Nucl - Phys.B v.773

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; 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 2007
Volumes 760787 are scheduled for publication in 84 issues. Publication frequency: weekly. 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
781797), Nuclear Physics B ISSN 0550-3213 (volumes 760787) and Nuclear Physics B Proceedings Supplements
ISSN 0920-5632 (volumes 164174) 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$ 17015 (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 773 (2007) 118
Relation between the pole and the minimally subtracted

mass in dimensional regularization and dimensional
reduction to three-loop order
P. Marquard a , L. Mihaila a , J.H. Piclum a,b, , M. Steinhauser a
a Institut fr Theoretische Teilchenphysik, Universitt Karlsruhe (TH), 76128 Karlsruhe, Germany
b II. Institut fr Theoretische Physik, Universitt Hamburg, 22761 Hamburg, Germany
Received 19 February 2007; accepted 5 March 2007

Available online 15 March 2007
Abstract
We compute the relation between the pole quark mass and the minimally subtracted quark mass in the
framework of QCD applying dimensional reduction as a regularization scheme. Special emphasis is put
on the evanescent couplings and the renormalization of the -scalar mass. As a by-product we obtain the
OS and Z OS in dimensional regularization and thus provide
three-loop on-shell renormalization constants Zm
2
the first independent check of the analytical results computed several years ago.
2007 Elsevier B.V. All rights reserved.
PACS: 12.38.-t; 14.65.-q; 14.65.Fy; 14.65.Ha
1. Introduction
In quantum chromodynamics (QCD), like in any other renormalizable quantum field theory, it is crucial to specify the precise meaning of the parameters appearing in the underlying
Lagrangianin particular when higher order quantum corrections are considered. The canonical
choice for the coupling constant of QCD, s , is the so-called modified minimal subtraction (MS)
scheme [1] which has the advantage that the beta function, ruling the scale dependence of the
coupling, is mass-independent. On the other hand, for a heavy quark besides the MS scheme also
other definitions are importantfirst and foremost the pole mass. Whereas the former definition
* Corresponding author.
E-mail address: piclum@particle.uni-karlsruhe.de (J.H. Piclum).

0550-3213/$ see front matter 2007 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2007.03.010
P. Marquard et al. / Nuclear Physics B 773 (2007) 118
is appropriate for those processes where the relevant energy scales are much larger than the quark
mass the pole mass is the relevant definition for threshold processes. Thus it is important to have
precise conversion formulae at hand in order to convert one definition into the other.
By far the most loop calculations performed within QCD are based on dimensional regularization (DREG) in order to handle the infinities which occur in intermediate steps. It is well
known, however, that it is not convenient to apply DREG to supersymmetric theories since it
introduces a mismatch between the numbers of fermionic and bosonic degrees of freedom in
super-multiplets. In order to circumvent this problem and, at the same time, take over as many
advantages as possible from DREG the regularization scheme dimensional reduction (DRED)
has been invented (see, e.g., Refs. [2,3]). However, as pointed out by Siegel himself, DRED in
its original formulation yields mathematical inconsistencies. One can get rid of them by interpreting the fields as living in an infinite dimensional space, which in turn might lead to violation
of supersymmetric Ward identities at higher loop order [4,5]. The application of DRED to supersymmetric theories leads to a relatively small price one has to pay at the technical level. An
elegant way is to introduce an additional scalar particle (the so-called scalar) at the level of the
Lagrangian and to proceed for the practical calculation of the Feynman diagrams as in DREG.
The situation becomes more complicated in case the symmetry between fermions and bosons in
the underlying theory is distorted, e.g., after some heavy squarks have been integrated out. In
such situations couplings involving the scalar, which are called evanescent couplings, renormalize differently from the gauge couplings and therefore one has to allow for new couplings in
the theory. The same is true if DRED is applied to QCD: in addition to s four new couplings
have to be introduced, each of which has their own beta function governing both the running and
the renormalization. In this paper we take over the notation from [6] and denote them by e (the
coupling between scalars and quarks) and i (i = 1, 2, 3), which describe three different four-
vertices.
The one-loop relation between the MS and pole mass has been considered long ago in Ref. [7].
At the beginning of the nineties the mass relation to two-loop order [8] was one of the first
applications of two-loop on-shell integrals. In a subsequent paper also the two-loop result for
the on-shell wave function counterterm has been obtained [9]. The three-loop mass relation has
been computed for the first time in Refs. [10,11] where the off-shell fermion propagator has been
considered for small and large external momenta. The on-shell quantities have been obtained with
the help of a conformal mapping and Pad approximation. The numerical results of Refs. [10,11]
have later been confirmed in Ref. [12] by an analytical on-shell calculation. The three-loop result
for Z2OS has been obtained in Ref. [13]. In this paper we provide the first independent check of
OS and Z OS in the DREG scheme.
the analytical results for Zm
2
The renormalization scheme based on DRED together with modified minimal subtraction is
called DR. As far as the relation between the DR and the pole mass is concerned one can find
the one-loop result in Ref. [14]. The two-loop calculation [15] has been performed in DRED,
identifying, however, the evanescent coupling e with s . In this paper we will provide the more
general result and furthermore extend the relation to three-loop order.
The paper is organized as follows: In Section 2 we outline the general strategy, provide technical details and derive the result for the on-shell mass and wave function counterterm in the case
of dimensional regularization. The peculiarities of dimensional reduction are explained in Section 3 where all relevant counterterms are listed. Finally, Section 4 contains the main result of our
paper: the relation between the pole mass and the minimally subtracted mass in the framework
of dimensional reduction up to three-loop order.
2. On-shell counterterms for mass and wave function in dimensional regularization

In order to compute the counterterms for the quark mass and wave function one has to put
certain requirements on the pole and the residual of the quark propagator. More precisely, in
the on-shell scheme we demand that the quark two-point function has a zero at the position of
the on-shell mass and that the residual of the propagator is i. Thus, the renormalized quark
propagator is given by
SF (q) =
q 2 Mq2
iZ2OS
q/ mq,0 + (q, Mq )
(1)
i
,
q/ Mq
(2)
where the renormalization constants are defined as

OS
Mq ,
mq,0 = Zm

0 = Z2OS .
(3)
(4)
is the quark field with mass mq , Mq is the on-shell mass and bare quantities are denoted by a
subscript 0. denotes the quark self-energy contributions which can be decomposed as

q mq )2 q 2 , mq .
(q, mq ) = mq 1 q 2 , mq + (/
(5)
OS and Z OS reduces all occurring
The calculation outlined in Ref. [8] for the evaluation of Zm
2
Feynman diagrams to the evaluation of on-shell integrals at the bare mass scale. In particular,
it avoids the introduction of explicit counterterm diagrams. At three-loop order we find it more
convenient to follow the more direct approach described in Refs. [12,13], which requires the calculation of diagrams with mass counterterm insertion. Following the latter reference, we expand
around q 2 = Mq2

q Mq )2 Mq2 , Mq
(q, Mq ) Mq 1 Mq2 , Mq + (/

2

2
q Mq2 +
+ Mq 2 1 q , Mq
q
q 2 =Mq2
2

Mq 1 Mq , Mq

2

2

2

+ 2 Mq , Mq + .
+ (/
q Mq ) 2Mq 2 1 q , Mq
(6)
q
2
2
q =Mq
Inserting Eq. (6) into Eq. (1) and comparing to Eq. (2) we find the following formulae for the
renormalization constants

OS
= 1 + 1 Mq2 , Mq ,
Zm
(7)

OS 1

Z2
(8)
= 1 + 2Mq2 2 1 q 2 , Mq
+ 2 Mq2 , Mq .
q
q 2 =Mq2
If we consider the external momentum of the quarks to be q = Q(1 + t) with Q2 = Mq2 , the
self-energy can be written as

/ Mq )2 q 2 , Mq + tQ
/ 2 q 2 , Mq .
(q, Mq ) = Mq 1 q 2 , Mq + (Q
(9)
Fig. 1. Sample three-loop diagrams. Solid lines denote massive quarks with mass mq and curly lines denote gluons. In
the closed fermion loops all quark flavours have to be considered.
Fig. 2. Three-loop master integrals. Solid lines denote massive and dashed lines are massless scalar propagators.
Q+M
/
Let us now consider the quantity Tr{ 4M 2q } and expand to first order in t
q

Q
/ + Mq
(q, Mq )
Tr
4Mq2

= 1 q 2 , Mq + t2 q 2 , Mq

+ 2 Mq2 , Mq t + O t 2 .
= 1 Mq2 , Mq + 2Mq2 2 1 q 2 , Mq
q
q 2 =Mq2
(10)
OS one only needs to calculate for q 2 = M 2 . To calculate Z OS , one has to

Thus, to obtain Zm
1
q
2
compute the first derivative of the self-energy diagrams. The mass renormalization is taken into
account iteratively by calculating one- and two-loop diagrams with zero-momentum insertions.
Sample diagrams contributing to the quark propagator are shown in Fig. 1. All occurring
(3)
(3)
Feynman integrals can be mapped onto J+ and L+ as given in Eq. (4) of Ref. [16] and to seven
more similar ones which will be listed in Ref. [17].
All Feynman diagrams are generated with QGRAF [18] and the various topologies are identified with the help of q2e and exp [19,20]. In a next step the reduction of the various functions
to so-called master integrals has to be achieved. For this step we use the so-called Laporta
method [21,22] which reduces the three-loop integrals to 19 master integrals. We use the implementation of Laportas algorithm in the program Crusher [23]. It is written in C++ and uses
GiNaC [24] for simple manipulations like taking derivatives of polynomial quantities. In the
practical implementation of the Laporta algorithm one of the most time-consuming operations
is the simplification of the coefficients appearing in front of the individual integrals. This task is
performed with the help of Fermat [25] where a special interface has been used (see Ref. [26]).
The main features of the implementation are the automated generation of the integration-by-parts
(IBP) identities [27] and a complete symmetrization of the diagrams.
The results for the master integrals can be found in Ref. [13]. We have checked the results
numerically with the MellinBarnes method [28,29] and the program MB [30]. We did, however,
find some differences to Ref. [13]: In addition to the 18 master integrals given in that reference,
we found the integral depicted in Fig. 2(a) where the result can be found in Eq. (A.2) of Ref. [16].
Furthermore, we already pointed out in Ref. [16], that the result for Fig. 2(b) (denoted by I16 in
Ref. [13]) is wrong. As it turns out, the result given in [13] corresponds to the integral depicted
in Fig. 2(c). Since there is a one-to-one correspondence between the two integrals with regard to
the IBP relations, it does not matter which is chosen to be the master integral.
Most of the master integrals of Ref. [13] were already calculated in Ref. [21]. However, it turns
out that there are differences in the order O() terms of the integrals I2 I7 of these references.
The expressions given in Ref. [13] agree with our numerical results.
The results for the renormalization constants can be cast into the following form

s () eE (1)
s () 2 eE 2 (2)
OS
Zi +
Zi
Zi = 1 +
4

s () 3 eE 3 (3)
+
Zi + O s4 ,
(11)
4
with i {m, 2}. It is convenient to further decompose the three-loop contribution in terms of the
different colour factors
Zi(3) = CF3 ZiF F F + CF2 CA ZiF F A + CF CA2 ZiF AA

+ CF TF nl CF ZiF F L + CA ZiF AL + TF nl ZiF LL + TF nh ZiF H L

+ CF TF nh CF ZiF F H + CA ZiF AH + TF nh ZiF H H ,
(12)
CF = (Nc2
where
1)/(2Nc ) and CA = Nc are the eigenvalues of the quadratic Casimir operators of the fundamental and adjoint representation of SU(Nc ), respectively. In the case of QCD
we have Nc = 3. TF = 1/2 is the index of the fundamental representation and nf = nl + nh is
the number of quark flavours. nl and nh are the number of light and heavy quark flavours, respectively. The former are considered to be massless, while the latter have mass mq . Although we
have nh = 1 in our applications, we keep a generic label which is useful when tracing the origin
of the individual contributions. s () is the strong coupling constant defined in the MS scheme
with nf active flavours.
The one- and two-loop contributions to the mass renormalization constant are

1
3
3
3
(1)
= CF
+ 1 + L + 2 + 2 + L + L2
Zm
4
4
16
8

1
1
1
1
1
+ 4 + 2 3 + 2 + 2 L + L2 + L3 2 ,
(13)
12
4
16
2
8
and
(2)
Zm

9
45
3
45
9
9
1 199 17 2 1 2
=
+
+ L
+
+ ln 2 3 + L + L2
2
64 16
128 64
2
4
32
16
32

677 205 2
135
7
1
+
+ 3 2 ln 2 2 ln2 2
3 + 4 ln4 2 12a4
256 128
16
40
2

199 17 2
3
45
3
+
+ 2 ln 2 3 L + L2 + L3 CF2
64
32
2
32
8
1111
1
11
97
1
3
185
11
+ 2 2 ln 2 + 3
L L2
2
192
384
12
4
8
96
32
32

8581
1 2 2
13
7 4 1 4
271 2 3 2
+
+
ln 2 + ln 2 + 3 + ln 2 + 6a4
768
1152
2
2
4
80
4

7
1463
1
3
229 2 11 3
+ 2 2 ln 2 + 3 L
L L C A CF
+
192
64
2
4
96 32
71
1
1
5
13
1
+
+ 2 + L + L2
+ 2+
48 96 12
24
8
8

97 2
3 2
581
103
17 2 1 3
+
+ 3 +
+ L + L + L C F TF n l
+
192 288
48
16
24
8

1
5
1 2
1133 227 2
143 1 2 13
+ 2+
+
+ L + L +
48
96
6
24
8
192
288
8

5
7
175
17
1
2 L + L2 + L3 CF TF nh ,
+ 2 ln 2 3 +
(14)
2
48
16
24
8
+
where L = ln(2 /Mq2 ). n is Riemanns zeta function with integer argument n and a4 =
Li4 (1/2). We give the one- and two-loop results to order O( 2 ) and O(), respectively. In general
these terms are necessary for three-loop calculations.
The individual three-loop terms read

1
27
9
63
457 111 2 3 2
FFF
+
L
+
ln 2
Zm
=
256 128
512 512
8
128 3
2

189
81 2 1 14225 6037 2
9
L
L
+ 5 2 ln 2
+ 3
16
256
256
3072
3072
5
153
73 4
1
5
1
3
2 3 + 5 ln4 2 3a4
+ 2 ln2 2 +
4
128
480
16
8
8

1371 333 2 9 2
27
567 2
81 3
+
ln 2 + 3 L
L
L ,
+
(15)
512
512
8
16
512 256

1
33
43 2
33
49
3311
3
F FA
L 2 +
+ 2 ln 2
+
Zm =
768 128
1536 512
16
128 3

379
33 2 1 100247 6545 2 25 2
9
L +
L
+
+
+ ln 2
3 +
32
256
256
9216
9216
36
89 2 2
3995
1867 4 19 2
45
35 4
35
3 +
3 + 5
ln 2 a4
ln 2
72
384
8640
16
16
144
6

14311 1135 2 71 2
71
1797 2 121 3
+ ln 2 3 L +
L +
L ,
+
(16)
1536
1536
48
32
512 256
1322545 1955 2 115 2
121
1679
11413
F AA

ln 2
Zm
=
+
124416
3456
72
576 3 3456 2 20736
1343
179 4 51 2
65
11 4
11
11
3
+ 3 5 +
ln 2 + a4
+ 2 ln2 2 +
36
288
3456
64
32
72
3

13243 11 2 11 2
11
2341 2 121 3
+ ln 2 + 3 L
L
L ,
+
(17)
1728
72
24
16
1152 576

1
3
7 2 1
3
5
65
23
3 2 1
+ L 2
+
+ 3 + L + L
=
+
192 32
384 128
4
64
64

32 3

575
2 2 2
145
119 4 1 4
1091 2 11 2
+
+
ln 2 + ln 2 +
3
+ ln 2
2304 2304
9
9
2160
9
96
5 2 1 2
8
497
1
117 2 11 3
+ ln 2 + 3 L
L L ,
+ a4
(18)
3
384 384
3
4
128 64

1 70763 175 2 11 2
11
121
139
1
F AL
=
+
Zm
+
+
+
+ ln 2
3
3
2
1296 4
15552 432
18
72
432
1 2 2
89
19 4
1 4
4
3 +

ln 2 a4
ln 2 +
9
144
2160
18
3

7 2 1 2
869
1
373 2 11 3
+ + ln 2 + 3 L +
L + L ,
+
(19)
216 72
6
2
288 72
2353
13 2
1
5
35
7
F LL
3
Zm
=
+
+
3
2
18
36
216 1296 7776 108

1 2
89
13 2
1 3
+ L L L ,
(20)
216 18
72
36
5917
13 2 2
1
5
35
FHL
+
+ 3
Zm
=
+
+
3
2
648
3888
108
9
18
108

1 2
143
13 2
1 3
L L L ,
(21)
108 18
36
18

281
1
1
3
17 2 1
3
5
23
3
FFH
+ L 2
+ 3 + L + L2
=
+
Zm
3
192
32
384
128
4
64
64

32

5257 1327 2
5
1
37
91 4 1 4
+ 2 ln 2 2 ln2 2 + 3 +
+ ln 2
2304 6912
36
9
96
2160
9

8
1145 221 2 1 2
1
117 2 11 3
+ ln 2 + 3 L
L L ,
+ a4
(22)
3
384
384
3
4
128 64

1 144959 449 2 32 2
1
11
121
139
F AH
+ 3
+
+ ln 2
=
+
Zm
3
2
1296 4

15552
144
9
72
432
1 2 2
109
43 4 1 2
5
1 4
4
3
+ 3 5
ln 2 a4
+ ln 2
18
144
1080
8
8
18
3

583 13 2 1 2
1
373 2 11 3
+
(23)
+ ln 2 + 3 L +
L + L ,
108 36
6
2
288 72
9481
4 2 11
1
5
35
FHH
+
+ 3
Zm
=
+
+
3
2
18
36
216 1296 7776 135

197 1 2
13 2
1 3
(24)
L L L .
216 9
72
36
FFL
Zm
(1)
(1)
For the wave function renormalization constant, we have at the one-loop level Z2 = Zm .
The two-loop contribution to the wave function renormalization constant reads

1 433 49 2
9
9
51
3
51
9
(2)
+ L
+
+ 2 ln 2 3 + L + L2
+
Z2 =
2
64 16
128 64
2
32
16
32

211 339 2 23 2
297
7
+ ln 2 2 2 ln2 2
3 + 4 ln4 2 24a4
+
256 128
4
16
20

433 49 2
51 2 3 3
2
+ 2 ln 2 33 L + L + L CF2
+
64
32
32
8
11
127
3
215
11
1705
5 2 1 2
+
+ ln 2 + 3
L L2
2
192
384
16
2
4
96
32
32

9907
129
7
1
769 2 23 2
+
+
ln 2 + 2 ln2 2 +
3 4 + ln4 2
768
1152
8
16
40
2

2057 109 2
3
259 2 11 3
2
+
ln 2 + 3 L
L L C A CF
+ 12a4 +
192
192
2
96 32
113
1
1
11
19
1
+
+ 2 + L + L2
+ 2+
48
96
12
24
8
8

3 2
851 127 2
145
23 2 1 3
+
+ 3 +
+ L + L + L C F TF n l
+
192 288
48
16
24
8

1
17971 445 2
1 947
3 2
1
5 2 11
+
+ L
+
+ L + L +
16 4
288 16
24
8
1728
288

1043 29 2
85
5
7
+ 2 2 ln 2 3 +
L + L2 + L3 CF TF nh . (25)
12
144
48
8
24

Again, we give the result to order O(), which is necessary for three-loop calculations.
The individual three-loop terms are given by

1
27
9
81
1039 303 2 3 2
+
L
+
ln 2
Z2F F F =
3
2
256 128
512
512
4
128

9
243
81 2 1 10823 58321 2 685 2
+ 3
L
L
+
ln 2
8
256
256
3072
9216
48
739
41 4 1 2
5
5 4
+ 3 2 ln2 2
3
+ 3 5
ln 2 10a4
128
120
8
16
12

3117 909 2 9 2
27
729 2
81 3
+
ln 2 + 3 L
L
L ,
+
(26)
512
512
4
8
512
256

1
33
33
95
1787 131 2 3 2
+ ln 2
Z2F F A =
+
+
3
2
768 128
512
512
8
128

5
469
33 2 1 136945 29695 2 755 2
L +
L
+
+

ln 2
3 +
8
256
256
9216
9216
288
235 2 2
6913
1793 4 45 2
145
55 4
55
ln 2
3 +
3 +
5
ln 2 a4
72
384
3456
16
16
144
6

25609 3335 2 71 2
37
2155 2 121 3
+ ln 2 3 L +
L +
L ,
+
(27)
1536
1536
24
8
512 256

3
3
121
2009
12793
1
1
4
+
+
3
Z2F AA =
+
+
+
3
20736 128
1080
768 256
576 3 3456 2

1 1654711 4339 2 325 2
1
127 2 2
4

ln 2 +
ln 2
4320

124416
3456
144
144
5857
3419 4 127 2
37
85 4
85
+
3
+
3 5 +
ln 2 + a4
576
23040
72
6
288
12

1 2
13
13
17
1 2
7
4

3 +
+
3 +
5
+
768 256
256
27648
144
384
36977 55 2 11 2
167
1 4
+ ln 2 +
3
+
3456
96
12
128
360

9
1
1
2671 2 121 3
3 +
4 L
L
L ,
+
(28)
256 256
1440
1152 576

1
1
3
7 2 1
3
19
235
41
3
+ L 2
+
+ 3 + L + L2
+
Z2F F L =
3
192 32
384 128
4
64
64

32

4 2 2
473
229 4 2 4
3083 2845 2 47 2
+
ln 2 + ln 2 +
3
+ ln 2
2304 2304
18
9
96
2160
9

16
1475 133 2 2 2
1
179 2 11 3
+
ln 2 + 3 L
L L , (29)
+ a4 +
3
384
384
3
4
128 64

1 111791 13 2 47 2
1
11
169
313
+
+
+ + ln 2
+
Z2F AL =
3
3
2
1296 4

15552
48
36
72
432
2 2 2
35
19 4 1 4
8
ln 2 a4
ln 2 3 +
9
72
1080
9
3

169
1
469 2 11 3
1 2 1 2
+
(30)
+ ln 2 + 3 L +
L + L ,
27
18
3
4
288 72
5767
19 2
1
11
5
7
3
Z2F LL =
+
+
3
2
18
36
216 1296 7776 108

1 2
167
19 2
1 3
+ L L L ,
(31)
216 18
72
36

1
1
1
1 2 1
1
5
1
+ L 2 +
+
L + L2
Z2F H L =
36 12
216 144
9
24

25 2
4721 19 2
1
329
7
5
+ 3 +
+
L L2 L3 ,
(32)
1296 54
36
108 144
12
72

1
1
3
7
707 15 2 29
15
+ L 2
+ L + L2
Z2F F H =
192 16
384 64
64
32

7
1
1763
31 4 1 4
76897 11551 2
+ 2 ln 2 2 ln2 2 +
3 +
+ ln 2
6912 20736
18
2
288
720
2

2891 233 2 2 2
143 2 19 3
+ 12a4 +
(33)
+
ln 2 + 3 L
L L ,
384
192
3
128 32

1
1
41
1
41 2
1
7
13
+
+ L 2 +
Z2F AH =
72 64
192 64
216 2304
192 3

83
1
1 2
3
3
35
35
+
+
+ L
+
L2
576 768
144 64
384 128

1
77
17 4 11 2
49901 36019 2 80 2
+
+ ln 2 + 2 ln2 2 3
+ 3
2592
5184
9
3
16 360
48
1 2
15
1 4
407
7
+

3
5 ln 2 8a4 +
16
3
1728 256
192

1 2
4141 641 2 1 2
1
35
+ ln 2 3
+
L
+
432
768
3
2
192 256
10
Z2F H H

35
247
9
3
2
+
+
L +
L3 ,
16 128
1152 128

1
5
7
1 2 1 8425
2
=
+
L
+ 2 + 3
2
432 12
2592 45
3
72

481
11
1
5
2 L L2 L3 .
216 24
72
6
(34)
(35)
Starting from the three-loop level, the wave function renormalization constant depends on the
gauge parameter, . The parameter in the above equations is defined through the gluon propagator
as

k k ab
i
ab
D (k) = 2 g 2 ,
(36)
k
k
where a and b are colour indices.
OS and Z OS agree with the literature [12,13].
We want to mention that our results for Zm
2
Whereas in [12,13] they are expressed in terms of the bare coupling we decided to use the
renormalized s as an expansion parameter which to our opinion is more convenient in practical applications.
The genuine three-loop integrals which appear in the DREG calculation are the same for
DRED. The main complication is the more involved renormalization which is discussed in more
detail in the next section.
3. Renormalization in DRED
Let us in this section collect the DRED counterterms needed for our calculation. In addition to
the strong coupling s also the evanescent coupling1 e has to be renormalized to two-loop order.
The evanescent couplings2 1 , 2 and 3 appear for the first time at three-loop order and thus
no renormalization is necessary. Both for the heavy quark mass, mq , and the -scalar mass, m ,
two-loop counterterms are necessary. Whereas the couplings are renormalized using minimal
subtraction the masses are renormalized on-shell. The corresponding counterterms are defined
through

2
s0,DR = 2 ZsDR sDR ,
OS,DR
mq0,DR = Mq Zm
,
e0 = 2 (Ze )2 e ,
0 2
OS
m = m2 Zm
.
(37)
We attach an additional index DR to the quark mass renormalization constant in order to remind that it relates the pole mass to the bare mass in the DRED scheme3 (in this context see also
Ref. [32]).
Recently the quantities Ze and ZsDR have been computed to three- and four-loop order [6,
32], respectively. The results have been presented in terms of the corresponding functions. For
1 We refer to [6,31] for a precise definition of the evanescent couplings.
2 For the SU(3) gauge group, which we exclusively consider in this paper, there are three independent such cou-
plings [31].
3 In principle such an index would also be necessary in Section 2. However, since the MS scheme in connection with
DREG constitutes the standard framework we refrain from introducing an additional index there.
11
completeness we present in the following the two-loop results for the renormalization constants

DR 2

sDR 1
1
11
s
1 121 2 11
=1+
CA + T F n f +
C C A TF n f

24
6
2 384 A 48

1 2 2
5
1
17 2
1
+ nf TF +
(38)
CA + CA TF nf + CF TF nf ,
24

96
48
16

sDR 1
1
1
3
1
e 1
Ze = 1 +
CF +
CA + CF + TF nf

4

4
2
4
DR 2

s
9 2 1
55
1 7 2
1 11
+
CA CF + CF CF TF nf +
C
CA CF
32
8
256 A 192
2 32

3
1
5
3
DR e 1 3
CF2 CA TF nf + CF TF nf
CA CF CF2
+ s
64
32
48
2 8
4

3
11
5
1 3 2 5
CF TF nf +
C CA CF + CF2 + CF TF nf
8
32 A 8
16
32

2

3 2
3
3
3 2 2
1 3 2 3
e
C
+
T
n
+
T
n
+
n
C
C
C
C
C
T
+
A F
A F f
F F f
8 F 16
8
32 F f
2 32 A 8

1
5
1
3
3
3
+
CA2 + CA CF CF2 + CA TF nf CF TF nf

32
16
4
32
16

2
2
2 5
3 3
2 1 15 1 3 1 9
1 1 27
e 1 1 9
+
+
+
32
16
16
256
16
64
2
3 1 21
(39)
.
128
ZsDR
Note that we set Nc = 3 in all terms containing the couplings i since our implementation of
these couplings is only valid for SU(3). We would also like to point out that the analytical form
of ZsDR is identical to the corresponding result in the MS scheme. This has been shown by an
explicit calculation in Ref. [3]. The one-loop result for Ze can be found in Ref. [31].
OS,DR
The one-loop corrections to Zm
can be found in Ref. [14] and the two-loop terms have
been computed in Ref. [15]. For our calculation we also need the O( 2 ) and O() parts of the
one- and two-loop terms, respectively.
OS,DR
DR .
In Section 4 we want to present the finite result obtained by considering the ratio Zm
/Zm
DR
The quantity Zm has been computed in Ref. [32] to three and in Ref. [6] even to four-loop order.
Whereas in [6,32] only the anomalous dimensions are given we want to present the explicit result
for the renormalization constant

DR 2

DR 1
9 2 1
3
s
1 11
DR
=1+ s
C
+
T
n
CF +
C
C
C
Zm
A F
F F f

4
32 F 8
2 32

1
3
5
91
DR e 3 1 2
+
CA CF CF2 + CF TF nf
CF
+ s

192
64
48
16
2
DR 3

e 1 1
1
1
s
1
121 2
+
CA CF CF2 CF TF nf +
C CF
16
8
16
576 A

12

33
9 3 11
3 2
1
2
2 2
CA CF
C + CA CF TF nf + CF TF nf CF TF nf
128
128 F 72
32
36

1 1613 2
295
9
59
29 2
+ 2
CA CF +
CA CF2 +
CF3
CA CF TF nf
C TF n f
768
256
216
192 F
3456

5
133
43 3
10255 2
1
+
CF TF2 n2f +
C A CF +
CA CF2
C
216

20736
768
128 F

281
23 1
35
1
2
2 2
+
+ 3 CA CF TF nf +
3 CF TF nf +
CF TF nf
2592 4
96 4
1296
DR 2

e 1
15 3
1 2
s
11
1 5 2
2
+
CA CF CF + CF TF nf +
C CF
2
192
64
48
256 A

7
9
3
9
DR e 2 1
9
+ CA CF2 + CF3 CF2 TF nf
+ s
CA CF2 + CF3
2
32
64
32
64
32

9 2
7
3
1
1 2
1
+ CF TF nf +
CA CF + CA CF2 CF3 CA CF TF nf
64

64
32
8
64
3

1
1
1
1
e
1
CF2 TF nf
+
CA2 CF + CA CF2 CF3
8
48
12
12
2

1
1 2
1
1
1 1 2
2 2
+ CA CF TF nf CF TF nf CF TF nf +
C CF CA CF2
24
12
48
32 A
8

1
1
5
1
1 e 2 1 1
+ CF3 CA CF TF nf + CF2 TF nf + CF TF2 n2f
8
24
48
96
8

2
2
2
2
1 e 3 1
3 e 1 1
5 e 2 1
5 e 2 1
+
+
+
36
12
64
12

2
1 e 1 3 1
7 e 3 1
.
+
(40)
96
16
In order to achieve the finite result for the relation between the pole and the DR quark mass
it is necessary to fix a renormalization scheme also for the mass of the scalar, m . Although,
there is in general no tree-level term in the Lagrangian, there are loop induced contributions to m
which require the introduction of corresponding counterterms. The relevant Feynman diagrams
contributing to the -scalar propagator show quadratic divergences and therefore, one needs to
consider only contributions from massive particles. Thus, in our case, only diagrams involving a
massive quark have to be taken into account. Some sample diagrams are shown in Fig. 3.
It is common practice to renormalize m on-shell and require that the renormalized mass is
zero to each order in perturbation theory [33]. This scheme is known as the DR scheme [33] and
offers the advantage that the -scalar mass completely decouples from the physical observables.
For supersymmetric theories the DR and DR renormalization schemes are the same, while for
theories with broken supersymmetry the latter one is most convenient. At one-loop order there is
only one relevant diagram (cf. Fig. 3(a)) which has to be evaluated for vanishing external momentum. A closer look at the two-loop diagrams shows that they develop infrared divergences in the
limit m 0 (cf., e.g., Fig. 3(e)). They can be regulated by introducing a small but non-vanishing
mass for the scalars. After the subsequent application of an asymptotic expansion [34] in the
limit q 2 = m2 Mq2 the infrared divergences manifest themselves as ln(m ) terms. Furthermore, one-loop diagrams like the ones in Fig. 3(b) and (c) do not vanish anymore and have to
13
Fig. 3. One- and two-loop Feynman diagrams contributing to the -scalar propagator. Dashed lines denote scalars, curly
lines denote gluons and solid lines denote massive quarks with mass mq .
be taken into account as well. Although they are proportional to m2 , after renormalization they
induce two-loop contributions which are proportional to Mq2 , partly multiplied by ln(m ) terms.
It is interesting to note that in the sum of the genuine two-loop diagrams and the counterterm
contributions the limit m 0 can be taken which demonstrates the infrared finiteness of the
on-shell mass of the scalar.
Taking the infrared finiteness for granted, it is also possible to choose q 2 = m2 = 0 from the
very beginning. Then the individual diagrams are infrared divergent, however, the sum is not. We
have performed the calculation both ways and checked that the final result is the same. It is given
by

Mq2 OS
2
e
1 2
2
+ 2 + 2L + 2 + + 2L + L
Z = 1 n h TF

6
m2 m

DR 2
31 1 3
1 3
sDR e
3
s
+ + L CA +
n h TF 2
CA + CF
n h TF
4 4 2

2
8

1 7
15
1
3
3
3
+
CA + CF +
CA + CF L +
+ 2 CA
8
2
4
2
8
16

3 1 2
7
3
3
3
+ CF +
CA + CF L +
CA + CF L2
+
2 8
4
2
4
4

2
1 1
1 1
1
1
e
CA CF TF n f +
CF
n h TF 2
+
2
2
2
4

1
1
1
5
1
+ 2 TF n f
(1 + L )TF nf CA + CF
2
2
2
2 24

1
1
1
1
1
CA 2CF + TF nf L
CA CF + TF nf L2
2
2
4
2
4

e 1
3 1
3
3
1 2 3
1 3
3 2
+
nh
+ L +
+ + L + L
+

16 2 16 8
16 32
8
8
14

5
5 2
2
5 2
5 1
1 25
+ 5 + L
nh
+
+ + 10L + L
4 2
2

2
24
2

7
7 2
3
7
7
7 2 7
7 1
1
+
+
+
nh
+
L
+
+
L
L
,
16 2
16 8
16 96
8
8
(41)
where the constants are defined after Eq. (14). The overall factor nh in front of the one- and twoloop corrections shows that the renormalization of m only influences those terms which contain
a closed heavy quark loop.
OS in dimensional reduction
4. Zm
The approach to extract the on-shell mass counterterm in DRED can be taken over from
DREG as described in Section 2, i.e. one considers the inverse quark propagator and requires
that it has a zero at the position of the pole. Again, the counterterm diagrams are generated
order-by-order in a generic way.
The major complication as compared to the calculation in DREG is the appearance of the
evanescent couplings and the scalars. In particular, there are three different four- vertices.
This leads to many more Feynman diagrams which have to be considered. Whereas in the case
of DREG about 130 diagrams contribute there are more than 1100 in the case of DRED. Typical
Feynman diagrams are shown in Figs. 1 and 4. The more involved renormalization has already
been discussed in Section 3.
OS,DR
but show the result for the
We avoid to present the result for the divergent quantity Zm
ratio to the DR quantity which is finite. We cast our result in the following form
OS,DR
=
zm
OS,DR
Zm
DR
Zm
mDR
q
MqOS
OS,DR
OS,DR
OS,DR
= 1 + (1) zm
+ (2) zm
+ (3) zm
.
(42)
Since our implementation of the couplings i is only valid for SU(3) we furthermore set Nc = 3.
Our one-, two- and three-loop results read:
OS,DR
(1) zm

sDR 4
1 e
=
+ L
,
3
3
(43)
Fig. 4. Sample three-loop diagrams contributing to the quark propagator which have to be considered additionally in
case dimensional reduction is used for the regularization. Solid lines denote massive quarks with mass mq , curly lines
denote gluons and the scalars are represented by dashed lines. In the closed fermion loops all quark flavours have to be
considered.
15
OS,DR
(2) zm
sDR 2 3143 2 2 1 2
1
151
7
=
ln 2 + 3
L L2
288
9
9
6
24
8

71
1 2
143 1 2 13
1 2 13
+
+ + L + L n l +
+ L
144 18
36
12
144 9
36

2

DR
1 2
3 1
1
s e
e
1
+ L nh +
+ L +
+ nf ,
12

8 3
6 48
(44)
OS,DR
(3) zm
DR 3
s
7
67
1160387 24707 2 38 2
=
ln 2 + 2 ln2 2 + 3
10368
2592
9
27
72

341 4 1331 2
1705
19 4
76
20089
+
+
3
5 +
ln 2 + a4
+ 2
2592
432
216
54
9
288

1 2
3
1475 2 21 3
42235 923 2 11 2
+ ln 2 3 L
L L +
+
+ ln 2
2
4
96 16
3888
648
81

2
707
61 4
1 4
8
3463
35 2
2 ln2 2 +
3

ln 2 a4 +
+
81
216
1944
81
27
432
108

1
7
35
2
2353
7
13 2
+ 2 ln 2 + 3 L + L2 + L3 nl
+
+ 3
27
9
16
9
23328 324
54

89
13 2
1 3 2
5917
2
1 2
13 2
+
+ L +
L +
L n l
3
648 54
216
108
11664 324
27

143
13 2
1
77065 13375 2
1
+
2 L +
L + L3 nl n h +
324 54
108 54
3888
1944
640 2
1
751
41 4 1 2
5
1 4
+
ln 2 + 2 ln2 2
3
+ 3 5
ln 2
81
81
216
972
4
4
81

8
4435 23 2
1
7
35
2
a4 +
+ 2 ln 2 + 3 L + L2 + L3 nh
27
432
54
27
9
16
9

9481
197
13 2
1 3 2
4 2 11
1 2
3 +
L +
L +
L nh
23328 405
54
648 27
216
108
DR 2
e 41105
1
7
35
7
2
s
+ 2 + 2 ln 2 + 3 + L + L2
+
20736 27
27
24
12
24

27
11
1
113
11
1
1
1
+ 2 + L + L2 nl
2 + L + L2 nh
64 54
54
36
192 27
54
36

2
1
55
1
5
5
1397
DR e
3 L +
+ 3 L n f
+ s
2592 36
6
1728 36
48
3
2

e
31
5 2
5
7
5 e 2
+
3
nf +
nf
144 216
576
576
24
2
2
2
9 e 1
15 e 2
7 e 3
3 e 1 3
(45)
+
+
.
256
16
128
64
For e = s the two-loop result agrees with [15], while the three-loop one is new.
16
There is a very strong cross check of the results in Eqs. (43), (44) and (45) for the limit
nh = 0. The starting point is the relation between the MS and the on-shell mass in DREG which
can easily be obtained from the results of Section 2 and the MS counterpart of Eq. (40) (see,
e.g., Ref. [35]). In this relation, which depends on sMS , both mMS and sMS are replaced by their
DRED counterparts using Eqs. (4.2) and (4.3) of Ref. [6]. In this way we could verify the results
DR,OS
|nh =0 . This provides a strong consistency check both on the results presented in this
for zm
paper but also on the approach used in [6] for the extraction of the conversion formulae between
the MS and DR quantities.
In order to get an impression of the numerical size of the corrections, both in DREG and
DRED, let us consider the relation between the minimally subtracted and the pole mass for the
(5),MS
case of the bottom and top quark. As input we use s

(MZ ) = 0.1189 [36], Mb = 4.800 GeV
and Mt = 171.4 GeV [37].
(5),MS
(5),MS
(MZ ) to s
(Mb ) and
In the framework of DREG it is straightforward to convert s
(6),MS
(Mt ) using four-loop accuracy4 leading to
bottom:
mMS
b (Mb ) = 3.953 GeV = 4.800(1 0.0929 0.0493 0.0342) GeV
= (4.800 0.445 0.236 0.164) GeV,
top:
(46)
mMS
t (Mt ) = 161.1 GeV = 171.4(1 0.0461 0.0109 0.0033) GeV
= (171.4 7.9 1.9 0.6) GeV,
(5),MS
(47)
(6),MS
with s
(Mb ) = 0.2188 and s
(Mt ) = 0.1085. The numbers given in the round brackets
of the above equations indicate the contributions from the tree-level, one-, two- and three-loop
conversion relation.
Within DRED the numerical analysis gets more involved since four more couplings appear
whose values are needed for = Mb and = Mt . As mentioned in the Introduction, DRED is
an appropriate scheme for supersymmetric theories where in the strong sector only one coupling
constant is presentlike in usual QCD using DREG. Since all supersymmetric particles are
heavier than the electroweak scale it is necessary to match the full theory to the Standard Model
at some properly chosen scale dec . At this step the additional couplings appear.
(5)
(5)
For the computation of e (Mb ) and i (Mb ) we use dec = MZ , evaluate in a first step
(5),DR
(MZ ) using the three-loop relation given in Ref. [6] and require
(5)
s(5),DR (MZ ) = e(5) (MZ ) = 1 (MZ ),
(5)
(5)
2 (MZ ) = 3 (MZ ) = 0.
(48)
In a second step the renormalization group functions, which are known to four (sDR ), three (e )
and one-loop order (i ) [6,32] are used to obtain
s(5),DR (Mb ) = 0.2241,
(5)
1 (Mb ) = 0.2152,
e(5) (Mb ) = 0.1721,

(5)
2 (Mb ) = 0.01798,
(5)
3 (Mb ) = 0.005777.
(49)
In the case of the top quark we choose dec = Mt . It is necessary to know the couplings for six
(6),MS
active flavours and thus we evaluate in a first step s
(MZ ) = 0.1178 and pose the analogue
4 We use RunDec [38] for the running and decoupling of in the MS scheme.
s
17
requirements as in Eq. (48) with (5) replaced by (6) and MZ by Mt . This leads to
(6)
s(6),DR (Mt ) = e(6) (Mt ) = 1 (Mt ) = 0.1096,
(6)
(6)
2 (Mt ) = 3 (Mt ) = 0.
(50)
Finally, we can insert the results of Eqs. (49) and (50) into Eq. (42) and obtain
bottom:
mDR
b (Mb ) = 3.859 GeV = 4.800(1 0.1134 0.0495 0.033) GeV
top:
mDR
t (Mt ) = 159.0
= (4.800 0.544 0.238 0.159) GeV,
(51)
GeV = 171.4(1 0.0581 0.0105 0.0030) GeV

= (171.4 10.0 1.8 0.5) GeV.
(52)
The perturbative expansion shows a similar behaviour as for the MS-on-shell mass relation:
the three-loop terms amount to 159 MeV and 500 MeV, respectively, and are thus far above the
current uncertainty for the bottom quark mass [39] and much larger than the expected uncertainty
for the top quark mass [40].
5. Conclusions
The main result of this paper is the relation between the pole and DR quark mass to threeloop order in QCD where dimensional reduction has been used as a regularization scheme. The
conversion formula has been obtained in analytical form where all occurring integrals have been
reduced to a small set of master integrals with the help of the Laporta algorithm. Due to the
occurrence of evanescent couplings when using DRED within QCD it is more advantageous
to use dimensional regularization in this case. However, the latter cannot be used in supersymmetric models. Thus, our result constitutes an important preparation for similar calculations in
supersymmetric extensions of the Standard Model.
As a by-product we have obtained the corresponding relation and the on-shell wave function
renormalization using dimensional regularization. This constitutes the first check of the analytical
results obtained about seven years ago. Furthermore, we can confirm the observation of Ref. [13]
that Z2OS depends on the gauge fixing parameter starting from three-loop order.
Acknowledgements
We would like to thank Stefan Bekavac for discussions about the MellinBarnes method and
checking some of our results. This work was supported by the Impuls- und Vernetzungsfonds
of the Helmholtz Association, contract number VH-NG-008 and the DFG through SFB/TR 9.
The Feynman diagrams were drawn with JaxoDraw [41].
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
W.A. Bardeen, A.J. Buras, D.W. Duke, T. Muta, Phys. Rev. D 18 (1978) 3998.
W. Siegel, Phys. Lett. B 84 (1979) 193.
D.M. Capper, D.R.T. Jones, P. van Nieuwenhuizen, Nucl. Phys. B 167 (1980) 479.
L.V. Avdeev, G.A. Chochia, A.A. Vladimirov, Phys. Lett. B 105 (1981) 272.
D. Stckinger, JHEP 0503 (2005) 076, hep-ph/0503129.
R.V. Harlander, D.R.T. Jones, P. Kant, L. Mihaila, M. Steinhauser, JHEP 0612 (2006) 024, hep-ph/0610206.
R. Tarrach, Nucl. Phys. B 183 (1981) 384.
N. Gray, D.J. Broadhurst, W. Grafe, K. Schilcher, Z. Phys. C 48 (1990) 673.
D.J. Broadhurst, N. Gray, K. Schilcher, Z. Phys. C 52 (1991) 111.
18
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
K.G. Chetyrkin, M. Steinhauser, Phys. Rev. Lett. 83 (1999) 4001, hep-ph/9907509.

K.G. Chetyrkin, M. Steinhauser, Nucl. Phys. B 573 (2000) 617, hep-ph/9911434.
K. Melnikov, T. van Ritbergen, Phys. Lett. B 482 (2000) 99, hep-ph/9912391.
K. Melnikov, T. van Ritbergen, Nucl. Phys. B 591 (2000) 515, hep-ph/0005131.
S.P. Martin, M.T. Vaughn, Phys. Lett. B 318 (1993) 331, hep-ph/9308222.
L.V. Avdeev, M.Y. Kalmykov, Nucl. Phys. B 502 (1997) 419, hep-ph/9701308.
P. Marquard, J.H. Piclum, D. Seidel, M. Steinhauser, Nucl. Phys. B 758 (2006) 144, hep-ph/0607168.
J.H. Piclum, Dissertation, Universitt Hamburg, in preparation.
P. Nogueira, J. Comput. Phys. 105 (1993) 279.
R. Harlander, T. Seidensticker, M. Steinhauser, Phys. Lett. B 426 (1998) 125, hep-ph/9712228.
T. Seidensticker, hep-ph/9905298.
S. Laporta, E. Remiddi, Phys. Lett. B 379 (1996) 283, hep-ph/9602417.
S. Laporta, Int. J. Mod. Phys. A 15 (2000) 5087, hep-ph/0102033.
P. Marquard, D. Seidel, unpublished.
C. Bauer, A. Frink, R. Kreckel, cs.SC/0004015.
R.H. Lewis, Fermats User Guide, http://www.bway.net/~lewis.
M. Tentyukov, J.A.M. Vermaseren, cs.SC/0604052.
K.G. Chetyrkin, F.V. Tkachov, Nucl. Phys. B 192 (1981) 159.
V.A. Smirnov, Phys. Lett. B 460 (1999) 397, hep-ph/9905323.
J.B. Tausk, Phys. Lett. B 469 (1999) 225, hep-ph/9909506.
M. Czakon, Comput. Phys. Commun. 175 (2006) 559, hep-ph/0511200.
I. Jack, D.R.T. Jones, K.L. Roberts, Z. Phys. C 62 (1994) 161, hep-ph/9310301.
R. Harlander, P. Kant, L. Mihaila, M. Steinhauser, JHEP 0609 (2006) 053, hep-ph/0607240.
I. Jack, D.R.T. Jones, S.P. Martin, M.T. Vaughn, Y. Yamada, Phys. Rev. D 50 (1994) 5481, hep-ph/9407291.
V.A. Smirnov, Applied Asymptotic Expansions in Momenta and Masses, Springer Tracts in Modern Physics,
vol. 177, 2002, p. 1.
K.G. Chetyrkin, Nucl. Phys. B 710 (2005) 499, hep-ph/0405193.
S. Bethke, hep-ex/0606035.
Tevatron Electroweak Working Group, hep-ex/0608032.
K.G. Chetyrkin, J.H. Khn, M. Steinhauser, Comput. Phys. Commun. 133 (2000) 43, hep-ph/0004189.
J.H. Khn, M. Steinhauser, C. Sturm, hep-ph/0702103.
M. Martinez, R. Miquel, Eur. Phys. J. C 27 (2003) 49, hep-ph/0207315.
D. Binosi, L. Theussl, Comput. Phys. Commun. 161 (2004) 76, hep-ph/0309015.
Direct detection of dark matter rates for various wimps

V.K. Oikonomou a , J.D. Vergados b, , Ch.C. Moustakidis a
a Department of Theoretical Physics, Aristotle University of Thessaloniki, 54124 Thessaloniki, Greece
b University of Ioannina, Ioannina GR 45110, Greece
Received 20 January 2007; received in revised form 6 March 2007; accepted 7 March 2007
Abstract
The event rates for the direct detection of dark matter candidates, originating from the universal extra
dimension scenario, are evaluated for a number of nuclear targets. Realistic form factors as well as spin ME
and response functions are employed. Due to LR + RL helicities contribution, the proton amplitude is found
to be dominant. Various other non-susy dark matter candidates are examined at the end.
PACS: 95.35.+d; 12.60.Jv
1. Introduction
Models with compact extra dimensions offer rich and interesting phenomenology [18]. In
such models fields propagating in extra dimensions at low energies appear as a tower of massive
particles corresponding to a given charge and spin. The massive states are nothing but modes of
the fields carrying quantized momentum in extra dimensions. This means that the spacing of the
towers is 1/R, i.e. the inverse of the characteristic size in extra dimensions. In this scheme the
ordinary particles are associated with the zero modes. In brane world models only fields interacting gravitationally can propagate in extra dimensions, i.e. the excitations are of the KaluzaKlein
(KK) type. Models with Universal Extra Dimensions (UED) can have stable particles, due to KK
parity, originating from higher-dimensional Poincar invariance [6]. Under this parity the even
modes, including the ordinary particles, are even and the odd modes are odd. Thus the lightest
E-mail address: vergados@cc.uoi.gr (J.D. Vergados).

doi:10.1016/j.nuclphysb.2007.03.014
20
V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942
odd mode particle is cosmologically stable. For a recent review we refer the reader to Ref. [9].
Like the neutralino, it must be a neutral a weakly interacting particle. It can thus serve as a viable dark matter candidate, which, together with dark energy, seems to dominate in the Universe
[1016].
The KK WIMPs (Weekly Interacting Massive Particles) velocity dependence can be assumed
to be the same with that used in neutralino calculations, since a particles rotational velocity is
independent of its mass. The kinematics involved is, by and large, similar to that of the neutralino, leading to cross sections which are proportional r , the WIMP-nucleus reduced mass.
Furthermore the nuclear physics input, which depends on r Amp , is expected to be the same.
There are appear two differences compared to the neutralino, though, both related to its larger
mass.
The density (number of particles per unit volume) of a WIMP falls inversely proportional to
its mass. Thus, since in KK theories the WIMP mass is much larger than that of the target, for
a given WIMP-nucleon gross section, the event rate is inversely proportional to the WIMP
mass. This means that the limits on the nucleon cross section extracted from the data must be
rising with the square root of the WIMP mass. This allows for nucleon cross sections larger
than those extracted for the neutralino.
The average neutralino energy is now quite higher. In fact for a MaxwellBoltzmann (MB)
velocity distribution one finds that TWIMP = 34 MWIMP 02 , with 0 the characteristic velocity
of the MB distribution, which coincides with the suns rotational velocity, 0 220 km/s.
Thus one finds
mWIMP
TWIMP 40
100 GeV

keV.
Furthermore, since the maximum allowed velocity is esc = 2.840 , we find that

mWIMP
Tmax 120
keV.
100 GeV
Thus for a KK WIMP with mass 1 TeV, the average WIMP energy is 0.4 MeV and the
maximum energy is 1.2 MeV. Thus in this case, due to the high velocity tail of the velocity
distribution it is reasonable to expect an energy transfer to the nucleus in the MeV region.
So one need not attempt to detect such a heavy WIMP the hard way, i.e. by measuring the
energy of the recoiling nucleus, as in the case of the neutralino. Many nuclear targets can
now be excited by the WIMP-nucleus interaction and the de-excitation photons can be easily
detected.
2. The KaluzaKlein boson as a dark matter candidate
Assuming small boundary terms we expect that the lightest exotic particle, which can serve as
a dark matter candidate, is a gauge boson B 1 having the same quantum numbers and couplings
with the Standard Model gauge boson B, except that it has KK parity 1. Thus its couplings
involve another negative KK parity particle. In this work we will assume that such particles are
the KK quarks, partners of the ordinary quarks, but much heavier [1,2].
(a)
21
(b)
Fig. 1. Two diagrams leading to the interaction of KK gauge boson B 1 with quarks at tree level mediated by KK quarks.
2.1. Intermediate KK quarks

In this case a generic Feynman diagrams are shown in Fig. 1. The corresponding amplitude
involving left-handed quarks is given by:

(p
/q p
/ B + mq (1) )
2
ML
(1)
(q)
=
i(g
Y
)
q
(x)
q
(x)
(pB ) (pB ),
1
L
L
L
1
(pq pB )2 m2q (1)

(p
/q +p
/ B + mq (1) )
2
(q)
=
i(g
Y
)
q
(x)
q
(x)
(pB ) (pB ),
ML
(2)
1 L
L
L
2
(pq + pB )2 m2q (1)

where g1 = g tan W = 4 2GF mW tan W , YL = 1/3 and (pB ), (pB ) are the helicities of
the KK bosons. For the right-handed quarks of the upper type we have,

(p
/q p
/ B )
R
2
M1 (u) = i(g1 4/3) uR (x)

(3)
uR (x) (pB ) (pB ),
(pq pB )2 m2q (1)

(p
/q +p
/B)
2
(u)
=
i(g
4/3)
(x)
u
(x)
(pB ) (pB ).
u
MR
(4)
1
R
R
2
(pq + pB )2 m2q (1)
For the right handed down quarks we have,

2
R
M1 (d) = i g1 (2/3) dR (x)
MR
2 (d) = i

2
g1 (2/3) uR (x)
(p
/q p
/ B )
(pq pB )2 m2q (1)
(p
/q +p
/B)
(pq + pB )2 m2q (1)

dR (x) (pB ) (pB ),
(5)

dR (x) (pB ) (pB ).
(6)
We also have RL interference terms, which seem to have been missed in the previous calculations
[1,2]. One finds:
MLR
1 (u)
= i(g1 ) 4/9 uL (x)

2
MLR
2 (u)
(pq pB )2 m2q (1)
= i(g1 ) 4/9 uL (x)

2
mq (1)
mq (1)
(pq + pB )2 m2q (1)
uR (x) + H.C. (pB ) (pB ),
(7)

uR (x) + H.C. (pB ) (pB ),
(8)
22
MLR
1 (d)

= i(g1 ) (2/9) dL (x)
2
mq (1)
(pq pB )2 m2q (1)
MLR
2 (d)

= i(g1 ) (2/9) dL (x)
2

dR (x) + H.C. (pB ) (pB ), (9)
mq (1)

dR (x) + H.C. (pB ) (pB ).
(pq + pB )2 m2q (1)
(10)
Since both the KK bosons and quarks are very massive and the momenta of the external particles
are quite small one can employ the non-relativistic limit. Thus the helicities of the KK bosons are
space like. Thus to leading order the amplitude corresponding to Eqs. (1) and (2) can be written:
1

2
( . ) 0 i( ) . 5 q(x)
ML
q = ig1 (1/9)q(x)
2

(Eq + mB (1) )
(Eq mB (1) )
+
.
(mB (1) + Eq )2 m2q (1)
(mB (1) Eq )2 m2q (1)
(11)
The first term in the square bracket is spin independent and it can lead to coherence. The second
term depends on the spin and it cannot lead to coherence. It may be important only in the case of
odd nuclear targets. The curly bracket in the last equation can be brought into the form:
Eq
f1 (),
(mB (1) )2
f1 () =
1 + + 2 /2
,
2 (1 + /2)2
mq (1)
mB (1)
1.
We see that the amplitude is very sensitive to the parameter (resonance effect). For values
of not very close to zero the event rate is perhaps unobserved since the mass of the KK
boson is expected to be large. In the case of the right-handed interaction we obtain an analogous
expression:
Eq
1

f1 (),
( . ) 0 + i( ) . 5 u(x)
2
(mB (1) )2

Eq
2

0

1
MR
f1 ().
d = ig1 (4/9)d(x) ( . ) + i( ) . 5 d(x)
2
(mB (1) )2
MR
u = ig1 (16/9)u(x)
(12)
(13)
In the case of the RL interference term there is no 5 term. Furthermore the amplitude to leading
order is now independent of the energy of the quark. We thus find:

2
( . ) u(x)
MRL
u = ig1 (4/9)u(x)
f2 (),
(mB (1) )

1
2

f2 (),
MRL
d = ig1 (2/9)d(x) ( . ) d(x)
(mB (1) )
(14)
(15)
with
f2 () =
1+
.
(1 + /2)
The results in this case are less sensitive to .

The next step involves going from the quark to the nucleon level. The only question concerns
the quark energy. It seems to us that the best procedure is to replace the quark energy with their
constituent mass 1/3mp , as opposed to adopting [1,2] a procedure related to the current mass
(a)
23
(b)
Fig. 2. On the left we show the ratio (LL + RR)/(LL + RR + LR + RL) of the various chirality amplitudes in the case
of the proton. On the right we show the ratio of the amplitude of the neutron divided by that of the proton. Both are
exhibited as functions of mB (1) in GeV for the values of = 0.05, 0.10, 0.20 and 0.40 (from top to bottom). We see that
due to the LR + RL contribution the amplitude associated with the proton is dominant.
encountered in the neutralino case [1719]. In the latter case the amplitude at the quark level is
proportional to the quark mass (see the intermediate Higgs exchange below). Thus in this case,
unlike the neutralino dark matter, the process is dominated by the quarks u and d. So the obtained
results do not critically depend on the quark content of the nucleon. We thus find:

Mcoh = i4 2GF mW tan2 W ( . )N

1 1
1 mp mW
mW
11 2
f1 () +
f2 () N.
+ 3
+ 3
(16)
18 3
3 (mB (1) )2
3 3
mB (1)
In Fig. 2 we present the ratio of the amplitude arising from LL and RR contributions alone divided
by the total (LL + RR + LR + RL) in the case of the proton. We see that the second term dominates
even slightly away from the resonance condition.
In the case of the spin contribution we find at the quark level that:
1 mp mW
Mspin = i4 2GF mW tan2 W
3 (mB (1) )2

17
5
5

f1 ()qi(
) .
(17)
u
5 u + d
5 d + s 5 s .
18
18
18
In going to the nucleon level we get the isoscalar part [19]
17
5
5
u + d + s,
18
18
18
while the isovector part is:
g0 =
17
5
u d.
18
18
The quantities q are given by [1,2,19]
g1 =
u = 0.78 0.02,
d = 0.48 0.02,
We thus find,
g0 = 0.26,
g1 = 0.41.
s = 0.15 0.02.
24
In the proton neutron representation we obtain:

ap = 0.67,
an = 0.15.
The picture is different for the neutralino case in which [20],

g0 = u + d + s = 0.15,
ap = 1.41,
g1 = u d = 1.26,
an = 1.11.
Thus at the nucleon level we get
1 mp mW
Mspin = i4 2GF mW tan2 W
3 (mB (1) )2

f1 ()qi(
) . N (g0 + g1 3 )N .
(18)
2.2. Intermediate Higgs scalars

The corresponding Feynman diagram is shown in Fig. 3. The vertex involving the interaction
of the Higgs particle with the KK boson is given by
1
1
LBBh = g12 (pB ) (pB )H H g12 (pB ) (pB )hH
.
4
4
The coupling of the Higgs scalar to the quark is given by:
mq
Lqqh =
h.
H
We thus obtain

mq
21

q(x) .
Mq (h) = ig1 . q(x)
4
m2h
(19)
(20)
(21)
In going from the quark to the nucleon level we follow a procedure analogous to that of the
neutralino [1719], i.e.

fq mp ,
N|mq q q|N
we thus get

1 mp
2
2

( . )N |N
fq .
MN (h) = i4 2GF mW tan W
4 m2h
q
In this case the proton and the neutron cross sections are about equal.
Fig. 3. The interaction of KK gauge boson B (1) with quarks at tree level mediated by Higgs scalars.
(22)
25
3. KK neutrinos as dark matter candidates

The other possibility is the dark matter candidate to be a heavy KK neutrino. We distinguish
the following cases
3.1. Process mediated by Z-exchange
The Feynman diagram associated with this process is shown in Fig. 4. The qqZ vertex is
given by,
1
g
J (qqZ),
2 2 cos W

1
2
2 sin W
J (qqZ) = q
+ 3 + (1 5 )3 q,
3

J (NNZ) = N 2 sin2 W (1 + 3 ) + (1 gA 5 )3 N.
f =
(23)
(24)
(25)
Thus in the case for the proton we encounter the combination,

gA 5 + 1 4 sin2 W gA 5
while in the case of the neutron:
1 + gA 5 ,
which exhibit the well-known fact the in the case of the neutral current interaction we can have
coherence over the neutrons, but no coherence over the protons. Thus,

1
J (NNZ) N gA 5 3 (1 3 ) N.
(26)
2
In the case of the neutrino vertex we write:

1
g
J (1) .
f (1) =
2 2 cos W
(27)
Regarding the neutrino current we now have two possibilities:

the KK neutrino is a Majorana particle. In this case the neutrino current is:

J (1) = (1) 5 (1) .
(28)
The Majorana neutrino has no electromagnetic properties (no neutral vector current interaction).
Fig. 4. The interaction of KK neutrino (1) with quarks at tree level mediated by Z-exchange.
26
The KK neutrino is a Dirac particle. In the presence of left-handed interaction only we have:

J (1) = (1) (1 5 ) (1) .
(29)
The amplitude associated with the diagram of Fig. 4 becomes:
M (1) =

1
1
1
g2
J (1) J (N N Z) = GF J (1) J (N N Z).
2
2
4 4 cos W mZ
2 2
(30)
In other words in the case of the Majorana neutrino we get:

1
M (1) = GF N gA 5 3 N (1) 5 (1) ,
2 2
while in the case of a Dirac neutrino the coherent contribution dominates, i.e.:
1 3
1
N (1) (1 5 ) (1) .
M (1) = GF N
2
2 2
(31)
(32)
3.2. Process mediated by right-handed currents via Z -boson exchange

The process is similar to that exhibited by Fig. 4, except that instead of Z we encounter Z ,
which is much heavier. We will assume that the couplings of the Z are similar to those of Z.
Then the above results apply except that now the amplitudes are retarded by the multiplicative
factor = m2Z /m2Z .
3.3. Process mediated by Higgs exchange
In this case in Fig. 4, Z is replaced by the Higgs particle. In this case the amplitude at the
quark level becomes:
mq m (1) (1) (1)

M (1) (h) = 2 2GF
(33)
qq.
m2h
Proceeding as above we find that the amplitude at the nucleon level is:

mp m (1) (1) (1)

M (1) (h) = 2 2GF
N
|N
fq .
m2h
q
(34)
4. Nucleon cross sections

In evaluating the nucleon cross section one proceeds as in the case of the neutralino. The
momentum transfer to the nucleon is q = 2r , where r = reduced mass mp , is the
dark matter candidate velocity and is the cosine of the angle between the initial dark matter
particle and the outgoing nucleus.
4.1. The KK boson case
To obtain the nucleon cross section, one must sum over the final spin and boson polarizations
and average over the initial ones. One finds,
N (coh) =

m2p 1 1
1
Mcoh + MN (h)2 ,
4 (mB (1) )2 2 3
pol,ms
(35)
N (spin) =
m2p 1 1
1
|Mspin |2 ,
4 (mB (1) )2 2 3
27
(36)
pol,ms
where,
11
= 1,
23
pol,ms
11
= 6,
23
pol,ms
for the spin independent and spin dependent parts respectively.

Unlike the neutralino case, where one has to live with an allowed SUSY parameter space
involving 4 parameters, see e.g. Ellis et al. [21], Bottino et al., and Arnowitt and Nath [22], the
situation here is much simpler. One encounters only three mass parameters, , mB (1) , mh , in
the case of the coherent process and only the first two parameters in the case of the spin cross
section. Admittedly, however, the cross section depends on large powers of these masses and,
therefore, the predictions of the event rates are not very accurate. The isoscalar spin cross section
and the Higgs contribution of the coherent process depend on the structure of the nucleon. The
uncertainties encountered here are no worse than those in the neutralino cross section. In the
evaluation of the parameters fq one encounters both theoretical and experimental errors. Thus
the nucleon cross section associated with the Higgs mechanism can vary within an order of
magnitude [19]. In the present calculation we will adopt an optimistic approach and employ:
fd = 0.041,
fu = 0.028,
fs = 0.400,
fc = 0.051,
fb = 0.055,
ft = 0.095.
The thus obtained results for the coherent process are shown in Fig. 5. The independent variable
in our plots is the mass of the dark matter candidate, since this has become standard in analyzing
the experimental searches. The cross sections associated with intermediate Higgs scalars only
are plotted in Fig. 13 below. In the case of the spin cross section we obtain the 3-dimensional
plots shown in Fig. 6. Since the cross sections, especially the spin cross sections, are sensitive
functions of their arguments, we will present the above results as a series of one dimensional
plots. This is done in Fig. 7 for the coherent mode and in Fig. 8 for that of the spin.
4.2. The KK neutrino case
In this case the expression for the cross section is quite simple. We will consider each case
separately.
4.2.1. Intermediate Z boson
In this case we have two possibilities:
Majorana neutrino. Now only the axial current contributes. The proton and the neutron cross
sections are equal and given by:
1 G2F 2 2
(37)
m 3g = 8.0 103 pb.
8 p A
Dirac neutrino. In this case we have again a contribution due to the axial current, but the
resulting nucleon cross section is twice as large compared to the previous case, i.e.:
N (spin) =
N (spin) =
1 G2F 2
2
= 1.6 102 pb.
m 3 2gA
8 p
(38)
28
Fig. 5. The coherent proton cross section on the left and that for the neutron on the right in units of 106 pb, as a function
of the gauge boson mass in the range of 6001200 GeV and the Higgs mass in the range of 100200 GeV. From top to
bottom = 0.05, 0.10 and 0.15, respectively.
29
Fig. 6. The spin proton cross section on the left and that for the neutron on the right in units of 106 pb, as a function of
the gauge boson mass in the range of 6001200 GeV and in the range 0.050.15.
In the case of the neutron we have, however, an additional spin independent contribution
given by:
n (coh) =
1 G2F 2
m 2 = 3.5 103 pb.
8 p
(39)
It is quite straightforward to compute the nuclear cross sections:

nuclear (spin) =
2r
N (spin)spin F11 (q).
m2p
(40)
Here F11 is the spin response function [19,23,24], which depends on the energy Q transfered to
the nucleus, u = mA Qb2 , with b the nuclear harmonic oscillator size parameter, and spin is the
nuclear static spin ME given by:

2
n
1
1
spin = p n
(41)
= [p n ]2
3
p
3
(p = n = N ). Here p and n are the nuclear spin ME associated with the proton and
neutron component, respectively.
The coherent cross section becomes
nuclear (coh) =
2
2r
n (coh)N 2 F (q) ,
2
mp
(42)
where N is the neutron number and F (q) the nuclear form factor.
4.2.2. The right-handed interaction
In this case the nucleon cross section is retarded compared to the previous case. The obtained
results are shown in Fig. 9.
4.2.3. The intermediate Higgs scalar
Naively one expects this process to be suppressed due to the small mass of the u and d quarks
[25], present in the nucleon. This is true in the naive quark model for the nucleon. We have
30
Fig. 7. The spin independent proton cross section on the left and that for the neutron on the right in units of 106 pb, as
a function of the gauge boson mass in the range of 6001200 GeV. From top to bottom = 0.05, 0.10, 0.20, 0.40 and
0.80. On each plot we show results for mh = 100, 125, 150, 175 and 200 GeV with mass increasing downwards. Note
that in the case of the proton, due to the RL + LR dominance, the effect of Higgs contribution is not visible.
31
Fig. 7. (continued)
Fig. 8. The spin proton cross section on the left and that for the neutron on the right in units of 106 pb, as a function of
the gauge boson mass in the range of 6001200 GeV. From top to bottom = 0.05, 0.10 and 0.20, respectively.
32
Fig. 9. The coherent nucleon cross section in the case of right-handed neutrino interaction for a Dirac (1) as a function
of the gauge boson mass responsible for this interaction.
Fig. 10. On the left we show the coherent nucleon cross section as a function of m (1) and mh in GeV. On the right we
show the same thing as a function of the mass of (1) for the indicated Higgs mass (from top to bottom 100, 125, 150, 175
and 200 GeV). We see that this mechanism excludes a heavy neutrino as a WIMP candidate, unless the Higgs mass is
quite large.
seen above and we know from the neutralino case that the heavy quarks contribute and in fact
dominate. One finds:

2
m2 (m (1) )2 2
8
2 2 p
mp
fq
GF mp
N (coh) =
m4h
q
= 1.1 10
pb

m2p (m (1) )2
m4h
2
fq
(43)

The value q fq = 0.67 is acceptable. Using this value we obtain the results shown in Fig. 10.
We see that this mechanism excludes a heavy neutrino as a WIMP candidate, unless the Higgs
mass is much larger. In the Standard Model this is possible and mh can be treated as a parameter
to be extracted from the data. In SUSY models, however, the lightest neutrino is expected to be
quite light, mh 120 GeV.
33
5. Other non-SUSY models

There exist extensions of the Standard Model not motivated by symmetry, which may have
a particle content similar to that of KK theories discussed above. Such models, however, have a
much lower predictive power than the KK scenarios discussed above. To see this we examine the
following cases:
Models which introduce extra Higgs particles and impose a discrete symmetry which leads
to a parity la R-parity or KK parity [26].
Extensions of the Standard Model, which do not require the ad hoc introduction of a parity, but introduce high weak isospin multiplets [27] with Y = 0. So the WIMP-nucleus
interaction via Z-exchange at tree level is absent and the dominant contribution to the
WIMP-nucleus scattering occurs at the one loop level.
We will consider here the first of the above possibilities [26]. All particles of the Standard Model
have parity +1, while the new exotic particles have parity 1. The Standard Model particles of
interest to us here, leptons and Higgs, are:
(i , li ) (2, 1/2),
+ 0
, (2, 1/2),
lic (1, 1),

Ni (1, 0),
+ 0
, (2, 1/2)
(44)
(45)
in the usual notation SU(2)L U (1)Y quantum numbers. Consider now the following minimal
extension of the SM with symmetry SU(2)L U (1)Y Z2 and particle content:
(i , li ) (2, 1/2; +),
+ 0
, (2, 1/2; +),
lic (1, 1; +),

Ni (1, 0; ),
+ 0
, (2, 1/2; ).
(46)
(47)
Note that the particles Ni and the scalar doublet (+ , 0 ) are odd under Z2 . This makes the
lightest exotic particle a viable dark matter candidate. In other words in this economic scenario
one introduces:
A new doublet of Higgs scalars , which have the same quantum numbers with the ordinary
Higgs, but parity 1. These are expected to be quite massive, but they do not develop a
vacuum expectation value.
Assign parity 1 to the usual isosinglet right-handed neutrinos.
In this scenario the see-saw mechanism for neutrino mass generation is not operative. One cannot
have a Dirac mass term iL 0 Nj R iL 0
Nj R since the parity forbids it. One can have
Majorana mass terms at the one loop level as shown in Fig. 11, involving two scalars. The two
scalars couple with the ordinary Higgs scalars with a quartic coupling . The net result is that
the isosinglet neutrino can be much lighter than that of the standard see-saw mechanism. We now
have two possibilities
The lightest of the heavy neutrinos is the dark matter candidate.
In this case the obtained results are the same as those discussed in the previous sections in
connection with the right-handed interaction (see Fig. 9).
The neutral component of the exotic Higgs scalars is the dark matter candidate. In this case,
since we have not introduced exotic quarks, the interaction of the dark matter candidate with
34
Fig. 11. The 1-loop diagram responsible for the Majorana neutrino masses. Note that, since the isosinglet neutrino has
negative Z2 parity, there is no Dirac mass. An adjustable quartic coupling is understood.
Fig. 12. A mechanism which may contribute to the direct detection for a scalar WIMP 0 , which is stable due to the
2 /4), with treated
fact that it has Z2 parity 1. This is similar with Fig. 3, except that an effective coupling eff = (g
1
phenomenologically, is understood.
the quarks is achieved only via the ordinary Higgs (see Fig. 12). The corresponding effective
2 /4) ( = 1 corresponds to the KK case).
h coupling is now parameterized as eff = (g
1
Applying the formalism of the previous section we obtained results like those shown in
Fig. 13.
As in the case of KK theories, only in the next generation of experiments such WIMPs can be
detected.
Before concluding this section we should mention another interesting extension of the Standard Model in the direction of technicolour [28]. In this case the WIMP is the neutral LTP
(lightest neutral technibaryon). This is a scalar particle, which couples to the quarks via derivative coupling through Z-exchange. This model, however, in its present form, leads to too large
nucleon cross sections and is excluded by the data.
6. Event rates
The event rate for the coherent WIMP-nucleus elastic scattering is given by [19,29]:

S
spin
(0) m 2
R=
v fcoh A, r (A) p,
0 + fspin A, r (A) p, 0 spin ,
m 0 mp
(48)
with
2

100 GeV r (A) 2 gcoh
(A, Z)
fcoh A, Z, r (A) =
tcoh (1 + hcoh cos ),
m 0
r (p)
A
and gcoh (A, Z) = A, A Z, Z for total, neutron and proton coherence, respectively.
(49)
35
Fig. 13. The nucleon cross section corresponding to the scalar being the dark matter candidate, as a function of the
mass. On each graph we present results for mh = 100, 125, 150, 175 and 200 GeV (the mass increases downwards).
From top to bottom and left to right = 1, 0.5, 0.25 and 0.1, respectively. Note that the special case = 1 it coincides
with the Higgs contribution discussed above in connection with the KK theories.

100 GeV r (A) 2 tspin
fspin A, r (A) =
(1 + hspin cos ),
m 0
r (p)
A
(50)
spin
S
with p,
0 and p, 0 the scalar and spin proton cross sections spin the nuclear spin ME.
In this work we will ignore the motion of the earth, i.e. hcoh = hspin = 0 no modulation.
6.1. The coherent contribution due to the scalar interaction

The number of events in time t due to the scalar interaction, which leads to coherence, is:

S
p,

0
m
v 2
(0)
3 t
fcoh A, Z, r (A) . (51)
R 1.60 10
3
1
6
1 yr 0.3 GeV cm 1 kg 280 km s 10 pb
The parameter t depends on the structure of the nucleus, the WIMP velocity distribution, the
WIMP mass and the energy cutoff imposed by the detector. In the case of 127 I this parameter is
shown in Fig. 14. The nucleon cross section depends on the particle model. We will consider the
following cases:
The WIMP is a KK boson.
In this case we will consider the viable possibility = 0.8 (see Fig. 7). Then one obtains the
event rates shown in Fig. 15. We see that, even further from the degeneracy and quite heavy
WIMPs, m 1 TeV, the event rates are detectable.
36
Fig. 14. The parameter t in the case of 127 I as a function of the WIMP mass in GeV for zero threshold (continuous curve)
and a threshold of 10 keV (dotted curve). For higher WIMP masses t remains approximately constant.
(a)
(b)
Fig. 15. The coherent event rate R per year per Kg of target in the case of 127 I, in the case of the KK gauge boson as
WIMP, plotted a as function of the WIMP mass in GeV for zero threshold (continuous curve) and a threshold of 10 keV
(dotted curve). Both figures show the same quantity except the WIMP masses range is different.
The WIMP is a KK Majorana neutrino.

The Dirac KK neutrino case is excluded, since, then, the Z-induced neutron coherent
contribution would be too large. In the case of Majorana neutrinos one can have coherence due to the amplitude obtained via the Higgs exchange. The obtained results are
shown in Fig. 16. From this figure we see that the lighter Higgs mass is allowed by
the data only for relatively light WIMPs. The heavier Higgs mass, however, is allowed
for all WIMPs. Clearly such heavy Higgs cannot occur in SUSY theories, since then,
mh 120 GeV.
6.2. The spin interaction
In this case, the event in time t rate can be cast into the form:

S
p,

0
t
m
v 2
(0)
f
R 1.60
A,
Z,
(A)
spin .
spin
r
1 yr 0.3 GeV cm3 1 kg 280 km s1 103 pb
(52)
37
Fig. 16. The same as in Fig. 15 when the WIMP is a KK Majorana neutrino (1) . The coherent process is mediated by
Higgs exchange, with mh = 300 GeV on the top and 500 GeV at the bottom. The lighter Higgs mass is allowed by the
data only for relatively light WIMPs. The heavier Higgs mass is allowed for all WIMPs. Clearly such heavy Higgs cannot
occur in SUSY theories (mh 120 GeV).
Note that there is a different normalization, since due to the lack of coherence, the nucleon spin
cross section must be larger to yield detectable results. In the above expression

2
1
an
p + n ,
spin =
3
ap
with ap , an the proton and neutron spin amplitudes normalized so that [19] p = |ap2 | and
n = |an2 |. p , n are the nuclear spin matrix elements arising from the protons and neutrons respectively, normalized so that = 1 for a single proton. The case of interest to us is
when the WIMP is a KK Majorana neutrino. In this case we have found that ap = an and
p = n = 8.0 103 pb. We will examine the following cases:
The target 19 F (p = 1.646, n = 0.30).
This light target is favored from the spin ME point of view [19,23], but for heavy WIMPs
is disfavored due to the small reduced mass. The parameter tspin is shown in Fig. 17. The
obtained rates are shown in Fig. 18.
The target 73 Ge (p = 0.036, n = 1.040).
This medium mass target, favored for the coherent process as well, is characterized by large
spin [19,30,31] induced rates. The parameter tspin is shown in Fig. 19. The obtained rates are
shown in Fig. 20.
The target 127 I (p = 1.460, n = 0.355).
This medium mass target, with which the DAMA experiment [32,33] claimed to have observed a signal, is favored for the spin contribution as well due to the large reduced mass,
38
Fig. 17. The parameter tspin in the case of 19 F as a function of the WIMP mass in GeV for zero threshold (continuous
curve) and a threshold of 10 keV (dotted curve). For higher WIMP masses tspin remains approximately constant.
(a)
(b)
Fig. 18. The spin event rate R per year per Kg of target in the case of 19 F, in the cases the WIMP is a KK Majorana
neutrino, plotted as a function of the WIMP mass in GeV for zero threshold (continuous curve) and a threshold of 10 keV
(dotted curve). Both figures show the same quantity except the WIMP masses range is different.
Fig. 19. The parameter tspin in the case of 73 Ge. Otherwise the notation is the same as in Fig. 17.
(a)
39
(b)
Fig. 20. The same as in Fig. 18 in the case of the 73 Ge target.
Fig. 21. The parameter tspin in the case of 127 I. Otherwise the notation is the same as in Fig. 17.
(a)
(b)
Fig. 22. The same as in Fig. 18 in the case of the 127 I target.
even though the spin ME is modest [30,31]. The parameter tspin is shown in Fig. 21. The
obtained rates are shown in Fig. 22. We see that a KK Majorana neutrino is not excluded by
the data.
40
7. Discussion
Even though the neutralino is the preferred WIMP candidate, in this work we concentrated
on non-SUSY WIMPs. Extensions of the Standard Model, not motivated by some symmetry,
involve many parameters and do not have much predictive power. So we will concentrate on KK
WIMPs, whereby the couplings involved are those of the Standard Model. Thus essentially one
encounters only one unknown parameter, namely the WIMP mass.
From the results of the previous section, in connection with the KK WIMPs as dark matter
candidates, one can conclude the following:
The KK neutrinos as CDM candidate.
In this case everything is under control, except of course, the fact that we do not know for
sure whether the KK neutrinos are Majorana or Dirac particles. Most authors expect them to
be Dirac neutrinos (see, e.g. Servant [9]). In the case of Higgs contribution the nucleon cross
section is the same both for Dirac and Majorana neutrinos. It is proportional to the [m (1) ]2
and excludes the KK neutrino as a viable WIMP candidate, unless the lightest Higgs is
very heavy. In all other cases the WIMP mass enters via r , both explicitly and implicitly
through the nuclear form factor. Anyway, since, essentially from cosmological requirements
[1,2], the KK neutrino mass is expected to be in the TeV region, r Amp . So the cross
section is independent of the WIMP mass. The number of WIMPs in our vicinity, for a given
density, is inversely proportional to the WIMP mass. Thus the rate will scale as follows:
R(mWIMP ) = R(A)
A GeV
.
mWIMP
(53)
Large nucleon spin cross sections ( 102 ) pb are possible via Z-exchange. Thus Dirac
neutrinos are excluded due to neutron coherence. No such coherence exists for Majorana
neutrinos, so these cannot be excluded from the data (16 events per year per kg of target
[34]). The precise predicted rates depend on nuclear physics assumptions.
The KK boson as CDM candidate.
For heavy WIMPs Eq. (53) holds in this case as well allowing larger nucleon cross-sections
to be consistent with the data.
1. The unknown parameters of the theory are the masses of KK quarks and gauge bosons as
well as the mass of the neutral Higgs. All relevant couplings are under control.
2. In the spin independent mechanism the proton cross section is dominant. This is due to
the RL + LR currents. This prediction can be consistent with the present data only away
from the resonance and/or large KK gauge boson masses. We should also take note of the
fact that the event rate will be down by a factor Z 2 /A2 compared to the analysis of the
neutralino case. This contribution (LR + RL) is absent in the spin mode. It is also absent
in the case of the neutron cross section.
3. Even in the other cases the proton cross sections are larger than those for the neutrons.
4. The results fall quite fast with increasing boson mass.
5. The obtained results, in particular those associated with the spin, are very sensitive functions of the mass difference between the KK quarks and the KK bosons.
6. For sufficiently small , the process involving the KK quarks is more important than the
Higgs induced cross section. Away from the resonance the Higgs contribution becomes
significant.
41
Acknowledgements
One of the authors (J.D.V.) is indebted to Ignatios Antoniadis and Geraldine Servant for discussions during his visit at CERN. His work and this visit were supported by European Union
under the contract MRTN-CT-2004-503369 as well as the program PYTHAGORAS-1. The latter
is part of the Operational Program for Education and Initial Vocational Training of the Hellenic
Ministry of Education under the 3rd Community Support Framework and the European Social
Fund. The author V.K. Oikonomou acknowledge support by the co-funded European UnionEuropean Social fund and National fund PYTHAGORASEEAEK II. The work of Ch.C.M.
was supported by the PYTHAGORAS II Research project (80861) of EEAEK and the European Union.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
G. Servant, T.M.P. Tait, Nucl. Phys. B 650 (2003) 391.

G. Servant, T.M.P. Tait, New J. Phys. 4 (2002) 99.
S.D.I. Antoniadis, N. Arkani-Hamed, Phys. Lett. B 436 (1998) 267, hep-ph/9804398.
G.R.D.N. Arkani-Hamed, S. Dimopoulos, Phys. Rev. D 59 (1999) 086004, hep-ph/9807344.
T.G.K.R. Dienes, E. Dudas, Nucl. Phys. B 47 (1999) 537, hep-ph/9806292.
B.A.D.T. Appelquist, H.C. Cheng, Phys. Rev. D 64 (2001) 035002, hep-ph/0012100.
M.Q.I. Antoniadis, S. Dimopoulos, Nucl. Phys. B 544 (1999) 503, hep-ph/9810410.
M.Q.I. Antoniadis, K. Benakli, Phys. Lett. B 331 (1994) 313, hep-ph/9403290.
See, e.g., G. Servant, in: B.C. Allanach, C. Grojean, P. Skands, et al. (Eds.), Les Houches: Physics at TeV Colliders
2005, Beyond the Standard Model Working Group: Summary Report, Section 25, p. 164, hep-ph/0602198.
S. Hanary, et al., Astrophys. J. 545 (2000) L5;
J.H.P. Wu, et al., Phys. Rev. Lett. 87 (2001) 251303;
M.G. Santos, et al., Phys. Rev. Lett. 88 (2002) 241302.
P.D. Mauskopf, et al., Astrophys. J. 536 (2002) L59;
S. Mosi, et al., Prog. Nucl. Part. Phys. 48 (2002) 243;
S.B. Ruhl, et al., astro-ph/0212229, and references therein.
N.W. Halverson, et al., Astrophys. J. 568 (2002) 38;
L.S. Sievers, et al., astro-ph/0205287, and references therein.
G.F. Smoot, et al., COBE Collaboration, Astrophys. J. 396 (1992) L1.
D.N. Spergel, et al., Astrophys. J. Suppl. 148 (2003) 175.
M. Tegmark, et al., Phys. Rev. D 69 (2004) 103501.
D.N. Spergel, et al., Three-year WMAP results: Implications for cosmology, astro-ph/0603449;
L. Page, et al., Three-year WMAP results: Polarization analysis, astro-ph/0603450;
G. Hinsaw, et al., Three-year WMAP observations: Implications temperature analysis, astro-ph/0603451;
N. Jarosik, et al., Three-year WMAP observations: Beam profiles, data processing, radiometer characterization and
systematic error limits, astro-ph/0603452.
A. Djouadi, M.K. Drees, Phys. Lett. B 484 (2000) 183;
S. Dawson, Nucl. Phys. B 359 (1991) 283;
M. Spira, et al., Nucl. Phys. B 453 (1995) 17.
T.P. Cheng, Phys. Rev. D 38 (1988) 2869;
H.-Y. Cheng, Phys. Lett. B 219 (1989) 347.
See, e.g., our recent review: J.D. Vergados, On the direct detection of dark matterExploring all the signatures of
the neutralinonucleus interaction, hep-ph/0601064, and references therein.
J. Ellis, M. Karliner, The strange spin of the nucleon, hep-ph/9501280.
J. Ellis, K.A. Olive, Y. Santoso, V.C. Spanos, Phys. Rev. D 70 (2004) 055005.
A. Bottino, et al., Phys. Lett. B 402 (1997) 113;
R. Arnowitt, P. Nath, Phys. Rev. Lett. 74 (1995) 4592;
R. Arnowitt, P. Nath, Phys. Rev. D 54 (1996) 2374, hep-ph/9902237;
V.A. Bednyakov, H.V. Klapdor-Kleingrothaus, S.G. Kovalenko, Phys. Lett. B 329 (1994) 5.
P.C. Divari, T.S. Kosmas, J.D. Vergados, L.D. Skouras, Phys. Rev. C 61 (2000) 054612.
42
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
J.D. Vergados, J. Phys. G 30 (2004) 1127, hep-ph/0406134.

G.S.G. Agashe, JCAP 0502 (2005) 002, hep-ph/0411254.
E. Ma, Mod. Phys. Lett. A 21 (2006) 1777, hep-ph/0605180.
M. Cirelli, N. Forengo, A. Sturmia, Minimal dark matter, Nucl. Phys. B 753 (2006) 178, hep-ph/0512090.
K.T.F. Sannino, Phys. Rev. D 71 (2005) 2005;
S.V. Gudnason, C. Kouvaris, F. Sannino, Dark matter from new technicolour theories, hep-ph/0608055.
J.D. Vergados, Phys. Rev. D 57 (2003) 103003, hep-ph/0303231.
M.T. Ressell, et al., Phys. Rev. D 48 (1993) 5519;
M.T. Ressell, D.J. Dean, Phys. Rev. C 56 (1997) 535.
E. Homlund, M. Kortelainen, T.S. Kosmas, J. Suhonen, J. Toivanen, Phys. Lett. B 584 (2004) 31;
E. Homlund, M. Kortelainen, T.S. Kosmas, J. Suhonen, J. Toivanen, Phys. At. Nucl. 67 (2004) 1198.
R. Bernabei, et al., Phys. Lett. B 389 (1996) 757.
R. Bernabei, et al., Phys. Lett. B 424 (1998) 195.
D.S. Akerib, et al., CDMS Collaboration, Phys. Rev. Lett. 93 (2004) 211301.
Resolving the AbFB puzzle in an extra dimensional

model with an extended gauge structure
Abdelhak Djouadi a , Grgory Moreau a, , Franois Richard b
a Laboratoire de Physique Thorique, CNRS and Universit Paris-Sud, Bt. 210, F-91405 Orsay Cedex, France
b Laboratoire de lAcclrateur Linaire, IN2P2-CNRS et Universit de Paris-Sud, Bt. 200, BP 34,
F-91898 Orsay Cedex, France

Received 6 November 2006; received in revised form 12 January 2007; accepted 7 March 2007
Abstract
It is notorious that, contrary to all other precision electroweak data, the forwardbackward asymmetry
for b quarks AbFB measured in Z decays at LEP1 is nearly three standard deviations away from the predicted value in the Standard Model; significant deviations also occur in measurements of the asymmetry off
the Z pole. We show that these discrepancies can be resolved in a variant of the RandallSundrum extradimensional model in which the gauge structure is extended to SU(2)L SU(2)R U(1)X to allow for
relatively light KaluzaKlein excitations of the gauge bosons. In this scenario, the fermions are localized
differently along the extra dimension, in order to generate the fermion mass hierarchies, so that the electroweak interactions for the heavy third generation fermions are naturally different from the light fermion
ones. We show that the mixing between the Z boson with the KaluzaKlein excitations allows to explain
the AbFB anomaly without affecting (and even improving) the agreement of the other precision observables,
including the Z bb partial decay width, with experimental data. Some implications of this scenario for
the ILC are summarized.
1. Introduction
The Standard Model (SM) of the strong and electroweak interactions of elementary particles
[1] has brilliantly passed almost all the experimental tests to date. These tests, performed at the
E-mail address: moreau@th.u-psud.fr (G. Moreau).

doi:10.1016/j.nuclphysb.2007.03.019
44
A. Djouadi et al. / Nuclear Physics B 773 (2007) 4364
per-mille level accuracy, have probed the quantum corrections and the structure of the SU(3)C
SU(2)L U(1)Y local symmetry [2,3]. However, it is notorious that there is an observable for
which the measured value significantly differs from the one predicted in the context of the SM.
The forwardbackward (FB) asymmetry for b quark jets AbFB [4] measured in Z boson decays
at LEP1 provides, together with the longitudinal asymmetry ALR measured at the SLC, the
lep
most precise individual measurement of the electroweak mixing angle sin2 eff , but the result
is 2.8 standard deviations away from the predicted value and the two individual measurements
differ by more than three standard deviations. In fact, the AbFB anomaly is present not only in
the data collected at the Z boson pole, but also a few GeV above and, to a lesser extent, a few
GeV below the Z pole. [There are also measurements of AbFB at energies below (from PEP to
TRISTAN) [57] and far above (LEP2) [8] the Z-pole which are in a reasonable agreement with
the SM expectations.]
This situation has led to speculations about a possible signal of new physics beyond the SM
in the Zbb vertex.1 However, it turns out that this discrepancy cannot be easily explained without affecting the Z bb partial decay width, which has also been very precisely measured at
LEP1 and found to be compatible with the SM expectation, and the FB longitudinal asymmetry
AbFB,LR measured at the SLC, although the corresponding experimental error is larger [3]. The
basic reason for this difficulty lies in the fact that a very large correction, O(30%), is needed
to alter the initially small right-handed component of the b-quark coupling to Z bosons without affecting significantly the left-handed b-quark coupling as well as the couplings of all other
light fermions. Note that this large effect can possibly not be generated through radiative corrections in reasonable models [this is, for instance, the case in supersymmetric extensions of the
SM where loop contributions of relatively light partners of the top quark and the W boson [10]
cannot account for the discrepancy] and one has to resort to large tree-level effects to explain the
anomaly.
Several new physics models fulfilling theses conditions have been proposed in the literature,
though. Three examples of such models are variations of leftright symmetric models in which
a new gauge boson Z occurs and has interactions only with third generation fermions [11],
models involving additional exotic (mirror) bottom-like quarks which strongly mix with the bquarks [7,12] and models where the electroweak symmetry breaking is induced by a new strongly
interacting sector coupled to the SM fields [13]. In most of the models listed above, a kind of
non-universality in which the third generation fermions are treated differently from the ones of
the first and second generations is assumed. In most cases, this assumption is made in an ad hoc
way and specifically to cure the AbFB anomaly.
Recently, it has been shown [14] that in different scenarios beyond the SM, such as technicolor scenarios, little Higgs theories, Higgsless models and models where the composite Higgs
boson arises as a pseudo-Goldstone boson [which are related to theories in five-dimensional antide Sitter space via the AdS/CFT correspondence], one can invoke a custodial O(3) symmetry
which protects at the same time the parameter and the ZbL bL coupling. Such an approach
might also account for the required shift of the ZbR bR coupling to explain the AbFB anomaly,
if one allows for a deviation of the Z bb partial decay width; this point will be discussed
later. The implications of this O(3) symmetry have been subsequently discussed within Higgsless models [15] and gauge-Higgs unification scenarios [16].
1 Of course, the possibility that this discrepancy could be simply due to some unknown experimental problem or
inaccuracy or to a large statistical fluctuation cannot be excluded [9].
45
In this paper, we point out that the discrepancy between the measured value of AbFB and the
theoretical prediction can be resolved in the context of variants of the RandallSundrum (RS)
extra-dimensional model [17] in which the fermion and bosonic fields are propagating in the
bulk. These models, besides the fact that they explain the large hierarchy between the Planck and
TeV scales without introducing any fundamentally new energy scale in the theory, allow for the
unification of the gauge coupling constants at a high energy Grand Unification scale [18], provide
a solution for proton decay and have a viable candidate of KaluzaKlein type for the dark matter
of the universe [19]. They also have the additional attractive feature of providing a geometrical
explanation of the large mass hierarchies prevailing among SM fermions [2022]. Indeed, when
the SM fermions are localized differently along the extra dimension depending on their nature,
their different wave functions which overlap with the Higgs boson [which is still localized on the
TeV-brane for its mass to be protected] generate large hierarchical patterns among their effective
four-dimensional Yukawa couplings. One can then naturally obtain electroweak interactions for
the heavy third generation fermions that are different from the ones of the light fermions.
More specifically, we will work in a variant of the RS model proposed in Ref. [23] where
the electroweak symmetry is enhanced to the leftright gauge symmetry SU(2)L SU(2)R
U(1)X in which the bulk fields are embedded, the right-handed fermions being promoted to
SU(2)R isodoublets. With this symmetry, the high-precision electroweak measurements can be
nicely fitted while keeping the masses of the first KaluzaKlein (KK) excitation modes of the
gauge bosons rather low, MKK of order of a few TeV, which is required if the gauge hierarchy
problem is to be addressed within the RS model with bulk matter. The ordinary Z boson will
then mix with its KK excitations and those of the new Z boson, with the possibility of choosing
their effective couplings in such a way that only the overall Z couplings to third generation
fermions are significantly altered. This would provide an explanation for the AbFB anomaly, while
keeping the electroweak observables involving light fermions as well as the Z bb decay width,
unaltered for MKK = O (TeV). In this context, we propose two scenarios: one in which the U(1)
group is U(1)B L (as originally discussed in Ref. [23]) and a second in which the right-handed
bR
b-quark has isospin I3R
= + 12 under the SU(2)R group. While the AbFB anomaly on the Z pole is
resolved in both scenarios, the experimental data for the asymmetry off the Z pole are reproduced
only in the second scenario.
The paper is organized as follows. In the next section, we briefly describe the extradimensional model of Ref. [23] based on the SU(2)L SU(2)R U(1)X symmetry and summarize the main features which allow for relatively light KK states; the role of the fermions and their
couplings to the Z boson are then summarized. In Section 3, we discuss two scenarios in which
the choice of Z-fermion couplings allows to explain the AbFB anomaly at energies on the Z pole
and reproduce the value of the Z bb partial decay width, without affecting the other precision
measurements. In Section 4, we analyze the asymmetry off the Z pole and show that the agreement between theory and experiment is satisfactory only in the scenario where the right-handed
bR
b-quark has isospin I3R
= + 12 under the SU(2)R group. In Section 5, we discuss the implications
of this scenario at the ILC. A brief conclusion is presented in Section 6.
2. The physical set-up
2.1. Theoretical framework
We consider the higher-dimensional RandallSundrum scenario in which the Standard Model
fields are propagating in the bulk like gravity, except for the Higgs boson which remains con-
46
fined on the TeV-brane in order not to re-introduce a gauge hierarchy. On the Planck-brane, the
gravity scale is equal to the reduced Planck mass MP = 2.44 1018 GeV, whereas the effective gravity scale on the TeV-brane, M = wMP , is suppressed by the exponential warp factor
w = exp(kRc ) where 1/k is the curvature radius of the anti-de Sitter space and Rc the compactification radius. For a small extra dimension, Rc 11/k with k being close to MP , one finds
w 1015 so that M = O(1) TeV, thus addressing the gauge hierarchy problem. For this Rc
value, the five-dimensional gravity scale M5 is close to the effective four-dimensional gravity
scale MP .
The mass of the first KaluzaKlein excitations of the SM gauge bosons, MKK = M (1) =
Mg (1) MZ (1) MW (1) , reads MKK = 2.45kw = 2.45kM /MP M = O (TeV) and, since
this parameter is more physical, we adopt it as our free input parameter instead of k. The maximal value of MKK is fixed by kRc and the theoretical consistency bound on the five-dimensional
curvature scalar by |R5 | = |20k 2 | < M52 leading to k < 0.105MP ; one can thus take a maximal
value of MKK 10 TeV which corresponds to the value kRc = 10.11. In fact, there exist a severe
indirect bound of typically MKK 10 TeV originating from electroweak precision data [24]. In
order to soften this bound down to a few TeV and, thus, to allow for a solution of the gauge hierarchy problem, several scenarios have been proposed. For instance, scenarios with brane-localized
kinetic terms for fermions and gauge bosons [25] allow to lower the previous bound down to
a few TeV [26]. Another possibility, that we retain here, is the model of Agashe et al. [23] in
which the electroweak gauge symmetry is enhanced to a leftright SU(2)L SU(2)R U(1)X
structure in the bulk and where right-handed fermions are promoted to SU(2)R doublet fields,
the new doublet components having no zero mode. The usual SM electroweak symmetry is recovered after the breaking of both SU(2)R and U(1)X on the Planck-brane, with possibly a small
breaking of SU(2)R in the bulk.
As mentioned in the introduction, the RS model with bulk matter provides naturally a geometrical interpretation of quark/lepton mass hierarchies [20]. The idea is to localize the SM
fermions differently along the extra dimension, depending on their nature. Then, the overlapping of their different wave functions with the Higgs boson generates large hierarchical patterns
among the effective four-dimensional Yukawa coupling constants. In this mass model, each
the interaction basis]
five-dimensional fermion field i [i = {1, 2, 3} is the family index
in
couples to a distinct mass mi in the fundamental theory as d 4 x dy Gmi i i where G
is the determinant of the RS metric and y parameterizes the fifth dimension. The localization
of fermions along the extra dimension is fixed by the dependence of mass mi on y. One can
take mi = sign(y)ci k [27] where ci are dimensionless parameters. Then the fields decompose

(n) i
as i (x , y) =
n=0 i (x )fn (y), with n labeling the tower of KK excitations, admitting
as a solution for the zero mode wave function f0i (y) = exp[(2 ci )k|y|]/N0i , where N0i is a
normalization factor. The Yukawa interactions with the Higgs boson H read then

(5)

(0) (0)
SYukawa = d 5 x G Yij H +i j + h.c. = d 4 x Mij Li Rj + h.c. + ,
(1)
(5)
where Yij are the five-dimensional Yukawa coupling constants and the dots stand for KK mass
terms. The effective fermion mass matrix is obtained after the integration
j
(5)
Mij = dy GYij Hf0i (y)f0 (y).
(2)
(5)
The dimensionful Yij couplings can be chosen almost universal so that the quark/lepton mass
i,j
hierarchies are essentially generated through the overlap between f0 (y) and H along y. Re-
47
markably, large fermion mass hierarchies can be created for fundamental parameters mi all of
the order of the unique scale of the theory, k M5 MP .
2.2. The couplings to fermions
Let us now discuss the fermions couplings to the Z boson in the context of this extended
leftright symmetric model. For each fermion fL/R of left/right-handed chirality, the couplings
to the Z boson which in the SM read
f
QZL/R = I3LL/R Qf sin2 W ,

f
(3)
where I3LL = 12 and I3LR = 0 are the third components of weak isospin and Q the electric
charge of the fermion f . One has to add the contributions due to the mixing between the Z
boson and its excitations and those of the new Z boson, which is a superposition of the state
W 3 associated to the SU(2)R group and B associated to U(1)X ; the other orthogonal state is the
SM hypercharge boson B. In terms of the coupling constant gZ of this new Z and the charges
f
QZL/R
of the fermion f and the Higgs boson QH

Z , the additional contributions to the Z boson
coupling to left- and right-handed fermions reads

f
f
QH
gZ2 QZL/R
MZ2
QZL/R
1
1

Z
=
1
1
+
+
F (cfL/R ), (4)
f
f
4F (cfL/R )
kRc
(0.4MKK )2
QZL/R
gZ2 QZL/R QH
Z
where gZ is the Z coupling constant. The first term comes from the mixing between the Z
boson with its KK excitations Z (n) , whereas the second term is due to the mixing with the Z (n)
excitations; the Z boson is coupled to a Planckian vev on the ultraviolet brane, which mimics
a (, +) boundary condition to a good approximation, so that it possesses no zero mode. In
f
Eq. (4), the Z charges are given, in terms of the new mixing angle , the isospin number I3RL/R
under the SU(2)R gauge group, and the SM hypercharge Y fL/R by
f
QZL/R
= I3RL/R Y fL/R sin2 ,
(5)
where the relation between the hypercharge and the charge QX under the U(1)X group is
f
Y fL/R = QXL/R + I3RL/R = Qf I3LL/R .
(6)
In the previous equation, sin g /gZ and gZ2 = g 2 + g 2 , where g and g are, respectively,
the SU(2)R and U(1)X couplings; the coupling g of the SM U(1)Y group reads g = g g /gZ .
From these relations, one easily derive the following relations:

2g 2 /gZ2 = 1 1 (2g /gZ )2 .
2g 2 /gZ2 = 1 1 (2g /gZ )2 ,
(7)
Hence, if the charge of the U(1) group is QX = 12 (B L) with B and L being respectively
the baryon and lepton numbers, that is exactly as in Ref. [23], one has I3L = Y and I3R = Y
H
for the charges of the neutral Higgs boson so that QH
Z /QZ 1 in the approximation that we
2
will consider later: sin 1.
Finally, the function F (c) in Eq. (4) reads

1 2c 5 2c
kRc
1
.
F (c) =
(8)
2
1 ekRc (2c1) 3 2c 4(3 2c)
48
Given the exponential term in Eq. (8), F (c) goes rapidly to zero above the value c = 0.5; it
remains everywhere negative and has no singularities and, in particular, it is finite at c = 0.5
with F (0.5) 0.25. The dependence of the F function on the fermion c parameters is due to
the fact that the effective four-dimensional couplings between KK gauge bosons and zero mode
fermions depend on the fermion localization which is given by the c parameters.
The couplings of the first KK excitations of the Z and Z bosons to SM fermions have expressions that can be found in Ref. [28] for instance. The ratio Q(c) (Q (c)) of the whole effective
coupling between the Z (1) (Z (1) ) boson and fermions over the same coupling for the Z (would
be Z ) boson zero mode, takes into account the wave function overlap between the
gauge boson KK excitations and the SM fermions. The ratios Q(c) and Q (c) tend both to 2kRc for
c , while one has Q(c) 0.2 and the other Q (c) 0 in the limit c +; for the
special value c = 0.5 one has Q(0.5) = 0 and Q (0.5) 0.2.
2.3. Two scenarios for U(1)X
In the present paper, we will discuss two possible scenarios for the Abelian group U(1)X . In
a first scenario, that we will call RSa, we assume that U(1)X U(1)B L , as in Ref. [23], so that
the fermion representations/charges under the SU(2)L SU(2)R U(1)B L group are:
1
ui
with I3RR = + ;
6
6
2
1
iR
i
i
LL (2, 1) 1 ;
R (1, 2) 1 with I3R =

2
2
2
for the left-handed SU(2)L doublets and the right-handed leptons and up-type quarks, and
QiL (2, 1) 1 ;
uiR (1, 2) 1
(9)
1
(10)
2
for the right-handed down-type quarks, i = {1, 2, 3} being a generation index. Here, the Yukawa
coupling terms for the quarks are written in terms of an invariant operator as,
dRi (1, 2) 1
RSa:
(2, 1) 1 (2, 2)0 (1, 2) 1

6
di
with I3RR =
(11)
the Higgs boson being embedded in a bidoublet of SU(2)L SU(2)R .

In the second scenario, that we will denote RSb, only the isospin assignment I3R [and thus
the QX charges as imposed by Eq. (6), the SM hypercharge Y being fixed] for the down-type
right-handed quarks are modified with respect to the previous scenario:
1
di
(12)
with I3RR = + .
6
2
This right-handed down quark assignment was mentioned recently in Ref. [14] (see the table).
The authors of Ref. [14] have suggested the possibility, considered here, that the isospin up-type
quarks (u, c, t) acquire masses through a Yukawa coupling of the type Eq. (11), whereas the
down-type quarks (d, s, b) become massive via another operator. In general, this latter operator
i
will violate the custodial symmetry protecting the charges QdZ , but it is natural to assume that its
coefficient is small (in particular, in order to generate the small ratio mb /mt ) so that the resulting
i
i
QdZ /QdZ is also small.
As mentioned previously, the scenario RSa has been discussed in detail in Ref. [23]. Because of the bulk custodial isospin gauge symmetry, an acceptable fit of the oblique parameters
RSb:
dRi (1, 2) 5
49
[29] S and T [including, after a redefinition, the coefficients x and y of fermion operators for
light fermions: u, c, d, s,
, ] can be achieved for MKK 3 TeV and clight sufficiently larger
than 0.5, irrespectively of the value of the coupling constant gZ of the new Z boson, as shown
in Ref. [23] for the case U(1)X = U(1)B L , that is, QX = 12 (B L) for all fields. The heavy
bottom and top quarks must be treated separately since, in order to reproduce the large value of
the top quark mass through the geometrical mechanism developed in the beginning of this section, one should typically have cQL < 0.5 [with cQL = cbL = ctL since bL and tL belong to the
same SU(2)L multiplet] and ctR < 0.5 [21] so that the coefficients x, y of operators for the b and
t quarks cannot be redefined into the S and T parameters [23]. Note that in order to maximize
the top quark mass, one can choose ctR 0.5; a smaller value cannot be taken otherwise the
of the SU(2) doublet would have a too light KK excitation which is
down-type component bR
R
experimentally ruled out.2
In the second scenario RSb, where the right-handed down-type quarks have different I3R values from RSa, the QZ charges of the quarks dRi are different than previously. However, in the
case of light quarks, clight values larger than 0.5 can always induce a decrease of | QdZR /QdZR |
sufficient not to generate unacceptable deviations in electroweak observables [the part independent of F (clight ) gives small contributions]. This is guaranteed by the condition MKK 3 TeV,
exactly as in scenario RSa. As before the bottom and top quarks, with c < 0.5, must be treated
separately.
The theoretical elements given in this section are sufficient to discuss the effect of the RS
scenario on the high precision measurements and show how the AbFB anomaly can be resolved,
while keeping the other electroweak observables in good agreement with experimental data.
3. Resolving the AbFB (MZ ) puzzle
3.1. The FB asymmetry on the Z pole
On top of the Z boson resonance, the forwardbackward asymmetry for b quarks, AbFB (MZ ),
can be written in the SM, in terms of the Z charges to the left- and right-handed chiralities
introduced in the previous section, as
AbFB (MZ ) =
e
e
b
b
3 (QZL )2 (QZR )2 (QZL )2 (QZR )2 3
Ae Ab .
4 (QeZL )2 + (QeZR )2 (QbL )2 + (QbR )2 4
Z
(13)
As discussed earlier, it is the electroweak observable for which the deviation between the experimentally measured value and the SM expectation is the highest [3]: AbFB = 0.0992 0.0016,
while the SM fit gives AbFB = 0.1037 0.0008. This effect, nearly at the 3 level, could be attributed to the special status of the Zbb vertex but this seems to contradict the nice agreement
2 In the present analysis, we will assume that the SU(2) partners of the right handed quarks, in particular b , are too
R
R
heavy to affect the electroweak precision data. This statement has to be quantified as the mass of this new quark is related
states lighter than those assumed here, one would need
to the masses of the KK excitations of gauge bosons. For bR
(1)
(0)
to include the fermionic b b
mixing which might affect the Zbb vertex as is intensively discussed in Ref. [16].
R
Nevertheless, note that the induced contribution to (QZL )2 /(QZL )2 can be compensated by (QZR )2 /(QZR )2 so that
the resulting deviation in Rb vanishes (as will be seen later). In a complete analysis, the mixing of both fermion and
gauge boson excitations might need to be taken into account.
50
observed for the partial width of the decay Z bb which reads in the SM
Rb
(QbL )2 + (QbZR )2
(Z bb)
= Z qL 2
qR 2
(Z hadrons)
q [(QZ ) + (QZ ) ]
(q = t)
(14)
and is measured to be Rb = 0.21629 0.00066 while in the SM, one has Rb = 0.2158, meaning
that the deviation is only at the +0.7 level [3].
In fact, if one uses the fitted value of Ae deduced from leptonic data assuming lepton universality, on obtains Ab = 0.882 0.017 which is 3 away from the value Ab = 0.9347 0.0001
predicted in the SM. In addition, if in the spirit of this paper one does not assume lepton universality and uses the most precise result for Ae obtained from the measurement of the longitudinal
polarization asymmetry ALR at SLC, Ae = 0.1514 0.0022, one has Ab = 0.874 0.019 which
is 3.2 away from the SM value. This (6 2.1) effect has to be contrasted with the agreement observed on Rb at the 0.3% level. Note that the direct measurement of Ab , using the
FB asymmetry with polarized electrons at SLC, gives Ab = 0.925 0.020, consistent with the
SM but not precise enough to rule out the LEP1 value; combining both measurements gives
Ab = 0.900 0.013 hence a 3.86 1.44% effect.
Using Eqs. (13) and (14) for the observables AbFB and Rb , since the new physics alters significantly only the left- and right-handed Zbb couplings (the corrections QeZ /QeZ will turn
2 and
out to be typically weaker as we must restrict to clight larger than 0.5), QbZL = 12 + 13 sW
bR
bL 2
bR 2
1 2
1
2
2
QZ = 3 sW with sW = sin W 4 , and noticing that (QZ ) /(QZ ) 30 which allows to
safely use the approximation (QbZL )2 (QbZR )2 , the cancellation of the new effects in Rb occurs
if (QbZR )2 (QbZL )2 while the effect on the asymmetry would be in this case
AbFB
AbFB
(QZL )2
(QZR )2
Ab
2

2
.
Ab
(QbL )2
(QbL )2
Z
(15)
One then obtains the needed deviations for the squared left- and right-handed Zbb couplings to
fully explain the 3 anomaly in AbFB (MZ )
(QbZR )2
(QbZR )2
(58 22)%,
(QbZL )2
(QbZL )2
(1.92 0.72)%,
(16)
where the uncertainties are due to the statistical errors.

Thus, as stated earlier, if new physics has to explain the AbFB anomaly while keeping Rb SMlike, one needs a drastic change, at least 30%, of the right-handed coupling QbZR and only a
small change, at the percent level, of the left-handed one. Moreover, the deviations in the two
couplings squared should be opposite in sign.
3.2. The FB asymmetry in scenario RSa
In the scenario RSa in which the Abelian group is U(1)X = U(1)B L as in Ref. [23], the new
contributions to the Zbb coupling, using the charges defined earlier and assuming sin2 1
are

QbZL
MZ2
1
(17)

0.09
F (cbL ),
1
+
4F (cbL )
(0.4MKK )2
QbL
Z
QbZR
QbZR

MZ2
1.5
1

+
1+
F (cbR ).
4F (cbR ) sin2
(0.4MKK )2
51
(18)
Thus, for moderate KK masses, MKK 3 TeV, and small values of the new mixing angle sin , one can generate a large correction QbZR /QbZR , while QbZL /QbZL remains small.
One must mention here the approximated perturbative limit sin 0.15 which corresponds
(cf. Eq. (7)) to impose that the five-dimensional loop expansion parameter, estimated at energy
scale k to be (gZ5D )2 k/16 2 = gZ2 kRc /16 2 (following Ref. [31]), is smaller than unity.3
A significant hierarchy between QbZR /QbZR and QbZL /QbZL can also be generated through the
F (cbL,R ) functions. However, since the latter function is always negative, the two contributions
QbZL /QbZL and QbZR /QbZR have always the same sign (negative) and this does not allow to
cure the AbFB anomaly while leaving Rb almost unaffected, Eq. (16).
A solution to this problem, as pointed out in Ref. [7] in a different context, would be to
generate an even larger correction QbZR to flip the sign of the overall ZbR bR charge QbZR +
1
QbZR since in the SM, the charge squared (QbZR )2 30
is naturally the smallest one. The needed
deviations would be in this case,
QbZR
QbZR
(230 10)%,
QbZL
QbZL
(1 0.4)%.
(19)
In Fig. 1, we show the values in the [cbL , cbR ] parameter space for which the predicted values
of AbFB and Rb in this RS scenario are equal and within 1 to the experimental measure-
Fig. 1. Contour plots in the plane [cbL , cbR ] for MKK = 3 TeV and sin = 0.1, for which AbFB and Rb in the scenario
RSa are equal to their experimental values (dotted lines) and are within their 1 bands; clight 0.5 for leptons and
light quarks.
3 A similar typical bound is obtained from the perturbativity condition for the KK Z coupling, which receives quantum
corrections through loop-exchanges of KK Z .
Indeed, the effective four-dimensional coupling of the first Z KK mode
to SM fields is increased by a factor as large as 2 kRc (see end of Section 2.2) relatively to gZ for SM fields near the
TeV-brane, like for example the Higgs boson.
52
ments. The other inputs are MKK = 3 TeV [which is allowed by the oblique S, T parameters]
and sin = 0.1 [i.e. roughly at the limit of the regime of a strongly coupled five-dimensional theory]. As can be seen, there is a set of [cbL , cbR ] assignments for which AbFB and Rb are equal to
their experimental values. Indeed, for cbL = 0.375 and cbR = 0.296, with again c
L = c
R 0.5,
one finds for the two observables AbFB = 0.0992 and Rb = 0.21629, i.e. the exact experimental
values.
For heavier KK excitations MKK > 3 TeV and fixed sin , lower cbR and cbL values are
2 decrease of | QbL,R /QbL,R |,
required to increase |F (c)| and thus to compensate the 1/MKK
Z
Z

Eq. (4). Alternatively, for sin larger than 0.1, lower c(bR ) values which compensate the decrease of the gZ effect in QbZR /QbZR are required [the gZ coupling is annihilated by the charge
QbZL in QbZL /QbZL ].
Thus, it is remarkable that both the measurements of AbFB and Rb can be reproduced for
cbL/R values of order unity, which means that no additional scale is introduced in the model as
explained in Section 2.1.
Given the above discussion on c parameters, the favored values for MKK and sin (the kRc
influence being weak), with regard to a realistic heavy quark mass spectrum, are those chosen for
Fig. 1. For the optimal c values, cbL = 0.47 and cbR = 0.305 which reproduce the experimental
values of AbFB and Rb [as noted earlier, the value ctR = 0.5 can be chosen as to maximize mt ,
while ctL = cbL ], the approximate mt and mb values which can be obtained via the geometrical
mechanism discussed in Section 2, are such that mb 40 GeV and mt 75 GeV. Both of these
generated masses are far from the experimentally measured values. However, these differences
can easily originate from the full three-flavor mixing effects, significant contribution from the
mixing with the first KK excitations as discussed in Ref. [32] in the case of the top quark, or from
a reasonably small hierarchy among the five-dimensional Yukawa couplings. Indeed, the quark
mass values given above correspond to a theory in which
the Higgs boson is exactly confined
on the TeV-brane with the natural choice Y (5) = / k, with (relating the five-dimensional
Yukawa coupling to the parameter k as usual [20,21,32]) taken equal to unity. The actual top
and bottom masses are reproduced for t 2 and b 0.1; such values are acceptable as they
correspond to almost universal Y (5) couplings and introduce mass parameters of an order of
magnitude relatively close to the fundamental energy scale k.
Note that within the full three-flavor treatment of the quark/lepton mass matrices (including
the determination of Yukawa couplings), which is beyond the scope of this study, one could
figure out the fermion mixing angles and should thus consider the experimental bounds on FCNC
processes translating into lower bounds on MKK [21,30]. It was shown [21,30] that this bound can
be softened down to MKK 1 TeV for certain geometrical configurations also reproducing all
fermion masses. In the present paper, cbL,R < 0.5 whereas cdL,R , csL,R > 0.5 so the couplings of
the first two generations of down quarks to KK gauge bosons are quasi universal [20] but different
from the b couplings (especially in RSa). Hence, transformation to the mass basis should induce
significant FC couplings mainly in the b sector but FCNC constraints for the third family are
generally weaker [21,33].
Finally, let us make a few remarks on the case of leptons
= e, and . The constraints from
LEP1 and SLC measurements are very tight since the decay widths
and the asymmetries A
and, in particular the measurement of Ae from the leftright asymmetry and A in polarized
decays, are determined at the percent level. However, it is known that Ae , which is the most precisely measured quantity, shows a 1.7 standard deviation with respect to the prediction in the
SM, ASLD
= 0.1516 0.0021 compared to the fitted value ASM
e
e = 0.1480 [the deviation drops
53
slightly in A
when all leptonic asymmetries are averaged and lepton-universality is assumed].
It is tempting to try to explain this discrepancy within the RS framework with the extended
gauge structure. To do so, one can choose a value ceR 0.56 [in agreement with the analysis of
Ref. [23] of oblique S and T parameters] while the value of ceL can be taken to be much larger,
ceL 0.5. Again for MKK = 3 TeV and sin = 0.1, one obtains then for the electron asymmetry Ae = 0.1504, which is only 0.5 away from the measured value by SLD. This choice of
parameters will also generate an additional deviation in AbFB 34 Ae Ab ; still, there is a choice of
c parameters in the b quark sector which keeps the FB asymmetry close to its measured value.
Thus, a judicious choice of the c parameters in both the bottom quark and electron sectors can
allow to relax the tension between the two measurements which provide the best individual determinations of the weak mixing angle sin2 W and, thus, provide a very nice global fit to the
precision data.
3.3. The FB asymmetry in scenario RSb
In the second scenario RSb, where only the I3R values and thus the QX charges for the downtype right-handed quarks are modified with respect to the previous scenario in which QX =
1
2 (B L), the contributions of the KK excitations to the Zbb coupling are

QbZL
QbZL
QbZR
QbZR

MZ2
1
0.09
F (cbL ),
1
+
4F (cbL )
(0.4MKK )2

MZ2
1.5
1

1
+
F (cbR ),
4F (cbR ) sin2
(0.4MKK )2
(20)
(21)
where we have used again the approximation sin2 1. The effect of the isospin choice
bR
I3R
= + 12 will be simply to reverse the sign of the contribution proportional to sin2 in the
charge QbZR and, therefore, one can have contributions QbZL /QbZL < 0 and QbZR /QbZR > 0 [the
attractive possibility of a positive sign for this shift in the ZbR bR coupling was first pointed out
in Ref. [14]] without flipping the sign of the charge QbZR in contrast to the previous situation. One
can then cure the AbFB (MZ ) anomaly and simultaneously leave Rb almost unchanged, Eq. (16),
without too large corrections for the bR coupling to the Z boson.
In Fig. 2, assuming MKK = 3 TeV (in order to address the gauge hierarchy problem), sin =
0.1 (as before) and using clight 0.5 (in such a way that the observables involving leptons and
light quarks are not affected), we show in the [cbL , cbR ] plane the predicted values of AbFB (MZ )
and Rb in this second scenario (the dotted lines); the bands represent the two observables when
the 1 experimental errors are included. One can see that, as previously, there is a set of [cbL , cbR ]
assignments for which the experimental values of AbFB and Rb are reproduced. However, here,
one needs values of cbR very close to 0.5 to resolve the AbFB (MZ ) anomaly. Note also that, exactly
as in scenario RSa, with a judicious choice of c parameters for the electron, one can also explain
the slight discrepancy of Ae with the value measured at SLD.
4. AbFB outside the Z-pole
To derive the FB asymmetry [and the production cross section] of the bottom quarks outside
the Z-pole, one has to take into account the usual s-channel virtual photon and Z boson ex-
54
Fig. 2. Contour plots in the plane [cbL , cbR ] for MKK = 3 TeV and sin = 0.1, for which AbFB and Rb in the scenario
RSb are equal to their experimental values (dotted lines) and are within their 1 bands; clight 0.5 for leptons and
light quarks.
changes but also, in principle, the exchange of the heavy KK states V (1) (1) , Z (1) , Z (1) as
well as their interference terms. At the tree level, the differential cross for the production of a
fermion pair in e+ e collisions, e+ e f f, can be written as

df
3
= 0 Nc f 1 + f2 cos2 Q1 + 1 f2 Q2 + 2f cos Q3
d cos
8
(22)
with f = (1 4m2f /s)1/2 the velocity of the final state fermion, specifying its direction with
respect to the incoming electron and Nc its color factor and where 0 = 4 2 /3s is the pointlike QED cross section for muon pair production. Integrating over the scatting angle, the total
f
production cross section f (s) and the FB asymmetry AFB (s) are then given by

3
1 2
2
f = 0 Nc f 1 + f Q1 + 1 f Q2
4
3

1

1 2
f
2
AFB = f Q3 1 + f Q1 + 1 f Q2
3
smf
smf
Nc 0 Q1 ,
(23)
3 Q3
.
4 Q1
(24)
In terms of the helicity amplitudes Qij with i, j = L, R, the charges Q1,2,3 are given by [34]

1
|QLL |2 + |QRR |2 + |QRL |2 + |QLR |2 ,
4

1
Q2 = Re QLL QRL + QRR QLR ,
2

1
Q3 = |QLL |2 + |QRR |2 |QRL |2 |QLR |2 .
4
Q1 =
(25)
55
Taking into account all exchanged gauge bosons, the helicity amplitudes Qij read
f
Qij = Qe Qf +
+

V
gV2
e2
QeZi QZj
2 c2
sW
W
e
MZ2
s
+ i Z MZ
QVi QVj QV (cei )QV (cfj )
s
,
s MV2
(26)
where gV = e, gZ = e/sW cW , gZ for V = (1) , Z (1) , Z (1) , Q (1) ,Z (1) = Q(c) and QZ (1) =
Q (c).
To evaluate the FB asymmetry for b-quarks just below or above the Z resonance, s MZ ,
one can safely neglect the contributions of the KK excitations that are exchanged in the s-channel
their couplings to the initial state
since these states are relatively heavy, MKK 3 TeV, and
electrons are rather small. In fact, even at LEP2 energies, s 200 GeV, the effects of the
exchanged heavy excitations can be also neglected as the experimental errors in the measurement
of AbFB (s) are rather large [8]. One then needs to consider solely the photon and the Z boson
exchange contributions and include the mixing effect of the Z boson with the gauge boson KK
states as discussed earlier. Note that one has to take into account the mixing between the B0 and
B 0 mesons when the b-quark is tagged through its semi-leptonic decays for instance as is the
case at experiments far below the Z-pole from PEP to TRISTAN; the bare asymmetry has to be
then multiplied by a mixing factor A
FB = (1 2)AbFB where the parameter is measured to be
0.125 [2].
In principle, to have a very accurate determination of the asymmetry, some higher order effects
[4] need to be included in Eq. (22). Besides the b-mass effects and the QCD radiative corrections
[involving again the mb effects] which can be readily implemented [4,35], very important effects
are the electroweak radiative correctionsand, in particular, initial state radiation (ISR) of photons
which are crucial especially at energies s MZ , as it allows for the return to the Z-pole where
the e+ e bb rate becomes huge, if no cut is applied on the photon energy. The latter corrections can only be implemented in a Monte Carlo based approach which involves a full treatment
of the kinematics.
At the Z-pole, these higher order effects have been implicitly taken into account as we have
evaluated small deviations with respect to the SM prediction given in Ref. [3] and in which
they are already incorporated. A few GeV above and below the Z pole, where high precision
measurements have also been performed by the LEP experiments [3], this approximation might
also hold, except for initial state radiation which needs to be explicitly implemented.
We have
evaluated the FB asymmetry for b-quarks at the three LEP1 running energies s = 89.55, 91.26
and 92.94 GeV in the SM and in the two scenarios RSa and RSb, including the Z and photon exchange and their interference as well as O() initial state radiation. A constant fudge
factor close to unity allows to reproduce exactly the SM curves given in Ref. [3], with the
Higgs boson mass taken to be MH = 100 GeV. Assuming the same fudge factor, the variation of AbFB with the c.m. energy around the Z pole in the two scenarios RSa [with cbL = 0.37,
cbR = 0.287 and clight 0.5] and RSb [with cbL = 0.36, cbR = 0.49 and clight 0.5] which
reproduce as closely as possible the experimental values of AbFB (s) and Rb , is shown in the
left-hand side of Fig. 3 and is compared to the one in the SM; see also Table 1. The quality
of the fit between theoretical and experimental values depends only mildly on the choice of
MKK and sin values. In the present paper, we have fixed MKK = 3 TeV and sin = 0.1. This
sin value corresponds approximatively to the limit of a strong coupling regime for the fivedimensional theory as discussed in Section 3.2, but equivalently, other values can be taken as
56
Fig. 3. The FB asymmetry AbFB at LEP1 (left) and outside the Z resonance (right) as a function of the c.m. energy.
Shown are the predictions in the SM and in the two RS scenarios RSa [with cbL = 0.37, cbR = 0.287, clight 0.5] and
RSb [with cbL = 0.36, cbR = 0.49, clight 0.5] for MKK = 3 TeV and sin = 0.1, as well as the various experimental
measurements with their error bars.
Table 1
The asymmetry AbFB (s) [corrected for B meson mixing] as measured at various c.m. energies, and the standard deviations
in the SM and in the two scenarios RSa and RSb. The measured values and the errors at LEP energies are from Refs. [3,8]
and at lower energies, they are taken from the summary given in Ref. [6]
s
Experimental value
SM sd
2
RSa sd
2
RSb sd
2
89.55 GeV
91.26 GeV
92.94 GeV
29 GeV
35 GeV
44 GeV
58 GeV
190.7 GeV
0.0560 0.0066
0.0982 0.0017
0.1125 0.0055
0.052 0.081
0.214 0.050
0.460 0.147
0.588 0.078
0.5055 0.058
+1.2
+2.5
+2.7
1.5
1.0
+0.3
0.2
+2.0
15.3
7.6
+2.8
+0.5
+0.4
0.8
+0.9
+1.3
+2.7
1.9
8.5
14
+0.4
0.1
+2.0
1.6
1.2
+0.1
0.4
+2.3
4.2
9.8
well (cf. [36]). For the considered sets of parameters corresponding to the best global fit, we
find: Rb = 0.21618, QbZL /QbZL = 0.79%, QbZR /QbZR = 227% [RSa] and Rb = 0.21629,
QbZL /QbZL = 0.85%, QbZR /QbZR = +27% [RSb].
For s = 91.26 GeV, i.e. very close to the Z pole, the experimental value for AbFB (s) lies
almoston top of the predictions in the two RS scenarios. The measurement of the asymmetry at s = 89.55 GeV is also perfectly reproduced in the RSb scenario, contrary to the SM
case in which the prediction is about
1.2 away and to RSa for which there is a large deviation, 2.8 . In turn, the point at s = 92.94 GeV is perfectly reproduced only in the RSa
scenario: the prediction in the SM is 2.7 away, while in the scenario RSb, the deviation from
the experimental central value is at +2 level which is acceptable, though. If one adds the three
2 4.2, than in the
measurements around the Z pole, the 2 is much better in the RSb model, RSb
2
2
2
SM, SM 15.3. In the scenario RSa, the is better than in the SM, RSa 8.5, despite of the
57
large deviation at s = 89.55 GeV. This is a consequence of the fact that the point at s MZ
has a much larger weight in the averaging procedure as its corresponding experimental error is
much smaller.4
Let us now turn to energies far below [from PEP to TRISTAN] and far above [LEP2] the
Z-pole for which the experimental values of AbFB with the error bars are given, respectively, in
Refs. [5,7,8]; these values are summarized in the lower part of Table 1. We have calculated the
asymmetry AbFB (s) as a function of the c.m. energy, taking into account only the effects of B0 B 0
mixing. The results for the FB asymmetry in the SM and the two RS scenarios are shown in the
right-hand side of Fig. 3 and are compared to the experimental values with the error bars that are
given in Table 1.
Here, although the experimental errors are rather large, the general trend is clearly that the
predictions for AbFB (s) in the RSb scenario as well as in the SM [which are almost identical] are
closer to the measured values. There are large deviations between
the predictions in scenario RSa
and the experimental values, in particular at the energy s = 58 GeV, where the difference is
at the level of 2.7 . Adding all the measurements outside the Z-pole region, the 2 in the SM
2 7.6 and 2 9.8, are much better than in the RSa scenario,
and in the RSb scenarios, SM
RSb
2
RSa 14.
In total, when all measurements around and off the Z pole are considered and if the measurement of Rb is included, the agreement between the theoretical prediction and the experimental
2 14, which corresponds to a
data is excellent in the RSb model where the overall 2 is RSb
probability5 of a few percent. In contrast, the prediction in scenario RSa is not satisfactory as
2 22.5 (which corresponds to a probability that is one order of magnitude worse than in
RSa
2 23.4. Thus, one can conclude that only in scenario RSb the
RSb) is as bad as in the SM, SM
b
AFB anomaly is resolved both at the Z-pole and outside the pole. We should however stress again
that, while our results for AbFB on the Z-pole, and to a lesser extent a few GeV around the pole,
are quite robust as we analyzed small deviations compared to the SM values which include all
refinements and higher order effects, the results off the Z-pole are less accurate. A full Monte
Carlo analysis, including eventually the experimental cuts [in particular for ISR] is required to
make very precise statements.
Before closing this section, let us make a brief comment on a scenario proposed in Ref. [14]
where, in addition to the possible modification of the SU(2)R assignment of the bR quarks as in
our scenario RSb, the assignments for the left-handed quarks are also changed, and one has the
representation QiL (2, 2)2/3 . In this case, the two multiplet embeddings for the isospin up-type
uiR quarks allowing invariant Yukawa couplings are: uiR (1, 1)2/3 or (1, 3)2/3 (3, 1)2/3 . In
this context, the necessary shift in the Z bR bR coupling ( QbZR /QbZR > 0) that could explain the
AbFB anomaly at the Z pole can be achieved, assuming that the smallness of the ratio mb /mt
comes from a strong hierarchy among coupling constants [14]. However, we find here that the
best fit of AbFB (s) and Rb corresponds to 2 19.4 which improves the SM case, but is not as
4 One should note that, in fact, the measurements at the two energies s = 89.55 and 92.94 GeV are included in the
averaged value of AbFB given by the LEP experiments [3].
5 To assess the statistical significance of this result, one should take into account the number of free parameters in the
model. While apparently AbFB is adjusted by varying cbR , MKK and sin , these parameters are highly correlated since
they intervene through the combination F (cbR )/(MKK sin )2 . There is, therefore, effectively only one free parameter,
since F (cL ) has no influence on AbFB but is only used to tune Rb .
58
good as our scenario RSb. This fit is obtained for cbR = 0.525 implying QbZR /QbZR = +8.5%
if MKK = 3 TeV and sin = 0.1, the dependence on MKK and sin being small once again.
The reason for this degradation of the fit is that in order to reproduce simultaneously Rb and
AbFB (s), one should have a correction | QbZL /QbZL | of order of the percent [cf. Eq. (16)] while
in this case, in contrast, the bidoublet nature of bL forces this correction to vanish. The breaking
due to the boundary conditions makes this vanishing imperfect but these effects lead typically to
too small deviations: | QbZL /QbZL | < 103 [16]. However, the fit can be improved significantly
by momentum-dependent corrections | QbZL /QbZL | which can reach the percent level for large
mixing between (t, b)L and KK states (if ctL 0.5) [14].
5. Expectations for the ILC
In this section, we analyze the prospects for observing the effects of the new KK states predicted in the RS scenarios discussed here at the ILC, as it is a straightforward generalization of
what has been discussed in the previous sections. At the hadron colliders Tevatron and LHC, the
situation is more delicate and the search will be more complicated as the main decay modes of the
KK states [contrary to conventional Z bosons from GUTs for which BR(Z e+ e , + ) is
at the level of a few percent] will be into top and bottom quark pairs which have large QCD backgrounds; see for instance the general analyses performed in Ref. [37]. A detailed account of the
expectations at these colliders in the particular model discussion here will be given in Ref. [36].
5.1. Expectations for bottom quarks
At the GigaZ option of the ILC, i.e. when running the collider at c.m. energies close to the Zpole, the LEP1 and SLC precision measurements discussed previously can be vastly improved, as
the planed luminosity will be much higher and the longitudinal polarization of the initial beams
available. This will allow to have a more significant determination of the Zbb vertex and thus,
to probe the mixing between the Z boson and the KK states with a much higher accuracy. The
accuracies for determining Rb and, using the polarization asymmetry Ae from ALR and Ab from
AbLR,FB by simply extrapolating the results from LEP1 and SLC are shown in Table 2. For the
normalized partial width Rb , we assume that the double b-tag method will be applied as for LEP1
but that, for a given purity, the b-jet tagging efficiency will be multiplied by a factor of three as
can be inferred from SLD. For Ab , we use the effective efficiency from SLD multiplied by a
factor of two given the improved algorithms already foreseen and the better angular coverage
of the micro-vertex detector at the ILC. Finally, for Ae , we have assumed that the dominant
systematical error from SLD, beam polarization, will be removed by using positron polarization.
Table 2
Deviations and accuracies expected at GigaZ in the b and electron sectors assuming the full accuracy provided by polar
ized electrons and positrons. For ILC operating at s = 500 GeV, one assumes a luminosity of 500 fb1 with polarized
electrons
Energy
Observables
Error %
Deviation %
GigaZ
Rb
Ab
Ae
ALR
0.015
0.03
0.01
0.3
0.3 0.3
3.9 1.4
2.4 1.4
+32
500 GeV
59
Fig. 4. Energy dependence of the leftright polarization asymmetry for b-quarks in the SM and the RSa (left) and RSb
(right) case including the KK contributions and in the pure mixing case. The blue curve is for the SM, the green curve
corresponds to mixing only while the red curve takes into account KK exchange. (For interpretation of the references to
colour in this figure legend, the reader is referred to the web version of this article.)
In view of the very high sensitivity of GigaZ, one can stand a reduction of the new effects
predicted in the RS scenario by up to two orders of magnitude. This would allow to set an upper
limit on the KK mass in the 1530 TeV range. As pointed out already in Ref. [38], GigaZ through
high precision measurements, can therefore already fully cover the RS scenario in a range of parameters which does not reintroduce fine-tuning. It can confirm without ambiguity the indication
observed on Ae at SLD and establish at a very sensitive level that light fermions are weakly
affected, ruling out alternative interpretations [note that in little Higgs and technicolor models,
for instance, there would be no large effects in the b-quark sector anyway]. The measurement of
Ae gives the electron coupling to the Z boson and, therefore, allows to estimate the interference
term at higher c.m. energy between the Z and photon exchanges and the KK excitations.
At higher energies, using the formulae derived in Section 4 and assuming c(eR ) = 0.56, one
caneasily predict the signals which can be observed at ILC. For the b-tagging efficiency/purity
at s = 500 GeV, we have assumed the same efficiency for double tagging. In addition to the
selection effect, one should worry about the determination of the integrated luminosity and, for
some part of the analysis, the knowledge of the beam polarization. For the luminosity, ILC intends to achieve the 104 level, which is sufficient to fulfill our goals. For the polarization, one
can use the e+ e W + W channel which has a strong polarization asymmetry, as a very effective and realistic analyzer of electron polarization. Given the huge cross section of the process,
there is no limitation coming from statistics.
The lower part of Table 2 shows the accuracy of the ALR measurement with L = 500 fb1 ; it
is lower than at GigaZ but the effect is much larger as one can already notice at LEP2 energy from
Fig. 4. The study of the energy dependence of appropriate observables will allow to estimate the
masses of the KK excitations. This is shown in Fig. 4 for the RSa case where one sees a clear
difference between a pure mixing effect [the flat green curve] which dominates a low energies
and the contribution of the KK states [the red curve which contains both contributions] which has
an energy slope which allows to estimate the KK mass. If one uses the high energy data alone,
60
without any theoretical input and without using the measurement of the mixing effect dominant
at low energy, the error on the KK mass would be at the 10% level for a 3 TeV KK state. One can
perform much better by noting that the mixing effect is proportional to F (cqR )/(sin MKK )2 .
For the heavy top quark, one can assume that F (ctR ) is close to the asymptotic value, 8, so
that the mixing determines the denominator. This allows to extract the cfR values for the electron
and the b quarks and thus the Z couplings; in this case one can estimate MKK to about a percent.
This argument indicates how ILC could allow to extract precisely the various parameters of the
RS model.
5.2. Expectations for top quarks
Excellent statistical accuracies are expected
for the channel e+ e t t since 4 105 events
1
can be collected with L = 500 fb at s = 500 GeV. A key issue will be the signal/background
ratio which drives systematic errors of the measurements. Since top quark pairs are mostly made
of six-jet events or four-jet events with one isolated lepton, with two of these jets being b quarks,
they can be very well separated from background events by using the following observations:
(i) W + W events which do not contain b-jets, have 4 jets and are very forward peaked; (ii) ZZ
events are an order of magnitude less numerous, have 4 jets and are forward peaked; (iii) bb
pairs with multi gluons give an incompressible but small background; (iv) ZW + W events are
an order of magnitude below the signal at 500 GeV; the channel Z bb can fake the signal
topology but is at the % level only.
One can thus clearly tag the latter background and eliminate it further by reconstructing the Z
boson mass using the resolution anticipated with particle flow methods developed for ILC detectors which should lead to a few percent resolution on the Z boson mass. This method will also
allow to reconstruct the W boson mass which will further reduce the 4 jet QCD background. Note
also that selecting isolated leptons would decrease the efficiency by a factor of two but allows
for a 100% purity. Therefore, one can conclude that the statistical accuracy achievable at the
ILC is unlikely to be jeopardized by an insufficient knowledge of the background contamination
or of the signal efficiency. This, however, needs confirmation with appropriate simulation and
reconstruction tools.
Table 3 displays the expected accuracies in the measurement of Rt and AtLR with s =
500 GeV and L = 500 fb1 , with longitudinally polarized beams. Taking as an example the LR
asymmetry, the energy dependence of which is shown in Fig. 5, a huge deviation is expected
from the SM case, if one assumes c(tR ) = 0.5 as previously discussed. This means that given
the accuracies expected at the ILC, one could as for GigaZ, allow for two orders magnitude
reduction in the KK effect and still clearly observe a significant deviation. Thus, in the top quark
sector, the sensitivity of the ILC is also sufficient to exclude or confirm the RS scenario; see also
Ref. [39] for instance.
Table 3
Deviations and accuracies expected at ILC in the top sector assuming s = 500 GeV and L = 500 fb1 and electron
polarization
Energy
Observables
Error %
Deviation %
500 GeV
Rt
ALR
0.2
0.1
+24
+200
61
Fig. 5. Energy dependence of the leftright polarization asymmetry for top quarks in the SM and in the RSa and RSb
scenarios. Conventions are as in the previous figure.
6. Conclusion
In this paper, we have analyzed the impact of a RS extra-dimensional model with the extended SU(2)L SU(2)R U(1)X gauge structure allowing relatively light KK excitations,
on the high precision electroweak data. In this model, in order to interpret the mass hierarchies, the fermions are localized differently along the extra dimension so that the electroweak
interactions for the heavy third generation fermions are naturally different from those of light
fermions. The mixing between the Z boson with its KK excitations and those of the new Z
boson, is then large enough to significantly affect the Zbb vertex, while keeping the Z coupling
to light fermions almost unchanged. We have shown that with a judicious choice of the parameter assignments in the b-quark sector and for order TeV KK states, so that the gauge hierarchy
problem is addressed, the 3 discrepancy of the forwardbackward asymmetry of b-quark
jets, AbFB , with measurements performed on the Z-pole can be resolved, while all other precision electroweak data including the normalized Z bb partial decay width Rb , stay almost
unaffected.
The key point of the analysis is to generate a rather large contribution to the right-handed
ZbR bR charge which is small in the SM, to provide the required negative shift in AbFB ; this
contribution is annihilated by a relatively smaller contribution to the left-handed ZbL bL charge,
leading to an almost constant Rb value. In this context, we have analyzed two possible scenarios. In a first scenario, in which the U(1) group is identified with U(1)B L , a large contribution
to the ZbR bR charge is needed as to flip its sign. However, at energies below and above the
Z resonance, although the experimental uncertainties are larger than at LEP1, the theoretical
predictions for AbFB (s) are in a rather poor agreement with the experimentally measured values. In a second scenario, the charges of the U(1) group have been chosen such that the bR
bR
quark has isospin I3R
= + 12 with respect to the new SU(2)R group. In this case, only a 30%
change of the ZbR bR charge is needed to resolve the AbFB anomaly. In this scenario, the predictions for the asymmetry outside the Z-pole are also in a good agreement with the experimental
data.
62
This scenario can be tested in detail at future high-energy colliders, and in particular at the
ILC. At GigaZ, i.e. when running ILC at energies close to the Z-pole, the LEP1 and SLC precision measurements of AbFB , Ae and Rb can be vastly improved, as longitudinal polarization will
be available and the planed luminosities will be much higher. With the expected high accuracies,
one could allow for two orders magnitude
reduction in the new effects and still clearly observe
a significant deviation. At high energies, s 500 GeV, measurements in heavy quark production, in particular in the process e+ e t t, will allow to perform precision measurement of the
masses of the KK states and their couplings to fermions. For instance, KK masses of the order of
1530 TeV can be probed and a mass MKK 3 TeV can be measured at the percent level. A detailed discussion of the impact of this model at the LHC and ILC will be given in a forthcoming
publication [36].
Acknowledgements
Discussions with Christophe Grojean and Ritesh K. Singh are gratefully acknowledged; G.M.
thanks Kaustubh Agashe for useful conversations. The work of A.D. and G.M. is supported by
the ANR for the project PHYSCOL&COS under the number NT05-1_43598.
References
[1] S. Glashow, Nucl. Phys. 22 (1961) 579;
S. Weinberg, Phys. Rev. Lett. 19 (1967) 1264;
A. Salam, in: N. Svartholm (Ed.), Elementary Particle Theory, Almqvist and Wiksells, Stockholm, 1969, p. 367.
[2] Particle Data Group, K. Hagiwara, et al., Phys. Rev. D 66 (2002) 010001;
Particle Data Group, S. Eidelman, et al., Phys. Lett. B 592 (2004) 1.
[3] LEP Collaboration, ALEPH Collaboration, DELPHI Collaboration, L3 Collaboration, OPAL Collaboration, LEP Electroweak Working Group, SLD Heavy Flavour Group, A combination of preliminary Electroweak measurements and constraints on the Standard Model, Phys. Rep. 427 (2006) 257, hep-ex/0509008,
http://lepewwg.web.cern.ch/LEPEWWG. We use the updated values given in the talk of C. Parkes at ICHEP 2006,
Moscow.
[4] A. Djouadi, J. Khn, P.M. Zerwas, Z. Phys. C 46 (1990) 411.
[5] TASSO Collaboration, M. Althoff, et al., Phys. Lett. B 146 (1984) 443;
TPC Collaboration, H. Aihara, et al., Phys. Rev. D 31 (1985) 2719;
JADE Collaboration, E. Elsen, et al., Z. Phys. C 46 (1990) 349;
TOPAZ Collaboration, A. Shimonoka, et al., Phys. Lett. B 268 (1991) 457;
Y. Inoue, et al., Eur. Phys. J. C 18 (2000) 213.
[6] For a review, see: K. Monig, Rep. Prog. Phys. 61 (1998) 999.
[7] D. Choudhury, T.M.P. Tait, C.E.M. Wagner, Phys. Rev. D 65 (2002) 053002.
[8] LEP Collaboration, LEP Electroweak Working Group, hep-ex/0511027.
[9] See for instance, M.S. Chanowitz, Phys. Rev. Lett. 87 (2001) 231802;
G. Altarelli, M.W. Grunewald, Phys. Rep. 403404 (2004) 189;
P. Gambino, Int. J. Mod. Phys. A 19 (2004) 808;
A. Djouadi, hep-ph/0503172, Section 1.
[10] M. Boulware, D. Finell, Phys. Rev. D 349 (1991) 48;
A. Djouadi, G. Girardi, C. Verzegnassi, W. Hollik, F.M. Renard, Nucl. Phys. B 349 (1991) 48.
[11] X.G. He, G. Valencia, Phys. Rev. D 66 (2002) 013004;
X.G. He, G. Valencia, Phys. Rev. D 68 (2003) 033011.
[12] D. Chang, W.F. Chang, E. Ma, Phys. Rev. D 61 (2000) 037301;
F. del Aguila, M. Perez-Victoria, J. Santiago, JHEP 0009 (2000) 011;
R.A. Diaz, R. Martinez, F. Ochoa, Phys. Rev. D 72 (2005) 035018.
[13] See for instance: E. Malkawi, T. Tait, C.P. Yuan, Phys. Lett. B 385 (1996) 304;
R.S. Chivukula, E.H. Simmons, Phys. Rev. D 66 (2002) 015006.
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
K. Agashe, R. Contino, L. Da Rold, A. Pomarol, Phys. Lett. B 641 (2006) 62.

G. Cacciapaglia, C. Csaki, G. Marandella, J. Terning, hep-ph/0607146.
M. Carena, E. Ponton, J. Santiago, C.E.M. Wagner, hep-ph/0607106.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 3370;
M. Gogberashvili, Int. J. Mod. Phys. D 11 (2002) 1635.
A. Pomarol, Phys. Rev. Lett. 85 (2000) 4004;
L. Randall, M.D. Schwartz, JHEP 0111 (2001) 003;
L. Randall, M.D. Schwartz, Phys. Rev. Lett. 88 (2002) 081801;
W.D. Goldberger, I.Z. Rothstein, Phys. Rev. D 68 (2003) 125011;
K. Choi, I.W. Kim, Phys. Rev. D 67 (2003) 045005;
K. Agashe, A. Delgado, R. Sundrum, Ann. Phys. 304 (2003) 145.
K. Agashe, G. Servant, Phys. Rev. Lett. 93 (2004) 231805;
K. Agashe, G. Servant, JCAP 0502 (2005) 002.
T. Gherghetta, A. Pomarol, Nucl. Phys. B 586 (2000) 141.
S.J. Huber, Q. Shafi, Phys. Lett. B 498 (2001) 256;
S.J. Huber, Nucl. Phys. B 666 (2003) 269;
S. Chang, et al., Phys. Rev. D 73 (2006) 033002;
G. Moreau, J.I. Silva-Marcos, JHEP 0601 (2006) 048;
G. Moreau, J.I. Silva-Marcos, JHEP 0603 (2006) 090.
Y. Grossman, M. Neubert, Phys. Lett. B 474 (2000) 361;
T. Appelquist, et al., Phys. Rev. D 65 (2002) 105019;
T. Gherghetta, Phys. Rev. Lett. 92 (2004) 161601;
G. Moreau, Eur. Phys. J. C 40 (2005) 539.
K. Agashe, A. Delgado, M.J. May, R. Sundrum, JHEP 0308 (2003) 050.
G. Burdman, Phys. Rev. D 66 (2002) 076003;
C.S. Kim, et al., Phys. Rev. D 67 (2003) 015001;
J.L. Hewett, F.J. Petriello, T.G. Rizzo, JHEP 0209 (2002) 030;
S.J. Huber, Q. Shafi, Phys. Rev. D 63 (2001) 045010;
S.J. Huber, C.A. Lee, Q. Shafi, Phys. Lett. B 531 (2002) 112;
C. Csaki, et al., Phys. Rev. D 66 (2002) 064021.
F. del Aguila, M. Perez-Victoria, J. Santiago, JHEP 0302 (2003) 051;
M. Carena, T.M.P. Tait, C.E.M. Wagner, Acta Phys. Pol. B 33 (2002) 2355.
M. Carena, E. Ponton, T.M.P. Tait, C.E.M. Wagner, Phys. Rev. D 67 (2003) 096006;
M. Carena, et al., Phys. Rev. D 68 (2003) 035010;
M. Carena, et al., Phys. Rev. D 71 (2005) 015010.
A. Kehagias, K. Tamvakis, Phys. Lett. B 504 (2001) 38.
H. Davoudiasl, J.L. Hewett, T.G. Rizzo, Phys. Rev. D 63 (2001) 075004.
M. Peskin, T. Takeuchi, Phys. Rev. Lett. 65 (1990) 964;
M. Peskin, T. Takeuchi, Phys. Rev. D 46 (1992) 381;
For a similar approach, see: G. Altarelli, R. Barbieri, Phys. Lett. B 253 (1991) 161.
K. Agashe, et al., Phys. Rev. Lett. 93 (2004) 201804;
K. Agashe, et al., Phys. Rev. D 71 (2005) 016002;
K. Agashe, et al., Phys. Rev. D 47 (2006) 053011;
K. Agashe, et al., hep-ph/0606293.
K. Agashe, R. Contino, A. Pomarol, Nucl. Phys. B 719 (2005) 165.
F. del Aguila, et al., Phys. Lett. B 492 (2000) 98;
D.E. Kaplan, T.M. Tait, JHEP 0111 (2001) 051;
F. del Aguila, J. Santiago, Phys. Lett. B 493 (2000) 175.
A. Delgado, A. Pomarol, M. Quiros, JHEP 0001 (2000) 030.
L.M. Sehgal, P.M. Zerwas, Nucl. Phys. B 183 (1981) 417;
A. Djouadi, A. Leike, T. Riemann, D. Schaile, C. Verzegnassi, Z. Phys. C 56 (1992) 289.
J. Jerzak, E. Laemann, P. Zerwas, Phys. Rev. D 25 (1982) 1218;
A. Djouadi, Z. Phys. C 39 (1988) 561;
63
64
[36]
[37]
[38]
[39]
A. Arbuzov, D. Bardin, A. Leike, Mod. Phys. Lett. A 7 (1992) 2029;

A. Djouadi, B. Lampe, P. Zerwas, Z. Phys. C 67 (1995) 123;
G. Altarelli, B. Lampe, Nucl. Phys. B 391 (1993) 3;
V. Ravindran, W. van Neerven, Phys. Lett. B 445 (1998) 214;
S. Catani, M.H. Seymour, JHEP 9907 (1999) 023;
W. Bernreuter, et al., Nucl. Phys. B 750 (2006) 83.
A. Djouadi, G. Moreau, F. Richard, R. Singh, in preparation.
T.G. Rizzo, JHEP 0306 (2003) 021, hep-ph/0610104;
See also: T. Han, G. Valencia, Y. Wang, Phys. Rev. D 70 (2004) 034002.
J.L. Hewett, et al., JHEP 0209 (2002) 030.
E. De Pree, M. Sher, Phys. Rev. D 73 (2006) 095006.
Radiative corrections to the reggeized quarkreggeized

quarkgluon effective vertex
A.V. Bogdan , A.V. Grabovsky
Budker Institute of Nuclear Physics, 630090 Novosibirsk, Russia
Received 18 January 2007; accepted 16 March 2007
Abstract
This paper is devoted to the calculation of the reggeized quarkreggeized quarkgluon effective vertex
in perturbative QCD in the next-to-leading order. The case of QCD with massless quarks is considered and
the correction is obtained in the D 4 limit. This vertex appears in the quark reggeization theory, which
next-to-leading order extension is now under construction.
1. Introduction
The construction of the quark reggeization theory in the next-to-leading approximation (NLA)
involves both the calculation of the next-to-leading order (NLO) corrections to the quark effective vertices and trajectory and the proof of the reggeization hypothesis. Some of these tasks
have already been done. Firstly, all multiparticle reggeon vertices required in the NLA were obtained [1]. It enabled one to prove the quark reggeization hypothesis in the quasi-multi-Regge
kinematics (QMRK) [2], important in the NLA. Next, NLO corrections to the effective particle
particlereggeon (PPR) vertices, appearing in the leading logarithmic approximation (LLA) were
found [3]. It allowed one to obtain the NLO correction to the quark Regge trajectory [4], which
was also a test of the hypothesis in the NLO but only for a particular elastic process.
Work supported by the Russian Fund for Basic Research, project 04-02-16685-a.
E-mail addresses: a.v.bogdan@inp.nsk.su (A.V. Bogdan), a.v.grabovsky@inp.nsk.su (A.V. Grabovsky).

doi:10.1016/j.nuclphysb.2007.03.017
66
A.V. Bogdan, A.V. Grabovsky / Nuclear Physics B 773 (2007) 6583
In this paper the NLO correction to the coupling of two reggeized quarks with external gluon
is calculated. The case of QCD with massless quarks is considered and the correction is obtained
in the (D = 4 2) 4 limit.
The paper is organized as follows. In Section 2 we introduce the reggeized form of the gluon
production amplitude required by analyticity, unitarity and crossing symmetry and calculate the
part of the effective vertex contributing to the discontinuities of the production amplitude in s1
and s2 channels. In Section 3 we obtain the correction to the gluon production amplitude in the
framework of dispersive approach in t1 channel. In Section 4 we calculate one-loop corrections
to quarkphotonreggeized quark vertex. In Section 5 we present the result for the reggeized
quarkreggeized quarkgluon effective vertex.
We perform calculations for massless quarks, but for generality we give the main formulae
with m kept.
2. Production amplitude in the multi-Regge kinematics
Let us consider the high energy process A + B A + G + B of a gluon G production in the
multi-Regge kinematics (MRK), which means that all participating particles are well separated
in the rapidity space and have limited transverse momenta (see Fig. 1):
(pA + pB )2 = s si |tj |,
s1 = (pA + k)2 ,
tj = qj2 ,
q2 = pB pB , k = q1 q2 ,
q1 = pA pA ,
s2 = (pB + k)2 ,
2
s1 s2 sk
.
(1)
Here k is the gluon momentum component transverse to the plane pA , pB :

k = pB + pA + k ,
2
k2 = k
(2)
and the last relation in Eq. (1) for the product s1 s2 is a consequence of the reality of the massless
gluon:
2
= 0.
k 2 = s + k
(3)
The behavior of the amplitude in the MRK is determined by the exchanges in qi channels. For
the case of quark quantum numbers in q1 and q2 channels, the multi-Regge form of the amplitude
proved for the LLA in [5] is
A23 = B B
s22
s11
G
Born
A A ,
m q2
m q1
(4)
where P P and P P are the particleparticlereggeon (PPR) effective vertices, describing P

P transitions due to the interaction with reggeons and taken in tree approximation; i (qi ) is
Fig. 1. Schematic gluon production diagram.
the quark Regge trajectory [6]

D2

d
r
(m q )
2
+ O g4 .
(q) = g CF
D1
2
(2)
(m r )(q r)
67
(5)
Here, the spacetime dimension D = 4 2, CF = N/2 1/(2N ), with the number of colors
N = 3 for QCD. Born =Born (q1 , q2 ) is the reggeonreggeonparticle (RRP) effective vertex,
describing the production of the gluon G at reggeized quark transitions [6]

2p
2p
G
= gt G eG + (m q1 ) A (m q2 ) B ,
Born
(6)
s1
s2
with t G being the color group generator. Amplitude (4) coincides with its real part because its
imaginary part is subleading in the LLA, but it is not quite so for amplitudes in the NLA.
If the reggeization hypothesis in the NLA is correct then, assuming Regge behavior in two
subchannels s1 and s2 (see Fig. 1) with the reggeized quarks in the corresponding crossing channels t1 and t2 , from the general requirements of analyticity, unitarity and crossing symmetry one
obtains [7] the multi-Regge form
1
1
A23 = B B
4
m q2

2 1

s 2
s 2
s1
s1 2
s1 1
s1
+
R
+
R
t2
t2
2
1
2
1
2 1
2
1
1

s2
s2
s2
s2
s 1
s
+
L
+
L
+
2
1
2
1
t1
t1
1
(7)
A A ,
m q1
where PPR vertices P P = P P (t) depend on the polarization of the particles P , P and the
squared momentum transfer t . They are real for t < 0. Hereafter, such expressions as (si )
should be read as (si io) so that for si > 0, ln(si ) = ln(si ) i . The RRP vertices
R = R(q1 , q2 ) and L = L(q1 , q2 ) depend on the polarization of the gluon and momenta q1 , q2 .
These vertices are real in all channels where ti < 0 and k2 > 0. Moreover, in NLO both PPR and
RRP vertices become dependent on the energy normalization scale. Since the amplitude A23 is
physical, it does not depend on this scale. Therefore, the energy normalization points in Eq. (7)
may be chosen in an arbitrary way supposing thatthe corresponding PPR and RRP vertices are
calculated at the same points. Our choice i = (ti )k2 yields a particularly symmetric real
part of the amplitude A23 in the NLA:

2
1
1
1
s1
s2

(R + L)
A A .
Re A23 = B B
(8)
2
2
m q2
m q1
k (t2 )
k (t1 )
Note, that because of freedom in choosing the energy normalization, one has to make sure that
all PPR and RRP vertices in the amplitude are calculated at the proper normalization points.
It is clear from Eq. (7) that the contribution of the sum R + L is leading, while that of the
difference R L is subleading. Indeed, in the LLA the imaginary part of the amplitude is negligible, therefore, only the sum R + L contributes to the effective vertex in Born approximation:
G
.
R(0) + L(0) = Born
(9)
68
On the contrary, if the reggeization is valid, the difference R(0) L(0) at the same order g
contributes to the amplitude (7) only as a radiative correction. It can be obtained together with
the order g 3 corrections to the sum R(1) + L(1) from the analysis of the NLO gluon production
amplitude. We compare projection of this amplitude on the color triplet state in ti channels taken
with the positive signature, with its multi-Regge form, assuming that the one loop corrections
(1)
P P are known. In fact, with such accuracy we get from Eq. (7)
1
1
1
1
G
G
(1)
Born
A(1)
Born
(0)
A + B B

m q2
m q1
m q2
m q1 A A

1
1 (0)
s(s)
s(s)
G
G
+ B B
1 ln
ln
Born
+ 2 Born
2
4
m q2
(t1 )
(t2 )2

G

s1 (s1 )
G
+ Born
ln
1 2 Born
s2 (s2 )

(0)

s1 (s1 )s2 (s2 )
(0)
(0)
(0)
+ R L 1 2 R L
ln
s(s)(k2 )2

1
(0)
+ 4 R(1) + L(1)
(10)
.
m q1 A A
A23 (one-loop) = B(0)

B
For massless QCD i is a scalar and (10) simplifies to

1 G 1 (1)
(1) 1 G 1
(0)
B B Born

q2 Born q1 A A
q2
q1 A A

1 (0) 1 G 1 (0)
s(s)
s(s)
s1 (s1 )
B B Born
A A 1 ln
ln
)
ln
+
+
(
2
1
2
4
q2
q1
s2 (s2 )
(t1 )2
(t2 )2

1 2
s1 (s1 )s2 (s2 )
1 (0)
(0) 1
= B B
.
ln
R(0) L(0)
+ R(1) + L(1)
q2
4
q1 A A
s(s)(k2 )2
(11)
(0)
A23 (one-loop) B B
Eq. (10) states that the difference R(0) L(0) contributes to the discontinuities of A23 in s1
and s2 channels. For small g we obtain

1 (0)
R L(0) 1
m q2

1
(0)
2 R(0) L(0)
.
m q1 A A
(0)
(discs1 + discs2 )A23 = iB B
(12)
On the other hand, the same expression can be found in the one-loop approximation using the
si -channel unitarity conditions [5,6]. Comparing these two approaches one obtains
(0)

R L(0) 1 2 R(0) L(0)
G
G
= Born
(q1 , q2 )1 + 2 Born
(q1 , q2 ) + g 2 (q2 m)
D2
d
r G
r m
N
gt eG C (q1 + r, q2 + r)
D1
2
2 m2 )(r + q )2
(2)
(r + q2 ) (r
1
D2
G
r [(q2 + r ) + m]Born (q1 + r, q2 + r)[(q1 + r ) + m]
1
d
+
2 [(r + q )2 m2 ]
N
(2)D1
[(r + q2 )2 m2 ]r
1
(q1 m),
(13)
69
which gives for the massless case

1 + 2
G
R(0) L(0) = Born
(q1 , q2 )
1 2
D2
2
d
r G
g N
q2 r q1
+
gt eG C (q1 + r, q2 + r) 2
1 2
(2)D1
r (r + q1 )2 (r + q2 )2
D2
G (q + r, q + r)(q + r ) q
d
r q2 (q2 + r ) Born
g2 1
1
2
1
1
,
+
2 (r + q )2 (r + q )2
1 2 N (2)D1
r
1
2
(14)
where

2
2
2q1
2q2
s2
s1
C (q1 , q2 ) = (q1 + q2 ) + pA
+
+
pB
s
s1
s
s2

p
p
= (q1 + q2 ) + 2 A k2 + t1 2 B k2 + t2 .
(15)
s1
s2
The integral over the transverse components of the virtual gluon momentum r is calculated in
Appendix A. Then, substituting the orthogonal momenta
s2 t1 t1
pA p B q 1
s
s
s1 t2 t2
pB + p A q 2 +
q2 = q2 +
s
s
we obtain
q1 = q1
s2
p ,
s A
s1
p
s B
R(0) L(0)
2

G
2g 2 (1 + ) 1
1 + 2 G
1 Born
k
=
Born +
+ ln 2 2
D/2
1 2
N 1 2 (4)

q1 q2
2

2

2
D/2
q2 (kq1 )
q1 (kq2 )
g (4)
G
ln 2
q2 Born q1 ln 2
4(N CF )
1 2
k
k 2 t2
k
k 2 t1

2

2
q
q
q1 q2 1
ln 12 t1 ln 22 t2
gt G eG C (q1 , q2 )
2
t1
t2 2k
k
k
2

2
(pA eG ) (pB eG ) k
q
q
+ gt G
ln 12 t1 ln 22 t2 .
s1
s2
k2
k
k
(16)
(17)
As for the sum R(1) + L(1) , it is clear that, contrary to the difference R(0) L(0) , it cannot
be defined by means of si -channel unitarity in the multi-Regge region. To calculate it we use
a more suitable here t -channel dispersive approach based on analyticity and t -channel unitarity,
developed in [810].
3. t-channel discontinuity
It is possible to calculate the corrections to the RRP vertex by considering gluon production in
various annihilation processes: quarkantiquark, gluongluon, quarkgluon or antiquarkgluon.
Moreover, we can consider photons instead of gluons as the scattered particles, which noticeably
reduces the number of contributing diagrams. Of course, if the reggeization hypothesis is valid,
70
Fig. 2. t1 -channel discontinuity.
we must obtain the same vertex. In the approach based on ti -channel unitarity it looks very
natural.
We choose to consider gluon production in the process of quarkantiquark annihilation into
two photons, therefore in Figs. 1, 2 the particles A , B are photons and G is a gluon. Let us take
into account the amplitude discontinuity in the t1 channel. The discontinuity in the t2 channel
could be considered analogously. The contribution of t1 channel is represented schematically in
Fig. 2. It is given by an ordinary Cutkosky rule from cut of the corresponding diagrams. After
this cut we come to consider the amplitudes being the left part (A) and the right part (B) of Fig. 2.
Since the external gluon is physical and the cut lines in Fig. 2 are on the mass shell, both A and B
are invariant under gauge transformations of the intermediate gluon polarization. Therefore, one
can use an arbitrary gauge. We choose Feynman gauge with the polarization tensor g .
After the convolution we substitute i(p 2 m2 + io)1 for the on-mass-shell -functions
1
2+ (p 2 m2 ) and perform the loop integration thus restoring the amplitude At23
. This amplitude has the same t1 -channel singularities as the complete one. That is enough to restore the
correct Regge asymptotic. Indeed, obtained amplitude has the same t1 singularity as the real
one and an arbitrary polynomial in t1 could be added to the final result in the framework of
this procedure but, such term would have a wrong asymptotic behavior incompatible with the
renormalizability of the theory [14], and, contrary to the case with massive quarks, where the
expression which is equal to the Born amplitude with some constant coefficient can be added,
for the massless limit considered in this paper in detail the correct analytical properties together
with the unitarity requirement in ti -channels determine the amplitude in an unambiguous way.
In order to construct the amplitude with true Regge behavior in MRK, we have to ensure that it
has correct quark quantum numbers, i.e. color triplet and positive signature in ti channels. While
the former is done by applying the projection operator (which is not actually necessary when A
and B are photons)
t c1 t c2
,
c1 |P 3 |c2
=
CF
(18)
the latter can be achieved via the signaturization procedure, which is naturally formulated
for truncated amplitudesthe amplitudes with omitted wave functions (polarization vectors and
Dirac spinors). The procedure is most simple for the massless caseto obtain positive signature
1
with respect to
states in t1 and t2 channels we have to symmetrize the truncated amplitude At23
the substitutions
pA pA ,
s1 s1 ,
s s,
(19)
71
and
pB pB ,
s2 s2 ,
s s.
(20)
So, we have to consider

BA
dD r
t1
,
A23 = S+
D
2
2
i(2) (p m + io)(r 2 + io)
p = q1 + r,
(21)
where the sum is taken over all polarizations of the intermediate particles, the convolution is
performed on-mass-shell (r 2 = 0, p 2 = m2 ), and S+ stands for the signaturization operator
returning the amplitude with positive signature.
On the one hand, we must use the exact expression for the amplitude A, as we integrate over
the momenta r and p of the intermediate particles. On the other hand, since s2 (pB + k)2 is
fixed and large, we take the amplitude B in the quasi-multi-Regge kinematics, which means that
gluon G with momentum k is produced in the fragmentation region of the intermediate quark
and gluon:
B = B B
(0)
1
{r,k}p .
m q2
(22)
Here B B = e B eB and e is a quarkphoton coupling constant. We use the light-cone gauge

for the external photon (eB pB ) = (eB pA ) = 0. One can show that the gauge invariant vertex
{r,k}p , initially calculated from an inelastic quarkgluon process [1], does not depend on the
nature of the particles B and B , so B can be a photon. This vertex has the form

1
1
+ t 1 t 2 (k1 )
{k1 ,k2 }p = g 2 t 2 t 1 (k2 )

p k1 m
p k2 m
[t 2 , t 1 ]
+
(k1 + k2 ) (k1 , k2 )
(k1 + k2 )2
1 2

pB pB
t 2t 1
t t
+
up e1 e2 ,
+ (q2 m)
(23)
(k1 + k2 , pB ) (k1 pB ) (k2 pB )
(0)
where

pB
,
ki ei = 0,
(rpB )
(k1 , k2 ) = (k1 k2 ) g 2k1 g + 2k2 g .
(r) = + (q2 m)
(24)
So, projecting the color state of B on quark quantum numbers (18), we get for massless quarks

p )
G r
2(eG
2(er pB )
1
1
1
B
2t t
{r,k}p = g
e
eG
e q2
CF eG + q2
CF
s2
q1 r 2N r
sB
p k

2(e pB )
N1
s2
+
+ q2 G
r k + q2 1 2
er eG + 2 eG
ker
2 l
uB
uB

p )
2(e pB )(eG
2(e pB )
1 1
N
B
2 er + q2 r
+
reG + q2 r
up . (25)
uB
s2
uB
N sB
Here we use some of the next denotations
(pA + r)2 = sA ,
(pA p)2 = uA ,
(pB r)2 = sB ,
72
(pB + p)2 = uB ,
(r + k)2 = l,
(p k)2 = n,
(26)
and the properties of our kinematics, obtained for the mass shell momenta r, p, and k:
sA + uA = t1 ,
sB + uB s2 ,
l + n = t2 t1 ,
2(rpA ) = sA ,
2(ppA ) = uA ,
2(rpB ) = sB ,
2(ppB ) = uB ,
2(q1 pB ) s2 ,
2(q2 pA ) s1 .
(27)
As for the amplitude A, it is very profitable here to decompose it into an asymptotic (as)
part, giving the asymptotic of the amplitude A in the Regge limit |sA | |uA | |t1 |, and a nonasymptotic (na) part:
A = A(as) + A(na) ,
where
(0)
A A
= eeA
uA ,
(as)
A(as) = pr
1
(0)
,
m q1 A A
A(na) = egt r u p
r e
eA

r
uA ,
sA m2
(28)

2(pA er )
(as)
r
pr = gt u p er + (q1 m)
.
sA m2
(29)
These parts are invariant under gauge transformations of the intermediate gluon polarization as
well as the complete amplitude. The asymptotic contribution to the full amplitude for massless
quarks takes the form:

(as)
{r,k}p pr 1 (0)
dD r
t1 (as)
(0) 1
.
A23 = S+ B B
(30)
q2 i(2)D
q1 A A
r 2 p2
The convolution of the vertices yields:

p A p q1
(as)
2
G
G 2(pA eG ) 1
p
+2
{r,k}p pr = g CF Born gt q1
s1
q1
sA

1
(V1 + V2 ) + N (V3 + V4 ) ,
+ g3t G
N
(31)
where
e p
e p q1 s
q2 (p k)
(p k)
G
G
2n
nsA sB

e p q1 q2 (p k)
e p p B
p A (p k)
G
G
,
+
nsA
nsB

)
2(pB eG
q2 p q1 s
V2 =
q2 p p B +
,
V3 = V2 (sB uB ),
s2 sB
sA

e p k
)

k p eG
q2 p k2(p
k p q1 2(pA eG
B eG )
G
+
V4 =
l
lsA
luB

e p q1 s1
p
q
q2 p q1
2 eG s2
pA e G
s2 pB eG
s1 +
G
+2
lsA uB
luB
lsA

)s
p
(r eG ) q2 p q1 2(reG
p A p q1 (reG ) q2 p p B (reG )
+
+
+
.
+2
lsA
luB
l
lsA uB
V1 =
(32)
(33)
(34)
73
The integrals appearing in Eq. (30) from the convolution (31) are calculated in Appendix B of
this paper and Appendix C of [8]. Due to its length we do not present the integrated expression
for (30) here and will use it for calculating the answer for R + L in Section 5. It is important
to stress that we integrate in the fixed order of limit taking, as it is done systematically in Regge
approach: first si , and only after it D 4. Of course, these two limits must commute
in final infrared stable results for observables, but at intermediate steps one must adhere to the
initially set order everywhere. The consequence of our choice is the prohibition to expand such
terms as s to series with respect to 0.
The integrated expression for (30) has a correct singularity (discontinuity) in the t1 channel.
It also has a t2 channel singularity, but the latter is correct only for the terms with both t1 and
t2 channel singularity. The amplitude with correct analytic properties in both channels can be
easily yielded by symmetry consideration. Using the uncertainty of the dispersion formula (30)
with respect to adding to the convolution (31) some terms proportional to r 2 or p 2 , we can restore
the expression which is symmetric (with respect to change left and right parts of the diagram (see
Fig. 1)), and vanishes after eG k substitution. We find such additional terms after analyzing
the integrated form of (30). They depend on how we treat hidden r 2 and p 2 in V s. In all cases
r into 2
r) r 2 e and set r 2 to 0. So our terms are
r (eG
we commute terms similar to r eG
G

1
p A p q1
2
2
G
G 2(pA eG ) 1
+2
p
p S12 2 g CF Born gt q1
s1
q1
sA
p

)
2
3
G
s q2 eG q1 2q2 (pB eG ) 2q1 (pA eG
r g t
3 G 1
+
V2 + N V3
+
+g t
N
n N
sA sB
sB
sA

)
2(p
e
p2 g 3 t G
B G
+ eG
t2 + q2 p B q2
N
2q2 p B eG
,
+
(35)
n
sB
s2
where S12 is the operator which flips gamma matrix order and makes exchange q1 q2
and pA pB thus yielding the corresponding terms from the contribution of the t2 channel
discontinuity. Expression (35) is applicable for arbitrary D. After adding Eqs. (35)(31) we do
not need to consider the contribution of the amplitude discontinuity in the t2 channel.
It is clear that regardless of energy normalization, R(1) + L(1) contributes only to the real part
of the l.h.s. of Eq. (11) and R(0) L(0) is unambiguously determined by its imaginary part. Since
G and A
both Born
23 vanish after the substitution of the gluon momentum k for the polarization
vector eG , we can conclude that both R(1) + L(1) and R(0) L(0) vanish as well.
t1 (na)
given by the product BAna . Since
Now, let us turn to the non-asymptotic contribution A23
the non-asymptotic part Ana (see (28)) does not contain terms of order sA for large values of
this invariant, the essential region of integration over r in this case is
sA uA r 2 p 2 ti ,
l n s1 ,
sB uB |ti |.
(36)
t1 (na)
,
A23
one may take the amplitude B

It implies that, in order to calculate the contribution of
in the multi-Regge asymptotic region instead of complicated expression (25). In that region one
may use the simple multi-Regge form (4) with the corresponding substitution of momenta and
color indices. Therefore, we obtain
1
1
(0)
G
Born
k(as) ,
B(na-region) = B B
m q2
m q1 rp

(ker )
k(as)
r
up .
rp = gt er + (q1 m)
(kr)
(37)
74

t (na)
1
Thus, the problem of calculating the A23
contribution reduces to a simpler problem for the
elastic case. Moreover, it is not necessary to calculate this contribution at all, since it is totally
absorbed in the correction to the PPR vertices P(1)
P . We stress that one-loop corrections to the
gluon production amplitude A23 (one-loop) include corrections to these vertices, as one can
see from Eq. (11). We will clarify this point in the next section.
4. Asymptotic correction to PPR vertex

(1)
The calculation of R(1) + L(1) includes the one-loop correction to the PPR vertex P P (see
Eq. (11)). Assuming the validity of the reggeization hypothesis, one may find P(1)
P from consideration of the amplitude of quarkantiquark annihilation into photons in Regge limit. We
compare the projection of this amplitude on the positive signature and its Regge form:
(q) (q)
1
1
s
s
A22 = B B (q, s0 )
(38)
+
A A (q, s0 ),
2
m q
s0
s0
where s0 is an energy normalization point. Eq. (7) is written for PPR vertex P P normalized at
s0 = t, so we calculate the correction to this vertex at the same point. The one-loop approximation gives us
A22 (one-loop)

1
1
1 (0) (q) (0)
s(s)
(1)
(0)
(0)
(1)
= B B
A A + B B
A A + B B
A A ln
.
m q
m q
2
m q
(t)2
(39)
Let us note that formula (39), i.e. the reggeization hypothesis at s2 order, is proved [4].
Similarly to the previous section we use t channel unitarity technique and, performing the
Cutkosky cut, deal with the left (A) and the right (B) parts of Fig. 3. After q1 q substitution
the amplitude A coincides with the ones presented in (28) and the amplitude B can be easily
obtained by its Hermitian conjugation and pA pB exchange. We mark its parts with letter B
B(as)
(e.g. pr ).
The contribution of the product B (as) A(na) represents a part of the corrections to the amplitude
(1)
connected with the piece of A A . Because of the factorization property of B (as) we have for the
massless case
B(as) (na)

rp A
dD r
(1)(asna)
= S+
,
A A
(40)
D
r 2 p2
i(2)
(1)
Fig. 3. Calculation of P P .
75
Here, as in the previous section we use the light-cone gauges for external photons
(eA pA ) = (eB pB ) = (eA pB ) = (eB pA ) = 0,
(41)
which means, in other words, that the final result in general gauges will be given by the replacements
eA eA pA
(eA pB )
,
(pA pB )
The integrand of (40) is

eB eB pB

B(as) (na)
rp
A
= eg 2 CF
(eB pA )
.
(pA pB )

q p eA r p B
p eA r
2
uA .
sA
sA sB
(42)
Necessary integrals can be found in Appendix C of this paper and in Appendix A of [8]. We
obtain

3 2
q(e
A q) 2 +
(1)(asna)
A A
(43)
= e(q) eA
+
uA ,
2(1 2)
t
1 2
where (q) is the quark trajectory (5) for m = 0.
(1)(asna)
B(as)
(40) into (11) we substitute pB k, which leads to rp
While inserting A A
k(as)
rp (cf. Eq. (37)). Therefore, due to the analogous factorization property of B (na-region)
represented in Eq. (37), the contribution of the product BA(na) to the gluon production amplitude
(1)(asna)
. So, actuA32 (one-loop) cancels the piece of the second term of (11) given by A A
(na)
to A32 , nor the contribution of
ally, we need to calculate neither the contribution of BA
(1)
B (as) A(na) to A22 . The symmetric situation takes place for the part of B B .
The totally non-asymptotic contribution to the A22 for massless quarks is given by the
product B (na) A(na) . For it, the essential values of (rpA ) and (rpB ) in the integration region are
small: r 2 p 2 sA sB t . Therefore, in Sudakov decomposition (2) parameters and are
suppressed t/s and can be omitted in the propagators:
2
r 2 r
,
p 2 (r + q)2 ,
2
sA r
+ s,
2
sB r
s.
(44)
It leads to the factorization of the corresponding integral (see (21)) into two blob integrals with
respect to and , which can be taken by residues. The corresponding contribution to the total amplitude is pure imaginary at high energies and corresponds to the lowest-order term of
the negative signature. Here we are interested in radiative corrections to the positive signatured
amplitude and therefore contribution of B (na) A(na) is not important for us.
(1)
(1)
Therefore, we have to consider only the piece of A A (and B B ) defined by B (as) A(as) and
subtract from the corrections to the gluon production amplitude A32 (one-loop) in Eq. (11)
as) ).
only the part-connected with this piece (A(1)(as
A
(asas)
(0) 1
A22 = S+ B B
q
dD r
i(2)D
pr pr 1 (0)
.
q A A
r 2 p2
B(as)
(as)
The integrand here has the form:

2p A p q 2q p p B
q p q
B(as) (as)
+
2s
.
pr
pr = g 2 CF p
sA
sB
sA sB
(45)
(46)
76
After the integration we obtain

1
1
3+
(1)(asas)
= e(q) eA +
+ (1) + (1 + ) 2(1 )
A A
2
2(1 2)

q(e
A q) 2
uA .
t
1 2
(47)
Finally, we present the full photonquarkreggeized quark vertex with photon polarization satisfying light-cone gauge conditions (41):

q(e
A q)
q uA ,
A A = e eA (1 + e ) +
(48)
t
where

1
1
3(1 + )
+ (1) + (1 + ) 2(1 ) + O g 4 ,
e = (q)
2
2(1 2)

e = (q)
(49)
+ O g4 .
1 2
This result coincides with the one, which can be easily obtained from the calculations of the
gluonquarkreggeized quark vertex, presented in Fadin and Fiores paper [3]. Indeed, taking
the coefficient only before 1/(2N ) in their result and multiplying it by CF corresponds to
omitting the triple gluon vertex and corrected color algebra.
5. Radiative correction to RRP vertex
Finally, using the table of integrals presented in Appendices A and B gathering results from
previous sections we find the one-loop correction to the RRP vertex. The answer contains terms
q (e.g. term q G q in (17)), which in our
proportional to three gamma-matrix structure q2 eG
1
2 Born 1
kinematics (do not forget about surrounding spinors in the amplitude, where the RRG vertex has
q , so we have
to be inserted) can be replaced by q2 eG
1

a b c
q1 = q1 eG
q2 + q2 eG
q1 (q1 q2 ) eG
+ i 5 d abcd q1
eG q2 ,
q2 eG
(50)
with the last term equal to zero. There is also a linear relation between the parts of the RRP vertex
resulting from the fact that eG can be decomposed into the basis vectors: q1 , q2 .

1
G
Born
= gt G eG C (q1 , q2 ) (q1 + q2 )k2 + (q1 q2 )(t1 t2 )

)
(pB eG
q1 q2
q1 q2 2
2t1 t2
G (pA eG )
+ gt
k
(t1 t2 ) ,
s1
s2
t1
t2
t1
t2
(51)
where = 4(q1 q2 )2 4t1 t2 . We used this relation to simplify our answer for R(1) + L(1) .
R(0) L(0)
1 2
c g 2

2

2
2
k
1
(k + t1 + t2 )2 2t1 (t1 + t2 )
G
G
+ (N CF ) Born
= 2NBorn
ln
ln k

t2
k 2 t1
2
k
(k2 + t1 + t2 )2 2t2 (t1 + t2 )
ln
+
2
t1
k t2
77

q1 + q2 + q1 q2 t1 + t2
V +
V
t1 t2
t1 t2
k2

)
)
(pB eG
(pA eG
q1 q2 (t1 + t2 )
q1 q2
V
V+
, (52)
+ 2gt G
s1
s2
t1
t2
t1
t2
k2
+ gt G eG C (q1 , q2 )
where
V + = ln

t1
t2
+
ln
t
t2 ,
1
k2
k2
(1 + ) (1 )2
,
c =
(1 2)(4)D/2
and the expressions for
G
Born
V = ln
1
Li2 (z) =

t1
t2
ln
t
t2 ,
1
k2
k2
dx
ln(1 zx),
x
(53)
(54)
and C are defined in Eqs. (6), (15).

1
R(1) + L(1) c g 2

2
t1
2
3k2
2 2
N
t1
G
= Born
ln
2 + ln k2 ln2 k2 ln2
+
2

t2
t 1 t2
t2
3

2
2
2
(k + t1 + t2 ) 2t1 (t1 + t2 ) 2 k
+ (N CF )
ln
t2
4k2 t1

2
(k + t2 )2 t12
(k2 + t1 + t2 )2 2t2 (t1 + t2 ) 2 k2
t1
+
ln
Li
+
1
2
t1
t2
4k2 t2
2k2 t1

2
2
2
2
(k + t1 ) t2
t1
t2
t1 + t2 2k
+
Li2 1
ln
+1
2
t1
2(t1 t2 )
t2
2k t2

1
q1 + q2 + q1 q2 (t1 + t2 )
G
W +
W
gt eG C (q1 , q2 ) (N CF )
2
t1 t2
t1 t2
k2

)
)
(pB eG
(pA eG
t1
3 q1 + q2
ln
gt G
+
N (t1 t2 )
t2
s1
s2

t1
q1 q2 +
q1 q2 3t1 t2
+
ln
+
+ (N CF )
W k2
N
t1
t2 t1 t2
t2
t1
t2

2
2
2
2
t1
q1 q2 (t1 + t2 )
(k )
2t1 t2 (k )
ln
(55)
W +
,
t1
t2
t1 t2 (t1 t2 )2 (t1 + t2 )
t2
k2
where

2

1
3t1 t2
t1
1 2 k2
t2
2 k
ln
W = t1 ln
+ ln
t2
+ t1 Li2 1
2
t1
2
t2
t 1 t2
t2
t1

t1
+ t2 Li2 1
,
t2

2

1
k 2 t1 t2
t1
1 2 k2
2 k
ln
W = t1 ln
ln
t2
2
t1
2
t2
t2
(t1 t2 )2

2
t2
t1
2k t1 t2
.
t2 Li2 1
+ 2
+ t1 Li2 1
t1
t2
t1 t22
+
(56)
78
Note that new spin structures appear in Eqs. (52), (55) in comparison with the Born effective
vertex (6) and that the corrections obtained are obviously gauge invariant. It is easy to verify that
poles at t1 = t2 in Eq. (55) cancel. Moreover, one can check that when |q1 | or |q2 | approaches
zero the coefficients at each spin structure remain finite.
Let us stress that the calculated amplitude A23 (one-loop) has the very same nontrivial
structure in energy variables si as predicted by Regge ansatz, which leads to cancellation of all
energy variables in Eq. (11) and finally gives us Eq. (55).
In all the above formulae we used the unrenormalized coupling constant g. Therefore, expressions (52) and (55) for D 4 contain singularities from ultraviolet as well as from infrared
divergences of Feynman integrals. We can remove the ultraviolet divergences from R L and
R + L expressing g in terms of the renormalized coupling constant, for example in the MS
renormalization scheme

2
g (2 D/2)
2
1 11
+
,
N nf
g = g 2D/2 1
(57)
2 3
3
(4)D/2
where nf is the number of flavors and g is the renormalized coupling constant at the renormalization point . At first sight, the absence of the singularity proportional to nf at D = 4 in our
answer for R + L may look suspicious, but one should realize that such a term with the ultraviolet singularity cancels the infrared singularity which arises [15] in the fermion part of gluon
self-energy P f (0) at m = 0.
Acknowledgement
We would like to thank V.S. Fadin for helpful comments and discussions.
Appendix A
Here we calculate the transverse momentum tensor integral appearing in (14) for D 4.
(D2)
) and scalar (I(D2) ) variants one can find in Appendix B
Expressions for its vector (I
of [8].

r r dD2 r
(D2)
=
I
2 (r + q )2 (r + q )2
r
1
2
2
2

q2
q1
g
(q1 k)
(q2 k)
= (1 + ) D/21
ln
ln
2
2
2
2
k t1
k
k t2
k2

q q
q q
1
1
12 1
ln q21 22 2
ln q22

k t1
k t2
2
2

q1
q2
(q q + q2 q1 )
1
1
ln
ln
+
+
O()
.
1 2
(A.1)
2
k 2 t2
k2
k 2 t1
k2
Let us explain how one can obtain this short formula. The method used is well known, and we
describe it shortly here to refer to it in the next section. First, we introduce our notation for
(D2)
:
coefficients in the tensor decomposition of the I
(D2)
I
(D2)
= I
[g]g[D2]
2

i,j =1
(D2)
[qi qj ]qi qj ,
(A.2)
79
where g[D2] is the metric tensor in D 2 dimensions. These coefficients can be expressed
(D2)
through the vector I
(D)
and scalar I integrals:
(D2)
1
(D2)
I
[g] P1 I
q1 (q1 q2 )
2
t1 t2 (q1 q2 )
(D2)
+ t1 P1 I
q2 ,
(D2)
1
(D2)
(D2)
I
[q1 q1 ] =
[g] P1 I
q1 t2
I
2
(q1 q2 ) t1 t2
(D2)
+ (q1 q2 ) P1 I
q2 ,
1 (D)
I(D2) [g] =
I ,
I(D2) [q2 q2 ] = I(D2) [q1 q1 ](q1 q2 ),
2
(D2)
[q1 q2 ] =
(A.3)
(D2)
qi ,
where P1 is the operator returning the coefficient of q1 in the vector integral I
P1 =
i.e.
2 q (q q )
q1 q2
2 1 2
2 q 2 (q q )2
q1
1 2
2
(A.4)
.
(D2)
and solve the equations obtained w.r.t. I

To find (A.3) we multiply (A.2) by qi
[qi qj ].
(D2)
(D)
[g] = I /(2) expressed in radicals
As one can see, relations (A.3) contain unpleasant I
(D2)
[g] is finite for D 4. Therefore, in the
and polylogarithms [11]. But we notice that I
decomposition of the (D 2)-dimensional metric tensor g[D2] in Eq. (A.2) into the two
dimensional metric tensor g[2] and the metric tensor in D 4 dimensions
g[D2] = g[2] + g[D4] ,

the last item can be neglected because it gives O() contribution. Then, the relation

(q1 q2 )2 t1 t2 g[2] = q1 q2 + q2 q1 (q1 q2 ) q1 q1
t2 q2 q2
t1
(A.5)
(A.6)
(D2)
helps us totally cancel I

[g] in (A.2). Finally, we eliminate all Gram determinants
(D2)
(D2)
[qi qj ] by the appropriate choice of a new I
[g],
(q1, q2 )2 t1 t2 from denominators of I
see (A.1).
Appendix B
Here we calculate in the Regge limit some of the integrals appearing in Eqs. (30) and (35).
Other necessary integrals can be found in Appendix C of [8] and Appendices A and B of [9]. We
use the following denotations:
(1 + ) (1 )2
(s1 )(s2 )
c (t)
,
(k12 )
,
I (t) =
,
D/2
(s)
(1 2)
(1 2)(4)

r
dD r
2(D 3) I (t1 )
I4 =
= pA
D
2
i(2) r p 2 sA sB
(D 4) st1

I4
2(D 3) I (t1 ) I (s2 )
2(D 3) I (t1 )
+ pB
q1
+
,
(D 4)
ss2
(D 4) st1
2
c =
(B.1)
(B.2)
80
where
I4 =

1
c
2
s
dD r
=
ln
+
(1)
()
i(2)D r 2 p 2 sA sB
s(t1 )1+
s2
(B.3)
may be found in [9] and (z) = (z)/ (z).

r
2(D 3) I (t1 ) I (t2 )
dD r
2(D 3) I (t1 )
1
I4A =
= pA
k
q1 I4A ,
D
2
2
i(2) r p nsA
(D 4) s1 t1
(D 4)s1
t 1 t2
2
(B.4)

D
r r
d r
I4A =
i(2)D r 2 p 2 nsA

(D 2)I4A
(D 2) I (t1 ) I (t2 )
= k q1 + k q1
+ q1 q1
(D 4)s1
t 1 t2
4(D 3)

2I (t1 )
2t1 (I (t1 ) I (t2 ))
k k I (t2 )

pA p A
+
2
s1
t1 t2
(D 4)s1 t1
(D 4)(t1 t2 )

)
I
(t
)
+
I
(s
)
I
t
I
(t
1
2
1
1 4A
I (t1 )
+ g
+
q1
pA q1 + pA
(4 D)s1
4(3 D)
s1 t1

)
2I
(t
)
I
(D
2)I
(t
t
(D
6)I (s1 )
1
2
1 4A
,
+ pA k + pA k
2(D 3)s1
(D 4)s12
(D 4)s12
(B.5)
where I4A may be found in [8]

1
dD r
I4A =
i(2)D r 2 p 2 nsA

c (1 2) 2
2
(s1 )(t1 )
=
ln
ln2 (t2 ) + 2 ln(s1 ) ln(t1 )
s1 t1 (1 )2 2
(t2 )

t2
2 Li2 1
(B.6)
2 + O().
t1
The most complicated of appearing integrals is a tensor pentabox, which we present in D 4
decomposition:

r r
dD r
1
1
c I5 = c
i(2)D r 2 p 2 nsA sB

g
(t1 + t2 + k2 )2 4t1 t2 2 k12
t2 t1 k2 2 k12
=
ln
ln
+
8s1 s2
t1 t2
t2
t1
k2

t 1 t2 k 2
k
t
12
1
+
ln2
ln2
t1
t2
t2

(t1 t2 )(t1 + t2 + k2 )
t2
2 (k2 )2 (t1 t2 )2
2 Li2 1
+
t1 t2
t1
3
t1 t2

2
2
pA pB + pB pA
(t1 t2 ) + k (t1 + t2 ) 2 k12
ln
+
4ss1 s2
t1 t2
k2

1 2 k12
1 2 k12
1 2 t1
ln
ln
+ ln
+ (t1 t2 )
t2
t1
t1
t2
t1
t2
81

t12 t22
t2
2 (t1 t2 )2 k2 (t1 + t2 )
2 Li2 1
+
t1 t2
t1
3
t1 t2

q1 q2 + q2 q1
t1
4
1 2 t1
+
ln
ln
4s1 s2
(t1 t2 )
t2
t1
t2

2 2
t
t
t2
2 t 1 + t2 k 2
1
2
+
ln (t1 ) ln2 (t2 ) +
2 Li2 1
+
t1 t2
t1 t2
t1
3
t1 t2

2
1
k12
1
k12
k + t1 + t2 2 k12
+ ln2
ln
+ ln2
t2
t1
t1
t2
t1 t2
k2

p q + pA q1 1
1
k12
1
k12
+ A 1
ln
ln2 (k12 ) ln2
ss1 t1

s1
2
2
k2

1
1
1
2
k12
s1
+ ln2
+ ln2 (s1 ) ln2
2
t1
2
2
t1
3

p p
1
1
1
+ A A 2 ln(t1 ) + ln2 (t1 )
ss1 t1

2

p q + pA q2 t1 + t2 + k2 2 k12
1 2 t1
1 2 k12
+ A 2
ln
ln
ln
+
4ss1
t1 t2
t1
t2
t1
t2
k2

2
2
1
t1 t2
t2
t1 + t2 k
k12
2 Li2 1
ln2
t2
t1
t1 t2
t1
3
t1 t2

q1 q1 1
1
t1
1 2 k12
t2
+
+
ln
)
ln
ln(k
12
s1 s2 t1 2 t1 t2
t2
2
t1

1
1
t2
k12
+ ln2 (k12 ) + ln2
ln2 (t1 ) ln2 (t2 )
2
2
2
2(t1 t2 )
k

t2
Li2 1
(B.7)
+ (1 2) + O(),
t1
+
where (1 2) means pA pB , and q1 q2 substitutions.

(D2)
To find (B.7) we perform a procedure similar to the one used for calculating I
in
Appendix A. Again, the coefficient of metric tensor is expressed through scalar pentabox in
D + 2 dimensions and, therefore, is finite, which allows one totally cancel it using the relations
analogous to (A.3) and (A.6). This cancellation looks much simpler in Sudakov basis: pA , pB ,
qi , there corresponding formulae are very similar to Eqs. (A.3) and
g[4]
p
p p + pA
B
= A B
+ g[2] ,
(pA pB )
(B.8)
with g[2] from (A.6). Then, most of remaining vector integrals are known [8,9], and the others
may be calculated by Sudakov decomposition technique exploited in [9]. Fortunately, the analysis
of I5 in Feynman parametrization shows that it does not contain terms proportional to si in
fractional power, such as I (s1 ) in I4A . That is why our answer coincides with the one obtained
in [12] where the limit D 4 was taken first. Then, we use freedom in choice of the coefficient
of g and choose it so (see Eq. (B.7)) that on the one side it eliminates all Gram determinants
(q1 q2 )2 t1 t2 in denominators, and on the other side, contains energy dependence compatible
with reggeization after its insertion in (11). Moreover, in calculating (17) the analogous freedom
82
was exploited (see Appendix A). So, to perform the necessary cancellations of terms proportional
to R(0) L(0) in (11) we have to choose in (B.7) the coefficient of g in a way consistent
(D2)
with corresponding coefficient of I
. Because of this freedom, formula (B.7) does not
straightforwardly coincide with the corresponding one presented in [13], there ln2 (si ) may be
found. Such ln2 (si ) are in contradiction with Regge asymptotic (11) but may be eliminated via
discussed trick. At last, we present one additional vector integral useful for the calculation of I5

r
2(D 3)
dD r
s2
I4k =
= pA
I (s1 )
I4k +
i(2)D p 2 nsA sB
2s
(D 4)ss1

2(D 3)
1
s1

pB
(B.9)
I (s2 ) q1 + q2 I4k ,
I4k +
2s
(D 4)ss2
2
where its scalar variant I4k also appears in (35) and may be found in [9]

1
2
c
(s)k2
dD r
=
ln
+ () (1 + ) .
I4k =
i(2)D p 2 nsA sB
(s1 )(s2 )
s(k2 )1+
Appendix C
(1)
Here we present vector and tensor integrals necessary to calculate A A (see (46)):

r r
dD r
I =
i(2)D r 2 p 2 sA sB

tI
2I (q)
I (t) + I (s)
= pA p A + p B p B
g
+
(D 4)st
(D 4)s
4(D 3)

I (q)

2(D 3)I (t) (D 2)I
+ q q
+
(pA pB ) q + (pA pB ) q
(D 4)st
4(D 3)
st

p p + pA pB (D 2)I (t) (D 6)I (s)

tI
+ A B
+
,
s
(D 4)s
2(D 3)

r
q
2(D 3) (pA q pB )
dD r
I =
(C.1)
=
+
I
I (t),

i(2)D r 2 p 2 sA sB
2
D4
st
where
1
dD r
i(2)D r 2 (r + q)2 sA sB

c
2
s
=
ln
2()
+
(1
+
)
+
(1)
,
t
s(t)1+
I =
was calculated in [8], and for c see Eq. (B.1).

References
[1] L.N. Lipatov, M.I. Vyazovsky, Nucl. Phys. B 597 (2001) 399, hep-ph/0009340.
[2] A.V. Bogdan, A.V. Grabovsky, hep-ph/0606132.
[3] V.S. Fadin, R. Fiore, Phys. Rev. D 64 (2001) 114012, hep-ph/0107010;
M.I. Kotsky, L.N. Lipatov, A. Principe, M.I. Vyazovsky, Nucl. Phys. B 648 (2003) 277, hep-ph/0207169.
[4] A.V. Bogdan, V. Del Duca, V.S. Fadin, E.W.N. Glover, JHEP 0203 (2002) 32, hep-ph/0201240;
A.V. Bogdan, V.S. Fadin, Yad. Phys. 9 (2005) 1659.
(C.2)
[5] A.V. Bogdan, V.S. Fadin, Nucl. Phys. B 740 (2006) 36, hep-ph/0601117.
[6] V.S. Fadin, V.E. Sherman, Pisma Zh. Eksp. Teor. Fiz. 23 (1976) 599, Sov. Phys. JETP Lett. 23 (1976) 548;
V.S. Fadin, V.E. Sherman, Zh. Eksp. Teor. Fiz. 72 (1977) 1640, Sov. Phys. JETP 45 (1977) 861.
[7] J. Bartels, Phys. Rev. D 11 (1975) 2977;
J. Bartels, Phys. Rev. D 11 (1975) 2989;
J. Bartels, Nucl. Phys. B 175 (1980) 365.
[8] V.S. Fadin, L.N. Lipatov, Nucl. Phys. B 406 (1993) 259.
[9] V.S. Fadin, R. Fiore, A. Papa, Phys. Rev. D 63 (2001) 034001, hep-ph/0008006.
[10] V.S. Fadin, R. Fiore, Phys. Lett. B 294 (1992) 286;
V.S. Fadin, R. Fiore, A. Quartarolo, Phys. Rev. D 50 (1994) 5893, hep-th/9405127.
[11] G. t Hooft, M.J.G. Veltman, Nucl. Phys. B 153 (1979) 365;
N.I. Usyukina, A.I. Davydychev, Phys. Lett. B 298 (1993) 363.
[12] Z. Bern, L.J. Dixon, D.A. Kosower, Nucl. Phys. B 412 (1994) 751, hep-ph/9306240.
[13] J.M. Campbell, E.W.N. Glover, D.J. Miller, Nucl. Phys. B 498 (1997) 397, hep-ph/9612413.
[14] V.S. Fadin, E.A. Kuraev, L.N. Lipatov, Phys. Lett. B 60 (1975) 50;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Zh. Eksp. Teor. Fiz. 71 (1976) 840, Sov. Phys. JETP 44 (1976) 443;
E.A. Kuraev, L.N. Lipatov, V.S. Fadin, Zh. Eksp. Teor. Fiz. 72 (1977) 377, Sov. Phys. JETP 45 (1977) 199.
[15] G. t Hooft, M. Veltman, Nucl. Phys. B 44 (1972) 189.
83
Isolated Minkowski vacua, and stability analysis

for an extended brane in the rugby ball
Burak Himmetoglu, Marco Peloso
School of Physics and Astronomy, University of Minnesota, Minneapolis, MN 55455, USA
Received 18 December 2006; received in revised form 25 January 2007; accepted 23 March 2007
Abstract
We study a recently proposed model, where a codimension one brane is wrapped around the axis of
symmetry of an internal two-dimensional space compactified by a flux. This construction is free from the
problems which plague delta-like, codimension two branes, where only tension can be present. In contrast,
arbitrary fields can be localized on this extended brane, and their gravitational interaction is standard 4d
gravity at large distances. In the first part of this work, we study the de Sitter (dS) vacua of the model. The
landscape of these vacua is characterized by discrete points labeled by two integer numbers, related to the
flux responsible for the compactification and to the current of a brane field. A Minkowski external space
emerges only for a special ratio between these two integers, and it is therefore (topologically) isolated from
the nearby dS solutions. In the second part, we show that the Minkowski vacua are stable under the most
generic axially-symmetric perturbations, and we argue that this is sufficient to ensure the overall stability.
1. Introduction
Models with two extra dimensions have been the object of many studies. Among them,
there are the first examples of supersymmetric compactifications to four-dimensional Minkowski
space [1], where a stable spherical internal space is achieved through the flux of a gauge field and
a cosmological constant. More recently, six-dimensional models have been reconsidered in the
brane-world scenario [2], since two is the minimum number of flat extra dimensions to have a
fundamental scale of order TeV and a compactification of (sub)millimeter range. The background
E-mail address: peloso@physics.umn.edu (M. Peloso).

doi:10.1016/j.nuclphysb.2007.03.024
B. Himmetoglu, M. Peloso / Nuclear Physics B 773 (2007) 84104
85
solution of [1] can be easily extended to a 6d braneworld. If one imposes that the internal and
the external coordinates are factorized, then the presence of a brane modifies the internal geometry by creating a deficit angle [3], in the same way as a point mass modifies a two-dimensional
space [4]. In this way, the spherical internal space of [1] is modified to a (so-called) rugby ball,
with conical singularities at the two poles generated by the codimension two brane and its Z2
image. It is manifest from the solution that the only quantity related to the brane tension is the
deficit angle in the bulk, so that one may hope that this construction can provide a self tuning
mechanism responsible for the vanishing of the 4d cosmological constant. Unfortunately, this is
not the case for a number of reasons [5]. Nonetheless, the construction of [3] remains of great
interest, since it is an extremely simple but complete solution of unwarped extra dimensions,
which can be employed in a number of applications of physical relevance [6].
There is, however, a further problem that these applications face, related to the localization of
matter and gauge fields on the codimension two branes. In field theoretical studies, matter and
gauge fields on the brane are usually treated as test fields, which do not contribute to the background geometry (namely, they are neglected in the Einstein equations of the system). However,
a more complete approach, with gravity also taken into account, shows that only tension can be
present on the brane, while fields with a different equation of state necessarily lead to worse than
conical singularities at the brane location [7]. Contrary to what happens for conical singularities
(where the scalar curvature is a delta function supported at the tip of the cone) such singularities
are not integrable, so that the codimension two brane cannot be consistently treated even in a
distributional sense. This appears as a specific example of the problem of defining sources of
codimension two and higher in general relativity [8] (see also [9] for a recent discussion focused
on exact codimension 2 solutions).
A way out of this problem was proposed in [10], through the addition of GaussBonnet terms
in the bulk, and in [11], where the codimension two brane emerges at the intersection between
two codimension one branes. A possibly more conservative approach is to replace the delta-like
strict codimension two brane with an extended codimension one defect [12] (extended defects
on a codimension two bulk were also discussed in [13]). In this model, a region close to the
singularity is replaced by a spherical cap, and a codimension one defect is placed at the junction
between the two spaces (the same is done for both singularities, so as to preserve the Z2 symmetry of the system; moreover, the defect wraps around the main axis of symmetry of the system;
the overall configuration can be seen in Fig. 1 of [12]). Fields with arbitrary equation of state
can be localized on the defect, and their zero modes (that is, the axially-symmetric ones) can be
identified with the 4d fields of the observable sector. It was shown in [12] that the gravitational
interaction between brane fields is described by Einstein 4d gravity at large distance. Therefore,
this construction appears as a phenomenologically viable and complete theory of two extra dimensions, without the need of invoking higher order gravity terms, or a more complicated system
of branes.
A natural extension of this construction is the study of more general background solutions.
For instance, Ref. [14] embedded the codimension one brane in a Minkowski compactification
characterized by a warped internal space. Alternatively, one can study cases of cosmological
relevance, where the brane is embedded in a time dependent background. We perform the first
step in this direction by studying and classifying the de Sitter (dS) vacua of this model. More
precisely, we look for solutions characterized by a dS external geometry, and a static internal
space. More general solutions can presumably be obtained analytically for small deviations from
the dS or Minkowski ones; in general, the values of the bulk cosmological constants and of
the brane tension determine both the dS expansion rate, and the bulk parameters, such as the
86
compactification radius and the deficit angle; if more general sources are present, we expect a
nearly static internal space, and an approximate FriedmannRobertsonWalker cosmology, as
long as their energy densities are smaller than the brane and bulk cosmological constants (this
is typical of extra-dimensional models with a stabilized internal space). At higher energies, we
expect a strong evolution of the entire space, which can be presumably studied only through
numerical computations [15].
We find two main results: first, the dS rate must be smaller than a given value, related to the
inverse of the compactification radius. This is in agreement with the findings of [16], where it
was observed that any compactification mechanism can stabilize the internal space only up to
some given expansion rate (by causality reasons, one cannot expect compactification when the
horizon size becomes parametrically smaller than the size of the internal space). A second, more
surprising, result, is that the Minkowski solutions, obtained for a special relation between the
bulk cosmological constants and the brane tension, are separated from the dS ones by topology.
Indeed, the possible vacua of the model are labeled by two integer numbers. The first integer N
is related to the winding number of bulk fields charged under the gauge symmetry responsible
for the compactification of the space (if present, these fields impose a quantization condition
analogous to the one taking place for the Dirac monopole). This can also be seen as quantization
of the flux of the gauge field in the compact space. The second integer n is related to the current
of a brane field; this current is necessary for matching the discontinuity of the gauge field across
the two sides of the brane, and it controls the position of the brane in the internal space. Ref. [12]
studied only the Minkowski solutions for the system, showing that they occur for N/n = 2 .
We find that a nonvanishing dS rate is instead achieved for greater ratios.
This result has been obtained assuming that the internal space has no warping. In principle,
it may be possible that a Minkowski unwarped solution, characterized by 2n/N = 1, can be
continuously deformed into a dS one, provided a nontrivial warping is also generated in the bulk.
However, we find that such warped solutions are forbidden by the Z2 symmetry across the equator of the internal space, which we impose in order to identify the two branes in the system. It is
indeed possible to show that, if nontrivial, the warping must be monotonic in the bulk. Therefore,
no warped Z2 symmetric solutions exist for our set-up, and the dS and Minkowski solutions are
indeed separated by the quantization conditions mentioned above. In the conclusions, we comment on the implications that this can have for the cosmological constant problem.
Clearly, for the construction of [12] to be of any interest, one should show that it is stable.
Only the zero modes of the perturbations of this geometry were studied in [12]. In the second part
of this work, we study the massive modes, to ensure that the system has no tachyonic instability
(we do so only for the Minkowski vacua; from what we already mentioned, and from the stability
study for the spherical compactification without branes [17], we expect that the system becomes
unstable at high H ). More precisely, we concentrate only on axially symmetric perturbations
(around the axis of symmetry of the background). The reason is mostly technical, since, as we
will see, already this system of modes is quite involved. However, we have a second (although
nonrigorous) justification for this restricted study. The construction is obtained by cutting in an
axially symmetric way the rugby ball and the spherical compactifications, and by joining them
across the brane. Before cutting them, both these configurations are stable [1], so we expect
that an instabilityif anywill be related to their interface. For instance, the rugby ball may
prevail over the spherical cap, so that the brane in between them would shrink towards the pole.
Alternatively, the spherical cap may be favored, and the string would then extend towards the
equator. Such instability would show up as a tachyonic axially symmetric mode; recently, the
stability of 6D chiral gauged supergravity, including the unwarped rugby-ball solution, was
87
studied in [18]. Also that analysis is restricted to axially symmetric perturbations; it is argued that
the study of more general modes is unnecessary, since any angular dependence would contribute
positively to the corresponding KaluzaKlein mass.
We decompose the perturbations into scalar, vector, and tensor modes (where the names refer to how they transform with respect to transformations along the noncompact coordinates).
These three sectors are decoupled at the linearized level, and they can be studied separately. The
equations for the zero modes studied in [12] could be solved analytically. Unfortunately, for massive modes, only the tensor mode can be obtained analytically, while the bulk equations for the
vector and scalar modes have to be solved numerically.1 We do so with a shooting method. No
tachyonic solution emerged from the (rather extensive) numerical computation, so that we can
conclude that the background solution of [12] is stable.
The paper is organized as follows. In Section 2 we review the model introduced in [12].
Section 3 is devoted to the study of the dS backgrounds. In Section 4 we show that no warped
solutions (with a Minkowski or dS external space) are possible in our set up, as a consequence
of the Z2 symmetry. In Section 5 we show that the system is stable under the most general
set of axially symmetric perturbations. The concluding Section 6 contains some remarks on the
phenomenological aspects of the model, with a particular focus on the possible implications of
the separation between the Minkowski and dS solutions.
2. Minkowski compactification
We summarize the construction of [12]. The action of the model is
S = S o + Si + Sb ,
4

1
M
R o,i F 2 ,
So,i = d 6 x g6
2
4

2
v
5
2
Sb = d x s + ( eA) ,
2
(1)
where Sb is the action of the brane, while So,i is the action in two bulk regions separated by the
brane. The line element in the two regions is
ds62 = dx dx + R 2 d 2 + R 2 2 cos2 d 2
ds62
= dx dx + R d + R cos d
2 2
2 2
out,
2
in,
(2)
where here and in the following x denotes the noncompact directions (notice the choice of the
Minkowski metric; in the next section we will instead consider a dS external space). The brane
is at the background position . The region at 0 < < , denoted as the out bulk, is a portion
of the so-called rugby ball compactification, characterized by the deficit angle 1 . The region
1 We are aware of two studies which are close to the present one, where the equations in the bulk could be decoupled,
and then studied analytically. For a spherical bulk, the modes can be decomposed on spherical harmonics, and then
decoupled due to orthogonality in the bulk [17]. This cannot be done for the composite bulk solution that we are
investigating. Second, in the study of [18] the modes could be decoupled by using the equation for the dilaton present in
the supergravity action. This mode, and the corresponding equation, is absent in our case (for other studies of linearized
gravity in 6d contexts, see [19]).
88
< < /2, denoted as in bulk, is a spherical cap, with the pole at = /2. A Z2 symmetry extends this geometry to the region /2 < < 0. In addition, the background is axially
symmetric (around the axis connecting the poles at = /2).
The internal compactification is achieved through the gauge field configuration
F = A = M 2 R cos
(3)
and the cosmological constants

2i =
M2
,
R

M2
20 =
.
R
(4)
If we suppress the external dimensions, the brane can be viewed as a string wrapped around
this axis of symmetry. The brane field acts as a Goldstone boson (v is most easily interpreted
as the vacuum expectation value of a field which breaks the U (1) symmetry on the brane). It
generates a current which is necessary to provide the discontinuity of the magnetic field between
the two bulk regions. From its own equation of motion, = n , where (due to the periodicity of
the coordinate) n is an integer. The brane position is then found to be [12]
= arctan
1
,
Rq 2
q ev,
(5)
while the deficit angle in the out bulk is

1 =
,
(6)
sin
where T is the four-dimensional (i.e., after an integration along ) energy density of the brane.
This result, in the limit /2 (when the brane shrinks to the north pole), reproduces the
known relation between the tension of a codimension two brane, and the deficit angle generated
by it.
To conclude the description of the background, we note that, if some field, with charge e under
the U (1) symmetry is present, the deficit angle must satisfy a quantization condition
2M 4
N
, N integer.
2eM 2 R
Such quantization is also known as flux quantization, since it can be recast in the form

2
B d d F =
N,
e
=
(7)
(8)
where the integral is performed over the entire internal space.

From the Einstein equations, one then finds [12] that this integer N must be related to the
winding n of the brane field by N/n = 2 . As we show in the next section, this relation is
actually due to the assumption of Minkowski noncompact space. Different ratios between these
two integers result in a dS external geometry.
3. De Sitter compactification
We now generalize the Minkowski solution described above to the case of a dS noncompact
space, characterized by the expansion rate H (namely, we replace by the dS metric in the
89
line elements (2)). The bulk compactification is achieved for

M4
1 + 9H 2 R 2 ,
2R 2
in the out bulk, and
0 =
i =

M4
1 + 9 2 H 2 R 2 ,
2
2
2R

F = M 2 R 1 3H 2 R 2 cos

F = M 2 R 1 3H 2 R 2 2 cos
(9)
(10)
in the in bulk. We observe that the expansion rate cannot exceed the value of 1/( 3R). This
is not surprising in the light of the findings of [16], where it was shown that, for any dS compactification, the expansion of the external coordinates has the generic effect of destabilizing the
internal space.
We still look for axially symmetric solutions, so that A = 0, while A depends only on .
Moreover, A = 0 at the poles (from regularity), and A is continuous across the two branes
(so that the brane action is well defined). This determines A in the bulk; the solution cannot be
provided on a unique chart. In the presence of a bulk field charged under this U (1) symmetry,
a consistent solution is possible only when the quantization condition
1/2
1/2

sin + 1 3H 2 2 R 2
(1 sin ) = N
2eM 2 R 1 3H 2 R 2
(11)
with N integer, holds (this computation can be performed exactly as in the Minkowski case;
see [12] for details). Also in this case, the quantization condition can be recast in the form (8).
Let us now discuss the brane equations. As for the Minkowski case, = n, where n is an
integer. By construction, the transverse metric components (g and g ) are already continuous across the brane. We are then left with three nontrivial brane equations (two second Israel
conditions, plus Amperes law, relating the discontinuity of F in the bulk to the current on the
brane). With some algebra, they can be recast in the form
1
= q 2 tan F 2 ,
R
(12)
2n
sin ( 1 3H 2 R 2 F )
1+
,
=
N
1 3H 2 R 2 2 (1 sin ) + 1 3H 2 R 2 sin
1
2s
tan = 4 ,
R
M
(13)
(14)
where q ev, and we have defined, for shortness,

F
1
.
(1 3H 2 R 2 )1/2
(1 3H 2 2 R 2 )1/2
(15)
To solve the above system of equations, we eliminate from (12) and (14), and we combine
the resulting equation with the two bulk expressions for 0 and i . We obtain
16q 2 (i 0 )
,
3(i 0 )2 + 2q 2 (5i 30 ) + 32 q 4
16q 2
R2 =
,
3(i 0 )2 2q 2 (3i 50 ) + 32 q 4
(i 0 )2 + 2q 2 (i + 0 ) 2 q 4
H2 =
,
48q 2
1 2 =
(16)
90
where we have rescaled

2i
20
0 ,
i ,
M4
M4
2s
.
M4
(17)
These solutions are valid only for (i 0 )/q 2 (this is because they are obtained
by
squaring some of the above equations). The maximal allowed value leads to H = 1/( 3R),
which, as we saw from Eq. (9), is the highest possible value that the system can have for the dS
expansion rate.
We see that the Minkowski compactification requires the tuning = ( i 0 )/q. For
a small deviation
i 0
=
+
(18)
q
Eqs. (16) give

0 3 0 ( i + 0 )q
=
+ O 2 ,
3/2
i
2i

1
3 i q
+ O 2 ,
R2 =
+
3/2
0 2 ( i 0 )
0

q 2 (i + i 0 + 0 ) 2
0 i q
2
H =
+ O 3 .
2
6( i 0 )
12( i 0 )
2
From either of (12) and (14), the brane position then satisfies

i 0
5
tan =
+
+ O 2 .
2
4q
q
(19)
(20)
The possible values for the above parameters are constrained by the two integer values N
and n. Eqs. (11) and (13) give

N = 2eM 2 / i + O H 2 ,

3( i 0 )2
2n

H2 + O H4 ,
=
1+
(21)
N
20 i ( i 0 )2 + q 4
where the expansion in has been replaced by an expansion in H 2 through the last of (19).
From the first of (21), we see that we cannot vary alone without also varying the bulk
cosmological constants. However, the Minkowski compactification ( = 0) can be still deformed
continuously into a dS one. However, once also the second condition is taken into account, the
Minkowski solution appears to be detached from the dS ones. Indeed, the choice 2n/N = 1 is
only compatible with H = 0, with the only exception of the trivial case of i = 0 (in which
case, = 1, and the brane is actually absent).2
We can gain further insight by estimating the parameters entering in Eq. (21). Neglecting the
subleading H 2 terms,
nonhierarchical values of the deficit angle (that is, and 1 of
and for
order one), both i and 0 , as well as their difference, are of order 1/R 2 . In addition, q 2 R
does not exceed one (as can be seen in Eq. (20)this value controls the ratio between the radius
2 This conclusion actually holds for arbitrary values of H , and not just at small , as can be seen by studying Eq. (13)
for 2n/N = 1.
91
of the brane and that of the internal space). Therefore, we find

2n
(22)
= O R2H 2 .
N
From the same reasoning, we also see that the expansion at small is actually an expansion for
RH 1.
In the concluding section we comment on the implication of these findings for the cosmological constant problem.
1+
4. Absence of Z2 symmetric warped solutions

It is worth examining whether there exist warped solutions which can continuously interpolate
between the Minkowski and dS compactifications found above. For instance, one can imagine
that a Minkowski unwarped solution, characterized by 2n/N = 1, can be continuously deformed into a dS one, provided a nontrivial warping is also generated in the bulk. There are
interesting solutions in the literature which suggest that this may be the case. Warped solutions,
in the absence of bulk scalar fields, were studied in Refs. [20,21]. More recently, Ref. [14] showed
how to embed codimension one defects of the type proposed in [12] in the warped background
of [21]. The solutions obtained in [12] are characterized by a Minkowski external space, but by
a different relation between N and n than the one holding in the unwarped case.
We note that none of the known warped solutions exhibit a Z2 symmetry in the bulk. We
now present a general argument against this possibility: it is indeed possible to show that, if
nontrivial, the warping must be monotonic in the bulk. This is incompatible with the requirement
of Z2 symmetry across the equator (which we choose to impose, to identify the two branes).
Therefore, such solutions are forbidden by our settings.
To show this, we follow the general method of [20] for obtaining bulk solutions. We start with
the general line element
ds 2 = (l)g dx dx + dl 2 + (l) d 2 ,
(23)
where g is the Minkowski or dS metric, while is the warp factor. Consistently with the axial
symmetry, both and depend only on l (notice that it is always possible to choose a normal
coordinate l, at least locally, in the bulk). Imposing that the bulk is Z2 symmetric around a given
position l implies that both = = 0 at that point (prime denotes derivative with respect to l).
For shortness, we denote this position as the equator of the internal space.
The nontrivial Maxwell equation for this background is solved by
Fl = C / 2 ,
(24)
where C is an integration constant. Once this is imposed, one can then combine two of the
nontrivial Einstein equations to give
+ 2 2 = 0.
(25)
There are two branches of solutions for this equation. One is = constant (where is then
specified by the remaining Einstein equation), which correspond to the unwarped case. The other
branch is
= A 2 /,
(26)
92
where A is another integration constant. We are interested in warped solutions, where vanishes
only at some discrete point in the bulk (namely at l , and possibly, but not necessarily, at some
other location). Such solutions belong to the second branch. We see from Eq. (26) that 0
as l l . Therefore, by continuity, (l ) = 0 for warped solutions (moreover, we also note
from (26) that = 0 implies = 0, which also must hold at the Z2 symmetric point).
Following [20], it is easy to construct Z2 symmetric solutions. In terms of the rescaled quantity
z 5/4 , the only remaining nontrivial Einstein equation can be written as an equation for a point
particle moving without friction in a potential. Consider the situation in which the particle,
due to an initial velocity, climbs up the potential up to some given height, and it then turns
backward. Denoting by l the turnaround time, the absence of friction in the equation for z
ensures that the solution is symmetric around l .
However, despite Z2 symmetric solutions can be easily obtained, we stress that = 0 at the
turnaround point. Therefore, l = l corresponds to a codimension two point in the internal space
[20,21]. The works [20,21] focus on the case where = 0 at two bulk locations, with codimension two branes ending the space there.3 In general, the geometry has two conical singularities
at these two locations, which are supported by the tension of the branes. Notice that there are
particular solutions for which the deficit angle is absent [20]; if this is the case, the location
l = l represents a regular codimension two point in the internal sphere, and the singularity in
the metric is only a coordinate singularity.
This analysis shows that a warped Z2 symmetric solution can only exist if either (i) a brane
with suitable tension is present at the equator (in this case, can change sign at the two sides
of the brane, without going continuously through zero) or (ii) the equator itself shrinks to a
point, and the bulk separates into two distinct regions. While the first case requires different
sources than those considered here, the second possibility is not a continuous deformation of the
solutions studied in this work. We therefore conclude that the separation between Minkowski
and dS compactification is protected by the Z2 symmetry imposed on the system.
5. Stability of the Minkowski compactification
The goal of this section is to obtain the massive perturbations of the system, to verify whether
the background solution described in Section 2 has tachyonic instabilities. For the reasons mentioned in the introduction, we focus on axially symmetric perturbations.4 The most general
perturbations of the geometry of this type are
ds62 = (1 + 2) dl 2 + 2A dl d + (1 + 2C) cos2 d 2 + 2(T + T ) dl dx

+ 2(V + V ) d dx + (1 + 2 ) + 2E, + E(,) + h dx dx , (27)
where we have defined the dimensionful angular coordinates

R d in,
dl
d R d,
R d out,
(28)
and E(,) E + E . The vector modes E , T , V are transverse, and the tensor mode
h is transverse and traceless. The remaining modes are scalar (the denomination refers to
3 In the notation of [21], = 0 corresponds to dA/dw = 0 or, equivalently, to f (r) = 0 in their rescaled metric.
4 Although involved, it is not hard to extend this analysis to general modes. Since the background is axially symmetric,
the general dependence of the modes on the angular coordinate can only be of the form exp(in), with n integer.
93
how these modes transform under 4d coordinate transformations). Simultaneously, one needs to
consider the perturbations of the gauge field,
A = a ,
Al = al ,
A = a + a ,
(29)
where a is a transverse vector mode. Following the discussion of [12], we fix part of gauge freedom by setting E = T = V = al a = 0 (here and in the following, prime denotes derivative
with respect to the rescaled coordinate l). We further impose
brane = ,
( ) = 0
(30)
that is, we require that the brane lies at the unperturbed background position, and that gll = 1
there (this choice includes the Gaussian normal coordinate choice at the brane location, which
is the most convenient one to interpret the gravitational effects measured by brane observers).
These choices do not fix the gauge completely (see [12] for details); however, we can still have a
general (and unambiguous) study if we perform our computation in terms of the combination of
modes which are invariant under the residual gauge freedom.
The tensor mode h , and the vector modes T , a , and V , are already invariant. For the
scalar sector, the invariant combinations are instead
= + E ,
C = C tan E ,
a = a + M 2 cos E ,
(31)
Tensor, vector and scalar modes are decoupled at the linearized level, so we can study the three
sectors separately. We do so in the next three subsections. The relevant equations were obtained
in [12], where the zero modes of the system were then studied. The derivation of these equations
is not repeated here.
5.1. Tensor modes
The axially symmetric tensor perturbation can be decomposed as

hn ( )C,n (x),
h (x, ) =
(32)
where C,n are 4d KaluzaKlein (KK) tensor modes, and hn their wavefunctions in the bulk.
Our goal is to find the allowed perturbations, and their 4d masses mn . The bulk equation
2 h + h tan h = 0
(33)
(where 2 denotes the dAlambertian operator in 4d) can be separated in

dhn
d 2 hn
tan
2 C,n = m2n C,n ,
+ 2n hn = 0,
(34)
d
d 2
where we have moved back to the coordinate, and where the parameter n is different in the
two bulk regions:
2 2 2
mn R
in,
2
n
(35)
2
2
mn R out.
94
From now on, we suppress the index n for brevity, understanding that we are studying one KK
mode at a time.
The bulk equations must be supplemented by a set of boundary and parity conditions. First, we
require regularity at the two poles, imposing that the first derivative of h vanishes there (we can
impose this condition either on the derivative with respect to or l, since the two variables are
simply related by a constant rescaling). Second, parity considerations impose that the modes are
even across the equator, h( ) = h( ). Finally, we must satisfy the junction conditions across
the brane,
hin ( ) = hout ( ),
hin ( ) = hout ( ) or
hin
hout
( ) =
( ).
(36)
The bulk equations

are solved by the Legendre functions P (x) and Q (x), where x = sin
and = 1/2 + 1/4 + 2 (we denote by i and o the values of this parameter in the in and
out bulk, respectively). The bulk solution which is regular at the poles, even across the equator,
and satisfies the first of (36) is

o
2
o
hin = Pi |x| ,
(37)
hout = A cos
Po (x) sin
Qo (x) ,
2
2
where
A=
Pi (|x|)
,
o
2
cos( 2 )Po (x)
sin( 2 o )Qo (x)
x = sin .
(38)
The only undetermined parameter is the mass square of the mode, which enters in the two
parameters i,o .5 It can be found by imposing the only remaining condition to be satisfied, namely
the second of (36). Specifically, for any fixed values of and , we (numerically) look for the
roots of

f m2 =
[hin hout ]|x .
x
(39)
As an example, Fig. 1 shows the behavior of f (m2 ) for the specific choices of = 0.9 and
= 85 . As can be seen in the figure, there are no tachyonic modes in the spectrum (this is the
case also for more negative values of m2 than those plotted). We verified that no tachyonic modes
appear also for several other values of the brane position (those reported in Fig. 2) and the deficit
angle (we varied = 0.2, . . . , 0.9 in steps of 0.1). We also notice the presence of the zero mode
already studied in [12].
Fig. 2 shows instead the mass spectrum for different brane positions, and for the specific
choice = 0.9. The points at = 90 have been obtained for a codimension two brane located
at the pole. We see that the mass spectrum for the codimension-1 model converges continuously
to the one of codimension-2 (we will see that this is not the case in the scalar sector). This can be
proven analytically, from the study of the equations which determine the allowed modes. In the
limit of /2, the in part of the bulk shrinks to zero, so that all the bulk geometry is described
by the rugby ball, as it is the case for the codimension two case; so, the bulk equations converge
to that of codimension two. Moreover, the boundary/parity conditions that we have discussed
5 There should also be an overall normalization, which cannot be determined by the linearized system; we have fixed
it by setting h = 1 at the two poles.
95
Fig. 1. f ((m2 )) for the specific choice = 0.9 and = 85 .
Fig. 2. Smallest masses for the tensor modes, for the specific choice = 0.9 and for different values of the brane position.
above reduce to
dh
= 0,

d /2
h( ) = h( ),
as /2,
(40)
which coincide with those of the codimension two case.

In the codimension two case, the spectrum does not depend on (this is strictly true for
axially symmetric perturbations); this also emerges from our numerical results (not shown here):
the dependence of the spectra on the deficit angle is very weak for any finite , and it disappears
as the codimension one brane is shrunk to the pole.
5.2. Vector modes
The linearized Einstein and Maxwell equations for the vector modes have been derived in [12].
The one for T simply reads
2 T = 0,
(41)
which immediately indicates that this perturbation has no massive modes. We decompose the
two remaining modes as we did for the tensor case,

,n (x)an ( ),
V =
w,n (x)Wn ( )/M 2 .
a =
(42)
n
96
Omitting the index n for brevity reasons, the bulk wavefunctions satisfy the following system of
equations
d 2a
da
1 dW
tan
+ 2 a
= 0,
d
cos d
d 2
dW
da
d 2W
+ tan
+ 2 W + 2 cos
= 0,
d
d
d 2
(43)
where is defined as in Eq. (35).

Due to the parity choice of the background, the mode a must be odd across the equator, while
the mode W must be even,
a( ) = a( ),
W ( ) = W ( ).
(44)
In addition, there are regularity conditions at the poles,

dW
da
=
=0

d /2
d /2
(45)
and junction conditions across the brane.

ain ( ) = aout ( ),

dW
+ 2a cos
d
Win ( ) = Win ( ),

dW
=
+ 2a cos

d
,in
da
a
d
tan

,out
da
a
=

d
tan
,in

,out
(46)
For massless modes, the bulk equations (43) form a system of two coupled first order differential equations in terms of da/d and dW/d . This system can be solved analytically. However,
for nonvanishing mass, these equations must be integrated numerically.
Therefore, to find the spectrum of vector modes, we resort to a shooting method, which is
appropriate for boundary value problems. Each mode is specified by its mass, and by a series
of parameters which determine the initial conditions at one of the poles. For definiteness, we
start from the south pole. As we discuss below, we actually need only one such parameter, which
we denote by C. We start from some guessed values for m2 and C, and we then solve the bulk
equations (43) (when we cross a brane, we impose the conditions (46)). If the resulting solution
turns out to be regular, and to have the correct parity assignment across the equator, then we have
managed to identify one physical mode of the system.
In practice, the bulk solutions that we obtain numerically never satisfy these properties, signaling that the initial guess for the parameters m2 and C was wrong. We can define the two
distances,

a( ) + a( )
W ( ) W ( )
d1
(47)
,
d2
,
a( ) a( )
W ( ) + W ( )
which indicate how far the solution is from being Z2 symmetric. We then proceed in two steps:
(i) we densely scan the parameter space {m2 , C} within some given range; the values leading to
the smallest distances are regarded as our best guesses; (ii) we use a Newtons method to find the
zeros of these distances, starting from the best guesses. Provided the initial conditions are dense
enough, Newtons method converges to all the physical solutions of the system, having values
of {m2 , C} not too far from the probed range of values.6 Indeed, the two-dimensional nature of
6 More accurately, the vanishing of d and d is a necessary, but not sufficient, condition for a mode to be a physical
1
2
perturbations of the system; indeed, in some cases those conditions were (accidentally) satisfied, although the modes
97
the initial parameter space, and the fact that the bulk geometry is regular everywhere, make the
numerical problem a relatively simple one. It is easy to verify (for instance, by increasing the
density of the initial scan) that all the solutions are reached with this method.
The main numerical difficulty occurs at the south pole, where the coordinate system used
is singular. To overcome this, we actually solve (by Taylor expansion) the bulk equations analytically in a neighborhood of the south pole. As for the tensor sector, there is one overall
normalization which cannot be determined by these linearized equations. We fix this by imposing a(/2) = 1.7 We then find

2 2
a(/2 + ) = 1 + C 2 + O 4 ,
(48)
W (/2 + ) = 2C +
+ O 4 .
2
The solutions of a system of second order equations, are usually specified by the values of the
functions and their first derivatives at a given point. In the present case, due to the coordinate
singularity at the poles, we also need to specify one of the second derivatives (we also note that
the linear terms in the expansions vanish, due to the regularity conditions (45)). We started our
numerical evolutions with = 106 , with the values of the wave functions and their derivatives
obtained from (48).
We performed the analysis for several values of (namely = 0.3, 0.6, 0.9) and ( =
60, 65, 70, 75, 80, 85). The initial region scanned was at 100 m2 100 and 20 C 20
(the density was progressively increased until the final roots did not change. The final density
we used was such that the 2 values are varied in steps of 0.2 and C values are varied in steps
of 0.02 in every iteration). In our opinion, this range is sufficiently large to probe the stability
of the model against tachyonic modes. Indeed, the first KK modes are expected to have a mass
of the order of the inverse compactification radius, corresponding to |2 | of order one. Also the
parameter C is naturally expected to be of order one (since it comes from a Taylor expansion).
Moreover, Newtons method can converge to solutions outside this range of starting values for m2
and C (this is indeed what happened in some cases). In all the cases studied, the two distances d1
and d2 strongly increased at negative m2 (they were typically several orders of magnitudes greater
than for positive m2 ), indicating that no tachyons are present in the model; to have a further check,
we started Newtons method also from some of the guessed values with negative m2 (despite the
corresponding distances d1 and d2 were always very high); the method never converged to any
tachyonic mode.
In Fig. 3 we show the obtained spectrum for several brane positions and for the specific choice
= 0.9. We note the presence of one massless mode. As for the tensor case, we also show the
results for the strict codimension two case ( = 90 ). We observe that, also for the vector sector,
the limit of shrinking the extended brane to a codimension two defect is continuous (this can also
be proven analytically from the study of the linearized equations, in the same way as we did for
the tensor modes). Moreover, also for the vector sector, we observed a weak dependence (not
shown here) of the spectrum on the value of the deficit angle (the reason is the same as for the
tensor sector).
did not have the correct parity all throughout the bulk. The easiest way to solve this problem (in an automated way) is
to check the parity condition also at other bulk positions. Specifically, we discharged all those solutions which satisfied
d1 = d2 = 0, but which had the wrong parity at the equator (physical solutions satisfy W = da/d = 0 at = 0). In all
the cases we attempted, this was enough to eliminate all the spurious solutions (we always verified, by direct inspection,
that all the modes which passed this second check had the correct parity all throughout the bulk).
7 This choice does not include the possibility of a = 0 at the pole. In this case, the numerical computation only leads
to the zero mode characterized by constant a and W in the entire bulk.
98
Fig. 3. Smallest masses for the vector modes, for various brane positions and for the specific value = 0.9.
5.3. Scalar modes

There are four scalar gauge invariant combinations, satisfying the set of linearized equations
derived in [12]. Two bulk constraint equations (containing at most first order derivatives) can be
used to express C and a in terms of the other two modes and . This leaves us with the two
bulk equations

2

d
d 2
d
2
3
+
tan
3
+
= 0,
d
d
2
2
d 2
d
2
2
d 2
+
tan
+
= 0,
(49)
d
2
2
d 2
where is related to the physical mass as in (35).
The parity assignment of the background imposes that both modes are even. Moreover, regularity at the poles requires
a /2 =
d a
d

/2
d
d
=
= 0.

d /2
d /2
(50)
Once we insert these conditions in one of the constraint bulk equations (which is legitimate, since
the involved quantities are continuous as we approach the poles), we find
( + )/2 = 0.
(51)
The junction conditions at the brane location were also expressed in [12] in terms of all 4
gauge invariant scalar combinations, plus the quantity E . This quantity can be identified with
the (scalar) perturbation of the brane position. In an arbitrary gauge, the brane is at the perturbed
position = + (x ). This quantity changes when we performed a change of coordinates
involving the bulk coordinate l. The combination which does not change under such change of
coordinate is = E , which can be then interpreted as the gauge invariant perturbation of
the brane position. Not surprisingly, this is the quantity which enters in the junction conditions,
when they are written in terms of the gauge invariant perturbations (31). We further restricted the
gauge freedom by choosing a system of coordinates where the brane remains at the background
position, that is = 0. In this case, = E , which is the quantity entering in the junction
conditions.
99
Fig. 4. Smallest masses for the scalar modes, for various brane positions and for the specific choice = 0.9.
We can combine the junctions conditions given in [12] to eliminate E . This requires the
use of the bulk equations (which is however legitimate, since the junction conditions relate bulk
quantities at the two sides of the brane). After some algebra, we find

[ ]J = 0,
1 + sin2 + sin cos ( ) J = 0,

= 0,
4 2 cos sin ( )
J

1 + 1/
(5 tan + )
(52)
= 0,
cos sin
J
where [f ]J fout fin denotes the difference of the quantity f between the two sides of the
brane.
Also for the scalar sector, the bulk equations (49) must be solved numerically. We therefore
perform a numerical analysis analogous to the one done for the vector modes. We first solve the
bulk equations analytically in a neighborhood of the south pole. Fixing the overall normalization
by setting = 1 at the south pole,8 and taking into account the regularity conditions mentioned
above, we find

(/2 + ) = 1 + C 2 + O 4 ,

2 + 4C 1 2
+ O 4 .
4
(53)
2
Also in this case, the mode is uniquely determined by the two parameters m and C. The
numerical investigation then proceeds in the same way as for vectors. Fig. 4 shows the lightest
masses in the spectrum, for the specific choice of = 0.9 and for different brane positions (the
small oscillatory behavior of the eigenmasses visible in the figure is probably due to numerical
errors, and it gives a measure of the precision of the computation). As for the other two sectors,
the computation does not show any tachyonic modes (the modes exhibit a very bad parity for
all negative values of m2 we have attempted). However, there are two interesting differences
between the scalar and the other two sectors.
The first difference is the absence of scalar zero modes (as can be also verified by solving the
equations analytically, which is possible for vanishing mass). This indicates that all the moduli
(/2
+ ) = 1 +
8 This choice does not include the possibility of = 0 at the pole (such modes could in principle exist, since they
could have a nonvanishing second derivative at the pole). We performed a separate numerical investigation for this case,
which however did not show the existence of any such mode.
100
of the model have been lifted (by the fluxes and tensions in the system), and that the compactification is stable (this is the case also for a codimension 2 brane at the pole [22]). The present
stability analysis is done in absence of (matter or gauge) fields localized on the brane. Ref. [12]
studied the gravitational interaction between brane sources; it was found that two zero modes are
then excited, and contribute to reproduce Einstein 4d gravity at large distances. A similar situation is also encountered in 5d models, for instance the RandallSundrum model with a single
brane [23]. The background solution of [23] has no scalar perturbations; however, when (matter)
fields are localized on the brane, a scalar zero modeoften denoted as brane bending [24]
is excited, and gives a relevant contribution to the gravitational interactions between the brane
fields.
The second peculiarity of the scalar sector is that the limit /2 is discontinuous. To
see this, we solve the linearized equations when the brane is close to the pole, at the position
= /2 + . The analytical solution (53) then accurately describes the modes in the in bulk
immediately before the brane. We can then expand the junction conditions (52) for small , and
obtain the values of the modes in the out part of the bulk immediately after the brane. They are

o = 1 + O 2 ,

do 8C + 3( 1)m2
=
+ O 3 ,
d
4

o = 2 + + O 2 ,
d o 2(1 )
(54)
=
+ O().
d

The limit 0 (i.e. /2) would be continuous if, these values converged to those which
must be imposed for a codimension two brane at the pole. The latter values are + =
d/d = d /d
= 0, which clearly shows that the limit is not continuous (the only exception is the trivial case of a vanishing deficit angle, = 1, when both cases collapse to a spherical
compactification with an empty brane at the pole).
This discontinuity, however, does not lead to any appreciable discontinuity on the lowest
eigenmasses, as can be observed from Fig. 4 (the values for = /2 refers to the codimension two brane). For small , the eigenfunctions, although starting from a different value on the
outside bulk, quickly approach the ones of the codimension two case, leading to nearly identical
eigenmasses (within the accuracy of the numerical computation).
We actually observe from the last of (54) that the limit /2 actually leads to a divergent
derivative of on the outside bulk. This results in divergent terms in the linearized Einstein
tensor. An analogous result is also encountered for the scalar modes excited by brane fields. It
was found in [12] that one scalar mode diverges when the codimension one brane is shrunk to
the pole. This singular limit in the scalar sector is what precludes the localization of matter fields
on a strict codimension two brane.
6. Conclusions
We studied a brane-world model in a six-dimensional spacetime, in which two of the dimensions are compactified by a flux. The model is characterized by a codimension one brane, with
one dimension extending inside the compact space. This construction avoids the singularities
which plague codimension two and higher defects, where only tension can be localized. Indeed,
it was shown in [12] that fields with arbitrary tension can be localized on this defect, and that
their gravitational interaction is described by Einstein 4d gravity at large distance.
We do not regard this construction as a regularized codimension two brane, in the sense that
one does not recover a well-behaved solution as the extended defect is shrunk to a zero size in the
101
bulk. More accurately, the linearized computation of gravity in this model breaks down (in the
scalar sector) as the size of the brane decreases. While the only safe claim that one can make is
that gravity becomes strong in this limit, it is hard to expect a regular nonperturbative limit, in the
light of the fact that the strict codimension two case is itself badly singular. The positive aspect
of this statement is that this construction leads to distinct and potentially testable predictions
for short-range gravitational and electroweak interactions (corrections to standard gravity would
show up at distances comparable to the compactification radius, while electroweak effects would
appear at energies close to the inverse size of the brane in the internal space). We expect that
other inequivalent regular constructions are possible, but that they would lead to different short
range observables.
To make clear predictions, one needs to obtain the massive spectrum of perturbations of the
model. While only the massless modes were studied in [12], we performed this computation
in the second part of this note. Even more importantly, the computation is mandatory to verify
that no tachyonic mode is present, so that the construction is stable. As also argued in [18], the
study of axially symmetric perturbations should be enough for this check (since the background
is itself axially symmetric, modes with a nontrivial angular dependence are expected to have a
higher mass). For this reason, we focused our investigation on these modes.
Due to technical difficulties that we have discussed in the paper, we were not able to decouple the system of bulk equations for the vector and scalar modes, so that we had to resort to a
numerical investigation. We did so with a shooting method (slightly modified, to cope with the
coordinate singularities at the poles). Clearly, numerical methods can only guarantee the stability
within the range probed. However, we conducted a rather extensive search. While KaluzaKlein
masses are naturally expected to be of the order of the inverse compactification radius, we densely
investigated a parameter space in the interval9 100/R 2 m2 100/R 2 , and for several bulk
parameters (brane position and deficit angle). In no case we found evidence for tachyonic solutions. Since this is a relatively easy numerical problem (the bulk is regular, and there are no strong
hierarchies present), we believe that the present analysis ensures the stability of the construction.
While the above considerations are valid for a Minkowski external space, in the first part of
this analysis we studied the dS solutions of this model. We found that the space of vacua is
characterized by discrete points labeled by two integer numbers, related to the quantized values
of the flux in the bulk (N ) and of the current of a brane field (n), which, in turns, controls the
position of the brane in the internal space. The Minkowski compactification requires10 N = 2n.
If this is the case, we can actually have different Minkowski compactifications, provided the
cosmological constant on the brane () and on the two sides of the bulk (i,0 ) satisfy
i 0
2e2 M 8
=
.
,
i =
(55)
q
N2
As long as these relations are satisfied, a change in the brane and bulk tensions leads to a different
internal space, but only to Minkowski external geometry. Since the ratio of the two integers N
and n cannot be varied continuously, it is tempting to ask whether this can be of some relevance
for the cosmological constant problem.
Discrete vacua typically arise in presence of fluxes, and several studies have already attempted
to use this as a solution of the cosmological constant problem. The original mechanism of [25]
9 We actually employed a Newton method which could converge to solutions also outside this interval, if present.
10 This is actually true only for an unwarped internal space, since a more general relation can be obtained in the warped
case [14]. However, we have seen in Section 4 that warped solutions do not survive the Z2 symmetry across the equator.
102
is realized with a 4-form in four dimensions. This form can acquire only quantized values in
units of a charge q, and its energy density behaves as a cosmological constant. The value of the
form can change through membrane nucleation; this is however a very slow process, and it could
be possible that the present universe is trapped in a metastable state, where the vacuum energy
and of the 4-form add up to the observed value of the cosmological constant tot . This mechanism allows for several possible values of tot , and, provided these values are densely packed
together, one may hope to reproduce the observed expansion rate for some value of the quantized
flux (even if this is not the case initially, one should simply wait until the flux tunnels to a value
compatible with observations). This, however, requires very small values of q, whichbesides
being unnaturaldo not lead to the known cosmology (they would lead to a reheating temperature much smaller than the one needed for primordial nucleosynthesis, see [26] for a discussion).
String theory can offer a big improvement in this respect. There, one requires an internal space
which can be stabilized by fluxes [27], and which can have a complex topology, with several
noncontractible cycles. There are various ways of wrapping fluxes around such internal space,
leading to several quantized fluxes, and to a multi-dimensional set of discrete vacua. Due to the
high dimensionality, it is natural to expect vacua with a value of tot compatible with the observed one, even if the distance between different vacua (namely, the value of the charge q) is
large [26]. This is one of the possible realizations of the landscape of string theory [28].
In the present context, inserting the measured value of H in Eq. (22), we find
2

2n
R
60
.
10
1+
(56)
N
0.1 mm
In the mechanism of [25], the present value of H is achieved at the price of an unnaturally
small charge, and very large flux. Our realization does not put a significant constraint of the
charge. However, the value (56) requires a very tiny mismatch of the relation 2n/N = 1,
which can be achieved only when the two winding numbers are themselves O(1060 ). Although
the corresponding request (very large flux) is usually not listed as a drawback of [25], it is hard
to regard such high windings as natural.
We can think of some possible improvement. The necessity of large windings is probably
due to the extreme simplicity of the model. As we mentioned, the problem of localizing fields
in general relativity is present for any defect of codimension higher than one, and not just for
codimension two. Assuming that such a construction must be done also for more realistic (and
richer) models, we can expect that the presence of more fluxes can allow for a solution without
too large winding, in a similar way as the string realization improves over the one of [26]. Alternatively, we may be satisfied in providing at least a partial solution to the cosmological constant
problem. Rather than requiring large windings, it is probably more natural to assume that we
are locked in one of the Minkowski vacua, with N = 2n. This relation (which by itself does
not appear unnatural) could possibly explain why the big cosmological constant vanishes. The
coincidence problem could instead be solved by some additional field, which would then play
the role of quintessence. This would still be an improvement with respect to the usual models of
quintessence, where the absence of a big cosmological constant is typically left unexplained.
To improve over these considerations requires a better understanding of the background solutions of the model. For instance, it may be possible to have N = 2n, even if the cosmological
constants do not satisfy the relations (55). This may be compatible with a more general solution than the restricted ansatz (Minkowski or dS external geometry, times a static internal space)
assumed here. Moreover, in order to study quintessence in this context, one needs to include
sources with a different equation of state than vacuum. It is usually hard, if not impossible, to
103
obtain analytical solutions with a time evolving internal space. However, such questions can be
possibly addressed analytically at low energies, or numerically along the lines of [15].
Acknowledgements
We thank Tony Gherghetta, Nemanja Kaloper, Lorenzo Sorbo, and Gianmasimo Tasinato
for very useful discussions. This work was partially supported by DOE grant DE-FG02-94ER40823, and by a grant from the Office of the Dean of the Graduate School of the University of
Minnesota.
References
[1] P.G.O. Freund, M.A. Rubin, Phys. Lett. B 97 (1980) 233;
S. Randjbar-Daemi, A. Salam, J.A. Strathdee, Nucl. Phys. B 214 (1983) 491.
[2] N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 429 (1998) 263, hep-ph/9803315;
I. Antoniadis, N. Arkani-Hamed, S. Dimopoulos, G.R. Dvali, Phys. Lett. B 436 (1998) 257, hep-ph/9804398.
[3] R. Sundrum, Phys. Rev. D 59 (1999) 085010, hep-ph/9807348;
J.W. Chen, M.A. Luty, E. Ponton, JHEP 0009 (2000) 012, hep-th/0003067;
S.M. Carroll, M.M. Guica, hep-th/0302067;
I. Navarro, JCAP 0309 (2003) 004, hep-th/0302129.
[4] S. Deser, R. Jackiw, G. t Hooft, Ann. Phys. 152 (1984) 220.
[5] Y. Aghababaie, C.P. Burgess, S.L. Parameswaran, F. Quevedo, Nucl. Phys. B 680 (2004) 389, hep-th/0304256;
I. Navarro, Class. Quantum Grav. 20 (2003) 3603, hep-th/0305014;
H.P. Nilles, A. Papazoglou, G. Tasinato, Nucl. Phys. B 677 (2004) 405, hep-th/0309042;
H.M. Lee, Phys. Lett. B 587 (2004) 117, hep-th/0309050;
J. Vinet, J.M. Cline, Phys. Rev. D 70 (2004) 083514, hep-th/0406141;
J. Garriga, M. Porrati, JHEP 0408 (2004) 028, hep-th/0406158.
[6] I. Antoniadis, K. Benakli, M. Quiros, New J. Phys. 3 (2001) 20, hep-th/0108005;
C. Csaki, C. Grojean, H. Murayama, Phys. Rev. D 67 (2003) 085012, hep-ph/0210133;
G. von Gersdorff, M. Quiros, Phys. Rev. D 68 (2003) 105002, hep-th/0305024;
T. Appelquist, B.A. Dobrescu, E. Ponton, H.U. Yee, Phys. Rev. Lett. 87 (2001) 181802, hep-ph/0107056;
B.A. Dobrescu, E. Poppitz, Phys. Rev. Lett. 87 (2001) 031801, hep-ph/0102010;
M. Fabbrichesi, M. Piai, G. Tasinato, Phys. Rev. D 64 (2001) 116006, hep-ph/0108039;
N. Borghini, Y. Gouverneur, M.H.G. Tytgat, Phys. Rev. D 65 (2002) 025017, hep-ph/0108094;
T. Asaka, W. Buchmuller, L. Covi, Phys. Lett. B 523 (2001) 199, hep-ph/0108021;
G. Burdman, B.A. Dobrescu, E. Ponton, hep-ph/0506334;
W.D. Goldberger, M.B. Wise, Phys. Rev. D 65 (2002) 025011, hep-th/0104170;
E. Dudas, C. Grojean, S.K. Vempati, hep-ph/0511001;
E. Dudas, C. Papineau, V.A. Rubakov, hep-th/0512276;
E. Dudas, C. Papineau, hep-th/0608054.
[7] J.M. Cline, J. Descheneau, M. Giovannini, J. Vinet, JHEP 0306 (2003) 048, hep-th/0304147;
P. Bostock, R. Gregory, I. Navarro, J. Santiago, Phys. Rev. Lett. 92 (2004) 221601, hep-th/0311074.
[8] R. Geroch, J.H. Traschen, Phys. Rev. D 36 (1987) 1017.
[9] N. Kaloper, D. Kiley, JHEP 0603 (2006) 077, hep-th/0601110.
[10] O. Corradini, Z. Kakushadze, Phys. Lett. B 506 (2001) 167, hep-th/0103031;
O. Corradini, A. Iglesias, Z. Kakushadze, P. Lengfelder, Phys. Lett. B 521 (2001) 96, hep-th/0108055;
H.M. Lee, G. Tasinato, JCAP 0404 (2004) 009, hep-th/0401221.
[11] N. Kaloper, JHEP 0405 (2004) 061, hep-th/0403208.
[12] M. Peloso, L. Sorbo, G. Tasinato, Phys. Rev. D 73 (2006) 104025, hep-th/0603026.
[13] J. Louko, D.L. Wiltshire, JHEP 0202 (2002) 007, hep-th/0109099;
B.M.N. Carter, A.B. Nielsen, D.L. Wiltshire, JHEP 0607 (2006) 034, hep-th/0602086;
M. Kolanovic, M. Porrati, J.W. Rombouts, Phys. Rev. D 68 (2003) 064018, hep-th/0304148;
S. Kanno, J. Soda, JCAP 0407 (2004) 002, hep-th/0404207;
J. Vinet, J.M. Cline, Phys. Rev. D 71 (2005) 064011, hep-th/0501098;
104
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
I. Navarro, J. Santiago, JHEP 0502 (2005) 007, hep-th/0411250;

C. de Rham, A.J. Tolley, JCAP 0602 (2006) 003, hep-th/0511138.
E. Papantonopoulos, A. Papazoglou, V. Zamarias, hep-th/0611311.
J. Martin, G.N. Felder, A.V. Frolov, M. Peloso, L. Kofman, Phys. Rev. D 69 (2004) 084017, hep-th/0309001;
J. Martin, G.N. Felder, A.V. Frolov, L. Kofman, M. Peloso, Comput. Phys. Commun. 171 (2005) 69, hepph/0404141.
J. Lesgourgues, S. Pastor, M. Peloso, L. Sorbo, Phys. Lett. B 489 (2000) 411, hep-ph/0004086;
A.V. Frolov, L. Kofman, Phys. Rev. D 69 (2004) 044021, hep-th/0309002;
C.R. Contaldi, L. Kofman, M. Peloso, JCAP 0408 (2004) 007, hep-th/0403270.
R. Bousso, O. DeWolfe, R.C. Myers, Found. Phys. 33 (2003) 297, hep-th/0205080;
J.U. Martin, JCAP 0504 (2005) 010, hep-th/0412111.
C.P. Burgess, C. de Rham, D. Hoover, D. Mason, A.J. Tolley, hep-th/0610078.
M. Giovannini, Phys. Rev. D 66 (2002) 044016, hep-th/0205139;
M. Giovannini, J.V. Le Be, S. Riederer, Class. Quantum Grav. 19 (2002) 3357, hep-th/0205222;
H. Yoshiguchi, S. Mukohyama, Y. Sendouda, S. Kinoshita, JCAP 0603 (2006) 018, hep-th/0512212;
Y. Sendouda, S. Kinoshita, S. Mukohyama, hep-th/0607189;
H.M. Lee, A. Papazoglou, Nucl. Phys. B 747 (2006) 294, hep-th/0602208.
C. Wetterich, Nucl. Phys. B 255 (1985) 480;
J.M. Schwindt, C. Wetterich, Phys. Lett. B 578 (2004) 409, hep-th/0309065.
S. Mukohyama, Y. Sendouda, H. Yoshiguchi, S. Kinoshita, JCAP 0507 (2005) 013, hep-th/0506050.
M.L. Graesser, J.E. Kile, P. Wang, Phys. Rev. D 70 (2004) 024008, hep-th/0403074.
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hep-th/9906064.
J. Garriga, T. Tanaka, Phys. Rev. Lett. 84 (2000) 2778, hep-th/9911055;
S.B. Giddings, E. Katz, L. Randall, JHEP 0003 (2000) 023, hep-th/0002091.
J.D. Brown, C. Teitelboim, Phys. Lett. B 195 (1987) 177;
J.D. Brown, C. Teitelboim, Nucl. Phys. B 297 (1988) 787.
R. Bousso, J. Polchinski, JHEP 0006 (2000) 006, hep-th/0004134.
S. Kachru, R. Kallosh, A. Linde, S.P. Trivedi, Phys. Rev. D 68 (2003) 046005, hep-th/0301240.
L. Susskind, hep-th/0302219.
Erratum
Erratum to: Order-s2 corrections to the polarized

structure function g1(x, Q2 )
[Nucl. Phys. B 417 (1994) 61100]
E.B. Zijlstra , W.L. van Neerven
Instituut-Lorentz, University of Leiden, P.O. Box 9506, 2300RA Leiden, The Netherlands
Received 2 March 2007
Eq. (A.2) penultimate line

1
1
(1 + 11z) ln z (1 11z) ln z.
3
3
(1)
Eq. (A.4) is correct up to the fifth line. After the fifth line it reads

760
11
1
(1 z) 20 ln2 (1 z) 88 ln(1 z) +
32 1 + z2 + z +
3
3
3z

16
Li2 (z) + ln z ln(1 + z) + 50 + z2 10z ln2 z 32(2 z) ln z ln(1 z)
3

32
4
+ (119 13z) ln z 72 + z2 40z (2) 8(3 + z) Li2 (1 z).
(2)
3
3
Eq. (A.5) is the most complicated one. The order s term does not change. The CF Tf part of
the order s2 term proportional to L2M does not change. However the CF Tf part of the term
proportional to LM does change and well into

8(1 2z) Li2 (1 z) 2 ln2 (1 z) + 3 ln z ln(1 z) ln2 z + 4 (2)

+ 4(17 20z) ln(1 z) 4(12 8z) ln z 68 + 52z LM .
DOI of original article: 10.1016/0550-3213(94)90538-X.
E-mail address: ellen.e.b.zijlstra@shell.com (E.B. Zijlstra).

doi:10.1016/j.nuclphysb.2007.03.002
(3)
106
E.B. Zijlstra, W.L. van Neerven / Nuclear Physics B 773 (2007) 105106
The non-log term of the CF Tf part also changes. Up to the sixth line it is okay but after the sixth
line it becomes equal to

1
+4(5 12z) Li2 (1 z) + 32 1 + z2 2z ln z Li2 (z) + 123 104z2 48z ln2 z
3
(88 96z) ln z ln(1 z) + 6(9 12z) ln2 (1 z) 32 (2)z2 ln(1 z)

1
4 31 4z2 26z ln(1 z) + 416 48z2 274z ln z 8 5 4z2 26z (3)
3

2
4
2
81 52z 108z (2) + (233 239z).
(4)
3
3
The CA Tf part remains unchanged.
Nuclear Physics B 773 [PM] (2007) 107136
Non-linear supersymmetry for non-Hermitian,

non-diagonalizable Hamiltonians: I. General properties
A.A. Andrianov a,b, , F. Cannata b , A.V. Sokolov a
a V.A. Fock Institute of Physics, Sankt-Petersburg State University, Russia
b Dipartimento di Fisica and INFN Bologna, Italy
Received 10 October 2006; received in revised form 18 February 2007; accepted 14 March 2007
Abstract
We study complex potentials and related non-diagonalizable Hamiltonians with special emphasis on formal definitions of associated functions and Jordan cells. The non-linear SUSY for complex potentials is
considered and the theorems characterizing its structure are presented. We define the class of complex potentials invariant under SUSY transformations for (non-)diagonalizable Hamiltonians and formulate several
results concerning the properties of associated functions. We comment on the applicability of these results
for softly non-Hermitian PT-symmetric Hamiltonians. The role of SUSY (Darboux) transformations in increasing/decreasing of Jordan cells in SUSY partner Hamiltonians is thoroughly analyzed and summarized
in the Index Theorem. The properties of non-diagonalizable Hamiltonians as well as the Index Theorem
are illustrated in the solvable examples of non-Hermitian reflectionless Hamiltonians. The rigorous proofs
are relegated to part II of this paper. At last, some peculiarities in resolution of identity for discrete and
continuous spectra with a zero-energy bound state at threshold are discussed.
1. Introduction
Quantum physics of open systems often deals with incomplete information on the influence
of an environment and can be adequately described by non-Hermitian Hamiltonians with a nonpositive imaginary part. This kind of effective description has been employed in condensed
matter, quantum optics and hadronic and nuclear physics [14] for many years. Non-self-adjoint
DOI of companion paper (part II): 10.1016/j.nuclphysb.2007.03.015.
* Corresponding author at: Dipartimento di Fisica and INFN Bologna, Italy.
E-mail address: andrianov@bo.infn.it (A.A. Andrianov).

doi:10.1016/j.nuclphysb.2007.03.016
108
A.A. Andrianov et al. / Nuclear Physics B 773 [PM] (2007) 107136
operators were also under mathematical investigations [57] and recently interesting examples
of non-Hermitian effective Hamiltonian operators have been found for the quantum many-body
equations [8].
The PT-symmetric quantum mechanics proposed in [9,10] and developed in [1014] and
its pseudo-Hermitian generalization [1517] describes a variety of non-Hermitian Hamiltonians
with real spectrum (but not all Hamiltonians with real spectrum are PT-symmetric [18,19]). There
is a progress in understanding some non-Hermitian but PT-symmetric Hamiltonians in terms of
Krein spaces [20]. This kind of quantum mechanics has attracted much interest as it may open
the way to give a solid probabilistic interpretation of non-Hermitian dynamics by means of a positive pseudo-norm [11,12]. PT-symmetry endows with a physical meaning the energy spectrum
of some Hamiltonians formally unbounded from below [9,10]. The latter possibility for anharmonic oscillators with potentials unbounded from below was observed long ago [21,22] but only
recently has been associated with a PT-symmetry [24].
For complex, non-Hermitian potentials the natural spectral decomposition involves biorthogonal states [25]. Moreover the Hamiltonians may not be diagonalizable [26] but can be reduced
only to a quasi-diagonal form with a number of Jordan cells [15]. This feature appears at level
crossing which, in fact, occurs under specific circumstances in atomic and molecular spectra [26]
and optics [27,28] as well as in PT-symmetric quantum systems [23,29,30]. There are also certain
links [31] to the occurrence of non-Hermitian degeneracies for essentially Hermitian Hamiltonians where the description has been developed for complex eigenvalue Gamow states (resonances)
unbounded in their asymptotics and, in general, not belonging to the Hilbert space of physical
wave functions. On the contrary, in what follows we examine non-Hermitian Hamiltonians with
normalizable bound and associated states. The subtleties of biorthogonality (the phenomenon
of self-orthogonality [28]) in resolution of identity and in definition of quantum averages of
observables have been thoroughly analyzed in our paper [32].
We find it certainly interesting and important to investigate the possible ways for quantum
design of such non-Hermitian quantum systems and in particular to extend the methods of nonlinear SUSY algebra [3364] in order to keep under control the emerging of non-diagonal parts
of those systems. In making a link to PT-symmetric systems we restrict ourselves with a soft type
of non-Hermiticity when the real part of a potential dominates over the imaginary one at both
infinities and asymptotically such a potential remains bounded from below. Respectively the
energy spectrum of a related system contains a number of bound states and possibly a continuum
part bounded from below. Thus having in mind the SUSY quantum design one can, for instance,
think of a chain of complex Hamiltonians produced by DarbouxCrum transformations from
a real one as a good representative of the class of softly non-Hermitian systems. The general
relations and theorems presented in Sections 24 certainly hold also for PT-symmetric potentials
with fixed asymptotics of ratio of imaginary and real part (semihard non-Hermiticity), say, for
potentials with leading asymptotics x 2n (ix) ; > 0 at infinities provided that the boundary
conditions for eigenvalue problem do not require to move to complex coordinates [10,23] (i.e.
for || < 1). However the lemmas and theorem of Sections 5 and 6 are proven for softly nonHermitian potentials and we pay hopes to extend them also on semihard non-Hermitian potentials
in a nearest future.
We start in Section 2 with the definitions and a summary of properties of non-Hermitian diagonalizable Hamiltonians and introduce the relevant biorthogonal expansions. Then we consider
non-diagonalizable non-Hermitian Hamiltonians with discrete spectrum and finite-size Jordan
cells and discuss the choice of the biorthogonal basis with diagonal resolution of identity. The
novel result of this section is the proof that a biorthogonal basis always exists which is made
109
of a set of eigenfunctions and associated functions of the initial Hamiltonian and a set of their
complex conjugates for the Hermitian conjugated Hamiltonian. Moreover if the Hamiltonian is
PT-symmetric and this symmetry is not spontaneously broken on states (eigenvalues are real)
then the elements of direct and conjugated bases are related by PT-reflection. In Section 3 the
origin of non-diagonalizable Hamiltonians is clarified to be level confluence.
Non-linear SUSY in QM is summarized and extended to complex potentials in Section 4 with
an emphasis to the possibility of conservation of PT-symmetry. Herein the important theorem
on the polynomial structure of SUSY algebra with transposition symmetry as well as the stripoff theorem describing the minimization of the differential order of intertwining operators are
adapted to the complex potentials. These theorems involve the zero-mode subspaces of supercharge componentsintertwining operators and their mapping by Hamiltoniansmatrices S.
They are well compatible with PT-symmetry (if any). The relationship between superpotentials
and Wronskians involving associated functions is discussed.
In Section 5 we present the class of complex potentials invariant under SUSY transformations for (non-)diagonalizable Hamiltonians: this class covers the systems with soft breaking of
Hermiticity and essentially real continuum spectrum (if any). For this case we formulate several
results characterizing the normalizability of associated functions at + and/or . The necessary conditions for SUSY transformation functions are found to provide a pre-planned Jordan
structure of a SUSY partner Hamiltonian. These results allow to unravel the relation between
Jordan cells in SUSY partner Hamiltonians the latter being described by the Index Theorem in
Section 6. It represents the main result of the present paper. The Index Theorem relates the dimensions of Jordan cells of super-partner Hamiltonians at any energy level with characteristics
of intertwining operator kernels (matrices S) and in fact exhaustively describes the quantum design options for softly non-Hermitian Hamiltonians. Needless to say that the latter theorem is
also compatible with PT-symmetry when non-Hermiticity, for instance, is introduced into P-even
potentials by shifting of coordinates into complex plane (see examples in [29]). The illustration
of properties of non-diagonalizable Hamiltonians as well as of the Index Theorem is thoroughly
performed in Section 7 by the solvable example of non-Hermitian reflectionless Hamiltonians
originated by SUSY transformations from the free particle Hamiltonian. The arising of nondiagonalizability is illuminated by an exactly solvable system with two coalescing bound states.
In conclusions we outline possible peculiarities of non-Hermitian Hamiltonians with continuous spectrum. The approaching to the continuum threshold yields more subtle problems with
normalizable eigen- and associated functions in continuum which may have zero binorm. As a
consequence it may cause serious problems with the resolution of identity investigated in detail
elsewhere [32].
All the new results on the structure of non-diagonalizable SUSY Hamiltonians presented in
this (part of) paper are rigorously proved and justified in the accompanying (second part of)
paper [65].
2. Non-Hermitian diagonalizable vs. non-diagonalizable Hamiltonians and biorthogonal
expansions
In our paper we deal with complex one-dimensional potentials V (x) = V (x) and respectively
with non-Hermitian Hamiltonians h of Schrdinger type,1 defined on the real axis,
1 Conventionally the system of units m = 1/2, h = c = 1 will be used with dimensionless energies, momenta and
coordinates.
110
h 2 + V (x),
(1)
which are assumed to be symmetric or self-transposed under the t transposition operation,

h = ht . The notation d/dx is employed. Only scalar local potentials will be analyzed which
are obviously symmetric under transposition (for some matrix non-diagonalizable problems, see
[66,67]). Taking into account possible applications in PT-symmetric QM we specify complex
potentials to give a semihard non-Hermiticity when the real part Re V is bounded from below
and the ratio Im V / Re V remains finite and sufficiently small for large x . In this case the
eigenvalue problem can be safely posed keeping the boundary conditions at x on the real
axis. Later on, in last two sections we restrict ourselves with softly non-Hermitian potentials with
vanishing asymptotic ratios Im V / Re V = o(1).
Let us first define a class of one-dimensional non-Hermitian diagonalizable Hamiltonians h
with discrete spectrum such that:
(a) a biorthogonal system {|n , | n } exists,
h | n = n | n ,
h|n = n |n ,
n |m = m | n = nm ;
(2)
(b) the complete resolution of identity in terms of these bases and the spectral decomposition of
the Hamiltonian hold (in the case of PT-symmetric potentials the necessary conditions for
that are formulated in [20]),

I=
(3)
|n n |,
h=
n |n n |.
n
In the coordinate representation,

n (x) = x| n ,
(4)
the resolution of identity has the form,

n (x ) n (x).
(x x ) = x |x =
(5)
n (x) = x|n ,
The differential equations,

hn = n n ,
h n = n n ,
(6)
and the fact that there is only one normalizable eigenfunction of h for the eigenvalue n (up to a
constant factor), allows one to conclude that
n (x) n n (x),
n = const = 0.
(7)
Hence the system {|n , | n } can be redefined

1
|n |n ,
n
| n

n | n ,
(8)
so that
n (x) n (x),
+
n (x)m (x) dx = nm .
(9)
111
We stress that the non-vanishing binorms in Eq. (9) support the completeness of this basis, i.e.
the resolution of identity,

n (x)n (x ).
(x x ) =
(10)
n
Indeed if some of the states in Eq. (10) were self-orthogonal (as it has been accepted in [66]),
i.e. had zero binorms in (9), the would-be unity in (10) would annihilate such states thereby
signaling the incompleteness.
The PT-symmetry of a potential entails a related symmetry of eigenfunctions,
V (x) = V (x)
n (x) = n (x) n n (x),
|n | = 1,
(11)
when the PT-symmetry is not spontaneously broken, i.e. n = n (further on, for clarity, we
restrict ourselves only with a case of the unbroken PT-symmetry although nearly all results can
be generalized to the case of spontaneous PT-symmetry breaking with pairs of eigenstates having
mutually complex conjugated eigenvalues). The normalization (8) leads to the value n = 1.
One can see that the biorthogonality does not, in general, provide positive binorms of states
related by the PT-symmetry,
+
n (x)m (x) dx = n nm = nm ,
(12)
bringing negative norm states in the PT-odd sector.

For non-Hermitian Hamiltonians one can formulate the extended eigenvalue problem, searching not only for normalizable eigenfunctions but also for normalizable associated functions for
discrete part of the energy spectrum. Some related problems have been known for a long time in
mathematics of linear differential equations (see, for instance, [68]).
Let us give the formal definition.
Definition 1. The function n,i (x) is called a formal associated function of ith order of the
Hamiltonian h for a spectral value n , if
(h n )i+1 n,i 0,
(h n )i n,i 0,
(13)
where formal emphasizes that a related function is not necessarily normalizable.

In particular, the associated function of zero order n,0 is a formal eigenfunction of h (a solution of the homogeneous Schrdinger equation, not necessarily normalizable).
Let us single out normalizable associated functions and the case when h maps them into
normalizable functions.2
Evidently this may occur only for non-Hermitian Hamiltonians. Then for any normalizable
associated functions n,i (x) and n ,i (x) the transposition symmetry holds
+
+
hn,i (x)n ,i (x) dx =
n,i (x)hn ,i (x) dx.
2 It takes place for a certain class of potentials described in Section 6, see part II of our paper.
(14)
112
Furthermore one can prove the following relations:

+

n,i (x)n ,i (x) dx n,i
, n ,i = 0,
n = n ,
(15)
where ( , ) is scalar product.

As well, let us take two normalizable associated functions n,k (x) and n,k (x) so that, in
general, k = k and there are two different sequences of associated functions for i k and i k
n,i (x) = (h n )ki n,k (x),
n,i (x) = (h n )k i n,k (x).
(16)
Then
+

n,i (x)n,i (x) dx = n,i
, n,i = 0,
i + i max{k, k } 1.
(17)
In particular, for some normalizable associated function n,l (x), the self-orthogonality [66] is
realized,
+
2
n,l
(x) dx = 0,

n,l (x) = (h )il n,i (x),
l = 0, . . . ,

i 1
.
2
(18)
All the above relations are derived from the symmetry of a Hamiltonian under transposition and
the very definition of associate functions.
We proceed to the special class of Hamiltonians for which the spectrum is discrete and there
is a complete biorthogonal system {|n,a,i , | n,a,i } such that,
(h n )|n,a,i = |n,a,i1 ,
h|n,a,0 = n |n,a,0 ,

h n | n,a,pn,a i1 = | n,a,pn,a i ,
h |n,a,pn,a 1 = n |n,a,pn,a 1 ,
(19)
where n = 0, 1, 2, . . . is an index of an h eigenvalue n , a = 1, . . . , dn is an index of a Jordan cell

(block) for the given eigenvalue, n ; dn is a number of Jordan cells for n ; i = 0, . . . , pn,a 1
is an index of associated function in the Jordan cell with indexes n, a and pn,a is a dimension of
this Jordan cell. We have taken a general framework which is applicable also for matrix and/or
multidimensional Hamiltonians. But the main results of this and the next sections are guaranteed
only for scalar one-dimensional Hamiltonians with local potentials.
We remark that the number dn is called a geometric multiplicity of the eigenvalue n . For a
scalar one-dimensional Schrdinger equation it cannot normally exceed 1 (but may reach 2 in
specific
cases of periodic potentials and of potentials unbounded from below). In turn, the sum

p
a n,a is called an algebraic multiplicity of the eigenvalue n .
The completeness implies the biorthogonality relations
n,a,i |m,b,j = nm ab ij ,
(20)
and the resolution of identity

I=
dn p
+
n,a 1

n=0 a=1
i=0
|n,a,i n,a,i |.
(21)
113
The spectral decomposition for the Hamiltonian can be constructed as well,

pn,a 1
pn,a 2
dn
+

h=
|n,a,i n,a,i | +
|n,a,i n,a,i+1 | .
n
n=0 a=1
i=0
(22)
i=0
It represents the analog of the block-diagonal Jordan form for arbitrary non-Hermitian matrices [69].
If existing such biorthogonal systems are not unique. Indeed the relations (19) remain invariant
under the group of triangle transformations,

=
ij |n,a,j ,
|n,a,i
0j i

=
| n,a,k
kl | n,a,l ,
(23)
klpn,a 1
where the matrix elements must obey the following equations,

ij = i+1,j +1 = ij,0 ij ,
00 = 0,
kl = k+1,l+1 = kl+pn,a 1 kl,pn,a 1 ,
pn,a 1,pn,a 1 = 0.
(24)
The biorthogonality (20) restricts the choice of pairs of matrices and in (23) to be,
= 1 .
(25)
This freedom in the redefinition of the biorthogonal basis is similar to Eq. (8) and it can be
exploited to define the pairs of biorthogonal functions n,a,i (x) x|n,a,i and n,a,i (x)
x| n,a,i in accordance with (9). However one has to take into account our enumeration of
associated functions n,a,i (x) vs. their conjugated ones n,a,i (x) as it is introduced in Eqs. (19)
n,a,i (x) = n,a,p

(x) n,a,pn,a i1 |x.
n,a i1
(26)
Then the analog of Eq. (9) reads,

+
n,a,i (x)m,b,pm,b j 1 (x) dx = nm ab ij .
(27)
We stress that this kind of biorthogonal systems is determined uniquely up to an overall sign.
In these terms it becomes clear that the relations (17) have the meaning of orthogonality of
some off-diagonal pairs in the biorthogonal system {|n,a,i , | n,a,j } as
n,a,i (x) = (h n )pn,a 1i n,a,pn,a 1 (x),
n,a,j
(x) = n,a,pn,a 1j (x) = (h n )j n,a,pn,a 1 (x).
(28)
When comparing with specification of indices in Eq. (17) one identifies pn,a 1 j i, i i .
In both cases k = k = pn,a 1. Then the inequality (17) singles out off-diagonal binorms,
i j 1. From Eq. (28) it follows that in order to have all diagonal binorms non-vanishing it is
sufficient to prove that at least one of them is not zero because
114
+
n,a,0 (x)n,a,pn,a 1 (x) dx
+

(h n )pn,a 1 n,a,pn,a 1 (x) n,a,pn,a 1 (x) dx
+
=
n,a,i (x)n,a,pn,a 1i (x) dx = 0.
(29)
The latter is necessary for the completeness of the basis because of the absence of biorthogonal
pairs of basis elements made of bound and associated functions in the diagonal resolution of
identity. If some of such pairs in resolution of identity (21) were biorthogonal then this operator
would be at best a projector but not an identity.
For a scalar one-dimensional Schrdinger equation the geometric multiplicity dn of the eigenvalue n cannot normally exceed 1. The latter possibility of non-degenerate eigenstates will be
implied throughout this paper. Thereby in the rest of the paper the index a = 1 = dn will be
omitted.
It is certainly interesting to examine what are specific features of biorthogonal bases for PTsymmetric systems. Let us restrict ourselves by one-dimensional systems with non-degenerate
spectrum of eigenstates and with the unbroken PT-symmetry, n = n . First one can easily find
that the functions n,pn i1 (x) are normalizable solutions of the initial Hamiltonian h and
therefore can be decomposed into a linear combination of its basis,
n,pn i1 (x) = n,i

(x) =
n,ij n,j (x),
(30)
0j i
where n,ij = n,ij due to equations conjugated to (19). By complex conjugation of Eq. (30)
and its further convolution with n,li (and after changing the sign x x) one comes to the
conditions on matrix elements,

2

(31)
n,li n,ij
= 0;
n,0
= 1.
j <l
j il
The analysis of the biorthogonality relations (20) and (27) for the PT-symmetric basis (30)
leads to the conclusion that all numbers n,ij are it real. Then one derives from (31) that
n,ij |j <i = 0 and the elements of a Jordan cell basis, the eigen- and associated functions are
simultaneously PT-even or PT-odd depending on the sign of n,0 = 1,
n,pn i1 (x) = n,i

(x) = n,0 n,i (x).
(32)
Thus a biorthogonal basis of eigen- and associated functions

is fully compatible
+exists which
with PT-symmetry. As a consequence of (32) the integrals n,0 (x)n,p

(x) dx remain
n 1
real.
We remark that in general case the existence and the completeness of a biorthogonal system is not obvious (especially if the continuum spectrum is present [32]) and needs a careful
examination.
115
3. The origin of non-diagonalizable Hamiltonians is level confluence

Let us demonstrate that one can generate a Jordan cell of a Hamiltonian in the process of coalescing levels of the Hamiltonian. In the simplest case one can consider a Hamiltonian depending
on the parameter ,
h = 2 + V (x; )
with two eigenfunctions 1,2 (x; ),
h 1,2 (x; ) = 1,2 ()1,2 (x; ).
Assume that the levels 1 () and 2 () coalesce for = 0 :
1 (x; 0 ) 2 (x; 0 ) = 0 (x),
1 (0 ) = 2 (0 ) = 0 .
1
(x; )
2
and
Let us also suppose that the functions
(x; ) are normalizable. Then, it is
evident that

1,2
V
=
1,2 () 1,2 ,
h 1,2 ()

1
2
(h0 0 )
(x; 0 )
(x; 0 ) = 1 (0 ) 2 (0 ) 0 (x)
and, thus, the functions

0 (x),
1 (x) =
1
2
(x; 0 ) (x; 0 )
1 (0 ) 2 (0 )
(33)
form a Jordan cell of the second order for the Hamiltonian h0 = h0 :

h0 0 = 0 0 ,
(h0 0 )1 = 0 .
Let us proceed now to the case with three coalescing levels of h :

h j () j (x; ) = 0, j = 1, 2, 3,
1 (x; 0 ) 2 (x; 0 ) 3 (x; 0 ) = 0 (x),
Let us introduce the auxiliary functions

(0)
1 (x; ) = 1 + ( 0 ) 1 (x; ),
(0)
j
(1)
j (x; ) =
(x; )
1 (0 ) = 2 (0 ) = 3 (0 ) = 0 .
(0)
j (x; ) = j (x; ),
j = 2, 3,
(0)
3
(x; )
,

j () 3 ()
j = 1, 2,
where the constant is chosen so that the associated functions of the first order
1
3
(x; 0 ) (x; 0 )
(1)
1 (x; 0 ) =
1 (0 ) 3 (0 )
and
2
3
(x; 0 ) (x; 0 )
(1)
2 (x; 0 ) =
2 (0 ) 3 (0 )
0 (x)
1 (0 ) 3 (0 )
116
(cf. with (33)) are identical.3 When using these auxiliary functions one can obtain the canonical
set of associated functions,
(1)
0 (x),
(1)
(1)
1 (x) = 1 (x; 0 ) 2 (x; 0 ),
2 (x) =
(1)
2
(x; 0 )
,
2[ 1 (0 ) 2 (0 )]
(x; 0 )
which form Jordan cell of the third order for the Hamiltonian h0 = h0 :
h0 0 = 0 0 ,
(h0 0 )j = j 1 ,
j = 1, 2.
Thus, we have shown that the confluence of two (three) levels (of algebraic multiplicity 1) leads
to appearance of a Jordan cell of the second (third) order. The described construction is illustrated by an example in Section 7.2. In the case of confluence of a larger number n of levels (of
algebraic multiplicity 1) one can construct the canonical chain of n normalizable eigenfunction
and associated functions in the same way.
One can examine also the confluence of levels of different algebraic multiplicity. Let us restrict
ourselves to the simplest case, when for the Hamiltonian h level 1 () of algebraic multiplicity 2 coalesces for = 0 with level 2 () of algebraic multiplicity 1:

h 10 = 1 ()10 ,
h 1 () 11 = 10 ,
h 2 = 2 ()2 ,
10 (x; 0 ) 2 (x; 0 ) = 0 (x),
1 (0 ) = 2 (0 ) = 0 .
Again we introduce the auxiliary functions

1 (x; ) = 10 (x; ),

(0)
2 (x; ) = 1 + ( 0 ) 2 (x; ),
1(1) (x; ) = 11 (x; ),
2(1) (x; ) =
(0)
(0)
(0)
2
(x; )
,

1 () 2 ()
(x; )
(1)
where one should choose so that the associated functions of the first order 1 (x; 0 )
11 (x; 0 ) and
(1)
2 (x; 0 )
10
2
(x; 0 ) (x; 0 )
1 (0 ) 2 (0 )
0 (x),
1 (0 ) 2 (0 )
are identical. Therefrom one can get the appropriate set of associated functions
0 (x),
(1)
(1)
2 (x) =
(1)
1 (x) = 1 (x; 0 ) 2 (x; 0 ),

2
(1)
1
(x; 0 )
,

2[2 (0 ) 1 (0 )]
(x; 0 ) 2
(34)
which form Jordan cell of the third order for the Hamiltonian h0 = h0 :
h0 0 = 0 0 ,
(h0 0 )j = j 1 ,
j = 1, 2.
3 The constant exists because in one-dimensional case the difference of normalizable associated functions of the first
order for the same eigenfunction is proportional to this eigenfunction. This freedom has been discussed in the previous
section, see Eq. (23).
117
4. Non-linear SUSY for complex potentials

Supersymmetric Quantum Mechanics (SUSY QM) in one dimension represents a concise way
for an almost isospectral transformation between two quantum systems [7078] (see the reviews
[7984]). Conventionally it can be built for a pair of Hamiltonians h+ and h assembled into a
super-Hamiltonian,
+

2
h
+ V1 (x)
0
0
H=
(35)
2 I + V(x),
=
0 h
0
2 + V2 (x)
where the potential V(x) is, in general, complex. The (almost) isospectral connection between
h+ and h is realized by the intertwining relations,
h+ q + = q + h ,
q h+ = h q ,
with q being components of the supercharges,

0 0
0 q+
=
,
Q
,
Q=
q 0
0 0
(36)
2 = 0.
Q2 = Q
(37)
The isospectral relations (36) result in the conservation of supercharges or the supersymmetry of
the super-Hamiltonian,
= 0.
[H, Q] = [H, Q]
(38)
In general, its algebraic closure is given, by a non-linear (deformed) SUSY algebra,

= P(H ),
{Q, Q}
(39)
where P(H ) is a function of the super-Hamiltonian [34,83,85].

The relevant supercharges are supposed to be generated by N th order differential operators
with smooth coefficient functions wk (x):
=
q qN
N
wk (x) k ,
wN
= const (1)N .
(40)
k=0
We focus our analysis on the non-Hermitian Hamiltonians interrelated by complex supercharges,

i.e. the supercharges with complex and smooth coefficient functions wk (x). In this paper we
+ t
connected to qN
by means of t transposition, qN
= (qN
) . Evidently the superchoose qN
t
Hamiltonian of Schrdinger type is self-transposed ( -symmetric) as only the scalar potentials
are under consideration.
The algebraic structure of a non-linear SUSY for local Hamiltonians is exhaustively determined by the following theorem,
Theorem 1. (On SUSY algebras with transposition symmetry.) Let us introduce two sets of N
linearly independent functions n (x) (n = 1, . . . , N) which represent complete sets of zeromodes of the supercharge components,
+ t

qN
(41)
n = 0,
qN
= qN
.
Then:
118
(1) the Hamiltonians h have finite matrix representations when acting on the set of functions
n (x),

Snm
m ;
h n =
(42)
m
= Qt takes the polynomial form,

(2) the SUSY algebra closure with Q

Q, Qt = det EI S+ E=H = det EI S E=H PN (H ).
(43)
The proof in [85] is based on the quasi-diagonalization of matrices S , i.e. on their reduction
to the Jordan canonical form S which is block-diagonal. Such a diagonalization can be realized
by non-degenerate linear transformations of the zero-mode sets which induce the similarity
transformations of matrices S ,
l =
N

lm
m ,
m=1
S = S

1
h l =
N

m ,
Slm
m=1
(44)
Evidently the canonical bases of zero-modes of intertwining operators qN

form the set of (in
general, formalnot necessarily normalizable) solutions and associated functions of the Hamiltonians h . These elements of canonical bases of intertwining operator kernels are named as
transformation functions. The proof in [85] can be easily generalized to the complex case with
transposition symmetry being built up along the same scheme.
For the polynomial SUSY algebra there is a possibility that the intertwining operators may be
trivially reduced by a factor depending on the Hamiltonian, without any changes in the Hamiltonians h , namely,

= P h p M
= pM
P h ,
qN
(45)
where P (x) is assumed to be a polynomial and N M + 2. Thus some of the roots of associated
polynomials may not be involved in determination of the structure of the potentials.
This problem of disentangling the non-trivial part of a supercharge and avoiding multiple
SUSY algebras generated by means of dressing can be systematically tackled with the help of
the following theorem which can be also extended for complex potentials.
Theorem 2 (Strip-off). Let us assume the construction of the theorem on SUSY algebras with
transposition symmetry. Then the requirement that the matrix S (or S + ) generated on the sub+
(or qN
) contains m pairs (and no more) of Jordan cells
space of zero-modes of the operator qN
with equal eigenvalues l in each pair and the sizes kl and kl + kl (kl being the size of a
smallest cell in the lth pair) is necessary and sufficient to ensure that the intertwining operator
+
(or qN
) can be factorized:
qN
qN
= pM
m

l h
l=1
kl
(46)
where pM
are intertwining operators of order M = N 2 m
l=1 kl =
j =1 kj which cannot
be decomposed further on in the product (46) type. Herein kj for m + 1 j n are sizes of
unpaired Jordan cells.
119
Remark 1. The matrices S cannot contain more than two Jordan cells with the same eigenvalue because otherwise the operator h would have more than two linearly independent
zero-mode solutions.
Remark 2. Theorem 2 together with Theorem 1 entails the essential identity of the Jordan forms
S and S + (up to transposition of certain Jordan cells).
Remark 3. The supercharge components cannot be stripped-off if the polynomial PN (x) does
not have degenerate zeroes. The latter is sufficient to deal with SUSY charges non-trivially factorizable, but not necessary because degenerate zeroes may well arise in the ladder (dressing
chain) construction giving new pairs of isospectral potentials.
Remark 4. In general, a super-Hamiltonian may commute with several different supercharges.
In this case few hidden-symmetry operators arise. The optimization of such a system of supercharges till one or two independent ones, their stripping-off and the minimal structure of a
symmetry operator has been investigated in [85] in details for Hermitian Hamiltonians. In so far
as in the quoted paper the transposition was used as a main conjugation operation, all essential
results of [85] remain to be valid also for complex potentials.
The intertwining operators (supercharge components) can be formally factorized into the
products of elementary Darboux operators. Let j j , j = 1, . . . , N , be4 the basis, in which
the S+ -matrix (see Theorem 1) has a canonical Jordan form and j is an eigenvalue of S+
corresponding to the Jordan cell, to which j belongs. Then adapting Lemma 1 from [85] for
non-Hermitian Hamiltonians one can prove the following statements:
(1) for the supercharge components the factorization holds, in particular,
= rN r1 ,
qN
(47)
where the Darboux operators

rj = + j (x),
j = 1, . . . , N,
can be chosen so that

rj r1 Nj +1 = 0,
j = 1, . . . , N ;
(48)
(2) the chain relations take place

t
t
rl+1 rl+1 + l+1 = rl rl + l hl , j = 1, . . . , N 1,
t
t
r1 r1 + 1 = h+ h0 ,
rN rN + N = h hN ;
(49)
(3) the intermediate Hamiltonians hl , l = 1, . . . , N 1, have Schrdinger form:

hl = 2 + vl (x),
2

vl (x) = l+1
(x) l+1
(x) + l+1 = l2 (x) + l (x) + l ,
V1 (x) v0 (x) = 12 (x) 1 (x) + 1 ,
V2 (x) vN (x) = N2 (x) + N (x) + N ,

(50)
4 In what follows, in the notations for , we omit their relation to a Hamiltonian h when it is only one of these sets
j
which is used, in order to avoid too heavy indices.
120
but, in general, with complex and/or singular potentials;

(4) the intertwining relations are valid:
t t
hl rl+1
= rl+1 hl+1 , rl+1
hl = hl+1 rl+1
, l = 0, . . . , N 1.
(51)
Let us introduce the generalized Crum determinants made of solutions of the initial
Schrdinger equation for the Hamiltonian h+ as well as of some of its formal associated functions,

(j 1)
(x)
N (x)
N
N (x)

(j 1)

N1 (x)
N 1 (x)
N1 (x)

, j = 1, . . . , N.
wj (x) =
(52)

..
..
..
..

.
.
.
.

(j 1)

Nj +1 (x) N
j +1 (x) Nj +1 (x)
Then in virtue of Eq. (48) one finds the representation for the intertwining operators,

(j )
(x)
N (x)
N
N (x)

..
..
..
..

1
.
.
.
.
, j = 1, . . . , N,
rj r1 =

(j
1)
wj (x) Nj +1 (x)
(x) Nj +1 (x)
N
j
+1

1
j
(53)
and, consequently,
rj r1 Nj =
wj +1 (x)
,
wj (x)
j = 1, . . . , N 1.
(54)
Hence the intermediate superpotentials are uniquely determined by the chosen basis of solutions
and formal associated functions of a given Hamiltonian h+ for a given ordering,
j (x) =
wj (x) wj 1 (x)
[wj (x)/wj 1 (x)]
=
+
,
wj (x)/wj 1 (x)
wj (x) wj 1 (x)
j = 1, . . . , N,
w0 (x) 1.
Thus, from Eq. (50) one obtains the chain relations between intermediate potentials,

wj (x)

vj (x) vj 1 (x) = 2j (x) = 2 ln
, j = 1, . . . , N;
wj 1 (x)
(55)
(56)
and furthermore

vj (x) v0 (x) = 2 ln wj (x) ,
j = 0, . . . , N.
(57)
It leads finally to the connection between the components of the potential in the superHamiltonian,

V2 (x) = V1 (x) 2 ln wN (x) .
(58)
The above set of relations is well known from the Crum theory [86]. However here we have extended them including not only solutions of a Schrdinger equation but also its formal associated
functions.
121
Remark 5. Let us comment the sufficient conditions to preserve PT-symmetry under SUSY
transformations, i.e. after intertwining with a DarbouxCrum operator. They consist in requirement for all transformation functions (all elements of the zero-mode basis of the intertwining operator) to be symmetric or antisymmetric in respect to PT-reflection. Indeed the PT(anti)symmetry of basis elements entails the PT-(anti)symmetry of their Wronskian wherefrom
one derives the PT-antisymmetry of W /W and the PT-symmetry of V2 = V1 2(ln W ) . This is
why under such conditions the above presented theorems and the forthcoming ones are certainly
valid.
5. Non-diagonalizable Hamiltonians and normalizability of associated functions
In this section we start examination of how the Darboux transformation may change the structure of a non-diagonalizable Hamiltonian with a Jordan cell spanned by a set of associated
functions. We keep in mind that an intertwining operator may annihilate part of them as zero
modes. As a result we find the additive composition of Jordan cells for partner Hamiltonians
h mediated by a Jordan cell for the Hamiltonian mapping of the zero-mode subspace of the
.
intertwining operator qN
The rigorous results can be obtained with a specification of the class of potentials invariant
under Darboux transformations. For such a class the asymptotic normalizability of associated
functions at one of the infinities is preserved by Darboux transformations and will be described
in certain lemmas and corollaries. When the normalizability on the whole axis is achieved we call
the related associated functions as normalizable in general. The detailed mathematical proofs are
given in part II, here we give only the general ideas of the construction as well as the formulations
of the theorems and lemmas and the related corollaries.
Let us investigate the Jordan structure of the Hamiltonians h . In what follows we restrict
ourselves to the particular class of potentials:
Definition 2. Let K be the set of all potentials V (x) such that:
(1) V (x) CR ;
(2) there are R0 > 0 and > 0 (R0 and depend on V (x)) such that for any |x| R0 the
inequality Re V (x) takes place;
(3) Im V (x)/ Re V (x) = o(1), x (this is sufficient to ensure the reality of the continuous
spectrum);
(4) functions
2
x

|V (x)|2 |V (x)|
V (x1 ) dx1
+
(59)
|V (x)|3
|V (x)|2
R0
are bounded respectively for x R0 and x R0 .

Remark 6. The last condition is not very rigid: it is fulfilled (if x +), for example, for
potentials:
(1)

V (x) = ax 1 + o(1) ,

V (x) = a x 1 1 + o(1) ,

V (x) = a ( 1)x 2 + o x 2 , a > 0, > 0;
122
(2)

V (x) = V0 + ax 1 + o(1) , V (x) = a x 1 1 + o(1) ,

V (x) = a ( + 1)x 2 1 + o(1) , V0 > 0, a C, Re > 0;
(3)

V (x) = ax ebx 1 + o(1) ,

V (x) = abx +1 ebx 1 + o(1) ,

V (x) = ab2 2 x +22 ebx 1 + o(1) , a > 0, b > 0, R, > 0;
(4)

V (x) = V0 + ax ebx 1 + o(1) , V (x) = abx +1 ebx 1 + o(1) ,

V (x) = ab2 2 x +22 ebx 1 + o(1) , V0 > 0, a C, Re b > 0, C, > 0.
A similar statement is valid for x . Thus one can find in this class of potentials both
representatives with purely discrete spectrum and the Hamiltonians with continuum spectrum.
Remark 7. This class of potentials only partially overlaps with PT-symmetric set of potentials investigated recently [917,23,24]. We call the related class of Hamiltonians as softly
non-Hermitian as they certainly do not involve complex coordinates in definition of asymptotic
boundary conditions.
The set K is closed under intertwining of Hamiltonians, that follows from the
Lemma 1. (On invariance of the potential set K.) Let:
(1)
(2)
(3)
(4)
h+ = 2 + V1 (x), V1 (x) K;
h = 2 + V2 (x), V2 (x) CR ;
+
h = h qN
, where qN
is differential operator of N th order with coefficients from CR2 ;
qN
+
each eigenvalue of S -matrix of qN
(see Theorem 1) satisfies one of the conditions: either
0 or Im = 0.
Then:
(1) V2 (x) K;
belong to CR ;
(2) coefficients of qN
+
+
+
t
+
(3) h+ qN = qN h , where qN
= (qN
) , and moreover coefficients of qN
belong to CR also.
Let us now analyze the normalizability properties of associated functions.
Definition 3. A function f (x) is called normalizable at + (at ), if there is R+ (R ) such
that
R

+

f (x)2 dx < +
f (x)2 dx < + .
(60)
R+
Otherwise f (x) is called non-normalizable at + (at ).
123
Using the asymptotics of formal associated functions (see Lemma 9, part II, one can show
that:
(1) when the potential in h (by which, in what follows, we imply either h+ or h ) belongs to the
class K, one can show that any formal associated function of nth order of h, normalizable
at + or respectively, for the spectral value , satisfying either 0 or Im = 0, can
be decomposed as follows
n
aj, j, (x),
aj, = const, an, = 0,
(61)
j =0
where j, (x), j 0, stand for either j, (x) or j, (x) and they form a sequence of
associated functions normalizable at + or respectively,
h0, = 0, ,
(h )j, = j 1, ,
j 1.
(62)
Correspondingly, any associated function of nth order of h, non-normalizable at the same

+ or , for the same spectral value can be presented as follows
n

bj, j, (x) + cj, j, (x) ,
(63)
j =0
where bj, , cj, = const, either bn, = 0, or cn, = 0 and j, (x), j 0, form a sequence of non-normalizable at + or respectively associated functions
h0, = 0, ,
(h )j, = j 1, ,
j 1;
(64)
(2) for the Hamiltonian with a potential from the class K, there are no degenerate eigenvalues, satisfying either 0 or Im = 0, i.e. the eigenvalues, whose geometric multiplicity
exceeds 1 (the eigenvalues, for which there are more than one linearly independent eigenfunctions). Hence, for the Hamiltonian with a potential from K there is no more than one
Jordan cell made of an eigenfunction and associated functions, normalizable on the whole
axis, for any given eigenvalue such that either 0 or Im = 0.
Properties of associated functions under intertwining are described by the
Lemma 2. Let:
(1) the conditions of Lemma 1 take place;
(2) n (x), n = 0, . . . , M, be a sequence of associated functions of h+ for the spectral value :
+

h+ 0 = 0 ,
h n = n1 , n 1,
where either 0 or Im = 0. Then:
(1) there is a number m such that 0 m min{M + 1, N },
qN
n 0,
n < m,
and
m+l ,
l = qN
l = 0, . . . , M m,
124
is a sequence of associated functions of h for the spectral value :

h l = l1 , l 1;
h 0 = 0 ,
(2) if the function n (x), for a given 0 n M, is normalizable at + (on ), then the
function qN
n is normalizable at + (on ) as well.
Corollary 1. Since h+ is an intertwining operator for itself and both eigenvalues of its S+ matrix are zero, then if n (x) is normalizable at + (at ), then j (x), j = 0, . . . , n 1, are
normalizable at + (at ) as well.
Corollary 2. If there is an associated normalizable function n (x) of nth order of the Hamiltonian h with a potential from K, for an eigenvalue , which is either 0 or Im = 0, then
for this eigenvalue there is an associated function j (x) of Hamiltonian h, normalizable on the
whole axis, of any smaller order j :
j = (h )nj n ,
j = 0, . . . , n 1.
Corollary 3. Let i,j

(x) be a canonical basis of zero-modes of the intertwining operator qN
, i.e.
+
such that S -matrix (from Theorems 1 and 2) has in this basis the canonical (Jordan) form:

+
= i i,0
,
h i i,j
= i,j
h+ i,0
1 , j = 1, . . . , ki 1,
+
+
+
where ki is a rank of a Jordan cell for i . Then there are numbers ki
and ki
, 0 ki,
ki ,
related to the Hamiltonian h+ such that for any i the functions
(x),
i,j
+
j = 0, . . . , ki,
1,
are normalizable at + or respectively and the functions
(x),
i,j
+
j = ki,
, . . . , ki 1,
are non-normalizable at the same + or . Thus one can derive that the number of func
+ +
(x) normalizable on the whole axis is equivalent to min{ki
, ki } and the number of
tions i,j
+ +
, ki }.
functions i,j (x) non-normalizable at both ends is given by ki max{ki
+
on a choice of the canonical basis, in the case of
Independence of these numbers ki,
stripped-off qN , follows from
Lemma 3. Let:
(1) conditions of Lemma 1 take place;
(2) qN
may not be stripped-off.
Then any two formal associated functions of h+ of the same order for the same spectral value ,
when being zero-modes of qN

, are either simultaneously normalizable at + or simultaneously
non-normalizable at +. The same takes place at .
We refer the reader for the proofs of Lemmas 13 to the forthcoming part II.
125
6. Interrelation between Jordan cells in SUSY partners

The first result on interrelation between Jordan structures of intertwined Hamiltonians and on
the behavior of transformation functions at follows from the
Lemma 4. Let us assume that:
(1)
(2)
(3)
(4)
conditions of Lemma 1 take place;
+
{i,j
} and {i,j
} are canonical bases of ker qN
and ker qN
respectively;
qN cannot be stripped-off ;
ki is an algebraic multiplicity of the eigenvalue i of S+ -matrix (see Theorems 1 and 2).
Then for any i and j function i,j

(x) is normalizable (non-normalizable) at + if and only if
+
i,ki j 1 (x) is non-normalizable (normalizable) at +. The same takes place at .
Corollary 4. In order that for the level i the Hamiltonian h+ does not have eigenfunctions and
associated functions normalizable on the whole axis and the Hamiltonian h has a Jordan cell
of multiplicity (i ) spanned by eigenfunction and associated functions normalizable on whole
axis (the same number (i ) measures the dimension of the subspace of non-normalizable zeromodes of q ) it is necessary and sufficient that

+ +

ki = max ki
, ki + (i ) = ki
ki
+ (i ),
(ki
) are numbers of functions i,j
(x) normalizable at + () (see the previous
where ki
section). Thus if there are no eigenfunctions and associated functions corresponding to the level
lambda i of the initial Hamiltonian h+ , normalizable on the whole axis and one wants to get
the final Hamiltonian h with a Jordan cell of rank (i ) spanned by an eigenfunction and
associated functions normalizable on the whole axis, one must choose transformation functions
such that they contain (i ) (and no more) associated functions of h+ non-normalizable at both
infinities.
A more precise result on interrelation between Jordan structures of intertwined Hamiltonians

and on the behavior of transformation functions is given in the
Theorem 3 (Index Theorem). (On relation between Jordan structures of intertwined Hamiltonians.) Let us assume that:
(2) () is the algebraic multiplicity of an eigenvalue of h , i.e. the number of independent
eigenfunctions and associated functions of h normalizable on the whole axis;
(3) if = i , where i is an eigenvalue of S (see Theorem 1), then n (i ) is a number of
normalizable functions at both infinities among ij (x), j = 0, . . . , ki 1, and n0 (i ) is a
number of functions normalizable only at one of infinities, among ij (x), j = 0, . . . , ki 1.
Then the equality
+ (i ) n+ (i ) = (i ) n (i )
126
takes place for any i. Moreover if n0 () > 0 for some = j , then for this j
+ (j ) n+ (j ) = (j ) n (j ) = 0;
and if is not an eigenvalue of S but 0 or Im = 0, then + () = ().
7. An example of non-diagonalizable Hamiltonians made by SUSY transformations
7.1. SUSY system with Jordan cell of rank 2
Let us start from the Darboux transformation of the free particle Hamiltonian (which is trivially PT-symmetric),
h+ = 2
(65)
and build an isospectral Hamiltonian (its SUSY partner) which is reflectionless [80,85,87] due
to spectral equivalence to a free particle system,
(x z) sh(2(x x0 )) 2 ch2 ((x x0 ))
,
[sh(2(x x0 )) + 2(x z)]2
> 0, x0 R, z C, Im z = 0,
h = 2 16 2
(66)
with the help of the intertwining operators q2 :

W (x)
1 W (x)
q2 = 2
2 +
,
q2 h = h q2 ,
W (x)
2 W (x)

W (x) = sh 2(x x0 ) + 2(x z).
t
q2+ = q2 ,
(67)
If x0 = Re z = 0 the Hamiltonian h reveals PT-symmetry. Otherwise PT-symmetry is not realized although the energy spectrum remains real.
The operator q2 can be factorized into two intertwining operators of first order in derivatives,

q2 = qb qa ,
qa = th (x x0 ) ,

W (x)
+ th (x x0 ) ,
qb =
(68)
W (x)
with the intermediate non-singular Hamiltonian of the ladder construction of Section 4,
h1 = 2
2 2
ch2 ((x x0 ))
(69)
The canonical basis of q2 consists of two non-normalizable functions:

0 (x) = ch (x x0 ) ,

(x z)
1
1 (x) =
sh (x x0 ) + 2 ch (x x0 ) ,
2
4

+
0
+
h 0 1 = 0 , 0 = 2 ,
S+ =
h 0 = 0 0 ,
1
(70)
0
0

.
(71)
On the other hand, the canonical basis of q2+ consists of two normalizable functions:
0+ (x) = (2)3/2
0 (x)
,
W (x)
1+ (x) = (2)3/2
1 (x)
,
W (x)
(72)
which form the Jordan cell for h corresponding to the level 0 :

0 0
h 0 1+ = 0+ ,
S =
h 0+ = 0 0+ ,
.
1 0
In relation to factorization (68) one can show that the zero-mode of qb becomes:

1
sh((x x0 ))
(x z)
W (x)

.
qa 1 =
2
2 ch((x x0 ))
4 ch((x x0 ))
2 0+
127
(73)
(74)
In turn the eigenfunctions of h for continuous spectrum read:

1
(x; k) =
q2 eikx
2
2
2 ( + k )

ik W (x)
1
1
W (x) ikx
1+ 2
=
e ,
+ k 2 W (x)
2( 2 + k 2 ) W (x)
2
k R, h (x; k) = k 2 (x; k).
(75)
One can check that eigenfunctions and associated functions of h obey the relations:
+

2
+
0,1
(x) dx
+
0+ (x)1+ (x) dx = 1,
= 0,
+
+
0,1
(x)(x; k) dx
= 0,
+
(x; k)(x; k ) dx = (k k ),
(76)
where the last relation is understood, as usual, in the sense of distributions.

One can also find that the functions 0+ (x), 1+ (x) can be obtained by analytical continuation
of (x; k) in k,

2

x0 +
2
lim k + (x; k) =
0 (x),
e
ki

x0 +
1 2z +
1 2
2
0 (x) .
k + (x; k) =
e
1 (x)
lim
ki 2k k
4 2
For this model resolution of identity made of eigenfunctions and associated functions of h
can be obtained by the conventional Green function method:
+
(x; k)(x ; k) dk + 0+ (x)1+ (x ) + 1+ (x)0+ (x ),
(x x ) =

(77)
or in the operator form,

I=
+
k| dk + |0 0 | + |1 1 |,
|, k,
(78)
where the Dirac notations have been used:

x|, k = (x; k),
k = (x; k),
x|,
(79)
128
+

x| 0,1 = 1,0
(x) ,

h 0 | 0 = | 1 ,
+
x|0,1 = 0,1
(x),
h | 1 = 0 | 1 ,
(80)
k = k 2 |,
k.
h |,
(81)
We stress that | 0,1 are analogs of | n,a,i from Section 2. In this notations the biorthogonal
relations take the form,
j |k = j k ,
k|0,1 = 1,0 |, k = 0,
,
k|, k = (k k ).
,
(82)
Accordingly, the spectral decomposition of h can be easily derived,

+
k| dk 2 |0 0 | 2 |1 1 | + |0 1 |.
k 2 |, k,
h =
(83)
are the
Let us remind that, for a given = i , ki is the number of zero-modes of q2 , ki,
numbers of zero-modes normalizable at one of labeled by , , () are the numbers of
eigenfunctions and associated functions of h normalizable on the whole axis, n () are the
numbers of eigenfunctions and associated functions among zero-modes i,j

normalizable on the
whole axis and n0 () is the number of eigenfunctions and associated functions among zero
modes i,j
normalizable only at one end. They are defined in Sections 5, 6. For this model they
take the following particular values (i = 0):

2, = 0 ,
+
k0 = 2,
k0, = 0,
+ () 0,
k0, = 2,
() =
0, = 0 ,

2, = 0 ,
n0 () 0,
n () =
n+ () 0,
0, = 0 ,
and the Index Theorem holds,

+ (j ) n+ (j ) = (j ) n (j ) = 0.
7.2. Coalescence of two levels
The Hamiltonian h with Jordan cell for bound state (66) is a particular limiting case of
the Hamiltonian h with two non-degenerate bound states (of algebraic multiplicity 1). The
former corresponds to the confluent case of the latter one. One again starts from the free particle
Hamiltonian,
h+ = 2
and obtains its SUSY partner
2
2

2
2 +
h = 16
sh 2(x x0 ) sh 2(x z)
2

2
2
2 ch (x x0 ) ch 2(x z) + 2 sh (x z)

2
sh 2(x x0 ) + sh 2(x z)
129
by intertwining with the operators

q2 = 2
1 W (x) + t

W (x)
2 + 2 +
= q2 ,
W (x)
2 W (x)
where

W (x) = sh 2(x x0 ) + sh 2(x z) ,
x0 R, Im z = 0, > 0 (or i > 0), 0 <
.
2 Im z
In the case = 0, = the canonical basis of q2 zero-modes (transformation functions) consists of

x0 z
,
(x) = ch k (x ) , k = , =
so that
h+ = ,
2
= k
= ( )2 .
One can check that in the case = 0 the function W (x) is a Wronskian of + (x) and (x)
divided by and in the case = 0 it is a product of 4 and of the Wronskian of 0 (x) and
1 (x).
For = 0 both functions (x) coincide with 0 (x) from (70), and as well = 0 . In
this case 1 (x) is a linear combination of + (x) and (x) in the following sense:

1
(+ )
1 =
+ 2 0 .

4
(+ ) =0
In the case = 0 the canonical basis of zero-modes of q2+ consists of
(x) =
(x)
.
W (x)
One can prove that

+
2
(x) dx =

2( )
(84)
(this formula is valid also in the case = 0). The fact that the integrals (84) vanish for = 0 is
in line with (76). It follows from (84) that the normalized eigenfunctions of h have the form

1
1
1 (x)
1 + (x)
+ (x) = 2i
+
,
(x) = 2
,
W (x)
W (x)
h = .
The eigenfunctions of h for continuous spectrum read,

(x)
1 W (x) ikx
[ 2 + 2 + k 2 + ik W
W (x) 2 W (x) ]e
,
(x; k) =

2 (k 2 + 2 + 2 )2 4 2 2
k R,
h (x; k) = k 2 (x; k),

(85)
130

where the branch of (k 2 + 2 + 2 )2 4 2 2 can be defined by the condition

2

k 2 + 2 + 2 4 2 2 = k 2 + o k 2 , k ,
in the plane with cuts, linking branch points, situated in the upper (lower) half-plane.
One can prove the following limits,

2

2i 1
1
2
2
2
2
2
lim
k + +
4 (x; k) =
+ e(x0 +z) + (x),
ki(+)

2

1
2
1
k 2 + 2 + 2 4 2 2 (x; k) =
e(x0 z) (x).
lim
ki()

In the limiting case = 0 the eigenfunction 0+ (x) and the associated function 1+ (x) of h (see
Eqs. (72)) can be derived from (x),

0+ (x) = 2i lim + (x) = 2 lim (x) ,

0
[ ( (x) + i+ (x))]
+
1 (x) = 2 lim
.
0
(+ )
Resolution of identity in the case = 0 takes the form:

(x x ) = + (x)+ (x ) + (x) (x ) +
+
(x; k)(x; k) dk
(86)
and one can show that in the case > 0

lim + (x)+ (x ) + (x) (x ) = 0+ (x)1+ (x ) + 1+ (x)0+ (x )
0
(cf. with (77)).

One can also check that the biorthogonal relations take place:
+
+ (x) (x) dx = 0,
+
+
(x)(x; k) dx = 0,
(x; k)(x; k ) dx = (k k ).
(87)
7.3. Symmetry operators

For the Hamiltonian h of Sections 1 and 2 there exists the antisymmetric symmetry operator
of the fifth order
R5 = q2 q2+ ,
R5 h = h R5 ,
R5t = R5 .
In the case of PT-symmetry, x0 = Re z = 0 it anticommutes with PT-reflection P T , R5 P T =

P T R5 .
131
The wave function 0+ (x), the associated function 1+ (x) and the wave function of zero-energy
bound state

1 W (x)
1
1 2
(x; 0) =
2 W (x)
2
+
are zero-modes of R5 because of 0,1
(x) are zero-modes of q2+ and q2+ (x; 0) = const. In
Section 2 (in the case = 0) the wave functions + (x), (x) and the wave function of
zero-energy bound state

(x)
]
[ 2 + 2 12 WW (x)
(x; 0) =
2( 2 2 )
are zero-modes of R5 because of (x) are zero-modes of q2+ and q2+ (x; 0) = const. In both
cases the spectrum of S-matrix of the symmetry operator (an analogue of S+ from Theorem 1)
consists of 0 and ( )2 .
We notice that if k = 0 then (x; k) and (x; k) are linearly independent wave functions for
the energy level E = k 2 . Simultaneously, (x; k) is an eigenfunction of the symmetry operator:
2

R5 (x; k) = ik k 2 + 2 (x; k),
or
2

R5 (x; k) = ik k 2 + 2 + 2 4 2 2 (x; k).
Thus the zeroes of the eigenvalue of R5 are related to eigenvalues of S-matrix of R5 in accordance
to [85].
8. Conclusions and perspectives: Peculiarities of non-Hermitian Hamiltonians with
continuous spectrum
In our paper we have investigated a bound state part of the spectrum for the class of potentials
among which one can find also those ones with continuum spectrum. The choice of this class
has allowed to keep all the analysis well below a possible continuum threshold. For this part
of the spectrum the relationship between Jordan cells of SUSY partner Hamiltonians is firmly
controlled by the Index Theorem (Theorem 3) and it was well illuminated by an exactly solvable
system with two coalescing bound states. We remind that the rigorous proofs of all new lemmas
and Theorem 3 are postponed to the second part of this paper [65].
Meantime the approaching to the continuum threshold yields more subtle problems with normalizable eigen- and associated functions in continuum which may have zero binorm. As a
consequence it may cause serious problems with the resolution of identity. This interesting problem we investigated in [32]. Here we would like only to draw attention to a class of models where
the continuous spectrum is involved and elaborate the resolution of identity.
Let us consider the model Hamiltonians
2
h = 2 +
, Im z = 0,
h+ = 2 ,
(88)
(x z)2
intertwined by the first-order operators q1 :
h q1 = q1 h ,
q1 =
1
.
x z
(89)
132
If Re z = 0 these Hamiltonians are PT-symmetric and respectively their eigenfunctions possess

definite PT-parities.
The eigenfunctions of h of continuous spectrum can be found in the form,

1
1
(x; k) =
(90)
eikx , k R\{0}, h (x; k) = k 2 (x; k).
1
ik(x z)
2
In addition, there is a normalizable eigenfunction of h on the lower end of continuous spectrum:
0 (x) =

1
= 2 lim ik(x; k) ,
k0
(x z)
h 0 = 0.
Isospectral relations between h+ and h take the form,

ikx

1
1
e
q
= ik(x; k), k = 0,
q
= 0 (x),
2
2
2
ikx
e
, k = 0,
q + 0 = 0, k = 0.
q + (x; k) = ik
2
(91)
k = 0;
(92)
(93)
The eigenfunctions of h satisfy the relations of biorthogonality,

+
02 (x) dx = 0,
+
0 (x)(x; k) dx = 0.
(94)
Resolution of identity made of eigenfunctions of h can be built as follows,

(x x ) = (x; k)(x ; k) dk,
(95)
where the contour L must be a proper integration path in the complex k plane which allows to
regularize the singularity in (90) for k = 0 circumventing it from up or from down. To reach an
adequate definition of resolution of identity one can instead use the NewtonLeibnitz formula
and rewrite (95) in the form

(x x ) =
+
+
(x; k)(x ; k) dk
2 sin2 ( 2 (x x ))
0 (x)0 (x ) sin (x x )
+
+
,
(x x )
(x z)(x z)
> 0.
(96)
One can show [32] that the limit of the 3rd term of the right side of (96) (as a distribution) under
0 is zero for any test function from CR L2 (R) but the limit of the last term of the right side
of (96) under 0 is zero only for test functions from CR L2 (R; |x| ), > 1. Thus for test
functions from CR L2 (R; |x| ), > 1 resolution of identity can be reduced to,
(x x ) = lim
0

+
+
0 (x)0 (x )
(x; k)(x ; k) dk

(97)
133
and for test functions from CR L2 (R) to,

(x x ) = lim
0
+
(x; k)(x ; k) dk

1
2

1 2 sin
(x x ) 0 (x)0 (x ) .
2

(98)
Decomposition (97) seems to have a more natural form than (98), but its right side obviously
cannot reproduce the normalizable eigenfunction

/ CR L2 R; |x| , > 1,
0 (x)
because of the orthogonality relations (94). One can show that
+
lim

2

sin2 (x x ) 02 (x)0 (x ) dx = lim eix 0 (x ) = 0 (x ).
0
(99)
Thus it is the 3rd term in the right side of (98) that provides the opportunity to reproduce 0 (x)
and thereby to complete the resolution of identity.
The spectral decomposition of h in this case reads
+
k| dk,
h =
k 2 |, k,
(100)
where
k = (x; k),
x|,
x|, k = (x; k),
k = k 2 |,
k.
h |,
For the Hamiltonian h of this model there is an antisymmetric symmetry operator of the 3rd
order
3
3
,
R3 = q1 q1+ = 3 +
(x z)2
(x z)3
R3 h = h R3 ,
R3t = R3 ,
R3 (x; k) = ik 3 (x; k).
We notice that this model is a limiting case of the example of Section 7.1 for 0 where z
must be taken as a half sum of x0 and z.
One can generalize this model constructing the Hamiltonian (by intertwining with the Hamiltonian of a free particle)
h = 2 +
n(n + 1)
,
(x z)2
for which there is a Jordan cell, spanned by [ n+1

2 ] normalizable eigenfunction and associated
functions:

n1
h0 = 0,
hj = j 1 , j = 0, . . . ,
,
2
(2(n j ) 1)!!
,
j (x) =
(2j )!!(2n 1)!!(x z)n2j
134
at the bottom of continuous spectrum. All these functions are mutually biorthogonal having also
zero binorm. The problem of a correct resolution of identity seems to be solvable in the same
way as for n = 1. However the rigorous analysis is postponed to a forthcoming paper.
Acknowledgements
The work of A.A. and A.S. was supported by Grant RFBR 06-01-00186-a and by Programs
RNP 2.1.1.1112 and LSS-5538.2006.2. A. Sokolov was partially supported by the INFN grant.
References
[1] H. Feshbach, Theoretical Nuclear Theory: Nuclear Reactions, Wiley, New York, 1992;
P.E. Hodgson, The Nucleon Optical Potential, World Scientific, Singapore, 1995.
[2] C. Itzykson, J.-M. Drouffe, Statistical Field Theory, vol. 1, Cambridge Univ. Press, Cambridge, 1989.
[3] N. Moiseyev, Phys. Rep. 302 (1998) 211.
[4] J.G. Muga, J.P. Palao, B. Navarro, I.L. Egusquiza, Phys. Rep. 395 (2004) 357.
[5] N. Dunford, Bull. Am. Math. Soc. 64 (1958) 217.
[6] B.S. Pavlov, Itogi Nauki i Tekhniki, Sovrem. Probl. Mat. Fund. Naprav. 65 (1991) 95;
B.S. Pavlov, Partial Differential Equations, vol. VIII, Encyclopaedia of Mathematical Sciences, vol. 65, Springer,
Berlin, 1996.
[7] E.B. Davies, Bull. London Math. Soc. 34 (2002) 513, and references therein.
[8] S.L. Yakovlev, Theor. Math. Phys. 102 (1995) 323;
S.L. Yakovlev, Theor. Math. Phys. 107 (1996) 513.
[9] C.M. Bender, K.A. Milton, Phys. Rev. D 55 (1997) R3255.
[10] C.M. Bender, S. Boettcher, Phys. Rev. Lett. 80 (1998) 5243;
C.M. Bender, S. Boettcher, P. Meisinger, J. Math. Phys. 40 (1999) 2201.
[11] C.M. Bender, D.C. Brody, H.F. Jones, Phys. Rev. Lett. 89 (2002) 270401;
C.M. Bender, D.C. Brody, H.F. Jones, Phys. Rev. Lett. 92 (2004) 119902, Erratum;
C.M. Bender, D.C. Brody, H.F. Jones, Am. J. Phys. 71 (2003) 1095;
C. Bender, J. Brod, A. Refig, M. Reuter, J. Phys. A: Math. Gen. 37 (2004) 10139, and references therein;
C. Bender, quant-ph/0501052;
C.M. Bender, J.-H. Chen, K.A. Milton, J. Phys. A 39 (2006) 1657.
[12] Z. Ahmed, C.M. Bender, M.V. Berry, quant-ph/0508117.
[13] M. Znojil, F. Cannata, B. Bagchi, R. Roychoudhury, Phys. Lett. B 483 (2000) 284.
[14] G. Lvai, F. Cannata, A. Ventura, Phys. Lett. A 300 (2002) 271.
[15] A. Mostafazadeh, J. Math. Phys. 43 (2002) 205;
A. Mostafazadeh, J. Math. Phys. 43 (2002) 6343;
A. Mostafazadeh, J. Math. Phys. 44 (2003) 943, Erratum.
[16] A. Mostafazadeh, Nucl. Phys. B 640 (2002) 419;
A. Mostafazadeh, J. Phys. A: Math. Gen. 36 (2003) 7081.
[17] G. Lvai, F. Cannata, A. Ventura, J. Phys. A: Math. Gen. 34 (2001) 839;
G. Lvai, F. Cannata, A. Ventura, J. Phys. A: Math. Gen. 35 (2002) 5041;
R.N. Deb, F. Khare, B.D. Roy, Phys. Lett. A 307 (2003) 215;
A. Mostafazadeh, J. Math. Phys. 46 (2005) 102.
[18] F. Cannata, G. Junker, J. Trost, Phys. Lett. A 246 (1998) 219.
[19] A.A. Andrianov, F. Cannata, J.-P. Dedonder, M.V. Ioffe, Int. J. Mod. Phys. A 14 (1999) 2675.
[20] T. Tanaka, J. Phys. A: Math. Gen. 39 (2006) 7757;
T. Tanaka, J. Phys. A: Math. Gen. 39 (2006) L369;
A. Mostafazadeh, quant-ph/0606173.
[21] A.A. Andrianov, Ann. Phys. 140 (1982) 82.
[22] V. Buslaev, V. Grecchi, J. Phys. A: Math. Gen. 26 (1993) 5541.
[23] P. Dorey, C. Dunning, R. Tateo, J. Phys. A: Math. Gen. 34 (2001) L391.
[24] H.F. Jones, J. Mateo, Phys. Rev. D 73 (2006) 085002;
C.M. Bender, D.C. Brody, J.-H. Chen, H.F. Jones, K.A. Milton, M.C. Ogilvie, Phys. Rev. D 74 (2006) 025016;
[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]
H.F. Jones, J. Mateo, R.J. Rivers, Phys. Rev. D 74 (2006) 125022;

H.F. Jones, R.J. Rivers, Phys. Rev. D 75 (2007) 025023.
T. Curtright, L. Mezincescu, quant-ph/0507015, and references therein.
E.A. Solovev, Usp. Fiz. Nauk 157 (1989) 228, and references therein.
M.V. Berry, J. Phys. A: Math. Gen. 31 (1998) 3493.
E. Narevicius, P. Serra, N. Moiseyev, Europhys. Lett. 62 (2003) 789.
M. Znojil, Phys. Lett. A 259 (1999) 220;
M. Znojil, Phys. Lett. A 264 (1999) 108;
A. Sinha, G. Lvai, P. Royc, Phys. Lett. A 322 (2004) 78.
B.F. Samsonov, P. Roy, J. Phys. A: Math. Gen. 38 (2005) L249.
A. Bohm, M. Loewe, S. Maxson, P. Patuleanu, C. Pntmann, J. Math. Phys. 38 (1997) 6072;
I.E. Antoniou, M. Gadella, G.P. Pronko, J. Math. Phys. 39 (1998) 2459;
E. Hernndez, A. Juregui, A. Mondragn, Phys. Rev. A 67 (2003) 022721;
O. Civitarese, M. Gadella, Phys. Rep. 396 (2004) 41.
A.V. Sokolov, A.A. Andrianov, F. Cannata, J. Phys. A: Math. Gen. 39 (2006) 10207.
A.A. Andrianov, M.V. Ioffe, V.P. Spiridonov, Phys. Lett. A 174 (1993) 273.
A.A. Andrianov, F. Cannata, J.-P. Dedonder, M.V. Ioffe, Int. J. Mod. Phys. A 10 (1995) 2683.
V.G. Bagrov, B.F. Samsonov, Theor. Math. Phys. 104 (1995) 1051.
B.F. Samsonov, Mod. Phys. Lett. A 11 (1996) 1563.
A. Gangopadhyaya, U. Sukhatme, Phys. Lett. A 224 (1996) 5.
U. Sukhatme, C. Rasinariu, A. Khare, Phys. Lett. A 234 (1997) 401.
A. Das, S.A. Pernice, Mod. Phys. Lett. A 12 (1997) 581.
D.J. Fernndez C., Int. J. Mod. Phys. A 12 (1997) 171.
G. Junker, P. Roy, Ann. Phys. 270 (1998) 155.
D.J. Fernndez C., V. Hussin, B. Mielnik, Phys. Lett. A 244 (1998) 309.
J.O. Rosas-Ortiz, J. Phys. A: Math. Gen. 31 (1998) 10163.
B. Bagchi, A. Ganguly, D. Bhaumik, A. Mitra, Mod. Phys. Lett. A 14 (1999) 27.
D.J. Fernndez C., J. Negro, L.M. Nieto, Phys. Lett. A 275 (2000) 338.
D.J. Fernndez C., V. Hussin, quant-ph/0011004.
M.S. Plyushchay, Int. J. Mod. Phys. A 15 (2000) 3679;
M.S. Plyushchay, Phys. Lett. B 485 (2000) 187.
H. Aoyama, M. Sato, T. Tanaka, M. Yamamoto, Phys. Lett. B 498 (2001) 117.
H. Aoyama, M. Sato, T. Tanaka, Phys. Lett. B 503 (2001) 423.
H. Aoyama, M. Sato, T. Tanaka, Nucl. Phys. B 619 (2001) 105.
H. Aoyama, N. Nakayama, M. Sato, T. Tanaka, Phys. Lett. B 519 (2001) 260;
H. Aoyama, N. Nakayama, M. Sato, T. Tanaka, Phys. Lett. B 521 (2001) 400.
T. Tanaka, Nucl. Phys. B 662 (2003) 413.
S.M. Klishevich, M.S. Plyushchay, Nucl. Phys. B 628 (2002) 217.
D.J. Fernndez C., R. Muoz, A. Ramos, quant-ph/0212026.
A.P. Veselov, A.B. Shabat, Funct. Anal. Appl. 27 (1993) 81.
V.E. Adler, Funct. Anal. Appl. 27 (1993) 140.
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, Phys. Lett. A 201 (1995) 103;
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, Theor. Math. Phys. 104 (1995) 1129;
A.A. Andrianov, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. 32 (1999) 4641.
A.A. Andrianov, F. Cannata, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. 30 (1997) 5037.
F. Cannata, M.V. Ioffe, D.N. Nishnianidze, J. Phys. A: Math. Gen. 35 (2002) 1389.
H. Aoyama, H. Kikuchi, I. Okouchi, M. Sato, S. Wada, Nucl. Phys. B 553 (1999) 644.
B. Bagchi, F. Cannata, C. Quesne, Phys. Lett. A 269 (2000) 79.
R. Sasaki, K. Takasaki, J. Phys. A: Math. Gen. 34 (2001) 9533.
S.M. Klishevich, M.S. Plyushchay, Nucl. Phys. B 606 (2001) 583;
S.M. Klishevich, M.S. Plyushchay, Nucl. Phys. B 616 (2001) 403.
P. Dorey, C. Dunning, R. Tateo, J. Phys. A: Math. Gen. 34 (2001) 5679.
A.V. Sokolov, math-ph/0610022.
N. Moiseyev, S. Friedland, Phys. Rev. A 22 (1980) 618.
S. Weigert, Czech. J. Phys. 55 (2005) 1183, and references therein.
M.A. Naimark, Linear Differential Operators, Frederick Ungar Publishing Co., New York, 1967.
135
136
[69]
[70]
[71]
[72]
[73]
F.R. Gantmacher, The Theory of Matrices, AMS Chelsea Publishing, Providence, RI, 1998.
H. Nicolai, J. Phys. A: Math. Gen. 9 (1976) 1497.
E. Witten, Nucl. Phys. B 188 (1981) 513.
F. Cooper, B. Freedman, Ann. Phys. (N.Y.) 146 (1983) 262.
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, JETP Lett. 39 (1984) 93;
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, Phys. Lett. A 105 (1984) 19;
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, Theor. Math. Phys. 61 (1985) 1078.
M.M. Nieto, Phys. Lett. B 145 (1984) 208.
B. Mielnik, J. Math. Phys. 25 (1984) 3387.
D. Fernndez, Lett. Math. Phys. 8 (1984) 337.
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, M.I. Eides, Phys. Lett. A 109 (1985) 143;
A.A. Andrianov, N.V. Borisov, M.V. Ioffe, M.I. Eides, Theor. Math. Phys. 61 (1985) 965.
C.V. Sukumar, J. Phys. A: Math. Gen. 18 (1985) L57;
C.V. Sukumar, J. Phys. A: Math. Gen. 18 (1985) 2917;
C.V. Sukumar, J. Phys. A: Math. Gen. 18 (1985) 2937.
L.E. Gendenshtein, I.V. Krive, Sov. Phys. Usp. 28 (1985) 645.
A. Lahiri, P.K. Roy, B. Bagchi, Int. J. Mod. Phys. A 5 (1990) 1383.
F. Cooper, A. Khare, U. Sukhatme, Phys. Rep. 251 (1995) 267.
G. Junker, Supersymmetric Methods in Quantum and Statistical Physics, Springer, BerlinHeidelberg, 1996.
A.A. Andrianov, F. Cannata, J. Phys. A: Math. Gen. 37 (2004) 10297.
B. Mielnik, O. Rosas-Ortiz, J. Phys. A: Math. Gen. 37 (2004) 10007.
A.A. Andrianov, A.V. Sokolov, Nucl. Phys. B 660 (2003) 25.
V.B. Matveev, M.A. Salle, Darboux Transformations and Solitons, Springer, BerlinHeidelberg, 1991.
S.P. Maydanyuk, Ann. Phys. (N.Y.) 316 (2) (2005) 440;
S.P. Maydanyuk, hep-th/0607125.
[74]
[75]
[76]
[77]
[78]
[79]
[80]
[81]
[82]
[83]
[84]
[85]
[86]
[87]
Non-linear supersymmetry for non-Hermitian,

non-diagonalizable Hamiltonians: II. Rigorous results
A.V. Sokolov
V.A. Fock Institute of Physics, Sankt-Petersburg State University, Russia
Received 11 October 2006; received in revised form 18 February 2007; accepted 14 March 2007
Abstract
We continue our investigation of the non-linear SUSY for complex potentials started in part I [A.A. Andrianov, F. Cannata, A.V. Sokolov, math-ph/0610024] and prove the theorems characterizing its structure in
the case of non-diagonalizable Hamiltonians. This part provides the mathematical basis of previous studies.
The classes of potentials invariant under SUSY transformations for non-diagonalizable Hamiltonians are
specified and the asymptotics of formal eigenfunctions and associated functions are derived. Several results
on the normalizability of associated functions at infinities are rigorously proved. Finally the Index Theorem
on relation between Jordan structures of intertwined Hamiltonians depending of the behavior of elements
of canonical basis of supercharge kernel at infinity is proven.
1. Introduction: Definitions and notation

In this part of the paper we continue the investigation of the non-linear SUSY [16] (see
the extended list of references in [7]) for complex potentials started in part I [7] and prove the
theorems characterizing its structure in the case of non-diagonalizable Hamiltonians. We use the
class of potentials invariant under SUSY transformations for non-diagonalizable Hamiltonians
and prove several results concerning the normalizability of associated functions at + or .
These results allow to unravel the relation between Jordan cells in SUSY partner Hamiltonians
which was described in the Index Theorem in Section 6 of [7].
DOI of companion paper (part I): 10.1016/j.nuclphysb.2007.03.016.

E-mail address: avs_avs@rambler.ru.
doi:10.1016/j.nuclphysb.2007.03.015
138
A.V. Sokolov / Nuclear Physics B 773 [PM] (2007) 137171
All the proofs and results of this part are safely applicable to PT symmetric non-Hermitian
Hamiltonians with soft type of non-Hermiticity when the real part of a potential dominates over
its imaginary one at both coordinate infinities. The latter property is embedded into the chosen
classes of potentials.
Let us summarize this part of the paper aimed to derive Theorem 3 and Lemmas 14 discussed
in part I [7]. First we introduce the relevant classes of potentials K (main) and K (auxiliary) as
well as we remind the notion of formal associated functions of a Hamiltonian. Next we provide
necessary estimates for potentials belonging to the class K (Lemma 5) and for auxiliary integrals
(Lemmas 6, 7). Furthermore, we derive the asymptotics of formal eigenfunctions (Lemma 8) and
associated functions (Lemma 9) of Hamiltonians with potentials belonging to the class K. Then
the invariance of the classes K and K under intertwining is proved (Lemmas 10 and 1 respectively). And finally the proofs of Lemma 2 (on properties of a sequence of formal associated
functions under intertwining), of Lemma 3 (on normalizability of elements of the canonical
basis of an intertwining operator), of Lemma 4 (on interrelation in (non-)normalizability of
canonical bases of mutually transposed intertwining operators) and of Theorem 3 (on relation
between Jordan structures of intertwined Hamiltonians depending on the asymptotic behavior
of elements of the canonical basis of an intertwining operator kernel at ) will be presented.
In the paper we use the following classes of potentials.
Definition 1. (Definition 2 of Ref. [7].) Let K be the set of all potentials V (x) such that:
(1) V (x) CR ;
(2) there are R0 > 0 and > 0 (R0 and depend on V (x)) such that for any |x| R0 the
inequality Re V (x) takes place;
(3)
Im V (x)/ Re V (x) = o(1),
x ;
(4) functions
x
2

|V (x)|2 |V (x)|
V (x1 ) dx1
+
|V (x)|3
|V (x)|2
(1)
(2)
R0

Definition 2. Let K be the set of all potentials V (x) such that:
(1) V (x) is a complex-valued (in particular, real-valued) function, defined on the real axis with
possible exception at some points;
2
,
(2) there are R0 > 0 and > 0 (R0 and depend on V (x)) such that V (x)|[R0 ,+[ C[R
0 ,+[
2
V (x)|],R0 ] C],R0 ] and for any |x| R0 the following inequality holds:
Re V (x) ;
(3)
(3)
Im V (x)/ Re V (x) = o(1),
x ;
(4)
139
(4) functions
2
x

|V (x)|2 |V (x)|
V (x1 ) dx1
+
|V (x)|3
|V (x)|2
R0

Let us clarify that K is a main class of potentialsthe class of physical potentials, and K is an
auxiliary, wider class of potentialsthe class, containing potentials of intermediate Hamiltonians
(corresponding to factorization of an intertwining operator in the product of intertwining operators of first order in derivative) in the case when potentials of the initial and final Hamiltonians
belong to K.
In what follows we shall use the functions of the form (V (x) ) , where V (x) K, > 0,
C and either 0 or Im = 0. Branches of these functions will be identically selected by
the condition

arg V (x) < .
(5)
In the case 0 this condition can be fulfilled in view of (3), and in the case Im = 0 because
of (4) the condition (5) can be satisfied for any |x| > R2 , where R2 R0 is such that for any
|x| R2 the inequality
| Im V (x)| 1 | Im |

,
| Re V (x)| 2 | Re |
|x| R2 ,
(6)
holds. For R0 < x < R0 (if 0) or for R2 < x < R2 (if Im = 0) the functions of the form
(V (x) ) will not be used.
The notation is adopted,
(x; ) =
1
(V (x))2
V (x)
5
,
16 (V (x) )5/2 4 (V (x) )3/2
5
1 |V (x)|
|V (x)|2
+
,
5/2
16 |V (x) |
4 |V (x) |3/2

x
R0 ,
(x; ) =
V (x1 ) dx1 , R1 =
R2 ,
(x;
) =
R1
(x) = (x; 0),

(x)
= (x;
0),
0,
Im = 0,
x

V (x1 ) dx1 ,
(x) =
R0

I1, (x; ) =
(x
1 ; ) dx1 ,
I1, (x) = I1, (x; 0),

I2, (x; ) =
(x
1 ; )e2 Re( (x1 ;) (x;)) dx1 ,
x
140
x
I3, (x; ) =
(x
1 ; )e2 Re( (x;) (x1 ;)) dx1 ,
R1
C = max
|x|R0
| Im V (x)|
.
| Re V (x)|
The notion of a formal associated function, used in this paper, is defined as follows.
Definition 3. (Definition 1 of Ref. [7].) The function n,i (x) is called a formal associated function of ith order of the Hamiltonian h for a spectral value n , if
(h n )i+1 n,i 0,
(h n )i n,i 0,
(7)
where the adjective formal emphasizes that a related function is not necessarily normalizable.
In particular, the associated function of zero order n,0 is a formal eigenfunction of h (a solution of the homogeneous Schrdinger equation, not necessarily normalizable).
In this paper we employ the normalizability of functions and in particular the normalizability
at + (at ), which is defined as follows.
Definition 4. (Definition 3 of Ref. [7].) A function f (x) is called normalizable at + (at ),
if there is R+ (R ) such that
+

f (x)2 dx < +
R+
R

f (x)2 dx < + .
(8)
Otherwise f (x) is called non-normalizable at + (at ).

2. Estimates on potentials and asymptotics of useful integrals
Lemma 5. If V (x) K, C and either 0 or Im = 0, then there are constants
C1 , C2 , C3 > 0 such that for any |x| R1 the inequalities are valid:
|V (x)|2
C2 ,
|V (x) |2

Re V (x) C3 V (x).
C1
(9)
(10)
Proof. Let us first consider the case 0. Then the right side of (9) is obvious. The left side of
(9) follows from the chain

|V |2 (1 / Re V )2 + Im2 V / Re2 V
2
+ C2.
1
|V |2
1 + Im2 V / Re2 V
The inequality (10) is derived from the chain

2 1 + Im2 V / Re2 V
|V |
2 1 + C2

.
1+1
(Re V )2 1 / Re V + (1 / Re V )2 + Im2 V / Re2 V
141
Let us now consider the case Im = 0. In this case the left side of (9) is provided by the chain
|V |2 (1 Re / Re V )2 + (Im V / Re V Im / Re V )2
|V |2
1 + Im2 V / Re2 V
2

| Im | 2
| Re |
+ C+
.
1+
The right side of (9) in the subcase Re V 2| Re | follows from the sequence of inequalities
1 + Im2 V / Re2 V
|V |2
2
|V |
(1 Re / Re V )2 + (Im V / Re V Im / Re V )2

1 + C2

4 1 + C2 ,
(1 | Re |/ Re V )2
and in the subcase Re V 2| Re | from the sequence1
|V |2
1 + Im2 V / Re2 V
2
|V |
1 + C2

(1 | Re |/ Re V )2 + (| Im |/ Re V | Im |/(2| Re |))2
1 + C2
4||2 (1 + C 2 )

.
2
2
||2 | Re |
Im2
1 Re 2 )2 + Im 2
2 (
Re
Re V
2||
4||
One can obtain (10) with the help of the inequality

|V |
(Re V )2

2 1 + Im2 V / Re2 V

2 1 + C2

. (11)
1 | Re |/ Re V + (1 | Re |/ Re V )2 + (| Im V |/ Re V | Im |/ Re V )2
In the subcase Re V 2| Re | the right side of (11) is less than or equal to 2 1 + C 2 , wherefrom
(10) follows, and in the subcase Re V 2| Re | the right side (11) is less than or equal to
2 1 + C2

1 | Re |/ Re V + (1 | Re |/ Re V )2 + (| Im |/ Re V | Im |/(2 Re ))2

2 1 + C 2 1 | Re |/ Re V

1

+ ||/| Re |
| Re |/ Re V 1/2 Re2 / 2||2 + Re2 Im2 / 4||4
.
1 Re / Re V +
(12)
Insofar as the function
||
f (y) = 1 y +
| Re |

y
1 Re2
2 2||2
1 For the derivation of this chain the inequality (6) is used.
2
+
Re2 Im2
(4||4 )
142
has a minimum at the point y = 1/2 + Re2 /||2 , (12) is less than or equal to
2 1 + C2
2||2

1 + C2.
2
1 Re2 /||2 + (||/| Re |) Re4 /(4||4 ) + Re2 Im2 /(4||4 ) Im
Thus, Lemma 5 is proved.
Corollary 1. If V (x) K, C and either 0 or Im = 0, then

|V (x)|
1
=
O
, x .
(x)
|V (x) |3/2
Lemma 6. If V (x) K, C and either 0 or Im = 0, then for any |x| R1 the integrals
I1, (x; ) and I2, (x; ) converge and the estimates hold:

1
I1, (x; ) = O
, x ,
(x)

1
I2, (x; ) = O 2
, x ,
(x)

1
, x .
I3, (x; ) = O 2
(x)
Proof. Because the proofs for the cases x + and x are similar, we shall consider
the case x + only. Due to V K and Lemma 5 there are positive constants C4 , . . . , C9 and
0 such that
+
I1, (x; ) C4
x
C6
|V |2
dx1 + C5
|V |5/2
+
x
dx1
2
+
x
|V |
dx1 C6
|V |3/2
+
|V |
dx1
2
x
C6
,
(x)
+ (x )

1 2 x1 Re V (x2 ) dx2
e x
dx1
I2, (x; ) C7
2
(x1 )
x
C7
+ (x )
1
x
C7
2 (x1 )
+ (x )
1
x
2 (x1 )
C7 /C8
,
= 2
(x)
eC8
x
1
x
|V (x2 )| dx2
dx1
eC8 ( (x) (x1 )) dx1
C7 C8 (x)
e
2 (x)
+
(x1 )eC8 (x1 ) dx1
x
x
I3, (x; ) C9
R0
(x1 )
(0 + (x1
))2
C8 ( (x) (x1 ))
dx1 = C9 e
C8 (x)
143
(x)
eC8 d
(0 + )2
(x)/2 (x)
eC8 d
C8 (x)
= C9 e
+
(0 + )2
0
(x)/2

1
C8 (x) C8 (x)/2 1
= C9 e
e
0 0 + (x)/2

1
1 C8 (x)
C8 (x)/2
+
e
e
(0 + (x)/2)2 C8

1 C8 (x)/2 4/C8
C9
,
e
+ 2
0
(x)
2
wherefrom Lemma 6 follows.

Lemma 7. Let:
(1) V (x) K;
(2) C and either 0 or Im = 0;
(3) the integral
R

+
1
dx1
dx1
V (x1 )
V (x1 )
(13)
R1
converges.
Then the integral
+
R0
dx1
|V (x1 )|
R
0
dx1
|V (x1 )|

(14)
converges too,
lim V (x) =
(15)
and
1
=O
V (x)
1
(x)

x

dx1
,
|V (x1 )|
x .
(16)
Proof. As the integral (13) converges, then the integral

+
+
Re V (x1 )
1
Re
dx1
dx1
|V (x1 ) |
V (x1 )
R1
R1
R
1
Re V (x1 )
dx1
|V (x1 ) |
144
converges
too. In view of Lemma 5 there are constants C1 and C3 such that Re V (x)
C3 |V (x)| and |V (x) | 1C |V (x)| for any |x| R1 . Hence the integral
1
+
R1
1
dx1
|V (x1 )|
C1 C3
R
1
+
R1
1
dx1
|V (x1 )|
C1 C3
Re V (x1 )
dx1
|V (x1 ) |
R
1
Re V (x1 )
dx1
|V (x1 ) |
converges as much as the integral (14).

Let us now check (15) in the case x + (examination of the case x is similar).
The integral
+
R1
V (x1 )
dx1
(V (x1 ) )2
converges owing to convergence of (14) and boundedness of |V |/|V |3/2 for x R1 (see
Corollary 1). Hence the limit of the function
1
1
=
V (x) V (R1 )
x
R1
V (x1 )
dx1 ,
(V (x1 ) )2
x R1 ,
for x + is finite. Moreover, because of convergence of (13) this limit is zero. Thus (15)
holds.
Validity of (16) for x + (consideration of the case x is similar) is justified by
the fact that for V K there is C4 > 0 such that |V |/|V |3/2 C4 / for any x R0 and by the
chain
+

+
+

1
C4
V (x1 )
dx1
dx1

dx1 C4

.
2
=
2

|V (x)|
V (x1 )
|V (x1 )| (x1 ) (x)
|V (x1 )|
x
3. Asymptotics of formal eigenfunctions of a Hamiltonian

Asymptotic behavior of formal eigenfunctions of a Hamiltonian with potential belonging to K
is described by the
Lemma 8. Let:
(1) V (x) K;
(2) C and either 0 or Im = 0.
Then there are functions 0, (x) normalizable at being zero-modes of h and functions
0, (x) non-normalizable at being zero-modes of h such that2
2 Leading terms of asymptotics (17) and (19) are well known (see for example [8]).
145

1
1
1
(x1 ; ) dx1 + O 2
e (x;) 1
,
0, (x) =
4
2
V (x)
(x)
x ,
(17)

1
1
V (x)
+O 2
(18)
= V (x) 1
, x ,
0, (x)
4 (V (x) )3/2
(x)

1
1
1
(x;)
, x ,
(x1 ; ) dx1 + O 2
e
1+
0, (x) =
4
2
V (x)
(x)

(x)
0,

1
1
V (x)
,
+O 2
= V (x) 1
0, (x)
4 (V (x) )3/2
(x)

0,
(x)
(19)
x .
(20)
Proof. We shall consider the case x + only because examination of the case x is
analogous. Let us show that the series
+
+

1
dx1 sh (x; ) (x1 ; ) (x1 ; )
0, (x) =
4
V (x) n=0
x
+

dx2 sh (x1 ; ) (x2 ; ) (x2 ; )
x1
+

dxn sh (xn1 ; ) (xn ; ) (xn ; )e (xn ;)
(21)
xn1
converges and gives the required function 0, (x). Convergence of (21) is provided by the fact
that the series (21) is majorized by the series
+
+
+

Re( (x1 ;) (x;))
dx1 e
(x
1 ; )
dx2 eRe( (x2 ;) (x1 ;)) (x
2 ; )
n=0 x
x1
+
dxn eRe( (xn ;) (xn1 ;)) (x

n ; ) eRe (xn ;)
xn1
Re (x;)
+
+
+
+

dx1 (x
1 ; )
dx2 (x
2 ; )
dxn (x
n ; )
n=0 x
x1
xn1
+
n
+

1
Re (x;)
=e
(x
1 ; ) dx1 eRe (x;)+I1, (x;) .
n!
n=0
From this estimate it follows also that due to Lemma 6 the asymptotics (17) for the function (21)
is valid. Insofar as the series of first and second derivatives of (21) are majorized for x belonging
146
to any segment [x1 , x2 ] [R1 , +[ by the series independent of x,

+

n=0

1
n
max (x; ) I1, (R1 ; )
n!
[x1 ,x2 ]
and
+

n=0

2 1

n
I1, (R1 ; )
max (x; ) + (x; )
[x1 ,x2 ]
n!

+ max (x; ) (x; )
[x1 ,x2 ]
n1
1
I1, (R1 ; )
,
(n 1)!
it is possible to differentiate twice the series (21) term by term. Calculation of h applied to
the function (21) allows to check that this function is a zero-mode of h . One can check also
(18), with the help of (21), using Lemma 6, Corollary 1 and the fact that the absolute value of
the derivative of nth term in (21) is less than or equal to

n
1
V (x) I1, (x; ) eRe (x;) .
n!
To prove normalizability of 0, (x) at + it is sufficient to prove normalizability at + of the
leading term of the asymptotics (17). The latter comes out of the fact that for V (x) K the chain
of inequalities holds due to Lemma 5,
4
4
e2Re (x;)
C2 2C3 Rx |V (x1 )| dx1
C2 2C3 Rx dx1
1
1

e
e
|V (x) |
|V (x)|
C2
e2C3 (xR1 ) ,
the right side of which is obviously normalizable at +.

Let us prove now that the required function 0, (x) can be written in the form
x
0, (x) = 20, (x)
R3
dx1
,
2 (x )
0,
1
(22)
where R3 R1 is a constant such that 0, (x) has no zeroes for x R3 (existence of R3 is

obvious because of (17)). The fact that the function (22) is a zero-mode of h follows from
elementary calculations. To prove that the asymptotics (19) and (20) for the function (22) are
valid it is sufficient to prove that
x
R3

+
dx1
1
1 2 (x;)
(x1 ; ) dx1 + O 2
= e
1+
2 (x )
2
0,
(x)
1
x
in view of (17), (18), (22) and the obvious formula

(x)
0,
0, (x)
(x)
0,
0, (x)

dx1
2 (x) x
0,
R3 2 (x1 )
0,
147
By virtue of (17)

+

1
1
2 (x;)
= V (x) e
.
(x1 ; ) dx1 + O 2
1+
2 (x)
0,
(x)
(23)
Because
of Lemma 6 the contribution of first and second terms of the right side of (23) at
x
2
dx
/
1
0, (x1 ) is given by
R3

x
+
2 (x1 ;)
V (x1 ) e
(x2 ; ) dx2 dx1
1+
R3
1
=
2
x
2 (x1 ;)
R3
x1
(x2 ; ) dx2 dx1
1+
x1
+
1 2 (x;)
=
e
1+
2
+
(x2 ; ) dx2 + O(1) +

x
1
= e2 (x;)
2
x
e
2 (x1 ;)
(x1 ; ) dx1
R3

+
(x1 ; ) dx1 + O
1+
x
1
2 (x)

,
x +.
2 (x) for x R the contribution of the third term of right side

Due to local boundedness of 1/0,
3
of (23) is less than or equal to the integral
x
|V (x1 ) | 2 Re (x1 ;)
C4
e
dx1
(0 + (x1 ))2
R3
C4 e
2 Re (x;)
x
|V (x1 ) | 2 xx Re V (x2 ) dx2
1
e
dx1 ,
(0 + (x1 ))2
(24)
R3
where C4 and 0 are positive constants. For some positive constants C5 and C6 the integral (24),
in view of Lemma 5, is less than or equal to the integral
x
C5 e
2 Re (x;)
R0
|V (x1 )|
eC6 ( (x) (x1 )) dx1
(0 + (x1 ))2
x
C5 e
2 Re (x;)
R0
(x1 )
(0 + (x1 ))2
eC6 ( (x) (x1 )) dx1 ,
which is equal (see the proof of Lemma 6) to O(e2 Re (x;) /2 (x)), x +. Thus, (19) and
(20) hold.
Finally in order to prove non-normalizability at + of function (22) let us first prove the
auxiliary inequality

V (x) C0 0 + (x) , x R0 ,
(25)
148
where C0 and are some positive constants. This equality for V (x) K follows from the sequence

x
V (x) = V (R0 )eln(V (x)/V (R0 )) V (R0 )e R0

x
V (R0 )e R0
(x )
1
0 + (x1 ) dx1
|V (x1 )|
|V (x1 )| dx1

x
V (R0 )e R0
|V (x1 )|
0 + (x1 ) dx1

(0 + (x))
= V (R0 )
.
(0 + (R0 ))
To prove non-normalizability at + of function (22) it is sufficient to prove non-normalizability

at + of the leading term of asymptotics (19). The latter is provided by the fact that in view of
(25) and Lemma 5 the chain holds,
4
4
C1 (x1 ) 2C ( (x) (R ))
e2 Re (x;)
C1 2C3 Rx |V (x1 )| dx1
0
1

e
e 3
|V (x)|
|V (x) |
|V (x)|
4
(x)
C2

e2C3 ( (x) (R0 )) ,
C0 (0 + (x))
the right side of which is non-normalizable at +. Lemma 8 is proved.
Remark 1. The asymptotics (19) and (20) are valid for any zero-mode of h linear independent
of 0, (x) (after its proper normalization).
Corollary 2. Let Hamiltonians h = 2 + V1,2 (x) be intertwined by q1 = + (x):
q1 h = h q1 ,
h = q1 q1 + = + 2 .
Suppose also (x) to be a zero-mode of q1 so that = /. Then

2
2

V V2 V1 = 2 2(ln ) 2
2 V1
.

At last suppose that V1 (x) K, C and either 0 or Im = 0. Then because of (18), (20),
Lemma 5 and Corollary 1

V (x)
V1 (x)
V = 1
(26)
+O 2
, x +,
V1 (x)
(x)
if at x + the asymptotics (17) or respectively (19) is valid for . For the case when at
x the asymptotics (17) or respectively (19) is valid for ,

V1 (x)
V1 (x)
V =
(27)
+O 2
, x .
V1 (x)
(x)
Finally,

2

2 V1 + V
V 2 V1 2

3/2
V (x)
= O 12
, x ,
(x)
(28)

2

1

V 2 V1 + V
+ V 2 V1 V 2 + V
2

V (x)
= O 21
, x ,
(x)
149
(29)
independently of asymptotics of .
It also follows from (26), (27), Lemma 5 and Corollary 1 that

V1 (x)
= o(V1 ) = o(V1 ), x ,
V = O
(x)
i.e. that

V2 = V1 + o(V1 ) = (V1 ) 1 + o(1) ,

V2 = V1 + o(V1 ) = V1 1 + o(1) , x .
x ,
(30)
(31)
Corollary 3. There are no degenerate eigenvalues of the Hamiltonian with a potential belonging to K, satisfying either 0 or Im = 0, i.e. eigenvalues, whose geometric multiplicity is more than 1 (eigenvalues, for which there are more than one linearly independent eigenfunction). Hence, for the Hamiltonian with a potential belonging to K there
are no more than one Jordan cell made of an eigenfunction and associated functions,
normalizable on the whole axis, for any given eigenvalue such that either 0 or
Im = 0.
4. Asymptotics of formal associated functions of a Hamiltonian
The asymptotic behavior of formal associated functions of a Hamiltonian h with a potential
belonging to K is characterized by
Lemma 9. Let:
(1) h = 2 + V (x), V (x) K;
(2) C and either
x 0or Im = 0;
(3) (x) = R0 dx1 / |V (x1 )|.
Then there are denumerable sequences: n, (x) of formal associated functions of h for
a spectral value , normalizable at , and n, (x) of formal associated functions, nonnormalizable at , such that:
h0, = 0, ,
(h )n, = n1, ,
n 1,
(32)
(h )n, = n1, ,
h0, = 0, ,

if R0 dx1 / |V (x1 )| < +, then for x ,
n 1;
(33)

n

x
dx1
1
1
1
(x;)
e
,
n, (x) =
1
+
O
2
(x)
n! 4 V (x)
V (x1 )
(34)
150

n

x
1
dx1
1
1
(x;)
n, (x) =
e
,
1
+
O
2
(x)
n! 4 V (x)
V (x1 )

n,
(x) =
and if

R0
1
1
4
V (x)
n!
2
x
dx1
V (x1 )
n
e
(x;)
1
1+O
(x)
(35)

; (36)
dx1 / |V (x1 )| = +, then
n, (x)

n

x
ln (x)
dx1
1
1
(x;)
e
=
,
1+O
2
(x)
n! 4 V (x)
V (x1 )
(37)
R1
n, (x)

n

x
ln (x)
dx1
1
1
(x;)
e
,
=
1
+
O
2
(x)
n! 4 V (x)
V (x1 )
(38)
R1

(x)
n,
n

x
ln (x)
1
dx1
1
4
(x;)
=
V (x)
e
.
1+O
n!
2
(x)
V (x1 )
(39)
R1
Proof. Let us prove the existence of n, (x) and n, (x) only, because the proof of existence of
n, (x) and n, (x) is analogous. The existence of 0, (x) and 0, (x) was proved in Lemma 8
and in view of V (x) K the estimate O(1/ (x)) = O(ln (x)/ (x)), x + (cf. Lemmas 9 and 8) follows from the chain
x
(x) =
R0
1
dx1
|V (x1 )|
x

V (x1 ) dx1 = 1 (x).
(40)
R0
Suppose now the existence of l, (x) and l, (x) and let us prove the existence of l+1, (x)
and l+1, (x). In this way the lemma will be completely proved.
+
Consider the case R0 dx1 / |V (x1 )| < +. One can check that in this case l+1, (x) and
l+1, (x) can be written in the form

x
x
1
l+1, (x) = 0, (x)
0, (x1 )l, (x1 ) dx1 0, (x)
0, (x1 )l, (x1 ) dx1 ,
2

l+1, (x) =
x
x
1
0, (x)
2
0, (x1 )l, (x1 ) dx1 0, (x)
(41)

0, (x1 ) l, (x1 ) dx1 .
R1
(42)
151
x
Convergence of + 0, (x1 )l, (x1 ) dx1 follows from the fact that due to (34), (35) and
Lemma 5 there is constant C4 > 0 such that

x
+
l
+

dx1
dx2

0, (x1 )l, (x1 ) dx1 C4

|V (x1 )|
|V (x2 )|
+
C4
=
l+1
x
+
x
x1
dx2
|V (x2 )|
l+1
< +.
Convergence of + 0, (x1 )l, (x1 ) dx1 can be proved analogously. Convergence of the intex
gral + 0, (x1 )l, (x1 ) dx1 is obvious. Thus the right sides of (41) and (42) are well defined.
The fact that the right sides of (41) and (42) satisfy (32) and (33) for n = l + 1 can be checked
by direct application of h to these sides. One must take into account here that the Wronskian
(x)

0,
0, (x) 0, (x)0, (x) 2, that follows from the asymptotics of Lemma 8.
x Now we transform the integrals in (41) and (42). In view of (34) and (35) the integrand of
+ 0, (x1 )l, (x1 ) dx1 reads
x
l

dx1
1
1
, x +.
1+O
0, (x)l, (x) = l
(x)
2 l! V (x)
V (x1 )
+
(43)
The first term of the right side of (43) in the integral is equal to
x
x1
l
l+1
x
dx1
dx2
dx1
2
1
1
=
,
2l l!
(l + 1)! 2
V (x1 )
V (x2 )
V (x1 )
+
(44)
and the absolute value of contribution of the second term is less than or equal to
l
+
+
1
dx1
dx2
C5
,
|V (x1 ) | (x1 )
|V (x2 ) |
x
x1
for some constant C5 > 0. From Lemma 5 the latter expression can be estimated in the following
way:
+
l
+
C5
dx1
dx2
(x)
|V (x1 ) |
|V (x2 ) |
x
C5
=
(l + 1) (x)
C6
(x)
+
x
+
x
x1
dx1
|V (x1 ) |
+
l+1
l+1

V (x1 )
Re V (x1 )
C6

dx1
dx1

|V (x1 ) |
(x)
|V (x1 ) |
l+1
+

dx1
C6

=
,

(x)
V (x1 )
x
l+1
(45)
152
for some constant C6 > 0. Thus

x
0, (x1 )l, (x1 ) dx1
+
l+1
x

dx1
2
1
1
=
,
1+
(l + 1)! 2
(x)
V (x1 )
x +.
(46)
In view of (34), V (x) K, Lemmas 5 and 7 the following estimate for the integral
l, (x1 ) dx1 is valid,3
x
+ 0, (x1 )
x

0, (x1 )l, (x1 ) dx1

+
+
C7
x
C8 e
dx1
e2 Re (x1 ;)
|V (x1 ) |
2 Re (x;)
+
e2
x
x
C8 e
2 Re (x;)
+
x
= C8 e
+
Re V (x3 ) dx3
x1
eC9 x |V (x3 )| dx3
|V (x1 )|
+
x
eC9 (x1 )
|V (x1 )|
+
x1
+
x1
+
C9 (x)
C8 2 Re (x;)+C9 (x) e
=
e
C9
|V (x)|

+
+
eC9 (x1 )
+
|V (x1 )|
x
1
2 Re (x;)+C9 (x)
x1
dx2
|V (x2 ) |
l
|V (x1 )|3/2
+
x1
dx2
|V (x2 )|
dx2
|V (x2 )|
dx2
|V (x2 )|
dx2
|V (x2 )|
dx2
|V (x2 )|
Re V (x1 ) Re V (x1 ) + Im V (x1 ) Im V (x1 )
|V (x1 )|3
l
l
dx1
l
dx1
l
dx1
l
l1
+
x1
dx2
|V (x2 )|
l
+

C8 2 Re (x;)
dx2
1

e
C9
|V (x)|
|V (x2 )|
x
3 The latter equality in (47) is obtained with the help of the same trick as in (45).
l

dx1
+
+ C10 eC9 (x)
x

=O e
2 (x;)
1
V (x)
+O e
=O
2 (x;)
1
(x)
+
2 (x;)
(x)

=O
+
x
eC9 (x1 ) 1
|V (x1 )| (x1 )
+
2 (x;)
(x)
dx2
|V (x2 )|
+
x
+
x1
dx2
V (x2 )
l

dx1
l
dx2
|V (x2 )|
dx2
|V (x2 ) |
dx2
|V (x2 )|
153
l+1
l+1
l+1
(47)
for some positive constants C7 , . . . , C10 . The asymptotics (34) and (36) for n = l + 1 follow from
(41), (46), (47), from (34) and (35) with n = 0 and from (18), (20) and Corollary 1.
x
The integral + 0, (x1 ) l, (x1 ) dx1 can be calculated in the same way and the result is
x
0, (x1 )l, (x1 ) dx1
+
l+1

x
dx1
2
1
1
=
,
1+O
(l + 1)!
2
(x)
V (x1 )
x +.
(48)
One can also obtain the estimate,

x
R1
e2 (x;)
0, (x1 )l, (x1 ) dx1 = O
(x)
x
+
dx1
V (x1 )
l+1
,
x +.
(49)
Then in view of (34), (35) for n = 0 and (42), (48), (49) the asymptotics (35) turns out to be valid
x
for n = l + 1. For the integral R1 0, (x1 )l, (x1 ) dx1 the following estimate can be derived for
some positive constants C11 , . . . , C15 , 0 ,
x

0, (x1 )l, (x1 ) dx1

R1
x
C11
R1
e2 Re (x1 ;)
|V (x1 ) |
x
C12 e
2 Re (x;)
R0
+
x1
dx2
|V (x2 ) |
eC13 ( (x) (x1 ))
|V (x1 )|
l
+
x1
dx1
dx2
|V (x2 )|
l
dx1
154
l x
+

dx2
C12 2 Re (x;)C13 (x) eC13 (x1 )

=
e

C13
|V (x1 )|
|V (x2 )|

x
eC13 (x1 )
R0
l
|V (x1 )|3/2
+
x1
R0
x1
dx2
|V (x2 )|
Re V (x1 ) Re V (x1 ) + Im V (x1 ) Im V (x1 )
|V (x1 )|3

C12 2 Re (x;)
1
e
C13
|V (x)|
+ C14 eC13 (x)
x
R0
+
x
eC13 (x1 )
|V (x1 )|
C12 2 Re (x;)
C15
e

C13
0 + (x)
+ C14 eC13 (x)
x
R0
dx2
|V (x2 )|
+
x1
+
x
l1
+
x1
l
dx2
|V (x2 )|
dx2
|V (x2 )|
l
dx2
|V (x2 )|
dx1
0 + (x1 )
l

dx1
l+1
+
l

eC13 (x1 )
dx2
dx1
,
|V (x1 )|
|V (x2 )| 0 + (x1 )
(50)
x1
with the help of (35) for n = 0 and n = l, V (x) K and Lemmas 5 and 7. Let us show that
l
x C13 (x1 ) +
e
dx2
dx1
C13 (x)
e
|V (x1 )|
|V (x2 )| 0 + (x1 )
R0
1
=o
(x)
+
x
x1
dx2
|V (x2 )|
l+1
,
x +.
(51)
Then using (50) and the fact that in accordance to Lemma 5

+

+
dx2
Re V (x2 )
=O
dx2
|V (x2 ) |
|V (x2 )|
x
+

+
V (x2 )
dx2
=O
dx2 = O
,
|V (x2 ) |
V (x2 )
x
the required estimate (49) would be proved.

Performing the change of variable = (x2 ), we get
l
x C13 (x1 ) +
e
dx2
dx1
C13 (x)
e
|V (x1 )|
|V (x2 )| 0 + (x1 )
R0
x1
x +,
=e
C13 (x)
(x)/2 (x)
+
0

e
C13 (x)
1 C13 (x)/2
e
0
1
+
0 + (x)/2
eC13 (x)/2
0
eC13 (x)/2
(l + 1)0
(x)
(x)/2
+ +
x1
R0
2eC13 (x)
+
(x)
=
(x)/2
x
R0
2eC13 (x)
+
(x)
R0
x
R0
(x)/2
+
x1 ( )
eC13
|V (x1 ( ))|
dx2
|V (x2 )|
eC13 (x1 )
|V (x1 )|
+
eC13
|V (x1 ( ))|
x1
dx1
|V (x1 )|
eC13 (x1 )
|V (x1 )|
x1 ( )
dx2
|V (x2 )|
x1 ( )
dx2
|V (x2 )|
+
l
+
+
l
dx2
|V (x2 )|
l
= O eC13 (x)/2 + o
1
(x)

d
dx1
|V (x1 )|
dx2
|V (x2 )|
l
dx1
l+1
+
x1
d
0 +
d
|V (x1 ( ))|
x1
dx2
|V (x2 )|
l
+
x
dx2
|V (x2 )|
(52)
dx1 .
x C13 (x1 ) +
It follows from (52) and from the estimate of R0 e|V (x )| ( x1
1
that
l
x C13 (x1 ) +
e
dx2
dx1
C13 (x)
e
|V (x1 )|
|V (x2 )| 0 + (x1 )
R0
l
155
dx2 )l dx1
|V (x2 )|
contained in (50),
l+1
.
(53)
As well it follows from (16) and (25) that

O e
C13 (x)/2
l(l+1)
= o
(x) = o

=o
1
(x)
+
x
l (x)
|V (x)|l+1
l+1
dx2
|V (x2 )|
x +.
(54)
Then the estimate (51) is derived from (54) and (53). Thus, (35) is valid for n = l + 1.
Finally let us show that functions n, (x) (n, (x)) for any n are normalizable (nonnormalizable) at +. For this purpose it is sufficient to prove that the leading term of (34)
156
((35)) is normalizable (non-normalizable) at +. Normalizability of the leading term of (34) is

owed to the fact that, because of Lemma 5, the following estimates take place,
2n
x

dx
1
1

e2 Re (x;)

|V (x) |
V (x1 )
+
(2n+1)/4
C
e2C3 ( (x) (R1 ))
2
|V (x)|
+
x
dx1
|V (x1 )|

(2n+1)/4 (x) 2C3 ( (x) (R1 ))
C2
e
|V (x)|
+
(2n+1)/4
C2
R0
dx1
|V (x1 )|
+
2n
R0
2n
dx1
|V (x1 )|
2n
(x)e2C3 ( (x) (R1 )) ,
x R1 ,
where the latter expression is normalizable at +. Non-normalizability of the leading term (35)
follows from the fact that in view of Lemma 5 and (16), (25) the estimates are valid for some
constant C16 > 0,
2n
x

dx1
1

2 Re (x;)
e

|V (x) |
V (x1 )
+
+
2n
4
C1 2C3 ( (x) (R1 ))
Re V (x1 )
e

dx1
|V (x1 ) |
|V (x)|
x
(4n+1)/4
C1
C32n
(x)e2C3 ( (x) (R1 ))
|V (x)|
(x)2n (x) 2C (x)

C16
e 3
|V (x)|2n+1
C16
+
x
dx1
|V (x1 )|
2n
(x)2n (x)
C02n+1 (0 + (x))(2n+1)
e2C3 (x) ,
x R1 ,
(55)
where the latter expression is non-normalizable

+
at +.
Let us now consider the case R0 dx1 / |V (x1 )| = + and prove that l+1, (x) and
l+1, (x) can be written in the form

x
x
1
l+1, (x) = 0, (x)
0, (x1 )l, (x1 ) dx1 0, (x) 0, (x1 )l, (x1 ) dx1 ,
2

l+1, (x) =
R1
x
x
1
0, (x)
2
R1
0, (x1 )l, (x1 ) dx1 0, (x)
(56)
0, (x1 )l, (x1 ) dx1 .
R1
(57)
157
x
Convergence of + 0, (x1 )l, (x1 ) dx1 follows from the fact that V (x) K and in view of
Lemma 5, and (37) for n = 0 and n = l there are positive constants C3 , C17 , such that

x

0, (x1 )l, (x1 ) dx1

+
C17 e2 Re (x;)+2C3 (x)
+
x
e2C3 (x1 )
|V (x1 )|
x1
R1
dx2
|V (x2 )|
l
dx1
+ l (x )
1
e2C3 (x1 ) dx1
|V (x1 )|
C17 2 Re (x;)+2C3 (x)

e
l
C17 2 Re (x;)+2C3 (x)
e
l+1
C17 2 Re (x;)+2C3 (x)
e
l+1
x
+
l (x1 ) (x1 )e2C3 (x1 ) dx1
x
+
l e2C3 d < +.
(x)
Let us now find

x the asymptotics of integrals, contained in (56) and (57). Due to (37) and (38) the
integrand of R1 0, (x1 )l, (x1 ) dx1 reads
0, (x)l, (x)
1
= l
2 l! V (x)
x
R1
dx1
V (x1 )
l
ln (x)
1+O
(x)

,
x +.
The first term of right side of (58) contributes into the integral as follows:
x
l
l+1
x x1
1
dx2
dx1
dx1
1
2
,
=
2l l!
V (x2 )
V (x1 ) (l + 1)! 2
V (x1 )
R1
R1
R1
and the absolute value of contribution of the second term is less than or equal to
l
x x1
ln(2 + (x1 ))
dx2
dx1
C18
,
2 + (x1 )
|V (x2 ) |
|V (x1 ) |
R1
R1
for a constant C18 > 0. In view of Lemma 5 the latter expression is less than or equal to
l
x x1
ln(2 + (x1 )) dx1
dx2
C19
2 + (x1 )
|V (x2 )|
|V (x1 )|
R0
R0
x
C19
R0
l1
(x1 ) 2 + (x1 )
ln 2 + (x1 ) dx1
(58)
(59)
158
C19
l+1

x
ln (x)
dx1
2 + (x) ln 2 + (x) = O
(x)
|V (x1 )|
R1
ln (x)
(x)
=O

ln (x)
=O
(x)
x
R1
x
R1
Re V (x1 )
dx1
|V (x1 ) |
dx1
V (x1 )
l+1
l+1
,
x +,
(60)
for a constant C19 > 0. Thus,

x
0, (x1 )l, (x1 ) dx1
R1
l+1
x

ln (x)
dx1
2
1
=
,
1+O
(l + 1)! 2
(x)
V (x1 )
x +.
(61)
R1
In the case l = 0 one may write (58), using (17), (19) and Lemma 6 in the form

1
1
0, (x)0, (x) =
,
1+O
(x)
V (x)
x +.
(62)
Respectively,
xdue to Lemma 5 and for V (x) K the contribution of the second term of (62) to
the integral R1 0, (x1 )0, (x1 ) dx1 is less than or equal to
x
C20
R1
dx1
|V (x1 ) |(0 + (x1 ))
x
C20 C2
R0
C20 C2
dx1
x
|V (x1 )|(0 + R01 |V (x2 )| dx2 )
x (x ) dx
1
1
R0
0 + (x1 )
1

= C20 4 C2 ln 1 + (x)/0 = O ln (x)

x
ln (x)
dx1
, x +,
=O
(x)
V (x1 )
R1
for constants C20 > 0 and 0 > 0.
159
5 and V (x) K, the following estimate holds for the integral

x In view of (37), (40), Lemma
4
(x
)
(x
)
dx
,
1
+ 0, 1 l, 1
x

0, (x1 )l, (x1 ) dx1

+
+ x1
C20
x
C21 e
R1
dx2
|V (x2 ) |
2 Re (x;)+C22 (x)
l
dx1
e2 Re (x1 ;)
|V (x1 ) |
+ x1
x
R1
dx2
|V (x2 )|
l
eC22 (x1 ) dx1
|V (x1 )|
+ +
l1
l (x1 )

C21 2 Re (x;)+C22 (x) eC22 (x1 ) l
C22 (x1 )

e
e
=
(x1 )
C22
|V (x1 )|
|V (x1 )|3/2
x
x

Re V (x1 ) Re V (x1 ) + Im V (x1 ) Im V (x1 ) l
(x1 ) dx1
|V (x1 )|3
l

+ C22 (x1 )
l (x1 )
l l1
e
C21 2 Re (x;) (x)
C22 (x)
e

+e
(x1 ) + C23
dx1
C22
|V (x)|
(x1 )
|V (x1 )|
x
C21 2 Re (x;) l

e
(x) + eC22 (x)
C22
=O
e2 Re (x;) l (x)

=O e
2 Re (x;)
x
R1

=O e
2 Re (x;)
x
R1
+
x
eC22 (x1 ) l1
(x1 )(l + C23 ) dx1
|V (x1 )|

+O e
2 Re (x;)+C22 (x)
dx2
|V (x2 )|
1
(x)
+
x
eC22 (x1 ) l
(x1 ) dx1
|V (x1 )|
l
dx2
V (x2 )
l
,
x +,
(63)
for positive constants C20 , . . . , C23 . The asymptotics (37) and (39) for n = l + 1 is derived from
(56), (61), (63) from (37) and (38) for n = 0 and from (18), (20), (40) and Corollary 1.

4 In (63) and (65) the estimate x dx /|V (x )| = O( x dx /V (x ) ) is used. Derivation of this estimate is
1
1
1
1
R1
R1
contained in (60).
160
x
The integral R1 0, (x1 )l, (x1 ) dx1 can be calculated in the same way as the following one,
x
R1 0, (x1 )l, (x1 ) dx1 and the result is
x
0, (x1 )l, (x1 ) dx1
R1
l+1

x
ln (x)
dx1
2
1
=
,
1+O
(l + 1)!
2
(x)
V (x1 )
x +.
(64)
R1
x
For the integral R1 0, (x1 )l, (x1 ) dx1 , due to (38), (40), Lemma 5 and V (x) K, the estimate
takes place

x

0, (x1 )l, (x1 ) dx1

R1
x x1
C24
R1
R1
dx2
|V (x2 ) |
l
e2 Re (x1 ;) dx1
|V (x1 ) |
x x1
C25 e2 Re (x;)C26 (x)

R0
R0
C25 2 Re (x;)C26 (x) l

=
(x)
e
dx2
|V (x2 )|
x
l
eC26 (x1 ) dx1
|V (x1 )|
(x1 )eC26 (x1 ) dx1
R0
C25 2 Re (x;) l
e
(x)
C26

x
=O e
2 Re (x;)
R1
x
= O e2 Re (x;)
R1
dx2
|V (x2 )|
l
dx2
V (x2 )
l
,
x +,
(65)
for positive constants C24 , . . . , C27 . The asymptotics (38) for n = l + 1 follows from (57), (64),
(65) as well as from (37) and (38) for n = 0.
Finally let us check that n, (x) (n, (x)) for any n is normalizable (non-normalizable) at
+. For this purpose it is sufficient to examine that the leading term of the right side (37) ((38))
is normalizable (non-normalizable) at +. Normalizability of the leading term of (37) follows
from the fact that due to V (x) K and Lemma 5 the following estimate is valid:
2n
x

dx
1
1

e2 Re (x;)

|V (x) |
V (x1 )
R1

(2n+1)/4
C
e2C3 ( (x) (R1 ))
2
|V (x)|
x
R0
dx1
|V (x1 )|
161
2n
(2n+1)/4
(x) 2C ( (x) (R )) 2n
C2
1
3
e
(x)
2n |V (x)|
(2n+1)/4
C2
2n (x) (x)e2C3 ( (x) (R1 )) , x R1 ,
2n+1
the right side of which is obviously normalizable at +. Non-normalizability of the leading
term in (38) follows from the fact that in view of (25), Lemma 5 and with the help of the trick in
(55) the following estimate holds:
2n
x

dx1
1

2 Re (x;)
e

|V (x) |
V (x1 )
R1
(4n+1)/4 2n
C3
C1
|V (x)|
(4n+1)/4
C1
C32n
C0
(x)
|V (x)|
x
2C3 ( (x) (R1 ))
(x)
(0 + (x))
R1
dx1
|V (x1 )|
x
2C3 ( (x) (R1 ))

R1
2n
dx1
|V (x1 )|
2n
,
x R1 ,
the right side of which is evidently non-normalizable at +. Lemma 9 is proved.
Corollary 4. In conditions of Lemma 9 any formal associated function of h of nth order normalizable at , for a spectral value such that either 0 or Im = 0, can be written in the
form
n

aj, j, (x),
aj, = const,
an, = 0
(66)
j =0
and any associated function of h of nth order, non-normalizable at , for the same spectral
value can be presented as follows
n

bj, j, (x) + cj, j, (x) ,
(67)
j =0
where bj, , cj, = const and either bn, = 0 or cn, = 0.

Corollary 5. For normalizable associated functions 1 (x) and 2 (x) of a Hamiltonian h K
of any orders, for eigenvalues 1 and 2 respectively such that either 1,2 0 or Im 1,2 = 0 the
equality
+

h1 (x) 2 (x) dx =
takes place.
+

1 (x) h2 (x) dx
(68)
162
5. Invariance of the potential sets K and K

Invariance of the potential sets K and K under intertwining is proved in Lemmas 10 and 1
respectively.
Lemma 10. Let:
(1)
(2)
(3)
(4)
h+ = 2 + V1 (x), V1 (x) K;
C and either 0 or Im = 0;
(x) be zero-mode of h+ ;
(x) = (x)/(x), q1 = + (x).
Then the potential V2 (x) of the Hamiltonian

h 2 + V2 (x) = + q1 q1+ ,
intertwined with h+ = + q1+ q1 by means of equalities
q1 h = h q1
belongs to K also.
> 0 such that
Proof. Let us first check that there is R02

V2 (x) V1 (x) 2 ln (x)
is twice continuously differentiable. For this purpose it is
(see Eq. (53) in [7]) for |x| R02

has not zeroes and is four
sufficient to show that there is R02 > 0 such that (x) for |x| R02

has not
times continuously differentiable. Existence of R02 > 0 such that (x) for |x| R02
zeroes follows from the fact that one of asymptotics of Lemma 8 is valid for (normalized) (x).
is so large that
Without loss of generality suppose that this R02

V1 (x)[R
02 ,+[
2
C[R

,
02 ,+[

2
V1 (x)],R ] C],R
].
02
(69)
02
follows from the

Then the fact that (x) is four times continuously differentiable for |x| R02
equality = (V1 ), from (69) and from the fact that (x) is twice continuously differen as a zero-mode of h+ .
tiable for |x| R02
and > 0 such
Let us now verify that Im V2 /ReV2 = o(1), x and there are R02 R02
2
that ReV2 (x) 2 for any |x| R02 . The former follows from (31) in view of Im V1 /ReV1 =
o(1), x . Moreover, since obviously

Re V2 (x) = Re V1 (x) 1 + o(1) ,
x ,
(70)
and > 0 such that for any |x| R , the value of [1 + o(1)] in (70) is
and there are R02 R02
1
02
more than or equal to 1/2 and Re V1 (x) 1 , so that for any |x| R02 the inequalities hold
Re V2 (x)
163
1
1
Re V1 (x) ,
2
2
wherefrom the existence of the required R02 and 2 = 1 /2 follows.

Finally we show that the function
x
2

|V2 (x)|
|V2 (x)|2
V2 (x1 ) dx1
+
|V2 (x)|3 |V2 (x)|2
(71)
R02
is bounded for x R02 (the case with a similar function for x R02 can be considered analogously). In view of V1 (x) K, (28), (29) and (31) we have
x

V2 (x1 ) dx1 = O 1, (x) ,

R02

|V2 (x)|
1
=O 2
,
|V2 (x)|2
1, (x)

|V2 (x)|2
1
=
O
,
2 (x)
|V2 (x)|3
1,
x

V1 (x) dx1 ,
x +, 1, (x) =
R02
wherefrom boundedness of (71) is derived. Lemma 10 is proved.
Corollary 6. Using (18), (20) and Remark 1, (34) and (36) ((37) and (39)), (25), (66) and estimations similar to the estimations in the proof of Lemma 9, one can easily check that under
conditions of Lemma 10 the operator q1 maps any formal eigenfunction or associated function
(of any order) of the Hamiltonian h+ normalizable at + (at ) to a function normalizable
at + (at ), for any spectral value such that either 0 or Im = 0.
Lemma 1. Let:
(1) h+ = 2 + V1 (x), V1 (x) K;
(2) h = 2 + V2 (x), V2 (x) CR ;
+
(3) qN
h = h qN
, where qN
is a differential operator of N th order with coefficients belonging
2
to CR ;
(see Theorem 1 in part I [7]) satisfies one of the condi(4) each eigenvalue of S+ -matrix of qN
tions: either 0 or Im = 0.
Then:
(1) V2 (x) K;
(2) coefficients of qN
belong to CR ;
+
+
+
t
+
(3) h+ qN = qN h , where qN
= (qN
) , and moreover coefficients of qN
belong to CR as well.
Proof. Let 1 (x), . . . , N (x) be a basis in ker qN

, in which S+ -matrix of qN
(see Theorem 1
in part I [7]) has the canonical form. Since, firstly, 1 (x), . . . , N (x) as eigen- and associated
functions of h+ belong to CR , secondly, the Wronskian W (x) of the functions 1 (x), . . . , N (x)
has not any zeros and, thirdly,
164
(x)
1

(x)
2
1 .
.
qN
=
W (x) .
N (x)

1
1 (x)
2 (x)
..
.
..
.
(x)
N
1 (x)

(N )
2 (x)

..
,
.

(N )
N (x)

N
(N )
qN
+
and thereby of qN
belong to CR . Belonging of V2 (x) to CR follows from
the coefficients of
the equality V2 (x) = V1 (x) 2(ln W (x)) (see Eq. (53) in [7]), from inclusion W (x) CR and
from absence of zeroes for W (x). Inclusion V2 (x) K follows from inclusion V2 (x) CR ,
from Lemma 10 and can be also justified by the factorization procedure described in Lemma 1
+
+
of [6]. The equality h+ qN
= qN
h is obvious. Lemma 1 is proved. 2
6. Proofs of Lemmas 24 and Theorem 3

The properties of associated functions under intertwining are described by the
Lemma 2. Let:
(2) n (x), n = 0, . . . , M, be a sequence of formal associated functions of h+ for spectral
value :
+
h+ 0 = 0 ,
h n = n1 , n 1,
where either 0 or Im = 0.
Then:
(1) there is a number m such that 0 m min{M + 1, N },
qN
n 0,
n < m,
and
l = qN
m+l ,
l = 0, . . . , M m,
is a sequence of formal associated functions of h for the spectral value :

h 0 = 0 ,
h l = l1 , l 1;
n is
(2) if a function n (x), for a given 0 n M, is normalizable at + (at ), then qN
normalizable at + (at ) as well.
Proof. Existence of m such that 0 m min{M + 1, N },
qN
n 0,
n < m,
and either m > M or
qN
m 0,
(72)
165
can be derived from linear independence of n and from the fact that dimension of ker qN
is N .
The fact that l = qN m+l , l = 0, . . . , M m, is a sequence of formal eigenfunction and associated functions of h (if m M):

h l = l1 , l 1,
h 0 = 0 ,
follows from the chains:
m = qN
h m = qN
(m + m1 ) = 0 , 1 0,
h 0 = h qN
h m+l = qN
m+l1 = l1 ,
h l = h qN m+l = qN
l 1,
+
= qN
h and (72) are used. Before the proof of the second statement of
if intertwining h qN
Lemma 2 let us note that with the help of similar arguments one can show that in the conditions of Lemma 10 the operator q1 maps any formal eigenfunction or associated function of
h+ for a spectral value either to the identical zero or to a formal eigenfunction or associated
function of h for the same spectral value . Thus, the second statement of Lemma 2 follows
from Lemma 10, Corollary 6 and the construction, described in Lemma 1 of [6]. Lemma 2 is
proved. 2
Corollary 7. (Corollary 1 of Ref. [7].) Since h+ is an intertwining operator to itself and both
eigenvalues of its S+ -matrix (see Theorem 1 in part I [7]) are zero, then if n (x) is normalizable
at + (at ), then j (x), j = 0, . . . , n 1, is normalizable at + (at ) as well.
Corollary 8. (Corollary 2 of Ref. [7].) If there is a normalizable associated function of nth
order n (x) of the Hamiltonian h with a potential belonging to K for an eigenvalue , which is
either 0 or Im = 0, then for this eigenvalue there is an associated function j (x) of the
Hamiltonian h, normalizable on the whole axis, of any smaller order j :
j = (h )nj n ,
j = 0, . . . , n 1.
(x) be a canonical basis of zero-modes of the

Corollary 9. (Corollary 3 of Ref. [7].) Let i,j
+
intertwining operator qN , i.e. such that S -matrix (in Theorem 1 of part I [7]) has in this basis
the canonical (Jordan) form:
= i i,0
,
h+ i,0
h+ i i,j
= i,j
1 ,
i = 1, . . . , n, j = 1, . . . , ki 1,
n

ki = N.
i=1
Then there are numbers
(x),
i,j
+
ki
and
+
ki
,
+
0 ki,
ki , such that for any i the functions
+
j = 0, . . . , ki,
1,
are normalizable at + or respectively and the functions
(x),
i,j
+
j = ki,
, . . . , ki 1,
are non-normalizable at the same + or .

+
on a choice of the canonical basis in the case, when the
Independence of these numbers ki,
intertwining operator qN cannot be stripped-off, follows from
166
Lemma 3. Let:
(2) qN
not be able to be stripped-off.
Then any two formal associated functions of h+ of the same order for the same spectral value
are either simultaneously normalizable at + or simultaneously

when being zero-modes of qN
non-normalizable at +. The same takes place at .
Proof. Assume that there are two sequences of a formal eigenfunction and associated functions
of h+ for the same spectral value :
+
h+ l,0 = l,0 ,
h l,j = l,j 1 , l = 1, 2, j = 1, . . . , j0 ,
such that 1,j0 is normalizable at +, 2,j0 is non-normalizable at + and
qN
l,j0 = 0,
l = 1, 2.
Let us show that it leads to contradiction.

Let us check first that
l,j = 0,
qN
l = 1, 2, j = 0, . . . , j0 1.
For j = j0 1 these equalities follow from the chain
+
qN
h l,j0 = h qN
l,j0 1 = qN
l,j0 = 0,
and for j < j0 1 they can be derived in the same way by induction. As for intertwining operator,
which cannot be stripped-off, there is only one zero-mode of h+ (up to a constant cofactor),
corresponding to a fixed eigenvalue of its S+ -matrix (see Theorems 1 and 2 in part I [7]), so
1,0 (x) and 2,0 (x) are proportional. Without loss of generality suppose that 1,j and 2,j are
normalized so that
1,0 (x) 2,0 (x).
Then the sequence 1,j 2,j represents a sequence of associated functions of h+ for the same
) and 1,j0 2,j0 is an associated function of the order

eigenvalue (being zero-modes of qN
j1 < j0 non-normalizable at +. But on the other hand there is an associated function 1,j1
of h+ normalizable at + (see Corollary 7 (Corollary 1 of Ref. [7])) of the order j1 for an
. Performing in the same way by induction, we come

eigenvalue which is a zero-mode of qN
ker(h+ ) (dimension of which is 1 in view of

to the conclusion that intersection ker qN
Theorem 2 of part I [7]) contains non-trivial functions normalizable and non-normalizable at
+, the latter being impossible. The consideration of the case is analogous. Lemma 3 is
proved. 2
The following Lemma 4 clarifies interrelation between the behavior at of elements of
canonical bases of mutually transposed intertwining operators.
Lemma 4. Let:
167
+
(2) {i,j
} and {i,j
} are canonical bases of ker qN
and ker qN
respectively;
(3) qN cannot be stripped-off ;

(4) ki is algebraic multiplicity of eigenvalue i of S+ -matrix (see Theorem 1 of part I [7]).
Then for any i and j the function i,j

(x) is normalizable (non-normalizable) at + if and only
+
if i,ki j 1 (x) is non-normalizable (normalizable) at +. The same takes place at .
has the
Proof. In accordance with Corollary 9 (Corollary 3 of Ref. [7]) for any i the basis i,j
following structure:
,
i,j
+
j = 0, . . . , ki,
1,
are normalizable at and
i,j
,
+
j = ki,
, . . . , ki 1,
+
are non-normalizable at , where 0 ki,
ki . Moreover in view of Lemma 3 the numbers
+
ki, are independent of a choice of a canonical basis. To prove Lemma 4 it is sufficient to
establish that for any i
+
,
i,j
+
j = 0, . . . , ki ki,
1,
(73)
are normalizable at and

+
i,j
,
+
j = ki ki,
, . . . , ki 1,
(74)
are non-normalizable at .
Let i,j, be a sequence of a formal eigenfunction and associated functions of h+ normalizable at , for a spectral value i . Then because of Lemma 3
i,j, 0,
qN
+
j = ki,
, . . . , ki 1.
On the other hand, by virtue of Lemma 2, the functions qN

i,j, form a sequence of a formal
eigenfunction and associated functions of h for the same spectral value i and for j ki 1
+
+
(since in virtue of Theorem 1 of part I [7], qN
qN is a polynomial
represent zero-modes of qN
+
+
k
i
of h , containing cofactor (h i ) ). Moreover in view of Lemma 2 these functions are
normalizable at . Thus, by virtue of Lemma 3 the functions (73) are normalizable at .
We prove now that the functions (74) are non-normalizable at . For this purpose, because
+
is non-normalizable
of Corollary 9 (Corollary 3 of Ref. [7]), it is sufficient to prove that i,k
k +
i
i,
in the product of intertwining operators of first order

at . Let us consider factorization of qN
in accordance with Lemma 1 of [6]:
= rN r1 ,
qN
where r1 , . . . , rki are chosen so that
= 0,
rj+1 r1 i,j
j = 0, . . . , ki 1.
+
there is a factorization
Then for qN

t
t
+
qN
= r1 rN ,
168
where the zero-mode of (rj )t is evidently

t

t +
rj+1 rN i,k
,
i j
+
Suppose that i,k
k +
i
rk+
i, +1
i,
j = 1, . . . , ki .
is normalizable at . Then

t +
rN i,k
k +
i
i,
(i.e. in view of Corollary 6 the zero-mode of (rk+ )t ) is normalizable at as well. But this
i,
statement contradicts to the fact that

rk+
i, 1
r1 i,k
+
i, 1
(i.e. the zero-mode of rk+ ) is normalizable (because of the same Corollary 6) at . Thus,
+
i,k
k +
i
i,
i,
is non-normalizable at and Lemma 4 is proved.
A more precise result on interrelation between Jordan structures of intertwined Hamiltonians

and the behavior of transformation functions is contained in
Theorem 3. Let:
(2) () is an algebraic multiplicity of an eigenvalue of h , i.e. the number of independent
eigenfunctions and associated functions of h normalizable on the whole axis;
(3) if is not an eigenvalue of S+ (see Theorem 1 of part I [7]), then n+ () = n () = n0 () =
0 and if = i , where i is an eigenvalue of S+ , then n (i ) is a number of functions among
i,j
(x), j = 0, . . . , ki 1, normalizable at both infinities and n0 (i ) is a number of functions
+
(x) (or i,j
(x)), j = 0, . . . , ki 1, normalizable only at one of infinities.
among i,j
Then for any such that either 0 or Im = 0 the equality
+ () n+ () = () n ()
takes place. Moreover if n0 () > 0 for some , then for this
+ () n+ () = () n () = 0.
Proof. Let us first notice that if for the level of the Hamiltonian h+ such that either 0 or
Im = 0 there is an associated function of the lth order normalizable on the whole axis, then for
the same level , any associated function of h+ of the lth order normalizable at one of infinities
is normalizable on the whole axis. This fact is easily verifiable in a way similar to the proof of
Lemma 3. Thus, in the case n0 () > 0 there is no any associated function of h+ normalizable
on the whole axis, of the order n+ () (and consequently of any greater order) for the level .
Hence in this case + () = n+ (). Moreover in view of Lemma 4 and of the symmetry between
h+ and h the equality () = n () holds for the case n0 () > 0 as well. Thus, for this case
Theorem 3 is proved.
169
In the case, when does not belong to the spectrum of S+ -matrix (see Theorem 1 in part I [7])
Theorem 3 follows from Lemma 2 and from the fact that an associated function of h , which
corresponds to under consideration, cannot be zero-mode of qN

(since in the opposite case this
function would be linear combination of formal eigenfunctions and associated functions of h ,
whose eigenvalues belong to the spectrum of S+ ).
Consider now the case, when = i belongs to the spectrum of S+ and n0 (i ) = 0. We shall
prove the inequality
(i ) + (i ) + n+ (i ) n (i )
(75)
only, because the opposite inequality

+ (i ) (i ) + n (i ) n+ (i )
follows from (75), Lemma 4 and the symmetry between h+ and h (the statement of the theorem is derived from these inequalities). Since in the subcase (i ) + (i ) + n+ (i ) 0 the
inequality (75) is trivial, we shall consider below the subcase
(i ) + (i ) + n+ (i ) > 0
(76)
only.
Let us show that there is a sequence i,j (x) such that,
+
h i i,j = i,j 1 , j = 1, . . . , (i ) + n+ (i ) 1,
h+ i,0 = i i,0 ,
i,j = 0,
qN
j = 0, . . . , (i ) + n+ (i ) 1,
and the functions i,j , j = + (i ), . . . , (i ) + n+ (i ) 1, are non-normalizable at both infinities. This sequence cannot contain more than ki terms, since in the opposite case associated
, the
functions of this sequence of orders greater ki 1 would be linear combinations of i,j
latter being impossible. Therefore, in view of Lemma 3, the number of associated functions of
the sequence non-normalizable at both infinities cannot be greater than the number of functions
non-normalizable at both infinities among i,j

with fixed i,
(i ) + (i ) + n+ (i ) n (i ),
that is required to be proved.
Consider a sequence of i,j, formal eigenfunction and associated functions of h+ normalizable at , for the level i (this sequence exists due of Lemma 9). First + (i ) functions of
this sequence are normalizable at both infinities (following the arguments used at the beginning
of this proof). By virtue of (66) any of the functions i,j

, j = 0, . . . , n+ (i ) 1, can be presented
as a linear combination of i,0, , . . . , i,n+ (i )1, . Moreover, due to linear independence of
i,0
, . . . , i,n
the reverse is valid as well. Hence,
+ (i )1
i,j, = 0,
qN
j = 0, . . . , n+ (i ) 1.
Moreover, in view of Lemmas 2 and 3 the functions
qN
i,j, ,
j = n+ (i ), . . . ,
are different from zero and form a sequence of a formal eigenfunction and associated functions
of h normalizable at for the level i . Applying the arguments at the beginning of this proof
one can show that the first (i ) terms of this sequence are normalizable at both infinities.
170
Using the sequence of formal associated functions i,j, one can construct another sequence
of formal associated functions of h+ for the same level i ,
i,j, =
j

Ai,k, i,j k, ,
Ai,k, = const,
Ai,0, = 0.
k=0
This sequence as well as the sequence i,j, has the following properties:
i,j, , j = 0, . . . , are normalizable at ;
i,j, , j = 0, . . . , + (i ) 1, are normalizable at both infinities;
i,j, = 0,
qN
j = 0, . . . , n+ (i ) 1;
(77)
qN
i,j, , j = n+ (i ), . . . , are different from zero and form the sequence of formal eigenfunction and associated functions of h normalizable at for the level i ;
i,j, , j = n+ (i ), . . . , (i ) + n+ (i ) 1, are normalizable at both infinities.

qN
One can choose constants Ai,k, so that the required sequence i,j can be written in the form
i,j = i,j, i,j, .
Indeed, notice that from inequalities (76) and + (i ) n+ (i ) it follows that (i ) > 0, i.e.
that there is a normalizable eigenfunction i,0 of h for the level i . As there is only one (up to
constant cofactor) normalizable eigenfunction of h for the level i the equalities
i,n+ (i ), = Ci, i,0

qN
take place for some constants Ci, = 0. The fact that the relation
i,j = 0
qN
(78)
holds for j = 0, . . . , n+ (i ) 1 follows from (77). The equality (78) holds for j = n+ (i ) if we
take Ai,0, = Ci, since
i,n+ (i ) =
qN
n+ (i )
Ai,k, qN
i,n+ (i )k, Ai,k, qN
i,n+ (i )k,
k=0
i,n+ (i ), Ai,0, qN
i,n+ (i ),
= Ai,0, qN
= (Ai,0, Ci, Ai,0, Ci, )i,0 = 0.

At last, one can attain validity of (78) for j = n+ (i ) + 1, . . . , (i ) + n+ (i ) 1, looking
through all j = n+ (i ) + 1, . . . , (i ) + n+ (i ) 1 and taking into account at every step that
qN
i,j being normalizable eigenfunction of h is proportional to i,0 . One has also to take into
i,j of Ai,j n+ (i ), is linear,

account that the dependence qN
i,j
qN
j

Ai,k, qN
i,j k, Ai,k, qN
i,j k,
k=0
j n
+ (i )

k=0
Ai,k, qN
i,j k, Ai,k, qN
i,j k,
j n
+ (i )1
Ai,k, qN
i,j k, Ai,k, qN
i,j k,
171
k=0
+ (Ai,j n+ (i ), Ci, Ai,j n+ (i ), Ci, )i,0 ,
i,j and i,0 is

and choose Ai,j n+ (i ), so that the proportionality coefficient between qN
vanishing. It happens that among i,j there are (i ) + n+ (i ) + (i ) functions, which
are non-normalizable at both infinities, because i,j, , j = + (i ), . . . , (i ) + n+ (i ) 1,
are normalizable at only. Thus, the required sequence is constructed and Theorem 3 is
proved. 2
Acknowledgements
I am grateful to my collaborators A.A. Andrianov and F. Cannata for invaluable remarks,
fruitful discussions and reading the manuscript. This work was supported by Grant RFBR 0601-00186-a and partially supported by the INFN grant.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
A.A. Andrianov, M.V. Ioffe, V.P. Spiridonov, Phys. Lett. A 174 (1993) 273.
A.A. Andrianov, F. Cannata, J.-P. Dedonder, M.V. Ioffe, Int. J. Mod. Phys. A 10 (1995) 2683.
V.G. Bagrov, B.F. Samsonov, Theor. Math. Phys. 104 (1995) 1051.
A.A. Andrianov, F. Cannata, J. Phys. A 37 (2004) 10297.
B. Mielnik, O. Rosas-Ortiz, J. Phys. A 37 (2004) 10007.
A.A. Andrianov, A.V. Sokolov, Nucl. Phys. B 660 (2003) 25.
A.A. Andrianov, F. Cannata, A.V. Sokolov, math-ph/0610024.
F.A. Berezin, M.A. Shubin, Schrdinger Equation, MSU Publ., Moscow, 1983.
Generalized hyper-Khler geometry and supersymmetry

Andreas Bredthauer
Department for Theoretical Physics, Uppsala University, Box 803, SE-751 08 Uppsala, Sweden
Received 6 October 2006; accepted 6 March 2007
Abstract
We propose the definition of (twisted) generalized hyper-Khler geometry and its relation to supersymmetric non-linear sigma models. We also construct the corresponding twistor space.
1. Introduction
It is well known that supersymmetry has a deep relation to geometry [1,2]. In the context of
sigma models the geometry of the target space is restricted by the amount of supersymmetry on
the worldsheet and its dimension. For example, for a two-dimensional worldsheet with manifest
N = (1, 1) supersymmetry the sigma model has N = (2, 2) supersymmetry if the target manifold is bi-Hermitian [3]. For around twenty years now the different possible geometries have
been studied [49]. For an overview over the different possibilities, see also [10]. Although this
classification is known, it was only recently that an elegant mathematical framework to deal with
these geometries was developed. Hitchin introduced the notion of generalized complex geometry in the context of generalized CalabiYau manifolds with fluxes [11]. Later, it was clarified by
Gualtieri [12]. The main idea is to replace the tangent bundle T M of a manifold M with the sum
of the tangent and the cotangent bundle T M T M. This puts metrics and two-forms of the
sigma model on an equal footing and unifies the concepts of complex and symplectic geometry.
It has been shown that bi-Hermitian geometry corresponds to a subset of generalized complex
geometry called generalized Khler geometry. The obvious question is how generalized complex geometries arise in the sigma model context. A lot of work has been done in this direction
[10,1318] and by now for the generalized Khler geometry, the picture is rather clear [19,20].
E-mail address: andreas.bredthauer@teorfys.uu.se.
doi:10.1016/j.nuclphysb.2007.03.004
A. Bredthauer / Nuclear Physics B 773 [PM] (2007) 172183
173
In the Lagrangian formulation of the sigma model generalized Khler geometry is completely
specified by the definition of a Lagrangian in a manifest N = (2, 2) supersymmetric formulation. The Lagrangian is the generalized Khler potential [21]. On the other hand, it has been
shown that it can be physically derived as the geometry admitting N = (2, 2) supersymmetry in
the phase space formulation of the sigma model [22]. In this work, we elaborate further in this
second direction and derive the conditions for N = 4 and N = (4, 4) supersymmetry. We show
how generalized hypercomplex and generalized hyper-Khler geometry arise in the phase space
formulation and construct the twistor space of generalized complex structures that is associated
with the N = (4, 4) supersymmetry.
This work is organized as follows: In Section 2, we review extended supersymmetry of the
N = (1, 1) supersymmetric sigma model and the relevant target space geometries. In Section 3,
we give an overview on the results for the phase space formulation of the N = (1, 1) sigma
model. In Section 4 we show generalized hypercomplex geometry arises in this notion and in
Section 5 we combine these results with those of a previous paper [22] and give the derive
generalized hyper-Khler geometry from N = (4, 4) supersymmetry. In Section 6 we comment
shortly on how this result relates to the Lagrangian formulation of the sigma model. In Sections 7
and 8 we define the twistor space for the generalized hyper-Khler geometry before concluding
with a short discussion in Section 9.
2. N = (1, 1) sigma model and extended supersymmetry
In this section, we review extended supersymmetry of the N = (1, 1) supersymmetric sigma
model and introduce the necessary notation. The action for the N = (1, 1) supersymmetric sigma
model is given by

1
S=
(1)
d2 d + d D+ D G () + B () .
2
Here, D are spinorial derivatives with
D = + i ++ ,
2
D
= i++ ,
{D+ , D } = 0.
The action is invariant under a manifest supersymmetry that is given by

0 () = i + Q+ + Q ,
(2)
(3)
where Q = iD +
++ are the supersymmetry charges. The action admits an extension
=
to N = (2, 2) supersymmetry if the target space geometry is bi-Hermitian [3]. The additional
supersymmetry transformation is of the form
1 () = + D+ J+ + D J ,
(4)
where J are complex structures that satisfy
= G ,
J G J
() J = 0.
(5)
The connections for the covariant derivatives are given by

()
(0)
=
G H ,
(6)
where (0) is the Levi-Civita connection for the metric G and H = dB is the torsion threeform. Explicitly, it is given by
H = 12 (B, + B, + B, ).
(7)
174
Indices separated by a comma define derivatives with respect to the corresponding spacetime
direction B, = B . We denote the Khler forms as = G J . In general, the nonmanifest supersymmetry transformation only closes on-shell

1 (), 1 ( ) = 2 + + ++ + 2 = .
(8)
The restrictions on the target manifold geometry increases if we consider N = (4, 4) supersymmetry. In that case, there are three additional supersymmetries of the form (4)
i () = + D+ J+i + D Ji ,
i = 1, 2, 3.
(9)
As an implication of the previous discussion, these are (on-shell) supersymmetry transformations

if the six tensors Ji are complex structures that satisfy (5). This is a consequence of the fact that
each of the additional supersymmetry transformations commute with the manifest supersymmetry (3). In addition, the transformations have to commute among themselves in order to satisfy
the supersymmetry algebra. Altogether, we have

i (), j ( ) = 2ij + + ++ + = , i, j = 0, 1, 2, 3.
(10)
We do not discuss the possibility of central charges here. These relations are satisfied on-shell if
the left- and the right-going complex structures anticommute among themselves
{J+i , J+j } = {Ji , Jj } = 2ij 1.
(11)
They form two hypercomplex structures. The target space geometry is bi-hypercomplex [3]. We
collect the requirements needed for N = (4, 4) supersymmetry of the action (1):
Ji G Ji
= G ,
() J = 0,
Ji Jj + Ji Jj = 2ij .
(12)
One way to achieve off-shell supersymmetry is if the left- and right-going complex structures
in (4) or (9) commute. A particular case is when the two complex structures are equal and the
extended supersymmetry transformations are of the form

i () = + D+ + D Ji .
(13)
The conditions for supersymmetry imply that this is only possible if the torsion H is zero. In
the case of N = (2, 2) supersymmetry, the target space geometry is Khler, while the action (1)
admits N = (4, 4) supersymmetry on a hyper-Khler manifold.
3. N = 2 extended supersymmetry in phase space
In [17] N = 1 supersymmetric phase space was introduced as the cotangent bundle T LM
(with parity reversed) of the superloop space LM = { : S 1,1 M}. Here, S 1,1 is a supercircle,
a circle with an additional Grassmann direction . is a superfield that embeds the supercircle into the manifold. The conjugate momenta S are spinorial fields. Therefore, the cotangent
bundle has reversed parity on its fibers. The superfields have the following expansions in the odd
coordinate
(, ) = X ( ) + ( ),
S (, ) = ( ) + ip ( ),
(14)
where p is the momentum conjugate to X . We use the notation of [22]. The phase space is
equipped with a symplectic structure

= i d d S ,
(15)
175
where is the de Rham differential on the manifold. The convention for is chosen in such
a way that its purely bosonic part is equal to the usual definition of the canonical symplectic
structure in bosonic phase space

|bos = d X p .
(16)
The symplectic structure yields a super-Poisson bracket

{F, G} = i d d F
G.
S S
can be twisted by a three-form H

1
H = i d d S H D .
2
(17)
(18)
Here, is the derivative with respect to . Consequently, this twists the Poisson bracket by H .
To keep things simple, we here work with the case H = 0 and only comment on the changes for
H = 0. The calculations in that case work out exactly in the same way.
The phase space has two natural operations
D = + i ,
Q = i .
(19)
They satisfy the algebra

D 2 = i,
Q2 = i,
{D, Q} = 0.
(20)
With this, the generator for manifest supersymmetry is given by

Q0 () = d d S Q ,
(21)
where is an odd parameter. It acts on the fields through the Poisson bracket

()S = S , Q0 () = iQS .

() = , Q0 () = iQ ,
(22)
Being a supersymmetry generator, it satisfies the supersymmetry algebra

P(a) = d d aS ,
Q0 (), Q0 ( ) = P(2 ),
(23)
where P(a) is the generator of -translations. Any additional supersymmetry that is generated
by some Q1 () has to satisfy the brackets

Q0 (), Q1 ( ) = 0,
(24)
Q1 (), Q1 ( ) = P(2 ).
The condition for which these conditions are satisfied was found in [17]. The form of Q1 () is
determined by dimensional arguments

1
d d 2D S J + D D L + S S P ,
Q1 () =
(25)
2
where the tensors can be conveniently combined into a map J : T T T T given by

J P
J=
(26)
.
L Jt
176
Q1 () is the generator of a supersymmetry transformation if the target space is generalized complex and J is a generalized complex structure.
For H = 0, we can construct the generators of supersymmetry in a similar way. With H = dB,
this can be achieved by replacing
S S B D
(27)
in the definitions of the supersymmetry generators, for example

Q0 = d d S B D Q .
(28)
Since we are forced to use the twisted version of the Poisson bracket as well, the transformation
on the fields (22) remains unchanged. Concerning the additional supersymmetry, we may afterwards rename the tensors in such a way that Q1 remains in the form (25). J is then a twisted
generalized complex structure.
4. N = 4 extended supersymmetry in phase space
In the previous section we reviewed the steps that lead to the condition that N = 2 extended
supersymmetry in phase space is possible if the target space is a generalized complex manifold.
The generator for the extended supersymmetry is given in (25). Here, we discuss the necessary
conditions for to have two such extended supersymmetries with generators Q1 () and Q2 () of
the form (25).
We show that the target space geometry has to be generalized hypercomplex. We define generalized hypercomplex geometry by three generalized complex structures that satisfy the algebra
of quaternions
J3 = J1 J2 .
(29)
Following the discussion of the previous section, Q1 () and Q2 () are generators of supersymmetry if we can relate them to two generalized complex structures J1 and J2 . In addition to
(23) and (24), the Poisson bracket of Q1 () with Q2 () has to vanish

Q1 (), Q2 ( ) = 0.
(30)
We can collect all the Poisson brackets in the following way:

Qi (), Qj ( ) = ij P(2 ),
i = 0, 1, 2.
(31)
The calculation of the bracket (30) is tedious but results in the two conditions for the generalized
complex structures
{J1 , J2 } = 0,
N (J1 , J2 ) = 0.
(32)
Here, N (J1 , J2 ) is the (generalized) Nijenhuis concomitant of J1 and J2 . It is given by

N (J1 , J2 ) = J1 (u + ), J2 (v + ) J1 u + , J2 (v + ) J2 J1 (u + ), v +
+ J1 J2 [u + , v + ] (J1 J2 ),
(33)
T M.
where the bracket is the Courant bracket and u + , v + are sections of T M

It follows that J3 = J1 J2 is the third generalized complex structure. Its integrability is
guaranteed by the vanishing of the (generalized) Nijenhuis concomitant and integrability of J1
and J2 . This proves that the target manifold is generalized hypercomplex.
177
The existence of J3 implies the existence of a third additional supersymmetry with generator
Q3 () of the form (25) and we conclude that a generalized hypercomplex manifold extends the
supersymmetry of the sigma model phase space to N = 4.
For the case H = 0 the calculations remain true but we have to replace the Poisson bracket by
its twisted version. The target space geometry is then twisted generalized hypercomplex.
5. Generalized hyper-Khler geometry
The previous discussion was completely model independent. In this section, we combine the
results with those of a previous paper [22]. There, it was shown that from the sigma model
point of view, the relation between generalized Khler and bi-Hermitian geometry follows from
the equivalence of the Hamilton and the Lagrange description of the sigma model. We start
with a short recapitulation of those results. The sigma model Hamiltonian for (1) is obtained by
performing one of the d -integrations.1 To this extend, we introduce new Grassmann coordinates
0 and 1 such that

1
1
1
D0,1 = (D+ iD ),
Q0,1 = (Q+ iQ ). (34)
0,1 = + i ,
2
2
2
We define the N = 1 component fields of by

= | 0 =0 ,
S = G D0 | 0 =0
(35)
0
and denote G ()
=0G ()|, D = D1 | and0 = . Performing the d integral in the usual
way by replacing d D0 and taking the = 0 component we obtain the action (1) in terms
of the N = 1 superfields

1
2

S = d d iS 0 i D + S DS G + S D S G .
(36)
2
Here, we focus on the case B = 0. The action has the typical form of a Legendre transformation
where the first term says that S is the conjugate momentum for and the second term yields
the Hamiltonian

1
H=
(37)
d d i D + S DS G + S D S G .
2
H is invariant under the manifest supersymmetry transformation (21) with Q Q1 |. The second
manifest supersymmetry of the original action (1) is non-manifest in this formulation. It is given
by
0 () = G S ,
0 ()S = iG + S S G .
(38)
In [22] it is shown that H is invariant under the additional supersymmetry if the target space
geometry is generalized Khler:

Q1 (), H = 0
(39)
implies that J commutes with the generalized metric

G1
.
G=
G
1 See also [23] for a mathematically more rigid derivation.
(40)
178
As a consequence, J = GJ is an additional generalized complex structure with supersymmetry

1 () of the form (25) such that
generator Q

1 ( ) = 2i H.
[J , J ] = 0,
Q1 (), Q
(41)
We now show that the Hamiltonian has N = (4, 4) supersymmetry if the geometry is generalized hyper-Khler. We call a manifold generalized hyper-Khler if it admits six generalized
complex structures Ji , Ji and a generalized metric G that satisfy the algebra of bi-quaternions
Cl2,1 (R)
Ji Jj = ij 12d + ij k Jk ,
Ji Jj = ij 12d + ij k Jk ,
Ji Jj = ij G + ij k Jk ,
Ji Jj = ij G + ij k Jk .
(42)
This definition coincides with [24,25]. However, we derive this definition from the sigma model.
The Hamiltonian is invariant under the three additional supersymmetries of the previous sections if it satisfies

Qi (), H = 0, i = 1, 2, 3.
(43)
As in (39), this is the case if the generalized complex structures Ji commute with G
[Ji , G] = 0.
(44)
According to the above, this induces three generalized complex structures Ji = GJ i . Each of
the triples {Ji , G, Ji } for i = 1, 2, 3 form a generalized Khler structure. The three generalized
i . The gencomplex structures Ji are associated to three additional supersymmetry generators Q
erators satisfy the algebra

j ( ) = 2iij H,
Qi (), Qj ( ) = ij P(2 ).
Qi (), Q
(45)
A straightforward calculation shows that these brackets are equivalent to the relations (42) and
integrability of Ji and Ji . We conclude that the sigma model Hamiltonian admits N = (4, 4)
supersymmetry if the target space is generalized hyper-Khler.
For H = 0, the Hamiltonian is given by

1
d d i D + S DS G + S D S G
H=
2

1
H
S S S + D D S H
(46)
3
and admits N = (4, 4) supersymmetry for a twisted generalized hyper-Khler geometry which is
defined in analogy to generalized hyper-Khler geometry but with twisted generalized complex
structures.
6. Relation to the Lagrangian formulation
The three different generalized Khler structures {Ji , G, Ji } correspond to bi-Hermitian
geometries, where the metric is read off from (40) and the complex structures are given via
the relation [12]
179

1
1
1 (J+i + Ji ) (+i
i
)
Ji =
,
t
(J
+
J
)
2
+i
i
+i
i

1
1
1 (J+i Ji ) (+i
+ i
)
,
Ji =
(47)
t
+
(J
J
)
2
+i
i
+i
i
where i = GJi are the Khler forms. From the bi-quaternion algebra it is easy to see that
J+i and Ji form two independent hypercomplex structures with
{J+1 , J+2 } = 0,
{J1 , J2 } = 0
(48)
but with nothing implied for the commutation relations of J+i and Jj . For the case H = 0 we
obtain in addition the relations of Section 2. We conclude that (twisted) generalized hyper-Khler
geometry is the phase space equivalent to bi-hypercomplex geometry.
7. Twistor space of generalized complex structures
In this section, we define the twistor space of generalized complex structures that is associated
to the N = (4, 4) supersymmetry of the sigma model Hamiltonian. The idea of a twistor space
is to encode the geometric properties of the target manifold M in the holomorphic structure of
a larger manifold, the twistor space. The original idea goes back to Penrose [26] and Salamon
[27,28]. We here follow the same approach as in the definition of the twistor space for hyperKhler geometry [6]. Twistor spaces of generalized complex structures and generalized Khler
structure are also discussed in [29,30] in order to find examples of generalized complex and
generalized Khler structures that are not induced by complex, symplectic and Khler structures.
Before discussing the twistor space for the generalized hyper-Khler geometry, we first review
the results for hyper-Khler geometry. Given a hypercomplex structure J1 , J2 , J3 the linear
combination
K = c 1 J1 + c 2 J2 + c 3 J3
(49)
is a complex structure if c lies on the unit sphere: c2
= 1. This sphere can be identified with CP 1 .

is usually represented as
with coordinates (, ) and the identification (, ) (, )
for = 0. Therefore, we can cover it with two sheets of coordinates (, 1) and (1, ) such that
= 1 in the overlapping region. In these coordinates,
CP 1
C2
K=
+

1
J1 +
J2 + i
J3 .
1 +
1 +
1 +
(50)
The twistor space of complex structures is the product space M S 2 , such that at any point
p M, S 2 parametrized the space of complex structures on Tp M. A complex structure for the
whole manifold is then given by the pair

1
+

J1 +
J2 + i
J3 , I ,
K =
(51)
1 +
1 +
1 +
where I is the standard complex structure on the sphere. This construction allows to define
hyper-Khler geometry in terms of an abstract parameter space.
We now define the twistor space of generalized complex structures in a completely analogous
way. Given the six generalized complex structures Ji and Ji of the previous section, we find that
the linear combinations that define generalized complex structures are given by the relation

1
1
K = ci + d i Ji + ci d i Ji ,
(52)
c 2 = d 2 = 1.
2
2
180
The space of generalized complex structures for a generalized hyper-Khler structure is parametrized by S 2 S 2 . In CP 1 CP 1 coordinates z, w, the vectors c and d are given by

1 w w w + w i(w w)
1 zz z + z i(z z )
,
,
,
d =
,
,
.
c =
(53)
1 + zz 1 + zz 1 + zz
1 + w w 1 + w w 1 + w w
Since the generalized complex structures Ji , Ji are a realization of the bi-quaternionic algebra,
it follows that K2 = 1 and K = GK where G is the generalized metric. The generalized metric
In the CP 1 coordinate w, this corresponds to
G acts on the parameter space by letting d d.
the anti-podal map
w : w w 1
(54)
that changes the orientation of the w-sphere. The ordinary complex structures for the two spheres
Iz and Iw define a complex structure JS for S 2 S 2 . This complex structure induces a generalized
complex structure on T (S 2 S 2 ) T (S 2 S 2 ) by

JS 0
.
JS =
(55)
0
JSt
A generalized complex structure for the combined space M S 2 S 2 is then given by

J = K(z, w), JS .
(56)
The proof that J is integrable is presented in the next section using the formulation of generalized
complex structures in terms of pure spinor lines. It is an interesting question, if I can be chosen
in a more general way in this context. Generalized complex structures for S 2 S 2 were explicitly
defined in [31].
The triples {K, G, K = GK} form different generalized Khler structures. The two spheres
parametrize the space of ordinary left- and right-complex structures on T M. We can clarify this
by introducing
1
1
Ji() = (Ji Ji ) = (1 G)Ji .
(57)
2
2
These are the projections of the generalized complex structures on the eigenspaces of G.
Explicitly and with relation (47), they are given by

1
1 Ji i
()
.
Ji =
(58)
t
Ji
2 i
With this, (52) becomes
(+)
K = ci Ji
()
+ d i Ji
(59)
We indeed find that c and d parametrize the two sets of complex structures J+i and Ji .
8. Pure spinors
It remains to show that J as defined in (56) is indeed a generalized complex structure. In order
to see this, we reformulate the previous discussion in the pure spinor language. Since
T M T M
always admits a Spin(d, d) structure that is isomorphic to the exterior algebra T M, we can
associate the +i eigenspace L of a generalized complex structure J1 with the annihilation space
181
of a spinor such that for the sections u + of L,

(u + ) = iX + = 0.
(60)
The spinor can in general only be defined locally. This is suitable for our purposes. More generally, we associate L with a pure spinor line U such that is locally a representative of U [12].
The spinor satisfies J1 = in, where n is the complex dimension of the manifold and the
multiplication is given by the action
1
tr(J ).
(61)
2
Here, J, P , L are the components of J as given in (26). Since J is integrable, the spinor is pure
and satisfies
J1 = L + iP J +
d = (u + )
(62)
T M.
If J1 was a twisted generalized complex structure then

for some section u + of T M
this equation would have been modified incorporating H :
dH = (d + H ) = (u + ) .
Given that is a pure spinor for the +i eigenspace of J1 , then

1 (+) 1
()
= 1 + zJ3 + wJ3
2
2
(63)
(64)
is a pure spinor for K. Since Ji and Ji are integrable by assumption, K is integrable as well.
This follows from the fact that the Nijenhuis concomitants vanish. Especially, for fix z, w,
d|z,w = (u + )
(65)
T M).
The bar indicates that the derivative is taken for fixed values
for some u + (T M
of z and w. The generalized complex structure JS is integrable by construction. We can associate
to it a pure spinor such that (A + b) = 0 for sections A + b of T (S 2 S 2 ) T (S 2 S 2 ).
Explicitly, is the top-holomorphic form = dz dw. Since is holomorphic in z, w, the spinor
= satisfies
d( ) = d|z,w + (1)|| d + dz z + dw w .
(66)

is a spinor.It is an element of the exterior algebra
T (M S 2 S 2 ) = ( T M)
2
( T S ) ( T S ). By construction the last two terms in (66) vanish such that
d = (X + ) + (1)|| (A + b)
= (X + + A + b) .
(67)
is a pure spinor for the almost generalized complex structure J = (K(z, w), JS ) and we conclude that J is integrable.
9. Discussion
In this short note we showed how generalized hypercomplex geometry emerges as the target
space geometry for the N = 4 supersymmetric sigma model phase space. We applied this result
to the Hamilton formulation of the N = (1, 1) supersymmetric sigma model and combined it
182
with the results of [22] to show that the Hamiltonian admits N = (4, 4) supersymmetry if the
target space is generalized hyper-Khler. We defined the twistor space of generalized complex
structures and clarified why the two parameterizing two-spheres do not parametrize the generalized complex structures but the supersymmetries in the Lagrangian formulation. Our results
fit the discussion in [24]. We also discussed the twistor space construction in terms of pure
spinors. This construction should be related to the deformation complex of generalized complex
structures [12]. Recently, it has been shown that generalized Khler geometry corresponds to a
manifest formulation of the N = (2, 2) supersymmetric sigma model [21]. It would be interesting
to relate our results to the harmonic superspace formulation of N = (4, 4) supersymmetry [32]
in the same way. Since hyper-Khler geometry is always CalabiYau, another interesting open
question is how generalized hyper-Khler geometry relates to generalized CalabiYau geometry.
Acknowledgements
The author thanks M. Zabzine and U. Lindstrm for many enlightening discussions and comments on this subject.
Appendix A. Generalized complex geometry
The notation of generalized complex geometry was first introduced by Hitchin [11] and later
clarified by Gualtieri [12].
The vector bundle T T on a complex d-dimensional manifold M has a natural pairing
1
X + , Y + = (iY + iX ).
2
(68)
The smooth sections of T T have a natural bracket, called the Courant bracket
1
[X + , Y + ]c = [X, Y ] + LX LY d(iX iY ).
2
(69)
It is a natural extension of the Lie-bracket [,] on the tangent bundle onto T T . Here, LX is
the Lie derivative with respect to X. This bracket has non-trivial automorphism parametrized by
2
a closed two-form b closed
(M) acting on the sections as
eb (X + ) = X + ( + iX b).
(70)
This transformation is called a b-transform and it acts on Courant bracket as

eb (X + ), eb (Y + ) c = b [X + , Y + ]c .
(71)
A generalized complex structure is the complex version of two complementary Dirac structures
We can define it as a map J : (T T ) C (T T ) C
with (T T ) C = L L.
satisfying
J t IJ = I,
J 2 = 12d ,

(X + ), (Y + ) c = 0,
where = 12 (12d J ) are projectors on L and L.
(72)
183
References
[1] B. Zumino, Supersymmetry and Khler manifolds, Phys. Lett. B 7 (1979) 203.
[2] L. Alvarez-Gaum, D.Z. Freedman, Ricci flat Khler manifolds and supersymmetry, Phys. Lett. B 94 (1980) 171.
[3] S.J. Gates, C.M. Hull, M. Rocek, Twisted multiplets and new supersymmetric nonlinear sigma models, Nucl. Phys.
B 248 (1984) 157.
[4] P.S. Howe, G. Sierra, Two-dimensional supersymmetric nonlinear sigma models with torsion, Phys. Lett. B 148
(1984) 451.
[5] T. Buscher, U. Lindstrm, M. Rocek, New supersymmetric sigma models with WessZumino terms, Phys. Lett.
B 202 (1988) 94.
[6] N.J. Hitchin, A. Karlhede, U. Lindstrm, M. Rocek, Hyper-Khler metrics and supersymmetry, Commun. Math.
Phys. 108 (1987) 535.
[7] P.S. Howe, G. Papadopoulos, Further remarks on the geometry of two-dimensional nonlinear sigma models, Class.
Quantum Grav. 5 (1988) 1647.
[8] P.S. Howe, G. Papadopoulos, Twistor spaces for HKT manifolds, Phys. Lett. B 379 (1996) 80, hep-th/9602108.
[9] S. Ivanov, I. Minchev, Quaternionic Khler and hyper-Khler manifolds with torsion and twistor spaces,
math.DG/0112157.
[10] U. Lindstrm, R. Minasian, A. Tomasiello, M. Zabzine, Generalized complex manifolds and supersymmetry, Commun. Math. Phys. 257 (2005) 235, hep-th/0405085.
[11] N. Hitchin, Generalized CalabiYau manifolds, Quart. J. Math. Oxford Ser. 54 (2003) 281, math.DG/0209099.
[12] M. Gualtieri, Generalized complex geometry, math.DG/0401221.
[13] U. Lindstrm, Generalized N = (2, 2) supersymmetric non-linear sigma models, Phys. Lett. B 587 (2004) 216,
hep-th/0401100.
[14] L. Bergamin, Generalized complex geometry and the Poisson sigma model, Mod. Phys. Lett. A 20 (2005) 985,
hep-th/0409283.
[15] R. Zucchini, A sigma model field theoretic realization of Hitchins generalized complex geometry, JHEP 0411
(2004) 045, hep-th/0409181.
[16] A. Bredthauer, U. Lindstrm, J. Persson, First-order supersymmetric sigma models and target space geometry,
JHEP 0601 (2006) 144, hep-th/0508228.
[17] M. Zabzine, Hamiltonian perspective on generalized complex structure, Commun. Math. Phys. 263 (2006) 711,
hep-th/0502137.
[18] I. Calvo, Supersymmetric WZPoisson sigma model and twisted generalized complex geometry, Lett. Math.
Phys. 77 (2006) 53, hep-th/0511179.
[19] U. Lindstrm, A brief review of supersymmetric non-linear sigma models and generalized complex geometry, hepth/0603240.
[20] M. Zabzine, Lectures on generalized complex geometry and supersymmetry, hep-th/0605148.
[21] U. Lindstrm, M. Rocek, R. von Unge, M. Zabzine, Generalized Khler manifolds and off-shell supersymmetry,
hep-th/0512164.
[22] A. Bredthauer, U. Lindstrm, J. Persson, M. Zabzine, Generalized Khler geometry from supersymmetric sigma
models, hep-th/0603130.
[23] F. Malikov, Lagrangian approach to sheaves of vertex algebras, math.AG/0604093.
[24] D. Huybrechts, Generalized CalabiYau structures, K3 surfaces, and B-fields, Int. J. Math. 16 (2005) 13, math.AG/
0306162.
[25] R. Goto, On deformations of generalized CalabiYau, hyper-Khler, G2 and Spin(7) structures I, math.DG/
0512211.
[26] R. Penrose, Nonlinear gravitons and curved twistor theory, Gen. Relativ. Gravit. 7 (1976) 31.
[27] S. Salamon, Quaternionic Khler manifolds, Invent. Math. 67 (1982) 143.
[28] S. Salamon, Differential geometry of quaternionic manifolds, Ann. Sci. cole Norm. Sup. Paris 19 (1986) 31.
[29] J. Davidov, O. Mushkarov, Twistor spaces of generalized complex structures, math.DG/0501396.
[30] J. Davidov, O. Mushkarov, Twistorial construction of generalized Khler manifolds, math.DG/0607030.
[31] N. Hitchin, Instantons, Poisson structures and generalized Khler geometry, Commun. Math. Phys. 265 (2006) 131,
math.DG/0503432.
[32] E. Ivanov, A. Sutulin, Diversity of off-shell twisted (4, 4) multiplets in SU(2) SU(2) harmonic superspace, Phys.
Rev. D 70 (2004) 045022, hep-th/0403130.
Asymptotics of the colored Jones polynomial

and the A-polynomial
Kazuhiro Hikami
Department of Physics, Graduate School of Science, University of Tokyo, Hongo 7-3-1, Bunkyo, Tokyo 113-0033, Japan
Abstract
We study Gukovs conjecture, which relates an asymptotics of the colored Jones polynomial to the Apolynomial, in the case of twist knots. We show that an asymptotics of the N -colored Jones polynomial with
q = exp(2 ir/N ) in large N limit is dominated by the NeumannZagier potential function which gives the
A-polynomial. We also discuss a case of torus knots.
1. Introduction
The N -colored Jones polynomial JK (N ) is a quantum invariant which is defined using the
quantum group Uq (sl(2)) as a generalization of the Jones polynomial. This can be constructed
by the Wilson loop observable of the ChernSimons theory [1] (see Ref. [2] for recent reviews);
when the link L in manifold M has knot components Ka , the quantum invariant for L is defined
as the correlation function
L

K
1
a
DA
WRa eiS .
WR1 ,...,RL (L) =
(1.1)
Z(M)
a=1
Here
WRK
is the Wilson loop operator

K
WR (A) = TrR P exp A
K
E-mail address: hikami@phys.s.u-tokyo.ac.jp.

URL: http://gogh.phys.s.u-tokyo.ac.jp/~hikami/.
doi:10.1016/j.nuclphysb.2007.03.022
K. Hikami / Nuclear Physics B 773 [PM] (2007) 184202
185
with a irreducible representation R. The ChernSimons action S and the partition function Z(M)
are respectively defined by

2
k
Tr A dA + A A A ,
S=
4
3
M

Z(M) = DAeiS ,
(1.2)
where k Z. Thus we have infinitely many quantum invariants based on Lie groups, and the
2i
N -colored Jones polynomial JL (N; e k+2 ) for link L is from SU(2) gauge group and the N dimensional representation. Studies on quantum invariants have been much developed last two
decades. Unfortunately geometrical interpretation of these quantum invariants is still missing.
Although, motivated by volume conjecture raised by Kashaev [3], it was pointed out that the
colored Jones polynomial at a specific value is related to the hyperbolic volume of knot complement [4].
Besides the colored Jones polynomial, we have the A-polynomial [5,6] which is also related
to SL(2; C). This is defined as an algebraic curve of eigenvalues of the SL(2; C) representation
of the boundary torus of knot, and contrary to the quantum invariants such as the colored Jones
polynomial it includes many geometrical information such as the boundary slopes of the knot.
Those two knot invariants are superficially independent. It is recently conjectured [7] that the
homogeneous difference equation of the N -colored Jones polynomial for knot K with respect
to N gives the A-polynomial for K (AJ conjecture). This fact was originally verified for both
the trefoil and the figure-eight knot with a help of computer algebraic system [7], and was later
proved for the torus knots [8] and the 2-bridge knots [9]. In Ref. [10] a recursion relation of the
summand of the colored Jones polynomial for twist knots was shown to give the A-polynomial.
See also Refs. [11,12].
Recently pointed out is still another connection between the colored Jones polynomial and
the A-polynomial, which we call generalized volume conjecture or Gukovs volume conjecture
(GVC) [13].
Conjecture 1 (GVC [13]). Let b be defined from a limit of the colored Jones polynomial evaluated at q = e2i/k by
b=
d
dr
lim
N,k
N/k=r
1
log JK N; e2i/k .
k
(1.3)
Then the pair (eb , eir ) is a zero locus of the A-polynomial of knot K.
We should note that different form of this conjecture is called generalized volume conjecture
in Ref. [13], but we rather call (1.3) GVC throughout this article.
GVC was demonstrated only for a case of the figure-eight knot [13,14]. Our purpose of this
article is to show that GVC is also fulfilled for a case of twist knots which include the figure-eight
knot using explicit forms of the colored Jones polynomials.
To study the limit (1.3) and check GVC, we set q, a variable of the N -colored Jones polynomial JK (N ) for knot K, as

2ir
q = exp
(1.4)
,
N
186
where r is a complex variable. Then we study a large N limit of JK (N ) with r being fixed;
N ,
r = fixed.
(1.5)
In studying the asymptotic limit of the colored Jones polynomial JK (N ), we recall a fact [15] that
the N -colored Jones polynomial can be rewritten by use of the q-hypergeometric function. Once
we obtain the invariant in such a form, it is rather straightforward to give an integral expression
of the asymptotic limit of JK (N ). Then we define the H -function for knot K as an integrand in
a limit (1.5);

dxi
N
2
exp
HK x, m + o(N ) .
JK (N ) =
(1.6)
xi
2ir
C
Here the path of integration C depends on an expression of the N -colored Jones polynomial
JK (N ). Concerning GVC, the term o(N ) does not matter, and the H -function can be defined as
a dominating term of the colored Jones polynomial. A parameter m is defined by
m2 = q N = e2ir .
(1.7)
In this article we demonstrate for twist knots Kp (see Fig. 1) and torus knots T2,2p+1 (see Fig. 4)
that the H -function is regarded as the NeumannZagier potential function [16,17] under a constraint
xi
HK (x, m2 )
= 0.
xi
(1.8a)
This constraint (1.8a) denotes saddle point equations for the integral (1.6) in large N limit. The
integral (1.6) is expected to be dominated by solutions {xi0 } of this set of saddle point equations,
and we have

0 2
N
HK x , m + o(N ) .
JK (N ) = exp
(1.9)
2ir
We should note that above estimation is not mathematically rigorous as we are not sure which
saddle point as solutions of (1.8a) dominates the integral. Although the H -function is not knot invariant due to that expression of the colored Jones polynomial is not unique, both the dominating
term of the integral (1.6) in N and the H -function with a constraint (1.8a) are invariants of
knot K by definition. To show that the H -function with (1.8a) is the NeumannZagier potential
function, we shall prove that the H -function gives the A-polynomial of knot. When we define
by
m2
HK (x, m2 )
= log
(m2 )
(1.8b)
we show that an algebraic equation of and m2 , which is calculated by eliminating x from a

set of Eqs. (1.8), coincides with the A-polynomial of knot K. As a result, we can conclude that
parameters (, m) defined from (1.7) and (1.8) denote the SL(2; C) representation of the meridian
and the longitude of the boundary torus of knot K by the upper triangular matrices,

m
() =
,
()
=
0 m1
0 1
187
up to conjugation. This proves GVC stated as Conjecture 1, and we may establish an intriguing
correspondence between the color N of the quantum knot invariant and the eigenvalue of the
SL(2; C) representation of the meridian.
This paper is organized as follows. In Section 2 we study the twist knots Kp . Using the
q-hypergeometric expression of the colored Jones polynomial derived skein theoretically in
Ref. [18] based on Ref. [19], we introduce the H -function, and we show that the H -function with
constraints (1.8) gives the A-polynomial for the twist knots which was computed in Ref. [20].
We also discuss numerically a relationship with the volume conjecture (a case of r = 1). We
study a limit p , and see that we obtain the hyperbolic volume of the Whitehead link. In
Section 3 we show that this correspondence also works for the torus knot.
Throughout this paper we use a standard notation [21]; the q-product and the q-binomial
coefficient are respectively defined as follows;
(x)n = (x; q)n =
n

1 xq
i1

(q)n
n
=
.
k q (q)nk (q)k
i=1
The Euler dilogarithm function for 0 x 1 is defined by

Li2 (x) =
n

x
n=1
n2
x
=
log(1 t)
dt
.
t
As usual we use the above integral expression for x C as the Euler dilogarithm function Li2 (x),
and we choose a branch of log(1 t) is on C \ [1, ) for which log(1 0) = 0 (see Ref. [22] for
detail). Our basic tool is the EulerMaclaurin formula; for analytic function f (x) on all points
of the straight line joining a to a + nh, we have
n

k=0
1
f (a + kh) =
h
a+nh
f (x) dx +
1
f (a) + f (a + n h)
2
m1

(1)r1
r=1
Br 2r1 (2r1)
f
(a + nh) f (2r1) (a) + Rm .
h
(2r)!
(1.10)
Here Br is the Bernoulli number and

Rm = (1)m+1
n1
Bm 2m (2m)
a + h(k + )
f
h
(2m)!
k=0
with 0 < < 1.

2. Twist knot
We study the N -colored Jones polynomial for the twist knot Kp . A case of p = 1 is the
figure-eight knot (see Fig. 1).
The N -colored Jones polynomial for the p-twist knot Kp was computed skein-theoretically
in Ref. [18] based on Ref. [19] as follows.
188
Fig. 1. Twist knots Kp are depicted. There is a p-full twist at the bottom of each figure. From left to right, figure-eight
knot (p = 1), left-hand trefoil (p = 1), and Stevedores ribbon knot (p = 2).
Proposition 1. (See [18].) The N -colored Jones polynomial JK (N ) for the twist knot K = Kp is
given in the form of the cyclotomic expansion as follows (we set p > 0);
JKp>0 (N ) =
sp s2 s1 0
p1
q (p1)sp (sp +1)+sp q 1N s q 1+N s

p
si2 (2sp +1)si
i=1
si+1
si

(2.1)
q
and
JKp<0 (N ) =
1
(1)sp q (p 2 )sp (sp +1) q 1N s q 1+N s
p
sp s2 s1 0
p1

i=1
si2 +(2sp +1)si
si+1
si

q 1

.
(2.1 )
Here the colored Jones polynomial is normalized such that Junknot (N ) = 1.

We should note that the infinite sums in above expressions terminate at finite order due to
(q 1N )sp = 0 for sp N , and that JKp (N ) is a Laurent polynomial of q 1 with integral coefficients.
We set a quantum parameter q as in (1.4), q = e2ir/N , and study an asymptotic behavior of
the quantum invariant in a limit N (1.5). A limit in a case of r = 1, i.e., q = exp(2i/N),
corresponds to the volume conjecture [3,4].
Proposition 2. In a limit (1.5) we have

dxp1
dx0
N
exp
HKp x0 , . . . , xp1 , m2 + o(N ) .
JKp (N ) =
x0
xp1
2ir
C
Here m is defined by (1.7), and we have the H -functions as

HKp>0 x0 , . . . , xp1 , m2
=
p1

i=1
2
log(xi /x0 ) + Li2 m2 + Li2 1/m2
(2.2)
189
2
Li2 x0 /m2 Li2 m2 x0 Li2 (x0 ) +
Li2 (xi /xi+1 ) (p 1)
6
p1
(2.3)
i=0
and

HKp<0 x0 , . . . , xp1 , m2
= (log x0 )2
p1

2
log(xi /x0 ) + Li2 m2 + Li2 1/m2
i=1
Li2 x0 /m
Li2 m x0 Li2 (x0 )

2
p1

Li2 (xi+1 /xi ) + (p 1)
i=0
2
,
6
(2.3 )
where we have set xp = 1. The integration path C is denoted as xi = e2(r)ti +2i (r)ti with
0 tp1 t1 t0 1.
Proof. From the definition of the q-product, we have in a limit N (see e.g. Refs. [21,23])
n
log 1 xe2i N r
log (xq)n =
k=1
2i Nn r
N
=
2ir

log 1 xet dt + o(N )
N
=
(2.4)
Li2 (x) Li2 xq n + o(N ).
2ir
In the second equality we have applied the EulerMaclaurin formula (1.10) for f (t) = log(1
xet ), and used the Euler dilogarithm function in the last equality. We thus have

m
nm
n
2
N
n
(2.5)
Li2 q
= exp
Li2 q + Li2 q
+ o(N )
m q
2ir
6
due to Li2 (1) = 2 /6. The error o(N ) comes from an analytic function of n and m as can be
seen from the identity (1.10).
We use (2.4) and (2.5) in (2.1). Setting
q N = m2
we have
JKp>0 (N ) =

Nsp s1 0

2ir sp
N
2
N
1
e
m
Li
exp
Li2
+
Li
2
2
2ir
m2
m2
p1

sp

Li2 m2 e2ir N +
2ir
i=1
+ Li2 e
2ir
s1
N
Li2 e
2ir
sp
N
si sp
N
2
si+1 si

+ Li2 e2ir N
2
(p 1)
6

+ o(N )
190

=
1t0 t1 tp1 0

2irt0

N
1
e
dti exp
Li2 m2 + Li2
Li
2
2
2ir
m
m2
2
Li2 m2 e2irt0 +
2ir(tpi t0 ) + Li2 e2ir(tp1i tpi )
p1
i=1
2
+ Li2 e2irtp1 Li2 e2irt0 (p 1)
+ o(N )
6

p1
dxi

N
1
x0
=
exp
Li2 m2 + Li2
Li
Li2 m2 x0
2
2
2
xi
2ir
m
m

i=0
p1

2
log(xi /x0 ) +
Li2
p2
i=1
2
(p 1)
6
j =0
xj
xj +1

Li2 (x0 ) + Li2 (xp1 )

+ o(N ) .
In the second equality, we have again applied the EulerMaclaurin formula (1.10), and set
si
tpi = .
N
In the last equality, we set
xi = exp(2irti )
and the integration path C is what is defined in Proposition 2. We then obtain the H -function (2.3).
For Kp<0 case, computation is essentially same with the above, and we omit the detail.
Finally we obtain the H -function defined in (2.3 ). 2
One of our main results is as follows.
Theorem 3. GVC (Conjecture 1) is true for twist knot Kp . Namely, the function HKp (x0 , . . . ,
xp1 , m2 ) defined in (2.3) and (2.3 ) is the NeumannZagier potential function [16] for the
p-twist knot Kp under a constraint (1.8a). An algebraic equation of and m, which is given
by eliminating x from (1.8a) and (1.8b), coincides with the A-polynomial of the twist knot Kp .
Note that a constraint (1.8a) denotes a saddle point condition of the integral (2.2) in the large
N limit, N .
To prove this theorem we rewrite a set of Eqs. (1.8) for Kp>0 as
1 m2 x0
= ,
x0 m2

p1

x0
x0
2(p1)
2
2
1
xi = x 0
(1 x0 ) 1 m x0 1 2 ,
x1
m
i=1

xi
xi1
x02 1
= xi2 1
for i = 1, . . . , p 1.
xi+1
xi
(2.6a)
(2.6b)
(2.6c)
191
The first equation is solved as

x0 =
1 + m2
.
+ m2
(2.7)
When we set
xpk = x0 Ck (x0 )
(2.8)
for k = 1, 2, . . . , p 1, we see that a rational function Ck (x) is recursively defined by

Ck+2 (x) = Ck+1 (x)
1
Ck+1 (x)
1
,
Ck (x)
(2.9)
where the first two rational functions are given by

1
1 (1 x)(1 m2 x)(1 x/m2 )
,
C1 (x) =
.
x
x
In the same way, a set of Eqs. (1.8) for a negative case, Kp<0 , is explicitly written as
C0 (x) =
1 m2 x0
= ,
x0 m2
p1

x1
x0
2p
xi2 1 2 1 m2 x0 (1 x0 ) = x0 1
,
x0
m
i=1

xi
xi+1
2
2
x0 1
= xi 1
, for i = 1, . . . , p 1.
xi1
xi
(2.6a)
(2.6b )
(2.6c )
In this case x0 is also fixed by (2.7). When we also set

(2.8 )
xpk = x0 Ck (x0 )
for k = 1, 2, . . . , p 1, a rational function Ck (x) is recursively defined from
Ck2 (x) Ck1 (x)
= Ck1 (x)Ck2 (x),
Ck1 (x) Ck (x)
(2.9 )
where the initial conditions are

C0 (x) =
1
,
x
C1 (x) =
m2 x 2
m2 x
.
(1 x)(1 m2 x)(m2 x)
As a result, a set of Eqs. (1.8) reduces to an algebraic equation of and m;

Cp (x0 ) = 1,
(2.10)
where the rational function Cp (x) is recursively computed as above, and x0 is defined in (2.7).
Then to prove Theorem 3, we must show that (2.10) gives the A-polynomial of the twist knot
Kp .
We introduce rational function C p (, m) by

1 Cp (x0 ) = (1 ) 1 m2 C p (, m).
(2.11)
192
The difference equations (2.9) and (2.9 ) are transformed into the recursion relation for C k (, m);
in a case of positive twist knot Kp>0 , we have
C k C k+1
= 1 (1 ) 1 m2 C k (, m) 1 (1 ) 1 m2 C k+1 (, m) (2.12)
C k+1 C k+2
with initial conditions
1
,
1 + m2
and for a negative case
C 0 (, m) =
C 1 (, m) =
+ m6
m2 ( + m2 )2

C k2 C k1
= 1 (1 ) 1 m2 C k1 1 (1 ) 1 m2 C k2
Ck1 Ck
(2.12 )
with
C 0 (, m) =
C 1 (, m)
=
1
,
1 + m2
+ m2 + m4 + 2m4 + 2 m4 + m6 m8

.
+ 2 + 2m2 2 m2 + m4 + 2m4 + 22 m6 + 3 m6 m8 + 22 m8 + m10 2 m10
Proposition 4. Rational function satisfying (2.12) is solved as

C p (, m) =
Ap (, m)
.
Bp (, m)
(2.13)
Here polynomials Ap (, m) and Bp (, m) are recursively defined as follows by use of a polynomial Z(, m) defined by
Z(, m) = + 2 + 2m2 2 m2 + m4 + 2 m4 m6 + 2m6 + m8 m8 :
a positive case, i.e., twist knots Kp>0

Ap+1 (, m)
Bp+1 (, m)

Z(, m)
=
m2 ( + m2 )2 (1 )(1 m2 )

Ap (, m)
M+
Bp (, m)
with
A1 (, m)
B1 (, m)

=
+ m6
2
m ( + m2 )2
(1 m2 )( m4 )
m2 ( + m2 )2

Ap (, m)
Bp (, m)
(2.14)
(2.15)

;
a negative case, i.e., twist knots Kp<0

Ap1 (, m)
Bp1 (, m)

m2 ( + m2 )2
=
(1 )(1 m2 )m2 ( + m2 )2
(1 m2 )( m4 )
Z(, m)

Ap (, m)
Bp (, m)

M
Ap (, m)
Bp (, m)
193
(2.15 )
with an initial condition

1
A0 (, m)
.
=
1 + m2
B0 (, m)
We note that we have

4
(M )t y M y = m4 + m2 ,
(2.16)

4
2 4
M+ M = m + m
(2.17)

with the Pauli spin matrix y = 0i i
, and that the characteristic polynomial of M is given by
0

F (x) = x 2 x 2 + m4 1 + m4 + 1 + 2m2 + 2m4 + 2m6 m8

4
+ m4 + m2 .
(2.18)
Proof of Proposition 4. We assume that C p (, m) is defined by (2.13), and that the polynomials
Ap and Bp satisfy (2.15) and (2.15 ). Hereafter we use Ap = Ap (, m) and Bp = Bp (, m) for
brevity.
We first prove a positive case p > 0. We see from (2.15) that

2
(1 ) 1 m2 Ap Bp = m2 + m2 Bp+1 ,

4
Ap Bp+1 Bp Ap+1 = m4 + m2 (Ap1 Bp Bp1 Ap ).
With these identities we find that

Ap+1 Ap+2
Bp (1 ) 1 m2 Ap Bp+1 (1 ) 1 m2 Ap+1
Bp+1
Bp+2
= Ap Bp+1 Bp Ap+1 ,
which proves (2.12).
For the negative case p < 0, we easily see from a recursion relation (2.15 ) that
2

Bp1 (1 ) 1 m2 Ap1 = m2 + m2 Bp ,

4
Ap1 Bp2 Bp1 Ap2 = m4 + m2 (Ap Bp1 Bp Ap1 ),
which leads

Ap Ap1
Bp1 (1 ) 1 m2 Ap1 Bp2 (1 ) 1 m2 Ap2

Bp
Bp1
= Ap1 Bp2 Bp1 Ap2 .
This is nothing but (2.12 ).
Proposition 5. Polynomial Ap (, m) defined by (2.13) coincides with the A-polynomial for the
twist knot Kp .
194
Proof. We define a polynomial

X(, m) = + 2 + 2m2 + m4 + 2m4 + 2 m4 + 2m6 + m8 m8 .
(2.19)
The recursion relation (2.15) gives (p > 0)
4
Ap+1 (, m) = X(, m)Ap (, m) m4 + m2 Ap1 (, m)
(2.20)
with
A1 (, m) = + m6 ,
A2 (, m) = 2 3 22 m2 m4 22 m4 + m6 + 2 m8
2m10 2 m10 2m12 m14 + m14 .
For a negative case (p < 0) we also see that the recursion relation (2.15 ) reduces to

4
Ap1 (, m) = X(, m)Ap (, m) m4 + m2 Ap+1 (, m)
(2.21)
with
A0 (, m) = 1,
A1 (, m) = + m2 + m4 + 2m4 + 2 m4 + m6 m8 .
These recursion relations, (2.20) and (2.21), coincide with those for the twist knot Kp derived
in Ref. [20], and we can conclude that the polynomial Ap (, m) defined by (2.13) is the Apolynomial for the twist knot Kp . 2
These propositions indicate that an algebraic equation (2.10), which is a consequence of (1.8)
for the twist knot Kp , is nothing but an algebraic equation of the A-polynomial for the twist knot
Kp ;
Ap (, m) = 0
(2.22)
if we suppose = 1 and
= 1. As we have defined the parameter , which now represents
the eigenvalue of the longitude of the boundary torus of knot, by a derivative of the function
HKp (x, m2 ) with respect to m, we can identify the H -function with a constraint (1.8a) as the
NeumannZagier potential function for the twist knot Kp . This proves the statement of Theorem 3, i.e., the H -function defined from the large-N limit of the N -colored Jones polynomial
(1.6) is the NeumannZagier potential function under a constraint (1.8a), and it gives the Apolynomial with a condition (1.8b).
This fact may support the volume conjecture [3,4] that the hyperbolic volume of the knot
complements dominates an asymptotics of the colored Jones polynomial, as (1.8a) denotes the
saddle point condition of the integral (1.6). To check the volume conjecture partly numerically,
we use a fact that under a constraint (1.8a) the H -function defined by (2.3) and (2.3 ) becomes
m2
|p|1

D(xi /xi+1 ),
Im HKp x0 , . . . , x|p|1 , m2 = 1 = 3D(1/x0 ) +
(2.23)
i=0
where (x0 , . . . , x|p|1 ) is a solution of (2.6b)(2.6c), or (2.6b )(2.6c ), with m2 = 1. Here we

have used the BlochWigner function D(z) defined by
D(z) = Im Li2 (z) + arg(1 z) log |z|,
which denotes the hyperbolic volume of the ideal tetrahedron with modulus z.
In the case of m2 = 1 we can simplify those equations as follows.
(2.24)
195
Proposition 6. We consider the saddle point equations (1.8a), i.e., (2.6b)(2.6c) or (2.6b )
(2.6c ), of the H -function in a case of m2 = 1. Let the polynomial Vk (z) be defined by (k > 0)
Vk (z) =

2k

k + j
j =0
z ,
Vk (z) = 1 +
2k1

j =1

k + j 1
2 zj .
j
(2.25)
Positive case (p > 0);

Eqs. (2.6b) and (2.6c) with m2 = 1 are solved as
xpk = x0
Vk (1 x0 )
,
Vk1 (1 x0 )
(2.26)
for k = 1, . . . , p, and x0 is a solution of Vp (1 x0 ) = Vp1 (1 x0 ), i.e.,

Vp (1 x0 ) = 0.
(2.27)
Negative case (p < 0);

Eqs. (2.6b ) and (2.6c ) with m2 = 1 are solved as
xpk = x0
Vk (x0 1)
,
Vk1 (x0 1)
(2.26 )
for k = 1, . . . , p, and x0 is a solution of Vp (x0 1) = Vp1 (x0 1), i.e.,

(2.27 )
Vp (x0 1) = 0.
Proof. We see that the polynomials Vp (z) satisfy the recursion relations;
Vp (z) Vp1 (z) = zVp (z),
Vp1 (z) Vp (z) = zVp (z),
which gives the 3-term relations

Vp+1 (z) z2 + 2 Vp (z) + Vp1 (z) = 0,

Vp1 (z) z2 + 2 Vp (z) + Vp+1 (z) = 0.
(2.28)
(2.29)
To complete the proof, we need to show that the solution of (2.9) and (2.9 ) with m2 = 1
Vk (x0 1)
k (1x0 )
is given by VVk1
(1x0 ) and Vk1 (x0 1) respectively, i.e., the polynomial satisfies the bilinear
equation

2
Vk+2 (z) Vk (z) Vk+1 (z) = Vk+1 (z) Vk1 (z) Vk (z) = z3 ,

2
Vk2 (z) Vk (z) Vk1 (z) = Vk1 (z) Vk+1 (z) Vk (z) = z3 .
This can be done easily by induction using (2.29).
We note that the polynomials Vk (z) are written as a sum of the hypergeometric functions,

z2
k, k + 1 z2
1 k, k + 1
;
;
+ kz 2 F1
,
Vk (z) = 2 F1
1
3
4
4
2
2

z2
1 k, k z2
1 k, k + 1
;
F
;
+
kz
.
Vk (z) = 2 F1
(2.30)
2 1
1
3
4
4
2
2
196
Table 1
Hyperbolic volume of the complement of the twist knot Kp coincides with the largest
value of Im HKp . Given are values of x0 which give the hyperbolic volume Vol(S 3 \ K)
by (2.31) or (2.31 ). Knot Kp=1 is the left-hand trefoil, which is not hyperbolic
p
Im HKp |m2 =1 = Vol(S 3 \ Kp )
x0
5
4
3
2
1
2
3
4
5
3.57388
3.52620
3.42721
3.16396
2.02988
2.82812
3.33174
3.48666
3.55382
0.99151 1.91177i
0.98405 1.86641i
0.96453 1.77530i
0.89512 1.55249i
0.50000 0.86603i
1.21508 1.30714i
1.05818 1.69128i
1.02317 1.82953i
1.01144 1.89257i
With these results we conclude that when x0 is a solution of (2.27) or (2.27 ) we have (p > 0)

Im HKp x0 , . . . , xp1 , m2 = 1

p1
Vj +1 (1 x0 )Vj 1 (1 x0 )
V1 (1 x0 )
D
+ D x0
= 3D(1/x0 ) +
(2.31)
V0 (1 x0 )
(Vj (1 x0 ))2
j =1
and

D
Im HKp x0 , . . . , xp1 , m2 = 1 = 3D(1/x0 ) +
p
j =1

(Vj (x0 1))2
.
Vj +1 (x0 1)Vj 1 (x0 1)
(2.31 )
See Table 1 for numerical computation. We have checked that the largest value of Im HKp (x0 ,
. . . , x|p|1 , m2 = 1) among solutions of (1.8a) coincides with the hyperbolic volume of the complement of Kp [24] as was proposed as volume conjecture (see Ref. [25] for an ideal triangulation
of the complement of the twist knots). We have plotted zeros of the polynomial Vk (z) in Fig. 2
for convention.
In a limit |p| , we may read off from both the above table and a numerical computation
that x0 1 2i (see Fig. 2). It is known geometrically that in the limit of |p| the hyperbolic volume of the twist knot Kp is that of the Whitehead link (Fig. 3) [26], which coincides
with the hyperbolic volume of the regular ideal octahedron 4D(i) = 3.66386237670887606 . . ..
Applying identities
Vp (2i) = (1)p (2p + 1 2pi),
Vp (2i) = (1)p (2p + 1 + 2pi)
(2.32)
> 0, which result from the ChuVandermonde identity, to (2.31) and (2.31 ), we may obtain
for p
formulae for the BlochWigner function;

1
i 2
4D(i) = 3D(2i) +
D k+ +
4 4
k=0

3
i 2
= 3D(2i)
D k +
.
4 4
k=0
(2.33)
(2.33 )
197
Fig. 2. Zeros of the polynomials Vk (z) for k = 5( ), 10( ), 30( ), 50( ). We only plot zero points in the upper half plane.
Fig. 3. Whitehead link.
In fact these identities follow from the pentagon identity,1

D z2 = 2 D(z) D(z + 1) .
We should remark that we do not have a mathematically rigorous proof whether we may apply
the saddle point method to the integral (2.2) and the saddle point dominates the asymptotics of
the colored Jones polynomial, though numerical experiments [27] supports our computations.
3. Torus knot
We apply above story to the colored Jones polynomial for the torus knot T2,2p+1 (we study a
case of p > 0. See Fig. 4); we reveal a relationship between the A-polynomial and the H -function.
We first recall the q-hypergeometric expression of the colored Jones polynomial for the torus
knot.
1 Anatol N. Kirillov kindly pointed out this fact.
198
Fig. 4. Torus knots T2,2p+1 with p > 0 are depicted. From left to right, p = 1 (right-hand trefoil), p = 2 (Solomons
seal knot), and p = 3.
Proposition 7. (See [8].) The N -colored Jones polynomial JK (N ) for the torus knot K = T2,2p+1
is written as
JT2,2p+1 (N ) = q p(1N
2)
p1
q (p1)sp (sp +1)
sp s2 s1 0
si2 (2sp +1)si
i=1
si+1
si
(q 1N )sp (q 1+N )sp

(q)sp

.
(3.1)
Here the colored Jones polynomial is normalized to be Junknot (N ) = 1.

Applying (2.4) to above expression, we easily obtain the H -function for the torus knot.
Proposition 8. The asymptotic behavior of the N -colored Jones polynomial for the torus knot
T2,2p+1 is written in an integral form in a limit N as

dxp1
dx0
N
2
x0 , . . . , xp1 , m + o(N ) ,
exp
HT
JT2,2p+1 (N ) =
x0
xp1
2ir 2,2p+1
C
(3.2)
where m is defined by (1.7), and

HT2,2p+1 x0 , . . . , xp1 , m2
2
log(xi /x0 )
= p log m2 +
p1
i=1

2
Li2 (xi /xi+1 ) p .
+ Li2 m2 + Li2 1/m2 Li2 x0 /m2 Li2 m2 x0 +
6
i=0
(3.3)
p1
The main result for the torus knots is as follows.

Theorem 9. GVC stated as Conjecture 1 is true for torus knot T2,2p+1 ; the H -function (3.3)
under a constraint (1.8a) is the NeumannZagier potential function for the torus knot T2,2p+1 ,
and an algebraic equation of and m, which is given by eliminating x from (1.8a) and (1.8b),
coincides with the A-polynomial for the torus knot T2,2p+1 .
199
Proof. A set of Eqs. (1.8) gives

1 m2 x0
= ,
m4p (x0 m2 )

p1
x0
x0
2p2
2
2
x0
1 m x0 1 2 = 1
xi ,
x1
m
i=1

xi1
xi
= x02 1
for i = 1, 2, . . . , p 1.
xi2 1
xi
xi+1
(3.4a)
(3.4b)
(3.4c)
In this case we have

x0 =
1 + m4p+2
m2 (1 + m4p2 )
(3.5)
and we see that

xk = x0 Cpk (x0 ),
(3.6)
where Ck is recursively solved as

(1 m2 x)(m2 x)
Ck (x),
Ck+1 (x) = 1 2
m (C1 (x) Ck (x))2
1
C0 (x) = .
x
Then the algebraic equation for and m may reduce to
(3.7)
Cp (x0 ) = 1,
(3.8)
where x0 is solved in (3.5). In this case we have

1 Cp (x0 ) =
(1 )(1 m2 )
1 + m2
and we obtain an unwanted solution = 1 or m2 = 1. This suggests that a solution of (3.4) is

rather given by
x0 = 0,
xi = 1,
for i > 0.
This gives an algebraic equation as

AT2,2p+1 (, m) = 1 + m4p+2 ,
which is the A-polynomial for the torus knot T2,2p+1 [5,20].
(3.9)
2
We note that studied in Ref. [28] is a rigorous treatment of asymptotics when r is near unity.
In the case of r = 1, i.e., an exact asymptotic expansion in N of the N -colored Jones
polynomial for the torus knot Ts,t with q being the N th root of unity, q = exp(2i/N), was
studied in Refs. [2931] (see also Refs. [3234]), and the invariant was identified with the Eichler
integral of the modular form with half-integral weight 1/2 which is related to the character of the
Virasoro minimal model M(s, t).
200
We shall check our theorems in detail by taking an example, the trefoil T2,3 . The N -colored
Jones polynomial for the right-hand trefoil was computed explicitly in Refs. [19,35], and collecting these results we have
q nN q 1N n
(3.10a)
q n(n+2) q 1N n q 1+N n
(3.10b)
JT2,3 (N ) = q 1N
n=0

n=0
= q 1N

(q 1N )n (q 1+N )n
n=0
(q)n
(3.10c)
All these infinite sums terminate at finite sums due to (q 1N )k = 0 for k N > 0. Note that
those q-hypergeometric type expressions are respectively from Refs. [8,35], (2.1) with p = 1
replacing q by q 1 , and (3.1) with p = 1.
We now compute the H -function from those different expressions of the colored Jones polynomial. By use of (2.4), we obtain the H -functions from three expressions (3.10) as follows;

1
x
2
Li2
,
Ha x, m = (log x) log m + Li2
(3.11a)
2
m
m2

1
x
+ Li2 m2 Li2
Li2 m2 x ,
Hb x, m2 = (log x)2 + Li2
(3.11b)
2
2
m
m

1
x
Hc x, m2 = log m2 + Li2
+ Li2 m2 Li2
Li2 m2 x
2
2
m
m
2
(3.11c)
.
6
See that the H -function depends on expression of the colored Jones polynomial, and that it is not
invariant itself. Although, a set of Eqs. (1.8) gives the invariant, the A-polynomial, as follows:
+ Li2 (x)
(a) We have from (3.11a)

1 + m2
= ,
(m2 x)x
m2 x
= 1,
m4
from which we have x = (1 m2 )m2 . We thus obtain an algebraic equation,

A(, m) = 0,
with A(, m) = 1 + m6 .
This is the A-polynomial for the (right-hand) trefoil [5].

(b) Substituting (3.11b) for (1.8), we get
1 + m2 x
= ,
m2 x
This gives x =
1+m2
,
m2 +
(m2 x)(1 m2 x)
= 1.
m2 x 2
and an equation of and m is written as
( + m2 )(1 + m6 )
= 0,
m2 (1 + m2 )3
which suggests (3.12).
(3.12)
201
(c) We have
1 m2 x
= ,
m4 (m2 x)
which gives x =
1+m6
m2 (1+m2 )
(m2 x)(1 m2 x)
= 1,
m2 (1 x)
and
(1 )(1 + m6 )
= 0.
(1 + m2 )(1 m4 )
We may assume = 1, and we obtain the A-polynomial (3.12).
To conclude, all three H -functions given from an asymptotics of three expressions (3.10), give
the A-polynomial for the trefoil with constraints (1.8).
4. Concluding remarks
We have studied generalized volume conjecture proposed by Gukov [13] based on explicit
expression of the N -colored Jones polynomial JK (N ) for twist knot K = Kp . We have shown
that the asymptotics of JKp (N ) in N is related to zero locus of the A-polynomial which is
computed explicitly in Ref. [20]. This result indicates that the color N may have a geometrical
information, and that it denotes the SL(2; C) representation of the meridian of the boundary
torus of knot. We have also checked numerically Kashaevs volume conjecture. Fruitful will be
further researches on geometrical interpretation of quantum invariants such as the colored Jones
polynomial.
Acknowledgements
The author would like to thank J. Kaneko, A.N. Kirillov, H. Murakami, T. Takata, and
Y. Yokota for discussions. This work is supported in part by Grant-in-Aid from the Ministry
of Education, Culture, Sports, Science and Technology of Japan.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
E. Witten, Quantum field theory and the Jones polynomial, Commun. Math. Phys. 121 (1989) 351399.
M. Mario, ChernSimons theory and topological strings, Rev. Mod. Phys. 77 (2005) 675720.
R.M. Kashaev, The hyperbolic volume of knots from quantum dilogarithm, Lett. Math. Phys. 39 (1997) 269275.
H. Murakami, J. Murakami, The colored Jones polynomials and the simplicial volume of a knot, Acta Math. 186
(2001) 85104.
D. Cooper, M. Culler, H. Gillet, D.D. Long, P.B. Shalen, Plane curves associated to character varieties of 3-manifolds, Invent. Math. 118 (1994) 4784.
D. Cooper, D.D. Long, Representation theory and the A-polynomial of a knotknot theory and its applications,
Chaos Solitons Fractals 9 (1998) 749763.
S. Garoufalidis, On the characteristic and deformation varieties of a knot, Geom. Topol. Monogr. 7 (2004) 291309.
K. Hikami, Difference equation of the colored Jones polynomial for the torus knot, Int. J. Math. 15 (2004) 959965.
T.T.Q. Le, The colored Jones polynomial and the A-polynomial of knots, Adv. Math. 207 (2006) 782804.
T. Takata, The colored Jones polynomial and the A-polynomial for twist knots, math.GT/0401068.
R. Gelca, On the relation between the A-polynomial and the Jones polynomial, Proc. Am. Math. Soc. 130 (2002)
12351241.
R. Gelca, J. Sain, The noncommutative A-ideal of a (2, 2p + 1)-torus knot determines its Jones polynomial, J. Knot
Theory Ramifications 12 (2003) 187202.
202
[13] S. Gukov, Three-dimensional quantum gravity, ChernSimons theory, and the A-polynomial, Commun. Math.
Phys. 255 (2005) 577627.
[14] H. Murakami, Y. Yokota, The colored Jones polynomials of the figure-eight knot and its Dehn surgery spaces,
math.GT/0401084.
[15] S. Garoufalidis, T.T.Q. Le, The colored Jones function is q-holonomic, Geom. Topol. 9 (2005) 12531293.
[16] W.D. Neumann, D. Zagier, Volumes of hyperbolic three-manifolds, Topology 24 (1985) 307332.
[17] T. Yoshida, The -invariant of hyperbolic 3-manifolds, Invent. Math. 81 (1985) 473514.
[18] G. Masbaum, Skein-theoretical derivation of some formulas of Habiro, Algeb. Geom. Topol. 3 (2003) 537556.
[19] K. Habiro, On the quantum sl2 invariants of knots and integral homology spheres, Geom. Topol. Monogr. 4 (2002)
5568.
[20] J. Hoste, P.D. Shanahan, A formula for the A-polynomial of twist knots, J. Knot Theory Ramifications 13 (2004)
193209.
[21] G.E. Andrews, The Theory of Partitions, AddisonWesley, London, 1976.
[22] L. Lewin, Polylogarithms and Associated Functions, North-Holland, New York, 1981.
[23] B. Richmond, G. Szekeres, Some formulas related to dilogarithms, the zeta function and the AndrewsGordon
identities, J. Aust. Math. Soc. (Series A) 31 (1981) 362373.
[24] J. Weeks, SnapPea, http://www.geometrygames.org/SnapPea/.
[25] M. Sakuma, J. Weeks, Examples of canonical decompositions of hyperbolic link complements, Jpn. J. Math. 21
(1995) 393439.
[26] J. Milnor, Hyperbolic geometry: The first 150 years, Bull. Am. Math. Soc. 6 (1982) 924.
[27] K. Hikami, Volume conjecture and asymptotic expansion of q-series, Exp. Math. 12 (2003) 319337.
[28] H. Murakami, Asymptotic behaviors of the colored Jones polynomials of a torus knot, Int. J. Math 15 (2004) 547
555.
[29] D. Zagier, Vassiliev invariants and a strange identity related to the Dedekind eta-function, Topology 40 (2001)
945960.
[30] K. Hikami, q-series and L-functions related to half-derivatives of the AndrewsGordon identity, Ramanujan J. 11
(2006) 175197.
[31] K. Hikami, A.N. Kirillov, Torus knot and minimal model, Phys. Lett. B 575 (2003) 343348.
[32] K. Hikami, Quantum invariant for torus link and modular forms, Commun. Math. Phys. 246 (2004) 403426.
[33] K. Hikami, On the quantum invariant for the Brieskorn homology spheres, Int. J. Math. 16 (2005) 661685.
[34] K. Hikami, A.N. Kirillov, Hypergeometric generating function of L-function, Slaters identities, and quantum knot
invariant, Algebra i Analiz 17 (2005) 190208.
[35] T.T.Q. Le, Quantum invariants of 3-manifolds: Integrality, splitting, and perturbative expansion, Topol. Appl. 127
(2003) 125152.
Current exchanges and unconstrained higher spins

D. Francia a , J. Mourad b , A. Sagnotti c,
a Department of Fundamental Physics, Chalmers University of Technology, S-412 96 Gteborg, Sweden
b AstroParticule et Cosmologie (APC) 1 , Universit Paris VII, Campus Paris Rive Gauche,
10, rue Alice Domon et Leonie Duquet, F-75205, Paris cedex 13, France
c Scuola Normale Superiore and INFN, Piazza dei Cavalieri, 7, I-56126 Pisa, Italy

Abstract
The (Fang) Fronsdal formulation for free fully symmetric (spinor-) tensors rests on ( -) trace constraints
on gauge fields and parameters. When these are relaxed, glimpses of the underlying geometry emerge: the
field equations extend to non-local expressions involving the higher-spin curvatures, and with only a pair of
additional fields an equivalent minimal local formulation is also possible. In this paper we complete the
discussion of the minimal formulation for fully symmetric (spinor-) tensors, constructing one-parameter
families of Lagrangians and extending them to (A)dS backgrounds. We then turn on external currents, that
in this setting are subject to conventional conservation laws and, by a close scrutiny of current exchanges in
the various formulations, we clarify the precise link between the local and non-local versions of the theory.
To this end, we first show the equivalence of the constrained and unconstrained local formulations, and
then identify a unique set of non-local Lagrangian equations which behave in the same fashion in current
exchanges.
E-mail addresses: francia@chalmers.se (D. Francia), mourad@apc.univ-paris7.fr (J. Mourad), sagnotti@sns.it

(A. Sagnotti).
1 Unit mixte de Recherche du CNRS (UMR 7164).
doi:10.1016/j.nuclphysb.2007.03.021
204
D. Francia et al. / Nuclear Physics B 773 [PM] (2007) 203237
1. Introduction
The (Fang) Fronsdal formulation of free higher-spin dynamics [1] rests on trace (or -trace)
constraints for gauge fields and corresponding gauge parameters.2 While these algebraic conditions do not conflict with Lorentz covariance, it is quite natural to try and forego them, aiming at
formulations that are closer in spirit to the familiar ones for low-spin fields. The issues are then
how to do it and what one gains from this extension.
A direct, if somewhat unconventional, way to forego the trace constraints of [1] is via a set of
non-local equations involving the higher-spin curvatures of de Wit and Freedman [6]. This was
first done in [7] for the set of higher-spin fields originally considered in [1], totally symmetric
tensors 1 ...s , that here will be loosely referred to as spin-s fields, and for corresponding totally
symmetric tensor-spinors 1 ...n , that here will be loosely referred to as spin-(n + 1/2) fields.
In exploring new directions for higher spins, it is customary to begin by restricting the attention
to this class of fields, although they do not exhaust the available choices in D > 4, since the
resulting analysis suffices to display some key features of the problem in a relatively handy
setting. The generalization to tensors of mixed symmetry is then an important further step, and
is also crucial for establishing a proper link with string theory. For the non-local formulation this
was first achieved to some extent in [8], and the resulting properties were recently explored in
further detail in [9].
A less direct, but more conventional, way to forego the trace constraints is via the introduction
of compensators. This was first achieved by Pashnev, Tsulaia, Buchbinder and others [10], using a
BRST [11] formulation inspired by string field theory but adapted to the description of individual
massless higher-spin modes. This is to be contrasted with the conventional BRST formulation
of string field theory [12], that associates to a given higher-spin field a whole family of other
fields belonging to lower Regge trajectories. The BRST formulation of individual higher-spin
fields of [10] is rather complicated, and makes use of O(s) fields to describe the propagation of a
single set of spin-s modes. On the other hand, it is possible to present compensator equations
for individual higher-spin fields that, while non-Lagrangian, are fully gauge invariant without
any trace constraints and involve only two fields. For instance, in the bosonic case, aside from
1 ...s they involve a single compensator field 1 ...s3 , which first emerges for s = 3. The
relation with free string field theory was clarified in [13], where the precise link between the
compensator equations for higher-spin fields and the triplet systems [1315] emerging from
free string field theory in the low-tension limit was displayed. In [16] these equations were then
related to the results in [10], from which they can be recovered by a partial gauge fixing, and
were also extended to (A)dS backgrounds. Interestingly, the simplest flat-space equation in this
set, for a spin-3 field and a scalar compensator ,
2 ( + ) + ( + ) = 3 ,
(1.1)
where, as in the rest of the present paper, primes denote traces, was first considered by
Schwinger long ago [17].3 More recently, a minimal Lagrangian formulation for the compensator equations was also obtained in [18]: rather than O(s) fields as in [10], it involves only
a Lagrange multiplier 1 ...s4 , that first emerges for spin s = 4, aside from the basic field
2 The web site http://www.ulb.ac.be/sciences/ptm/pmif/Solvay1proc.pdf contains the Proceedings of the First Solvay
Workshop on Higher-Spin Gauge Theories [2], with some contributions closely related to the present work [35] and
many references to the original literature.
3 We are grateful to G. Savvidy for calling Schwingers result to our attention.
205
1 ...s and the compensator 1 ...s3 . In contrast with the non-local case, the generalization of
this minimal local formulation to tensors of mixed symmetry is not known at the present time,
although its key features may be anticipated from the known constrained gauge transformations.
While it is certainly interesting and instructive to explore them, it is difficult to assess the
relative virtues of different formulations for free higher-spin fields before the systematics of
higher-spin interactions is further clarified. There have been many attempts in this direction over
the years [19], that have marked the history of the subject as a result of the unexpected difficulties that were readily met, but this line of approach deserves further efforts and is being further
explored, with the help of more powerful techniques, in recent times [20]. It is fair to say that
we do not possess yet a general understanding of higher-spin interactions and of the underlying
geometry, but we do have at our disposal two important paradigmatic examples that are based
on a well-motivated algebraic setting. These are the Vasiliev equations, in their four-dimensional
form based on spinor oscillators [21] and in their more recent D-dimensional form based on
vector oscillators [22]. The Vasiliev equations are a set of first-order differential constraints involving a one-form master field A, that encodes an infinite family of 1 ...s via corresponding
generalized vielbeins and spin connections, and a zero-form master field , that collects their
Weyl curvatures and covariant derivatives, together with corresponding data for a scalar mode.
Due to the presence of the zero-form , the Vasiliev constraints generalize the more conventional
notion of free-differential algebra [23] in a non-trivial fashion, and embody a description of free
higher-spin modes in an unfolded form, via infinitely many auxiliary fields that subsume their
local data. Both the four-dimensional formulation of [21] and the D-dimensional formulation of
[22] are not Lagrangian, but while the former is fully based on the constrained Fronsdal form
of the free dynamics, this is not quite true for the latter. More precisely, the spinor oscillators
build consistent interactions for the frame version of Fronsdals formulation, which was first
discussed in [24], and can be gauge-fixed to its metric version but does not leave room for the
compensator of [13,16,18] or for the wider gauge symmetry of [10]. On the other hand, the field
equations of [22], as pointed out in [5], allow a non-dynamical off-shell variant devoid of trace
constraints and, if completed with a strong projection, turn into dynamical equations that at the
free level reduce precisely to a frame version of the compensator equations of [13,16,18]. The
potential open problems with the resulting interactions are still not fully sorted out at the present
time, and we refer the reader to [4,5] for further details, but there are reasons to believe that fully
interacting equations can be defined in this fashion, thus encoding naturally a gauge symmetry
not subject to Fronsdals trace constraints.
To summarize, even with our current incomplete grasp of the systematics of higher-spin interactions the unconstrained formulation of free higher-spin fields presents a number of attractive
features. First, and most notably, it embodies a direct link with higher-spin geometry, since its
minimal, non-local form of [7] rests directly on the higher-spin curvatures of [6] rather than on
lower connections. In addition, it links naturally to the BRST form of free string field theory, from
which it emerges, after a suitable truncation to the leading Regge trajectory, in the low-tension
limit [13,16]. Finally, and most importantly for the purposes of the present paper, it couples to
conserved currents, to be contrasted with the partially conserved currents of the FangFronsdal
formulation.
This paper is devoted to exploring some key features of the minimal unconstrained Lagrangian formalism of [18] that manifest themselves in the presence of external currents. However, we begin in Section 2 by reconsidering and simplifying somewhat the results of [18] for
bosons and fermions, that here are also extended to the interesting cases of (A)dS backgrounds.
206
While the extension of flat-space results to these more general backgrounds is an interesting
and well-defined problem in its own right, this is perhaps the place to spend, as in [5], some words
of caution against the naive identification of the spin-2 modes of the Vasiliev equations with the
gravitational field. The problem arises since, by virtue of their very field content, the Vasiliev
equations are naturally an effective description of the first Regge trajectory of the open bosonic
string, albeit in an unconventional and largely unexplored regime where it would experience a
gigantic collapse of all its massive modes to the massless level. It should be stressed that little is
presently known about this stringy regime, although the AdS/CFT correspondence, in the weak
gauge-coupling limit, provides indirect arguments for it [25]. The Vasiliev equations also allow a
ChanPaton [26] extension to matrix-valued modes, precisely as is the case for the open bosonic
string. Hence, they generally involve not a single graviton, but rather a whole multiplet of
spin-2 modes, so that the minimal Vasiliev model, whose interest was clearly stressed in [27]
and whose first non-trivial cosmological solution was recently presented in [28], is somehow
the counterpart in this context of the O(1) open string discussed by Schwarz in his 1982 review
[29], whose open sector contains indeed only even excitation levels. Should one then associate
the singlet spin-2 Vasiliev mode with gravity, that lies outside the open spectrum? Our present
knowledge does not allow sharp statements to this effect, and caution is called for. Indeed, while
the distinction between open and closed spectra is strictly in place for tensile strings, in the
largely unexplored low-tension limit a mixing between string states at different levels, and in
particular between the singlet part of the spin-2 open-string field and the closed-string graviton,
might take place, although a clear understanding of this important issue from a string vantage
point is lacking at present.
In a similar spirit, a closed-string analogue of the Vasiliev equations, which should rest on
a more subtle algebraic structure in order not to allow a ChanPaton extension, is not available
at the present time, and the search for it might provide important clues on low-tension string
regimes. These are clearly deep issues on which, unfortunately, we shall have little to say in the
present paper. Taking a close look at current exchanges, however, in Sections 3 and 4 we shall be
able to investigate in some detail to which extent the available options for formulating the free
higher-spin dynamics lead to the same counting of degrees of freedom. Thus, in Section 3 we
shall establish the direct equivalence between the constrained Fronsdal formulation of [1] and
the unconstrained local formulation of [18]. On the other hand, the non-local counterpart of the
unconstrained free theory is not fully determined. Many possible options for a non-local spin-s
Lagrangian equation exist and, as we shall see, the simple choice made in [7] for an Einsteinlike tensor does not provide the direct counterpart of the local Lagrangian equations of [18]. In
Section 4, however, we shall identify a different, unique form of the non-local Lagrangian field
equations which behaves exactly like the local formulations, thus arriving at a precise link between the local and non-local unconstrained forms of free higher-spin gauge theory. Let us stress
that the issue here is the correct coupling to external currents via the natural source term, J ,
and the Lagrangian equations that couple correctly to external currents will be more complicated
than those proposed in [7]. Still, in the absence of external currents they can be turned into the
non-local non-Lagrangian equations of [7],
1
R[p] ;1 2p+1 = 0
2p
for odd spins s = 2p + 1, and
1
R[p] ;1 2p = 0
2p1
(1.2)
(1.3)
207
for even spins s = 2p. Section 5 contains our conclusions, while Appendix A collects some
useful results concerning our implicit notation for symmetric tensors and our conventions.
2. Minimal Lagrangians for unconstrained higher spins
In this section we review the construction of the minimal unconstrained free Lagrangians for
fully symmetric higher-spin tensors and tensor-spinors in flat space presented in [18]. Our aim is
to streamline both the derivation and the resulting presentation, by properly stressing the role of a
few gauge invariant constructs playing the role of building blocks for the dynamical quantities
of interest. As we shall see, the resulting simplified form of the Lagrangians proves quite helpful
in extending the previous results to the interesting cases of (A)dS backgrounds.
2.1. Bosons
The definition of fully gauge invariant kinetic operators is a very convenient starting point for
the construction of the minimal bosonic Lagrangians, that we would like to reconsider and extend
here. For a rank-s fully symmetric tensor, that in the index-free notation of [18] (see Appendix A
for details and some examples) can be simply denoted by , one can begin by considering the
Fronsdal operator
F = 2 + 2 ,
(2.1)
where, here as elsewhere in this paper, a prime denotes a trace. Under the gauge transformation
= ,
(2.2)
F varies according to
F = 3 3 ,
(2.3)
denotes the trace of the gauge parameter . From F one can build the
where, as anticipated,
fully gauge invariant tensor
A = F 3 3 ,
(2.4)
where the spin-(s 3) compensator transforms as

=
(2.5)
under the tensor gauge transformation in Eq. (2.2), and one can then show that A satisfies the
Bianchi identity
3
1
A A = 3 ( 4 ).
(2.6)
2
2
With the mostly-positive spacetime signature that we shall use throughout, the minimal
bosonic Lagrangians of [18] can thus be conveniently recovered starting from

1
1
L0 = A A ,
(2.7)
2
2
which, on account of (2.1) and (2.6), varies as

3 s
s

L0 =
[ 4 ]
A 3
4
4 3
(2.8)
208
under the tensor gauge transformations (2.2) and (2.5). These terms can be compensated adding

3 s
s

[ 4 ],
A + 3
L1 =
(2.9)
4
4 3
so that the end result can be presented in the rather compact form

1
s
1
3 s
L = A A
[ 4 ],
A + 3
4
2
2
4 3
(2.10)
where, as in [18], the Lagrange multiplier transforms as

= .
(2.11)
We would like to stress that working insofar as possible with the gauge-invariant tensor A
has streamlined both the derivation and the final form of the Lagrangian (2.10) with respect to
[18]. One can actually do better defining, for the bosonic case, three independent gauge-invariant
tensors in terms of which all dynamical quantities of interest can be expressed. The first member
of this set is the tensor A introduced in (2.4), the second,
C 4 ,
(2.12)
defines the constraint, and finally the third,

1
B ( 22 ),
(2.13)
2
relates the Lagrange multiplier to the single combination of and that possesses an identical
gauge transformation. Actually, the very definition of B suggests that the Lagrangian (2.10) is
not the only possibility, and may be generalized by turning the coefficient of C into

1
s
1
3 s

Lk = A A
(2.14)
[ kB]C,
A + 3
4
2
2
4 3
so that (2.10) corresponds to the particular choice k = 0.
A more general analysis should involve additional quadratic terms mixing A and C tensors, as
well as terms quadratic in the tensor C, thus exhausting all possible gauge-invariant terms with at
most two derivatives for the physical field and at most four derivatives for the compensator .
In particular, given the identity
A = 32C + 3 C + 2 C ,
(2.15)
and since the only possibility to combine A and C without increasing the number of derivatives
in is via A C, it turns out that the only possibilities left are quadratic terms in the tensor C,
C2,
C2C,
C C,
C C,
(2.16)
together with their traces. Nonetheless, all these terms would only generate additional terms
linear in C in the equations for and whose role would be irrelevant once C is set to zero by
the field equation for .
We have thus arrived at a one-parameter family of gauge-invariant, unconstrained Lagrangians, whose field equations can be nicely expressed in terms of the three tensors A, B,
and C, and read
1
1
E (k) A A + (1 + k) 2 C + (1 k)2 B = 0,
2
4

3 s
k + 1
A
2 + 2 C + (k 1)(2 + )B = 0,
E (k)
2 3
2

s
E (k) 3
(1 k)C = 0,
4
209
(2.17)
where it should be noted that, in terms of the tensors (2.17), Eq. (2.14) can be cast in the particularly compact form
1
1
1
Lk = E (k) + E (k) + E (k).
(2.18)
2
2
2
For generic values of k the field equation for can be reduced to the compensator form A = 0,
while B = 0 determines the Lagrange multiplier in terms of the other fields, as was shown to
be case in [18] for k = 0. One is eventually left with the non-Lagrangian compensator equations
of [13,16],
A F 3 3 = 0,
C 4 = 0,
(2.19)
and these can be finally gauge-fixed to the Fronsdal form F = 0 and to Fronsdals double-trace
constraint = 0, making use of the trace of the gauge parameter to remove the compensator . This indicates that the unconstrained Lagrangians (2.14) provide a proper description of
higher-spin dynamics, just like Fronsdals constrained formulation.
The value k = 1 appears particularly interesting, since in this case the tensor C disappears
from the field equations for and . Finally, the choice k = +1 is somewhat degenerate, since in
this case disappears from the Lagrangian, and this implies, in its turn, that B disappears from
the field equations for and . Whereas naively this could be seen as a simplification, in this
case one can only derive directly from the field equations the condition
2C = 0,
(2.20)
which is not sufficient to remove the double trace.

In all cases, however, the divergence of E (k) vanishes when and are on-shell, as demanded by the Noether relation
1
1
E (k) = s E (k) s 3 E (k),
3 3
4 4
(2.21)
that implies that external sources coupled to the dynamical field are to be conserved.
As anticipated, this presentation of the Lagrangians has the virtue of leading rather directly
to their (A)dS deformations. These draw their origin from the basic (A)dS commutator for two
covariant derivatives acting on a vector field V , that reads
1
(g V g V ),
(2.22)
L2
where L denotes the (A)dS radius. For brevity, in the following we shall refer only to the AdS
case, since the corresponding results for dS backgrounds can be simply obtained by the formal
continuation of L to imaginary values.
[ , ]V =
210
In discussing higher-spin gauge fields in these curved backgrounds, it is convenient to begin

by considering the deformed AdS Fronsdal operator [30] (see also, for instance, [16])

1
(3 D s)(2 s) s + 2g ,
2
L
where D denotes the spacetime dimension and
FL = F
(2.23)
F = 2 + 2
(2.24)
is the AdS-covariantized Fronsdal operator, that transforms according to

4
g
L2
under the gauge transformation
FL = 3 3
(2.25)
= .
(2.26)
Even in this more general AdS setting the structure of the theory can be fully encoded in three
gauge invariant tensors, AL , BL and CL , that reduce to the previous expressions in the flat limit.
Assuming for the auxiliary fields the straightforward generalizations of the gauge transformations (2.5), (2.11),
= ,
= ,
(2.27)
the AdS generalizations of A and C,

4
g,
L2
CL = 4 ,
AL = FL 3 3 +
are rather simple, while the covariantization of B,

1
1
2
BL =
2
2

1

2 2 + 2g + (s 3)(5 s D)
L
(2.28)
(2.29)
(2.30)
is somewhat more tedious to obtain, since new terms appear in the gauge variation of
in this curved background.
The starting point for the construction of the bosonic Lagrangians is now

e
1
L0 = AL gAL ,
(2.31)
2
2
where e and g denote the determinant of the vielbein and the AdS metric. Proceeding as in flat
space and taking into account the deformed Bianchi identity,
1
3
2
AL AL = 3 CL + 2 gCL ,
2
2
L
one can finally arrive at the gauge invariant Lagrangians

e
1
s
4
3e s

L = AL gAL
2 CL
AL + 3e
4
2
2
4 3
L
(2.32)
(2.33)
211
that, as in the flat case, are special members of a whole family of gauge invariant Lagrangians,

e
1
s
4
3e s
Lk = AL gAL
2 kBL CL , (2.34)
AL + 3e
4
2
2
4 3
L
defined introducing in L the gauge-invariant coupling BL CL . From these Lagrangians one can
then derive the AdS field equations: for the fields and , they have the form of those obtained
in flat case, aside from the natural substitutions
(A, B, C, , ) (AL , BL , CL , g, ),
(2.35)
while the field equation for contains additional terms that depend on the tensor CL . The final
result reads
1
1
EL (k) AL gAL + (1 + k)g 2 CL + (1 k)g 2 BL = 0,
4
2

3 s
k+1
4

EL (k)
AL
2 CL + (k 1)(2 + g)BL 2 CL = 0,
2 3
2
L

s
(1 k)CL = 0.
EL (k) 3
(2.36)
4
More explicitly, in the case k = 0 the field equation for takes the form

1
1 2

EL AL g AL ( 4 )
2
2

2
1
1
+ g 2 + + 2 + 2
2
2
L

2
1
+ 2 g + 2 (s 3)(5 s D) + 4 .
L
L
(2.37)
Current conservation is guaranteed in this case by the AdS generalization of Eq. (2.21),
1
1
EL (k) = s gEL (k) s 3 EL (k),
3 3
4 4
(2.38)
a result that reflects the Noether identity signalling the gauge invariance of the action, which
indeed implies that

Lk
Lk
D
Lk
+
+
= 0.
d x s
(2.39)
2.2. Fermions
The fermionic case is more involved but does not add substantial novelties with respect to
what we have seen for bosons. As in the preceding subsection, we can now begin by simplifying
and generalizing the construction of gauge invariant Lagrangians for unconstrained free fields of
[18]. In doing so, we shall be able to stress the key role of a few gauge-invariant blocks, a step
that will prove again quite convenient when constructing the AdS deformation of the flat-space
results. Let us therefore begin by recalling the definition of the FangFronsdal operator for a
totally symmetric rank-n tensor-spinor [1],
S = i(/
/
),
(2.40)
212
where
/ denotes the -trace of the gauge field. Under the gauge transformation
= ,
(2.41)
S varies according to
S = 2i 2 /,
(2.42)
where / denotes the -trace of the gauge parameter . In analogy with what we have seen for
bosons, one can build from S the fully gauge invariant operator
W S + 2i 2 ,
(2.43)
where the rank-(n 2) compensator transforms as

= /
(2.44)
under the gauge transformation (2.41).

The Bianchi identity for W,
1
1
W W / W
/ = i 2 [/
2 // ],
2
2
leads naturally to a second gauge-invariant tensor-spinor,
Z i{/
2 / / },
(2.45)
(2.46)
directly related to the triple -trace constraint on the fermionic gauge field , that is absent in the
FangFronsdal formulation. The minimal flat-space Lagrangians of [18] can then be recovered
starting from the trial Lagrangians

1
1
1
L0 = W W W
(2.47)
/ + h.c.,
2
2
2
and compensating the remainders in their gauge transformations with new terms involving the
field and the tensor Z. The complete Lagrangians are finally

1
1
1
3 n
L = W W
/ W
/ W
2
2
2
4 3

1 n
3 n
+ h.c.,
+
/+
(2.48)
W
Z
2 2
2 3
where, as in [18], we have introduced a Lagrange multiplier , whose gauge transformation is
determined to be
=
(2.49)
in order for L to be gauge invariant. As in the bosonic case, it is useful to introduce an additional
gauge-invariant tensor, now involving ,

1

Y i ( 2 / / )
(2.50)
2
since only this gauge-invariant combination can actually enter the field equations.
213
The identification of the Y tensor suggests again that (2.48) is but a member of a family of
Lagrangians depending on a parameter k,

1
1
3 n
1
L = W W W
/
/ W
2
2
2
4 3

1 n
3 n
+ h.c.,
(2.51)
+
/+
W
( ik Y)Z
2 2
2 3
where the previous case of Eq. (2.48) corresponds to the choice k = 0. Other possible quadratic
terms in Z are excluded as irrelevant, for the same reasons as in the bosonic case. The field
equations following from (2.51) are then
1
1
1
1
E (k) W W
/ W (1 + k)Z (1 k) Y = 0,
2
2
4
2

n
1
1

E (k)
W
/ + W + (1 k) + / + Y
2
2
2

1+k
+
(2 + )Z = 0,
4

3 n
E (k)
(1 k)Z = 0.
2 3
(2.52)
As in the bosonic case, they can be reduced in general to the compensator equations of [13,16]
W S + 2i 2 = 0,
Z i{/
2 // } = 0,
(2.53)
and eventually to the FangFronsdal form upon partial gauge fixing.

In complete analogy with the bosonic case, when is coupled to a conserved external source
the consistency of the system is guaranteed by the field equations for the two auxiliary fields
since
and ,
1
1
E (k) = n 2 E (k) n E (k).
3 3
2 2
(2.54)
Moreover, in terms of the tensors defined in (2.52), the Lagrangian finally takes the simple form
1
1
1
(k) + E (k) + E (k) + h.c.
Lk = E
(2.55)
2
2
2
Generalizing these expressions to an AdS background entails a few complications. First of all,
the gauge transformation of acquires an additional contribution [31], a phenomenon that has a
well-known counterpart in gauged supergravity, so that now
1
.
2L
Hence, with our mostly plus convention for the spacetime signature,
= +
(2.56)
1
(2.57)

.
2L
The fermionic gauge transformation lends naturally to define a modified covariant derivative,
=
+
1
,
2L
(2.58)
214
in terms of which the basic relations take a simpler form. The starting point is again the covariantized FangFronsdal operator, that in an AdS background is also modified by the addition of
mass-like terms and reads
/ /
) +
SL = i(
i
i
(n 2) + .
/
L
L
One can now show that SL transforms as

2
2
SL = i / / ,
L
(2.59)
(2.60)
under the gauge transformation and satisfies the Bianchi identity

1
1
1
1
i
2

2
/ S/ L SL = (n 1) S/ L SL + 2 / ,
SL
2
2
2L
2L
2
L
(2.61)
a simpler expression than the corresponding one in terms of the more conventional derivative ,
1
1
1
1
SL
/ L SL = SL +
/L
/S
(D 2) + 2(n 1) S
2
2
4L
4L

i
1
3
+
g
/ ,
2 2
2
L
2L2
(2.62)
where we are correcting a couple of misprints present in Eq. (5.37) of [16]. One may notice, in
particular, that the spacetime dimension D never appears explicitly in Eq. (2.61).
As in the bosonic case, one is led to define the deformed gauge-invariant structures
2i
WL = SL + 2i2 ,
L

1
1
+ 2g/
,
YL = i ( 2 / / ) 2 (n 3)(5 n D)/
2
L

1
1
ZL = i
/ 2
/ + (n 3) / ,
L
L
(2.63)
where in particular the YL tensor is gauge-invariant provided the field transforms according to
= .
(2.64)
The Bianchi identity satisfied by WL ,

1
1
1
1
1
2
WL
/W
/ L WL = (n 1)W
/ L WL + 22 ZL ,
2
2
2L
2L
2
L
(2.65)
or, what is equivalent, the divergence of the kinetic operator constructed from it,

1
1
1
1
2
1
/ L gWL = W
/ L g WL + 22 ZL ,
WL W
2
2
2
2
2
L
(2.66)
is then the main ingredient to build the one-parameter family of gauge-invariant Lagrangians

e
1
1
3e n
L = WL W
/ L gWL
/ WL
2
2
2
4 3

e n
3e n
2
+
/L+
W
ikYL ZL + h.c.
2 2
2 3
L
215
(2.67)
From these one can finally derive the AdS field equations,
1
1
1
1
/ L gWL (1 + k)gZL (1 k) gYL = 0,
EL
(k) WL W
2
4
2
2

n
1
1

E L (k)
W
/ L + WL + (1 k) YL + ( YL )
2
2
L

2
1+k
+ ZL +
(ZL ) = 0,
L
4

3 n
E L (k)
(1 k)ZL = 0.
2 3
(2.68)
The conservation on an external current coupled to is finally implied by the analog of (2.54),
1
1
2
EL
(k) = n E L (k) n E L (k).
6 3
2 2
(2.69)
3. Current exchanges in the local formulations

In this section we consider the response of unconstrained higher-spin gauge fields to external
currents, focussing in particular on the current exchange. This is a convenient device to compare
different formulations in the simplest possible setting, and the ensuing analysis is indeed rather
rewarding. As we shall see, the constrained Fronsdal formulation agrees directly, in this respect,
with the minimal unconstrained form discussed in the previous section, but not with the non-local
geometric formulation of [7,13]. The lack of direct agreement between the source couplings in
the local and non-local forms of the theory, however, is an interesting fact that can be turned
to our own advantage: it determines a unique form for the non-local theory, selecting a specific
form for the corresponding Lagrangian.
In the absence of sources, as stressed in [13], the iterative procedure of [7] builds a sequence
of pseudo-differential operators
F (n+1) = F (n) +
1
1
2 (n)
F
F (n) ,
(n + 1)(2n + 1) 2
n+12
(3.1)
turning the first of the compensator equations (2.19) into a sequence of non-local equations,
2n+1 [n1]
(3.2)
.
2n1
Eventually one thus arrives at an irreducible form involving the gauge field alone, that is to be
expressible in terms of higher-spin curvatures. This gauge invariant form, however, is clearly not
unique. For instance, after has disappeared, further iterations of (3.1) will produce additional
gauge invariant equations. However, as we shall see in the next section, external sources select a
unique form of the non-local equations, that should be regarded as the proper counterpart of the
Lagrangian equations of the previous section, although they will be somewhat more complicated
than the Lagrangian equations of [7]. We can now begin to investigate these issues in some detail
F (n) = (2n + 1)
216
for the local formulation, in order to show the equivalence of its constrained and unconstrained
forms.
3.1. Bose fields in flat space
In order to motivate our procedure, let us begin by rephrasing a simple and familiar example,
the comparison between the light-cone and covariant forms of the Maxwell theory. In the former,
one has the D 2 transverse components Ai of the gauge field, that in momentum space couple
to an external current ji according to
p 2 Ai = ji .
(3.3)
The currentcurrent interaction in this physical gauge is thus sized by the product of two purely
transverse currents, since
p 2 ji Ai = ji ji .
(3.4)
In a similar fashion, in the covariant formulation one would start from the Maxwell equation for
the full vector potential,

2
p p p A = J ,
(3.5)
where, however, consistency demands that the current be conserved. As a result, in this case the
currentcurrent interaction is sized by the full scalar product of a pair of covariant conserved
currents, since
p 2 J A = J J .
(3.6)
Incidentally, here one faces a cute, albeit well-known, fact: folding the kinetic operator into a
conserved current has made it possible to effectively recover the propagator without a gaugefixing procedure. The issue is now to proceed to the (singular) on-shell limit, in order to compare
the number of degrees of freedom exchanged in the two formulations. These are encoded in the
Euclidean product ji ji for Eq. (3.4) and in the Lorentzian product J J for Eq. (3.6), which,
differently from the propagators, are well defined on-shell.
An arbitrary on-shell current J (p) can be made transverse upon multiplication by the projector
= p p p p ,
(3.7)
where p is the exchanged on-shell momentum, so that p2 = 0, and p is a second vector such that
p 2 = 0 and p p = 1. Notice that the projector satisfies the conditions
= ,
= D 2,
p = 0.
(3.8)
The same number of polarizations is thus exchanged in both cases: in the former one has directly
(D 2) of them in D spacetime dimensions, while the latter involves J J , so that the
trace of the projector , again equal to D 2, leads to the same result even starting with full
D-dimensional covariant currents.
One can repeat almost verbatim the exercise for a spin-2 field. For all dimensions D 3, its
degrees of freedom in the light-cone exchange fill the D(D3)
independent components a sym2
metric traceless tensor in (D 2) dimensions, while now the covariant Lagrangian equation is

p 2 h p p h p p h + p p h p 2 h p p h = J (p).
217
(3.9)
Combining this equation with its trace then gives

p 2 h p p h p p h + p p h = J

J
D2
(3.10)
and finally folding it, as above, with the conserved current J (p),
p 2 J h = J J
1
(J )2 ,
D2
(3.11)
or equivalently
p J
1
= J
J
D2
2
,
(3.12)
where the expression within brackets is the traceless and transverse projection of J . In deriving
this result we have taken into account that, as in the spin-one case, the conservation of J forces
the projector into the expression. The end conclusion is, therefore, that the degrees of freedom
interchanged in the covariant formulation fill a traceless symmetric matrix in D 2 dimensions,
just like their light-cone counterparts.
The previous discussion gives us the flavor of the general case, although it inevitably leaves
out its key subtleties. Thus, in D dimensions the degrees of freedom carried by a massless fully
symmetric tensor 1 ...s should fill an irreducible representation of the little group SO(D 2)
corresponding to a traceless symmetric tensor ja1 ...as . Standard properties of Young tableaux
determine the dimension of a traceful rank-s symmetric tensor of SO(D),
P (D, s) =
(D + s 1)!
,
(D 1)!s!
(3.13)
while for a traceless symmetric tensor of the same rank in D 2 dimensions the corresponding
number is
P (D 2, s) P (D 2, s 2) =
(D + 2s 4)(D + s 5)!
.
(D 4)!s!
(3.14)
Only for s = 1, 2, however, can this result be also expressed as P (D, s) 2P (D, s 1), the
relation suggested by the two simple examples above. In both these cases, one can indeed first
subtract P (D, s 1) degrees of freedom to account for the Lorentz (or de Donder) gauge conditions, one for s = 1 and a full vector for s = 2, and then again the same number, P (D, s 1), to
account for a second, on-shell, gauge transformation that preserves the first. This is at the origin
of the simple description of the s = 1 (Maxwell) and s = 2 (Einstein) cases by unconstrained
fields with a local gauge invariance, while the discrepancy for s > 2 reflects the novel features of
higher-spin fields, the need for Fronsdals trace conditions as in [1] or for the compensators of
the previous section.
Returning to our main task we would like to discuss, in a Lorentz-covariant formalism and for
an arbitrary tensor 1 ...s , the current exchange for a pair of totally symmetric currents J1 ...s .
As anticipated, the light-cone construction forces the result to equal ja1 ...as j a1 ...as on shell, and
the issue at stake is whether, in the various available formulations, the basic equality
J1 ...s P 1 ...s ;1 ...s J1 ...s = ja1 ...as j a1 ...as ,
(3.15)
218
holds, with P the spin-s analogue of the tensor defined by the r.h.s. of Eq. (3.10). In the following
we shall describe how this identity, that guarantees that only physical degrees of freedom are
exchanged, can be recovered in the local, constrained or unconstrained, formulations of higherspin gauge fields. In Fronsdals construction, where both the double trace of the current J and
the traceless part of its divergence vanish, this result was presented in [1], and the derivation is
repeated here for completeness, while simply extending it to D dimensions.
We can now build the l.h.s. of Eq. (3.15). To this end, let us begin by noticing that, given
a generic totally symmetric current J , its projection onto a traceless symmetric tensor can be
attained via the sum
N

Ts J =
n (D, s)n J [n] ,
(3.16)
where J [n] denotes the nth trace of J and N is the smallest integer such that the next trace J [N+1]
cannot be defined. The coefficients of this expansion depend on the spin s of J and on the space
time dimension D. They are determined by the results collected in Appendix A, that lead to the
one-term recursion relation
n (D, s)
,
n+1 (D, s) =
(3.17)
D + 2(s n 2)
with the initial condition 0 (D, s) = 1. Notice that n (D 2, s) = n (D, s 1).
As we have seen, a generic tensor can be rendered transverse by the application of the projector . In order to build the tensor P that defines the current exchange, one can first project onto
the transverse part of J to then extract the traceless part of the resulting tensor, thus obtaining
PJ =
N

n (D 2, s) n J [n] .
(3.18)
n=0
This is just what we have seen in Eq. (3.16), but for a key novelty: here depends on D 2, as
a result of the presence of .
We have thus made our way backwards to the tensor P of Eq. (3.15), that projects a covariant
current J onto its transverse traceless part, but at the price of an explicit dependence on both
the physical momentum p and the additional vector p.
In some notable cases, however, the
dependence on p disappears. For instance, if J is conserved,
J P J =
N

n (D 2, s)
n=0
s!
J [n] J [n] ,
n!(s 2n)!2n
(3.19)
so that P can be effectively replaced by

Pc J =
N

n (D 2, s)n J [n] ,
(3.20)
n=0
and one can now show that

(Pc J ) = 2
N1

n+1 (D 2, s)n J [n+1] ,
(Pc J ) = 0.
(3.21)
219
The momentum dependence of P can be also suppressed in Fronsdals constrained formulation [1]. The Lagrangian equation is in this case
1
F F = J,
2
and the double-trace condition may be used to show that
J =
1
F ,
21 (D 2, s)
(3.22)
(3.23)
while the double trace of J vanishes,

J = 0,
(3.24)
so that Eq. (3.22) can be also presented in the alternative form

F = J + 1 (D 2, s)J .
(3.25)
In addition, Eq. (3.22) implies that

1
p J = p F ,
(3.26)
2
so that, using Eq. (3.23), one can see that only the traceless part of the divergence of J vanishes
in general. The last condition can be also written
p J + 1 (D, s 1)p J = 0,
(3.27)
so that, introducing a new tensor , the combination

J + 1 (D 2, s)J + p
(3.28)
is actually both traceless and transverse provided is traceless and satisfies the condition
p = 1 (D 2, s)J .
(3.29)
However, p yields a vanishing result upon contraction with J since, as we have seen, the
traceless part of p J vanishes. The end conclusion is that in Fronsdals constrained theory PJ
can be actually replaced with
Pc J = J + 1 (D, s 1)J ,
(3.30)
and therefore the corresponding current exchange is determined by the relatively simple expression
1 (D 2, s)s(s 1)
J J +
(3.31)
J J .
2
In order to compare with the unconstrained formulation of the previous section, let us begin
by noticing that, once the third of Eqs. (2.17), the constraint C = 0, is enforced, the coupling to
an external source is described by
1
A A + 2 B = J
2
while the double trace A vanishes identically. As a result, the quantity
K = J 2 B
(3.32)
(3.33)
220
is somehow the counterpart of Fronsdals current in this case, since on shell K = 0 as a result
of the condition A = 0, that as we have stated follows from the constraint C = 0. This condition
also determines B, and thus the Lagrange multiplier , in terms of J .
One can now write
K =J +
N

n n J [n] ,
(3.34)
n=2
with the coefficients determined recursively by the condition that the double trace K vanish.
This apparently results in a three-term recursion relation,

n + D + 2(s n 3) 2n+1 + D + 2(s n 4) n+2 = 0,

(3.35)
with the conditions

2 D + 2(s 3) D + 2(s 4) = 1,
22 + 3 D + 2(s 5) = 0.
However, introducing

un = n + D + 2(s n 3) n+1 ,
(3.36)
(3.37)
one can turn Eq. (3.35) into the simpler two-term recursion relation,
un+1 = un
1
,
[D + 2(s n 3)]
(3.38)
which, taking into account the value of u2 determined by Eqs. (3.36),

u2 =
1
,
[D + 2(s 3)][D + 2(s 4)]
(3.39)
is actually solved by
un = n (D 2, s),
(3.40)
where the n are defined in Eq. (3.17).

Making use of (3.38), the defining relation for the un becomes
un
n+1 ,
un = n
un+1
(3.41)
or, in terms of vn = n /un ,

vn+1 vn + 1 = 0,
(3.42)
whose solution is vn = v2 (n 2) = n + 1. In conclusion,

n = (n + 1)n (D 2, s).
(3.43)
Since A is doubly traceless, Eq. (3.32) can be turned into

A = K + 1 (D 2, s)K ,
and one can verify that
K + 1 (D 2, s)K =

n
(3.44)
n (D 2, s)n J [n] .
(3.45)
221
This determines PJ , and the current exchange is finally

N

n (D 2, s)
n=0
s!
J [n] J [n]
n!(s 2n)!2n
(3.46)
which agrees with Eq. (3.19), with the correct number of degrees of freedom, since the exchange
involves, as expected, a pair of traceless conserved currents built from the original conserved
current J .
3.2. Fermi fields in flat space
In the previous subsection we have recalled how, in D dimensions, the physical degrees of
freedom carried by a massless symmetric rank-s tensor 1 ...s fill a symmetric traceless rank-s
tensor in D 2 transverse dimensions. For massless Fermi fields the situation is similar, if technically more involved, so that a D-dimensional spinor-tensor field 1 ...m carries physical degrees
of freedom corresponding to an on-shell -traceless spinor-tensor in D 2 transverse dimensions. The on-shell condition is of course stronger for a Fermi field: not only does it select a
light-like momentum, but via the Dirac equation it further halves the number of its degrees of
freedom. This is reflected in the familiar presence, in spinor propagators, of the matrix p
/ , that for
a light-like momentum p has precisely this effect.
In discussing current exchanges for Fermi fields, it is useful to begin by defining the projection
of 1 ...m to its -traceless part. One can verify that this is effected by
Tm = +
N

n (D, m + 1) n [n] + n1
/ [n1] ,
(3.47)
n=1
where the coefficients n (D, s) were introduced in the previous subsection, since they also determine the traceless projection for bosonic fields.
Using Tm , one can also construct the projectors to the doubly and triply -traceless parts of .
The first is simply the traceless projector that was introduced in the previous subsection, but here
we can also relate it to the projector of Eq. (3.47), according to
Tm(2) =
n (D, m)n [n] = Tm +
1
/
Tm1 .
D + 2m 2
(3.48)
In a similar fashion, one can define a triply -traceless projector,

1
Tm(3) = Tm(2) +
Tm2
D + 2m 4

1

=
/ + .
(D + 2m 6)(D + 2m 4)
(3.49)
In deriving these results, use has been made of the relations

n (D 2, m + 1) = n (D, m),
n (D, m 1) =
n+1 (D, m)
,
1 (D, m)
which follow from Eq. (3.17), and of the initial condition 0 (D, s) = 1.
(3.50)
222
If the propagator is now denoted by

BJ = J +
N

p
/
B,
p2
the previous considerations lead to

n (D 2, m + 1) n (J )[n] + n1 (J )[n1] ,
(3.51)
n=1
which projects the external current J to its transverse and -traceless part. This is achieved
in two steps: the current J is first projected to its transverse part J T using the projector
introduced for bosonic fields, and then the trace is eliminated from the resulting expression via
Eq. (3.47). Notice that, as in the bosonic case, the presence of brings about the replacement
of D with D 2 in this expression.
Simplifications are again possible in special circumstances, and in particular if J is transverse:
Eq. (3.51) reduces to
Bc J = J +
N

n (D 2, m + 1) n J [n] + n1 J [n1] ,
(3.52)
n=1
without the need for any explicit projectors, and hence with no explicit dependence on p.
Notice that Bc J can also be written in the form

Bc J = Tm(2) J + 1 (D, m) Tm(3) J ,
(3.53)
which makes it manifest that the current thus projected is triply -traceless. This is the counterpart of the condition (3.21) obtained for bosonic fields. As we shall see shortly, this presentation
of the result is particularly useful when comparing with the propagator for the local unconstrained
formulation of the previous section.
In the FangFronsdal theory with an external source J the field equation is
1
1
S S S
/ =J,
2
2
which via the Bianchi identity implies the condition
1
1
/.
J = S S
2
2
As a result only the divergence of the -traceless part of J vanishes in general,
Tm1 ( J ) = 0.
(3.54)
(3.55)
(3.56)
In addition, the FangFronsdal constraint on the triple -trace of implies that S, and hence J
on account of Eq. (3.54), are also triply -traceless. In analogy with what was done for bosons,
Eq. (3.54) can thus be inverted and BJ can be replaced with
Bc J = J + 1 (D 2, m + 1)(J + J ),
(3.57)
which in four spacetime dimensions agrees with the expression given in [1].
In the unconstrained local formulation of the previous section the situation is similar to some
extent, since once the third of Eqs. (2.52) is enforced, the first, the field equation for , takes the
form
1
1
1
W W
/ W = J Z K.
2
2
4
(3.58)
223
Just as for bosons the gauge invariant tensor A was doubly traceless on shell, so one can show
that the gauge invariant spinor-tensor W is triply -traceless on shell,
W
/ = 0.
(3.59)
Hence, the same property holds for K, so that the combination of Z and J results in
K = Tm(3) J ,
(3.60)
that is therefore an effective FangFronsdal current. One can then solve Eq. (3.58), obtaining
W = K + 1 (D 2, m + 1)[ K
/ + K ],
(3.61)
and therefore the currentcurrent amplitude is determined by
/ + K ].
K + 1 (D 2, m + 1)[ K
(3.62)
In terms of J , using (3.49), one can now conclude that

W = Tm(2) J + 1 (D, m) Tm(3) J .
(3.63)
This expression is the direct counterpart of Eq. (3.57), since, as we have stressed, 1 (D, m) =
1 (D 2, m + 1). It finally shows that in the unconstrained formulation the current exchange
has the form (3.53), which indeed guarantees that the correct number of degrees of freedom
propagates on-shell.
3.3. Current exchanges in an AdS background
In the last two subsections we have shown that the constrained and unconstrained formulations
for Bose or Fermi fields in flat spacetime are equivalent even in the presence of external currents.
This result reflects the occurrence, in both settings, of tensors (F and A, or S and W) that onshell are effectively subject to the (Fang) Fronsdal ( -) trace constraints and satisfy the same
Bianchi identities. We have also identified effective currents (K or K) that behave exactly like
the FangFronsdal currents. We now want to show briefly how this structure carries over to AdS
backgrounds.
Let us begin by deriving the current exchange for unconstrained Bose fields in AdS. The key
observation is that, when is on shell, the first of Eqs. (2.36) takes the form
1
AL gAL + g 2 CL = J,
2
(3.64)
where g denotes the AdS metric and, again, AL = 0. The same steps followed in the preceding
subsections for the flat-space analysis then lead to
AL =
N

n (D 2, s)g n J [n] .
(3.65)
n=0
In order to proceed further, it is very convenient to introduce the Lichnerowicz operator [32],
that in an AdS background takes the form
2L = 2 +
1
s(D + s 2) 2g ,
2
L
(3.66)
224
and is particularly convenient, since it commutes with contraction and covariant differentiation
and allows one to write the AdS Fronsdal operator as

2
2L 2 (s 1)(D + s 3) + 2 .
(3.67)
L
When contracted with a conserved current, clearly only the first term is relevant. As a result, the
current exchange is finally determined by

1
N

s!
2
[n]
(3.68)
n (D 2, s)
J
(s
1)(D
+
s
3)
J [n] ,
L
n!(s 2n)!2n
L2
n=0
which reduces to the flat space amplitude in the limit L .

Fermi fields in AdS, as we have seen, entail a few additional complications, since the current
is now subject to a modified conservation law,
1
J = 0.
2L
When the auxiliary fields are on shell, the relevant field equation reads
J +
(3.69)
1
1
1
WL W
(3.70)
/ WL = J ZL KL ,
2
2
4
with, again, the triple -trace constraint W
/ L = 0. As in the flat case, one can first solve for ZL
and then obtain WL from
WL = KL
1
/ L + KL ],
[ K
D + 2(m 2)
(3.71)
(3)
where the effective current KL is subject to the triple -trace constraint KL = Tm J , with Tm
defined as in (3.49) but for the replacement of the flat metric and the Dirac matrices with their
AdS counterparts. Notice that KL satisfies the modified conservation law

1
K
/ L = 0.
Tm KL +
(3.72)
2L
One thus faces, again, an effective current KL which is built from the conserved current J
but nonetheless behaves as a constrained current of the FangFronsdal theory. Constrained and
unconstrained local formulations agree, the latter being effectively a FangFronsdal theory with
a partially conserved current which is built, as stressed above, from the conserved current J .
4. Current exchanges in the non-local formulation
We can now turn to the non-local formulation. Confining our attention to bosonic fields, we
begin by analyzing the current exchange in the non-local theory with reference to the Lagrangian
equations proposed in [7], in order to display the problem. As anticipated, and as we shall see
shortly, the Lagrangian equation proposed in [7] is not the proper counterpart of the local one
discussed in Section 2 since, with standard couplings of the J type, it does not result in the
correct counting of degrees of freedom in current exchanges. We shall then present a unique set
of non-local Lagrangian equations that, for all s, reproduce the current exchange of the previous
section. The result of this analysis will thus be a precise map, for all s, between the local and
non-local formulations of the unconstrained theory. In the absence of external currents, however,
225
these novel Lagrangian equations can also be turned into the non-Lagrangian equations of [7],
Eqs. (1.2) and (1.3). Here we confine our attention, for brevity, to the case of bosonic fields, but
similar results are expected to hold for fermionic fields.
4.1. The problem
The iterative procedure of [7], reviewed in the previous section, was meant to terminate, for a
given value s of the spin, s = 2n 1 or s = 2n, after first reaching a non-local gauge invariant
extension F (n) of the Fronsdal operator. From this, as shown in [7], one can build a divergencefree Einstein-like tensor by simply combining F (n) with its traces according to
G (n) =
n

(1)p (n p)!
p F (n)[p] .
2p n!
p=0
(4.1)
The form of this Einstein-like tensor is determined by the Bianchi identity satisfied by the F (n) ,

1
1 2n+1 [n+1]
F (n) F (n) = 1 +
(4.2)
,
2n
2n 2n1
where for the two relevant values of s associated to a given n, the term on the right-hand side
vanishes identically. In this case Eq. (4.2) has a number of interesting consequences, that can be
derived taking successive traces:
F (n)[k]
1
F (n)[k+1] = 0
2(n k)
(k n 1).
(4.3)
In particular, if the spin s is odd, so that s = 2n 1, one can see that

F (n)[n1] = 0.
(4.4)
If the system is coupled to an external current J , the Lagrangian field equations proposed in
[7] thus read
G (n)
n

p=0
(1)p
(n p)! p (n)[p]
=J,
F
2p n!
(4.5)
where, as we have anticipated, s = 2n 1 or s = 2n. These equations can now be inverted

noticing that
1 (D 2n, s)J =
2 (D 2n, s)2 J =
n

p=1
n

(1)p
(1)p
p=2
p(n p)! p (n)[p]

,
F
2p n!
p(p 1) (n p)! p (n)[p]

,
F
2
2p n!
(4.6)
(4.7)
and continuing in this fashion one readily obtains

F (n) =
n

n (D 2n, s)n J [n] ,
n=0
where the coefficients n were introduced in Section 3.
(4.8)
226
One should now contract Eq. (4.8) with a conserved current, and the occurrence of a key
simplification may be simply anticipated: the relations J F (k) = = J F hold, up to terms
involving the vanishing divergence of J . The current exchange in the non-local theory of [7],
based on the Einstein-like tensor of Eq. (4.5), is thus determined by
n

n (D 2n, s)n J [n] ,
(4.9)
n=0
and clearly disagrees with Eq. (3.18) whenever s > 2, i.e. for all interesting cases of higher-spin
fields.
In order to better appreciate the nature of the problem, it is instructive to take a closer look at
the relatively simple but still non-trivial case of a spin-3 field. The kinetic operator of [7],
1
( F + F + F ),
32
is then determined by a single iteration of Eq. (3.1), and satisfies the Bianchi identity
(2)
= F
F
1
F (2) F (2) = 0,
4
to be compared with the conventional Bianchi identity for the Fronsdal operator
(4.10)
(4.11)
1
F F = 0.
(4.12)
2
In both cases the Lagrangian field equations would couple the corresponding divergence-free
Einstein-like tensors to divergence-free currents, according to
1
G F F = J,
2
1
G (2) F (2) F (2) = J ,
4
(4.13)
(4.14)
but the definitions of G and G (1) involve different coefficients, reflecting the differences between
the Bianchi identities of Eqs. (4.11) and (4.12). Inverting these equations, one would then arrive,
in the two cases, at the current exchanges
3
(4.15)
J J ,
D
3
J J
(4.16)
J J ,
D2
which are clearly different, as special cases of Eqs. (3.18) and (4.9). The lesson to be drawn from
this example is that the differences between the current exchanges in the local formulation of the
previous sections and in the non-local formulation of [7] based on Eq. (4.5) reflect those between
the modified Bianchi identity satisfied by the non-local kinetic operators F (n) and the original
Bianchi identity satisfied by the Fronsdal operator F .
Interestingly, the two results obtained in this spin-3 example can actually be mapped into
one another provided the currents J and J are related by a suitable non-local field redefinition.
Indeed, defining
J J
2
,
2
(4.17)
a non-local extension of which is divergence free, if J and J are related according to
3D 3D 2 6D
J =J +
J
,
3D(D + 1)
227
(4.18)
Eq. (4.16) turns precisely into (4.15). Notice that the map between the two constructions thus
obtained is compatible with the conservation of both J and J .
This result is adding a useful piece of information: while the tensor G (1) does not couple as
expected to an external current, a suitable non-local combination of this tensor with its trace does.
These more singular objects satisfy Bianchi identities that are closer to that satisfied by F and A:
hence, they are precisely the types of non-local kinetic operators we are after, for all values of s.
As a side technical remark, let us stress again that, strictly speaking, neither nor the F (n) are
well-defined quantities on shell. Hence, it should be understood that, in all the present treatment
of current exchanges in the non-local formulation, one is actually working off shell all the way
and only the final results are continued on-shell to arrive at the correct counting of degrees of
freedom.
Actually, in [7] a more singular class of non-local operators was also considered. For all s,
they can be obtained combining the F (n) with their traces and are quite interesting, since they
lead in general to field equations of the type
F F 3 3 H = 0,
(4.19)
that have the same form as the non-Lagrangian compensator equations (2.19). In particular, for
s = 3 one can define

(1)
F
(4.20)
F
F ,
22
which actually satisfies the same Bianchi identity as F , so that the corresponding Einstein tensor
is
1
G = F F .
(4.21)
2
Since F and F differ by terms that vanish upon contraction with a conserved current, with
this type of non-local Lagrangian equation the current exchange clearly agrees with the result
obtained in the previous subsection for the local theory. Notice that in this case one could also
arrive at Eq. (4.20) starting from the two conditions
1
A A = 0,
2
A = 0,
(4.22)
where the second guarantees that the first couple consistently to a conserved current.
The lesson to be drawn from this example is, therefore, that for a given s, equal to 2n1 or 2n,
the Einstein-like tensors that correspond to those of the local theory are not directly expressible
in terms of the single F (n) operator, but are to involve other more singular operators F (m) , with
m > n. The issue is now to determine the form of these Einstein-like tensors for all s and to
elucidate the corresponding geometrical structure. As in other circumstances, for instance for the
Lagrangians of Section 2, one has to reach the s = 4 case to first uncover the relevant pattern. In
this case, letting
A = F 3 3 ,
228
and insisting on the condition A = 0, leads to

=
F
F .
2
32
423
(4.23)
On the other hand, combining it with its trace, one can turn the non-local equation F (1) = 0 into
T F 3 3 H = 0,
(4.24)
where
1
F
F .
(4.25)
322
422
Notice that and H do not coincide. Their difference, however, is determined by a gauge
invariant quantity according to
H =
3
( 4 ),
42
a relation that can also be presented in the more symmetric form
H =
1
4 = [ 4 H ],
4
and implies that the two choices result in the two inequivalent Einstein tensors

1
1 2
1
A 2 A ,
E = A A
2
32
6

1
1 2
5

T
+ 2 T .
EH = T T
2
32
24
(4.26)
(4.27)
(4.28)
Notice that none of the two choices results in the correct current exchange. The first indeed gives
J A=J J
3(D + 6)
6
(J )2 ,
J J +
D+2
(D + 2)(D 2 + 6D + 2)
(4.29)
6
3(6 5D)
(J )2 ,
J J +
D+2
(D + 2)(5D 2 + 6D + 8)
(4.30)
while the second gives

J T =J J
to be compared with the correct result, given in Eq. (3.19).

4.2. Nonlocal equations with a proper current exchange
The arguments of the previous subsection show clearly that the Einstein tensor is the crucial
ingredient behind the current amplitude, whose form is determined by the Bianchi identity for
Anl , the gauge-invariant extension of the Fronsdal operator F . Actually, a closer look at the
discussion of Section 3 for the local theory shows that the non-local formulation can reproduce
the correct currentcurrent amplitude provided its Lagrangian field equations
Gnl = J
(4.31)
lead to the solution

Anl = Pc J ,
(4.32)
229
where Pc was defined in Eq. (3.20). Eqs. (3.21), however, imply a pair of consistency conditions
for this statement,
Anl = 0,
1
Anl Anl = J + 2 ( ),
(4.33)
2
which, as we have seen, are precisely met by the local construction.
One is thus led to search for an Einstein tensor that differs from that of Eq. (4.5), and is rather
of the form
1
Gnl = Anl Anl + 2 B,
(4.34)
2
for some tensor B, and where Anl = 0. Conversely, given this form of the Einstein tensor, with a
doubly traceless Anl , one can see that the solution of Eq. (4.31) is bound to take the form (4.32).
On the other hand, the Einstein tensor can take the form of Eq. (4.34) only if Anl obeys the
Bianchi identity
1
Anl Anl = 0,
(4.35)
2
and if the divergence of the trace of Anl is a pure gradient, a condition that we shall write in the
form
Anl = 2Dnl ,
(4.36)
with Dnl an arbitrary non-local tensor. These two requirements are in fact necessary to guarantee
the consistency of Eq. (4.31) with the conservation of the current J .
Eqs. (4.35) and (4.36) are also sufficient if they are combined with the additional demand that
Anl be of the form
Anl = F 3 3 nl ,
(4.37)
for some nl to be determined. In order to prove that this is actually the case, notice first that the
Bianchi identity reads
1
3

),
Anl Anl = 3 ( 4 nl nl
2
2
so that it vanishes provided nl solves the constraint

Cnl ( 4 nl nl
) = 0.
(4.38)
(4.39)
Since the double trace of Anl can be expressed as

,
Anl = 32Cnl + 3 Cnl + 2 Cnl
(4.40)
the condition (4.39) also guarantees that Anl be doubly traceless on shell. However, Eq. (4.36)
and its successive traces also imply that
[n]
1
2
n+2 Dnl
Gnl = Anl Anl + Dnl + +
(4.41)
,
2
2n n!
where n is the integer such that s = 2(n + 2) or s = 2n + 5, is indeed conserved and is of the
form (4.34).
230
To summarize, the currentcurrent amplitude is correctly reproduced by the modification

(4.37) of the Fronsdal operator provided the two conditions (4.39) and (4.36) are met. Notice
that (4.39) and its consequence (4.40) are the counterparts of the equation of motion in the
local formulation (the third equation in (2.17)). On the other hand, Eq. (4.36) can be regarded
as the counterpart of the equation of motion of the local unconstrained formulation, while the
very form of the Einstein tensor (4.34) is clearly tailored after the one entering the equations
of motion in the local formulation.
Now we would like to show that a solution to these two conditions exists and is unique. To
this end, we shall first determine nl in terms of . We shall then conclude the present subsection
by displaying the geometrical structures underlying Anl .
Let us first notice that
Anl = 32 2 322 nl

92 nl + 3 nl

3
1

+ 2nl
+ 2 nl
2 .
2
2
(4.42)
Requiring that this expression be equal to 2Dnl gives the two conditions
32 2 322 nl = f,

3
1

92 nl + 3 nl + 2nl
+ 2 nl
2
2
2
(4.43)
= f 2Dnl ,
where f is an arbitrary tensor, that will be determined shortly by Eq. (4.39). The first equation
gives indeed
1
2

f,

2
2
32
322
and inserting this expression in Eq. (4.39) yields an equation for f ,
nl =
2
3
f = 2 + 3
4
2
3
1
3
1 2
+

f
f .
4
22
22
22
One can now look for a solution of the form

f=
n fn ,
(4.44)
(4.45)
(4.46)
where for a spin-s field the sum terminates at n = s 4, to be determined by successive iterations,
and the result is
3
2
f0 = 2 + 3 ,
(4.47)
4
2
3
2
6
,
f1 = +
(4.48)
4
2
522

1
1

(n 2).
2n(n + 2) fn1 + n(n 1)fn2
fn =
(4.49)
(n + 1)(n + 4) 2
231
This truncation has an interesting consequence: if s denotes the expression for nl for a spin-s
field, taking into account the last terms leads to the recursion relation
s 2 s2
fs3 .
3 22
The solutions for the first few cases are
1
2
,
3 =
2
2
32

1
2

4 = 3 + 2
2 +
+ ,
32
2 4

2 1
4
4

,

+
5 = 4 2
32
2 2
522
s+1 = s
(4.50)
(4.51)
(4.52)
(4.53)
and can also be expressed in terms of the Fronsdal operator as

1
(4.54)
F ,
322
1
1
F
F +
F ,
4 =
(4.55)
2
3
12 22
32
32
1
2 2
1 1 2

5 =
F
+
F
+
F
F . (4.56)
5 24
12 22
5 23
322
323
The Einstein tensor is finally determined by nl and Dnl given in (4.43). For spin s = 4 and
s = 5, for instance, the results read

1 1

F F ,
D4 =
(4.57)
2 2

1 1
D5 =
F F 2 F + F
2 2
2
2

1
1

=
1
F F .
(4.58)
2
2
2
3 =
There is actually a more illuminating way to proceed. The key idea is to present the solution
in a manifestly gauge invariant fashion by expressing Anl in terms of the higher-spin curvatures.
To this end, let us begin by considering the geometric operators

1
[n+1] ,
s = 2(n + 1),
nR
(n+1)
= 21
F
(4.59)
[n]
2n R , s = 2n + 1.
As discussed in [7], these operators are directly related to the F (n) of Eq. (3.1), and hence satisfy
the sequel of identities
F (n+1)[k] =
1
F (n+1)[k+1] ,
2(n k + 1)
(4.60)
so that, in the odd case, F (n+1)[n] 0.

As a result, all divergences of the tensors Fn+1 can be expressed in terms of traces, so that the
only available independent structures are
F (n+1) ,
F (n+1) ,
F (n+1) ,
...,
F (n+1)[k] ,
...,
F (n+1)[q] ,
(4.61)
232
where q = n + 1 or q = n depending on whether the rank is s = 2(n + 1) or s = 2n + 1. One can

now construct the most general linear combination of all these terms, with arbitrary coefficients
(the first may be set to one, up to an overall normalization)
2
2k
Anl = F (n+1) + a1 F (n+1) + + ak k F (n+1)[k] +
2
2

2(n+1) (n+1)[n+1]
an+1 2n+1 F
, s = 2(n + 1),
+
2n (n+1)[n]
,
s = 2n + 1.
an 2n F
(4.62)
One can first determine a1 by requiring that Anl be of the form (4.37). The remaining coefficients
may then be fixed, recursively, demanding that the proper Bianchi identity,
1
Anl Anl = 0,
2
(4.63)
hold for Anl or, equivalently, that Anl be doubly traceless. Indeed, demanding that all terms in 2
disappear yields the condition
a1 =
n
,
n+1
(4.64)
while the double trace A vanishes provided

n+k+1
n+k
ak+2 =
ak + 2ak+1 ,
nk
nk+1
(4.65)
so that there is one and only one solution to our conditions.

The explicit solution of this difference equation may be foreseen after a few iterations, and
reads
ak = (1)k+1 (2k 1)
k1
n+2 n+j
,
n1
nj +1
(4.66)
j =1
so that
Anl =
n+1

(1)
k=0
k+1

k1
n+2 n+j
2k (n+1)[k]
(2k 1)
F
.
n1
n j + 1 2k
(4.67)
j =0
Alternatively, the Bianchi identity (4.63) holds provided

ak+1 = ak
(2k + 1)(n + k)
,
(2k 1)(n k + 1)
(4.68)
which is also solved by Eq. (4.66).

It is remarkable that (4.68) holds without any assumption on the form of the tensor Anl , aside
from the condition that no s be present in the linear combination (4.62). This means that by
simply imposing the Bianchi identity on Anl both the double-tracelessness and the compensator
form Anl = F 3 3 nl emerge as direct consequences.
To these results one should finally add the explicit solution for Dnl ,
Dnl =
233
n+1
1
2n + 4k + 1
2(k1) (n+1)[k+1]
1 2(k2) (n+1)[k]
ak
F
+
F
2
2k 3 2k2
2(2k 1)(n k + 1) 2k1
k=2

n+k+1
2k (n+1)[k+2]
+
F
(4.69)
2(n k)(n k + 1) 2k
which allows to complete the construction of non-local Einstein-like tensors leading to correct
current exchanges. In order to show that these tensors are indeed divergence-free, in the odd-spin
case, after all possible traces of Dnl are computed and all divergences are turned into further
traces via the Bianchi identities, one can notice that the result only involves divergences of
F (n+1)[n] , that vanish identically because of Eq. (4.60).
As an example, the modified Fronsdal tensors for the cases of spin s = 3, 4 and 5 read
1
2
R ,
R +
2
222
1
1 2
4 [4]
R
3
R ,
A4 = R +
2
2 22
23
1
2 2
4
3
R[4] ,
A5 = 2 R +
3 23
2
24
while the corresponding D tensors are similarly given by
A3 =
(4.70)
(4.71)
(4.72)
3 1 [4]
R ,
(4.73)
82
5
D5 = 2 R[4] .
(4.74)
82
In the absence of sources, taking successive traces of the Lagrangian equation of motion coming
from (4.34)
D4 =
1
Anl Anl + 2 B = 0,
(4.75)
2
one can reduce it to Anl = 0. This last equation, in turn, can be shown to imply the nonLagrangian equations of [7], Eqs. (1.2) and (1.3), by making careful use of the Bianchi identities
(4.60). For example, for the spin-4 case, once B and Anl are shown to vanish one is left with
1 2 (2)
4
F 3 2 F (2) = 0,
22
2
whose trace implies
F (2) +
F (2)
2 (2)
= 0.
F
2
(4.76)
(4.77)
Taking the divergence of this last relation, and using the identity F (2) = 12 F (2) it is then possible to show that F (2) = 0, which implies, via Eq. (4.77), that F (2) = 0 and finally F (2) = 0,
Eq. (1.3), as previously advertised.
5. Conclusions
In this paper we have examined a number of problems that present themselves when higherspin gauge fields interact with external currents, the key motivation for our analysis being the
234
precise comparison between Fronsdals constrained formulation of [1] and the unconstrained
Lagrangian formulations of [7] and [18]. To this end, we have began by streamlining the results
of [18], that are here presented in terms of a few invariant structures which also extend rather
simply to the interesting cases of AdS backgrounds. The subsequent sections have dealt with
the precise comparison between the different available formulations in the presence of external
sources. These provide an important testing ground for the Lagrangian equations, and indeed
while the local formulation of [18] is directly equivalent to Fronsdals constrained formulation
of [1], this is not the case for the Lagrangian equation proposed in [7]. In Section 4 we have thus
displayed the precise map between the minimal unconstrained local Lagrangian formulation for
fully symmetric tensors and spinor-tensors presented in [18] and the non-local formulation of
[7], and we have proposed a new set of non-local Lagrangian equations that lead to a proper
current exchange and, in the free case, can be reduced to the non-Lagrangian equations (1.2) and
(1.3). Let us stress that the non-local geometric form thus identified is uniquely determined by
the procedures described in Section 4.
Note added
The current exchange of Eq. (3.46) and its fermionic counterpart determine the van Dam
VeltmanZakharov discontinuity [33] in flat space for these whole classes of (bosonic and
fermionic) higher-spin fields. The expressions depend on the dimension D of spacetime, and
the discontinuity follows directly from the comparison of the D-dimensional result with the
corresponding one in D + 1 dimensions. This is the case since the massless theory in D + 1
dimensions, after a suitable reduction la ScherkSchwarz [34], describes irreducibly a massive
field in the Stueckelberg mode. The extension of this result to an AdS background, along the lines
of what was done by Higuchi and Porrati for s = 2 [35], is quite interesting and will be discussed
elsewhere.
Acknowledgements
We are grateful to the CPhT of the Ecole Polytechnique, to the APC-Paris VII, to the LPTOrsay, to the Universit Roma Tre and to the CERN Theory Division, where the stay of A.S.,
while most of this work was being done, was supported in part by the Scientific Associates
Program, for the kind hospitality extended to us. D.F. is also grateful to the MPI-Albert Einstein
Insitute, where part of this work was done. The present research was also supported in part
by INFN, by the MIUR-PRIN contract 2003-023852, by the EU contracts MRTN-CT-2004503369 and MRTN-CT-2004-512194, by the INTAS contract 03-51-6346, and by the NATO
grant PST.CLG.978785.
Appendix A. Notation and conventions
We use the mostly positive convention for the spacetime signature, and typically omit sym
metrized indices in tensor relations. Hence, our gamma matrices satisfy 0 = 0 , i = + i ,

0 0 = . In addition, a prime always denotes a trace: U is thus the trace of U , while
U is its double trace. A generic multiple trace, however, is denoted by a bracketed suffix, so
that U [n] is the nth trace of U . This notation is a very convenient method of streamlining the
presentation, but it also results in an effective calculational procedure. This is especially true in
flat space, but also in AdS backgrounds provided some care is clearly exercised with the resulting
235
non-commuting covariant derivatives. To take full advantage of the compact notation, one needs
to make repeated use of a number of identities, which reflect some simple combinatorics. These
rest on our convention of working with symmetrized objects not of unit strength, which is convenient in this context but is not commonly used. For instance, given of vectors A and B , AB
here stands for A B + A B , without additional factors of two. The key identities are then:
p
= 2 p2 + 2 p1 + p ,

p + q p+q
p q
=
,
p
p
= 2 p1 + p ,
k = k1 ,
k
= D + 2(s + k 1) k1 + k ,
(U V ) = U V + U V + 2U V ,
n1 = nn .
(A.1)
As anticipated, the basic ingredient in these expressions is the combinatorics, which is simply determined by number of relevant types of terms on the two sides. Thus, for a pair of flat derivatives,
= 2 2 reflects the fact that, as a result of their commuting nature, the usual symmetrization is
redundant precisely by the overall factor of two that would follow from the second relation. In a
similar fashion, for instance, the last identity reflects
the different numbers of terms generated by
the naive total symmetrization of the two sides: 2n
2 (2n 1)!! for the expression on the l.h.s,
and (2n + 1)!! for the expression on the r.h.s.
References
[1] C. Fronsdal, Phys. Rev. D 18 (1978) 3624;
J. Fang, C. Fronsdal, Phys. Rev. D 18 (1978) 3630.
[2] R. Argurio, G. Barnich, G. Bonelli, M. Grigoriev (Eds.), Higher-Spin Gauge Theories, Proc. I Solvay Workshop,
Brussels, 1214 May 2004, Int. Solvay Institutes, 2006.
[3] M. Bianchi, V. Didenko, hep-th/0502220;
D. Francia, C.M. Hull, hep-th/0501236;
N. Bouatta, G. Compere, A. Sagnotti, hep-th/0409068.
[4] X. Bekaert, S. Cnockaert, C. Iazeolla, M.A. Vasiliev, hep-th/0503128.
[5] A. Sagnotti, E. Sezgin, P. Sundell, hep-th/0501156.
[6] B. de Wit, D.Z. Freedman, Phys. Rev. D 21 (1980) 358;
T. Damour, S. Deser, Ann. Poincar Phys. Theor. 47 (1987) 277.
[7] D. Francia, A. Sagnotti, Phys. Lett. B 543 (2002) 303, hep-th/0207002.
[8] X. Bekaert, N. Boulanger, Phys. Lett. B 561 (2003) 183, hep-th/0301243;
P. de Medeiros, C. Hull, JHEP 0305 (2003) 019, hep-th/0303036.
[9] X. Bekaert, N. Boulanger, hep-th/0606198.
[10] A. Pashnev, M.M. Tsulaia, Mod. Phys. Lett. A 12 (1997) 861, hep-th/9703010;
A. Pashnev, M.M. Tsulaia, Mod. Phys. Lett. A 13 (1998) 1853, hep-th/9803207;
C. Burdik, A. Pashnev, M. Tsulaia, Nucl. Phys. B (Proc. Suppl.) 102 (2001) 285, hep-th/0103143;
I.L. Buchbinder, A. Pashnev, M. Tsulaia, Phys. Lett. B 523 (2001) 338, hep-th/0109067;
I.L. Buchbinder, A. Pashnev, M. Tsulaia, hep-th/0206026;
I.L. Buchbinder, V.A. Krykhtin, A. Pashnev, Nucl. Phys. B 711 (2005) 367, hep-th/0410215.
[11] C. Becchi, A. Rouet, R. Stora, Commun. Math. Phys. 42 (1975) 127;
C. Becchi, A. Rouet, R. Stora, Ann. Phys. 98 (1976) 287;
I.V. Tyutin, preprint FIAN Nr. 39 (1975);
E.S. Fradkin, G.A. Vilkovisky, Phys. Lett. B 55 (1975) 224;
236
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
I.A. Batalin, G.A. Vilkovisky, Phys. Lett. B 69 (1977) 309;

M. Henneaux, Phys. Rep. 126 (1985) 1.
W. Siegel, Nucl. Phys. B 263 (1986) 93;
W. Siegel, B. Zwiebach, Nucl. Phys. B 263 (1986) 105;
T. Banks, M.E. Peskin, Nucl. Phys. B 264 (1986) 513;
M. Kato, K. Ogawa, Nucl. Phys. B 212 (1983) 443;
N. Ohta, Phys. Rev. D 33 (1986) 1681;
N. Ohta, Phys. Lett. B 179 (1986) 347;
A. Neveu, H. Nicolai, P.C. West, Nucl. Phys. B 264 (1986) 573;
A. Neveu, P.C. West, Nucl. Phys. B 268 (1986) 125;
E. Witten, Nucl. Phys. B 268 (1986) 253.
D. Francia, A. Sagnotti, Class. Quantum Grav. 20 (2003) S473, hep-th/0212185.
A.K. Bengtsson, Phys. Lett. B 182 (1986) 321;
M. Henneaux, C. Teitelboim, in: C. Teitelboim, J. Zanelli (Eds.), Quantum Mechanics of Fundamental Systems,
vol. 2, Plenum Press, New York, 1988, p. 113.
G. Bonelli, Nucl. Phys. B 669 (2003) 159, hep-th/0305155;
G. Bonelli, JHEP 0311 (2003) 028, hep-th/0309222.
A. Sagnotti, M. Tsulaia, Nucl. Phys. B 682 (2004) 83, hep-th/0311257.
J. Schwinger, Particles, Sources, and Fields, AddisonWesley, Reading, MA, 1970.
D. Francia, A. Sagnotti, Phys. Lett. B 624 (2005) 93, hep-th/0507144;
For a review see: D. Francia, A. Sagnotti, J. Phys. Conf. Ser. 33 (2006) 57, hep-th/0601199.
S. Weinberg, Phys. Rev. 133 (1964) B1318;
S. Weinberg, Phys. Rev. 134 (1964) B882;
S. Weinberg, Phys. Rev. 181 (1969) 1893;
A.K.H. Bengtsson, I. Bengtsson, L. Brink, Nucl. Phys. B 227 (1983) 31;
A.K.H. Bengtsson, I. Bengtsson, L. Brink, Nucl. Phys. B 227 (1983) 41;
A.K.H. Bengtsson, I. Bengtsson, N. Linden, Class. Quantum Grav. 4 (1987) 1333;
F.A. Berends, G.J.H. Burgers, H. Van Dam, Z. Phys. C 24 (1984) 247;
F.A. Berends, G.J.H. Burgers, H. Van Dam, Nucl. Phys. B 260 (1985) 295;
S. Deser, Z. Yang, Class. Quantum Grav. 7 (1990) 1491.
X. Bekaert, N. Boulanger, S. Cnockaert, J. Math. Phys. 46 (2005) 012303, hep-th/0407102;
X. Bekaert, N. Boulanger, S. Cnockaert, JHEP 0601 (2006) 052, hep-th/0508048;
N. Boulanger, S. Leclercq, S. Cnockaert, Phys. Rev. D 73 (2006) 065019, hep-th/0509118;
X. Bekaert, N. Boulanger, S. Cnockaert, S. Leclercq, Fortschr. Phys. 54 (2006) 282, hep-th/0602092.
M.A. Vasiliev, Phys. Lett. B 243 (1990) 378;
M.A. Vasiliev, Class. Quantum Grav. 8 (1991) 1387;
M.A. Vasiliev, Phys. Lett. B 285 (1992) 225;
For a review, see: M.A. Vasiliev, Int. J. Mod. Phys. D 5 (1996) 763, hep-th/9611024.
M.A. Vasiliev, Phys. Lett. B 567 (2003) 139, hep-th/0304049.
D. Sullivan, Publ. Math. IHS 47 (1977) 269;
R. DAuria, P. Fre, Nucl. Phys. B 201 (1982) 101;
R. DAuria, P. Fre, Nucl. Phys. B 206 (1982) 496, Erratum;
L. Castellani, P. Fre, F. Giani, K. Pilch, P. van Nieuwenhuizen, Phys. Rev. D 26 (1982) 1481;
R. DAuria, P. Fre, Print-83-0689 (TURIN), Lectures Given at the September School on Supergravity and Supersymmetry, Trieste, Italy, 618 September 1982;
P. Fre, Class. Quantum Grav. 1 (1984) L81.
M.A. Vasiliev, Yad. Fiz. 32 (1980) 855;
M.A. Vasiliev, Ann. Phys. 190 (1989) 59;
V.E. Lopatin, M.A. Vasiliev, Mod. Phys. Lett. A 3 (1988) 257.
B. Sundborg, Nucl. Phys. B (Proc. Suppl.) 102 (2001) 113, hep-th/0103247;
E. Witten, Talk at J.H. Schwarz 60, Caltech, November 2001;
E. Sezgin, P. Sundell, Nucl. Phys. B 644 (2002) 303;
E. Sezgin, P. Sundell, Nucl. Phys. B 660 (2003) 403, hep-th/0205131, Erratum;
R.R. Metsaev, Phys. Lett. B 531 (2002) 152, hep-th/0201226;
I.R. Klebanov, A.M. Polyakov, Phys. Lett. B 550 (2002) 213, hep-th/0210114;
U. Lindstrom, M. Zabzine, Phys. Lett. B 584 (2004) 178, hep-th/0305098;
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
237
L. Girardello, M. Porrati, A. Zaffaroni, Phys. Lett. B 561 (2003) 289, hep-th/0212181;

R.G. Leigh, A.C. Petkou, JHEP 0306 (2003) 011, hep-th/0304217;
R.G. Leigh, A.C. Petkou, hep-th/0309177;
H.J. Schnitzer, hep-th/0310210;
M. Bianchi, J.F. Morales, H. Samtleben, JHEP 0307 (2003) 062, hep-th/0305052;
N. Beisert, M. Bianchi, J.F. Morales, H. Samtleben, JHEP 0402 (2004) 001, hep-th/0310292;
N. Beisert, M. Bianchi, J.F. Morales, H. Samtleben, JHEP 0407 (2004) 058, hep-th/0405057.
J.E. Paton, H.M. Chan, Nucl. Phys. B 10 (1969) 516;
J.H. Schwarz, CALT-68-906-REV, Presented at 6th Johns Hopkins Workshop on Current Problems in High-Energy
Particle Theory, Florence, Italy, 24 June 1982;
N. Marcus, A. Sagnotti, Phys. Lett. B 119 (1982) 97;
N. Marcus, A. Sagnotti, Phys. Lett. B 188 (1987) 58;
For a review on open strings, see: C. Angelantonj, A. Sagnotti, Phys. Rep. 371 (2002) 1, hep-th/0204089;
C. Angelantonj, A. Sagnotti, Phys. Rep. 376 (2003) 339, Erratum.
E. Sezgin, P. Sundell, JHEP 0207 (2002) 055, hep-th/0205132.
E. Sezgin, P. Sundell, hep-th/0508158;
E. Sezgin, P. Sundell, hep-th/0511296.
J.H. Schwarz, Phys. Rep. 89 (1982) 223.
C. Fronsdal, Phys. Rev. D 20 (1979) 848.
J. Fang, C. Fronsdal, Phys. Rev. D 22 (1980) 1361.
A. Lichnerowicz, Propagateurs et commutateurs en Relativit Gnrale, Publ. Math. lIHES 10 (1961).
H. van Dam, M.J.G. Veltman, Nucl. Phys. B 22 (1970) 397;
V.I. Zakharov, JETP Lett. 12 (1970) 312, Pisma Zh. Eksp. Teor. Fiz. 12 (1970) 447.
J. Scherk, J.H. Schwarz, Nucl. Phys. B 153 (1979) 61.
A. Higuchi, Nucl. Phys. B 282 (1987) 397;
M. Porrati, Phys. Lett. B 498 (2001) 92, hep-th/0011152.

Nucl - Phys.B v.773

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Nucl - Phys.B v.773

Загружено:

Авторское право:

Доступные форматы

NUCLEAR PHYSICS B

Journal devoted to the experimental and theoretical study of the fundamental

Nuclear Physics B 773 (2007) 118

Relation between the pole and the minimally subtracted

Received 19 February 2007; accepted 5 March 2007

E-mail address: piclum@particle.uni-karlsruhe.de (J.H. Piclum).

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

2. On-shell counterterms for mass and wave function in dimensional regularization

where the renormalization constants are defined as

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

OS one only needs to calculate for q 2 = M 2 . To calculate Z OS , one has to

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

case of the bottom and top quark. As input we use s

(Mt ) using four-loop accuracy4 leading to

s(5),DR (MZ ) = e(5) (MZ ) = 1 (MZ ),

e(5) (Mb ) = 0.1721,

active flavours and thus we evaluate in a first step s

(MZ ) = 0.1178 and pose the analogue

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

s(6),DR (Mt ) = e(6) (Mt ) = 1 (Mt ) = 0.1096,

= (4.800 0.544 0.238 0.159) GeV,

GeV = 171.4(1 0.0581 0.0105 0.0030) GeV

P. Marquard et al. / Nuclear Physics B 773 (2007) 118

K.G. Chetyrkin, M. Steinhauser, Phys. Rev. Lett. 83 (1999) 4001, hep-ph/9907509.

Nuclear Physics B 773 (2007) 1942

Direct detection of dark matter rates for various wimps

E-mail address: vergados@cc.uoi.gr (J.D. Vergados).

V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942

V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942

2.1. Intermediate KK quarks

M1 (u) = i(g1 4/3) uR (x)

= i(g1 ) 4/9 uL (x)

(pq pB  )2 m2q (1)

= i(g1 ) 4/9 uL (x)

uR (x) + H.C.  (pB ) (pB ),

V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942

(pq + pB )2 m2q (1)

The results in this case are less sensitive to .

V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942

V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942

In the proton neutron representation we obtain:

The picture is different for the neutralino case in which [20],

Thus at the nucleon level we get

2.2. Intermediate Higgs scalars

V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942

3. KK neutrinos as dark matter candidates

Thus in the case for the proton we encounter the combination,

Regarding the neutrino current we now have two possibilities:

V.K. Oikonomou et al. / Nuclear Physics B 773 (2007) 1942

In other words in the case of the Majorana neutrino we get:

3.2. Process mediated by right-handed currents via Z  -boson exchange

(pq pB )2 m2q (1)

uR (x) + H.C. (pB ) (pB ),

3.2. Process mediated by right-handed currents via Z -boson exchange