)i dui
2
1
(,0)
(0,)
,
+ u(,0)i dui
(,) = u(0,)i dui
2

1
(0,0)
(,0)
(,0)
ui
du(+,0)i ,
1 = u(+,0)i dui
2

1
(0,0)
(0,)
(0,)
ui
du(0,+)i .
2 = u(0,+)i dui
(2.21)
2
These differential forms will be further used in the construction of the superparticle Lagrangian
on N = 4 harmonic superspace with USp(4) harmonic variables.
3. N = 4 harmonic superparticle model
In this section we construct the N = 4 harmonic superparticle model extending the conventional N = 4 superparticle by the Lagrangian for the harmonic variables on the coset
USp(4)/(U (1) U (1)). Then we develop the Hamiltonian formulation and analyze the constraints.
We assume everywhere that any Lagrangian L defines an action S by the rule S =

d L, where is worldline parameter. The derivatives of all superspace variables over are
denoted by dots, e.g., x m = dx m /d , i = di /d , etc.
214
3.1. N = 4 superparticle Lagrangian

The Lagrangian for a massive superparticle moving in N = 4 superspace can be written in
terms of Cartan forms (2.7) in a standard way [1417],4
LN =4 = L1 + L2 ,

1
L1 = e1 m m + em2 ,
2

L2 = Z ij i j + Z ij i j .
(3.1)
(3.2)
(3.3)
Here L1 is a kinetic term with the mass m and L2 is the WessZumino term with the central
charges Z ij , Z ij . Further we will consider only the case when the central charges are given
by (2.5), so the Lagrangian L2 takes the following form

L2 = z ij i j z ij i j .
(3.4)
The superparticle action is invariant under supertranslations,

i = i ,
i = i ,
x m = ii m i + ii m i ,
(3.5)
which lead to the conserved charges (supercharges),

Qi = 2ie1 m m i + 2z ij j ,

i = 2ie1 m i m 2zij j .
Q
(3.6)
Upon quantization, the supercharges (3.6) turn into the differential operators (2.1) with the superalgebra (2.2).
If the BPS condition (2.4) is satisfied, the Lagrangian (3.1) respects also -symmetry,

i = ipm m i z ij j ,

i = ipm i m + z ij j ,

e = 4 i i + i i ,
(3.7)
where i , i are anticommuting local parameters. Despite the relation (2.4) defines the central
charge z only up to a phase, we fix this freedom without loss of generality as
m = iz = i z .
(3.8)
As it will be shown further, it is the relation (3.8) that provides us with the correct Dirac equations
for physical spinors in massive supermultiplets.
4 Here we assume that Cartan forms M are pulled back M =
M ( ) d on the superparticle worldline Z M =
M
M
Z ( ), where Z is a set of superspace coordinates. We hope that our notations for the Cartan forms M and their
values M do not lead to misunderstandings. The same concerns the Cartan forms for harmonics (2.21) and their values
appearing in the Lagrangian (3.20).
215
3.2. Momenta, constraints and Hamiltonian

We introduce the canonical momenta for the superspace variables as follows,
LN =4
= e1 m ,
x m

LN =4
= ipm m i + z ij j ,
i =

pm =
(3.9)
(3.10)
i =

LN =4
j
= ipm i m + z ij = i .
i
(3.11)
The spinorial momenta (3.10), (3.11) do not allow one to express the corresponding velocities
and therefore they are considered as the constraints,

Di = i + ipm m i + z ij j 0,

j
D i = i ipm i m z ij 0.
(3.12)
As usual, the first and second class constraints are entangled in (3.12) and their separation can be
done, e.g., with the help of spacetime harmonic variables [31,32]. However, we do not need this
technique here since the correct accounting of these constraints will be done in the next section
by applying GuptaBleuler quantization method in harmonic superspace.
To get the Hamiltonian for N = 4 superparticle we perform the Legendre transform,

e
HN =4 = x m pm + i i + i i LN =4 = p 2 m2 .
(3.13)
2
The Hamiltonian is proportional to the first-class mass-shell constraint
p 2 m2 0
(3.14)
with the Lagrange multiplier e.

The Poisson brackets are defined in a standard way,
m

j
i
j
x , pn P = nm ,
i , P = i ,
, j P = ji .
(3.15)
With the help of the Poisson brackets we write down the variation for -symmetry transformations,

= i [i , }P + i i , P ,
(3.16)
where
i = ipm m D i + z ik Dk ,
i = ipm m Di z ik D k
(3.17)
are the generators of -transformations (3.7).

3.3. Lagrangian for N = 4 harmonic superparticle
Let us consider now the N = 4 harmonic superspace with USp(4) harmonics, ZH =
{x m , i , i , u}, where the harmonics u are USp(4) matrices defined in (2.8). The superparticle Lagrangian (3.1) should be supplemented by the harmonic term,
L = LN =4 + LUSp(4) ,
(3.18)
216
where
LUSp(4) = L + LWZ + L ,
(3.19)

2R 2 (++,0) (,0)
+ (0,++) (0,) + (+,+) (,) + (+,) (,+) ,
e
is2 (0,+)i (0,)
is1 (+,0)i (,0)
(,0)
(0,)
,
u i
u (+,0)i ui
u i
u (0,+)i ui
LWZ =
u
u
2
2

(+,0) (,0)i
(0,+) (0,)i
(+,0) (0,+)i
u
1 + 2 ui
u
1 + (,) ui
u
L = 1 ui
L =
(+,0) (0,)i
+ (,+) ui
(0,+) (,0)i
+ (+,) ui
(,0) (0,)i
+ (+,+) ui
(3.20)
(3.21)
(3.22)
Here L is built out from the Cartan forms (2.21) and describes the kinetic term for harmonics,
LWZ is the WessZumino term and L takes into account the constraints (2.17), (2.18) with the
Lagrange multipliers 1 , 2 , (+,+) , (,) , (,+) , (+,) . R, s1 , s2 are some constants.
The main advantage of using the harmonic superspace ZH is the possibility of passing to
the harmonic projections for all objects with USp(4) indices. For instance, for the Grassmann
variables we have
I = uI i i ,
I = uI i i ,
(3.23)
where the index I takes the following values

I = (+, 0), (, 0), (0, +), (0, ) .
(3.24)
One can promote the conjugation (2.20) to such objects,

(,0)
(0,)
=
,

(0,)
= (,0) ,

(0,)
(,0)
=
,

(,0)
= (0,) .
(3.25)
Analogously, we project the constraints (3.12) with harmonics,

DI = uI i Di = uI i i + ipm m I zI 0,

D I = uI i D i = uI i i ipm I m + z I 0.
(3.26)
They are also related by the conjugation as

(,0)
(0,)
D
= D
,

(,0)
= D(0,) ,
D

(0,)
(,0)
D
= D
,

(0,)
= D(,0) .
D
(3.27)
Let us give here also the harmonic projections of the -symmetry constraints (3.17),
I = ipm m D I + z DI 0,
I = ipm m D I + zD I 0,
(3.28)
where index I takes the values (3.24). However, apart from (3.26) and (3.28) there are harmonic
constraints originating from the symmetries of the Lagrangian for harmonics (3.19). We analyze
them in details in the next subsection.
217
3.4. Constraints and Hamiltonian for harmonic variables

Let us define the canonical momenta for harmonic variables,
v (,0)i =
=
v (0,)i =
LUSp(4)
(,0)
u i
is1 (,0)i
R 2 (,0)i (,0)
+ u(0,)i (,) + u(0,)i (,)

,
2u
u
e
2
LUSp(4)
(0,)
u i
is2 (0,)i
R 2 (,0)i (,)
+ u(,0)i (,) + 2u(0,)i (0,)

.
u
u
e
2
They are used in constructing of covariant harmonic momenta,
=
(+,0) (,0)i
S1 = ui
(,0) (+,0)i
= is1 ,
(3.30a)
(0,) (0,+)i
ui
v
= is2 ,
2
2R (,0)
(,0) (,0)i
D (,0) = ui
v
=
,
(3.30b)
ui
(3.29)
(0,+) (0,)i
v
S2 = ui
(0,) (0,)i
D (0,) = ui
e
2R 2 (0,)
=
,
2R 2 (,)
,
e
2R 2 (,)
(,0) (0,)i
(0,) (,0)i
v
ui
v
=
.
D (,) = ui
e
There are six constraints with the canonical momenta (3.29)
(,0) (0,)i
D (,) = ui
(+,0) (,0)i
C1 = ui
(0,) (,0)i
+ ui
(,0) (+,0)i
+ ui
(3.30c)
(3.30d)
(3.30e)
(3.30f)
= 0,
(0,+) (0,)i
(0,) (0,+)i
v
+ ui
v
= 0,
C2 = ui
(,0) (0,)i
(0,) (,0)i
v
ui
v
= 0,
C (,) = ui
(,0) (0,)i
(0,) (,0)i
(,)
= ui
v
+ ui
v
= 0.
C
(3.31)
Eqs. (3.30), (3.31) considered together allow one to express all sixteen harmonic momenta (3.29)
through the variables S, D, C in the lhs in (3.30), (3.31). Therefore we will use further the
covariant momenta (3.30) instead of canonical ones.
There are also the following constraints for the harmonic variables which appear by varying
(3.19) over Lagrange multipliers,
LUSp(4)
(+,0) (,0)i
= ui
u
1 = 0,
1
LUSp(4)
(0,+) (0,)i
= ui
u
1 = 0,
2 =
2
LUSp(4)
(,0) (0,)i
(,) =
= ui
u
= 0,
(,)
1 =
218
LUSp(4)
= ui(,0) u(0,)i = 0.
(3.32)
(,)
Let us introduce the Poisson brackets for the harmonic variables and corresponding momenta,
(0,) (0,)j
(,0) (,0)j
j
j
,v
= i ,
ui
,v
= i , other brackets vanish. (3.33)
ui
P
P
(,) =
It is easy to see that the functions (3.30a), (3.30b) commute weakly under these brackets with
the constraints (3.31), (3.32) and therefore they belong to the first class according to the Diracs
terminology. Another first-class constraint appears as the equation of motion for the einbein field,
0=
L 1 m
= p pm m2 + D (++,0) D (,0) + D (0,++) D (0,)
e
2

+ D (+,+) D (,) + D (+,) D (,+) .
(3.34)
This equation is a modification of the mass-shell constraint (3.14) with the covariant momenta
for the harmonic variables.
Let us now turn to the second-class constraints for the harmonic variables. We denote the
covariant momenta (3.30), (3.31) and constraints (3.32) as

D M = S1 , S2 , D (++,0) , D (,0) , D (0,++) , D (0,) , D (+,+) , D (,) , D (+,) , D (,+) ,

C I = C1 , C2 , C (+,+) , C (,) , C (+,) , C (,+) ,

I = 1 , 2 , (,) , (+,+) , (,+) , (+,)
(3.35)
and calculate their Poisson brackets,

I J
I
IJ M
C , C P = FM
D ,
C , J P = 2JI ,
(3.36)
I J is some constant matrix. Eq. (3.36) show that the constraints (3.31), (3.32) are secondwhere FM
class and therefore should be taken into account with the help of the Dirac bracket,
[A, B]D = [A, B]P +

1
1
A, C I P [I , B]P [A, I ]P C I , B P
2
2
1
IJ M
[A, I ]P FM
D [J , B]P ,
(3.37)
4
were A, B are arbitrary two functions on a phase space for harmonic variables.
Now we are ready to define the Hamiltonian for the particle on USp(4)/(U (1) U (1)) coset
as a Legendre transform for the Lagrangian (3.19),
(+,0) (,0)i
HUSp(4) = u i
(,0) (+,0)i
u i
(0,) (0,+)i
u i
v
LUSp(4) .
(0,+) (0,)i
u i
(3.38)
Expressing the velocities for harmonic variables from (3.29) through the momenta and substituting them into (3.38) we find

HUSp(4) = 2R 2 e1 (++,0) (,0) + (0,++) (0,) + (+,+) (,)

(0,0)
(0,0)
+ (+,) (,+) 1 (S1 is1 ) 2 (S2 is2 ) L ,
(3.39)
where L is given by (3.22). Since the constraints (3.32) are accounted by the Dirac bracket
(0,0)
(0,0)
are arbitrary, we treat them as
(3.37), we omit L further. Note that the functions 1 , 2
the Lagrange multipliers and denote further as , , respectively. As a result, the Hamiltonian
219
(3.39) is given by
e (++,0) (,0)
D
D
+ D (0,++) D (0,) + D (+,+) D (,)
2R 2

+ D (+,) D (,+) (S1 is1 ) (S2 is2 ).
HUSp(4) =
(3.40)
The Hamiltonian describing the dynamics of both superspace and harmonic variables reads
H = HUSp(4) + HN =4 ,
(3.41)
where HN =4 and HUSp(4) are given by (3.13) and (3.40), respectively.

4. GuptaBleuler quantization of massive harmonic superparticle with central charge
term
According to the canonical quantization, one replaces the canonical momenta (3.9)(3.11),
(3.29) with the following differential operators,
pm im ,
v (,0)i
i i
(,0)
ui
,
i
v (0,)i
i = i
(0,)
ui
,
i
(4.1)
The covariant harmonic momenta (3.30) turn into the harmonic derivatives (2.16), while the
Grassmann constraints (3.26) correspond to the following covariant spinor derivatives5
D(,0) =
(,0)

+ i m (,0) m + iz(,0) ,

+ i m (0,) m + iz(0,) ,
(0,)

(,0)
(,0)
= (,0) i (,0) m m i z
,
D

= (0,) i (0,) m m i z (0,)

,
D (0,)
D(0,) =
with non-trivial anticommutation relations given by

(+,0) (,0) (0,+) (0,)
D , D
= D , D
= 2iz ,
(+,0) (,0) (0,+) (0,)
= D
= 2i z ,
, D
, D
D
(+,0) (,0) (0,+) (0,)
= D , D
= 2im m .
D , D
(4.2)
(4.3)
The operators (4.1) should be realized in some Hilbert space formed by the superfunctions

= x m , i , i , u ,
(4.4)
which should satisfy some equations of motion and constraints originating from the superparticle constraints. The superparticle has both first- and second-class constraints. The first-class
constraints (3.30a), (3.30b), (3.34) form closed algebra under the Poisson or Dirac bracket.
5 The operators D i , D
i are multiplied here by i for convenience.
220
Therefore, they all should be imposed on state vectors,6

S (s1 ,s2 ) = s1 (s1 ,s2 ) ,
S2 (s1 ,s2 ) = s2 (s1 ,s2 ) ,

1
1
m m 2 X + m2 (s1 ,s2 ) = 0,
R
(4.5)
(4.6)
where7
X = D (,0) D (++,0) + D (0,) D (0,++) + D (,) D (+,+) + D (+,) D (,+) .
(4.7)
The second-class constraints should be accounted either by constructing the corresponding

Dirac bracket or by applying GuptaBleuler method. In our case the second-class harmonic constraints (3.31), (3.32) are taken into account by the Dirac bracket (3.37), while the spinorial ones
(3.26) should be accounted la GuptaBleuler. It means that they have to be divided into two
complex conjugate subsets with weakly commuting constraints in each subset. As follows from
the algebra (4.3), there are two ways of separating the derivatives (4.2) into such subsets:
(+,0) (0,+) (+,0) (0,+) (,0) (0,) (,0) (0,)
D , D , D
(4.8)
D , D , D
,
, D
, D
(+,0) (0,) (+,0) (0,) (,0) (0,+) (,0) (0,+)
D , D , D
.
, D
, D
D , D , D
(4.9)
Both these lines (4.8), (4.9) lead to the equivalent superfield realizations of supersymmetry. Here
we consider in details only (4.8) and give short comments on the second case (4.9) in the end of
this section on a particular example. Therefore we require the superfield (s1 ,s2 ) to be analytic,
(+,0) (s1 ,s2 )
(0,+) (s1 ,s2 )
= D
= 0.
D(+,0) (s1 ,s2 ) = D(0,+) (s1 ,s2 ) = D
(4.10)
Note that the physical states should respect the first-class -symmetry constraints (3.28)
which are not accounted so far. The problem is that the harmonic part of the superparticle
Lagrangian (3.19) violates the -symmetry (3.7). As follows from (4.6), the states (s1 ,s2 ) acquire additional masses due to the eigenvalues of the operator X and the BPS condition (2.4)
is violated. This is not surprising as we consider a superparticle in the harmonic superspace
ZH = {x m , i , i , u}, where the harmonic variables have non-trivial dynamics rather than playing auxiliary role. To resolve this problem and to obtain the physical states describing irreducible
representations of supersymmetry algebra we have to freeze the harmonic dynamics by imposing additional harmonic constraints on the superfield (s1 ,s2 ) . Such constraints should be
first-class and should be compatible with the second-class constraints (4.10). Since the harmonic
derivatives D (++,0) , D (0,++) , D (+,+) , D (,+) , D (+,) leave the set of Grassmann derivatives
in (4.10) invariant, we require them to annihilate the state,
D (++,0) (s1 ,s2 ) = D (0,++) (s1 ,s2 )
= D (+,+) (s1 ,s2 ) = D (,+) (s1 ,s2 ) = D (+,) (s1 ,s2 ) = 0.
(4.11)
It is easy to see that the state under constraints (4.11) belongs to the kernel of the operator X, i.e.
X (s1 ,s2 ) = 0. Therefore the mass-shell constraint (4.6) reduces to

m
m + m2 (s1 ,s2 ) = 0,
(4.12)
6 Further we label the functions by the values of U (1) charges as (s1 ,s2 ) .
7 Here we use a particular ordering of the harmonic derivatives although other orderings are also possible.
221
that is nothing but the usual KleinGordon equation. The BPS condition (2.4) is now restored
and the state respects the constraints of -symmetry which eliminate the unphysical degrees of
freedom.
Upon quantization, the generators of -symmetry (3.28) turn into the differential operators,
I = m m D I + z DI ,
I = m m D I + zD I .
Owing to the analyticity (4.10), we have to impose only the following constraints,
m

m D (,0) + z D(,0) (s1 ,s2 ) = 0,

m
m D (0,) + z D(0,) (s1 ,s2 ) = 0,
m
(,0) (s1 ,s2 )
m D (,0) + zD
= 0,
m

(0,)
m D (0,) + zD
(s1 ,s2 ) = 0.
(4.13)
(4.14)
The relations (4.14) can be brought to more useful form. For this purpose we introduce the
operators

i (,0) (,0)
z D (,0) D(,0) ,
zD
D
4

i (0,) (0,)
(0,)
= zD
z D (0,) D(0,) ,
Y
D
4
which have the following commutators with (4.13)
Y (,0) =
(4.15)
1 (+,0) (,0)
= m m D (,0) + z D(,0) ,
,Y
D
z
1 (0,+) (0,)
= m m D (0,) + z D(0,) ,
,Y
D
z
1 (+,0) (,0)
(,0)
D
= m m D (,0) + zD
,Y
,
z
1 (0,+) (0,)
(0,)
D
(4.16)
= m m D (0,) + zD
,Y
.
z
Therefore instead of (4.14) we impose the following first-class constraints on the superfield,
(,0) (,0)

zD
z D (,0) D(,0) (s1 ,s2 ) = 0,
D
(0,) (0,)

zD
(4.17)
z D (0,) D(0,) (s1 ,s2 ) = 0.
D
In fact, the constraints (4.17) are stronger than (4.14). Nevertheless, as is argued in [18,19] for
N = 2 superparticle, such first-class constraints should be imposed on the state since they appear
as functions of second-class constraints (3.26). Moreover, they can be considered as the generators of symmetries of the superparticle Lagrangian as is shown in Appendix B for the massless
case.
Let us summarize all the equations for the superfield (s1 ,s2 ) in a single list
S1 (s1 ,s2 ) = s1 (s1 ,s2 ) ,
S2 (s1 ,s2 ) = s2 (s1 ,s2 ) ,

(+,0) (s1 ,s2 )
(0,+) (s1 ,s2 )
= D
= 0,
D(+,0) (s1 ,s2 ) = D(0,+) (s1 ,s2 ) = D
D (++,0) (s1 ,s2 ) = D (0,++) (s1 ,s2 ) = D (+,+) (s1 ,s2 ) = 0,

D (,+) (s1 ,s2 ) = D (+,) (s1 ,s2 ) = 0,
222

D (,0) z D (,0) D(,0) (s1 ,s2 ) = 0,
(0,) (0,)

zD
z D (0,) D(0,) (s1 ,s2 ) = 0,
D
m

m + m2 (s1 ,s2 ) = 0.
zD
(,0)
(4.18)
Note that due to the algebra (A.3) the derivatives D (+,) , D (,+) commute as [D (+,) , D (,+) ] =
S2 S1 . Hence, the operator S2 S1 also annihilates the state, (S2 S1 ) (s1 ,s2 ) = 0 and the resulting superfield has equal U (1) charges, s1 = s2 .
We point out that the constraints D (,+) (s1 ,s2 ) = D (+,) (s1 ,s2 ) = 0 restrict effectively the
superfield to depend on USp(4)/(SU(2) U (1)) harmonic variables rather than the ones on
the USp(4)/(U (1) U (1)) coset. Therefore these constraints can be effectively resolved by
considering the USp(4)/(SU(2) U (1)) harmonics. Nevertheless, in the present work we do not
follow this way and work only with the USp(4)/(U (1) U (1)) harmonic variables introduced
above.
4.1. N = 4 massive vector multiplet
Now we consider particular examples of superfields (s1 ,s2 ) with the lowest values of U (1)
charges satisfying (4.18). It is easy to see that the case s1 = s2 = 0 is trivial since such chargeless
superfield is just a constant, (0,0) = const. Therefore the first physically interesting example
appears when s1 = s2 = 1,
(1,1) W (+,+) .
(4.19)
One can check that for this superfield the equations in the last three lines in (4.18) follow from
the other ones, while the relations in the first line are satisfied automatically. As a result, the
superfield (4.19) obeys
D(+,0) W (+,+) = D(0,+) W (+,+) = D (+,0)
W (+,+) = D (0,+)
W (+,+) = 0,
(++,0)
(,+)
(+,+)
(+,+)
=D
=D
(0,++)
(+,)
(+,+)
(+,+)
Eq. (4.20) mean that the superfield

make the change of coordinates
=D
(+,+)
(+,+)
= 0,
= 0.
W (+,+)
(4.20)
(4.21)
(4.22)
is analytic. To resolve these constraints we first
xAm = x m i (,0) m (+,0) i (+,0) m (,0) i (0,) m (0,+)

i (0,+) m (0,) ,
(4.23)
and then pass from to frame,8

DI DI = eZ DI eZ = DI DI Z ,

D I D I = eZ D I eZ = D I D I Z ,
W (+,+) W (+,+) = eZ W (+,+) ,
(4.24)
8 Here we use the terminology of [6] used for similar constructions in N = 2 harmonic superspace. In fact, such
passing from to frame is nothing but the use of two different representations for Grassmann and harmonic derivatives
which are related by the unitary operator eZ .
223
where
(+,0) (,0)
(0,+) (0,)
+ i z
Z = iz (+,0) (,0) + iz (0,+) (0,) + i z
(4.25)
is a bridge superfield. In the analytic coordinates (4.23) the derivatives (4.2) read
D(+,0) =
D(0,+) =
D (+,0)
=
D(,0) =
,
(0,)
D(0,) =
(,0)
,
(,0)
(0,+)
= (0,) ,
D

(0,+)

+ 2i m (,0) m + 2iz(,0) ,

+ 2i m (0,) m + 2iz(0,) ,
(0,+)

(,0)
(,0)
D
= (+,0) 2i (,0) m m 2i z
,

(0,)
(0,)
D
= (0,+) 2i (0,) m m 2i z
.

(4.26)
(+,0)
(0,+)
(+,0) (0,+)
Since the derivatives D , D , D
, D
are short, the superfield W (+,+) depends
only on the analytic coordinates,

(0,+)
W (+,+) = W (+,+) xAm , (+,0) , (0,+) , (+,0)
(4.27)
,
,u .
To solve the constraints (4.21), (4.22) we rewrite the covariant harmonic derivatives in the
analytic coordinates in frame (omitting the terms vanishing on the analytic superfields),

2
2

D(++,0) = D (++,0) 2i (+,0) m (+,0)
iz (+,0) i z (+,0) ,
m
xA

2
2

iz (0,+) i z (0,+) ,
D(0,++) = D (0,++) 2i (0,+) m (0,+)
m
xA

D(+,+) = D (+,+) 2i (+,0) m (0,+) + (0,+) m (+,0)
xAm
(+,0) (0,+)
(+,0) (0,+)
2i z
,
2iz
D(,+) = D (,+) (0,+)

D(+,) = D (+,) + (+,0)
(+,0)
(0,+)
(0,+)

(+,0)
(+,0)

.
(0,+)
(4.28)
The full component structure of the superfield (4.27) is very long. However, on the equations of
motion (4.21), (4.22) all auxiliary fields vanish and the component decomposition appears pretty
short,
W (+,+) = u[i(+,0) uj(0,+)
f ij + (+,0) i ui(0,+) (0,+) i ui(+,0)
]
+ (+,0)
i ui(0,+) (0,+)
i ui(+,0)
2
(+,0) 2 ij (,0) (0,+)

(+,0) (0,)
+ iz
f u[i uj ] + iz (0,+) f ij u[i uj ]
2
2

(,0) (0,+)
(+,0) (0,)
+ i z (+,0) f ij u[i uj ] + i z (0,+) f ij u[i uj ]
(+,0) (,0)

(0,+) (0,)
+ iz (+,0) (0,+) + i z (+,0) (0,+) f ij u[i uj ] u[i uj ]
+ 2i (+,0) m (+,0) m f ij u[i
(,0) (0,+)
uj ]
+ 2i (0,+) m (0,+) m f ij u[i
(+,0) (0,)
uj ]
224
(+,0) (,0)

(0,+)
(+,0) m
(0,+) (0,)
+ i (+,0)
+ (0,+)
m f ij u[i uj ] u[i uj ]
(0,+)
(+,0) (0,+)
(0,+)
(+,0)
A + (0,+)
+ (+,0)
F +
A
G + (+,0)
2
2 (0,+) i (,0)

(,0)
iz (+,0) (0,+) i ui
iz (+,0)
ui
(0,+) 2 (+,0) i (0,)
(0,+) 2 (+,0) i (0,)
+ iz
ui
+ iz
ui
(+,0) 2 (0,+) i (,0)

(+,0) 2 (0,+) i (0,)
i z
ui
i z
ui
(0,+) 2 (+,0) i (0,)

(0,+) 2 (+,0) i (0,)
+ i z
ui
+ i z
ui
+ 2i (+,0) (0,+) (+,0) m m i ui
(,0)
+ 2i (+,0) (0,+) (0,+) m m i ui
(0,)
(,0)
+ 2i (+,0) (0,+) (+,0) m m i ui
+ 2i (+,0) (0,+) (0,+) m m i ui

2
2

2
2
(,0) (0,)
(,0) (0,)
z2 (+,0) (0,+) f ij u[i uj ] z 2 (+,0) (0,+) f ij u[i uj ]
2
2
2
2

(,0) (0,)
zz (+,0) (0,+) + (0,+) (+,0) f ij u[i uj ]
2
2

(0,+)
(+,0) m
(,0) (0,)
2z (+,0) (0,+)
+ (0,+) (+,0)
m f ij u[i uj ]
2
2

(0,+)
(+,0) m
(,0) (0,)
2z (+,0) (0,+)
+ (0,+) (+,0)
m f ij u[i uj ]
(0,)
(0,+) (+,0) (0,+) m n

(,0) (0,)
m n f ij u[i uj ] .
+ 4(+,0)
(4.29)
The scalar fields in (4.29) satisfy KleinGordon equation,

( + zz)f ij = 0,
(4.30)
while for the spinors we have Dirac equations with the correct signs in the mass terms owing
to (3.8),
im m i iz i = 0,
im m i i z i = 0.
(4.31)
, A are related to each other as

The fields F , A and G

i m m F + izA = 0,
+ i z A = 0,
i m m G
i m m A + i z F = 0,
= 0,
i m m A + izG
m m A
= 0.
= 0,
m m A
(4.32)
Eq. (4.32) mean that the fields A and A obey the relations
( + zz)A = 0,
( + zz)A = 0,
m m A = 0,
m m A = 0
(4.33)
and, hence, describe complex massive vector field. We conclude that (4.29) represents a superfield realization of N = 4 massive vector multiplet,

5 complex scalars,
+ m2 f ij = 0 ij f ij = 0 ,
im m i + m i = 0,
4 Dirac spinors,
225
im m i mi = 0,

+ m2 A = 0,
m m A = 0,

m m A = 0.
+ m2 A = 0,
1 massive complex vector,

(4.34)
Let us now consider in short the second type of separation of superparticle constraints (4.9)
on the examples of a superfield W (+,) . The Grassmann derivatives in (4.9) should annihilate
this superfield,
D(+,0) W (+,) = D(0,) W (+,) = D (+,0)
W (+,) = D (0,)
W (+,) = 0.
(4.35)
We impose also the following harmonic constraints

D (++,0) W (+,) = D (0,) W (+,)
= D (+,) W (+,) = D (,) W (+,) = D (+,+) W (+,) = 0,
(4.36)
since harmonic derivatives in (4.36) leave the set of the constraints (4.35) invariant.
To solve the constraints (4.35) we consider this superfield in analytic coordinates,
xA m = x m i (,0) m (+,0) i (+,0) m (,0) + i (0,) m (0,+)
+ i (0,+) m (0,)
(4.37)
and pass from to frame by the rules (4.24) with the bridge
Z = iz (+,0) (,0) iz (0,+) (0,) + i z
(+,0) (,0)
i z
(0,+) (0,)
(4.38)
The component structure of the superfield W (+,) is found in a similar way,

(+,0) (0,) ij
(0,)
(+,0)
uj ] f + (+,0) i ui
(0,) i ui
(+,0) i (0,)
(0,) i (+,0)
+
ui

ui
2
(+,0) 2 ij (,0) (0,)

(+,0) (0,+)
+ iz
f u[i uj ] iz (0,) f ij u[i uj ]
2
2

(,0) (0,)
(+,0) (0,+)
+ i z (+,0) f ij u[i uj ] i z (0,) f ij u[i uj ]
(+,0) (,0)

(0,+) (0,)
+ iz (+,0) (0,) + i z (+,0) (0,) f ij u[i uj ] u[i uj ]
(,0) (0,)
(+,0) (0,+)
+ 2i (+,0) m (+,0) m f ij u[i uj ] 2i (0,) m (0,) m f ij u[i uj ]
m

+ i (+,0) (0,)
+ (0,) (+,0)
m f ij u[i(+,0) uj(,0)
u[i(0,+) uj(0,)
]
]
(0,)
(+,0) (0,)
(0,)
(+,0)
+ (+,0)
F +
A
A + (0,)
G + (+,0)
W (+,) = u[i
+ ,
(4.39)
where dots stand for the terms with s to the third and fourth powers. All components here
depend on xA m and satisfy the free equations of motion (4.34).
Analogously, one can consider the superfield W (,+) constrained by
(,0) (+,)
(0,+) (+,)
D(,0) W (+,) = D(0,+) W (+,) D
W
= D
W
= 0,
D (,0) W (+,) = D (0,++) W (+,) = D (,+) W (+,)

= D (,) W (+,) = D (+,+) W (+,) = 0.
The component structure of
swapped.
W (,+)
(4.40)
is similar to (4.39), but the values of the U (1) charges are
226
5. GuptaBleuler quantization of massless superparticle

We turn to the massless N = 4 superparticle by considering the model (3.1) in the massless
limit, i.e.
m = z = z = 0.
(5.1)
The Lagrangian (3.1) reduces to

1
L1 = e1 m m .
(5.2)
2
We point out that the massless superparticle has U (4) R-symmetry group rather than USp(4).
Therefore, it is naturally to extend this model with the SU(4) harmonics. This case was already
studied in [18], where the superfield description of N = 4 vector multiplet was given. Therefore
in the present work we study the N = 4 massless superparticle extended by USp(4) harmonic
variables. Exactly this feature allows us to get some new insight on the problem of superfield
formulation of N = 4 SYM model in harmonic superspace.
The Lagrangian of N = 4 harmonic superparticle reads
LN =4 = L1 + LUSp(4) ,
(5.3)
where L1 and LUSp(4) are defined in (5.2) and (3.19), respectively. The further quantization of
this model is straightforward and can be read from the above considerations in the limit (5.1).
Here we mention only the new features.
In the massless case the algebra of covariant spinor derivatives (4.2) has non-trivial anticommutation relations given only in the last line in (4.3). Therefore, in the GuptaBleuler
quantization approach, there are eight ways of separation of corresponding constraints,
(0,+) (0,) (+,0) (,0) (+,0) (,0) (0,+) (0,)
D , D , D
(5.4a)
D , D , D
,
, D
, D
(+,0) (0,+) (+,0) (0,+) (,0) (0,) (,0) (0,)
D , D , D
,
, D
, D
D , D , D
(5.4b)
(+,0) (0,) (+,0) (0,) (,0) (0,+) (,0) (0,+)
D , D , D
,
, D
, D
D , D , D
(5.4c)
(+,0) (0,) (,0) (0,) (+,0) (0,+) (,0) (0,+)
D
,
, D , D
, D
, D
, D
D , D
(5.4d)
(+,0) (0,+) (,0) (0,+) (+,0) (0,) (,0) (0,)
D
,
, D , D
, D
D , D , D , D
(5.4e)
(+,0) (0,+) (+,0) (0,) (,0) (0,+) (,0) (0,)
D , D , D
(5.4f)
D , D
,
, D
, D
, D
(,0) (0,+) (,0) (0,) (+,0) (0,+) (+,0) (0,)
D
,
, D , D , D
, D
, D , D
D
(5.4g)
(+,0) (0,+) (,0) (0,) (+,0) (0,+) (,0) (0,)
D
.
, D
, D
, D
D , D , D , D
(5.4h)
As before, we assume that the state is described by a superfield (s1 ,s2 ) subject to (4.5). To take
into account the constraints (5.4) we claim
{D, } (s1 ,s2 ) = 0,
(5.5)
where {D, } are covariant spinor derivatives from one of the subsets in (5.4). We have to impose
also the harmonic constraints,
A (s ,s )
D 1 2 = 0,
(5.6)
227
where {D A } is a subset of harmonic derivatives (2.16) which leave the set of Grassmann derivatives {D, } invariant. Apart from these constraints there are also -symmetry ones with the
generators (4.13). In the massless case they lead to the following equations
m m D I (s1 ,s2 ) = 0,
m m D I (s1 ,s2 ) = 0.
(5.7)
Using the algebra of covariant spinor derivatives one can show that the constraints (5.7) follow
from
D I D J (s1 ,s2 ) = 0,
D I DJ (s1 ,s2 ) = 0,
(5.8)
where the indices I, J have the values (3.24). As explained in Appendix B, the constraints (5.8)
originate from the bosonic version of -symmetry. We point out that not all the constraints (5.6),
(5.8) are independent, however they are all first-class. Some of them become trivial when definite
subset in (5.4) is chosen.
As a result, all the first-class and one half of second-class constraints are taken into account by
Eqs. (5.5)(5.8) for the superfield (s1 ,s2 ) . In the following subsections we consider the particular
examples of massless representations of N = 4 superalgebra on such superfields with lowest
values of U (1) charges s1 , s2 , which correspond to different ways of separations of constraints
(5.4) and describe different multiplets.
5.1. Supergauge multiplet
Regarding to the separations of constraints (5.4a)(5.4c), there are two essentially different
superfield realizations of N = 4 supergauge multiplets. In the next subsections we consider both
of them separately.
5.1.1. Chargeless superfield representation
According to the separation of constraints (5.4a), the superfield (s1 ,s2 ) should be annihilated
by the following Grassmann derivatives
(+,0) (s1 ,s2 )
(,0) (s1 ,s2 )
= D
= 0.
D(0,+) (s1 ,s2 ) = D(0,) (s1 ,s2 ) = D
As follows from (5.8), the linearity conditions read

(+,0) 2 (s ,s ) (,0) 2 (s ,s ) (+,0) (,0) (s ,s )
1 2 = 0,
1 2 = D
1 2 = D
D
D
(0,+) 2 (s ,s ) (0,) 2 (s ,s ) (0,+) (0,) (s ,s )
D
1 2 = 0.
1 2 = D
1 2 = D
D
(5.9)
(5.10)
We have to impose also the harmonic constraints, to fix the dynamics over the harmonic
variables. Only the operators D (++,0) , D (,0) , D (0,++) , D (0,) leave the set of Grassmann
derivatives in (5.9) invariant. Therefore we claim
D (++,0) (s1 ,s2 ) = D (,0) (s1 ,s2 ) = D (0,++) (s1 ,s2 ) = D (0,) (s1 ,s2 ) = 0.
(5.11)
According to the algebra (A.3), [D (++,0) , D (,0) ] = S1 , [D (0,++) , D (0,++) ] = S2 , the constraints (5.11) imply s1 = s2 = 0. Hence, the state under considerations is realized by chargeless
superfield (0,0) W1 .
However, the constraints (5.11) do not fix the harmonic dynamics completely. Therefore we
impose also the quadratic harmonic constraint,
D (+,+) D (+,+) W1 = 0.
(5.12)
228
As a consequence of the algebra (A.3), the operators D (,) D (+,+) , D (+,) D (,+) annihilate
the state as well and the whole operator (4.7) vanishes on this superfield,
XW1 = 0.
(5.13)
Therefore we are convinced that the constraints (5.11), (5.12) are sufficient to fix the dynamics
over the harmonic variables and to eliminate all unphysical degrees of freedom in this superfield.
Let us rewrite all these constraints for the superfield W1 in a single list,
(+,0)
(,0)
W1 = D
W1 = 0,
D(0,+) W1 = D(0,) W1 = D
D (++,0) W1 = D (,0) W1 = D (0,++) W1 = D (0,) W1 = 0,

D (+,+) D (+,+) W1 = 0.
(5.14)
The linearity constraints (5.10) are not in this list since they are not independent but follow
from (5.14).
To solve the constraints in the first line of (5.14) we pass to the analytic coordinates,
xAm = x m i (0,) m (0,+) + i (0,+) m (0,) i (+,0) m (,0)
+ i (,0) m (+,0) ,
(5.15)
in which the Grassmann derivatives in (5.9) become short,
.
(,0)
Therefore the superfield W1 depends only on one half of Grassmann variables,

(0,+) (0,)
W1 = W1 xAm , (+,0) , (,0) ,
,
,u .
D(0,) =
(0,)
(,0)
D
=
(5.16)
(5.17)
Let us rewrite also the harmonic derivatives in (5.14) in such analytic coordinates (we omit the
terms acting trivially on W1 ),
(,0)
DA
= D (,0) + (,0)
(0,)
= D (0,) + (0,)
DA
(,0)
(0,)
,
,
(5.18)

= D (+,+) 2i (+,0) m (0,+) (0,+) m (+,0) m
+ (0,+) (,0) + (+,0) (0,) .

(5.19)

Since the operators (5.18) do not contain the spatial derivatives, the constraints in the second line
of (5.14) are not kinematical but purely algebraical for the components of W1 . In fact, these constraints show that W1 depends effectively on USp(4)/(SU(2) SU(2)) harmonic variables since
the derivatives in (5.11) form two su(2) algebras. In principle, such harmonic constraints (5.11)
can be solved manifestly by choosing the appropriate harmonic variables. As a result, only the
constraint in the last line of (5.14) is true equation of motion for the superfield W1 .
Finally, we impose also the reality constraint,
(+,+)
DA
W 1 = W1 ,
where the conjugation is defined in (2.20).
(5.20)
229
Now we give the solution of all the constraints for the superfield W1 ,
(+,0) (,0)
(0,+) (0,)
W1 = + if ij u[i uj ] u[i uj ]
(,0)
+ i (+,0) i ui
(,0)
+ (+,0)
+ 4(+,0)
(+,0)
i (,0) i ui
+ i
(0,+)
(0,+) (0,) ( )
F () +
F
(0,+)
(,0) (0,)
uj ]
(+,0) (0,)
ij (,0) (0,+)
4
f u[i uj ]
+ 4(,0)
(0,)
(0,) i (0,+)
i
i ui
ui
(+,0) (0,+)
uj ]
(,0) (0,+)
ij (+,0) (0,)
4
f u[i uj ]
i (0,)
i (0,+)
2 (+,0) (,0) (0,+) ( )
ui
+ 2 (+,0) (,0) (0,) ( )
ui
f ij u[i
(0,)
f ij u[i
i (,0)
i (+,0)
2 (+,0) (0,+) (0,) ( )
+ 2 (,0) (0,+) (0,) ( )
ui
ui

(,0) (0,+) (0,)
(+,0) (,0)
(0,+) (0,)
. (5.21)
+ 4(+,0)
if ij u[i uj ] u[i uj ]
All component fields here depend only on xAm and satisfy the corresponding free equations of
motion,
= 0,
f
ij
=0
1 real scalar,
f ij = 0 ,
ij
= 0,
i
m
m F()
= 0,
5 real scalars,
m m i = 0,
m m F( )
= 0,
4 Weyl spinors,
1 Maxwell field
(5.22)
and reality conditions,

= ,
fij = f ij ,
i = i ,
F() = F( )
.
(5.23)
As a result, the N = 4 SYM multiplet is embedded into the superfield W1 .

Let us consider now the Grassmann derivatives in the second subset of (5.4a). We denote
the superfield annihilated by these variables as W2 . Similarly as W1 , it satisfies the following
constraints,
(0,+)
(0,)
D
W2 = D
W2 = D(+,0) W2 = D(,0) W2 = 0,
D (++,0) W2 = D (,0) W2 = D (0,++) W2 = D (0,) W2 = 0,

D (,) D (,) W2 = 0,
W 2 = W2 .
(5.24)
The Grassmann constraints in (5.24) are solved by passing to the corresponding analytic coordinates,

(,0) (0,+) (0,)
W2 = W2 xA m , (+,0)
(5.25)
,
,
,
,u ,
where
xA m = x m + i (0,) m (0,+) i (0,+) m (0,) + i (+,0) m (,0) i (,0) m (+,0) .
(5.26)
The component structure of W2 is analogous to (5.21), but the values of U (1) charges of superspace coordinates should be changed appropriately. Therefore, W2 also describes the N = 4
SYM multiplet.
230
5.1.2. Charged superfield representation

Consider the separation of constraints (5.4b) on the example of a superfield W (+,+) with U (1)
charges s1 = s2 = 1. According to (5.4b) and (4.11), it satisfies similar equations as the massive
vector multiplet considered in Section 4.1,
D(+,0) W (+,+) = D(0,+) W (+,+) = D
(+,0)
(++,0)
(,+)
(+,+)
(+,+)
=D
=D
(0,++)
(+,)
(+,+)
(+,+)
=D
W (+,+) = D
(+,+)
(0,+)
(+,+)
W (+,+) = 0,
(5.27)
= 0,
(5.28)
= 0.
(5.29)
We point out that the harmonic constraints (5.29) are not kinematical, but they just show that
W (+,+) depends effectively on USp(4)/(SU(2) U (1)) harmonic variables. These constraints
can be solved manifestly by the appropriate choice of harmonics, but we prefer to work here with
the harmonic variables on the USp(4)/(U (1) U (1)) coset, introduced above.
There are also linearity conditions (5.8),
(,0) 2 (+,+) (0,) 2 (+,+)
D
W
= D
W
(,0) 2 (+,+) (0,) 2 (+,+)
= D
W
= D
W
= 0,
(,0) (0,) (+,+)
(,0) (0,) (+,+)
D
W
=D
W
= 0.
D
(5.30)
D
Finally, in the massless case we impose also the reality condition,

(+,+) = W (+,+) .
W
(5.31)
The solution of these constraints can be obtained from (4.29) by putting the central charges to
zero, z = z = 0 and taking into account the reality (5.31),
(+,0) (0,+) ij
(0,+)
(+,0)
uj ] if + i (+,0) i ui
i (0,+) i ui
(+,0) i (0,+)
(0,+) i (+,0)
+ i
i
ui
ui
(+,0) (+,0) m
ij (,0) (0,+)
2
m f u[i uj ]
(0,+) m
(+,0) (0,)
2(0,+)
m f ij u[i uj ]
m

(+,0) (0,+)
+ (0,+) (+,0)
m f ij u[i(+,0) uj(,0)
(+,0) (0,+) ( )
(+,0) (0,+) ()
F
+
F

(+,0) m (0,+)
+i

+ (0,+) m (+,0) mnrs Gnrs
(,0)
2 (+,0) (0,+) (+,0) m m i ui
(0,)
2 (+,0) (0,+) (0,+) m m i ui
(,0)
2 (+,0) (0,+) (+,0) m m i ui
W (+,+) = u[i
u[i(0,+) uj(0,)
]
2 (+,0) (0,+) (0,+) m m i ui
(0,)
(0,+) (+,0) (0,+) m n

(,0) (0,)
m n f ij u[i uj ] ,
+ 4i(+,0)
(5.32)
where all components depend on the analytic coordinates (4.23). Owing to the reality condition
(5.31) we have
fij = f ij ,
i = i ,
F() = F( )
,
Gmnr = Gmnr .
(5.33)
These components satisfy free equations of motion,

5 real scalars,
f ij = 0 f ij ij = 0 ,
m
i
m
i
m = 0,
m = 0,
4 Weyl spinors,
m
m
m F() = 0,
m F = 0, 1 real Maxwell field,
( )
mnr
m G
= 0,
nrs
mnrs G
= 0,
1 antisymmetric tensor field.
231
(5.34)
(5.35)
(5.36)
(5.37)
It is well known that the antisymmetric tensor field (5.37) is classically equivalent to the scalar
field on-shell. Indeed, introducing the field
1
Lm = mnrs Gnrs ,
6
Eqs. (5.37) can be rewritten as
m Lm = 0,
m Ln n Lm = 0.
(5.38)
(5.39)
A general solution of (5.39) is expressed through the scalar field subject to the KleinGordon
equation,
Lm = m ,
= 0.
(5.40)
As a result, we see that the superfield (5.32) describes the N = 4 vector multiplet, in which
one of the scalars is dualized and represented by the antisymmetric tensor field. Here Gmnr is a
strength of the antisymmetric tensor field which can always be expressed through its potential as
Gmnr = m B nr + n B rm + r B mn .
In conclusion of this subsection we comment briefly on the other ways of separation of constraints (5.4c) on the examples of the superfields W (+,) , W (,+) . Similarly as in the massive
case, these superfields are constrained by (4.35), (4.36) and (4.40), respectively. The component
structure can be read from (4.39) by putting the central charges to zero, z = z = 0. Therefore,
these superfields also describe the N = 4 vector multiplet (5.34)(5.37) in which one of the
scalars is represented by the antisymmetric tensor field.
5.2. Gravitino multiplet
Let us consider the separation of constraints (5.4d) and require the superfield (0,+) to be
annihilated by the following Grassmann derivatives,
(0,+) (0,+)
(,0) (0,+)
D (+,0)
(0,+) = D
= D
= D(0,+) (0,+) = 0.
(5.41)
The harmonic derivatives commuting with the spinor derivatives in (5.41) are D (++,0) , D (,0) ,
D (0,++) , D (+,+) , D (,+) . Therefore we require them to annihilate the state,
D (++,0) (0,+) = D (,0) (0,+) = D (0,++) (0,+) = D (+,+) (0,+)
= D (,+) (0,+) = 0.
(5.42)
Finally, the superfield satisfies the linearity conditions (5.8),

(0,) 2 (0,+) (+,0) 2 (0,+) (,0) 2 (0,+) (0,) 2 (0,+)
= D
= D
= D
= 0,
D
D (,0) D (0,) (0,+) = D (+,0) D (0,) (0,+) = D (+,0) D (,0) (0,+) = 0.
(5.43)
232
To solve the constraints (5.41) we pass to the chiral-analytic coordinates,

y m = x m + i (,0) m (+,0) i (+,0) m (,0) i (0,) m (0,+)
i (0,+) m (0,) ,
(5.44)
in which the derivatives in (5.41), (5.42) are given by (as usual, we omit here the terms acting on
(0,+) trivially)
D (,0)
=
(,0)
DA
(0,++)
DA
(,+)
DA
,
(,0)
D (0,+)
=
= D (,0) + (,0)
= D (0,++) 2i
,
(0,)
D (0,+) =
(0,)
(5.45)
(,0)
(0,+) m (0,+)
(5.46)
m ,
= D (,+) 2i (,0) m (0,+) m (0,+)
(,0)
(5.47)
Since the derivatives (5.45) are short, the constraints (5.41) are manifestly solved by

(0,+)
,u .
(0,+) = (0,+) y m , (+,0) , (,0) , (0,+) ,
(5.48)
Using the expressions for the harmonic derivatives (5.46), (5.47) one can easily check that the
following component expression for (0,+) solves (5.42),
(0,+)
(0,+) = f i ui
(,0) (0,+)
uj ]
(+,0) ij u[i
(+,0) (0,+)
uj ]
+ (,0) ij u[i
(+,0) (,0)
1
(0,+) (0,)
(0,+)
+
+ (0,+) (0,+) ij u[i uj ] u[i uj ]
2
(+,0) (,0)
(0,+)
(+,0) (0,+)
(,0)
(,0) (0,+)
(+,0)
+ ( ) F i ui
( ) F i ui
+ ( ) F i ui
+ 2i (+,0) m (0,+) m f i ui
(,0)
2i (,0) m (0,+) m f i ui
(+,0)
+ 2i (0,+) m (0,+) m f i ui
(0,)
m
+ (+,0) (,0) (0,+) T( ) 2i (+,0) (,0) (0,+) (
m )

ij
(+,0) (,0)
(0,+) (0,)
m
i (+,0) (,0) (0,+) (
uj ] u[i uj ]
m ) u[i
m
+ 2i (,0) (0,+) (0,+) (
m ) u[i
All the
motion,
ij (+,0) (0,)
uj ]
ij (,0) (0,)
(+,0) (0,+) (0,+) m
2i
( m ) u[i uj ]
i (0,)
+ 2i (+,0) (,0) (0,+) (0,+) m m F
ui
.
component fields in (5.49) depend on y m given by (5.44) and
f i = 0,
4 complex scalars,
m m = 0,
ij

ij = 0 ,
m = 0,
m ij = 0
i
= 0,
m m F()
m
m
2 Weyl spinors,
5 Weyl spinors,
4 real Maxwell fields,
(5.49)
satisfy free equations of
m m T( ) = 0,
233
1 gravitino.
(5.50)
As a result we obtain the N = 4 gravitino multiplet.

The other ways of separations of constraints (5.4e), (5.4f), (5.4g) also lead to the gravitino
multiplets realized by the superfields (0,) , (,0) , (+,0) , respectively. The constraints for
these superfields can be easily read from the general expressions (5.5)(5.8). The component
structure is analogous to (5.49) with appropriate change of U (1) charges.
5.3. N = 4 supergravity multiplet
Consider the separation of constraints (5.4h) on the example of chargeless superfield
(0,0) . We impose the derivatives in the second set in (5.4h) as the constraints on ,
(+,0)
(0,+)
(,0)
(0,)
D
= D
= D
= D
= 0,
(5.51)
which show this superfield to be chiral,

D i = 0.
(5.52)
This constraint is explicitly solved by passing to the chiral coordinates

= y m , i , u .
ym
= xm
+ ii
m i ,
(5.53)
Next, we impose the harmonic constraints (4.11)

D (++,0) = D (0,++) = D (+,+) = D (,+) = D (+,) = 0,
which mean that is harmonic independent,

y m , i , u = y m , i .
(5.54)
(5.55)
Finally, there are the linearity constraints (5.8) originating from -symmetry (see Appendix B
for details),
D i Dj = 0.
(5.56)
It is well known that the solution of Eq. (5.56) describes the N = 4 supergravity multiplet
[11,33]. Its component structure is given by
[ij ]
= + i i + i j F() + i j k ij kl Tl( ) + i j k l ij kl C( ) ,
(5.57)
where the component fields satisfy

= 0,
1 complex scalar,
m i = 0,
[ij ]
m m F() = 0,
m m Ti( ) = 0,
m m C( ) = 0,
4 Weyl spinors,
6 Maxwell fields,
4 gravitini,
1 Weyl tensor.
As a result, this superfield describes the multiplet of N = 4 supergravity.
(5.58)
234
6. Applications of USp(4)/(U (1) U (1)) harmonic superspace to the N = 4 SYM model

The quantization of N = 4 harmonic superparticle has demonstrated an important role of the
specific N = 4 harmonic superspace with USp(4)/(U (1) U (1)) harmonic variables9 and the
corresponding superfields. We have seen that this superspace is naturally associated with N = 4
superparticle and therefore it is instructive to study its properties and to try to develop field theory
in it.
The purpose of this section is to apply the USp(4)/(U (1) U (1)) harmonic superspace to
the N = 4 SYM model and show that the superfields, obtained in quantizing the superparticle,
appear naturally in the solution of N = 4 SYM constraints. We consider also some possibilities
of constructing the invariant actions depending on these superfields in harmonic superspace and
discuss their relevance to the N = 4 SYM model.
6.1. Harmonic superspace analysis of N = 4 SYM constraints
Let us consider standard N = 4 superspace Z M = {x m , i , i } with supercovariant spinor
derivatives
m
m ,
D i
ii m m .
Di = + i i
(6.1)
=
i
x
x
i
According to the generic procedure of superspace formulation of the extended supersymmetric
models [33], one introduces the gauge connections for these derivatives,
Di i = Di + Vi ,
D i i = D i + Vi
and defines the superfield strengths by the following anticommutators,

i j
{ i , j } = 2 W ij .
, = 2 W ij ,
(6.2)
(6.3)
It is well known that the following N = 4 SYM constraints put the superfield strengths onshell [33],

1 j
D i W j k = i D l W lk ik D l W lj ,
3
Di W j k + Dj W ik = 0,
1
W ij = W ij = ij kl W kl .
2
(6.4)
(6.5)
(6.6)
Let us project the strengths W ij with harmonics,

W ij W I J = uI i uJ j W ij ,
(6.7)
where the indices I, J take the values (3.24). We denote these superfields also as
(0,+) (0,) ij
uj
W ,
(+,0) (0,+) ij
uj
W ,
W (+,+) = ui
(+,0) (0,)
uj
Wij ,
W (+,) = ui
W1 = ui
(+,0) (,0) ij
uj
W ,
(,0) (0,) ij
W (,) = ui
uj
W ,
(,0) (0,+)
W (,+) = ui
uj
Wij .
W2 = ui
9 Such harmonic variables were introduced in [25].
(6.8)
235
Contracting equations (6.4), (6.5) with harmonics we find a number of Grassmann analyticity
constraints for these superfields,
D(0,+) W1 = D(0,) W1 = D
(+,0)
W1 = D
(,0)
W1 = 0,
(0,+)
(0,)
D(+,0) W2 = D(,0) W2 = D
W2 = D
W2 = 0,
(+,0)
(0,+) (+,+)
(+,0) (+,+)
(0,+) (+,+)
(+,+)
D W
= D W
= D
W
= D
W
(,0)
(0,)
(,0) (,)
(0,) (,)
(,)
= D W
= D
W
= D
W (,)
D W
(0,) (+,)
D(+,0) W (+,) = D(0,) W (+,) = D (+,0)
W (+,) = D
W
(,0)
(0,+)
(,0) (,+)
(0,+) (,+)
(,+)
D W
= D W
= D
W
= D
W (,+)
= 0,
= 0,
= 0,
= 0.
(6.9)
Moreover, by construction, the superfields (6.8) are annihilated by the following harmonic derivatives,
2

D (++,0) W1 = D (,0) W1 = D (0,++) W1 = D (0,) W1 = D (+,+) W1 = 0,
2

D (++,0) W2 = D (,0) W2 = D (0,++) W2 = D (0,) W2 = D (,) W2 = 0,
D (++,0) W (+,+) = D (0,++) W (+,+)
= D (+,+) W (+,+) = D (+,) W (+,+) = D (,+) W (+,+) = 0,
D (,0) W (,) = D (0,) W (,) = D (,) W (,)
= D (+,) W (,) = D (,+) W (,) = 0,
D (++,0) W (+,) = D (0,) W (+,) = D (+,) W (+,)
= D (+,+) W (+,) = D (,) W (+,) = 0,
D (,0) W (,+) = D (0,++) W (,+)
= D (,+) W (,+) = D (+,+) W (,+) = D (,) W (,+) = 0.
W (+,+) ,
W (,) ,
W (+,) ,
(6.10)
W (,+)
introduced in
We see that the superfield strengths W1 , W2 ,
Sections 5.1.1, 5.1.2 satisfy the same constraints (6.9), (6.10) and therefore give the solutions of
N = 4 SYM constraints (6.4), (6.5).
Let us now consider the reality constraint (6.6). Applying the following identities with harmonics
(0,+) (0,)
ul] ,
(+,0) (,0)
= 2u[k ul] ,
(+,0) (0,)
= 2u[k ul] ,
(0,+) (,0)
= 2u[k ul] ,
(+,0) (0,+)
= 2u[k ul] ,
(,0) (0,)
= 2u[k ul] ,
u(+,0)i u(,0)j ij kl = 2u[k

u(0,+)i u(0,)j ij kl
u(+,0)i u(0,)j ij kl
u(0,+)i u(,0)j ij kl
u(+,0)i u(0,+)j ij kl
u(,0)i u(0,)j ij kl
(6.11)
we find that (6.6) leads to the reality properties of superfield strengths,

W
1,2 = W1,2 ,

(+,+) = W (+,+) ,
W

(+,) = W (,+) ,
W

(,) = W (,) ,
W

(,+) = W (+,) .
W
(6.12)
236
These conjugation rules for superfield strengths coincide with (5.20), (5.31) which were
previously introduced in the superparticle considerations. This establishes the correspondence
between the N = 4 superfield strengths (6.3) and the superfields obtained by the superparticle
quantization.
6.2. Actions in USp(4)/(U (1) U (1)) harmonic superspace
We have seen that the superfields (6.8) satisfying the constraints and equations of motion (6.9),
(6.10) appear naturally in the solution of N = 4 SYM constraints (6.4)(6.6). Now we address a
question, whether it is possible to construct with the help of these superfields any N = 4 invariant
superfield functionals in USp(4)/(U (1) U (1)) harmonic superspace, which can be treated as
the actions of some N = 4 supersymmetric field models. For instance, similar superfields on
N = 4 harmonic superspace with SU(4) harmonic variables were used in the construction of
different integral invariants in [12].
Each of the superfields (6.8) lies in its own analytic subspace parameterized by eight Grassmann variables, as is seen in each line in (6.9). We restrict ourself to two superfields, W (+,+)
and W1 , the others can be studied in a similar way.
Let us consider first W (+,+) , which lies in the analytic subspace with the coordinates
(+,0) (0,+) (+,0) (0,+)
m
,
,
,
, u}, where xAm is given by (4.23). In general, one can consider
{xA ,
the following functional depending on this superfield

d (4,4) F W (+,+) ,
(6.13)
where d (4,4) is the dimensionless analytic measure defined in (C.14). Since such a function
F should have definite U (1) charges, s1 = s2 = 4, it can be only quartic,

4
S4 = g d (4,4) W (+,+) ,
(6.14)
where g is a coupling constant of mass dimension 4. Using the component decomposition
(5.32) for W (+,+) it is easy to see that the action (6.14) contains the following term in components,

S4 g d 4 x F F F F + ,
(6.15)
where F , F are the spinorial components of the Maxwell strength and the dots indicate the
terms for all other component fields. Since the term in the rhs of (6.15) appears in the fourth order
of the decomposition of the BornInfeld action (e.g. see [35,36]), one can treat the action (6.14)
as a part of an N = 4 supersymmetric generalization of BornInfeld theory.
Consider now the chargeless superfield W1 in the analytic superspace with the coordinates
m
, (0,)
, u}, where xAm is given by (5.15). In general, one can write the
{xA , (+,0) , (,0) , (0,+)
following functional with it,

d F(W1 ),
(6.16)
where d is the analytic measure given by (C.19). Since the measure in (6.16) is chargeless, there
are no restrictions on the function F . Of particular interest is the quartic potential,

d (W1 )4 d 4 x F F F F + ,
(6.17)
237
as it may have some relations to the N = 4 supersymmetric generalization of BornInfeld theory

as mentioned above. One can easily check the presence of F 4 term in the component structure
of (6.17) by applying the component decomposition for W1 given by (5.21). Another interesting
example is given by the logarithmic potential,

F4
d ln(W1 /) d 4 x 4 ,
(6.18)
where is some scale which makes the combination W1 / dimensionless. Clearly, the expression in lhs of (6.18) is scale independent despite the manifest presence of dimensionful
parameter. In components, such functional reproduces the first leading term in the low-energy
effective action of N = 4 SYM model (see, e.g., [34]). Note that the N = 4 superfield description given by (6.18) for such terms is even simple than the one with N = 2 superfields [34]. In
the work [37] we noticed that it is much more difficult to write down similar terms within N = 3
harmonic superspace approach.
In this subsection we saw that the superfield strengths (6.8) allowed us to construct some actions with the gauge fields in the fourth power while the N = 4 superfield description for F 2
terms remains unclear. Moreover, the strengths (6.8) are on-shell objects constrained by (6.10).
Therefore the classical action for N = 4 SYM model, if it admits an N = 4 superfield description, should be constructed in terms of some potentials for these superfield strengths. One such
attempt is undertaken in Appendix C, where we introduce the analytic prepotentials and study
some superfield action with it. However, this question requires further independent studies and
lies beyond the frame of the present work.
7. Summary
In this paper we have studied the construction of massive and massless N = 4 superparticle
models in N = 4 harmonic superspace and their quantization. The crucial point of our considerations is the use of USp(4) harmonic variables since exactly this group corresponds to
the R-symmetry of N = 4 superalgebra with central charge. Since the mass of the superparticle should be equal to its central charge, as required for the construction of BPS supermultiplets, both massive and massless superparticles can be studied and quantized in such an N = 4
USp(4)/(U (1) U (1)) harmonic superspace. The quantization leads straightforwardly to the
superfield realization of physically interesting multiplets of N = 4 supersymmetry. Namely,
in the massive case, the N = 4 massive vector multiplet is described by four analytic superfields W (+,+) , W (,) , W (+,) , W (,+) with different types of analyticity obeying also harmonic shortness constraints, which serve as the equations of motion for these superfields. In the
massless case these superfields reduce to the usual N = 4 SYM multiplet if additional reality
constraints are imposed. Moreover, there are also two chargeless superfields W1 and W2 with
specific Grassmann and harmonic shortness constraints describing the same multiplet. All these
six strength superfields are shown to appear naturally in the solution of N = 4 SYM constraints
with the help of USp(4)/(U (1) U (1)) harmonic variables. Apart from the N = 4 SYM multiplet, the superparticle leads to the superfield realizations of N = 4 gravitino multiplet (with
highest helicity 3/2) and N = 4 supergravity multiplet (with highest helicity 2). These multiplets are represented by chiral-analytic and N = 4 chiral superfields in harmonic superspace,
respectively, with appropriate Grassmann and harmonic constraints.
The quantization of the N = 4 harmonic superparticle shows some new possibilities for studying the N = 4 SYM theory directly in N = 4 harmonic superspace with USp(4)/(U (1) U (1))
238
harmonics. The USp(4) group is very suitable for this purpose, particularly because of its invariant antisymmetric 2-tensor, which raises and lowers the R-symmetry indices, similarly as
the -tensor in the SU(2) group. The corresponding USp(4) harmonic superspace possesses a
specific conjugation generalizing the usual complex conjugation, and harmonic projections of
N = 4 superfield strengths appear real under this conjugation. Therefore the USp(4) harmonics
seem to be very useful for studying the N = 4 SYM model in harmonic superspace. Moreover, as
is sketched in the last subsection, it is very straightforward to build some gauge invariant actions
in N = 4 harmonic superspace with USp(4) harmonics. We propose actions which contain F 4
term in the bosonic sector. These actions are written in an analytic superspace with an integration over half of the Grassmann variables of N = 4 superspace, such that the analytic measure is
dimensionless. The F 4 term may be interpreted as the quartic term in an N = 4 BornInfeld action. It is interesting to note that it allows for a very simple scale-invariant generalization which
corresponds to the leading term in the low-energy effective action of the N = 4 SYM model.
However, these issues require deeper investigations.
To conclude, we have constructed the N = 4 massive and massless superparticle models and
developed their quantization. As a result we found superfields in N = 4 harmonic superspace
with USp(4)/(U (1) U (1)) harmonics which describe the basic on-shell N = 4 multiplets.
These superfields allow one to write some N = 4 superfield actions corresponding to N = 4
SYM theory.
It would be very interesting to develop systematically the ideas presented in Appendix C,
where we show that the analytic potentials, which serve as gauge connections for the harmonic
derivatives, can be used for constructing some superfield actions.
Acknowledgements
The authors are grateful to E.A. Ivanov, N.G. Pletnev and D.P. Sorokin for stimulating discussions. The present work is supported particularly by INTAS grant, project No. 05-1000008-7928,
by RFBR grants, projects Nos. 06-02-16346 and 08-02-90490, by DFG grant, project No. 436
RUS/113/669/0-3 and grant for LRSS, project No. 2553.2008.2. I.B.S. acknowledges the support
from INTAS grant, project No. 06-1000016-6108.
Appendix A. Commutation relations in usp(4) algebra
Let us introduce the operators
T ij = k(i uj ) l
,
uk l
(A.1)
where ui j are the harmonic variables (2.8). It well known, that the operators (A.1) obey the
commutation relations of usp(4) algebra (see, e.g., [27]),
ij kl
T , T = i(k T l)j + j (k T l)i .
(A.2)
In the present work we prefer to use the operators (2.16) which differ from (A.1) only by constants and therefore also span the usp(4) algebra. Their commutation relations are given by
(++,0) (,0)
(0,++) (0,)
D
= S1 ,
= S2 ,
,D
D
,D
(+,+) (,)
(+,) (,+)
D
= S 1 + S2 ,
= S2 S1 ,
,D
D
,D

D (+,) , D (+,+) = 2D (++,0) ,
(+,) (++,0)
D
= 0,
,D
(+,) (0,++)
D
= D (+,+) ,
,D
(+,+) (++,0)
D
= 0,
,D
(+,+) (0,++)
D
= 0,
,D
(++,0) (0,++)
D
= 0,
,D
(+,+) (,+)
D
= 2D (0,++) ,
,D
(++,0) (,)
D
= D (+,) ,
,D
(0,++) (,)
D
= D (,+) ,
,D
(,+) (0,)
D
= D (,) ,
,D
(+,) (0,)
D
= 0,
,D
(,) (0,)
D
= 0,
,D

(++,0)
S1 , D
= 2D (++,0) ,

S1 , D (0,++) = 0,

S1 , D (,0) = 2D (,0) ,

S1 , D (0,) = 0,

S1 , D (+,+) = D (+,+) ,

S1 , D (,) = D (,) ,

S1 , D (+,) = D (+,) ,

S1 , D (,+) = D (,+) ,
239

D (+,) , D (,) = 2D (0,) ,
(+,) (,0)
D
= D (,) ,
,D
(,) (,0)
= 0,
D
,D
(+,+) (,0)
D
= D (,+) ,
,D
(+,+) (0,)
D
= D (+,) ,
,D
(++,0) (0,)
D
= 0,
,D
(++,0) (,+)
= D (+,+) ,
D
,D
(0,++) (,+)
= 0,
D
,D
(0,++) (,0)
= 0,
D
,D
(,+) (,)
= 2D (,0) ,
D
,D
(,+) (,0)
D
= 0,
,D
(,0) (0,)
D
= 0,
,D

(++,0)
= 0,
S2 , D

S2 , D (0,++) = 2D (0,++) ,

S2 , D (,0) = 0,

S2 , D (0,) = 2D (0,) ,

S2 , D (+,+) = D (+,+) ,

S2 , D (,) = D (,) ,

S2 , D (+,) = D (+,) ,

S2 , D (,+) = D (,+) ,

[S1 , S2 ] = 0.
(A.3)
Appendix B. Comment on the D D constraint and -symmetries

Here we explain the origin of the quadratic spinor constraints (5.8), (5.56), (5.30), (5.43)
appearing in the massless case. All the considerations are valid for arbitrary N , however only
N = 4 superparticle was studied in this work.
Consider the Lagrangian of massless superparticle (5.2) with the constraints (3.12), (3.14),
which in the massless case read
p 2 0,
Di
= i
m i
+ ipm
0,
D i = i ipm i
(B.1)
0.
(B.2)
The transformations of -symmetry (3.7) in the massless case

i = ipm m i ,
i = ipm i m ,

e = 4 i i + i i ,
(B.3)
240
are generated by

= i [i , .}P i i , . P ,
(B.4)
where [., .}P is the graded Poisson bracket and

i = ipm m D i 0,
i = ipm m D i 0
(B.5)
are the generators of -symmetry which are first-class constraints.

We point out that the transformations (B.3) leave the superparticle Lagrangian invariant
( L1 = 0) for arbitrary local parameters , . In particular, we can take
i = k ij j ,
i = kij ,
j
(B.6)
where k ij = k j i , kij = kj i are new local bosonic parameters. The transformations (B.3) read now
k i = i kij pm m j ,
k i = ik ij pm j m ,

k e = 4 k ij i j + kij i j .
k x m = p m k ij i j + kij i j ,
They are generated by the variation

k = k ij [Kij , .]P + kij K ij , . P ,
(B.7)
(B.8)
where Kij , K ij are new first-class constraints,

Kij = ipm (i m j) 0,
K ij = ipm (i m j ) 0.
(B.9)
Of course, these constraints are not independent, but follow from (B.5) upon contractions with
Grassmann variables. With the use of (B.1), (B.2) the constraints (B.9) can be rewritten as
i j 0,
i j 0,
(B.10)
or as
D ij D i D j 0,
D ij D i Dj 0.
(B.11)
Therefore the constraints (B.11) can also be imposed on the states upon quantization.
Note also that the generators of -symmetries appear from the commutators of the constraints
(B.2) and (B.9),

ij
(i j )
Kij , Dk P = (i jk) ,
(B.12)
K , D k P = k ,

ij
j
)
(i
k
k
D ij , D P = 4(i j ) ,
(B.13)
D , D k P = 4 k .
Therefore one can consider only the constraints (B.2) and (B.11) since the -symmetry ones
(B.3) follow from their algebra.
As a result, there is a bosonic version of -symmetry if we choose the parameters of transformations being proportional to Grassmann coordinates. However, not all the parameters
kij , k ij are independent. Indeed, a symmetric N N matrix has N (N + 1)/2 independent elements that is too much in comparison with the number of first-class superparticle constraints.
However, there is U (N ) R-symmetry which rotates indices i, j of all objects. It is well known,
that any symmetric complex matrix k ij can be brought to the diagonal form by U (N ) rotations [30],
ukuT = diag(d1 , . . . , dN ),
u U (N ).
241
(B.14)
After such a rotation, there are only N independent real parameters in the matrix k.
We point out the importance of accounting the constraints (B.11) in the GuptaBleuler
quantization of a superparticle despite they are not independent but follow from spinorial constraints (B.2) and -symmetries (B.5). Recall that the spinorial constraints (B.2) consist of 2N
first-class and 2N second-class constraints (which cannot be separated explicitly) while the symmetry constraints (B.5) have effectively 2N first-class constraints (since they are infinitely
reducible). Indeed, on the example of a chiral superfield (5.55) it is easy to see that the constraints (5.7) are not sufficient and the constraint (5.56) must be imposed to achieve the correct
component structure (5.57). This is a feature of GuptaBleuler quantization approach in which
we take into account only the constraint D i 0 corresponding to the chiral superfield but not
Di 0. As a result, some of the first class constraints originating from Di 0 remain unaccounted until one has not imposed the constraints (5.56).
Finally, we note that the analogous quadratic spinorial constraints (4.17) also originate from
the -symmetry if one passes to the bosonic parameters as in (B.6).
Appendix C. F 2 term in N = 4 USp(4)/(U (1) U (1)) harmonic superspace
It is well known that the classical actions in N = 2 and N = 3 SYM models can be written in terms of unconstrained superfields (prepotentials) in harmonic superspace [3,5]. A similar
superfield formulation for N = 4 SYM model is very desirable since it may be fruitful for studying quantum aspects of this model. One can hope that the N = 4 harmonic superspace with
USp(4)/(U (1) U (1)) harmonic variables may be useful for this purpose. In particular, here we
propose some functional built of analytic prepotentials which gives F 2 term in components.
In principle, one can try to construct some actions in different subspaces of full N = 4 superspace. However, the superspaces with eight Grassmann variables are more promising since the
corresponding integration measure is dimensionless. Indeed, if one tries to construct an action
in full N = 4 superspace one needs the dimensionful constant since the corresponding measure
is of dimension +4. Therefore we consider the analytic subspaces of N = 4 superspace which
are singled out by the covariant spinor derivatives in different lines in (6.9). Our aim now is to
develop the differential geometry in one of such analytic subspaces and to build possible gauge
invariant superfield functionals.
Consider the analytic subspace in N = 4 harmonic superspace with coordinates
m , (+,0) , (0,+) , (+,0) , (0,+) , u}, where x m is given by (4.23). As is shown in (5.27), the
{xA
A
Grassmann derivatives
D(+,0) ,
D(0,+) ,
(+,0)
D
,
(0,+)
D
(C.1)
single out the analytic superfields. The set of derivatives (C.1) is invariant under the commutators
with the following harmonic derivatives
D (++,0) ,
D (0,++) ,
D (+,+) ,
D (,+) ,
D (+,) .
(C.2)
Therefore we can introduce five analytic gauge connections for the harmonic derivatives (C.2),
D (++,0) (++,0) = D (++,0) + V (++,0) ,
D (0,++) (0,++) = D (0,++) + V (0,++) ,
D (+,+) (+,+) = D (+,+) + V (+,+) ,
242
D (+,) (+,) = D (+,) + V (+,) ,

D (,+) (,+) = D (,+) + V (,+) .
(C.3)
Clearly, these prepotentials contain too much component fields even in the physical sector and we
have to introduce some constraints. One of the possible ways to impose the off-shell constraints
is to set the prepotentials in the last two lines in (C.3) to zero,
V (+,) = 0,
V (,+) = 0.
(C.4)
We point out that it is the constraints (C.4) which do not lead to the equations of motion for the
component fields, but just reduce the number of independent components. Since the derivatives
D (+,) , D (,+) form su(2) subalgebra in (A.3), the constraints (C.4) can be naturally resolved if
one uses the harmonic variables on USp(4)/(SU(2) U (1)) coset rather than USp(4)/(U (1)
U (1)). However, for our considerations it is sufficient to work in the USp(4)/(U (1) U (1))
harmonic superspace with the constrained prepotentials (C.3).
The covariant derivatives (C.3) should satisfy the same algebra (A.3) even with imposed constraints (C.4). It leads to the following constraints for the prepotentials,
D (+,) V (++,0) = 0,
D (,+) V (0,++) = 0,
D (,+) D (,+) V (+,+) = 0,

D (+,) D (+,) V (+,+) = 0,
(+,+)
(,+) (+,)
D
+2 V
= 0.
D
(C.5)
Moreover, the prepotentials V (++,0) , V (0,++) , V (+,+) can be expressed through each other and
only V (+,+) is independent,
D (,+) V (++,0) = V (+,+) ,
D (+,) V (0,++) = V (+,+) ,
D (+,) V (+,+) = 2V (++,0) ,

D (,+) V (+,+) = 2V (0,++) .
(C.6)
There are also the following reality properties for the prepotentials,
(+,+) = V (+,+) ,
V
(++,0) = V (0,++) ,
V
(0,++) = V (++,0) .
V
The analytic prepotentials define the harmonic field strengths,

F (2,2) = (++,0) , (0,++)

= D (++,0) V (0,++) D (0,++) V (++,0) + V (++,0) , V (0,++) ,

F (3,1) = (++,0) , (+,+)

= D (++,0) V (+,+) D (+,+) V (++,0) + V (0,++) , V (+,+) ,

F (1,3) = (+,+) , (0,++)

= D (+,+) V (0,++) D (0,++) V (+,+) + V (+,+) , V (0,++) ,
which are analytic and gauge covariant (or invariant in the Abelian case),

F (2,2) = , F (2,2) ,
F (3,1) = , F (3,1) ,
F (1,3) = , F (1,3)
(C.7)
(C.8)
(C.9)
under the following gauge transformations of the prepotentials

V (++,0) = (++,0) ,
V (+,+) = (+,+) .
V (0,++) = (0,++) ,
(C.10)
243
Here is a real analytic gauge parameter constrained by

D (+,) = D (,+) = 0.
(C.11)
Owing to the constraints (C.5) these strengths are related to each other,
D (+,) F (2,2) = F (3,1) ,
D (,+) F (2,2) = F (1,3) ,
D (+,) F (1,3) = D (,+) F (3,1) = 2F (2,2) .
(C.12)
We stress that the constraints (C.11) are purely algebraical and mean that depends on
USp(4)/(SU(2) U (1)) harmonics. One can avoid all these constraints by using the superfields
in USp(4)/(SU(2) U (1)) harmonic superspace.
Now we apply the strength superfields (C.8) to build a gauge invariant action in analytic
subspace,

1
(4,4) (2,2) (2,2)
S2 = tr d
(C.13)
F
F
= tr d (4,4) F (3,1) F (1,3) ,
2
where d (4,4) is the analytic measure,
2
2
2
2

1 4
d xA du D (,0) D (0,) D (,0) D (0,) .
8
2
The integration over harmonic variables is defined by the following rules

du 1 = 1,

du f (s1 ,s2 ) = 0, if s1 = 0, s2 = 0,

du (irreducible harmonic tensor) = 0,
d (4,4) =
(C.14)
(C.15)
where f (s1 ,s2 ) is some function of USp(4)/(U (1) U (1)) harmonic variables. The rigorous
grounds for such rules of harmonic integrals are given in the book [6] for the case of SU(2)/U (1)
harmonic variables and in [5,38] for SU(3)/(U (1) U (1)) harmonics. Here we just generalize
these constructions for the USp(4)/(U (1) U (1)) coset.
The action (C.13) is supersymmetric and gauge invariant by construction and contain the component fields with the spins (helicity) not higher than one. In particular, it is easy to find the vector
field in its component structure by considering the following term in the prepotential V (+,+) ,
V (+,+) =

i (+,0) m (0,+)

+ (0,+) m (+,0) Am + .
(C.16)
The real vector field Am turns into the Maxwell strength in the F (2,2) superfield (in the Abelian
case),

F (2,2) = (+,0) m (+,0) (0,+) n (0,+) Fmn +
(C.17)
and leads to the Maxwell term in the action (C.13),

1
d 4 x F mn Fmn + .
S2 =
(C.18)
4
The dots stand here for the other component terms in the action. As a result we see that the action
contains N = 4 vector multiplet in component decomposition and may have some relation to the
244
N = 4 supergauge theory. However, one can check that the action (C.13) contains much more
propagating degrees of freedom than a single N = 4 SYM multiplet. Therefore to make the
action (C.13) physical one needs more superfield constraints for the prepotentials and this issue
requires further studies.
Another way in the seek of unconstrained N = 4 SYM action in the USp(4) harmonic superspace is the use of other analytic subspaces in N = 4 superspace. Of particular interest may by
(+,0) (,0) (0,+) (0,)
,
,
,
, u}, where xAm is given
the analytic subspace with coordinates {xAm ,
by (5.15). The corresponding analytic measure is chargeless,
d =
(+,0) 2 (,0) 2 (0,+) 2 0, 2

1 4
D
D
D
d
x
du
D
.
A
28
(C.19)
One can use other prepotentials and field strengths in this subspace for constructing the invariant
actions. We leave these questions for further studies.
References
[1] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity, Phys.
Rep. 323 (2000) 183, hep-th/9905111.
[2] S.J. Gates, M.T. Grisaru, M. Rocek, W. Siegel, Superspace: Or One Thousand and One Lessons in Supersymmetry,
Benjamin/Cummings, 1983, 548 p.
[3] A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic superspace: Key to N = 2 supersymmetric theories,
JETP Lett. 40 (1984) 912;
A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, Unconstrained N = 2 matter, YangMills and
supergravity theories in harmonic superspace, Class. Quantum Grav. 1 (1984) 469.
[4] A. Galperin, E.A. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic supergraphs. Green functions, Class. Quantum
Grav. 2 (1985) 601;
A. Galperin, E.A. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic supergraphs. Feynman rules and examples, Class.
Quantum Grav. 2 (1985) 617.
[5] A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, N = 3 Supersymmetric gauge theory, Phys. Lett.
B 151 (1985) 215;
A. Galperin, E. Ivanov, S. Kalitzin, V. Ogievetsky, E. Sokatchev, Unconstrained off-shell N = 3 supersymmetric
YangMills theory, Class. Quantum Grav. 2 (1985) 155.
[6] A. Galperin, E. Ivanov, V. Ogievetsky, E. Sokatchev, Harmonic Superspace, Cambridge Univ. Press, UK, 2001,
306 p.
[7] E. Ahmed, S. Bedding, C.T. Card, M. Dumbrell, M. Nouri-Moghadam, J.G. Taylor, On N = 4 supersymmetric
YangMills in harmonic superspace, J. Phys. A 18 (1985) 2095.
[8] I.A. Bandos, Solution of linear equations in spaces of harmonic variables, Theor. Math. Phys. 76 (1988) 783, Teor.
Mat. Fiz. 76 (1988) 169.
[9] G.G. Hartwell, P.S. Howe, (N, p, q) harmonic superspace, Int. J. Mod. Phys. A 10 (1995) 3901, hep-th/9412147;
P. Heslop, P.S. Howe, On harmonic superspaces and superconformal fields in four dimensions, Class. Quantum
Grav. 17 (2000) 3743, hep-th/0005135.
[10] L. Andrianopoli, S. Ferrara, E. Sokatchev, B. Zupnik, Shortening of primary operators in N extended SCFT(4) and
harmonic superspace analyticity, Adv. Theor. Math. Phys. 3 (1999) 1149, hep-th/9912007.
[11] S. Ferrara, E. Sokatchev, Short representations of SU(2, 2/N ) and harmonic superspace analyticity, Lett. Math.
Phys. 52 (2000) 247, hep-th/9912168.
[12] J.M. Drummond, P.J. Heslop, P.S. Howe, S.F. Kerstan, Integral invariants in N = 4 SYM and the effective action
for coincident D-branes, JHEP 0308 (2003) 016, hep-th/0305202.
[13] D.P. Sorokin, Superbranes and superembeddings, Phys. Rep. 329 (2000) 1, hep-th/9906142.
[14] R. Casalbuoni, The classical mechanics for BoseFermi systems, Nuovo Cimento A 33 (1976) 389.
[15] A.I. Pashnev, D.V. Volkov, Supersymmetric Lagrangian for particles in proper time, Theor. Math. Phys. 44 (1980)
770, Teor. Mat. Fiz. 44 (1980) 321.
[16] L. Brink, J.H. Schwarz, Quantum superspace, Phys. Lett. B 100 (1981) 310.
245
[17] J.A. de Azcrraga, J. Lukierski, Supersymmetric particles with internal symmetries and central charges, Phys. Lett.
B 113 (1982) 170;
A. Frydryszak, J. Lukierski, N = 2 massive matter multiplet from quantization of extended classical mechanics,
Phys. Lett. B 117 (1982) 51;
J.A. de Azcrraga, J. Lukierski, Supersymmetric particles in N = 2 superspace: Phase-space variables and Hamiltonian dynamics, Phys. Rev. D 28 (1983) 1337.
[18] V.P. Akulov, D.P. Sorokin, I.A. Bandos, Particle mechanics in harmonic superspace, Mod. Phys. Lett. A 3 (1988)
1633.
[19] V.P. Akulov, I.A. Bandos, D.P. Sorokin, Particle in harmonic N = 2 superspace, Sov. J. Nucl. Phys. 47 (1988) 724,
Yad. Fiz. 47 (1988) 11361146.
[20] D.P. Sorokin, V.I. Tkach, D.V. Volkov, Superparticles, twistors and Siegel symmetry, Mod. Phys. Lett. A 4 (1989)
901.
[21] L. Lusanna, B. Milewski, N = 2 super-YangMills and supergravity constraints from coupling to a supersymmetric
particle, Nucl. Phys. B 247 (1984) 396;
J.A. Shapiro, C.C. Taylor, Superspace supergravity from the superstring, Phys. Lett. B 186 (1987) 69.
[22] I.L. Buchbinder, I.B. Samsonov, N = 3 Superparticle model, arXiv: 0801.4907 [hep-th].
[23] P. Fayet, Spontaneous generation of massive multiplets and central charges in extended supersymmetric theories,
Nucl. Phys. B 149 (1979) 137.
[24] S. Ferrara, C.A. Savoy, B. Zumino, General massive multiplets in extended supersymmetry, Phys. Lett. B 100
(1981) 393.
[25] E. Ivanov, S. Kalitzin, N. Ai Viet, V. Ogievetsky, Harmonic superspaces and extended supersymmetry: The calculus
of harmonic variables, J. Phys. A 18 (1985) 3433.
[26] E. Sokatchev, An action for N = 4 supersymmetric selfdual YangMills theory, Phys. Rev. D 53 (1996) 2062,
hep-th/9509099.
[27] S. Ferrara, E. Sokatchev, Superconformal interpretation of BPS states in AdS geometries, Int. J. Theor. Phys. 40
(2001) 935, hep-th/0005151;
S. Ferrara, E. Sokatchev, Conformal superfields and BPS states in AdS(4/7) geometries, Int. J. Mod. Phys. B 14
(2000) 2315, hep-th/0007058;
S. Ferrara, E. Sokatchev, Representations of (1, 0) and (2, 0) superconformal algebras in six-dimensions: Massless
and short superfields, Lett. Math. Phys. 51 (2000) 55, hep-th/0001178;
S. Ferrara, E. Sokatchev, Universal properties of superconformal OPEs for 1/2 BPS operators in 3 D 6, New
J. Phys. 4 (2002) 2, hep-th/0110174.
[28] B.M. Zupnik, ChernSimons D = 3, N = 6 superfield theory, Phys. Lett. B 660 (2008) 254, arXiv: 0711.4680
[hep-th];
B.M. Zupnik, ChernSimons theory in SO(5)/U (2) harmonic superspace, arXiv: 0802.0801 [hep-th].
[29] P.J. Heslop, P.S. Howe, A Note on composite operators in N = 4 SYM, Phys. Lett. B 516 (2001) 367, hep-th/
0106238;
P.J. Heslop, P.S. Howe, OPEs and three-point correlators of protected operators in N = 4 SYM, Nucl. Phys. B 626
(2002) 265, hep-th/0107212;
B. Eden, E. Sokatchev, On the OPE of 1/2 BPS short operators in N = 4 SCFT(4), Nucl. Phys. B 618 (2001) 259,
hep-th/0106249;
G. Arutyunov, F.A. Dolan, H. Osborn, E. Sokatchev, Correlation functions and massive KaluzaKlein modes in the
AdS/CFT correspondence, Nucl. Phys. B 665 (2003) 273, hep-th/0212116;
E. DHoker, P. Heslop, P. Howe, A.V. Ryzhov, Systematics of quarter BPS operators in N = 4 SYM, JHEP 0304
(2003) 038, hep-th/0301104.
[30] B. Zumino, Normal forms of complex matrices, J. Math. Phys. 3 (1962) 1055.
[31] E. Sokatchev, Light cone harmonic superspace and its applications, Phys. Lett. B 169 (1986) 209;
E. Sokatchev, Harmonic superparticle, Class. Quantum Grav. 4 (1987) 237;
F. Delduc, A. Galperin, E. Sokatchev, Lorentz harmonic (super)fields and (super)particles, Nucl. Phys. B 368 (1992)
143.
[32] I.A. Bandos, Superparticle in Lorentz harmonic superspace, Sov. J. Nucl. Phys. 51 (1990) 906, Yad. Fiz. 51 (1990)
1429;
I.A. Bandos, Spinor moving frame, M0-brane covariant BRST quantization and intrinsic complexity of the pure
spinor approach, Phys. Lett. B 659 (2008) 388, arXiv: 0707.2336 [hep-th];
I.A. Bandos, D = 11 massless superparticle covariant quantization, pure spinor BRST charge and hidden symmetries, Nucl. Phys. B 796 (2008) 360, arXiv: 0710.4342 [hep-th].
246
[33] R. Grimm, M. Sohnius, J. Wess, Extended supersymmetry and gauge theories, Nucl. Phys. B 133 (1978) 275;
M.F. Sohnius, Bianchi identities for supersymmetric gauge theories, Nucl. Phys. B 136 (1978) 461;
M.F. Sohnius, Supersymmetry and central charges, Nucl. Phys. B 138 (1978) 109;
P. Howe, K.S. Stelle, P.K. Townsend, Supercurrents, Nucl. Phys. B 192 (1981) 332;
P. Howe, K.S. Stelle, P.K. Townsend, The relaxed hypermultiplet: An unconstrained N = 2 superfield theory, Nucl.
Phys. B 214 (1983) 519.
[34] I.L. Buchbinder, S.M. Kuzenko, B.A. Ovrut, On the D = 4, N = 2 nonrenormalization theorem, Phys. Lett. B 433
(1998) 335, hep-th/9710142;
I.L. Buchbinder, E.I. Buchbinder, S.M. Kuzenko, Non-holomorphic effective potential in N = 4 SU(n) SYM, Phys.
Lett. B 446 (1999) 216, hep-th/9810239;
E.I. Buchbinder, I.L. Buchbinder, E.A. Ivanov, S.M. Kuzenko, B.A. Ovrut, Low-energy effective action in N = 2
supersymmetric field theories, Phys. Part. Nucl. 32 (2001) 641;
I.L. Buchbinder, E.A. Ivanov, Complete N = 4 structure of low-energy effective action in super-YangMills theories, Phys. Lett. B 524 (2002) 208, hep-th/0111062;
I.L. Buchbinder, E.A. Ivanov, A.Yu. Petrov, Complete low-energy effective action in N = 4 SYM: A direct N = 2
supergraph calculation, Nucl. Phys. B 653 (2003) 64, hep-th/0210241;
I.L. Buchbinder, N.G. Pletnev, Hypermultiplet dependence of one-loop low-energy effective action in the N = 2
superconformal theories, JHEP 0704 (2007) 096, hep-th/0611145.
[35] A.A. Tseytlin, BornInfeld action, supersymmetry and string theory, in: M.A. Shifman (Ed.), The Many Faces of
the Superworld, pp. 417452, hep-th/9908105.
[36] E.A. Ivanov, B.M. Zupnik, N = 3 supersymmetric BornInfeld theory, Nucl. Phys. B 618 (2001) 3, hep-th/0110074.
[37] I.L. Buchbinder, E.A. Ivanov, I.B. Samsonov, B.M. Zupnik, Scale invariant low-energy effective action in N = 3
SYM theory, Nucl. Phys. B 689 (2004) 91, hep-th/0403053.
[38] F. Delduc, J. McCabe, The quantization of N = 3 super-YangMills off-shell in harmonic superspace, Class. Quantum Grav. 6 (1989) 233.

General CP violation in minimal leftright symmetric

model and constraints on the right-handed scale
Yue Zhang a,b, , Haipeng An b , Xiangdong Ji b,a , Rabindra N. Mohapatra b
a Center for High-Energy Physics and Institute of Theoretical Physics, Peking University, Beijing 100871, China
b Maryland Center for Fundamental Physics, Department of Physics, University of Maryland,
College Park, MD 20742, USA

Received 27 December 2007; received in revised form 13 April 2008; accepted 21 May 2008
Abstract
In minimal leftright symmetric theories, the requirement of parity invariance allows only one complex
phase in the Higgs potential and one in the Yukawa couplings, leading to a two-phase theory with both
spontaneous and explicit CP violations. We present a systematic way to solve the right-handed quark mixing
matrix analytically in this model and find that the leading order solution has the same hierarchical structure
as the left-handed CKM matrix with one more CP-violating phase coming from the complex Higgs vev.
Armed with this explicit right-handed mixing matrix, we explore its implications for flavor changing and
conserving processes in detail, low-energy CP-violating observables in particular. We report an improved
lower bound on the WR mass of 2.5 TeV from MK and MB , and a somewhat higher bound (4 TeV) from
kaon decay parameters , , and neutron electric dipole moment. The new bound on the flavor-changing
neutral Higgs mass is 25 TeV.
1. Introduction
The physics beyond the Standard Model (SM) has been the central focus of high-energy
phenomenology for more than three decades. Many proposals, including supersymmetry, technicolor, little Higgs, and extra dimensions, have been made and studied thoroughly in the literature,
tests are soon to be made at the Large Hadron Collider (LHC). One of the earliest proposals, the
* Corresponding author at: Center for High-Energy Physics and Institute of Theoretical Physics, Peking University,
Beijing 100871, China.
E-mail address: yzhang@physics.umd.edu (Y. Zhang).
doi:10.1016/j.nuclphysb.2008.05.019
248
Y. Zhang et al. / Nuclear Physics B 802 (2008) 247279
leftright symmetric model (LRSM), was motivated by the hypothesis that parity is a perfect
symmetry at high-energy, and is broken spontaneously at low-energy due to the asymmetric vacuum [1]. Asymptotic restoration of parity has a definite aesthetic appeal [2]. The model has a
number of additional attractive features, including a natural explanation of weak hyper-charge
in terms of baryon and lepton numbers, existence of right-handed neutrinos and entailed seesaw
mechanism for neutrino masses, possibility of spontaneous CP (charge-conjugation-parity) violation, and natural solution for the strong CP problem. The model can be constrained strongly by
low-energy physics and predicts clear signatures at colliders. It so far remains a decent possibility
for new physics.
The LRSM is best constrained at low-energy by flavor-violating mixing and decays, particularly CP violating observables. In making theoretical predictions, the major uncertainty comes
from the unknown right-handed quark mixing matrix, similar in spirit to that of the left-handed
quark CabibboKobayashiMaskawa (CKM) mixing. The new mixing is a unitary matrix, depending on 9 real parameters: 6 CP violation phases and 3 rotational angles. All are physical
after the left-handed CKM mixing is rotated into a standard 4-parameter form.
Historically, two special CP violation scenarios in LRSM have been considered. The first one,
the manifest leftright symmetry, assumes that there is no spontaneous CP violation i.e. all
Higgs vacuum expectation values (vevs) are real. Then the quark mass matrices are hermitian,
and the left- and right-handed quark mixings become identical, modulo the sign uncertainty of the
elements from negative quark masses. The second scenario, pseudomanifest leftright symmetry, assumes that the CP violation comes entirely from spontaneous symmetry breaking (SSB)
of the vacuum and all Yukawa couplings are real [3]. Here the quark mass matrices are complex
and symmetric, implying that the right-handed quark mixing is related to the complex conjugate
of the CKM matrix multiplied by additional CP phases. There are few studies of the model with
general CP violations in the literature [4,5]. It has been noted that there are problems with both
manifest and pseudo-manifest scenarios. The manifest LRSM with real potential and vevs always provides more minimization conditions than the number of vevs, thus has a fine-tuning
problem. In any case, in the non-supersymmetric LRSM, the assumption of all parameters of the
model being real is not technically natural since loop corrections lead to one of the parameters
being complex. On the other hand, the pseudo-manifest LRSM, where exact CP is assumed before symmetry breaking and all couplings in the Higgs potential are real, leads in the decoupling
limit to either a model with light triplet Higgs, which is already excluded by SM precision test,
or a two Higgs doublet model, which is excluded by experiment due to large tree level flavorchanging neutral current [6]. Therefore, neither of the two special scenarios can be realistic.
In a recent paper [7], we reported a systematic approach to analytically solving the righthanded quark mixing in the minimal LRSM where only the requirement of parity invariance
is imposed prior to symmetry breaking, leaving automatically only one CP phase in the Higgs
potential and one in the Yukawa couplings and leading to a theory with both explicit and spontaneous CP violations. This model therefore falls in-between the above two extreme cases and
is free of the problems described above. Our approach is based on the observation that in the
absence of any fine tuning, mt mb implies that the ratio of the two vevs of the Higgs bidoublet, = /, is small and is of the order of mb /mt . In the leading-order in , we find
a linear equation for the right-handed quark mixing matrix which can be readily solved. We
present an analytical solution of this equation valid to O(3 ), where = sin C is the Cabibbo
mixing parameter. The leading right-handed quark mixing is nearly the same as the left-handed
CKM matrix, except for additional phases which are fixed by , spontaneous CP phase , and
the quark masses.
249
Our work is similar in spirit to the detailed numerical study of the general CP LRSM made
by Kiers et al. [5], with the two vevs ratio / fixed exactly to mb /mt . It is interesting that the
gross feature of the right-handed CKM was obtained already in the appendix of that paper, in
particular the hierarchical structure of flavor mixing and the magnitude of the Dirac CP phase.
However, realistic studies were made numerically for lack of an explicit solution with known
precision. In fact, much of the numerical work of Ref. [5] goes into solving the right-handed
CKM in the presence of 11 input parameters, which must be scanned through using Monte Carlo
to obtain the physical quark masses and the left-handed CKM mixing. Because the extensive
nature of numerical study, it is difficult to see some of the physics in a clear way, in particular,
the interplay between the explicit and spontaneous CP violations in physical observables. Our
explicit analytic solution for the right-handed CKM makes extensive analytical studies simple
and straightforward.
In this paper, we first give the detailed method obtaining the analytical solution to righthanded quark mixing [7]. With this explicit right-handed mixing, we study the neutral kaon and
B-meson systems and the neutron electric dipole moment (EDM) to obtain the lower bound
on right-handed W -boson mass scale. The neutral kaon mass mixing provides a rigorous lower
bound MWR > 2.5 TeV, with the use of the new lattice QCD calculations of the four-quark matrix elements and the strange quark mass. The indirect CP violation receives large contributions
from both explicit and spontaneous CP phases, and from both the gauge-boson box diagram and
flavor-changing neutral Higgs (FCNH). There are strong cancellations among all the contributions, which in turn constrain severely the relation among the spontaneous CP phase, MWR , and
FCNH mass MH . We use the cancellation condition to fix the spontaneous CP phase, which is
then used to predict the neutron EDM in terms of the model parameters. Using the experimental
bound on the EDM, we obtain a strong lower bound on MWR , which can be improved with better
calculations of the hadronic matrix elements and more precise experimental data. Furthermore,
we study implications of direct CP violation in K 0 and B 0 decays. In the former case, a strong
lower bound on MWR is obtained under the factorization assumption for the four-quark matrix
elements. The CP violating observables in the kaon and B systems and the neutron EDM provide
competitive or even stronger bounds than the well-known kaon mass mixing. We also present a
detailed study of the Higgs sector in the presence of the spontaneous CP phase, including the
mass spectrum, the neutral and charged couplings to the quarks, and the bound on the FCNH
mass. One of the main motivations of this work is the possibility that a TeV mass WR can be
searched for at the LHC.
In the LRSM, there is a natural explanation for the small neutrino masses discovered in recent
year, which arise from both type I [8] and type II seesaw [9] mechanism. As far as the type I
contribution goes, parity scale in the few TeV range requires a DiracYukawa coupling of order
105.5 to give neutrino masses of the right order to fit the atmospheric as well as solar neutrino
observations. To make the type II contribution that arises from the left triplet vev can a priori
be large for TeV scale parity violation. However, if we set the parameters i = 0 in the Higgs
potential (see Eq. (30) below), that leads to vL = 0 which in turn makes the type II seesaw
contribution to neutrino masses zero. We do not discuss the detailed fits to neutrino masses and
mixings here but it is clear that appropriate choice of the fermion Yukawa couplings of the model.
The paper is organized as follows. We first briefly introduce the minimal LRSM in Section 2.
In Section 3, we report our method in solving for the right-handed CKM matrix in the scenario of
generic CP violation, including both spontaneous and explicit phases as is the case with only parity invariance. In Section 4, we study the Higgs sector, including the mass spectrum and Higgs
couplings to quarks. In Section 5, we explore the well-known neutral kaon mass difference to
250
find out updated constraints on the right-handed W mass and the FCNH mass. We also consider
similar constraints from the neutral B-meson system. The CP violations in various processes
are discussed in Section 6, including , , neutron EDM and CP asymmetry in B J /KS ,
to constrain the mass of WR and the spontaneous CP phase . We find that they place consistently strong lower bounds on MWR and MH . We conclude the paper in Section 7, with a table
summarizing various constraint on the right-handed scales.
2. The minimal leftright symmetric model
The minimal LRSM is based on the gauge group SU(2)L SU(2)R U (1)BL . Parity is
assumed to be a good symmetry at the Lagrangian density level, and is broken spontaneously by
vevs of Higgs fields. The electric charge formula can be written as a generalized Gell-Mann
Nishijima formula including the third component of the right-handed isospin, T3L and T3R , and
the difference between baryon and lepton numbers [10],
Q = T3L + T3R + (B L)/2.
(1)
This in turn gives an explicit explanation for the standard model U (1)-hypercharge in terms of
physical quantum numbers rather than an arbitrarily adjustable quantum number Y .
In the matter sector, the left-handed fermions form fundamental representations of the SU(2)L
gauge group, while right-handed ones form the representations of the SU(2)R gauge group. The
right-handed neutrinos are introduced automatically so that the right-handed leptons also form
doublets under SU(2)R

1
1
uL
uR
QL =
,
QR =
,
2, 1,
1, 2,
dL
dR
3
3

LR = R (1, 2, 1),
LL = L (2, 1, 1),
lL
lR
where the quantum numbers are those of the above gauge groups. From these, we can easily
write down the gauge-coupled Lagrangian density for fermions,

gL
g
WL B QLi
L fermion = Q Li i
2
6

gR
g
+ Q Ri i
WR B QRi
2
6

gL
g
+ LLi i
WL + B LLi
2
2

gR
g
+ LRi i
WR + B LRi ,
2
2
where the index i = 1, 2, 3 labels fermion generation with all the fields being in flavor eigenstates.
R,L and B are the gauge fields associated with the above gauge groups, with corresponding
W
couplings gL = gR and g . are Pauli matrices for isospins. The right-handed currents couple to
the gauge bosons WR in a way symmetric to the left-handed counterparts.
The Higgs sector contains a bidoublet , belonging to the (2, 2, 0) representation of the gauge
group, which is the leftright symmetric version of the SM Higgs doublet and two triplets L,R
belonging to (3, 1, 2) and (1, 3, 2), respectively,

+
0
L++
L / 2
1 2+
,
,
=
=
L
1 20
L0
L+ / 2

+
R++
R / 2
R =
.
R0
R+ / 2
251
(2)
The Higgs bosons kinetic energy and coupling to the gauge fields are canonical,

L Higgs kin = Tr (D L ) D L + Tr (D R ) D R + Tr (D ) D , (3)
where the covariant derivatives are
gL
gR
WR ,
D = + i W
L i
2
2
g(L,R)
D (L,R) = (L,R) + i
[W(L,R) , (L,R) ] + ig B (L,R) .
2
Again, leftright symmetry is explicit.
Like the SM, we use the vevs of neutral Higgs fields to break the gauge symmetry. The
symmetry group SU(2)L SU(2)R U (1)BL is first broken to SU(2)L U (1)Y by the vev
R
= vR at TeV or multi-TeV scale. Then electroweak symmetry breaking is induced by the
vevs of . With a generic Higgs potential, all the neutral components of the Higgs get vevs,
i

e 1
0

=
,
0
ei2

0
0
0
0
L
=
(4)
,

,
=
R
vL ei1 0
vR ei2 0
where there are 4 neutral complex components and so generically 4 phases. At first it appears
that 3 of them can be eliminated through gauge symmetry because we have 3 generators, T3L ,
T3R , and B L, commuting with the electromagnetic charge operator Q. In reality, however, we
can eliminate only 2. If the transformation parameters associated with the above 3 operators are
L , R and BL , the Higgs fields transform as,

eiT3L L
eiT3R R ,
L
eiT3L L L
eiT3L L eiBL ,
R
eiT3R R R
eiT3R R eiBL .
(5)
This leads a transformation of the vev phases,

1
1
1
1
2 2 L + R ,
1 1 + L R ,
2
2
2
2
1,2 1,2 L,R + BL .
(6)
It is clear that there are two independent combinations of transformation parameters, allowing
removing two phases only. Conventionally the phases of and vR are set to zero, and thus the
general form of Higgs vevs is simplified to

0

=
,
0 ei
252

L
=
0
vL eiL

0
,
0

R
=
0
vR

0
.
0
(7)
If one chooses vL = 0, the only remaining phase is physically relevant.

Since the bidoublet transforms non-trivially under both SU(2)L and SU(2)R , the gauge
bosons WL and WR are not the mass eigenstates after SSB. From (3) we get the W -boson mass
matrix,
1 2 2
2
2
2
i W +
L
2 g ( + + 2vL ) g e
L W-Mass = ( WL
(8)
WR )
+ ,
2
2

i
2
g e
g vR
WR
1
2 )/ 2, and g = g = g by parity. Since v is at most on the
= (WL,R
iWL,R
where WL,R
R
L
L
2

2
order of the left-handed neutrino masses and + on the SM scale, vL 2 + 2 , and
we can neglect vL2 in Eq. (8). The mass matrix can be diagonalized by a unitary transformation
+
+
W1
WL
cos
sin ei
(9)
,
+ = sin ei
cos
WR
W2+
where W1+ and W2+ are mass eigenstates (1 and 2 subscripts shall not be confused with the
Cartesian components of the isospins), with masses MW1 MWL = g 2 + 2 / 2 and MW2
MWR = gvR . The parameters and are related to the CP phase and the masses of W1 , W2 ,
and

MWL 2
= ,
tan = 2 2
,
(10)
MWR
vR
where again = /. If there is no cancellation in generating quark masses, has a natural
size mb /mt , and thus is suppressed by both (MWL /MWR )2 and mb /mt and is smaller than
4 105 , which is much smaller than the current experimental bound 102 . Even so this tiny
mixing will be the dominating contribution to the neutron EDM, as we will discuss below. In
terms of the mass eigenstates, the charged current couplings in the quark sector are

gL
+
+
L W -current = u Li cos W1
dLi
sin ei W2
2

gR
+
+
dRi + h.c.
u Ri sin ei W1
+ cos W2
(11)
2
The above expression is in quark flavor basis. In the next section, we will consider the quark
mass basis in which it will be modified by the CKM mixing matrices.
3. Right-handed quark mixing matrix
In this section, we will focus on the quark sector, solving for the right-handed quark mixing
matrix compatible with the observed quark masses and the left-handed mixing. The solution is
valid in the general CP violation scenario in which both explicit and spontaneous CP breakings
are allowed. The only significant assumption we will make is that bidoublet Higgs vevs and
related Yukawa couplings have a hierarchical structure and there is no large cancellation in generating quark masses, and the solution can be made in a systematic expansion of the relevant
small parameter.
The most general Yukawa coupling of the quark fields with the Higgs bidoublet is given by
Li (hij + h ij )Q
Rj + h.c.,
LY = Q
(12)
253
where = i2 i2 and flavor indices i, j = 1, 2, 3. Parity symmetry, under which

and QLi QRi , constrains h and h be hermitian matrices. After SSB, the above Lagrangian
density yields the following up- and down-type quark mass matrices,
MU = h + ei h,
MD = ei h + h.
(13)
There are two terms in each, and we assume that there is no fine-tuned cancellation to generate
a quark mass scale. Since the top quark mass is much larger than that of the bottom quark, the
and should not be on the same order. Without loss of
assumption implies that h and h,
generality, we take and h h. To leading order in /, we have
MU h,
MD = ei h + h.
(14)
The two terms in the down-type quark masses can be on the same order, however. Because of
the flavor independence of the gauge coupling and the hermiticity of h, we can work in the basis
where MU is diagonal,

mu
MU = S U
(15)
SU M U ,
mc
mt
in which SU = diag{su , sc , st } is the sign of the up-type quark masses. It is present because the
eigenvalues of a hermitian matrix can either be positive or negative, and by convention we take
all mi positive. In this basis, MD is not diagonal and is related to M D = diag{md , ms , mb } via
left- and right-handed CKM rotations. Since the phase factor ei in MD is generically non-zero,
VLCKM = VRCKM
MD = VLCKM M D VRCKM SU .
(16)
For simplicity, we will omit the superscript CKM henceforth. From Eqs. (14), (15) and (16), we
have

(17)
SU M U ei .
Two comments are in order. First, through the phase transformations that are chirally independent
i
i , one can bring V to a standard
but isospin-dependent, uiL,R eii uiL,R and dL,R
eii dL,R
L
form with only 4 parameters (3 rotations and 1 CP violation phase) and the above equation
remains the same. The h matrix, however, will be subjected to a unitary transformation and
remains hermitian. Second, after the transformation, all parameters in the unitary matrix VR
must be physical, including 3 rotations and 6 CP-violating phases.
which leads to the followTo make further progress, one uses the hermiticity condition for h,
ing equation,
h = VL M D VR SU
M D VR VR M D = 2i sin VL M U SU VL ,
(18)
where VR is the quotient between the left and right mixing VR = SU VL VR . There are a total
of 9 equations in the above expression, which is just enough to solve 9 parameters in VR . It is
interesting to note that if there is no spontaneous CP violation, = 0, we recover the solution
254
VR = SU VL SD , in which SD = diag{sd , ss , sb } is the sign matrix for down-type quark masses.

This is just the manifest LRS case.
The above linear equations can be readily solved utilizing the hierarchy between down-type
quark masses. We find an analytical expression for VR up to order of O(3 ), where is sine of
Cabibbo angle. We begin directly from (18) and the left side is anti-hermitian, and we can write
it explicitly
2imd Im VR11
ms VR12
mb VR13
ms VR12
2ims Im VR22
mb VR23 ,
(19)
mb V
mb V
2imb Im VR33
R13
R23
where we have used md ms mb . The right-hand side of Eq. (18) depends on the physical
quark masses, the standard CKM matrix, and the spontaneous CP violation parameter sin .
Thus we can solve VR in terms of these up to O(3 )

mt 2 4
mb mc 2
sc + st
A (1 )2 + 2 ,
Im VR11 = r sin
(20)
md mt
mc

mt 2 4
mb mc
A ,
sc + st
Im VR22 = r sin
(21)
ms mt
mc
Im VR33 = r sin st ,
(22)

mt 4 2
mb mc
VR12 = 2ir sin
(23)
sc + st
A (1 + i) ,
ms mt
mc
VR13 = 2ir sin A3 (1 + i)st ,
(24)
VR23 = 2ir sin A2 st ,
(25)
where r (mt /mb ) , and , A, , and are Wolfenstein parameters for VL . The above solution exists only when |r sin | 1, which is an interesting and unexpected constraint. Since
the natural size of is mb /mt , r 1, allowing angle 1. Given the physical values of various parameters, we find the following power counting: Im VR11 , Im VR22 , VR12 2 ,
VR13 3 and VR23 2 . Using unitarity condition for VR , we can solve all other elements.
Defining new phases sin i = SDii Im VRii , where i = 1, 2, 3, we have up to O(3 ),
VRii = SDii eii ,
ei(1 +2 ) ,
VR21 = sd ss VR12
ei(1 +3 ) ,
VR31 = sd sb VR13
VR32 = ss sb VR23
ei(2 +3 ) .
(26)
Therefore, we can write the right-handed CKM matrix in a more compact form
VR = PU VL PD ,
in which PU = diag(su , sc e2i2 , st e2i3 ), PD = diag(sd ei1 , ss ei2 , sb ei3 ), and
1 2 /2
A3 ( i)
VL =
1 2 /2
A2 e2i2 ,
3
2
2i
2
A (1 i) A e
1
which differs from VL by a small phase in 23 and 32 elements.
(27)
(28)
255
The phases i are functions of parameter r sin and the signs of the quark masses si and si .
Numerically we have,

1 = sin1 0.31(sd sc + 0.18sd st )r sin ,

2 = sin1 0.32(ss sc + 0.25ss st )r sin ,
3 = sin1 [sb st r sin ],
(29)
where we have taken from the Particle Data Group the central values of the quark masses mu =
2.7 MeV, md = 5 MeV, ms = 98 MeV, mc = 1.25 GeV, mb = 4.2 GeV, and mt = 174 GeV at
scale 2 GeV [11]. It shall be noted that since only the quark mass ratios enter the mixing matrix
and the quark masses run multiplicatively, the result is independent of quark mass scale. The
parameters for the left-hand quark mixing are taken as = 0.2272, A = 0.818, = 0.221, and
= 0.34.
A few remarks about the above result are in order. First, the hierarchical structure of the righthanded mixing is similar to that of the left-handed CKM, namely 12 mixing is of order , 13
order 3 and 23 order 2 . Second, every element now has a substantial CP phase. When r is of
order 1, the elements involving the first two families have CP phases of order , and the phases
involving the third family are of order 1. These phases are all related to the single spontaneous
CP-violating phase , and generate rich phenomenology for K and B meson systems as well as
the neutron EDM. Finally, from (27) and (29), it is clear that the final solution is a function of
sign bi-products si sj . We can always fix one of them, say su , to be positive, then we are left with
25 = 32 distinct sectors. The actual physical choice must be determined by phenomenology, as
we will illustrate in the following sections.
4. Higgs potential, mass spectrum and couplings
In this section, we discuss several issues related to the Higgs sector. In particular, we consider
the possibility of spontaneous CP violation from the Higgs potential, the mass spectrum of the
Higgs bosons, and the Higgs couplings to the quark sector. The results are useful for phenomenological studies in the following sections. Some of the results presented here have appeared in the
literature before, and we include them for completeness.
The most general renormalizable Higgs potential invariant under parity is given by [12]
V (, L , R )

+ Tr 23 Tr L L + Tr R R
= 21 Tr 22 Tr

2 2
2

+ 1 Tr + 2 Tr
+ 3 Tr
Tr
+ Tr

2
2

+ 4 Tr Tr
+ Tr + 1 Tr L L + Tr R R

+ 2 Tr(L L ) Tr L L + Tr(R R ) Tr R R

+ 3 Tr L L Tr R R + 4 Tr(L L ) Tr R R + Tr L L Tr(R R )

+ 1 Tr Tr L L + Tr R R

Tr L L + Tr Tr R R + h.c.
+ 2 ei2 Tr

+ 3 Tr L L + Tr R R + 1 Tr R L + Tr L R

,
R + Tr L + 3 Tr R + Tr L
+ 2 Tr
L
R
L
R
(30)
256
where there are a total of 18 parameters, 21,2,3 , 1,2,3,4 , 1,2,3,4 , 1,2,3 , and 1,2,3 . Due to the
leftright symmetry (LRS), only one of them, 2 can become complex and all other couplings
are real. We have included an explicit phase ei2 in 2 , introducing an explicit CP violation in
the Higgs potential.
After SSB, the Higgs fields acquire vevs, and the potential is minimized with respect to them.
The six minimization conditions are
V
V
V
V
V
V
=
=
= 0,
= =
=
vL vR
L
which lead to six relations among the vevs and coefficients in the Higgs potential [12,13]

2

21 1
vL2
vL2
3 2
2
1
+
+
2
=
cos

1
+
1
+
1
4
2
2(1 2 )
vR2
vR2
vR2

vL /vR
+ 2 cos L 3 2 cos(L 2)
,
1 2

22
vL2
vL2
2
3
=
)
+
cos(
)
cos(
+
+
1
+
2
2 2
2 cos
4(1 2 ) cos
vR2
vR
vR2
2

1
+ 22 cos 2 + 3 + 4 1 + 2 cos
2
cos

2
+ 1 1 cos(L ) 22 cos L + 23 cos(L 2)
vL /vR
,
4(1 2 ) cos

23
vL2
1
=
1
+
+ 1 1 + 2 + 3 2 2
1
2
2
2
vR
vR

vL2
2
+ 22 cos( + 2 ) cos( 2 ) 2
,
vR 1 vL2 /vR2

82 2 sin sin 2 vL
(21 3 )
vR
1 vL2 /vR2

= 1 cos(L ) + 2 cos L + 3 2 cos(L 2) 2 ,
0 = 1 sin(L ) + 2 sin L + 3 sin(L 2),

22 1 2 1 vL2 /vR2 sin 2

vL
= 2 sin(L )(2 + 3 ) + sin L + 2 sin(L 2) 1
vR

+ sin 3 1 + vL2 /vR2 + (43 82 ) 1 2 2
2
(31)
(32)
(33)
(34)
(35)
(36)
(37)
where = /vR represents a hierarchy in symmetry breaking. The above equations can be solved
for the Higgs vevs in terms of the parameters in the Higgs potential.
Historically, two special cases of the general potential have been studied in the literature,
namely manifest and pseudo-manifest LRS limits. The manifest LRS assumes real Higgs
potential i.e. 2 = 0, and in addition, no spontaneous CP violation, = L = 0. The only source
of CP asymmetry is from the Yukawa couplings. In this case, the quark mass matrices are hermitian due to parity invariance, and the left- and right-handed CKM matrices are identical up
257
to quark mass signs. Most early studies were made based on this simplification. At the level of
Higgs potential, this scenario necessitates fine-tuning: From the neutrino and quark mass hierarchy, we have vL < vR . Taking all the phases to zero in Eqs. (32)(37), the following
relations are found at leading order in 2 [12]
21
vR2
1
3 2
,
2
2(1 2 )
22
vR2
2
3 2
,
+
2
4(1 2 )
23
vR2
= 1 .
(38)
There are three equations for only two vevs vR and , implying a relation among parameters in
the Higgs potential, which can only be achieved through fine-tuning.
On the other hand, pseudo-manifest LRS requires P and CP invariance of the Lagrangian
(2 = 0), with the complex vev phase alone to explain the source of CP violation in the quark
sector. The Higgs potential is real when 2 = 0, but the vev could be complex. The Yukawa couplings are real and symmetric. The right-handed CKM matrix is related to the complex conjugate
of its left-handed counterpart with additional diagonal phase matrices multiplied on both sides.
mWL 2
However, when Higgs potential is real, the spontaneous CP phase is proportional to ( mW
)
R
and therefore goes to zero in the vR limit [14]. If one allowed for fine tuning of parameter,
one can generate a large enough phase [15] but at the price of large flavor changing neutral current. Phenomenology of these models have been extensively studied in literature [16,17], and it
has been established that this scenario fails to produce large enough CP asymmetry in B K
decay in the decoupling limit even when the maximum spontaneous CP phase is allowed. Away
from the decoupling limit, the scenario has been ruled out by the sign correlation between and
the above B-decay CP asymmetry [16].
In view of these results, for the minimal LRSMs to be realistic and natural, both explicit and
spontaneous CP phases must be taken into account. Anyway, as noted this is precisely what
happens in the minimal model. In Ref. [13], an approximate relation was derived between the
spontaneous CP phase and the explicit CP phase 2 in the Higgs potential,

2|2 | sin 2
sin1
(39)
,
3
where small requires a hierarchy between 2 and 3 , and/or small 2 . Clearly, when 2 = 0, one
has 0. A pioneering numerical study of the general CP scenario has been made in Ref. [5].
We will consider the Higgs spectrum and coupling in this general case in the remainder of the
section.
4.1. Higgs mass spectrum
With the Higgs vevs in Eq. (7) and the minimization conditions in Eqs. (32)(37), the Higgs
mass spectrum can be found in the presence of the CP phase 2 in the Higgs potential as well
as the spontaneous phase . We further restrict to the case , vL = 0. Thus L becomes
irrelevant and all i decouple. We will keep only terms linear in and for simplicity.
In the minimal LRSM, there are 20 scalar degrees of freedom in the Higgs fields , L and
R , including 2 double-charged, 4 single-charged and 4 complex neutral Higgs bosons. After
SSB, the mass eigenstates are linear combinations of those. Two single-charged and two real
neutral Higgs bosons get absorbed and become longitudinal components of WL , WR , Z and Z .

i +
2 + 1+ ,
G+
L = e
258

1
+
+
G+
=
R
R ,
2
2
G0Z = 2 Im R0 ,

G0Z = 2 Im 10 + ei 20 ,
(40)
where we have neglected terms of order 2 and 2 . Among the remaining 14 fields, only one real
and neutral component h0 acquires mass at the electroweak scale , identified as the SM Higgs
boson, while the other Higgs fields have masses of order vR . The physical Higgs states and their
masses are collected in Table 1. In the limit 0, 2 0 and , vR , our results agree
with those in Ref. [18], except for the SM Higgs mass.
The calculation of SM Higgs mass is a bit involved and warrants a little further discussion.
From the (tree-level) Higgs potential, one can write down the mass matrix for 8 neutral Higgs
components in the basis of {Re 10 , Im 10 , Re 20 , Im 20 , Re L0 , Im L0 , Re R0 , Im R0 }. The rows
and columns containing Re L0 , Im L0 and GZ = Im R0 are already diagonal and hence decouple.
The remaining 5 5 sub-matrix M 2 is somewhat complicated. Since the major components
have vR -scale masses, we can work in perturbative expansion with respect to 3 parameters: the
spontaneous phase , = / and = /vR . As we shall see later, in order to satisfy the
CP constraints and to have the right-handed scale at TeV, they must be approximately of the
same order: , , O(102 ). Naively, the SM Higgs mass squared should be of electroweak
scale 2 = 2 vR2 , and we must work up to the order 2 in diagonalization: M 2 (, , ) =
2 + M 2 + M 2 + . At zeroth order, in the basis of {Re 0 , Im 0 , Re 0 , Im 0 , Re 0 },
M(0)
R
1
1
2
2
(1)
(2)
2
M(0)
= vR2
0
3
3
(41)
41
which is already diagonal. Re 10 and Im 10 have zero mass and the other three have masses at
scale vR . This means Re 10 and Im 10 are the dominant components of the SM Higgs h0 and
Table 1
Physical Higgs states and mass spectrum at leading order in minimal LRSMs. We assume vL = 0 and keep only linear
terms in = /vR and = /. All fields but h0 have masses on the vR scale. h0 is identified as the SM Higgs boson
Higgs state
h0 = 2 Re(10 + ei 20 )
H10 = 2 Re( ei 10 + 20 )
0
H20 = 2 Re R
0
H30 = 2 Re L
0
A1 = 2 Im( ei 10 + 20 )
0
A02 = 2 Im L
+
H1+ = L
+
H2+ = 2+ + ei 1+ + 1 R
2
++
R
++
L
Mass2
2
2 2
(41 1 ) 2 + 3 vR
1
2
3 v R
2
41 vR
2
(3 21 )vR
2
3 v R
2
(3 21 )vR
2 + 1 2
(3 21 )vR
2 3
2 + 1 2)
3 (vR
2
2 + 2
42 vR
3
2 + 2
(3 21 )vR
3

2 are relevant,
the Goldstone boson GZ . At the first order, only the first two rows of M(1)

0 0 0
0
21
2
M(1)
= vR2
,
0
0 0 0 3
2 is needed
and at the second order, only the upper 2 2 block of M(2)

3 2 + 4 2 1
0
2
2
M(2) = vR
.
0
3 2
259
(42)
(43)
Using the standard formula in perturbation theory, the SM Higgs boson state and mass up to the
second order in the expansion are

h0 2 Re 10 + ei 20 ,

12 2
2
mh0 = 41
(44)
+ 3 2 vR2 .
1
There are two contributions to mh0 . One is the usual electroweak breaking mass 41 2 , and
2
the other is the mass shift 3 2 vR2 11 2 , due to additional couplings in the Higgs potential.
Because = mb /mt = /vR , the two parts are roughly comparable when 1 3 O(1).
4.2. Charged and neutral currents
In this subsection, we present the charged and neutral couplings between quarks and Higgs
mass eigenstates. From the Yukawa coupling term (12), we can express h and h in terms of the
vevs and quark mass matrices
h=
Mu Md ei
,
2 2
Using
= 2 2 =
20
2
Md Mu ei
h =
.
2 2
1+
10
(45)

,
(46)
one can write

Rj + h.c. LN + LC .
LY = Q Li (hij + h ij )Q
The neutral Higgs-quark coupling part is

LN = u Li hij 10 + h ij 20 uRj + dLi hij 20 + h ij 10 dRj + h.c.

= ( 2GF )1/2 u Li M U ii h0 iG0Z 2 ei H10 iA01 uRi

+ dLi M Dii h0 + iG0Z 2 ei H10 + iA01 dRi

+ ( 2GF )1/2 u Li VL M D VR ij H10 iA01 uRj

+ dLi VL M U VR ij H10 + iA01 dRj + h.c.,
(47)
(48)
where VL,R are left- and right-handed CKM matrices. h0 has the known SM Higgs couplings
to the quark fields. The second term in Eq. (48) changes the quark flavors through H10 and A01
260
bosons. They are called the flavor changing neutral Higgs (FCNH) bosons in LRSMs. The dominant contribution comes from the intermediate top, bottom and charm quark masses. Taking into
account the CKM hierarchy, we found that the transitions from d to s with intermediate charm
quark mass and from b to s with intermediate top quark masse are most significant.
The charged Higgs-quark coupling part of Lagrangian density is
LC = u Li (hij 1+ h ij 2+ )dRj + dLi (hij 2 h ij 1 )uRj + h.c.

= ( 8GF )1/2 u Li M U VR 2 ei VL M D ij dRj H2+

u Ri VR M D 2 ei M U VL ij dLj H2+

u Li (VL M D )ij dRj G+
Ri (M U VL )ij dLj G+
L +u
L + h.c.
(49)
Again, the couplings are proportional to quark masses and hence the heavy-quark contributions
stand out. With the above couplings, we will study their contributions to various flavor changing
and conserving processes in the minimal LRSM.
5. K 0 K 0 and neutral B-meson mass mixing
In this section, we consider the neutral kaon and B-meson mass mixing in the minimal LRSM,
using the right-handed quark mixing matrix obtained in the previous section. We first study the
WL WR mixing-box contribution to the KL KS mass difference mK and derive an improved
bound (2.5 TeV) on the mass of right-handed gauge boson WR , using the updated hadronic
matrix element and strange quark mass. Historically, the kaon mass mixing provided the most
stringent constraint upper bound (1.6 TeV) on the mass scale of the right-handed WR boson [19].
With our new right-handed CKM mixing, the conclusion does not change significantly, although
in the literature, quite different mixings have been speculated upon and the result did change
dramatically, and we rule these possibilities out. In the past few years, significant progress has
been made in hadronic physics through lattice QCD simulations, helping to tighten the bound. We
also consider the contribution from the FCNHs and constraint on their masses. Because the box
and FCNH contributions are additive, the bounds are valid independently. In the last subsection
we explore Bd B d and Bs B s mass mixing. Because the hadronic contributions arise dominantly
from short distance, the bounds on MWR and MH turn out to be significant as well.
It is useful to provide our convention for neutral meson mixing at the beginning. For a pair
of neutral mesons, |P 0
and |P 0
, we assume under CP transformation CP|P 0
= |P 0
. If the
effective Hamiltonian in the basis of |P 0
and |P 0
is Hij =Mij iij /2, a pair of eigenstates
i /2)/(M i /2).
are |P1,2
= p|P 0
q|P 0
. The ratio q/p is chosen to be (M12
12
12
12
In the CP symmetric limit, it is possible to have q/p = 1, and P1 is then CP-even and P2 is CPodd. The mass difference is M2 M1 = 2 Re(q/p(M12 i12 /2)), and the width difference
2 1 = 4 Im(q/p(M12 i12 /2)).
5.1. Kaon mixing and the boxing diagram
In SM, the leading-order short-distance S = 2 process comes from the box diagram with
WL boson and up-type quark exchanges. Flavor change happens at the vertices via the CKM
mixing matrix. The short distance contribution comes mainly from internal loop momentum
flow at the scales around c and t quark masses, whereas the momentum region around WL -boson
mass is suppressed due to the celebrated GlashowIliopoulosMaiani (GIM) mechanism. The
261
Fig. 1. Dominant contribution to KL KS mass difference in LRS model.
long-distance contribution comes from one or two up-quark exchanges and must be calculated
using non-perturbative methods. It has been generally accepted that this latter contribution does
not dominate over the short distance one. In fact, a chiral perturbation calculation [20] puts the
long distance contribution at about half of the experimental mass difference.
In the LRSM, there are new box-diagram contributions which turn out to be quite large [19,
21]. The dominant one comes from WL WR interference with one internal vector-boson being
the left-handed WL and the other right-handed WR . As such, the chirality of the internal as well
as external quarks must be flipped, and the contribution is proportional to the internal quark
masses, as shown in Fig. 1. There is no GIM suppression even in the manifest LRS scenario:
VL = VR . The effective Lagrangian from the WL WR box-diagram is
LLR =
2
G2F MW
L
4 2
RL
LR
xi xj
i j
ij

(4 + xi xj )I1 (xi , xj , ) (1 + )I2 (xi , xj , ) O LR + h.c.,
(50)
MW
mi 2
L 2
in which LR
i = VLi2 VRi1 , = ( MW ) , and xi = ( MW ) , i = u, c, t. The four quark operator is
2
O LR MW
= (s PL d)(s PR d),
R
(51)
where PL,R = (1 5 )/2 are chiral projection operators. The loop-related integrals are
ln(1/)
(1 )(1 xi )(1 xj )

xi ln xi
+ (i j ) ,
+
(xi xj )(1 xi )(1 xi )
ln(1/)
I2 (xi , xj , ) =
(1 )(1 xi )(1 xj )

xi2 ln xi
+
+ (i j ) .
(xi xj )(1 xi )(1 xi )
I1 (xi , xj , ) =
(52)
The above contribution is evaluated in t HooftFeynman gauge. Generally, we have other contributions from the exchange of single-charged Higgs bosons as well as the Goldstone bosons
if the calculation is not in unitary gauge. In fact, to maintain gauge invariance, we have also to
include the triangle diagrams with the FCNH bosons and the related self-energy corrections [22].
It turns out that in the gauge we use, the box diagram with WL and WR interchanges alone gives
the dominant contribution.
The Wilson coefficient in Eq. (50) is proportional to the internal quark masses and the corresponding CKM matrix elements. The up-quark contribution is negligible and the top-quark
262
contribution is suppressed by the off-diagonal CKM elements. Therefore the charm-quark exchange dominates the effective interaction which simplifies

2
G2F MW
1
LR RL
L
HLR =
2
x
+
ln
1
+
ln
x
c
c
c
c
4
4 2

2
2
(s d) (s 5 d) .
(53)
The above effective operator is obtained at the electroweak scale MWL and by including QCD radiative corrections, the operator can be evolved to the hadronic scale where the non-perturbative
matrix element is calculated. We will include this correction in the next subsection. The hadronic
matrix element of the four-quark operator at the hadronic scale is expressed in the vacuum saturation form,

2
mK
5 )s d(1
+ 5 )s|K 0
= 2MK FK2 B4 ()
K0 |d(1
(54)
,
ms () + md ()
where the kaon decay constant FK = 113 MeV and B4 = 1 corresponds to vacuum saturation
approximation. In this form, the matrix element diverges in the chiral limit, an important reason
for the enhanced contribution of the box diagram. The correction factor, B4 , can be and has been
calculated using lattice QCD. In a recent calculation [23], the domain-wall fermion was used,
and B4 = 0.81 was found at = 2 GeV in naive dimensional regularization scheme. In the same
scheme and scale, the strange quark mass is ms = 98(6) MeV, which is smaller than what one
has naively expected in the past.
5.2. An improved lower bound on MWR
Considering only the box diagram, the new contribution to the mass difference of KL KS
can be expressed as

mK = 2 Re 4 () K 0 H S=2 ()K 0 ,
(55)
where 4 is a factor characterizing the QCD radiative correction in scale running from MW1
to 2 GeV [2426]. There are several enhancement factors here comparing to the SM box
diagram. First, due to absence of the GIM mechanism, the Wilson coefficient is about a factor of
30 larger. Second, the hadronic matrix element is chirally enhanced by a factor of 20. Finally, the
short-distance QCD correction 4 = 1.4 gives another enhancement. The only suppression comes
MW
from the difference between the left and right-handed symmetry breaking scales, = ( MWL )2 .
R
Therefore the new contribution can be approximated by

MWL 2
MKLR 103
(56)
MKSM .
MWR
The sign can both be positive or negative depending on the product st sc . With the standard criteria
that the new contribution should not exceed the experimental value [19], we find a lower bound
for MWR ,
MWR > 2.5 TeV.
(57)
On the other hand, the SM contributions from both long and short distances have the same sign
as the experimental number and account for more than one-half of its value. Therefore, a less
263
Fig. 2. S = 2 effective interaction induced by flavor-changing neutral Higgses.
conservative bound is obtained if requiring the new physics contribution is less than one-half
of the experimental data. If this new standard is adopted, the above bound change to 4 TeV.
Giving the long-distance uncertainty in MK , this bound shall be used at less confidence level.
Nonetheless, as we shall see in the next section, the CP violating observables are providing
equally competitive bounds albeit with a significant hadronic physics uncertainty.
5.3. Tree-level FCNH contribution and A lower bound on MH
In the LRSM, there is also a new contribution to the K 0 K 0 mixing mediated by the FCNH.
The FCNH boson is a complex field and can be expressed in terms of the two real fields H10
and A01 . The effective Lagrangian follows from Eq. (48)
2

LR
GF RL
(s d)2 (s 5 d)2
i + i
mi
LFCNH =
2
m2 0
m2 0
2
i
H1
A1
RL
2

LR
2
i i
(s 5 d)2
(s d)
(58)
mi
.
2
2
2
m
m
0
0
i
A1
H1
The corresponding Feynman diagram is shown in Fig. 2. According to our previous discussion,
the two scalar fields H10 and A01 have the same masses, roughly corresponding to the right-hand
scale, m2 0 m2 0 3 vR2 . Therefore, it is convenient to rewrite Eq. (58) in a more compact
form
H1
A1

GF
RL
(s d)2 (s 5 d)2 .
mi mj LR
HFCNH 2
i j
2m 0 i,j
(59)
H1
It is easy to check that the FCNH and the box diagram contributions have the same sign
because 4(1 + ln xc ) + ln < 0, and thus they cannot cancel each other, even allowing possible
freedom in choosing the quark mass sign. Therefore, the lower bound on the right-handed-W
boson mass remains. One can also derive a lower bound on the masses of H10 and A01 using
MK . A straightforward calculation shows that if demanding the FCNH contribution is less
than the experimental data,
MH 0 , MA0 > 15 TeV,
1
(60)
which is about twice as large as in [5]. One can obtain this value presumably by a large 3
parameter in the Higgs potential. However, one cannot make 3 arbitrarily large. As we shall
discuss later, large 3 not only causes naturalness problem, but also leads to a large SM Higgs
mass which threatens the perturbative unitarity [27].
264
Fig. 3. Bd B d and Bs B s mass differences from the WL WR box diagram with top-quark intermediate state plus the
SM contribution. The shaded regions are within 20% of the experimental values.
5.4. Constraints from neutral Bd and Bs mass mixing

The physics of Bd B d and Bs B s mixing in SM is similar to that of K 0 K 0 mixing, coming
from the W -boson box diagram and FCNH. However, in the former case, the intermediate top
quark contribution dominates almost entirely due to its mass and CKM couplings. Because of
this, the SM calculation can be done quite accurately with the help of the lattice QCD matrix
elements. In fact, in many global CKM fits, the mass differences MB and MBs have been
used to determine the top CKM couplings Vtd and Vts . However, there are still appreciable
uncertainties in the lattice calculations and global CKM fits, and the beyond SM physics could
contribute as much as 20% of the mass differences without running into conflict with the present
SM calculations and experimental data. We will use this possible discrepancy as a constraint on
the LRSM.
The LRSM contribution to the mixing can be taken directly from (50)
q
HLR (b )
2
G2F MW
L

2S VtbL VtqR VtbR VtqL xt

L q bP
R q + h.c.,
4 + xt2 I1 (xt , xt , ) (1 + )I2 (xt , xt , ) bP
4 2
(61)
where the leading-logarithmic running factor S = 2.112 [24] and the functions Ii are defined in
Eq. (52). The hadronic matrix element can be defined using a factorization,

2
mBq
1
1
q
Bq |qP
(62)
L bqP
R b|B q
= MBq fB2q B4 ()
+
,
12 2 mb + mq
where the minus arises from our definition of the CP transformation for the meson states. The decay constants have been calculated in lattice QCD: fBd = 216 MeV, fBs = 1.20fBs [28] and the
non-perturbative B-factors are B4d = 1.16 and B4s = 1.17 [29]. The ratio of the new contribution
2 /M 2 , which is smaller than the kaon mixing case in the absence of
to the SM one is 102 MW
WR
L
chiral enhancement.
The experimental values for the mass differences of Bd and Bs are (0.507 0.12) ps1 and
(17.77 0.12) ps1 , respectively [11], with central values shown as horizontal lines in Fig. 3.
The shades around the central values are within 20%, and are considered as the combined experimental and SM theory error. In the same figure, we also plot the mass differences as a function
of MWR , calculated as a sum of the LRSM contribution and the experimental central values. The
agreement between experiment and theory is reached only when MWR is larger than 2.5 TeV.
The sign of the LRSM contribution is related to that of sd sb and ss sb , which we have chosen to
be +1. If taking as 1, approaching to the central value comes from below.
265
Thus the constraint on the MWR mass from the neutral B-meson mixing is roughly comparable to the kaon case, due to a better theoretical understanding of the SM physics. The future
improvement can come from a better determination of Vtd from other sources and a better determination of B-parameter and decay constants.
We have also studied the constraint on the FCNH mass, MH . We find that from MBd , the
bound is 12 TeV, and for MBs , the bound is 25 TeV.
6. CP-violating observables in LRSM
As mentioned before, in a generic yet minimal LRSM we have both explicit and spontaneous
CP violations. A prominent feature of right-handed quark mixing VR obtained in the previous
section is its phases, which are entirely determined by the Dirac phase CP and the spontaneous
CP phase . These physical phases generate interesting effects in various CP-violating observables. The phenomenology of CP violation in LRSM is rich, which in turn constraints the model
severely. In this section, we will explore a number of CP violating observables including , ,
neutron EDM and CP asymmetry in B J /KS , to place constrains on the mass of WR as well
as the spontaneous CP phase . Some results here have appeared before in a Rapid Communication paper [7] and there are important updates in this more extensive study.
The indirect CP violation receives large contributions from both explicit and spontaneous
CP phases. Unless there is a strong cancellation, the right-handed WR mass must be larger than
15 TeV. We use the cancellation condition to fix the spontaneous CP phase, which is then used to
predict the neutron EDM. Using the experimental bound on the EDM, we obtain a strong lower
bound on MWR , which can be improved with better calculations of the hadronic matrix elements
and more precise experimental data. We obtain a strong lower bound on MWR from the direct CP
violation parameter , calculated under the factorization assumption for the four-quark matrix
elements. Therefore, we find that the CP violating observables in the kaon system and neutron
EDM provide competitive bounds as the well-known kaon mass mixing. These bounds can be
improved further with better knowledge of the non-perturbative hadronic physics.
6.1. Indirect CP violation in kaon decay
We first study the CP violating parameter in kaon mixing. This indirect CP violation parameter is related to the flavor mixing interaction by,
ei/4 Im K 0 |H S=2 |K 0
,
=
(63)
mK
2
where we have neglected the direct CP contribution 0 from kaon decay, which can be justified
posteriori [5]. In LRSM, according to the previous section, both the WL WR box diagram in
Eq. (53) and the tree level FCNH exchange in Eq. (59) can make significant contributions. In
the present case, their signs can be different due to both charm and top quark contributions, in
contrast to the mass mixing [5]. For simplicity, we ignore that latter contribution and consider
the constraint from the box diagram alone.
There are two sources of CP phases in VR which enter the WL WR box diagram: the Dirac
phase CP inherited from VL , and the spontaneous phase . In the manifest LRS case, only CP
is present and there is a very tight lower bound on mass of WR which we find no lighter than
1520 TeV (see below). If the spontaneous CP phase is also present, one can seek for certain
cancellation between the two contributions to lower the bound on MWR . In fact, one can roughly
266
estimate the size of r sin for a cancellation. The Dirac phase CP appears in the expression
proportional to VtsR VtdR 5 sin CP via top quark exchange in the box diagram, while the
R 2 r sin . Hence r sin should be of order
spontaneous CP phase contributes through VcsR Vcd
3
sin CP 0.01 when cancellation happens.
The present experiment value is ||expt = (2.232 0.007) 103 [11]. In SM, can be calculated quite accurately because the top quark dominates the box diagram and the main contribution
is due to short-distance QCD physics. The only large uncertainty comes from the CKM matrix
element Vtd and the hadronic matrix element related to K 0 K 0 mixing. Because of this, we
assume that the new contribution accounts for less than 1/4 of the experimental value.
Using the box-diagram result in the previous section, we find an approximate expression for
LR valid for MWR > 200 GeV,

1 TeV 2
LR = 1.58
(64)
ss sd Im g(MWR , 2 , 3 )ei(1 +2 ) ,
MWR
where

MWL 2
g(MWR , 2 , 3 ) = 2.22 + 0.076 + (0.030 + 0.013i)sc st cos 2(2 3 ) ln
,
MWR
(65)
where i are given in Eq. (29) and the QCD running correction has been taken into account.
When i = 0, the dominant CP violating contribution comes from charm-top interference in the
box diagram.
We search for the allowed region in MWR r sin plane shown in Fig. 4. The spontaneous CP
violation effect can always cancel the effect of the Dirac phase, thus itself places no bound
Fig. 4. Typical scenarios from the constraint: The first and two figures correspond to sd = ss = 1 and sd = ss = 1,
respectively (small r sin solution). The third and fourth correspond to sd = ss = 1 and sd = ss = 1 respectively
(large r sin solution). In all cases, su = sc = st = sb = 1.
267
Fig. 5. Constraint on MWR and the spontaneous phase from kaon decay parameter , with sd = ss = 1 and all other
si = 1. Red triangles are for MH = , blue squares are for MH = 75 TeV, and green dots for MH = 20 TeV. (For
interpretation of the references to colour in this figure the reader is referred to the web version of this article.)
on MWR . The cancellation depends on choices of the quark mass signs. The 32 choices of the
signs (for fixed su = 1) roughly correspond to two scenarios: small r sin solutions for ss = sd
and large r sin solutions for ss = sd . The existence of two scenarios can be readily seen
through the above approximate expression. According to (29), 1 and 2 are similar in size and
their relative sign depends on the bi-product sd ss . When ss = sd , 1 and 2 have the same sign
and add constructively in the exponential in (64). So itself must be small in order for 1 + 2 to
cancel the phase in g(MWR , 2 , 3 ). This generates the small r sin solutions. On the other hand,
if ss = sd , 1 + 2 cancels and is proportional to r sin multiplied with a small coefficient.
So r sin has to be large to cancel the phase in g(MWR , 2 , 3 ). This corresponds to the large
r sin solutions. For the small r sin case, we find r sin 0.05, but for large r sin , it can
take several different positive and negative values. As we will see in the next subsection, if one
includes the constraint from neutron EDM, only small r sin is phenomenologically viable.
Of course, one has to consider the FCNH contribution which has been known to be large [30].
In fact, with just the Dirac CP phase in the FCNH contribution, places a limit on the Higgs
mass on the order of 100 TeV. With the new spontaneous CP contribution, there is a possibility
of cancellation. In fact, one can make similar plots as in Fig. 4, in which the bound can be
satisfied even for very low MH 1 TeV. However, the required r sin for the cancellation,
0.2, is very different from that needed for the box diagram. The conclusion is that there is no
bound on MH coming from when FCNH contribution is considered alone.
Because of the conflict in the spontaneous CP phase required for individual cancellations in
the box and FCNH contributions, one might expects their combined contribution places a joint
bound on MWR and MH . This, however, is not the case, because the two contributions again
cancel each other, as was first found in [5]. In fact, with the analytical solution for the righthanded mixing, we arrive at an even stronger conclusion: For any given pair of MWR and MH , we
can always find a r sin such that the total contribution to vanishes. This situation is extremely
interesting, because it implies that itself, unlike the kaon mass difference, is completely useless
in constraining the individual parameters in the new contribution. It does, however, provide a
correlation among different parameters, as shown in Fig. 5, where for several different values of
MH = , 75, 25 TeV, we plotted the allowed regions in the MR and r sin plane. Because of
the FCNH contribution, the pattern of the correlation changes considerably as MH changes. The
268
Fig. 6. Dominant quark-level effective operators contributing to neutron EDM in LRSM.
general trend is that the spontaneous CP parameter r sin increases toward 0.2 as MH becomes
smaller, consistent with the cancellation pattern in FCNH contribution to found above.
6.2. Neutron EDM
The neutron EDM imposes another constraint on the new CP phases in VR and MWR . Nonzero EDM implies both P and T (or equivalently CP in local quantum field theories) violations.
At the quark level, sources of flavor-neutral CP violation are mainly from the penguin diagrams
in the SM, and from the tree level WL WR exchange in the LRSM [31,32]. Generally, there are
several contributions to the neutron EDM, including valence quark EDM, quark chromomagnetic
dipole moment (CDM) induced EDM, dimension-6 pure gluonic operator contribution, as well
as the contributions at hadronic level. The present experimental upper bound on neutron EDM is
3.1 1026 e cm [11].
In the SM, contributions to the neutron EDM mainly come from the CP violating penguin
diagrams. The flavor changing nature of CKM CP violation means that the leading contribution
is at least second order in weak interaction ( G2F ). The predicted neutron EDM is well within
the experimental boundabout 1033 e cm.
In the LRSM, the flavor-conserving CP violating four-quark operator arises from the tree-level
diagram with WL WR mixing exchange, as shown in Fig. 6
q
q
L R
Luquq = 2 2GF sin ei Vuq
(66)
Vuq O O+ + h.c.,
where q = d, s, and
2
q
O+ = u
PL q q
PR u + uP
L q,
R uqP
3
2
q
L q.
R uqP
O = uP
3
(67)
q
At low energy, short-distance QCD effect enhances the operator O and suppresses O+ . The
effective Lagrangian reduces to [33]

2 2GF
L R
uu
5 d)
5 udd
Vud (u
d
Lud(s)ud(s) = i
sin Im ei Vud
3

L R
5 us s uu
Vus (u
s 5 s) ,
+ Im ei Vus
(68)
where the leading-log QCD factor = (
S (2 ) 89
2 ))
S (MW
L
3.5. The CP violating nn coupling is
proportional to the hadronic matrix element g nn = n|Lud(s)ud(s) |n

which, in the factor-
269
Fig. 7. A dominant contribution to the neutron EDM through chiral pion exchange. The shaded blob is CP violating
coupling g nn , and the black dot is the strong coupling g nn . The neutron couples to photon via its anomalous magnetic
moment.
ization approximation, is
n|q
5 q q q |n
|q
5 q|0
n|q q |n
,
(69)
for q = u, d and q = u, d, s. From the SSB of the chiral symmetry, 2mu |u

5 u|0
=
5 d|0
= iF m2 , with F = 93 MeV. n|uu|n
4 and n|dd|n
5 can be fixed
2md |d
dd|n

from the neutronproton mass difference and the N -term: N = 12 (mu +md ) n|uu+
45 MeV [34], where we use quark masses mu = 2.7 MeV, md = 5.0 MeV at = 2 GeV. We neglect n|s s|n
n|uu|n
.
The neutron EDM from the Feynman diagram in Fig. 7 is

2
egnn g nn N
m
F
,
dn =
(70)
2
2mN
8
m2N
where N = 1.91 is the neutron anomalous magnetic dipole moment and the loop function is

3
s s 2 /4
3s s 2
s(5s s 2 ) 4s

F (s) = s
(71)
arctan
ln s +
.
2
2
s/2
2 s s 2 /4
The contribution is suppressed by the mixing angle between WL and WR , but is enhanced by
the chiral logarithmic factor ln( m )2 . Putting in all the known physical parameters, we arrive at
an approximate formula
e

d 3 1019 sin Im ei V L V R e cm,
(72)
n
ud ud
which is approximately a function of r sin for small .
For a fixed MH , the neutron EDM and can be used to give a lower bound for MWR as well
as a corresponding solution for r sin . On the left panel in Fig. 8, we have shown the neutron
EDM constraint as a function of r sin and MWR , and it is obvious that the EDM limit prefers
small r sin . The smallest spontaneous CP phase is obtained when MH is large and decouples,
and then
MWR > 8 TeV,
(73)
which is a very tight bound. At this point, we also fix the product
r sin 0.05.
(74)
For a lower MH , the spontaneous CP phase must be large from the constraint, and the corresponding lower bound on MWR increases considerably. For example, when MH = 75 TeV, r sin
is now greater than 0.1, and the lower bound on MWR becomes 18 TeV.
270
Fig. 8. Constraints on the mass of WR and the spontaneous CP violating parameter from kaon decay parameter
(MH = , red triangle; MH = 75 TeV, blue square; MH = 20 TeV, large green dots) and neutron EDM (yellow dots).
In the right panel, the theoretical EDM result is reduced by a factor of 5. (For interpretation of the references to colour in
this figure the reader is referred to the web version of this article.)
Fig. 9. Higgs exchange contributions relevant to neutron EDM. The dashed lines include both charged and neutral Higgs
bosons exchanges. The two-loop diagrams contain closed top-quark loops.
We note that there is considerable hadronic uncertainty in the evaluation of CP-violating coupling g nn and, to the less extent, chiral perturbation expansion. However, even one allows a
factor of 5 over-estimate in the hadronic calculation, the combined and EDM will still provide
a strong constraint on MWR on the order of order 4 TeV, as shown on the right panel in Fig. 8. If
MH = 25 TeV, the lower bound on MWR becomes 8 TeV. A future improvement on the neutron
EDM data and theoretical calculation can strengthen this bound considerably.
We finally comment on the Higgs boson exchange contributions to the neutron EDM. According to Section 4.2, the Higgs bosons H10 , A01 and H2+ have CP violating couplings (Eqs. (48)
and (49)) to the quark fields. The valence quark EDM receives contributions from virtue H10 ,
A01 and H2+ exchange as shown in Fig. 9. Potentially large contribution also comes from neutral Higgs boson exchange and virtual top-quark effect at two loops [35], as well as two loop
pure gluonic operators due to charged/neutral Higgs exchange [36]. A complete analysis of these
contributions has been carried out in a pseudomanifest LRS limit with two doublets Higgs fields
271
Fig. 10. One scenario (all sq = 1) for the Higgs exchange contribution to neutron EDM as a function of r for different
values = 0.05 (solid line), = 0.5 (long dashed line) and = 1 (short dashed line). We choose the FCNH mass to be
15 TeV and the charged Higgs mass equal to 3 TeV.
instead of triplets considered here [37]. With the explicit form of right-handed CKM mixing, we
re-evaluate these contributions in the general case of CP violation. As discussed in Section 5.3,
H10 and A01 must be heavy enough to suppress their contribution to kaon mixing, and we take
their masses to be 15 TeV. In this case, the charged-Higgs H2+ exchange dominates, whose contribution to d-quark EDM is approximated by
m2 +

2e mt GF
m2t
H2
2 2 ln
dd
d Im VtdL VtdR ei ,
2
3 4 2 2 m +
mt
(75)
H2
where the scaling factor d 0.12. The d-quark CDM fd come from a similar diagram with
photon replaced by gluon leg, efd 32 fd dd . The contribution to u-quark EDM and CDM is
suppressed by a factor mb /mt and is negligible compared to that of d-quark. Meanwhile, the
two loop diagrams are found to be negligibly small. To a certain level of approximation, the
neutron EDM can be related to the quark EDM and CDM through the SU(6) relation:

1
1 4
2
fd + fu .
dne = (4dd du ) +
(76)
3
3 3
3
A more accurate relation would use the tensor charges of the nucleon.
In Fig. 10, we plot the Higgs exchange contributions to the neutron EDM as a function of
r for different values of . We choose the FCNH mass to be 15 TeV, and the charged Higgs
mass be 3 TeV. We find the contributions are always smaller than 1026 e cm, well within the
experimental bound. Therefore one can neglect the Higgs exchange contribution without altering
the W -mass bound for the neutron EDM.
6.3. Direct CP violation
The direct CP violation in neutral kaon to decay is calculated via

i
Im A2 Im A0 i(2 0 )
q
,
e
=
(77)
p
Re A2 Re A0
2
where the decay amplitudes A0 and A2 are defined as the matrix elements of the S = 1 effective
Hamiltonian between the neutral-K meson and the isospin I = 0 and 2 states,

(2)I (i)HS=1 K 0 = AI eiI .
(78)
272
Fig. 11. New tree-level contribution to the S = 1 interaction from LRSM.
I is the strong phase for scattering, A2 /A0 and p, q are the mixing parameters for
K 0 K 0 . To an excellent approximation, can be taken as real and q/p = 1. Therefore, we
focus on calculating the imaginary part of the decay amplitudes.
In the SM, the contributions to come from both QCD and electromagnetic penguin diagrams
[38]. The QCD penguin contributes exclusively to I = 1/2 decay, whereas the electromagnetic
penguin is mainly responsible for I = 3/2 decay. Both contributions are important but have
opposite signs. Therefore, the final result depends on precision calculation of hadronic matrix
elements. The state-of-art chiral perturbation theory [26,3941] and lattice QCD calculations
[42,43] have not yet been sufficiently accurate to reproduce the experimental result [44]. A review
of the standard model calculation can be found in Refs. [45,46].
In LRSM, each element in the right-handed CKM matrix has a substantial CP phase. As a
consequence, there are tree level contributions to the phases of A2 and A0 . Following closely
the work by Ecker and Grimus [25], the S = 1 effective Hamiltonian from Eq. (11) and the
tree-level Feynman diagram in Fig. 11 is
2
4

S (2 ) b LL
S (2 ) b LL
O
()
+
O
()
+
S (ML2 )
S (ML2 )
2
4

ML2 RR
S (2 ) b RR
S (2 ) b RR
O+ () +
O ()
+ 2GF 2 u
MR
S (MR2 )
S (MR2 )
8
1

S (2 ) b LR
S (2 ) b LR
i
+ 2 2GF sin LR
e
O
()
O
()
u
+
S (ML2 )
S (ML2 )
8
1

S (2 ) b RL
S (2 ) b RL
RL i
+ 2 2GF sin u e
O ()
O+ () ,
S (ML2 )
S (ML2 )
(79)
where we have taken into account the leading-logarithm QCD corrections with renormalization
scale taken to be around 1 GeV, and b = 11 2Nf /3 with Nf the number of active fermion
CKM CKM
flavors. The mixing coupling AB
u = VAud VBus , A, B are L, R. The four quark operators are
tree
HS=1
=

2GF LL
u
LL,RR
PL,R uu
PL,R s u
= d
PL,R s d
PL,R u,
O
2
LR,RL
PL,R uu
= d
PR,L s + dP
L,R u,
O+
R,L s uP
3
2
LR,RL
= dP
L,R u,
O
(80)
R,L s uP
3
where PL,R are projection operators. There are also new penguin diagrams involving the righthanded gauge boson contributing to . However, these contributions are suppressed by loop
factors and are neglected here.
273
The hadronic matrix elements of the four-quark operators are calculated using the factorization assumption,
LL,RR 0
X
K =
(2)I =0 O+
,
3 6
LL,RR 0

K = 2X
(2)I =2 O+
,
3 3
LL,RR 0

X
K =
,
(2)I =0 O
2 6
LL,RR 0

K = 0,
(2)I =2 O
LR,RL 0
K = 4X
(2)I =0 O+
,
9 6
LR,RL 0

K = 2X
(2)I =2 O+
,
9 3

LR,RL 0

K = 1 X + Y + Z ,
(2)I =0 O
2
6
6 18

X
1
LR,RL 0
(2)I =2 O
Z ,
K =
6 3 6

where the parameters X, Y and Z are

0
5 u|0
+ u
s K ,
X |d
2

= i 2F mK m2 0.03i GeV3 ,

5 s K 0 ,
Y + |uu|0
0|

2
0.273i GeV3 ,
= i 2FK A2 1 m2K /m2

5 u|0
+ |us
Z |d
K 0

2
0.18i GeV3 ,
= i 2F A2 1 m2 /m2
(81)
(82)
where A = m2K /(ms + md ), F = 93 MeV and FK = 1.22F .

The numerical estimates are taken from Ref. [17]. Y and Z are much bigger than X due to
chiral enhancement. Clearly the factorization approximation must be improved as indicated by
the empirical I = 1/2 rule, which is beyond the scope of this paper. We note, however, that for
our estimation of the bound on MWR , a multiplicative uncertainty factor on the matrix elements
is reduced by a square-root.
To calculate the weak phases of the decay amplitudes A0 and A2 , we use the experimental
value for the real parts of A0 and A2 : Re A0 3.33 107 and 1/22. The dominant new
contribution is from the WL WR mixing term due to enhanced hadronic matrix elements and
larger CP violation phase . In fact, because the phase in the apparent factor ei is much
which is typically of order 1 and 2 given in Eq. (29), is
larger than the phase in RL,LR
u
approximately a function of r sin , rather than r and sin independently. Since r sin has been
fixed by and dne in the previous subsections, is approximately a function of MWR only. In
1
Fig. 12, we plot as a function of MWR for = arcsin 10
, r = 0.5 and sd ss = 1 which is required
by the neutron EDM calculation. [All si = 1.] Requiring that the new contribution should be no
274
1 , r = 0.5 and all s = 1. The shaded part is allowed by the experimental

Fig. 12. as a function of MWR for = arcsin 10
q
data.
Fig. 13. Tree level Feynman diagrams for Bd J /Ks from SM and LRSM.
| = 3.92 106 [11], we get a lower bound
larger than |expt
MWR > 4.2 TeV.
(83)
Here we obtain a slightly tighter bound on MWR than that from kaon mixing. However, because
of the I = 1/2 rule, the factorization assumption might have overestimated the phase of A2 . If
we take r sin = 0.15, as required by low MH , the bound changes to 7.4 TeV.
Finally, there are also tree-level FCNH contributions to HS=1 , as one can see from the Lagrangian in Eq. (48) [17,25,32]. Since the relevant coupling is suppressed by either the Cabibbo
angle or the quark masses, their contribution is negligible.
6.4. CP violation in Bd J /KS decay: SJ /K
The CP violation in B-meson decay was first observed in Bd J /KS . In SM, the decay
proceeds mainly through the tree-level b ccs
and the penguin contributions are expected to
be suppressed by CKM and/or loop factors. The tree-level diagram is shown in the left panel in
Fig. 13 with an intermediate WL exchange.
The relevant CP asymmetry is defined as
SJ /KS =
where
d =
2 Im d
,
1 + |d |2

q
A (Bd J /KS )
,
p B A (Bd J /KS )
A is a decay amplitude and (q/p)B is from the B-meson mixing.
(84)
275
The magnitude of d is close to 1 and thus SJ /KS Im d . In the SM, SJ /KS predominantly
comes from BB mixing, and is related to the angle of the unitary triangle because the ratio
of the decay amplitude is independent of the hadronic matrix element when the tree operator
dominates. Experimentally, sin 2expt = 0.673 0.028 [11].
In the LRSM, the effective angle will receive new contributions from initial and final neutral meson mixings [3] and from the new tree-level decay operators through WR exchange and
WR WL mixing, as shown in Fig. 13. We will not consider the kaon mixing contribution for
the following reason: Since KS is dominantly CP-even |KS
= p|K 0
+ q|K 0
and B 0 decay
involves K 0 , d is proportional to (q/p)K . The imaginary part of (q/p)K is proportional to the
imaginary part of which is known to be on the order of 103 , much smaller than the phase in
(q/p)B . The constraint on the LR symmetric model has already been studied independently
and we decouple the kaon-mixing effect from SJ /KS . Thus we write

M Bd ,LR
2 eff 2 + arg 1 + 12
B ,SM
M12d

J /K0 |H LR |B0
1
J / K 0 |H LR |B 0
1+
.
+ arg 1 +
(85)
J /K0 |H SM |B0
J / K 0 |H SM |B 0
The H LR operator is similar to that in Eq. (79), with substitutions s b, u c, and d s.

Bd ,LR
Bd ,SM
/M12
can be calculated from Eq. (58) and a corresponding exThe ratio M12
2 /M 2
pression from the SM. Its magnitude is around 102 MW
WR and carries a phase factor
L
R
R
L
L
i(
+
)
1
3
. The hadronic matrix element for the decay is less known. In
Vtd Vtb /Vtd Vtb = sd sb e
the naive factorization approximation for J / K 0 |H SM |B 0
, the decay rate is under predicted
by an order of magnitude [47]. Therefore, the non-factorization contribution must be significant
[48]. In the ratio of matrix elements, we expect the factorization approach work better. Using this
approximation, we find
J / K 0 |H LR |B 0
J / K 0 |H SM |B 0
2
MW
RR 2[S (mb )/S (MWL )]2/b [S (mb )/S (MWL )]4/b
= 2 L cLL
,
MW c 2[S (mb )/S (MWR )]2/b [S (mb )/S (MWR )]4/b
(86)
A B
RR LL
i(2 +3 ) . The non-perturbative
where AB
c = Vcs Vcb for A, B = L, R, so c /c = ss sb e
2
B K form factors F+ (mJ / ) has been canceled out and the result is independent of hadronic
2 /M 2 ) and hence is much smaller than the
parameters. The magnitude of the ratio is O(MW
WR
L
mixing contribution to 2eff . This conclusion remains valid even if we underestimate this ratio
by an order of magnitude.
The modified CP asymmetry in LRSM depends on the right-handed scale MWR and r sin .
We take r sin 0.05 as determined from the constraint. There are two independent choices
for quark mass signs which generate different predictions. We take either st = +1 or st = 1,
and the results are shown in Fig. 14. Demanding sin 2eff to lie within the experimental error bar,
we get a moderate lower bound on MWR ,
MWR > 0.7 TeV.
(87)
As we have commented above, this constraint comes predominantly from the Bd B d mixing
contribution.
276
Fig. 14. Predicted CP asymmetry in Bd0 J /KS as a function of MWR , for r sin = 0.05. The solid line corresponds
to the sign choice si = 1 and the dashed line corresponds to si = 1 except st = 1. The shaded part is allowed by the
experimental data.
The new spontaneous CP phase will contribute to many CP asymmetries in B-decays. In

particular, the contribution might become significant when the tree-level contribution in SM is
absent. We will leave this to a future study.
7. Conclusion
In this paper, we have made a comprehensive study of CP violating observables in the lowenergy sector of the minimal LRSM with the only assumption of parity invariance imposed on
the theory. This is made possible by an explicit solution for the right-handed quark mixing matrix with explicit dependence on the spontaneous CP violation phase . Although the hadronic
physics uncertainty is still large, the CP observables do provide significant and strong constraint
on the right-handed W -boson mass scale WMR and the FCNH mass scale MH . In fact, a new
experiment result and/or improved theoretical calculation on EDM might provide the strongest
bound yet on the right-handed gauge boson mass.
We stress the point that CP must be violated both explicitly and spontaneously in the minimal
model, with one bidoublet and two triplets. Up to O(3 ), we can write VRCKM in a compact form
as in Eqs. (27), (28) and (29). We find that VR has the same hierarchical structure as VL , i.e.
elements of the two mixing matrices are suppressed by the same orders of Cabibbo angle =
0.22. And because of the spontaneous CP phase , each element in VRCKM acquires a phase angle
proportional to r sin . Therefore, the phenomenology related to CP violation in the minimal
LRSM turns out to be very rich. We explored mass differences of neutral kaon and B-meson
with updated lattice results, and found an updated lower bound on right-handed W -boson mass:
MWR > 2.5 TeV. With the CP violating processes, we find a combined bound MWR > 4 8 TeV
from and neutron EDM constraints when the FCNH contribution is ignored. And we can also fix
r sin 0.05 from the combined bound. When the FCNH contribution is added to , the bound is
stronger when the MH becomes small. We go on to study and CP asymmetry in Bd J /KS
decay. By applying the experimental constraints and theoretical uncertainty, we conclude that the
lower bound on right-handed W -boson mass surviving from all above experimental constraint is
about 4 TeV.
If the right-handed gauge boson WR has a mass close to this lower bound, it is possible to
detect its presence in the up-coming LHC experiments. If the right-handed neutrino NR is lighter
than WR -boson, a typical event would consist of two isolated leptons and two jets carrying high
277
Table 2
A summary of bounds on MWR and MH from different physical observables. Stars on the items indicate input values.
For the neutron EDM case, the numbers in the parenthesis are direct result from Eq. (72) and those without are obtained
by reducing theoretical values by a factor of 5
mK
mBd
mBs
dne &

B J KS
MWR (TeV)
MH 0 , MA0 (TeV)
|r sin |
2.5
2.5
2.7
4 (8)
8 (20)
4.2
7.4
0.8
1.3
15
12
25
* 100
* 25
0.05
0.15
* 0.05
* 0.15
* 0.05
* 0.15
transverse momenta. Due to the Majorana nature of NR , half of the lepton pairs will have the
same sign. For an integral luminosity of 30 fb1 , there will be a 5 discovery of WR with MWR
up to 4 TeV and MNR up to 2.4 TeV [49].
We also find that the lower bound on MH 0 > 25 TeV is tighter than the bound previous bounds
1
[21,30]. Perhaps, this suggests that one should have two bidoublets so that one can invoke cancellation between them, as in the spontaneous CP violation model discussed in Ref. [50]. In that
case for our analytic solution to remain valid, the second bi-doublet should develop vev. All
bounds are shown in Table 2 for easy reading of the results of the paper.
Finally, we would like to comment on the constraint on r arising from the consideration of the
mass shift of SM Higgs boson in LRSM. According to discussions in Section 4, the SM Higgs
mass is

12 2
2
mh = 41
(88)
+ 3 2 vR2 ,
1
to second order in , and . The shift in mass due to LRS is 3 2 vR2
expressed in terms of the masses of FCNH
mh0 MH 0 .
1
12 2
1
and can be
(89)
From the discussions below Eq. (59), FCNH masses have to be large enough to suppress the tree
level contribution to kaon mixing. On the other hand, the SM Higgs mass should not exceed TeV
scale in order to preserve perturbative unitarity. A recent analysis [27] yields
m2h0 < 870 GeV.
(90)
The lower bound on the FCNH boson mass MH 0 > 25 TeV yields an upper bound on r =
1
mt /mb < 1.44. From discussions on CP violating observables, 0.05 < |r sin | < 0.15, this
translates into a lower bound on the spontaneous phase || > 0.035.
Acknowledgements
This work was partially supported by the US Department of Energy via grant DE-FG0293ER-40762. Y.Z. acknowledges the hospitality and support from the TQHN group at University
278
of Maryland and a partial support from NSFC grants 10421503 and 10625521. X.J. is supported
partially by a ChangJiang Scholarship at Peking University. R.N.M. is supported by NSF grant
No. PHY-0652363.
References
[1] R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 566;
R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 2558;
G. Senjanovic, R.N. Mohapatra, Phys. Rev. D 12 (1975) 1502;
G. Senjanovic, R.N. Mohapatra, Phys. Rev. D 23 (1981) 165;
For a review R.N. Mohapatra, in: C. Jarlskog (Ed.), CP Violation, World Scientific, 1989.
[2] T.D. Lee, Talk given at the Center for High-Energy Physics, Peking University, November 2006.
[3] R.N. Mohapatra, F.E. Paige, D.P. Sidhu, Phys. Rev. D 17 (1978) 2462;
D. Chang, Nucl. Phys. B 214 (1983) 435;
G.C. Branco, J.M. Frere, J.M. Gerard, Nucl. Phys. B 221 (1983) 317;
G.C. Branco, L. Lavoura, Phys. Lett. B 165 (1985) 327.
[4] P. Langacker, S. Uma Sankar, Phys. Rev. D 40 (1989) 1569;
G. Barenboim, J. Bernabeu, J. Prades, M. Raidal, Phys. Rev. D 55 (1997) 4213.
[5] K. Kiers, J. Kolb, J. Lee, A. Soni, G.H. Wu, Phys. Rev. D 66 (2002) 095002.
[6] G. Barenboim, M. Gorbahn, U. Nierste, M. Raidal, Phys. Rev. D 65 (2002) 095003.
[7] Y. Zhang, H. An, X. Ji, R.N. Mohapatra, Phys. Rev. D 76 (2007) 091301.
[8] P. Minkowski, Phys. Lett. B 67 (1977) 421;
M. Gell-Mann, P. Ramond, R. Slansky, in: P. van Nieuwenhuizen, et al. (Eds.), Supergravity, North Holland, Amsterdam, 1980, p. 315;
T. Yanagida, in: O. Sawada, A. Sugamoto (Eds.), Proceedings of the Workshop on the Unified Theory and the
Baryon Number in the Universe, KEK, Tsukuba, Japan, 1979, p. 95;
S.L. Glashow, The future of elementary particle physics, in: M. Lvy, et al. (Eds.), Proceedings of the 1979 Cargse
Summer Institute on Quarks and Leptons, Plenum Press, New York, 1980, p. 687;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. Lett. 44 (1980) 912.
[9] G. Lazarides, Q. Shafi, C. Wetterich, Nucl. Phys. B 181 (1981) 287;
R.N. Mohapatra, G. Senjanovic, Phys. Rev. D 23 (1981) 165.
[10] R.E. Marshak, R.N. Mohapatra, Phys. Lett. B 91 (1980) 222.
[11] W.M. Yao, et al., Particle Data Group, J. Phys. G 33 (2006) 1.
[12] N.G. Deshpande, J.F. Gunion, B. Kayser, F.I. Olness, Phys. Rev. D 44 (1991) 837.
[13] K. Kiers, M. Assis, A.A. Petrov, Phys. Rev. D 71 (2005) 115015.
[14] A. Masiero, R.N. Mohapatra, R.D. Peccei, Nucl. Phys. B 192 (1981) 66.
[15] J. Basecq, J. Liu, J. Milutinovic, L. Wolfenstein, Nucl. Phys. B 272 (1986) 145.
[16] G. Ecker, W. Grimus, Phys. Lett. B 153 (1985) 279;
G. Barenboim, J. Bernabeu, M. Raidal, Nucl. Phys. B 478 (1996) 527;
P. Ball, J.M. Frere, J. Matias, Nucl. Phys. B 572 (2000) 3.
[17] J.M. Frere, J. Galand, A. Le Yaouanc, L. Oliver, O. Pene, J.C. Raynal, Phys. Rev. D 46 (1992) 337.
[18] P. Duka, J. Gluza, M. Zralek, Ann. Phys. 280 (2000) 336, hep-ph/9910279.
[19] G. Beall, M. Bander, A. Soni, Phys. Rev. Lett. 48 (1982) 848.
[20] M. Neubert, Z. Phys. C 50 (1991) 243.
[21] R.N. Mohapatra, G. Senjanovic, M.D. Tran, Phys. Rev. D 28 (1983) 546;
G. Ecker, W. Grimus, H. Neufeld, Phys. Lett. B 127 (1983) 365;
G. Ecker, W. Grimus, H. Neufeld, Phys. Lett. B 132 (1983) 467, Erratum;
F.J. Gilman, M.H. Reno, Phys. Rev. D 29 (1984) 937;
S. Sahoo, L. Maharana, A. Roul, S. Acharya, Int. J. Mod. Phys. A 20 (2005) 2625.
[22] D. Chang, J. Basecq, L.F. Li, P.B. Pal, Phys. Rev. D 30 (1984) 1601;
W.S. Hou, A. Soni, Phys. Rev. D 32 (1985) 163;
J. Basecq, L.F. Li, P.B. Pal, Phys. Rev. D 32 (1985) 175.
[23] R. Babich, N. Garron, C. Hoelbling, J. Howard, L. Lellouch, C. Rebbi, Phys. Rev. D 74 (2006) 073009;
D.J. Antonio, et al., hep-lat/0702026.
[24] A.J. Buras, S. Jager, J. Urban, Nucl. Phys. B 605 (2001) 600.
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
[48]
[49]
[50]
279
G. Ecker, W. Grimus, Nucl. Phys. B 258 (1985) 328.

A.J. Buras, hep-ph/9806471.
A. Djouadi, hep-ph/0503172.
N. Tantalo, hep-ph/0703241.
D. Becirevic, et al., Nucl. Phys. B 634 (2002) 105, hep-ph/0112303;
D. Becirevic, V. Gimenez, G. Martinelli, M. Papinutto, J. Reyes, JHEP 0204 (2002) 025, hep-lat/0110091;
H. Wittig, Eur. Phys. J. C 33 (2004) S890, hep-ph/0310329.
M.E. Pospelov, Phys. Rev. D 56 (1997) 259, hep-ph/9611422.
G. Beall, A. Soni, Phys. Rev. Lett. 47 (1981) 552;
G. Ecker, W. Grimus, H. Neufeld, Nucl. Phys. B 229 (1983) 421;
J.M. Frere, J. Galand, A. Le Yaouanc, L. Oliver, O. Pene, J.C. Raynal, Phys. Rev. D 45 (1992) 259.
X.G. He, B.H.J. McKellar, S. Pakvasa, Phys. Rev. Lett. 61 (1988) 1267.
X.G. He, B. McKellar, Phys. Rev. D 47 (1993) 4055.
J. Donoghue, E. Golowich, B. Holstein, Dynamics of the Standard Model, Cambridge Univ. Press, Cambridge,
1994.
S.M. Barr, A. Zee, Phys. Rev. Lett. 65 (1990) 21;
S.M. Barr, A. Zee, Phys. Rev. Lett. 65 (1990) 2920, Erratum.
S. Weinberg, Phys. Rev. Lett. 63 (1989) 2333.
D. Chang, X.G. He, W.Y. Keung, B.H.J. McKellar, D. Wyler, Phys. Rev. D 46 (1992) 3876, hep-ph/9209284.
M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Sov. Phys. JETP 45 (1977) 670;
M.A. Shifman, A.I. Vainshtein, V.I. Zakharov, Zh. Eksp. Teor. Fiz. 72 (1977) 1275.
G. Buchalla, A.J. Buras, M.K. Harlander, Nucl. Phys. B 337 (1990) 313.
G. Buchalla, A.J. Buras, M.E. Lautenbacher, Rev. Mod. Phys. 68 (1996) 1125, hep-ph/9512380.
S. Bosch, A.J. Buras, M. Gorbahn, S. Jager, M. Jamin, M.E. Lautenbacher, L. Silvestrini, Nucl. Phys. B 565 (2000)
3, hep-ph/9904408.
T. Blum, et al., RBC Collaboration, Phys. Rev. D 68 (2003) 114506, hep-lat/0110075.
D. Pekurovsky, G. Kilcup, Phys. Rev. D 64 (2001) 074502, hep-lat/9812019.
H. Burkhardt, et al., NA31 Collaboration, Phys. Lett. B 206 (1988) 169;
V. Fanti, et al., NA48 Collaboration, Phys. Lett. B 465 (1999) 335, hep-ex/9909022;
A. Alavi-Harati, et al., KTeV Collaboration, Phys. Rev. Lett. 83 (1999) 22, hep-ex/9905060.
S. Bertolini, J.O. Eeg, M. Fabbrichesi, Phys. Rev. D 63 (2001) 056009, hep-ph/0002234.
A.J. Buras, M. Jamin, JHEP 0401 (2004) 048, hep-ph/0306217.
J. Chay, C. Kim, hep-ph/0009244;
H. Cheng, K. Yang, Phys. Rev. D 63 (2001) 074011, hep-ph/0011179.
C. Chen, H. Li, Phys. Rev. D 71 (2005) 114008, hep-ph/0504020.
S.N. Gninenko, M.M. Kirsanov, N.V. Krasnikov, V.A. Matveev, Phys. At. Nucl. 70 (2007) 441.
Y.L. Wu, Y.F. Zhou, arXiv: 0709.0042 [hep-ph].
NUCLEAR PHYSICS B
Journal devoted to the experimental and theoretical study of the fundamental
constituents of matter and their interactions
Editorial Board
Supervisory Editors: G. ALTARELLI, Geneva, Switzerland; W. BARTEL, Hamburg, Germany; C. BECCHI, Genova,
Italy; R.H. DIJKGRAAF, Amsterdam, The Netherlands; D. KUTASOV, Chicago, IL, USA; L. MAIANI, Rome, Italy;
H. MURAYAMA, Berkeley, CA, USA; T. NAKADA, Geneva, Switzerland; N. NEKRASOV, Bures-sur-Yvette, France;
H. OOGURI, CalTech, Pasadena, CA, USA; H. SALEUR, USC, Los Angeles, CA, USA; A. SCHWIMMER, Rehovot,
Israel
Associate Editors for Particle Physics: S.J. Brodsky, SLAC; M. Dine, UC, Santa Cruz; G. Dvali, New York; G.W.
Gibbons, Cambridge, UK; D. Gross, St. Barbara; S.S. Gubser, Princeton; L. Ibanez, Madrid; R.L. Jaffe, MIT; K. Kajantie,
Helsinki; M. Luescher, CERN; E. Rabinovici, Jerusalem; L. Randall, Princeton; A. Sen, Bombay; G. Steigman, Ohio
State; G. t Hooft, Utrecht; Y. Totsuka, KEK; P. van Baal, Leiden; S. Weinberg, Texas; J. Zinn-Justin, Saclay
Associate Editors for Field Theory and Statistical Systems: B. Duplantier, Saclay; A.W.W. Ludwig, Santa Barbara;
S. Lukyanov, Rutgers, Piscataway; T. Miwa, Kyoto; G. Mussardo, Trieste; M. Oshikawa, Tokyo; G. Parisi, Frascati;
A.M. Polyakov, Princeton; R. Sasaki, Kyoto; M. Zirnbauer, Cologne
Aims and Scope
Nuclear Physics B focuses on the domain of high energy physics and quantum field theory, and includes three main sections, i.e. particle physics, field theory and statistical systems and physical mathematics. The emphasis is on original
research papers. Conference Proceedings on the topics covered by Nuclear Physics B are published in the (separately
available) journal Nuclear Physics B-Proceedings Supplements.
Abstracted/Indexed in:
Chemical Abstracts/Current Contents: Physical, Chemical & Earth Sciences. Also covered in the abstract and citation
database SCOPUS . Full text available on ScienceDirect .
Subscription Information 2008
Volumes 788805 are scheduled for publication in 54 issues. Publication frequency: three times a month. Subscription
prices are available upon request from the Publisher or from the Regional Sales Office nearest you or from this journals
website (http://www.elsevier.com/locate/nuclphysb). A combined subscription to Nuclear Physics A ISSN 0375-9474
(volumes 798814), Nuclear Physics B ISSN 0550-3213 (volumes 788805) and Nuclear Physics B Proceedings Supplements ISSN 0920-5632 (volumes 174185) is available at a reduced rate. Subscriptions are accepted on a prepaid
basis only and are entered on a calendar year basis. Issues are sent by standard mail (surface within Europe, air delivery
outside Europe). Priority rates are available upon request. Claims for missing issues should be made within six months
of the date of dispatch.
Orders, claims, and journal enquiries: please contact the Customer Service Department at the Regional Sales Office
nearest you:
Orlando: Elsevier, Customer Service Department, 6277 Sea Harbor Drive, Orlando, FL 32887-4800, USA; phone: (+1)
(877) 8397126 [toll free number for US customers], or (+1) (407) 3454020 [customers outside US]; fax: (+1) (407)
3631354; e-mail: usjcs@elsevier.com
Amsterdam: Elsevier, Customer Service Department, PO Box 211, 1000 AE Amsterdam, The Netherlands; phone:
(+31) (20) 4853757; fax: (+31) (20) 4853432; e-mail: nlinfo-f@elsevier.com
Tokyo: Elsevier, Customer Service Department, 4F Higashi-Azabu, 1-Chome Bldg, 1-9-15 Higashi-Azabu, Minato-ku,
Tokyo 106-0044, Japan; phone: (+81) (3) 5561 5037; fax: (+81) (3) 5561 5047; e-mail: jp.info@elsevier.com
Singapore: Elsevier, Customer Service Department, 3 Killiney Road, #08-01 Winsland House I, Singapore 239519;
phone: (+65) 63490222; fax: (+65) 67331510; e-mail: asiainfo@elsevier.com
Advertising information
Europe, USA, Canada and ROW: Advertising orders and enquiries can be sent to: Miss Katrina Barton, phone: (+44)
(0) 20 7611 4117; fax: (+44) (0) 20 7611 4463; e-mail: commercialsales@elsevier.com.
South America: Advertising orders and enquiries can be sent to: Mr Tino DeCarlo, The Advertising Department,
Elsevier Inc., 360 Park Avenue South, New York, NY 10010-1710, USA; phone: (+1) (212) 633 3815; fax: (+1) (212)
633 3820; e-mail: t.decarlo@elsevier.com.
US mailing notice Nuclear Physics B (ISSN 0550-3213) is published weekly by Elsevier B.V., P.O. Box 211, 1000
AE Amsterdam, The Netherlands. The annual institutional subscription price in the USA is US$ 13612 (valid in North,
Central and South America), including air speed delivery. Periodical postage paid at Rahway NJ and additional mailing
offices. USA Postmaster: Send change of address: Nuclear Physics B, Elsevier, 6277 Sea Harbor Drive, Orlando, FL
32887-4800. Airfreight and mailing in the USA by Mercury International Limited, 365, Blair Road, Avenel, NJ 07001.
Nuclear Physics B 802 [FS] (2008) 281306

Shear viscosity of boost invariant plasma

at finite coupling
Alex Buchel a,b,
a Department of Applied Mathematics, University of Western Ontario, London, Ontario N6A 5B7, Canada
b Perimeter Institute for Theoretical Physics, Waterloo, Ontario N2J 2W9, Canada
Received 30 January 2008; accepted 17 March 2008

Available online 20 March 2008
Abstract
We discuss string theory corrections in the dual description of the expanding boost invariant N = 4
supersymmetric YangMills plasma at strong coupling. We compute finite t Hooft coupling corrections to
the shear viscosity and find that it disagrees with the equilibrium correlation function computations. We
comment on the possible source of the discrepancy.
1. Introduction
Recent work by Janik and collaborators [13] (see also [4]) opened a possibility to study nonstationary/non-equilibrium processed in gauge theories using their dual string theory formulation
[5,6]. In fact, one might try to use gauge theory/string theory correspondence to define nonequilibrium quantum field theory dynamics.1 For example: in principle, one can use AdS/CFT
correspondence to extend MullerIsraelStewart theory of dissipative relativistic hydrodynamics
[8,9] to arbitrary order in deviation from the equilibrium.
Even though the framework proposed in [1] cannot be consistently implemented within a supergravity approximation of the string theory dual to N = 4 supersymmetric YangMills (SYM)
plasma in the Bjorken flow [10] (see [11]), if correctly reproduces the equilibrium and near* Corresponding author at: Department of Applied Mathematics, University of Western Ontario, London, Ontario N6A
5B7, Canada.
E-mail address: abuchel@uwo.ca.
1 For an approach alternative to [1] see [7].
doi:10.1016/j.nuclphysb.2008.03.009
282
A. Buchel / Nuclear Physics B 802 [FS] (2008) 281306
equilibrium properties of the N = 4 SYM plasma obtained from the equilibrium correlation
functions, i.e., the equilibrium equation of state [12], its shear viscosity at infinite t Hooft coupling [13], and its relaxation time2 [17]. Furthermore, this agreement appears to be unaffected by
the subtleties associated with the breakdown of the supergravity approximation. Given substantial potential of the Janiks framework,3 we believe it is important to further explore its regime
of validity.
As we already mentioned, the shear viscosity of the N = 4 plasma at infinite coupling is a
robust prediction within Janiks framework [2]. In this paper we compute finite t Hooft coupling
corrections (corresponding to string theory corrections in the dual formulation) to the N = 4
plasma. Since physically equivalent results extracted from the equilibrium correlation functions
are already available [18,19], our analysis provides a test of the framework [1] beyond the supergravity approximation. We find that the finite coupling corrections to the shear viscosity of
the N = 4 plasma extracted from the string theory dual to its Bjorken flow disagrees with the
equilibrium correlation function computations.
The paper is organized as follows. In the next section we review the results of [18,19], the dual
string theory model within which such computation were performed, and emphasize the selfconsistency of the obtained results.4 We further summarize our new results for shear viscosity
at finite t Hooft coupling. In Section 3 we describe our string theory computational framework
introduced in [20]. The analysis is incredibly complicated and thus we focus on explaining the
consistency checks on our analysis (fortunately, there are lots of such consistency checks). In
Section 4 we summarize equations of motion describing correction to the supergravity background dual to late-time Bjorken flow of the N = 4 plasma. We also present analytic solutions
to these equations. In Section 5 we map the results of the string theory computation to the nearequilibrium MullerIsraelStewart (MIS) theory and extract the leading correction to the SYM
plasma shear viscosity. We conclude in Section 6. Appendices A and B contain some technical
details; further computation details are available from the author upon request.
2. Shear viscosity of the N = 4 SYM plasma at finite t Hooft coupling
Consider N = 4 SU(nc ) SYM plasma in the (t Hooft) large-nc limit: we take the YangMills
2 0 and n in such a way that the t Hooft coupling ,
coupling gYM
c
2
nc ,
= gYM
(2.1)
remains finite. We are interested in the transport properties of the plasma at strong coupling, i.e.,
1. In this limit the useful description of the N = 4 plasma is in terms of holographically dual
background of near-extremal black 3-brane geometry in the supergravity approximation of the
type IIb string theory. The supergravity approximation is exact at = , in which case one finds
[13,21,22] the speed of sound cs , the shear viscosity , and the bulk viscosity correspondingly
1
cs = ,
3
2 3
n T ,
8 c
= 0.
2 Related work appeared in [1416].

3 One can formulate and study questions such as dynamical hadronization in this approach.
4 For a recent review of the Janiks framework and its clarification see [11].
(2.2)
283
In the supergravity approximation the entropy density s is

2 2 3
n T ,
2 c
leading to the viscosity-to-entropy ratio
s=
(2.3)
1
(2.4)
=
.
s
4
In a hydrodynamic approximation to near-equilibrium dynamics of hot gauge theory plasma
there are several distinct ways to extract the transport coefficients (2.2). First [13], the shear
viscosity can be computed from the two-point correlation function of the stress-energy tensor at
zero spatial momentum via the Kubo formula

1
dt d x eit Txy (x), Txy (0) .
= lim
(2.5)
0 2
Second [21], diffusive channel two-point retarded correlation function of the stress energy tensor,
for example,

1
Gtx,tx (, q) = i dt d x eit+iqz (t) Ttx (x), Ttx (0)
(2.6)
,
i Dq 2
have a pole at
= iDq 2 ,
(2.7)
where the shear diffusion constant D is
.
D=
(2.8)
sT
Finally, all the transport coefficients (2.2) can be read off from the dispersion relation of a pole
in the sound wave channel two-point retarded correlation function of the stress energy tensor, for
example,

Gtt,tt (, q) = i dt d x eit+iqz (t) Ttt (x), Ttt (0) ,
(2.9)
as
(q) = cs q i

2q 2
3
1+
.
3T s
4
(2.10)
In [18], using the Kubo formula, the finite t Hooft coupling correction to the ratio (2.4) was
computed. Such corrections were further extracted from the two-point retarded correlation functions (2.6) and (2.9) in [19]. Much like as for the infinite t Hooft coupling, all three approaches
led to the same result:

1
135
3/2
+ .
=
1+
(3)
(2.11)
s
4
8
The computation leading to (2.11) were performed in the supergravity approximation to type
IIb string theory including the leading correction

1
1
1
10
2
2
32
d x g R ()
W + , (2.12)
I=
(F5 ) + + e
16G10
2
4 5!
284
where
1
= (3)( )3 ,
8
corresponding to
1
= (3)3/2
8
in terms of the SYM t Hooft coupling , and
1
W = C hmnk Cpmnq Ch rsp C q rsk + C hkmn Cpqmn Ch rsp C q rsk .
2
(2.13)
(2.14)
A somewhat uncontrollable approximation5 used in [18,19] (and earlier in [23]) was an assumption that in a chosen scheme (2.12), the self-dual F5 form does not receive order ( )3 corrections.
Though the agreement of the shear viscosity extracted from different boundary correlation functions at order O( ) computationally appears to be rather dramatic,6 a recent work [24] suggests
that such an agreement might be automatically encoded in any gravitational model, perhaps similar to the way the thermodynamic relation between the energy (E), free energy (F ) and the
entropy densities F = E T s is encoded in the horizon geometries [25].
In [2] the ratio of the shear viscosity to the entropy density at infinite t Hoof coupling was
computed for the Bjorken flow of the N = 4 SYM plasma, and the result was found to agree
with (2.4). In this paper we compute leading order finite t Hooft coupling correction to /s for
the N = 4 plasma in the Bjorken flow. We use string theory effective action (2.12), which is
expected to reproduce (2.11) obtained from the equilibrium correlation function computations.
Unfortunately, we report here a disagreement with (2.11):

1
120
3/2
(2.15)
=
+ .
1+
(3)
s Bjorken flow 4
8
In principle, analysis is very straightforward:
we compute corrections to the gravitational background dual to boost invariant Bjorken
expansion on the boundary;
as in [2], we fix some of the parameters7 of the corrected geometry by requiring nonsingularity of the string theory background to order O( 4/3 ) in the late proper time expansion;
we use corrected holographic renormalization (see [20]) to evaluate the energy density of
the expanding N = 4 plasma;
as in [11] we interpret the string theory computation in the framework of the MullerIsrael
Stewart theory of dissipative relativistic hydrodynamics.8
5 This is related with difficulties of deriving corrected string theory effective action in the presence of nontrivial RR
fluxes.
6 For example, using (2.9), (2.10) one has to take into account corrections to the black three brane temperature [23].
7 Amusingly, not all parameters can be fixed from the nonsingularity condition [1]. Despite this, our answer for the
viscosity ratio (2.15) is unambiguous.
8 In [17] a correction to Bjorken expansion within MIS theory was pointed out. Such a correction affects the secondorder hydrodynamic description only and thus though it will modify the interpretation of the plasma relaxation time, it
will not change the computation of its shear viscosity (determined by first-order hydrodynamics) compare to the analysis
of [11].
285
Given the computational complexity of the analysis and the disagreement we claim to exist
between equilibrium correlation function computation and the Bjorken flow computation of the
plasma shear viscosity at finite t Hooft coupling, the remainder of the paper focuses on explaining the consistency checks we performed.
3. Computational framework
The most difficult step in the analysis is to obtain -corrected supergravity equations governing the SYM Bjorken flow at finite t Hooft coupling. As in [20], we first derive effective action
for the warp factors of the gravitation background, and then determine -corrected equations of
motion. As explained in [11], choosing a Bjorken frame, fixes reparametrization of a radial and
a proper time coordinates, which leads to two first order differential constraints (at each order in
the late proper time expansion) on the warp factors. Not only these constraints are important consistency checks on the obtained system of equations, but they are also crucial to pick up correct
solutions.9 If one naively obtains effective action for the warp factors, the constraints are lost. We
explain how to properly derive effective action, so that the reparametrization constraints are kept.
As in the supergravity approximation, the corrected constraints will provide vital consistency
checks.
3.1. Supergravity approximation: Effective action, constraints and the solution
The supergravity background holographically dual to a Bjorken flow of the N = 4 plasma
takes form [1,11]
2
d s10
= g MN d M d N

2
= e2(,z) g (x) dx dx + e6/5(,z) dS 5 ,
(3.1)
where (dS 5 )2 is the line element for a 5-dimensional sphere with unit radius, and
ds 2 = g dx dx
=
dz2
1 2a(,z) 2
2
e
+ 2 ,
d + e2b(,z) 2 dy 2 + e2c(,z) dx
2
z
z
(3.2)
2 dx 2 + dx 2 . The 5-form F takes form10

where dx
5
1
2
F5 = F5 + F5 ,
F5 = 4S 5 ,
(3.3)
where S 5 is the 5-sphere volume form. Moreover, the dilaton is = (, z).

Notice that choosing the five-dimensional metric as in (3.2) fixes the (, z) reparametrization
invariance so that
gzz =
1
,
z2
g z = 0.
9 These constraints fix some of the integration constants.

10 We normalize the five-form flux so that the asymptotic AdS radius is one.
(3.4)
286
Substituting the ansatz (3.1)(3.3) into type IIb supergravity action11 (the = 0 approximation
of (2.12)) one obtains naive effective action for the scalars {a, b, c, , }:

Seff,naive a(, z), b(, z), c(, z), (, z), (, z) = I =0 .
(3.5)
Clearly, from (3.5) one obtains only five second order PDEs, instead of seven, as in [11]. The
reason that is happening is because the reparametrization constraints (3.4) are imposed directly
on the action, instead on the equations of motion.
In order the keep constraints within the effective action approach we deform the five-dimensional metric (3.2) as follows
ds 2 = g dx dx

1
2
= 2 e2a(,z) d 2 + e2b(,z) 2 dy 2 + e2c(,z) dx
z

dz2
+ h(, z) d dz + 1 + g(, z) 2 .
z
(3.6)
Constraints (3.4) now correspond to

h(, z) 0,
g(, z) 0.
Next, we evaluate the supergravity action on (3.6), (3.3)

Seff [a, b, c, , ; h, g] = I =0 .
(3.7)
(3.8)
While it is important to maintain full nonlinear contributions for the physical scalars
{a, b, c, , } in Seff , it is sufficient to evaluate (3.8) to linear order in {h, g}. The equations
of motion derived from Seff , followed by fixing the reparametrization invariance as in (3.7)

Seff
Seff
,
0
=
,
0=
a(, z) h(,z)=g(,z)=0
b(, z) h(,z)=g(,z)=0

Seff
,
0=
c(, z) h(,z)=g(,z)=0

Seff
Seff
,
0=
,
0=
(, z) h(,z)=g(,z)=0
(, z) h(,z)=g(,z)=0

Seff
Seff
,
0
=
,
0=
(3.9)
h(, z) h(,z)=g(,z)=0
g(, z) h(,z)=g(,z)=0
are precisely equivalent to the full set of equations derived in [11] (see Eqs. (3.12)(3.18)).
Eqs. (3.9) are solved as a series expansion in the late proper time , but exactly in the
scaling variable [1]
v
z
1/3
(3.10)
11 As usual, care should be taken with the self-dual five-form. The correct contribution to the effective action from the
1 F 2 = 8e6(,z) , see [11].
five-form is 45!
5
287
We find it useful (especially when we generalize our effective action approach to the full action (2.12)) to rewrite Seff as a functional of

a(, v), b(, v), c(, v), (, v), (, v), h(, v), g(, v) ,
i.e., fields depending on the scaling variable (3.10). When we do this, it is important to note the
modification of the effective action measure
dt dz t 1/3 dt dv .
Solving (3.9) in the scaling variable v and with the boundary conditions

a(, v), b(, v), c(, v), (, v), (, v) v0 = 0,
(3.11)
we find [1,2]

a2 (v) + O
4/3
b2 (v) + O
4/3
a1 (v) +
2/3
b(, v) = b0 (v) +
b1 (v) +
2/3

2 ,
a(, v) = a0 (v) +

2 ,

1
1
c(, v) = c0 (v) + 2/3 c1 (v) + 4/3 c2 (v) + O 2 ,

(, v) = O 2 ,
(, v) = O 2 ,
(3.12)
with

1 (1 v 4 /3)2
1
,
b0 = c0 = ln 1 + v 4 /3 ,
ln
4
2
2
1 + v /3
4
4
4
(9 v )v
v
3 v 4
a1 =
,
c
=
,
ln
1
9 v8
3 + v4 2 3 + v4
v4
2c1 ,
b1 = 3
3 + v4
a0 =
a2 =
(3.13)
(3.14)
(9 + v 4 )v 4
(1053 171v 4 + 9v 8 + 7v 12 )v 4
(9 + 5v 4 )v 2
C
+ 2
8
8
12(9 v )
72(9 v )
6(9 v 8 )2
3 v2 3 2 3 v4
1
+ ln
ln
,
3 + v4
8 3
3 + v2 4
4
8 4
v4
v 2 (9 + v 4 )
2
2 (9 + 54v + 7v )v
+
C
+
72(3 + v 4 )
6(3 + v 4 )(9 v 8 )
288 3 12(9 v 8 )
3 v4
3 v2
1
1
+
C + 66 2 ln
+ ln
3 + v4
8 3
3 + v 2 72

3 v 2 ( 3 v 2 )( 3 + v 2 )3
1
( 3 v 2 )2
+ ln
ln
li2
,
4(3 + v 4 )2
24 3
3 + v2
( 3 + v 2 )2
4 4
v4
3 v2
v2
1
2 (39 + 7v )v
+
C
+
ln
+
b2 = 2c2 +
4
4
4
2
4(3 + v )
24(3 + v )
2(3 + v )
8 3
3 + v2
c2 =
3
3 v4
+ 2 ln
,
4
3 + v4
for arbitrary parameters {,
C}.
(3.15)
288
3.2. Beyond the supergravity approximation

To go beyond the supergravity approximation we deform proper late time expansion for the
physical fields (3.12) as

1
1
a(, v) = a0 (v) + a 0 (, v) + 2/3 a1 (v) + a 1 (, v) + 4/3 a2 (v) + a 2 (, v) ,

1
1
b(, v) = b0 (v) + b0 (, v) + 2/3 b1 (v) + b1 (, v) + 4/3 b2 (v) + b2 (, v) ,

1
1
c(, v) = c0 (v) + c0 (, v) + 2/3 c1 (v) + c1 (, v) + 4/3 c2 (v) + c2 (, v) ,
1
1
(, v) = 0 (, v) + 2/3 1 (, v) + 4/3 2 (, v),
1
1
(, v) = 0 (, v) + 2/3 1 (, v) + 4/3 2 (, v),
(3.16)
where we assumed that as

a i (, v) a i (, v),
i (, v) i (, v),
bi (, v) bi (, v),
i (, v) i (, v),
ci (, v) ci (, v),
i = 0, 1, 2.
(3.17)
Substituting (3.16) into (2.12) we can obtain effective action for

{a i , bi , ci , i , i }.
(3.18)
We need to expand the supergravity part of (2.12) to quadratic order in the fields (3.18), while
it is sufficient to evaluate the e3/2 W term in (2.12) to linear order in (3.18). As explained in
the previous section, in order to obtain reparametrization constraints we further need to evaluate
the whole action (2.12) to linear order in {h, g}. Resulting effective action can be organized as
follows

Seff [a i , bi , ci , i , i ; h, g; ,
C] = d dv Leff [a i , bi , ci , i , i ; h, g; ,
C],

1/3
1/3
3/3
3/3
5/3
5/3
Leff = 2 L[2] + L[1]
+ 2 L[2] + L[1]
+ 2 L[2] + L[1] , (3.19)
where, given (3.17), the superscript in L indicates the leading scaling behaviour as , i.e.,

Lq = O q .
(3.20)
Notice that we suppressed explicit dependence on {,

C} in the decomposition of Leff .
In the rest of this subsection we present further decomposition of Lq , and explain the structure
of the consistency checks on the example of the equation of motion for a 2 .
In the following expressions all Ls do not scale as the explicit scaling dependence is factored out. A superscript indicates whether a term comes from the supergravity part
of (2.12), or from the O( )-correction in (2.12); the subscripts indicate order of fields on which
a Lagrangian term depends. For example, the subscript in L(2,1) implies that L(2,1) is strictly
bilinear in order-2 and order-1 fields:

L(2,1) = L(2,1) (a 2 , b2 , c2 , 2 , 2 ), (a 1 , b1 , c1 , 1 , 1 ) ,
while the subscript in L(1,1) implies that such a term is strictly quadratic in order-1 fields

L(1,1) = L(1,1) (a 1 , b1 , c1 , 1 , 1 ), (a 1 , b1 , c1 , 1 , 1 ) .
289
A single subscript indicates the order of fields on which the corresponding Lagrangian term
depends linearly. Finally, no subscript indicates that the Lagrangian terms are independent of
(3.18), but are linear in either h or g. Notice that Lagrangian terms with a single subscript coming
from the supergravity part of (2.12) are necessarily linear12 in either h or g. The latter implies that
reparametrization constraints at leading order in appear as O( ), and thus in the Lagrangian
terms which are either bilinear or quadratic in the fields (3.18) we can set h = g = 0. For the
same reason we can set h = g = 0 in the Lagrangian terms of order O( 2 ) coming from the
O( )-correction in (2.12).
3.2.1. Order-0
We find
1/3
L[2]
1/3
L[1]
=
=
1 [1/3]
1/3

LSUGRA
+ [1/3] LW
(0,0)
(0) ,
1 [1/3] SUGRA [1/3] W

L(0)
+
L .
1/3
(3.21)
3.2.2. Order-1
We find
1 [3/3] SUGRA [3/3] W
L(0,1) +
L(1)

1
+ 5/3 [5/3] LSUGRA
+ [5/3] LSUGRA
+ [5/3] LW
(0,1)
(1,1)
(1) ,

1
3/3
+ [3/3] LSUGRA
+ [3/3] LW .
L[1] = [3/3] LSUGRA
(0)
(1)
3/3
L[2]
(3.22)
(3.23)
3.2.3. Order-2
We find
5/3
L[2]
1 [5/3]
5/3
LSUGRA
+ [5/3] LW
(0,2)
(2)
1 [7/3] SUGRA [7/3] SUGRA [7/3] W

L(0,2) +
L(1,2) +
L(2)
7/3

1
+ [9/3] LSUGRA
+ [9/3] LSUGRA
+ [9/3] LW
+ 9/3 [9/3] LSUGRA
(0,2)
(1,2)
(2,2)
(2) ,
1 [5/3] SUGRA [5/3] SUGRA [5/3] SUGRA [5/3] W

5/3
L(0)
+
L(1)
+
L(2)
+
L .
L[1] = 5/3
3.2.4. Equations of motion for a 2 and consistency check on lower order solution
Consider a variation

.
dt dv Leff
EOM[a 2 ]
a 2 (, v)
h=g=0
(3.24)
(3.25)
(3.26)
12 This is simply a check (which we verified to be true) that the supergravity background (3.13)(3.14) solves supergravity equations of motion.
290
Given (3.19), (3.24), we have13

EOM[a 2 ] =
1 [5/3]
5/3
+
I(0) + [5/3] I
1 [7/3]
I(0) +
[7/3]
I(1) + [7/3] I
7/3

1
+ 9/3 [9/3] I(0) + [9/3] I(1) + [9/3] I(2) + [9/3] I ,
with obvious notations. For example:

1 [7/3] SUGRA
1 [7/3]
dt
dv
I
=
L
.
(0)
(0,2)
a 2 (, v)
7/3
7/3
(3.27)
(3.28)
Thus, a single variation (3.26) not only will produce the equation of motion for a 2 , namely
0 = [9/3] I(0) + [9/3] I(1) + [9/3] I(2) + [9/3] I,
(3.29)
but also would produce a consistency tests on the order-0 and order-1 solutions:
0 =[5/3] I(0) + [5/3] I,
0 = [7/3] I(0) + [7/3] I(1) + [7/3] I,
(3.30)
correspondingly. It is easy to understand the origin of consistency constraints (3.30): a field

deformation at order-2 can be considered as a variation of a field deformation at order-0 or
order-1; thus, once equations of motion at lower orders are solved, such variations must vanish.
Altogether, there are 10 consistency constraints when deriving equations for order-2 field deformations, and 5 consistency constraints when deriving equations for order-1 field deformations.
We explicitly verified that all such constraints are satisfied.
4. Equations of motion and solutions for (3.18)
We obtain equations of motion (including the constraints) at each order i = 0, 1, 2 for (3.18)
as explained in the previous section. It is important that in deriving these equations of motion a
field is assumed to depend both on {, v}. Once the equations of motion are computed, we can
assume
{a i , bi , ci , i , i }(, v) {a i , bi , ci , i , i }(v).
All the equations must be solved with the boundary conditions

a i (v), bi (v), ci (v), i (v), i (v) v0 = 0.
4.1. Order-0
To leading order we find the following system of equations
13 The term [5/3] LSUGRA does not contribute as it can only be bilinear in both a and either h or g.
2
(2)
(4.1)
1
5v 4 9
5v 4 9
c0 +
0 = c0 + b0 +
b
4
2
(3 + v )v
2(3 + v 4 )v 0
933120(4v 8 33v 4 + 36)v 10
+
,
(3 + v 4 )8
1
5v 4 + 9
5v 8 + 27
a
c0
0 = c0 + a 0 +
+
2
2(v 4 3)v 0 v(v 8 9)
311040v 10 (4v 8 45v 4 + 36)
,
(3 + v 4 )8
5v 4 + 9
5v 8 + 27
5v 8 + 27
a 0 +
c
0 = c0 + a 0 + b0 + 4
b0 +
8
(v 3)v
v(v 9)
v(v 8 9) 0
622080v 10 (4v 8 45v 4 + 36)
,
(3 + v 4 )8
1
3(v 8 10v 4 3) 3(5v 4 3)
3(v 4 3)
c
a
0 = c0 + b0 +
b
0
0
2
v(v 8 9)
v(v 8 9)
2(3 + v 4 )v 0
3732480v 10 (v 4 3)2
+
,
(3 + v 4 )8
1
(v 4 3)2
a
0 = c0 + b0 +
4
2
2
2(v + 2v + 3)(v 4 2v 2 + 3) 0
2799360(v 4 3)v 15
4
,
(v 2v 2 + 3)(v 4 + 2v 2 + 3)(3 + v 4 )7
5v 8 + 27 32
2332800v 14
a
+
,
0 = 0 +
0
0
v(v 8 9)
v2
(3 + v 4 )8
5v 8 + 27 5598720v 14
0 = 0 +
.

v(v 8 9) 0
(3 + v 4 )8
291
(4.2)
(4.3)
(4.4)
(4.5)
(4.6)
(4.7)
(4.8)
Although the system (4.2)(4.8) is overdetermined (we have 7 ODEs for the 5 functions), it is
straightforward to verify that it is consistent.
Most general solution to (4.2)(4.8), subject to the boundary conditions (4.1), is parametrized
by three arbitrary integration constants14 {1 , 1 , 2 }:
72576
585144
216
1714608
864
1714608
4
3
4
4
4
4
5
4
2
(3 + v )
(3 + v )
3+v
(3 + v )
(3 + v )
(3 + v 4 )6

1
1
1
+
+ 4
(288 + 1 ) + 24 + 1 ,
4
12
12 + 4v
2v 6
223560
25920
664848
664848
b0 =
(3 + v 4 )4 (3 + v 4 )3 (3 + v 4 )6 (3 + v 4 )5
1
1
+ 24 + 1 ,
12
4(3 + v 4 )
c0 = b0 ,
a 0 =
(4.9)
(4.10)
(4.11)
14 Altogether we expect 10 integration constants; setting nonnormalizable components of the fields to zero fixes 5
integration constants; the two constraints fix another 2 integration constants.
292

1
9
1
25 4
v4
3 v4
3 v 4 75
+
+
+
+
+
v
ln
ln
4
4
4
864 96v
144
16
8
3+v
v
3 + v4
225
675
22275
36450
36450
+
+
,
2(3 + v 4 )2 (3 + v 4 )3 2(3 + v 4 )4 (3 + v 4 )5 (3 + v 4 )6

405
2 45
135
3 v4
0 =
ln
24
4
3 + v 4 2(3 + v 4 ) 2(3 + v 4 )2
810
25515
87480
87480
+
.
(3 + v 4 )3 (3 + v 4 )4 (3 + v 4 )5 (3 + v 4 )6

0 = 1
(4.12)
(4.13)
For a generic choice of 1 the curvature invariants of the ten-dimensional geometry (3.1) will
1/4
be singular as v 3 . For example,

14
(1350 + 1 ) ln 3 v 4 + finite.
3
However, once we chose
R R =
(4.14)
1 = 1350,
(4.15)
the curvature invariant (4.14) is finite. We verified that for (4.15) all other curvature invariants
(including those discussed in [11]) are finite.
Notice that unlike studies of the string theory dual of the Bjorken flow in the supergravity
approximation [13,11], here the background geometry is not completely fixed from the nonsingularity conditiontwo arbitrary integration constants {1 , 2 } remain. Though 2 does not
affect the holographic stress energy tensor of the background geometry, the latter does depend
on 1 . Additional arbitrariness will appear at higher orders, however, the string theory computation when properly matched to the near-equilibrium SYM dynamics provides unambiguous
predictions for plasma transport properties. We further comment on the physical origin of the
arbitrariness in Section 5.
4.2. Order-1
Using results of the previous subsection, including (4.15), we find the following set of equations for the next-to-leading order in the late proper time expansion at order O( )
1
5v 4 9
5v 4 9
0 = c1 + b1 +
c1 +
b + S(1,1) ,
4
2
(3 + v )v
2(3 + v 4 )v 1
1
5v 4 + 9
5v 8 + 27
a 1 +
c + S(1,2) ,
0 = c1 + a 1 +
4
2
2(v 3)v
v(v 8 9) 1
5v 4 + 9
5v 8 + 27
5v 8 + 27
a 1 +
c + S(1,3) ,
0 = c1 + a 1 + b1 + 4
+
b
1
(v 3)v
v(v 8 9)
v(v 8 9) 1
1
(v 4 15)v 3 5v 8 30v 4 27
c1
0 = c1 + b1 +
b1 +
2
v8 9
v(v 8 9)
3(v 4 3)
24v 2
48v 2
a
c1 + S(1,4) ,
b
1
2(3 + v 4 )v 1 v 8 9
v8 9
(v 4 3)2
1
a 1 + b1 + S(1,5) ,
0 = c1 +
4
2
4
2
2
2(v 2v + 3)(v + 2v + 3)
(4.16)
(4.17)
(4.18)
(4.19)
(4.20)
293
5v 8 + 27 32
1 + S(1,6) ,
(4.21)
v(v 8 9) 1 v 2
5v 8 + 27
0 = 1 +
(4.22)
+ S(1,7) ,
v(v 8 9) 1
where the source terms {S(1,1) , . . . , S(1,7) } are given in Appendix A. As before, while the system
(4.16)(4.22) is overdetermined, we explicitly verified that it is consistent.
Most general solution to (4.16)(4.22), subject to the boundary conditions (4.1), is parametrized by three new arbitrary constants {2 , 3 , 4 }:
0 = 1 +
72013536 61725888 31189536 3(288
+ 1 )
+
+
+
4
6
4
7
4
5
4
(3 + v )
(3 + v )
(3 + v )
(v 3)2
2
1
5987520 3(1728
+ 1 ) 576 + 3 2
3 2 576
+
+
4
4
4
2
4
(3 + v )
2(3 + v )
2(v 3)
2(3 + v 4 )
1
2
445824

1+ ,
+
4
3
6
18
(3 + v )

1 + 62
1
2
3 v 4 864 12
1+
ln
b1 = 120
6
18
3 + v4
3 + v4
a 1 =
(4.23)
1
144 + 12
3024
3
1
11687328
107136
4
4
2
4
2
4
5
v 3
(3 + v )
2(3 + v )
(3 + v )
(3 + v 4 )3
23934528 2091744 27597024 1
2
+
+

1+ ,
4
7
4
4
4
6
6
18
(3 + v )
(3 + v )
(3 + v )

1
4
72
+
2
1
3v
4 1

1 ln
c1 = 60 +
36 12
3 + v4
(v 4 3)
(4.24)
1 1
432 62 14
179712
23934528

1
4
4
3
6
(v + 3)
(3 + v )
(3 + v 4 )7
2336688 28086912 (1512 + 32 1 ) 12177216
2
+
+
+
+ ,
18
(3 + v 4 )4
(3 + v 4 )6
(3 + v 4 )2
(3 + v 4 )5

4

1
9
1
25 4
v
3 v4
3 v4
1 = 3
+
+
+
+
v
ln
ln
864 96v 4
4
3 + v 4 144
v4
3 + v4
(4.25)
75 450v
8 (v 4 3)(v 8 + 24v 4 + 9)
+
,
(4.26)
+
2
(3 + v 4 )7

2
4 2
2
3 v4
1 =
ln
4
4
24
12
3+v
4(3 + v ) 4(v 4 3)
135
405
135
1487160 3149280
+
4
5
4
7
4
2
4
(3 + v )
(3 + v )
(3 + v )
2(3 + v ) 2(v 4 3)
218700
3674160
2430
+
+
.
(4.27)
4
3
4
4
(3 + v )
(3 + v )
(3 + v 4 )6
For a generic choice of 3 the curvature invariants of the ten-dimensional geometry (3.1) will
1/4
be singular as v 3 . For example,
R R =

1 14
ln 3 v 4 + finite.
(3 5400)
2/3 3
(4.28)
294
However, once we chose

3 = 5400
(4.29)
the curvature invariant (4.28) is finite. We verified that for (4.29) all other curvature invariants
(including those discussed in [11]) are finite.
At this stage we have four arbitrary integration constants: {1 , 2 , 3 , 4 }.
4.3. Order-2
Using results of the previous two subsections, including (4.15) and (4.29), we find the following set of equations for the next-to-next-to-leading order in the late proper time expansion at
order O( )
1
5v 4 9
5v 4 9
0 = c2 + b2 +
(4.30)
c2 +
b + S(2,1) ,
4
2
(3 + v )v
2(3 + v 4 )v 2
1
5v 8 + 27
9 + 5v 4
c
a + S(2,2) ,
0 = c2 + a 2 +
(4.31)
+
2
2
v(v 8 9)
2v(v 4 3) 2
9 + 5v 4
5v 8 + 27
5v 8 + 27
a
c + S(2,3) ,
0 = c2 + a 2 + b2 +
(4.32)
+
+
b
2
2
v(v 4 3)
v(v 8 9)
v(v 8 9) 2
1
3(v 4 3) 2v 8 15v 4 9
a +
0 = c2 + b2
b2
2
2(3 + v 4 )v 2
v(v 8 9)
7v 8 30v 4 45
48v 2
96v 2
c
b
c2 + S(2,4) ,
+
(4.33)
2
2
v(v 8 9)
v8 9
v8 9
1
(v 4 3)2
a + S(2,5) ,
0 = c2 + b2 +
(4.34)
2
2(v 4 2v 2 + 3)(v 4 + 2v 2 + 3) 2
5v 8 + 27 32
2 + S(2,6) ,
0 = 2 +
(4.35)
v(v 8 9) 2 v 2
5v 8 + 27
0 = 2 +
(4.36)
+ S(2,7) ,
v(v 8 9) 2
where the source terms {S(2,1) , . . . , S(2,7) } are given in Appendix B. As in previous subsections,
while the system (4.30)(4.36) is overdetermined, we explicitly verified that it is consistent.
Solving (4.30)(4.36) is quite complicated. Fortunately, we do not need a complete solution.
Our ultimate goal is to determine from the nonsingularity of the ten-dimensional metric curvature invariants to order O( ) and to order O( 4/3 ) in the late proper time expansion. Thus we
1/4
evaluate metric invariants first, find what field combinations affect the singularity as v 3 ,
and then solve just for those combinations of field.
Up to now, all our equations and solutions are exact in .
We assume now that
2
= 0 + 1 + O ,
(4.37)
and evaluate background curvature invariants to order O( ) near
x 31/4 v.
(4.38)
We use explicit solutions (3.15), (4.9)(4.13) and (4.23)(4.27) to evaluate curvature invariants.
At this stage we use equations of motion (4.30)(4.36) to eliminate the derivatives (if possible)
of {a 2 , b2 , c2 , 2 } from the curvature invariants.
4.3.1. R
For Ricci scalar we find (there is no dependence on order-2 fields here)

3 18 02 3 1
2 1
13 3 1
31/4 5 1
R=
+
+ finite,
2
6 x2
2 x
4/3
x 4 31/4 x 3
295
x 0+ .
(4.39)
From (4.39) we find that Ricci scalar of the string theory geometry is nonsingular as x 0+
when
0 =
1
21/2 33/4
(4.40)
which is precisely the condition found from the nonsingularity of Riemann tensor squared [2],
as well as higher curvature invariants [11], in the supergravity approximation to the string theory
dual of the N = 4 SYM Bjorken flow. The difference here (compare to [2,11]) is that 0 is
already fixed by requiring the nonsingularity of the Ricci scalar.15
4.3.2. R R
A bit more work is necessary to determine the nonsingularity condition of Riemann tensor
squared at order O( ). Generalizing the notation of [11]
I [2] R R

1
= I0[2]SUGRA (v) + I0[2]W (v) + 2/3 I1[2]SUGRA (v) + I1[2]W (v)

1
+ 4/3 I2[2]SUGRA (v) + I2[2]W (v) + O 2 + O 2 .
(4.41)
With (4.40) and the lower order solutions, all terms in the decomposition of I [2] in (4.41) except
1/4
for I2[2]W (v) are finite as v 3 . We find
2304v 5 (v 4 3)
f2 (v)
(3 + v 4 )3
192(5v 16 + 60v 12 + 54v 8 + 540v 4 + 405)
2 (v) + I2 (v),
(3 + v 4 )4
I2[2]W =
(4.42)
where
1
f2 (v) = c2 (v) + b2 (v),
2
and I2 (v) does not depend on order-2 fields, and can be evaluated explicitly

3 1
5 3
23/2
2 1
1
5/2 5/4
1/4 3
1 3/4 2 + 2 3 1
I2 = 399 3 +
4
6
6 x2
3
x
3 x

31/4 (72 2 33/4 1 4 2 33/4 2 + 39 1 270) 1
+
+ finite,
2 (7182 + 151 4 2 33/4 2 + 72 2 33/4 1 ) x

as x 0+ .
15 Notice that Ricci scalar vanishes in the supergravity approximation [11].
(4.43)
(4.44)
296
From (4.30) one can obtain a decoupled equation for f2 as defined in (4.43). The resulting
equation is possible to explicitly solve for f2 :

24126004224 2 v 3 13961808v 3 2
2v + 1932 2 C
+
4
9
4
8
(v + 3)
(v + 3)

34992v 3
4
665v 2 + 320520 2 399C
7
(v + 3)

243v 3
2
2
+ 4
28221v
+
8578560
20600C
(v + 3)6

81v 3
20823v 2 + 3918720 2 18560C

4
5
2(v + 3)

27v 3
+
7561v 2 + 481 2 + 921600 2 11520C
4
4
8(v + 3)

v3
26307v 2 + 8v 2 1 41 C + 407808 2
16(v 4 + 3)3

v3
35831808 2 1990656v 2
+ 192
2 481 2 + 4
(v + 3)2
(v 4 3)5
f2 (v) =
1327104v 2 + 29859840 2 8709120 2 290304v 2

+
(v 4 3)4
(v 4 3)3

1
3 2
2
2
2
+ 4
23472v + 1029024 v 1 271
2
(v 3)2

1
1 2
333 2
2
2
+ 4
2+
v
91 v 1 + 25488 6
2
2
v 3

v2
10425 3
arctan + 3 ,
+
16
3
+
(4.45)
where 3 is a new arbitrary integration constant. The solution (4.45) is exact in .

Unfortunately, we cannot present an explicit solution for 2 (v). We can prove though that
2 (v) can be chosen to be finite as v 31/4 , while having a vanishing nonnormalizable mode
as v 0+ . Indeed, the most general inhomogeneous solution of (4.36) as x 0+ takes form
3/4

3
125 3/4
875 1/4
C1 +
3 C+
3
+ O x 2 ln x ,
a 2 (x) = C0 + C1 ln +x
(4.46)
6
1728
864
where C0 and C1 are the two arbitrary integration constants. On the other hand, the homogeneous
solution to (4.36) is

4

8
1
1
v
3 v4
hom
hom v + 9
hom
+ C1
+
+
,
ln
a 2 (v) = C0
(4.47)
864 96v 4
v4
3 + v 4 144
leading to asymptotic expansion as v 0+
a 2hom (v) = 9C0hom

C1hom 8
1
hom 4
+
C
v
v + O v 16 ,
0
4
972
v
(4.48)
297
and
1 hom
1 hom
ln 2
ln 3
C
C
144 1
576 1
1 hom
1 hom
+
+
ln x + O(x),
C
C
144 1
144 1
as x 0+ . Thus, given (4.46), we can take
a 2hom (x) = 6C0hom +
C1hom = 144C1 ,
(4.49)
(4.50)
and tune16 C0hom so that

a 2v + a 2hom (v)
both have only a normalizable mode as v 0+ , and be finite (along with its derivatives) as
x 0+
We now have all the necessary ingredients to determine 1 from the nonsingularity of
1/4
[2]W
I2 (v). We find that I2[2]W (v) is finite as v 3 , provided

31/4 2
7182 151 + 25/2 33/4 2 .
1 =
(4.51)
432
Notice that while 0 (4.40) is determined unambiguously from the nonsingularity condition of
the background geometry, the absence of singularities is not a powerful enough constraint to
fix 1 . As we already mentioned, this fact will not preclude us from computing a definite ratio of
shear viscosity to the entropy density.
4.3.3. R R
Analysis of the square of the Ricci tensor can be performed in the same way as for the Riemann tensor square. We find
R R = finite
1920 2 (v)
,
4/3
1/4
v 3 ,
(4.52)
were we explicitly indicated dependence on order-2 fields. We argued above that 2 (v) can be
1/4
chosen to be finite as v 3 ; this would guarantee the nonsingularity of R R to orders
O( 4/3 ) and O( ).
4.3.4. Higher order curvature invariants
As in [11] we denote
R[2
n]
R[2
n1 ]
1 1
R[2
n1 ]
1 1
(4.53)
where
R[0] R .
(4.54)
We further define higher curvature invariants I [2 ] , generalizing (4.41):

n
16 This is always possible since S

8
(2,6) = O(v ), see (B.9).
298
I [2 ] R[2 ] R[2 ]
n

n
n
1 n
= I0[2 ]SUGRA (v) + I0[2 ]W (v) + 2/3 I1[2 ]SUGRA (v) + I1[2 ]W (v)

1 [2n ]SUGRA
[2n ]W
+ 4/3 I2
(v) + I2
(v) + O 2 + O 2 .
n1
n1
(4.55)
Given complexity of analysis, we checked at order O( 4/3 ) only17 nonsingularity of I [4] .

Keeping explicit only fields at order-2, we find

55296v 5 (v 4 3) 16
v 4v 12 + 198v 8 36v 4 + 81 f2 (v)
4
7
(3 + v )
663552v 9 (v 4 3)3
1536 32
+
5v + 120v 28 + 6876v 24
2 (v)
(3 + v 4 )7
(3 + v 4 )8
I2[4]W (v) =
+ 33480v 20 72738v 16 + 301320v 12 + 556956v 8 + 87480v 4

+ 32805 2 (v) + I4 (v).
(4.56)
No new analysis of order-2 fields are necessary here to verified that (4.56) is nonsingular precisely for (4.51).
5. Shear viscosity for the Bjorken flow of N = 4 SYM plasma
In previous section we analytically evaluated -corrected supergravity background dual to a
Bjorken flow of N = 4 SYM plasma at finite coupling to order O( 2/3 ) in the late proper time
expansion. We can now extract the boundary energy density ( ) from the one-point correlation
function of the boundary stress energy tensor using the -corrected holographic renormalization
developed in [20]. We confirmed that the final expression for the energy density can be evaluated
as in the supergravity approximation [1,2,11]:
( ) =
N2
2a(v, )
lim 4 4/3 .
2
2 v0 v
(5.1)
Using (3.16), (4.9), (4.23), (4.37), (4.40) and (4.51) we find

( ) =
N 2 (6 + 576 + 1 ) 1
12 2
4/3
2
1/2
1/4

N 2 3 (1566 + 8 + 1 ) 1
+ O 8/3 .
2
2
48
(5.2)
Notice that even though 1 depends on an arbitrary integration constant 2 , see (4.51), such a
dependence disappears in (5.2).
The string theory result (5.2) should now be interpreted within MullerIsraelStewart theory
[8,9]. For the Bjorken flow of the N = 4 SYM plasma we expect [11]

3
1
1
gauge ( ) = 2 N 2 4 (1 + 15 ) 4/3 2 N 2 3 A(1 + 15 ) 2 + O 8/3 ,
8
17 We studied more general higher curvature invariants at lower orders in the late proper time expansion.
(5.3)
299
where is an arbitrary scale, related to the initial energy density of the expanding plasma, and
A= ,
(5.4)
s
is the ratio of shear viscosity to entropy density. Notice that in (5.3) we have to use N = 4
equation of state at finite t Hooft coupling,18 see [23]. Matching the leading terms in the late
time expansion in (5.2) and (5.3) we find

2
4
1
4 =
(5.5)
1
+
81
+
1 +O ,
4
6
3
which implies that

1
(1 + 120 ) + O 2 ,
(5.6)
4
leading to the result quoted in (2.15). Notice that the remaining arbitrary integration constant 1
got absorbed into a definition of an arbitrary scale . The physical observable, i.e., the ratio of
shear viscosity to entropy density (5.6) is evaluated unambiguously.
A=
6. Conclusion
In this paper we evaluated finite t Hooft coupling correction to strongly coupled N = 4
SYM plasma undergoing (1+ 1)-dimensional Bjorken expansion. Such finite coupling correction
corresponds to string theory corrections in the supergravity dual to the boost invariant plasma
expansion proposed by Janik and collaborators [13]. We extracted the ratio of shear viscosity
to entropy density at large, but finite, t Hooft coupling. Our result is in conflict with analysis
of shear viscosity of strongly coupled SYM plasma extracted from the equilibrium two-point
correlation functions of the boundary stress energy tensor [18,19].
Though we do not currently understand the source of the discrepancy, there are two potential
outcomes.
First, the nonsingularity approach of [13] is simply inconsistentit was shown in [11] that
it is definitely inconsistent within the supergravity approximation at high orders in the late
proper time expansion (though it appears to be well defined in the first two orders necessary
to compute shear viscosity of N = 4 plasma in the supergravity approximation).
The discrepancy might be due to using incorrect effective string theory action (2.12), as a
dual to N = 4 dynamics at large, but finite, t Hooft coupling. Specifically, the full set of
corrections to RR fluxes is unknown. It is conceivable that proper inclusion of corrections
to the RR fluxes would reconcile two results (as it must be). A starting point for such analysis
could be the effective action proposed in [26].
We believe resolution of the puzzle presented here deserves further study.
Acknowledgements
My research at Perimeter Institute is supported in part by the Government of Canada through
NSERC and by the Province of Ontario through MRI. I gratefully acknowledge further support
18 We would like to thank Romuald Janik for pointing this out.
300
by an NSERC Discovery grant and support through the Early Researcher Award program by the
Province of Ontario.
Appendix A. Source terms for Eqs. (4.16)(4.22)
S(1,1) =
6
72v
6 1
1119744(20v 16 340v 12 + 1010v 8 555v 4 9)v
+
,
(3 + v 4 )4
(3 + v 4 )9
(A.1)
6 1
72(v 16 + 54v 8 + 72v 4 + 81)v
373248v
6
(v 8 9)4
(3 + v 4 )9 (v 4 3)4

113319v 16 55647v 4 + 159732v 8 180090v 12 37251v 20

+ 2187 532v 28 + 12v 32 + 6750v 24 ,
(A.2)
6 1
144(v 16 + 6v 12 + 54v 8 + 18v 4 + 81)v
746496v
6
8
4
4
9
(v 9)
(3 + v ) (v 4 3)4

24v 32 604v 28 + 6840v 24 37017v 20 + 113751v 16

179118v 12 + 139806v 8 30861v 4 + 2187 ,
(A.3)
S(1,2) =
S(1,3) =
S(1,4) =
4
12v 6 (v
+ 15)(v 8
(3 + v 4 )4 (v 4
+ 9)1
+ 4
3)2
(v 3)2 (3 + v 4 )9
186624v
6

84v 24 2312v 20 + 15663v 16 50778v 12 + 74304v 8

37098v 4 1215 ,
S(1,5) =
24v 7
(A.4)
(v 8 9)(v 4 2v 2 + 3)(v 4 + 2v 2 + 3)
373248v 7
(v 4 + 2v 2 + 3)(v 4 2v 2 + 3)(v 4 3)(3 + v 4 )8

40v 20 520v 16 + 1455v 12 1269v 8 189v 4 81 ,
(A.5)
S(1,6) =
259200(71v 8 + 639 456v 4 )v 14

,
(3 + v 4 )9 (v 4 3)
(A.6)
S(1,7) =
8 9v 4 + 27)(v 8 3v 4 + 3)
288v
10 2 44789760v 14 (v
.
(v 8 9)3
(v 4 3)3 (3 + v 4 )9
(A.7)
Appendix B. Source terms for Eqs. (4.30)(4.36)

We find it convenient to express the source terms at this order implicitly in the background
warp factors {a2 , b2 , c2 }, while using equations of motion for the supergravity background
to eliminate higher than second derivatives.19 For convenience, we present these equations
here [2,11]
19 The highest derivative is the fourth.
301
9 + 5v 4
16v 3
32v 3
4(2v 4 + 9)v 4
a2 8
b2 8
c2 +
4
v(v 3)
v 9
v 9
3(v 4 3)(v 8 9)
432v 6 2 (v 16 + 24v 12 + 30v 8 + 216v 4 + 81)
,
(v 8 9)4
5v 8 8v 4 + 27
16v 3 432v 6 2 (v 12 + 11v 8 + 63v 4 27)
b
c
b2 =
+
2
v(v 8 9)
v8 9 2
(v 8 9)3 (3 + v 4 )
8
4
4
4(2v 9v 63)v
,
3(v 4 3)2 (v 8 9)
8v 3 5v 8 16v 4 + 27
b
c2
c2 = 8
v 9 2
v(v 8 9)
432v 6 2 (v 12 13v 8 9v 4 27) 4(v 8 9v 4 18)v 4
.
+
(3 + v 4 )(v 8 9)3
3(v 4 3)2 (v 8 9)
a2 =
(B.1)
(B.2)
(B.3)
Of course, (3.15) solves (B.1)(B.3)

S(2,1) =

4320v 7 (119v 16 837v 12 + 4776v 8 7533v 4 + 9639)
2v 3 1
a
+
2
4
4
7
(v 3)(3 + v )
(3 + v 4 )2

864v 3
125v 28 15725v 24 + 145095v 20 669570v 16
(3 + v 4 )7 (v 4 3)3

+ 1303911v 12 1275669v 8 + 96957v 4 + 8748 (b2 + 2c2 )

432v 6 (2v 12 21v 8 + 36v 4 81) 2
(2v 8 + 3v 4 + 9)v 4
+
1
+ 4
(v 3)3 (3 + v 4 )2
(3 + v 4 )5 (v 4 3)3

2
72(v 8 2v 4 + 9)v 6
373248v 6 2
+
87058152v 16
4
2
4
4
4
6
4
10
(v 3) (3 + v )
(v 3) (3 + v )

22973922v 24 + 61231383v 20 39438171v 8 + 92195901v 12 + 7794468v 4

+ 177147 + 6804783v 28 1074783v 32 + 126421v 36 + 132v 44 6007v 40

288v 4
v 40 + 2386v 36 + 5073v 32 + 50304v 28 + 2354787v 24
(3 + v 4 )7 (v 4 3)6
3096360v 20 + 26563545v 16 11763144v 12 + 26305236v 8 118098v 4

118098 ,
(B.4)

S(2,2) =

2v 3 1
288v 3
535v 24 5190v 20
+
(v 4 3)2 (3 + v 4 )7 (v 4 3)2
+ 34755v 16 92925v 12 + 160542v 8 109593v 4 + 8748

a2

288v 7
473v 24 18797v 20 + 177291v 16 787422v 12
(3 + v 4 )7 (v 4 3)3

+ 1595619v 8 1522557v 4 + 344817 b2

24v 7 1
576v 7
41v 24 14909v 20 + 162387v 16
+ 8
+
2
4
7
4
3
(v 9)
(3 + v ) (v 3)
+
302
751134v 12 + 1471203v 8 1184301v 4 + 47385
c2
v 4 (4v 16 6v 12 171v 8 648v 4 567)

9(v 4 3)4 (3 + v 4 )2

144v 6 2 20
16
12
8
4
2v 99v 36v 1458v 2430v 2187 1
+ 8
(v 9)5
+
24v
6 (3v 16 8v 12 + 90v 8 + 216v 4 + 243)2
(v 8 9)4

124416v 6 2
4
86541048v 16 23414778v 24 + 65133639v 20
6
4
10
(v 3) (3 + v )
43348527v 8 + 96948981v 12 + 13463172v 4 531441 + 7098399v 28

958419v 32 + 104165v 36 + 60v 44 2587v 40
+

32v 4
4v 44 + 21v 40 + 11406v 36 3843v 32 + 51372v 28
(3 + v 4 )7 (v 4 3)6
+ 14188743v 24 27797580v 20 + 159619653v 16 94425912v 12

+ 135077868v 8 5314410v 4 2480058 ,

S(2,3) =

4v 3 1
576v 3
535v 24 5190v 20
+
(v 4 3)2 (3 + v 4 )7 (v 4 3)2
+ 34755v
16
92925v
12
+ 160542v 109593v + 8748

8
(B.5)
a2

24v 7 1
576v 7
+ 8
+
41v 24 14909v 20 + 162387v 16
2
4
7
4
3
(v 9)
(3 + v ) (v 3)

12
8
4
751134v + 1471203v 1184301v + 47385 b2

+

24v 7 1
1152v 7
257v 24 16853v 20
+
2
4
2
4
7
4
3
3) (3 + v )
(3 + v ) (v 3)

+ 169839v 16 769278v 12 + 1533411v 8 1353429v 4 + 196101 c2
(v 4
2v 4 (v 16 + 12v 12 + 99v 8 + 162v 4 + 162)

9(v 4 3)4 (3 + v 4 )2
6
2
288v 20
4v + 99v 16 + 252v 12
8
(v 9)5

16 + 16v 12 + 90v 8 + 243)2
48v 6 (3v
+ 1458v 8 + 2187 1
(v 8 9)4

248832v 6 2
4
87964056v 16
6
4
10
(v 3) (3 + v )
+
24187842v 24 + 60165747v 20 30830139v 8 + 80345277v 12 + 3542940v 4

531441 + 6773211v 28 1132371v 32 + 126641v 36 + 168v 44 6439v 40
64v 4
(3 + v 4 )7 (v 4
3)6
303

v 44 + 12v 40 + 3873v 36 + 57582v 32 870975v 28
+ 10771434v 24 39529377v 20 + 117880758v 16 124357194v 12

+ 97483338v 8 + 1417176v 4 + 708588 ,
(B.6)

v 3 (v 4 3)1
9v 3
16v 28 + 192v 24 57909v 20
+
4
2
4
7
4
4(3 + v )
2(3 + v ) (v 3)

+ 492612v 16 3870990v 12 + 5165316v 8 13920741v 4 + 69984 a2
S(2,4) =

+

v 3 (v 12 33v 8 + 27v 4 675)1
3v 3
16v 36 + 288v 32
8
2
4
7
4
3
12(v 9)
2(3 + v ) (v 3)
+ 54939v 28 6400314v 24 + 62833977v 20 333665244v 16 + 710087229v 12

8
4
937477962v + 81104895v + 15746400 b2

+
v 3 (v 12 + 3v 8 27v 4 + 459)1
6v 3
+
8
2
4
3(v 9)
(3 + v )7 (v 4 3)3

16v 36 + 144v 32 11157v 28 + 3087459v 24 31309542v 20 + 166984254v 16

12
8
4
355695705v + 476607807v 55436076v 10707552 c2

72(v 4 + 1)v 2 1
144v 2
16v 28 + 240v 24 + 1371v 20
8
2
4
7
4
2
(v 9)
(3 + v ) (v 3)

16
12
8
4
48555v + 48465v 196425v 186624v 69984 (b2 + 2c2 )
+
28
6 2 v 2 (2v 8 9v 4 27)1
324 2 v 2
8v + 120v 24 + 623v 20
8
2
4
7
4
2
(v 9)
(3 + v ) (v 3)

4v
3 v4
2 2
ln
+ 17985v 16 21555v 12 + 36315v 8 + 13608v 4 5832 8
v 9
3 + v4

4
8
6
2

(v 9)v
12v
1917v 8 + 12v 16 81v 12
+ 4
+
3
4
2
4
5
4
3
(v 3) (3 + v )
(3 + v ) (v 3)

12 + 12v 8 + 27v 4 + 108)2
16v 6 (v
2511v 4 8019 + 4v 20 1
(v 4 3)2 (3 + v 4 )4
6
2

216v
+ 4
100639406448v 20 + 177616911942v 12
6
4
10
(v 3) (3 + v )
+
+ 19659065472v 4 + 931084632 84883350843v 8 148217241528v 16

+ 165520368v 36 1463056992v 32 + 10018391364v 28 34345317222v 24

2v 2
+ 98934v 44 6831063v 40 + 312v 4 8 + 32v 52 + 4
(v 3)6 (3 + v 4 )7

17006112v 6 289315557v 26 3796250733v 18 + 74994660v 22
304
4935413835v 10 + 931858830v 14 13966290v 30 476019v 34

207648v 38 144v 42 ,

S(2,5) =
(B.7)
(v 4 3)v 4 1
6(v 4 + 2v 2 + 3)(v 4 2v 2 + 3)
28
48v 4
v + 12v 24 9v 20 6948v 16
4
2
4
2
4
6
(v 2v + 3)(v + 2v + 3)(3 + v )

12
8
4
+ 15435v 65934v 10935v 4374 a2

+
+
v 4 (v 4 + 1)1
6(v 4 2v 2 + 3)(v 4 + 2v 2 + 3)
(v 4
2v 2
+ 3)(v 4
48v 4
+ 2v 2 + 3)(3 + v 4 )6 (v 4 3)2

v 36 + 14v 32 + 24v 28 + 1782v 24 136782v 20 + 553644v 16 1220832v 12

+ 161838v 8 2187v 4 + 13122 (b2 + 2c2 )

(v 4 6)v 9
9(v 4 3)2 (v 4 + 2v 2 + 3)(v 4 2v 2 + 3)

72v 7 (v 8 18) 2
1
+ 4
(v 2v 2 + 3)(v 4 + 2v 2 + 3)(v 8 9)2
24v
7 2
+ 4
2
4
(v 2v + 3)(v + 2v 2 + 3)(v 8 9)
20736v 7 2
4
5
4
9
(v 3) (3 + v ) (v 4 + 2v 2 + 3)(v 4 2v 2 + 3)

1062882 + 18680058v 24 + 59248017v 16 + 7853517v 8 + 1417176v 4
33907248v 12 36367380v 20 + 1161v 40 46404v 36 + 732348v 32

4038876v 28 + v 48 + 12v 44
32v 9
(v 4 2v 2 + 3)(v 4 + 2v 2 + 3)(3 + v 4 )6 (v 4 3)5
40
v + 6v 36 117v 32 12276v 28 112041v 24 + 345492v 20 7735095v 16

+ 6890508v 12 21065184v 8 + 3346110v 4 + 236196 ,
(B.8)
S(2,6) =
180v 7 (163v 16 42v 12 2106v 8 378v 4 + 13203)

a2
(3 + v 4 )7 (v 4 3)

180v 7
163v 24 2556v 20 + 34557v 16
+
4
7
4
3
(3 + v ) (v 3)

165528v 12 + 311013v 8 207036v 4 + 118827 (b2 + 2c2 )
+
(v 4

207360v 14 2
166212v 12 393660v 4 + 166941 + 451413v 8
4
4
10
3) (3 + v )

240v 8
+ 229v 24 4860v 2 0 + 50157v 16
4
(3 + v )7 (v 4 3)4

173v 24 + 183v 20 + 18162v 16 2160v 12 + 195615v 8

+ 18225v 4 + 7290 ,
305
(B.9)

466560v 11 (2v 8 11v 4 + 18)
v 3 2
a2
v8 9
(3 + v 4 )7 (v 4 3)

3
155520v 11 (2v 8 63v 4 + 18)
v 2
+
(b2 + 2c2 )
+
v8 9
(3 + v 4 )7 (v 4 3)

12v 4 (v 4 + 1)
10368v 10 2
288v
10 4
+
+
+
2
(v 4 3)4 (3 + v 4 )
(v 8 9)4
(v 8 9)3
14
2

13436928v
4
11v 24 192v 20 + 1853v 16 5832v 12 + 16677v 8
4
4
10
(v 3) (3 + v )

124416v 16
15552v 4 + 8019 +
(3 + v 4 )7 (v 4 3)4

10v 12 + 33v 8 + 63v 4 + 79 .
(B.10)

S(2,7) =
References
[1] R.A. Janik, R. Peschanski, Asymptotic perfect fluid dynamics as a consequence of AdS/CFT, Phys. Rev. D 73
(2006) 045013, hep-th/0512162.
[2] R.A. Janik, Viscous plasma evolution from gravity using AdS/CFT, Phys. Rev. Lett. 98 (2007) 022302, hepth/0610144.
[3] M.P. Heller, R.A. Janik, Viscous hydrodynamics relaxation time from AdS/CFT, Phys. Rev. D 76 (2007) 025027,
hep-th/0703243.
[4] S. Nakamura, S.J. Sin, A holographic dual of hydrodynamics, JHEP 0609 (2006) 020, hep-th/0607123.
[5] J.M. Maldacena, The large N limit of superconformal field theories and supergravity, Adv. Theor. Math. Phys. 2
(1998) 231, Int. J. Theor. Phys. 38 (1999) 1113, hep-th/9711200.
[6] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity, Phys.
Rep. 323 (2000) 183, hep-th/9905111.
[7] S. Bhattacharyya, V.E. Hubeny, S. Minwalla, M. Rangamani, Nonlinear fluid dynamics from gravity, arXiv:
0712.2456 [hep-th].
[8] I. Muller, Z. Phys. 198 (1967) 329.
[9] W. Israel, J.M. Stewart, Transient relativistic thermodynamics and kinetic theory, Ann. Phys. 118 (1979) 341.
[10] J.D. Bjorken, Highly relativistic nucleusnucleus collisions: The central rapidity region, Phys. Rev. D 27 (1983)
140.
[11] P. Benincasa, A. Buchel, M.P. Heller, R.A. Janik, On the supergravity description of boost invariant conformal
plasma at strong coupling, arXiv: 0712.2025 [hep-th].
[12] S.S. Gubser, I.R. Klebanov, A.W. Peet, Entropy and temperature of black 3-branes, Phys. Rev. D 54 (1996) 3915,
hep-th/9602135.
[13] G. Policastro, D.T. Son, A.O. Starinets, The shear viscosity of strongly coupled N = 4 supersymmetric YangMills
plasma, Phys. Rev. Lett. 87 (2001) 081601, hep-th/0104066.
[14] M. Natsuume, T. Okamura, Causal hydrodynamics of gauge theory plasmas from AdS/CFT duality, arXiv:
0712.2916 [hep-th].
[15] M. Natsuume, T. Okamura, Comment on: Viscous hydrodynamics relaxation time from AdS/CFT correspondence, arXiv: 0712.2917 [hep-th].
[16] M. Natsuume, T. Okamura, A note on causal hydrodynamics for M-theory branes, arXiv: 0801.1797 [hep-th].
[17] R. Baier, P. Romatschke, D.T. Son, A.O. Starinets, M.A. Stephanov, Relativistic viscous hydrodynamics, conformal
invariance, and holography, arXiv: 0712.2451 [hep-th].
306
[18] A. Buchel, J.T. Liu, A.O. Starinets, Coupling constant dependence of the shear viscosity in N = 4 supersymmetric
YangMills theory, Nucl. Phys. B 707 (2005) 56, hep-th/0406264.
[19] P. Benincasa, A. Buchel, Transport properties of N = 4 supersymmetric YangMills theory at finite coupling,
JHEP 0601 (2006) 103, hep-th/0510041.
[20] A. Buchel, Higher derivative corrections to near-extremal black holes in type IIB supergravity, Nucl. Phys. B 750
(2006) 45, hep-th/0604167.
[21] G. Policastro, D.T. Son, A.O. Starinets, From AdS/CFT correspondence to hydrodynamics, JHEP 0209 (2002) 043,
hep-th/0205052.
[22] G. Policastro, D.T. Son, A.O. Starinets, From AdS/CFT correspondence to hydrodynamics. II: Sound waves,
JHEP 0212 (2002) 054, hep-th/0210220.
[23] S.S. Gubser, I.R. Klebanov, A.A. Tseytlin, Coupling constant dependence in the thermodynamics of N = 4 supersymmetric YangMills theory, Nucl. Phys. B 534 (1998) 202, hep-th/9805156.
[24] M. Brigante, H. Liu, R.C. Myers, S. Shenker, S. Yaida, Viscosity bound violation in higher derivative gravity, arXiv:
0712.0805 [hep-th].
[25] O. Aharony, A. Buchel, P. Kerner, The black hole in the throatThermodynamics of strongly coupled cascading
gauge theories, Phys. Rev. D 76 (2007) 086005, arXiv: 0706.1768 [hep-th].
[26] S. de Haro, A. Sinkovics, K. Skenderis, On corrections to D-brane solutions, Phys. Rev. D 68 (2003) 066001,
hep-th/0302136.

Solving topological defects via fusion

Z. Bajnok , Zs. Simon
Theoretical Physics Research Group of the Hungarian Academy of Sciences,
H-1117 Pzmny s. 1/A, Budapest, Hungary
Received 4 February 2008; received in revised form 7 April 2008; accepted 7 April 2008
Available online 10 April 2008
Abstract
Integrable defects in two-dimensional integrable models are purely transmitting thus topological. By
fusing them to integrable boundaries new integrable boundary conditions can be generated, and, from the
comparison of the two solved boundary theories, explicit solutions of defect models can be extracted. This
idea is used to determine the transmission factors and defect energies of topological defects in sinh-Gordon
and LeeYang models. The transmission factors are checked in Lagrangian perturbation theory in the sinhGordon case, while the defect energies are checked against defect thermodynamic Bethe ansatz equations
derived to describe the ground-state energy of diagonal defect systems on a cylinder. Defect bootstrap
equations are also analyzed and are closed by determining the spectrum of defect bound-states in the Lee
Yang model.
1. Introduction
Recently, there has been an increasing interest in integrable quantum field theories including
defects or impurities. This is motivated both by the realistic physical applications in statistical
and solid state physics and also by the need of theoretical understanding of this so-far unexplored
field.
The community of integrable systems have not payed much attention to defect theories at the
beginning due to the no-go theorem formulated by Delfino, Mussardo and Simonetti in [1,2].
The theorem, formulated originally for diagonal theories and extended later for a large class of
non-diagonal ones in [3], states that a relativistically invariant theory with a non-free integrable
E-mail address: bajnok@elte.hu (Z. Bajnok).

doi:10.1016/j.nuclphysb.2008.04.003
308
Z. Bajnok, Zs. Simon / Nuclear Physics B 802 [FS] (2008) 307329
interaction in the bulk can allow only two types of integrable defects: the purely reflecting and
the purely transmitting ones. (Although some effort has been made to overcome this obstacle by
giving up Lorentz invariance, see for instance [4] and references therein, in the present paper we
restrict ourselves to the relativistically invariant case.)
The analysis of boundary integrable theories was initiated in [5] by formulating, in an axiomatic way, the properties of the reflection matrix: unitarity, boundary crossing unitarity and
boundary bootstrap equation. The boundary bootstrap framework was completed by introducing
boundary ColemanThun mechanism [6] and the bulk bootstrap equations [7]. Later this framework got a sound basis by developing boundary quantum field theories from first principles in
[8,9]. The success of the boundary bootstrap approach resulted in a large class of closed bootstrap theories in which the boundary reflection factors together with the spectrum of boundary
excited states were determined [5,6,1014]. The solutions, obtained by the bootstrap method, are
not connected, however, to other formulations of the model such as to the classical field theory or
to the perturbed conformal field theory (both defined by a Lagrangian). To connect the different
descriptions one either has to check the reflection factors perturbatively, like in [15], or solve
the theories in finite volume. The boundary thermodynamic Bethe ansatz (BTBA), developed
in [16], systematically sums up the finite size corrections by taking into account the scatterings
and reflections. By analyzing its small volume limit the needed link between the bootstrap and
perturbed conformal field theoretical descriptions can be established.
As the no-go theorem showed non-free integrable defect theories are purely transmitting.
This fact kept back the researchers for some time to analyze these models until new life was
put into the subject due to their explicit Lagrangian realizations [17]. Following the original idea
many integrable defect theories were constructed at the classical level [1820]. The basis for the
quantum formulation of defect theories is provided by the folding trick [21] by which one can
map any defect theory into a boundary one. As a consequence defect unitarity, defect crossing
symmetry and defect bootstrap equations together with defect ColemanThun mechanism are
derived. Despite of these results the explicitly solved relativistically invariant defect quantum
field theories are quite rare, containing basically the sine-Gordon and affine Toda field theories
[2224] and even in these cases the explicit relation to their Lagrangian have not been worked
out yet.
One may think that purely transmitting theories are too simple and there is no point to analyze
them, but we would like to argue that they carry very important information about an integrable
quantum field theory, without which our knowledge cannot be complete. Purely transitivity implies the conservation of momentum from which the topological nature of the defect follows.
Thus, such defects can freely be transported in space without affecting the physics of the theory.
We can either move them close to each other or move them to integrable boundary conditions
and, as a result, new integrable boundary conditions can be generated. These ideas were successfully applied in conformal field theories [2527], in integrable lattice models [28,29] and the aim
of the present paper is to exploit it in solving integrable defects in the sinh-Gordon and LeeYang
theories.
The paper is organized as follows: In Section 2, on the example of the sinh-Gordon theory, we
show how new boundary conditions can be obtained by fusing integrable defects to boundaries
at the classical level. We focus on two cases in detail: fusing the integrable defect to Dirichlet
boundary condition (DBC) the perturbed Neumann boundary condition (PNBC) can be obtained,
while fusing it to the Neumann BC a new class of time-dependent integrable BCs can be generated. Section 3 recalls the quantum version of the fusion method together with the properties of
transmission factors. By comparing the already known reflection factors and boundary energies
309
of the DBC to those of the PNBC solutions for the defect transmission factor and defect energy
can be extracted. The same method is then used to determine defect energies and transmission
factors of the scaling LeeYang model. This method not only provides the explicit solutions of
the sinh-Gordon defect theory but also relates its parameter to that of the Lagrangian. Since the
relation obtained here is different from the suggestion of [23] we perform a one-loop perturbative analysis in Section 4. (In the subsequent paper [24] the same authors raised the possibility
of the quantum renormalization of the transmission parameter, which we calculate at one-loop
here.) In Section 5 the calculated defect energies are subject to another consistency check. For
this we derive a TBA equation to describe the groundstate energy of a diagonal defect systems
on a cylinder. From a careful UV analysis defect energies are extracted and the previous results
are verified. By completing the defect bootstrap program we analyze the singularity structure of
the transmission factors in Section 6. Since the sinh-Gordon transmission factor does not contain any singularity in the physical strip we analyze the LeeYang model only. For each pole of
the transmission factor in the physical strip we associate either a defect boundstate or a defect
ColemanThun diagram and calculate the excited transmission factors in the former case from
the defect bootstrap equation. Finally, we conclude in Section 7 and give directions for further
research.
2. Fusion method at the Lagrangian level
In this section we demonstrate, on the example of the sinh-Gordon (ShG) theory, how new integrable boundary conditions can be obtained by the fusion method at the Lagrangian or classical
level [30].
The ShG theory is defined on the whole line by the following Lagrangian
m2
1
1
LShG () = (t )2 (x )2 2cl cosh b.
(1)
2
2
b
It describes an integrable field theory and by restricting the theory to the half line this property
can be only maintained, whenever the following boundary potential is introduced [5]
b
(2)
B() = M0 cosh ( 0 ).
2
By varying M0 from 0 to the arising boundary condition interpolates between the Neumann
x |x=0 = 0 and the Dirichlet (0, t) = 0 ones.
The most general integrable defect condition can be obtained by the analytical continuation
of the sine-Gordon result [23]. The Lagrangian in the ShG case reads as
LBShG = (x)LShG () (x)B();
LDShG = (x)LShG ( ) (x)D( , + ) + (x)LShG (+ )
(3)
where the defect potential contains just one single parameter:

b
2D( , + ) = + + + Mcl e cosh (+ + )
2
b
+ Mcl e cosh (+ ).
2
cl
.
Here are the fields living on the left/right half-line, respectively and Mcl = 4m
b2
The fusion idea is based on the integrability of the defect: Integrability guaranties the existence of an infinite number of commuting conserved charges which results in the possibility of
310
shifting the trajectories of particles, without changing the amplitude of any scattering process.
The shifting of all the trajectories can alternatively be described by shifting the location of the
defect, which then does not alter the physics. Same is true for a system in which an integrable
defect is placed in front of an integrable boundary. Integrability is maintained thus particle-like,
localized, finite energy excitations do not radiate, instead, purely transmit on the defect and reflect on the boundary. Any particle like excitation for large enough negative/positive time has no
overlap with the defect and the boundary so their effect can be recognized only by the total time
delay acquired which is independent of the position of the defect.
At the level of the Lagrangian this observation can be formulated in the following way: The
spectrum of the system, which contains a defect in the origin, x = 0, in front of a boundary,
located at x = a,
LDBShG = (x)LShG ( ) (x)D( , + ) + (x)(a x)LShG (+ )
(x a)B(+ )
does not actually depend on a. Thus we can perform the a 0 limit and represent the same
system as a boundary one, but with a different (dressed) boundary condition:
LDBShG = (x)LShG ( ) (x)B ( , + );
B ( , + ) = D( , + ) + B(+ ).
The field + lives on the boundary only and can be thought naively to be a boundary degree
of freedom. It does not have, however, any kinetic term so it merely implements a new timedependent integrable boundary condition. Let us specify these findings in two concrete examples
that will be used later on.
If the original boundary condition is the Dirichlet one with (a, t) = 0 , then the arising
dressed boundary condition is
Mcl e
Mcl e
b
b
cosh ( + 0 ) +
cosh ( 0 ) (4)
2
2
2
2
where we dropped the total time derivatives. This BC is exactly of the form of (2) with parameters

b
b
Mcl cosh 0 = M0 e 2 0 .
2
B ( , + ) = D( , 0 ) =
Thus by fusing the integrable defect to the DBC we can reconstruct the most general (two parameter family of) PNBCs. Interestingly 0 and are the classical analogues of the parameters
in which the boundary reflection is factorized [45], see also (13, 14) in Section 3.
By fusing the defect to the NBC we obtain the boundary potential
B ( , + ) = D( , + ).
Variation of action provides BCs in the form:
t |x=0 =
B (+ , )
;
+
(t + x )|x=0 =
B (+ , )
.
(5)
By expressing + in terms of and t then plugging back to the second equation we

obtain a highly non-trivial boundary condition for containing its second time derivative,
which is nevertheless integrable as it follows from the construction. Obviously, this solution was
not covered by the two parameter family of (time-independent) integrable boundary conditions
311
determined in [5], thus by the fusion method we were able to construct a new type of integrable
BC. By fusing other integrable defects with new free parameters to this dressed boundary we can
generate integrable BCs with as many parameters as we want. What is nice in the construction,
that the solution of the defects transmission factor will provide, via the fusion method, solutions
for these general integrable BCs, too, as we will show in the next section.
Finally, we note that similar construction can be used in the case of the LeeYang model,
however, the explicit form of the integrable defect perturbation has not been identified at the
Lagrangian level yet (for details see the next section).
3. Fusion method in the bootstrap
For simplicity we present the fusion idea in the case of an integrable diagonal scattering theory
with one particle type of mass m. The general discussion can be found in [21].
In integrable bulk theories multi-particle scattering processes factorize into the product of
two particle scatterings, S(12 ), where 12 = 1 2 is the rapidity difference of the scattering
particles whose momenta are parametrized as pi = m sinh i . In relativistically invariant theories
the two particle S-matrix satisfies unitarity and crossing symmetry
S( ) = S 1 ( );
S(i ) = S( ).
Once boundaries are introduced the basic process is the multi-particle reflection. Integrability
ensures its factorization into pairwise scatterings S(ij ) and individual reflections R(i ), where
i is the rapidity of the reflected particle. The reflection matrix satisfies unitarity and boundary
crossing unitarity [5]

i
i
R
= S(2 )R
+ .
R( ) = R 1 ( );
2
2
Integrable non-free defects are severely restricted: they are either purely reflecting (thus
boundaries, like above) or purely transmitting. The latter case can be described by the two, left
() and right (+), transmission matrices T ( ) and T+ ( ). We parametrize T+ such a way
that for its physical domain ( < 0) its argument is always positive. Transmission factors satisfy
unitarity and defect crossing symmetry [21]
T+ ( ) = T1 ( );
T ( ) = T+ (i ).
(6)
If we place a defect with transmission matrices T ( ) in front of a boundary with reflection

matrix R() then the fused boundary will also be integrable and have reflection factor R ( ):
R ( ) = T+ ( )R( )T ( ).
(7)
R ( )
The correspondence (7) between the original R() and the fused
reflection factors can be
used either to generate new BCs or, if the two BCs are already known, to solve defect transmission factors. This will be illustrated in the next subsections for the sinh-Gordon and LeeYang
models.
3.1. Solution of defect sinh-Gordon theory
The spectrum of ShG theory defined by (1) consists of one particle type with mass [32]

1
m2cl
(1 + b2 ) 2+2b2
b2
4
;
B
=
.
m = 1B
b2
(b2 )
8 + b2
( 2 )
(1 + B2 )
(8)
312
The two particle scattering matrix is given by

S=
sinh i sin B
= (B)(1 + B),
sinh + i sin B
(x) =
sinh( 2 +
sinh( 2
ix
2 )
.
ix
2 )
It has no poles in the physical strip: 0 < i and is invariant under the weak-strong duality,
b2
8
8 b2 . The bulk energy density turns out to be [33]
m2
.
(9)
8 sinh B
Integrable boundary conditions can be either Dirichlet type with (0, t) = 0 or PN type (2).
The corresponding reflection factors can be obtained from the analytical continuation of the sineGordons first breathers one [5,34]. In the Dirichlet case it reads as
bulk =
RDir (, ) =
iBDir
B
2)(
iBDir
B
2) (
( 12 )(1
12 )
( 32
+ 12 )
(10)
where the reflection parameter is related to 0 as

4
0 .
b
The boundary energy has been also calculated [16]

m
B
B
Dir
(Dir ) =
2 cosh BDir sin
cos
1 .
bdry
4 sin B
2
2
Dir =
(11)
(12)
In the PN case the reflection factor turns out to be

RPN (, , ) =
iB
B
2)(
iB
B
2) (
( 12 )(1
1
12 ) ( iB
2)
( 32
1
+ 12 ) ( iB
+ 2)
while the relation of , to the parameters of the Lagrangian is [45]

b
2
b2
20
.
; M = mcl
( ) = M0 e
M cosh
2
8
b sin(b2 /8)
The boundary energy has been also determined [31,45] as

m
B
B
PN
2 cosh B + 2 cosh B sin
cos
1 .
bdry (, ) =
4 sin B
2
2
(13)
(14)
(15)
We note that the resultsboth for the UVIR relation and for the boundary energywere obtained in the framework of perturbed BCFT in which the perturbing operator is normal-ordered
to have a definite scaling dimension.
The integrable defect potential for the sinh-Gordon model can be written as in (3). Let us
denote the transmission factors by T (, ). Fusing classically this defect to a DBC a PNBC can
be obtained (4). The quantum analogue of this statement in view of (7) is
RPN (, , ) = T+ (, )RDir (, Dir )T (, ).
Comparing the reflection factor of the PNBC (13) to that of the Dirichlet one (10) and taking
into account defect unitarity and defect crossing symmetry (6) we can extract the transmission
313
factors for the defect. The simplest possible solution corresponds to = Dir and
T ( ) = i
sinh( 2
sinh( 2 +
i
4
i
4
+
+
B
2 )
;
B
2 )
T+ ( ) = i
sinh( 2
sinh( 2 +
i
4
i
4
B
2 )
.
B
2 )
(16)
All other solutions contain additional CDD type factors satisfying (6). Actually the solution (16)
itself is a CDD factor, therefore it is the simplest non-trivial solution of (6).
To find the correspondence between the parameter of the quantum transmission factor and
the Lagrangian parameter we follow the following strategy: since the boundary results are
derived in the perturbed BCFT normalization we allow not only the parameter but also Mcl to
renormalize. Their renormalized quantum values are determined from the requirement that when
fusing the defect to the DBC with (11) we obtain the PNBC with (14). The unique solution turns
out to be
b2
Me 8 = Mcl e .
(17)
The renormalization of Mcl may depend on the scheme in which the quantum potential is defined.
The b 0 limit, in which M Mcl , shows that is the quantum renormalized version of .
Also the defect energy can be extracted as the difference of the boundary energies corresponding
to the PN (15) and to the Dirichlet (12) one:
m cosh B
(18)
.
2 sin B
Summarizing, by the fusion method we were able to solve the defect theory defined by the
Lagrangian (3): The transmission factors are (16), the defect energy is (18), and the bootstrap
parameter parametrizes the Lagrangian as (17). We spend the next section to provide consistency checks of this solution.
Once the defect theory is solved we can use it to generate new integrable BCs from known
ones. In the example presented in Section 2 the defect with parameter was fused to the NBC
to generate a more general integrable BC (5). The quantum version of this fusion dresses up the
Neumann reflection factor
PN
Dir
Def () = bdry
(, ) bdry
() =
RN ( ) =
( 12 )(1
( 32
B
1
2 ) (2
B
1
2 ) (2
B
2)
B
2)
(19)
to the reflection factor

( 1 )(1 B ) ( 1
R(, ) = T+ , () RN ( )T , () = 2 3 B 2 21
(2 2 ) (2 +
iB
B
2)(
B
iB
2)(
12 )
+ 12 )
(20)
Thus we solved the more general integrable BC defined by (5) without doing any serious calculation. It is important to note, that the extra factor appearing in (20) compared to (19) is a CDD
factor. Consequently, we have determined the physical meaning of the CDD factors appearing in
the reflection factors: they represent integrable defects of the form of (5) standing in front of integrable boundaries. In principle, we can fuse as many integrable defects with various parameters
as we want, the resulting theory can be solved and its reflection factor contains the corresponding
boundary CDD factors.
Finally, we note that placing two defects with parameters = i(1 2B

) after each other
both the scattering matrix and the energy of a standing particle can be reproduced. Thus the
defect with imaginary parameter can be considered as a half particle. Similar phenomena was
observed in [23] at the classical level.
314
3.2. Solution of defect scaling LeeYang model

The scaling LeeYang model can be defined as the perturbation of the M(2,5) conformal
minimal model with central charge c = 22
5 . It contains two modules of the Virasoro algebra
corresponding to the Id and the (z, z ) primary fields with weight (0, 0) and ( 15 , 15 ), respectively. The only relevant perturbation by the field results in the simplest scattering theory with
one neutral particle of mass m and scattering matrix [35]

sinh + i sin 3
1
2
=
.
S( ) =
sinh i sin 3
3
3
The pole at = i
3 shows that the particle can form a bound-state. The relation

S i
= S( )
S +i
3
3
however, implies that the bound-state is the original particle itself and the bulk bootstrap is closed.
1
m2 .
The bulk energy constant is given by bulk =
4 3
We can impose two conformal invariant boundary conditions in the model [36,37]. They can
be labeled by I and and correspond to the highest weight representations of a single copy of
the Virasoro algebra with weight 0 and 15 , respectively. Introducing the integrable bulk perturbations with the I conformal invariant boundary condition the integrability is maintained and the
reflection factor of the particle can be written as

1
2
1
RI ( ) =
2
6
3
I
while the boundary energy is given by bdry
= m2 ( 3 1). If the conformal invariant boundary
condition corresponds to then additionally to the bulk perturbation we can introduce a oneparameter family of integrable boundary perturbations and the corresponding reflection factor
turns out to be

1
2
1
b1
b+1
5b
5 b
Rb ( ) =
2
6
3
6
6
6
6
m
b
while the boundary energy is bdry = 2 ( 3 1 + 2 sin 6 ). The boundary bound-states were
analyzed in [6] where the boundary bootstrap program was carried out.
The LeeYang model has two types of conformal defects [38], but only one of them admits
relevant chiral defect fields. They have weights ( 15 , 0), (0, 15 ). We conjecture that perturbing
in the bulk and with a certain combination of these defect fields we can maintain integrability and
arrive at a purely transmitting theory. We plan to analyze this issue systematically in a forthcoming publication. Let us denote the transmission factors of this integrable defect by T ( ). Using
that the fusion of the defect to the I boundary results in the perturbed boundary we have
Rb ( ) = T+ (, b)RI ( )T (, b).
This is supported by the fact that fusing the conformal defect to the I boundary we obtain the
boundary. Since the particle appears as a bound-state in the two particle scattering process the
transmission matrix satisfies the defect bootstrap equation [3]:

i
i
T
= T ( ).
T +
(21)
3
3
315
Using this relation together with the defect unitarity and defect crossing symmetry (6) we can fix
the transmission factor as
T ( ) = [b + 1][b 1];
[x] = i
sinh( 2 + i x
12 )
sinh( 2 + i x
12 i 2 )
(22)
(Actually the inverse of the solution is also a solution but the two are related by the b 6 + b
transformation.) The defect energy, as in the sinh-Gordon case, can be obtained as
b
.
6
We are going to recover this expression from the UV analysis of defect TBA in Section 5. We
also note that the defect with parameter b = 3 behaves as a standing particle both from the energy
and from the scattering point of view.
I
def = bdry bdry
= m sin
4. Perturbative calculations
In this section we check the exact solution of the ShG defect system in the free/classical
(b 0) limit, and develop a systematic perturbative expansion. The parameter b2 plays the
same role as h which can be seen by scaling it out from the Lagrangian via b. Since the
classical groundstate = 0 is invariant under this scaling, the b2 0 limit corresponds both to
the free and also to the classical limit. Moreover, the perturbative expansion in b2 is equivalent
both to the loop expansion and to the semi-classical approximation.
4.1. Classical/free limit
As a first step we identify the b 0 limit of the defect Lagrangian (1) as:

m2 2
m2 2
1
1
+ (x) ( + )2 cl +
L = (x) ( )2 cl
2
2
2
2

2

(x)
2
+ 2(sinh ) + .
+ + + mcl (cosh ) + +
2
Then we expand the fields on the two sides of the defect in terms of creation/annihilation operators

dk 1
+
(k)eikx+i(k)t ; (k) = k 2 + m2cl
a (k)eikxi(k)t + a
(x, t) =
2 2(k)
a, a +
where the
operators are adjoint of each other with commutators:

+
a (k), a
(k ) = 22(k)(k k ).
Imposing the defect condition (obtained by varying the action) at the origin
t x = mcl (sinh ) + mcl (cosh )
we can connect the creation/annihilation operators as
a (k) = T (k)a (k);
T (k) =
mcl sinh i(k)

; k > 0.
mcl cosh ik
316
This shows that the defect is purely transmitting, that is we do not have any reflected wave. The
transmission factor in the rapidity parametrization (k = mcl sinh ) can be written also in the
following form:
T ( ) = i
sinh( 2
sinh( 2 +
i
4
i
4
+ 2 )
+ 2 )
Clearly it has exactly the same form as the exact quantum one (16) except the B replacement. Having observed this coincidence the authors in [23] suggested that they might be the same
B = . Using our defect UVIR relation (17) we can perform the expansion:

b2
B = 1
(23)
+ .
8
The term of first order shows that our exact solution is correct in the classical limit, i.e. for
b 0. The term of second order shows that the B = relation suggested in [23] is not valid:
B acquires non-trivial quantum correction. Since the renormalization of the parameter is
crucial to decide about the two proposals we perform a perturbative check at order b2 .
4.2. Perturbation theory
As a first step we collect the free propagators. If the fields are on the same side of the defect
we have

i
d 2q

eiq(yy )
0 0|T (x, t) (x , t ) |0
0 =
2
2
2
(2) q mcl + i

= G
(y, y )
where q = (k, ) and y = (x, t). The absence of an eik(x+x ) term shows the absence of reflection.
The other two point functions are

i
d 2q

0|T
(x,
t)
(x
,
t
)
|0
=
T (, k)eiq(yy )
0
0
(2)2 q 2 m2cl + i

= G
(y, y )
where
T (, k) =
mcl sinh i
.
mcl cosh ik
In the final equations we used the fact that the contour can be closed on the upper/lower half
plane. As it was shown in [8] the reflection factor can be read off from the T ( )
propagator
317
of the fields. The defect/boundary equivalence [21] then implies that the transmission factor can
be read off from the T ( )
propagator.
The perturbation at order b2 follows from (1):
2 2

2 2

mcl b 4
mcl b 4
L = ( )
( )
+
4!
4!

2

4
mcl b2
2 2
4
2
( )
+ 4(sinh )+ +
.
+ 6+
+
+
(cosh ) +
4 4!
We calculate the propagators upto first order in b2 :

0|T (x, t) (x , t ) |0

= 0 0|T (x, t) (x , t ) 1 i d d L + |0
0 .
Using Wicks theorem we obtain the contribution of the following diagrams:
We have two bulk diagrams presented on Fig. 1, where the bulk interaction point, denoted by
an empty circle, represents z = (, ) and we have to integrate over the whole left/right space
time. Thus the contribution of the first diagram is
m2cl b2
2
Clearly
0
d
G
(z, z)
G
(z, z) =
+

dz G
(y, z)G (z, z)G (z, y ).
is divergent and we have to regularize it by introducing a cutoff :

i
d 2q
=
2
2
(2) q m2cl + i
dk
= (mcl ).
k 2 + m2cl 2
Its contribution can be absorbed into the renormalization of the mass parameter m2cl m2cl
2
m2cl b2 (mcl ) which results in extra counter term diagrams presented on Fig. 2.
The contributions from the defect terms can be grouped in two sets of diagrams. The first
contains the same divergent loop integral and consists of those on Figs. 3 and 4, where the
interaction vertex is even together with the odd diagrams presented on Fig. 4.
We have to integrate in time over the real axis and the left/right part of the defect represents
the contraction with the operators + and , respectively. They all contain the divergent and
regularized (mcl ) loop integral which can be absorbed into the renormalization of the defect
2
parameter mcl mcl mcl b4 (mcl ), which is consistent with the bulk renormalization. The
resulting counter-terms produce the diagrams on Fig. 5.
Fig. 1. The two bulk diagrams with prefactor
m2cl b2
2 .
318
The fact that all these singularities can be absorbed into the renormalization of mcl is a nontrivial statement, since we have eight divergent diagrams having different propagators on the
outer legs those we canceled just by renormalizing one single parameter in the original Lagrangian. The form of the renormalized Lagrangian is the same as the original one thus the
integrable/topological nature of the defect is not spoiled by quantum effects, there is no anomaly.
Observe also that the bulk mass term m2cl and the boundary term mcl renormalizes the same way
so the bulk is the square of the other.
The last group of the diagrams is the one which really contributes to the transmission factor.
They are presented on Fig. 6.
Fig. 2. Bulk counter-term diagrams with prefactor m2cl b2 (mcl ).
Fig. 3. Divergent even defect loop diagrams with prefactor
mcl b2
8 cosh .
Fig. 4. Divergent odd defect loop diagrams with prefactor
mcl b2
8 sinh .
Fig. 5. Defect counter terms with prefactors mcl b4 (mcl ) cosh and mcl b4 (mcl ) sinh , respectively.
Fig. 6. Defect diagrams contributing to the transmission factor. They have prefactors
respectively.
319
mcl b2
mcl b2
4 sinh and
4 cosh ,
The contribution of the first diagram is

mcl b2
sinh
4
+
+

d G
(y, z)G (z, z)G (z, y ).
Each of the terms on Fig. 6 contains the finite contribution of the propagator

i
d 2q
T (, k) = .
(z,
z)
=
G
2
(2)2 q 2 m2cl + i
This term together with the prefactor can be interpreted as the finite renormalization of the pa2
rameter + , where = b
8 . Summing up all the contributions and taking into
account the different transmission factor dependent contributions on the outer legs we obtain the
correction of order b2 to the transmission factor as

b2
1
T , B() = T (, ) +
+ O b4
8 1 i sinh( + )
which is in complete agreement with (23). Thus we confirmed the renormalization of the parameter , but we have not checked the renormalization of the parameter M which has not shown
up at this order. We suspect, however, that in this perturbative scheme the boundary parameter
mcl renormalizes as the square-root of the m2cl term and only renormalizes as B . It would be
interesting to perform a two-loop perturbative calculation to decide about the renormalization of
Mcl in the perturbative scheme.
We note that we have also performed a perturbative calculation of order b2 of the reflection

factor, which can be extracted from G+
+ (y, y ), and confirmed the absence of reflection at this
order. In [39] the form of the sine-Gordon Lagrangian was fixed at order b6 by demanding the
absence of particle creation. Following a similar line it would be tempting to see how the absence
of reflection restricts the form of the defect potential in perturbation theory.
5. Defect thermodynamic Bethe ansatz
In this section we would like to check the consistency between the transmission factors and
defect energies. In doing so we derive a DTBA to describe the ground state energy of a purely
transmitting diagonal integrable defect on the circle of perimeter L. In boundary and defect
systems there are two inequivalent ways to derive TBA equations. We can either focus on the
groundstate energy or on the so-called g-factors which is related to the finite volume normalization of defect/boundary states. Both BTBA equations have been analyzed in [16] although the
g-function type required further refinement [40]. Some details of the g-function type DTBA can
320
be found in [41] and references therein. Here we consider the groundstate DTBA in the periodic
setting as opposed to the strip geometry analyzed in [21].
5.1. Derivation of DTBA
In order to derive the DTBA equation for the ground state energy we compactify the time-like
direction with period R and calculate the partition function in two inequivalent ways by changing
the role of the space and time coordinates.
In the original description the defect is located in space as drawn in Fig. 7.
By taking the R limit the groundstate energy can be extracted from the partition function
as

lim Z(L, R) = lim Tr eH (L)R = eE0 (L)R + .
In the alternative description when the role of time and space is exchanged as shown on Fig. 8,
the defect becomes an operator of the form [21]

d
i
+
T+
a ( )a( ) ,
D = exp
2
2
Fig. 7. Defect is located in space.
Fig. 8. Defect is located in time acting as a defect operator.
321
which acts on the Hilbert space of the periodic model, H. The partition function can be calculated
as
n|D|n
eEn (R)L

Z(L, R) = Tr eH (R)L D =
(24)
.
n|n
|n
H
The Hilbert space consists of multi-particle states

|1 , 2 , . . . , n
= a + (1 )a + (2 ) a + (n )|0
;
1 > 2 > > n
on which the defect operator collects non-trivial diagonal matrix elements from
D|1 , 2 , . . . , n

i
i
i
1 T+
2 . . . T+
n |1 , 2 , . . . , n
+
= T+
2
2
2
while the energy operator acts as

H |1 , 2 , . . . , n
= m cosh(1 ) + m cosh(2 ) + + m cosh(n ) |1 , 2 , . . . , n
.
We can introduce another energy operator via H = H L1 log D such that the partition function
can be written as

eEn (R)L .
Z(L, R) = Tr eH (R)L D = Tr eH (R)L =
|n
H
This partition function can be calculated in the R limit by standard saddle point approximation taking into account the scattering of the particles. The calculation follows the usual route
of TBA calculations, but now the kinetic term is shifted m cosh m cosh L1 log T+ ( i
2 ).
As a consequence we obtain the following DTBA equations for the pseudo energy

(
) = mL cosh log T+

d
i

( ) log 1 + e ( )
2
2
(25)
where
d
() = i d
log S( ). Once is known the ground state energy can be expressed as

E0 (L) = m

d
cosh log 1 + e () .
2
Here we do not have to shift the cosh term since it comes from the derivative of the momentum
(sinh ), which appears in the quantization condition. For the result in this generality see e.g. [42].
Alternatively we can redefine the pseudo energy as ( ) = ( ) log T+ ( i
2 ) to obtain

( ) = mL cosh

d
i

( ) log 1 + T+
e( )
2
2
from which the ground state energy turns out to be

E0 (L) = m

d
i
cosh log 1 + T+
e() .
2
2
322
i
The ground state energy is real which can be easily seen form (6) since T+ ( i
2 ) = T+ ( 2 + ).
As a simple consistency check we can see that the DTBA equation for the trivial defect, T+ = 1,
reduces to the periodic TBA equation [43]. We analyze the large and small volume limits separately in the next two subsections.
5.2. Lscher type correction in defect systems

If the volume L is large then ( )
= mL cosh is large and we can expand the logarithm to
obtain

d
i
cosh T+
emL cosh + O e2mL .
E0 (L) = m
2
2
This result can be calculated directly from (24) by taking the large L limit there. In that case,
however, only the one-particle transmission term contributes which is universal for any quantum
field theory. The result obtained is the analogue of boundary Lscher type correction to the
ground state energy [44] and is valid in any theory even in non-integrable ones.
5.3. Defect energy
Here we analyze the small volume behavior of the ground state energy. Its normalization
depends on the scheme in which the quantum field theory is defined. If we would like to compare the DTBA normalization to that of a perturbed defect conformal field theory, in which the
perturbing operators have dimensions h on the defect and (h, h) in the bulk, then we have:
E0 (L) = def bulk L +
2
cn l n(1h) ;
L
l = mL.
n=0
Only the perturbative terms, cn , are present in a perturbed rational defect CFT. (In non-rational
CFT-s, like the UV limit of the boundary sinh-Gordon theory, we expect terms with logarithmic
behaviour, see [45] for the details.) By calculating the small volume limit of E0 (L) from DTBA
def and bulk can be extracted exactly. The computation is analogous to the boundary one [31,36],
so we sketch only here. In the L 0 limit the solution for in (25) develops two kink regions
around = log 2l and a breather region around the origin. The behaviour of the solutions are
determined by the asymptotics of the integral kernel and defect source term:

() = Ce|| + O e2|| ;

i
as .
= A e + O e2
log T+
2
The two kink functions are responsible for the terms giving the central charge and the bulk energy
constant, while the central/breather part gives the defect energy in the following form:
m2
m(A+ + A )
;
def =
.
(26)
2C
2C
We note that the kink type behaviour does not exists for the whole parameter range of C and A .
The results is understood that we analytically continued it from a range where the calculation is
reliable.
bulk =
323
Let us concretes the result for the two cases in question. In the sinh-Gordon model
C = 4 sinh B;
A = 2eB
so using (26) we recover (9) and (18).

In the LeeYang case we have

b+1
b1
A = 2i ei 6 + ei 6 .
C = 2 3;
Plugging these expressions back to (26) the results confirms the bulk energy density and the
defect energy.
We emphasize that the agreement obtained in the two cases confirm the solutions on one side
and the DTBA equation on the other.
The other perturbative coefficients cn can be calculated from the DTBA only numerically.
In the LeeYang case, however, one can gain further analytical information. On has to define
Y ( ) = e() and to show from (21) that it satisfies the LeeYang Y -system relation

i
i
Y +
= 1 + Y ( )
Y
3
3
from which the Y ( ) = Y ( + 5i
3 ) periodicity follows. Similarly to the boundary case this gives
the exponent of the perturbative expansion to be 65 showing that the dimension of the perturbing
operator is h = 15 .
6. Defect bound-states and bootstrap closure
In this section we analyze the analytic structure of the transmission factors for the whole
range of their parameters. Since the sinh-Gordon transmission factors (16) never have poles
in the physical strip we focus on the LeeYang model only. Recall that in our convention the
physical strip of the transmission factors T ( ) are [0, i
2 ].
6.1. Pole analysis on the ground-state defect
We analyze the pole structure of both
T ( ) = [b + 1][b 1]
and
T+ ( ) = [5 b][5 b]
as the function of the parameter b, simultaneously. We note that by folding the theory to a boundary one (with two particles) we could analyze its bootstrap in the usual boundary formulation.
Here, however, we present the results in the defect language since the diagrams are more clearcut. (The whole procedure, going from the reflection factor to the defect transmission factor, can
be interpreted as taking a sort of square root of the boundary theory. By closing the defect bootstrap we would like to show that such a theory is indeed sensible. Since on the I boundary there
is no boundstate any pole of the reflection factor appears either in T or in T+ so the bootstrap
will be very similar to the boundary one [6].)
In determining the fundamental range of the parameter b we can see that b b + 12 is a
symmetry. Moreover b 6 b exchanges T T+ so we can restrict ourselves to the range
b [3, 3]. We will see by analyzing the defect excited states that the fundamental range is even
smaller, only b [3, 2], as in the boundary case, since at b = 2 the role of the ground-state and
the first excited state is exchanged.
324
The poles and zeros of the transmission factor T ( ) are
poles of T are at = i (b 5);

6
The analogous expressions for T+ are
zeros of T are at = i (b 1).

6
(b 1);
zeros of T+ are at = i (b 5).
6
6
They can be drawn as the function of the parameter b as shown on Fig. 9.
For b [1, 2] there is a pole in the transmission factor T+ at = iu = i 6 (b + 1), for which
we associate a defect boundstate and denote it by |1+
. Its energy is m cos 6 (b + 1) and the
corresponding excited transmission factor can be calculated from the defect bootstrap equation
shown on Fig. 10.
poles of T+ are at = i
|1+
( ) = T ( )S( + iu).
|1+
From the defect crossing symmetry (6) we can calculate T+

|1+
T+
|1+
( ) = T
( ) as
(i ) = T (i )S(i + iu) = T+ ( )S( iu)
which is consistent with the other bootstrap equation where the second particle arrives from the
right. The resulting transmission factors are
|1+
( ) = [b + 1][b + 3];
|1+
T+
( ) = [5 b][3 b].
Fig. 9. Poles and zeros of T+ and T as function of b. Solid lines correspond to poles, while dashed ones to zeroes. The
dotted lines show the fundamental range.
Fig. 10. Defect bootstrap equations.
325
|1+
They are related to the groundstate ones as T (b 4 b, ) = T (b, ). This symmetry

together with the defect energies indicate that when b exceeds 2 the role of the ground-state
and the excited state |1+
are exchanged. This confirms that the fundamental range is indeed
b [3, 2].
In the range b [1, 2] the transmission factor T+ ( ) has another pole at = i 6 (b 1) for
which we associate the defect boundstate |2+
. It has energy m cos 6 (b 1) and transmission
factor

|2+
T ( ) = T ( )S i (b 1) .
6
Now we turn to the pole analysis of excited defect states.
6.2. Pole analysis on the exited defect state |1+
The poles and zeros of the transmission factors on the state |1+
are indicated on Fig. 11.
The state exist in the b [1, 2] domain so we have to explain the poles in this range only.
|1+
The pole of T+ labeled by 1 on Fig. 11 is at the same location as the one which creates the
excited state itself, namely at = i 6 (b + 1) in the full range b [1, 2]. It can be explained by
the first of the defect ColemanThun diagrams on Fig. 12.
|1+
The pole of T+ labeled by 2 on Fig. 11 is at = i 6 (b + 3) and can be explained by the

second diagram on Fig. 12. Observe that by applying the Cutkosky rules [9] we would obtain a
Fig. 11. Poles and zeros on the |1+

defect excited state. Dotted line shows the range where the excited state |1+
exists.
Fig. 12. Defect ColemanThun diagrams for the |1+

state. The angles are measured in units of i
6 .
326
pole of second order but the transmission factor T has a first order zero at = i 6 (1 b) which,
in this way, reduces the order of the pole to one.
|1+
The pole of T labeled by 3 on Fig. 11 is at = i 6 (3 b). In the range b [0, 1] it can

be explained by the third diagram on Fig. 12. Since the transmission factor T on the ground
state has a zero at = i 6 (1 b) the order of the diagram is reduced to one again. In order
for the diagram to exist the particle has to travel towards the defect, that is 1 b > 0. This
explains the pole in the range b [0, 1]. In the range b [1, 2] the particle creates a defect
boundstate which is nothing but |2+
. This can be seen both from the energy of the excited state
m cos 6 (b + 1) + m cos 6 (3 b) = m cos 6 (b 1) and from the transmission factor. If the left
particle creates a defect boundstate at rapidity = iu then the excited states transmission factors
are Tex ( ) = T ( )S( iu). Now we can see from the bulk bootstrap equation that

|1+
T ( )S i (3 b) = T ( )S i (b + 1) S i (3 b)
6
6
6

|2+
= T ( )S i (b 1) = T ( )
6
that is the transmission factors also supports the identification.
6.3. The pole analysis on the excited defect state |2+
The defect boundstate labeled by |2+

has transmission factor
|2+
( ) = [b 1][b + 1]2 [b + 3];
|2+
T+
( ) = [3 b][5 b]2 [7 b].
The singularity structure can be summarized as follows.

|2+
The pole labeled by 5 on Fig. 13 is in T+ ( ) at = i 6 (b 1) and can be explained in the

full range b [1, 2] by the first diagram on Fig. 12, if we replace |1+
by |2+
.
|2+
The pole labeled by 6 on Fig. 13 is in T ( ) at = i 6 (3 b) and can be explained by a

diagram similar to the third one of Fig. 12 in which the |1+
state is replaced by |2+
and the
vacuum |0
is replaced by |1+
.
|2+
The pole labeled by 7 on Fig. 13 is a second order one in T+ ( ) at = i 6 (b + 1) and can

be explained by the two diagrams on Fig. 14. Clearly the transmission factor T ( ) does not have
zeros neither at = i 6 (3 b) nor at = i 6 (b 1) so the pole is of second order.
By now we explained all the poles of all the transmission factors of the ground and excited
defect states. We used either the creation of a new defect boundstate or presented the appropri-
Fig. 13. Singularity structure of the transmission factors on the |2+

state. Bold straight lines represent poles of second
order. The relevant interval where the boundstate |2+
exists is indicated by dotted lines.
327
Fig. 14. Defect ColemanThun diagrams for the excited state |2+
.
ate defect ColemanThun diagram which was responsible for the singularity. By finishing this
procedure the spectrum become complete and we managed to define a sensible defect theory. It
would be nice to check these findings by the defect truncated conformal space approach (TCSA).
7. Conclusions
We have demonstrated how the fusion idea can be used to solve topological defects in the sinhGordon and LeeYang models. In the sinh-Gordon case we determined the transmission factors
and the defect energy as a function of a bootstrap parameter whose relation to the Lagrangian
was also given. We checked these results in perturbation theory and against the newly derived
DTBA.
In the LeeYang case we determined the transmission factors together with the defect energy
and checked them in DTBA. For certain range of the parameter the transmission factor admits
poles in the physical strip. We closed the defect bootstrap programme: we explained all poles
either by associating new defect boundstates or by giving the appropriate defect ColemanThun
mechanism both for the groundstate and for excited defect states.
The relation obtained between the transmission parameter and that of the Lagrangian in the
sinh-Gordon theory can be analytically continued to describe the analogues relation in the sineGordon theory. This result also passes the test of first order perturbation theory and together with
the transmission factors obtained in [22,23] gives the complete solution of defect sine-Gordon
model. We have checked this solution by performing the fusing procedure on the solitonic transmission factors. This is analogous to the dressing procedure in the XXZ spin chain developed
in [46].
The perturbation theory developed here can also be used in higher rank affine Toda theories to
connect the parameters of the transmission factor of the bootstrap solution [24] to the parameters
of their Lagrangians [18].
The derivation of the DTBA generalizes to any diagonal scattering theory. A large and small
volume analysis analogous to the one presented in the paper will provide the leading finite size
correction to the groundstate energy and give the bulk/defect energies, respectively.
In the present paper we were concerned with the bootstrap (IR) description of our models.
There is a need, however, to understand their UV behavior which probably can be described by
perturbed defect CFTs. To connect these alternative descriptions we can use methods starting
328
either from the IR side, like DTBA, or starting from the UV side, like defect TCSA. There are
works in progress in both directions.
Acknowledgements
Z.B. thanks Laszl Palla, Gbor Takcs, Gerard Watts and Cristina Zambon for the useful
discussions. Z.B. was supported by a Bolyai Scholarship, OTKA K60040 and the EC network
Superstring.
References
[1] G. Delfino, G. Mussardo, P. Simonetti, Statistical models with a line of defect, Phys. Lett. B 328 (1994) 123.
[2] G. Delfino, G. Mussardo, P. Simonetti, Scattering theory and correlation functions in statistical models with a line
of defect, Nucl. Phys. B 432 (1994) 518.
[3] O.A. Castro-Alvaredo, A. Fring, F. Ghmann, On the absence of simultaneous reflection and transmission in integrable impurity systems, hep-th/0201142.
[4] V. Caudrelier, M. Mintchev, E. Ragoucy, P. Sorba, Reflection-transmission quantum YangBaxter equations,
J. Phys. A 38 (2005) 34313442.
[5] S. Ghoshal, A. Zamolodchikov, Boundary S-matrix and boundary state in two-dimensional integrable quantum field
theory, Int. J. Mod. Phys. A 9 (1994) 38413886;
S. Ghoshal, A. Zamolodchikov, Int. J. Mod. Phys. A 9 (1994) 4353, Erratum.
[6] P. Dorey, R. Tateo, G. Watts, Generalisations of the ColemanThun mechanism and boundary reflection factors,
Phys. Lett. B 448 (1999) 249256.
[7] A. Fring, R. Kberle, Factorized scattering in the presence of reflecting boundaries, Nucl. Phys. B 421 (1994)
159172.
[8] Z. Bajnok, G. Bhm, G. Takcs, Boundary reduction formula, J. Phys. A 35 (2002) 93339342.
[9] Z. Bajnok, G. Bhm, G. Takcs, On perturbative quantum field theory with boundary, Nucl. Phys. B 682 (2004)
585617.
[10] V. Riva, Boundary effects in two-dimensional critical and off-critical systems, hep-th/0106268.
[11] E. Corrigan, A. Taormina, Reflection factors and a two-parameter family of boundary bound states in the sinhGordon model, J. Phys. A 33 (2000) 87398754.
[12] P. Mattsson, P.E. Dorey, Boundary spectrum in the sine-Gordon model with Dirichlet boundary conditions, J. Phys.
A 33 (2000) 90659094.
[13] Z. Bajnok, L. Palla, G. Takcs, G.Zs. Tth, The spectrum of boundary states in sine-Gordon model with integrable
boundary conditions, Nucl. Phys. B 622 (2002) 548564.
[14] G. Zsolt Tth, Investigations in two-dimensional quantum field theory by the bootstrap and TCSA methods, arXiv:
0707.0015.
[15] E. Corrigan, On duality and reflection factors for the sinh-Gordon model, Int. J. Mod. Phys. A 13 (1998) 27092722.
[16] A. LeClair, G. Mussardo, H. Saleur, S. Skorik, Boundary energy and boundary states in integrable quantum field
theories, Nucl. Phys. B 453 (1995) 581618.
[17] P. Bowcock, E. Corrigan, C. Zambon, Classically integrable field theories with defects, Int. J. Mod. Phys. A 19S2
(2004) 8291.
[18] P. Bowcock, E. Corrigan, C. Zambon, Affine Toda field theories with defects, JHEP 0401 (2004) 056.
[19] J.F. Gomes, L.H. Ymai, A.H. Zimerman, Classical integrable super-sinh-Gordon equation with defects, J. Phys.
A 39 (2006) 74717483;
J.F. Gomes, L.H. Ymai, A.H. Zimerman, Classical Integrable N = 1 and N = 2 super-sinh-Gordon models with
jump defects, arXiv: 0708.2407.
[20] V. Caudrelier, On a systematic approach to defects in classical integrable field theories, arXiv: 0704.2326.
[21] Z. Bajnok, A. George, From defects to boundaries, Int. J. Mod. Phys. A 21 (2006) 10631078.
[22] R. Konik, A. LeClair, Purely transmitting defect field theories, Nucl. Phys. B 538 (1999) 587611.
[23] P. Bowcock, E. Corrigan, C. Zambon, Some aspects of jump-defects in the quantum sine-Gordon model, JHEP 0508
(2005) 023.
[24] E. Corrigan, C. Zambon, On purely transmitting defects in affine Toda field theory, JHEP 0707 (2007) 001.
[25] V.B. Petkova, J.-B. Zuber, The many faces of Ocneanu cells, Nucl. Phys. B 603 (2001) 449496.
329
[26] K. Graham, G.M.T. Watts, Defect lines and boundary flows, JHEP 0404 (2004) 019.
[27] I. Runkel, Perturbed defects and T-systems in conformal field theory, arXiv: 0711.0102.
[28] R.E. Behrend, P.A. Pearce, Integrable and conformal boundary conditions for sl(2) A-D-E lattice models and unitary
minimal conformal field theories, hep-th/0006094.
[29] C.H. Otto Chui, C. Mercat, W. Orrick, P.A. Pearce, Integrable lattice realizations of conformal twisted boundary
conditions, Phys. Lett. B 517 (2001) 429435.
[30] P. Bowcock, Boundaries and defects, 3rd EUCLID Network School, 2006 Trieste, Italy.
[31] Al. Zamolodchikov, unpublished;
Z. Bajnok, L. Palla, G. Takacs, Finite size effects in boundary sine-Gordon theory, Nucl. Phys. B 622 (2002) 565
592.
[32] Al. Zamolodchikov, Mass scale in sine-Gordon and its reductions, Int. J. Mod. Phys. A 10 (1995) 11251150.
[33] C. Destri, H. deVega, New exact results in affine Toda field theories: Free energy and wave function renormalizations, Nucl. Phys. B 358 (1991) 251294.
[34] S. Ghoshal, Bound state boundary S matrix of the sine-Gordon model, Int. J. Mod. Phys. A 9 (1994) 48014810,
hep-th/9310188.
[35] J.L. Cardy, G. Mussardo, S matrix of the LeeYang edge singularity in two dimensions, Phys. Lett. B 225 (1989)
275278.
[36] P. Dorey, A. Pocklington, R. Tateo, G. Watts, TBA and TCSA with boundaries and excited states, Nucl. Phys. B 525
(1998) 641663, hep-th/9712197.
[37] P. Dorey, M. Pillin, R. Tateo, G.M.T. Watts, One-point functions in perturbed boundary conformal field theories,
Nucl. Phys. B 594 (2001) 625659, hep-th/0007077.
[38] T. Quella, I. Runkel, G.M.T. Watts, Reflection and transmission for conformal defects, hep-th/0611296.
[39] P. Dorey, Exact S-matrices, hep-th/9810026.
[40] P. Dorey, D. Fioravanti, C. Rim, R. Tateo, Integrable quantum field theory with boundaries: The exact g-function,
Nucl. Phys. B 696 (2004) 445467.
[41] O. Castro-Alvaredo, A. Fring, From integrability to conductance, impurity systems, Nucl. Phys. B 649 (2003) 449
490.
[42] F. Woynarovich, O(1) contribution of saddle point fluctuations to the free energy of Bethe Ansatz systems, Nucl.
Phys. B 700 (2004) 331360.
[43] Al.B. Zamolodchikov, Thermodynamic Bethe ansatz in relativistic models. Scaling three state Potts and LeeYang
models, Nucl. Phys. B 342 (1990) 695720.
[44] Z. Bajnok, L. Palla, G. Takcs, Finite size effects in quantum field theories with boundary from scattering data,
Nucl. Phys. B 716 (2005) 519542.
[45] Z. Bajnok, C. Rim, Al. Zamolodchikov, Sinh-Gordon boundary TBA and boundary Liouville reflection amplitude,
arXiv: 0710.4789.
[46] Z. Bajnok, Equivalences between spin models induced by defects, J. Stat. Mech. 0606 (2006) P010.

Four loop MS mass anomalous dimension

in the GrossNeveu model
J.A. Gracey
Theoretical Physics Division, Department of Mathematical Sciences, University of Liverpool, PO Box 147,
Liverpool, L69 3BX, United Kingdom
Received 18 March 2008; accepted 7 April 2008
Abstract
We compute the four loop term of the mass anomalous dimension in the two-dimensional GrossNeveu
model in the MS scheme. The absence of multiplicative renormalizability which results when using dimensional regularization means that the effect of the evanescent operator, which first appears at three loops in
the 4-point Greens function, has to be properly treated in the construction of the renormalization group
function. We repeat the calculation of the three loop MS -function and construct the -function of the
evanescent operator coupling which corrects earlier computations.
1. Introduction
The GrossNeveu model is a two-dimensional asymptotically free renormalizable quantum
field theory whose basic interaction is a simple quartic fermion self-interaction, [1]. It has been
widely studied since its introduction in [1], as it has many interesting properties which are readily accessible given the spacetime dimension the model is defined in. For instance, unlike the
same interaction in four dimensions it is renormalizable in two dimensions and the generation of
mass dynamically has been observed and studied in the large N expansion, [1]. Moreover, one
property of interest is that it possesses an S-matrix whose exact form is known, [2,3], whence
the mass gap is known exactly, [4], in terms of the basic mass scale of the theory, MS . Aside
from these features the model itself underpins several problems in condensed matter physics. For
instance, in the replica limit it is equivalent at the critical point to the two-dimensional random
E-mail address: gracey@liv.ac.uk.
doi:10.1016/j.nuclphysb.2008.04.002
J.A. Gracey / Nuclear Physics B 802 [FS] (2008) 330350
331
bond Ising model. (See, for example, the review [5].) Necessary to study the fixed point properties for such physical problems is knowledge of the renormalization group functions in some
renormalization scheme, such as MS. These have been computed to three loops in MS over a period of years. In [1] the one loop -function was computed demonstrating asymptotic freedom.
This was extended to two loops in [6], whilst the three loop -function appeared more or less
simultaneously in [7] and [8]. Though the method of computation in both articles was significantly different. For instance, given the quartic nature of the sole interaction it can be rewritten
in terms of an auxiliary field producing a trivalent interaction with the introduction of an auxiliary field. This was the version of the theory used in [8], as well as at two loops in [6], not only
to deduce the -function but also to study the effective potential of the auxiliary field at three
loops. However, renormalization effects will generate a quartic interaction. So [8] had in effect to
handle the intricate problem of renormalizing a version of the theory with two independent couplings. The three loop MS -function of the original theory was eventually extracted when the
effect of the newly generated interaction was properly accounted for in the renormalization group
equations. By contrast, in [7] the purely quartic version of the theory was treated with a massive
fermion. The agreement of both three loop results was a reassuring non-trivial check on the final
expression. At four loops only the wave function renormalization has been computed in [9] and
later verified in [10]. Although apparently one loop further than the -function of [7,8], or mass
anomalous dimension, [11], the fermionic nature of the interaction means that the anomalous dimension begins at two loops since the one loop snail graph is zero in the wave function channel
of the 2-point function. Thus in effect the four loop wave function is a computation on a par with
the three loop mass anomalous dimension and -function.
Given the range of problems which the GrossNeveu model underlies, it is the purpose of
this article to start the programme of completing our knowledge of the four loop structure by
computing the mass anomalous dimension at this order in the MS scheme. This may appear
to be the least important of the two outstanding quantities. However, as will become apparent
from the calculation we will describe, the nature of the model is such that there are several difficult technical issues to be dealt with en route which do not arise in other four loop calculations
in other important theories. Therefore, in one sense we are testing the viability of computing
renormalization constants at four loops in an example in the GrossNeveu model which contains a moderate number of Feynman diagrams rather than the 1000 plus graphs which will
occur in the 4-point function renormalization and with a sensible investment of time. Moreover,
the experience gained in such an exercise will be invaluable for any future such coupling constant renormalization. It will transpire that we will require properties of Feynman integrals in
two spacetime dimensions higher than the one we compute in, in order to determine the final
renormalization group function. Whilst this may not be the unique way to determine this, it will
suggest the importance of the structure of higher dimensional Feynman diagrams in complementing practical lower dimensional calculations and potentially equally vice versa at high loop
order. Our main tool of computation will be the use of dimensional regularization in d = 2
dimensions where the relevant part of the 2-point function is written in terms of basic massive
vacuum bubble graphs. Such a calculation could only be completed with the use of computer
algebra and invaluable in this was the symbolic manipulation language F ORM, [12]. At this level
of loop order automatic Feynman diagram computation using computers is the only viable way
of keeping a reliable account of the algebra within a reasonable time. Several other computer
packages were also required.
As this is the first four loop calculation which involves four terms of the renormalization
group functions, we will have to revisit and redo the earlier calculations using the same approach
332
consistently as that which will be used here at four loops. For reasons which will become evident
later this has entailed us carrying out the full renormalization of the 4-point function in the
theory with a massive fermion, where the mass will act as a natural infrared regulator. In [7]
and [8] these articles centered on the derivation of the three loop -function itself, which unlike
most field theories, is not the same as fully renormalizing the underlying n-point function. For
the GrossNeveu model this was first observed in [13,14] with explicit three loop calculations
for the full 4-point function given for a massless version of the theory discussed in the series
of articles [1517] and examined in [18] for the massive version. However, neither computation
was in complete agreement as to the final structure of the 4-point function renormalization. Prior
to considering any four loop mass anomalous dimension this discrepancy needs to be resolved in
one way or another which is a secondary consideration for this article. Any four loop -function
computation would also require this resolution but constructing the mass anomalous dimension
in a consistent way is an easier environment in which to check any final explanation.
The paper is organized as follows. We review the current understanding of the renormalization
properties of the GrossNeveu model in Section 2. Given this we discuss the three loop vacuum
bubble integrals needed for the full three loop renormalization in Section 3. There we resolve the
discrepancy between [17] and [18] in the 4-point renormalization in Section 4. Section 5 extends
the discussion of the relevant vacuum bubble computations to four loops with the four loop mass
anomalous finally being derived in Section 6. Concluding remarks are given in Section 7.
2. Preliminaries
We turn now to the specific properties of the GrossNeveu model. The bare two-dimensional
Lagrangian is, [1],
2
1
L = i 0i / 0i m0 0i 0i + g0 0i 0i ,
(2.1)
2
where the subscript 0 denotes a bare quantity and g is the coupling constant which is dimensionless in two dimensions. Unlike [7,11] which used the symmetry group O(N) we take the
SU(N ) version of the theory so that the fermion field i is a complex Majorana fermion with
the former property deriving from the fields taking values in the group SU(N ) with 1 i N .
An advantage of the choice of the group SU(N ) is that the Feynman rule of (2.1) involves two
terms unlike the three of the O(N) case. At four loops this reduces the number of terms needed
to be substituted when the Feynman rules are implemented in F ORM which speeds up the calculation. We choose to work with the massive version where m is the mass. This is primarily
because we will need to redo the three loop renormalization of the coupling and the presence of
a non-zero mass will ensure that any resulting divergences are purely ultraviolet and not deriving
from spurious infrared infinities which could arise when external momenta are nullified in the
basic divergent 4-point Greens function. In two dimensions the theory (2.1) is renormalizable to
all orders in the coupling constant and is asymptotically free. Specifically we note that the MS
scheme renormalization group functions of the model, as they currently stand are, [1,69,11,17],
g2
g3
(N
1)(2N
1)
8 2
16 3
4

g
+ 4N 2 14N + 7 (2N 1)
+ O g5 ,
4
128

g
g3
g2
m (g) = (2N 1)
+ O g4 ,
+ (2N 1) 2 + (4N 3)(2N 1)
2
8
32 3
(g) = (2N 1)
333

g2
g4
g3
+ O g 5 , (2.2)
+ (N 1) 2 + (N 1)(2N 7)
4
2
16
where (g), m (g) and (g) are respectively the field and mass anomalous dimensions and
the -function. Their formal definitions will be discussed later. Although several terms were
determined for the O(N) version of the model, we have converted the previous computations to
the SU(N ) model whence the free field case emerges when N = 12 as indicated by the vanishing
of (g) and m (g) for this value.
In the computations deriving (2.2) the main strategy was to dimensionally regularize (2.1) in
d-dimensions and determine the renormalization constants as poles in the deviation from two
dimensions. Here we will take d = 2 with being regarded as small. Whilst the correct
renormalization group functions emerged at three loops, [8] overlooked a novel feature of the
dimensionally regularized Lagrangian which was explicitly discussed in [15,16] after the observation in [13,14]. Basically (2.1) ceases being multiplicatively renormalizable in d-dimensions
but crucially retains renormalizability. This is not a property solely restricted to the GrossNeveu
model but is a feature of any two-dimensional model with a 4-fermi interaction such as the
Abelian and non-Abelian Thirring models, [18]. At a certain loop order, which is different for
different models, evanescent operators are generated through the renormalization which are nontrivial in d-dimensions but which are absent or evaporate in the limit to strictly two dimensions
which corresponds to the lifting of the regularization. A comprehensive study of this problem
was provided for general 4-fermi theories in [13,14] and we recall those features which are relevant for our ultimate goal. The same problem in four dimensions has been considered in [19,20].
Though there 4-fermi operators are of course non-renormalizable and treated in the context of
effective field theories.
First, in d-dimensions the basis of -matrices based on the Clifford algebra

, = 2
(2.3)
(g) = (d 2)g (N 1)
...
has to be extended to the set of objects (n)1 n for integer n 0, [1317], which is totally
antisymmetric in the Lorentz indices and defined by
2 ...n
(n)1
= [1 2 . . . n ] ,
(2.4)
where we use the convention that the square brackets include division by n! when all possible
...
permutations of the -strings are written explicitly. Then (n)1 n form a complete closed basis
for -matrices in d-dimensions where (0) is the unit matrix. Hence one can immediately see that
the most general multiplicatively renormalizable 4-fermi theory using dimensional regularization
in d-dimensions is, [1317],
L = i 0i / 0i m0 0i 0i +
1
...
g(n) 0 0i (n)1 n 0i 0i (n) 1 ...n 0i ,
2
(2.5)
n=0
where there is an infinite number of (bare) couplings g(n) 0 with g(0) g identified as the original
one of the GrossNeveu model (2.1). Though the GrossNeveu model strictly will correspond to
the case where g(1) = g(2) = 0 as (1) and (2) are not evanescent. Given (2.5), there are several
points of view depending on the problem in hand. If (2.5) is the most general renormalizable
theory in d-dimensions, then in principle for the GrossNeveu model one must begin with (2.5)
but omit g(1) and g(2) . This will produce renormalization group functions dependent, in principle,
on all evanescent couplings. The true renormalization group functions of the original theory
would eventually emerge from this multiplicatively renormalizable theory by setting g(n) = 0 for
334
n 3 at the end, [1317]. Clearly this would involve a significant amount of calculation much of
which would be redundant in the production of the final renormalization group functions. From
a practical point of view there is a less laborious route to follow if one abandons the insistence
on multiplicative renormalizability, [17,18]. Then operators such as
1
...
On = i (n)1 n i i (n) 1 ...n i
(2.6)
2
for n 3 will be generated with g(0) g dependent coefficients. The problem for this point
of view then becomes one of how to extract the true two-dimensional renormalization group
functions. It turns out that a formalism was developed in [13,14] and used in [10,18] for this
evanescent operator issue. In essence the true renormalization group functions are not strictly
determined from what we term the naive renormalization constants. By these we mean those
required to render 2- and 4-point functions finite. Instead these naive renormalization group
functions need to be amended by the effect the evanescent operators have on the divergence
structure in d-dimensions relative to two dimensions. In [13,14] such a projection formula was
(k)
introduced which involves projection functions, (k) (g), m (g) and C (k) (g), where the index
k ranges over the evanescent range k 3. These functions quantify the effect the evanescent
operators have on the derivation of the renormalization group functions. The derivation of the
projection formula is given in [14] and applied additionally in [10,18]. We recall that it is

d

m
+ 2gO0 ]
d x N [Ok ] g =0,d=2 = d d x (k) (g)N [i /
(i)

(k)
m
(2.7)
(g)N [m]
+ C (k) (g)N [O0 ] g =0,d=2 ,
(i)
where the normal ordering symbol, N , is included, [13,14,21,22]. The relation strictly only has
meaning when inserted in a 2- or 4-point Greens function. In other words one inserts the evanescent operator of the left side of (2.7) in a Greens function and evaluates it using the naive
renormalization constants to yield a finite expression. Equally one inserts the left side of (2.7)
into the same Greens function to the same loop order and renders it finite. Then the coefficients
of the perturbative expansion in the coupling constant g are chosen order by order to render the
equation consistent at each loop order after one has set d = 2. This procedure is denoted by the
restriction {g(i) = 0, d = 2} on both sides of (2.7). Once the explicit projection functions have
been determined to the appropriate order, then the true renormalization group functions are given
by, [13,14],
(g) = (g)
+
(g) = (g) +

k=3
C (k) (g)k (g),

(k) (g)k (g),
k=3
m (g) = m (g) +
(k)
m
(g)k (g),
(2.8)
k=3
where denotes the naive renormalization group functions.

For the GrossNeveu model the first appearance of an evanescent operator is at three loops
which was originally observed in [14,17]. Whilst this postdates the three loop MS -functions
of [7,8] the latter are unaffected by the generation of O3 since it occurs with a coupling dependence of g 3 . So that coupled with C (3) (g) it will only affect the -function itself at four loops.
335
Equally the mass anomalous dimension of [11] does not feel this evanescent operator presence
until four loops either. We refrain from quoting the value of the associated -function, 3 (g),
until later. This is primarily because there are two competing values given in [17] and [18]. In the
former the renormalization was deduced in a massless version of (2.1) where is was claimed that
only ladder style diagrams were the origin of O3 . In that instance the nullification of two external
momenta in the associated 4-point function should not have resulted in spurious infrared singularities. Whilst a 3 (g) was determined, it involved (3) which was not found in [18] which used
the massive version, (2.1), where (x) is the Riemann zeta function. This clearly avoided infrared
singularities when all the external momenta were nullified in the 4-point function. Moreover it
was claimed that the diagrams leading to O3 were akin to those analyzed in [17] but with no
(3) appearing in the published value of 3 (g). Though both calculations agreed on the rational
part of 3 (g). The discrepancy between both computations needs to be resolved and a four loop
calculation which requires 3 (g) explicitly to obtain the true renormalization group function will
provide a non-trivial forum in which to achieve this. The correct expression for 3 (g) will be
crucial for the four loop -function. Given this structure of the GrossNeveu model we can now
write down the renormalized form of (2.1) we will use. It is, [17,18],

2
1
L = iZ i / i mZ Zm i i + g Zg Z2 i i
2

2
1
+ g Z33 Z2 i (3) i ,
(2.9)
2
where the renormalized quantities are defined from their bare counterparts by
1
0 = Z2 ,
m0 = mZm ,
g0 = gZg
(2.10)
in d-dimensions and Z33 absorbs the infinity associated with the generation of O3 at this order.
Unlike Z , Zm and Zg its coupling constant expansion does not commence with unity. Though
we stress that (2.9) is valid only for 2-point calculations to four loops. Only by renormalizing the
4-point function at four loops would the full evanescent operator structure at that order emerge.
For instance, it is not inconceivable given the -matrix structure of the four loop 4-point function
that an O4 evanescent operator will be generated. From these renormalization constants the naive
renormalization group functions are given by
ln Z ,
m (g) = (g)
ln Zm ,
g
(g)
= (d 2)g g (g)
(2.11)
ln Zg
g
to the order we are working to. For 3 (g) one deduces its explicit form from the simple pole in
via standard methods, [14]. Thus in the context of (2.8) and these observations, we note that for
(3)
the mass anomalous dimension the result of [18] for m
(g) is

g
(3)
(g) = + O g 2 .
m
(2.12)
The higher terms are not required since the first term of 3 (g) is O(g 3 ).
(g) =
3. Three loop calculations

We begin this section by discussing the computational strategy. To determine the mass anomalous dimension for (2.1) we consider the 2-point function for the massive theory. In [9] the four
336
Fig. 1. One loop contribution to the 2-point function.
loop MS anomalous dimension was calculated and independently verified in [10]. Therefore,
we assume that result for Z . However, this is effectively a three orders calculation since the
one loop snail of Fig. 1 corresponding to (p) (p) has no non-zero contributions in the
p
/ channel for the massive or massless Lagrangians where p is the external momentum. Moreover, since for this case one is interested only in Z , it sufficed to consider the massless theory
whence one only has to determine massless Feynman integrals. The component involving p
/
can be deduced by multiplying all diagrams by p
/ and taking the spinor trace. For the mass dimension one cannot follow this strategy. Not only because the one loop diagram contributes but
also because its contribution to Zm requires the presence of the mass itself. Therefore unlike
the determination of Z one cannot neglect the snail graph of Fig. 1 at one loop as well as
the graphs where snails appear as subgraphs at higher order. However, given that one is only
interested in the m channel of (p) (p) the Greens function can be analyzed by nullifying the external momentum. Taking the spinor trace produces Lorentz scalar integrals but
with tensor structure resulting from internal momenta contractions. The presence of the common mass m automatically protects against the appearance of spurious infrared infinities and
relegates the problem of determining the ultraviolet structure to mapping the integrals with internal momenta contractions to a set of basic master vacuum bubbles at each loop order. The
problem of studying massive vacuum bubbles in four dimensions has received much attention
over the years, culminating in, for example, the M ATAD package at three loops, [24], and the
comprehensive study by Broadhurst of all combinations of massive and massless propagators
in the Benz or tetrahedron topology, [25]. The analogous problem relative to two dimensions
has not been treated as systematically. Though the main results to three loops have appeared
within various articles. Additionally, at four loops we will have to handle new integrals for
topologies which do not simply break into products of lower loop vacuum bubbles. The main
difficulty lies in having to handle the tensor structure emanating from the fermion propagator.
Throughout we have made extensive use of the symbolic manipulation language F ORM, [12],
in which to code our algorithm where the contributing Feynman diagrams to the 2- and 4-point
functions are generated automatically with the Q GRAF package, [23]. To summarize there are
1 one loop, 2 two loop, 7 three loop and 36 four loop graphs for the 2-point function. For
the 4-point function there are 3 one loop, 18 two loop and 138 three loop graphs to renormalize.
At three loops we now summarize the few master (massive) integrals which will be of interest
to us, in our notation and conventions. First, Fig. 2 denotes the basic vacuum bubbles to two
loops. We define the first graph of Fig. 2 as

I =i
k
[k 2
1
.
m2 ]
(3.1)
337
Fig. 2. One and two loop vacuum bubbles.
We work in Minkowski space and choose to include a factor of i with each integration measure
which is abbreviated by

dd k
=
.
(3.2)
(2)d
k
The integral I is trivial to deduce from the Euler -function as

I=
(1 d/2) 2 d/21
m
.
(4)d/2
Hence the middle graph of Fig. 2 is I 2 .

The final graph we denote by (0), [11], where

J (k 2 )
(0) = i
[k 2 m2 ]
(3.3)
(3.4)
and

J p2 = i

k
[k 2
1
p)2 m2 ]
m2 ][(k
(3.5)
is the basic one loop self-energy bubble. There is a sequence of integrals related to J (p) defined
by

1
J (p) = i
(3.6)
,
[k 2 m2 ] [(k p)2 m2 ]
k
where we choose J21 (p) K(p 2 ). In this form a Gauss relation of the hypergeometric functions
gives the relation

2
p 4m2 K p 2 = J (0) (d 3)J p 2
(3.7)
with
J (0) =
(2 d/2) 2 d/22
m
(4)d/2
since explicit calculations produce
2
(2 d/2) 4m2 p 2 d/22
p2
d 1 3
J p =
2 F1 2 , ; ; 2
4
2 2 2 p 4m2
(4)d/2
(3.8)
(3.9)
338
Fig. 3. Three loop vacuum bubbles.
and
K p
(3 d/2) 4m2 p 2 d/23

p2
d 1 3
=
.
2 F1 3 , ; ; 2
4
2 2 2 p 4m2
2(4)d/2
Likewise, at the subsequent level

2
p 4m2 J22 p 2 + 2J31 p 2 = 2K(0) 2(d 5)K p2
(3.10)
(3.11)
whence
(d 6) 2

(p 2 2m2 )
+
K
p
J31 p 2 =
K(0) (d 5)K p 2 ,
2
2
2
2
2p
p (p 4m )
2

(d 6)
4m2
J22 p =
K p2 + 2 2
K(0) (d 5)K p 2 .
2
2
p
p (p 4m )
(3.12)
These rules are used extensively for the two and higher loop Feynman integrals. The integral
(0) is finite in two dimensions and can be evaluated in an expansion in powers of as
9s2
+ O(),
(3.13)
16 2 m2
where s2 = (2 3/9) Cl2 (2/3) with Cl2 (x) the Clausen function. The analogous fourdimensional vacuum bubble also contains s2 in its finite part but is divergent. In principle the
O() term of (3.13) can be deduced. However, throughout our computations we left (0) itself
unevaluated since on renormalizability grounds it must be absent from the final renormalization constants at higher loops. This is because if the 2-point function did not have its external
momenta nullified then the integral (p) would emerge, where

J (k 2 )
.
(p) = i
(3.14)
[(k p)2 m2 ]
(0) =
Clearly such a non-local function of the external momenta could not be retained when all the
counterterms are included.
At three loops there are several more basic topologies. If one ignores for the moment the
complication due to the presence of internal momenta contractions in integral numerators, then
the basic graphs are given in Fig. 3. The first involves I and its derivatives with respect to m2 .
Also the second is a variation on (0) and we note that the two loop subgraph is

J (k 2 )
(d 3)
=
(0)
i
(3.15)
2
2
2
[k m ]
3m2
k
339
which is established by differentiating (0) with respect to m2 . Aside

from the Benz topology
the

remaining vacuum bubbles of Fig. 3 are related to the integrals i k J 2 (k 2 )/[k 2 m2 ], i k J 2 (k 2 )

and i k J (k 2 )K(k 2 ). Similar to (0) these are finite in two dimensions but their values are
required at four loops when multiplied by counterterms. As only the leading term in is required
in each case it transpires that we can set d = 2 and use the fact that in Euclidean space, denoted
by the subscript E,

2

J (0)
JE k d=2 =
(3.16)
sinh
d=2
upon the change of variables k 2 = 4m2 sinh2 (/2) where J (0)|d=2 = 1/(4m2 ). Then, for
instance,

2
m2 2
2
J k = J (0)
d
+ O().
2
sinh
d=2
2
i
k
(3.17)
This can be evaluated from standard integrals to give

7 (3)
i J 2 k2 =
+ O().
64 3 m2
(3.18)
Equally we find

11 (3)
J 2 (k 2 )
=
+ O().
i
2
2
[k m ] 576 3 m4
(3.19)
Though in d-dimensions one can derive the relation

2 2 (3d 8)
i J 2 k2 .
i J k K k =
8m2
k
(3.20)
K(k 2 )
in several of the master integrals with topology similar to those of Fig. 3

The presence of
produces similar finite integrals whose finite part is required and which is determined in an
analogous way. We note that

ln(2)
J (k 2 )

=
+ O(),
J
(0)
i

2
2
2
[k 4m ]
d=2
k

2

7
J 2 (k 2 )
J (0)
i
(3.21)
=
+ O().
(3)
ln(2)
2
2
8
2 d=2
[k 4m ]
k
For the mass anomalous dimension these basic vacuum bubbles suffice to determine the renormalization constants to three loops. Given the nature of the 4-point interaction in (2.1) the Benz
topology does not occur in the 2-point function at this loop order.
Having discussed the basic scalar master integrals which result we briefly note the algorithm
dealing with the numerator structure of the integrals. This has been systematically quantified
in [11]. However, we note that repeated use of
kl =

1 2
k + l 2 (k l)2 m2 m2
2
(3.22)
340
and then

k 2 = k 2 m2 + m2 ,

l 2 = l 2 m2 + m2
(3.23)
in each contributing topology where there are [k 2 m2 ], [l 2 m2 ] and [(k l)2 m2 ] propagators
already. This is done in such a way that powers of kl can remain when all mixed [(k l)2 m2 ]
propagators are absent and one does not then continue substituting for kl. In such integrals one
can use Lorentz symmetry in the k and l subgraph integrals to redefine even powers of kl as
proportional to k 2 l 2 or zero if there are an odd number of factors of kl. Then (3.23) is repeated.
Consequently several variations in the basic bubble graphs of Fig. 3 emerge and we note that

4 3 4 2

2 2 2
i k J k = I + m i J 2 k2 ,
3
3
k

i
2 2 2 2
k J k =

8m2
3
2
2 2
(5d 6)I + 2dm i J k ,
3(3d 4)
(3.24)
where these are exact and no finite parts have been omitted since these are crucial for the next loop
order. In essence this summarizes the key ingredients in the algorithm for evaluating the three
loop mass anomalous dimension which has been coded in F ORM and reproduces the previous
three loop MS GrossNeveu mass anomalous dimension.
4. Three loop 4-point function renormalization
At this point we turn to our secondary aim which is to resolve the discrepancy in the renormalization associated with the generation of O3 . This requires the complete determination of
the 4-point function divergence structure at three loops. Whilst the algorithm to do this is very
similar to that of the 2-point function there are several key differences. First, the 4-point function
divergences will be independent of the external momenta which means that they can be immediately nullified. The mass again protects against spurious infrared divergences. However, we
cannot now take the Lorentz traces since that would prevent one from seeing the emergence of
any (3) (3) -matrix structure. Instead we have to retain -strings and also unentangle
the internal momenta within them. Hence one decouples the Feynman diagrams into -strings
and Lorentz tensor vacuum bubbles. At one and two loops the resulting tensor integrals for the
whole integral can be straightforwardly reduced by noting that at one loop

(4.1)
k k f1 k 2 =
k 2 f1 k 2 ,
d
k
where k is the sole loop momentum and at two loops

k1 1 k2 2 k3 3 k4 4 f2 (k, l)
kl
1
d(d 1)(d + 2)

f2 (k, l) (d + 1)k1 . k2 k3 . k4 k1 . k3 k2 . k4 k1 . k4 k2 . k3 1 2 3 4
kl

+ (d + 1)k1 . k3 k2 . k4 k1 . k2 k3 . k4 k1 . k4 k2 . k3 1 3 2 4

+ (d + 1)k1 . k4 k2 . k3 k1 . k2 k3 . k4 k1 . k3 k2 . k4 1 4 2 3 ,
341
(4.2)
where ki {k, l} and in the Lorentz tensor of the integrand all possible combinations of the two
internal momenta are covered. The functions fi ({ki }) represent the various possible propagator
combinations. For clarity we have included the dot of the scalar products explicitly. At three
loops the situation is complicated by the observation that the extension of both these formula
gives

k1 1 k2 2 k3 3 k4 4 k5 5 k6 6 f3 (k, l, q)
klq
1 2 3 4 5 6
=
d(d 1)(d 2)(d + 2)(d + 4)

f3 (k, l, q) d 2 + 3d 2 k1 . k2 k3 . k4 k5 . k6
klq
(d + 2)k1 . k2 k3 . k5 k6 . k4 (d + 2)k1 . k2 k3 . k6 k4 . k5
(d + 2)k1 . k3 k2 . k4 k5 . k6 + 2k1 . k3 k2 . k5 k6 . k4
+ 2k1 . k3 k2 . k6 k4 . k5 (d + 2)k1 . k4 k2 . k3 k5 . k6
+ 2k1 . k4 k2 . k5 k6 . k3 + 2k1 . k4 k2 . k6 k3 . k5
+ 2k1 . k5 k2 . k3 k4 . k6 + 2k1 . k5 k2 . k4 k6 . k3
(d + 2)k1 . k5 k2 . k6 k3 . k4 + 2k1 . k6 k2 . k3 k4 . k5
+ 2k1 . k6 k2 . k4 k5 . k3 (d + 2)k1 . k6 k2 . k5 k3 . k4
+ 14 similar terms,

(4.3)
where ki {k, l, q}. The full decomposition is clearly quite large. However, it is the appearance
of the 1/(d 2) factor which is novel. In [17,18] the full set of three loop graphs in both massless
and massive cases where a divergent (3) (3) structure emerged, was noted. The sets of
graphs appear to be the same. Though in [17] it is not fully clear which the actual ladder graphs
referred to are. However, the seemingly finite graph of Fig. 4 was regarded as fully finite in all
-string channels, [18]. In our present reconsideration it transpires that within the integral of
the graph of Fig. 4 there is a divergent contribution to the (3) (3) channel but not for
the (0) (0) one. This derives from the pole 1/(d 2) in (4.3) producing the massive Benz
integral corresponding to the final graph of Fig. 3. The key part is then

1
1
,
2
2
2
2
2
2
2
(d 2)
[k m ][l m ][q m ][(k l) m2 ][(k q)2 m2 ][(l q)2 m2 ]
klq
(4.4)
where the actual integral itself is finite in two dimensions. It remains after repeated application
of (3.22) and (3.23) in the scalar integrals of (4.3). However, to have the complete divergence
structure the integral needs to be evaluated since it will contribute to Z33 . The remaining integrals
with this 1/(d 2) pole in the (3) (3) channel correspond to (4.4) but with one or more
propagators removed after application of (3.22) and (3.23). These can be evaluated from the three
loop techniques discussed earlier. In [18] this contribution, (4.4), was overlooked since it was
assumed that the parent integral with the internal momenta contracted was finite without noting
the possibility of the 1/(d 2) factor deriving from the tensor decomposition. In relation to [17]
we can only comment that in the massless version of (4.4) the integral will be zero. However,
342
Fig. 4. Three loop contribution to 4-point function.
given the totally different method of calculating the 4-point function of [18] in the massless case,
a contribution analogous to (4.4) could possibly arise elsewhere.
There remains the task now of evaluating the integral of (4.4). Although finite it clearly cannot
be reduced to any of the three loop master vacuum bubbles already discussed even using, say,
integration by parts. Instead we have had to resort to the more extensive experience of fourdimensional vacuum bubble diagrams and specifically the Benz graphs discussed in [25]. To
promote (4.4) to four dimensions we exploit Tarasovs observation of relating d-dimensional
integrals to (d + 2)-dimensional ones, [26,27]. Moreover, this is straightforward to do via the
TARCER package, [28], written in M ATHEMATICA for the basic two loop self energy topology
given in Fig. 5. Specifically one feature of TARCER is that one can relate this two loop self
energy graph in d-dimensions to that in (d + 2)-dimensions. This is a subgraph of Fig. 4 with
nullified external momenta and given that this is a three loop vacuum bubble, the final three loop
integration measure can be rewritten as

dd k
d d+2 k 1
(4.5)
=
2d
d
(2)
(2)d+2 k 2
in our conventions since the two loop subgraph will clearly be a function of k 2 only. From (4.5) a
massless propagator will appear in the higher dimensional integral. Since all the lines of Figs. 4
and 5 are massive and both final integrations involve functions of the square of the momentum,
then we find the relation between the d-dimensional massive Benz graph and similar topologies
in two dimensions higher is

Be 1, 1, 1, 1, 1, 1, m2 , m2 , m2 , m2 , m2 , m2 , d

1
J 2 (k 2 )
3
2 2
=
k
i
J
i
12m4
4m2
[k 2 m2 ]
k

d(d 1)(d 2)
Be 1, 1, 1, 1, 1, 1, m2 , m2 , m2 , m2 , m2 , m2 , d + 2
+
m6

Be 1, 1, 1, 1, 1, 1, 0, m2 , m2 , m2 , m2 , m2 , d + 2 ,
(4.6)
where we define

Be , , , , , , m21 , m22 , m23 , m24 , m25 , m26 , d

1
= i3
2
2
2
klq
[k 2 m1 ] [l 2 m2 ] [q 2 m3 ] [(k l)2 m24 ] [(k q)2 m25 ] [(l q)2 m26 ]
(4.7)
343
Fig. 5. Two loop self-energy topology central to TARCER.

and emphasize that k indicates a d-dimensional integration. The key part is the piece which
represents the difference in two Benz topologies in (d + 2)-dimensions where one is completely
massive and the other has one massless line. However, since these are multiplied by (d 2) then
in our expansion relative to two dimensions we note that the leading term of each is O()
meaning that

1
3
i
[k 2 m2 ][l 2 m2 ][q 2 m2 ][(k l)2 m2 ][(k q)2 m2 ][(l q)2 m2 ]
klq
(3)
(4.8)
+ O()
192 3 m6
from (3.18) and (3.19). This is because whilst each of the two (d + 2)-dimensional integrals are
divergent in four dimensions due to the presence of a simple pole in the regularizing parameter, the difference in (4.6) is finite and the residue is independent of the masses in either Benz
topology, [25].
With this observation all the ingredients are assembled to repeat the full three loop renormalization of the 4-point function of (2.1). In [7] only the (0) (0) part was isolated but this
was sufficient to deduce the -function at three loops. It is satisfying to record that we have
verified the previous three loop MS result of [7,8]. However, by contrast we find that a different
renormalization constant from [17] and [18] emerges for Z33 . We find

(3)
1 g3
+ O g4
Z33 =
(4.9)
3
64
48
=
whence

3 (3)
1 g3
+ O g4 .
3 (g) =
64
16 3
(4.10)
Though there is universal agreement on the rational part of (4.10), [17,18], only the contribution
from the diagram of Fig. 5 to the (3) (3) channel produces the irrational piece thereby
confirming the overall structure observed in [17]. However, rather than finding that we produce
one of the previous values for Z33 we are in the seemingly unfortunate position of finding a new
alternative. To determine which is correct and consistent it will transpire that the four loop mass
anomalous dimension is the correct testing ground for this in the context of (2.8).
5. Four loop vacuum bubbles
In this section we return to out initial aim and summarize the evaluation of the underlying four
loop vacuum bubbles required for the mass anomalous dimension. For the 2-point function there
are 18 distinct topologies and 36 Feynman diagrams to be considered. Of these topologies 14
involve snail insertions in one way or another and hence their determination is in effect relegated
344
Fig. 6. Several four loop diagrams contributing to the 2-point function.
to the straightforward extension of the three loop topology discussion. One effect of a snail
is to produce two propagators on a line of a three loop graph but this can be reproduced by
differentiating that line with respect to m2 . This is also a reason why the three loop vacuum
bubbles were required to be evaluated to the finite part exactly or left in terms of (0) and
other known integrals whose expansion could be substituted when required, if at all. Several
topologies contributing to the 2-point function, however, have a more demanding evaluation.
These are illustrated in Fig. 6 and we concentrate on these for the main part. Essentially the main
complication now derives from rewriting the scalar products of internal momenta in terms of the
propagator structure. For all the integrals which result we used several interconnected techniques.
First was the use of the TARCER package, [28], again, particularly for the third and fourth
graphs of Fig. 6. Clearly the last graph contains the two loop self-energy topology of Fig. 5 as
a subgraph and the third has a similar two loop subgraph but with one line removed. Unlike
the properties of TARCER we described previously, the feature exploited in this instance is the
ability to relate diagrams with different powers of the propagators in Fig. 5 to that with unit
power. Further, TARCER reduces integrals involving powers of the scalar products kl, kp and
lp where k and l are internal and p is the external momentum in Fig. 5. The point is that the
Lorentz tensor reduction for these situations can only be performed by this route. Any one loop
subgraph of Fig. 5 will involve three external legs and the invariant decomposition in this case is
too intricate. By contrast where possible we did exploit the Lorentz structure of subgraphs with
one internal momentum flowing through it which can be regarded as a 2-point function external
momentum for that subgraph. Then integrals can be rewritten using results such as

1
ll
1
=
i
i
2
2
2
2
2
2
[l m ][(k l) m ] (d 1)
[l m ][(k l)2 m2 ]
l

kl 2
kl 2
kk
l 2 2 2 l 2 d 2
.
k
k
k
(5.1)
The outcome of the TARCER implementation is to reduce these more complicated tensor integrals
to a set of master scalar four loop vacuum bubbles since the resulting combination of internal
momenta allows for the repeated application of (3.22) and (3.23).
345
The use of (5.1) and TARCER though may appear to introduce potential infrared difficulties.
However, it transpires that in the full sum of all contributing pieces to a Feynman graph it can
be checked that no integral retains an unprotected factor of 1/k 2 which would give an infrared
divergence upon integrating over the internal momentum k. For one instance checking this proved
to be a tedious non-trivial exercise which we document for completeness. In all bar the second
graph of Fig. 6 the following combination of integrals emerges

2
(k)
V = i
(5.2)
J k J (0)
.
k2
k
Clearly each could be infrared divergent but the above combination always appears. Defining

k
K (p) = i
(5.3)
2
2
2
[k m ] [(k p)2 m2 ]
k
then one can show

1

1
p K (p) = 2m2 K p 2 (d 4)J p 2 = p 2 K p 2 + J p 2 J (0) .
(5.4)
2
2
Using this and integration parts in (5.3) one finds

2
1 2 2
(k 2 m2 )J (k 2 )J (l 2 )
2
V = (d 3)i

k
2i
+
i
J
k
(k)K
k
. (5.5)
2
k
l 2 [(k l)2 m2 ]2
k
kl
The final integral can be reduced using TARCER if one regards the k momentum as external to the
self-energy graph of Fig. 5 and J (l 2 ) is replaced by the Feynman integral of (3.5). Consequently,
TARCER produces

J (l 2 )
i
l 2 [(k l)2 m2 ]2
l

1
(d 2)2 (d 4)I 2
(d 2)(d 3)
i
2(d 3)m2 [k 2 m2 ]2
2m2 [k 2 m2 ]
l 2 [(k l)2 m2 ]
l

1
(d 4)(3d 8) 2
i
+
[k 2 m2 ]2
[l 2 m2 ][(k q)2 m2 ][(l q)2 m2 ]
lq
[(d

i2
lq
2)[k 2
m2 ] 8(d
4)m2 ]
[k 2 m2 ]2
1
.
[l 2 m2 ]2 [(k q)2 m2 ][(l q)2 m2 ]
(5.6)
The benefit of rearranging the two loop integral of the left-hand side is to isolate the potential
infrared singularity into a simple term on the right-hand side. Moreover, the appearance of powers
of 1/[k 2 m2 ] will lead to simplifications when substituted back into the expression for V and
the term with the singularity will actually combine with the first term of (5.5) to produce V but
with a factor of (d 3). Hence, evaluating the remaining integrals of (5.6) in the context of (5.5)
346
one arrives at the expression

(d 2)2 I 2 (0)
V = i (k)K k 2
2(d 3)m2
k
(3d 8)i
k
J (k 2 )(k)
+ 8m2 i 2
[k 2 m2 ]

kl
[k 2
J (k)K(l)
m2 ][(k l)2 m2 ]
(5.7)
which has no potential infrared singular term. Moreover, each term of the right side of this is
ultraviolet finite in two dimensions. So, in fact, when the combination V appears in our computation, it can actually be dropped as there is no contribution to the renormalization of the mass at
four loops.
Having completed the tensor reduction of the scalar propagators, all that remains is the evaluation of a set of divergent four loop master integrals akin to those discussed earlier. Most of
these are elementary given the results of (3.7), (3.11) and (3.12) and the observation that as all
integrals are infrared finite then one can ignore those four loop ones which are clearly ultraviolet
finite by the usual counting rules. Though one integral is worth recording and that is

2 2 3 2

(7d 13)m2 I
k J k = 2I 4 +
i J 2 k2
i
(2d 5)
k
k

2(d 1)(d 3) 4
+
(5.8)
m i J 3 k2
(2d 5)(d 2)
k
because in the determination of this relation a pole in (d 2) emerges in the standard ddimensional manipulations such as differentiating the original integral with respect to m2 and
using (3.7). This pole gives rise to a problem similar to that discussed for (4.4) but with a simpler
resolution since we merely apply the technique used to deduce (3.18) and (3.19), to find

3 (3)
+ O().
i J 3 k2 =
(5.9)
256 4 m4
k

The need to evaluate i k J 2 (k 2 ) to the finite part as well is also illustrated by this equation.
This completes the discussion of the construction of the relevant basic Feynman integrals.
For each of the eighteen topologies a F ORM module was created within which the algorithm
to break the original Feynman graphs up into its basic components was encoded. The tedious
identification with the above results together with the remaining more elementary ones were
also contained in each module. Finally, prior to summing all the results from the 36 four loop
diagrams, the expansion of I and other integrals were evaluated to the appropriate order in .
The resulting sum produced the divergent part of the mass component of the 2-point function to
the simple pole in as a function of the bare parameters.
6. Four loop renormalization
The final piece of the calculation rests in determining the overall renormalization constant Zm
at four loops in MS. However, prior to this we must consider the full theory. To this point we
have tacitly assumed that only the basic 4-point vertex of (2.1) is responsible for all the Feynman
diagrams we have discussed. The presence of the generated evanescent operator in (2.9) needs
347
Fig. 7. Contribution from evanescent operator to the four loop mass renormalization.
to be included. As noted earlier since the operator appears with a coupling g 4 the effect of this
operator cannot arise before four loops. Therefore, we now have to include the additional graph
of Fig. 7 where the circle with a cross in it denotes the insertion of the operator O3 with its
associated renormalization constant Z33 . The integration routine to determine its contribution is
the same as that for the original vertex except that one has to first replace the (3) matrices by
the corresponding string of ordinary -matrices.
With this additional graph included the overall renormalization constant is extracted using
the standard method for automatic Feynman diagram computations developed in [29]. Briefly
one computes the Greens function of interest as a function of all the bare parameters such as
the coupling constant and the mass. Then the renormalized parameters are introduced by the
rescaling defined by the renormalization constants. In the present context these are the renormalization constants leading to the naive anomalous dimensions as defined in (2.9) and (2.10).
This rescaling in effect reproduces the counterterms to remove subgraph divergences. Moreover
the Greens function is multiplied by the associated renormalization constant which in our case
and conventions is Z Zm . As the former is already known, [9], then the divergences which remain in the 2-point function are absorbed by the unknown pieces of Zm . We recall that at four
loops the anomalous dimension of [9] corresponds to the naive anomalous dimension (g) since
there is no contribution from the graph of Fig. 7 in the wave function channel. Therefore, having
followed this procedure we find the naive mass anomalous dimension in MS is
g
g3
g2
+ (2N 1) 2 + (4N 3)(2N 1)
m (g) = (2N 1)
2
8
32 3

3
2
3
+ 48N 384N + 492N 138 (3) 40N
72N 2 + 160N 81
g4

+ O g5 .
4
384
(6.1)
At this stage several comments are necessary. First, there are several checks on the underlying
renormalization constant itself. Whilst the evanescent operator issue arises at four loops, it will
manifest itself in the simple pole in of Zm . Therefore, the quartic, triple and double poles
in are in fact already predetermined by the structure of previous loop order poles from the
renormalization group equation. For (6.1) we have verified that this is in fact correct. One other
useful check was the explicit cancellation of divergences of the form (0)/ n for n = 1 and 2 at
four loops. This is non-trivial since, for instance, (0) arises at three loops both associated with
a simple pole in and in the finite part. Therefore, one needs to write (0) as a formal expansion
in powers of prior to the rescaling of the bare quantities. This is because the O(1) piece at three
loops will be multiplied by 1/ poles. Moreover, since (0) has dependence (m2 )d3 , then this
has to be explicitly factored off since this mass is bare and needs to be renormalized too. Once
written in this way we have checked that the poles in involving the O(1) and O() residues
stemming from the expansion of (0) do indeed cancel completely.
348
Again one can partially check part of (6.1) from another point of view. In [9] the structure
of the mass anomalous dimension has been given in the large N expansion to O(1/N 2 ) based
on the results from a series of articles [3036]. Again at this level of expansion the evanescent
operator is not manifested and so the O(1/N 2 ) coefficients of the mass anomalous dimension
which are given there at four and higher loops in fact equate to those of the naive mass anomalous
dimension m (g). In other words if it were possible to compute the critical exponent corresponding to the mass anomalous dimension at the d-dimensional fixed point of the theory at the next
order in large N , O(1/N 3 ), then unless the effect of the evanescent operator could be included,
it would not correspond to the true mass anomalous dimension, [9]. From the expression given
in [9] we note that when the same convention is used, that part of (6.1) at four loops which corresponds to the O(1/N 2 ) piece agrees precisely with [9]. This is a reassuring cross-check on a
significant part of our four loop computation since, within the computer setup, one can examine the N -dependence multiplying all the basic integrals which we have had to compute for all
topologies. The vast majority are at least touched by a quadratic or cubic in N which are related
respectively to the O(1/N 2 ) or O(1/N) large N piece already determined in [9]. For the small
number of remaining pieces which have linear factors in N we have been careful in evaluating
the corresponding, though invariably simple, vacuum bubbles. Therefore, we are confident that
(6.1) is correct.
One clear problem remains which is related to the structure of the expression (6.1). Unlike
the previous orders the four loop part does not vanish when N = 12 which corresponds to the free
field theory. Moreover, it transpires that of the eighteen underlying topologies only the graphs
for one do not vanish for this value for N . (Though actually the parts from the second and third
graphs of Fig. 6 cancel between each other which is similar to what occurs at three loops for
analogous graphs.) The topology which gives a contribution for N = 12 is the final graph of
Fig. 6. However, given our discussion in several places concerning the evanescent operator, the
resolution is clearly straightforward. More concretely one can see the evidence for this if one
evaluates (6.1) at N = 12 to find

g4

m (g)N= 1 = 3 (3) 4
(6.2)
+ O g5 .
4
2
64
This is the piece which needs to be cancelled in order to have a mass dimension consistent with
a free field theory. Indeed this is the relative combination of rationals and (3) which our three
loop 4-point function renormalization reevaluation produced. Therefore, using (2.8) and (4.10)
we can derive the true mass anomalous dimension as
g
g3
g2
+ (2N 1) 2 + (4N 3)(2N 1)
2
8
32 3

g4
+ 12(2N 13)(N 1) (3) 20N 2 46N + 57 (2N 1)
384 4
5
+O g .
m (g) = (2N 1)
(6.3)
Clearly this has the correct expected N = 12 property and given our earlier checks on (6.1) we will
regard (6.3) as the completion of our original aim. Also, it is worth stressing that the discrepancy
in the 4-point function renormalization has now been crucially resolved simultaneously. It turns
out that neither of the previous expressions for 3 (g), [17,18], could be correct to preserve the
vanishing of m (g) in the free field case. So we can regard this mass anomalous dimension
calculation as also a non-trivial check on the full three loop MS renormalization.
349
7. Discussion
We have completed the four loop renormalization of the mass anomalous dimension of the
GrossNeveu model in the MS scheme. Despite the lack of multiplicative renormalizability when
the Lagrangian is regularized dimensionally, it has been possible to compute an expression which
passes all possible internal checks. Not least of these is the correct implementation of the projection formula formalism of [13,14] which has been justified by the consistency with the free
field case. Concerning this the previous attempts to deduce 3 (g) appear to indicate that the only
approach which is truly reliable for renormalizing the model is the one where there is a non-zero
mass. This seems to be the conclusion one must draw from the origin of the necessary (3) part
missing from [17] required to balance the discrepancy of (6.2). Given these remarks one possible
extension would now be to repeat the derivation of the mass anomalous dimension at four loops
in other 4-fermi models in two dimensions. Whilst considering the most general possible interactions involving O0 , O1 and O2 of [13,14] would perhaps be too ambitious, there is the interesting
case of the non-Abelian Thirring model, [37,38]. The seed interaction involves O1 but includes
color group generators too. It has been renormalized at three loops in MS in [18] and the four
loop wave function is also known, [10]. Though in light of our comments on the 4-point function
in the GrossNeveu model, the corresponding 4-point function renormalization would clearly
need to be reconsidered to deduce the correct evanescent operator -functions. One motivation
for determining the mass anomalous dimension in the non-Abelian Thirring model would be to
examine the color group Casimir structure of the final expression since, given the similarity with
QCD, it is thought that it should involve the same structures as the corresponding expression for
the quark mass anomalous dimension, [39,40]. This was the case for the wave function, [10].
Acknowledgements
The author thanks the Max Planck Institute for the Physics of Complex Systems, Dresden,
Germany where part of this work was carried out. Also the author is grateful to Dr. R. Mertig for
assistance with setting up TARCER.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
D. Gross, A. Neveu, Phys. Rev. D 10 (1974) 3235.

A.B. Zamolodchikov, A.B. Zamolodchikov, Ann. Phys. 120 (1979) 253.
A.B. Zamolodchikov, A.B. Zamolodchikov, Nucl. Phys. B 133 (1978) 525.
P. Forgacs, F. Niedermayer, P. Weisz, Nucl. Phys. B 367 (1991) 123.
B.N. Shalaev, Phys. Rep. 237 (1994) 129.
W. Wetzel, Phys. Lett. B 153 (1985) 297.
J.A. Gracey, Nucl. Phys. B 367 (1991) 657.
C. Luperini, P. Rossi, Ann. Phys. 212 (1991) 371.
N.A. Kivel, A.S. Stepanenko, A.N. Vasilev, Nucl. Phys. B 424 (1994) 619.
D.B. Ali, J.A. Gracey, Nucl. Phys. B 605 (2001) 337.
J.A. Gracey, Nucl. Phys. B 341 (1990) 403.
J.A.M. Vermaseren, math-ph/0010025.
A. Bondi, G. Curci, G. Paffuti, P. Rossi, Ann. Phys. 199 (1990) 268.
A. Bondi, G. Curci, G. Paffuti, P. Rossi, Phys. Lett. B 216 (1989) 349.
A.N. Vasilev, M.I. Vyazovskii, S. Derkachov, N.A. Kivel, Theor. Math. Phys. 107 (1996) 441.
A.N. Vasilev, M.I. Vyazovskii, S. Derkachov, N.A. Kivel, Theor. Math. Phys. 107 (1996) 359.
A.N. Vasilev, M.I. Vyazovskii, Theor. Math. Phys. 113 (1997) 1277.
J.F. Bennett, J.A. Gracey, Nucl. Phys. B 563 (1999) 390.
350
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
M.J. Dugan, B. Grinstein, Phys. Lett. B 256 (1991) 239.

S. Herrlich, U. Nierste, Nucl. Phys. B 455 (1995) 39.
W. Zimmermann, Ann. Phys. 77 (1973) 536.
W. Zimmermann, Ann. Phys. 77 (1973) 570.
P. Nogueira, J. Comput. Phys. 105 (1993) 406.
M. Steinhauser, Comput. Phys. Commun. 134 (2001) 335.
D.J. Broadhurst, Eur. Phys. J. C 8 (1999) 311.
O.V. Tarasov, Phys. Rev. D 54 (1996) 6479.
O.V. Tarasov, Nucl. Phys. B 502 (1997) 455.
R. Mertig, R. Scharf, Comput. Phys. Commun. 111 (1998) 265.
S.A. Larin, J.A.M. Vermaseren, Phys. Lett. B 303 (1993) 334.
S.. Derkachov, N.A. Kivel, A.S. Stepanenko, A.N. Vasilev, hep-th/9302034.
A.N. Vasilev, S.. Derkachov, N.A. Kivel, A.S. Stepanenko, Theor. Math. Phys. 94 (1993) 179.
A.N. Vasilev, A.S. Stepanenko, Theor. Math. Phys. 97 (1993) 364.
J.A. Gracey, Phys. Lett. B 297 (1992) 293.
J.A. Gracey, Int. J. Mod. Phys. A 6 (1991) 395;
J.A. Gracey, Int. J. Mod. Phys. A 6 (1991) 2755, Erratum.
J.A. Gracey, Int. J. Mod. Phys. A 9 (1994) 567.
J.A. Gracey, Int. J. Mod. Phys. A 9 (1994) 727.
R. Dashen, Y. Frishman, Phys. Lett. B 46 (1973) 439.
R. Dashen, Y. Frishman, Phys. Rev. D 11 (1975) 2781.
J.A.M. Vermaseren, S.A. Larin, T. van Ritbergen, Phys. Lett. B 405 (1997) 327.
T. van Ritbergen, J.A.M. Vermaseren, S.A. Larin, Phys. Lett. B 400 (1997) 379.

Separation of variables in the open XXX chain

Holger Frahm , Alexander Seel, Tobias Wirth
Institut fr Theoretische Physik, Leibniz Universitt Hannover, Appelstr. 2, 30167 Hannover, Germany
Received 11 March 2008; accepted 11 April 2008
Abstract
We apply the Sklyanin method of separation of variables to the reflection algebra underlying the open
spin- 12 XXX chain with non-diagonal boundary fields. The spectral problem can be formulated in terms of
a TQ-equation which leads to the known Bethe equations for boundary parameters satisfying a constraint.
For generic boundary parameters we study the asymptotic behaviour of the solutions of the TQ-equation.
PACS: 02.30.Ik; 75.10.Pq
1. Introduction
Exact solutions of spin chain models have provided many insights into the properties of interacting many-body systems subject to strong quantum fluctuations. Various methods have been
established to study the spectrum and nature of their low-lying excitations of such models as
well as their thermodynamical properties and even correlation functions without the need to revert to perturbational approaches. At the same time various problems concerning systems with
open boundaries are still not solved completely. Even for the prototype spin- 12 XXZ chain with
general open boundary conditions techniques for the solution of the spectral problem have been
developed only recently [15]. This model, apart from being the simplest starting point for studies of boundary effects in a correlated system, allows to investigate the approach to a stationary
state in one-dimensional diffusion problems for hard-core particles [6,7] and transport through
E-mail addresses: frahm@itp.uni-hannover.de (H. Frahm), alexander.seel@itp.uni-hannover.de (A. Seel),

tobias.wirth@itp.uni-hannover.de (T. Wirth).
doi:10.1016/j.nuclphysb.2008.04.008
352
H. Frahm et al. / Nuclear Physics B 802 [FS] (2008) 351367
one-dimensional quantum systems [8]. Its Hamiltonian is given by

H=
L1
j =1

y y
jx jx+1 + j j +1 + jz jz+1 1 cosh cosh

2 x
y
z
cosh + i1 sinh
sinh
+ 1 coth +
sinh 1

2 + x
y
L cosh + + iL sinh + sinh ,
+ Lz coth + +
+
sinh
(1.1)
where j , = x, y, z, denote the usual Pauli matrices at site j and the parameters , ,
refer to the left and right boundary respectively. The Hamiltonian (1.1) is a member of a commuting family of operators generated by a transfer matrix based on a representation of Sklyanins
reflection algebra [9]. While this establishes the integrability the actual solution of the spectral
problem through application of Bethe ansatz methods has been impeded by the absence of a
reference state, such as the ferromagnetically polarized state with all spins up for the case of diagonal boundary fields. Interestingly, by imposing certain constraints obeyed by the left and right
boundary fields, the eigenvalues of the spin- 12 XXZ chain and of the isotropic spin-S model can
be obtained by means of the algebraic Bethe ansatz [1,3]. In an alternative approach, Nepomechie
et al. have been able to derive Bethe type equations whose roots parametrize the eigenvalues of
the Hamiltonian for special values of the anisotropy = i/(p + 1) with p a positive integer and
where the transfer matrix obeys functional equations of finite order [4]. Their approach relies on
the periodicity of the underlying trigonometric R-matrix of the model which is missing in the
rational limit 0 of the isotropic chain. For generic values of the anisotropy the spectral problem has been formulated as a TQ-equation assuming that the large j -limit of the transfer matrices
with spin j in auxiliary space exists [10]. No such constraints are needed in the derivation of a
different set of recursion relations for the diagonalization of (1.1) based on the representation
theory of the q-Onsager algebra [10a]. Very recently, Galleas has formulated another functional
approach to determine the eigenvalues of (1.1) in the generic case without a reference state [5].
Taking certain matrix elements of the transfer matrix involving the ferromagnetic pseudo vacuum and the unknown eigenstates he derives the Bethe equations of Nepomechie et al. without
the need to restrict the anisotropy to roots of unity. Unfortunately, this approach yields no information on the eigenstates, therefore further studies of correlation functions are out of reach for
now.
In this paper we approach the problem by means of a different method which circumvents
the difficulties of the algebraic Bethe ansatz in the absence of a reference state. Within Sklyanins functional Bethe ansatz (or separation of variables method) [11] the eigenvalue problem is
formulated using a suitably chosen representation of the underlying YangBaxter algebra on a
space of certain functions. This approach was independently confirmed by use of Manin matrices
[12] and has proven its strength in particular in models with non-compact target space lacking
a reference state, e.g., the Toda chain [13] or the sinh-Gordon model [14,15] where the spectral
problem can be formulated in terms of separable functional equations on this space which then
have to be solved by exploiting the analytical properties of the problem.
The article is organized as follows. In Sections 2 and 3 we will review the basic properties of
the reflection algebra and its representation for the open XXZ spin chain. Then in Section 4 we
will implement the functional Bethe ansatz for the restriction to the open XXX spin chain following Sklyanins original article [11]. Finally in Section 5 we address the problem of extracting the
353
solution of the spectral problem from the resulting second order difference equations and study
the analytical properties of their eigenfunctions. Section 6 is devoted to a concluding summary.
2. Integrable boundary conditions
Sklyanins construction [9] of integrable systems involving boundaries is valid for a general class of integrable systems characterized by an R-matrix of difference form R(, ) =
R( ) End(V V ) (V is a vector space with dim V N) which not only satisfies the Yang
Baxter equation
R12 ( )R13 ( )R23 ( ) = R23 ( )R13 ( )R12 ( )
(2.1)
but also several conditions such as symmetry with respect to the permutation operator P on
V V (P x y = y x),
R() = P R()P ,
(2.2)
unitarity involving some complex function (),

R()R() = ()
(2.3)
and crossing unitarity for another complex function (),
R t1 ()R t1 ( 2) = ().
(2.4)
The parameter characterizes the R-matrix and the superscript tj denotes the transposition with
respect to the j th space in the tensor product V V . Here we will need the well-known 6-vertex
model solution
1
0
0
0
0 b(, ) c(, ) 0
R(, ) =
,
0 c(, ) b(, ) 0
0
0
0
1
sinh
sinh( )
, c(, ) =
b(, ) =
(2.5)
sinh( + )
sinh( + )
of the YangBaxter equation (2.1). It generates the Hamiltonian of the spin- 21 XXZ chain with
() = 1,
()
sinh sinh( + 2)
sinh2 ( + )
(2.6)
Each solution R() of the YangBaxter equation fixes the structure constants of a YangBaxter
algebra
R12 ( )T1 ()T2 () = T2 ()T1 ()R12 ( )
(2.7)
with generators T (), , = 1, 2; where T1 () = T () I , T2 () = I T () are the embeddings of the monodromy matrix T ().
Sklyanins construction of open spin chains is based on the representations of two algebras
T (+) and T () defined by the relations
R12 ( )T1() ()R12 ( + )T2() ()
()
= T2
()
()R12 ( + )T1
()R12 ( ),
(2.8)
354

(+)t1
R12 ( + )T1
(+)t2
= T2
(+)t2
()R12 ( 2)T2
(+)t1
()R12 ( 2)T1
()
()R12 ( + ).
(2.9)
We shall call T (+) and T () right and left reflection algebras respectively. The transfer matrix
t () = tr T (+) ()T () ()
(2.10)
as a trace in auxiliary space is the central object under consideration because it generates with
[t (), t ()] = 0 a commuting family of operators.
The explicit construction of integrable open boundary conditions for models arising from the
YangBaxter algebra starts with the 2 2 matrix

1
sinh( + )
e sinh(2)
K(, ) =
sinh cosh e sinh(2) sinh( )
= I2 + z tanh coth +
2e sinh + 2e sinh
+
sinh
sinh
(2.11)
originally found by de Vega et al. [16]. It constitutes the known c-number representations
K (+) () = 12 K( + , + ) and K () () = K(, ) of the reflection algebras with the obvious properties
tr K(, ) = 2,
K () (0) = I2 ,
tr K (+) (0) = 1.
(2.12)
The Hamiltonian (1.1) involving two neighbouring sites for the interaction is connected, up to
a factor, to the first derivative of t () by looking at the expansion t () = 1 + 2H + around
the point = 0. Considering local L-matrices building up the two representations T (+) () =
LL () LM+1 () and T () () = LM () L1 () of (2.7) then by construction
T () () = T () ()K () ()T ()1 (),
t

T (+)t () = T (+)t ()K (+)t () T (+)1 ()
(2.13)
are representations of the reflection algebras such that the normalized transfer matrix
t () = K (+) ()T ()K () ()T 1 (),
t (0) = 1
(2.14)
is independent of the factorization of T () = T (+) ()T () (). Thus we are free to choose
T (+) () = K (+) (),
T () () = T ()K () ()T 1 ().
(2.15)
In order to gain more symmetric arguments and to avoid inconvenient scalar functions after
applying the inversion formula
T 1 () =
1
y T t ( ) y
(dq T )( /2)
(2.16)
it is instructive to define the new object U ( + /2) T () ()(dq T )( /2) consisting of

U () = T ( /2)K () ( /2) y T t ( /2) y .
(2.17)
It is still a representation of the left reflection algebra with a 2 2 matrix in auxiliary space,

A() B()
U () =
.
(2.18)
C() D()
355
The quantum determinant (dq T )() is the central element (Casimir) of the YangBaxter al
gebra (2.7). With the one-dimensional projector P12
onto the antisymmetric (singlet) state in the
tensor product V V of auxiliary spaces the definition reads
(dq T )() = tr12 P12

T1 ( /2)T2 ( + /2)
= A( + /2)D( /2) B( + /2)C( /2).
(2.19)
Here, the trace tr12 is to be taken in both auxiliary spaces 1 and 2 of the tensor product V V
and the monodromy matrix T enters with the usual representation

A() B()
.
T () =
(2.20)
C() D()
3. Quantum determinants
Like the Casimir for the YangBaxter algebra there exists a similar object for the left reflection
algebra. It is defined as
(q U )() = tr12 P12

U1 ( /2)R12 (2 )U2 ( + /2).
(3.1)
To express (q U )() in terms of the generators A(), B(), C() and D() it is instructive
to use the combinations
D()
sinh(2)D() sinh A(),
sinh(2 + )C()
C()
(3.2)
borrowed from the algebraic Bethe ansatz. Then the suggestive form of the quantum determinant
reads
/2) B( + /2)C(
/2).
(q U )() = A( + /2)D(
(3.3)
In case of the c-number representation K( /2, ) for U () connected to the left reflection
algebra the relation
sinh(2 2) cosh
(3.4)
det K(, )
cosh( )
holds. Note that this relation is only valid for the shifted argument /2 because the arising
expressions are no longer of difference form.
Appropriately transforming the boundary parameters and according to the mapping [2]
(q K)( /2, ) =
cosh
sinh
,
cosh sinh =
(3.5)
2
2
the determinant det K(, ) factorizes and its quantum version decomposes to product form
sinh cosh =
(q K)( /2, )
sinh( ) cosh( ) sinh( + ) cosh( + )
(3.6)
.
sinh cosh cosh( ) sinh cosh cosh
It is obvious that in the parametrization (3.5) the model is invariant under the simultaneous transformations and i . As the quantum determinant respects co-multiplication,
applying it to the full representation (2.17) of the left reflection algebra with monodromy matrices T yields
= sinh(2 2)
(q U )() = (dq T )( /2)(q K)( /2, )(dq T )( /2).
(3.7)
356
Example. Consider the inhomogeneous periodic chain with inhomogeneities sj C at each

lattice site j = 1, . . . , L. Then the quantum determinant of a fundamental L-matrix Lj () =
R( sj ) takes the scalar value
(dq Lj )() = sinh( sj + 3/2) sinh( sj /2)
(3.8)
L
yielding (dq T )( /2) = j =1 sinh( sj + ) sinh( sj ) for a chain of L local
spins 1/2.
4. Functional Bethe ansatz
The entries of the monodromy matrix U () can be expressed in terms of the operators A(),
B(), C() and D() of the periodic monodromy matrix T (), i.e.
sinh( /2 + ) sinh(2 )
B( /2)A( /2)
cosh( /2) sinh sinh(2)
sinh( + /2 ) sinh(2 )
B( /2)A( /2)
cosh( /2) sinh sinh(2)
B() =
e sinh(2 )
A( /2)A( /2)
sinh cosh( /2)
e sinh(2 )
B( /2)B( /2).
sinh cosh( /2)
(4.1)
Obviously there is no easy pseudo vacuum |0 for an algebraic Bethe ansatz to work
(B()|0 = 0) for non-diagonal boundaries ( = 0) in K () . Instead we can apply Sklyanins
functional Bethe ansatz [11].
Henceforth let us restrict to the rational case. Then we can choose the boundary matrices
diagonal in favour of a twisted monodromy matrix T () as demonstrated in the following
Example. Applying the rational limit1 to the parametrization of the boundary matrix K(, )
yields the similarity transformation independent of the spectral parameter
+

0
1
K(, ) = MS
(4.2)
(MS)
0
with the diagonal matrix to be a solution to the reflection algebra and the 2 2 number matrices
/2
+/2

1
e
0
e
e/2
M=
(4.3)
,
S=
.
/2
e/2
0
e/2
2 cosh e
Thus in the transfer matrix t () = tr K (+) ()T ()K () ()T 1 () we are free to consider
a diagonal outer boundary matrix K (+) together with the c-number twist (M (+) S (+) )1 T ()
of T ().2
1 Rational limit means to scale the arguments by a small parameter 0 and calling ic. Here only is
rescaled whereas remains unchanged. This immediately follows from (3.5).
2 The gl(2) symmetry of the rational model allows to remove the twist of the monodromy matrix in favour of a twisted
boundary matrix K () = (M (+) S (+) )1 K () M (+) S (+) , see Ref. [3].
357
4.1. Operator-valued zeroes

The main goal of the functional Bethe ansatz is to treat the spectral problem of the transfer
matrix in a representation space of symmetric functions manipulated by some shift operators,
which descent from operator-valued zeroes of the B-operator. Starting from B() in the rational
limit reading

2 ic + ic/2
B() =
B( ic/2)A( ic/2)
2
+ ic/2
+
B( ic/2)A( ic/2)
2
e A( ic/2)A( ic/2)

+ e

B( ic/2)B( ic/2)
(4.4)
we observe the expression in the square brackets to be symmetric with respect to and
having no pole at = 0. As the operators A, B, C and D are polynomials3 of degree L with the
known asymptotics
+
) L
exp(
2
,
A
2 cosh +
+
) L
exp( +
2
,
C
+
2 cosh
) L
exp(
2
B
,
2 cosh +
+
) L
exp( +
2
D
+
2 cosh
(4.5)
we can factorize the square brackets in terms of 2 . The asymptotic pre-factors arise from the
twist S 1 M 1 of the periodic monodromy matrix T (). Thus B() is polynomial with a simple
zero at = ic/2 and operator-valued coefficients assembling

L

2 ic 1 2 sinh( + + ) 2
2
xl .
B() =
(4.6)
(1)L
2 cosh +
l=1
As [B(), B()] = 0, according to the reflection algebra, we can deduce [xj2 , xk2 ] = 0 for all
j, k = 1, . . . , L with the spectrum shown in the next
Example. Setting L = 1 in the explicit expression (4.6) for a spin 1/2 representation with inhomogeneity s1 yields for the argument = 0 the form

(s1 + ic/2)2
2
x1 =
(4.7)
(s1 ic/2)2
of the operator-valued zero x12 reading in a diagonalized form. This fixes discrete sets j
{sj ic/2, sj + ic/2} representing the spectra of the coordinates xj except for a global sign and
resembling XL 1 L .
For the forthcoming relations, to work with the simple zeroes xj instead of xj2 , we refer to the
supplemental
3 From here the L -dependence of all periodic chain operators is suppressed.
358
Remark. Let xj be the operator-valued zeroes satisfying =xj |B() = 0. Then all xj2 can be
simultaneously diagonalized such that for all j, k = 1, . . . , L

s + ic/2
.
[xj , xk ] = 0, xj = j
(4.8)
sj ic/2
4.2. Conjugated momenta
With the operators xj the next problem is to calculate the expression for the transfer matrix
in the x-representation.
For this let us introduce first the conjugated momenta to the coordinates xj .
Considering A() and D()

as polynomials and inserting the operator valued zeroes of B()
by substitution from the left yields the new operators
p
xj Ap Xj ,
=xj |A() =
p
=xj |D() =
xj D p Xj+ .
p
(4.9)
Here Ap and D p denote operator-valued expansion coefficients. The commutation relations with
the coordinates xj are summarized in
Theorem 1. Let xj be the operator-valued zeroes of B() and Xj their conjugated momenta
related by the reflection algebra. Then
Xj xk = (xk icj k )Xj .
(4.10)
Proof. Consider the commutation relation

sinh( + ) sinh( )
B()A()
A()B() =
sinh( + ) sinh( )
sinh sinh(2 )
+
B()A()
sinh(2) sinh( + )
sinh
(4.11)
B()D()
sinh(2) sinh( + )
of A() and B() and multiply it by sinh( + ) sinh( ). Then take its rational limit and
insert the coordinates xj by substitution from the left. The expression reduces to
(xj )(xj + )Xj B() = (xj + ic)(xj ic)B()Xj .
(4.12)
Replacing B() by its factorized form, cancelling the constant asymptotics and multiplying by
the inverse (xj2 2 )1 from the left the commutation relation

L

L

2

2

2
2
2
2
Xj
(4.13)
xl
= (xj ic)
xl Xj
l=1
l=1
l=j
remains. Implying all expressions to be symmetric in xj to act on we arrive at the desired relation. Analogously the elementary commutation of Xj+ with the coordinates arises from the
2
commutation of B and D.
359
The next natural step would be establishing the commutation relation between two X s.
However, it cannot be done directly because Xj exceed per definition the representation space.
4.3. Representation space
Following Sklyanins approach the square brackets in operator (4.6) can be expanded into
2L b1 2(L1) + bL with commuting operators bj thus sharing a common system of
eigenfunctions f ,
bj f = bj f ,
= 1, . . . , 2L ,
(4.14)
) BL CL there corresponds
where 2L represents spin-1/2. To every point b = (b1 , . . . , bL
only one eigenfunction f and the representation space W (e.g., for the XXX chain we have
W = (C2 )L ) is isomorphic to the space Fun BL .
Example. A possible realization of the eigenfunctions f is

(bj f ) b = bj f b ,
(4.15)
where bj act as multiplication operators. Let {t (C2 )L | (t ) = f (b )} be a basis of W
then, with the constraint f (b ) = , it is indeed orthonormal and complete.

Since bn are the symmetric polynomials of the roots xj2 we are led to consider the mapping
: C L CL ,
x b
(4.16)
given by the formula bn (x) = sn (x). The sn (x) are the elementary symmetric polynomials of
degree n = 1, . . . , L of c-number variables
s1 (x) = x12 + x22 + + xL2 ,
..
.
sL (x) = x12 x22 xL2 .
(4.17)
The diagram
g
XL
BL
{0, 1} C
of the combined mapping f reveals the isomorphism between Fun BL

= W and the space of
symmetric functions SymFun XL . The set {0, 1} is the range of f in the example above. Thus the
operator roots xj2 can be thought of multiplication operators
xj2 g(y1 , . . . , yL ) = yj2 g(y1 , . . . , yL )
(4.18)
in an extended representation space Fun XL

= W which is a non-physical one. Recall all the
results should only use the original space SymFun XL
= W , as the operators A, B, C and D
map SymFun XL SymFun XL or more sloppy W W . So for plainness we will use in the
following the terms W and W for the representation spaces instead.
360
For the action of Xj on a function s SymFun XL we need first to extend the operators from
W to W by the constant function
(x) = 1
for all x XL .
(4.19)
Obviously is symmetric and thus belongs to the representation space W . Now define the action
of Xj on by

Xj (x)
(4.20)
j (x).
Then the functions

j (x) uniquely determine the action of Xj on any vector s which is identified due to the isomorphism with some symmetric function s(x1 , . . . , xL ) = (s )(x) created
from the cyclic vector by the operator s = s(x1 , . . . , xL ). Thus

Xj s (x) = Xj s (x) = s Ej x Xj (x) = s Ej x
(4.21)
j (x).
Here we introduced the shift operators

Ej : CL CL : (x1 , . . . , xj , . . . , xL ) (x1 , . . . , xj ic, . . . , xL )
(4.22)
acting on some L-tuple of c-numbers. In the extended representation space W of not necessarily
symmetric functions the action of Xj then reads
Xj =
j Ej
(4.23)
with
j = j (x). By the operator relation (4.23) we can now calculate the commutations of the
momenta in
Theorem 2. Let Xj be the conjugated momenta related to the coordinates xj by the reflection
algebra. Then

Xj , Xk = 0 for all j, k = 1, . . . , L,
+
Xj , Xk = 0 for all j, k = 1, . . . , L but j = k.
(4.24)
Proof. Let us start with X where the first assertion is obvious for j = k. Then it is enough to
consider the cases j = 1, k = 2. Taking the rational limit of

A(), A() =

sinh
B()C() B()C()
sinh( + )
(4.25)
and inserting = x1 and = x2 by substitution from the left the RHS turns into zero and for
the LHS we get

x1m x2n Am An =
x2n x1m Am An
=x1 ,=x2 |A()A() =
m,n

n
m,n
x2n X1 An
= X1
x2n An = X1 X2 .
(4.26)
In the same way starting from A()A() one obtains X2 X1 and the assertion is proven. The
commutation of X + s and mixed commutators excluding the cases j = k can be treated analogously by considering
361

D(),
D()
= sinh(2 + ) sinh(2 + ) A(), A() ,

sinh( + ) sinh(2 + )
A(), A()
D(),
A() =
sinh( )
(4.27)
in the rational limit.
The remaining commutation involving the quantum determinant q is summarized in

Theorem 3. Let xj and Xj be the coordinates and conjugated momenta related by the reflection
algebra and q () is the quantum determinant. Then
Xj Xj = q (xj ic/2) for all j, k = 1, . . . , L.
(4.28)
Proof. Substituting the operator-valued zeroes xj into the quantum determinant (3.7) one obtains

q (xj ic/2) =
xjm (xj ic)n Am D n =
(xj ic)n xjm Am D n
m,n
m,n
n
= Xj Xj+

=
(xj ic)n Xj D n = Xj
xjn D n
(4.29)
and analogously q (xj + ic/2) = Xj+ Xj exerting the reflection algebra.
Remark. The remaining zero ic/2 of B() is an exception and renders the operators A(ic/2) =
dq (ic/2) and D(ic/2)

= 0 to be constant yielding q (ic) = 0.
4.4. Representation of

Applying Xj Xj =
j Ej j Ej to an arbitrary function g W induces the sequence

Xj Xj g (x) =
j Ej j Ej g (x) = j (x) Ej j Ej g (x)

=
j (x) j Ej g Ej x = j (x)j Ej x Ej g Ej x

=
j (x)j Ej x g(x)
!
= q (xj ic/2)g(x)
(4.30)
relating the representations

j to the quantum determinant q . In the case of a finite dimenL
sional representation of the generators {xj , Xj }L
j =1 such that the spectrum X shows no multiple
points the problem of constructing such a representation is equivalent to that of determining the
L
L
functions {
j }j =1 on X satisfying

m (x)n Em x = n (x)m En x
+

+
+
m (x)n Em x = n (x)m En x

q (xj ic/2) =
j (x)j Ej x
for all n, m,
for all n, m but n = m,
for all j
(4.31)
362
arising from Theorems 2 and 3. The above relations are not defined when the shifts Ej move the
L
point x out of XL = 1 L . This means {
j }j =1 have to vanish on the boundary

L
L
L
X
(4.32)
j x X | Ej x C \X
of the set XL . For the open XXX chain with j = {sj ic/2, sj + ic/2} this is clear from the
following
Example. The vanishing of

j (x) on the boundary Xj can be directly seen from the explicit
factorization of q () = ( + /2)+ ( /2) into

L

ic/2
+
() =
( sl + ic/2)( + sl + ic/2) ,
(1)L
l=1

L
+ ic/2
+
() = (2 ic)
(4.33)
( sl ic/2)( + sl ic/2)
(1)L
l=1
considered in the rational limit indicated by 0.

4.5. Spectral analysis
Now let us return to the original problem, the spectral analysis of the new transfer matrix
() tr K( + /2, + )U ()/2 in the rational limit
() =
( + ic/2)( + + ic/2)
1 + + ic/2
D()
A()
+
2
4 +

( + ic/2) + +
+
+
e C() + e B()
+
(4.34)
and mind the scaling factor 0. To plug in the zeroes xj by substitution from the left we have
to get rid of C() by diagonalizing K (+) . Thus only the first line remains. The diagonalization
does not change the quantum determinant q () and the eigenvalue problem () = ()
can be solved by substitution from the left reading
=xj | () =
(xj + ic/2)(xj + + ic/2) 1 xj + + ic/2 +

Xj
Xj .
2xj +
4xj +
(4.35)
With this representation at hand one observes separation of variables suggesting the product
ansatz

L

Q(xl )
=
(4.36)
l=1
for the eigenfunction SymFun XL

= W symmetric in its arguments xl . To explicitly apply
the operator-valued expression (4.35) one should clarify its behaviour by the following
Lemma. The action of the combined expression xj Xj by substitution from the left onto a
symmetric function s = s(x1 , . . . , xL ) is given by

xj Xj s(x) = xj Xj s (x) = xj Xj s (x) = xj
(4.37)
j (x)s Ej x .
363
Then applying (4.35) to only the j th argument is affected such that the problem separates
and
(xj )Q(xj ) =
(xj + ic/2)(xj + + ic/2)

(xj )Q(xj ic)
2xj +
1 xj + + ic/2 +
(xj )Q(xj + ic)
4xj +
(4.38)
holds. Here we used (4.37) with the allowed arguments xj j = {sj ic/2, sj + ic/2} on the
grid entering
j (x) = (xj ).
Remark. Along with the normalization (dq T )(), cf. (3.8), and a shift in the arguments the
original transfer matrix t () is related to () via () = (dq T )()t ( /2).
5. TQ-equations
The eigenvalue problem as formulated in Eq. (4.38) reduces to a system of homogeneous
linear equations due to the fact that (xj ) = 0 at the points xj = sj ic/2:
+
+
+ + (xj + ic/2)(xj + + ic/2) +
xj Q xj ,
xj Q x j =
2xj+ +
+

1 xj + ic/2 + +
xj Q x j .
xj Q xj =
4xj +
(5.1)
For pairwise different inhomogeneities, sj = sk for j = k, these linear equations allow for a
non-trivial solution provided that the following functional equation for the eigenvalues are
satisfied4
(sj + ic/2)(sj ic/2)
=
sj + ic sj +
sj + +
q (sj ),
2 (2sj ic) + (2sj + ic) +
j = 1, . . . , L.
(5.2)
Using the known asymptotic form of the even polynomial () = () we are led to the ansatz
() =
(1)L sinh + sinh + cosh( + ) 2L+2
+
cosh + cosh
+ a2L 2L + a2L2 2L2 + + a0 .
(5.3)
The (L + 1) unknown coefficients aj are determined by Eqs. (5.2) and the constraint (ic/2) =
dq (ic/2) with the quantum determinant dq () of the periodic chain. Thus the solution of the
() L
spectral problem amounts to finding the common roots {a2j
}j =0 , = 1, . . . , 2L , of these polynomial equations. This task is of the same complexity as finding the eigenvalues of the spin chain
Hamiltonian directly and therefore this approach is limited to small system sizes where we have
checked numerically that this approach does indeed yield the complete spectrum.
4 If n of the inhomogeneities coincide the (n 1) derivatives of this equation at this value of s have to be taken into
j
account in addition.
364
To compute the eigenvalue of the transfer matrix or the spin chain Hamiltonian in the thermodynamic limit L the functional equations introduced above need to be analyzed beyond
the set XL using explicitly the analytic properties of the functions therein.
Remark. Note that the functional equations (5.2) hold at the discrete points sj only: treating
the sj as a continuous variable and applying standard Fourier techniques one can compute
ln () and thereby the corresponding eigenvalue of the spin chain Hamiltonian (1.1), i.e.
H = ic( ln )(ic/2) = ict (0) in the homogeneous limit. For | | > c/2 one obtains

ln
(ic/2) = + /2c + /2c + 1/2 + c/ +
ic

+ /2c /2c + 1/2 + c/
+ 2 ln 2 1 + (2 4 ln 2)L
(5.4)
which is for imaginary

the known energy eigenvalue of the XXX spin chain with diagonal
boundary fields [17] ( is the digamma function). However, the non-diagonal contributions and
corrections of the order 1/L are not included. This is a consequence of neglecting the corrections
to Eqs. (5.2) away from the points sj .
Instead we go back one step and consider Eqs. (4.38) for general arguments xj . Formally,
this is a second order difference equation reminiscent of Baxters TQ-equation [18]. The analysis
above leading to Eqs. (5.2) has been based on the singular points of the TQ-equation at the
boundaries XL , i.e., points where one of the coefficients vanishes. Away from these points
there exist two independent solutions to (4.38) and one needs some information on the properties
of the unknown functions Q() in this formulation of the spectral problem which has to be solved
for polynomial eigenvalues () of the transfer matrix.
In cases where a pseudo vacuum exists and the algebraic Bethe ansatz is applicable to
the
Msolution of the problem the Q-functions are known to be symmetric polynomials Q() =
=1 ( v )( + v ) with roots v satisfying Bethe ansatz equations. Note that only in these
cases the constant function = 1 introduced in the construction of the representation of the
YangBaxter algebra on the space W can be identified with the pseudo vacuum |0. In general,
the TQ-equation can be rewritten as a recursion relation
Q( + ic) = a()Q() + b()Q( ic)
for the function Q() or equivalently, with the auxiliary function P ( + ic) Q(),

Q( + ic)
a() b()
Q()
=
.
P ( + ic)
1
0
P ()
(5.5)
(5.6)
The coefficients a() and b() are obtained from the TQ-equation (4.38) and show constant
asymptotics for large values of their arguments
4 +
sinh + sinh + cosh( + )
()
,
2
+ () + + ic/2
cosh + cosh
() 2( + ic/2)( + + ic/2)
b() = +
1.
()
+ + ic/2
This allows to solve the recursion relations in the asymptotic regime || 1 yielding
a() =
Q( + inc) =
n1 n1
n1 n2
2
Q( + ic) 1 2 1
Q().
1 2
1 2
(5.7)
(5.8)
365
Here n is an integer and 1,2 = e are the eigenvalues of the asymptotical matrix of coefficients in (5.6), cosh = (sinh + sinh + cosh( + ))/(cosh + cosh ). Ordering the
eigenvalues as |1 | > |2 | we obtain for fixed and large n the leading term Q( + inc) n1
suggesting the following ansatz for the asymptotic form

Q() exp
(5.9)
polynomial in .
ic
Here the polynomial form of the subleading part assures that the eigenvalue () of the transfer
matrix remains polynomial. Note, that since () is an even function of its argument there exists
always a second solution Q() to the TQ-equation which decays exponentially for .
Only in two cases, namely = 0 and i or, equivalently,

cosh + = cosh + ,
(5.10)
the exponential factor disappears and the TQ-equation can be solved by an even polynomial:
(+)
in the first case Eq. (5.9) implies that Q() = M
=1 ( v )( + v ). For = i , the exponential factors can be removed by the transformation Q() = exp(i/ic)Q () resulting in a
TQ-equation for Q :
(xj )Q (xj ) =
(xj + ic/2)(xj + + ic/2)

(xj )Q (xj ic)
2xj +
1 xj + + ic/2 +
(xj )Q (xj + ic).
4xj +
(5.11)
Again, it follows from the asymptotic analysis that this equation allows for a polynomial solution
()
Q () = M
=1 ( v )( + v ) whose existence has been verified by numerical analysis for
small systems.
In both cases the spectrum is determined by the roots of these polynomials. To guarantee
analyticity of the transfer matrix eigenvalues () the vj , j = 1, . . . , M () , have to satisfy the
Bethe ansatz equations
L

vj + ic/2 vj + + ic/2 vj s + ic/2 vj + s + ic/2
vj + ic/2 vj + + ic/2
vj s ic/2 vj + s ic/2
=1
M ()

vj vk + ic vj + vk + ic
=
(5.12)
.
vj vk ic vj + vk ic
k=1
k=j
Note that Eq. (5.10) is equivalent to the constraint that the boundary matrices K () can be simultaneously diagonalized or brought to triangular form. In this case Eqs. (5.12) can be obtained by
means of the algebraic Bethe ansatz [3] or in the rational limit from the TQ-equation approach for
the open XXZ chain [10]. In this trigonometric case the complete set of eigenvalues is obtained
from two sets of Bethe equations which both reduce to (5.12). This is due to the invariance of
the model under the change of parameters and i which maps = 0 i ,
see Eq. (3.5). As another difference to the situation in the XXZ model the number of Bethe roots
is not restricted by the constraint on the boundary fields and we have to consider solutions of the
TQ-equations beyond the equator, M () > L/2.
366
6. Summary
In this paper we have extended Sklyanins functional Bethe ansatz method to the open XXX
chain with non-diagonal boundary fields. Within this framework we have derived a single TQequation (4.38) which determines the spectrum of this model for any values of the boundary
parameters. This supports the approach of Yang et al. [10] who obtain a TQ-equation for the
XXZ model assuming the existence of a certain limit in the auxiliary space.
The TQ-equation allows for a solution in terms of polynomials for the function Q provided
that a constraint between the left and right boundary field is satisfied. In this case the solution
is parametrized by the roots of one set of Bethe ansatz equations. In our numerical study of
small systems we obtain the complete spectrum from these equations, when solutions beyond
the equator, i.e., polynomial Q with degree > L/2, are taken into account. This is in contrast to
the trigonometric case where two types of Bethe equations are required and the degrees of the
corresponding Q-functions are fixed by the constraint. We suppose that this feature of the XXX
case is a consequence of the rational limit.
If the constraint between the boundary fields is missing only the asymptotic (exponential)
behaviour of the Q-functions is obtained from the TQ-equation, the subleading terms have to
be chosen such that the eigenvalues of the transfer matrix remain polynomial. In Sklyanins
approach the Q-functions contain all the information on the eigenstates of the model. Therefore, their determination for generic boundary parameters is necessary to tackle the problem
of computing norms and scalar products within this approach. To make progress in this direction it should be useful to investigate how the recent construction of Galleas [5] connects to
the TQ-equation (4.38). His solution of the spectral problem for XXZ chains with non-diagonal
boundaries is given in terms of the zeroes of the transfer matrix eigenvalues and two complementary sets of numbers which parametrize matrix elements of certain elements of the YangBaxter
algebra. They satisfy equations reminiscent of the nested Bethe ansatz used to solve models of
higher-rank symmetries. Further studies are necessary to see whether this parametrization of
the spectrum in terms of O(L) complex numbers can be used to obtain a closed expression for
generic (non-polynomial) Q-functions. This would be of great importance for the applicability
of the TQ-equation to solve the spectral problem of integrable quantum chains.
Acknowledgements
The authors would like to thank A. Klmper, J. Teschner and S. Niekamp for helpful discussions. This work has been supported by the Deutsche Forschungsgemeinschaft under grant
numbers Se 1742/1-1 and FR 737/6.
References
[1] J. Cao, H.-Q. Lin, K.-J. Shi, Y. Wang, Exact solution of XXZ spin chain with unparallel boundary fields, Nucl. Phys.
B 663 (2003) 487519, cond-mat/0212163.
[2] R.I. Nepomechie, Bethe ansatz solution of the open XXZ chain with nondiagonal boundary terms, J. Phys. A 37
(2004) 433440, hep-th/0304092.
[3] C.S. Melo, G.A.P. Ribeiro, M.J. Martins, Bethe ansatz for the XXX-S chain with non-diagonal open boundaries,
Nucl. Phys. B 711 (2005) 565603, nlin/0411038.
[4] R. Murgan, R.I. Nepomechie, C. Shi, Exact solution of the open XXZ chain with general integrable boundary terms
at roots of unity, J. Stat. Mech. (2006) P08006, hep-th/0605223.
[5] W. Galleas, Functional relations from the YangBaxter algebra: Eigenvalues of the XXZ model with non-diagonal
twisted and open boundary conditions, Nucl. Phys. B 790 (2008) 524542, arXiv: 0708.0009.
367
[6] J. de Gier, F.H.L. Essler, Bethe ansatz solution of the asymmetric exclusion process with open boundaries, Phys.
Rev. Lett. 95 (2005) 240601, cond-mat/0508707.
[7] J. de Gier, F.H.L. Essler, Exact spectral gaps of the asymmetric exclusion process with open boundaries, J. Stat.
Mech. (12) (2006) P12011, cond-mat/0609645.
[8] J. Cao, Y. Wang, Spin current in quantum XXZ spin chain, Nucl. Phys. B 792 (2008) 284299.
[9] E.K. Sklyanin, Boundary conditions for integrable quantum systems, J. Phys. A 21 (1988) 23752389.
[10] W.-L. Yang, R.I. Nepomechie, Y.-Z. Zhang, Q-operator and TQ relation from the fusion hierarchy, Phys. Lett.
B 633 (2006) 664670, hep-th/0511134;
(a) P. Baseilhac, K. Koizumi, Exact spectrum of the XXZ open spin chain from the q-Onsager algebra representation
theory, J. Stat. Mech. (2007) P09006, hep-th/0703106.
[11] E.K. Sklyanin, Quantum inverse scattering method. Selected topics, in: M.-L. Ge (Ed.), Quantum Group and Quantum Integrable Systems, in: Nankai Lectures in Mathematical Physics, World Scientific, Singapore, 1992, pp. 6397,
hep-th/9211111.
[12] A. Chervov, G. Falqui, Manin matrices and Talalaevs formula, math.QA/0711.2236.
[13] E.K. Sklyanin, The quantum Toda chain, in: N. Sanchez (Ed.), Non-linear Equations in Classical and Quantum
Field Theory, in: Lecture Notes in Physics, vol. 226, Springer, Berlin, 1985, p. 196.
[14] A.G. Bytsko, J. Teschner, Quantization of models with non-compact quantum group symmetry: Modular XXZ
magnet and lattice sinh-Gordon model, J. Phys. A 39 (2006) 1292712981, hep-th/0602093.
[15] J. Teschner, On the spectrum of the sinh-Gordon model in finite volume, hep-th/0702214.
[16] H.J. de Vega, A. Gonzlez-Ruiz, Boundary K-matrices for the six vertex and the n(2n 1)An1 vertex models, J.
Phys. A 26 (1993) L519L524, hep-th/9211114.
[17] H. Frahm, A.A. Zvyagin, The open spin chain with impurity: An exact solution, J. Phys.: Condens. Matter 9 (1997)
99399946.
[18] R.J. Baxter, Exactly Solved Models in Statistical Mechanics, Academic Press, London, 1982.

Occupation numbers from functional integral

C. Wetterich
Institut fr Theoretische Physik, Universitt Heidelberg, Philosophenweg 16, D-69120 Heidelberg, Germany
Received 2 January 2008; received in revised form 2 April 2008; accepted 14 April 2008
Abstract
Occupation numbers for non-relativistic interacting particles are discussed within a functional integral
formulation. For bosons at zero temperature the Bogoliubov theory breaks down for strong couplings as
well as for low-dimensional models. We find that the leading behavior of the occupation numbers for small
momentum is governed by a quadratic time derivative in the inverse propagator that is not contained in
the Bogoliubov theory. We propose to use a functional renormalization group equation for the occupation
numbers in order to implement systematic non-perturbative extensions beyond the Bogoliubov theory. We
also discuss interacting fermions, in particular the issue of pseudogaps.
1. Introduction
Occupation numbers are the central tool for the statistical description of non-interacting particles or quasi-particles. Extending this concept to interacting particles allows for the description of
many interesting phenomena, as condensates, the formation of gaps and pseudogaps or nontrivial scaling for the momentum dependence of the occupation numbers. Most of these phenomena
occur at low temperatures T and we will concentrate on this case.
For non-relativistic bosons and small couplings the use of perturbation theory, within a Hamiltonian framework with operator valued fields, has led to the successful Bogoliubov theory [1].
For d > 2 and weak interactions the behavior for T 0 is well understood since a long time
[2]. This theory breaks down, however, in low dimensions and for strong couplings. In one dimension (d = 1) and for T = 0, the occupation number n(
q ) for a mode with momentum q
diverges proportional to the inverse momentum, n(
q ) q 1 , q = |
q |. The particle density is the
momentum integral over the occupation numbers and diverges logarithmically in this case. FurE-mail address: c.wetterich@thphys.uni-heidelberg.de.
doi:10.1016/j.nuclphysb.2008.04.010
C. Wetterich / Nuclear Physics B 802 [FS] (2008) 368404
369
thermore, the Bogoliubov theory is built on a non-zero condensate density or order parameter 0 .
Such an order parameter leads to spontaneous symmetry breaking of the global U (1)-symmetry
associated to the conserved particle number. It is known that for d = 1 and T = 0 no spontaneous
symmetry breaking of a continuous symmetry is possiblethe order parameter 0 has to vanish
in the infinite volume limit. Also for d = 2 the situation is problematic. Now one finds a nonzero 0 for T = 0 and the Bogoliubov approximation to the particle density shows no infrared
divergence anymore. However, for any non-zero temperature the order parameter must again
vanish due to the MerminWagner theorem [3], 0 (T > 0) = 0. This reveals again a problem for
the perturbative treatment where n(
q ) 0 /q for small q. For q > 0 the occupation numbers
should be continuous functions of T and this indicates a contradiction between the Bogoliubov
approximation and the MerminWagner theorem.
Going beyond leading order perturbation theory (Bogoliubov approximation) becomes quite
intricate in the operator formalism. A good alternative is the functional integral formulation
which has already given a rather detailed picture of BoseEinstein condensation for interacting
bosons, largely by use of the renormalization group [4]. In this paper we address the computation
of occupation numbers within a functional integral approach.
Already at the very basic level of the definition of n(
q ) one needs to be careful to subtract
counterterms that diverge in the ultraviolet limit. We work with a momentum cutoff q2 < 2
and show that a too naive definition of n(
q ) would lead to a divergent particle density for
. One needs to introduce counterterms that arise from the transition from the Hamiltonian formalism to the functional integral. With this correct starting point the functional integral
formalism is well suited to solve the infrared problems of the Bogoliubov theory, as well as
potential ultraviolet problems of extensions of it.
For T = 0 we show how the infrared problems of the low-dimensional Bogoliubov theory
can be resolved by an investigation of the frequency and momentum dependence of the full
propagator. In particular, it is crucial that the frequency dependence of the inverse propagator
is quadratic in the range of small q [5,6], and not linear as for the Bogoliubov theory. We use
the propagator computed from the functional renormalization group in [6], which removes the
infrared problems. For q 0 the particle number behaves now V 1/2 q 1+ instead of
0 q 1 in the Bogoliubov approximation. Here V is the coefficient of the term 2 in the inverse
propagator, with the frequency. The anomalous dimension vanishes for d = 2, 3, while it is
in the range 0 < 1/2 for d = 1. This renders the particle number finite for all dimensions.
Furthermore, the leading behavior of n(
q ) for q 0 becomes independent of 0 and is therefore
not affected by the vanishing of 0 at T = 0 (for d = 1) or T > 0 (for d = 2). In three dimensions
and for a small coupling the behavior n V 1/2 q 1 is relevant only for very small q, whereas
1/2
the validity of the Bogoliubov approximation n 0 /q extends to a large range of q.
Extensions beyond the Bogoliubov theory have also to cope with ultraviolet problems. They
occur if the frequency dependence of the propagator is not taken into account properly. We propose here a renormalization group treatment based on the average action [79]. An infrared
cutoff k for the fluctuations is introduced in order to extrapolate between the microscopic action
for k = and the full quantum effective action for k = 0. Correspondingly, we study flow equations for k-dependent occupation numbers nk (
q ). They yield the physical occupation numbers
n(
q ) as the IR-cutoff is removed for k = 0. The renormalization group flow permits an improvement in the computation of n(
q ) since for every momentum range only the effective renormalized
couplings in this momentum range matter.
In particular, we use the renormalization group flow in order to show two properties. The first
is the non-renormalization of the condensate density for non-relativistic bosons. It is given by
370
the unrenormalized order parameter, nc = 02 . For an interaction theory, the condensate density
differs from the total density n by a condensate depletion n nc even for T = 0. We explicitely
compute the occupation numbers n(
q ) for q = 0 that contribute to this depletion. Second, this
method is free of ultraviolet problems. Our method can be extended to coupled systems of bosons
and fermions (cf. Section 12 and Appendix B) which are relevant for the BEC-BCS crossover for
ultracold fermions. There the issue of the condensate density is more complicated [10]. A renormalization group flow for the total density in such systems has already been investigated in [11].
This paper is organized as follows. In Section 2 we discuss the occupation numbers in a functional integral context. They are related to a q0 -integral over the propagator, with q0 the Euclidean
frequency (Matsubara frequency for T > 0). This section also introduces the (ultraviolet divergent) counterterms. Section 3 specializes to zero temperature and Section 4 deals with the case
where the frequency dependence of the inverse propagator is linear. This yields the Bogoliubov
approximation and first extensions of it. In Section 5 we extend the general form of the inverse
propagator by taking terms quadratic in the frequency (or also higher orders) into account. We
show that this solves the infrared problems of the Bogoliubov theory in low dimensions. We
present differential equations for the momentum dependence of the occupation numbers and the
particle density. In Section 6 we turn to a heuristic renormalization group improvement. This
helps for a proper understanding of the occupation numbers for large momentum.
The following sections introduce an exact renormalization group equation for the occupation
numbers. It motivates the simpler heuristic renormalization group improvements. We start in Section 7 by introducing a generating functional for the occupation numbers. This can be made scale
dependent, and an exact flow equation is derived in Section 8. We perform a non-perturbative
approximation by quadratic truncation in Section 9. The flow shows a mutual influence of
occupation numbers for different momentait cannot be separately described for individual
momentum modes. The corresponding density transfer function is discussed in Section 10. In
Section 11 we apply the functional renormalization group equations to non-relativistic bosons
for T = 0. Section 12 discusses gaps and pseudogaps in the light of renormalization group improvement and we give a short summary of our main results in Section 13.
2. Occupation numbers and renormalization
We start with the partition function for a non-relativistic bosonic particle

Z = D exp S[] ,

1
S[] =
+ Sint .
+
2M
(1)
The complex field may be expressed by its Fourier modes

i qx
(x) = (, x) = e (, q) = eiqx (q),
q
with

q = (q0 , q),
q
(2)
= (2)d

d d q,
= (2)(d+1)
q

dq0
d d q.
(3)
371
For non-zero temperature T the Euclidean time parameterizes a circle with circumference

= T 1 and the Matsubara frequencies q0 = 2nT , n Z, are discrete, with q0 = T n .
The second thermodynamic parameter is the effective chemical potential . We regularize the
theory by a momentum cutoff q2 < 2 and take when appropriate.
Furthermore, we assume the invariance of the classical action S under a global Abelian symmetry of phase rotations ei , corresponding to a conserved total particle number

x ) = d n(
q ),
N = n(
(4)
x
q
q ) is
with d the volume of d-dimensional space (d+1 = d ). In our normalization n(
dimensionless. The momentum distribution n(
q ) denotes the fraction of particles with momentum q . This is the quantity we want to investigate in this paper. For definiteness, we may consider
a torus with finite d and discrete momenta, taking d at the end. For discrete momenta
we write

N=
(5)
n(
q)
q
and identify n(

q ) with the occupation number in momentum space.
Following the Noether construction we can express n(
x ) as

1
d d q
1
d (, x)(, x)
n(
x) =
2
(2)3
and, correspondingly

1
1
(q0 , q)(q0 , q) .
n(
q) =
d+1
2
(6)
(7)
q0
Note that n(

q ) is not the Fourier transform of n(
x ).
The constant part 1/2 in n(
q ) arises from the connection between the functional integral and
the operator formalism [10]. For a single degree of freedom it is the symmetric combination of
annihilation and creation operators (a a + aa )/2 that translates into . Therefore the number
operator n = a a = (a a + aa )/2 1/2 acquires an additional contribution and this generalizes
to all momentum modes.
In the following we will work with a basis of real fields 1 , 2 defined by (x) = 1 (1 (x) +
2
i2 (x)), such that a (q) = a (q) and

1
1
n(
q) + =
(8)
1 (q)1 (q) + 2 (q)2 (q) .
2 2d+1
q0
The connected part of the two point function describes the propagator G

a (q)b (q ) = Gab (q, q ) + a (q) b (q ) .
(9)
For a translation invariant setting G is diagonal in momentum space

ab (q)(q q ),
Gab (q, q ) = G
(10)
with (q q ) = (2)d+1 (q0 q0 )d d (

q q ). Also, translation invariance implies for a possible
order parameter a (q) = 20 (q)a1 with real 0 . Here we have chosen the expectation value
372
in the one-direction without loss of generality. We define

g(q)

1
22 (q)
G11 (q) + G
2
and express n(q) in terms of the full propagator G(q)

and the order parameter 0

1
n(
q ) = g(q
0 , q) + 02 (
q ) = n p (
q ) + 02 (
q ).
2
(11)
(12)
q0
In the absence of spontaneous symmetry breaking ( 0 = 0) there is no difference between 1

and 2 . Since the expectation values of the type and vanish in this case, it is often
convenient to work in the complex basis of a one component complex field. The propagator is
then associated with
(
(13)
q )(q) = G(q)(q
q )
and G(q)
will typically be a complex function. For free bosons one has

1
q2
,
G0 (q) = iq0 +
2M

1
0 (q) = T
2inT + q 2 /2M
G
n
q0
=T

2 1
q2 /2M 4 2 n2 T 2 + q2 /2M
.
(14)
Performing explicitely the Matsubara sum yields

2

1

0 (q) 1 = exp q 1
n(
q) = G
2
2MT
T
(15)
q0
and we recover the usual occupation number for free bosons. Notice that the subtraction of the
term 1/2 in Eq. (12) is crucial for the vanishing of n(
q ) for q 2 .
For fermions the situation is similar. There is no fermionic expectation value and the additive
constant changes sign according to the anticommuting properties 12 (aF aF aF aF ) = nF 12 ,
i.e.

F (q0 , q) + 1 .
q) = G
nF (
(16)
2
q0
The Matsubara sums for T = 0 involve now half-integer n and we recover for a free theory the
Fermi distribution
2

1
q
nF (
(17)
q ) = exp
.
+1
2MT
T
In presence of interactions, however, the exact propagator G(q)

can be a rather complicated
object. We distinguish between the vacuum and occupied states where n(
q ) differs from zero.
Often, but not always, an occupied state will go along with a condensate 0 = 0. For nonrelativistic interacting particles the propagator becomes complicated only for the occupied states.
373
The complexity of the occupied states is due to the presence of a new scale related to the density.
In contrast, the vacuum of non-relativistic theories is characterized by a simple propagator [6,13],
even in the presence of interactions. For the vacuum, we will use a general non-renormalization
property of the wave function renormalization in order to clarify some relations between the
microscopic formulation of the theory and macroscopic properties.
Let us consider T = 0 where the q0 integration is continuous. (More precisely, we define

0
the result of a q0 -integral as the principal value and q0 f (q0 ) = lim0 0 0 dq
2 f (q0 ).) The
vacuum is defined by the limit n(
q ) 0. Since no condensate exists in vacuum ( 0 = 0) we
will work in the complex basis. For a stable particle and q 2 > 0 the full vacuum propagator G(q)
must have a pole for purely imaginary q0 with positive definite imaginary part. (For free particles
the location of the pole is q0 = i q 2 /(2M) for = 0.) Often we can write the q0 -integration as
1
=
2
q0
1
dq0 =
4

+
upper
(18)
lower
where upper denotes the integral closed on the upper half plane in complex q0 -space. As long as
the pole remains in the upper half plane the result of the q0 integration is given by the residuum
of the pole

0 , q) = 1 Z1 (
G(q
q ).
(19)
2
q0
q ) by the derivative of the inverse propaHere we define the wave function renormalization Z (
gator at the location of its zero
1
q) = G
(, q)|0 (q ) ,
Z (

1 0 (
q ), q = 0,
= iq0 .
G
(20)
For a pole in the lower half plane the r.h.s. of Eq. (19) picks up a minus sign.
Eq. (19) constitutes an important relation between the occupation number and the behavior of
the full propagator at its pole

1 1
(21)
q) 1 .
Z (
2
From Eq. (21) we can derive immediately a crucial non-renormalization property for the vacuum
of a theory for non-relativistic particles. Since for the vacuum n(
q ) must vanish one concludes
q ) is not renormalized. It keeps its classical value even in presence of interactions
that Z (
n(
q) =
lim
T 0,n0
Z (
q ) = 1.
(22)
This non-renormalization property resolves several puzzles that would arise if the interaction
effects would lead to a non-trivial renormalization Z = 1. The first concerns the normalization of the particle number. Indeed, the global U (1)-symmetry implies a conserved total particle
number N , but it does not fix the normalization of N . Obviously, ZN is also conserved for
arbitrary constant Z. The question of the proper normalization of N can be addressed from a
microscopic and a macroscopic point of view. In the operator language the normalization of the
particle number follows directly form the normalization of the creation and annihilation operators via the commutation relation [a, a ] = 1. Constructing the functional integral results in a
374
normalization of the particle number through the normalization of the term in the action.
(For an arbitrary normalization of the fields the term linear in the frequency will take the form
q )q0 (q)(q) instead of the normalization Z (
q ) = 1 employed in Eq. (1). The particle
iZ (
q ).)
number reflects then this normalization and the r.h.s. of Eq. (7) is multiplied by a factor Z (
This microscopic point of view relates the particle number to the full propagator G(q)
for the
bare or microscopic fields.
On the other hand, from a macroscopic point of view it should be possible to express all
physical quantities relating to macroscopic scales in terms of the n-point functions for renormalized fields, without explicit reference to the microscopic formulation. This is a basic concept in
quantum field theory and leads to the concept of renormalizable theories when the characteristic
macroscopic length scale is many orders of magnitude larger than the microscopic length 1 . In
relativistic quantum field theories the renormalized fields R = Z 1/2 often involve a wave function renormalization Z that diverges for . Our definition of the particle number seems at
first sight to be at odd with this general property of quantum field theory. It is expressed in terms
of the bare propagator G(q)

and not in terms of the two point function for the renormalized field
which is Z G(q).
Therefore the particle number seems to keep some memory of the microscopic
normalization.
In view of Eq. (22) the resolution of the problem is simple: in vacuum the renormalized fields
q )1/2 (q) are identical to the microscopic fields. Only for nonzero T and n one will
R (q) = Z (
q ) = 1. However, this quantity can now be expressed in terms of macroscopic quantities
find Z (
which describe the difference between a thermodynamic equilibrium state and the vacuum. As it
should be, no microscopic physics is involved.
A second puzzle is related to a possible dependence of the microscopic wave function renorq ) on the scale (in some given arbitrary normalization of the field ). Indeed,
malization Z (
the definition of the functional integral does not only involve the action S but also a regularization
scale . Changing results in a change of S such that for a given model the action S depends
on the choice of . If for a given normalization of the fields the wave function Z (
q ) would run
q ) should be used for the definition of n. Fortunately, the
with , it would not be clear which Z (
q ) = 1 does not depend on .
non-renormalization property in vacuum (22) implies that Z, (
We close this section by a formal side remark concerning the q0 -integral of the vacuum propagator. For large real q0 the renormalization effects vanish such that
1 (q0 , q) = iq0 .
lim G
|q0 |
Using the identity (for > 0)

1
= (iq0 + )1
2
(23)
(24)
q0
we can write the q0 -integral in a manifestly convergent form

1
0 , q) (iq0 + )1 .
G(q0 , q) =
G(q
2
q0
(25)
q0
3. Occupation numbers at zero temperature

For T = 0 and n(
q ) > 0 we may employ Eq. (21) (or generalizations thereof) in order to relate
q ). One recovers the well known
the occupation number to the wave function renormalization Z (
375
results of the Landau Fermi liquid theory. (Partly this section serves for setting notations and as
a preparation for more complicated situations.) Before exploiting this relation in Sections 9 and
10 for bosons, we first discuss the analogous case of fermions. In the absence of spontaneous
F . For a pole in the upper half plane
symmetry breaking one has a single complex propagator G
the q0 -integration yields
q) =
nF (

1
q) ,
1 Z1 (
2
Z (
q) =
1
,
1 2nF (
q)
(26)
whereas for a pole in the lower half plane one finds

q) =
nF (

1
q) ,
1 + Z1 (
2
Z (
q) =
1
.
2nF (
q) 1
(27)
For a free theory and > 0 the pole remains in the upper half plane for q2 > kF2 , kF2 = 2M.
q ) = 1. In contrast, for q 2 < kF2 the pole is in the lower
For these momenta n(
q ) = 0 implies Z (
half plane. Now n(
q ) = 1 results again in Z (
q ) = 1. Thus a sharp Fermi surface is equivalent to
q ) = 1 for all q . In contrast, the presence of interactions can smoothen the Fermi surface
Z (
and lead to Z (
q ) = 1, even for T = 0.
The jump in the particle number nF (
q ) is an interesting example how quantum mechanical
discreteness arises from a perfectly continuous functional integral. Let us for simplicity consider
a finite box such that the momenta are discrete. (This discreteness in the number of integration
variables is not related to quantum mechanicsit is the same for classical physics.) The integration measure in the functional integral (1) depends continuously on . One may therefore expect
that also the expectation values like the average value nF (
q ) depend continuously on . We have
just demonstrated that this is not the case for T = 0. If for a free theory kF2 crosses the value q2
for one of the discrete momentum levels the pole in the propagator moves from the upper half
plane to the lower half plane and nF (
q ) jumps by one unit. For stable particles this generalizes
to the interacting theory. The number of particles with a given momentum q is discrete.
For bosons the situation at T = 0, n = 0 is more subtle due to the presence of a BoseEinstein
condensate, 0 = 0. We now employ the real basis of fields 1 and 2 . In principle, one has
11 for the radial mode and of G
22 for the
to consider separately the poles of the propagator G
Goldstone mode. Let us denote by G the 2 2 matrix with elements Gab and investigate the
1 . By Euclidean time reversal symmetry it has the general form
inverse matrix P = G

a,
q0 Z
P=
(28)
q0 Z ,
b
with a, b, Z depending only on q 2 and q02 . This implies that g is an even function of q0

b,
q0 Z
1 a+b
2 2 1
G = ab + q0 Z
.
,
g =
2 ab + q02 Z2
q0 Z ,
a
22 is the same, namely for
11 and G
Therefore the location of the poles for G

ab
ab
,
0 =
.
q0 = i
2
Z
Z2
(29)
(30)
Here we use that a, b, Z2 are real and positive and we assume Z > 0. We recall that a, b, Z
may depend on q02 such Eq. (30) is an implicit equation for q0 (
q ).
376
We notice that in the real basis g has poles both in the upper and
lower half plane. There
If the solution of
residua have an opposite sign such that both contribute equally to q0 g(q).
Eq. (30) is unique we define
q) =
Z (
1 g 1 (, q)
1 g 1 (, q)
=
.

0 (
0 (
2
q )>0
q )<0
(31)
This normalization ensures consistency with the complex basis in the absence of spontaneous
symmetry breaking. Relations of the type (19) and (21) continue to hold for 0 > 0. The
residuum of the pole at 0 > 0 is given by Z1 , whereas the one for the pole at 0 < 0 has
the opposite sign. In consequence, both poles contribute an equal amount to the q0 -integral. This
yields for bosons at T = 0 and q2 > 0
1
.
(32)
1 + 2n(
q)
Thus for single poles we recover Eq. (21), while generalizations occur if Eq. (30) has more than
one solution.
q) =
Z (
4. Boson occupation numbers for linear dynamic term

In this section, we compute the occupation numbers for bosons at T = 0 in the approximation that the q0 -dependence of the inverse propagator is linear, as for the classical action. This
amounts to an ansatz where the coefficients a, b, Z in Eq. (28) depend on q , but not on q0 . For
Z = 1, a = q 2 /(2M) + 2 02 , b = q 2 /(2M) this amounts to the Bogoliubov approximation.
We will encounter infrared problems in low dimensions (d = 1). We show in the next section how
those are cured by a more general form of the propagator. In fact, for d = 1, 2 the q0 -dependence
of the inverse propagator becomes quadratic in the regime with non-zero density, where the fluctuations of the Goldstone bosons dominate [6]. We also encounter ultraviolet problems if Z
differs from one. They are cured by a renormalization group improved treatment that we discuss in Sections 610. These issues show that a functional integral computation of occupation
numbers is not trivial and needs a thorough understanding of the many body dynamics.
In the limit of q0 -independent a, b and Z one finds
2Z2 |0 | 2Z ab
=
.
Z =
(33)
a+b
a+b
In the absence of spontaneous symmetry breaking one has a = b and therefore Z = Z (we
assume Z > 0). In this case possible deviations of Z from one can only arise from Z = 1.
In the presence of spontaneous symmetry breaking, however, Z deviates from one even for
q0 -independent a, b and for Z = 1. We may parameterize

2
q2
q
2
+ 2 0 ,
b = A
a=A
(34)
2M
2M
where A and may depend on q2 . This reflects the vanishing of a b for 0 0 and the
Goldstone-character of 2 , i.e. limq2 0 b = 0. We will employ later that P is related to the
second functional derivative of the effective action . The coefficients (34) follow from a simple
form of , i.e.

2
A
+ 02
Z A
.
=
(35)
2M
2
x
They can also arise from a more general form of .

With the ansatz (28), (34) we find for q 2 > 0

1
1 1
(a b)2
1
q ) = Z 1 =
1+
np (
1
2
2 |Z |
4ab
1/2

2 04
1 1
=
1 .
1+ 2 2
q
q
2 Z
2
2M ( 2M + 2 0 )
377
(36)
For Z = 1 and constant this is the occupation number for the Bogoliubov theory. Writing
the density n as a condensate density 02 and a particle density n p

2
n = 0 + n p , n p = np (
q)
(37)
q
the Bogoliubov theory yields a finite result for n p only for d > 1. Indeed, for small q the occupation number diverges q 1 in the Bogoliubov theory

M 0
np (
(38)
q) =
.
2q
For d = 2, 3 the particle density (37) is dominated by modes with momenta q 2 4M 02 .
q ) for
We will see that quantum fluctuations are responsible for important modifications of np (
small q = |
q |. The dominant effect concerns the q0 dependence of the propagator discussed in
1 on q0 and show that
the next section. In this section we first retain the linear dependence of G
the dependence of and Z on q does not resolve the infrared problems. Evaluating Eq. (36)
for q2 0 the occupation number diverges according to

M (
q ) 0
q) =
.
np (
(39)
2Z (
q ) |
q|
The detailed behavior depends on the q -dependence of and Z . This can be extracted from
a recent functional renormalization group study of quantum phase transitions [6]. For the far
infrared behavior q 2 0 the Goldstone regime applies. The relevant physics strongly depends
q|
on the space dimension d. For d = 3 one finds that Z vanishes logarithmically with q = |
2
k
Z ln1 02 .
(40)
q
(In the notation of [6] one has Z = S = AS if appropriate units for length and time are used.)
Also vanishes logarithmically, Z . The overall behavior for d = 3
1
ln 2 (k02 /q 2 )
n(
q)
q
(41)
differs from the free BoseEinstein distribution

which vanishes for q = 0 and T = 0. The infrared
contribution to the total particle number dq q 2 n(q) remains finite.
For d = 1 and d = 2 the issue is more involved. It is convenient to introduce renormalization
fields

= A .
(42)
378
The renormalized order parameter 0 is related to the superfluid density ns by ns = 0 = 02 . For

T = 0 the Galilei invariance for non-relativistic bosons yields a Ward identity which implies that
the total particle density n equals 0 . Since A depends on momentum the unrenormalized order
parameter 0 becomes effectively momentum dependent. In particular, 0 vanishes for d = 1 and
q 0, while 0 remains different from zero. Writing
Z = AS,
S q S ,
A q
(43)
one finds S 1 for d = 1, 2. For d = 2 the anomalous dimension vanishes for q 0, while
for d = 1 one has a range 0 < 1/2. We also introduce the renormalized coupling = /A
and note that s = S/ goes to a constant for q 0.
The resulting expression for the occupation numbers reads for q 0

S
M0 1/2
1
n(
q) =
(44)
q 1++ 2 .
s
2 S(q)A(q)q

For d = 2 the integral dq q d1 n(q) remains infrared finite only for S > 2, while for d = 1
finiteness requires S > 2. For d = 1 this condition is not obeyedin this case already the
classical approximation with vanishing and S leads to a logarithmic divergence. We will
see in the next section how this too strong infrared increase of n(
q ) is cured by a modification of
the q0 -dependence of the propagator.
We next turn to the behavior for large q as given (for all dimensions) by Eq. (36). We may use
Z = 1 + Z where Z vanishes for q . This results in
n(
q) =
M 2 2 04
q 2 (q 2 + 4M 02 )
Z
.
2
(45)
For a first evaluation of Z we may use the flow equation of [6], replacing k by q. For the
regime q 2 4M 02 one finds
q
32vd 2 2 2
Z = ( + S )Z =
M 0 ( q)d4
q
d
(46)
with vd1 = 2d+1 d/2 (d/2). In consequence, the leading order (with constant , 0 )
Z =
32vd M 2 2 02
d(4 d)
( q)d4
(47)
is positive and decreases q d4 . A similar result follows from a direct evaluation of the one loop
contribution to Z , if Z is defined as the coefficient of the term linear in q0 of an expansion of
the inverse propagator in powers of q0 cf. Appendix A.
The simple estimate (45) for Z cannot be true. The decay for large q is too slow. The
term Z would dominate n(
q ) for large q and lead to negative occupation numbers. Also
the q -integral for the total particle density would not converge for d > 2. The origin of this
problem is again the neglection of the non-linear q0 dependence in the inverse propagator. It
is not justified to evaluate the q0 -dependence by the q0 -derivative of the inverse propagator at
q0 = 0. If one wants to use an approximation linear in q0 for the inverse propagator, Z should
be evaluated at the location of the pole rather than near q0 = 0. We will develop in Sections 610
a renormalization group treatment for the computation of n(
q ). As a result, one finds that there
is actually no contribution from Z in Eq. (45). Omitting this term the ultraviolet contribution
379
to the total particle number becomes UV-finite

nUV =
q
M 2 2 04
q 2 (q 2 + 4M 02 )
(48)
as for the Bogoliubov approximation.

5. Boson occupation numbers for general dynamic term
Let us next address the infrared problem for d = 1. For the Bogoliubov approximation the
occupation numbers diverge 1/q and the particle density n does not remain finite. This problem is exacerbated if the q -dependence of 02 /Z2 is taken into account. The remedy of this
problem is related to the q0 -dependence of a and b in Eq. (28), that was neglected so far. Indeed,
the investigation of [6] reveals a crossover to a relativistic kinetic term for small momenta. For
q qr one may neglect the off-diagonal term q0 Z in Eq. (28). Due to the strong decrease of
Z for small q (43) the q0 -dependence is now dominated by a contribution quadratic in q02 in a
and b

2
q2
A
q
2
b=
+ 2 0 + V q02 ,
+ V q02 .
a=A
(49)
2M
2M
In this region the Goldstone mode dominates
1
1
g G
22
2
2b
and the location of the pole is now at

2
Aq
0 =
.
2M V
(50)
(51)
With
Z = 2V |0 | =
1/2
2AV
q
M
the particle density for q 0 (q = 0) obeys

M 1/2 1
q .
n(
q) =
8A V
(52)
(53)
The result (53) for the low momentum behavior of the occupation numbers differs qualitatively from the Bogoliubov theory (38). It involves the new coupling V , such that the coefficient
in front of q 1 does not vanish anymore for 0 0. For q 0 one finds [6] that V = V /A goes
to a constant such that (cf. Eq. (43)) for d = 1, 2
n(
q ) q (1) .
(54)
In consequence, the integral dq q d1 n(q) remains infrared convergent both for d = 2, where
= 0, and for d = 1, where > 0. We note that the infrared convergence for d = 1 arises only as
a consequence of a non-vanishing anomalous dimension . The same effect guarantees that the
380
correlation function in position space decays with a power law for large separation (d = 1, =
it)

r2 + v 2 2 /2 .
G
(55)
Away from the infrared limit q 0 one has to follow the location of the poles of the general
0 = A 2 , 0 = 2 , S =
propagator given by Eqs. (49), (29). They obey, with = /A,
0
0
Z /A,

2
2
1
q2
2 1/2
2
2
2 q
q0 = 2 S + 2V 0 + V
(56)
S + 2V 0 + 2S V
.
M
M
2V
We discuss here the case where S, V and are independent of q0 . (A generalization for mildly q0 dependent couplings will result in implicit equations where these quantities have to be evaluated
on the location of the poles.) On the other hand, the couplings depend on q. For large q the
fluctuation effects are small, such that V 0, S 1, while for small q one has S 0, V = 0.
All solutions of Eq. (56) correspond to negative q02 and therefore to pairs of purely imaginary q0 .
Let us start with V = 0 where the poles at

q 2 q 2
1
q0 = iS
(57)
+ 20
2M 2M
can be associated with the Goldstone mode. We observe positive and negative frequencies for the
excitations, corresponding to particles and antiparticles, the latter being associated to holes in
the condensate. For small non-zero V , the poles (57) get slightly shifted. Two new poles appear
at large |q0 |,
S
.
(58)
V
As S/V decreases towards zero, these modes correspond to the radial modes of the relativistic
model, i.e. for S 0

1 q2
S2
q 2
2
q0,R =
(59)
+ 20 +
1+
.
V 2M
V
4M0
q0 = i
The Goldstone poles take for S 0 the standard relativistic values

q2
S2
2
1
.
q0,G =
2MV
2V 0
(60)
The radial mode corresponds to collective fluctuations related to the size of the condensate 0 ,
whereas the Goldstone mode is associated to phase changes of the condensate. More generally,
we refer to the negative (positive) sign of the square root in Eq. (56) as the Goldstone (radial)
poles.
For a computation of the occupation number with the propagator (49) we evaluate for q = 0

1
0 , q)
0 , q) =
tr G(q
n(
q ) = g(q
2
q0
1
=
2

q0
q0
a+b
1
=
2
2
A
ab + Z q0

q0

q2
+ V q02 + 0 det1 (q)
2M
(61)
with
381
2

q2
q
+ V q02 + 20
+ V q02 + S 2 q02
2M
2M

1
1
2
= V 2 q02 +
q
+
,
0
2V 2
2V 2
det(q) =
(62)
and
= S 2 + 2V 0 +
V q2
,
M
2

q2
= S 2 + 2V 0 + 2S 2 V .
M
(63)
Writing

1
1
1

,
det1 (q) =
2 + +
q 2 +
q
2
2
0
0
2V
2V
we recognize the contribution from the radial and Goldstone poles

1
1 S2/ 1 + S2/
+
n(
q) =
.
2AV
q02 +
q02 + +
q
2V 2
2V 2
(64)
(65)
The q0 -integration is performed easily

1
1 S2/ 1 + S2/

n(
q) =
+
2 2A

+
1/2
M 2 2
F,
q q + 4M0
=
2 2AV
(66)
with

S2
F = + + +
(67)
+ .
For the limiting case V 0 one has = S 2 +V (q 2 /M +20 ), F 2(V /S)(q 2 /M +

2
2 ) and we recover Eq. (36).
In the
limit S 0 we can use = (q + 2M0 )V /M,
0
2
2
= 2V 0 and F = V /M( q + q + 4M0 ) such that we recover (53).
For an estimate of the leading behavior for q we need F /V and therefore the qdependence of V . A renormalization group estimate [6] yields
lim V (q) =
10vd
2 0 q d6 .
d(6 d)
(68)
In consequence, the terms V q 2 in Eq. (63) become subleading and the ultraviolet behavior is
well described by the linear dynamic term in Section 4.
In the spirit of a renormalization group improvement, which will be motivated in the next
sections, we may derive a differential equation for the q-dependence of np
np (q) n(q)
=
= (q),
ln q
ln q

1/2
q2
F
M 2 2
.
1+ 2
F
q q + 4M0
(q) =
ln q
q + 4M0
2 2AV
(69)
382
Here F/ ln q should be evaluated for fixed couplings. The advantage of such an equation is that
the ultraviolet problem associated with the additive counterterm is now absent. For large q the
function (q) decays q 4 . We can therefore integrate Eq. (69), starting with an initial value for
large q = for which the perturbative result Eq. (36) with Z = 1) can be trusted. Furthermore,
S, , 0 , V will
the renormalization group improvement of using q-dependent couplings A,
make the solution of Eq. (69) rather accurate. It can be used for all dimensions and is free of
infrared or ultraviolet problems.
Eq. (69) can directly be used to compute the particle density or condensate depletion np =

and k a suitable
q ) by solving a second order differential equation. With x = ln(q/k)
q np (
) from the differential equation
momentum unit, one may compute np = k d n(x
n
2 n
d
= vd edx (x),
2
x
x
(70)
using the initial condition n()
= 0 (or a similar condition for x = ln(/k)).

6. Renormalization group improvement
In view of the rather complicated momentum dependence of the full propagator it is useful
to consider the contributions from fluctuations with different momenta p separately. This will
also allow us to use a renormalization group improvement by employing effective couplings
appropriate for a given momentum range. In this section we will follow a heuristic approach.
The second part of this paper will present a more formal development, based on exact functional
renormalization group equations.
We start by adding to the inverse propagator an infrared cutoff piece such that

1
k (q0 , q) = P (q0 , q) + Rk (
q) .
G
(71)
The infrared cutoff is a 2 2 diagonal matrix and suppresses all modes with q 2 < k 2 . We choose
Rk (
q) =

A 2
k q 2 k 2 q2
2M
(72)
such that the modes with q 2 > k 2 are not affected. On the other hand, the increase of the classical
free propagator for small momenta, G q 2 , is stopped at q 2 = k 2 . We define a k-dependent
occupation number (for q = 0)

1
k (q) 1
q) =
Tr G
np,k (
(73)
2
2
q0
and consider the derivative

1
2k (q0 , q) .
q) =
Tr k Rk (
q )G
k np,k (
2
(74)
q0
We note that k Rk vanishes for q 2 > k 2 , such that the running stops once k q. Furthermore,
for the classical propagator the r.h.s. of Eq. (74) becomes independent of q for q2 < k 2 . The
q0 -integration on the r.h.s. of Eq. (74) will be dominated by k-dependent poles. The physical
occupation number obtains by solving Eq. (74) for k 0. Indeed, the absence of the cutoff for
q ) = np,0 (
q ).
k = 0 implies np (
More explicitely, the flow equation reads

2
k A 2
2
11 + G
222 + 2G
12 G
21 .
k np,k (
1
G
q) =
k q
2M
2
383
(75)
q0
We will neglect the anomalous dimension
(76)
ln A
k
in this section. Inserting the ansatz (28) and approximating a, b and Z to be q0 -independent one
finds
= k
kk np,k (
q) =
4M 2 2 04
Z
k(k 2
+ 4M 02 )3/2

k 2 q2 .
(77)
Using for k the initial condition np,k = 0 an integration for k-independent , 0 , Z

yields for k = 0 Eq. (36).
The renormalization group improvement uses k-dependent parameters , 0 and Z . These
k-dependent couplings are in turn computed in presence of the cutoff scale k, such that they
approximate best the relevant couplings in the momentum range p2 k 2 . We will motivate this
procedure more explicitely in the second part of the paper. A crucial property for the ultraviolet
issue is Z (k ) = 1. This simply follows from the absence of fluctuation effects for an
infinitely large cutoff. This property guarantees the correct initial value for the flow np,k (q) =
0 for all q . Indeed, the classical contribution to np,k for very large k yields Eq. (36) with q2
replaced by k 2 and all parameters evaluated at the scale k. It is obvious that Z (k ) = 1 is
needed for a vanishing occupation np,k in this limit. The initial value Z (k ) = 1 is well
compatible with a nonzero value Z (k 0). While only the latter is used in Eq. (36), the flow
equation (74) involves the value of Z (k) for all k 2 > q 2 . We will see that the approach with
a varying Z (k) avoids in a simple way the difficult ultraviolet issues encountered in the direct
evaluation of Eq. (36).
Let us consider the range of high momenta q 2 4M 02 . In the relevant k-range k 2 q2
we can approximate

q ) = 4M 2 2 (k) 04 (k)Z1 (k)k 5 k 2 q2 .
k np (
(78)
This yields

np (
q ) = M 2 2 04 Z1 q 4
(79)
where we have defined the weighted average

2 04 Z1 =

q
dk 2 (k) 04 (k)Z1 (k)k 5

dk k 5 .
(80)
k 5
strongly enhances the weight of the range k q and we may approximate

The factor
2 4 1
0 Z 2 (q) 04 (q)Z1 (q).
(81)
Comparison with Eqs. (36), (45) yields Z = 0 and therefore effectively Z = 1 for the range
q 2 2M2 02 . No difficult ultraviolet issue needs to be addressed anymore. In particular, the
384
renormalization group improvement avoids the need to evaluate the propagator at non-zero q0 ,
corresponding to the location of the pole. We have also found that the parameters , 02 , Z
appearing in Eq. (36) should be replaced by q-dependent effective couplings. We emphasize that
our argument in favor of Z (q) = 0 is valid for q 2 2M2 02 and does not necessarily apply
to small q 2 .
7. Occupation numbers from source terms
In Sections 810 we address the formulation of the flow equations for k-dependent occupation
numbers more formally. We will derive an exact functional differential equation and propose
non-perturbative approximations. The solution of the flow equations offers an alternative to the
direct computation of n(
q ) from the propagator (12). The q0 -integration in Eq. (12) depends in
a critical way on the precise knowledge of the q0 -dependence of the propagator which is quite
involved. In contrast, the flow equations are less sensitive to this dependence. Only a small q0
range influences the running at a given renormalization scale and rather crude approximations
are often sufficient. The prize to pay is the numerical integration of the flow equations. In this
paper we will not perform detailed quantitative computations. We rather use the flow equations
for a justification of the renormalization group improvement discussed in Section 6.
In the present short section we set up the formal tools for the computation of the momentum
distribution n(
q ). For this purpose we add appropriate source terms to the action (1)

S S h(
(82)
q ) (q)(q).
q
The partition function Z depends then on h(

q ) and we can formally extract n(
q ) as
n(
q)
1
1
= d+1
ln Z|h=0 .
2
h(
q)
(83)
Here we set the source term to zero after the differentiation and the minus sign applies for
fermions. We note that the source multiplies a composite operator and is non-local in the fields
(x)

x y) (, x)(, y),
(84)
h(
q ) (q)(q) =
h(
q
x y
z) the Fourier transform of h(

with h(
q ). Eq. (84) shows that the source term remains translation
invariant.
which generates the 1P I correlation
We will conveniently work with the effective action []
functions. It obtains by introducing local linear sources j (x) for (x) and performing a Legendre
transform of ln Z[j ], with (x)

= (x) |j in the presence of sources. The non-local sources
We also add
h(
q ) are not affected by this transformation and depends now on h(
q ) and (x).
a field-independent piece linear in h(
q ) which takes care of the additive renormalization 1/2
on the l.h.s. of Eq. (83)

d+1
= ln Z[j ] +
[]
(85)
h(
q ).
(q)j (q) + j (q)(q)
2
q
q
385
It is easy to verify that the occupation numbers obey

n(
q) =
.
d+1 h(
q ) 0 (x),h=0
(86)
Here the partial derivative /h(

q ) is taken at fixed (x)
and evaluated for the solution of the
field equation 0 (x) which obeys
(x)

= 0.
(87)
0 (x),h=0
h] in linear order in h. Expanding

Our goal will be the computation of [;

q]h(
h] = []
N [;
q ) + 0 h2
[;
(88)
q
the occupation numbers follow directly from N

n(
q) =
1
N [ 0 ; q].
d+1
(89)
The field equation for 0 only involves [].

8. Exact flow equation for occupation numbers
q] from its classical value results from quantum and thermal fluctuThe difference of N [;
ations. We include these fluctuation effects stepwise by introducing first an infrared cutoff Rk
which suppresses the fluctuations with momenta q2 < k 2 . (For fermions we may choose to suppress the fluctuations with |
q 2 kF2 | < k 2 .) At the end we lower k to zero such that all fluctuations
are included. We achieve this strategy by adding to the action (1) an infrared cutoff term [7].

k S = Rk (
(90)
q ) (q)(q).
q
In turn, the effective action is now replaced by the average action k which depends on k
[8,9]. With Rk (
q ) diverging for k all fluctuations are suppressed in this limit and one
has k = S. On the other hand Rk (
q ) = 0 for k 0 implies k0 = . The average action therefore interpolates smoothly between the classical action for k (or k = ) and the
effective action for k 0.
Its dependence on k obeys an exact flow equation [8]
k k [; h] =
(2)
1
1
,
Tr k R k [; h] + R
2
(91)
q )(q q ). The second functional derivative k is given by the full inwith R(q, q ) = Rk (
For a homogeneous background field
verse propagator in the presence of background fields .
(2)

)
one has k (q, q ) = P (q)(q q ), with P a matrix in the space of complex fields (,
1
or real fields (1 , 2 ), with, = (1 + i 2 ). The trace involves a momentum integration and

2
trace over internal indices. We will choose here the basis ( 1 , 2 ). For homogeneous background
(2)
386
fields Eq. (91) takes the explicit form (with tr the internal trace)

1
d+1
k k =
q ) 2 (q) + Rk (
q) .
tr k Rk (
2
(92)
The precise shape of the cutoff function Rk is, in principle, arbitrary. As an example, one may
consider the optimized cutoff [12]

A 2
(93)
k q 2 k 2 q 2 .
2M
We observe that for Rk depending only on q the non-local source term (82) may be viewed as a
q ) Rk (
q ) h(
q ) such that a variation with respect to h can also be interpreted as a
shift Rk (
variation in the space of cutoff functions.
The presence of the non-local source h(
q ) does not change the general structure of the exact
flow equation (91). Employing again the expansion (88) (now with all quantities depending on
obtains from Eq. (91) by simply omitting the h-dependence, corresponding
k) the flow for k []
an evaluation for h = 0. The solution of the field equation (87) 0 (k) will now depend on k. The
q] obtains from Eq. (91) by taking a derivative with respect to
exact flow equation for Nk [;
h(
q)
Rk (
q) =

1
1
1
k Nk = Tr k R k(2) + R Nk(2) k(2) + R
.
2
(94)
Here Nk (q , q ) is the second functional derivative

(2)
Nk,ab (q , q ) =
(2)
2 Nk
.
a (q ) b (q )
(95)
Eq. (94) will be our basic starting point for a computation of n(
q ) through a flow equation. We
will consider a k-dependent nk (
q ) as defined by Eq. (89) and extract the occupation numbers for
k 0.
(2)
For a homogeneous 0 (k) one has Nk (q , q ) = 2 (q )(q q ) and the flow equation for
Nk simplifies

2
d+1
k Nk [ 0 ] =
(96)
q )2 (q ) P (q ) + Rk (
q ) ,
tr k Rk (
2
q
where 2 is evaluated for = 0 . This results in the final flow equation for the occupation numbers

2 nk (
q)
1
k nk (
(97)
q ) = tr k Rk (
q )2 (q ; q) P (q ) + Rk (
q )
+
k 02 .
2
2
0
Here the second term accounts for the k-dependence of the solution 0 (k).
For k the fluctuation effects can be computed exactly. For bosons and 0 we start
for k = with the initial value

1
N [; q] =
(98)
a (q0 , q) a (q0 , q)
2 a
q0
387
and
q](q , q ) = (
q q )(q q )ab .
N [;
(2)
(99)
More precisely, the initial value for k has a classical term and a one loop contribution.
For bosons and 0 the latter precisely cancels the classical contribution from the last term in
Eq. (85). Since there is essentially no running between k and k = we may begin the flow
at k = . (For fermions and > 0 the one loop effect is less simple, see Eq. (140).) According
to Eq. (98) one has for 0 the initial value
n (
q ) = 02 ()(
q ).
(100)
The only possible contribution arises from a condensate 0 () = 0. For q = 0 a non-zero occupation number has to be built up as a result of the flow.
We want to extract n(
q ) from a solution of the flow equation (94) for k 0. This will only
be possible in suitable approximations. One may imagine to solve first the flow equation for k
and then use the result in Eq. (95). For this purpose we will need 2 (q ; q).
q ) in terms of the kThe quantity obeys a type of Ward identity. We have defined np (
dependent propagator. Using the exact flow equation for the exact propagator [9] yields a second
exact equation for the flow of np . The comparison of the two exact flow equations yields the
Ward identity.
Let is demonstrate this for a homogeneous setting where

1
0 , q) 1 ,
np,k (
(101)
q) =
tr G(q
2
2
q0
the k-dependent propagator matrix. For q = 0 this yields the flow equation
with G
q ) = n (
q) +
k nk (
with
q) =
n (
1
2
q0
1
2
q)
nk (
k 02 ,
2
0
(102)
0 , q)
tr k G(q

tr (P + Rk )2 k (P + Rk ) (q0 , q).
(103)
q0
If we decompose
q q) + p (q , q)
(q ; q) = (
(104)
we obtain by comparing Eqs. (97) and (102)

2 (q ) = tr k P (q0 , q)G
2 (q0 , q).
tr k Rk (q )p (q ; q)G
q
(105)
q0
(2) (q, q ) = P (q)(q
With
the exact flow equation for P follows from the second functional derivative of Eq. (91) [8]. It can be written as a one loop expression

2 (q )H(q ; q),
k P (q) = tr k Rk (q )G
(106)
q
q )
388
acts (and over which the trace in Eq. (106) is taken)

with H a matrix both in the space where G
as well as in the space where P acts. We identify

2 (q0 , q),

p (q ; q) = tr H(q ; q0 , q)G
(107)
q0
acts.
with
tr denoting the trace in the space where G
9. Quadratic truncation
The flow equation (97) is exact. However, even if P (q ) would be known from a solution to
the r.h.s. still involves the unknown second functional derivative 2 .
the flow equation for k [],
One may derive an exact flow equation for 2 , but in turn it will involve even higher functional
derivatives of N . We need to approximate the solution by a suitable ansatz (truncation) for the
general form of N .
As a first approach we consider bosons and truncate N in the number of fields. We keep only
terms with up to two powers of
q] = d+1 np,k (
Nk [;
q)

1
Zc,k (q0 , q) a (q0 , q) a (q0 , q) + Zp,k (q , q) a (q ) a (q )
+
2 a
q0
d+1 Zp,k (0, q) 02 .
(108)
q q) are acHere Zp,k is a regular function of q , q , while possible singular parts (
counted for by a separate function Zc,k . (Both contributions can be combined to a single function
Zh (q , q), see below.) This truncation still contains three functions np,k , Zc,k and Zp,k .
The initial values are
q ) = 0,
np, (
Zc, (q0 , q) = 1,
Zp, (q , q) = 0.
(109)
q ) = 0 (q ) the momentum distribution reads

For 0 (
q ) + Zc,0 (0, 0) 02 (
q ).
n(
q ) = np,0 (
(110)
The condensate density

nc = Zc,0 (0, 0) 02
(111)
differs from the bare order parameter 02 only if Zc (0, 0) = 1. The total density n and condensate
fraction c read

q ),
c = nc /n.
n = nc + np (
(112)
q
We need to understand the meaning of the functions Zc,k and Zp,k and start by defining them
properly. Let us first use a combined density transfer function Zh
Zh (q , q) = Zp (q , q) + Zc (q0 , q)(
q q)
(113)
389
and define Zp and Zc by their behavior for q q . We want to choose Zp (q , q) such that it
remains finite for q = q . We therefore define

1

Zc (q0 , q) = d
(114)
Zh (q , q)(
q q),
q
which yields Zp implicitly by subtraction in Eq. (113). (One may replace (

q q) by a

smoothened -function and change Zp (q , q) correspondingly. This concerns the practical definition which particles are counted as having zero momentum.)
In terms of Zh the second functional derivative reads
N (2) (q , q ) = Zh (q , q)(q q ),
2 (q , q) = Zh (q , q).
(115)
All quantities need to be evaluated for the solution of the field equation 0 (k). The substraction
of a constant term Zp,k 02 in the truncation (108) guarantees that the expansion in powers of
fields is correctly performed around 0 (k) and that Zp does not contribute to the particle density.
The flow equation for n(
q ) (97) has both a contribution from Zp and Zc

2
1
k n(
q ) = tr k Rk (
q )Zc (q0 , q) P (q0 , q) + Rk (
q)
2
q0
1
tr
2

2 n(
q)
k Rk (
q )Zp (q , q) P (q ) + Rk (
q )
+
k 02 .
02
(116)
In the remainder of this section we argue that the condensate density is precisely given by 02 .
This means that Zc (0, 0) = 1 is not renormalized, as will be shown more precisely in the next
section. We observe that both contributions to Eq. (116) from Zp and Zc are regular for q 2 0
and therefore do not contribute to the running of the condensate density nc . The latter simply
obeys for 0 = 0
k nc =
nc
k 02
02
(117)
and k nc = 0 for 0 = 0.
We will still need the 0 -dependence of n(
q ). Again, we derive a flow equation for
q) =
k (
q)
nk (
,
2
(
q ) = (
q ).
(118)
It reads (G 1 = (2) + R)

1

Tr k R 2 GN (2) G
2d+1
0

(2)

1
(2)
= tr
Gk Rk G
,
GZh + Zh G
2
02
02
q) =
k (
(119)
where we have omitted a term Zh / 02 in the last equation. The only dependence on q arises
from Zh and we observe that the r.h.s. of Eq. (119) remains regular for q2 0. In consequence,
390
we can use for all k a non-singular k (

q)
(
q ) = (
q ) + (
q)
(120)
and the condensate density obeys

nc
= 1,
2
k nc = k 02 ,
nc,k = 02 (k).
(121)
This closes the argument that the condensate density is not renormalized and shows that np (
q)
has no contribution (
q ). For the SSB regime with 0 (k)2 > 0 we may define Zc (0, 0) = nc / 02
such that
k Zc (0, 0) =
Zc (0, 0)
k 02 = 0.
02
(122)
A simple solution with the initial condition of a independent Zc, (0, 0) = 1 is Zc (0, 0) = 1 for
all k.
10. Density transfer function
The right-hand side of the density flow equation (116) involves the density transfer function
Zh (q , q) (113). In order to understand its role we neglect for a moment the dependence of Zh
on q0 . The first two terms in Eq. (116) can then be written as

k n(
(123)
q ) = Zh (
q , q)k n(
q ) +
q
q ) accounts for the contribution of a loop with momentum q

where k n(

2
1

k n(
q ) = tr k Rk (
q ) P (q0 , q ) + Rk (
q ) .
2
(124)
q0
For a q0 -dependent Zh the folding (123) is in a certain sense averaged over q0 . The density
transfer function describes how a loop with momentum q influences the occupation number
n(
q ).
The flow equation for Zh (q , q) obtains by taking the second functional derivative of Eq. (94)
In the truncation (108) N (2) does not depend on and one finds with G =
with respect to .
k
(2)
(k + R)1

1
2 (2)
(2)
k Nab (q , q ) = Tr Gk RGN (2) G + GN (2) Gk RG
2
a (q ) b (q )

(2)
(2)
1
Tr Gk RG GN (2) G
2
a (q )
b (q )

1
(2)
(2)
(2)
Tr Gk RG
GN G
2
b (q )
a (q )

1
Tr G N (2) Gk R + k RGN (2) G

2
(2)
(2)
(2)
(2)
G
+
G
a (q ) b (q ) b (q ) a (q )
391

.
(125)
The first term involves a quartic vertex and the two last terms arise from cubic vertices present
for 0 = 0. We will approximate the vertices here by pointlike vertices
cd (p , p )
= abcd (p p + q q ),
a (q ) b (q )
(2)
cd (p , p )
= acd 0 (p p + q ),
a (q )
(2)
cd (p , p )
= bcd 0 (p p q ).
b (q )
(2)
(126)
For a quartic coupling one has
acd = acd = 2(a1 cd + c1 ad + d1 ac ),

abcd = (ab cd + ac bd + ad bc ).
(127)
For a homogeneous setting G is diagonal in momentum space. One obtains the flow equations
(2)
(q , q ; q)
k Ncc

2

= (q q )

3
k Rk (p ) ccab G
(p )Zh (p , q)
ba

a,b=1 p
2
2

02 2
G
(p ) cbd cea G
(p + q )Zh (p + q , q)
ab
de
2
d,e=1
2
+ cbd cea G
(p q )Zh (p q , q)
de
3
de (p + q )
+2 G
(p )Zh (p , q) cbd cea G
ab

de (p q ) .
+ cbd cea G
(128)
(2)
Since k N11 gets, in principle, also contributions from higher order terms neglected in our truncation we identify
k Zh (q , q)(q q ) = k N22 (q , q ; q).
(2)
(129)
The r.h.s. of Eq. (128) remains finite for q q . The contribution from Zc changes the q integration to an integration over q0 with values of p fixed by the -function in Zh , i.e. p = q
1 . This
or p = q q . The q0 -integration is well behaved due to the infrared cutoff Rk in G
leads to the important conclusion that renormalization effects for the condensate contribution are
absent
k Zc (q0 , q) = 0.
(130)
The condensate density is therefore given by the bare order parameter

nc = 02 .
(131)
392
This is consistent with the discussion after Eq. (122).

In contrast, the build up of the occupation number for uncondensed particles is influenced by
the renormalization of Zp

2
1
k np (
q ) = tr k Rk (
q ) P (q0 , q) + Rk (
q)
2
q0
1
tr
2

2 np (
q)
k Rk (
q )Zp (q , q) P (q ) + Rk (
q )
+
k 02 .
2
(132)
Only the first term in Eq. (132) is present in the classical contribution that neglects the renormalization effects for Zh . For the second term the initial value Zp = 0 does not remain a solution
of the flow equation. A non-zero Zp is generated by the contribution (
q q) in Zh (q , q) in the
r.h.s. of Eq. (128). Whereas the first term involves only on shell momenta where the propagator
in the loop is fixed to q = q , the second term gets also contributions from off shell momenta, i.e. from q different from q . This reflects the broadening of the density transfer function
Zh (q , q) due to the renormalization flow.
11. Occupation number flow at low temperature
In order to get some practice with the flow equations we first check simple situations. We start
with the vacuum. The term 02 in Eq. (128) is absent. We want to compute the contribution
from Zc to the flow of Zp
k Zp (p , q) =

22aa k Rk (
q)
a=1
3
G

aa
(q0 , q).
(133)
q0
For large |q0 | the integrand decays

It is convenient to work in the complex basis with diagonal G.
|q0 |3 such that we can close the integral either on the upper or lower half plane. The integral
has only one pole in the upper half
vanishes, as most easily seen in the complex basis where G
plane and we can evaluate the closed integral in the lower half plane. In consequence, Zp stays
zero for all k and the density transfer function remains a -function
Zh (p , q) = (p q).
(134)
Inserting this result in Eq. (132) only the first term contributes

1
2 (q0 , q).
k np (
q ) = k Rk (
q ) tr G
2
(135)
q0
2 (q , q) decays sufficiently fast for large |q | and the r.h.s. of Eq. (135) vanishes, as it
Again, G
0
0
should be.
Next we may have a look on the free theory for T = 0. In the absence of interactions the density transfer function does not flow and Eq. (135) remains valid, with (2) the classical inverse
propagator. Let us use for 0 the explicite cutoff (93) such that
2

k2 2
q 2
k 2 q 2 2
2

2
q ) = k q
.
+
k q
iq0 +
kk n(
(136)
M
2M
2M
qo
393
The occupation number n(

q ) flows only as long as k 2 > q 2 , then the flow stops. In the range
k 2 > q2 the flow equation simplifies
2

1
k2

np =
= 0.
iq0 +
(137)
2M
2M
k 2
q0
q ) vanishes we conclude that only the condensate nc

Since for large k = the initial value np, (
contributes to the particle number, n = nc .
For fermions a positive chemical potential > 0 is relevant. Now Eq. (136) is ill defined
when max (
q 2 , k 2 ) = 2M. This reflects that the cutoff (93) is not well adapted to the problem.
Indeed, instead of cutting out the small momentum fluctuations an efficient cutoff should rather
suppress the fluctuations close to the Fermi surface, i.e. with |
q 2 2M| < k 2 . For the fermions
we will therefore employ the cutoff

1 2
(F )
Rk =
k sgn k 2 ,
2M
= q2 ().
(138)
(F )
For positive the propagator in presence of Rk

1
2
k2
G = iq0 +
,
sgn q 2M
2M
becomes in the vicinity of the Fermi surface

(139)
(F )
such that the pole in presence of Rk remains in the same half plane in complex q0 -space as for
Rk(F ) = 0. As a result, the initial value of the flow is given by the Fermi distribution

nF, (
(140)
q ) = 2M q 2 .
In turn, the flow equation (137) is replaced by

2
2
sgn(
q 2 2M)
k2

n
(
q
)
=
+
2M
=0
sgn
q

iq
F
0
2M
2M
k 2
(141)
q0
and the Fermi distribution is maintained for k 0. For T > 0 the r.h.s. does not vanish and the
flow accounts for the smoothening of the Fermi distribution.
We next turn on the interactions. The initial values np, (
q ) = 0, nF, (
q ) = (2M q 2 )
are not changed. However, the flow of n(
q ) can now account for non-vanishing np (
q ) or for a
smoothening of the Fermi surface due to fluctuations. There are two possible effects. The first
concerns the modifications of the propagator. Furthermore, a second effect arises from the build
up of a non-vanishing Zp .
Let us first consider bosons and employ the cutoff (93). The flow of n(
q ) (cf. Eq. (124) be
comes ( = kk ln A)

2
k A 2
21 .
11 + G
222 + 2G
12 G
k n(
(142)
G
q) =
k q2 1
2M
2
q0
We use the general form (28)

211 + G
222 + 2G
12 G
21 =
G
(a + Rk )2 + (b + Rk )2 2q02 Z2
[(a + Rk )(b + Rk ) + q02 Z2 ]2
(143)
394
with momentum dependent coefficients a, b, Z . In order to understand the general structure we

concentrate on large q2 where perturbation theory should remain valid. We want to see in which
limit we recover the Bogoliubov theory and what are the extensions of it. For this purpose we
, Z this
may and employ the ansatz (34). Neglecting and a possible q0 -dependence of A,
yields (Z = AS)

k
k 2 q 2
k n(
q) =
2M A
=

q0
(k 2 /2M + 2 02 )2 + (k 2 /2M)2 2q02 S 2

[(k 2 /2M)(k 2 /2M + 2 2 ) + q 2 S 2 ]2
4M 2 2 04
Z k 2 (k 2 + 4M 02 )3/2

k 2 q2 .
(144)
For k-independent , 02 and Z the integration of Eq. (144) yields

n(
q) =
dk
q
1
=
2Z
4M 2 2 04
Z k 2 (k 2 + 4M 02 )3/2

1+
+ n (
q)
4M 2 2 04
q 2 (q 2 + 4M 02 )
1/2

1 {q } + n (
q ).
(145)
q ) = 0 (we only consider here q = 0) the sum of the last two terms vanishes,
For , n (
such that
1/2

4M 2 2 04
1
1
.
1+
n(
q) =
(146)
2Z
q 2 (q 2 + 4M 02 )
If we replace n by n this is almost Eq. (36)the difference vanishes for Z = 1. The Bogoliubov
result therefore obtains if we use the classical transfer function Zc = 1, Zp = 0 and neglect the
dependence of a, b, Z on q0 , q and k.
Extensions beyond the Bogoliubov theory take into account the running and the momentum
dependence of the parameters as well as a non-trivial transfer function. In a first step we may
use in Eq. (146) effective parameters , 0 , Z which depend on q 2 . Based on an approximate
solution of the flow equation for the occupation numbers we argue that one should use the values
(
q , k 2 = q 2 ) and 0 (k 2 = q 2 ) instead of the values at k = 0. This is justified since the build up
of a non-vanishing n(
q ) is dominated by the range k 2 q 2 .
Indeed, the flow (144) stops for k 2 < q 2 , while for k 2 q2 it is suppressed by inverse powers
of k. Using also Z = Z (
q , k 2 = q 2 ) we note that Z multiplies now a term that vanishes for
q , in distinction from Eq. (36). As already noted in Section 6 this renormalization group
improvement gets rid of the dangerous term Z in Eq. (45). Indeed, the replacement of Z (k)
by a k-independent constant is only allowed if Z multiplies a quantity that vanishes for large q,
as in Eq. (146). This is necessary since Z multiplies in the flow equation (144) a term k 5
for large k. The properly implemented renormalization group flow takes care of the ultraviolet
problems, even if the resolution of the q0 -dependence of the propagator is only poor. This has
direct consequences for the -dependence of the relation between particle density and chemical
potential, n(). For large the momentum integration (for q < ) of n(
q ) yields n d4 ,
with a coefficient that depends on . For T = 0 the cutoff dependence of n() is suppressed by
(2M/2 )(4d)/2 .
395
A second improvement beyond Bogoliubov theory becomes necessary if we consider small q.

Now the flow of the occupation numbers should also contain the term V q02 in the inverse
propagator, as we have discussed in Section 4.
Let us finally discuss the flow of the density transfer function which generates a non-vanishing
Zp according to Eq. (128). The cubic couplings must have an even number of indices taking the
value two. They are given by
212 = 221 = 122 = 212 = 221 = 122 = 2 ,
111 = 111 = 3 2
(147)
and the flow induced by these cubic couplings becomes
k Zp,3 (q , q)
2 k 02 2
2
2
2
2
G
k p 2
(p ) G
(p + q ) + G
(p ) G
(p + q )
=
11
22
22
11
M
p
2
2
2

212 (p ) G
(p + q ) + G
(p ) G
(p + q ) Zh (p + q , q)
+G
12
21
21
3
3
22 (p + q ) + G
11 (p + q )
+ G
(p )G
(p )G
11
22
3

12 (p + q ) + G
3 (p )G
21 (p + q ) Zh (p , q)
+ G
(p )G
12
21

+ (q q ) .
We evaluate Eq. (148) for q = 0, with the ansatz (28) and a = a + Rk , b = b + Rk ,

2 a b + p0 2 Z2 3 .
k Zp,3 (0, q) = 42 k 02 k 2 p 2 Zh (p , q)(a + b)
(148)
(149)
For the high momentum region we use Eq. (34) and the cutoff Rk (72) such that
A 2
k ,
b =
2M
The flow
02 .
a = b + 2A
(150)
k Zp,3 (0, q)

12M 2 2 02 (k 2 + 2M 02 )2

2
2
2
2
q)
=
k q + k p Zp (p,
Z A 2 k 4 (k 2 + 4M 2 )5/2
0
(151)
p
stops for q 2 < k 2 and is dominated by k 2 q 2 . As long as Zp remains small we can neglect the
second term in the curled bracket. Omitting also the contribution from the four point vertex one
finds for large q
Zp (0, q) =
3M 2 2 02
.
Z A 2 q 4
(152)
We conclude that Zp appears only in higher order in perturbation theory ( 2 ) and is suppressed for large q. For a discussion of the ultraviolet behavior it can be neglected. For lower q,
however, it may become necessary to include a non-trivial transfer function Zp if one aims for
quantitative accuracy. It may be evaluated alternatively from Eq. (106).
396
12. Gaps and pseudogaps

Besides a formal theoretical framework for systematic non-perturbative approximations, the
exact flow equation for the occupation numbers can also be employed to motivate comparable
simply renormalization group improvements. Examples are the discussion in Section 6 or the
computation of the occupation numbers by a solution of the differential equation (69), with parameters evaluated at k 2 = q 2 . In this section we discuss gaps and pseudogaps in the light of such
renormalization group improvements.
Let us first discuss fermions in presence of an order parameter or mean field. For definiteness,
we consider a system of fermionic atoms, interacting with an effective bosonic di-atom fields .
This field can represent microscopic molecules as well as Cooper pairs or other collective states
with atom number two. For our purpose it is sufficient to consider a rather simple effective action
for fermion fields with two spin states and

h T
=
(153)
.

2M
2
x
The Yukawa or Feshbach coupling h couples the fermions to bosons. For a constant real order
parameter (x)
= 0 this action is Gaussian and reads in momentum space

(q)
1
T
=
(154)
(q) =
.
(q)PF (q)(q),
2
(q)
q
The inverse propagator PF is a 4 4 matrix [10]

PF (q) =
h 0 ,
(iq0 +
q2
2M
(iq0
) ,
q2
2M
+ )
h 0
(155)
where , = 1, 2 are spin indices and 12 = 21 = 1. Correspondingly, the propagator matrix

reads
!
q2
(iq0 + 2M
)
h 0 ,
1
detF PF =
(156)
,
q2
(iq0 2M
+ ) ,
h 0
with

q2
q2
detF = iq0 +
iq0 +
+ h 2 02 .
2M
2M
(157)
The fermionic occupation number obtains as

0,
1
1
nF (
q) =
tr
PF + 1
2
,
0
q0

=1

q2
2
det1
F .
M2
q0
For 0 = 0 we recover twice the result (17), corresponding to the two spin states.
(158)
Let us next evaluate nF (

q ) for T = 0. The zeros of detF occur for
2
2

q
2
2 2
q0 =
+ h 0 .
2M
One finds
q2
nF (
q) = 1
2M

q2
2M
2
+ h 2 02
397
(159)
1/2
,
(160)
such that the Fermi surface is smoothened with a typical width set by the gap = h 0 . This
generalizes to T = 0

1
1 q2
2 q2

q) = 1
+
exp
+1
nF (
(161)
2M
2M
T
with positive square root

2
2
q
=
+ 2 ,
2M
= h 0 .
(162)
In presence of a gap the fermion density nF does not vanish for T = 0 and any finite . For
T = 0 and = 0 the occupation numbers approach
nF (
q) = 1
q2
q 4 + 4M 2 2
q2
.
2M
According to Eq. (159) the dispersion relation

(
q ) = (q)
(163)
(164)
is characterized by a minimal frequency set by the gap ||min = . For a constant gap we have
recovered the results of mean field theory (BCS-theory).
In a more accurate treatment the Yukawa coupling in Eq. (153) becomes momentum dependent. In consequence, the diagonal entries in Eq. (155) depend on momentum. One ends with
a momentum dependent gap (
q ) which modifies Eqs. (161)(164). We present here a simple
way of computing (
q ) without explicit momentum dependent truncations.
Renormalization group improvement implies that for any q the gap (
q ) should be evaluated
at a characteristic momentum scale k. For fermions we associate

(
q ) = h (k) 0 (k), k 2 = q2 2M.
(165)
Effectively, the fermions with momenta distant from the Fermi surface, q = qF + q, qF2 =
2M, are influenced by an effective volume with linear size k 1 , where k 2 = |2
qF q + q2 |.
2
(The energy needed for a change q in momentum is k /(2M), and k is a typical momentum of
a boson whose absorption/emission can induce such a change.) The quantities h (k) and 0 (k)
are associated with the running couplings evaluated at the renormalization scale k. (This is
precisely what happens if one investigates a k-dependent particle density as in [11].) In summary,
this allows us to use the flow of h and 0 , evaluated in a momentum independent tuncation for
q ).
h , in order to estimate the momentum dependent gap (
Let us define (k) = h (k) 0 (k). A true non-vanishing gap corresponds to
= (k = 0) = h (k = 0) 0 (k = 0) = h(k = 0) 0
(166)
398
(We omit in our discussion the wave function

with renormalized Yukawa coupling h = h / A.
renormalization for the fermions.) We observe that a gap does not necessarily require a non
zero order parameter 0 . A non-vanishing renormalized order parameter 0 > 0 is sufficient,
provided the renormalized Yukawa coupling remains also non-vanishing for k 0.
Furthermore, one may encounter situations where (k) differs from zero for some range
k > kSR , while (k < kSR ) = 0. For small enough kSR this is a typical situation for a pseudogap. Even though a true gap is absent ((
q = 0) = 0), the momentum dependent gap (
q ) is
non-zero for q = 0. Except for a very narrow region around the Fermi surface the occupation
numbers nF (
q ) behave as if a gap were present. Such a situation has been discussed in detail for
antiferromagnetic order in the two-dimensional Hubbard model [14]. A pseudogap occurs for
temperatures between the critical temperature Tc and the pseudocritical temperature Tpc > Tc .
The presence of a pseudogap (k) simply reflects the existence of local order in domains with
linear size k 1 .
Similar considerations
hold for bosonic occupation numbers. For non-relativistic bosons we
may interprete r = 20 as a gap for the radial mode, while the Goldstone boson remains
gapless. More complex gaps may occur if different bosonic degrees of freedom play a role.
A gap for the bosons also exists in the symmetric phase where the effective potential U () has a
minimum at 0 = 0, with m2 = U/|0 > 0. In this case one replaces in Eq. (15) m2 .
13. Conclusions
In this paper we have concentrated on the more formal aspects of the functional integral approach to occupation numbers n(
q ) for interacting non-relativistic particles. Our study of bosons
at zero temperature reveals pitfalls for computations beyond leading order perturbation theory
the Bogoliubov approximation. They are connected to the infrared and ultraviolet behavior of
n(
q ). If insufficient care is taken, no meaningful result for the particle density n will be obtained.
This is a very practical problem for precision computations for ultracold atoms, where the relation between the particle density and the chemical potential is a crucial quantity. One of the
potential problems arises from an unphysical dependence of the relation between density and
chemical potential, n(), on the ultraviolet cutoff .
A second class of problems concerns the infrared behavior of the occupation numbers. For
T = 0 we have found that the behavior of n(
q ) for q 0 differs qualitatively from the Bogoliubov theory for d = 1, 2 and mildly for d = 3. This is due to the dominance of a term quadratic
in the frequency in the full inverse propagator. Denoting with V the corresponding coupling we
find
n(
q ) = 02 (
q ) + np (
q ),
np (
q ) V 1/2 q (1) .
(167)
The anomalous dimension relates to the dense regime or Goldstone regime [6]. It is positive
for d = 1 and vanishes
for d = 2, 3. As one of the consequences we find that the condensate

depletion np = q np (
q ) is not dominated by the extreme infrared modes with q 0. This allows
for a clear conceptual separation between the condensate density nc = 02 and np . For d = 3
the asymptotic behavior (167) sets in only for extremely small q, such that for most practical
purposes the Bogoliubov approximation remains valid.
For practical calculations with a good accuracy we propose to use the differential equation
(69) for the q-dependence of np (
q ). Correspondingly, np can be computed from Eq. (70). The
benefit of such a calculation is the absence of infrared or ultraviolet problems and the renor-
399
malization group improvement due to the use of running couplings. We postpone a numerical
solution of Eqs. (69), (70) to future work.
We have derived an exact functional renormalization group equation for scale-dependent occupation numbers. They can be employed to motivate a renormalization group improvement in
simpler settings. The effective couplings appearing in the evaluation of the occupation number
n(
q ) for a particular momentum q correspond to running couplings, evaluated at a renormalization scale k = q (with modifications in presence of a Fermi surface). This allows for an easy
description of pseudogaps.
Appendix A. One loop calculation of Z
We define the wave function renormalization Z (, q) by the -derivative of the 21element of the inverse propagator
Z (, q) = i

P21 .
(A.1)
For non-zero 0 the one loop contribution to P21 has a contribution from effective cubic couplings 0

11 (p0 , p)
21 (p0 + i, p + q) + G
12 (p0 i, p q)
3G
P21 (, q) = 2 02
G
p0 p

12 (p0 + i, p + q) + G
21 (p0 i, p q)
22 (p0 , p)
G
+G

2

p
= 2i2 02
02 det1 (p) det1 (p + q) + det1 (p q) ,
2M
p0 p
(A.2)
with

1
(p + q)2 (p + q)2 + 4M 02 + (p0 + i)2 .
2
4M
We first consider the special case q = 0

2

p

2 2
2
0
J (p,
) + J (p,
) ,
Z (, 0) = 2 0
2M
det(p + q) =
(A.3)
(A.4)
p
where
J (p,
) =
det1 (p) det1 (p + q)
p0
p02 + B
1
(p0 + i)2 + B
1
p0

1 2 2
p p + 4M 02 .
B(p)
=
2
4M
(A.5)
The integrand of the p0 -integration has simple poles at p0 = i B and p0 = i B i

and
we take 0 without loss of generality. The first pole in the upper half plane at p0,1 = i B
400
has residuum
i
,
r1 =
2 B(2 B + )
(A.6)
i
.
r2 =
2 B(2 B )
(A.7)
whereas a second pole in the upper half plane at p0,2 = i( B ) exists only for < B, with
residuum at
This yields

1
1
1
J ( > 0) =

B
2 B 2 B + 2 B
or, for arbitrary ,

K() = J () + J ()
for || < B,
2 2
B(4B )
=
sign()
for || > B.
(A.8)
(A.9)
B(2 B+||)
Both K and K/ are discontinuous at || = B.

At first sight the situation seems to simplify for = 0 where
1
K
=
,

=0 2B B
(A.10)
resulting in
Z (0, 0) = 2 02

p

p2
2
0 B 3/2 (p).

2M
(A.11)

With = 4M 02 we have to solve the remaining integral over y = |p|
Z (0, 0) = 4vd M 2 2 02 I (),

I () =
dy y d4

2
1/2 3 2
3/2
y +
y +

1/2
= 1 + 3
dy y d4 y 2 +
0

d4
3
d 5 2 Fd ,
=
2

Fd =
0

1/2
dz zd4 z2 + 1
.
(A.12)
401
This yields
Z (0, 0) = 2d3 (10 3d)vd Fd ( M)d/2 0d2 .
(A.13)
However, the integral Fd is infrared finite only for d > 3 and ultraviolet finite for d < 4. For
the realistic cases with d 3 the limit
0 has to be done more carefully. Even for very small
|| Eq. (A.10) can only be used for B > || or
z>
2M||
.
(A.14)
This modifies effectively the factor Fd in Eq. (A.13), which now becomes a function of /.
The contribution to Fd from this range is approximately
1 2M
(
)d3 for d < 3,
+
Fd = 3d
(A.15)
for d = 3
ln 2M
3/2
and diverges for 0. In consequence, one has Z (, 0) 0 for d 3 with logarithmic
corrections for d = 3. The range z < 2M/ (for
> 0) gives an infrared finite contribution
for d > 1 since for z 0 one has K/ 1/(2 B) and the r.h.s. of Eq. (A.11) is multiplied by a factor 2B/2 z2 2 /(2M 2 2 ). For d = 1 the one loop expression for Z (, 0)
remains infrared divergent even for > 0. In any case, the negative one loop correction (A.13)
drives Z (, 0) rapidly towards zero as decreases, making the use of renormalization group
improvement mandatory. This has been done in [6].
Let us now turn to the evaluation of Z (, q) for general q and . By suitable momentum
shifts Eq. (A.2) can be written as
P21 =

H=

i2 02
M
p
p 2M 02 +
2

q 2
H,
4

q
q
1
1
det
det
p+
p
2
2
(A.16)
p0
where

2
q
i
det p
= B + p0 ,
2
2

2

1
q
q 2
+
.
p
p
B =
2
2
4M 2
(A.17)
For the p0 -integration we need the poles which are situated at p0,1 = +i B+ 2i , p0,2 =
+i B + 2i , p0,3 = i B+ 2i , p0,4 = i B + 2i . For || < 2B one finds

H = i(r1 + r2 ),

2 1
ir1 = 2 B+ B B+
,

2 1
ir2 = 2 B B+ B +
.
For q = 0 one recovers the formulae discussed above.
(A.18)
402
For q = |
q | we can neglect such that

1
q 2 2
B =
.
p
2
4M 2
(A.19)
We may evaluate Eq. (A.18) for = 0

H 1 = 2 B+ B (B+ + B )

2 2
q

q 2 2
1
2
2
2
(pq)
+ (pq)2 .
p +
p +
=
4
4
2 2M 3
(A.20)
This yields
Z = 2
22 02 M 2
p +
2

q 2 2
4
q 2
p +
4

p

+ (pq)
q 2
p +
4
2
(pq)
1
1/2
.
(A.21)
The p-integral

is finite for d < 4 and dominated by values p2 q2 /4. As a result it is propord4
tional vd q
/(d 4) and we find an expression as in Eq. (47). Solving the integral explicitely
determines the coefficient in this equation.
Appendix B. Fermion boson coupling
In this appendix we briefly discuss the effects of fermion fluctuations on the bosonic occupation numbers in a coupled system as given by Eq. (153). For this purpose the effective action has
to be extended by a piece quadratic in the boson field and by suitable boson interactions. We
1 . We consider here an approximation where
denote the inverse boson propagator by P = G
P is given by a classical part and a part P induced by the fermion fluctuations

q 2
2 + P .
P = iq0 +
(B.1)
4M
(The boson mass is twice the fermion mass M and the chemical potential twice the fermion
chemical potential. This corresponds to di-atom molecules.) This inverse propagator has been
extensively studied in [11]. We are interested here in its value for analytically continued momenta
q0 = i and at T = 0. We want to show here that the fermion fluctuation contribution to the
boson (or molecule) occupation number nM (
q ) vanishes in absence of a gap = h 0 = h 0
(for T = 0).
Using the shorthands
= M( + 2),
q0 = 2Mq0 ,
q0 = Mq0
one finds
P (q0 + i, q)

1
h 2 M
q 2

=
d q0
+ i q0 + i q0
q
2
q
(B.2)

q
q +
2

2
+ i q0 i q0
1

( 0, q 0, q0 0) .
The q0 integration has poles at

q 2
q 2

q0 = i q
q0 , i q +
+ q0
2
2
and can be performed by closing the integral either in the upper or lower half plane
1

2
1
q2
2
2
P = h M
q 2M
+ i q0
q +
.
4
403
(B.3)
(B.4)
(B.5)
q
We concentrate here on d = 3 and 0

P =
h 2 M
4 2
dq

2M
A2
+
,
(q iA)(q + iA) (q i 2M)(q + i 2M)
with A2 = q 2 /4 + i q0 . We choose A to be the root with positive real part. (This always exists
for q0 = 0.) The q -integration yields
1/2

2

h 2 M 3/2
q
P =
(B.6)
2
+ iq0 2
4
4M
+
1/2
where ( )+ indicates the root with positive real part. We therefore obtain the propagator in the
molecule phase ( < 0)

1/2
1
h 2 M 3/2
q 2
q2
G(q0 , q) = iq0 +
(B.7)
(2)1/2
.
+
iq0 +
2
4M
4M
+
We want to show that the bosonic occupation number vanishes

q ) = G(q0 , q) = 0.
nM (
(B.8)
q0
q )/ h 2 . For this quantity we can close the integral on the

For this purpose we consider nM (
lower half plane. The integral is analytic in the lower half plane, implying nM (
q 2 )/ h 2 = 0.
Since for h 2 = 0 the theory is free we infer for all q2 > 0 that nM (
q ; h 2 = 0) = 0. Therefore
nM (
q ) vanishes for all h 2 .
For the atom phase one has = 0 and the bosonic propagator becomes

h 2 M 3/2
q 2
q 2 1/2 1
G(q0 , q) = iq0 +
(B.9)
.
+m
2 +
iq0 +
4M
4
4M +
The argument showing nM (
q ) = 0 proceeds similar as above.
In the limit (B.1) the boson occupation number vanishes for T = 0 as long as there is no
gap. A non-vanishing fermion contribution to the boson numbers only arises from modifications
of the fermion propagator in Eq. (B.3) due to a non-vanishing gap = h 0 . Of course, the
fermion fluctuation contribution to nM (
q ) is non-zero for T > 0 even in absence of a gap [10].
Furthermore, one has the contribution from boson fluctuations as discussed in the main part of
this paper.
404
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
N. Bogoliubov, J. Phys. (Moscow) 11 (1947) 23.

P. Weichmann, Phys. Rev. B 38 (1988) 8739.
N.D. Mermin, H. Wagner, Phys. Rev. Lett. 17 (1966) 1133.
G. Baym, J.-P. Blaizot, M. Holzmann, F. Lalo, D. Vautherin, Phys. Rev. Lett. 83 (1999) 1703;
G. Baym, J.-P. Blaizot, J. Zinn-Justin, Europhys. Lett. 49 (2000) 150;
S. Ledowski, N. Hasselmann, P. Kopietz, Phys. Rev. A 69 (2004) 061601(R);
J.-P. Blaizot, R. Mendez-Galain, N. Wschebor, Phys. Rev. E 74 (2006) 051116;
J.-P. Blaizot, R. Mendez-Galain, N. Wschebor, Phys. Rev. E 74 (2006) 051117.
C. Castellani, C. Di Castro, F. Pistolesi, G.C. Strinati, Phys. Rev. Lett. 78 (1997) 1612;
F. Pistolesi, C. Castellani, C. Di Castro, G.C. Strinati, Phys. Rev. B 69 (2004) 024513.
C. Wetterich, Phys. Rev. B 77 (2008) 064504.
C. Wetterich, Nucl. Phys. B 352 (1991) 529.
C. Wetterich, Phys. Lett. B 301 (1993) 90;
C. Wetterich, Z. Phys. C 57 (1993) 451.
J. Berges, N. Tetradis, C. Wetterich, Phys. Rep. 363 (2000) 223.
S. Diehl, C. Wetterich, Nucl. Phys. B 770 (2007) 206.
S. Diehl, H. Gies, J. Pawlowski, C. Wetterich, Phys. Rev. A 76 (2007) 021602(R).
D. Litim, Phys. Lett. B 486 (2000) 92.
S. Sachdev, Quantum Phase Transitions, Cambridge Univ. Press, Cambridge, 1999.
T. Baier, E. Bick, C. Wetterich, Phys. Lett. B 605 (2005) 144.

Double-horizon limit,
AdS geometry and entropy function
Hessamaddin Arfaei a,b , Reza Fareghbal a,b,
a Department of Physics, Sharif University of Technology, PO Box 11365-9161, Tehran, Iran
b Institute for Studies in Theoretical Physics and Mathematics (IPM), PO Box 19395-5531, Tehran, Iran
Received 31 October 2007; received in revised form 25 February 2008; accepted 18 April 2008
Abstract
We start from a generic metric which describes four-dimensional stationary black holes in an arbitrary
theory of gravity and show that the AdS2 part of the near-horizon geometry is a consequence of the doublehorizon limit and finiteness. We also show that the field configurations of the near horizon are determined
if the same conditions are applied to the equations of motion. This is done by showing that in the doublehorizon limit field equations at the horizon decouple from the bulk of the space. Solving these equations
gives the near-horizon field configurations. It is shown that these decoupled equations can be obtained from
an action derived from the original action by applying the double-horizon condition. Our results agree with
the entropy function method.
PACS: 04.60.Cf; 11.25.-w
Keywords: String theory; Black holes; Attractor mechanism; Entropy function
1. Introduction
It is well known that in a large class of the gravity theories coupled to a number of scalars
and gauge fields, the value of the scalars at the horizon of the extremal black holes is independent of the values at the large distance. This phenomenon called attractor mechanism, has
been first shown for the supersymmetric theories [13] and was later studied and proved for the
* Corresponding author. Tel.: +98 21 22280692; fax: +98 21 22280415.
E-mail addresses: arfaei@mail.ipm.ir (H. Arfaei), fareghbal@theory.ipm.ac.ir (R. Fareghbal).

doi:10.1016/j.nuclphysb.2008.04.017
406
H. Arfaei, R. Fareghbal / Nuclear Physics B 802 [FS] (2008) 405420
non-supersymmetric cases [4]. Application of the attractor mechanism in string theory and its
importance in the counting of the states of black holes is discussed in [5] and [6].
A remarkable progress in understanding of the attractor mechanism came with the works of
Sen who introduced the entropy function method [7,8]. He showed that not only scalar fields but
all parameters of the near horizon of an extremal black hole is fixed by extremizing a function,
called the entropy function, which is evaluated near the horizon. Sens method resulted in a
generalized attractor mechanism and was shown to work in a number of cases.
In the entropy function method, one starts with characterization of the extremal black holes
with their near-horizon geometry. It is taken to have an AdS2 factor. We would like to address
the question of deeper physical properties that leads to such geometric characterization in most
known cases. It is also of our interest to examine whether one can find similar method for the
non-extremal cases which do not have AdS2 space in their near-horizon geometry. There has been
several attempts to answer these questions [911].
In this paper we show that both questions can be approached from a single point of view
which is imposing finiteness on the physical quantities at the horizon of the double-horizon
black holes. For most of the known four-dimensional black hole solutions, extremality coincides
with the double-horizon limit, i.e. the radii of the event inner and outer horizons coincide. This
results in zero surface gravity on the horizon and zero temperature for the black hole. Moreover,
the singularities at the horizons are coordinate singularities and can be removed by coordinate
transformation, thus we expect that the scalars constructed by the metric are finite at the horizons.
The physical reason for the attractor mechanism is the infinite distance to the double horizon
that prevents the information to affect the horizon physics [12]. This distance from an arbitrary
point to a simple horizon is finite and hence allows the bulk information reach the horizon and
therefore blocks the attractor mechanism. The double-horizon property is essential in the divergence of the distance in the first and finiteness of the latter.
This work follows and completes the work done in [9]. In Section 2, we start from a generic
metric which describes four-dimensional stationary black holes. This form is completely general
and does not depend on any particular theory. Then we explore the consequences of the finiteness
of scalars constructed from the metric. The finiteness is justified since the singularities of the
horizon must be removable. We demand the finiteness of R, R R and R R and
find certain restrictions on the components of the metric. The double-horizon condition imposes
further restrictions. Using these restrictions and following the method of [13], we obtain the nearhorizon geometry of the stationary black holes which is the starting point of [14] in generalizing
the entropy function method to the rotating black holes. Our result clearly shows that the AdS
near-horizon geometry is a consequence of the finiteness at the horizon and the double-horizon
limit.
In Section 3 we use our results in [9] and the finiteness assumption and the double-horizon
condition on the equations of motion to cast the equations to a set of equations that are decoupled from the bulk. We do this by starting from the general ansatz for the metric which describes
both the extremal and non-extremal black holes. The decoupling occurs due to vanishing of all
r-derivative terms. First we show the decoupling of the dynamics in Einstein gravity coupled to a
number of scalars and Abelian gauge fields and then extend it to f (R) gravities. For f (R) gravities we use the technique of the equivalence of these theories with EinsteinScalar gravity [18].
These equations are non-linear ordinary differential equations with respect to . The solution(s)
to these equations specifies the field configuration(s) at the near horizon. However concluding
the attractor mechanism from this decoupling is possible if we prove that our equations have a
unique solution.
407
Our analysis also shows that if we do not apply the double-horizon limit, the field equations
at the horizon do not decouple from the bulk and thus we cannot expect an attractor mechanism
for the case of simple horizon. In particular we observe that it is not possible to reproduce the
dynamics on the simple horizon by extremizing a function which is only defined at the horizon.
Recently [1517] has found examples of non-extremal black holes for which the entropy
function method works. These results are not in contradiction with ours, since the original metric
which describes black holes in those cases are not the same as one we consider in this work.
Our method for dealing with the rotating black holes is the reduction of -coordinate in
KaluzaKlein style. Because of axisymmetry of the stationary black holes, parameters of the
metric do not have dependence. Moreover is a periodic coordinate, therefore similar to the
KaluzaKlein reduction, we can reduce this direction and find a three-dimensional gravity theory in which apart from the original scalar and gauge fields additional scalar and gauge fields
origination from components of four-dimensional metric and fields exist. The interesting point is
that the charge of the new gauge field is exactly the angular momentum of the four-dimensional
black hole. This technique simplifies the derivation and the form of the decoupled equations.
In Section 4 we establish the relation of the equations to the entropy function method. We
reproduce the decoupled equations by extremizing a function at the horizon. The difference with
entropy function method is that this function is calculated using the general form for the fields
of the theory and imposing the double-horizon and finiteness conditions without any particular
assumption on the geometry. Whereas in the standard entropy function method the AdS geometry
is part of the assumption.
This results show that the possibility of derivation of the equations of motion by extremizing
a function at the horizon, is a direct consequence of the double-horizon limit. We also investigate
the result of our assumed conditions on the Walds entropy formula. Following the method of [14]
we find the results previously obtained by the entropy function method. The authors of Ref. [19]
have used the zero surface gravity property of the extremal black holes to simplify the Wald
formula for static solutions. Their results are in agreement with our proposal that the decoupled
physics of the horizon is a consequence of the double-horizon limit.
2. Double-horizon limit and AdS geometry
We consider 4-dimensional stationary black holes with axial symmetry. They are described
by the generic metric
ds 2 = a(r, ) dt 2 +
b(r, ) 2
dr + e(r, ) d 2 + f (r, ) d 2 + 2c(r, ) dt d,
S(r)
(1)
where a(r, ), b(r, ), c(r, ), e(r, ) and f (r, ) are assumed to be regular functions. Event
horizons are located at r = rH where S(rH ) = 0. In writing this metric, it is assumed that event
horizons are Killing horizons too. This is a general form based on the assumed symmetries. At
this stage we consider only the symmetry which is a reflection of the topology of the horizon
in four dimension and no particular theory of the gravity is assumed in this part. Therefore the
result is applicable to a wide class of theories. We consider the cases that S(r) has non-zero roots
and exclude the naked singularities.
Let us first impose the finiteness assumption on the determinant of this metric. The determinant g is a coordinate dependent quantity. The quantity which is invariant and coordinate
independent is V = g d 4 x. If we assume that the volume of any finite neighborhood near
408
the horizon is finite then from finiteness of d 4 x we conclude that

minant is,
g=
b(r, )e(r, )(a(r, )f (r, ) + c(r, )2 )

.
S(r)
g is also finite. The deter-
(2)
Finiteness requires,
a(r, )f (r, ) + c(r, )2 = S(r)v(r, )
(3)
where v(r, ) is a regular function.

Finding a(r, ) from (3) and substituting it in (1) and redefining new regular functions, we
obtain,

2
S(r)
dr 2
ds 2 = A(r, )
(4)
dt 2 +
+ E(r, ) d 2 + F (r, ) d + C(r, ) dt .
B(r, )
S(r)
Using this metric we can calculate scalars such as R, R R and R R . Assuming
that the singularities of the horizons are coordinate singularities, we conclude that these scalars
are all finite there. However, computations of these scalars show that they include terms that have
S(r) factor in their denominator which will diverge unless the nominators also develop similar
factors. Investigation of all these terms shows that some have

B(r, ) and some
C(r, )
factor in their nominator. If these derivatives are proportional to S(r) then all of them become
finite at the horizons. Thus finiteness imposes the conditions,

(5)
= 0,
= 0.
C(r, )
B(r, )
r=rH
r=rH
This means that these two functions B and C must have the forms,
B(r, ) = S(r)B1 (r, ) + B2 (r),
(6)
C(r, ) = S(r)C1 (r, ) + C2 (r).
(7)
For our argument it is sufficient to take the first power of S(r) in the right-hand side. One can
absorb higher powers in the functions B1 (r, ) and C1 (r, ). The regularity of the derivatives of
B(r, ) and C(r, ) at the horizon do not allow that power of S(r) be less than one. It is obvious
that at the horizons B(r, ) and C(r, ) are constants and do not have -dependence. We show
their values at the outer horizon by B and C.
For black holes with the double-horizon where the inner and outer horizons coincide, we have
S(r) = (r rH )2
(8)
where rH is the radius of the horizons.

Using (7) and applying the double-horizon condition, it is not difficult to see that defined as

C(r, )
(9)
r
r=rH
is a constant.
Now we are ready to find the near-horizon geometry of the double-horizon black holes. First,
we expand the functions specifying the metric around rH in terms of r rH . Using (5) and (9),
409
we get

A
(r rH ),
r r=rH

B
B(r, ) B +
(r rH ),
r
A(r, ) A( ) +
r=rH
C(r, ) C + (r rH ),

E
(r rH ),
E(r, ) E( ) +
r r=rH

F
F (r, ) F ( ) +
(r rH ).
r
(10)
r=rH
are defined as [13];

The near-horizon coordinates (t, r , , )
B
B
= C
t,
r = rH + r ,
t
t=
(11)
in the limit 0.
Substituting (10) in (4) and using (11), we obtain the metric in terms of the near-horizon
coordinates,

d r 2
ds 2 = A( ) r 2 d t2 + 2 + E( ) d 2 + F ( )(d + r d t )2
(12)
r
where
=
B.
(13)
In this coordinate system which is suitable for the near horizon, g component of the metric has
only dependence. This results a reparametrization freedom which allows us to fix g up to a
constant by defining as
1
=

d E( )
(14)

d E( ).
(15)
where
1
=

0
It is clear that goes between 0 to . Near-horizon geometry finally takes the form

d r 2
ds 2 = A( ) r 2 d t2 + 2 + 2 d 2 + F ( )(d + r d t )2 .
r
(16)
The main point is that (16) is a result of the finiteness at the horizon and the double-horizon
condition. This result is general and does not depend on the action of the theory. It is also independent of the asymptotic behavior of the metric. Hence it can be applied even to metrics
embedded in non-asymptotically flat space. Eq. (16) is the starting point of [14] in generalizing
entropy function method for the rotating black holes. In [14] it is taken as the beginning point
410
for the rest of the argument but we have shown it as a result of physically simple assumptions.
This shows that AdS2 near-horizon geometry which is used in entropy function method as the
definition of the extremal black holes is a consequence of the finiteness and the double-horizon.
In the next sections we will show that in order to find the parameters of the near horizon it is
not necessary to go to the near-horizon geometry and apply entropy function method. Starting
from the metric (4) and writing the equations at the horizon and applying the double-horizon
condition, we can find the parameters of the near-horizon geometry directly.
Another point is that values of B and C at the horizon are absorbed in the definition of t
This means that there is a reparametrization freedom at the horizon and the values of B
and .
and C at the horizon cannot be determined by equations of motion and are not physical. We can
only find which is sufficient for specifying the near-horizon geometry. This will be clear in
the next sections where by using double-horizon condition we get a set of decoupled equations
at the horizon for A( ), F ( ), and .
3. At the horizon of double-horizon black holes
Analysis of the previous section shows the importance of the finiteness and the double-horizon
condition in finding the near-horizon geometry of the double-horizon black holes. Furthermore,
in this section we would like to show that these assumptions are enough to fix the field configuration of the near-horizon geometry. The interesting point here is that this fixing is possible without
directly using the near-horizon field configurations. It is done by applying double-horizon limit
on the field equations come from variation of a generic action and using generic ansatz for the
field content of the theory. The result applies to a large class of theories and is independent of the
details of the dynamics in the bulk.
We first consider rotating charged black holes in a theory of gravity with scalars I (I =
1, 2, . . .) and Abelian gauge fields A(K) (K = 1, 2, . . .) described by the action

1
(K) (L)
F
V () ,
S = 2 d 4 x g R hI J () I J wKL ()F
(17)
where wKL () determines the coupling of the scalars to the gauge fields and V () is a potential
term for the scalars. After analyzing this case we will consider more general theories. As shown
in the previous section the form of the metric is

2
S(r)
1
2
2
2
dt +
dr + E(r, ) d 2 + F (r, ) d + C(r, ) dt ,
d s = A(r, )
B(r, )
S(r)
(18)
where
S(r) = (r r+ )(r r )
(19)
r+ and r are the radii of the outer and inner horizons. It is clear that for the case of doublehorizon black holes i.e. r+ = r , not only S(r) but also dS(r)
dr vanishes at the horizons and for all
2
cases d drS(r)
2 = 2.
Using axial symmetry of the rotating stationary black holes, we choose following ansatz for
the scalars and gauge fields:
I = I (r, ),
A(K)
dx
(K)
= At (r, ) dt
(20)
(K)
+ A (r, ) d.
(21)
411
It is assumed that this black hole has angular momentum J , electric charges Q(K) and magnetic charges P (K) defined through the relation

(K)

(K)
(K)
P (K) = d d F = 2 A () A (0) .
(22)
As we saw in the previous section, requirement of the finiteness of the scalars constructed by
the metric imposes conditions (5) on the functions B(r, ) and C(r, ).
Now we demand that at the horizons, not only the scalars constructed from the metric but
also all the terms that appear in the action to be finite. It is possible
to consider (17) as the KK
reduction of a higher-dimensional pure gravity which only has g (d) R (d) term. The singularity
of the horizons in this higher-dimensional picture is a coordinate singularity too and therefor R (d)
must be finite. After the reduction, this term breaks to three terms of the 4d action (17) and thus
sum of these terms must be finite at the horizons. However, all of these terms must be separately
finite. The R term is finite since the singularity is coordinate singularity. The other two terms
are both positive and hence each of them must be separately finite. hI J and wKL are also finite
since are reduced from the higher-dimensional regular metric components. Finiteness of F F
requires that

(K)
(K)
Ft (r, ) C(r, )F (r, )r=r = 0.
(23)
H
Using (5) and (21) this condition takes the form:

(K)
(K)
At (r, ) C(r, )A (r, ) r=r = 0.
H
Therefore the left-hand side is proportional to S(r) and we can write

(K)
(K)
A (r, ) C(r, )A (r, ) = S(r)f (K) (r, )
t
(24)
(25)
where f (K) (r, ) is a regular function.1 By integrating both sides with respect to we obtain
(K)
At
(K)
(K)
(r, ) C(r, )A (r, ) = S(r)f1
(K)
(r, ) + f2
(r).
(26)
It is obvious from this relation that (K) s defined as

(K)
(K)
(27)
At (r, ) C(r, )A (r, ) r=r
H
r
are constant for the double-horizon black holes. It is not difficult to see that in the near-horizon
coordinate (11), we have
(K)
Fr t = B (K)
(28)
(K) =
thus our conditions guarantee that this component of the field strength is constant in the nearhorizon coordinate.
In dealing with the rotating black holes it is easier to apply a reduction similar to KK-reduction
in the -direction. The axisymmetry of the rotating black hole background guarantees that non of
the metric components depends on which allows such reduction. Since this direction is periodic
1 The power of S(r) in the right-hand side of (25) can be n (n 1). We absorb S(r)n1 in f (K) (r, ). This power
cannot be less than one since makes F F infinite.
412
with period of 2 , we can look at it as a compact direction with unit radius of compactification.
The four-dimensional rotating black hole will be converted to a non-rotating but charged black
object represented in this new three-dimensional picture. This black object is the source of the
new scalar fields and a new gauge fields. Angular momentum of the 4d rotating black hole will
be charge of the new 3d black object.
Let us define
M(r, ) F (r, )A(r, ),
N (r, ) F (r, )E(r, ).
(29)
Reducing the action (17) in the -direction and changing to Einstein frame, we obtain

2
(a) (b)
d 3 x g R H ij () i j Wab ()F
F
U () ,
S= 2
where the three-dimensional metric takes the form,

S(r)
dr 2
2
2
ds = M(r, )
dt +
+ N (r, ) d 2 .
B(r, )
S(r)
(30)
(31)
The scalars are given as

1
2 ln F (r, )
(K)
.
=
A
I (r, )
(32)
The first two components are from the metric component and -component of the gauge
potential.
The three-dimensional gauge potentials are

C(r, )
At =
(33)
(K)
(K)
At (r, ) C(r, )A (r, )
and the moduli metric is
2
0
2
wKL
H = 0 F (r,)
0
0
0
0 .
hI J
The gauge field couplings are given by

1
(K) (L)
F (r, ) + wKL A A
W = F (r, ) 4
(L)
wKL ()A
(34)
A(L) w LK
wKL

(35)
and the potential is

U () =
1
V ().
F (r, )
(36)
It is seen that only the t -component of the new gauge fields is non-zero. Using (5) and (24),
we see that at the horizon Ft is zero for the both gauge potentials. From the definitions (9)
and (27), we deduce that in the case of the double-horizon we have,

.
Frt |r=rH =
(37)
(K)
413
In the three-dimensional picture C(r, ) becomes a component of the new gauge filed. Because of the gauge freedom, it is not possible to determine this function by using the equations
of motion. This gauge fixing freedom of 3d picture is a consequence of the reparametrization
freedom of the original four-dimensional theory.
Another subtle point in the new 3d theory is that coordinate goes from 0 to and does not
cover the complete 3d space. In order to solve this problem, we use the axial symmetry of the
original solution and extend the new theory to the whole of the space by demanding that
X( ) = X(2 )
(38)
where X stands for any field of the 3d theory and covers the complete cycle between 0 to 2 .
In this form the three-dimensional theory is on the whole of R 3 . Requirement of the smoothness
of the solution at the poles i.e. = 0, implies,

X
= 0.
(39)
=0,
Now we want to impose the double-horizon condition on the equations of motion. Writing
equations of motion at the horizon is equivalent to removing all the terms which have S(r)
factor. Imposing the double-horizon condition is done by setting dS(r)
dr factor to zero. Variation
of the metric in (30) gives:
R H ij () i j

a
b
a
= Wab () 2g F
F
g F
F b + g U ().
Setting S(r) and
dS(r)
dr
(40)
to zero in this equation one obtains equations,
4N 2 ( ) + 2N ( )M ( ) N ( )M ( ) + 4N 2 ( )M( )U () = 0,
2

4M( )N ( ) + M ( )
(41)
= 2M 2 ( )H ij ()i j + 4N ( )Ve () 2N ( )M 2 ( )U (),
(42)

Ve () = B W11 2 + W1K (K) + WK1 (K) + WKL (K) (L)
(43)
where
and derivatives are respect to . Note that the R equations have resulted only the two above
equations. The equations of motion from the variation of the scalars are:

2
gH ij () j
g
=
Wab () a b U ()
H kj ()
k j +
F F
+
i
i
i
(44)
which at the horizon of the double-horizon black holes take the form:

M()
M() ij
H ()j
N ( )
N ( )
=
M 2 ( ) H j k () Ve () 1 2 U ()
j k
+ M ( )
.
2N ( ) i
i
2
i
(45)
414
Finally the variation of the gauge fields gives;

gWab ()F b = 0.
(46)
The only non-trivial equation in (46) is when = t . Using (33) in this equation and integrating
with respect to in the interval [0, ] gives;

rr tt
b
(47)
d gg g Wab ()Frt + gg g tt Wab ()Ftb 0 = 0.
r
0
Using (39), Ft = 0 at = 0, , thus the second term vanishes. The first term introduces a
constant which for a = 1 is proportional to the angular momentum and for a = K + 1 to the
charge Q(K) of the 4-dimensional black hole. At the horizon it takes the form,

B d

N ( )
W11 + W1K (K) = 2J,
M()
(48)

N ( )
B d
WK1 + WKL (L) = 2Q(K) ,
M()
(49)
where J and Q(K) are the angular momentum and charges of the black hole. The coefficients of
proportionality at the right-hand sides are fixed by use of the known KerrNewman solution.
Defining

,
= B
(50)
(K)
Eqs. (41), (42), (45), (48) and (49) take the following compact forms,
4N 2 ( ) + 2N ( )M ( ) N ( )M ( ) + 4N 2 ( )M( )U () = 0,
2
N ( )
M ( )
N ( )
= 2( )T H + 4 2 T W 2N ( )U (),
+
4
M()
M()
M ( )

N ( ) M()
(H )i
M()
N ( )
H
W
1
N ( )
N ( ) U ()
= ( )T

()T
+
,
2
2
i
i
2
i
M()

N ( )
d
W = 2Q,
M()
(51)
(52)
(53)
(54)
where
Q=
J
Q(K)

.
(55)
Eqs. (51)(54) are a set of differential equations which are decoupled from the bulk in the sense
that they do not have any r-derivative term. They involve only functions of angular variables
and respective derivatives. Solving these equations provides the information only on the horizon
with no reference to the bulk of the space and in particular asymptotic values of the fields. This
415
guarantees the decoupling of the dynamics of the horizon from the bulk. If these equations have
unique solutions, this decoupling will prove the attractor mechanism in its strong form. It implies
that the field configurations on the horizon are determined uniquely by these equations and the
behavior at the infinity dose not enter in specifying their values. If the equations admit more than
one solution, the problem needs further analysis and one must explore which class of asymptotic
conditions corresponds to a particular solution at the horizon [4].
It is notable that solutions of these equations determine M(), N ( ), ( ) and which are
required for specifying the near-horizon geometry (16). We find them without using directly the
equations in the near horizon. The boundary values for solving these equations are given by (38)
and (39).
Our analysis clearly shows that this decoupling occurs for the double-horizon black holes and
for distinct-horizon cases there is not such a decoupling. If we write the field equations for the
black holes with the distinct horizons, a number of terms with r-derivative factors will survive.
The presence of these terms obstruct the decoupling of the equations and the behavior of the fields
will depend on the boundary conditions at infinity. One expects that as the distance between the
inner and outer horizon decreases and the black hole approaches the double-horizon case, the
decoupling violating terms become smaller and correspondingly less important.
3.1. Generalization to f (R) gravities
Our methodand argument can be simply generalized
to include f (R) gravities. In these theories the term GR in the action is replaced by the Gf (R) resulting in the action,

1
I = 2 d 4 x Gf (R).
(56)
The other terms of the action remain unchanged and therefore we only study the effect of this
term on the above considerations. General analysis of Section 2 which is independent of any
particular action shows that the black hole solutions of this theory also have the same form
as (18).
The simplest way of studying f (R) gravities is using their equivalence to Einstein gravity
coupled to a scalar field. If we define by

3 2 df (R)
ln
=
(57)
2
dR
and make a conformal transformation,

2 df (R)
2

g =
G = exp G .
dR
3
The field equations derived from (56) are equivalent to those derived from the action

1
S = 2 d 4 x g R 2g V ( )
(58)
(59)
where

4
df (R)
V = exp
R
f (R)
dR
3
(60)
416
and

=
if
1 if
df (R)
dR
df (R)
dR
> 0,
(61)
< 0.
It is seen that V ( ) is the Legendre transformation of f (R).

Using (18) and (58) we can write g as

) S(r) dt 2 + 1 dr 2 + E(r,
) d 2 + F (r, ) d + C(r, ) dt 2
d s 2 = A(r,
B(r, )
S(r)
(62)
where

2
) = exp
A(r, ),
A(r,
3

2
F (r, ) = exp F (r, ).
3

2
) = exp
E(r,
E(r, ),
3
Hence the problem is reduced to the previous case and we can repeat our method.
Reduction of this theory in -direction results a theory with the action

2
d 3 x g R 2g 2g w()F F
S = 2

U (, ) ,
(63)
(64)
where
1
w() = e4 ,
4
U (, ) = e2 V ( )
and
(65)
(66)

S(r)
1
d s 2 = M(r, )
dt 2 +
dr 2 + N (r, ) d 2 ,
B(r, )
S(r)
2
e
= F (r, ),
(68)
At = C(r, ),
(69)
),
M(r, ) = F (r, )A(r,
).
N (r, ) = F (r, )E(r,
(70)
(67)
with
Field equations from variation of (64) with respect to the metric, the scalars and the gauge
filed are given by
R 2 2

= w() 2F F g F F + g U (, ),

1 w()
1 U (, )
1
g
=
F F +
,
4
(71)
(72)

1 U (, )
g
=
,
4
g

gw()F
= 0.
417
(73)
(74)
At the horizon of the double-horizon black holes these equations take the forms,
4N 2 ( ) + 2N ( )M ( ) N ( )M ( ) = 4N 2 ( )M( )U (, ),
2

4M( )N ( ) + M ( ) = 4M 2 ( )( )2 + 4M 2 ( )( )2 + 4 2 N ( )w()

2N ( )M 2 ( )U (, ),
w() 1 2 U (, )
1
= 2
+ M ( )
,
2
M()
M()
N ( )
N ( )

M()
M() 1 2 U (, )
= M ( )
N ( )
N ( )

N ( )
d
w() = 2J.
M()
(75)
(76)
(77)
(78)
(79)
The derivatives are with respect to and we have defined
= B.
(80)
Again these equations are a set of decoupled equations from the bulk. Solving them gives
M(), N ( ), (), () and which determine the near-horizon geometry of the doublehorizon black holes in f (R) gravity.
We note that a special case is when f (R) = R + where is a constant. In this case the black
hole is asymptotically AdS. Hence our analysis is valid even in the presence of the cosmological
constant.
4. Double-horizon limit and entropy function method
Our analysis in the previous sections clearly shows the decoupling of the dynamics at the
horizon from the bulk. This decoupling occurs at the level of the equations of motion, but an
interesting question is to see what happens at the level of the action. We consider this point in
this section.
Let us consider charged rotating black holes in the theory of gravity which is described by the
action (17). The form of the metric, gauge field and scalar field are given by (18), (20) and (21).
A proper way to deal with this theory is the reduction in -direction and studying the new threedimensional theory which is described by (30)(35). At the horizon of the double-horizon black
holes, field equations derived from the action of this new theory take the forms (51)(54).
We define
a
F b U ()
L = R Hij () i j Wab ()F
thus from (30) it is seen that

1
d 3 x gL
S=
8
where we have chosen = 16 .
(81)
(82)
418
By using the double-horizon condition, at the horizon we have

l() L|r=rH =
1
2N 2 ( )M 2 ( )
2

N ( ) M ( ) + 2M()N ( )M ( )
4N ( )M( )M ( ) 4N 2 ( )M( )

2N ( )M 2 ( )( )T H + 4N 2 ( ) T W U ().
(83)
It is obvious that l() does not have any r-derivative term and it is decoupled. Due to this decoupled property of l(), Eqs. (51)(54) are given respectively as
f
= 0,
M()
f
= 0,
N()
f
= 0,
i
f
= Qi ,
i
(84)
(85)
(86)
(87)
where
1
d gl( ).
(88)
8
Generating equations of motion by extremizing a function which is defined at the horizon, is similar to the Entropy Function Method which determines parameters of the near-horizon geometry
[7,8]. The basic point here is that in the our case this extremization is a direct consequence of the
double-horizon limit. It is not difficult to see that F which is defined as
f=
F = i Qi f
(89)
is exactly the entropy function if we started from the near-horizon geometry.

Our analysis shows that why we cannot provide a method like the entropy function method
for the black holes with distinct horizons. For these cases it is not possible to cancel r-derivative
terms of the action at the horizon and thus we do not have a decoupled action which gives the
equations of motion.
Using Eqs. (84)(87) we can simplify Walds entropy formula. We follow the method of [14].
The entropy of the black hole is given by
L(2)
SBH = 8 h (2) hrr htt
Rrtrt
(90)
where h with , = r, t is a two-dimensional metric defined as

h
1
=
2

d sin g
0
(91)
and hL(2) is the two-dimensional Lagrangian density, related to the three-dimensional Lagrangian density via the formula:

hL(2) = d gL.
(92)
419
It follows from (31) and (91) that at the horizon, after imposing double-horizon limit we have

(2)
(93)
hrr htt = BRrtrt
thus we can express (90) at the horizon as

L(2) (2)
SBH = 8 B h (2) Rrtrt .
Rrtrt
Following the method of [14], one can use Eqs. (84)(87) to simplify (94) and obtain

SBH = 2 B T Q f ,
(94)
(95)
where f is evaluated at the

extremized values. This is the same result as the entropy function
method. The extra factor B is a result of this point that we did not use near-horizon geometry.
We can cancel it in the first term by the definition of and in the second term by 1 factor of
B
g, hence it does not enter in the calculations.

5. Conclusion
The analysis provided in this paper has two aspects. First it gives a deeper physical reason for
AdS part of the near-horizon geometry of the double-horizon (extremal) black holes which has
been the beginning point of the entropy function method. Second it puts our earlier result about
the decoupling of the dynamics of the horizon for double-horizon black holes on a firm ground.
It also opens venues for further investigation of its properties.
Our analysis also clarify that why black holes with distinct horizons do not enjoy a decoupling
or attractor mechanism. The nature of approach to the decoupling limit is also of interest. The
decoupling given as a set of equations on the compact sphere of the horizon which can be solved
consistently. If it has a unique solution then the attractor mechanism works like the non-rotating
case. Even if the solution is not unique we expect it to result in a discrete set of solutions closely
related to the minima of the potential in the non-rotating case. The relation of different solutions
of such discrete set to the large distance boundary condition and quantum transition between
these solution are not clear yet.
Solution to the set of the decoupled equations of the horizon provides sufficient information
for the physical properties of the black hole in particular the entropy. Hence we may find a way
to understanding of the fact that all the information hidden in a black hole is distributed on the
surface of the horizon.
The other interesting direction for further investigation is to generalize this method to the
higher dimensions where the topology of the horizon is more complicated than a simple sphere.
These questions and other unclear properties of the double-horizon limit is under investigation.
Acknowledgements
We would like to thank M. Alishahiha, M.M. Sheikh Jabbari, A.E. Mosaffa, A.R. Tavanfar and
A. Ghodsi for their useful comments and discussions. We would also like to thank the Iranian
chapter of TWAS for partial support.
420
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
S. Ferrara, R. Kallosh, A. Strominger, Phys. Rev. D 52 (1995) 5412, hep-th/9508072.

A. Strominger, Phys. Lett. B 383 (1996) 39, hep-th/9602111.
S. Ferrara, R. Kallosh, Phys. Rev. D 54 (1996) 1514, hep-th/9602136.
K. Goldstein, N. Iizuka, R.P. Jena, S.P. Trivedi, Phys. Rev. D 72 (2005) 124021, hep-th/0507096.
D. Astefanesei, K. Goldstein, S. Mahapatra, hep-th/0611140.
A. Dabholkar, A. Sen, S.P. Trivedi, JHEP 0701 (2007) 096, hep-th/0611143.
A. Sen, JHEP 0509 (2005) 038, hep-th/0506177.
A. Sen, JHEP 0603 (2006) 008, hep-th/0508042.
H. Arfaei, R. Fareghbal, JHEP 0701 (2007) 060, hep-th/0608222.
H.K. Kunduri, J. Lucietti, H.S. Reall, Class. Quantum Grav. 24 (2007) 4169, arXiv: 0705.4214 [hep-th].
D. Astefanesei, H. Yavartanoo, arXiv: 0706.1847 [hep-th].
R. Kallosh, N. Sivanandam, M. Soroush, JHEP 0603 (2006) 060, hep-th/0602005.
J.M. Bardeen, G.T. Horowitz, Phys. Rev. D 60 (1999) 104030, hep-th/9905099.
D. Astefanesei, K. Goldstein, R.P. Jena, A. Sen, S.P. Trivedi, JHEP 0610 (2006) 058, hep-th/0606244.
R.G. Cai, D.W. Pang, JHEP 0705 (2007) 023, hep-th/0701158.
M.R. Garousi, A. Ghodsi, JHEP 0705 (2007) 043, hep-th/0703260.
M.R. Garousi, A. Ghodsi, arXiv: 0705.2149 [hep-th].
K.i. Maeda, Phys. Rev. D 39 (1989) 3159.
R.G. Cai, L.M. Cao, Phys. Rev. D 76 (2007) 064010, arXiv: 0704.1239 [hep-th].

Phase transition properties of 3D Potts models

Alexei Bazavov a , Bernd A. Berg b,c, , Santosh Dubey b,c
a Department of Physics, University of Arizona, Tucson, AZ 85721, USA
b Department of Physics, Florida State University, Tallahassee, FL 32306, USA
c School of Computational Science, Florida State University, Tallahassee, FL 32306, USA
Received 13 April 2008; accepted 22 April 2008

Abstract
Using multicanonical Metropolis simulations we estimate phase transition properties of 3D Potts models for q = 4 to 10: The transition temperatures, latent heats, entropy gaps, normalized entropies at the
disordered and ordered endpoints, interfacial tensions, and spinodal endpoints.
1. Introduction
Potts models were introduced as a footnote in the thesis by Potts [1] on clock models, whose
investigation had been proposed to him by his adviser Domb. We define their energy function by
E (k) = 2
1
ij

(k) (k)
,
qi , q j
(1)
where (qi , qj ) is the Kronecker delta function, the sum ij is over the nearest neighbor lattice
sites, the index (k) refers to the configuration, and qi(k) is the Potts spin at site i. For the q-state
model Potts spins take the values 0, 1, . . . , q 1. Following the notation of Ref. [2] the factor of
two in front of the sum is introduced to match for q = 2 on Ising model conventions. Expectation
values of the Gibbs canonical ensemble are calculated with the Boltzmann (or Gibbs) factor
exp(E), = 1/(kT ).
* Corresponding author at: Department of Physics, Florida State University, Tallahassee, FL 32306, USA.
E-mail address: berg@hep.fsu.edu (B.A. Berg).

doi:10.1016/j.nuclphysb.2008.04.020
422
A. Bazavov et al. / Nuclear Physics B 802 [FS] (2008) 421434
Potts, and not so much clock models, received considerable attention up to the day. Developments till 1982 were reviewed by Wu [3]. In 2D the models have second order phase transitions
for q = 2 to 4, and first order transitions for q 5. In a work by Baxter [4], critical temperatures,
latent heats and entropies were analytically derived. Equations for interface tensions followed
later [5] and played a role in the verification of the multicanonical approach [6] to Markov Chain
Monte Carlo (MCMC) simulations. In 3D the q = 2 Ising model phase transition is second and
the q 3 transitions are first order. The strength of the first order transitions, measured by their
latent heat, increases with q. The q = 5 transition in 2D and the q = 3 transition in 3D are weakly
first order.
Potts models proved amazingly versatile to grasp the essence of physically interesting situations, many of them are described in Ref. [3]. When they are generalized by introducing quenched
random variables as exchange constants, the Ising case of the EdwardsAnderson spin glass [7]
and quadrupolar Potts glasses [8] are obtained. Particular choices give anti-ferromagnets and the
fully frustrated Ising model [9].
The 3D q = 3 Potts model shares the center symmetry of SU(3) gauge theory [10] and plays
a role in our understanding of the QCD deconfining phase transition. This and other motivations
led to a number of numerical investigations of the 3D 3-state Potts model [1118]. However,
when we recently looked out for a strong first order transition in 3D, we found only a few papers
[1921] dealing with 3D Potts for q 4. The purpose of this paper is to fill this gap in the
literature for 3D Potts models up to q = 10. Thereby, we will closely follow the outline of a
previous investigation of the equilibrium statistical properties of the 3D 3-state Potts model by
two of the present authors [18].
Next, we briefly summarize our simulation method and give an overview of our assembled
statistics. In Section 3 we calculate and analyze transition temperatures, specific and latent heats.
Section 4 deals with entropy and energy across the phase transition, Section 5 with interface tensions and spinodal endpoints. A brief summary and conclusions are given in the final Section 6.
2. Simulation method and overview of data
We want to calculate expectation values O in the Gibbs canonical ensemble. They are given
by the ensemble average
O = O()
= O = Z 1
K
O(k) eE ,
(k)
(2)
k=1
where
Z = Z() =
K

k=1
eE
(k)
n(E)eE
(3)
is the partition function. The index (k), k = 1, . . . , K, labels the configurations (microstates) of
the system and E (k) is the (internal) energy of configuration (k). In the last equation n(E) is the
number of configurations with energy E.
We consider Potts models on cubic lattices of size L3 with periodic boundary conditions.
There are N = L3 Potts spins. Each microstate k defines a unique arrangement of Potts spins and
vice versa:
(k)
(k)
k = q 1 , . . . , qN .
(4)
As each Potts spin can take on q values, there are

nL (E) = q N ,
KL = ZL (0) =
423
(5)
microstates. Even for rather small numbers of L, K is a very large, so that one will not be able
to sum the partition function explicitly. Instead, one can use statistical methods.
MCMC simulations [2] are a suitable approach to evaluate equilibrium properties of the
canonical ensemble. Off the phase transition temperatures canonical Metropolis or heatbath simulations with weight factor exp(E) work sufficiently well, provided an ordered start is used
for simulations in the ordered phase, and (somewhat less important) a disordered start for simulations in the disordered phase. However, canonical simulations deteriorate quickly when it
comes to the (most interesting) investigation of phase transition properties. For first order transitions the reason is that the relevant transition states are in the canonical ensemble suppressed
exp(2od LD1 ), where od is the orderdisorder interface tension.
Multicanonical simulations [2,6] are a remedy for this supercritical slowing down. One performs for Emin E Emax MCMC simulations with a working approximation of the weight
factors
1
wmuca (E) = eb(E)E+a(E) =
(6)
,
n(E)
supplemented by the canonical weights exp(max E) for E < Emin and exp(min E) for
E > Emax . Here n(E) is the number of states with energy E as introduced by Eq. (3). With the
weights (6) all energies in the range Emin E Emax are sampled with the same probability, so
that the Markov process will perform some kind of random walk in the range Emin E Emax .
The microcanonical inverse temperature b(E) and the dimensionless free energy a(E) follow
uniquely from n(E) due to the relation

a(E ) = a(E) + b(E ) b(E) E,

a(Emax ) = 0,
(7)
where is the step to the next energy. However, n(E) is a priori unknown, so that some iterative
procedure needs to be used. In this context working approximation of the weights means that any
weights that enable cycling (also called tunneling)
Emax Emin
and back
(8)
are considered to be acceptable. Actually it is known that not the weights (6), but some modifications of them, can be optimal for that purpose [22,23] and it should be noted that there is a
residual exponential slowing down [24].
We find working estimates of the multicanonical weights by finite size (FS) extrapolations
from a smaller lattice to a next larger lattice, a method which was already used in [6]. Obviously, this requires that the FS behavior of the system under consideration is well defined. In
particular, for complex systems like spin glasses or proteins this is not the case and more sophisticated recursion approaches need to be used [2], most noted is presumably the one by Wang and
Landau [25].
Let e = E/N = E/L3 be the energy density. For our Potts models we simply convert the
microcanonical inverse temperature bL1 (E) for a given lattice size L1 to the bL2 (E) of our next
larger lattice size L2 > L1 via the interpolation
bL2 (E) = w+ bL1 (E+ ) + w bL1 (E )
with w =
|e e |
,
e + e
(9)
424
Table 1
Analyzed statistics per production run in sweeps: 32 the number given
L
q =4
q =5
q =6
q =7
q =8
q =9
q = 10
max =
2
3
4
6
8
10
12
14
16
18
20
22
24
26
28
30
0.45
256
1024
1024
4096
4096
16384
65536
262144
262144
1048576
4194304
45 105
9 106
14 106
15 106
0.41
1024
2048
2048
8192
32768
10 104
25 104
45 104
85 104
13 105
19 105
39 105
1 107
3 107
0.43
6 103
9 103
9 103
1 104
4 104
16 104
3 105
5 105
15 105
21 105
30 105
1 107
2 107
0.45
10 103
10 103
10 103
15 103
45 103
25 104
45 104
14 105
20 105
29 105
50 105
1 107
3 107
0.47
15 103
15 103
20 103
20 103
5 104
3 105
55 104
17 105
26 105
70 105
14 106
3 107
0.49
15 103
15 103
20 103
30 103
75 103
4 105
7 105
22 105
32 105
2 107
8 107
0.50
16384
4000
4000
65536
262144
1048576
4194304
16777216
2 107
4 107
8 107
where e = E/(L2 )3 and e < e and e+ > e are the values closest to e, so that we have for the
corresponding E+ and E values entries in bL1 (E). For our purposes this simple procedure
turned out to be sufficiently accurate. Better extrapolations can be expected by taking into account details of the shape of nL (E), as for a magnetic field driven phase transition discussed in
Ref. [26].
Suitably, the Emax , Emin , min and max parameters, which accompany the weights (6) are
chosen so that they embrace the phase transitions and that
Emax = Emin ,
Emin = Emax
(10)
hold. From a simulation with these weights, canonical expectation values are obtained by
reweighting for the temperature range min max . This property has coined the name
multicanonical. To calculate the partition function (3), from which the normalized entropy and
free energy follow, one has to include = 0 in this temperature range. So we choose min = 0,
for which our normalization of the energy (1) implies Emax = 0. For max we chose the values
given in Table 1, each of them well above the transition value c . Our simulations of multicanonical ensembles defined by the weights (6) rely on the Metropolis algorithm. We update
sequentially with one Metropolis update per spin during one sweep through the lattice. This is
more efficient [2] than picking spins at random for the updates.
In Table 1 we give an overview of the statistics per run. We followed the outline of Potts
model MCMC simulations in [2]. First, we performed the number of sweeps listed in Table 1
for reaching equilibrium. Data from these sweeps are excluded from the statistics for which
measurements were performed. Subsequently, we collected for each run 32 histograms, each
relying on the number of sweeps listed in the table. All error bars are then calculated with respect
to these 32 bins (32 jackknife bins when nonlinear operations on the data are involved). From the
student distribution it is known that error bars from 32 independent Gaussian data give almost
Gaussian confidence probabilities at the level of two standard deviations.
425
Table 2
Number of cycling events for the first production runs
L = 2:
L = 3:
L = 4:
L = 6:
L = 8:
L = 10:
L = 12:
L = 14:
L = 16:
L = 18:
L = 20:
L = 22:
L = 24:
L = 26:
L = 28:
L = 30:
q =4
q =5
q =6
q =7
q =8
q =9
q = 10
3
32
109
68
21
20
77
173
125
353
711
441
903
867
356
6
135
139
117
125
186
249
279
313
322
327
84
43
639
4
350
377
80
106
190
195
178
301
228
172
1
116
2
246
358
98
76
189
169
242
139
69
48
27
27
1
239
550
89
71
122
99
102
47
38
11
6
1
104
531
87
92
109
52
42
12
10
5
1
15
88
164
113
135
107
86
11
3
2
Table 3
Number of cycling events for the second production runs
L = 2:
L = 3:
L = 4:
L = 6:
L = 8:
L = 10:
L = 12:
L = 14:
L = 16:
L = 18:
L = 20:
L = 22:
L = 24:
L = 26:
L = 28:
L = 30:
q =4
q =5
q =6
q =7
q =8
q =9
q = 10
139
288
72
69
22
42
93
209
130
323
921
521
986
869
1451
378
126
100
154
180
256
251
308
301
307
321
460
5918
1275
410
109
125
223
221
219
366
347
329
487
708
7091
1167
331
127
117
295
228
427
396
342
358
405
910
2614
1436
578
117
105
243
218
384
352
585
590
779
7985
1209
497
165
110
273
246
394
359
1056
2168
6339
1098
324
311
349
598
1142
2423
1542
1707
1591
For most data points we performed two runs with the statistics of Table 1. The first runs are
based on the weights iterated from the closest smaller lattice. These data are taken to refine the
weights for the lattices at hand. The refined weights are used for the second production runs
on these lattices. Exceptions from this procedure are iterations from a smaller to a larger lattice
immediately after the first run. This speeds up the process of getting to larger lattices and has
often been done when the cycling frequency of the first run was already satisfactory.
Table 2 collects the number of cycling (8) events obtained in the first production runs. For
L = 2, some cycling is already achieved by a canonical simulation at = 0. This allows one
to determine multicanonical weights for the second run on 23 lattices, which have large cycling
rates as shown in Table 3, and to start off FS iterations of the weights. From the 23 lattices
426
Fig. 1. Internal energies.
we extrapolate weights for the first runs on 33 lattices, refine them for the second runs on 33
lattices, iterate to the next larger lattice, and so on (up to the before mentioned exceptions). The
calculations were carried out on PC clusters at FSU. Our present lattice sizes are limited by the
deterioration of cycling with increasing L, the computational power of a single PC (here 23 GHz
per PC), and the limitation of the total length of one run to a few months.
During the simulations we collect histograms of the energy in the multicanonical ensemble,
hmu (E), and calculate functions of the energy f (E) from them by reweighting to the canonical
ensemble:

f (E)hmu (E) exp[E + b(E)E a(E)]
f = E
(11)
,
E hmu (E) exp[E + b(E)E a(E)]
where the sums are over all energy values for which hmu (E) has entries. The computer implementation of this equation requires care, because the differences between the largest and the
smallest numbers encountered in the exponents can be large. We rely here on the logarithmic
coding of Ref. [2]. Whenever the function f (E) is non-linear jackknife binning is employed.
3. Energies, transition temperatures and latent heats
In Fig. 1 we show internal energies as functions of , for each q from the largest available
lattice (error bars are not resolved on the scale of this figure). This gives a rough estimate of
the latent heats and the inverse transition temperatures t . Accurate results follow from FS extrapolations of indicators, which are defined on finite lattices, so that they converge in the limit
L (quickly) towards the infinite volume value of the desired physical quantity.
We calculate specific heats via the fluctuationdissipation theorem

()2 2
E E2 .
(12)
3
L
For first order phase transitions the finite volume specific heats are regularization of Dirac delta
functions, which are the infinite volume extrapolations. The multicanonical approach allows to
calculate specific heat values for a continuous range of values. Consequently, the locations of
the maxima can be accurately determined.
C=
427
Fig. 2. Transition temperature fits for q = 4.
Finite lattice indicators for the transition temperatures are called pseudo-transition temperatures, and there are various options to define them. On finite volumes their values differ, while
they all converge to the same L limit. We use here three definitions of pseudo-transition
1 , the value at which equal heights are achieved in the double peaked energy
temperatures: pt
2
3 , the location of the
histogram, pt , the position of the central energy of the latent heat, and pt
2
maximum of the specific heat. The first two definitions are explained in more detail later, pt
1
plays a role in determining the entropy gaps in Section 4 and pt for the interface tensions in
Section 5 (they are labeled in this order for consistency with Ref. [18]).
For q = 4 Fig. 2 shows our linear fits
i
pt
(L) = ti +
ci
,
L3
(13)
which determine the infinite volume transition temperatures. The vertical order of the fits agrees
i (L) differ, the infinite volume
with that in the legend. While the finite volume estimators pt
extrapolations are consistent with one another. The smallest lattices have been omitted from the
fits to ensure an acceptable goodness of fit Q [2] in each case. This q = 4 pattern repeats for
all q: The quality of the fits from our three definitions of pseudo-transitions temperatures are
similar, and the final estimates as well as their error bars are consistent with one another. To give
one best number for each q, we simply average over the three estimates. We average also their
error bars, because all three estimators rely on the same simulation, so that one does not expect
error bar reduction when averaging over them. The thus obtained transition values t (q) are
collected in Table 4. For the convenience of the reader we have included q = 3 and 2 estimates
from Refs. [18] and [27], respectively. So far we have not found a simple formula for the q
dependence like t = ln(1 + q )/2, which holds in 2D [4] (given here in our convention).
Following Ref. [28], we extract the latent heat e = E/N by fitting the maxima of the
specific heat to the form
Cmax (L) = a1 + a2 L3
(14)
428
Table 4
Estimates of observables (q = 2 from Ref. [27] and q = 3 from Ref. [18])
q
e
e(t )
e+
2
3
4
5
6
7
8
9
10
0.2216544 (06)
0.2752827 (29)
0.3143041 (17)
0.3447205 (12)
0.3697070 (15)
0.3909657 (17)
0.4094959 (23)
0.4259432 (23)
0.4407371 (18)
0
0.3286 (15)
1.16294 (61)
1.84619 (20)
2.36442 (17)
2.76430 (12)
3.08039 (15)
3.33628 (12)
3.547570 (87)
0.9957 (14)
1.3470 (74)
1.719 (36)
1.987 (20)
2.177 (26)
2.316 (22)
2.421 (32)
2.503 (32)
2.567 (25)
e(t )
1.1826 (73)
1.1367 (64)
1.063 (20)
0.995 (26)
0.934 (22)
0.881 (32)
0.835 (32)
0.794 (25)
e(t )
1.5112 (79)
2.3019 (64)
2.910 (20)
3.359 (26)
3.698 (22)
3.961 (32)
4.171 (32)
4.341 (25)
and using the relation

1
a2 = (t )2 (e)2 .
2
The results are also included in Table 4.
(15)
4. Entropy and energy across the phase transitions

The entropy density is
s = (e f ),
(16)
where f is the free energy density, which is continuous at the phase transition. So, the entropy
gaps across the phase transitions are
s = t e,
(17)
or s = 2 a2 with a2 from the fits (14). The entropy and energy density endpoints in the disordered (+) and ordered () phases are given by

s + = t e+ f (t ) ,
(18)

s = t e f (t ) ,
(19)
and are more difficult to compute than the gaps, because the additive normalization constants no
2 as given
longer drop out. We follow the method of Ref. [18], which relies on the definition of pt
below.
In multicanonical simulations, the normalization constant for the entropy is determined by the
known value at = 0:

S0 = ln q N
S0
(20)
= ln q.
N
Fig. 3 shows normalized entropy densities of our models for our largest lattices.
To calculate the endpoints of the entropy and energy on the ordered and disordered sides of
2 (L) by the relation
the transitions, we define pt
and
s0 =
2 1 + 2
2
= eL pt + eL
pt ,
eL pt
2
(21)
429
Fig. 3. Normalized entropy densities.
2 (L) for q = 4.
Fig. 4. Energy density histograms at pt
2
2 (L). For q = 4
where eL
(pt ) are the locations of the maxima of the double peak histogram at pt
these histograms are shown in Fig. 4 (we excluded the L = 30 lattice to keep a reasonable scale in
the figure). This construction ensures that the energy endpoints eL

are positioned symmetrically
2
about the central energy density eL (pt ):
2 1
eL
(22)
eL
= eL pt
2
and one finds that
2 1
sL = sL pt
(23)
sL
2
holds as well. As in [18] we use jackknife estimators and arrive at the values for e(t ), e+ and
e compiled in Table 4 and those for s(t ), f (t ), s + and s compiled in Table 5. Although
430
Table 5
Estimates of observables (q = 3 from Ref. [18])
q
s(t )
f (t )
s+
w+
2
3
4
5
6
7
8
9
10
0.55715 (31)
0.8491 (21)
0.983 (11)
1.0680 (70)
1.1324 (93)
1.1858 (85)
1.233 (13)
1.274 (14)
1.312 (11)
3.50956 (23)
4.431364 (50)
4.846358 (41)
5.084679 (34)
5.239933 (50)
5.349346 (60)
5.430753 (66)
5.493609 (42)
5.543856 (50)
s(t )
0.8943 (21)
1.166 (11)
1.3862 (70)
1.5695 (93)
1.7262 (85)
1.863 (13)
1.984 (14)
2.094 (11)
s(t )
0.8038 (22)
0.800 (11)
0.7498 (70)
0.6954 (93)
0.6455 (85)
0.602 (13)
0.563 (14)
0.530 (11)
1.75
2.45
3.21
4.00
4.80
5.62
6.44
7.27
8.11
1.75
2.24
2.23
2.12
2.00
1.91
1.83
1.76
1.70
Fig. 5. Equal heights energy density histograms q = 10.
there are simple relations between these values, we have to list them separately, because these
relations do not determine error bars. The estimates are correlated and the jackknife procedure
takes care of correct error bars. Also it should be noted that t in the arguments of e, s and f is
t2 , defined as the extrapolation of the pseudo-transition temperatures as defined by (21). These
values are consistent with the t values listed in Table 4, but not identical, as t of Table 4 is the
average of the extrapolations from our three definitions of pseudo-transition temperatures.
In the last two columns of Table 5 we give the first few digits of the effective number of states
on the disordered and ordered sides of the phase transition,

w = exp s .
(24)
Amazingly, the effective number of states per spin at the ordered endpoint goes down by increasing q.
5. Interface tensions and spinodal endpoints
For L the interface tension between ordered and disordered phases is [29]

1
Pmax (L)
2od (L) = 2 ln
,
Pmin (L)
L
(25)
431
Table 6
Estimates of interface tensions (yes/no refers to capillary waves, q = 3 from Ref. [18])
q
2od yes
2od no
2od yes
2od no
3
4
5
6
0.001806 (35)
0.0224 (11)
0.0632 (21)
0.1054 (50)
0.001602 (35)
0.0221 (13)
0.0628 (22)
0.1050 (45)
7
8
9
10
0.1484 (15)
0.1897 (56)
0.2308 (87)
0.2628 (40)
0.1478 (17)
0.1891 (56)
0.2302 (84)
0.2688 (47)
Fig. 6. Microcanonical temperature b(E) for q = 10.
where Pmax (L) represents the value of the maxima when the energy histogram is reweighted to
equal heights and Pmin (L) the minimum in between the peaks. For q = 10 we show our equal
heights histograms in Fig. 5. Including capillary waves [3032], we perform 2- and 3-parameter
fits to the form (compare Eq. (16) of [33])
ln(L)
c2
c3
= 2od + 2 + 3 .
(26)
2
2L
L
L
In case of the 2-parameter fits we set c3 = 0. While the 3-parameter fits are somewhat unstable,
consistent 2-parameter fits are limited to the largest three lattices. The differences between these
fits exhibit systematic errors, which show that larger lattices are needed for high precision results.
The results of the 2-parameter fits are compiled in Table 6, where the differences to the less stable
3-parameter fits are used to estimate systematic errors, which are, in these cases, substantially
larger than statistical errors of the fits.
We also include in Table 6 results from fits without the capillary wave contribution
ln(L)/(2L2 ). With the exception of the q = 3 case from Ref. [18], the difference between the
two fits is always smaller than the expected error from other sources.
The double peak histograms at first order phase transitions are intimately related to a Maxwell
construction [34,35] for the inverse microcanonical temperature b(E) defined by Eq. (6). For
q = 10 and two lattice sizes this is shown in Fig. 6. The areas above and below the infinite
volume line are identical and for L = 20 one sees that a small fraction of the curve joins this line.
sp
sp
The minimum min and the maximum max of the (E) curve are the inverse spinodal
temperatures. Equilibration at with dissipative model A (Glauber) dynamics [36] encounters
2od (L) +
432
Fig. 7. FS behavior of the spinodals.

sp
metastability in the range t < < max after a disordered start, whereas after an ordered start
sp
it encounters metastability in the range t > > min . For the (H ) of a magnetic field driven
phase transition [26] this would already be the entire metastability picture. In case of the temperature driven phase transitions of Potts model it is more complicated, because metastability after
sp
a disordered start persists for equilibration at > max due to orderorder domain walls, which
are for q = 3 investigated in Ref. [37].
sp
sp
In the past there may have been some hesitation in identifying min and max , as defined
here, with the spinodal endpoints. The reason is that their values agree in the infinite volume
limit with t [34,35] as is illustrated by the dotted line in Fig. 6. So the metastability disappears in the infinite volume limit, whereas the opposite is the case for the mean field spinodal,
which is introduced in many textbooks [38]. However, the recent finite volume analysis [39] of
KolmogorovJohnsonMehlAvrami (KJMA) theory demonstrates that the mean field approach
is a conceptually wrong starting point for describing the infinite volume limit of phase conversion. Within the KJMA framework one gets for V always spinodal decomposition1 and
sp
sp
never metastability. Our definitions of min and max are consistent with this picture as well as
with studies of magnetic field driven phase transitions by Rikvold et al. [40].
For large L the areas in the Maxwell construction are known [35] to shrink 1/L. Therefore,
the leading order 2-parameter fit for sp (L) is

a1
a2
sp (L) t = 1 +
(27)
.
L
L
Using t from Table 4 we show in Fig. 7 the fits to this form. Together with their goodness of
fit Q [2] the fit parameters are collected in Table 7. The Q values are a bit on the high side, as
a relatively flat (E) curve tends to give rather large statistical errors for the spinodal estimates
sp (L). In Table 7 this is reflected by a2 parameters, which are mainly statistical noise about
zero. For the q = 3 data [18], (E) is altogether too flat to allow for reasonably accurate sp (L)
estimates (larger lattices would be needed).
1 Here we use the terminology spinodal decomposition in a broader sense than some statistical physicists do.
433
Table 7
Estimates of the fit parameters of Eq. (27)
q
a1,max
a2,max
a1,min
a2,min
4
5
6
7
8
9
10
0.0193 (21)
0.0357 (20)
0.0543 (20)
0.0764 (22)
0.0969 (10)
0.1154 (19)
0.13631 (93)
1.6 (3.0)
3.0 (1.5)
2.07 (89)
0.53 (57)
0.01 (19)
0.14 (30)
0.58 (12)
0.57
0.19
0.84
0.67
0.85
0.58
0.05
0.0053 (24)
0.0154 (19)
0.0200 (29)
0.0315 (27)
0.0412 (27)
0.0475 (25)
0.0639 (19)
21 (21)
5.5 (3.7)
7.5 (4.3)
2.3 (2.0)
0.8 (1.4)
0.6 (1.1)
2.00 (41)
0.83
0.84
0.81
0.95
0.69
0.82
0.21

For 3D, q-state Potts models in the range q = 4, . . . , 10 we have estimated a number of observables by multicanonical MCMC calculations and supplemented them with q = 2 and q = 3
results from the literature. Transition temperatures, latent heats and energy endpoints of the
phases are given in Table 4, entropy and free energy values in Table 5.
Less accurate are our interface tension estimates of Table 6. They could possibly be improved
by using simulation techniques similar to those, which led to high-precision estimates of the
order-order interface tension in the 3D Ising model [41].
Minima and maxima of the microcanonical inverse temperature curve b(E) are identified
as adequate definition of spinodal endpoints. As expected [39,40] the thus defined regions of
metastability disappear in the infinite volume limit.
We hope that future investigations of 3D first order phase transitions will benefit from the
results collected in this paper.
Acknowledgements
This work was in part supported by DOE grants DE-FG02-97ER-41022 and DE-FC02-06ER41439 and by NSF grant 0555397.
References
[1] R.B. Potts, Proc. Cambridge Philos. Soc. 48 (1952) 106.
[2] B.A. Berg, Markov Chain Monte Carlo Simulations and Their Statistical Analysis, World Scientific, Singapore,
2004.
[3] F.Y. Wu, Rev. Mod. Phys. 54 (1982) 235.
[4] R.J. Baxter, J. Phys. C 6 (1973) L445.
[5] C. Borgs, W. Janke, J. Phys. I France 2 (1992) 2011, and references therein.
[6] B.A. Berg, T. Neuhaus, Phys. Rev. Lett. 68 (1992) 9.
[7] S.F. Edwards, P.W. Anderson, J. Phys. F 5 (1975) 965.
[8] K. Binder, Quadrupolar glasses and random fields, in: A.P. Young (Ed.), Spin Glasses and Random Fields, World
Scientific, Singapore, 1997.
[9] J. Villain, J. Phys. C 10 (1977) 1717.
[10] B. Svetitsky, L.G. Yaffe, Nucl. Phys. B 210 (1982) 443.
[11] R.V. Gavai, F. Karsch, B. Petersson, Nucl. Phys. B 322 (1989) 738.
[12] M. Fukugita, H. Mino, M. Okawa, A. Ukawa, J. Stat. Phys. 59 (1990) 1397.
[13] N. Alves, B.A. Berg, R. Villanova, Phys. Rev. B 43 (1991) 5846.
[14] M. Schmidt, Z. Phys. B 95 (1994) 327.
[15] W. Janke, R. Villanova, Nucl. Phys. B 489 (1997) 679.
434
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
F. Karsch, S. Stickan, Phys. Lett. B 488 (2000) 319.

R. Falcone, R. Fiore, M. Gravina, A. Papa, Nucl. Phys. B 767 (2007) 385.
A. Bazavov, B.A. Berg, Phys. Rev. D 75 (2007) 094506.
A. Gendia, T. Nishino, Phys. Rev. E 65 (2002) 046702.
A.K. Hartmann, Phys. Rev. Lett. 94 (2005) 050601.
M. Hellmund, W. Janke, Phys. Rev. E 74 (2006) 051113.
B. Hesselbo, R. Stinchcombe, Phys. Rev. Lett. 74 (1995) 2151.
S. Trebst, D.A. Huse, M. Troyer, Phys. Rev. E 70 (2004) 046701.
T. Neuhaus, J.S. Hager, J. Stat. Phys. 113 (2003) 47.
F. Wang, D.P. Landau, Phys. Rev. Lett. 86 (2001) 2050.
B.A. Berg, U.H. Hansmann, T. Neuhaus, Z. Phys. 90 (1993) 229.
A.L. Talapov, H.W.J. Blte, J. Phys. A: Math. Gen. 29 (1996) 5727.
M.S.S. Challa, D.P. Landau, K. Binder, Phys. Rev. B 34 (1986) 1841.
K. Binder, Phys. Rev. A 25 (1982) 1699.
E. Brzin, J. Zinn-Justin, Nucl. Phys. B 257 (1985) 867.
M.P. Gelfand, M.E. Fisher, Physica A 166 (1990) 1.
J.J. Morris, J. Stat. Phys. 69 (1991) 539.
A. Billoire, T. Neuhaus, B.A. Berg, Nucl. Phys. B 413 (1994) 795.
A. Hller, Z. Phys. B 95 (1994) 63.
W. Janke, Nucl. Phys. B (Proc. Suppl.) 63AC (1998) 631.
R.J. Glauber, J. Math. Phys. 4 (1963) 294;
Model A in the classification of P.M. Chaikin, T.C. Lubensky, Principles of Condensed Matter Physics, Cambridge
Univ. Press, Cambridge, 1997, Table 8.61.1, p. 467.
B.A. Berg, H. Meyer-Ortmann, A. Velytsky, Phys. Rev. D 70 (2004) 054505;
A. Bazavov, B.A. Berg, A. Velytsky, Phys. Rev. D 74 (2006) 014501.
E.g., D.P. Landau, K. Binder, A Guide to Monte Carlo Simulations in Statistical Physics, Cambridge Univ. Press,
2000, p. 41.
B.A. Berg, S. Dubey, Phys. Rev. Lett. 100 (2008) 165702.
P.A. Rikvold, H. Tomita, S. Miyashita, S.W. Sides, Phys. Rev. E 49 (1994) 5080;
M.A. Novotny, G. Brown, P.A. Rikvold, J. Appl. Phys. 91 (2002) 6908.
M. Caselle, M. Hasenbusch, M. Panero, JHEP 0709 (2007) 117, and references given therein.

Lschers -term and finite volume bootstrap principle

for scattering states and form factors
B. Pozsgay
Institute for Theoretical Physics, Etvs University, Budapest, H-1117 Budapest, Pzmny Pter stny 1/A, Hungary
Received 4 April 2008; accepted 23 April 2008
Abstract
We study the leading order finite size correction (Lschers -term) associated to moving one-particle
states, arbitrary scattering states and finite volume form factors in (1 + 1)-dimensional integrable models.
Our method is based on the idea that the -term is intimately connected to the inner structure of the particles,
i.e., their composition under the bootstrap program. We use an appropriate analytic continuation of the
BetheYang equations to quantize bound states in finite volume and obtain the leading -term (associated
to symmetric particle fusions) by calculating the deviations from the predictions of the ordinary BetheYang
quantization. Our results are compared to numerical data of the E8 scattering theory obtained by truncated
fermionic space approach. As a by-product it is shown that the bound state quantization does not only yield
the correct -term, but also provides the sum over a subset of higher order corrections as well.
1. Introduction
The knowledge of the properties of finite volume QFT is of central importance in at least two
ways. On one hand, numerical approaches to QFT necessarily deal with a finite volume box and
in order to interpret the results correctly a reliable theoretical control of finite size corrections is
needed. On the other hand, working in finite volume is not necessarily a disadvantage. On the
contrary, the volume dependence of the spectrum can be exploited to obtain (infinite volume)
physical quantities like the elastic scattering phase shifts [1,2] or resonance widths [3,4].
Finite size mass correction were first derived by Lscher [5]. The (1 + 1)-dimensional formulas relevant to integrable models together with a generalized F-term formula for moving particles
E-mail address: pozsi@bolyai.elte.hu.
doi:10.1016/j.nuclphysb.2008.04.021
436
B. Pozsgay / Nuclear Physics B 802 [FS] (2008) 435457
were obtained in [6]. Finite size corrections have recently become important in the context of
AdS/CFT correspondence as well. The generalized -term and F-term formulas for moving particles with arbitrary dispersion relation were derived in [7].
Besides the volume dependence of the spectrum itself, finite volume form factors (matrix
elements of local operators) also represent a central object in finite volume QFT. Apart from the
obvious relevance to lattice QFT they are also important in (1 + 1)-dimensional models, where
they can be used to construct a systematic low-temperature expansion of correlation functions at
finite temperature [8]. The connection to infinite volume form factors is given by a simple (though
non-trivial) proportionality factor [911] which is exact to all orders in the inverse of the volume.
There are however finite size corrections that decay exponentially with the volume and they play
a crucial role whenever the numerical simulation is limited to small volumes. Moreover, they can
produce huge deviations even in relatively large volumes provided that the exponent is small.
This work was partly motivated by such an example which can be found in [11] (Section 4.1.2).
In this work we present a method to obtain the leading -term associated to arbitrary multiparticle energy levels and finite volume form factors in (1 + 1)-dimensional integrable models.
Our approach is based on the idea that the -term is associated to the inner structure of the
particles, i.e., their composition under the bootstrap program. It is supposed that the leading term is caused by a symmetric particle fusion Aa Aa Ac . The results can also be applied in
nonintegrable models for states below the first inelastic threshold.
The outline of the paper is as follows.
We begin our analysis in Section 2 by giving a new interpretation of Lschers -term and
extending it to describe moving particles. The scaling Ising model serves as a testing ground
for our calculations: the analytic predictions are compared to numerical data obtained by the
Truncated Conformal Space Approach (TCSA) developed by Yurov and Zamolodchikov [12].
The extension of our results to arbitrary multi-particle scattering states is presented in Section 3.
Section 4 deals with the -term of finite volume form factors and Section 5 is devoted to the
conclusions.
2. One-particle states
2.1. Bound-states in finite volume
Let us consider an integrable QFT with diagonal scattering. The spectrum consists of particles
Ai , i = 1, . . . , N , with masses mi which are assumed to be strictly non-degenerate. Asymptotic
states are denoted by
|1 , 2 , . . . , n i1 i2 in ,
where the indices i1 in denote the particle species. Multi-particle scattering processes are described by the products of two-particle phase shifts Sij (ij ) where ij is the relative rapidity of
the incoming particles Ai and Aj .
The explicit formulas for the leading finite size mass corrections in a periodic box of volume L
read [6]
m()
a =

c 2 c L
m2a m2b m2c cab ab
e ab ,
b,c
(2.1)
(a) Fusion angles
(b) The mass triangle
437
(c) A triple bound state (see Section 2.2.3 for explanation)
Fig. 1. Pictorial representation of particle fusions.
)
m(F
a

b

P

d mb L cosh()
mb cosh( ) Sab ( + i/2) 1 ,
e
2
(2.2)
c )2 is the residue of
cab is the altitude of the mass triangle with base mc (see Fig. 1) and (ab
Sab ( ) corresponding to the formation of the bound state.
Here we determine the leading -term associated to a moving one-particle state Ac . Based
on the description of mass corrections it is expected that this contribution is associated to the
fusion Aa Ab Ac with the smallest cab . We assume that a = b, i.e., the fusion in question is a
symmetric one. This happens to be true for the lightest particle in models with the 3 -property
and for other low lying states in most known models. At the end of this section we comment on
the possible extension to nonsymmetric fusions.
The bootstrap principle for a symmetric fusion consists of the identification

|c + i u aac , i u aac aa ,
(2.3)
resulting in

mc = 2ma cos u aac ,
2
m2c
.
4
Smallness of caa means that mc is close to 2ma , in other words the binding energy is small.
For a moment let us lay aside the framework of QFT and consider quantum mechanics with
an attractive potential. Bound states are described by wave functions
caa
= m2a
(x1 , x2 ) = eiP (x1 +x2 ) (x1 x2 ),

where P is the total momentum and (x) is the appropriate solution of the Schrdinger equation
in the relative coordinate. It is localized around x = 0 and shows exponential decay at infinity.
Except for the region x1 x2 , the wave function can be approximated with a product of plane
waves with imaginary momenta p1,2 = P ik. The interaction results in the quantization of the
allowed values of k.
The theory in finite volume is described along the same lines. There are however two differences:
The total momentum gets quantized.
(x) (and therefore k) obtains finite volume corrections.
438
This picture also applies to relativistic integrable theories. We consider Ac as a simple quantum mechanical bound state of two elementary particles and use the infinite volume scattering
data to describe the interaction between the constituents. To develop these ideas, let us consider
the spectrum of the theory defined on a circle with circumference L. We state the identification

Ac ( ) Aa (1 )Aa (2 ) ,
(2.4)
L
L
where the 1,2 are complex to describe a bound-state; this idea also appeared in [13,14]. Relation (2.4) can be regarded as the finite volume realization of (2.3). The total energy and
momentum of the bound state have to be purely real, constraining the rapidities to take the form
1 = + iu,
2 = iu,
(2.5)
where the dependence on L is suppressed. Energy and momentum are calculated as

p = 2ma cos(u) sinh( ).
E = 2ma cos(u) cosh( ),
(2.6)
The momenta of two-particle states in finite volume are quantized by a relation involving
the scattering phase shift [1]. This procedure can be extended in (1 + 1)-dimensional integrable
models to arbitrary multi-particle scattering states. The quantization condition for an n-particle
state is given by the BetheYang equations
eipj L
n

Sij ik (j k ) = 1,
j = 1, . . . , n.
k=1
k=j
To quantize the bound state in finite volume, an appropriate analytic continuation of the above
equations with n = 2 can be applied. This procedure is justified by the same reasoning that leads
to original BetheYang equations: one assumes plane waves (with imaginary momenta) except
for the localized interaction, which is described by the S-matrix of the infinite volume theory.
Inserting (2.5) and separating the real and imaginary parts
eima cos(u) sinh()L ema sin(u) cosh()L Saa (2iu) = 1,
(2.7)
Saa (2iu) = 1.
(2.8)
ima cos(u) sinh()L ma sin(u) cosh()L
Multiplying the two equations and making use of S(2iu) = S(2iu)1 one arrives at
2I
(2.9)
,
L
which is the quantization condition for the total momentum. I is to be identified with the momentum quantum number of Ac . The quantization condition for u is found by eliminating
from (2.7):
e2ima cos(u) sinh()L = 1
or

2
ma L sin(u) 1+( ma LI
cos(u) )
2ma cos(u) sinh( ) =
Saa (2iu) = (1)I .
(2.10)
The exponential factor forces u to be close to the pole of the S-matrix associated to the formation
of the bound-state. For the case at hand it reads
c )2

i(aa
,
Saa iucaa
iucaa
(2.11)
with ucaa = 2u aac . Note the appearance of (1)I on the rhs of (2.10), which is a natural consequence of the quantization of the total momentum. This sign determines the direction from
439
which the pole is approached. The exact solution of (2.10) can be developed into a power series
c
in eaa L , where the first term is found by replacing u with u aac in the exponent:

1 c 2 caa L 1+( m2I
c
)2
cL
= (1)
e
+ O e2aa L .
2 aa
First order corrections to the energy are readily evaluated to give
u u aac
c 2 caa mc caa E0 L

c
e mc
+ O e2aa L ,
E = E0 (1)I aa
E0
where E0 is the ordinary one-particle energy

2I 2
2
E0 = mc +
.
L
(2.12)
(2.13)
In the case of zero momentum the former result simplifies to the leading term in (2.1). For large
volumes we recover
2I
u u aac ,
arsh
.
mc L
Having established the quantization procedure we now turn to the question of momentum
quantum numbers inside the bound state. For the phase shift let us adopt the convention introduced in [11]
Sab ( ) = Sab (0)eiab () ,
where ab ( ) is defined to be continuous on the real line and is antisymmetric by unitarity and real
analyticity. ab ( ) can be extended unambiguously to the imaginary axis by analytic continuation
apart from the choice of the logarithmic branch. For a generic rapidity one has
ab ( ) = ab ( ) ,
ab ( ) = ab ( ).
Note that ab (iu) is purely imaginary.

With this choice of the phase shift the BetheYang equations in their logarithmic form
l sinh( + iu) + 11 (2iu) = 2I1 ,
l sinh( iu) + 11 (2iu) = 2I2 ,
imply I1 = I2 . Quantization of the total momentum on the other hand requires I1 = I2 = I /2.
Note that different conventions for ab would result in a less transparent rule for dividing I among
the two constituents. The only disadvantage of our choice is the appearance of the unphysical
half-integer quantum numbers.
Let us denote a multi-particle state in finite volume as

{I1 , . . . , In }
,
i i ...i ,L
1 2
where the quantum numbers Ii serve as an input to the BetheYang equations. The bound-state
quantization can be written in short-hand notation as

{I }
{I /2, I /2} aa,L .
c,L
The example of mass corrections suggests that in order to obtain the total -term it is necessary to include a sum in (2.13) over the different fusions leading to Ac . However, the case of
nonsymmetric fusions requires special care. Here we outline the difficulties of the nonsymmetric
bound state quantization.
440
A straightforward application of the BetheYang equations to an |Aa ( + i u bac )Ab ( i u abc )L
ma
2I
bound state yields a nontrivial phase factor e ma +mb instead of (1)I in (2.10). This in turn
implies that the rapidity difference of the constituents cannot be purely imaginary. The real parts
of the rapidities thus get different finite size corrections and the total energy of the bound state becomes complex. A possible solution would be to take twice the real part of the energy correction,
corresponding to the sum of the contributions coming from the Aa Ab Ac and Ab Aa Ac
fusions. This is however only a guess and a more systematic treatment is needed. In fact, the
infinite volume bootstrap principle suggests that the states

Aa i u b Ab + i u a
Aa + i u b Ab i u a
ac
ac
bc L and
bc L ,
should be handled on an equal footing. A possible way to accomplish this would be to develop a
multi-channel BetheYang quantization scheme. However, this is beyond the scope of the present
work.
As a conclusion of this section (2.12) is compared to the lowest order results of the TBA
approach. The general discussion of excited states TBA equations in diagonal scattering theories
is not available. For simplicity we restrict ourselves to the LeeYang model which was considered
in the original paper [15]. In this nonunitary model there is only one particle and the scattering
is described by
S( ) =
sinh( ) + i sin(2/3)
.
sinh( ) i sin(2/3)
The exact TBA equations for moving one-particle states read

E = im(sinh 0 sinh 0 )
d
m cosh( )L( ),
2
S( 0 )
( L)( ),
( ) = mR cosh + log
S( 0 )
(2.14)
(2.15)
where

L() = log 1 + e()

and (f g)( ) =
d
f ( )g( ).
2
Here the volume is denoted by R to avoid confusion with L(). The complex rapidity 0 satisfies
the consistency equation

(0 ) = mR cosh 0 + i log S(2i Im 0 ) ( L)( ) = i(2n + 1).
(2.16)
The convolution term in (2.16) can be neglected and one obtains Im 0 = /6 + where is
exponentially small. To zeroth order one also has

mR cosh(Re 0 ) = 4n + (1 sign ) ,
which is the ordinary one-particle quantization condition with I = 2n + 12 (1 sign ). Neglecting the contribution of the integral in (2.14) and substituting 0 = + iu one has E =
2m sin(u) cosh( ). This is exactly the energy of an AA bound state with the imaginary rapidities
iu. Separating the real and imaginary parts of (2.16) and still neglecting the convolution
441
Table 1
a
ma
a11
a )2
(11
1
2
1
1.6180
0.86603
0.58779
205.14
120.80
1
0.95106
(m1 )
(212 )
1.9890
0.10453
0.20906
(2311 )
1.0819
term (which is responsible for the F-term) one obtains Eqs. (2.10) and (2.9), thus proving the
consistency of the two approaches.
It seems plausible that by exactly solving the bound state quantization condition (2.10) one
c
obtains all higher order corrections that go as enaa L with n N. In the next subsection we
present numerical evidence to support this claim.
2.2. Numerical analysis
We investigate the famous E8 scattering theory [16], which is the relativistic integrable field
theory associated to the scaling limit of the Ising model in the presence of a magnetic field.
The infinite volume spectrum of the model consists of 8 particles. The first three particles lie
below the two-particle threshold and they all show up as A1 A1 bound states. These fusions are
responsible for the leading -term. The corresponding parameters (in units of m1 ) are listed in
Table 1. The exponent of the next-to-leading correction (the error exponent) is denoted by .
In [6] Klassen and Melzer performed the numerical analysis of mass corrections. The analytic predictions were compared to TCSA data and to transfer matrix results. They observed the
expected behaviour of mass corrections of A1 and A2 ; in the former case they were also able
to verify the F-term. On the other hand, the precision of their TCSA data was not sufficient
to reach volumes where the -term for A3 could have been tested. This limitation is a natural
consequence of the unusually small exponent 311 : the next-to-leading contribution is of order
e211 L , still very slowly decaying.

Here we employ the TFCSA (Truncated Fermionic Conformal Space Approach, see [17])
routines that were successfully used in [8,11]. Calculations are performed for I = 0, 1, 2, 3 at
different values of the volume. One-particle states of A1 , A2 and A3 are easily identified: they
are the lowest lying levels in the spectrum, except for I = 0 where the lowest state is the vacuum.
We use the dimensionless quantities l = m1 L and e = E/m1 .
The results are extrapolated from ecut = 2030 to ecut = using the procedure developed
in [18]. Our experience shows that this extrapolation technique reduces the numerical errors by
an order of magnitude. However, our attempts to develop an adequate method to estimate these
errors have failed. Several examples were encountered where the actual numerical deviation was
either underestimated or overestimated, no matter which estimate was used. We therefore resign
from quantitatively monitoring the TCSA errors and constrain ourselves to a range of the volume
parameter where it is safe to neglect truncation effects.
3
2.2.1. A3
We begin our analysis with the most interesting case of A3 . At each value of l and I the
following procedure is performed.
The energy correction is calculated according to (2.13).
442
The quantization condition (2.10) is solved for u and the energy correction is calculated by
e = 2 cosh( ) cos(u) e0
where is determined by the total momentum quantization (2.9) and e0 is the ordinary oneparticle energy.
The exact correction is calculated numerically by e = eTCSA e0 .
The choice for the range of the volumes is limited in two ways. On one hand, l has to be
sufficiently large in order to reduce the contribution of the F-terms and other higher order finite
size corrections. On the other hand, numerical errors grow with the volume and eventually become comparable with the finite size corrections, resulting in an upper bound on l. The window
l = 3040 is suitable for our purposes.
The results are shown in Fig. 2. It is clear that the -term yields the correct prediction in the
L limit. However, higher order terms cause a significant deviation for l < 40, which is in
turn accurately described by the bound state prediction. The sign of the correction depends on
the parity of I as predicted by (2.13).
Based on the success of this first numerical test we also explored the region l < 30. Inspecting
the behaviour of u as a function of l reveals an interesting phenomenon. It is obvious from
(2.12) that u(l) is monotonously increasing if I is odd, with the infinite volume limit fixed to
u aac . However, the complex conjugate pair 1,2 approaches the real axis as l is decreased and
they collide at a critical volume l = lc . For l < lc they separate again but stay on the real line,
providing a unique solution with two distinct purely real rapidities. The same behaviour was also
observed in [13,14].
The interpretation of this phenomenon is evident: if the volume is comparable to the characteristic size of the bound-state, there is enough energy in the system for the constituents to become
Fig. 2. Finite size corrections to A3 one-particle levels (in sectors I = 0, . . . , 3) as a function of the volume. The TCSA
data are plotted against theoretical predictions of the single -term associated to the A1 A1 A3 fusion (solid curve)
and the exact solution of the bound-state quantization (dotted curve).
443
Fig. 3. A3 one-particle levels in sectors I = 0, . . . , 3 as a function of the volume. Dots represent TCSA data, while the
lines show the corresponding prediction of the A1 A1 bound state quantization. In sectors I = 1 and I = 3 the bound
state dissociates at lc and for l < lc a conventional A1 A1 scattering state replaces A3 in the spectrum. The values of lc
are shown by the two arrows.
unbound. Therefore the A3 one-particle level becomes an A1 A1 scattering state for l < lc . We
call this phenomenon the dissociation of the bound state. The same result was obtained also in
the boundary sine-Gordon model by a semiclassical analysis [19].
The value of lc can be found by exploiting the fact that the Jacobian of the BetheYang
equations (viewed as a mapping from (1 , 2 ) to (I1 , I2 )) vanishes at the critical point. A straightforward calculation yields

lc = 411 (0)2 I 2 2 ,
( ). The numerical values for the case at hand are
where 11 ( ) = 11
lc = 27.887 (I = 1) and lc = 26.434 (I = 3).

We are now in the position to complete the numerical analysis. The Bethe-equations are solved
at each value of l, providing two distinct real rapidities for l < lc (with I being odd), and a
complex conjugate pair otherwise. The energy is calculated in either case as
e = cosh(1 ) + cosh(2 ),
which is compared to TCSA data. The results are exhibited in Fig. 3.
The agreement for the upper two curves (I = 1 and I = 3) is not as surprising as it may seem
because what one sees here are conventional A1 A1 scattering states. The BetheYang equation

determining their energy is exact up to O(e L ) where = 111 is the smallest exponent that
occurs in the sequence of finite volume corrections of A1 . On the other hand, the energy levels
are analytic functions of L, which leads to the conclusion that the prediction of (2.10) is correct
1
up to O(e11 L ) even for L > Lc . Comparing the numerical values one finds 111 > 8311 . We
444
Fig. 4. Finite size corrections to A1 one-particle levels in sectors I = 0, . . . , 3, log10 e is plotted as a function of the
volume. Dots represent TCSA data, while the lines show the -term corresponding to the A1 A1 A1 fusion.
conclude that the bound state picture indeed accounts for finite volume corrections up to the first
c
few orders in eaa L (the first 8 orders in the case at hand).
2.2.2. A1 and A2
Particles A1 and A2 also appear as A1 A1 bound states. However, there is no point in applying
the complete bound state quantization to them, because the error terms dominate over the higher
order contributions from (2.10): the exponents of the subleading finite size corrections m1 and
212 are smaller than 2111 and 2211 . Nevertheless, the leading -term can be verified by choosing
suitable windows in l.
In Figs. 4 and 5 log(|e|) is plotted against the prediction of (2.13) for l = 616 and l = 622
(the sign of e was found to be in accordance with (2.13) for both A1 and A2 ).
In the case of A1 perfect agreement is observed for l = 1018 in the sectors I = 0 and I = 1.
For I = 2 and I = 3 the energy corrections become too small and therefore inaccessible to TCSA
(note that the prediction for I = 3 is of order 106 ).
In the case of A2 precise agreement is found for l = 1422 in all four sectors.
2.2.3. A5
Here we present an interesting calculation that determines the leading mass corrections of
A5 . The standard formulas are inapplicable in this case, because m5 lies above the two-particle
threshold. However, it is instructive to consider the composition of A5 under the bootstrap principle and to evaluate the -term prediction.
There are two relevant fusions
A1 A3 A5 ,
with 513 = 0.2079,
A2 A2 A5 ,
with 522 = 0.6581.
445
Fig. 5. Finite size corrections to A2 one-particle levels in sectors I = 0, . . . , 3, log10 e is plotted as a function of the
volume. Dots represent TCSA data, while the lines show the -term corresponding to the A1 A1 A2 fusion.
Numerical evaluation of (2.1) shows that the contribution of the second fusion is negligible
for l > 30. The first fusion on the other hand yields a significant discrepancy when compared
to TCSA data. This failure is connected to the two-particle threshold and it can be explained in
terms of the bound state quantization. Experience with A3 suggests that one should first take into
account the energy corrections of A3 and consider the A1 A3 A5 fusion afterwards. A3 can
be split into A1 A1 leading to the triple bound state A1 A1 A1 A5 . In infinite volume one has
(see also Fig. 1(c))

|5 2i u 311 , , + 2i u 311 111 .
The finite volume realization of this identification is most easily carried out in the I = 0 sector
with

{0}
{0, 0, 0} 111,L .
5,L
Setting up the three-particle BetheYang equations with rapidities (iu, 0, iu):
em1 sin(u)L S11 (iu)S11 (2iu) = 1,
e
m1 sin(u)L
S11 (iu)S11 (iu) = 1,
S11 (iu)S11 (2iu) = 1.
The second equation is automatically satisfied due to unitarity and real analyticity, whereas the
first and the third are equivalent and they serve as a quantization condition for u. The finite
volume mass of A5 is given in terms of the solution by
m5 (l) = 2 cos(u) + 1.
(2.17)
In the large L limit the infinite volume mass is reproduced by u 2u 311 . Fig. 6 demonstrates the
agreement between TCSA and the prediction of (2.17).
446
Fig. 6. Finite size mass of A5 as a function of the volume. The squares represent the TCSA data which is compared to the
leading -term (solid curve) and the solution of the quantization condition for the triple bound state A1 A1 A1 (dotted
curve). The straight line shows the infinite volume mass.
The possibility of solving the quantization of the triple bound state in a moving frame looks
very appealing. In the general case the rapidities are expected to take the form (1 + iu, 2 , 1
iu) where 1 and 2 do not necessarily coincide. However, the numerical precision of our TCSA
data was not sufficient to check our predictions.
3. Multi-particle states
3.1. BetheYang quantization in the bound state picture
Multi-particle states in finite volume are denoted by

{I1 , . . . , In }
,
i i i ,L
n
1 2
where the quantum numbers Ij serve as an input to the BetheYang equations

Qj (1 , . . . , n ) = mij sinh(j )L +
ij ik (j k ) = 2Ij , j = 1, . . . , n.
(3.1)
k=j
The energy is calculated as

E=
n
mij cosh(j ) + ,
(3.2)
j =1
where ( , 1 , . . . , n ) is the solution of (3.1). The dots indicate exponentially decaying finite size
corrections.
Let us consider a scattering state |{I, I1 , . . . , In }cb1 bn ,L composed of n + 1 particles, the
first one being Ac . We determine the leading part of the -term by considering Ac as an Aa Aa
bound state inside the multi-particle state. Therefore we write

{I, I1 , . . . , In }
{I /2, I /2, I1 , . . . , In } aab b ,L .
(3.3)
cb b ,L
1
The energy is then determined by analytic continuation of the n + 2 particle BetheYang equations. They read
ema sin(u) cosh()L eima cos(u) sinh()L Saa (2iu)
n

j =1
Sabj ( + iu j ) = 1,
(3.4)
ema sin(u) cosh()L eima cos(u) sinh()L Saa (2iu)
n

447
Sabj ( iu j ) = 1,
(3.5)
j =1
imbj sinh(j )L
Sabj (j iu)Sabj (j + iu)
n

Sbj bk (j k ) = 1.
(3.6)
k=1
k=j
The ordinary n + 1 particle Bethe-equations are reproduced in the L limit by multiplying

(3.4) and (3.5) and making use of the bootstrap equation

Scbj ( ) = Sabj + i u aac Sabj i u aac .
We now proceed similar to the previous section and derive a formula for the leading correction.
The shift in the imaginary part of the rapidity can be calculated by making use of (3.4) and (2.11)
as

( c )2
= aa e cosh()L eimc sinh()L/2

Sabj + i u aac j .
2
n
u = u u aac
(3.7)
j =1
Multiplying (3.4) and (3.5)

ei2ma cos(u) sinh()L
n

Sabj ( iu j )Sabj ( + iu j ) = 1,
(3.8)
j =1
imbj sinh(j )L
Sabj (j iu)Sabj (j + iu)
n

Sbj bk (j k ) = 1.
(3.9)
k=1
k=j
Let us define
Sabj ( iu j )Sabj ( + iu j ) Scbj ( j )e
iu cbj (j )
where

cbj ( ) = icbj + i u aac icbj i u aac

with ab ( ) = ab
( ).
Using 2ma cos(u) mc 2caa u the logarithm of (3.8) and (3.9) can be written as

n

c
cbj ( j ) u,
Q0 (, 1 , . . . , n ) = 2aa sinh( )L
j =1
Qj (, 1 , . . . , n ) = cbj ( j )u.
The lhs can be expanded around the n + 1 particle solution ( , n , . . . , n ) to arrive at

c
2aa sinh( )L nj=1 cbj ( j )

cb1 ( 1 )
1 1 (n+1) 1
. = J
u,
..
..
.
n n
cbn ( n )
where
(n+1)
Jkl
Qk
.
l
(3.10)
448
The final result for the energy correction reads

m sinh( )
c
mb1 sinh(1 ) n+1 1
c
J
E = 2aa cosh( )u +
..
m sinh(n )
bn
c
2aa sinh( )L nj=1 cbj ( j )
cb1 ( 1 )
u,
..
cbn ( n )
(3.11)
with u given by (3.7).

Based on the previous section it is expected that there is a similar contribution for every fusion
leading to each one of the constituents of the multi-particle state.
3.2. Multi-particle statesNumerical analysis
We first consider finite size corrections to A1 A3 states. They are not the lowest lying twoparticle states in the spectrum, but they possess the largest -term which is connected to the
A1 A1 A3 fusion. Given a particular state |{I1 , I3 }13,L the following procedure is performed
at each value of the volume:
The two-particle BetheYang equation for |{I1 , I3 }13,L is solved and the -term is calculated according to (3.11).
The exact three-particle BetheYang equation is solved for |{I1 , I3 /2, I3 /2}111,L .
The results for different A1 A3 levels are shown in Fig. 7. The situation is similar to the case of
the A3 one-particle levels: the bound state quantization yields a remarkably accurate prediction,
whereas the single -term prediction only becomes correct in the L limit.
In Table 2 we present a numerical example for the dissociation of the bound state inside the
two-particle state. In this case an A1 A3 state turns into a conventional A1 A1 A1 three-particle
state at lc 30.
Finite size corrections to A1 A1 and A1 A2 states are also investigated, the leading -term
given by the fusions A1 A1 A1 and A1 A1 A2 , respectively. In the former case we calculate
separately the contribution associated to both A1 particles and add them to get the total correction. Results are exhibited in Figs. 8 and 9 and formula (3.11) is verified in both cases.
4. Finite volume form factors
The connection between finite volume and infinite volume form factors was derived in [11] as

j1 ...jm ,L {I1 , . . . , Im } O(0, 0) {I1 , . . . , In } i i ,L
1
F O (m

+ i, . . . , 1 + i, 1 , . . . , n )jm j1 i1 in
i1 in (1 , . . . , n )j1 jm (1 , . . . , m )

+ O e L ,
(4.1)
449
Fig. 7. Finite size corrections to A1 A3 scattering states as a function of the volume. Dots represent TCSA data, the solid
line shows the -term corresponding to the A1 A1 A3 fusion. The dotted lines are obtained by the exact solution of
the quantization condition for the A1 A1 A1 three-particle system.
Table 2
An example for the dissociation of the A1 A1 bound state inside a scattering state. |{2, 0}31,L is identified with
|{1, 1, 0}111,L and the corresponding BetheYang equations is solved. For l < 30 there is a real A1 A1 A1 three-particle
state in the spectrum, whereas at l 30 two of the rapidities become complex and the two-particle state A1 A3 emerges
l
e(l) (predicted)
e(l) (TCSA)
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
0.44191
0.42574
0.60523
0.56934
0.38910
0.50092
0.46686
0.42947
0.38646 + 0.02332i
0.37059 + 0.04042i
0.35572 + 0.05104i
0.34182 + 0.05891i
0.32876 + 0.06513i
0.31650 + 0.07022i
0.30498 + 0.07446i
0.68235
0.64280
0.41155
0.39931
0.53475
0.38119
0.37634
0.37739
0.38646 0.02332i
0.37059 0.04042i
0.35575 0.05104i
0.34182 0.05891i
0.32876 0.06513i
0.31650 0.07022i
0.30498 0.07446i
0.58742
0.55190
0.51892
0.48831
0.45989
0.43350
0.40899
0.38622
0.36505
0.34537
0.32706
0.31002
0.29415
0.27936
0.26556
3.51877
3.46201
3.41239
3.36890
3.33071
3.29709
3.26743
3.24122
3.21801
3.19741
3.17908
3.16275
3.14816
3.13511
3.12341
3.51900
3.46219
3.41255
3.36907
3.33089
3.29729
3.26767
3.24164
3.21832
3.19775
3.17945
3.16313
3.14856
3.13547
3.12235
where the rapidities are solutions of the corresponding BetheYang equations and it is supposed
that j = k whenever ij = ik . The extension of (4.1) to include disconnected terms can be found
in [8]. The proportionality factor in (4.1) is given by the density of states
(n)
i1 ...in (1 , . . . , n ) = det J (n) ,
(n)
Jkl =
Qk
,
l
k, l = 1, . . . , n,
450
Fig. 8. Finite size corrections to A1 A1 scattering states, log10 e is plotted as a function of the volume. Dots represent
TCSA data, while the solid line show the sum of the two -terms corresponding to the A1 A1 A1 fusions.
Fig. 9. Finite size corrections to A1 A2 scattering states, log10 e is plotted as a function of the volume. Dots represent
TCSA data, while the solid line show the -term corresponding to the A1 A1 A2 fusion.
which can be used to identify formally the finite volume and infinite volume states as

{I1 , . . . , In }
i1 i2 in ,L
1
(n)
i1 in (1 , . . . , n )
|1 , . . . , n i1 in .
Based on general arguments it was shown in [11] that where is determined by the
pole of the S-matrix closest to the physical line. A systematic finite volume perturbation theory
(Lschers method applied to form factors) is not available. However, it is expected that the
actual value of depends on what diagrams contribute to the form factor in question. Apart
from the insertion of the local operator they coincide with the diagrams determining the finite
size corrections of the multi-particle state. Therefore is associated to the bound state structure
of the constituents of the multi-particle state. In this section it is shown that the leading correction
term can be obtained by the bound state quantization.
451
4.1. Elementary one-particle form factors

(4.1) yields a simple prediction for the elementary one-particle form factor:

FO

FcO (I, L) 0|O(0, 0)|{I }c,L = c + O e L ,
EL
(4.2)
where E is the one-particle energy, and FcO = FcO ( ) is the infinite volume one-particle form
factor, which is constant by Lorentz symmetry.
The -term associated to (4.2) is derived by employing the bound state quantization. We gain
some intuition from the previous sections where it was found that the bound state Aa Aa may
dissociate at a critical volume Lc . For L < Lc there is no one-particle level of type Ac in the
given sector of the spectrum, however an Aa Aa scattering state appears instead. Finite volume
form factors of this state are calculated using (4.1) as
F O (1 , 2 )aa
FcO (I, L) =
aa (1 , 2 )
(4.3)
for L < Lc .
The generalization to L > Lc seems to be straightforward: one has to continue analytically

(4.3) to the solutions of the BetheYang equation with imaginary rapidities 1,2 = iu. However, note that Eqs. (4.2) and (4.3) are valid up to a phase factor. In order to continue analytically
to imaginary rapidities we also need to fix this phase.1
The two-particle form factor satisfies
F O (1 , 2 )aa = Saa (1 2 )F O (2 , 1 )aa .
The simplest choice for the phase is therefore

F O (1 , 2 )aa = Saa (1 2 )F O (1 , 2 )aa .
(4.4)
This choice is dictated by CPT symmetry [9,10,20], and it is respected by all known solutions of
the form factor bootstrap axioms. There is a sign ambiguity caused by the square root, but it can
be fixed by demanding (Saa (0))1/2 = i and continuity. Using (4.4)
Saa (2 1 )F O (1 , 2 )aa
FcO (I, L) =
aa (1 , 2 )
and upon analytic continuation
Saa (2iu)F O ( + iu, iu)aa

O
Fc (I, L) =
aa ( + iu, iu)
for L > Lc .
(4.5)
It is easy to see that the result (4.2) is reproduced in the L limit. First observe that
F O ( + iu, iu)aa
c
aa
F O ( ).
2(u u caa ) c
1 The phase of a (nondiagonal) infinite volume form factor is unphysical in the sense that it may be redefined by a complex rotation of the state vectors and physical quantities, e.g., correlation functions, do not depend on such redefinitions.
However, the bootstrap program uniquely assigns a phase to each form factor.
452
The residue of aa is determined by aa (2iu) and it reads

( + iu, iu)aa 2ma L cos(u) cosh( )(i)
(2iu)
Saa
1
= mc L cosh( )
.
Saa (2iu)
2(u u aac )
The singularities in the numerator and denominator of (4.5) cancel and indeed
FcO (I, L)
FcO
.
mc L cosh( )
We emphasize that it is crucial to include the extra normalisation factor Saa (2iu) to obtain a
meaningful result.
Expression (4.5) can be developed into a Taylor-series in u u aac . The first order correction
is evaluated in Appendix A and it reads

FO
2i c (F O )
FcO (I, L) = c aa aa u u aac
Ec0 L
Ec0 L
2

c,0

FcO
mc
ma 0 2
2Saa
+
+ 0 2
E 0 L u u aac + O e2L ,
c )2
2 c
0
(
(E
)
m
E
Ec L
aa
c
c
c
(4.6)
where
O
Faa =
lim
=2i u aac
c,0
Saa
= lima
uu ac
d O
F (, )aa ,
d

Saa (2iu)
c )2
(aa
.
2(u u aac )
Ec0 is the ordinary one-particle energy and the rapidity shift u u aac is given by (2.12).
4.2. One-particle form factorsnumerical analysis
Let us introduce the dimensionless form factors as
fi (I, l) =
0|(0, 0)|{I }i,L

.
m1
The machinery of [11] is used to determine fi (I, l) for i = 1, 2, 3 and I = 0, 1, 2, 3. The numerical results are compared to the exact infinite volume form factors [21,22].
We start our investigation with f3 (I, l), for which relatively large exponential corrections
were already reported in [11]. It is convenient to consider
f3 (I, l) = (e0 l)1/2 f3 (I, L) with lim f3 (I, l) = F3 .
l
(4.7)
The numerical results are demonstrated in Fig. 10. Note, that this is exactly the same figure as
Fig. 4.6(c) in [11], but this time the interpretation of the huge deviations from F3 is also provided.
We also tried to verify the predictions for f1 and f2 . In the latter case reasonably good
agreement was found with TCSA, the results are demonstrated in Fig. 11. In the case of f1
we encountered the unpleasant situation that the F-term decays slower than the TCSA errors
grow, thus making the observation of the -term impossible.
453
Fig. 10. Elementary finite volume form factors of A3 one-particle levels. Here the normalization (4.7) is applied to obtain
a finite l limit, which is given by the infinite volume form factor F3 = 0||A3 (). The TCSA data are plotted
against the bound state prediction. The ordinary evaluation of f3 is simply the constant F3 .
It is straightforward to generalize (4.5) to matrix elements between two different one-particle

states. For b = c one has for example

F (b + i, + iu, iu)baa
.
{Ib } {Ic } c,L =
b (b )aa ( + iu, iu)

b
Numerical examples are presented in Fig. 12(a)(c) for c = 3 and b = 1, 2.

The most interesting case is the one shown in Fig. 12(d) where the matrix element between
two different A3 one-particle states are investigated. This can be done by considering both A3
particles as the appropriate A1 A1 bound states and then calculating the finite volume form factor

3 {I }||{I }3,L as

F ( + iu + i, iu + i, + iu , iu )1111
{I /2, I /2} {I /2, I /2} 11,L =
.
11 ( + iu , iu )11 ( + iu, iu)

11
Once again we find complete agreement with the TCSA data.

4.3. Elementary multi-particle form factors
The generalization of (4.5) to multi-particle states is straightforward, the only task is to find
the appropriate phase factor. Similar to the one-particle case one has

Sbi bj (i j )F O (1 , . . . , m )b1 ...bm .
F O (1 , . . . , m )b1 bm =
i<j
A general n particle finite volume form factor with real rapidities can thus be written as

i<j Sbi bj (j i ) O
F (1 , . . . , n )b1 bn .
n (1 , . . . , n )b1 bn
454
Fig. 11. Finite size corrections to the elementary form factors of A2 . Dots represent TCSA data, while the lines show the
-term prediction corresponding to the A1 A1 A2 fusion.
Fig. 12. One-particleone-particle form factors, dots correspond to TCSA data. The solid lines represent the ordinary
evaluation of the finite volume form factors, while the dotted lines show the bound state prediction.
Substituting the solution of the Bethe-equation for the state |{I /2, I /2, I1 , . . . , In }aab1 bn ,L and
making use of the real analycity condition

Sab (j iu)Sab (j + iu) = 1,
j
j
one gets
0|O|{I /2, I /2, I1 , . . . , In }aab1 bn ,L
Saa (2iu)|F O ( + iu, iu, 1 , . . . , n )aab1 bn |

=
,
(n+2) ( + iu, iu, 1 , . . . , n )aab1 bn
(4.8)
455
Fig. 13. Elementary form factors of A1 A3 scattering states, dots correspond to TCSA data. The solid lines are obtained
by a naive evaluation of the finite volume form factors, while the dotted line represents the bound state prediction (in
this case A1 A1 A1 form factors at the appropriate rapidities).
up to a physically irrelevant phase.

It is easy to show once again that the naive result is reproduced in the L limit. To do
so, we first quote the dynamical pole equation of the infinite volume form factor:
F O ( + iu, iu, 1 , . . . , n )aab1 ...bn =
c
aa
F O (, 1 , . . . , n )cb1 bn + O(1).
2(u u aac )
The singularity of (n+2) is given by

1
Resuu aac (n+2) ( + iu, iu, 1 , . . . , n )aab1 bn = (n+1) (, 1 , . . . , n )cb1 bn .
2
The naive formula is now recovered by inserting the last two equations into (4.8).
The leading exponential corrections can be obtained by plugging (3.7) and (3.10) into (4.8)
and expanding to first order in u u aac . This procedure is straightforward but quite lengthy,
therefore we refrain from giving the details of the calculations.
In Fig. 13 two examples are presented for the evaluation of (4.8) applied to A1 A3 two-particle
states.
5. Conclusions
In this work we determined the generalization of Lschers -term associated to moving oneparticle states, arbitrary scattering states and finite volume form factors. Our method is based
on the bootstrap principle of the infinite volume theory which states that a particle which is a
bound state of two others is indeed indistinguishable from the two-particle state with the appropriate (imaginary) momenta. An analytic continuation of the BetheYang equations was used to
quantize the bound states in finite volume.
The analytic results were tested by comparing the predictions to high-precision TCSA data
of the Ising model; a satisfactory agreement was observed in each case. We also demonstrated
that the bound state quantization goes beyond the leading mu-term corresponding to the fusion
c
and gives a resummation of all powers of eaa L . This is a substantial improvement over the
inclusion of the leading Lscher term when other sources of finite volume corrections (other
fusions, F-terms) can be neglected. Our main example was provided by particle A3 from the E8
scattering theory.
456
As an analytic check the calculations were compared to the leading order results of the exact
TBA equations. The agreement between the two approaches was explicitly demonstrated in the
case of the one-particle levels of the LeeYang model.
Our results can also be applied in nonintegrable models for states below the first inelastic
threshold. However, the calculations only apply to symmetric fusions of the type Aa Aa Ac .
The nonsymmetric case requires an extension of the BetheYang equations to incorporate multichannel scattering. Such a quantization scheme could also be used to describe finite volume states
in nonintegrable theories above the two-particle threshold.
Acknowledgements
The author is grateful to G. Takcs for his support during the completion of this work and for
correcting the manuscript. The author would also like to thank Z. Bajnok for interesting discussions. This research was partially supported by the Hungarian research fund OTKA K60040.
Appendix A. -term for the one-particle form factor
Here we develop the first order correction to (4.5). Using the exchange axiom
FcO (I, L) =
F O ( iu, + iu)aa
.
Saa (2iu)aa ( + iu, iu)
(A.1)
The form factor axioms imply that

lim F O ( iu, + iu)aa =
uu aac
1 O
F .
c c
aa
The simple pole of aa (2iu) in aa is cancelled by Saa (2iu), therefore both the numerator and
the denominator of (A.1) have continuous limits as u u aac .
The form factor F O ( iu, + iu)aa only depends on u by Lorentz-symmetry. Therefore
F O ( iu, + iu)aa =

O
1 O
u u aac + ,
Fc 2i Faa
c
aa
where
O
d O
Faa =
F (, )aa | =2i u aac .
d
Expanding the S-matrix element into a Laurent-series in the vicinity of the pole
c )2
(aa
c,0
+ ,
+ Saa
2(u u aac )

2(u u aac )
2(u u aac ) 2 c,0
Saa (2iu) =
Saa + .
c )2
c )2
(aa
(aa
Saa (2iu) =
Expanding the denominator:

S(2iu)aa ( + iu, iu)

2
= S(2iu)E1 E2 L2 i(E1 + E2 )LS (2iu) S(2iu)

Ec L
2E1 E2 L2
4Ec L c,0
=
+
Saa u u aac + ,
c
2
c
2
c
4
(aa )
(aa )
(aa )
457
where
Ec = E1 + E2 = 2ma cos(u) cosh( ).
Putting all this together
FcO (I, L) =

c (F O )
2iaa
FcO
FcO
2Ec L c,0
aa
2
E1 E2 L +
+
+
S
3
c )2 aa
(aa
Ec L
Ec L
Ec L

u u aac + .
Note that in the precedingformulas Ec does include the leading order correction to the usual
one-particle energy Ec0 = m2c + (2I )2 /L2 . Using
Ec = Ec0 2

mc
m2
u u aac + O e2L and E1 E2 = a2 Ec2 2 ,
0
Ec
mc
the final result is given by

FO
2i c (F O )
FcO (I, L) = c aa aa u u aac
Ec0 L
Ec0 L

2
c,0

2Saa
FO
mc
ma 0 2
L u u aac + O e2L ,
+ c
+
c
c
2
0
2
2
0
(Ec )
mc
Ec
Ec0 L (aa )
(A.2)
with
u u aac =
1 c 2 caa L
e
2 aa
I
2
a )
a L cos(u ac )
1+( m
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
M. Lscher, Commun. Math. Phys. 105 (1986) 153.

M. Lscher, U. Wolff, Nucl. Phys. B 339 (1990) 222.
M. Lscher, Nucl. Phys. B 364 (1991) 237.
B. Pozsgay, G. Takcs, Nucl. Phys. B 748 (2006) 485, hep-th/0604022.
M. Lscher, Commun. Math. Phys. 104 (1986) 177.
T.R. Klassen, E. Melzer, Nucl. Phys. B 362 (1991) 329388.
R. Janik, T. Lukowski, Phys. Rev. D 76 (2007) 126008, arXiv: 0708.2208 [hep-th];
M. Heller, R. Janik, T. Lukowski, arXiv: 0801.4463 [hep-th];
Y. Hatsuda, R. Suzuki, arXiv: 0801.0747 [hep-th].
B. Pozsgay, G. Takcs, Nucl. Phys. B 788 (2008) 209251, arXiv: 0706.3605 [hep-th].
L. Lellouch, M. Lscher, Commun. Math. Phys. 219 (2001) 31.
C.-J.D. Lin, G. Martinelli, C.T. Sachrajda, M. Testa, Nucl. Phys. B 619 (2001) 467.
B. Pozsgay, G. Takcs, Nucl. Phys. B 788 (2008) 167, arXiv: 0706.1445 [hep-th].
V.P. Yurov, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 6 (1991) 3419.
H. Kausch, G. Takcs, G. Watts, Nucl. Phys. B 489 (1997) 557, hep-th/9605104.
Z. Bajnok, L. Palla, G. Takcs, Nucl. Phys. B 614 (2001) 405, hep-th/0106069.
P. Dorey, R. Tateo, Nucl. Phys. B 482 (1996) 639, hep-th/9607167.
A.B. Zamolodchikov, Int. J. Mod. Phys. A 4 (1989) 4235.
V.P. Yurov, Al.B. Zamolodchikov, Int. J. Mod. Phys. A 6 (1991) 4557.
Z. Bajnok, L. Palla, G. Takcs, F. Wgner, Nucl. Phys. B 587 (2000) 585, hep-th/0004181.
Z. Bajnok, L. Palla, G. Takcs, Nucl. Phys. B 702 (2004) 448, hep-th/0406149.
L. Maiani, M. Testa, Phys. Lett. B 245 (1990) 585.
G. Delfino, P. Simonetti, Phys. Lett. B 383 (1996) 450, hep-th/9605065.
G. Delfino, P. Grinza, G. Mussardo, Nucl. Phys. B 737 (2006) 291, hep-th/0507133.

Numerical study of entanglement entropy in SU(2)

lattice gauge theory
P.V. Buividovich a,b, , M.I. Polikarpov b
a JIPNR, National Academy of Science, Acad. Krasin str. 99, 220109 Minsk, Belarus
b ITEP, B. Cheremushkinskaya str. 25, 117218 Moscow, Russia
Received 29 February 2008; accepted 24 April 2008

Abstract
The entropy of entanglement between a three-dimensional slab of thickness l and its complement is
studied numerically for four-dimensional SU (2) lattice gauge theory. We find a signature of a nonanalytic
behavior of the entanglement entropy, which was predicted recently for large Nc confining gauge theories in
the framework of AdS/CFT correspondence. The derivative of the entanglement entropy over l is likely to
have a discontinuity at some l = lc . It is argued that such behavior persists even at finite temperatures, probably turning into a sort of crossover for temperatures larger than the temperature of the deconfinement phase
transition. We also confirm that the entanglement entropy contains quadratically divergent l-independent
term, and that the nondivergent terms behave as l 2 at small distances.
PACS: 12.38.Aw; 03.65.Ud
1. Introduction
The application of the concepts of quantum information theory to quantum field theories in
continuous spacetimes and on the lattices has led recently to many important advances in our
understanding of their quantum behavior [16]. One such concept is the quantum entanglement
of states of systems with many degrees of freedom, which turned out to be a very useful modelindependent characteristic of the structure of the ground state of quantum fields. In particular, for
quantum fields on the lattice in the vicinity of a quantum phase transition (i.e. a phase transition
* Corresponding author at: JIPNR, National Academy of Science, Acad. Krasin str. 99, 220109 Minsk, Belarus.
E-mail addresses: buividovich@tut.by (P.V. Buividovich), polykarp@itep.ru (M.I. Polikarpov).

doi:10.1016/j.nuclphysb.2008.04.024
P.V. Buividovich, M.I. Polikarpov / Nuclear Physics B 802 [FS] (2008) 458474
459
which occurs at zero temperature when some parameters of the system, such as the coupling
constants, are varied) the ground state is a strongly entangled superposition of the states of all
elementary lattice degrees of freedom (such as spins in Heisenberg model, or link variables in
lattice gauge theory), and different phases of lattice theories can be characterized by different
patterns of entanglement [16]. Quantum entanglement thus appears to be an adequate concept
for the description of the emergence of collective degrees of freedom in quantum field theories
[15].
A commonly used measure of quantum entanglement of the ground state of quantum fields in
(D 1) + 1-dimensional spacetime is the entropy of entanglement S[A] between some (D
1)-dimensional region A and its (D 1)-dimensional complement B, which characterizes the
amount of information shared between A and B [2,4,7]. Entanglement entropy is defined as the
usual von Neumann entropy for the reduced density matrix A associated with the region A:
S[A] = TrA (A ln A ).
(1)
The reduced density matrix is obtained from the density matrix of the ground state of the theory,
AB = |00|, by tracing over all degrees of freedom which are localized outside of A, i.e. within
B [2,7]:
A = TrB AB = TrB |00|.
(2)
This density matrix describes the state of quantum fields as seen by an observer who can only
perform measurements within A. In other words, the region B is inaccessible for such an observer, as if it was separated from A by a sort of event horizon. Thus the entanglement entropy
is in some sense similar to the entropy of black holes [8,9].
As was demonstrated by Bekenstein and Hawking, the entropy of a black hole is proportional
to the area of its horizon [10,11]. This fact has recently received an interesting development in the
works [1215], where this entropic area law was conjectured to hold also for the entanglement
entropy of field theories which are dual (in the sense of Maldacena, or gauge-gravity, duality
[16]) to supergravity on AdS spaces (or their thermal modifications). More precisely, the original
conjecture of [12,13] is that for (D 1) + 1-dimensional conformal field theories living on the
boundary of (D + 1)-dimensional AdS space, the entropy of entanglement between the region
A and its complement is proportional to the minimal area of hypersurface in AdS space which is
spanned on the boundary of A. This was explicitly demonstrated for two-dimensional conformal
field theories living on the boundary of AdS3 , for which the entanglement entropy is proportional
to the minimal length of a line which joins the ends of A in the bulk of this AdS3 (for field
theories in 1 + 1 spacetime dimensions A is just a line). For higher-dimensional conformal
field theories which have supergravity duals the entanglement entropy is not known exactly, thus
the conjecture of [12,13] cannot be directly verified. Nevertheless, it was found in [12,13] that
the area law prescription gives at least a reasonable approximation both for zero and nonzero
temperatures also in these cases.
To be more specific, the results of [1215] are obtained for the case when A is a slab of
thickness l in (D 1)-dimensional space, and the entanglement entropy is normalized per unit
area of the boundary A of A. A typical behavior of entanglement entropy for field theories in
(D 1) + 1-dimensional spacetime is [1215]:
1
1
1
2D
f (l),
S(l) =
SUV +
Sf (l) = D2
UV l
|A|
|A|
|A|
(3)
460
where is some constant which is different for different theories, UV is the UV cutoff scale
and f (l) is some function which is finite as l 0. The l 2D behavior of the UV finite term
in (3) can be fixed by the requirement that the entropy does not diverge at l 0, which is
reasonable for a local field theory. Indeed, the ultraviolet divergence in (3) can only be cancelled
if Sf (l) tends to l 2D at small distances of order of 1
UV . This means that the entanglement
entropy associated with a set of points of zero measure is finite. If the entanglement entropy
is calculated as the minimal area of hypersurface with boundary A in the bulk geometry, the
ultraviolet divergence and the l 2D behavior at small distances are automatically reproduced due
to the singularity of the metric near the boundary of AdSD+2 . Thus the nontrivial information
about the entanglement due to field interactions is contained in the function f (l). In terms of
the dual geometries the function f (l) is a probe of the nonsingular structure of the bulk space.
For the sake of brevity we will restrict further discussion to the case D = 4. Following [14], it is
convenient to introduce the so-called entropic C-function as follows:
C(l) =
l3
S(l) = 2f (l) l f (l).

|A| l
l
(4)
Since the divergent part of the entropy (3) does not depend on l, ultraviolet divergences cancel in
(4), and the entropic C-function is UV finite. The factor l 3 ensures that C(l) is also finite at l 0.
For conformal field theories, this function is proportional to the central charge and thus counts the
number of effective degrees of freedom of the theory, that is, the number of distinct field species
weighted with their central charges (such as 1 for a free boson, 1/2 for a free fermion, etc.). An
important property of the entropic C-function, which follows from Zamolodchikov C-theorem
[17], is that it should decrease as the energy scale degreases, i.e. as its argument l grows.
Simplicity and elegance of the area law for the entanglement entropy suggest that it has
a fundamental nature and can be applied also to nonconformal field theories which have dual
supergravity descriptions. This was done in [14,15], where the entanglement entropy of some
confining large Nc gauge theories was studied basing on the conjecture of [12,13]. These theories
are dual to supergravity on certain SUSY-breaking five-dimensional geometries [18,19]. One of
the most interesting results of [14,15] is that if the entropic area law is assumed to hold also for
such theories, their entanglement entropy should be nonanalytic in l. This nonanalytic behavior
is a consequence of the existence of two distinct minimal hypersurfaces in five-dimensional bulk
geometry, a connected one and a disconnected one [14,15]. For small regions A the area of
the connected hypersurface is less than that of the disconnected one, and it is this area which
determines the entanglement entropy, while for sufficiently large A the situation is reversed. At
some lc the areas of these two hypersurfaces become equal, and as a result the derivative of the
entanglement entropy over l and the entropic C-function are discontinuous at l = lc . For example,
for gauge theories which are dual to D3 or D4 branes on a circle [15], C(l) is proportional
to the step function: C(l) Nc2 (lc l). This means that the effective number of degrees of
freedom rapidly changes from being of order of Nc2 to Nc0 as the region A in (1) becomes smaller
(in the large Nc limit Nc0 is just zero as compared with Nc2 ). It was conjectured in [15] that
this nonanalytic behavior is similar to the confinement-deconfinement phase transition at finite
temperatures. Namely, when the size of the region A is sufficiently small and the corresponding
energy scale is sufficiently high, we are in the domain of asymptotic freedom and the number
of particles which contribute to the entropy (1) is proportional to Nc2 . As the region A becomes
larger and the infrared effects become strong, coloured states are confined and we are left with
a theory of colourless glueballs and hadrons, for which the number of states and the entropy are
461
much smaller. It was argued in [15] that such behavior of the entanglement entropy is generic for
confining large-Nc gauge theories because of the Hagedorn-like structure of their spectrum.
Thus the behavior of entanglement entropy of confining gauge theories can be an interesting
new test of Maldacena duality between string theories on (D + 1)-dimensional curved spaces and
D-dimensional gauge theories which live on their boundary. It can be also used as an alternative
probe which indicates the emergence of the colourless degrees of freedom in the low-energy limit
of gauge theories. The results of [14,15] imply that colourless degrees of freedom always emerge
instantly, so that in some sense gluons and glueballs cannot coexist at one energy scale. For finite
Nc the arguments of [14,15] remain valid at the qualitative level, however, it is not clear how the
nonanalytic behavior may be modified in this case, and different phenomenological descriptions
of confining gauge theories give different predictions. For instance, in the AdS space with hard
wall [20], which is one of the most popular phenomenological stringy models of confinement,
there is only one minimal hypersurface for all l [15], and according to the conjecture of [1215]
the entanglement entropy should be analytic in l. On the other hand, if the entanglement entropy is calculated using the approximate MigdalKadanoff decimations in lattice gauge theory,
it appears to be non-analytic [21].
Unfortunately, up to now the infrared dynamics of confining gauge theories is not understood
from the first principles, and no method is known which can be used to calculate explicitly
the entanglement entropy of confining gauge theories for sufficiently large regions A. The aim
of the present work is to investigate the entanglement entropy (1) in pure SU(2) gauge theory
numerically, using lattice Monte Carlo simulations. The same geometry of A as in [1215,21]
is used, namely, A is a slab of three-dimensional space which has thickness l in one direction
and which extends maximally in all other directions, and the entropy is normalized per unit
area of the boundary of A. Simulation results suggest that the derivative of the entanglement
entropy over l is indeed discontinuous at some l = lc . In addition, we present some indications
that this nonanalytic behavior may persist even at finite temperatures above Tc , probably turning
into a crossover line in the lT plane. We also confirm that the entanglement entropy diverges
quadratically and that at small distances Sf (l) indeed behaves as Sf (l) l 2 .
The structure of the paper is the following. In Section 2 we discuss the path-integral representation of the entanglement entropy, which is suitable for Monte Carlo lattice simulations, and
introduce the spaces of special topology on which these simulations should be performed. In
Section 3 we present the results of our measurements of the action density distributions on lattices with such special topology and confirm that the divergent term in the entanglement entropy
scales as 2UV and does not depend on l. In Section 4 the dependence of the nondivergent part
of the entanglement entropy on l is investigated and the discontinuity of the derivative l Sf (l)
is detected. Nonanalytic behavior of the entanglement entropy with respect to l at T > Tc is
discussed and studied numerically in Section 5. Finally, Section 6 contains some concluding remarks and the discussion of the obtained results. Some technical details of our calculations are
relegated to Appendices A and B.
2. Measurement procedure
In this section we present some important details of the numerical procedures which we have
used to measure the entanglement entropy. Highly technical details are relegated to Appendices A
and B.
Entanglement entropy cannot be expressed as an expectation value of some field operator, thus
theoretical and especially numerical calculations of such entropy in field theories are far from
462
Fig. 1. The topology of space on which the free energy F [A, s, T ] in (5) is calculated, an example for s = 2. Dashed
lines with arrows denote identification of cut sides, i.e. periodic boundary conditions in time direction.
being straightforward. For example, most investigations of quantum entanglement of the ground
states of lattice spin systems were carried out using rather special transfer-matrix methods, which
are applicable only to some particular models [1]. In [4] a path-integral representation of the
entanglement entropy was derived, which relates the entanglement entropy to the set of partition
functions on the spaces of some special topology, namely, the spaces with an integer number
of cuts. In two dimensions such spaces can be implemented as multi-sheeted Riemann surfaces,
which allows one to find explicitly the entanglement entropy in conformal field theories [2,4].
For higher-dimensional spaces, it is also relatively easy to implement the spaces of required
topology, at least on the lattice. For the purposes of lattice simulations it is convenient to represent
the result of [4] in the following form:

S[A] = lim lim F [A, s, T ] F (T ) ,

(5)
T 0 s1 s
where F (T ) = ln Z(T ) is the free energy of the theory at the temperature T . Z[A, s, T ] and
F [A, s, T ] = ln Z[A, s, T ] are the partition function and the free energy obtained by integrating over all fields with the following boundary conditions [2,4]: the fields are periodic in
time with the period sT 1 if the spatial coordinates lie within A, otherwise (i.e. for spatial
coordinates within B) the period is T 1 . Such topology of space is schematically shown on
Fig. 1 for s = 2. Thus F [A, s, T ] is the free energy of the theory on the space with s cuts [2,4].
In four-dimensional spacetime these cuts are the three-volumes B at the time slices t = kT ,
k = 0, 1, . . . , s, which have a common boundary A = B. At this boundary the branching points
of the cuts are situated. The expression (5) is derived in Appendix A, where the original derivation of [4] is briefly revised.
Since free energies cannot be directly measured in lattice simulations, one usually looks at the
average plaquette action, which is related to the derivative of the free energy over the coupling
constant = 4g1 2 of SU(2) lattice gauge theory:
s

1
F (),
1 Tr gp =
2
ln
(6)
463
where summation goes over all lattice plaquettes. The average plaquette action and the free energy have the same power-like ultraviolet divergences, which follows from the fact that the lattice
spacing a() = 1
UV depends on as a() exp(), where the coefficients and can
be obtained from the GellmanLow -function. Thus the derivative in (6) can only change logarithmic dependence on the UV cutoff scale, which is usually too weak to be observed in lattice
simulations. In Section 3 we use the expression (6) to find the UV divergent part of the entanglement entropy (3).
In practice it turns out that Eq. (6) can only be reliably used to find the divergent part of the
entanglement entropy, while extremely precise measurements are required to study the behavior
of the nondivergent part. For this reason it is more convenient to measure the derivative of the
entanglement entropy (5) over l, which is related to the free energy as:

S[A] = Sf (l) = lim lim

F [A, s, T ].
T 0 s1 s l
l
l
(7)
The derivative l F [A, s, T ] can be estimated from finite differences of free energies for close
values of l. Such differences can be directly measured in lattice simulations using the method
proposed recently in [22,23]. The results of these measurements are presented in Section 4, and
the method itself is described in Appendix B.
Thus the expressions (5), (6) and (7) allow one to calculate the entanglement entropy by
performing lattice simulations on lattices with integer number of cuts. In practice such simulations can be done only for several smallest s, after which some numerical interpolation method
should be used to find the derivative over s. In this work we use the simplest possible way to
estimate this derivative. Namely, from lattice simulations on lattices with two cuts we obtain
the values of limT 0 F [A, 2, T ] or limT 0 l F [A, 2, T ]. Taking into account that by definition
l F [A, 1, T ] = 0 and F [A, 1, T ] = F (T ), we estimate the derivative over s at s = 1 in (5) as the

following finite difference:

lim lim F [A, s, T ] F (T ) lim F [A, 2, T ] 2F (T ) .

(8)
T 0 s1 s
T 0
The results that we obtain using such an estimation are in a rather good agreement with theoretical expectations [1215], which gives us a hope that the approximation (8) is not unreasonable.
It should be also noted that in fact any of the quantities PA (s) = exp(F [A, s, T ] + sF (T )) =
Tr As (see Appendix A for a proof of this equality) for s > 1 can be used as a measure of quantum entanglement. In particular, the quantity P = exp(F [A, 2, T ] + 2F (T )) = Tr A2 , although
it does not have such a straightforward physical interpretation as the entanglement entropy, is
also widely used in quantum information theory and has a special name of the purity of quantum
states. For pure states the purity is equal to 1, for impure states it is less than 1, and the smaller it
is, the more impure is the quantum state described by A .
3. Average action density on lattices with cuts
In this section we present the results of our measurements of the average action density on
lattices with cuts, and demonstrate that the entanglement entropy contains quadratically divergent
l-independent term.
Lattices with s cuts are obtained from sNt NL3 lattices with periodic boundary conditions
in all space-like directions by identifying lattice sites with periods Nt and sNt in time for spatial
coordinates within A and B, respectively. These sites are then linked taking into account this
464
Fig. 2. The distributions of average action density in the (x 0 , x 1 ) plane for different lattices and cut sizes: at the upper
left: 24 203 lattice with two cuts, a 1 l = 8, a = 0.12 fm, at the upper right: 24 203 lattice with two cuts, a 1 l = 6,
a = 0.12 fm, at the lower left: 28 243 lattice with two cuts, a 1 l = 6, a = 0.10 fm, at the lower right: 18 123 lattice
with three cuts, a 1 l = 6, a = 0.17 fm. The cuts are shown as thick solid lines. The topology of lattices with two cuts is
illustrated on Fig. 1.
periodical identification, and lattice gauge theory is defined on the obtained links. It turns out that
lattices with such topology can be entirely mapped on sNt NL3 hypercubic lattices so that all
sDNt NL3 links and s D(D1)
Nt NL3 lattice plaquettes are used. To reduce the anisotropy due
2
to the cuts, in our simulations at each Monte Carlo iteration the heatbath and the over-relaxation
procedures are applied to 2sDNt NL3 randomly chosen links.
To check the physical scaling of the first, UV divergent term in (3), we have performed lattice
Monte Carlo simulations on lattices with cuts at different values of l and various lattice spacings,
lattice sizes and numbers of cuts. We have used from 30 to 50 lattice configurations obtained
on lattices with sizes from 28 243 to 18 123 and with lattice spacings from a = 0.06 fm to
a = 0.17 fm. The number of cuts was equal to 2 for all lattices except for the 18 123 lattice,
which had three cuts. The action density for a given lattice site was obtained by averaging the
action of all plaquettes to which this site belongs. The distributions of the average action density
in the (x 0 , x 1 ) plane of the sNt NL3 toric lattices which cover the original lattices with cuts
are plotted on Fig. 2 for different lattice parameters. The cuts are shown as thick solid lines.
465
Fig. 3. Average excess of action on the lattice plaquettes which are closest to the branching points on different lattices
with two and three cuts.
Numerically the average plaquette action appears to be much higher in the immediate vicinity
of the branching points, namely, for the lattice plaquettes which are closest to them. Since in
four dimensions the branching points are situated on the two-dimensional boundary of A, one
can expect that this excess of action near the branching points is related to the first, quadratically
divergent term in the entanglement entropy (3). In order to check this, the average action density
was measured for those sites on the sNt NL3 toric covering lattice which are closest to each
branching point (see Fig. 2, where the action density is maximal in these points) for different
lattices with two and three cuts. According to (5), (6) and (8), the average action density for
lattices without cuts was subtracted. The results of these measurements are plotted on Fig. 3 as
the function of lattice spacing a. One can see that the average action excess practically does not
depend on the lattice spacing, and that for two and three cuts the results are the same within the
range of statistical errors. Since at fixed lattice size the branching points cover the surface with
total area sNL2 a 2 on the covering sNt NL3 lattice, the total action excess associated with the
boundary of A includes the quadratically divergent term sa 2 |A|. According to (6), the same
is true for F [A, s, T ]. Differentiating over s, we thus recover the first, quadratically divergent
term in (3). Since the divergence in the total average action comes from the vicinities of the
branching points, it is clear that the divergent term in the free energy (6) and the entanglement
entropy (5) does not depend on l.
4. UV finite part of the entanglement entropy
In this section we present the results of our measurements of the derivative of the UV finite
part of the entanglement entropy and point out a signature of its discontinuity.
As discussed in Section 1, only the nondivergent part Sf (l) of the entanglement entropy depends on l. In our simulations we have found that it is very difficult to extract Sf (l) from the
measurements of the average action excess using (6) and integrating the average action over
(see Section 2). To overcome this difficulty, we implement the method proposed in [22,23]
and directly measure the differences of the free energies F (l + a, 2, T ) F (l, 2, T ) in (8) at
different l. Another advantage of this method is that all ultraviolet divergences are automatically
canceled because the UV divergent part of the entanglement entropy does not depend on l, and no
466
Table 1
S (l)
The data points which were used to estimate the derivative l
f
a (fm)
a 1 l
# conf.
a (fm)
a 1 l
# conf.
2.35
2.35
2.35
2.40
2.40
2.40
0.1394
0.1394
0.1394
0.1193
0.1193
0.1193
2
4
5
3
4
5
38
19
97
98
98
98
2.425
2.45
2.45
2.45
2.50
0.1113
0.0996
0.0996
0.0996
0.0854
5
2
4
5
2
98
49
97
141
18
subtractions of F (T ) should be made in (8). The method itself is briefly described in Appendix
B. All measurements were performed on the 24 123 lattice, and the distance l was varied both
by changing the lattice spacing a at fixed l/a and by changing l/a at fixed a. Lattice spacings
and the lengths l used in our simulations are listed in Table 1, together with the number of lattice
configurations used to obtain each data point. The method of [22,23] involves integration of the
outcome of Monte Carlo simulations over a certain parameter [0, 1] (see Appendix B). To
perform this integration we have calculated the integrand at = 0, 0.2, 0.4, 0.6, 0.8, 1.0, and the
numbers of lattice configurations in Table 1 are the numbers of configurations for each of these
values of .
Finite differences F (l + a, 2, T ) F (l, 2, T ) were then used to estimate the derivative
1
|A| l Sf (l) in (7):
F (l + a, 2, T ) F (l, 2, T )
1
Sf (l + a/2)
.
|A| l
a|A|
(9)
1
The derivative |A|
l Sf (l) estimated from these measurements is plotted on Fig. 4. It can be
seen that this derivative grows rapidly at small distances. For comparison with the asymptotic
behavior Sf (l) l 2 at l 0, we have fitted these results by the function Cl 3 (solid line on
Fig. 4). For the data points with the smallest l the finite differences (9) were found at fixed l/a
at different a, so that the finite differences (l + a/2)2 (l a/2)2 still behave as l 3 a 3 .
The estimated ratio of derivatives l Sf (2a2 )/ l Sf (2a1 ) = (2.0 0.4) at a1 = 0.0996 fm and
a2 = 0.0854 fm is indeed close to the ratio (a1 /a2 )3 = 1.586, which is an indication that at small
l the derivative l Sf (l) indeed scales in physical units of length.
At larger l l Sf (l) goes to zero faster than l 3 , and seem to approach a kind of plateau for the
values of l between 0.3 fm and 0.5 fm. Here the values of l Sf (l) obtained for different values
of lattice spacing differ rather significantly, which indicates that for our lattice parameters finitevolume and finite-spacing effects may still be rather strong. Nevertheless, at least qualitatively
all data points for different values of a display the same behavior. We have also tried to plot the
entropic C-function as the function of l (see Fig. 6). It also has a distinct discontinuity, however,
it grows slowly for intermediate values of l and thus its behavior seems to disagree with general
theoretical expectations (see Section 1). In our opinion this growth is simply the artefact of the
finite differences that we have used to estimate the derivative of the entanglement entropy over
2
l. For example, if the finite difference
x
x with
x = 1 is used to estimate the function
d 2
2
x , we obtain the function x 3
x
x =x
x 3 dx
x3
,
(x+1)2
which indeed grows slowly.
What is most interesting, however, is that at lc 0.5 fm the estimated derivative l Sf (l)
and the entropic C-function rapidly go to zero, and remain equal to zero within error range
for larger values of l. Data points near this lc are plotted with larger scale on Fig. 5. This is
S (l)
f
1
Fig. 4. The dependence of the derivative of the entanglement entropy |A|
l
3
the function Cl .
467
on l. Solid line is the fit of the data by
Fig. 5. The discontinuity of the derivative of the entanglement entropy over l near lc 0.5 fm.
a clear signature of the discontinuity of the derivative of the entanglement entropy over l, thus
the entanglement entropy indeed seems to be nonanalytic in l even at finite number of colours.
Of course, it should be remembered that these results were obtained under some simplifying
assumptions, in particular, the contributions of the free energies F [A, s, T ] with s > 2 to the
entanglement entropy (5) were neglected. One cannot completely exclude the possibility that
once higher s are included in the analysis, the dependence of the entanglement entropy on l may
be changed. It could be therefore interesting to perform similar measurements on larger lattices
with larger number of cuts.
5. Confinementdeconfinement phase transition w.r.t. the size of A at finite temperatures
If the expression (5) is applied to theories at finite temperatures, very straightforward arguments suggest that the nonanalytic behavior of the entanglement entropy with respect to the size
of A may persist even at temperatures above the temperature of the confinementdeconfinement
468
l 3 S(l).
Fig. 6. Entropic C-function C(l) = |A|
l
phase transition. In this case it is also very simple to demonstrate the relation between the deconfinement phase transition at T = Tc and the discontinuity of the derivative of the entanglement
entropy over l. Indeed, consider gauge theory at some temperature T > Tc . It is clear that when
A becomes very large and extends through the whole physical space, the free energy F [A, s, T ]
is equal to the free energy of the theory at the temperature T /s, and one can always find such
s that T /s < Tc and the theory is in the confinement phase. On the other hand, if the size of A
tends to zero, F [A, s, T ] is just the free energy of s copies of the gauge fields at the temperature T , each of which is in the deconfinement phase. Since for the confinement-deconfinement
phase transition in pure SU(2) lattice gauge theory there is an exact order parameter associated
with Z2 center symmetry, namely, the Polyakov loop, one can expect that the transition between
confinement and deconfinement phases should be associated with some discontinuity in l, so
that the center symmetry is broken spontaneously for l < lc . If this is true, at this transition
the free energy F [A, s, T ] should change as at the deconfinement phase transition at T = Tc ,
which is of the second order for pure SU(2) YangMills theory. Thus the second derivative of
F [A, s, T ] over l should change stepwise at some l = lc (s), and for the values of l close to lc
we can write F (A, s, T ) (lc (s) l)2 (lc (s) l) + ( ), where ( ) denotes the analytic part
(0)
of F (A, s, T ). Differentiating over s and l, we find that if the limit lc = lims1 lc (s) is different from zero, the derivative of the entanglement entropy over l should have a discontinuity of
(0)
lc (s). However, even if lims1 lc (s) = 0 and the enthe form l Sf [A] (lc l) lims1 s
tanglement entropy is analytic in l, one can still discuss the nonanalytic behavior of the purity
P = Tr A2 and its generalizations like PA (s) = TrA As , which also indicates that near lc (s) quantum entanglement between the regions A and B is quickly destroyed. In general all the quantities
PA (s) are nonanalytic in l at different lc (s), which is a typical situation when a phase transition
turns into a crossover and different order parameters undergo transitions at different temperatures. It is also interesting to note that such transitions with respect to the size of A can occur in
any theories with phase transitions at finite temperatures.
To study this transition from deconfinement to confinement, we have measured the expectation
value of the Polyakov line which winds around the cycle of length (aT )1 = 4 on the 8 203
lattice with two cuts of variable length (see Fig. 1) at = 2.40. If the cut goes through the whole
lattice, we have two lattices with Nt = 4, for which the confinementdeconfinement transition
469
Fig. 7. The expectation value of the Polyakov loop which winds around the short cycles of the lattice as the function
of l for 8 203 lattice with two cuts, a = 0.12 fm, T /Tc = 1.43.
occurs near = 2.29 [24], therefore at = 2.40 we are indeed in the deconfinement phase.
On the other hand, with no cuts the time extent of the lattice is Nt = 8 and the deconfinement
transition occurs at = 2.51 [24], thus we are in the confinement phase. Expectation value of
the Polyakov loop as the function of l is plotted on Fig. 7. We see that at l approximately equal
to 1.5 fm, the expectation value of the Polyakov loop has a sharp bend, after which it is equal to
zero within error range. Thus center symmetry is restored and we are indeed in the confinement
phase. Such dependence of the Polyakov loop on l is qualitatively the same as its dependence on
temperature at the usual confinementdeconfinement phase transition in pure SU(2) YangMills
theory, in agreement with the predictions of [1215].
6. Conclusions
In this work we have studied numerically the behavior of the entropy of entanglement between a three-dimensional slab A of thickness l and its complement for SU(2) lattice gauge
theory. This entropy is found to be a sum of the quadratically divergent and the UV finite terms
SUV and Sf (l), both of which are proportional to the area of the boundary of A. The behavior
of the nondivergent term was compared with the results of recent theoretical works [14,15,21],
where a discontinuity of the derivative of the entanglement entropy at some critical value l = lc
was predicted. We have indeed found a signature of such discontinuity at lc 0.5 fm, where the
entanglement entropy extracted from finite differences (9) and (8) rapidly goes to zero. Thus the
predictions of [14,15] are confirmed for finite Nc , at least qualitatively. It seems that such behavior of entanglement entropy is a common feature of all confining gauge theories, similarly to the
confinementdeconfinement phase transition at high temperatures. Since the entropic C-function
C(l) measures the number of effective degrees of freedom at the energy scale E l 1 , the results presented in [14,15] and in this paper imply that the transition from asymptotic freedom to
confinement is always rapid, either when the temperature or simply the energy scale are varied.
In other words, gluons and glueballs cannot coexist at one energy scale, and the properties of
gauge theories should be described either in terms of coloured or colourless particles, but never
in terms of both.
470
Fig. 8. A conjectured line of phase transitions in the lT plane.
At finite temperatures above Tc we have found that for lattices with cuts there is a real phase
transition from deconfinement at small l to confinement at large l. This transition is associated
with spontaneous breaking of the Z2 symmetry of the Polyakov loop, just as in the case of the
usual confinementdeconfinement phase transition. As is argued in Section 5, such transition
should also be associated with a nonanalytic behavior of the purity P = Tr A2 of the reduced
density matrix A . More generally, for lattices with s cuts at T > Tc the deconfinement phase
transition w.r.t. l at some lc (s) results in a nonanalytic behavior of the quantity PA (s) = Tr As .
Since in general all lc (s) are different, different measures of entanglement yield different transition points, which is a typical behavior for crossover. Thus it reasonable to conjecture that the line
of phase transitions w.r.t. l in the lT plane can be continued above Tc as a sort of crossover line
(see Fig. 8), which is associated with rapid but continuous destruction of quantum entanglement.
It can be interesting to investigate how such finite-temperature phase transition for the entanglement entropy can be described in the framework of AdS/CFT correspondence, and how the
geometry of the minimal hypersurface in the bulk which spans on the boundary of A changes at
this transition. On the other hand, a more detailed study of the line of confinementdeconfinement
phase transitions in the lT plane in lattice or continuum gauge theories can also provide some information on higher-dimensional geometries which are dual to confining gauge theories. Finally,
we would like to explain why, in our opinion, the confirmation of the predictions of [1215]
is important methodologically. Until recently the development of the AdS/QCD phenomenology was based on the attempts to fit some known data with the results of calculations of some
geometric quantities in the dual higher-dimensional geometries. In contrast, the nonanalytic behavior of the entanglement entropy was not previously known for confining gauge theories and
was originally found in [14] using the dual description, which demonstrates the predictive power
of the AdS/QCD approach.
Acknowledgements
The authors are grateful to V.I. Zakharov and T. Takayanagi for interesting discussions and for
bringing the papers [15] and [14] to our attention, and to E.T. Akhmedov and F.V. Gubarev for
useful remarks. This work was partly supported by grants RFBR 06-02-04010-NNIO-a, RFBR
08-02-00661-a, DFG-RFBR 436 RUS, grant for scientific schools NSh-679.2008.2 and by Federal Program of the Russian Ministry of Industry, Science and Technology No. 40.052.1.1.1112
and by Russian Federal Agency for Nuclear Power.
471
Appendix A. Path integral representation of the entanglement entropy

This appendix revises the derivation of the path-integral representation for the entanglement
entropy, which was originally obtained in [4]. We first represent the entanglement entropy in the
following form:
ln TrA As .
s1 s
SA = lim
(A.1)
To obtain the reduced density matrix A = TrB AB = TrB |00| of the ground state of the
theory, it is convenient to consider the theory at some finite temperature T , so that AB =
Z 1 (T ) exp(T 1 H), where H is the Hamiltonian of the theory and Z(T ) = Tr exp(T 1 H)
is the partition function. At the end of the calculation one can take the limit T 0.
We denote the set of all fields in the theory as (
x ), where x denotes spatial coordinates,
omitting all indices for the sake of brevity. For gauge theories one should also include into
the appropriate ghost sector. According to the FeynmanKac path integral formula [25], matrix
between the eigenstates |
(
elements of exp(T 1 H)
x ), |
(
x ) of the field operators are
given by the path integral over all fields on a section of four-dimensional Euclidean space of
time extent T 1 , with corresponding eigenvalues set as boundary conditions at t = 0 and t = T 1
[25]:

AB
(
x ),
(
x ) =
(
x )AB
(
x)
= Z 1 (T )

T 1

3
D(
x , t) exp
dt d x L (
x , t) ,
(
x ,T 1
)= ( x )
(
x ,0)=
(
x)
(A.2)
where L[(
x , t)] is the Lagrangian of the theory.
In order to find the reduced density matrix for the region A, we should somehow trace over all
degrees of freedom in the complement B to the region A. Assume that the theory is regularized
by discretizing the spatial coordinates
x . In this case it is always possible to rewrite the states

|(
x ) as a direct product x |x , where |x are the eigenstates of the field operator in the
point x . It is then natural to define the trace over all degrees of freedom in B as a sum over all
products of basis states |x with x B, so that the reduced density matrix A reads:

x ),
(
x) =
D(
x ) (
x )
(
x )AB
(
x ) (
x)
A
(
x B
=Z

(T )

T 1

3
D(
x , t) exp
dt d x L (
x , t) ,
(A.3)
x ),
where in the last path integral the following boundary conditions are imposed: (
x , 0) =
(
(
x , T 1 ) =
(
x ) for x A, and (
x , 0) = (
x , T 1 ), if x B. Matrix multiplication can also
be defined in terms of the path integral over (
x ):

A2
(
(A.4)
x ),
(
x ),
(
x) =
D(
x )A
(
x ), (
x ) A (
x) .
x A
472
It is now straightforward to calculate the trace and the matrix contractions in the definition of the
function PA (s) = Tr As :

D1 (
x ) Ds (
x)
Tr As =
x A
x ), 2 (
x ) A 2 (
x ), 3 (
x ) A s (
x ), 1 (
x) .
A 1 (
(A.5)
Tr As
Putting together the expressions (A.3) and (A.5), one sees that to calculate
in (A.5), one
should impose periodic boundary conditions with period sT 1 in time if x A and with period
T 1 if x B, and find the partition function by integrating over all such fields. We thus arrive at
the following representation for Tr As :
Z[A, s, T ]
(A.6)
,
Z s (T )
where Z[A, s, T ] is the partition function obtained by integrating over all fields which satisfy the
x -dependent periodic boundary conditions, as discussed above. The topology of space on which
the fields which enter the path integral for Z[A, s, T ] live is schematically shown on Fig. 1 for
s = 2.
Inserting the expression (A.6) into (A.1) and taking the limit T 0, we arrive at the following
representation of the entanglement entropy:

S[A] = lim lim F [A, s, T ] F (T ) ,

(A.7)
T 0 s1 s
where F (T ) = ln Z(T ) and F [A, s, T ] = ln Z[A, s, T ] are the free energies which correspond to the partition functions Z(T ) and Z[A, s, T ].
Tr As =
Appendix B. Direct Monte Carlo measurements of the differences of free energies

In this appendix we briefly review the method proposed in [22,23] for direct measurements
of the differences of free energies in lattice Monte Carlo simulations. The works [22,23] deal
mainly with the case of finite-temperature lattice gauge theories, so it may be useful to repeat
their arguments in a somewhat more general context.
Assume that we have two different actions S1 [] and S2 [] for some set of fields defined on
the lattice of some fixed size. Again, here denotes any lattice fields, which can live on lattice
sites, links, cubes, etc., as long as the number of these lattice simplices which enter both actions
remains the same. For instance, this is true for gauge fields on the lattices of special topology,
which are described in Section 3.
We would like to calculate the difference F2 F1 of the free energies F1 = ln Z1 and
F2 = ln Z2 , where Z1 and Z2 are the partition functions calculated by integrating over all
fields with the weights exp(S1 []) and exp(S2 []), respectively. To this end, define the
interpolating partition function Z(), so that Z(0) = Z1 and Z(1) = Z2 :

Z() = D exp (1 )S1 [] S2 [] .
(B.1)
Now the difference F2 F1 = ln Z1 ln Z2 can be represented in the following form:
1
F2 F1 =
d
0
ln Z().
(B.2)
473
Fig. 9. The dependence of the integrand in (B.3) on for the difference of free energies on 24 123 lattices with two
cuts of length 10a and 11a, lattice spacing a = 0.085 fm.
Calculating the derivative over from (B.1), we arrive at the following result:
1
F 2 F1 =

d S2 [] S1 [] ,
(B.3)
where the average is defined by integrating over all fields with the weight Z 1 ()
exp((1 )S1 [] S2 []). For any [0, 1], the average S2 [] S1 [] can be calculated
with arbitrary precision by increasing the number of Monte Carlo iterations. After that some
numerical integration method should be used to find the integral in (B.3). In practice it turns out
that although for each the expectation value S2 [] S1 [] is numerically rather large, the
contributions from < 1/2 and > 1/2 cancel almost exactly. For example, the dependence of
the integrand in (B.3) on for the difference of free energies on 24 123 lattices with two cuts
with l = 2a and l = 1a is plotted on Fig. 9. Thus the expectation value S2 [] S1 [] should
be found with a very good precision, which typically requires rather large number of Monte Carlo
iterations. However, as discussed in the body of the text, in our case the described method seems
to be the best way to measure free energies.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
G. Vidal, J.I. Latorre, E. Rico, A. Kitaev, Phys. Rev. Lett. 90 (2003) 227902, quant-ph/0211074.
P. Calabrese, J. Cardy, Int. J. Quantum Inf. 4 (2006) 429, quant-ph/0505193.
T.J. Osborn, M.A. Nielsen, Quantum Inf. Proc. 1 (2002) 45, quant-ph/0109024.
P. Calabrese, J. Cardy, J. Stat. Mech. 0406 (2004) 002, hep-th/0405152.
A. Kitaev, J. Preskill, Phys. Rev. Lett. 96 (2006) 110404, hep-th/0510092.
T.J. Osborne, M.A. Nielsen, Phys. Rev. A 66 (2002) 032110, quant-ph/0202162.
C.H. Bennett, H.J. Bernstein, S. Popescu, B. Schumacher, Phys. Rev. A 53 (1996) 2046, http://link.aps.org/abstract/
PRA/v53/p2046.
M. Srednicki, Phys. Rev. Lett. 71 (1993) 666, hep-th/9303048.
D.V. Fursaev, Phys. Rev. D 73 (2006) 124025, hep-th/0602134.
S.W. Hawking, Commun. Math. Phys. 31 (1973) 161.
J.D. Bekenstein, Phys. Rev. D 12 (1975) 3077, http://prola.aps.org/abstract/PRD/v12/i10/p3077.
S. Ryu, T. Takayanagi, Phys. Rev. Lett. 96 (2006) 181602, hep-th/0603001.
474
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
S. Ryu, T. Takayanagi, JHEP 0608 (2006) 045, hep-th/0605073.

T. Nishioka, T. Takayanagi, JHEP 0701 (2007) 090, hep-th/0611035.
I.R. Klebanov, D. Kutasov, A. Murugan, Entanglement as a probe of confinement, http://arxiv.org/abs/0709.2140.
J.M. Maldacena, Int. J. Theor. Phys. 38 (1997) 1113, hep-th/9711200.
A.B. Zamolodchikov, JETP Lett. 43 (1986) 730.
E. Witten, Adv. Theor. Math. Phys. 2 (1998) 253, hep-th/9803131.
I.R. Klebanov, M.J. Strassler, JHEP 0008 (2000) 052, hep-th/0007191.
J. Polchinski, M.J. Strassler, Phys. Rev. Lett. 88 (2002) 031601, hep-th/0109174.
A. Velytsky, Entanglement entropy in d + 1 SU(N ) gauge theory (2008), http://arxiv.org/abs/0801.4111.
Z. Fodor, PoS LAT2007 (2007) 011, http://arxiv.org/abs/0711.0336.
G. Endrodi, Z. Fodor, S.D. Katz, K.K. Szabo, PoS LAT2007 (2007) 228, http://arxiv.org/abs/0710.4197.
J. Fingberg, U. Heller, F. Karsch, Nucl. Phys. B 392 (1993) 493, hep-lat/9208012.
J. Kapusta, Finite-Temperature Field Theory, Cambridge Univ. Press, 1989.

Towards the construction of local logarithmic conformal

field theories
Anne-Ly Do, Michael Flohr
Institute for Theoretical Physics, University of Hannover, Appelstrae 2, D-30167 Hannover, Germany
Received 15 December 2007; received in revised form 1 May 2008; accepted 5 May 2008
Abstract
Although logarithmic conformal field theories (LCFTs) are known not to factorise many previous findings
have only been formulated on their chiral halves. Making only mild and rather general assumptions on the
structure of an chiral LCFT we deduce statements about its local non-chiral equivalent. Two methods are
presented how to construct local representations as subrepresentations of the tensor product of chiral and
anti-chiral Jordan cells. Furthermore we explore the assembly of generic non-chiral correlation functions
from generic chiral and anti-chiral correlators. The constraint of locality is studied and the generality of our
method is discussed.
1. Introduction
Since the spadework of Gurarie [1], an enormous amount of work was done to evolve logarithmic conformal field theories (LCFTs). Their applications to other fields in physics, for example
to the theory of percolation and critical disordered systems have been explored. Many structural
aspects have been studied in detail. In particular the powerfull techniques from non-logarithmic
CFT have been ported to LCFT. The decoupling of the conformal symmetry algebra in two independent sectors of opposite chirality gives reason to a especially elegant and beneficial technique
known from the non-logarithmic case: Many statements about CFTs can be derived considering
only one sector, either the chiral or the anti-chiral. This means that objects are considered which
transform effectively in a left- or right-handed representation. In case of such a constriction on
half objects it is common to speak about chiral theories. Results of one sector can be trans* Corresponding author.
E-mail addresses: lydo@itp.uni-hannover.de (A.-L. Do), flohr@itp.uni-hannover.de (M. Flohr).

doi:10.1016/j.nuclphysb.2008.05.001
476
A.-L. Do, M. Flohr / Nuclear Physics B 802 [FS] (2008) 475493
ferred to the other one by mirroring the chirality. Subsequently both halves have to be glued
together to form the full, non-chiral theory.
Among the findings which were up to now stated for chiral LCFTs only, the following topics
are of central interest: Correlation functions are calculated for instance in [211], fusion rules are
investigated, among others, in [1216] and some studies on null vectors can be found in [17,18].
Enlarging their scope to non-chiral theories and implementing the constraints of locality is a task
of elementary importance, as only local theories have physical interpretation. Unfortunately, the
construction of local theories turned out to be non-trivial: Unlike conventional, non-logarithmic
CFTs, LCFTs were found to be non-factorisable. Insofar speaking about chiral halves in the
context of LCFTs might be delusive. As we will see, assembling a local theory out of a chiral
LCFT and its anti-chiral counterpart is evocative of screwing them into each other rather than of
combining two halves. Anyhow, we will stick with the familiar naming convention and refer in
abuse of language to the chiral theories as chiral halves.
So far, only few attempts have been made to close the gap between the well-known chiral
and the almost unknown non-chiral LCFTs. Gaberdiel and Kausch succeeded in constructing
a non-chiral local theory at c = 2 by solving the conformal bootstrap [19]. This construction
was interpreted in terms of symplectic fermions [20] and enabled a detailed comparison of the
two-dimensional Abelian sandpile model with the local triplet theory at c = 2 [21]. The logarithmic triplet theory with boundary was studied in [22] for c1,2 = 2 and the generalisation
to rational c1,p models was achieved in [23]. Stating remarkable structural similarities between
the local triplet theory and supergroup WZNW models, Schomerus et al. suggested that consulting the better understood local WZNW models might promote the construction of generic
local LCFTs. Local logarithmic bulk correlation functions for the GL(1|1) WZNW model have
been computed in [24], non-chiral indecomposable representations on which the zero-mode of
the energy-momentum tensor is not diagonalizable were investigated by means of the WZNW
model on the supergroup PSU(1, 1|2) [25]. A rather general discussion on conclusions for local
LCFTs from the supergroup WZNW point of view can be found in [26].
In this paper, we will choose an approach which rests solely upon the analysis of the conformal
symmetry and aims for a high degree of generality. Beside some rather general assumptions on
the structure of an chiral LCFT the deduced statements about its local non-chiral equivalent can
be called generic.
The paper is organized as follows: In Section 2 we fix the naming convention. Section 3
presents two methods how to construct the space of states of a local LCFT. Both methods predict
a non-chiral local theory to possess the same rank as the halves it is composed of. In other
respects the results of both methods turn out to be mutually excluding. The consideration of a
specific model suggests one of the proposed methods to be more intuitive. From Section 4 we
turn our attention to LCFT correlation functions. After a brief recapitulation about generic chiral
correlation functions, we explore the possibility to compose generic non-chiral correlators out
of generic chiral ones. Beside the constraint of locality, invariance under the global conformal
group, duality and monodromy invariance have to be implemented. We propose a construction
method, discuss the generality of our solution and check its consistency with previous findings.
More details on this work can be found in [27].
2. Definitions and preliminaries
Logarithmic conformal field theories feature indecomposable but reducible representations of
the chiral symmetry algebra. In this paper we will consider only the case where such indecompos-
477
able representations occur with respect to the Virasoro zero modes. Through appropriate choice
of the basis, the generators L0 and L0 can be transformed into Jordan normal form. A rank r
Jordan cell is spanned by r fields {(h,r1) , . . . , (h,1) , (h,0) }. The action of the zero mode L0
of the Virasoro algebra is then given by
L0 (h,k) (0)|0 = h(h,k) (0)|0 + (1 k,0 )(h,k1) (0)|0,
where h is the conformal weight and |0 as usual denotes the SL(2, C) invariant vacuum. The
parameter k grades the fields within the Jordan cell and will be referred to as Jordan level. In the
scope of this paper, only reducible representations are regarded whose irreducible subrepresentations (hi ,0) (zi ) accord to proper primaries, i.e. to fields whose OPE among each other does
never produce logarithmic fields on the right-hand side. Furthermore, we assume the logarithmic
partner fields of the proper primary to be quasi-primary, i.e. Ln (hi ,ki ) = 0 n 1. Of course,
our assumptions limit the generality with which our results are valid. This is particularly true for
our assumption on the quasi-primarity of the logarithmic partner fields. We will discuss these
limitations and how they may be overcome in the conclusions. Thus, our paper should be understood as a first step towards a full treatment of the question of local logarithmic conformal field
theories.
If an LCFT accommodates more than one indecomposable representation one can consider
every Jordan cell to be of rank r(h) = r, with r being the rank of the largest Jordan cell. Hypothetically emerging smaller Jordan cells can be padded with fields that are to be set zero
afterwards.
The following decomposition of the non-diagonal action of the Virasoro modes on LCFT
n-point correlators will turn out to be advantageous:

n

q
zi (q + 1)hi ,
(2.1)
Lq = Oq +
i=1
where Oq abbreviates the diagonal part of the action as known from ordinary non-logarithmic
CFT
Oq =
n

q
zi zi i + (q + 1)hi .
i=1
The off-diagonal, nilpotent part is generated by operators hi . They act on a logarithmic field by
reducing its Jordan level by one, on a primary by annihilating the field:
hj (hi ,ki ) (zi ) = ij (1 0,ki )(hi ,ki1 ) (zi ).
Below we will augment the introduced glossary with entities marked with a bar. These can be
obtained from those without a bar by complex conjugation of all variables z and providing all
parameters h, k and g with a superscript line. For h, k and g this line is not related to complex
conjugation, but only indicates that parameters of the anti-chiral theory are denoted.
To distinguish quantities of a full local theory from those living on its right- or left-handed
half we will apply the pair of concepts non-chiralchiral. If used in this sense the latter shall
cover anti-chiral quantities, too.
478
2.1. Locality constraints

The fundamental postulation on a non-chiral theory is locality of the fields or, in terms of
correlation functions, singlevaluedness of the amplitudes. In [19] it was shown that this imposes
the following constraints:
hi h i Z,
Si i = 0,
(2.2)
where i abbreviates (hi ,h i ,ki ) and Si shortens hi h i . Here we introduced ki in foresight

as non-chiral Jordan level. As the notion Jordan level rests upon the Jordan block structure of
the zero modes of the chiral symmetry algebra, we need a redefinition in terms of non-chiral
representations, which will be given in Section 3.
3. Non-chiral local representations
Non-chiral irreducible representations can be obtained as diagonal tensor product of irreducible chiral representations [28]. This course of action fails in case of non-chiral indecomposable representations. A chiral indecomposable representation Rh of Jordan rank r + 1 is
generated by states |h, ki , ki running from r to null. The various tensor products included in
Rh Rh can be endowed with a gradation: Starting with the tensor product of those states with
highest Jordan level, all other possible tensor products can be obtained by repeated application
of h and h . The resulting structure is summarised in Fig. 1.
Here the following abbreviations were introduced: An element of the product space Rh Rh
is uniquely denoted by its chiral and anti-chiral Jordan level
k .
k k := |h, k |h,
Let
indicate that the state the arrow points at is an image of the state the arrow starts at
under the action of h . The symbol does the same for the action of h . Level n = r m is
subsequently assigned to states that are obtained from r r by m-fold descending, i.e. by m-fold
application of delta operators. Starting with one state on level r, the number of states with level
Fig. 1. Gradation of Rh Rh .
479
n increases gradually with decreasing n 0. Level zero possesses the highest number of states,
namely r + 1. For level n 0 those states located on the fringes of the above graphic are deleted
either by h or by h , hence the number of states on level n decreases with decreasing n 0:
number of states in Rh Rh with level n = r + 1 |n|,
|n| r.
Due to the field-state isomorphism |h, k := limz0 (h,k) (z)|0 Eq. (2.2) enforces that the action of the operator S = h h has to vanish on all non-chiral states belonging to local theories.
Fig. 1 shows that generally the image of a tensor state under the action of h differs from its image under h . As a consequence states of the local non-chiral theory span only a subspace of the
space generated by the diagonal tensor product of the chiral and anti-chiral Jordan cells [19]. This
local subspace Lhh can be constructed in two ways. Firstly, we will revisit the quotient space
construction which Gaberdiel and Kausch introduced for the c = 2 local triplet theory [19]
and enlarge the scope of their findings to theories whose chiral halves are of arbitrary rank. We
would like to point out that a generalisation was achieved earlier but under a different viewpoint:
Starting off at a single boundary condition Gaberdiel and Runkel constructed the compatible
local space of bulk states [23]. Their course of action holds for rational CFTs, logarithmic and
non-logarithmic, and was shown to reproduce the known local bulk theory at c = 2 if applied
to the c1,2 triplet model. Our approach towards an enhanced quotient space construction differs
from [23] inasmuch it does not use any information about boundaries.
Secondly, we will introduce the kernel construction and compare the results of both methods.
3.1. Quotient space construction
Locality requires that the image of a local state under the action of h has to equal the image
of the same state under the action of h (2.2). The quotient space construction presented here
approaches the problem by identifying both images modulo elements of a subspace Nhh :
LhhQSC
(Rh Rh )/Nhh .
(3.1)
The chiral version of the c = 2 triplet theory possesses two indecomposable representations of
rank two. For these cases, Gaberdiel and Kausch identified the subrepresentation spanned by S
acting on the tensor product of states with Jordan level one and the descendants of this state with
respect to h and h to be a minimal choice for Nhh [19]:

, h = h = 0, 1.
Nhh = (h )p (h )q S(1 1)
This result can be generalised to local theories whose chiral halves exhibit Jordan cells of arbitrary rank r + 1. For that purpose we repeat the procedure used to gain the gradation of Rh Rh
with a different starting point: S acting on the tensor product of those states with highest Jordan
level. The structure of the resulting subrepresentation Shh is depicted on the right-hand side of
the chart given below (Fig. 2).
On the left-hand side the gradation of Rh Rh is recapitulated: Every marks an element of
the product space. As per construction the formation on the right-hand side is generated by one
state S(r r ) = (r 1) r r (r 1) on level r 1. This state and every state that emanates
from it by application of delta operators are pictured as . Counting the states Shh with level n
yields
number of states in Shh with level n = r + 1 |n + 1|,
|n| r 1.
480
Fig. 2. Rh Rh versus Shh .
Fig. 3. Comparison of Rh Rh and Shh .
Level n = 1 accommodates the maximum number of states. Level n 1 states that are located on the fringes vanish either under the action of h or of h .
Comparing the state content of both representations one finds: For |n| r 1, the difference
of two adjacent level n states in Rh Rh is element of Shh . Additionally, for level n 1 the
skirting states of both representations are pairwise identical. This is summarized in Fig. 3.
Hence dividing the subspace Shh out of Rh Rh accords with identification of all those states
Rh Rh that exhibit identical level:
(h )pi (h )qi r r (h )pj (h )qj r r
i, j |pi + qi = pj + qj .
The action of h on these equivalence classes equals that of h . Thus indeed with Shh we constructed a minimal choice for Nhh .
481
The equivalence classes in (Rh Rh )/Shh are parameterised and arranged in order by their
level. A non-chiral level n state shall be defined as a representative of the equivalence class with
level n. As standard representative we choose the symmetric sum over all elements:

n :=
|h, h,
level n states.
(3.2)
To avoid confusion with the Jordan level naming of a chiral state, we will use bold numbers
to denote states of the non-chiral theory by their level. Since the equivalence classes with level
n 1 include representatives Shh , they are entirely removed from (Rh Rh )/Shh , i.e. states
Rhh at level zero are annihilated by both delta operators.
3.2. Kernel construction
The kernel construction defines LhhKC
Rh Rh as the kernel of S. The key idea for
determining the kernel of S is to use telescoping series. If we sum up all states Rh Rh with
same level and act with S on it, every term except the first and the last cancels with either the
preceding or succeeding term.

(n r ) (r n)
for n 1,
h
h
S
level n states =
for n 0.
h (0 (r n)) ((r n) 0)
h
As for level 0 the surviving terms vanish, we can conclude:

=
level n states | n 0 .
LhhKC
3.3. Discussion and summary

In the preceding sections we presented two methods how to construct local indecomposable
representationsthe quotient space construction (QSC) and the kernel construction (KC). According to both methods Lhh possesses the same rank as the chiral halves it is composed of:
rank(Lhh ) = rank(Rh ) = rank(Rh ).
The state content of a local representation depends on the method used to construct it:

LhhQSC
=
level n states | n 0 ,

LhhKC
=
level n states | n 0 .
Above, specifications were made such that for QSC the term non-chiral Jordan level is in perfect
accordance to the chiral Jordan level. For KC it is of avail to slightly adapt the definition of the
non-chiral Jordan level: Redefining level n to be level n + r
n n = n + r
we achieve the familiar situation of non-negative integer values for the non-chiral Jordan level.
Furthermore the redefinition guarantees that level zero states are annihilated by h and h :
n KC = |h, h,
n KC = 0 for n = 0.
h |h, h,
h
Computable predictions in QSC may depend on the choice of the considered representative.
The vanishing Shapovalov form of |0 for example can be shown by means of the representatives
482
However, rank r + 1 theories with r + 1 > 1 and r an even integer possess a

0 r and r 0.
level zero representative 2r 2r . According to Eq. (4.2) the corresponding Shapovalov form is
r r r r r r r r
= = 0.

(3.3)
2 2
2
2
2 2 2 2
In a rank r + 1 theory with r + 1 > 2 and r an odd integer the same problem occurs but on
r+1
level n = 1: Such a theory possesses a level one representative r+1
2 2 with non-vanishing
Shapovalov form. The crucial point is that though representatives of QSC equivalence classes
are as per construction equivalent with respect to the action of L0 and L 0 in other respects their
equivalence is not guaranteed. In every case is true: Representatives of QSC equivalence classes
may be chosen such that observables in both formulationsQSC and KCare identical. That is,
nQSC is equivalent
= LhhKC
both representations are isomorphic and |h, h,
even though LhhQSC
to |h, h, n KC . Hence, at this point it is both impossible and unnecessary to finally rule on the
question wether one method has to be preferred.
However, the known symplectic fermion realisation of the LCFT at c = 2 may be interpreted
as a hint on the KC to be the more natural method. A symplectic fermion a la Zamolodchikov is
a two-component fermionic field of spin zero [29]. The stress energy tensor of its free theory
is given by T (z) = 12 : :(z). The mode expansion of the component fields reads:

,n zn + ,0 log(z) + .
(3.4)
=
n=0
The s are Grassmann numbers and act as creation operators for the chiral logarithmic partner
of the identity:
1
|h = 0, k = 1 = |h = 0, k = 0 =: |0.
(3.5)
2
Of course an analogue identity holds for the anti-chiral half. Therewith we can give explicit
expressions for the sum of all states R0 R0 with equal level n:
It is possible to choose a basis such that = , i.e. the state |n = 1 vanishes due to the
nilpotency of . This coincides with the prediction of the kernel construction.
4. Correlation functions
The prominent role that correlation functions play in CFTs results from two facts: On the one
hand, they are related to observables and therefore represent a connection between theory and
accessible experimental data. On the other hand, they are considered fundamental from a pure
theoretical point of view: As shown in [30], a CFT is completely constituted if all correlation
functions are known. Given the two- and three-point functions of the fundamental fields, all
other amplitudes can actually be derived from these. Furthermore, the consistency conditions of
all amplitudes can be traced back to those obeyed by the four-point functions.
483
The calculation of correlation functions in LCFTs holds two major difficulties that do not arise
in case of non-logarithmic CFTs. Both have their origin in the non-diagonal action of the generators of the chiral symmetry algebra. The off-diagonal contribution enters the global conformal
Ward identities in the shape of an inhomogeneity and results in the aforementioned challenges:
Firstly the identification of the generic structure of chiral correlation functions compatible
to global conformal invariance is remarkably hindered. A hierarchical solution scheme for the
inhomogeneous Ward identities allows to explore the texture of the subset of correlators which
contain chiral fields residing in indecomposable representations whose irreducible subrepresentation corresponds to a proper primary field. For these cases, it is possible to fix the generic
structure of n-point functions up to structure functions of n 3 SL(2, C) invariant crossratios,
however only within sets of other correlators. Possible extensions of the hierarchical solution
scheme to pre-logarithmic fields and non-quasi-primaries are discussed in [3] and [9].
Secondly, correlation functions of an LCFT do not generally factorise into chiral and antichiral parts. This is also an immediate consequence of the inhomogeneous Ward identities which
is mirrored by the fact that only those correlators are factorisable that solve Ward identities
with vanishing inhomogeneity. Gurarie pointed out that even amplitudes not explicitly involving
logarithmic fields do not necessarily fall in this category [1]. Two attempts have been made to
adapt the knowledge about chiral correlation functions for non-chiral ones. One of us provided
a rule of thumb, how to generalise known chiral sets of correlation functions to local sets by
replacing all emerging variables zi zj =: zij by |zij |2 [31]. This approach preserves the full
generality of the chiral sets but obscures the interrelationship between chiral, anti-chiral and nonchiral amplitudes. Gaberdiel and Kausch succeeded in constructing a consistent set of amplitudes
for the local theory at c = 2 [19]. As their course of action rests crucially upon model specific
information like the operator product expansion (OPE) of the fundamental fields, it cannot be
transfered to the generic case.
Our proceeding will be as follows: We first recapitulate the necessary assumptions under
which generic chiral n-point functions can be calculated and briefly describe the hierarchical
solution scheme for these cases. Subsequently, we summarise the generic structure of the npoint functions found this way. In Section 4.2 we will give a short proof for the statement that a
non-chiral correlation function factorises if and only if the inhomogeneity of the Ward Identities
vanishes for the chiral correlators it is composed of. Finally we present an ansatz built solely
out of quantities enclosed in the chiral and anti-chiral sets of n-point correlators that allows the
construction of local n-point amplitudes.
Correlators of fields residing in the chiral (anti-chiral) half of an LCFT will be named chiral,
anti-chiral respectively. Even though chiral intrinsically describes propagation properties of
fields, we prefer this term to the adjunct holomorphic which is often chosen to indicate that a
function only depends on the formal variable z but not on z .
4.1. Chiral correlation functions
Correlation functions in LCFTs are invariant under the global conformal group. This stipulates
the generic texture of n-point correlators up to structure functions of n 3 SL(2, C) invariant
crossratios that are a priori undetermined. Evaluating equation (2.1) for q = 1, 0, 1 exhibits the
inhomogeneous global conformal Ward identities (GCWIs):
Oq =
n

i=1
zi (q + 1)hi ,
for q {1, 0, 1}.
(4.1)
484
From Eq. (4.1) follows immediately that correlation functions containing logarithmic fields
cannot be determined independently: Due to the action of hi , the generic structure of an npoint function including fields (hi ,ki ) (zi ) can only be specified within a framework of other
n-point functions containing fields (hi ,ki 1) (zi ). For a given set of n conformal weights hi
with r(hi ) = r exists a hierarchy of r n n-point functions with s = 0, . . . , n logarithmic fields
displaying varying Jordan levels ki = 1, . . . , r 1. The number of different correlation functions
of a set is actually reduced, because a correlator is non-zero only, if the sum over the Jordan
levels of the logarithmic fields it contains equals at minimum r 1 [3]:
k1 kn = 0
K :=
n
ki r 1.
(4.2)
i=1
To improve lucidity we shortened the naming of the correlators in Eq. (4.2):

k1 kn := (h1 ,k1 ) (z1 ) (hn ,kn ) (zn ) .
The identity (4.2) serves as starting point for a recursive construction of solutions of the
GCWIs (4.1). For total Jordan level K = r 1, the GCWIs are homogeneous and can be solved
as known from ordinary CFT. Successive increase
of the total Jordan level yields differential
equations for k1 kn with the inhomogeneity i hi k1 kn determined in foregoing steps
of the recursion.
One finds that n-point correlators which contain fields of rank r Jordan cells possess the
generic form:
k1 kn =
max
(zi zj )
ij
i<j
l
G
F{q}
(xa ) PG (lmn ),
lmn := ln(zmn ),
(4.3)
G=0
where PG denotes a sum over monomials of degree G:

PG =

|g()=G
j

c (lmi ni )gi =:
c p ,
i=1
j
gi = g().
(4.4)
i=1
The constraint of global conformal invariance (4.1) connects a coefficient c multiplying a monomial p in a correlator A to the coefficient c which multiplies a monomial p in a correlator

i hi A, with p being the image of p under the action of O0 . Four-point or higher correlation functions that exhibit logarithmic fields at every vertex may feature polynomials KG PG ,
whose multiplicities are special in the following sense: They are not cross-linked to any coefficients in other n-point functions of the set. Linkage to correlators with lower total Jordan level
is canceled if KG resides in the kernel of the operator O := (O0 , O1 ) [9]. No linkage to higher
correlators of the set has to be required separately. It follows that kernel terms KG may only arise
in the highest correlator of a set, i.e. in the correlator where all inserted fields are of maximum
Jordan level kn = r 1.
The occurring monomials p are subject to selection rules [31]: Logarithms in the correlators
stem from contractions of logarithmic fields. Hence, only such logarithms may arise whose indices refer to positions of fields with Jordan level k 1 within a correlator. Two more restrictions
rule the logarithmic terms:
[S1] The total logarithmic degree G in (4.3) is bounded above as follows:
G K r + 1 =: l max .
485
[S2] Each index mi may arise at most r 1 times within one monomial.
G (x ) denominate a family of functions, which solely depend on n 3 anharmonic
Let F{q}
a
ratios xa . The subscript {q} denotes a set of n indices qi which take integer values between zero
and ki :
G
F{q}
:= FqG1 q2 ...qn ,
qi {0, 1, . . . , ki }.
For fixed superscript index G, only those combinations {q} emerge that fulfill

qi + G = K.
i

Due to the cluster decomposition property all structure functions that satisfy qi = r 1 have
max
G
to be identified [9] and will be referred to as F l . For n 3 F{q} does not depend on the values

qi but only on i qi [3].
i (x) which repThe structure functions F{q} (x) may be decomposed in conformal blocks F{q}
resent the internal propagators:

i
F{q}
(x).
F{q} (x) =
i
During the main part of the paper this decomposition will not play any role for the presented
argumentation. To keep things simple we will abstain from making it explicitly where it is not
necessary.
The exponents ij in Eq. (4.3) solve

ij = 2hj .
i=j
So far we have stated properties of chiral correlation functions only. It is clear that analogues
propositions hold for anti-chiral correlators.
4.2. Assembling local amplitudes from the chiral sets
We can now bring the original query into sharper focus. Non-chiral amplitudes shall be obtained by multiplying suitable chiral and anti-chiral amplitudes. Therefore we have to revisit the
constraints of locality (2.2). As in the frame of this paper only the case hi = h i is considered, the
first condition does not cause concern. The second condition shall, for our purpose, be restated
as constraint on correlation functions. Using Eq. (4.1), one finds:
n
Si 1 n = (O0 O 0 )1 n = 0.
i=1
In non-logarithmic conformal field theories, the non-chiral amplitudes A can be achieved by
multiplying the chiral and anti-chiral amplitudes A and A:
A = AA.
(4.5)
In LCFTs, factorisation is contradictory to locality constraints except for correlators satisfying

homogeneous Ward identities:
486
(O0 O 0 )A = (O0 A)A A(O 0 A)

hi A A + A
h i A = 0
=
i

i
hi A =
h i A 0
(4.6)
where the first identity follows from the fact, that the operators O0 and O 0 act as derivatives with
respect to z (z respectively) on the function space, i.e. O 0 acting on the chiral amplitude does
not yield a contribution and vice versa. The second identity arises out of the chiral amplitudes
satisfying the GCWIs. The last step is based on the fact, that the maximum logarithmic degree
of hi A is reduced by one compared to the maximum logarithmic degree of A.
The given argumentation is not affected if the conformal block decomposition of the amplitudes is taken into account, i.e. if Eq. (4.5) is substituted by

Xqp Aq A p ,
A=
qp
where Aq denotes the contribution of a conformal block q to A. As the conformal blocks do only
depend on the crossratios, adjustment of their linear combination cannot cancel the mismatch of
logarithmic powers in Eq. (4.6).
Subsequently, we present an ansatz that admits the construction of generic local n-point functions out of the known chiral correlators:
k1 kn = k1 kn k1 kn |selection rules + GOL,
(4.7)
where GOL stands for guarantor of locality and lives up to its name by providing the desired
behavior of k1 kn under the action of O0 O 0 . This fixes (O0 O 0 )GOL as follows:
(O0 O 0 )GOL =
n

hi k1 kn k1 kn k1 kn h i k1 kn .
(4.8)
i=1
The contribution of
k1 kn k1 . . . kn =

i<j
|zij |
2ij
l max
l max
G
F{q}
(x)PG (lms ns )
g=0
PG (lmr nr )
F{q}
(x)
g=0
in Eq. (4.7) is constricted to terms satisfying selection rules. The selection rules for the arising
logarithmic terms in chiral correlators have been resumed in Section 4.1. Generalising them to
the non-chiral case is straightforward: The highest logarithmic degree to appear in a correlator
was shown to depend on its total Jordan level and the rank of the theory, [S1]. As demonstrated
in Section 3 the rank of the non-chiral theory equals the rank of the chiral halves it is composed
of. Furthermore, as per construction the total Jordan level of the left-hand side of Eq. (4.7) equals
the total Jordan level of the chiral correlators k1 kn and k1 kn , i.e.
g + g l max ,
max
max
max
lnon-chiral
= lchiral
= lanti-chiral
.
The highest multiplicity for one index to appear within a monomial was stated to solely depend on the rank of the theory, [S2]. Thus, according to the aforementioned reasoning, it can be
adopted from the chiral case as its stands.
Therewith the left-hand side of the ansatz (4.7) can in principle be calculated. For that purpose,
we have to expand the product of the chiral correlators, implement [S1] and [S2] and add an
expansion of GOL. The latter is given as follows:
GOL =
max
|zij |
2ij
l

G=0 ,
i<j
t
u

487
(lms ns )gs (lmr nr )gr ,
s=1 r=1
u
t
(gs + g r ) = G.
(4.9)
s=1 r=1
It is clear that the coefficients c depend on the chiral and anti-chiral structure functions, i.e.
c = c (F{q} , F{q}
). For further convenience, we will subsequently use the naming convention
introduced in Eq. (4.4):
p :=
t

gs
(lms ns )
t
s=1
gs = g() =: g .
s=1
Herewith the expansion of GOL (4.9) reduces to the form:

GOL =
max
|zij |
2ij
l

c p p .
(4.10)
G=0 ,
i<j
Of course, the logarithmic terms of GOL are subordinated to [S1] and [S2], too. Further
restrictions on the structure of GOL arise from its claimed behaviour under the action of (O0
O 0 ): Eq. (4.8) changes sign under complex conjugation. It follows that
GOL = GOL
c = c ,
in particular c R.
(4.11)
In addition the coefficients c are coupled to linear combinations of the structure functions
and F{Gq}
by two sets of constraints. The first set arises out of the condition (4.8). The
(x)
second emanates from the logarithmic identities governing the assembly of local monomials:
Monodromy invariance of Eq. (4.7) enforces arguments of emerging logarithms to be real, i.e.
logarithmic terms have to be of the shape
G (x)
F{q}
t

g
p |lms ns |2 :=
|lms ns |2 s
=
=
s=1
t

(lms ns + lms ns )gs
s=1
gs
t

s=1
i=0

gs
i
gs i
(lms ns ) (lms ns )
.
i
(4.12)

Hence, coefficients of terms ts=1 (lms ns )i (lms ns )gs i with 0 i gs are fixed up to an overG (x)F G (x)
all factor. These coefficients are proportional to F{q}
if a monomial stems from
{q }
k1 kn k1 kn or else given by c , which establishes the aforementioned coupling. Recalculating one finds, that every solution of the resulting set of constraints solves equation (4.8).
Without loss of generality
the free choice of an overall factor of Eq. (4.12) can be absorbed in
the factors of contributions ts=1 (lms ns )gs (lms ns )0 . This lightens our ansatz (4.7) to
k1 kn =

i<j
max
|zij |
2ij
l

G=0
G
F{0q } F{q}
c + c0 p .

=:C
(4.13)
488
Within this generic approach the monodromy properties of the local correlators cannot be
completely explored. The structure functions F{q } (x)
and F{q} (x) in Eq. (4.13) are linear combinations of conformal blocks:

j

j
F i (x),
=
F{q} (x) =
F (x),
F{q } F{q} =
Xij F i F .
F{q } (x)
{q }
{q }
{q}
{q}
ij
Enforcing monodromy invariance of the amplitude (4.13) determines the coefficients Xij , a task
that cannot be performed within the generality aimed here.
It is worth pointing out that the claims asserted so far suffice to guarantee all coefficients
C being real: The constraint (4.12) in particular demands that the i = 0 term and the i = gs
term arise with the same multiplicity. The coefficient of the former can easily be shown to equal
0
F{Gq}
F{q } c + c0 . Using Eq. (4.11) we can conclude:
0
F{Gq}
F{q } c + c0 = C = C
C R.
Reimplementing the GCWIs establishes dependencies between coefficients C of different

local correlators of a set. The occurring cross-linkage of coefficients is in perfect
analogy
to the
chiral case: C emerging in a correlator A is connected to C in a correlator i hi A = i h i A
if p is the image of p under the action of O0 as well as under the action of O 0 . It follows that
the obtained solution for a generic chiral set of correlation functions, obeying locality and global
conformal invariance is not unique.
The number of degrees of freedom a local correlator possess equals the number of those
the corresponding chiral correlator shows. This matches with the predictions of the substitution
method [31]. Anyhow it is astonishing as one could have expected that implementing the condition of locality would confine the number of free parameters. Investigating duality [28,32] of the
obtained generic correlation functions yields that this constraint too does not reduce the number
of free parameters any further.
4.3. Discussion and summary
Non-uniqueness of the solution
Generic local sets of correlators cannot uniquely be determined by virtue of locality, duality
and global conformal invariance. The most general form of GOL preserves the number of free
parameters a set of correlation functions exhibits. The minimal choice for GOL fulfilling the
constraints (4.8) and (4.12) is build up of mixed terms only, i.e. it does not exhibit any contributions p p with g = 0 or g = 0. As a consequence, c0 in Eq. (4.13) vanishes identically
for all . In this case, reimplementing the GCWIs fixes all but one conformal block of a set. We
are confronted with a situation very similar to non-logarithmic CFT: Although in the LCFT case
a whole hierarchy of n-point functions emanates from a set of n conformal weights hi = h i the
structure of each correlator is fixed up to one shared structure function F which solely depends
on n 3 crossratios and the conformal weights hi but not on ki .
Consistency check
Our method allows us to connect the results of [9] to the results of [19]. According to [9] the
rank two chiral set of four-point functions with hi = 0 i reads
1000 = F1000 ,

1
1100 = PS2 F1100 l12 F1000 ,
2
(4.14a)
(4.14b)

1
1 2
1
F1110 +
l23 l12 F0110 + l12 l23 l12
F1000 ,
6
2
2

1
1
1
F1111 +
l23 l12 F0111
1111 = PS4
24
6
3

1
1
1 2
+
l14 l23 l12 l34 + l12 l23 l24 l23 l12
F0011
2
2
4

1 2
1
+ l12
l34 + l12 l23 l13 l12 l23 l34 F1000 .
2
3
489
1110 = PS3
(4.14c)
(4.14d)
Missing correlators of the set can be obtained by permutation of the inserted fields. Let PSx
denote the sum over all permutations of indices generated by the group Sx . As all four fields
posses the same conformal weight hi = hj it follows from the associativity of the fusionalgebra
that
Fq1 q2 q3 q4 (x) = Fq (1) q (2) q (3) q (4) (x) S4 ,
(4.15)
i.e. PSx does only affect the

indices of logarithmic terms lij . The symmetry (4.15) admits to
identify all Fq with equal qi = p and to abbreviate them by the pth capital letter of the Latin
alphabet. As h=0,k=0 denotes the identity field the amplitudes (4.14a)(4.14c) are identical to
the one-, two- and three-point functions of the field h=0, k=1 . Consequentially, in the following
expressions, A, B and C are constant, D = D(x). We can now arrange Eq. (4.14) more clearly:
1000 = A,
(4.16a)
1100 = B 2A(l12 ),
(4.16b)
1110 = C B(l12 + l13 + l23 )

2

2
2
,
+ l13
+ l23
+ A 2(l12 l23 + l13 l23 + l12 l13 ) l12

2
1111 = D C + 2B (2 + 1) + =
3
+ 2A[= + ].
(4.16c)
(4.16d)
The correlation function (4.16d) exhibits a kernel term KG whose multiplicity is not fixed
by global conformal invariance. Moreover in support of clarity we chose a graphical notation
to depict the logarithmic terms in this correlator. Reading the diagrams is straightforward: Each
position denotes a variable indexvalue. The symbol filling a position designates the variable
which carries the index. We will use for z and for |z|2 . A line between two symbols of the
same kind indicates the logarithm of the symbolically represented variable. Each graph stands for
the sum over all those identifications of variable indexvalues with numerical indexvalues which
provide expressions that are not equivalent, e.g.

= l12 l34 + l13 l24 + l14 l23 , lij := ln |zij |2 .
Combining local amplitudes according to the method described above yields:
1000 = CAA ,
(4.17a)
1100 = CBB 2CBA (l12 ),
(4.17b)
1110 = CCC CCB (l12 + l13 + l23 )

+ CCA 2(l12 l23 + l13 l23 + l12 l13 ) l212 + l213 + l223 ,
(4.17c)
490

2
1111 = CDD CDC + 2CDB (2 + 1) + =
3
+ 2CDA [= + ],
(4.17d)
+ cxy with cxy being a complex number corresponding to c0 in formula
where CXY = XY
(4.13). The reimplementation of the GCWIs enforces the identification all coefficients CXY with
same second index: CXY CY X. It is worth noticing that in the course of assembling local
amplitudes the arbitrary multiplicities and of the chiral kernel term are set to zero. The kernel
term KG = (2 + =) in the correlator (4.17d) is exclusively composed
of contributions of GOL.
We can now resume: Eq. (4.17) constitute a generic set of correlators containing fields from
a rank two reducible representation with h = h =
0. One explicitly known representative of this
case is given through the set of n-point functions n hn =0,h n =0,kn =1 , n = 1, . . . , 4, of the local
LCFT at c = 2 [19]. Except for an overall sign of the correlator (4.17d) both setsthe generic
and the concreteare consistent: For the local triplet theory, the undetermined coefficients of the
generic set take the values
CA = C0 = 1,
CB = 8 ln(2),
CC = 48 ln2 (2),
CD = 256 ln3 (2).
Furthermore the kernel multiplicity is fixed to 12 in virtue of the operator product expansion.
5. Conclusions
Exploiting the conformal symmetry allows first steps towards the construction of generic local
LCFTs. The generality of our approach is confined by the following assumptions: Indecomposable representations are considered only with respect to the Virasoro zero mode. Irreducible
subrepresentations are assumed to correspond to proper primaries, logarithmic partners to quasiprimaries. Furthermore only the diagonal case hi = h i is regarded.
For future work, generalising our findings by relaxing these assumptions would be an interesting task: As mentioned before the scope of Section 4 can possibly be extended to pre-logarithmic
fields and non-quasi-primaries [3,9]. Remarkably, it seems that even abdicating the hi = h i condition does not seriously damage the proposed method for the construction of generic local
correlation functions out of generic chiral ones.
The assumption that irreducible subrepresentations correspond to proper primary fields, as
made in our work, is more of a technical nature to simplify the deductions. Furthermore, it is
the only case for which the general form of four-point functions has been explicitly computed.
However, the algorithm presented in [9] can easily be adapted to deal with a more general setting.
Essentially, this amounts into setting a less simple initial condition. We stress that this is of
relevance as the irreducible subrepresentation with weight h = 5/8 in the augmented c = c2,3
model provides a counter-example to our assumption [15].
The assumption of quasi-primarity of logarithmic partner fields is the only assumption we
really need and which certainly is violated in many cases. For example, representations of all
types but type A in the notation of [15] do violate the assumption. It is also violated in general
in the cp,1 models for p > 2. Even the standard example, the c2,1 = 2 model, contains one
reducible but indecomposable representation, where this assumption is broken, R1 . However, as
argued in [3], a fermionic zero mode content can be defined for the c = 2 theory. It can then be
shown that the part of the field which violates quasi-primarity differs in its fermionic zero mode
491
content by one modulo two. As a consequence, this part does not affect correlation functions,
see [3] for details. As all cp,1 models enjoy structurally similar fermionic sum representations of
their partition function and characters as for the c = 2 model [33], in particular a symplectic
structure, it seems very plausible that one can define a fermionic zero mode content for all cp,1
models. This might indicate that the condition of quasi-primarity could be relaxed for these
models in the same way as for the c = 2 model. However, we do not know whether this might
generalize to the augmented minimal models.
On the other hand, the representation R1 in the c = 2 model actually is a representation of
a larger symmetry than the Virasoro algebra, namely a W-algebra representation of the triplet
algebra. In this particular case, it turns out that the non-quasi-primary state, let it call us |,
has the property that L1 | W10 |. Thus, this degeneracy in the space of states can be used
to eliminate contributions from the non-quasi-primarity of |. Thus, the problem of non-quasiprimarity occurs in the c = 2 model on the level of W-descendants of W-primaries and their
W-log-partners. As correlation functions of descendant fields are uniquely determined through
the W-symmetry, as soon as the correlation functions of the basic fields are known, one may
argue that the problem of non-quasi-primarity ison the level of the representation theory of the
maximally extended symmetry algebranot that relevant. The same argument should hold for
the structurally very similar cp,1 models with their triplet algebras [34]. As the augmented minimal models also possess extended symmetry algebras, we may hope that non-quasi-primarity
does not play a role for the basic fields in the full W-algebra representations.
Nevertheless, a matter of particular interest would be to investigate the local space of states
for a theory whose chiral halves violate the assumption of quasi-primarity of the logarithmic
partner fields. In case of chiral Jordan cells which contain at least one logarithmic field h,k
( h,
k respectively) with the property
n > 0:
lim Ln h,k (z)|0 = 0
z0
the equivalence of QSC and KC is broken. This shall be illustrated by means of a rank three
non-chiral representation where the chiral Jordan level one fields are not quasi-primary.
According to both methods non-quasi-primarity of the chiral level k states induces non-quasiprimarity of the non-chiral level k state (here n = 1, n = 1 respectively). Pursuant to QSC
By
additionally states with lower level are affected (here the level n = 0 representative 1 1).
contrast after KC the non-quasi-primarity is passed on states with higher level (here the n = 2
Two cases of the sketched scenario can be distinguished:
state 0 2 + 1 1 + 2 0).
492
Rank r 3: In KC non-quasi-primarity (NQP) of a chiral state with Jordan level k (k respectively) encroaches upon all local states with n k. In QSC the spread of NQP can be
suppressed by suitable choice of the considered representative unless the non-quasi-primary
logarithmic partner is the field with highest Jordan rank.
Rank r = 2: In KC the only possible non-quasi-primary is identical to the field with highest Jordan level, i.e. the NQP does not spread. In QSC every level zero representatives is
affected by NQP. We deal with a situation that could be called converse quasi-primarity
as the behaviour under L0 and L 0 equals the behaviour of a proper primary but for every
representative
+ (1 1 0 1),
|n = 0 = (1 0 + 0 1)
exists a positive Virasoro mode such that

(L n )i (Ln )j |n = 0 = 0,
i, j {0, 1}, i + j = 0.
Actually, a promising possibility to cope with the spreading of NQP in the framework of
QSC might be to extend Nhh . For an r = 2 example we refer to [23].
Exploring the first case might bring light to a problem we have already sketched in Section 3.3.
It seems desirable to understand how the uncertainty can be mastered which enters computable
data in the QSC formulation through the arbitrary but non-equivalent choice of the regarded
representative.
Furthermore the broken QSCKC equivalence provides the opportunity to treat them as competing models. On the basis of an explicitly known realisation exhibiting the claimed properties
it could become possible to determine wether the predicted spread of NQP is reasonable. And finally the comparison of the explicit realisation with the predictions of QSC and KC could resolve
the question if one of the proposed methods can be adapted to the NQP case.
Acknowledgements
We would like to thank Matthias Gaberdiel, Kirsten Vogeler and Hendrik Adorf for valuable
discussions. The work of M.F. is partially supported by the European Union network HPRN-CT200200325 (EUCLID).
References
[1] V. Gurarie, Logarithmic operators in conformal field theory, Nucl. Phys. B 410 (1993) 535549, hep-th/9303160.
[2] M.A.I. Flohr, Null vectors in logarithmic conformal field theory, JHEP Proc. Sect., PRHEP-tmr2000/004, hepth/0009137.
[3] M.A.I. Flohr, Operator product expansion in logarithmic conformal field theory, Nucl. Phys. B 634 (2002) 511545,
hep-th/0107242.
[4] A.M. Ghezelbash, V. Karimipour, Global conformal invariance in D dimensions and logarithmic correlation functions, Phys. Lett. B 402 (1997) 282289, hep-th/9704082.
[5] M.R. Rahimi Tabar, A. Aghamohammadi, M. Khorrami, The logarithmic conformal field theories, Mod. Phys. Lett.
A 12 (1997) 13491353, hep-th/9610168.
[6] J. Rasmussen, On logarithmic solutions to the conformal Ward identities, Nucl. Phys. B 730 (2005) 300311, hepth/0507177.
[7] J. Rasmussen, Affine Jordan cells, logarithmic correlators and Hamiltonian reduction, Nucl. Phys. B 736 (2006)
225258, hep-th/0508179.
[8] M.A.I. Flohr, M. Krohn, A note on four-point functions in logarithmic conformal field theories, Fortschr. Phys. 53
(2005) 456462, hep-th/0501144.
493
[9] M.A.I. Flohr, M. Krohn, Four-point functions in logarithmic conformal field theories, Nucl. Phys. B 743 (2006)
276306, hep-th/0504211.
[10] S. Moghimi-Araghi, S. Rouhani, M. Saadat, Use of nilpotent weights in logarithmic conformal field theories, Int. J.
Mod. Phys. A 18 (2003) 47474770, hep-th/0201099.
[11] P. Mathieu, D. Ridout, From percolation to logarithmic conformal field theory, arXiv: 0708.0802.
[12] M.R. Gaberdiel, H.G. Kausch, Indecomposable fusion products, Nucl. Phys. B 477 (1996) 293318, hep-th/
9604026.
[13] M.R. Gaberdiel, An algebraic approach to logarithmic conformal field theory, Int. J. Mod. Phys. A 18 (2003) 4593
4638, hep-th/0111260.
[14] J. Rasmussen, P.A. Pearce, Fusion algebras of logarithmic minimal models, arXiv: 0707.3189.
[15] H. Eberle, M.A.I. Flohr, Virasoro representations and fusion for general augmented minimal models, J. Phys. A 39
(2006) 1524515286, hep-th/0604097.
[16] M.A.I. Flohr, H. Knuth, On verlinde-like formulas in c(p, 1) logarithmic conformal field theories, arXiv: 0705.0545.
[17] M.A.I. Flohr, Singular vectors in logarithmic conformal field theories, Nucl. Phys. B 514 (1998) 523552, hepth/9707090.
[18] H. Eberle, M.A.I. Flohr, Notes on generalised nullvectors in logarithmic CFT, Nucl. Phys. B 741 (2006) 441466,
hep-th/0512254.
[19] M.R. Gaberdiel, H.G. Kausch, A local logarithmic conformal field theory, Nucl. Phys. B 538 (1999) 631658,
hep-th/9807091.
[20] H.G. Kausch, Symplectic fermions, Nucl. Phys. B 583 (2000) 513541, hep-th/0003029.
[21] M. Jeng, G. Piroux, P. Ruelle, Height variables in the Abelian sandpile model: Scaling fields and correlations, J. Stat.
Mech. 0610 (2006) P015, cond-mat/0609284.
[22] M.R. Gaberdiel, I. Runkel, The logarithmic triplet theory with boundary, J. Phys. A 39 (2007) 1474514780, hepth/0608184.
[23] M.R. Gaberdiel, I. Runkel, From boundary to bulk in logarithmic CFT, arXiv: 0707.0388.
[24] V. Schomerus, H. Saleur, The GL(1|1) WZW-model: From supergeometry to logarithmic CFT, Nucl. Phys. B 734
(2002) 211245, hep-th/0510032.
[25] G. Gtz, T. Quella, V. Schomerus, The WZW model on PSU(1, 1|2), JHEP 0703 (2007), hep-th/0610070.
[26] T. Quella, V. Schomerus, Free fermion resolution of supergroup WZNW models, JHEP 0709 (2007), arXiv:
0706.0744.
[27] A.-L. Do, Diploma thesis.
[28] P. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, 1997.
[29] V. Gurarie, M.A.I. Flohr, C. Nayak, The HaldaneRezayi quantum hall state and conformal field theory, Nucl. Phys.
B 498 (1997) 513538, cond-mat/9701212.
[30] M.R. Gaberdiel, P. Goddard, Axiomatic conformal field theory, Commun. Math. Phys. 209 (2000) 549594, hepth/9810019.
[31] M.A.I. Flohr, Bits and pieces in logarithmic conformal field theory, Int. J. Mod. Phys. A 18 (2003) 44974592,
hep-th/0111228.
[32] M.B. Green, J.H. Schwarz, E. Witten, Superstring Theory, An Introduction, vol. 1, Cambridge Univ. Press, 1987.
[33] M. Flohr, C. Grabow, M. Khn, Fermionic expressions for the characters of cp,1 logarithmic conformal field theories, Nucl. Phys. B 768 (2007) 263276, hep-th/0611241.
[34] N. Carqueville, M. Flohr, Nonmeromorphic operator product expansion and C2 -cofiniteness for a family of Walgebras, J. Phys. A: Math. Gen. 39 (2006) 951966, math-ph/0508015.

Harmonic measure and winding

of random conformal paths:
A Coulomb gas perspective
Bertrand Duplantier a, , Ilia A. Binder b
a Institut de Physique Thorique, CEA/Saclay, F-91191 Gif-sur-Yvette cedex, France
b Department of Mathematics, University of Toronto, Toronto, Ontario, M5S 3G3, Canada
Received 19 February 2008; accepted 13 May 2008

Available online 9 June 2008
Abstract
We consider random conformally invariant paths in the complex plane (SLEs). Using the Coulomb gas
method in conformal field theory, we rederive the mixed multifractal exponents associated with both the
harmonic measure and winding (rotation or monodromy) near such critical curves, previously obtained by
quantum gravity methods. The results also extend to the general cases of harmonic measure moments and
winding of multiple paths in a star configuration.
Crown Copyright 2008 Published by Elsevier B.V. All rights reserved.
1. Introduction
1.1. Historical perspective
The subject of conformally invariant (CI) random curves in two dimensions has seen spectacular progress in recent years thanks to the invention of the Stochastic Loewner Evolution
(SLE) [13]. This represents the crowning achievement of studies of 2D critical systems undertaken more than thirty-years ago. The first advance came with the introduction of the so-called
Coulomb-gas (CG) formalism. The critical properties of fundamental two-dimensional statistical
E-mail address: bertrand@spht.saclay.cea.fr (B. Duplantier).

0550-3213/$ see front matter Crown Copyright 2008 Published by Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2008.05.020
B. Duplantier, I.A. Binder / Nuclear Physics B 802 [FS] (2008) 494513
495
models, like the O(N) and Potts models, then found an analytic description within that formalism, which led to a number of exact results [4,5].
This was soon followed by the conformal invariance breakthrough that occurred in 1984 with
the celebrated BPZ article [6]. It was followed by many studies in conformal field theory (CFT),
which became an essential source of applications to 2D statistical mechanics [7,8]. That finally
caught the attention of mathematicians [9], first through the peculiar cases of Cardys formula
for crossing probabilities in percolation [10] and of intersection properties of planar Brownian
paths [11], resulting several years later in the advent of the SLE era [1214].
Meanwhile, the relationship between the CG and CFT approaches was brought to light
when the Coulomb gas representation appeared as an explicit model for a continuum of twodimensional CFTs [15,16]. A Gaussian free field theory in a given domain, modified by a
background charge 20 (the charge at infinity) that couples the field to the domain or boundary curvature, provides a concrete representation of abstract conformal field theories with central
charge c = 1 2402 1 [16]. Many of the critical properties of statistical models could then be
obtained from the fusion of these approaches.
Another advance came in 1988 with the intrusion into statistical mechanics of twodimensional quantum gravity (QG) [17]. The so-called KPZ relation between conformal weights
in presence of a fluctuating metric and those in the Euclidean complex plane [17,18] could then
be checked by explicit calculations [1921]. Though it is not the subject of this article, one cannot avoid mentioning in passing the important body of knowledge accumulated since then in the
related field of random matrix theory.
A decade later, just before the advent of SLE, it became clear, first from the reinterpretation of
an independent rigorous study in probability theory on intersection properties of planar Brownian
paths [22], that an underlying quantum gravity structure played a unifying role in conformal
random geometry in the complex plane [23]. In particular, revisiting QG allowed the prediction
of fine geometrical, i.e., multifractal, properties of random critical curves.
Those concerned the multifractal spectra associated with the moments of the harmonic measure [24], i.e., the electrostatic potential, near CI curves, which could be derived exactly within
that approach for any value of the central charge c [25]. A generalization concerned the peculiar
mixed multifractal spectrum that describes both the harmonic measure moments and the indefinite winding or rotation (i.e., as logarithmic spirals) [26] of a CI curve about any of its points
or of the Green lines of the potential [27]. Higher multifractality spectra were also introduced
and calculated, concerning multiple moments of the harmonic measure and winding in various
sectors of multiple random (SLE) paths in a star configuration [28,29].
All these studies resorted to quantum gravity, by using a probabilistic representation of
harmonic measure moments in terms of collections of Brownian paths, and by performing a
transmutation of the latter paths into multiple (mutually-avoiding) SLEs, the rules of which
are established within the QG formalism. The mixed harmonic-rotational spectrum was then
obtained by blending this method with an earlier Coulomb gas study of the winding angle distributions of critical curves [30].
In mathematics, the same multifractal harmonic measure or mixed spectra are the subject of
present studies and can be obtained rigorously via the probabilistic SLE approach [31,32].
Recently, the same problem of the geometrical properties of critical curves was addressed in
the physics literature via the Coulomb gas approach alone [33,34]. That work in particular used
the statistical equivalence between correlation functions of conformal operators in the complex
plane C and correlation functions of a subset of these operators in presence of SLEs, i.e., in C
cut by the latter [3537]. Riemann uniformizing conformal maps were then devised to unfold
496
the random paths onto the outside of some smooth domain such as the unit disk. Standard CFT
transformation rules of primary conformal operators were then the tool of choice. They allowed
in particular to recover the multifractal exponents associated with harmonic measure moments
near critical curves, originally obtained from quantum gravity.
This ingenious approach thus raised the open question of how to generalize it to mixed harmonic and winding properties of random conformal curves, and recover the associated mixed
multifractal spectra. The aim of the present article is to present such a generalization, still solely
within the Coulomb gas formalism.
Not surprisingly, the main ingredient is the use of arbitrary chiral primary operators, instead
of the more familiar spinless ones. Their associated CFT transformation rules under arbitrary
conformal maps will then yield information about the rotations that occur along Green lines near
conformally invariant random paths.
A technical subtlety will arise here, since one no longer uses correlation functions of operators alone to extract geometrical information about windings of random paths. To study such
windings, it is convenient to resort to products of vertex operators, without statistical averaging,
and study their monodromy properties at short distance. These encode the asymptotic geometrical rotation properties of the random paths and can be analyzed via operator product expansions,
as will be shown below.
1.2. Harmonic measure and winding moments
Definitions
Let us consider a random conformal path S, such as the trace of a Stochastic Loewner Evolution (SLE), or a critical scaling curve in a two-dimensional critical statistical system (Fig. 1).
Consider a point 0 S and a point z
/ S. Define (0,
z) as the harmonic measure of the ball
Bz (0) of radius |z|, centered at 0. This harmonic measure is the probability that a Brownian
motion, started at infinity, first hits S inside the ball Bz (0). Alternately, one can consider an
electrostatic potential, with Dirichlet boundary condition 0 on the equipotential S and value 1 on
a large circle at infinity. The harmonic measure (0,
z) is then the electric charge induced on
S Bz (0). Let (z) be the (possibly indefinite) rotation angle of the Green line (i.e., electrostatic
field line) passing through point z, when z tends to 0 while avoiding S; equivalently, this is the
rotation angle of the equipotential passing through z (i.e., path S for z 0).
We shall evaluate the (typical) asymptotic behavior for z 0 of the mixed moment,

,
h (z) exp p(z) |z|x(h,p)
(1)
Fig. 1. A random conformal path S, equipped with a ball Bz (0) centered at point 0 S and passing through point z
/ S.
The winding angle (z) essentially measures the (possibly multiple or indefinite) rotation of the path S up to point z,
when z 0.
497
also expected when averaging over configurations of the random path. The critical exponent
x(h,
p) depends on the two arbitrary parameters h and p explicitly [27].
It is also interesting to consider in general multiple SLE paths with a n-arm star topology, and
to define in a similar way the multiple mixed moments [29]:
n

i };p)
ihi (z) exp p(z) |z|x({h
.
(2)
i=1
Here i (z) is the harmonic measure in the ith sector of the star, within the ball Bz (0) of radius
|z| centered at apex 0, while (z) is the common monodromy angle describing the rotation for
z 0 of the Green lines between the arms of the random conformal star. The previous single
path case can be seen as a two-star configuration with i=1,2 (z) the harmonic measures of the
two sectors cut by the path inside the ball Bz (0).
Here again, the critical exponent x({h
i }, p) depends explicitly on all arbitrary exponents hi
and rotation parameter p [29], as we shall see in Section 3.
Riemann map
A known way of computing these moments is to consider the conformal map w(z) which
transforms the exterior of the random path S into the exterior of the unit disk D. The derivative
of this map, w (z), encodes all the relevant geometrical information. For z 0, one has the
equivalence for the harmonic measure (z)
|zw (z)|, while the winding angle is given asymptotically by (z) arg w (z). Thus the mixed moments (1) above can be studied as well via
the moments associated with the derivative of the conformal map
h

w (z) exp p arg w (z) |z|x(h,p) ,
(3)
with a now obvious shift of the critical exponent x(h, p) := x(h,

p) h.
In the case of a n-arm star, one takes a set of points zi each in a separate sector i = 1, . . . , n,
and with distances |zi | |z| all scaling in the same way. Then the windings in each sector are
asymptotically equivalent so that for all i, arg w (zi ) (z), whence
n
n

w (zi )hi exp pi arg w (zi ) w (zi )hi exp p(z)
i=1
i=1
|z|x({hi };p) ,
(4)

i }; p) i hi .
where p := ni=1 pi and x({hi }; p) := x({h
To keep the formalism and technical notations to a minimum, and to avoid confusing the
reader, we have simply adopted the notations of previous work by the Chicago group [34], of
which the present study can be considered as a generalization. The reader is thus referred to their
article which contains some relevant introductory details. The connection to our previous results
and notation more familiar in quantum gravity will be seen at the end of this article.
498
2. Derivation
2.1. Star and operator products
Vertex functions
In complex Gaussian free field theory [38,39], the basic objects considered here are operator
products, i.e., products of so-called vertex functions:
n

P :=
(5)
Oi , i (zi , z i ) ,
C
i=1
where each vertex function or operator O, (z, z ) is formally made of two respectively holomorphic and anti-holomorphic components
O, (z, z ) = V (z) V (z),
V (z) := ei
2(z)
, V (z) := ei
with Gaussian free-field correlators:

(z)(z ) = log(z z ),

z ) = 0.
(z)(
z)
2 (
(6)

z)(
z ) = log(z z ),
(
(7)
The holomorphic and the anti-holomorphic weights of the vertex operators are found in a standard way by applying the stressenergy tensor to them:
h = h = ( 20 ),
h = h = (
20 ),
(8)
where 20 is the background charge acting in the Coulomb gas representation of the Gaussian
free field theory. A given weight h thus corresponds to two possible charges:

h = 0 02 + h,
(9)
an equation which applies separately to holomorphic and anti-holomorphic components. A vertex
if either = or = 20 . In the sequel, since we
operator is spinless (meaning that h = h)
are interested in the rotation of conformally invariant curves, the main tool is the consideration of
chiral operators, as opposed to the usual case of spinless ones. Therefore we explicitly consider
in (5) operators Oi , i with different conformal weights:
hi := hi
= h i =: h i .
Star operator product
To study the harmonic or rotation properties of a star configuration of n critical curves, i.e.,
SLEs, the main mathematical object is the star operator product [34]

n

Pn {zi , z i } := 0,n/2 (0)
(10)
Oi , i (zi , z i ) ,
i=1
where the operator 0,n/2 (0) in (10) is, in the conformal field theory parlance, the operator corresponding to the existence of n critical curves originating at point z = 0 in the plane. It represents
499
the seed of a star configuration of n such curves, i.e., SLEs. The test operators Oi , i are a priori
chiral, so that i
= i (or
= 20 i ).
We shall need the standard CG notations for holomorphic charges [39]:
1
1
1
r,s := 0 (r+ + s ) = (1 r)+ + (1 s) ,
2
2
2
where the basic charges are given in terms of the SLE parameter [0, 8] by [34]
2
,
= ,
+ =
2
2
20 = + + =
.
2
(11)
(12)
(13)
The charges satisfy the simple relations

= 0 02 + 1,
thus both correspond to a conformal weight h = 1, leading to the possibility of using them as
fundamental screening charges to build up the algebra of screening operators [16].
The central charge of the CFT associated with the Gaussian free field theory, modified la
FeiginFuchs and DotsenkoFateev by the background charge 20 , is given by
3 ( 4)2
,
2
while the conformal weights corresponding to charges r,s are

c = 1 2402 = 1
(14)
(r 4s)2 ( 4)2
(15)
.
16
In this representation, the holomorphic (or anti-holomorphic) charge of the bulk curvecreating operator 0,n/2 can be taken to be 0,n/2 = 0 n4 (with conjugate 0,n/2 ), and
corresponds in the Coulomb gas formalism to a combination of electric and magnetic charges
[34,40]. The corresponding operator is spinless with conformal weight [40]
hr,s := r,s (r,s 20 ) =
4n2 ( 4)2
.
(16)
16
Notice that operator products such as (5) or (10) are understood as objects to be inserted in correlation functions, in the same way as the formal holomorphic vs anti-holomorphic factorization (6)
is meaningful only in such correlations (6). Now, in the Coulomb gas formulation of conformal
field theory, correlation functions are to be evaluated for a set of charges which overall respect
electroneutrality, so that the sum of the latter always equals 20 , to compensate for the existence of the background charge 20 . So ultimately, the set of charges introduced by the operator
product Pn near the origin, namely the set {0,n/2 , (i , i ), i = 1, . . . , n} has to be compensated
by other charges located at the observation points of the other remaining vertex functions.
When dealing with conformal blocks, namely treating separately the holomorphic and antiholomorphic parts in correlation functions, electroneutrality should apply separately to both
sectors. Furthermore, if holomorphic (or anti-) electroneutrality does not apply in the arguments
of the correlation functions of the original vertex operators, it is known that supplementary integral screening operators, with screening charges + and (12), can be introduced to extend
the domain of definition of the theory.
h0,n/2 =
500
This caveat here is to explain why, when dealing with local operator products, such as that
(5) or (10), one does not have to enforce Coulomb gas neutrality rules locally, but only globally.
We are here actually concerned only with the short distance and monodromy properties of the
star operator product (10) near the origin, which result from the analysis of the short-distance
expansion of chiral operators O , in presence of the spinless curve-creating one 0,n/2 , and
the CG formalism will be kept at a minimum. It would be interesting to make a more complete
study of the algebra of holomorphic operators, including screening ones, for the present problem
and in the context of SLE theory [16,41].
2.2. Star operator product expansion
OPE for vertex functions
Let us first consider an arbitrary product of vertex functions in the complex plane:

P=
Oi , i (zi , z i ) C =
Vi (zi ) V i (zi ) C ,
iI
(17)
iI
where we have made explicit the formal factorization into holomorphic and anti-holomorphic
vertex operators.
When a subset P of I of points i P is contracted towards a common point, like the origin
here, i.e., i P, zi 0, it is well known that the limiting object is given by the so-called
operator product expansion (OPE), with coefficients that are singular functions of the set of
points P.
In a Gaussian free-field theory, and for a product of vertex functions, i.e., complex exponentials of the field, these coefficients are precisely given by the average correlators of the set
of vertex functions located at the contracted set P. The simplest and most convenient way to
express this result is probably to use the normal ordering of operators :( ): such that [42,43]

Oi , i (zi , z i ) C =
Oi , i (zi , z i ) :
Oi , i (zi , z i ) C :.
(18)
iP
C iP
iP
In this notation, the vertex operators appearing inside :( ):, when inserted in any correlation
function, are to be Wick-contracted only with operators located outside of the contracting set P.
In the complex plane, because of the form (7) of Gaussian averages (or Wick contractions,
or also free-field propagators), the holomorphic and anti-holomorphic sector contributions to
a correlation function decouple, hence factorize:

Vi (zi )V i (zi )

Oi , i (zi , z i ) =
iP
iP
2i j
(zi zj )

(zi z j )2 i j .
(19)
i,j P ,i<j
This in turn gives the explicit form of the OPE (18).

Star OPE
Applying the above OPE result to the star-operator product

n

Oi , i (zi , z i ) ,
Pn {zi , z i } := 0,n/2 (0)
i=1
(20)
501
Fig. 2. Curve-creating operator 0,1 (0) and chiral test operator O, (z, z ), with monodromy of z around base point 0.
when the set of points zi , i = 1, . . . , n all contract to the origin z = 0, yields

20,n/2 i 20,n/2 i

zi
z i
(zi zj )2i j (zi z j )2 i j
Pn {zi , z i } =
i
i<j
:Pn {zi , z i } :,
(21)
a result that will be our main tool in the analysis of the multifractal properties of a set of critical
paths in a star configuration.
2.3. Monodromy and harmonic measure of a single path
Original operator product
Let us consider the standard case, where the star reduces to a single random curve passing
through the origin z = 0, with n = 2 arms, i.e., a set of two semi-infinite curves arising at the
origin. The associated operator is 0,1 with holomorphic charge 0,1 = + /2.
Let us place first one arbitrary test operator O , (z, z ) near the origin, so as to measure
the harmonic measure moments and rotation on one side of the critical curve (Fig. 2). The case
of two test operators on both sides of the path will be treated later as the peculiar n = 2 case of
the general n-star geometry. In the present geometrical situation, the relevant operator product is

P(z, z ) := 0,1 (0)O , (z, z ) C ,
(22)
and we are interested in its rotation, i.e., monodromy properties at short-distance (Fig. 2). Its
short-distance behavior when z 0 is a simple case of (21)

P(z, z ) = z20,1 z 20,1 :P(z, z ):,

such that the SDE coefficient (23) is explicitly
+ ( )/2
z

P(z, z ) z+ z + = (zz)+ ( + )/2
.
z
(23)
(24)
Operator in path geometry

One first writes the identity in law of the operator product P (22) in C, of the path creating
vertex operator 0,1 by the test operator, to the same test operator but now in presence of the
random path S originally represented by 0,1 , hence to the same test operator in the complex
plane C\S slit by the curve (Fig. 3):
(in law)

P(z, z ) := 0,1 (0)O , (z, z ) C = O , (z, z ) C\S .
(25)
502
Fig. 3. Illustration of the statistical identity (25) of the product of operators 0,1 and O, with the latter one put in
presence of the stochastic path.
Fig. 4. Conformal map of the complement C\S of the random path S in C to the exterior of the unit disk D. A mirror
image of w(z) by inversion with respect to the unit circle D appears underneath the boundary.
This identity in law means that averaging within correlation functions the left-hand side over
the complex Gaussian free field (GFF), or the right-hand side over the GFF in presence of S
and over the configurations of the random path S yield the same result [34,35]. [As discussed
in [34], the precise boundary conditions on S for the geometrical random fields associated with
the complex GFF may depend on the phase of the critical system (dilute for simple SLE paths,
4, or dense for > 4), and are not specified here. In the holomorphic formalism, analytic
continuation in allows one to pass from one phase to the other.]
Conformal map
The complex plane slit by S has the topology of the disk. One introduces the Riemann uniformizing map z w(z) that opens the slit S into the unit disk D centered at i, so that w(0) = 0
(Fig. 4) [33,34]. This map naturally depends on the random path S. Under this conformal map,
vertex operators transform like primary fields, whence [39]:

h
h

O , (z, z ) C\S = w (z) w (z)
O , w(z), w(z) C\D ,
(26)
where the holomorphic and anti-holomorphic weights of operator O , are respectively

h = h = ( 20 ),
such that
= h := 0
h = h = ( 20 ),
(27)
02 + h .
Associating together the holomorphic and anti-holomorphic weights h and h , we can write
identically

,
(h h )/2

h +h dw

O , w(z), w(z) C\D ,
(z, z ) C\S = w (z)
(z)
d w
(28)
503
where
dw
w (z)
.
(z) :=
d w
w (z)
The final disk configuration in [O , (w(z), w(z) )]C\D is a boundary configuration [41],
where an image charge appears at the conformal image of point w(z) with respect to the disk
boundary, namely its image by inversion with respect to the unit circle. In the limit z 0, since
w(0) = 0, the latter image coincides with the image with respect to the tangent line, i.e., the
complex conjugate w(z), so the same notation is kept here for simplicity. The vertex operator
with anti-holomorphic charge in [O , ]C\D then becomes an holomorphic vertex function of
charge taken at image point w(z), so that the holomorphic and anti-holomorphic sectors get
coupled [39].
SDE in the w plane
When z 0 in the original domain, w(z) 0 as well as w(z) 0, so the two image points
pinch the unit circle at the origin in the w plane. The coefficient in the short-distance expansion
(SDE) of the right-hand side operator in Eq. (28) is given by the Gaussian averaged correlation:

2
O , w(z), w(z) C\D = w(z) w(z)
.
Then the short-distance expansion of Eq. (26) is
(h h )/2

h +h dw

2
O , (z, z ) C\S w (z)
w(z) w(z)
.
(z)
d w
(29)
After uniformization to the w plane, since the two image points w(z) and w(z) pinch the unit
circle while staying either in the exterior or the interior of the disk, they cannot wind about the
origin w(0) = 0 indefinitely for |z| 0, so arg w(z) remains bounded, e.g.,
arg w(z) [, +].
(30)
We conclude that w(z) w(z) |z||w (z)|, up to a non-winding (or non-monodromic) complex
phase factor. This finally gives the SDE (29)
(h h )/2

h dw

O , (z, z ) C\S |z|2 w (z)
(31)
,
(z)
d w
where the overall (harmonic measure) derivative exponent h is defined as
h := h + h + 2 .
(32)
Owing to (27), it is also

h = ( + )( + 20 ).
It is the weight of an operator whose conformal charge h = 0
conjugate):

+ = h = 0 02 + h.
(33)
02 + h is just + (or its
(34)
Notice also that

h h = ( )( + 20 ).
(35)
504
Winding and uniformizing map

In the z plane, the curve S can wind or rotate indefinitely about any of its points, e.g. about
the origin 0 when z 0, so arg z . We are thus especially interested in the monodromy
properties of the operator O , (z, z ).
By analyticity of the conformal map w onto the unit disk, w(z) w (z)z, so that arg w(z)
arg w (z) + arg z. We just have seen in (30) that arg w(z) remains bounded, so that asymptotically
arg z arg w (z) under the conformal map.
The winding angle (z) of the Green lines of the random curve, asymptotically close to z = 0
on the curve, is given by
(z) = arg z = arg w (z) + O(1).
(36)
Therefore one can also write asymptotically

(z) = log w (z) =

1
1
w (z)
log w (z) log w (z) = log
,
2i
2i
w (z)
so that the exponential winding is

1/2
1/2
1/2
w (z)
dw
z
i(z)
=e

=
,
(z)
z
d w
w (z)
(37)
(38)
where here means equality within a non-winding (non-monodromic) phase factor. We therefore arrive at the expression for the SDE (31)
h

O , (z, z ) C\S |z|2 w (z) ei(h h )(z) .
(39)
Mixed moments
Let us now return to the original OPE (22), (23). We can rewrite (23), (24) as a complex
scaling

P(z, z ) = 0,1 (0)O , (z, z ) C |z|+ ( + ) ei+ ( ) arg z .
(40)
Because of the identity in law (25), identifying (40) and the short-distance expansion (39) in the
transformed slit domain yields the equivalence
h i[h h + ( )](z)

+
w (z) e
|z|+ ( + )2 .
(41)
This scaling equivalence is an SDE result, which can be interpreted as describing typical statistical behavior, also expected to hold true in a weaker form after averaging over the configurations
of the stochastic path S:
h

w (z) eis(z) |z|x ,
(42)
where
s := h h + + ( ),

x := + ( + ) 2 .
(43)
(44)
Let us now express exponent x := x(h, is) solely in terms of weight h and winding conjugate
parameter s. Using (33), (34) and (35) we have
s = ( )(+ + + 20 ) = ( )(+ + h 20 ),
(45)
505
Fig. 5. Conformal map of the complement C\S3 of the random 3-star S3 in C to the exterior of the unit disk D. Three
mirror images of points w(zi ), i = 1, 2, 3, appear by inversion with respect to the unit circle D.
x = + h 2 .
(46)
We can further write trivially

4 = ( + )2 ( )2 = h2 ( )2 ,
to eliminate in (45) and (46)
p
=
,
+ + h 20
(47)
1
1
s2
x(h, is) = + h h2 +
.
2
2 (+ + h 20 )2
(48)
To get the mixed harmonic measure-rotation exponents one has first to analytically continue s:
s = ip in (42) so that

h
w (z) ep(z) |z|x(h,p) ,
(49)
with now
1
1
p2
x(h, p) = + h h2
.
2
2 (+ + h 20 )2
(50)
Finally, one has to choose the value of h that vanishes with h, namely the dilute phase or
simple SLE path one, where 4 and 0 0 (see (13))

h = 0 + 02 + h.
(51)
The exponent x(h, p) above is identical to the scaling exponent obtained by quantum gravity
in [27] for the hth power h (z)ep(z) of the harmonic measure with rotation conjugate parameter p (up to a natural shift by h, due to the local scaling (z)
|z||w (z)| for the harmonic

measure (z)
in a ball of radius |z|, cf. Eqs. (1), (3)). [See Section 3.]
2.4. Monodromy and harmonic measure of multiple paths
Operator product in star geometry
We first use the identity in law of the star-operator product (20) with the operator product of
the test vertex operators in presence of the stochastic star Sn in the complex plane:

(in law)
Oi , i (zi , zi )
.
Pn {zi , z i } =
(52)
i
C\Sn
506
Conformal map
The plane slit by the star having the topology of the disk, a conformal map w(z) transforms
the open set C\Sn into the exterior of the unit disk D [34]. The vertex operator product Pn (52)
in presence of Sn is transformed, according to primary operator rules, as:

Pn =
Oi , i (zi , zi )
C\Sn

h
h

w (zi ) i w (zi ) i
Oi , i w(zi ), w(zi )

i
(53)
C\D
Associating for each i the holomorphic and anti-holomorphic weights hi and h i , we can write it
identically
(h h )/2

i
i
h +h dw

Pn = w (zi ) i i
(54)
.
(zi )
d w
C\D
i
Short-distance expansion
Now let all points converge to the origin in the original z plane, zi 0, i = 1, . . . , n. As in
Section 2.3 above, in the disk geometry, the operators on the right-hand side of (54) are boundary
operators, each of them made of a pair of holomorphic vertex operators taken at w(zi ) and at
its inverted mirror image with respect with the unit circle (Fig. 5). When zi 0, each image
coincides with the local complex conjugate w(zi ) with respect to the local tangent to the unit
circle, so we keep the same notation (a slight abuse).
As usual, the original anti-holomorphic operators with charges i now become holomorphic
vertex operators located at points w(zi ), and get coupled to the original holomorphic vertex operators with charges i located at points w(zi ) [39]. Under the contraction zi 0, i = 1, . . . , n,
the SDE of the transformed operator product on the right-hand side of (54) therefore scales as
the Gaussian free-field average taken at all pairs of points and images [34]:

i
C\D
w(zi ) w(zi )
2
i i
(55)
2
2
2
w(zi ) w(zj ) i j w(zi ) w(zj ) i j
w(zi ) w(zj ) i j .
i<j
(56)
i
=j
Among all these factors, the diagonal (i) terms give the short-distance behavior. The i
= j terms
are finite in the disk geometry, or give subleading contributions for configurations of the star in
which a pair of points (zi , zj ) lie in the same sector, since the relative affixes w(zi ) w(zj ),
w(zi ) w(zj ), w(zi ) w(zj ) then tend to zero. The short-distance behavior of the operator
product (54) is thus
(h h )/2

i
i
h +h dw

2

i
i

Pn
w(zi ) w(zi ) i i .
w (zi )
(zi )
d w
i
When zi 0, the transformed complex coordinate w(zi ) and its image w(zi ) pinch the unit
circle, so that w(zi ) w(zi ) while the winding angle arg w(zi ) stays bounded; whence we can
507
simply take: w(zi ) w(zi ) |zi ||w (zi )|, up to a non-monodromic complex phase factor. This
gives the final short-distance behavior of (54)
(h h )/2

i
i
h +h +2 dw

|zi |2i i w (zi ) i i i i
.
(zi )
Pn
(57)
d w
i
Hence the following exponents appear for the harmonic measure derivative terms:
hi := hi + h i + 2i i = (i + i )(i + i 20 ),
(58)
which are the conformal weights of operators whose conformal charges simply result from the
addition of holomorphic and anti-holomorphic charges

hi = 0 + 02 + hi = i + i .
Windings and uniformizing map
Let us now return to the SDE (21) in the original z plane

20,n/2 i 20,n/2 i

zi
z i
(zi zj )2i j (zi z j )2 i j
Pn {zi , z i }
i
(59)
i<j
and, as in Eq. (24), let us separate the complex modulus and argument parts, by rewriting it as:

2 ( + ) zi 0,n/2 (i i )
|zi | 0,n/2 i i
Pn {zi , z i }
z i
i

zi zj i j i j
|zi zj |2i j +2 i j
.
(60)
z i z j
i<j
We are interested in the monodromy properties of expression (60) when all points in their
own sectors converge to the star apex z = 0 while avoiding touching the star, which can wind
indefinitely about its origin. Let us then introduce a common scaling |z| for all distances |zi | to
the star apex, as well as a common asymptotic winding angle (z) for all arguments arg zi :
zi = |z|ei(z) i ,
arg zi = (z) + arg i ,
(61)
where all moduli |i | and arguments arg i remain bounded when |z| 0. Since trivially
(zi /zi )1/2 = ei(z) (i /i )1/2 ,
(62)
the short-distance expansion (60) obeys the simple identity

Pn {zi , z i } = |z|20,n/2 i (i + i )+ i<j (2i j +2 i j )

ei(z)[20,n/2

Pn {i , i } .
)+2

i (i
i<j (i j
i
i j )]
(63)
Let us now consider the geometrical setting in the transformed w-plane. Since all arguments
arg w(zi ) stay bounded there, we have for zi 0, as in Eq. (36), arg w (zi ) = arg zi + O(1),
so that, as in (38)
1/2
1/2
dw
zi

.
(zi )
(64)
z i
d w
508
Identity (62) then yields immediately the common asymptotic behavior for all i
1/2

dw
ei(z) ,
(zi )
i,
d w
(65)
in terms of the unique rotation angle (z) (61), and up to non-monodromic phase factors. Applying this to the SDE result (57) gives the asymptotic formula

h
Pn |z|2 i i i ei(z) i (hi hi ) w (zi ) i .

(66)
i
Multiple moments and rotation

Because of the identity in law (52), we can now identify expressions (63) and (66) and get
the fundamental scaling formula for the multiple harmonic measure factors and (indefinitely)
rotating phase factor

h

ei(z)[ i (hi hi )+20,n/2 i (i i )+2 i<j (i j i j )] w (zi ) i

20,n/2
|z|
)+
i (i +
i

i j )
i<j (2i j +2
i

i 2i
i
(67)
We obtained this scaling relation from a short-distance operator product expansion, which yields
the leading (typical) scaling behaviour of the product of harmonic measure factors and rotation
phase factor. It can naturally be expected to also give the weaker result on the statistical average
of the same quantity over the configurations of the multiple random paths of star Sn :

h
w (zi ) i eis |z|x({hi };is) ,
(68)
i
with the parametric representation in terms of holomorphic and anti-holomorphic charges

hi := hi + h i + 2i i ,
hi
s
(69)
h i
= i (i
20 );
= i ( i 20 ),

:=
(hi h i ) + 20,n/2
(i i ) + 2
(i j
i
i
i<j
x := 20,n/2
(i + i ) +
(2i j + 2 i j )
i<j
(70)
i j ),
2i i .
(71)
(72)
It remains to find the explicit expression of the scaling exponent x := x({hi }; is) in terms of the
set of weights hi and the Fourier variable s conjugate to rotation . Let us introduce the notations
h = ( 20 ),
h = ( 20 ),

i ,
:=
i ,
:=
i
hi = i + i ,
{h} := + =

(i + i ) =
hi .
i
A little bit of algebra first shows that s (71) can be written in the compact form
s = hi i + 20,n/2
i hi i 20,n/2
509
i
= h h + 20,n/2 ( )
= ( )( + 20 + 20,n/2 )
= ( )({h} 20 + 20,n/2 ).
(73)
One can also check that the exponent x (72) can be written in the compact form
2

x = 20,n/2
hi +
hi
h2i 2
i
2
= 20,n/2 {h} + {h}
h2i
2 .
(74)
Note that after this compaction of formulae, the expressions (73) and (74) are similar to expressions (45) and (46). We therefore eliminate in the same way as above and arrive at a
formula similar to (47) and (48)
=
s
,
{h} 20 + 20,n/2
(75)

1 2
1
s2
h2i +
,
x {hi }; is = 20,n/2 {h} + {h}
2
2 ({h} 20 + 20,n/2 )2
(76)

where we recall that {h} = i hi , hi = 0 + 0 + hi , hence exponent x has now an explicit
form in terms of the set of weights {hi } and of s.
It remains to analytically continue s into s = ip to get the expectation of the multiple harmonic measure moments and Laplace transform of the rotation:

h
w (zi ) i ep |z|x({hi };p) ,
(77)
i

1 2
1
p2
h2i
.
x {hi }; p = 20,n/2 {h} + {h}
2
2 ({h} 20 + 20,n/2 )2
(78)
The CG formula (78) for p = 0 coincides naturally with the result found in Ref. [34].
3. Comparison to quantum gravity results
In previous work [29] we introduced for SLE the KPZ relation
1
h = U () := ( + 4 )
4
between conformal weights in a fluctuating metric with a conformal factor given by a Gaussian
free field (two-dimensional quantum gravity (QG)) and conformal weights h in the complex
plane. Its inverse reads

1
1
4
1
2
= U (h) =
16h + ( 4) +
1
.
2
2
510
Using (13) and (51) we identify
1
h =
U (h).
2
(79)
The CG results above can be written as

p2

,
x {hi }; p = x {hi }; 0
2 L2{h}
L{h} := ({h} 20 + 20,n/2 ),

1 2
h2i .
x {hi }; 0 = 20,n/2 {h} + {h}
2
(80)
Using (79) we find their equivalent quantum gravity expressions [29]

p2
x {hi }; p = x {hi }; 0
(81)
,
2 L2{h}
1
U (hi ) + n,
L{h} =
(82)
2
i

4
1 2
x {hi }; 0 +
(83)
hi = 2U
L{h} + 1
2
i

4
1 2
2U
(84)
n+1
.
2

[As seen in (4), the term i hi simply comes from the passage from derivative moments to
harmonic measure ones.]
Result (50) for the multifractal exponent x(h, p) of a single-sided SLE simply corresponds to
n = 2 and (h1 , h2 ) = (h, 0) in QG formulae (81)(84).
The interpretation of these formulae is quite clear:
Eq. (81) is the Coulomb gas formula obtained a long time ago for the exponent governing
the winding angle distribution of a star made of a number L{h} of random paths (SLEs) [27,
29,30];
L{h} (82) represents, in a star topology, the effective number of SLEs that are exactly equivalent through QG to a collection of hi , i = 1, . . . , n, Brownian paths (which represent the set
of powers hi of the harmonic measure), in addition to the n original SLEs [28,29];
The scaling dimension (83) is the result obtained through QG construction rules for (twice)
the conformal weight in C of a random star made precisely of this effective number L{h}
of SLEs, while (84) is (twice) the weight of the original n-SLE star without the auxiliary
harmonic Brownian paths. Eq. (83) is therefore also exactly (twice) the conformal weight
h = ( 20 ) corresponding to the holomorphic charge = 0,L{h} /2 of the curvecreating operator 0,L{h} /2 , which gives the final and rather elegant CG formula for (80):

hi = 20,L{h} /2 (0,L{h} /2 20 ) 20,n/2 (0,n/2 20 ).
x {hi }; 0 +
i
511
Conclusion
Results (81)(84) can be obtained by quantum gravity construction rules almost immediately,
and have a natural interpretation in that formalism. As we saw, they can also be recovered under the form (78) in the fully developed Coulomb gas and conformal field theory approach,
but though the latter is quite interesting, it also appears to be significantly more combersome.
Its most important aspect is probably the renewed suggestion that a systematic yet rigorous
representation of the stochastic properties of SLEs via a Coulomb gas driven by a Gaussian
free field must exist [44], mimicking the (heuristic) physics CG approachs long-established
predictive power. Let us finally mention that the mixed multifractal moments, introduced here
to study the mixed harmonic measure-rotation spectrum of critical random paths, are amenable
to an approach using only the Stochastic Loewner Evolution, that should allow one to establish
rigorously the present results [31].
Acknowledgements
It is a pleasure to thank Vladimir Dotsenko for generously sharing his scholarly knowledge
of the Coulomb gas representation, David Kosower and Vincent Pasquier for their interest and
a critical reading of the manuscript, and Emmanuel Guitter for help with the figures. We also
thank the Institute for Pure and Applied Mathematics at UCLA, the IAS/Park City Mathematics Institute and B.D. thanks the School of Mathematics of the Institute for Advanced Study,
for their graceful hospitality in successive stays during which this work was done. B.D. wishes
especially to thank Michael Aizenman and Tom Spencer for their generous hospitality in Princeton.
While writing up this manuscript, we learned of related work by A. Belikov, I.A. Gruzberg
and I. Rushkin.
References
[1] O. Schramm, Israel J. Math. 118 (2000) 221288.
[2] W. Werner, Lecture Notes from the 2002 Saint-Flour Summer School, Lecture Notes in Mathematics, vol. 1840,
Springer, 2004, 107195, math.PR/0303354;
See also: W. Werner, Some Recent Aspects of Conformally Invariant Systems, Les Houches Summer School LXIII
Mathematical Statistical Physics, 429 July 2005, Elsevier, Amsterdam, 2006, math.PR/0511268.
[3] G.F. Lawler, Conformally Invariant Processes in the Plane, Mathematical Surveys and Monographs, vol. 114, Amer.
Math. Soc., 2005.
[4] M. den Nijs, J. Phys. A 12 (1979) 18571868;
M. den Nijs, Phys. Rev. B 27 (1983) 16741679.
[5] B. Nienhuis, Phys. Rev. Lett. 49 (1982) 10621065;
B. Nienhuis, J. Stat. Phys. 34 (1984) 731761;
B. Nienhuis, in: C. Domb, J.L. Lebowitz (Eds.), Phase Transitions and Critical Phenomena, vol. 11, Academic
Press, London, 1987.
[6] A.A. Belavin, A.M. Polyakov, A.B. Zamolodchikov, Nucl. Phys. B 241 (1984) 333380.
[7] See: C. Itzykson, H. Saleur, J.-B. Zuber (Eds.), Conformal Invariance & Applications to Statistical Mechanics,
World Scientific, Singapore, 1988.
[8] J.L. Cardy, in: C. Domb, J.L. Lebowitz (Eds.), Phase Transitions and Critical Phenomena, vol. 11, Academic Press,
London, 1987.
[9] R. Langlands, P. Pouliot, Y. Saint-Aubin, Bull. Am. Math. Soc. 30 (1994) 161.
[10] J.L. Cardy, J. Phys. A 25 (1992) L201L206.
[11] B. Duplantier, K.-H. Kwon, Phys. Rev. Lett. 61 (1988) 25142517.
512
[12] G.F. Lawler, O. Schramm, W. Werner, Acta Math. 187 (2001) 237273, math.PR/9911084;
G.F. Lawler, O. Schramm, W. Werner, Acta Math. 187 (2001) 275308, math.PR/0003156;
G.F. Lawler, O. Schramm, W. Werner, Ann. Inst. Henri Poincar PR 38 (2002) 109123, math.PR/0005294.
[13] G.F. Lawler, O. Schramm, W. Werner, Acta Math. 189 (2002) 179201, math.PR/0005295;
G.F. Lawler, O. Schramm, W. Werner, Math. Res. Lett. 8 (2001) 401411, math.PR/0010165.
[14] S. Smirnov, C. R. Acad. Sci. Paris Sr. I Math. 333 (2001) 239244.
[15] B.L. Feigin, D.B. Fuchs, Funct. Anal. Appl. 17 (1982) 114.
[16] Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 240 (1984) 312;
Vl.S. Dotsenko, V.A. Fateev, Nucl. Phys. B 251 (1985) 691.
[17] A.M. Polyakov, Mod. Phys. Lett. A 2 (1987) 893898;
V.G. Knizhnik, A.M. Polyakov, A.B. Zamolodchikov, Mod. Phys. Lett. A 3 (1988) 819826.
[18] F. David, Mod. Phys. Lett. A 3 (1988) 16511656;
J. Distler, H. Kawai, Nucl. Phys. B 321 (1988) 509527.
[19] V.A. Kazakov, Phys. Lett. A 119 (1986) 140144.
[20] B. Duplantier, I.K. Kostov, Phys. Rev. Lett. 61 (1988) 14331436;
B. Duplantier, I.K. Kostov, Nucl. Phys. B 340 (1990) 491541.
[21] I.K. Kostov, Mod. Phys. Lett. A 4 (1989) 217226;
M. Gaudin, I.K. Kostov, Phys. Lett. B 220 (1989) 200206;
I.K. Kostov, Nucl. Phys. B 376 (1992) 539598, hep-th/9112059.
[22] G.F. Lawler, W. Werner, Ann. Probab. 27 (1999) 16011642.
[23] B. Duplantier, Phys. Rev. Lett. 81 (1998) 54895492;
B. Duplantier, Phys. Rev. Lett. 82 (1999) 880883, cond-mat/9812439;
B. Duplantier, Phys. Rev. Lett. 82 (1999) 39403943, cond-mat/9901008.
[24] N.G. Makarov, Fine structure of harmonic measure, Ross. Akad. Nauk Alg. Anal. 10 (1998) 162.
[25] B. Duplantier, Phys. Rev. Lett. 84 (2000) 13631367, cond-mat/9908314.
[26] I.A. Binder, Harmonic Measure and Rotation of Simply Connected Planar Domains, Ph.D. Thesis, Caltech, 1997.
[27] B. Duplantier, I.A. Binder, Phys. Rev. Lett. 89 (2002) 264101, cond-mat/0208045.
[28] B. Duplantier, J. Stat. Phys. 110 (2003) 691738, cond-mat/0207743.
[29] B. Duplantier, Conformal fractal geometry and boundary quantum gravity, in: M.L. Lapidus, M. van Frankenhuysen
(Eds.), Fractal Geometry and Applications: A Jubilee of Benot Mandelbrot, in: Proceedings Symposia Pure Math.,
vol. 72, Part 2, Amer. Math. Soc., Providence, R.I., 2004, pp. 365482, math-ph/0303034;
See also: Conformal Random Geometry, in: Les Houches Summer School LXIII Mathematical Statistical Physics,
429 July 2005, Elsevier, Amsterdam, 2006, math-phys/0608053.
[30] B. Duplantier, H. Saleur, Phys. Rev. Lett. 60 (1988) 23432346.
[31] I.A. Binder, B. Duplantier, Multifractal properties of harmonic measure and rotations for stochastic Loewner evolution, in preparation (2008).
[32] D. Beliaev, Harmonic Measure on Random Fractals, PhD thesis, KTH, Stockholm, Sweden, 2005;
D. Beliaev, S. Smirnov, Harmonic measure and SLE, arXiv: 0801.1792.
[33] E. Bettelheim, I. Rushkin, I.A. Gruzberg, P. Wiegmann, Phys. Rev. Lett. 95 (2005) 170602, hep-th/0507115.
[34] I. Rushkin, E. Bettelheim, I.A. Gruzberg, P. Wiegmann, J. Phys. A: Math. Gen. 40 (2007) 21652195, condmat/0610550.
[35] See: M. Bauer, D. Bernard, 2D growth processes: SLE and Loewner chains, Phys. Rep. 432 (2006) 115, mathph/0602049, and references therein.
[36] J. Cardy, SLE for theoretical physicists, Ann. Phys. 318 (2005) 81118, cond-mat/0503313.
[37] W. Kager, B. Nienhuis, J. Stat. Phys. 115 (2004) 11491229, math-ph/0312056.
[38] Vl.S. Dotsenko, Srie de cours sur la thorie conforme, unpublished.
[39] Ph. Di Francesco, P. Mathieu, D. Snchal, Conformal Field Theory, Springer, New York, 1997.
[40] H. Saleur, J. Phys. A 19 (1986) L807L810;
H. Saleur, J. Phys. A 20 (1987) 455470;
B. Duplantier, Phys. Rev. Lett. 57 (1986) 941944, 2332;
B. Duplantier, J. Stat. Phys. 54 (1989) 581680;
B. Duplantier, H. Saleur, Phys. Rev. Lett. 57 (1986) 31793182;
B. Duplantier, H. Saleur, Nucl. Phys. B 290 (1987) 291326.
[41] For the boundary case, see: J. Cardy, Nucl. Phys. B 240 (1984) 514532;
J. Cardy, Nucl. Phys. B 324 (1989) 581596;
J. Schulze, Nucl. Phys. 489 (1997) 580602;
S. Kawai, J. Phys. A: Math. Gen. 36 (2003) 68756893.
513
[42] B. Simon, The P ()2 Euclidean (Quantum) Field Theory, Princeton Series in Physics, Princeton Univ. Press, 1974.
[43] See the multi-local OPE in: F. David, B. Duplantier, E. Guitter, Renormalization theory of Edwards models for
self-avoiding polymerized membranes, cond-mat/9702136.
[44] O. Schramm, S. Sheffield, A contour line of the continuum Gaussian free field, in preparation.

Nuclear Physics B

Uploaded by

Copyright:

Available Formats

Nuclear Physics B

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Nuclear Physics B

Uploaded by

Copyright:

Available Formats

NUCLEAR PHYSICS B

Journal devoted to the experimental and theoretical study of the fundamental

Nuclear Physics B 802 (2008) 139

Algebra of transfer-matrices and YangBaxter equations

Received 18 January 2008; accepted 28 April 2008

E-mail addresses: andrei@theory.caltech.edu (A. Mikhailov), ss299@theory.caltech.edu (S. Schfer-Nameki).

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

or inside the Wilson line:

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

In terms of the coordinates of the coset space

The gauge invariance (2.5) acts on g as follows:

There are two versions of the transfer matrix. One is T

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

and same rules for w , .

where a is 0 if ea is an even element of the superalgebra, and 1 if ea is an odd element of the

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

T2 [C2 ]T1 [C1 ],

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

where r is field dependent (dynamical). In fact r is of the order h.

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

Fig. 1. Fusion of transfer matrices.

Fig. 2. Exchange of transfer matrices.

2.2.2. Relation to Poisson brackets

and the equation:

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

Fig. 3. An infinite Wilson line with an operator insertion.

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

Here means the representation dual to .

This is gauge invariant because ad ad ad as a representation of psu(2, 2|4) and therefore

5 If is a trivial (zero-dimensional) representation, then the Wilson line terminates:

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

Fig. 4. Fusion operations F+ , F and G+ .

2.3.2. Split operators

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

2.3.3. Switch operators

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

We will use the notation

2.3.4. Intersecting Wilson lines

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

2.4. Outline of the calculation

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

of this action with the exchange relation is verified in Section 7.

3. Short distance singularities in the product of currents

The collective notations for the generators of psu(2, 2|4) are

e1 (1t) e2 (1t ) e3 (1t ) |0 |0

1 (t1+()F t) 2 (t 1+()F t ) 3 (t 1+()F t )

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

Generally speaking 1 1 1 ta 1 1 means

()j F ()j F ()j F ta 1 1.

We also use the following abbreviations:

We will also use this notation:

Using these notations we can write, for example

3.2. Short distance singularities using tensor product notations

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

4.2. Contribution of triple collisions to 

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

In the calculation of the contribution of J0 J2 we take an average of first taking an integral

A. Mikhailov, S. Schfer-Nameki / Nuclear Physics B 802 (2008) 139

4.2. Contribution of triple collisions to

Adding this to dbl + trpl we get

4.5. The structure of