Received 24 April 2006; received in revised form 12 August 2006; accepted 18 August 2006
Available online 7 September 2006
Abstract
Using a peculiar version of the SU(3)L U(1)N electroweak model, we investigate the production of
doubly charged Higgs boson at the Large Hadron Collider. Our results include branching ratio calculations
for the doubly charged Higgs and for one of the neutral scalar bosons of the model.
2006 Elsevier B.V. All rights reserved.
PACS: 11.15.Ex; 12.60.Fr; 14.80.Cp
1. Introduction
In the last few years we have seen a tremendous experimental progress in the realm of the
weak interactions. However, these advances do not attain the scalar sector yet. This is the sense
in which LHC (Large Hadron Collider) facilities may shed some light especially on the Higgs
boson. One of the main ingredients of the Standard Model is the Higgs mechanism which, in
principle, explains how the particles gain masses through the introduction of an isodoublet scalar
field. The scalar field is the responsible for the spontaneous breakdown of the gauge symmetry.
After electroweak symmetry breaking, the interaction of this scalar with the gauge bosons and
* Corresponding author.
fermions generate the mass of the particles. In this process there remains a single neutral scalar,
manifesting itself as the Higgs particle.
The Standard Model possibly is a low energy effective theory which must be generalized by
some GUT (grand unified theory). However, there are several motivations to extend the electroweak theory below the GUT scale. Supersymmetric models, for example, provide a solution
to the hierarchy problem through the cancellation of the quadratic divergences via fermionic and
bosonic loop contributions [1]. The little Higgs model, recently proposed, predicts that the Higgs
boson is a pseudoGoldstone boson of some approximate broken global symmetry [2]. Therefore, this model is also able to solve the naturalness problem of the Higgs mass. One of the main
motivations for the leftright models and for the study of their phenomenological consequences
is that in which the Higgs triplet representation furnishes a seesaw type neutrino mass matrix associated with the presence of a doubly charged Higgs boson [3]. Therefore, these models suggest
a route to understanding the pattern of the neutrino masses.
Doubly charged Higgs bosons appear in several popular extensions of the Standard Model
such as leftright symmetric models [3] and Higgs triplet models [4]. However, there is another
interesting class of electroweak models which also predict particles like that. This class of models
is called 331 models [5,6]. This is the simplest chiral extension of the Standard Model. It
is able to solve the fermion familys replication problem through of a simple relation between
the number of colors and the anomaly cancellation mechanism. It is important to notice that a
solution to this problem is not furnished even in the context of the GUTs. The 331 models have
other interesting features, as for example the upper bound on the Weinberg mixing angle, through
the relation sin2 W < 1/4. This feature does not happen in any kind of others electroweak models
except GUTs, where the value of sin2 W is predicted. This result leads to an upper bound for
the energy scale of the model when this parameter is evolved to high values [7]. In a similar
fashion as occurs in leftright model, the seesaw mechanism can be naturally incorporated in
some versions of the 331 models [8].
No Higgs bosons have yet been found. In the meantime, it is the last brick that is lacking to
finish the construction of the building of the standard electroweak theory. Besides, it is possible
that the Higgs sector brings to light a nonstandard physics.
Since that the 331 models are good candidates for physics beyond the Standard Model,
it is interesting to evaluate if the future accelerators will produce events in sufficient numbers
to detect some of the 331 Higgs bosons. In particular, there is an increasing interest in the
phenomenology associated with doubly charged Higgs bosons, a kind that appears in models that
admit scalars in triplet representation of the gauge group [9]. Here we are interested in one of such
version of the 331 models for which the scalar fields come only in triplet representation [6,8].
It predicts three new neutrals, four single charged and two doubly charged Higgs bosons. These
scalars can be disclosed in relatively low energies, which make them interesting for searches in
the next generation of particle accelerators.
Our work is organized as follows. In Section 2 we summarize the relevant features of the
model, in Section 3 we present the cross section calculations and in Section 4 we give our conclusions.
2. Overview of the model
The underlying electroweak symmetry group is SU(3)L U(1)N , where N is the quantum
number of the U(1) group. Therefore, the lefthanded lepton matter content is (a a La )TL transforming as (3, 0), where a = e, , is a family index (we are using primes for the interaction
eigenstates). LaL are lepton fields which can be the charge conjugates aR C [5], the antineutrinos
C [10] or heavy leptons P + (P + = E + , M + , T + ) [6].
La
aL
aL
L
L
L
The model of Ref. [6] has the simplest scalar sector for 331 models. In this version the
charge operator is given by
Q 1
(1)
= (3 38 ) + N,
e
2
where 3 and 8 are the diagonal GellMann matrices and e is the elementary electric charge. The
righthanded charged leptons are introduced in singlet representation of SU(3)L as
aR (1, 1)
+
(1, 1).
and PaR
The quark sector is given by
u1
d
2
1
Q1L = d1 3,
(2)
,
QL = u 3 , ,
3
3
J1 L
J L
where = 2, 3, J1 and J are exotic quarks with electric charge 53 and 43 respectively. It must
be notice that the first quark family transforms differently from the two others under the gauge
group, which is essential for the anomaly cancellation mechanism [5].
The physical fermionic eigenstates rise by the transformations
L,R
aL,R = Aab bL,R ,
UL,R
=U
L,R
UL,R ,
L,R +
+
PaL,R
= Bab
PbL,R ,
DL,R
=D
L,R
DL,R ,
(3a)
JL,R
=J
L,R
JL,R ,
(3b)
where UL,R = (u c t)L,R , DL,R = (d s b)L,R , JL,R = (J1 J2 J3 )L,R and AL,R , B L,R , U L,R ,
DL,R , J L,R are arbitrary mixing matrices.
The minimal scalar sector contains the three scalar triplets
0
+
= 1 (3, 0),
(4)
=
(3, 1),
= (3, 1).
2+
++
The most general, gauge invariant and renormalizable Higgs potential, which conserves the leptobaryon number [11], is
2
2
2
V (, , ) = 21 + 22 + 23 + 1 + 2 + 3
+ 4 + 5 + 6 + 7
1
+ 8 + 9 + f ij k i j k + H.c. .
(5)
2
The neutral components of the scalars triplets (4) develop nonzero vacuum spectration val2 = (246 GeV)2 . This mechaues 0 = v , 0 = v and 0 = v , with v2 + v2 = vW
nism generate the fermion and gauge boson masses [12]. The pattern of symmetry breaking
,
using the eigenstates and masses (see Appendix B) of Ref. [12]. For others analysis of the 331
Higgs potential see Ref. [13].
Symmetry breaking is initiated when the scalar neutral fields are shifted as = v + + i ,
with = 0 , 0 , 0 . Thus, the physical neutral scalar eigenstates H10 , H20 , H30 and h0 are related
to the shifted fields as
0
H1
1 c s
(6a)
ih0 ,
,
H30 ,
vW s c
H20
and in the charged scalar sector we have
1+ = s H1+ ,
2+ = s H2+ ,
+ = c H1+ ,
+ = c H2+ ,
++ = s H ++ ,
(6b)
++ = c H ++ ,
(6c)
+ G Q 1L J1R +
j
j
G QL JR
+ H.c.
(7a)
(7b)
D i = i ig W .
j ig N i B ,
2 i
(8)
c V
+
P H2 + c e V eP H + v + H30 + ih0 P + M E GR P +
v
+ H.c.,
(9a)
1
c s
c
s c
s
u
0
LQ =
+
+
U GR 1 +
V H1 + +
V u H20 M u U
2
v
v
v
v
v v
R 1 + c + s c V D H 0 + s c + s V D H 0 M d D
+ DG
1
2
v
v
v
v
v
v
c s
s
+ U GR
VCKM H1 +
V ud H1+ M d D
v
v
v
R c VCKM H + + s c V ud H M u U + H.c.,
+ DG
(9b)
1
1
v
v v
1
LJ = JGR J L N U L M u U + RDL M d D
2
L X2 + JJ L X0 J L M J GR J + H.c.,
+ U U L X1 + DD
(9c)
where GR = 1 + 5 , VLU VLD = VCKM , the CabibboKobayashiMaskawa mixing matrix, U e ,
U P e , V e , V eP , V u = VLU VLU , V d = VLD VLD and V ud = VLU VLD are arbitrary mixing
matrices, M e = diag(me m m ), M P = diag(mE mM mT ), M u = diag(mu mc mt ), M d =
diag(md ms mb ) and M J = diag(mJ1 mJ2 mJ3 ). In Eq. (9c) we have defined
s H2+ /v
N =
0
0
s H /v s H /v ,
s H /v s H /v
1 0 0
v + H30 + ih0
0 1 1,
X0
v
0 1 1
s H ++ /v
0
0
0
s H2 /v s H2 /v ,
R=
0
s H2 /v s H2 /v
0
0
c H2
1
X1 =
0
c H ++ c H ++ ,
v
0
c H ++ c H ++
c H
0
0
1
0
c H2+ c H2+ .
X2 =
v
0
c H2+ c H2+
(10a)
(10b)
(10c)
We call attention to the fact that nonstandard field interactions violate leptonic number, as can
be seen from the Lagrangians (7b) and (8). However the total leptonic number is conserved [5].
3. Cross section production
The main mechanism for the production of Higgs particles in pp collisions occurs through
the mechanism of DrellYan and gluongluon fusion as shown in Fig. 1. In all calculations we
are considering that the charged fermionic mixing matrices [see Eqs. (3)] are diagonals. Using
the interaction Lagrangians (5) and (9) we first evaluate the differential cross section for the
DrellYan process, i.e., pp H ++ H through the exchange of , Z, Z , H10 and H20 in the
Fig. 1. Feynman diagrams for production of doubly charged Higgs bosons via (a) DrellYan process, where
X = , Z, Z , H10 , H20 and (b) gluongluon fusion.
schannel. Therefore we obtain the differential cross section for this reaction as
d
1
M 2 + MH 0 2 + MH 0 2 + MZ 2 + MZ 2 + 2 Re MH 0 MH 0 ,
=
2
1
2
1
2
d
64 s
where
d
H [1 (1) (s )mq v v ]2 + [2 (2) (s )]2 (v4 m2u + v4 m2d )
=
d cos
24
(2vW v v )2
2
8 mq + m2H 4u 4t
2 Qq 2
2
2
2
2
+ 3
s 2mq s 4mH s + 2mq (t u)
sin W
s
+
H 2 2Z(Z )
q(q )2
8m4H gV
+ gA
q(q )2
q(q )2
q(q )2
q(q )2
+ 8m2H 2m2q t u gV
+ 8m4q gV
+ gA
+ gA
q(q )2
q(q )2
q(q )2
8m2q (t + u)
+ 8m2q s gA
gV
+ gA
q(q )2
q(q )2
q(q )2
q(q )2
+ 2s 2 gV
.
+ 2(t + u)
2 gV
+ gA
+ gA
Z,Z
(11)
Here, s is the CM (center of mass) energy of the q q system. For the Standard Model parameters
we assume PDG values, i.e., mZ = 91.19 GeV, sin2 W = 0.2315, and mW = 80.33 GeV [14].
Z(Z ) are the total width of the boson Z(Z ) [15,16]. The velocity of the Higgs in the CM of the
process is denoted through H . The i (i = 1, 2) are the vertex strengths for H10 H H ++
and H20 H H ++ , respectively, while is one for H H ++ and the Z(Z ) is for the
bosons Z(Z )H H ++ . The analytical expressions for these vertex strengths are
(12a)
+ c 2 5 c2 + 4 s2 v + f c s ,
2
2 = ic 2 (5 + 6 + 9 )c + 2(22 4 9 )s v + f c s ,
(12b)
2
= e c s2 (p q) ,
(12c)
Z = e
4 sin2 W (v2 v2 ) v2
4(v2 + v2 ) sin W cos W
Z = e
(p q) ,
(12d)
(12e)
The Higgs parameters i (i = 1, . . . , 9) must run from 3 to +3 in order to allow perturbative calculations. For H0 ( = 2, 3) we take mH0 = (0.23.0) TeV, while we assume
mH 0 = 230 GeV for the Standard Model scalar one. It must be notice that here there is no con1
tribution from the interference between the scalar particle H10 and a vectorial one ( , Z or Z ).
The kinematic invariant t and u are,
2
4m
4m2H
s
q
1
1 cos 1
,
t = m2q + m2H
(13a)
2
s
s
2
4m
4m2H
s
q
2
2
u = mq + mH
(13b)
1
1 + cos 1
.
2
s
s
The total cross section ( ) for the process pp H ++ H is related to the cross section ( ) of
the subprocess qq H ++ H through
=
1
ln min
qi =u,d,s,c min ln
min
d dy fqi
y y
e fqi e (, s ),
(14)
d
MH 0 2 + MH 0 2 + MH 0 2
=
1
2
3
d
64 2 s
+ 2 Re MH 0 M 0 + Re MH 0 M 0 + Re MH 0 M 0 .
1
H2
H3
H3
(15)
In order to make explicit the different contributions to the elementary cross section, we will
present them separately. The quarkloop contributions involve the Higgs Hi , where i = 1, 2, 3,
which are exchanged in the schannel. For the H10 we have
H 0
1
2
2
H ++
(1) 2 1
2
s 1
=
q1 mq Id + s
q1 q 2 + (4q 1)Iq
s q
8192 3
q
r
r
+q
+q
2
m2q q ln
ln2
2 .
r
r
q
q
q
(16)
We fix the scale parameter = 0.2 GeV and the appropriate scale where the strong coupling
constant s is evaluated as being equal to the center of mass energy of the subprocess, both, used
in our numerical calculations. The sums in Eq. (16) run over all generations where q = m2q /s
and rq = 1 1 4q . The q1 is the q qH
10 vertex strength,
mq
GR .
2vW
q1 = i
(17)
1
,
s m2Hi + imHi Hi
(18)
where H 0 are the total width of the H10 and H20 boson, with i = 1 for q = u, c, t, d, s, b and
i
i = 2 for u(c, t) or d(s, b), separately [16]. Eq. (18) defines also the propagator for H30 with
i = 3. However, H30 does not contribute to the DrellYan process.
For the Higgs H20 we have
2
(2) 2 1
H ++
2 v
2 v
H 0 =
(
s
)
+
m
Iq 2
m
s 2
q2
u
d
2
s q
v
v
8192 3
2
v v 2
+ s
+
q2 q 2 + (4q 1)Iq
v
v q
v2
v2
r+q
2
2
+ 4
q2 ln
(19)
mu 2 + md 2 q ,
rq
v
v
q
20 vertex, i.e.:
where q2 is the strength of the q qH
v
v
i
q2 =
GR mu + md
.
2vW
v
v
(20)
2
H
2 3
J 3 J 2 + (4J 1)IJ ,
H 0 =
3
8192 3
(21)
where J 3 describe the quark vertex with the Higgs H30 and 3 for the H30 with H , i.e.,
mJ
GR ,
2v
3 = 2iv (6 9 )s2 + (23 + 9 )c2 .
J 3 = i
(22a)
(22b)
We note that H30 decays only into exotic leptons and quarks because it becomes massive at the
first symmetry breaking. Therefore, the H30 total width is obtained from
H30 all = J0 J + J0 J + J0 J + 3P0 P + H0 0 H 0 + H0 H
1 1
+ H0 H
2
2
2 2
3 3
+ H0 H
1
2
+ U0 U ,
(23)
+ q
+ J
+ e
H all = J
P + +
q
q
u,c,t J2,3
W H
H H
2,3 u,c,t
1 d,s,b
+ U
+ U
h0
+ U
Z
+ U
Z
H10
U
+ V
H
1
H20
H30
(24)
with XY
= (H XY ) (see Appendix A for the partial widths).
The contribution for the interference of the H10 and H20 is given by
d H 0 H 0
1
4 1 2 k k (2) (s )ab
s
1
2
8
q ,q
u
ka1 kb2 ab
+i
2 + (4 1)Iq
2
s 2
(a) (b)
k1 k2 k1 k2 v 2
v 2
m Iq m Iq
v u u v d d
s 2
ka1 kb2 v
+i
,
U 2 + (4U 1)Iqu D 2 + (4D 1)Iqd
2
v
v
(a)
k1 k(b)
k1 k2
2
m2q Iq
(25)
where , are the polarizations, k1,2 are the gluon momentum vectors, mqu are the masses of
the u, c, t quarks (5 MeV, 1.5 GeV and 175 GeV respectively), mqd are the masses of the d, s,
b quarks (9 MeV, 150 MeV and 5 GeV respectively) [14], is referred to all quarks and U for
the quark u, c, t and D for d, s, b respectively and is the antisymmetric tensor.
The contribution for the interference of H10 and H30 gives
d H 0 H 0
1
s4 1 3 q1 q3 (1) (s ) (3) a b qa qb ab
1
2
i
2 + (4 1)Iq
=
2
256
q
a b
q q
q(a) q(b)
2
i 1 2 2 + (4 1)Iq 1
,
2
s 2
(26)
and finally we have the following for the H20 and H30
d H 0 H 0
2
s4 2 3 q2 q3 (2) (s ) (3) (s )a b qa qb ab
1
2
i
2 + 4(4 1)Iq
=
2
256
q
v 2
v 2
(a) (b)
q1 q2
m Iq m Iq
v u u v d d
qa1 qb2 v
+i
.
u 2 + (4u 1)Iqu d 2 + (4d 1)Iqu
2
v
v
(27)
10
The loop integrals involved in the evaluation of the elementary cross section can be expressed in
terms of the function Iq (q ) Iq which is defined through
dx
(1 x)x
Iq =
ln 1
x
q
0
q > 1/4,
1 4 arcsin2 [1/(2 q )],
=
2
2 [ln(r+q /rq ) + 2i] ln(r+q /rq ) , q < 1/4.
1
(28)
Here, q stands for the quarks running in the loop (Fig. 1(b)).
The total cross section ( ) for the process pp gg H H ++ is related to the total cross
section ( ) through the subprocess gg H H ++ , i.e.,
1
ln min
d dy G ey , Q2 G ey , Q2 (, s ),
=
min
(29)
ln min
where G(x, Q2 ) is the gluon structure function, for which the factorization scale is taken equal
to the center of mass energy of the subprocess and used in our numerical calculations.
4. Results and conclusions
In this work we have calculated the pair production of doubly charged Higgs by computing
the contributions due the DrellYan and quark loop processes. In Section 3 we have given the
analytical expressions that allow the numerical evaluations of these contributions and it was
showed that the dominant contribution come from the wellknown DrellYan process. We have
presented the cross section for the process pp H H ++ involving the DrellYan mechanism
and the gluongluon fusion, to produce such Higgs bosons at the LHC (14 TeV).
Taking into account that the masses of the gauge bosons, Higgs and some other parameters
must satisfy the limits imposed by the model (see Section 2), besides the approximations in the
calculations of masses (Appendix B) and eigenstates (Section 2) given in Ref. [12], we have
considered two possibilities: f 0 and f = 99.63 GeV (see Table 1). For both possibilities
we have assumed the values v = 195 GeV, v = 1300 GeV, 1 = 1.2, 2 = 3 = 6 =
8 = 1, 4 = +2.98, 5 = 1.56, 7 = 3, but for 9 we have used 9 = 1 when f 0 and
9 = 1.9 when f = 99.63 GeV.
The masses of the exotic bosons in Table 1 are in accordance with the estimated values of
CDF and D experiments, which probe their masses in the range from 500 GeV to 800 GeV
[17], while the reach of the LHC will be in the range 1 TeV < mZ 5 TeV [18].
In Fig. 2 we show the cross section for the process pp H ++ H for f 0 GeV. Considering that the expected integrated luminosity for the LHC will be of order of 3 105 pb1 /yr
Table 1
Approximated values of the masses (see Appendix B) used in this work. All the values in this table are given in GeV
f
mE mM
mT
0
194 1138 2600
99.63
mH 0 mH 0
1
2
mH 0
mh 0 mH mH mV
1
2
mU
mZ
m J1
m J2
m J3
874
2600
0
520
601
2220
1300
1833
1833
1322
426
218
1315
1295
603
Fig. 2. Total cross section for the process pp H H ++ as a function of mH for f = 0 GeV at
for DrellYan (solid line) and gluongluon fusion (dotdash line).
11
s = 14 000 GeV
Fig. 3. Total cross section for the process pp H H ++ as a function of mH for f = 99.63 GeV at
s = 14 000 GeV for DrellYan (solid line) and gluongluon fusion (dotdash line).
then the statistics give a total of 2 104 events per year for DrellYan and 0.2 events per
year for gluongluon fusion, for mH = 1309 GeV.
In Fig. 3, we show the results for the same process when f = 99.67 GeV. This value was
calculated considering the exotic boson masses in the range from 500 to 800 GeV and v having a
minimum value of 194.5 GeV, which assure the values of the i between 3 to +3, so we obtain
the mass of the doubly charged Higgs, i.e., mH ++ = 1780 GeV. Considering the same integrated
luminosity as above this gives a total of 585 events per year for DrellYan and 0.13 events per
12
Table 2
Branching ratios for the H30 decay with mH 0 = 2600 GeV. The notation used in this table is BR0XY = BR(H30 XY )
3
f (GeV)
105 BR0
108 BR0
BR0
BR0
BR0E + E
BR0M + M
0
99.63
3.35
3.14
2
4
no
4 107
0.9999
2 107
2 106
H10 H20
H1+ H1
H2+ H2
H1+ H2
Table 3
3 BR(H XY )
Branching ratio for the H decay with mH = 1309 GeV. Here BR
XY = 10
f (GeV)
BR
BR
E
BR
M
0
99.63
0.001
2
0.08
0.001
0.005
0.004
f (GeV)
BR
BR
BR
BR
0
99.63
no
0.09
19
13
no
329
no
6
J1 q
W H2
V H1
BR++
+ E +
3
0.4
H1 H2
BR++
+ M +
6
4
H10
BR
U Z
444
146
BR
U
29
2
BR
h0
2
0.5
year for gluongluon fusion. We present in Table 2 the branching ratios for H30 all with f 0
and f = 99.63 GeV and we can observe that, due to the coupling constant, the largest width
corresponds to H30 H1+ H2 decay. In Table 3 we present the branching ratios for H
all. From Table 3 it can also be noticed that, as the branching ratio (BR) for H H1 H2 ,
H U Z and H U are large, these channels could lead to some interesting
signal.
We emphasize that the window for varying the free parameters is small because of the constraints imposed on the model. In summary, the analysis of the values in Tables 13 show that,
although a large number of doubly charged Higgs can be produced by the DrellYan mechanism,
the decays of these particles into ordinary quarks and leptons do not lead to a good signature for
its detection even for energies which the LHC can reach.
Appendix A. Partial widths
In this appendix we present the partial widths for Higgs decays from Eqs. (23) and (24). We
define
1
R(a, b; x) =
(A.1)
x (ma + mb )2 x (ma mb )2
16x 3/2
mJi 2 2
J0 J = 3R(Ji , Ji ; s)
mH 0 2m2Ji ,
i i
v
3
2
mP 2
mH 0 2mP ,
P0+ P = R P + , P ; s
v
3
0 0
2
0
H 0 H 0 = R H1 , H2 ; s 4(5 6 )c v v + f (c v s v ) ,
1 2
2
H0 H + = R H1+ , H1+ ; s 2v (5 s v + 6 c v ) + f c v ,
1
(A.2a)
(A.2b)
(A.2c)
(A.2d)
2
H0 H + = R H2+ , H2+ ; s 2v (5 c v + 6 s v ) + f s v ,
2
2
0
H H + = R H1+ , H2+ ; s (6 5 )c v v + f (s v c v ) ,
1
m 0 2
m 0 2
g 4 v
H3
H3
1
,
R U ,U ;s 3
2
mU ++
2mU ++
13
(A.2e)
(A.2f)
U0 U ++ =
(A.2g)
H0 H ++
(A.2h)
s = mH ++ ,
mq s 2 2
mH ++ m2J1 m2d,s,b ,
= 3R(J1 , q; s)
1 d,s,b
v
mJ2,3 c 2 2
++
qu,c,t J2,3 = 3R(J2,3 , q; s)
mH ++ m2J2,3 m2u,c,t ,
v
2
mqu,c,t s 2 2
J++
mH ++ m2J2,3 m2u,c,t ,
=
3R(J
,
q;
s)
2,3
2,3 qu,c,t
v
2
R(e, P ; s) me s 2
e++
mH ++ m2e m2P ,
P =
4
v
R(e, P ; s) mP c 2 2
++
e+ P + =
mH ++ m2e m2P ,
4
v
2
R(W, H2 ; s)
ec c
++
W H =
32
sin W mW
2
2
2
mH ++ m2W + m2H + m2H + m2H ++ + m2W ,
2
2
H++
(7 + 9 )s2 + (7 + 8 )c2 s s
= R H2 , H2 ; s
H
1
2
(8 9 )c c v + f s s
,
+
v2 + v2
J++
q
U++
++
++
3
=
4mH ++
H10
H20
e 2 c v
2
mU ++
1
mH ++
(A.3b)
(A.3c)
(A.3d)
(A.3e)
(A.3f)
(A.3g)
2
,
sin2 W
2
R(U , H10 ; s)
ec v
=
32
mU ++ vW sin W
2
2
2
mH ++ mU ++ + m2H 0 m2H 0 2 m2H ++ + m2U ,
=
(A.3a)
2
R(U , H20 ; s)
ec v
32
mU ++ vW sin W
2
2
2
mH ++ mU + m2H 0 m2H 0 2 m2H ++ + m2U ++ ,
(A.3h)
(A.3i)
(A.3j)
14
++
H30
2
R(U , H30 ; s)
ec
32
mU ++ sin2 W
2
mH ++ m2U ++ + m2H 0 m2H 0 2 m2H ++ + m2U ++ ,
3
3
2
2
c
v
e
U++
, Z; s
Z = R U
cos
2
(m ++ m2Z )2 + m2U ++ (m2U ++ 2m2H ++ )
5+ H
,
2m2U ++ m2Z
ev v (10 sin2 W 1) 2
R(U , Z ; s)
U++
=
Z
12(sin2 W 1)(4 sin2 W 1)
sin2 W
2
2
2
2
2
(m ++ mZ ) + mU ++ (mU ++ 2m2H ++ )
5+ H
,
2m2U ++ m2Z
2
ev s
=
R
V
,
H
;
s
V++
H
1
vW sin W
1
2
2 2
mH ++ mV + m2H 0 m2H + 2 m2H ++ + m2V .
(A.3k)
(A.3l)
(A.3m)
(A.3n)
++
(H ++ XY ).
Here, XY
2 v4 21 v4
m2H 0
v2 v2
2
2
f v 2
v v
m2h =
vW +
,
v v
v
1
mH =
v2 + v2
m2H =
1
m2H 0 43 v2 ,
2
vW
(f v 27 v v ),
2v v
m2H =
(B.1a)
v2 + v2
(f v 29 v v ),
2v v
2v v
2 2
2 2
v + v2
v + v2
1 evW 2
e
e
2
2
2
,
mV =
,
mU =
,
mW =
2 sW
sW
2
sW
2
2 )
ev 2
ev 2 2(1 sW
1
,
m2Z
.
m2Z
2 )
sW
2(1 sW )
sW
3(1 4sW
2
(f v 28 v v ),
2
vW
v2 ,
2v v2
(B.1b)
(B.1c)
(B.1d)
(B.1e)
2 v2 1 v2
v2 v2
5 v2 + 26 v2
v v
.
2
(B.2)
15
G1u v v
,
G2u v + G3u v
G1d v v
G2d v + G3d v
mc =
G1c v + G2c v
v ,
G3c v + G4c v
mt = G1t v + G2t v ,
(B.3a)
ms =
G1s v + G2s v
v ,
G3s v + G4s v
mb = G1b v + G2b v ,
(B.3b)
and from the Lagrangian (9c) the heavy quark masses are mJ1 , mJ2 , mJ3 v . In Eqs. (B.3) the
q
parameters G , q = u, c, t, d, s, b and = 1, 2, 3, 4 are functions of the Yukawa couplings.
References
[1] J. Wess, B. Zumino, Nucl. Phys. B 70 (1974) 39;
J. Wess, B. Zumino, Phys. Lett. B 49 (1974) 52;
J. Iliopoulos, B. Zumino, Nucl. Phys. B 76 (1974) 310;
S. Ferrara, J. Iliopoulos, B. Zumino, Nucl. Phys. B 77 (1974) 413;
E. Witten, Nucl. Phys. B 188 (1981) 513.
[2] N. ArkaniHamed, et al., JHEP 0208 (2002) 021;
M. Scmaltz, Nucl. Phys. B (Proc. Suppl.) 117 (2003) 40.
[3] J.C. Pati, A. Salam, Phys. Rev. D 10 (1974) 275;
R.N. Mohapatra, J.C. Pati, Phys. Rev. D 11 (1975) 566;
G. Senjanovic, R.N. Mohapatra, Phys. Rev. D 12 (1975) 1502;
T. Rizzo, Phys. Rev. D 25 (1982) 1355.
[4] H. Georgi, M. Machacek, Nucl. Phys. B 262 (1985) 463;
J.F. Gunion, R. Vega, J. Wudka, Phys. Rev D 42 (1990) 1673;
J.F. Gunion, C. Loomis, K.T. Pitts, hepph/9610237.
[5] F. Pisano, V. Pleitez, Phys. Rev. D 46 (1992) 410;
P.H. Frampton, Phys. Rev. Lett. 69 (1992) 2889;
R. Foot, O.F. Hernandez, F. Pisano, V. Pleitez, Phys. Rev. Lett. 47 (1993) 4158.
[6] V. Pleitez, M.D. Tonasse, Phys. Rev. D 48 (1993) 2353.
[7] A.G. Dias, R. Martinez, V. Pleitez, Eur. Phys. J. C 39 (2005) 101;
P. Jain, S.D. Joglekar, Phys. Lett. B 407 (1997) 151.
[8] N.V. Cortez, M.D. Tonasse, Phys. Rev. D 72 (2005) 073005;
J.C. Montero, C.A. de S. Pires, V. Pleitez, Phys. Rev. D 65 (2002) 095001.
[9] For some recently papers see A.G. Akeroyd, M. Aoki, Phys. Rev. D 72 (2005) 035011;
CDF Collaboration, D. Acosta, et al., Phys. Rev. Lett. 95 (2005) 0151801;
CDF Collaboration, D. Acosta, et al., Phys. Rev. Lett. 93 (2004) 221802;
D Collaboration, V.M. Abazov, et al., Phys. Rev. Lett. 93 (2004) 141801;
L3 Collaboration, P. Achard, et al., Phys. Lett. B 576 (2003) 18;
OPAL Collaboration, G. Abbiendi, et al., Phys. Lett. B 577 (2003) 93.
[10] J.C. Montero, F. Pisano, V. Pleitez, Phys. Rev. D 47 (1993) 2918;
H.N. Long, Phys. Rev. D 59 (1996) 437;
R. Foot, H.N. Long, T.A. Tran, Phys. Rev. D 50 (1994) 34.
[11] V. Pleitez, M.D. Tonasse, Phys. Rev. D 48 (1993) 5274.
[12] M.D. Tonasse, Phys. Lett. B 381 (1996) 191.
[13] P.V. Dong, H.L. Long, D.T. Nhung, Phys. Rev. D 73 (2006) 035004;
R.A. Diaz, R. Martinez, F. Ochoa, Phys. Rev. D 69 (2006) 095009, and references cited therein.
[14] Particle Data Group, S. Eidelman, et al., Phys. Lett. B 592 (2004) 1.
[15] J.E. Cieza Montalvo, M.D. Tonasse, Nucl. Phys. B 623 (2002) 325.
[16] J.E. Cieza Montalvo, M.D. Tonasse, Phys. Rev. D 71 (2005) 095015.
[17] M. Carena, A. Daleo, B.A. Dobrescu, T.M.P. Tait, Phys. Rev. D 70 (2004) 093009;
D Collaboration, V.M. Abazov, et al., Phys. Rev. Lett. 87 (2001) 061802;
CDF Collaboration, T. Pratt, Talk at the SUSY 2004 Conference, June 2004.
[18] A. Freitas, Phys. Rev. D 70 (2004) 015008.
Abstract
Fundamental superstrings (Fstrings) and Dstrings may be produced at high temperature in the early
Universe. Assuming that, we investigate if any of the instabilities present in systems of strings and branes
can give rise to a phenomenologically interesting production of gravitons. We focus on Dstrings and find
that Dstring recombination is a far too weak process for both astrophysical and cosmological sources. On
the other hand if Dstrings annihilate they mostly produce massive closed string remnants and a characteristic spectrum of gravitational modes is produced by the remnant decay, which may be phenomenologically
interesting in the case these gravitational modes are massive and stable.
2006 Elsevier B.V. All rights reserved.
1. Introduction
After the dismissal of cosmic string as candidate seeds for the formation of structure, they
received a fair share of attention in the last few years from both the theoretical and observational
perspective.
Scenarios of inflation involving branes have become popular recently [1,2], because of the
appealing feature of providing a working model of inflation based on a fundamental theory on one
side, and because they can represent a phenomenological handle for string theory on the other.
* Corresponding author.
17
In this kind of scenarios the production of Dstrings as cosmic defects, by Kibble mechanism, is
a generic prediction [3], even if their (meta)stability is not at all ensured [4].
In general, cosmic defects like monopoles, strings and walls can be problematic as their redshift is slower than the one of radiation: if present in the early Universe and if cosmologically
stable they tend to overclose the Universe, unless they are diluted by some release of entropy into
the Universe after their creation. While this is certainly true for stable monopoles, whose annihilation rate is negligible for standard choice of parameters [5], in the standard scenario cosmic
strings loose energy by chopping off small loops and finally reach scaling solutions on which
they make up a constant fraction of the energy density of the Universe during both the era of
radiation and matter domination [6]. This allows the density of long strings to be always roughly
few per Hubble volume, while the small loops disappears by emitting gravitational radiation.
The situation is less clear for domain walls, as the existence of scaling solutions and the time
required for the solution to fall into these kind of attractors in presently under investigation [7].
In this work we concentrate on cosmic strings. In particular we want to identify cosmic strings
with the Dstrings of string theory.
The production of lowerdimensional defect can result for instance from Dp Dpbrane annihilation, which will produce D(p 2n)branes as remnants. In a type II theory Dpbranes
are BPS only if p has a definite parity (even in IIA, odd in IIB), Dpbranes with the wrong
dimensionality will decay immediately (on time scale given by the string scale).
Actually in type I string theory non BPS D0branes can be stable, because the unstable mode
is projected out by the orientation projection taking type IIB into type I. Still it has to be remembered that the presence of a type I excitation spectrum is related to the presence of orientifolds.
Even if the starting setup is endowed with a space filling orientifold, since some of the dimensions
may get small during cosmological evolution, Tdualities over those directions getting smaller
than the string length confine the orientifold hyperplane to subspace of strictly positive codimension, so the bulk physics, away from the orientifolds, will be again unoriented, i.e. of type IIA or
IIB. Thus even taking a full type I configuration as a starting point, the general assumption that
only the BPS Dpbranes are stable is justified on general grounds.
Strictly speaking BPS Dstrings are present in type IIB and type I theories and not in type IIA,
but we can consider Dpbranes with arbitrary p wrapped around (p 1)dimensional cycles to
be Dstrings from the 4dimensional point of view, allowing a variety of values for the tension
of the effective Dstring. For instance, the tension of a Dpbrane is p (p+1)/2 gs1 (being
ls2 the square of the string fundamental length) and assuming that a D(2 + n)brane wraps
around a ndimensional flat torus with radii R1 , . . . , Rn the effective tension of the resulting
4dimensional Dstring is
=
n
1
Ri
.
ls
2ls 2 gs
i=1
Moreover the internal dimensions do not need to be flat, but can be warped ( la Randall
Sundrum [8], for instance), so that their tension can be redshifted (or blueshifted) by huge factors
and in principle it can take any value.
One important and not completely settled issue is the stability one. The (effective) Dstrings
are generically created but for them to play a role in cosmology they must be stable, or at least
metastable over cosmological time scale.
An instability already pointed out in [9] is the following. Dstrings couple electrically to a
2form, whose dual in 4 dimensions is a scalar, axionlike field. Once supersymmetry is broken
18
the axion acquires a periodic potential with several degenerate minima, so the axion is expected
to condense to different minima in different regions of space, separated by domain walls. Finite
domain walls are bound by Dstrings, and the domain wall tension, unless it is exceedingly small,
will make Dstrings shrink and collapse.
Another source of instability is breakage, which is present whenever systems of Dp and
Dqbranes are present at the same time. Let us define as the number of dimensions in which
the Dpbranes are extended and the Dq are not or viceversa, i.e. the total number of Neumann
Dirichlet (ND) plus DN dimensions. If = 2 the axion symmetry is Higgsed, there are no axion
domain walls, so the branes can break. Strings can in general break or annihilate if they meet
a brane with = 2, if = 4 or = 8 the system is stable and supersymmetric, but we fall
back into the domain wall problem when supersymmetry is broken (if = 6 branes repel each
others).
The branes come with orientation, so meeting a brane with opposite orientation they can also
annihilate, whereas if they have exactly the same orientation and = 0, 4, 8 the full system is
still BPS. Of course intermediate situation are possible, when branes meet at angle. A description
of the microscopic mechanism leading to recombination is given in [10,12].
The stability issue is thus far from settled but it can be solved in specific models. Here we will
focus on the effects triggered by two of the aforementioned sources of instability, the meeting of
a Dstring at angles and the annihilation of Dstrings. This kind of processes can be addressed
in the context of string theory and the dynamic of the instability can be studied through the time
evolution of a specific tachyonic mode of the system [13].
Since everything is coupled to gravity such instabilities can be expected to give rise to some
gravitational radiation, which we study quantitatively in the present work. Modeling the string
sources with an effective energy momentum tensor the amount of emitted gravitational radiation
can be computed, but a negligible result is found.
Actually, rather than the emission by a time dependent dynamics we can consider the closed
string remnant left by a braneantibrane annihilation [14], whose decay can be an interesting
source of gravitational radiation, as we will show.
It is also worth noticing that no cosmic string has ever been observed directly and that the
cosmic microwave background constrains cosmic string tension to be GN < 107 , where
GN is the Newton gravitational constant.
The outline of the paper is the following. In Section 2 the salient features of the analysis of [10]
of the D string reconnection are reproposed and the GW emission in such a process is computed,
then generalised to a cosmological setup like in [15]. In Section 3 the annihilation process be string is analysed, finding again a negligible gravitational radiation
tween a D string and a D
emission. In Section 3.2 we study the gravitational radiation emitted by the decay of the massive
19
Dstring Poincar group, if polarised along (orthogonally to) the brane. The system is described
by a (1 + 1)dimensional YangMills theory coupled to as many scalars as transverse direction.
The scalar fields parametrise the displacements of the brane in the orthogonal directions.
2.1. Almost parallel Dstrings
2.1.1. Reconnection dynamics
Following the analysis of [10], let us consider a system made of 2 almost parallel Dstrings at
an angle 2 , see Fig. 1. The low energy dynamics of the system is described by the action
1
1
F F + D i D i ,
S = (2 )2 Tr dx dt
(1)
2
4
2
where = 0, 1 and i {2 . . . d} runs over the dimension orthogonal to t and x.
In order to describe two Dstrings at an angle the previous action (1) has to be expanded
(b)
(b)
around the background i , A
(b)
= i,2 qx 3 ,
A(b)
= 0,
(2)
with q tan /( ) and i the Pauli matrices which form the algebra of SU(2), plus the fluc(f )
tuations (f ) , A
(f )
)
2
A(f
i = i,2 (t, x) 3 + (t, x) 1 ,
(3)
= ,1 a(t, x) .
To lowest nontrivial order in interactions, the resulting Lagrangian density is,1
1
q2
1
1
L a 2 + 2 + 2 2 + 2
q a (qx + )
2
2
2
2
1
+ a(q + ) a 2 q 2 x 2 + 2qx ,
2
and it takes to the equations of motion
(4)
a + qx q + aq 2 x 2 = 0,
1 The set of fluctuations 2 , A 1 decouple from 1 , A 2 and its Lagrangian is another copy
2
1
2
2
of (4).
20
a qx 2aq = 0,
+ a + 2a + a 2 qx = 0,
(5)
where it is understood that a dot stands for a time derivative and a prime for a derivative with
respect to x.
We note that consistency of the equations requires that a, are of the same order, while
is O(a 2 ), so that the first two equations of (5) can be solved for a and and then the result
plugged into the equation for . This justifies a posteriori the choice made in (4) to drop O(a 3 )
terms as well as O(a 2 ).
For the coupled a, system it is convenient to take the anstze
a(t, x) =
An eimn t an (x),
n>0
(t, x) =
Fn eimn t an (x),
(6)
n>0
where the equation of motions force the condition m2n = (2n 1)q.
Both a and have an unstable lowest lying mode, with a Gaussian profile, so that neglecting
the positive mass squared mode the solution can be written as [10]
a(t, x) = A0 e
qt qx 2 /2
= (t, x).
(7)
The expression for is slightly more complicated, but again its nontrivial part is concentrated
in an area of size q 1/2 around x = 0
e
(9)
,
2
0
q 2 x 2 + 2qx + 02
where in the previous (9) a gauge transformation has been performed to diagonalise the field
and only the first order in has been kept. The dynamics of the YangMills system tells the
Dstrings to recombine smoothly around the intersection point. This analysis neglects in the
Lagrangian terms of the order a 4 and higher, thus it is valid for
A0 e qt eqx /2 < q.
(10)
By a shift in time we can set A0 = q (as we will do in the rest of the paper), so the solution (7)
can be expected not to be valid for t > 0.
21
Here we neglected the effect of a relative velocity v between the Dstrings, which affects the
probability of the reconnection taking place. It has been studied in [12], with the result that the
probability P 1 if v 1, which takes to the condition
qt
(11)
qe
q.
Again this restriction suggests that this picture is a good description only for negative times.
2.1.2. Energymomentum tensor
Having understood the dynamics of the recombination via the effective action of the lightest string modes, it is then possible to compute the energy momentum tensor of the system to
estimate the amount of gravitational radiation emitted during the process. We take the definition
2 S
.
T =
g g
(12)
Considering the system described by Eq. (1), localized on the Dstring world volume, the t, x
components of the energy momentum tensor can be explicitly written on the solutions (7) as
2
q /2 q
q
T = 2 2 2
(13)
.
q
a 2 q q q 2 /2
In Eq. (13) we recognise the constant contribution from the background configuration of the
tilted Dstrings, which is Tdual to a constant magnetic field on a D2brane. The time dependent
part is the source of the gravitational radiation. This energy momentum tensor is conserved, but
it would not have been conserved without the inclusion of in the anstze (3), like in [10], where
the solution considered have = 0. This should not come as a surprise as forcing the solution (2)
with = 0 would imply an explicit breaking of Lorentz invariance, which would necessarily let
no right to expect a conserved energymomentum tensor. The inclusion of turns the Lorentz
symmetry breaking of the background into a spontaneous one, thus allowing the conservation of
the energy momentum tensor.2
By linearising general relativity over a flat background
g = + h ,
(14)
we have
2 h = 16GN T ,
h h 1/2 h
(15)
T (, k) = T (t, x )ei(tkx ) dt d 3 x,
(16)
2 Note that in [11] the energy momentum tensor of the recombining Dstrings is derived with a different method and
additional terms to (13) are found. Such terms are not localized on the Dstrings world volume but in the region between
the Dstrings. Here we do not take them into consideration, limiting our consideration to the physical system defined
by (1) consisting of (1 + 1)D YangMills with a scalar field. As we will show the production of gravitational wave is
phenomenologically negligible, and no important enhancement is due to the insertion of the new terms obtained in [11].
22
(17)
where n is the unit vector pointing into the direction of x and the local wave zone approximation
x .

x x 
x x n has been used, with x 
Given (13) and the explicit solution (7)
a(t, x) =
qe
qt qx 2 /2
(18)
for t < 0
2
h11 (t, x ) = 4 GN q 3/2 2 3/2 d 1
ei(tx) e 4q + + e 4q
x
2 q + i
4
= GN q 3/2 2 5/2 e2 q(tx) f (t x, , ),
x
(19)
valid for t < x, where f (t x, , ) takes account of the angular dependence of the radiation
and its explicit form is
q
q
2
2
f (u, , ) e+ 1 + erf
(20)
u + + + e 1 + erf
u + ,
+
where =  sin (cos q sin ) and , are the usual polar angles. Actually the details
of the angular distribution are of little importance in view of the intensity of the radiation, as we
will now show.
To compare with experimental sensitivity it is convenient to express the previous result in
terms of the (double sided) spectral density Sh (f ) which for a short burst of (Fourier transformed)
) and duration t reduces to
amplitude h(f
Sh (f )
h 2 (f )/t.
(21)
From the computation (19) we can estimate a spectral density function, neglecting the angular
dependence in the exponential,
(GN )2 q 5/2 4
Sh (f )
103
x2
e 4q
1 + 2 /(4q)
2
2
3/2
4q
GN 2
x
120
1
5/2 e
10
Hz (tan )
,
2
7
1 Mpc
TeV2
1 + 10
4q
(22)
where redshift factors has been neglected. The typical signal frequency is
q/((1 + z)). For
presently working experiments Sh 1042 Hz1 [18], so the result (22) is ridiculously small.
This is due to the inherently microscopic origin of the dynamics, whereas to generate a consistent
amount of radiation a coherent motion of a macroscopic mass is needed. Let us now investigate
if a cosmological population of such recombining Dstrings may help in enhancing the signal.
23
L=
t,
2(1 )
(24)
allowing the string network to track the dominant energy source of the Universe and to make up
a constant fraction of it.
String loops have typical size lcl and density ncl roughly given by
lcl
t,
ncl 1 t 3 .
(25)
The value of is not known, as an indication we can take the value given in [20]
GN
where
50, so that it can well be that 1 (we remind that smoothness of cosmic microwave
background requires GN < 107 ). A relation similar to (25) between lop , nop and t can be
expected to hold for long, open strings with
1. Given (25) the number of reconnection events
between open strings per unit of spacetime volume can be estimated to be
opop (t) = n2op vopop
P t 4 ,
(26)
(27)
which is slightly higher than (26), where it has been used that the average distance traveled by a
1/3
string before meeting another one is ncl .
We want now to relate the time variable t to the redshift factor (1 + z) as in [15]. Let us
consider the usual FRW metric
ds 2 = dt 2 + a 2 (t) dr 2 + r 2 d 2 ,
(28)
24
the comoving distance r between an object emitting light at a time t (z), where (1 + z) denotes
the redshift factor, and the observer at present time t0 is given by
t0
r(z) =
dt
.
a(t )
(29)
t (z)
(30)
where ti mark the epoch of the beginning of radiation domination and teq denotes the time of
matterradiation equality, the redshiftcomoving distance relation is obtained
1/2 (1 + z )1/2 1 + 2 zzeq , z > z > z ,
eq
i
eq
3t0 (1 + zeq )
3 1+z
r(z) =
(31)
(1+z)1/2 1
a0
,
zeq > z > 0,
(1+z)1/2
which, neglecting numerical factors, can be well interpolated by
t0 z
r(z)
.
a0 1 + z
The previous formulae allow us to write the volume element as
dV (z) = 4a 3 (t0 )r 2 dr
102 t03
z2
1
dz.
13/2
(1 + z)
(1 + z/zeq )1/2
(32)
(33)
Let us introduce N(z), the reconnection rate for events occurring at a given redshift z. Its differential is given by
dV
102 3 (1 + z/zeq )3/2
z
d ln z,
d ln z
dz
t0
(1 + z)3/2
where the timeredshift relation
dN = (1 + z)1 (z)z
(34)
(35)
has been used. The observer at r = 0 sees a burst with duration t0 = t (1 + z), being t the
proper time burst duration. By comparing (t0 )1 with dN/d ln z we can verify if the signals
actually overlap to make a continuous background, or if they are resolved in time. Since N (z)
is a growing function of z, it turns out that for z > z the signal is made by a collection of
bursts whereas signals originated at higher redshift than z combine into a stochastic background.
Comparing (34) with t0 one obtains
1/2 1/4
.
z
1010 (tan )1/8 1/4
(36)
TeV1
The average over introduces a numerical factor
0.9. In the case of continuous background,
i.e. for events generated at redshift z > z , taking into account the cosmological redshift, the
spectral density function can be estimated as
zin
2 dN (z )
Sh (f ) = h f (1 + z )
d ln z ,
d ln z
z
(37)
25
(1 + z)) stands
where zin stands for the redshift of the set in of the scaling solution and h(f
for a sum over metric polarisation and average over directions n of h (f (1 + z), n).
Since
dN (z)/d ln z z3 at large z, the integral in Eq. (37) can be approximated by the value of the
integrand at the upper limit of integration region
3
2 102 zin
,
Sh (f ) = h f (1 + zin )
3/2
t0 zeq
(38)
) can be read from (22) with the aid of (21) after averaging over orientations. With
where h(f
3 /(t z3/2 )
respect to the signal from an astrophysical source given by (22) there is a factor 102 zin
0 eq
1/2
(1 + z), leading to an enhancement for high enough redshift, but again the
instead of t0
q
final result is negligible
2
GN 2
zin 3
2 2 2
Sh (f )
10144 Hz1 e f zin
(39)
,
107
1010
TeV2
where an average over Dstring meeting at angle 2 has been performed and the distance x from
the source has been expressed as a function of the redshift according to (32). Moving to higher
redshift brings competing effects: a higher distance from the source, see Eq. (32), and a decreased
integration volume, see Eq. (33), both of which tend to suppress the signal, and increased event
rate, see Eq. (26), which tends to increase the signal and it is actually the dominant effect. We
remind the relation between redshift z and temperature T during standard cosmological evolution
T
9
z
10
(40)
.
MeV
Actually before the scaling solution sets in, the density can be higher than (25) and consequently the rate of string encounter higher than (26). One could consider the very extreme case
of an initial density of cosmic Dstrings
nex 3/2
(41)
at the epoch of cosmic string creation (inflation ending), characterised by a redshift zex . Such a
density would correspond to a Hagedorn energy density h 2 , which is generally a limiting energy density for open string systems. The corresponding rate of reconnection events per
volume is
(ex)
2 4
op
,
op
(42)
while the standard scaling solution is recovered in a few Hubble times. The highly nonstandard
form of the Dbrane tachyon action, which suppresses gradients of the field, allows us to assume
a far higher density of defects than the standard Kibble argument would allow as we will see in
the next section, see [19]. With this assumption we can reconsider the integral (37), noting that
it will be dominated by the early epoch z zex thus giving, analogously to (39),
2 3 5
(GN )2 2 2 f 2 zex
2 10 t0 zex zeq
e
x 2 (zex )
2 4
2
zex 5
2 2 2
1091 Hz1 e f zex
,
1010
TeV2
1/2
Sh (f ) 103
(43)
26
which is still way too small for detection. It has also to be noted that the Kibble density is smaller
than the extremal density nex given in (41) only if
1/3 4/3
ls
.
(44)
< GN
With the process considered so far one cannot obtain a large enough result, as an inherently
microscopic mechanism is involved, which affects a portion of the string of size the order of the
string length in a huge Universe. For the gravitational emission to give a sizeable effect we need
it to be produced in a coherent motion of all its world volume. In Section 3 we move to the study
D Dstring
case.
2.2. Almost antiparallel case
2.2.1. System dynamics
If the system of two Dstrings is close to a braneantibrane pair it can still be analysed
through a YangMills low energy effective action, but a scalar tachyon has to be added to the
spectrum, as it comes from the lowest lying excitations of the string stretched between the brane
and the antibrane [10]. The action is then
1
1
1
m2 2
2
i i
S = (2 )
dt dx Tr F F + D D + D T D T
T
,
4
2
2
2
(45)
with m2 = 1/(2 ). Now the expansion has to be taken around the background (again q
tan /( ))
(b)
= i,2 qx 3 ,
A(b)
= 0,
T (b) = 0,
(46)
= i,2 (t, x) 3 ,
)
A(f
= 0,
T (f ) = (t, x) 2 .
(47)
In terms of the anstze (46) and (47) the Lagrangian, up to lowest order interaction, can be written
as
1
1 2 1
1
1
1
1
L 2 q 2 q 2 x 2 + 2qx 2 + 2 + 2 2 .
(48)
2
2
2
2
4
2
2
The analysis follows straightforwardly as in the previous case to obtain the equations of motion
+ q 2 x 2 /(2 ) = 0,
+ 2 qx = 0.
(49)
27
which will be discussed in Section 3.2, whereas a kink profile will let a Dp Dpbrane
system
decay into a D(p 1)brane, a vortex configuration into a D(p 2)brane and so on. We show
next that in the case of a trivial spatial configuration, important conclusions can be drawn for the
production of gravitational radiation.
3.1. Recombination and annihilation
For ordinary phase transitions the presence of defects is related to the global configuration
of some field triggering the transition. In an expanding Universe with Hubble rate H , the defect
separation d must be smaller than the Hubble length H 1 , as any superhorizon correlation
would violate causality, thus leading to the lower bound of one defect per Hubble volume (Hubble
density), and the actual number of defects per Hubble volume is as large as a few.
However the density of defects can be much larger than the Hubble one as this picture does
not apply to the tachyon condensation, whose dynamics is described by the Lagrangian
S = d p+1 x V (T ) 1 + T T g ,
(53)
where the V (T ) is a symmetric, positive potential, which vanishes asymptotically for T
as eT [23] (it is not very important here the exact form of such potential).
As shown in [19], the nonstandard form of the kinetic terms implies that once the tachyon
has rolled to large values, gradients are exponentially suppressed, essentially eliminating the
restoring force which would tend to eliminate gradients within a casual volume, thus allowing a
higher defect density than the Kibble value.
28
We assume that the density nex as in (41) can be achieved, corresponding to an energy density
of the order of the Hagedorn scale, which is a limiting energy density for open strings [26].
Actually in the evolution of an ordinary string network, the initial conditions are not crucial
as the network reaches the scaling solution in a few Hubble time, long strings loose energy by
chopping off small loops, which in standard simulations then looses energy by emitting gravitational wave until disappearance. If we take the extremal case of Hagedorn energy density like
in the end of Section 2.1.3 the long string interseparation is dex 1/2 rather than the Kibble
1/2
value dK H 1
GN 1 , which is eventually reached after few Hubble times.
The high starting density of strings is anyway not enough in itself to produce a large amount
of gravitational wave if the usual picture is assumed, i.e. that small loop radiates as in [15] via
the formation of cusps and kinks. Again this is due to the smallness of the physical size of the
radiating objects, in this case the small loops, whose typical
size is some orders of magnitude
smaller than the Hubble scale, see Eq. (25), where H GN
lPl / ls2 , taking for the extremal Hagedorn density.3 In this picture the emission is originated in the early Universe, when
the size of cosmic string is small as they had no time to stretch with cosmological expansion, so a
small amount of radiation is produced. In [15] for instance, a typical signal close to experimental
sensitivity at a frequency f , with rate N , originated at redshift z is obtained from a kink in a
cosmic loop of size lcl roughly given by
8/3
z3 (1 + z/zeq )11/6
f 2/3
.
kHz
(1 + z)7/6
106
(55)
So far we have dealt with an effective field theory description of Dstring Dstring
annihilation, the excitations of the branes have been described as classical fields subject to ordinary
second order differential equations (for a quantum description of creation of open and closed
strings by Dbrane recombination see [21]).
On the other hand the annihilation process admits an exact description in terms of the fundamental string theory, i.e. in terms of the twodimensional worldsheet theory, by adding to the
standard world sheet action
1
1
a
Sws =
(56)
d d
a X X g + g ,
4
a boundary term
0
Sb d 0 eX /(2ls ) 1 .
(57)
3 Here we have a limiting density different from [19], where the energy scale of the string tension and the inverse
distance between them are identified.
29
The worldsheet (disk) partition function Zws , with the world sheet couplings interpreted as
spacetime fields, corresponds to the space time action S [25], given by
S S
S = Zws d p x g
(58)
dX d e ws b .
The boundary term describes the condensation of the tachyon to a profile of the type
2ls )
T = et/(
1 .
(59)
Giving an expectation value to the tachyon shifts the mass squared of the open string modes by
an amount m2 T 2 , eventually dropping them out of the physical spectrum as the tachyon
increasing expectation value makes them more and more massive.
A Dbrane in the free theory can be described by a specific boundary state B, the B B
vacuum amplitude is written as the so called cylinder amplitude
ABB = BP B,
(60)
where P is the closed string propagator. When the internal particle running in the propagator
goes on shell the amplitude ABB picks an imaginary part. Once written the expansion of B into
a set of complete states as
B =
(61)
U (f )f ,
f
1
U (f )2 ,
2f
(62)
where f is the energy of the state f . In the unperturbed theory, where the tachyon has vanishing
expectation value, the brane boundary state can be written as
B = O, k,
(63)
where O is an operator build out of fermionic and bosonic matter oscillators and ghost oscillators
and , k is the Fock vacuum with momentum k. The specific form of the vacuum will not be
needed here, as we shall simply quote known results, see [22,23]. Taking the initial condition to
be that of Dstrings, the addition of the boundary term (57) turns B into BT , whose energy
momentum tensor is obtained from (58) via (12)
T = 2 diag 1, g(t), 0, . . . , 0 ,
(64)
with
g(t) =
1 + 2 2 2 e 2t/ ls
(65)
The boundary state BT at t corresponds to B, and at t tends to a state without
the original Dstrings but endowed with the same amount of energy and vanishing pressure.
30
Applying the optical theorem as in (62) one can find, see [14,27], that the expansion coefficient UT (f ) relative to BT turns out to be just the Fourier transform of g(t), which is4
/ 2
UT (f ) =
(66)
.
sinh(ls f / 2)
The imaginary part of T BBT is related to the average numerical density n and energy density
of particles emitted by
n =
1
U (f )2 ,
2f
(67)
1
U (f )2 ,
2
(68)
where the sums run over a basis of closed string states, implicitly including both the sum over
excitation level N of the closed string and the momenta of the spatial directions transverse to the
brane. For large N the degeneracy of states G(N ) of a closed string in D dimensions is
G(N ) = N (D+1)/2 e
D2
6
(69)
and here we assume D = 10 from superstring theory. Given the degeneracy of the states and the
mass formula for closed strings
1
m2 = 4 N
(70)
,
2
the exponential in Eqs. (66) and (69) combine to leave a power law spectrum for the radiated
energy according to
dE E p/2 ,
(71)
for Dpbranes, with a total emitted energy which is ultraviolet divergent for p 2. Of course the
emitted energy cannot be larger then the mass of the initial brane, so backreaction is expected
to set in and cut off the emission, with the result that the total radiated energy is finite and of
the order of the initial mass of the brane. In particular cutting off the integral in Eq. (71) at
Ecut lop would give all the energy into closed string modes, emitting most of their energy
into very massive strings.
Actually Eq. (71) seems to suggest that for p > 2 the amount of energy of the initial Dpbrane
going into closed string modes is rather small. However the decay described by the boundary
term (57) is spatially homogeneous and it is not expected to realistically describe the decay
process. Perturbations with a nontrivial spatial profile can still be tachyonic and thus contribute
the decay. We should expect the decay to take place rather inhomogeneously and the original
Dpbrane can be thought of decomposing into a collection of small patches. In each of this small
patches, disconnected one from the other, the behaviour of a Dpbrane is similar to that of a
collection of D0branes [24].
4 This is actually the coefficient for the expansion over NeveuSchwarz states, which is the interesting sector for
gravitons.
31
1
A2 df1 ,f2
2MG(N )
i
gc2
2M
(N N1 N2 )2
G(N1 )G(N2 )
df1 ,f2 .
G(N )
(72)
k
d,
16 2 M
(73)
where k is the modulus of the momentum of the decay states and the solid angle. Considering
for simplicity flat internal compact dimensions, the inclusion of momentum and winding modes
in the mass formula (70) adds a sum over momenta and windings in the compact directions.
The closed string mass formula which generalises (70) in the presence of discrete momenta and
windings is
2
2
1
ni
2 Ri
+
2 + wi ,
m =4 N
2
Ri
(74)
(75)
This allows to approximate the summations over ni and wi as Gaussian integrals in the case
respectively m ls Ri2 and m Ri2 / ls3 . Using the asymptotic formula for the density of
32
where a = 4/ 3, dc the number of compact dimensions along which the closed strings have
momenta and windings, the kinetic energy of the decay objects
M
k2
2m1 m2
has been introduced and the approximate relation
N N1 N2
M k 2 + m21 m21
m1 m2
E M m1 m2
(77)
(78)
(81)
clrad gc2 M.
(82)
Thus in a time much shorter than the string scale a Dstring and a Dstring
decay efficiently into
radiation.
If it is so, one of the two decay objects can be massless from the 10dimensional point of view
but it will be massive from the effective 4dimensional theory. In this case the decay width (81)
5 We assume that the closed strings have enough energy to possess momenta and windings along the compact dimen
R
sions. If this were not the case the righthand side of (76) should be divided by 31/4 ls m 2li for each direction i along
s
ls
which the momentum modes are not excited and 31/4 ls m 2R
for each direction with unexcited winding modes.
i
6 Note that we obtain a dependence on M and m different from [29]. This happens because here we have summed over
KaluzaKlein momenta and winding modes of the decaying object, thus obtaining the expected dependence on m of (80)
instead of the weird m(D1+dc )/2 factor of [29].
33
Fig. 3. Recombination of two almost antiparallel Dstrings. It is highlighted the region of space where tachyon condensation occurs.
dc
als k 2 +m2
2 e
ls M(ls k)d1 dk
Ri dki
dclKK = gc
2
2
1 eals k +m
i
2
2
gc2 eals k +m mdc 1 dm.
(83)
Then integration over the momenta inthe compact dimensions have the effect of multiplying
the differential decay width by a factor Ri / ls , for Ri ls , because of the opening up of new
decay channels, and it does not modify the result (81) for Ri ls .
If two Dstrings are not exactly antiparallel but meet at an angle 2 , as in Fig. 2, the tachyon
condensation makes the Dstring disappear not over the entire worldvolume of the Dstrings,
but only as long as the Dstrings are separated by a distance shorter than 2ls , which for strings
meeting at angle 2 happens for a region of size Lx = 2ls / tan , see Fig. 3.
The energy density in gravitational waves gw is related to the spectral density function by [31]
GN gw =
2
df f 2 Sh (f ).
(84)
The energy a single annihilation of Dbranes make available for radiation can be estimated
as Lx which eventually expands spherically. From (81) and (84) one can then compute the
strain due to a gravitational wave emitted at a distance x from us (here we neglect redshift for
34
1 eals f
x2
als f G 3/2 x 2
N
108
1 e
10
Hz
.
tan 107
Mpc
TeV2
Sh (f )
(85)
Of course Sh (f ) cannot diverge for 0, as the naive estimation for Lx does, but for small
one has to substitute 2ls / tan with lop , the (open) cosmic string length. In this optimal case of
perfect antialignment one has
(opt)
Sh
x2
1 eals f (1+z)
(1 + z)1/2
GN
1070 Hz1 eals f (1+z) 2
,
z (1 + z/zeq )1/2 107
TeV2
(f )
(86)
where it has been used that ls f (1 + z) 1, as a frequency ls1 at an epoch characterised by a temperature T = ls1 , once redshifted corresponds to 100 GHz today, assuming standard adiabatic
evolution in between.
Assuming that the radiation is produced by an annihilation event during cosmological evolution when the scaling solution has set in, the expected rate of such an event is
N=
P
P
(1 + z)3/2 (1 + z/zeq )1/2 .
t
t0
(87)
It is clear that here again we have to assume an extremal initial density of strings well above
the Kibble density to hope to have a detectable signal. Let us then assume an initial density of
annihilation (42) at an epoch characterised by a redshift zex . Annihilation will thus take place in
a region of volume V (zex ), see Eq. (33), and for a time duration
dt
dt
z
zex
t (zex ) =
dz z=zex
dz z=zex
to give
(ex)
Sh (f ) =
dgw
2
GN
ex V t
2
df
f
t04 z4
2 GN lop ls f (1 + z)eals f (1+z) 1
t02 z2 /(1 + z)2
2 4 (1 + z)9 (1 + zzeq )
1 eals f (1+z)
1
3
z2 eals f (1+z)
50
1
Hz
,
10
(1 + z)17/2 (1 + zzeq )3/2
TeV2
107 G1
N
(88)
where we have considered the optimal case Lx = lop . Since the extremal case of a Hagedorn
energy density has been assumed, zex has to be high enough to accommodate such a high density phase. Assuming standard adiabatic expansion after this extremal density phase the relation
between the redshift factor z and the string scale is Eq. (40)
1/2
z = 1015
(89)
,
TeV2
35
which takes the estimate (88) very far from both experimental sensitivity and phenomenological
bound. In particular the nucleosynthesis bound requires that not too much radiation is present at
the time of the synthesis of the primordial elements, in order not to spoil the beautiful agreement
between theory and observation of primordial element abundances. Such bound constraints
3
51
1 kHz
Hz
.
Sh (f ) < 10
(90)
f
Substituting (89) into (88) we can obtain
eals f (1+z)
Sh (f ) = 1072 Hz1
1 eals f (1+z)
1/2
5/2
f
.
1011 Hz
TeV2
107 G1
N
(91)
In our toy model the spectral strength can saturate the nucleosynthesis bound only around
10100 GHz, i.e. at the end of the spectrum. This is hardly surprising as most of the energy
of the Fstring decay is concentrated for frequencies f 1 ls (1 + zex ).
On the other hand if KaluzaKlein modes are created and if they are absolutely stable, they
can rapidly overclose the Universe. In this case of extremal production their initial energy density
is a nonnegligible fraction of the total energy density, so their decay is necessary otherwise they
would start dominating as soon as they become nonrelativistic. Their final abundance is then set
by the annihilation cross section [32], which may not be Planck scale, as gravity can be strong
much below the Planck scale in large extra dimension scenarios.
If instead we consider the standard scenario with the spacetime rate of open stringopen
string encounters given by (26), we can estimate the fraction of the critical energy density going
into the production of massive gravity modes, which may eventually play the role of dark matter
(see [33] for an accurate computation of the decay amplitude of closed strings into KaluzaKlein
modes).7 Let us suppose that each encounter give rise to the production of massive modes for a
total energy EKK = Lx , thus giving an energy density per Hubble volume KK
Lx /t 3 . By
normalizing it by the critical energy density c = 3H 2 /(8GN )
(8GN t 2 )1 and integrating
over the history of the Universe, using Eq. (35), we have
KK (znr )
ls
= 32 2 GN
KK (znr )
c (znr )
t0
10
2
50 zin
zeq
GN
107
zin
(1 + z )1/2 (1 + z /zeq )1/2 dz
znr
TeV2
1/2
,
(92)
where zin denotes as in Eq. (37) the time of the onset of the scaling solution. Note that the
integral in (92) has been halted at znr , the redshift at which the KaluzaKlein modes become
non relativistic. For znr > z > zeq , KK keeps earning a factor of z over c thus resulting into a
present fractional energy density KK KK (z = 0) given by
1
znr
GN
1011
,
KK = KK (znr )
(93)
zeq
107
TeV2
7 See also [34] and [35] for different methods of creating dark matter candidates from cosmic strings.
36
where for simplicity we have assumed the mass of the KaluzaKlein modes to be 1/2 , see
Eq. (83), and substituted the relation (89) for z = zin = znr .
This result can be phenomenologically relevant still for a moderate value of the string scale.
4. Conclusions
Motivated by the recent revival in cosmic strings we have studied cosmic strings by modeling
them as Dbranes, the solitonic objects of string theory and computed the amount of gravitational
waves emitted in recombination processes which can take place in the moderately early Universe.
We have found that in general such processes are well below present sensitivities. Still the process
leading to string annihilation can result into a production of gravity modes of fundamental strings
which can be massive at an effective 4D level and can thus have a cosmological effect. Such
effect depends ultimately, in our simple model, on the string scale value, and it can be interesting
even if the string scale is well below the Planck scale. From a more theoretical point of view, this
mechanism represents a way for producing out of equilibrium, very massive particles.
Acknowledgements
It is a pleasure to thank F. Dubath, R. Durrer, M. Kunz and M. Sakellariadou for very useful
discussions. J.L.C. and E.P. thank the theoretical physics department of the University of Geneva
for kind hospitality during the completion of part of this work.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
S. Kachru, R. Kallosh, A. Linde, J. Maldacena, L. McAllister, S.P. Trivedi, JCAP 0310 (2003) 013, hepth/0308055.
N. Jones, H. Stoica, S.H.H. Tye, JHEP 0207 (2002) 051, hepth/0203163.
S. Sarangi, S.H.H. Tye, Phys. Lett. B 536 (2002) 185, hepth/0204074.
E.J. Copeland, R.C. Myers, J. Polchinski, JHEP 0406 (2004) 013, hepth/0312067.
J. Preskill, Phys. Rev. Lett. 43 (1979) 1365.
A. Vilenkin, E. Shellard, Cosmic Strings and Other Topological Defects, Cambridge Univ. Press, Cambridge, 1994.
P.P. Avelino, J.C.R. Oliveira, C.J.A. Martins, Phys. Lett. B 610 (2005) 1, hepth/0503226;
P.P. Avelino, C.J.A. Martins, J. Menezes, R. Menezes, J.C.R. Oliveira, Phys. Rev. D 73 (2006) 123519, astroph/
0602540.
M. Gogberashvili, Int. J. Mod. Phys. D 11 (2002) 1635, hepph/9812296;
L. Randall, R. Sundrum, Phys. Rev. Lett. 83 (1999) 4690, hepth/9906064.
E. Witten, Phys. Lett. B 153 (1985) 243.
K. Hashimoto, S. Nagaoka, JHEP 0306 (2003) 034, hepth/0303204.
K. Hashimoto, W. Taylor, JHEP 0310 (2003) 040, hepth/0307297;
K. Hashimoto, JHEP 0404 (2004) 004, hepth/0401043.
A. Hanany, K. Hashimoto, JHEP 0506 (2005) 021, hepth/0501031.
A. Sen, Int. J. Mod. Phys. A 20 (2005) 5513, hepth/0410103.
J.L. Karczmarek, H. Liu, J. Maldacena, A. Strominger, JHEP 0311 (2003) 042, hepth/0306132.
T. Damour, A. Vilenkin, Phys. Rev. D 64 (2001) 064008, grqc/0104026;
T. Damour, A. Vilenkin, Phys. Rev. D 71 (2005) 063510, hepth/0410222.
M.G. Jackson, N.T. Jones, J. Polchinski, JHEP 0510 (2005) 013, hepth/0405229.
S. Weinberg, Gravitation and Cosmology, Wiley, New York, 1972.
B. Abbott, et al., LIGO Collaboration, Phys. Rev. Lett. 95 (2005) 221101, astroph/0507254.
N. Barnaby, A. Berndsen, J.M. Cline, H. Stoica, JHEP 0506 (2005) 075, hepth/0412095.
D.P. Bennett, F.R. Bouchet, Phys. Rev. Lett. 60 (1988) 257.
T. Sato, Nucl. Phys. B 682 (2004) 117, hepth/0304237.
F. Larsen, A. Naqvi, S. Terashima, JHEP 0302 (2003) 039, hepth/0212248.
37
Abstract
Nontopological solitons, Qballs can arise in many particle theories with U (1) global symmetries. As
was shown by Cohen et al. [A.G. Cohen, S.R. Coleman, H. Georgi, A. Manohar, The evaporation of Qballs,
Nucl. Phys. B 272 (1986) 301], if the corresponding scalar field couples to massless fermions, large Qballs
are unstable and evaporate, producing a fermion flux proportional to the Qballs surface. In this paper we
analyse Qball instabilities as a function of Qball size ans fermion mass. In particular, we construct an exact
quantummechanical description of the evaporating Qball. This new construction provides an alternative
method to compute Qballs evaporation rates. We shall also find the new expression for the upper bound
on evaporation as a function of the produced fermion mass and study the effects of Q balls size on particle
production.
2006 Elsevier B.V. All rights reserved.
1. Introduction
A scalar field theory with an unbroken continuous global symmetry admits a remarkable class
of solutions, nontopological solitons or Qballs. These solutions are spherically symmetric nondissipative solutions to the classical field equations [13]. In a certain way they can be viewed
as a sort of BoseEinstein condensate of classical scalars. The construction of these solutions
uses the fact that they are absolute minima of the energy for a fixed value of the conserved
U (1)charge Q. So in the sector of fixed charge the Qball solution is the ground state and all its
stability properties are due to charge conservation. An important amount of work has been done
on Qball dynamics and on their stability versus decay into scalars [1,4]. Apart some existence
39
theorems that depend on the type of symmetry and the potentials involved [3], the stability of
Qballs is due to the fact that their mass is smaller than the mass of a collection of scalars.
In realistic theories the scalar field has a coupling with fermionic fields. The addition of this
coupling modifies the criterions of Qball stability since they can now produce fermions. This
fact will have an important interest for cosmology since Qballs can play the role of dark matter
[5,6]. Particle production from the Qball will reduce its charge Q and at a certain point the
Qball will become unstable versus decay into scalars to finally disappear. This problem has
been considered in [2,3] for the production of massless fermions by a large Qball. The method
used was to construct the quantum field as a superposition, with operator valued coefficients, of
the classical solutions. In most cases we can express the general solution as a superposition of
partial waves. The next step will be to use the asymptotic behaviour of the fields, considering the
far past and the far future where the fields can be identified to free ones. In most configurations
the fields in the far past and in the far future are free fields and the relation, the Smatrix, linking
them together contains all the information needed to answer the question of particle production.
The problem we have here is that the Qball is a timedependent configuration, so we need to be
careful when we identify the asymptotic states to free (static) ones. We must make sure that the
identification is made before the interaction is turned on and after it is turned off.
To avoid the problems linked to the time dependence of the Qball we shall propose an alternative method. This uses Heisenbergs picture of quantum mechanics. We shall construct the
time independent state representing particle production, it is done by solving the condition that
no fluxes are coming from infinity (no particles are moving towards the Qball). We can then
build the Heisenberg field operator containing all the relevant information and time dependence.
Particle creation is then computed by using the number operator, but any other operator valued
quantity can be calculated. The use of this method needs no limit calculations on the Qballs
size, so it can be used to study small Qball as well. The difficulty now lies in solving the production condition. This condition can be solved considering asymptotics of the fields far away
from the Qball.
The major difference between the standard construction and this one lies in the kinematical
conditions used. The standard Smatrix based method will solve matching conditions to compute
all the reflection and transmission amplitudes of an incident wave. The approach used here is
different in the sense that we construct a state having no incoming wave, making the computation
of scattered amplitudes useless. Using this construction implies that all the particles are created
inside the Qball, since they all move away from the Qball.
The other important fact to investigate is the production of massive fermions. Can a Qball
produce any type of fermions, and is the fermion mass relevant for the Qballs life time? To
answer this question we can use both pictures, the problem is now that solutions to massive field
equations have twice as many degrees of freedom. Even if the construction of the Heisenberg
field operator is possible and not very difficult, the resolution of the particle production condition
is a complex task to achieve, so we shall use the Smatrix picture. The partial wave expansion of
the solution can easily be obtained when working with one space dimension. In fact using this
simple picture allowed us to obtain analytical results describing a fermion field being scattered
by a Qball. Computing the evaporation rate as a function of the produced fermion mass will lead
to a new definition for the absolute upper bound of evaporation rate. An other important question
we can ask is the role of the Qballs size on the different particle creation regimes. To answer
this question we shall study both very big and small Qballs.
The paper is organised as follows. We first give a review of the simplest 3dimensional Qball
model to build up its basic properties and we then reduce it to 1 space dimension. We then con
40
sider interaction of Qballs with massless fermions and construct a solution where there are only
fermions moving away from the Qball, this construction will give us the evaporation rate of
Qballs for production of massless fermions. We then consider production of massive fermions,
using the incident wave outside the Qball. This requires the knowledge of the reflection and
transmission amplitudes on the Qballs surface. Finally we extend our results to 3 1 dimensions.
2. A simple Qball model
We review here the basic properties of a 3dimensional Qball using the simplest possible
model. As we mentioned in the introduction, the Qball is the ground state of a scalar theory
containing a global symmetry. We can now build the simplest model in 3 1 dimensions having
a Qball solution: it is a SO(2) invariant theory of two real scalar fields (in fact it is the U (1) theory of one complex scalar field) [1]. We start by writing down the Lagrangian and the equations
of motion for the scalar field, to obtain the conserved charge and current. The Lagrangian of the
scalar sector is given by:
L = U  .
(1)
The U (1) symmetry is
ei .
The conserved current is
j = i ,
(2)
(3)
We build a solution with the minimal energy: if U (0) = 0 is the absolute minimum of the potential, = 0 is the ground state and the U (1) symmetry is unbroken. It was shown in [1] that new
U
particles (Qballs) appear in the spectrum, if the potential is such that the minimum of 
2 is at
some value 0 = 0.
Min 2U/2 = 2U0 /0 2 < 2 = U (0).
(4)
The charge and energy of a given field configuration are:
1
t + c.c. d 3 x,
Q=
2i
1 2 1
2
E=
 +  + U () d 3 x.
2
2
(5)
The Qball solution is a solution with minimum energy for a fixed charge, we thus introduce the
following Lagrange multiplier
1
t + c.c. d 3 x .
= E + Q
(6)
2i
41
(7)
where (r) is a monotonically decreasing function of distance to the origin, and zero at infinity.
Inserting the Qball ansatz in the equations of motion gives in spherical coordinates
d 2
2 d
(8)
=
2 + U ().
2
r dr
dr
If we interpret as a particle position and r as time this equation is similar to a Newtonian
equation of motion for a particle of unit mass subject to viscous damping moving in the potential
1 2 2
2 U . We are searching for a solution in which the particle starts at t = 0 at some position
(0), at rest, d
dr = 0, and comes to rest at infinite time at = 0. Solving this problem is not
difficult (see [1] for details). One of the solutions can be the localised step function. Although
we can solve exactly the equation of motion, we will not do it in this work.
This construction is the Qball we where looking for, in the sense that it is the ground state
of the theory with constant charge. We used only one field to describe it but it is in fact made
of a collection of scalars. Qballs rotate with constant angular velocity in internal space and are
spherically symmetric in position space. As the charge Q goes to infinity, approaches
0 = 2U0 / 2 ,
(9)
where U0 is the value of the potential at the minimum. In this limit, resembles a smoothedout step function. The two regions (r < R and r > R) are separated by a transition zone of
thickness 1 . This leads to the consideration of two approximations the thick and thin wall
regime (see [3] and [7] for details). The radius of the Qball can easily be calculated using the
definition of charge:
4
Q = R 3 0 02 .
(10)
3
This calculation has been done (r) = 0 if r > R. All the properties of the Qball are now
known except the exact profile of the Qball field. We shall now build up the Qball solution to
our problem. The energy is given by the integral (5) and using the previous properties and taking
the limit V , where V is the volume of the Qball, the energy becomes
1
E = 2 2 V + U V .
2
The charge becomes
Q = 0 2 V .
(11)
(12)
We wish to minimise the energy with fixed charge. Using Eq. (12) to eliminate , we have in the
limit of Q ,
E=
1 Q
+ UV.
2 2 V
22 U
(13)
(14)
42
(15)
With some little modifications this construction can be adapted to any dimensions. The model
we shall use in the next section is a 1 1dimensional model. In one space dimension the Qball
profile will be the standard step function localised in a space region of size l.
3. Production of massless fermions
The solution to the problem of particle creation by a Qball can be solved using two different
pictures. The first one is based on the Smatrix formalism, using the idea that the field is free
for t . This construction is done by finding the solution to the equations of motion for
a fermion interacting with a Qball, in terms of a superposition of classical solutions. The quantisation is made by upgrading expansion coefficients to operators, this will give us the Heisenberg
field operator. The Smatrix will then be constructed by identifying the fields in the far past and
in the far future to fields having the exact positive and negative frequency behaviour. This method
was widely used to solve the problem for particle creation. This method was used to compute the
evaporation rate of Qballs [2,3] where the expansion was made using rotational eigenfunctions.
Once the total solution is known it is simple to build the transformation from the far past to the far
future. In the far past only the incoming wave will survive and in the far future only the outgoing
ones.
The construction we are going to use here is different, we shall in the first place solve the
equations of motion and obtain the Heisenberg field operator representing a fermion interacting
with a Qball. In one space dimension this solution will be expressed in the form,
+
1
(
, t, z)A(
) + Q
(
, t, z)B(
) ,
d
Q
Q =
4
where the Q
(
, t, z) are a basis of the solution to the Dirac equation for fermions interacting with a Qball of charge Q. A(
) and B(
) are operators depending on energy, their
anticommutation relations are the standard ones if the solutions satisfy proper orthogonality conditions. The next step we shall use is consider the space asymptotics of this solution. Far
away from the Qball (z = for one space dimension) the solution is the standard free field
solution. This identification will give us a relation between the solution operators A(
), B(
) and
the free asymptotic ones a(p), b(p). The only difficulty in this identification is that the quantisation of the solution was made using energy (due to the time dependence of interaction) while
the asymptotical operators depend on momentum. The next step will be to define and solve the
particle production condition, saying that no particles are moving towards the Qball. In terms of
asymptotic operators it is
aL (p)
= bL (p)
= 0 for p > 0, on the left,
aR (p)
= bR (p)
= 0
(16)
The last step of the resolution will consist in using the total Heisenberg operator , and the
particle productive state to compute the fermionic flux giving evaporation rate. The main idea
used here is to construct a solution having no incoming wave, so we do not need to compute
any reflected or transmitted coefficients. This way all the particles come from inside the Qball.
43
As mentioned before this new construction gives a good alternative to the standard scattering
kinematics. The other advantage of this picture is to allow us giving a consistent treatment of
time continuity.
In the next two subsections we shall build the solution and the relation to asymptotic operators, while in the two last subsections we shall solve the production condition and compute the
evaporation rate.
3.1. Solutions to the equations of motion
Writing down the Lagrangian of a massless fermion having a Yukawa interaction with a scalar
field gives in one spatial dimension,
+ g C + h.c. ,
Lferm. = i
(17)
where the C superscript indicates the charge conjugated fermion. The equations of motion and
their solutions are fully described in literature on the subject [13]. Instead of treating separately
the fermion and the antifermion, we shall construct the exact global solution to this problem,
this solution will be made of different parts first the solution inside the Qball (for z [l, l]).
Equations of motion for the two components of the field are:
(i0 + iz )1 g2 = 0,
(i0 iz )2 g 1 = 0,
and = 0 ei0 t in the zone from l to +l and zero everywhere else. Using the ansatz:
0
A i
t+i(k+ 20 )
1
0
ei 2 t
e
,
=
0t
i
B
2
0
e 2
(18)
(19)
M
A
= 0.
M
(k +
)
B
and k = k
, we obtain the solution inside the Qball:
k
+
1
1
ik
z
M
+
B
=
A
e
eik
z ,
Q =
(20)
k
+
2
1
M
where M = g0 , g is the coupling constant and 0 the value of the scalar field. The second part
is the solution when 0 = 0 (outside the Qball) it is,
L,R i
z
C1 e
1
i
t
=
(21)
=e
,
2
C2L,R ei
z
where superscripts L, R indicate the left and right side of the Qball. In order to solve Diracs
equation everywhere the solution needs to be continuous in space. Space continuity gives at
z = l:
C1L = Aei(k
+
)l + B
ei(k
)l ,
C2L = A
ei(k
)l + Bei(k
+
)l ,
44
and at z = +l,
C1R = Aei(k
+
)l + B
e+i(k
)l ,
C2R = A
ei(k
)l + Be+i(k
+
)l .
These matching relations are used to express the solution only using the parameters coming from
the inner part of the solution. This construction will allow us to build a state where there is no
incoming fermion, all the fermions are now produced inside the Qball. Putting together all these
parts gives the full solution continuous in space and time:
0
0
i
l
e (Aeik
l + B
eik
l )ei(
+ 2 )t ei(
+ 2 )z
z
<
l
ei
l (A eik
l + Beik
l )ei(
20 )t ei(
20 )z
0
0
ik
z + B eik
z )ei(
+ 2 )t ei 2 z
(Ae
(22)
= d
l z +l
,
i(
20 )t i 20 z
ik
z
ik
z
2
(A
e
+ Be )e
e
ei
l (Aeik
l + B
eik
l )ei(
+ 20 )t ei(
+ 20 )z
z > +l
0
0
i(
)t
i(
)z
i
l
ik
l
ik
l
2
2
e (A
e
+ Be
)e
e
where
k
+
(23)
.
M
A little work on the solution and on its orthogonality properties leads to a solution of the form:
+
0
1
Q =
(24)
(
)A(
) + Q
(
)B(
) ei 2 z (t),
d
ei
t Q
4
with
(t) =
and
ei 2
0
i 20 t
(25)
f1 (
, l)ei
z
z < l
(f2 (
, l)) ei
z
ik
z
ik
z
+
e )
(e
= N
l z +l ,
ik
z
ik
z
(
e
+ e
)
i
z
f1 (
, l)e
z
>
+l
(f2 (
, l)) ei
z
(26)
1
ei
l eik
l +
eik
l ,
4N
1
f2 (
, l) =
ei
l
eik
l + eik
l .
4N
f1 (
, l) =
N = 4 cosh Im[k
]l 1 + 
2 cos Re[k
]l Re[
] .
(27)
(28)
45
Finally the timedependent matrix was introduced for simplicity, the N are the normalisation
constants. Quantisation of solution (24) was made using equal time anticommutation relations
for . If the functions satisfy dz ( (
)) (
) = (
) we can show that,
+
+
A(
), A (
) = dz dz Q
(z,
) Q
(z ,
) Q , ( Q ) = (
).
(z z)
The Q solution we obtained has now being upgraded to a Heisenberg field operator describing
fermions interacting with a Qball.
3.2. Relation to asymptotic operators
First we conjugate the second component of the above solution, in order to compare it with the
standard free solution for a massless fermion in 1 1 dimensions (see [8] for details). We look at
the asymptotic behaviour of the Qball solution (24). On the left and righthand side of the Qball,
it has to be the standard free solution since the interaction is zero outside the Qballs volume.
After elimination of integrals and standard manipulations and variable changes, we obtain at
z :
1
(p)
(p)
a(p) +
b (p)
(p)
2 (p)
+
f1 (
, l)A(
)
=p 0
f1 (
, l)B(
)
=p 0
2
2
=
(29)
+
.
f2+ (
, l)A (
)
=p+ 0
f2 (
, l)B (
)
=p+ 0
2
(p) is the heavyside function, (p) = 0 if p is negative and (p) = 0 when p is positive.
(p)
(p)
Multiplying Eq. (29) by (p) and by (p) we obtain the two following equations:
1
aL (p) = f1+ (
, l)A(
) + f1 (
, l)B(
)
=p 0 (p)
2
2
+
+ f2 (
, l)A (
) + f2 (
, l)B (
)
=p+ 0 (p),
2
1
(p) = f1+ (
, l)A(
) + f1 (
, l)B(
)
=p 0 (p)
bL
2
2
+
+ f2 (
, l)A (
) + f2 (
, l)B (
)
=p+ 0 (p).
(30)
In these two equations the l subscript indicates we are on the lefthand side of the Qball, the
same manipulations on the righthand side lead to:
1
aR (p) = f1+ (
, l)A(
) + f1 (
, l)B(
)
=p 0 (p)
2
2
+
+ f2 (
, l)A (
) + f2 (
, l)B (
)
=p+ 0 (p),
2
1
(p) = f1+ (
, l)A(
) + f1 (
, l)B(
)
=p 0 (p)
bR
2
2
+
+ f2 (
, l)A (
) + f2 (
, l)B (
)
=p+ 0 (p).
(31)
Checking the anticommutation relations of operators aL,R and bL,R is a tedious task but using
46
equations will be the basis of the construction of particle productive state, since they give a
relation between free operators (lower case) and solution operators (upper case). These relations
will give the solution in terms of free operators, so the next task we need to achieve is to define
and construct the particle productive state.
3.3. Construction of particle productive state
As mentioned before, the construction of this quantum state will be done using the fact that
there are no particles moving towards the Qball. These are negative momentum particles on the
left and positive momentum particles on the right. In terms of aL,R , and bL,R operators:
aL (p)
= bL (p)
= 0 for p > 0, on the left,
aR (p)
= bR (p)
= 0
(32)
This construction will lead to the opposite sign of the fermionic current on the left and on the
righthand side of Qball using Eqs. (30), (31). We then obtain four equations. For positive p, we
have:
+
f1 (
, l)A(
) + f1 (
, l)B(
)
=p 0 
= 0,
2
+
f1 (
, l) A (
) + f1 (
, l) B (
)
=p 0 
= 0,
(33)
2
(34)
(35)
< 20
we have:
+
f1 (
, l) A (
) + f1 (
, l) B (
) 
= 0,
+
f2 (
, l)A (
) + f2 (
, l)B (
) 
= 0.
(36)
(37)
(f2 (
, l)) (f2 (
, l))
47
is always different from zero. The only solution for an evaporating state in this range is the trivial
solution given by:
A(
)
= B(
)
= 0 for
>
0
,
2
0
(39)
.
2
In fact these two equations are the same, because we can always use the transformation A(
) =
A (
) (
) + B (
) (
), all equations will have vacuum solutions. Here the anticommutation
relations are trivial to check because of the two different energy ranges. For the middle range
[ 20 , + 20 ] things are a little more complicated, this being the range where particle production
occurs as first shown in [2]. Taking a look at solution (26) in this range, only particles are created
and changing the sign of 0 changes the particle type. We now need to check normalisation
of these new operators describing the evaporating state and their anticommutation relations.
Defining the evaporation operators in all the energy ranges, we have
(
),
< 20 ,
ae (
) =
(40)
8(f1+ (
, l)A(
) + f1 (
, l)B(
)),
[ 20 , + 20 ],
0
A(
),
>+ 2 ,
A (
)
= B (
)
= 0 for
<
and
(
),
be (
) =
8(f1+ (
, l) A (
) f1 (
, l) B (
)),
B(
),
1
where the 8 factor is the normalisation +
< 20 ,
[ 20 , + 20 ],
f1 (
,l)2 +f1 (
,l)2
use the fact that f1 2
(41)
> + 20 ,
1
4 .
This simple relation gives the ground state for a Qball producing fermions. We could now compute lots of different properties of these fermions but we compute their number. The state defined
by this relation has no incident fermion, it is exactly the state we wanted to build. The final step
will now be to compute the fermionic flux and obtain the particle production rate.
3.4. Particle production rate
(x), which in
The particle production rate is given by the current operator j (x) = (x)
our case is
1 1 2 2 = j(x)
(43)
that we shall apply on the evaporating state defined by the vacuum for ae and be operators. First
we invert the systems (40) and (41) to obtain:
"
0
A(
) = ae (
),
< ,
(44)
2
B(
) = be (
),
48
"
0
A(
) = ae (
),
,
>
B(
) = be (
),
2
A(
) =
0
0
,
, +
B(
) =
2
2
(45)
1
(a (
) + be (
)),
82f1+ (
,l) e
1
(a (
) be (
)).
8 2f1 (
,l) e
(46)
Now we can compute the first term of the current on the lefthand side of the Qball: using anticommutation relations and the separate range of integrals and the definition of A(
) and B(
) in
terms of evaporation operators ae (
), be (
) we obtain:
0
2
01 1 0
=
2
2
d
f1+ (
, l)
0ae (
)ae (
)0
+ f1 (
, l)
0be (
)be (
)0
1
8
0
2
0
2
+
f (
, l) f1 (
, l) 2
0be (
)b (
)0
.
d
1
e
2
2
(47)
The other term of the current, proportional to 2 2 , is very similar but we need to be careful
with the fact that 2 is proportional to f2 (
, l). Applying the same method we obtain:
+
02 2 0
=
1
8
0
2
0
2
+
+
+
+
(f2 (
, l)) (f2 (
, l)) 2
0ae (
)a (
)0
d
+
e
2f1+ (
, l)
2f1 (
, l)
2
d
f2+ (
, l)
0ae (
)ae (
)0
0
2
2
+ f (
, l)
0be (
)b (
)0
.
2
(48)
jL =
0
2
0
2
+ 2
+
(f2 (
, l)) (f2 (
, l)) 2
sinh[2ik
l] 2
=
d
d
+
e2ik
l 2 e2ik
l .
2f1+ (
, l)
2f1 (
, l)
(49)
0
2
k
+
,

2 = 1
k
= i M 2
2 ,
=
M
and the current is then:
+
0
2
jL =
20
sinh2 [2 M 2
2 l]
d
2
2
2
2
e2 M
l
2 e2 M
l 2
(50)
In the limit Ml we can neglect the negative exponentials and the only factor we are left
with is the 12 coming from the hyperbolic sine. The righthand side the current is the same except
49
0 = 0.5.
Fig. 1. Particle production rate for small values of Ml and for a fixed value of 2M
for the sign, so the total current is equal to one half. Using the continuity equation we can write:
j0 j1
dQ
(51)
+
=0
= z j (z) dz = jL jR = 2jL .
t
z
dt
The current does not depend on z so the value of the current is constant on both sides, so after
integration of the current over
we obtain:
dQ
1
(52)
=
0 .
dt
4
This expression gives the particle production rate as a function of 0 when 0 is smaller than M
in the limit of big Ml. It is in fact [2] an evaporation rate since it does not depend on the Qballs
size. The other importance of this result is that it gives a absolute upper bound on evaporation
rate. The case when the imaginary part of k
is equal to zero is a bit more complicated to solve.
In this case we need to explicitly compute the current integral, so this case will be studied numerically. The value obtained for the evaporation rate in (52) is the 1 1 equivalent of the results
found in literature on the subject.
3.4.1. Production rate in function of size
We shall first consider the limit where l is small. In this case we write
2
2
2
2 2
2
2
2
sinh 2 M
l = 4 M
l = 4(Ml) 1
,
M
2 M 2
2 l
2
2 2
2
2
2
2
e
2 e2 M
l = e4 M
l + e+4 M
l 2 Re
2
2
2
M 2
=2 1 2 .
=2 1
M2
M
These two terms will simplify to give after integration over
:
0
jL = l 2 M 2 ,
8
leading to the particle production rate:
(53)
(54)
1
dQ
= l 2 M 2 0 .
(55)
dt
4
This result ensures us the fact that when Ml = 0, the Qball does not exist, the evaporation rate
is zero. This behaviour is shown on Fig. 1. The next limit we shall study is the very large Qball
50
0 = 0.5.
Fig. 2. Particle production rate in function of Ml and for a fixed value of 2M
0
dN
Fig. 3. Particle production rate 2
M dt as a function of 2M in the limit of very big Ml parameter.
limit. To do so we take a look at the production rate for large values of the size parameter, Ml and
observe that the production rate becomes constant for big values of the size parameter (see Fig. 2
and Fig. 3). These considerations also stand for all the possible values of the frequency parameter
0
the only difference is when 2M
gets bigger the stability of the evaporation rate comes for bigger
values of Ml.
3.4.2. Energy flux far away form the Qball
The next calculation we can do is the calculation of the energy flux far away from the Qball.
In the case where we consider the observer very far from the Qball the only relevant coordinate
is the distance to the Qball, we are in a one spacial dimension case. The energy flux a distant
observer can measure is given after normalisation by M,
dE
=
M dt d
+ 2M0
2M0
M
( M
) sinh[2i k( M
) l]
e
2ik(
l
M)
(2
) e
2ik(
2
2
,
l
M
)
(56)
this expression is obtained by computing the energy flux through a sphere containing the Qball.
When the real part of k
equals zero the fraction becomes equal to one. The result in this range
will be proportional to 03 [2,3] (see Fig. 4).
51
0
1
dN
Fig. 4. Normalised particle production rate 2
M dt upper bound as a function of 2M in the limit of a very big Ml parameter.
dN
dt
0
2
d
= 0 ,
20
this absolute upper bound will be used to normalise the evaporation rate.
4. Production of massive fermions
The evaporation of a Qball into massive fermions is more complicated than the previous case.
We can quite easily obtain the Heisenberg field operator but solving the evaporation condition
is a difficult task, even with only one space dimension. So the method we are going to use
is the same Smatrix based method used in [2,3]. This picture will need as starting point the
expression of the solution as a superposition of wave packets. It is done by expressing the motion
equations in matrix form and then expanding the solutions over the eigenfunctions. This gives
the separation into left and right movers. Choosing which wave is the incident one, we can write
the solution as
L = B1 ei p1 x up1 + B2 ei p2 x up2 + r1 ei p1 x up1 + r2 ei p2 x up4 ,
(57)
i p x
i p 2 x
1
R = t1 e
(58)
up1 + t2 e
up2 ,
where the us and ps
describe the solution away from the Qball and the L, R subscripts stand
for the left or righthand side of the Qball. This solution has two incident waves associated
with particles or antiparticles moving towards the Qball, giving two solutions. The reflected
and transmitted waves are associated with particles moving away from the Qball. The same
construction is done on the other side of the Qball, to give four solutions. Finally we obtain the
total solution as a superposition of these four solutions with the expansion coefficients becoming
operators. This canonical quantisation does not introduce any big problem and can be done in
a straightforward way. The next step will be to consider that in the far past only the incoming
wave survives, giving us a relation between the operators in the far past and in the far future
(where only outgoing waves survives). The last step we shall do is compute the number operator.
52
The only difficult task is the computation of reflection and transmission amplitudes appearing
in the solutions. Will shall provide two methods for calculating these amplitudes. One method
will consist in calculating all the scalar products of the motion eigenvectors, while the other one
will consist in the diagonalisation of the motion matrices. The results are fully consistent, and the
two methods serve to illustrate a variety of physical insights. The main objective we shall reach
is the computation of the new value of the upper bound on evaporation rate. This bound will now
depend on the produced fermion mass and not only on the internal Qball frequency.
The first three subsections describe the solutions in terms of eigenvectors of 4 4 matrices.
In the three next subsections we compute the transmission and reflection amplitudes for massive waves on the Qballs surface. In the last subsections we compute the evaporation rate and
simplify the problem by diagonalisation of the motion matrices.
4.1. Preliminaries
Using the same Lagrangian as for the massless case and adding a Dirac mass coupling for
massive fermions, gives the fermionic Lagrangian:
+ g C + h.c. + MD (x).
L = i
(59)
The equations of motion using two component field, in 1 1dimensions are
(it + iz )1 Mei
0
2 t
(it iz )2 + Mei
0
2 t
2 + MD 2 = 0,
1 + MD 1 = 0.
(60)
These equations can be solved if we use fields with four degrees of freedom. Solving this system
will give us the solution inside the Qball, which is the first step. To solve these equations of
motion we use the following ansatz
1 = f1 (z)ei(
2 = g1 (z)e
i(
0
2 )t
+ f2 (z)ei(
+
0
2 )t
0
2 )t
i(
+
0
2 )t
+ g2 (z)e
(61)
Due to the separation of time components this ansatz will lead to four equations. These equations
can easily be modified to reduce the numbers of parameters: we divide all equations by M = 0.
We shall now rewrite these equations, taking
f1 (z) = Aeipz ,
g1 (z) = Ceipz ,
(62)
f1 g2 + MD g1 = pf1 ,
g1 f2 MD f1 = pg1 ,
(63)
0
2 .
and
+ =
+
This arrangement has the advantage that we can now write
where
=
the field in terms of four component spinors as:
f
1
g1
=
(64)
.
f2
g2
53
The idea of using fourcomponent spinors is that now the fermion field contains both energy
components, just like the solution used in the previous chapter. The other advantage is that this
rearrangement leads to the standard four component spinor solution. The equations of motion
become in matrix form,
A
A
0
1
MD
1
0 B
B
MD
(65)
= Mp .
C
C
0
1
+ MD
1
0
MD
+
D
D
All the parameters are normalised by M. The fact that we have the M factor one the righthand
side will allows us to simplify the space components and replace l by Ml. All the parameters we
are left with now are all dimensionless, we should read the matrix elements to be all divided by
M and thus dimensionless. These satisfy
M = M T ,
(66)
with = Diagonal Matrix [1, 1, 1, 1]. This symmetry will be used to perform the normalisation of eigenvectors.
4.2. Solution inside the Qball
Using the dimensionless parameters and the eigenvectors of the motion matrix we can write
the time independent solution inside the Qball in the form:
Q =
4
#
Cj vj eipj z .
(67)
j =1
The vs are the eigenvectors of the motion matrix while the ps are its eigenvalues. We shall not
compute here the exact form of these eigenvectors, but we can prove that the eigenvalues are
either purely imaginary or purely real. For reasons that will become clear later on, the first two
terms of this solution have positive momentum while the two last have negative momentum. This
arrangement does not modify the shape or any properties of the solution, but will greatly simplify
the rest of the work. Inside the Qball the timedependent solution is:
=
4
#
up
Cj ei(
)t vpj ei pj z + Cj ei(
+)t vpdown
ei pj z ,
j
(68)
j =1
where the up superscript stands for the first two components of the eigenvectors, while the down
one indicates we take the two last components.
4.3. Solution without Qball background
Outside the Qball the solution is given by the eigenvalues and eigenvectors of the following
matrix:
A
A
0
0
MD
0
0 B
B
MD
(69)
= p .
0
0
+ MD
C
C
0
0
MD
+
D
D
54
All its parameters are normalised by M, so they are all dimensionless. The eigenvalues are given
by:
2 M 2 p ,
p 1,3 =
1
D
2 M 2 p .
p 2,3 =
+
(70)
2
D
Let us stress that all parameters are dimensionless since we have to read them as being divided
by M, the Majorana mass coupling. The exact eigenvectors can be easily calculated but we do
not need them. Here the p 1,2 momentum can be complex or real. If we want some particle to
propagate outside Qball we need both p 1,2 to be real, it gives for

 MD ,
to solve this we must identify two cases. The first case is,
0
,
0
2
0 0
>
,
M D
M D +
2
2
the last inequality is verified if 20 MD . The second case is,
0
,
2
0 0
MD
,
MD
2
2
once more the last inequality is valid when 20 MD . A similar calculation for 
+  MD gives:
0
0
0
(71)
and
MD
.
2
2
The only way to avoid the gaps and have the two waves (both p1 and p2 ) is for
to be in the
range (see Fig. 5):
0 0
,
MD .
MD
(72)
2 2
MD
This range will mix both the particles and antiparticles and thus lead to the nontrivial Bogolubov transformation. We also have:
0
(73)
MD .
2
This range is the equivalent as the range defined for the massless case. Following the same
construction as in the previous section, the static solution outside the Qball can also be written
55
in the form
0 =
4
#
Aj uj ei pj z ,
(74)
Bj uj ei pj z ,
(75)
j =1
0 =
4
#
j =1
this time the Bj coefficients are on the lefthand side of Qball while the Aj are on the righthand
side. The timedependent solution is once more given by,
4
#
up
i p j z
(Aj , Bj )ei(
)t upj ei pj z + Aj , Bj ei(
+)t udown
.
=
pj e
(76)
j =1
These considerations allow us to find the particle production energy range where particles can
propagate outside the Qball. The first part of our calculation is over, the next task is to compute
the reflection and transmission amplitudes. It is in fact solving the matching equations in matrix
form. The solution will give a relation from the far left to the far right of the Qball and the centre
part (the Qball itself) will not appear directly.
4.4. Construction of scattering matrix
We want to construct the matrix linking the solution at z = to the solution at z = +.
We are searching for the matrix:
A1
B1
B
A
2
2
(77)
= V .
B3
A3
B4
A4
As a reminder the Bs are on the left while the As are the righthand side of the Qball. This
construction will not contain directly any inside parameters so the A and B can be considered
as free parameters while the reflection and transmission amplitudes will be contained in the V
matrix. To compute the matrix elements we need to solve the matching equations.
4.5. Matching in space
We first start by matching the solutions at z = l we have:
B1 up1 ei p1 l + B2 up2 ei p2 l + B3 up3 ei p3 l + B4 up4 ei p4 l
= C1 vp1 eip1 l + C2 vp2 eip2 l + C3 vp3 eip3 l + C4 vp4 eip4 l .
(78)
Ci
C i = .
Ni
(79)
Multiplying Eq. (80) by u Ti , orthogonality property coming from the symmetry of the motion
matrices:
B i uTpi upi =
4
#
j =1
uTi vj eipj l C j .
(80)
56
Doing the same for all the Bs and writing down all the relations in matrix form we obtain:
C1
B1
C2
B2
U = SE ,
(81)
B3
C3
B4
C4
with,
uT1 v1 uT1 v2
uT v1 uT v2
2
2
S=
uT v1 uT v2
3
3
uT4 v1 uT4 v2
ip1 l
e
0
eip2 l
0
E(l) =
0
0
0
0
and
uT u
p 1
p 1
0
U =
0
0
uT1 v3
uT2 v3
uT3 v3
uT4 v3
0
0
eip3 l
0
0
T
up2 up2
0
0
uT1 v4
uT2 v4
,
uT v4
3
uT4 v4
0
0
0
(82)
(83)
0
.
0
T
up4 up4
(84)
eip4 l
0
0
uTp3 up3
0
1
= U SE .
B3
C3
B4
C4
At z = +l we have using the same definitions as for before:
A1
C1
A2
C2
U = SE(l) .
A3
C3
A4
C4
(85)
(86)
Mixing up these two relations we obtain for the total transformation matrix V the following
relation:
B1
A1
B
A
2
1
1 1 2
(87)
= U SEE S U .
B3
A3
B4
A4
Using the definition
E(l)E(l)1 E,
(88)
(89)
57
As we shall find out later on the last form is a transformation that allows us to diagonalise the
motion matrix inside the Qball. Using this matrix we shall construct all the reflection and diffusion coefficients for all the waves moving inside and outside of the Qball. Before we continue
we need to remember that p3 = p1 and p4 = p2 for both sets of ps (barred ones and no bar
ones). We see here that the choice for normalisation of eigenvectors will just act on the U matrix that can be either the identity or the matrix or any other choice we can make. Finally the
diffusion matrix V we where looking for is given by:
V = U 1 SES 1 U.
(90)
In fact nothing we shall do depends on these matrices but we have shown the full procedure for
completeness.
4.6. Construction of reflection and transmission amplitudes
We shall first construct the reflection and transmission amplitudes from the left side to the
righthand side of the Qball. To do so we shall use the definition of the V matrix given by
Eq. (77). The first two coefficients are linked to rightmoving waves while the last two coefficients are linked to leftmoving waves. This choice for arranging the waves was done for
simplicity, we can show that any other arrangement leads to same results. In fact, we choose the
simplest possible arrangement. Due to the shape of the u spinors and the matrix in front the
first and the third coefficients of the free solution have the same energy while the second and the
fourth coefficients correspond to another energy wave. We shall identify these two energy ranges
to the type one particles (1) and type two particles (2). Using Eq. (77) and separating the matrix
into four two by two blocs we can write:
t1
V11 V12 t2
(91)
,
=
0
r1
V21 V22
0
r2
where the two arrows stand for the incoming waves, the first two coefficients will be replaced by
one. Using the bloc separation of the matrix we find:
t
= V11 1 ,
(92)
t2
t
r1
= V21 1 ,
(93)
r2
t2
leading to
t1
1
= V11
,
t2
r1
r2
.
1
= V21 V11
(94)
(95)
The R and T matrices will give the reflection and transmission amplitudes when they are applied
on the incoming wave coefficients. These two matrices are two by two the first line corresponding
to transmission or reflection of particles with two different incoming waves, while the second
58
Fig. 6. Sketch of both cases used to build the solution: we have each time two incident particles, two reflected and two
transmitted. It is an effect of massive particles because we cannot identify any more the particles with the antiparticles.
line gives the coefficients for antiparticles. We shall construct the transmission and reflection
coefficients from the right to the lefthand side of Qball. Using the same method as before we
have this time:
0
r1
V11 V12 r2
0
(96)
=
,
t1
V21 V22
t2
= V11 1 + V12
,
0
r2
rp
t1
= V21
+ V22
.
rap
t2
(97)
(98)
Solving these two equations gives the reflection and transmission matrices for incoming particles
from the left, they are:
= V12 V 1 ,
R
11
1
T = V22 V21 V11
V12 .
(99)
(100)
Now that all the coefficients are known we can construct and quantise the solution.
4.7. Construction of standard solution
Using the transmission and reflection coefficients we can identify two different cases the first
case is when incident particles are on the lefthand side of the Qball, while the other case stands
for incident particles coming from the righthand side (see Fig. 6). We shall treat separately the
solution on the left and the solution on the right, the matching coefficients are those found in the
previous section, they link the expansion coefficients on the right to those on the left. Writing
down these two possibilities we have:
L = B1 ei p1 x up1 + B2 ei p2 x up2 + r1 ei p1 x up1 + r2 ei p2 x up4 ,
(101)
i p x
R = t1 e 1 up1 + t2 ei p2 x up2 ,
(102)
for the first case, the two incident particles coming from the lefthand side of the Qball and
L = t1 ei p1 x up1 + t2 ei p2 x up2 ,
(103)
i p x
i p 2 x
i p 1 x
i p 2 x
1
up1 + r2 e
up2 + A1 e
up1 + A2 e
up2 ,
R = r1 e
(104)
59
(105)
(106)
To clearly understand the construction, the Bs are the incident amplitudes from the left while
the As are the amplitudes from the right. The 1 and 2 subscript indicate the type of particle we
are dealing with, we have two different exponentials in (t). We then need to take the complex
conjugate of the terms corresponding to the two last components of spinors. To continue building
the solution we still need to separate each of these two cases in two, considering only one type
of incident particle at the time. This construction leads to the four following pieces, that will be
identified to the four degrees of freedom that our solution has:
L = ei p1 x up1 + r11 ei p1 x up1 + r12 ei p2 x up2 ,
R = t11 ei p1 x up1 + t12 ei p2 x up2 ,
(107)
for the one incident type one particle from the left and
L = r21 ei p1 x up2 + ei p2 x up2 + r22 ei p2 x up2 ,
R = t21 ei p1 x up1 + t22 ei p2 x up2 ,
for an incident type two particle. The coefficients are given by:
1
t11
1
r11
=R
,
=T
,
0
0
r12
t12
0
t21
0
r21
=R
,
=T
.
1
1
r22
t22
(108)
(109)
(110)
The two other pieces for particles incident from the right we have:
L = t11 ei p1 x up1 + t12 ei p2 x up2 ,
R = r11 ei p1 x up1 + ei p1 x u3 + r12 ei p2 x u2 ,
(111)
for one incident type one particle from the right and
L = t21 ei p1 x up1 + t22 ei p2 x up2 ,
R = r21 ei p1 x up1 + r22 ei p2 x up2 + ei p2 x up2 ,
(112)
for an incident type two particle and finally the coefficients are given by:
1
r11
t11
1 ,
=R
= T
,
0
0
r12
t12
0
r21
t21
0 ,
=R
= T
.
1
1
r22
t22
(113)
(114)
If we want to easily remember the coefficients there is an easy trick. For the coefficients without
the tilde we read the subscript from the left to the right, r21 is the coefficient for an incident
type two particle being reflected as a type one particle. The same lecture stands also for the tilde
coefficients representing incident particles from the right.
60
4
#
(Aj , Bj )ei(
0
2 )t
0
up
i p i z
upj ei pj z + Aj , Bj ei(
+ 2 )t udown
,
pi e
(115)
j =1
in our case only the eigenvectors corresponding to the p1 eigenvalue have up components, while
only the eigenvectors corresponding to p2 have down components.
0
0
up
up
L = ei(
2 )t ei p1 x up1 + r11 ei p1 x up1 + ei(
+ 2 )t r12 ei p2 x up2 ,
(116)
incident particle is the wave containing up1 , all this superposition must be represented using the
same operator so we are sure to have only four degrees of freedom. Quantisation will be done
using energy, for the incident wave we have:
0
MD ,
2
0
,
MD +
2
leading for this first wave to
ei(
0
2 )t
u1
0
up
up
d
ei(
2 )t ei p1 x u1 + r11 ei p1 x u3
L =
MD +
(117)
0
2
+ ei(
+
0
2 )t
r12 ei p2 x udown
b (p 1 ),
p2
i(
+ 20 )t
we have,
0
up
up
d
ei(
2 )t ei p1 x up1 + r11 ei p1 x up1 bp1
L =
MD +
(118)
0
2
+ ei(
+
0
2 )t
i p x down
r12 e 2 up2
bp1 .
(119)
Applying the same method to all the terms we finally obtain for the total solution having four
degrees of freedom:
0
up
up
d
ei(
2 )t ei p1 x up1 + r11 ei p1 x up1 bp1
L =
MD +
0
2
+ ei(
+
+
MD
0
2
0
2 )t
i p x down
r12 e 2 up2
bp1
0
up
d
ei(
2 )t r21 ei p1 x up1 ap2
+ ei(
+
0
2 )t
0
2
0
up
i(
+ 20 )t i p 2 x down
t22 e
up2
ap2 .
d
ei(
2 )t t21 ei p1 x up1 a
p 2 + e
+
MD
i p x
i p 2 x down
e 2 up2 + r22
up2
ap2
e
0
up
i(
+ 20 )t i p 2 x down
t12 e
up2
bp1
d
ei(
2 )t t11 ei p1 x up1 b
p 1 + e
+
MD +
61
0
2
0
2
0
2 MD
0
up
d
ei(
2 )t r21 ei p1 x up1 ap2
L =
MD
0
2
+ ei(
+
0
2 )t
i p x down
i p 2 x down
e 2 up2
up2
ap2
+ r22
e
0
2 MD
0
up
d
ei(
2 )t t21 ei p1 x up1 a
p 2
+
MD
0
2
+ ei(
+
0
2 )t
i p x down
t22 e 2 up
ap2 .
2
(120)
One interesting result can be found here is that like in the massless case if we change the sign of
0
2 we change the particle type since we change the operator type. For the moment the a coefficients are only expansion coefficients since we have not quantised the wave yet. The solution on
the righthand side of the Qball is given by:
0
2 MD
R =
i p x
t12 ei p1 x up1 ap2 + t22
e 2 up2 ap2
0
2
0
2 MD
MD
i p x
i p 1 x
1 u
r12 ei p1 x up1 a
up2 ap 2 .
p2 + e
p 2 + r22 e
(121)
MD 20
At t = + only the terms without any incident wave will survive we have then
r22
a(p 2 ) + t22
a(p 2 ) + r21 a (p 2 ) + t12 a (p 2 ) = aout (p 2 ),
(122)
this is the Bogolubov transformation we where looking for. If we want the aout coefficient to be
a operator we need to check that it satisfies the same anticommutation relations as ain . We have:
aout , aout = r21
ap2 + r22
ap + t21
ap2 + t22
ap
2
2
+
(t
)a
(r12 )ap2 + (r22 )ap2 + (t21 )a
22 p 2
p 2
r21 + r22
r22 + t21
= r21
t21 + t22
t22 ain , ain .
62
This relation can also be obtained if we set that the incident current is equal to the outgoing one,
or even with the normalisation of wave packets. At this stage it can be important to use some
normalised eigenvectors. As will shall show later on it is always the case if we diagonalise the
matrix outside the Qball. The number of created particles is now given by
r21 2 + t21 2
(123)
(
).
in
0aout aout 0
in =
r21 2 + r22 2 + t21 2 + t22 2
We need to smooth out this result, to do so we shall use the same argument as [2] to finally obtain
dN
1
=
dt
2
0
2 MD
MD 20
r21 2 + t21 2
d
.
r21 2 + r22 2 + t21 2 + t22 2
(124)
0
up
d
ei(
2 )t r21 ei p1 x up1 ap2
L+R =
MD
0
2
+ ei(
+
0
2 )t
i p 2 x down
ei p2 x udown
up2
ap2 ,
+ r22
e
p2
0
2 MD
0
up
d
ei(
2 )t t21 ei p1 x up1 a
p 2
+
MD
0
2
+ ei(
+
0
2 )t
i p 2 x down
t22
up2
ap2 ,
e
(127)
0
2 MD
MD 20
r21 2 + t21 2
d
,
r21 2 + r22 2 + t21 2 + t22 2
(128)
after normalisation of operators. These results seem to be correct because when the fermions
become massless there is identification of both types of produced particles so the total coefficient
becomes equal to one as in the previous section and there is total reflection. We could stop our
calculations here since we have got the expression of particle production state and all the matrix
elements are known. But we can greatly simplify this problem so the expressions we obtain
become simpler and more readable.
63
(129)
with,
0
cosh(x1 ) sinh(x1)
0
sinh(x1 ) cosh(x1 )
v1 =
0
0
cosh(x2 )
0
0
sinh(x2 )
0
0
,
sinh(x2 )
cosh(x2 )
v1T v1 = .
(130)
(131)
The last equation ensures us the fact that v T = v 1 and that the symmetry of the problem is
conserved. Setting x1 and x2 being solutions of:
sinh(2x1 )
,
MD
+ sinh(2x2 )
cosh(2x2 ) =
,
MD
cosh(2x1 ) =
k1
0
M0 =
0
0
0
0
0
0
k1 0
,
0
0
k2
0
0 k2
(132)
(133)
where
2
2 ) sinh(x )
(MD
1
,
MD
2
2 ) sinh(x )
(MD
2
+
.
k2 =
MD
k1 =
(134)
All the parameters we find after this transformation are real, the ks that we find represent the
momentum of the particles in this new base. The same transformation applied on the M1 matrix
defined by Eq. (65) gives:
k1
0
sinh(x)
cosh(x)
cosh(x)
sinh(x)
k1
0
M1 =
(135)
,
0
sinh(x) cosh(x)
k2
cosh(x) sinh(x)
0
k2
64
with x = x1 + x2 . We can check that this new matrix has the same symmetry properties as M1 .
This simple transformation allows us to eliminate the Dirac coupling of our equations but in
presence of the Qball it is replaced by a double Majorana coupling. Since this transformation is
made every where, it will not change any of the properties. A solution of Eqs. (132) is easy to
construct it is:
cosh(2x1 )
(136)
, 2x1 = argcoth
= coth(2x1 ) =
.
sinh(2x1 )
MD
MD
Once this transformation is made inside and outside of the Qball the eigenvectors outside of the
Qball only contain one nonzero component, the important fact here is that the M0 matrix is now
self adjoint so its eigenvectors have the standard orthogonality properties without the matrix.
When we do the matching in space Eq. (80) instead of multiplying by uTi we multiply by ui
and the Smatrix is made of the components of M1 eigenvectors, so the only thing we did was to
diagonalise the transformed matrix. To continue the diagonalisation process we now transform
M1 in the way:
M1 = (s1 s2 )T M1 (s1 s2 )
with,
cosh[y/2]
0
0
sinh[y/2]
0
cosh[y/2] sinh[y/2]
0
s1 =
,
0
sinh[y/2] cosh[y/2]
0
sinh[y/2]
0
0
cosh[y/2]
cos[z/2]
0
sin[z/2]
0
0
cos[z/2]
0
sin[z/2]
s2 =
.
sin[z/2]
0
cos[z/2]
0
0
sin[z/2]
0
cos[z/2]
As before we have:
(137)
(s1 s2 )T (s1 s2 ) = ,
(138)
(139)
(140)
to preserve the symmetry of the problem. This set of transformations looks more complicated
then the simple boost we used to start, it is the case for the parameters we shall need to use. But
it is of great use for the final simplifications and results. Setting y and z to be solutions of:
2 cos(z) cosh(y) sinh(x)
k 1 k2
2 cosh(y) cosh(x)
sinh(y) =
k1 + k2
2 cosh(y) sinh(x)
tan(z) =
,
k1 k2
2 cosh(x)
,
tanh(y) =
k 1 + k2
we have
A M
0
0
0
0
M A
M1 =
,
0
0
B
M
0
0
M B
sin(z) =
(141)
(142)
(143)
65
with,
4 cosh(x)2
1
(k1 k2 ) cos(z) + k1 + k2
cosh(y)
2
k 1 + k2
4 cos(z) sinh2 (x) cosh2 (y)
+
(144)
,
k1 k2
4 cosh(x)
1
(k1 + k2 ) cos(z) + k1 + k2
cosh(y)
B=
2
k1 + k2
4 cos(z) sinh2 (x) cosh2 (y)
(145)
,
k1 k2
cosh(y) sinh(2x)
.
M =
(146)
k1 + k2
Taking a look at this M1 matrix we see it has the same form as the M0 matrix so we shall
diagonalise it using the same boost transformation. This time the transformation will not be
a boost since some parameters can be complex. In fact before going any further we have to find
the solution to Eq. (142) that night be complex, k1 + k2 is small so the fraction on the righthand
side is always bigger than one. We shall have to set y = i 2 + , this little trick allows us to easily
solve all these equations. Finally to finish the diagonalisation we transform using the v1 matrix:
A=
(147)
with v3 = v2 v1 . This last transformation can also be done using a slightly different matrix the v1
matrix defined by:
cosh(a) sinh(a)
0
0
0
0
sinh(a) cosh(a)
v1 =
(148)
,
0
0
cosh(b) sinh(b)
0
0
sinh(b) cosh(b)
T
v1 v1 = .
(149)
We finally have for the M1 matrix the form
1
0
0
0
0
0 1 0
M1 =
,
0
0
0
2
0
0
0 2
with
1 = A2 M 2 sinh(2a),
M
2 = B 2 M 2 sinh(2b),
M
(150)
(151)
where
M cosh[2a] = A sinh(2a),
M cosh[2b] = B sinh(2a).
(152)
Using these transformation the diffusion matrix V can be expressed in terms of the diagonalisation matrices in the way:
V = (s1 s2 v1 )T E(s1 s2 v1 ).
(153)
66
Fig. 7. Particle production rate for small values of the size parameter.
Fig. 8. Particle production rate for small values of the size parameter.
This form will be in fact far more simple then all the other possible ones, so this is the reason why
we decided to use it rather then the form with the scalar products of the eigenvectors. What we
do is exactly the same since we work in a base where the matrix of motion equations is diagonal.
If we keep the scalar products with the matrix we see that the only to vectors having negative
values are the negative moving ones, with a simple calculation we could link the matrix to
the helicity operator. This final expression we obtained will lead to simple results and is used to
compute all the reflection and transmission amplitude the results are given in the next section.
4.9.1. Small sized Qballs
Using the results of previous section we where able to compute all the amplitudes for
small sized Qballs. The method used was to replace the exponentials in the E matrix by:
(1 ip1,2,3,4 l). These amplitudes still have complicated expressions but all of them except t22
are proportional to the size parameter Ml l. In the limit where l goes to zero all the amplitudes
fall to zero except t22 going to one. The t22 amplitude representing the probability of a fermion
remaining a fermion. This probability is obviously one if the Qball disappears. If the size parameter is small the amplitudes will become proportional l 2 as for massless particle production.
This quadratic behaviour is shown on Fig. 7.
5. Results of numerical integration
We first tested the stability of production rate in function of size to see if like in the previous
case the particle production rate becomes constant and stable for big values of the size. If the
production becomes constant above a certain size then we do not need to care about complex
averaging processes. Fig. 8 shows the stability of evaporation rate for large Qballs. The little
67
dN
Fig. 9. Evaporation rate, 2
M dt for infinite (very big) Qballs in function of the frequency parameter for different values
of the fermion mass parameter.
oscillations are due to numerical instabilities that vanish for very big values of size. We can now
compute the evaporation rate for values of the Dirac mass smaller then the Majorana mass (the
coupling inside the Qball). The next task we need to do is test the stability of our computations
when the fermion mass parameter is bigger then the Majorana coupling inside the Qball, it is
the case when MD 1. This case shows exactly the same behaviour of the other one except for
the fact that it takes more computer time to obtain the plot of evaporation. The results are on
Fig. 9 and Fig. 10. A quick analysis of these results shows that there is a superior limit for all
parameter sets, this limit does not depend on the mass parameter. Is seems to be normal since
an infinite Qball with an infinite internal frequency can produce any mass fermions. The last
words we shall say about these results is that the angle in the curve correspond to the value of
the frequency where the imaginary part of the impulsion inside the Qball becomes zero, it is the
point where particles start to propagate inside the Qball. The normalisation of evaporation by its
upper bound will lead to the same shape as the massless case but there will be a gap from zero to
the value of the fermion mass.
68
dN
Fig. 10. Evaporation rate, 2
M dt for infinite (very big) Qballs in function of the frequency parameter for different values
of the fermion mass parameter bigger then one.
69
dE
=
M dt d
+ 2M0
MD
M
M
2M0 + MD
r21 2 + t21 2
2 d
,
r21 2 + r22 2 + t21 2 + t22 2
(154)
the transmission amplitudes disappear when the Qballs size is very big. This integration can be
done numerically and we can also introduce the Qballs size to see its influence on the energy
flux (see Fig. 11). The only difference is that a small Qball will produce less energy until the
value of the frequency parameter becomes big. We could normalise these figures with the absolute upper bound. This normalisation does not introduce any new features since the normalised
curve for very big Qball would start with a constant part to then fall down. For the small one we
will not find any constant part in the normalised curve.
7. Conclusions
As we have seen the coupling between the scalar field and fermionic field leads to particle production from the Qball [2]. To study this particle production we constructed the exact
quantummechanical state describing the particle producing Qball. We used Heisenbergs picture of quantum mechanics, the state describing the produced fermions is fully characterised by
the fact that no fluxes are moving towards the Qball. This condition is solved considering the asymptotics of the fields far away from the Qball. Using this state we constructed the Heisenberg
field operator describing massless fermions produced by a Qball. This construction allowed us
to prove that for large Qballs in one space dimension the particle production does not depend on
the Qballs size. While for small Qballs the particle production rate is proportional to l 2 . The
extension of these results to three space dimensions is simple. For large Qballs particle production is an evaporation, while for small Qballs it depends on the size. The other result we need
to point out is that we can consider a variety of kinematical constructions to compute the evaporation rates. The first one is the standard one where we compute the reflection and transmission
amplitudes for an incoming wave. The second one we used is based on the fact that no particles
are moving towards the Qball. We proved that these two pictures are equivalent.
The fact that fermions acquire a Dirac mass does not introduce many changes. In 1 1 dimensions the particle production rate does not depend on the Qballs size for sufficiently big
ones. This result is not very surprising, since taking the limit m 0 leads to the same results
as the coupling with massless fermions. In this case the only difference is that evaporation occurs in a different range. The internal frequency 0 , the energy of one single scalar forming
the Qball, must be bigger than the produced fermion mass. This result is also quite intuitive,
the scalars forming the Qball desintegrate into fermions so their energy must be bigger than
the fermion mass. The second fact is that particle production occurs in the range mixing positive and negative frequency terms. In this range the Bogolubov transformations we build are
non trivial. Using these two results we proved that evaporation can only take place in the range:
[MD 20 ; 20 MD ], with 20 MD . This result is in total accordance with the previous work
70
done on the subject [2,3], and extends it a significant way. This new definition for the evaporation
range will introduce a new upper bound for the evaporation rate.
When the Qballs size becomes small there is no more evaporation since the production rate
depends on the size. For small sized Qballs the particle production rate is proportional to l 2 .
The size will also slow down the energy flux a distant observer can measure. Taking the limit
l does not need any complex averaging processes since the evaporation rate is constant
and l independent for big values of the size.
We also computed all the transmission and reflection coefficients for a massive fermion being
scattered by a large Qball. This construction allowed us to compute the exact profile of the
evaporation rate. Using these profiles we proved that both constructions are equivalent. The last
result we have proved is that evaporation rate is proportional to 0 in one space dimension while
it is proportional to 03 in three dimensions. In fact in three dimensions it is proportional to
( 20 MD )3 . These reflection and transmission amplitudes will be used to study the behaviour
of Qballs in matter.
Acknowledgement
The author would like to thank M. Shaposhnikov for suggesting this problem and for useful
discussions.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
Received 20 July 2006; received in revised form 21 August 2006; accepted 30 August 2006
Available online 20 September 2006
Abstract
We study the properties of configurations from which Pvortices on one hand or Abelian monopoles on
the other hand have been removed. We confirm the loss of confinement in both cases and investigate in what
respect the modified ensembles differ from the confining ones from the point of view of the complementary
confinement scenario.
2006 Elsevier B.V. All rights reserved.
1. Introduction
There are two popular phenomenological explanations of confinement in lattice gluodynamics: the monopole [1] and the center vortex [28] confinement mechanism that have been critically discussed in Ref. [9]. Monopoles and center vortices are defined by projection to U (1)N1
or Z(N) gauge fields, respectively. Hence, the center vortices are called Pvortices in order to
emphasize the difference from extended (continuum) vortices.1 These two types of excitations
reproduce, respectively, about 90% [10] and about 70% [11] of the nonAbelian string tension
and are able to explain other nonperturbative properties. Therefore, removing monopole fields
* Corresponding author.
05503213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.08.025
72
[12,13] or Pvortices [14] should leave only inert gauge field configurations, unable to confine
external charges or to break chiral symmetry.
It was very attractive to conjecture [15,16] that monopoles and Pvortices are geometrically
interrelated. Indeed, this was found to be the case in SU(2) gluodynamics: more than 90% of
monopole currents are localized on the Pvortices [1618]. One of our goals in the present paper
is to elaborate on this interrelation between monopoles and vortices.
In Ref. [14] the operation of center vortex (Pvortex) removal has been invented, and it was
demonstrated that the quark condensate and the topological charge are destroyed thereby. Later it
has been found [19] that the number of links that have to be modified in this operation can be essentially reduced when the Z(2) gauge freedom is used to minimize the number of negative trace
links. The remaining negative links can be interpreted in terms of threedimensional domains.
The operation of monopole removal has been invented by Miyamura et al. in Refs. [12,13].
It has been shown that not only the quark condensate and the topology is destroyed thereby, but
that the Abelian monopole field alone, completed again by the nondiagonal gluons, carries all
necessary degrees of freedom in order to allow the reconstruction of chiral symmetry breaking
and topological charge.
All technical details (including those of gauge fixing) relevant for our study are relegated to
Appendix A.
de Forcrand and DElia [14] were the first to ask what happens to the Abelian monopoles when
Pvortices are removed. We have started our work from the opposite question: what remains
from the center vortex degrees of freedom when Abelian monopoles are subtracted from the
Abelian part of the gauge fields. Originally, it was only for completeness that we also repeated
the study of Ref. [14]. Not unexpectedly, we found that the results depend on the method of
gauge fixing and details of its algorithmic implementation. But turning again to de Forcrands
and DElias problem with the perfectionated gauge fixing has resulted in a paradox that has led
us to a closer investigation what properties the network of monopole world lines must possess in
order to confine.
Most of the above mentioned results in the literature was obtained for the direct center gauge
(DCG) [4] and for the indirect center gauge (ICG) [2]. In the present paper we also use both of
them. The Laplacian center gauge [20], also used for studies of the Pvortices, is not studied here.
Although the absence of the Gribov ambiguity is a virtue of this gauge, we consider the absence
of scaling of the resulting density of Pvortices [21] as a less appealing feature. We have been
doing all gauge fixings by means of simulated annealing, repeating however the procedure for a
number of copies. In this respect we have done our best (according to the present state of the art)
to circumvent the infamous Gribov ambiguity.
In all studies described in this paper we have used a confining ensemble of 100 configurations
on a 243 6 lattice, generated at = 2.35 using the Wilson action. This corresponds to a lattice
where the ratio Tc / = 0.7091(36) (obtained for SU(2) gluodynamics with Wilson action in
Ref. [23]) has been used. Adopting the standard value of the string tension, = (440 MeV)2 , the
estimated lattice spacing would be a = 0.139(1) fm, resulting in a spatial volume of (3.336 fm)3 .
2 We thank Mike Teper for confirming that this value actually refers to = 2.35.
73
In order to demonstrate the confining property or the loss of confinement in the case of finite
temperature adopted here, we use the correlator of the Polyakov loop,
aV (r) =
1
log P (
x )P (
y)
N
(1)
with r = 
x y. Strictly speaking, this is the excess free energy related to the presence of a
quarkantiquark pair. For simplicity we will call it potential, however. For the full nonAbelian
potential the Polyakov loop
1
Ux,t;4
Tr
2
P (
x)=
(2)
t=1
is used. When the monopolegenerated potential shall be sorted out, the monopole part of the
Abelian Polyakov loop,
N
N
mon
mon
mon
(
x ) = Re exp
ix,t,4 = cos
x,t,4 ,
P
(3)
t=1
t=1
P vort (
x)=
N
Zx,t;4 ,
(4)
t=1
74
Nmon
4Ns3 N
and vort =
Nvort
,
6Ns3 N
(5)
where Nmon is the number of dual links occupied by monopole currents and Nvort the number of
dual plaquettes belonging to the vortex area.
The observed scaling of the monopole density has later found to be premature. Moreover,
the density lacks scaling and universality [35,36] and requires to be separated into ultraviolet and infrared component, the scaling behavior of which is different, meaning that the UV
component reflects only short distance artefacts without relevance for confinement. Only the
IR component turned out to be scaling and relevant for condensation [37] and for confinement.
Relatively soon after the first quantitative study of mon [34], the focus has turned to structural properties of the network of monopole currents, which has found to be clearly undergoing a
change at the deconfinement phase transition [38]. The change concerned the distribution of connected clusters with respect to the length L (i.e. the number of participating monopole currents).
This distribution is denoted as N (L) and counts the number of connected clusters of given size L
in the ensemble of configurations. In [38], and later in Refs. [39] and [40,41] it has been clearly
demonstrated that, in the confinement phase, each configuration contains a single macroscopic
cluster (in modern parlance the infrared monopole component) of many connected dual links
carrying nonvanishing monopole current, apart from a big number of small clusters (the ultraviolet monopole component). The macroscopic clusters were found to be percolating. This is
meant in the sense that they are impossible to enclose in a 4D cuboid smaller than the periodic
lattice in one ore more directions or in the sense of a cluster twopoint function [37] not to vanish
at infinite distance. Obviously, this notion of percolation does not take into recognition the
orientation of the monopole currents. The eventual occurrence of more than one macroscopic
cluster was found to be an effect of smaller volume.
75
Of course, it is not difficult just to evaluate Wilson loops under the influence of the Abelian
field due to all monopole currents [1,42,43], a procedure that clearly takes the orientation into
account. In this way, monopole dominance [1,10] has been demonstrated. Later it has been found
that the macroscopic clusters (and only these) build up the Abelian string tension [41]. The
essence of the deconfinement transition in this light had already earlier been identified [38] as
the decomposition of the macroscopic cluster into many smaller ones, i.e. a qualitative change
of the size distribution of the clusters, accompanied by the generation of an anisotropy of the
current density. In the Euclidean time direction, the percolation property of the remaining larger
clusters persists above Tc , whereas there is no percolation anymore in the spatial directions.
All this contributed to the common belief that spacelike percolation is a necessary and sufficient condition for confinement. In this paper we find an interesting counter example showing
that a percolating cluster does not necessarily confine, as the direct evaluation of the monopole
contribution to the Wilson loops reveals.
In the case of the Pvortex mechanism, the direct evaluation of their contribution to Wilson
loops (by a factor (1)linking number for each vortex), the fact of center dominance (of the string
tension) and the scaling property of the Pvortex density vort (being equal to the density of negative plaquettes) have been emphasized from the very beginning [5]. Also here, the existence of
one macroscopic Pvortex, a closed surface consisting of many dual plaquettes, was found to be a
prerequisite of confinement, and this Pvortex itself was found percolating [44] in all directions.
The deconfinement transition was later identified, in analogy to the monopole mechanism, with
a depercolation transition [7,45,46], i.e. with the macroscopic vortex being replaced by smaller
ones. If big enough, these are percolating in the time direction, but no more in spatial directions.
In our investigation, for Pvortices we did not find any indication that vortices would persist
to be spatially percolating in the case confinement has been destroyed (e.g. by the removal of
monopoles).
3. Indirect center gauge
ICG fixing includes two steps. First, the maximally Abelian gauge (MAG) is fixed and Abelian
projection is made. Then the Abelian gauge fields (and the remaining Abelian gauge invariance)
are used to fix the (Abelian) maximal center gauge. After the center projection is made, the
emerging Z(2) configuration is looked up for negative links Zx, = 1 (see Appendix A for
further details). In Fig. 1 the length spectrum N (L) of the monopole clusters, obtained after
fixing the MAG, is compared with the same spectrum after the Pvortices have been removed.
More precisely, here and later N (L) represents the number of clusters of size L found in all 100
configurations. Note, that the vortex removal operation consists in changing the sign of links with
negative cos(x, ) in the (Abelian) maximal center gauge, and that this operation preserves the
MAG [10].
The original monopole density (per dual link) is mon = 0.0357 0.0002 and the original
vortex (area) density (per dual plaquette) is vort = 0.0805 0.0002.
It is seen that the percolating monopole clusters with length around 7 000 that are present
before in each lattice configuration, are absent from the length spectrum after the vortices have
been removed. Only small monopole clusters survive. The density of monopoles mon is reduced
by a factor 0.500 0.003. Combining this observation with the abovementioned fact of strong
correlation between monopoles in MAG and vortices in ICG we conclude that the removal of
Pvortices gives rise to removal of almost all monopoles belonging to the infrared part of the
monopoles. It is well known that the string tension vanishes in the modified ensemble.
76
(a)
(b)
Fig. 1. The monopole clusters length distribution N (L) for Wilson = 2.35 on 243 6 lattices before (a) and after (b)
removing the center vortices found according to ICG.
(a)
(b)
Fig. 2. The area spectrum of center vortices for Wilson = 2.35 on 243 6 lattices (a) for the original ensemble put into
ICG, and (b) after monopoles have been removed.
In Fig. 2 the area spectrum of Pvortices, as it is found after ICG of the original ensemble, is compared with the spectrum that is obtained when the singular (monopole) part is
subtracted from the Abelian projected gauge field after MAG, before the ICG is finally accomplished. Thereby the original vortex density vort = 0.0805 0.0002 is reduced to one fifth,
vort = 0.0162 0.0003, without monopoles.
As long as the monopole fields are retained there is one big percolating vortex with an area
of about 40 000 plaquettes present in each configuration (in addition to a lot of small size ones,
with area 100). After the monopole contributions to the Abelian field are removed the percolating Pvortex has disappeared from all configurations, the largest vortices being smaller in
area than 6 000 plaquettes. Thus we observe that removing monopoles does not lead to the loss
of all big Pvortices (which would be analogous to what happens to the monopoles after vortex
removal), but rather to splitting of the single percolating Pvortex into a number of still relatively
extended Pvortices.
77
The above results are in agreement with both scenarios of confinement, claiming that a
nonvanishing string tension is due to percolating clusters of monopoles [40,41] or due to the
existence of percolating center vortices [44], respectively.
4. Direct center gauge: removing center vortices
When DCG fixing and center projection is completed, the vortex density is vort = 0.0605
0.0002. As we mentioned already in the introduction we found something unexpected for the case
of DCG, when center vortices are removed from the configurations. This is done in the following
way. Each Monte Carlo configuration is gaugefixed twice, to the MAG and to the maximal
center gauge by DCG. The nonAbelian gauge field is then projected either to an Abelian U (1)
gauge field (from MAG) or to an Z(2) gauge field (from DCG). In DCG, the remaining Z(2)
gauge freedom is exploited to reduce the number of negative (Zx, = 1) links.
In Ref. [14], de Forcrand and DElia noticed already the existence of a few percolating monopole clusters after the removal of vortices, clearly reduced in comparison to the amount of
huge monopole clusters in the normal case. He mentioned, that only a handful of large clusters survives, whereas most of them are broken into pieces. Moreover, even the remaining largest
clusters, according to his observation, did not contain monopole loops that wind around the periodic lattice. For the case of asymmetric lattices (finite temperature) considered in our work
this observation would imply the absence of spatial winding. Qualitatively our results are in
agreement with this expectation. Here we do not consider the winding number of the emerging
monopole network clusters. Instead, we shall apply the decomposition of the clusters into nonintersecting monopole loops and find that none of them is spatially winding around the lattice.
Note that certain quantitative differences are possible due to the fact that DCG and MAG both
are afflicted by their Gribov ambiguities. Our method differs from the method in Ref. [14] that we
(i) always used the simulated annealing version of the gauge fixing procedure in question, i.e. we
obtained higher local maxima for respective gauge fixing functionals, and (ii) the Z(2) Landau
gauge is realized. Both modifications are expected to lead to a minimal number of nonvanishing
monopole currents and a minimal number of negative Z(2) links, respectively. The realization of
the Z(2) Landau gauge has no effect on the monopole configuration. The procedure of Pvortex
removal then requires the multiplication of each link by Zx, . This is applied not only to the
original nonAbelian configuration but also to the Abelian projected configuration. In this way
the monopole content of configurations with Pvortices removed can be easily compared with
the monopole content of the original configuration. The monopole density mon is enhanced
to mon = 0.0551 0.0003. We found, in particular, that the monopole clusters still contain a
percolating component in agreement with Ref. [14].
The apparent paradox is demonstrated in Fig. 3 where the monopole cluster length spectrum
before and after center vortex removal is compared. On the other hand, the nonAbelian string
tension vanishes as it should [14] although only a minimal number of links has actually been
changed. As expected in such a case, the monopole string tension also vanishes as it is shown in
Fig. 4 where the nonAbelian and the monopole potentials before and after vortex removing are
depicted.
Thus, monopole percolation is not sufficient to generate a confining potential. The monopole
string tension, and consequently the Abelian string tension, may vanish if the percolating monopole cluster is highly correlated at small distances. In the following we will present several pieces
of numerical evidence pointing out that the geometrical properties of monopole clusters after the
removal of center vortices in DCG are indeed very different from the confining case.
78
(a)
(b)
Fig. 3. The same as Fig. 1, but (b) shows the situation after the center vortices found according to DCG have been
removed.
(a)
(b)
Fig. 4. The nonAbelian potential (aV (r), denoted by squares) and the monopole potential (aVmon (r), denoted by circles)
at = 2.35 on 243 6 lattices before (a) and after (b) removing of center vortices.
(i) The number of selfintersections of the percolating cluster after removing of center vortices
is substantially larger than before, as can be seen in Fig. 5(a).
(ii) A useful quantity to discriminate between confinement and deconfinement phases at finite
temperature is suggested by the following fact. The monopole part P mon (
x ) of the Polyakov
loop in any threedimensional point x is defined in Eq. (3). Using Eq. (A.7) one can express the
monopole part P mon (
x ) as
x ) = ei(x )/2 ,
P mon (
(6)
where (
x ) is the solid angle of all monopole trajectories (projected onto a single time slice,
with proper regard of the direction of the monopole current) as monitored from the locus x of
the Polyakov loop (thought to be located in the same time slice).
(
x ) has been introduced as an observable characterizing the difference between the confinement and deconfinement phase in Ref. [47]. It has turned out that the distribution of the local
(a)
79
(b)
Fig. 5. The ratio of the total length of the largest cluster to the number of crossings in it (a) and the solid angle
distribution (b) for normal configurations (full circles) and configurations free of center vortices (empty circles). The
error bars in (b) are less than symbols.
values (
x ) (the histogram summed over all configurations of an ensemble) is flat in confinement and has a Gaussian form in the case of deconfinement. Fig. 5(b) shows the distributions
of (
x ) as found for normal confining configurations and for the ensemble cleaned from center
vortices. The distribution for the latter is qualitatively very similar to that in the deconfinement
phase.
(iii) In order to describe the correlations within the apparently percolating monopole clusters, it is useful to decompose the cluster by subtracting from bigger clusters closed, oriented
and irreducible (nonselfintersecting) loops, stepping up in loop size (lengths of 4, 6, 8, etc.)
until nothing remains to be subtracted. The result of this procedure can be presented as a his
togram N(L)
of irreducible loops (with respect to the loop size) in place of the histogram N (L)
of connected clusters (with respect to the length of monopole currents inside the clusters). Fig. 6
is the analog to Fig. 3, showing now the respective loop length distributions. It turns out from
comparing Figs. 6 and 3 that the percolating but nonconfining monopole clusters can be completely decomposed into loops of comparably small size. The irreducible monopole loops of the
original ensemble (before removing center vortices) ranges to three times larger loops. Among
them are such wrapping around the lattice.
This result is corroborated by the following random simulation of nonconfining, yet percolating monopole clusters. We have tried to model the monopole configurations left over after removal of vortices by a random gas of small monopole loops. Such configurations were
created by the following procedure. A set of NDPL Dirac plaquettes were put on the lattice
with randomly chosen location in a randomly chosen plane, with random sign of mx, . In
Fig. 7 the resulting distribution of (
x ) and the static potential are shown for NDPL = 6 000.
For this number of Dirac plaquettes we find as the result of the procedure that a large connected monopole cluster of a length of O(10 000) has grown up and the total number of
monopole currents (including also smaller clusters) is close to that in the Pvortexremoved
80
(a)
(b)
Fig. 6. The length distribution N (L) of irreducible loops emerging from the decomposition of monopole clusters, for
= 2.35 on 243 6 lattices in DCG before (a) and after (b) removing the center vortices.
(a)
(b)
Fig. 7. The solid angle distribution (a) for a random monopole configurations obtained as described in the text and the
static monopolegenerated potential (b) for these configurations (full circles) compared with the ensemble free of center
vortices (empty circles). Error bars are within symbol size.
(a)
81
(b)
Fig. 8. The same as Fig. 2 but with vortices defined in DCG, (a) before and (b) after monopoles have been removed.
82
Fig. 9. The center vortex contribution to the quarkantiquark potential, rising for the original ensemble put into DCG
(denoted by squares), flat for DCG after Abelian monopoles have been removed (denoted by circles).
gauge is fixed within the residual Abelian gauge symmetry, before the projection to the center
subgroup Z(2) is made, and (ii) through the direct maximally center gauge achieved within the
nonAbelian gauge freedom. Abelian monopoles were throughout studied within the maximally
Abelian gauge. The gauges were iteratively fixed with the help of a simulated annealing algorithm, whichwhen applied to the consecutive step of complimentary gauge fixinghas erased
the memory with respect to the first step of gauge fixing.
We have found a clear correlation between both types of objects. We confirm that the removal
of each of these types leads to a loss of confinement described in terms of the other type, respectively. However, in case of removing Pvortices detected within the direct center gauge, we found
much more monopole currents and larger (spanning) monopole clusters than seen before Pvortex
removal. A closer inspection showed that these monopole clusters are strongly correlated at short
range such that they are reducible into a large number of considerably short monopole loops, the
latter not contributing to a linearly rising confinement potential. This somewhat surprising observation shows, that spatial percolation of infrared monopole clusters alone is not a sufficient
condition for real monopole condensation and thus for quark confinement (for the connection of
the latter see Refs. [33,48]).
We have restricted our investigations to the case of just one temperature value within the
confinement phase. We are convinced that this does not represent a real loss of generality as long
as T < Tc .
Acknowledgements
The authors want to thank M.N. Chernodub for valuable discussions and Ph. de Forcrand for
useful remarks on a first version of this paper. The work was partially supported by grants RFBRDFG 060204010 and DFGRFBR 436 RUS 113/739/2. The ITEP group (A.I.V., B.V.M., M.I.P.,
V.G.B., A.V.K. and P.Yu.B.) was partially supported by grants RFBR 060216309, 050216306,
050217642 and 040216079, and by the EU Integrated Infrastructure Initiative Hadron Physics
(I3HP) under contract RII3CT2004506078. The work of E.M.I. is supported by DFG through
the Forschergruppe FOR 465 (Mu932/2).
83
Appendix A
Here we give the standard definitions for SU(2) lattice gauge theory which we have used in
the studies described in the text. We perform our analyses both in the direct [4]and in the
indirect [2]maximal center gauges (DCG and ICG). The DCG in SU(2) lattice gauge theory is
defined by the maximization of the functional
2
F1 (U ) =
(A.1)
Tr g Ux, ,
x,
with respect to gauge transformations g SU (2). Ux, is the lattice gauge field and g Ux, =
the gauge transformed lattice gauge field. Maximization of (A.1) fixes the
g (x)Ux, g(x + )
gauge up to Z(2) gauge transformations, and the corresponding, projected Z(2) gauge field is
defined as:
Zx, = sign Tr g Ux, .
(A.2)
After this identification is made, one can make use of the remaining Z(2) gauge freedom in order
to maximize the Z(2) gauge functional
z
F2 (Z) =
(A.3)
Zx,
x,
with respect gauge transformations z(x) Z(2), Zx, z Zx, = z (x)Zx, z(x + ).
This is the
Z(2) equivalent of the Landau gauge. In distinction to Ref. [14], this final step is automatically
understood here before the vortex removal operation (to be defined below) is done.
The ICG goes an indirect way. At first, one fixes the maximally Abelian gauge (MAG) by
maximizing the functional
F3 (U ) =
(A.4)
Tr g Ux, 3 g Ux, 3 ,
x,
with respect to gauge transformations g SU (2). The maximization fixes the gauge up to
g U (1). Therefore, the following projection to an U (1) gauge field through the phase of the
diagonal elements of the links, x, = arg((g Ux, )11 ), is not unique. The nonAbelian link field
is split according to Ux, = ux, Vx, in an Abelian (diagonal) part ux, = diag{exp(ix, ),
exp(ix, )} and a coset part Vx, SU (2)/U (1), the latter representing nondiagonal gluons. Exploiting the remaining U (1) gauge freedom, which amounts to a shift x, x, =
one can maximize the Abelian gauge functional
(x) + x, + (x + ),
2
F4 (u) =
(A.5)
cos x, ,
x,
that serves the same purpose as (A.1). Finally, the projected Z(2) gauge field is defined as
Zx, = sign cos x, .
(A.6)
This second step in ICG is completely analogous to the DCG case, in other words one maximizes F1 (U ) in Eq. (A.1), however for all Ux, restricted to the projected links ux, =
diag{exp(ix, ), exp(ix, )}, admitting only U (1) gauge transformations denoted by . The
Z(2) gauge variables are used to form Z(2) plaquettes. Most of them are equal to +1, typically 30 000 plaquettes are equal to 1 after center projection via DCG (and 40 000 after
84
center projection via ICG, respectively). The Pvortex surfaces are actually formed by plaquettes dual to the negative plaquettes. The vortex density is therefore vort = 0.0605 0.0002 and
vort = 0.0805 0.0002 in DCG and ICG, respectively.
In order to fix the maximally Abelian and the direct maximal center gauge we have created
10 randomly gauge transformed copies of the original gauge field configuration and applied the
Simulated Annealing algorithm [10] to find the optimal nonAbelian gauge transformation g. We
have used for further analyses always that copy g U (or ) which corresponds to the maximal
value of the respective gauge fixing functional. In order to perform the second step of the indirect
maximal center gauge via configurations, that have been first fixed to the maximally Abelian
gauge and projected to U (1) fields u, we have started again from 10 random Abelian gauge
transforms of the latter.
We have used the standard DeGrandToussaint definition [49] of monopole currents defined
by the phase x, of ux, . The typical number of monopole currents per configurations amounts
to 12 000 dual links, corresponding to a density of mon = 0.0357 0.0002. The part of the
Abelian gauge field originating from the monopoles is
mon
x,
(A.7)
= 2
D(x x ) mx , .
x
Here D(x) is the inverse lattice Laplacian, and is the lattice backward derivative. The Dirac
sheet variable, mx, , is defined as the integer multiple of 2 part of the plaquette angle x, ,
whereas the reduced plaquette angle x, (, ] is the fractional part: x, = 2mx, +
x, . The Abelian gauge field with monopoles removed is defined as [13]:
monopole removed
ux,
= umon
x, ux, ,
(A.8)
mon
mon
where umon
x, = diag{exp(ix, ), exp(ix, )}. Upon multiplication with the coset field Vx, ,
this holds also for the nonAbelian links
monopole removed
= umon
Ux,
(A.9)
x, Ux, .
Analogously the gauge fields with the Pvortices removed are defined as [14]:
vortex removed
= Zx, Ux, ,
Ux,
(A.10)
85
Abstract
We find a rotating KaluzaKlein black hole solution with squashed S 3 horizons in five dimensions. This
is a Kerr counterpart of the charged one found by Ishihara and Matsuno [H. Ishihara, K. Matsuno, Kaluza
Klein black holes with squashed horizons, hepth/0510094] recently. The spacetime is geodesic complete
and free of naked singularities. Its asymptotic structure is a twisted S 1 fiber bundle over a fourdimensional
Minkowski spacetime. We also study the mass and thermodynamics of this black hole.
2006 Elsevier B.V. All rights reserved.
1. Introduction
Very recently, Ishihara and Matsuno [1] have found a black hole solution in the fivedimensional EinsteinMaxwell theory. It is a charged static black hole with a nontrivial asymptotic topology. Horizons of the black hole are in the shape of squashed S 3 , and the spacetime is
asymptotically locally flat, approaching a twisted S 1 bundle over a fourdimensional Minkowski
spacetime. For simplicity, we will refer to this solution as an IshiharaMatsuno black hole
or an IM1 black hole for short, and use the name KerrIM for the Kerr counterpart with the
same asymptotic geometry, etc.
The IshiharaMatsuno black holes deserve further investigations for several reasons.
First, string theory tells us that spacetime has ten dimensions, and the idea of large extra
dimensions suggests a possibility to produce black holes smaller than the extra dimensions at
colliders [2,3]. The IshiharaMatsuno black holes bring us a new toy model to study. In this
Email address: wangtao218@itp.ac.cn (T. Wang).
1 Maybe RNIM is a more suitable name to adopt, but anyhow we will not do it here.
05503213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.09.001
87
spacetime, one extra dimension can be smaller than four uncompactified dimensions, but is
much lager than other dimensions. The size of this large extra dimension is characterized by an
adjustable parameter r [1].
Second, it is a longstanding problem to define and compute conserved quantities in a space
time with a nonflat asymptotic structure or a nontrivial boundary topology. Some breakthroughs
have been made [46], and more attention was attracted on this problem in recent years [712].
IshiharaMatsuno black holes have a nontrivial asymptotic structure, hence provide a new arena
to study diverse techniques we have got. According to calculations performed by Cai et al. [13],
the boundary counterterm method [10] and the generalized AbbottDeser method [7] result in
the same mass (energy) for IM black holes, which also satisfies the first law of thermodynamics.
However, as the compactified dimension expands, such a mass increases, even in the absence of
fivedimensional black holes. This is a little counterintuitive but may be explained by comparing
IM black holes with KaluzaKlein monopoles [14].
Third, from the discussion in [1], it is clear that horizons of IshiharaMatsuno black holes
are deformed owing to the nontrivial asymptotic structure. There is an interesting problem to
ask: what will happen if we switch on angular momenta, the cosmological constant and Chern
Simons terms [15] or replace the black hole by a black ring [16]? We have tried to answer this
question, but only bite a bit, that is, we obtain a KerrIM black hole solution with equal angular
momenta in Einstein theory whose cosmological constant is zero. The KerrIM black hole with
nonequal angular momenta is too complicated to struggle with. When looking for the metric of
dSIM black holes we lose our way. As far as we know, in fivedimensional asymptotically flat
spacetime, charged rotating black holes have not been obtained analytically in pure Einstein
Maxwell theory yet [17]. So a KerrNewmanIM black hole solution seems to be equally hard
to get. The solution for KaluzaKlein multiblack holes see [18].
The organization of this paper is as follows. We produce the KerrIM black hole and identify
its asymptotic spacetime in the next section. Then we prove in Section 3 that such a KerrIM
spacetime is geodesic complete and free of naked singularities. Sections 4 and 5 are dedicated
to computing the thermodynamic quantities as well as checking the first law. In Section 5, we
also analyze the results and make some comments on the relation between IM black holes and
KaluzaKlein monopoles. Finally, a brief summary is given in Section 6. Through the entire
paper we exploit the abbreviations introduced at the beginning of this section.
2. KerrIM black holes
2.1. A prescription to squash black holes
By observing the KaluzaKlein black hole with squashed horizons [1], and comparing it with
the fivedimensional ReissnerNordstrm black hole, we guess that there is a general prescription which transforms some known fivedimensional black hole into a new one with squashed
horizons. We further conjecture a detailed squashing transformation as the following
(1) Write the original black hole metric in terms of the leftinvariant MaurerCartan 1forms
on S 3
1 = sin d + cos sin d,
2 = cos d + sin sin d,
3 = d + cos d,
(1)
88
where
0 < < ,
0 < < 2,
0 < < 4
(2)
are Euler angles [13]. The Euler angles are related to ordinary angles (o , o , o ) via [15]
o o = ,
o + o = ,
k1 ,
1
o = .
2
2
k2
(3)
(4)
2 r 2 )(r 2 r 2 )
(r
+
,
2 r 2 )2
(r
(5)
r = r+ corresponds to the outer horizon and r = r the inner horizon. r characterizes the size
of a S 1 fiber at infinity. We impose to them a constraint 0 r r+ < r .
When we apply this procedure to Kerr black holes with nonequal angular momenta, it gives
a metric too complicated to struggle with. So in the rest parts of this article, we focus on the
squashed Kerr black holes with equal angular momenta, to which we sometimes refer as Kerr
IM black holes.
2.2. From Kerr to squashed Kerr
In terms of MaurerCartan 1forms, a fivedimensional Kerr black hole with two equal angular
momenta takes the form
2
r 2 + a2 2
m
a
ds 2 = dt 2 + dr 2 +
(6)
.
1 + 22 + 32 + 2
dt
4
2
r + a2
The parameters are given by
= r 2 r 2 + a2 ,
2
= r 2 + a 2 mr 2 .
(7)
Taking the second step of the squashing transformation we have described in Section 2.1,
with
k(r) =
2 + a 2 )2 mr 2
(r
2 r 2 )2
(r
(8)
2
m
2 2 r 2 + a2 2
a
.
k dr +
k 1 + 22 + 32 + 2
dt
4
2
r + a2
(9)
It is straightforward to check that metric (9) satisfies the vacuum Einstein equation
1
R Rg = 0.
2
(10)
89
We set the order of parameters as 0 < r r+ < r , with an outer horizon at r = r+ and an
inner horizon at r = r , and restrict r within the range 0 < r < r , then metric (9) is geodesic
complete and has no naked singularity, as we will establish in Section 3. In particular, if we take
the limit r and at the same time concentrate on a region with a finite value of rrr , the
KerrIM metric (9) will reduce to a fivedimensional Kerr metric. This is similar to the IM black
hole, which reduces to a fivedimensional ReissnerNordstrm black hole in the same limit [1].
2.3. Asymptotic structure
In order to see the asymptotic structure, we introduce a new radial coordinate as
= 0
r2
2 r2
r
(11)
with
k0 2
r + a 2 ,
4
2
(r 2 + a 2 )2 mr
k0 = k(r = 0) =
4
r
02 =
(12)
(13)
+ 0
+
m
a2
2 +a 2 0
r
K 2,
2 + a2
r
r 2 + a2
W2 = 2
4K
m
,
4V
( + 0 )2
,
R2 =
K2
2
02
r2
.
U= 2 2
r2
r + a
W 2 4 +
V
0
(14)
12
+ 22
2
2 + a2
r
m
a
2
+
3 + 2
dt 3 .
4
2
r + a 2
(15)
At first sight, the crossterm between dt and 3 appears to imply a pathology of the space
time. Nevertheless, if we change the coordinates as
=
2
(r
2ma
t,
+ a 2 )2 + ma 2
90
t =
2 + a 2 )2 mr 2
(r
t
2 + a 2 )2 + ma 2
(r
(16)
and take the notation 3 = d + cos d, the asymptotic spacetime is actually all right
(r 2 + a 2 )2 + ma 2 2
3 .
ds 2 = d t 2 + d 2 + 2 12 + 22 + 2
4(r + a 2 )
(17)
This asymptotic topology is the same as that of the IM spacetime [1]: a twisted S 1 bundle over a
fourdimensional Minkowski spacetime. We have expected it because the asymptotic structure
should not be affected by angular momenta or any other charges of black holes. From the same
point of view, we predict that the asymptotic geometry ought to be a twisted S 1 bundle over a
fourdimensional antide Sitter spacetime for a AdSIM black hole if it really exists.
At infinity, the size of the compactified dimension is controlled by r , m and a together
rather than r alone, implied by metric (17). One should not be bothered by this illusion of
2 for
parametrization. A better parameter may be obtained if we trade r
2
r
=
2 + a 2 )2 + ma 2
(r
2 + a2
r
(18)
in which the size of the compactified dimension is apparently decoupled from m and a, but that
will only make our formulas more scattered. In our expressions, for brevity, we will always use
. But one should keep in mind that the geometric interpretation is
the parameter r instead of r
clearer for r than for r .
We stress that the coordinate transformation (11) is valid only for a finite r value, i.e.,
r < . In the limit r , one merely finds 0. Thus we will carry out most computations based on the metric in the form (9) instead of (13). But the calculations of mass and
angular momenta are inconvenient from (9), so we will in Section 5 resort to the coordinate
introduced in (11). We believe the calculations should always be right for a finite r , given that
the boundary counterterm method and the generalized AbbottDeser method work.
3. Absence of naked singularities
3.1. Ingoing Eddington coordinates
The metric (9) beaks down at = 0, i.e., r = r . These singularities can be removed by a
coordinate transformation
k(r 2 + a 2 )2
dr,
2ak(r 2 + a 2 )
dr.
d = d +
m
r 2 + a2 2
a 2
ds 2 = dv 2 +
k 1 + 22 + 3 2 + 2
dv
4
2 3
r + a2
a
+ 2k dv 3 dr
2
(19)
(20)
91
in which we make use of a notation 3 = d + cos . The new coordinates are nothing but the
ingoing Eddington coordinates. The metric is regular now at r = r .
In fact, r = r are where Killing horizons sit. We will only discuss outer horizon r = r+ in
this article. By exchanging r+ r one immediately gets the results for inner horizon r = r .
From (20) we can write down the Killing vector of the event horizon at r = r+
= 0 (v + H ),
2a
,
H = 2
r+ + a 2
(21)
(22)
dx () dx () 1/2
d g x()
.
d
d
(23)
= mpart
r 2 + a2 2 r 2 + a2
dt 1 k 2 r 2
k
k sin2 2
4
4
2 1/2
+ a2
m
a a
2
( + cos ) 2
1 cos
4
2
2
r + a2
r
dr
r 2 + a2
= mpart
1 k 2 r 2
k sin2 2
r
4
r2
2 1/2
r 2 + a2
m
a
a
2
cos
( + cos )
4
2
2
r 2 + a2
(24)
in which a dot denotes a derivative with respect to t , and we have assumed = 0. As will be
shown later, this assumption is consistent with geodesic equations.
92
dx () dx () 1/2
u = g x()
d
d
2
2
r +a
r 2 + a2
2
= 1 k 2 r 2
k sin2 2
( + cos )
4
4
2 1/2
m
a a
2
.
1 cos
2
2
r + a2
(25)
mpart 2 2 r 2 + a 2
r 2 + a2
2
=
k r +
k sin2 2 +
( + cos )
u
4
4
m
a
a
a a
cos
+ 2
cos
+ mpart u
1
2
2
2
2
r + a2
mpart
m
a
a
cos .
=
1 2
u
2
2
r + a2
H =E=
(26)
For a stationary spacetime the energy E is a constant, and the first integrals of and are
found to be
P =
mpart r 2 + a 2
r 2 + a2
k sin2 +
cos ( + cos )
u
4
4
a a
m a
,
cos
cos
2
2
2
r + a2 2
2
2
L mpart r + a
m a
a a
=
P =
( + cos ) 2
1 cos .
u
4
2
2
r + a2 2
=
(27)
We will assume P = 0, P = 0 and show the consistency a little later. Using these assumptions
to solve (27), we get a clean result
= 0,
=
(r 2
2ma
.
+ a 2 )2 + ma 2
(28)
(29)
93
r
dx () dx () 1/2
dr
d =
d g x()
=
u
d
d
r
r
2
mpart
(r 2 + a 2 )2 mr
dr 2
=
2
2
E
(r r )
0
E 2 r 2 (r 2 + a 2 )
.
2
2
2
(E mpart )(r + a 2 )2 + m(m2part r 2 + E 2 a 2 )
(30)
For E 2 > m2part this integration is divergent when and only when r approaches r , that is, for a
particle with enough energy, their geodesics can be extended to all values of . Accordingly we
can expect the KerrIM spacetime is geodesic complete, and it is clear now that r r indeed
corresponds to spatial infinity.
In the above, we have assumed = 0, P = 0 and P = 0. To be rigorous, we now demonstrate that these assumptions are consistent with geodesic equations. Christoffel connections are
easy to derive from metric (9), and we find the expected vanishing components
=
=0
(, = t, r, ).
(31)
4
4(r 2 + a 2 )
= 0 (t + H )
(32)
ds 2 = g 00 d t 2 +
and
2 + a 2 )2 + ma 2
(r 2 + a 2 )2 mr 2
(r
,
2 + a 2 )2 mr 2
(r 2 + a 2 )2 + ma 2 (r
2
(r + a 2 )2 + ma 2
2ma
2ma
2
,
=
2
2
2
2
2
2
2
2 + a 2 )2 mr 2
(r + a ) + ma
(r + a ) + ma
(r
2 + a 2 )2 mr 2
(r
,
0 = 0
2 + a 2 )2 + ma 2
(r
2
(r + a 2 )2 + ma 2
2ma
.
H = H 2
2 + a 2 )2 mr 2
(r + a 2 )2 + ma 2
(r
g 00 =
(33)
94
This form of metric has the merit that g 00 1 and 0 at spatial infinity r r .
Correspondingly the twodimensional Euclidean Rindler spacetime is
dsE2 =
2 2
k dr + g 00 d 2 .
(34)
= 2
2
2
(r + a 2 )2 mr
r+ + a 2 k(r+ )
2 r2
2 + r r )2 + r r (r + r )2
r
(r+ r ) (r
+
+ +
+
=
.
2 r2 )
2 r2
r+ (r+ + r )(r
r
2
,
with
(35)
The same result can also be derived from (9) and (22), using r=r = , and
choosing a normalization to ensure g 00 t t 1 as r r , that is
2 + a 2 )2 + ma 2
(r
.
0 =
(36)
2 + a 2 )2 mr 2
(r
Having the surface gravity on the outer horizon, one immediately obtains the temperature of
the black hole
2 r2
2 + r r )2 + r r (r + r )2
r
(r+ r ) (r
+
+ +
+
T=
(37)
.
=
2 r2 )
2 r2
2
2r+ (r+ + r )(r
r
.
H =
(38)
2 r2 )
2 + r r )2 + r r (r + r )2
r+ + r
r+ (r
(r
+
+ +
It is not surprising that we should take into account asymptotic properties of metric at infinity when computing thermodynamic quantities on the event horizon. The key point is that all
thermodynamic quantities are seen from infinity.
In [13] it has been found that the area formula of entropy still survives in spite of a deformation
of the horizon. Hereby we calculate the entropy of KerrIM black hole according to the area
formula. Starting with (9), (13), (20) or (32), one easily obtains a unique consequence
A = 2 2 r+ (r+ + r )2
2 r2
r
2 r2
r
+
S=
2 r2
A 2 r+ (r+ + r )2 r
.
=
2 r2
4
2
r
+
95
(39)
(40)
in which U , R and K are given by (14). g 00 and are the same as those in (33), also taking the
form
2 + a 2 )2 m(r 2 + a 2 )K 2 + ma 2 K 4
(r
(r 2 + a 2 )2 + ma 2
2
,
2
2
2
2
4
2
(r + a ) + ma K
(r + a 2 )2 mr
2
(r + a 2 )2 + ma 2
2ma
2maK 4
2
.
=
2
2
2
2
4
2
2
2
2 + a 2 )2 mr 2
(r + a ) + ma K
(r + a ) + ma
(r
g 00 =
(41)
In Abbott and Desers definition, conserved charges are associated with isometries of the
asymptotic geometry which is supposed to be the vacuum of the system [4]. In KerrIM space
time, corresponding to (40), the asymptotic geometry is described by (17).
Taking = t as the canonically normalized timelike Killing vector, and repeating the
computation done in [13], one is led to the generalized AbbottDeser mass of the KerrIM black
hole
MKerrIM =
=
2 + a 2 )2 + m(r 2 + 2a 2 )
(r
(r 2 + a 2 )2 ma 2
2 + a 2 )2 + ma 2
2 + a 2 ) (r 2 + a 2 )2 mr 2
4
(r
(r
(r 2 + r+ r )2 r+ r (r+ + r )2
2 + r r )2 + r r (r + r )2
4
(r
+
+ +
2 + 2r r + r 2 )(r 2 + 2r r + r 2 )
(r
+
+
+
.
2
2
2
2
2
(r + r+ r ) (r r+ )(r r )
(42)
This mass and the mass of IM black hole obtained in [13] will be analyzed in Section 5.4.
5.2. Angular momentum and the first law
Starting with (40), after a few calculations, the Komar integrals give the angular momenta
J = 0,
J = J =
r+ r (r+ + r )2 .
ma =
4
4
(43)
96
(44)
involves five parameters. All of them have been obtained now. To check the first law, one might
view r+ , r as variables while keep r as a constant. But that will break (44). Remember that for
, as we have explained
KerrIM spacetime the geometric parameter at infinity is not r but r
in Section 2.3. In deed, if we fix r rather than r , the first law of black hole thermodynamics
(44) is obeyed.
5.3. The counterterm method
For a spacetime with an asymptotic boundary topology R S 1 S 2 , a simple counterterm
in gravitational action is introduced by Mann and Stelea [10]. As a matter of fact, asymptotic
structures are the same for IM black holes and KaluzaKlein monopoles, the latter of which
are treated in [10]. An advantage of the counterterm method over the background subtraction
method is that it is independent of reference backgrounds hence leads to a unique result. The
stress tensor for the asymptotically flat spacetimes was proposed for the first time by Astefanesei
and Radu in [9]. For different boundary topologies there exist different counterterms, but the
form of the stress tensor is generic.
We still start with (40). After a substantial amount of calculations, we find the nonvanishing
components of stress tensor
2 + a 2 )4 + m(r 2 + a 2 )3 m2 a 2 (r 2 + 2a 2 )
(r
1
1
+
O
8GT tt =
,
2
2 + a 2 )2 + ma 2 ] (r 2 + a 2 )3 mr 2 (r 2 + a 2 ) r
r3
4[(r
2 + a2
r
cos
ma
1
t
+O 3 ,
8GT =
2
2
2
2
2
4 (r + a ) + ma r
r
2 + a2
r
1
ma
1
+
O
,
8GT t =
2 + a 2 )2 + ma 2 r 2
4 (r
r3
8GT =
2 + a 2 )4 m(r 2 5a 2 )(r 2 + a 2 )2 m2 (r 4 + 3a 2 r 2 + a 4 ) 1
(r
2 + a 2 )3 mr 2 (r 2 + a 2 )]
32[(r
r3
1
+O 4 ,
r
2 + a 2 )4 m(r 2 5a 2 )(r 2 + a 2 )2 m2 (r 4 + 3a 2 r 2 + a 4 ) 1
(r
2 + a 2 )3 mr 2 (r 2 + a 2 )]
32[(r
r3
1
+O 4 ,
r
3/2
2 + a2
1
r
1
+
O
8GT t = ma
,
2
2
2
2
2
(r + a ) + ma
r
r3
2 + a 2 )4 m(r 2 + a 2 )3 + m2 a 2 (r 2 a 2 )
cos
2(r
1
+O 3 ,
8GT =
2
2
2
2
2
2
2
3
2
2
2
r
r
4[(r + a ) + ma ] (r + a ) mr (r + a )
2 + a 2 )4 m(r 2 + a 2 )3 + m2 a 2 (r 2 a 2 )
1
2(r
1
+
O
8GT =
(45)
.
2
3
2
2
2
2
2
2
3
2
2
2
r
r
4[(r + a ) + ma ] (r + a ) mr (r + a )
8GT =
97
Associated with the Killing vector t, the conserved mass is the same as (42). Corresponding
to and respectively, the angular momenta are given by (43).
5.4. Comments and conjectures
In a recent paper [13], Cai et al. have got an expression similar to (42). In their notations, the
IM black hole mass is
MIM = MCT = MAD =
4 3r 2 r 2 + r 2 (r 2 + r 2 ))
(r
+
+
.
2 r 2 )(r 2 r 2 )
4 (r
+
(46)
Besides a background subtraction method, i.e., the generalized AbbottDeser method [7], they
also employed the counterterm method [10].
Both (42) and (46) grow as r increases, that is, as the size of the compactified dimension
increases. This is a little counterintuitive. More surprisingly, when the black hole disappears, i.e.,
in the empty limit r+ = r = 0, both masses tend to
2
r
(47)
.
4
We observe that (47) is just the mass expression for KaluzaKlein monopoles in four dimensions [10,14]. Furthermore, if we set r+ = 0 and r = 0, the metrics for both IM and KerrIM
black holes simply reduce to the metric for KaluzaKlein monopoles [14], whose geometry is
perfectly regular in five dimensions. This indicates the physical meaning of (42) and (46): they
express the masses of the KaluzaKlein black holes in the sight of four dimensions. The Kaluza
Klein charges [14] account for their counterintuitive behaviors. In a fivedimensional point of
view, there are no KaluzaKlein charges, and (47) is the contribution of energy from background
geometry. A further observation is that if we subtract (47) from (46), the first law of thermodynamics checked in [13] will not be affected, for they have fixed the value of r .
Strictly speaking, the expressions (42) and (46) for mass are reliable only in the case of a
finite r value, as we remarked in the end of Section 2.3. Yet we want to extrapolate the result to
the limit r r . Then we propose the following conjectures for an arbitrary value of positive
r .
Mmon =
(1) Formula (46) represents the mass of IM black holes observed in four dimensions. Subtracting the monopole contribution (47), we conjecture the mass welldefined in five dimensions
MIM =
4 3r 2 r 2 + r 2 (r 2 + r 2 ))
(r
r 2
+
+
.
4
2 r 2 )(r 2 r 2 )
4 (r
+
(48)
(2) Similarly, the formula (42) represents the mass of KerrIM black holes observed in four
dimensions. Subtracting the monopole contribution (47), we conjecture the mass welldefined in
five dimensions
2
2 + a 2 )2 + m(r 2 + 2a 2 )
(r
(r + a 2 )2 ma 2
MKerrIM =
2
2
2
2
2
2
2
2
4
(r + a ) (r + a 2 )2 mr
(r + a ) + ma
(r 2 + a 2 )2 + ma 2
2
(49)
.
r + a 2
98
instead
Here we have considered that the size of the compactified dimension is controlled by r
of r .
Obviously, the masses MIM and MKerrIM also satisfy the first law of black hole thermodynamics in five dimensions. To inspect and improve our conjectures, one may check the law in four
variable. It
dimensions with MIM and MKerrIM , by including monopole charges and allowing r
3(r 2 +r 2 )
6. Summary
In this paper, we have extended the recently found static KaluzaKlein black holes with
squashed horizons [1] to rotating counterparts. Our investigations are restricted to black holes
with two equal angular momenta. These counterparts are geodesic complete and free of naked
singularities. The thermodynamic quantities have been calculated and the first law has been
checked. The angular momenta are computed using Komar integrals, then reproduced with the
counterterm method. We calculated the mass by both background subtraction and counterterm
method, and found a counterintuitive behavior as the compactified dimension expands. This behavior is also hidden in the mass of static black holes. We attributed it to KaluzaKlein charges
and gave some comments on it.
Acknowledgements
We would like to thank LiMing Cao, Miao Li, JianXin Lu, Wei Song and Yushu Song
for useful discussions and kind help. We are especially grateful to RongGen Cai for a careful
reading of the manuscript and many helpful discussions.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
99
Abstract
We consider the decoupling of neutrinos in the early Universe in presence of nonstandard neutral current neutrinoelectron interactions (NSI). We first discuss a semianalytical approach to solve the relevant
kinetic equations and then present the results of fully numerical and momentumdependent calculations,
including flavor neutrino oscillations. We present our results in terms of both the effective number of neutrino species (Neff ) and the impact on the abundance of 4 He produced during big bang nucleosynthesis.
We find that the presence of neutrinoelectron NSI may enhance the entropy transfer from electron
positron pairs into neutrinos instead of photons, up to a value of Neff 3.12 for NSI parameters within
the ranges allowed by present laboratory data, which is almost three times the effect that appears for standard weak interactions. Thus nonstandard neutrinoelectron interactions do not essentially modify the
density of relic neutrinos nor the bounds on neutrino properties from cosmological observables, such as
their mass.
2006 Elsevier B.V. All rights reserved.
Keywords: Early Universe; Neutrinos; Nonequilibrium kinetics
* Corresponding author.
101
1. Introduction
In the early Universe, neutrinos were kept in thermal contact with the electromagnetic primordial plasma by rapid weak interactions with electrons and positrons. When the temperature
dropped below a few MeV, these weak processes became ineffective and the process of neutrino
decoupling took place, while shortly after the e pairs began to annihilate almost entirely into
photons thus producing a difference between the temperatures of the relic photons and neutrinos.
This difference can be easily calculated if we assume that neutrinos were completely decoupled
when the e pairs transferred their entropy to photons, leading to the wellknown temperature
ratio T /T = (11/4)1/3 1.40102. Indeed, this simplified picture should be improved since
some relic interactions between e and neutrinos exist all along the e annihilation stage, leading to a slightly smaller increase of the comoving photon temperature and to small distortions (at
the percent level) of the neutrino momentum distributions.
Presently, there exist compelling evidences for flavor neutrino oscillations from a variety of
experimental data on solar, atmospheric, reactor and accelerator neutrinos (see e.g. [1,2]). These
results are well understood by assuming that neutrinos have masses and mix, which in turn seems
to point out the necessity of some new physics beyond the Standard Model (SM) of fundamental
interactions. Interestingly, nonzero neutrino masses usually come with nonstandard interactions
(NSI) that might violate leptonic flavor and/or break weak universality. Recent analyses [36]
have considered the neutral current NSI in a phenomenological way, showing that they can be
bound using measurements of neutrinoelectron scattering, as well as data from LEP and from
related charged lepton processes.
The aim of this paper is to study the neutrino decoupling process in presence of additional
interactions between neutrinos and electrons, a possibility already noted in [3,5]. In this case,
neutrinos could be kept in longer contact with e and thus share a larger amount of the total entropy transfer than in the SM. Actually, if the nonstandard neutrinoelectron interactions
were large enough, the neutrino momentum distribution would be significantly different from
the standard case. In turn, this would modify the final yield of light nuclei during the epoch of
big bang nucleosynthesis (BBN), as well as the radiation content of the Universe, affecting the
anisotropies of the cosmic microwave background (CMB) and the power spectrum of large scale
structures (LSS). Our goal is to calculate how the decoupling is modified taking into account
NSI with couplings which are still allowed by present laboratory data, and to discuss the possibility that cosmological observations can be used as a complementary way to bound these exotic
scenarios.
The paper is organized as follows. We begin in Section 2 by describing the formalism adopted
for the nonstandard electronneutrino interactions and summarize the current bounds from a
variety of experimental data. We then consider the process of relic neutrino decoupling in the
presence of nonstandard electronneutrino interactions, giving first an estimate by using a semianalytical approach in Section 3. Finally, in Section 4 we report the results of the full momentumdependent numerical calculations for the neutrino spectra and the effect on the primordial 4 He
yield and other cosmological observables. We present our conclusions in Section 5.
2. Nonstandard neutrinoelectron interactions
The longstanding evidence of flavor change in atmospheric and solar neutrino experiments
represents a strong indication of some new physics beyond the SM of fundamental interactions.
For several years these experimental results have been interpreted in terms of neutrino masses
102
(flavor oscillations) or by introducing new neutrino interactions. Recently, reactor and accelerator data have confirmed that neutrino masses are indeed the main mechanism explaining the
atmospheric and solar anomalies, while the role of NSI can be only subleading. In any case,
most extended particle physics models that account for neutrino masses naturally lead to new
NSI, whose value strongly depends on the model. For instance, NSI may arise from the structure
of the charged and neutral current weak interactions in seesawtype extended models [7].
In the present analysis we will follow [3,5] and assume that new physics induces NSI only
through the four fermion operators ( )(ff ), where f is a charged lepton or quark, but not
new charged lepton physics at tree level. In particular, since we are interested in the decoupling
process of relic neutrinos, we consider only the NSI related to electrons which, together with the
standard weak interactions, are described by the effective Lagrangian
LNSI
Leff = LSM +
(1)
,
LSM = 2 2GF e Le (e
Le) +
P e) ,
gP L (e
(2)
P ,
P
LNSI = 2 2GF
L (e
P e)
(3)
for energies much smaller than the Z boson mass, as in our case, with GF the Fermi constant and
P = L, R = (1 5 )/2 the chiral projectors. Greek indices label lepton flavors (, = e, , )
and the Z couplings are gL = 12 + sin2 W and gR = sin2 W .
P can induce a breaking of lepton universality ( = ) or rather a
The NSI parameters
flavorchanging contribution ( = ). Their values can be constrained by a variety of laboratory
experiments, as discussed in [3,5,6]. In what follows we summarize the present bounds and refer
the reader to the analyses mentioned above for all details. When considering the bounds on the
P parameters, it is important to notice that they are usually obtained taking only oneatatime,
L,R
). This implies that the derived constraints
or at most combining two of them (such as the pair ee
are expected to be weaker but more robust if many NSI parameters are simultaneously included,
since cancellations may occur.
A remark is in order. Since isospinchanging weak interactions converting neutrons into protons and vice versa are important for the BBN yields, one might wonder if it is legitimate
to neglect the neutrino NSI with quarks. We argue that this approximation is well justified
since, while relic neutrino distributions affect n p processes in a crucial way, vice versa is
not true. The highly suppressed density of baryons with respect to electromagnetic particles
(nb /n 6 1010 ) allows one to neglect completely neutrino scattering on quarks for calculating the neutrino momentum spectrum. In addition, what contributes to the n p rates are
neutrino chargedcurrent interactions, where possible nonstandard terms are severely bounded
by the accurate agreement of SM calculations for several processes with the data (e.g. leptonic
and hadronic weak decays). For baryon thermalization, neutral current neutrino reactions are
completely negligible with respect to electromagnetic interactions with e and photons. Also
note that any tiny nonstandard effect in the weak rates is effectively taken into account by our
prescription of rescaling the theoretically predicted neutron lifetime to the experimentally measured value (see [8] for details).
103
3
2 1 2
L + R
+
(4)
e
3 e
=e
G2F me E
.
(5)
In [5] this result was compared with the SM prediction = 1.0967G2F me E / including
electroweak corrections and best fit value for sin2 W to obtain the following allowed regions
around the SM value, assuming only one operator at a time
L
R
0.07 (0.14) < ee
< 0.11 (0.16),
0.99 (1.14) < ee
< 0.53 (0.67),
R
L
< 0.7 (0.9)
< 0.4 (0.5),
e
e
(6)
(7)
(8)
104
When leaving the corresponding flavorchanging NSI parameters free, the bounds are relaxed
R
to the numbers in square parentheses: note the significant reduction especially for negative ee
values.
We also have data on e scattering from the CHARM II collaboration [14]. The obtained
results for the vector and axial vector e couplings can be translated into the following 90%
CL bounds on NSI parameters (see e.g. [5])
L
< 0.03,
0.025 <
P
< 0.1,
R
0.027 <
< 0.03,
(9)
(10)
which imply that the e interactions must be very close to the SM predictions.
LEP data on e+ e
As pointed out in [3], neutrino NSI can be also constrained by measuring the e+ e
cross section. This is actually the only way to get bounds on the L,R
parameters from laboratory
data. The approximate limits for these parameters can be extracted from Fig. 3 of [4],
0.7 L 0.5,
0.5 R 0.6
(11)
at 99% CL when the other parameter is left free. For the e e NSI the bounds are comparable to
those obtained from neutrinoelectron scattering.
2.2. One loop effects
The nonstandard neutrinoelectron interactions also lead to corrections at the oneloop level
to processes such as the decays of the electroweak gauge bosons or lepton flavor violating decays
of charged leptons. These corrections arise from effective nonrenormalizable interactions and
in principle their computation requires the knowledge of the complete theory leading to these
105
effective terms in the lowenergy regime. Nevertheless for mW , where is the energy scale
setting the limit of validity of the effective theory, the leading term is independent of the specific
theory and has a logarithmic behavior, ln(/mW ) [5]. Using conservatively ln(/mW ) 1, one
gets the following 90% CL bounds from the decay rates of the electroweak gauge bosons
L,R
0.5
(12)
P as shown in Fig. 1;
restrict the ee
P
force the to be very small and negligible for the purposes of the present paper;
allow large P , almost of order unity, see (11);
P , (7), moderate P , (10),
as far as flavor changing parameters are concerned, allow large e
P
and very tiny e , the latter being too small to have any impact on neutrino decoupling
dynamics.
Again it is useful to remind the reader that the bounds on NSI parameters are relaxed when more
than one parameter is allowed to vary simultaneously, as shown in [6].
As noted in [5], the leptonic measurement of sin2 W at a neutrino factory will be sensitive
to values of the e, e NSI parameters of O(103 ). Additional information on the NSI can be
extracted from a consistency comparison of the results of solar neutrino experiments with those
of an experiment detecting reactor neutrinos such as KamLAND. The former depend on the NSI
parameters through the matter potential felt by solar neutrinos and/or via the neutrino neutral
current detection. Instead, KamLAND data is essentially unaffected by the NSI. Ref. [5] found
that the combination of SNO and KamLAND data1 will provide bounds on the P comparable
but slightly better to those in Eqs. (11) or (12). However, it has been noted in several works [15
19] that while NSI are expected to play a subdominant role on the oscillations of atmospheric
and solar neutrinos, there exist degenerate directions in the parameter space where large NSI are
still allowed. Future data from short and medium baseline neutrino beams will help to resolve
these degeneracies [20].
3. Neutrino decoupling in presence of NSI
3.1. Delayed neutrino decoupling in the instantaneous limit
In the early Universe, thermally produced neutrinos were in equilibrium with other particles down to temperatures of few MeV, when weak interactions became ineffective and neutrino
1 The Borexino detector was considered in [4].
106
decoupled from the plasma. As a first approximation, the thermal FermiDirac momentum spectrum is preserved after the freezeout of weak interactions, since neutrinos decoupled when
ultrarelativistic and both neutrino momenta and temperature redshift identically with the Universe expansion. A neutrino chemical potential would exist in the presence of a neutrino
antineutrino asymmetry, but it was shown in [21] that the stringent BBN bounds on e (for an
updated limit see e.g. [22]) apply to all flavors, since neutrino oscillations lead to flavor equilibrium before BBN. Thus we shall ignore a relic neutrino asymmetry in the following.
It proves useful to define the following dimensionless variables instead of time, momenta and
photon temperature
x mR,
y pR,
z T R,
(14)
where m is an arbitrary mass scale which we choose as the electron mass me and R is the Universe scale factor. The function R is normalized without loss of generality so that R(t) 1/T
at large temperatures, T being the common temperature of the particles in equilibrium far from
any entropytransfer process. With this choice, R 1 can be identified with the temperature of
neutrinos in the limit of instantaneous decoupling. After decoupling, neutrinos enter cosmological observables mainly via their energy density. Since they also remain relativistic for most of
the cosmological evolutiondown to a temperature of the order of their subeV mass scaleit
is customary to parameterize their contribution to the radiation energy density R in terms of the
effective number of neutrinos Neff [23,24]
4
3
+X 0
z0
Neff
= 3+
,
0
z
(15)
where z0 (x) describes the photon to neutrino temperature ratio in the limit of instantaneous neutrino decoupling. For x 1 the e annihilation phase is over, and z0 (11/4)1/3 1.4010.
The energy densities 0 and 0 refer respectively to the photon plasma and to a single neutrino
species in the limit of instantaneous decoupling, while is the actual energy content of the
photon plasma and +X the total energy content of weakly interacting particles (including possible exotic contributions). The second equality in Eq. (15) follows when only the three active
neutrinos contribute to +X ; eventually, the actual photon temperature evolution accounts for
the second factor in Eq. (15) and the possible energydensity distortion in the th neutrino flavor
is given by ( 0 )/0 . Note that from Eq. (15) it follows
7 4
R = 1 + z0 Neff ,
8
(16)
which, replacing z0 with (11/4)4/3 , is often used in the literature to define Neff in the asymptotic
limit x 1. Clearly, well after e annihilation, three thermally distributed neutrinos correspond
to Neff = 3 in the instantaneous decoupling limit.
In order to estimate analytically how large would be the impact of NSI on the decoupling
mechanism, let us discuss a simple toymodel, which we shall later compare with the results
obtained solving the relevant kinetic equations. Consider the general case where 3 N neutrinos
interact via standard weak processes, while the remaining N have extra contributions from NSI
which enhance the interaction rates over e . In the instantaneous decoupling limit for both the
neutrino species, the plasma can be described by a single additional parameter, the nonstandard
107
Fig. 2. Contribution to the radiation energy density parameterized in terms of Neff , when 1, 2 or 3 neutrino species are
coupled to the electromagnetic plasma until a given value of the decoupling temperature Td .
z0 4
z0 4
Neff = (3 N ) + N w 4
= 3 + N w4 1
,
z
z
(17)
where the second step reflects the fact that, for a single NSI neutrino species, the distortion is
given by = (w 4 1). If we denote with xd Td R the epoch of the standard neutrino decoupling, and parameterize with xd Td R the decoupling temperature of the nonstandard species,
we can easily calculate the functions w(x) and z(x) from the conservation of entropy per comoving volume,
w(x) = z(x) = 1,
x < xd ,
,
xd x xd ,
w(x) = z(x) = N s (x)+s (x)+se (x)
N s (x )+s (x )+s (x )
1/3
,
w(x) = wd N s (xd )+s (xd )+se (xd )
d
d
e
d
x > xd ,
s (x)+se (x)
where s , s and se are respectively the specific entropy of one species, of the photons and
of the e .
In Fig. 2 we show the results for Neff in the instantaneous decoupling limit as function of the
decoupling temperature Td and for various choices of N . As expected, Neff grows if neutrinos
decouple at a smaller temperature and shows two regimes. For very small NSI contribution to
the neutrino interaction rates we have wd 1 and Neff 3 since neutrinos decouple long before
any e annihilation. On the other hand, for N neutrinos tightly coupled to the electromagnetic
component down to very small temperatures when all e pairs have already annihilated,
22 + 7N 1/3
z
(18)
,
w z.
8 + 7N
108
Thus, in the tightly coupled limit, one finds Neff {7.0, 9.7, 11.5} for N = {1, 2, 3} respectively, as shown in Fig. 2. Note that one obtains a contribution to the radiation energy density of
order Neff 4 if neutrinos are kept in equilibrium down to temperatures of 0.20.3 MeV.
3.2. Decoupling temperature with NSI
In order to dynamically predict the decoupling temperature Td when NSI are present, we have
to solve the relevant kinetic equations taking into account the new couplings . For a first analytical estimate, we can compare the interaction rate with e with the expansion rate of the
Universe given by the Hubble parameter H . Since H T 2 , and (1 + )2 T 5 , the decoupling
temperature at which H = decreases as Td (1 + )2/3 . As we have argued in the previous
section, in order to produce changes of O(1) in Neff the decoupling temperature should be lowered down to Td 0.20.3 MeV, i.e. should be one order of magnitude smaller than for ordinary
neutrinos. This implies 20, which would largely exceed present laboratory bounds. We thus
expect at most changes in Neff of O(0.010.1). To improve further our treatment, we can use an
approximate solution of the kinetic equation for the neutrino distribution function f (x, y),
Hx
f (x, y)
= Icoll ,
x
(19)
where Icoll is the collision term that includes all relevant neutrino interaction processes. Following [25], we shall keep only the direct reaction term in the collision integral using the appropriate
matrix elements for neutrino elastic scattering and the inverse annihilation e+ e in the
relativistic limit (me 0). Assuming that all particles that interact are close to equilibrium, the
collision integral can be estimated in the Boltzmann approximation leading to
Hx
(20)
with gL2 (1 + gL )2 for the case of e . This equation can be easily integrated in x, leading to
the values Td (e ) 1.8 MeV and Td (, ) 3.1 MeV for an average momentum of y = 3.15.
This exercise can be repeated in the presence of neutrinoelectron NSI in order to find their
L,R
or L,R
influence over the decoupling of neutrinos, adding ee
to the couplings gL and gR . The
L,R
calculated decoupling temperature is shown in Fig. 3 for the case of nonzero ee
and L,R
.
In general, a significant increase of the NSI parameters from the SM prediction leads to a larger
interaction of neutrinos with e and thus to a lower decoupling temperature. However, for a
region close to the pair of values (L , R ) that minimize the interaction by accidental cancellation
with SM couplings, the decoupling temperature is significantly raised. Finally, one can relate
the value of the decoupling temperature in presence of NSI with an estimate of the change in
Neff following the instantaneous decoupling approximation previously described. The results are
shown in Fig. 4. As already noticed, large modifications of Neff require that the neutrinoelectron
interactions are much larger than ordinary weak processes and are thus excluded by laboratory
bounds.
Nonetheless, this study suggests that additional distortions in the neutrino spectra comparable
to or larger than the ones predicted in the SM are possible, and it is still interesting to assess
exactly how large these effects might be. However, the treatment followed till now assumes a
series of approximations (such as me 0, no QED corrections to the plasma, no effect of the
active neutrino oscillations, etc.) that in the next section we shall check using a fully numerical
and momentumdependent calculation.
109
Fig. 3. Contours of equal neutrino decoupling temperature in MeV for different values of the parameters characterizing
the e e and the e diagonal NSI.
Fig. 4. Contours of equal Neff for different values of the NSI parameters as found in the instantaneous decoupling
approximation.
110
differential kinetic equations that are difficult to solve numerically. In the early 1990s several
works [2729] performed momentumdependent calculations assuming some approximations,
such as Boltzmann statistics for neutrinos, while the full numerical computation was later carried out in [3033]. Finally, a further refinement involves the inclusion of finite temperature QED
corrections to the electromagnetic plasma, [34,35], while the last works also include the effect of
flavor neutrino oscillations [36,37].
Here we will follow our previous work [37]. In particular, we will describe the neutrino ensemble in the usual way by generalized occupation numbers, i.e. by 3 3 density matrices for
neutrinos and antineutrinos as described in [38,39], with elements where , = e, , . The
diagonal elements correspond to the usual occupation numbers of the different flavors, while the
offdiagonal terms are nonzero in the presence of neutrino mixing or flavorchanging NSI. The
equations of motion for the density matrices relevant for our situation of interest in an expanding
Universe are [38]
i(t Hpp )p =
M 2 8 2GF p
E
,
p + C[p ],
2p
3m2W
(21)
where mW is the W boson mass. We use the notation p = (p, t) and [, ] denotes the commutator. The vacuum oscillation term is proportional to M 2 , the masssquared matrix in the flavor
basis that is related to the diagonal one in the mass basis diag(m21 , m22 , m23 ) via the neutrino mixing matrix U , which in turn depends on the neutrino mixing angles 12 , 23 and 13 (we assume
CP conservation). From a global analysis of experimental data on flavor neutrino oscillations, the
values of mixing parameters can be extracted. As a reference, we take the bestfit values from [1]
m221
m231
,
105 eV2 103 eV
, s2 , s2 , s2
2 12 23 13
(22)
where s2ij = sin2 ij for i, j = 1, 2, 3. In Ref. [37] we have shown that the results are not essentially modified for other values within the allowed regions, in particular for nonzero 13 close to
the experimental upper bound. Since we assume maximal 23 and zero 13 the spectra of and
will be the same after decoupling.
The effects of the medium in Eq. (21) are contained in the collision term C[] and in a refractive term that corresponds in Eq. (21) to the term proportional to the diagonal matrix E,
that represents the energy densities of charged leptons. See [37] for a description of both terms.
The kinetic equations (21) should be numerically solved along with the equation governing the
evolution of the electromagnetic temperature during the process of e annihilations. This can be
found from the continuity equation for the total energy density ,
d
= 3H ( + P ),
dt
(23)
with P the pressure, which can be cast into a first order differential equation for z as function of
x, see e.g. [35].
We have generalized the results of [37] calculating the full evolution of neutrino during decoupling in presence of nonstandard interactions with electrons, which modify the equations in
L,R L,R
L,R
, or e
,
two different ways. First, the couplings are modified inserting the parameters ee
which implies a variation on the collisional term in Eq. (21). Second, neutrino refraction in the
111
Fig. 5. Evolution of the distortion of the e and , spectra for a particular comoving momentum (y = 10) with standard
weak interactions (solid line) and with e NSI (dotted line). The line labeled with T corresponds to the distribution
of a neutrino in full thermal contact with the electromagnetic plasma.
medium is modified by the NSI, which we took into account in the refractive term in Eq. (21)
1 + ee 0 e
8 2GF p
e
Hmatt =
(24)
0
0 0 ,
3m2W
0
L + R ,
where e is the energy density of the background electrons and positrons and =
since matter effects are sensitive only to the vector component of the interaction. The above
equation includes only those NSI parameters that are not constrained to be smaller than O(0.1)
(see Section 2), and thus represents the leadingorder correction to the refractive index.
We have numerically solved the kinetic equations Eq. (21) using a discretization in a grid
of 100 dimensionless neutrino momenta in the range yi [0.02, 20]. We start to compute the
evolution of the system at a value xin , when weak interactions were effective enough to keep
neutrinos in equilibrium with the electromagnetic plasma but flavor oscillations are suppressed
by medium effects. The system of equations is then solved from xin until a value of x when both
the neutrino distortions and the comoving photon temperature z are frozen, approximately at
xfin 35. In some cases, we made the calculation for different initial conditions and concluded
that the results are stable within per cent accuracy unless we start the numerical evaluation at
xin 0.2. For other technical details, we refer the reader to [37].
112
and in presence of large NSI between tau neutrinos and e , corresponding to L = R = 2. The
behavior of f has been previously described e.g. in [31,33,37]. At large temperatures or x 0.2,
neutrinos are in good thermal contact with e and their distributions only change keeping an
equilibrium shape with the photon temperature [exp(y/z(x)) + 1]1 (the T line in the figure).
In the intermediate region 0.2 x 4, the standard weak interactions become less effective
in a momentumdependent way, leading to distortions in the neutrino spectra which are larger
for e s than for the other flavors. Finally, at larger values of x neutrino decoupling is complete
and the distortions reach their asymptotic values. In the case of large e NSI, neutrinos are
kept in thermal contact with e for a longer time, leading to larger distortions for all neutrino
L,R
L,R = 0, the electron and muon neutrinos also
=
flavors, in particular for , (although ee
acquire significant distortions due to the effect of flavor oscillations). For the particular neutrino
momentum in Fig. 5, if L = R = 2 the final values of the distribution are 9.8% for the e s and
14.3% for the , s larger than in the limit of neutrino decoupling before any e annihilation.
For comparison, the corresponding values for standard weak interactions are 4.4% for the e s
and 2% for the , s.
As in Ref. [37], we will summarize the results in terms of the following frozen values: dimensionless photon temperature zfin , the fractional changes in the neutrino energy densities
and the asymptotic effective number of neutrinos Neff as defined in Eq. (15). We have considered
four main sets of NSI parameters:
L,R
(1) Only ee
: our results are summarized in Table 1. We have calculated various combinations of the two e e NSI parameters, that correspond to the points shown in Fig. 1. For
L = R = 4 (already
illustrative purposes, we considered the case with extremely large NSI ee
ee
excluded by laboratory bounds), where the distortions in the distribution functions of neutrinos
lead to a significant change in Neff .
(2) Only L,R
: our results are summarized in Table 2 for values of both parameters up to order
unity in absolute value. Again, we also show one case with larger NSI L = R = 2, disfavored
by terrestrial experiments.
L,R
and L,R
(3) Combinations of ee
: our results are summarized in Table 3. Here we varied
L,R
the values of the ee pair, while the e NSI parameters were fixed to a value close to the
current upper bound from laboratory experiments, L = R = 0.5.
L,R
(4) Cases where the flavorchanging NSI parameters e
are nonzero, that for simplicity
we consider real. We consider both cases with vanishing diagonal NSI and with all possible NSI
parameters with the largest allowed values according to Eqs. (8).
The last column (Yp ) in Tables 14 reports the variation in the mass yield of 4 He synthesized
during BBN, which predicts typically Yp 0.248 for the baryon density b = 0.0223 0.0008
favored by CMB anisotropy studies [40]. Neutrinos affect the outcome of BBN in a flavor independent way through their contribution to the radiation energy density (change in Neff ). In
addition, the role played by electron neutrinos and antineutrinos in the weak reactions that
convert neutrons and protons is very important for fixing the primordial production of 4 He,
which is thus the nucleus mostly sensitive to nonstandard physics in the neutrino sector. To
predict quantitatively the change Yp induced by NSI, we take exactly into account the modified thermodynamical quantities, while treating perturbatively the change in the weak rates. This
is justified given the relatively small effects we are considering. We closely follow the treatment
given in [37], to which we refer the reader for further details.
113
R
ee
zfin
e (%)
, (%)
Neff
1.3978
0.73
0.52
3.046
2.1 104
0.1
1.0
1.5
1.5
1.0
0.07
0.5
1.0
0.1
1.0
1.5
1.0
1.3969
1.3966
1.3976
1.3969
1.3977
1.3973
1.20
1.32
0.81
1.17
1.45
0.99
0.72
0.77
0.56
0.71
0.83
0.64
3.062
3.067
3.048
3.061
3.060
3.055
2.8 104
3.0 104
2.0 104
2.7 104
2.9 104
2.6 104
4.0
1.3812
9.47
3.83
3.357
11.7 104
Yp
4.0
Yp
Table 2
Same as Table 1 but for e NSI
L
zfin
e (%)
, (%)
Neff
1.3978
0.73
0.52
3.046
2.1 104
0.5
1.0
1.0
0.5
1.0
1.0
1.3970
1.3966
1.3952
0.90
1.24
1.24
0.79
1.43
1.45
3.059
3.079
3.092
4.0 104
4.8 104
5.3 104
2.0
2.0
1.3911
2.04
2.96
3.168
10.0 104
Table 3
Same as Table 1 when both the e e and e NSI are nonzero. In all the cases we have fixed L = R = 0.5
L
ee
R
ee
0.1
1.0
1.5
1.0
0.5
1.0
1.0
0
zfin
e (%)
, (%)
Neff
Yp
1.3963
1.3960
1.3963
1.3977
1.30
1.41
1.28
0.54
0.95
0.99
0.94
0.66
3.073
3.077
3.073
3.047
3.6 104
3.7 104
3.4 104
2.6 104
Table 4
L,R
Same as Table 1 when the flavorchanging NSI parameters e
are nonzero. The last two cases correspond to the
results for all NSI parameters nonzero with values given by the bounds in Eqs. (8)
L
ee
0
0
0.12
0.61
0.12
R
ee
0
0
0.15
1.58
1.58
0
0
0.5
0.5
0.5
0
0
0.5
0.5
0.5
L
e
0.4
0.85
0.85
0.85
0.85
R
e
0.7
0.38
0.38
0.38
0.38
zfin
e (%)
, (%)
Neff
Yp
1.3960
1.3956
1.3949
1.3948
1.3937
1.32
1.46
1.68
1.69
2.21
1.03
1.16
1.38
1.41
1.66
3.077
3.085
3.098
3.100
3.120
3.9 104
4.4 104
5.0 104
5.1 104
6.0 104
The results are generally in agreement with the approximate ones predicted in Section 3 (see
in particular Fig. 4), apart for minor differences due to the improved treatment. Also, as one
L,R
enhance mostly the distortions in the electronic flavor, while large
naively expects, large ee
L,R
lead to larger distortions in the tau neutrino spectra. However, the differences between the
spectra of the three neutrino flavors are reduced by the partial reshuffling of the entropy transfer
due to the effects of neutrino oscillations.
114
In general, we find that the presence of nonzero NSI enhances the transfer of entropy from
e to neutrinos, leading to values up to Neff 3.12 when all NSI parameters are close to the
boundaries of the allowed regions discussed in Section 2, which is almost three times the departure of Neff from 3 that already exists for SM weak interactions (the first row in Table 1). We
also obtained some significant departures of Neff from the standard value, such as Neff 3.36 for
L = R = 4, but only for NSI parameters well beyond the allowed regions. On the other hand,
ee
ee
for particular choices of the diagonal NSI parameters that minimize the e interaction we find
a reduction in the final value of the distortions or Neff with respect to the SM value, as expected
from our discussion in Section 3.2.
Finally, we mentioned in Section 2 that neutrinoneutrino interactions are poorly constrained.
In the case of interactions involving lefthanded neutrinos only, they can still be twice as effective
as in the SM, as reported in [9]. Since enhanced neutrinoneutrino interactions will redistribute
the distortions more efficiently, they can modify the results described in this section. In order
to check this, we doubled the intensity of the contributions of the neutrinoneutrino interactions
to the collision integral and repeated the calculations for some of the cases considered above.
We found that the results are basically unchanged, confirming the unimportant role played by
neutrinoneutrino collisions in the process of neutrino decoupling.
5. Conclusions
The process of relic neutrino decoupling in the early Universe is sensitive to the strength of
the interactions between neutrinos and the plasma formed by electrons and positrons. If neutrinos were kept in longer thermal contact with them than in the standard case, they would share
a larger fraction of the entropy release from e annihilations. This would affect the predicted
characteristics of the cosmic background of neutrinos (CB), which in turn could modify the late
evolution of the Universe and the bounds on neutrino properties from the analysis of cosmological observables.
In this paper we have considered how the decoupling of relic neutrinos is modified in presence
of nonstandard neutral current neutrinoelectron interactions. First, we have provided a rough
estimate of the size of NSI couplings needed to affect in a relevant way the properties of the CB
and of related observables. We find that needed couplings are by far larger than the existing laboratory bounds, as also confirmed by a semianalytical solution of the relevant kinetic equations.
Thus, NSI can only play a minor role in shaping the CB. However, NSI might still contribute
with comparable or larger distortions in the neutrino spectra than predicted in the SM. In order
to quantify these deviations, we have performed fully numerical and momentumdependent calculations of the densitymatrix equations relevant for neutrino evolution in the early Universe,
including the effects of flavor neutrino oscillations.
Typically, we find that the enhancement in observables like Neff , which measures the change
in the radiation content, or the 4 He mass yield in the BBN can be up to three times larger than
the ones found in the standard case [37], for values of all NSI parameters close to the limits
placed by laboratory experiments. Nevertheless, even the variation of up to 0.2% that we found
in the 4 He abundance is yet too small (about one order of magnitude) when compared to the
observational error on Yp , which is unfortunately dominated by systematics, and as such difficult
to pin down in the near future. Instead, a value of Neff 3.12, which is still possible within the
present parameter space for NSI couplings, might even be barely detectable. Indeed, it has been
shown that the CMB satellite PLANCK will soon provide temperature and polarization data that
will probably measure the value of Neff with an uncertainty of order (Neff ) 0.2 [41,42] (for
115
a previous, more optimistic forecast see [43]), or better when combined with data from a large
galaxy redshift survey such as SDSS [44]. Future CMB missions may reach the sensitivity of
0.040.05 needed to test the standard scenario [42], and might achieve a 2 hint for Neff 3.12
and eventually test the effect of large neutrinoelectron NSI.
However, barring some extreme cases, we conclude that the prediction of Neff 3 within a
few % is quite robust even when taking into account neutrinoelectron NSI of the fourfermion
type. Thus, their existence cannot modify in a significant way the bounds on neutrino properties from cosmological observables, in particular on their masses, as recently reviewed in [45].
Since it is likely that future experiments may narrow further the allowed range of NSI couplings,
their role in a cosmological context will be even smaller, at most a subleading correction to the
standard prediction of the CB properties. Turning the argument around, however, one can conclude that a significant deviation of Neff from 3 may require major revisions of the cosmological
model, like the introduction of new relativistic relics and/or of an exotic thermal history, or both
(see e.g. [46,47]).
Acknowledgements
We thank Omar Miranda and Clio A. de Moura for fruitful discussions. This work was
supported by a SpanishItalian AI, the Spanish grants FPA200501269 and GV/05/017 of Generalitat Valenciana, as well as a MECINFN agreement. S.P. was supported by a Ramn y Cajal
contract of MEC. P.S. acknowledges the support by the Deutsche Forschungsgemeinschaft under grant SFB 375 and by the European Network of Theoretical Astroparticle Physics ILIAS/N6
under contract number RII3CT2004506222.
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
M. Maltoni, T. Schwetz, M.A. Trtola, J.W.F. Valle, New J. Phys. 6 (2004) 122, hepph/0405172.
G.L. Fogli, E. Lisi, A. Marrone, A. Palazzo, Prog. Part. Nucl. Phys. 57 (2006) 742, hepph/0506083.
Z. Berezhiani, A. Rossi, Phys. Lett. B 535 (2002) 207, hepph/0111137.
Z. Berezhiani, R.S. Raghavan, A. Rossi, Nucl. Phys. B 638 (2002) 62, hepph/0111138.
S. Davidson, C. PeaGaray, N. Rius, A. Santamaria, JHEP 0303 (2003) 011, hepph/0302093.
J. Barranco, O.G. Miranda, C.A. Moura, J.W.F. Valle, Phys. Rev. D 73 (2006) 113001, hepph/0512195.
J. Schechter, J.W.F. Valle, Phys. Rev. D 22 (1980) 2227.
P.D. Serpico, et al., JCAP 0412 (2004) 010, astroph/0408076.
M.S. Bilenky, A. Santamaria, hepph/9908272.
A. de Gouva, J. Jenkins, Phys. Rev. D 74 (2006) 033004, hepph/0603036.
L.B. Auerbach, et al., LSND Collaboration, Phys. Rev. D 63 (2001) 112001, hepex/0101039.
F. Reines, H.S. Gurr, H.W. Sobel, Phys. Rev. Lett. 37 (1976) 315.
Z. Daraktchieva, et al., MUNU Collaboration, Phys. Lett. B 564 (2003) 190, hepex/0304011.
P. Vilain, et al., CHARMII Collaboration, Phys. Lett. B 335 (1994) 246.
A. Friedland, C. Lunardini, C. PeaGaray, Phys. Lett. B 594 (2004) 347, hepph/0402266.
M.M. Guzzo, P.C. de Holanda, O.L.G. Peres, Phys. Lett. B 591 (2004) 1, hepph/0403134.
O.G. Miranda, M.A. Trtola, J.W.F. Valle, hepph/0406280, JHEP, in press.
A. Friedland, C. Lunardini, M. Maltoni, Phys. Rev. D 70 (2004) 111301, hepph/0408264.
A. Friedland, C. Lunardini, Phys. Rev. D 72 (2005) 053009, hepph/0506143.
A. Friedland, C. Lunardini, Phys. Rev. D 74 (2006) 033012, hepph/0606101.
A.D. Dolgov, et al., Nucl. Phys. B 632 (2002) 363, hepph/0201287.
P.D. Serpico, G.G. Raffelt, Phys. Rev. D 71 (2005) 127301, astroph/0506162.
V.F. Shvartsman, Pisma Zh. Eksp. Teor. Fiz. 9 (1969) 315, JETP Lett. 9 (1969) 184.
G. Steigman, D.N. Schramm, J.R. Gunn, Phys. Lett. B 66 (1977) 202.
A.D. Dolgov, Phys. Rep. 370 (2002) 333, hepph/0202122.
116
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
[38]
[39]
[40]
[41]
[42]
[43]
[44]
[45]
[46]
[47]
Abstract
The manifestly gauge invariant formulation for free symmetric higherspin partially massless fields in
(A)dSd is given in terms of gauge connections and linearized curvatures that take values in the irreducible
representations of (o(d 1, 2)) o(d, 1) described by tworow Young tableaux, in which the lengths of the
first and second row are, respectively, associated with the spin and depth of partial masslessness.
2006 Published by Elsevier B.V.
PACS: 11.15.q; 11.10.Kk
Keywords: Higherspin fields gauge theory; Partiallymassless fields; (anti)de Sitter space
1. Introduction
An interesting property of massive higherspin (HS) fields in spaces with nonzero curvature
like de Sitter (dS) and antide Sitter (AdS) is that for certain values of the parameter of mass,
the HS field equations acquire gauge invariance. This phenomenon was originally discovered by
Deser and Nepomechie in [1] and investigated further in [28]. The special values of the mass
parameter are scaled in terms of the curvature of the background space. The first special value of
the mass corresponds to the usual massless fields with the richest gauge symmetries. The other
special values correspond to different fields called partially massless fields. A gauge transformation law of a partially massless field has less parameters but more derivatives compared to the
massless fields. A number of degrees of freedom carried by a partially massless field is interme* Corresponding author.
118
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
diate between the massless case (two for s > 0 in 4d) and the massive case (2s + 1 in 4d). In
the flat space limit all special values of the mass shrink to zero so that one is left with the usual
massless fields, i.e., partially massless theories do not exist in Minkowski space.
In AdSd spacetime, the phenomenon of partial masslessness has clear interpretation in terms
of representations of the AdSd algebra o(d 1, 2). Namely, a space of singleparticle states of a
given relativistic field with the energy bounded from below forms a lowest weight o(d 1, 2)module, E0 , s. Its lowest weight is defined in terms of the lowest energy E0 and spin s
associated with the weights of the maximal compact subalgebra o(2) o(d 1) o(d 1, 2).
For quantummechanically consistent fields, the modules E0 , s correspond to unitary representations of o(d 1, 2) that requires in particular E0 E0 (s) where E0 (s) is some function of the
spin s found for the case of d = 4 in [9] and for the general case in [10] (see also [1115]). At
E0 = E0 (s) null states appear that form a submodule to be factored out. These signal a gauge
symmetry of the system. The corresponding fields are the usual massless fields. Modules with
lower energies E < E0 (s) correspond to nonunitary (ghost) massive fields. At certain values
Ei (s) of E the corresponding nonunitary module may contain a submodule that again signals
a gauge symmetry in the fieldtheoretical description. Such modules correspond to the partially
massless fields. Unfortunately, the lack of unitarity [8] makes their physical applications problematic in the AdS case.
In the dSd case with the symmetry algebra o(d, 1) the analysis is different because it admits
no unitary lowest weight (i.e., bounded energy) representations at all. Relaxing the concept of
bounded energy, it is possible to have unitary partially massless fields in that case [2,4,8].
In this paper we address the question what is a framelike formulation of the symmetric partially massless fields, that generalizes the framelike formulation of ordinary symmetric massless
fields elaborated in [1619]. Such a reformulation operates in terms of a gauge field (connection),
that takes values in some module of the (A)dSd algebra and makes gauge symmetries of the system manifest allowing to write down manifestly gauge invariant geometric actions. As shown
in [19], a usual spins massless field in (A)dSd is described by a 1form connection, which takes
values in the finitedimensional o(d 1, 2)module spanned by traceless tensors of o(d 1, 2)
that have a symmetry of a length s 1 tworow rectangular Young tableau
. The main
result of this paper is that a spins partially massless symmetric gauge fields is analogously described by a 1form connection, which is a o(d 1, 2)(o(d, 1))tensor that has the symmetry
properties of a tworow Young tableaux with the first row of length s 1 and the second row of
an arbitrary length t = 0, 1, . . . , s 1,
.
Since the gauge fields of the framelike formalism are interpreted as connections of a nonAbelian HS algebra of a full HS system, the results of such a reformulation may be useful to rule
out HS algebras incompatible with unitarity by requiring that physically relevant algebras considered as o(d 1, 2)modules under its adjoint action should not contain the modules associated
with (nonunitary) partially massless fields. (Note that, as shown in [19,20], gauge fields associated with true massless fields in AdSd carry o(d 1, 2) indices of various o(d 1, 2)modules
described by Young tableaux, that have first two rows of equal lengths.) Although in this paper
we consider integer spins, our approach admits a straightforward extension to fermions.
The layout of the rest of the paper is as follows. In Section 2 we recall the metriclike formulation of massive, massless and partially massless fields. The framelike formalism for massless
HS fields is recalled in Section 3. The main results of our approach are presented in Section 4 and
then illustrated in Section 5 by the spin2 and spin3 examples. Possible further developments
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
119
are discussed in Section 6. Notation, Young tableau convention and explicit expressions for some
coefficients are collected in Appendices AC.
2. Metric formalism for free higherspin fields
In this section we summarize known results on the description of massive, massless and partially massless HS fields, which are relevant to our analysis.
2.1. Massive higherspin fields
A massive spins totally symmetric field in the flat Minkowski space can be described by a
{s}
ranks symmetric tensor field m1 ...ms (x) that satisfies the equations [21]1
+ m2 {s} = 0,
{s} = divergence {s} = 0, s 1,
{s} = trace {s} = 0, s 2,
(2.1)
which form a complete set of local Lorentzinvariant conditions that can be imposed on {s} . The
first equation of (2.1) is the massshell condition. The third condition, that the field {s} is traceless, is an algebraic constraint. As pointed out by Fierz and Pauli in the pioneering paper [22], the
second condition, which is a differential equation, can be derived for s > 1 from a Lagrangian
without introducing new degrees of freedom by adding auxiliary fields that are expressed in terms
of (derivatives of) the dynamical fields by virtue of equations of motion. For totally symmetric
massive fields of integer spins, the Lagrangian formulation with a minimal set of auxiliary fields
was worked out by Singh and Hagen in [23]. For a spin s they introduced the set of auxiliary
fields that consists of symmetric traceless tensors of ranks s 2, s 3, . . . , 0.
For example, to describe a spin2 field, one auxiliary scalar field is needed. It can be identified
with the trace of a traceful rank2 field mn satisfying the FierzPauli equation
G2 = 2 + m2 (mn mn ) + m n + mn l p lp 2(m k kn) = 0.
(2.2)
One obtains
G2 = m2 ( )
(2.3)
and
2 G2
d 1
m2
G = m4
.
d 2 2
d 2
(2.4)
From Eq. (2.4) it follows that, if m2 = 0, is traceless (i.e., the auxiliary field is zero onshell).
Then from (2.3) it follows that = 0.
If m2 = 0, Eq. (2.3) expresses Bianchi identities which manifest the spin2 gauge symmetry.
Let us note that it is important that the higherderivative part of the massive equations is the
same as for massless fields just to make it possible to express the auxiliary fields in terms of
the dynamical fields by virtue of algebraic constraints, that would not be possible for a different
choice of the coefficients in front of the second derivative terms in (2.2). This means that, for
1 Our notations and conventions are explained in Appendix A.
120
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
a different choice of the coefficients, the corresponding equations describe a dynamical system
with the propagating trace field, that possesses more degrees of freedom. A similar argument will
play crucial role in the subsequent analysis of a proper action for a partially massless field.
Note that equivalent dynamical systems can be formulated with the aid of different sets of
auxiliary fields. Particularly useful approaches to massive fields with different sets of auxiliary
fields include the gauge invariant formulation of [5,24] and the BRST approach of [25,26].
2.2. Massless higherspin fields
The free action for a symmetric massless field of an arbitrary integer spin in the flat Minkowski
space was obtained by Fronsdal [27]. A symmetric spins massless field is described by a totally
symmetric double traceless tensor field {s} m1 ...ms (x), {s} = 0. The equations of motion
(2.5)
(2.6)
with a gauge parameter {s1} (x) being a symmetric rank(s 1) traceless tensor field. The
gauge invariant Lagrangian is
1
1
Ls = {s} F s s(s 1) {s} F s .
(2.7)
2
8
Note that alternative metriclike formulations of symmetric massless HS fields were also developed in [26,28,29].
2.3. Partially massless higherspin fields
The (A)dSd space is described by a metric gm n satisfying the equation
Rm n,p q =
2
(gpm gnq gpn gmq ),
(d 1)(d 2)
(2.8)
where Rm n,p q is the Riemann tensor and the normalization of the cosmological constant is chosen as in [4,5]. is (negative) positive for (anti)de Sitter space. Later on we use the notation
2
, where 1  is the (A)dSd radius.
2 = (d1)(d2)
2.3.1. The approach of Deser and Waldron
In this subsection, we consider the case of d = 4 following to [4] (modulo the adjustment of
the coefficients to the mostly minus signature conventions used in this paper). To illustrate the
effect of partial masslessness let us start with the simplest example of spin two.
The deformation of the massive field Eq. (2.2) to (A)dSd is achieved by the addition of some
masslike terms proportional to along with the replacement of the flat derivatives by the covariant ones D associated with gm n .
The (A)dSd deformation of Eq. (2.2) is
G2 = (2) + m2 ( g ) + DD DD + gD D = 0,
(2.9)
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
121
(2.10)
(2.11)
(2.12)
One observes that, for the critical value m2 = 23 , cannot be algebraically expressed in
terms of the other field and Eq. (2.12) becomes Bianchi identity signaling the appearance of a
gauge invariance which is
m n = D(m Dn) gm n
(2.13)
3
with a scalar gauge parameter . This invariance kills the helicity zero degree of freedom leaving
spin polarizations 1, 2. Note that although cannot be expressed in terms of the other field,
this does not imply the appearance of extra degrees of freedom because it is pure gauge.
At m = 0, (2.11) is the Bianchi identity of the usual spin2 massless theory in (A)dSd .
A spin3 massive field requires a scalar auxiliary field and an auxiliary vector, which combines with a rank3 traceless field into a rank3 traceful tensor m n k . The equations of motion of
the massive spin3 field m n k are [4]
16
DD + DD
G3 = (3) + m2
3
11
1
m
g (1) + m2
(2.14)
D D + DD gD = 0,
3
2
4
3
G0 = (0) + 4m2 8 mD = 0
(2.15)
2
with
2
g ,
3
The following identities hold
3 = + 5
(1) = + ,
(0) = .
1
[D G3 ] = m (DD mD + mD ) ,
2
m
m2
5
2
G3 = m 3m2 4 D m ,
D 2 G3 + DG0
4
4
8
3
5
5
D 3 G3 m 3m2 4 G0 = m 3m2 4 m2 2 ,
12
2
where [. . .] is the projector to the traceless part.
At m = 0, (2.17) becomes the Bianchi identity associated with the gauge invariance
rank2 traceless parameter {2}
= D {2} ,
= 0.
(2.16)
(2.17)
(2.18)
(2.19)
with a
(2.20)
122
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
In this case, the system decomposes into two independent subsystems: massless spin three with
two spin projections 3 is described by the Fronsdal field and massive spin zero is described
by .
At m2 = 43 , (2.18) becomes the Bianchi identity associated with the gauge invariance with
the rank1 gauge parameter {1} , which has the form
2
2
{1}
=
D {1} .
= D g ,
(2.21)
3
3 3
This model describes a field with spin projections 2, 3.
At m2 = 2, (2.19) becomes the Bianchi identity that manifests the gauge invariance with the
scalar parameter {0}
2
10 {0}
3
{0}
= D gD ,
=
(2.22)
DD
,
2
3 3
3
leaving out the helicity zero degree of freedom. At noncritical m2 the scalar and vector are
auxiliary. They are zero on shell and D = 0 as a consequence of the constraints (2.17), (2.18)
and (2.19). A generic noncritical system describes a massive spin3 field with spin projections
(0, 1, 2, 3) (may be ghost and/or tachionic for negative m2 ).
For general spin s, the full set of fields contains a ranks double traceless field along with
the traceless fields of ranks from 0 to s 3. As argued in [4], there exist s critical masses mt 2
(t = 0, 1, . . . , s 1) and the following properties are true for m2 = mt 2 :
The equations of motion are invariant under the gauge transformation
D {t} +
{s} = D
(2.23)
st
k=s
Lkcov + Ls ,
(2.24)
k=0
where
L =
s
k=s
ak {k1} D {k} + bk {k} D {k1} + ck D {k} {k1}
k=0
2
2
+ dk {k} + ek {k} + fk {k} {k2} .
(2.25)
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
123
The coefficients in Ls are fixed by the condition that Ls is invariant under some gauge transformations of the form
1
{k} = D {k1} + k {k} + k g {k2}
k
(2.26)
with rankk traceless gauge parameters {k} (k = 0, 1, . . . , s 1). The gauge invariance condition
expresses all the coefficients in terms of one free mass parameter m2 .
For generic m2 , the Lagrangian (2.24) describes a spins massive field. In that case all gauge
symmetries are Stueckelberg (i.e., algebraic for some of the fields). It is easy to see that the
gauge fixing of all gauge symmetries reproduces the set of fields of Singh and Hagen [23], i.e.,
the SinghHagen theory deformed to (A)dSd is a gaugefixed version of the theory of Zinoviev.
In this approach it is still unclear, however, what is the fundamental principle that underlies the
gauge symmetry of the model, thus fixing the coefficients in the action (2.24).
At the critical values of the mass [5]
m2 t =
2
(s t 1)(d + s + t 4),
(d 1)(d 2)
t = 0, 1, . . . , s 1
(2.27)
some of the coefficients in the gauge transformation law (2.26) and in Ls vanish so that
the Lagrangian (2.24) splits into two independent parts. One contains the fields {k} with
k = t + 1, . . . , s and another one contains those with k = 0, . . . , t. The former Lagrangian describes a partially massless theory, including the usual massless theory as a particular case of
t = s 1, while the latter describes a spint massive field. In the flat limit 0 all critical
masses tend to zero and all lower derivative terms in the Lagrangian vanish so that Ls reduces to
the sum of the flat space Fronsdals Lagrangians of massless fields 0 , . . . , s of all spins from 0
to s.
3. Massless higherspin fields in the framelike formalism
3.1. Lorentz covariant approach
The Minkowski or (A)dSd geometry can be described by the frame field (vielbein) ea =
and Lorentz spinconnection 1form a,b = dx m m a,b (a,b = b,a ), which obey the
equations
dx m em a
T a dea + a, b eb = 0,
(3.1)
(3.2)
where d = dx m xm is the de Rham operator, 2 is the cosmological constant and the exterior
product symbol will be systematically omitted in what follows. Eq. (3.1) is the zerotorsion
constraint that expresses the Lorentz connection m a,b via vielbein em a , provided that the vielbein is nondegenerate, which is required. Eq. (3.2) then describes the AdSd space with the
symmetry algebra o(d 1, 2) if 2 > 0, dSd space with the symmetry algebra o(d, 1) if 2 < 0
and flat Minkowski space if 2 = 0. The latter corresponds to the contraction of either o(d 1, 2)
or o(d, 1) to iso(d 1, 1) and can be described by m a,b = 0 and em a = m a in Cartesian coordinates. Dynamical gravity is described by the vielbein and Lorentz connection which satisfy the
zerotorsion condition (3.1) and Eq. (3.2) relaxed to the Einstein equations.
124
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
(3.3)
C a(n),b(m),... (x)
where
Let us now recall how this approach extends to massless HS fields in flat and (A)dS spaces.
The generalization to HS massless fields in the Minkowski space [16] starts with the gauge fields
em a(s1) and m a(s1),b with the tangent Lorentz indices possessing the symmetry properties
of traceless Young tableaux of the form
coordinates, the gauge transformations are
and
, respectively.3 In Cartesian
(3.5)
where the traceless gauge parameters a(s1) , a(s1),b and a(s1),bb have the symmetry properties of Young tableaux
,
and
, respectively.
In these terms the gauge invariant action can be written in the following simple form
1 m n1 ...ns2 p,
a1
ad3
n1 ...ns2 p
q,r
S=
dx dx
a1 ...ad3 pqr de
+ dx
.
m n1 ...ns2
2
d
M
(3.6)
It is easy to check its gauge invariance, although it is not manifest.
Further analysis [17,18] shows that, both in Minkowski and in (A)dSd cases, HS fields are
conveniently described by a set of 1forms m a(s1),b(t) , t = 0, 1, . . . , s 1, possessing the
symmetry properties of the tworow traceless Young tableaux4
with t 2 called extra fields are gauge fields for higher gauge symmetries the
tions m
simplest of which is that with the gauge parameter a(s1),bb in the gauge transformations (3.5).
This leads to the gauge field a(s1),bb which in its gauge transformation contains a new gauge
parameter thus leading to the next extra field and so on.
The gauge transformation law for em a(s1) m a(s1) is
a(s1),b(t)
(3.7)
where Dm is the Lorentz covariant derivative defined with respect to the background spin connection a,b and vielbein ea that describe either Minkowski or (anti)de Sitter geometry. Raising
a form index by the vielbein a(s1)b = m a(s1) emb one observes that this field decomposes
into the following irreps of o(d 1, 1)
a(s1)b =
(3.8)
2 As explained in more detail in Appendix A, upper or lower indices denoted by the same letter are automatically
symmetrized.
3 I.e., the 1form a(s1),b is symmetric in the indices a, traceless in the indices a and b, and the symmetrization over
all indices gives zero: a(s1),a 0. See also Appendix B.
4 I.e., a(s1),b(t) is symmetric in indices a and b separately, traceless in all indices and a(s1),ab(t1) 0.
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
125
The first two components combine into the Fronsdal field: a ranks double traceless tensor decomposes into traceless tensors of ranks s and (s 2). The last component is Stueckelberg. It
is gauged away by the traceless gauge parameter a(s1),c that has the same symmetry type.
As a result only the gauge parameter a(s1) is left and (3.7) gives rise to the Fronsdal gauge
transformation law (2.6).
With some rescaling of fields by the gauge transformations of the other connections have
the form
m a(s1),b(t)
= Dm a(s1),b(t) + emc a(s1),b(t)c + as,t (d)P a(s1),b(t1) em b ,
(3.9)
where as,t (d) are some coefficients and the projector P restores the tracelessness and the proper
Young symmetry. All the parameters a(s1),b(t) with t > 0 are Stueckelberg, eliminating some
components of the connections.
The fieldstrengths, which have analogous form5
a(s1),b(t1) b
a(s1),b(t)
a(s1),b(t)
a(s1),b(t)c
R{2}
= D{1}
+ ec {1}
+ as,t (d)P {1}
e ,
(3.10)
are demanded to be gauge invariant,
a(s1),b(t)
R{2}
= 0,
(3.11)
(3.12)
+ .
x
For example, in the spin2 case, the field a,b is the Lorentz connection. It is expressed in terms
of the first derivatives of the dynamical field, the vielbein ea . The field a(s1),b is analogous
to the Lorentz connection and is called auxiliary. The condition that the free action contains at
most two spacetime derivatives of the dynamical field is equivalent to the extra field decoupling
condition that requires the variation of the free action with respect to the extra fields to vanish
identically
S
0, t 2.
a(s1),b(t)
A general P even, gauge invariant action in (A)dSd has the form [17,18]
S=
k=s2
1
c1 a(sq2),c2 b(s2)
aq
c1 ...cd ec5 ecd R{2}
R{2} c3 a(sq2) ,c4 b(s2) .
2
k=0
Md
(3.13)
(3.14)
126
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
The extra field decoupling condition fixes the coefficients aq modulo a spindependent normalization factor. The resulting action is manifestly gauge invariant by (3.11), containing just two
derivatives of the dynamical field upon excluding the auxiliary field by virtue of its equation
of motion. As such, it is equivalent to the Fronsdal action in AdSd . (In fact, the free action for
massless fields in AdSd for any d was originally obtained in this form in [18].)
3.2. (A)dSd covariant approach
Eqs. (3.1) and (3.2) are equivalent to the zerocurvature condition for the 1form connection
W0 A,B of o(d 1, 2) or o(d, 1)
dW0 A,B + W0 A, C W0 C,B = 0.
(3.15)
For definiteness, let us consider the AdSd case of o(d 1, 2). The Lorentz connection and frame
1form are identified as
W0 a,b = a,b ,
W0 a, = ea .
(3.16)
denotes the extra time value of the o(d 1, 2) vector index compared to the o(d 1, 1) vector
index. (For more detail see Appendix A.) The dimensionful constant to be identified with
the inverse (A)dS radius is introduced to make the vielbein dimensionless. The curvature R =
dW0 + W0 2 decomposes as R = T a Pa + 12 R a,b Ma,b , where the o(d 1, 2) generators Pa and
Ma,b are associated with (A)dSd translations and Lorentz rotations, respectively. T a and R a,b
are the torsion tensor (3.1) and the modified Riemann tensor (3.2), respectively. Eq. (3.15) can
be rewritten as the nilpotency condition D0 2 = 0 for D0 = d + W0 .
As shown by MacDowell and Mansouri [30], modulo the topological GaussBonnet term, the
action6
1
S= 2 2
R a1 ,a2 R a3 ,a4 ea5 ead a1 a2 ...ad
(3.17)
4
Md
in d = 4 describes Einstein gravity with the cosmological term. The case of arbitrary d was
discussed in [19], where it was shown in particular that this action still describes a massless
spin two field at the linearized level although differs from the pure Einstein action with the
cosmological term at the nonlinear level (for more detail see also [31]).
A covariant generalization of the MacDowellMansouri approach proposed by Stelle and
West [32] is achieved by introducing a compensator field V A (x) that has a fixed norm
V A VA = 1
(3.18)
and carries no physical degrees of freedom. A sign (+) corresponds to the (A)dS case. The
Lorentz algebra o(d 1, 1) o(d 1, 2)(o(d, 1)) is identified with the stability subalgebra of
the compensator. The vielbein and spin connection admit the covariant definition
E A = DV A = dV A + W0 A, B V B ,
W0L A,B = W0 A,B E A V B E B V A .
(3.19)
(3.20)
6 The coefficient 2 instead of is due to a somewhat different dimension of fields describing background space in
comparision with dynamical ones. 2 is the gravitational constant.
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
127
(3.21)
and the Lorentz covariant derivative of the compensator with respect to the spin connection (3.20)
is zero
DL V A = dV A + W0L A, B V B = 0.
(3.22)
A convenient gauge choice, referred to as standard gauge, which breaks local o(d 1, 2)
(o(d, 1)) down to the local Lorentz algebra, is
V A = A
(3.23)
with the compensator vector pointing along the extra (d + 1)th direction. In the standard gauge,
the nonzero components E a of E A are identified with the vielbein ea .
The idea of the approach to the description of HS gauge fields proposed in [18,19] was to
replace the (A)dSd isometry algebra h = o(d 1, 2) or o(d, 1) by some its extension g called
HS symmetry algebra in such a way that the dynamics of massless HS fields is described in terms
of the connection 1forms of g similarly to as gravity is described by the connections of o(d, 1)
(o(d 1, 2)). Note that g turns out to be infinitedimensional.
The free field approximation results from the following perturbative expansion. A zeroorder
vacuum (background) field 0 takes values in h g, i.e., it identifies with the 1form W0A,B =
W0B,A . It is required to satisfy the zerocurvature equation for h thus describing (A)dSd space
time (or Minkowski in the flat limit).
1 describes firstorder fluctuations of the fields of all spins including gravity. The HS field
strength is
R = R0 + R1 + o(1 ),
(3.24)
where
R0 = d0 + [0 , 0 ],
(3.25)
R1 = d1 + [0 , 1 ].
(3.26)
Here [ , ] is the Lie product of g. Since h g, the zeroorder part of the curvature of the full HS
algebra g is zero R0 = 0. The firstorder part is R1 = D0 1 , where D0 is the background covariant derivative built of the connection 0 . Since D0 2 = 0, R1 is invariant under the linearized
gauge transformations
1 = D0 ,
(3.27)
where is an arbitrary gauge parameter that takes values in g. This is the linearized HS gauge
transformation. Any action constructed in terms of R1 is therefore automatically HS gauge invariant at the linearized level.
In the known cases, g decomposes into an infinite direct sum of finitedimensional hmodules
under the adjoint action of h g. Gauge fields 1s , corresponding to different irreducible submodules labeled by the integral weights s of h, describe fields of different spins s. In other
words, the connections 1s take values in irreducible tensor hmodules.
As an example, a spins symmetric massless field in (A)dSd can be described [19] by the
A(s1),B(s1)
with tangent indices that possess the symmetry of the tworow traceless
1form Wm
128
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
(3.28)
R{2}
A(s1),B(s1)
= D0 W{1}
(3.29)
hmodule
k = 0, 1, . . . , (s 1),
(3.30)
A(s1),B(s1)
(3.31)
q=s2
1
s
bq
A1 ...Ad+1 V A5 E A6 E Ad+1
2
q=0
R{2}
Md
A1 B(s2),A2 C(q)D(sq2)
(3.32)
where VC(p) = VC VC . The coefficients bqs are fixed [19] modulo an overall factor
p
bqs = bs
(d 5 + 2q)!!(q + 1)
q!
(3.33)
by the condition that the variation with respect to the fields contracted with less than s 2
compensators is identically zero
S
L V C(k)
(3.34)
0, k < s 2,
W A(s1),B(sk1)C(k)
where L is the projector to the V A transversal part of a tensor. By virtue of (3.31), this condition
is equivalent to the extra field decoupling condition (3.13). As a result, modulo total derivatives,
the action depends only on the vielbeinlike and Lorentz connectionlike fields (3.31), being free
of higher derivatives.
Now we are in a position to apply this approach to the description of partially massless fields.
For more detail on the HS gauge theory we refer the reader to [31,33].
7 I.e., the field W A(s1),B(s1) is symmetric in indices A and B separately, traceless and the symmetrization over
m
any s indices gives zero: Wm A(s1),AB(s2) 0.
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
129
A(s1),B(t)
A(s1),B(t)
W{1}
= D0 {0}
(4.1)
(4.2)
Since D0 2 = 0, the field strength is gauge invariant and satisfies Bianchi identities
A(s1),B(t)
R{2}
= 0,
A(s1),B(t)
D0 R{2}
= 0.
(4.3)
In the rest of this section we sketch the main results of our analysis leaving some details of
their derivation to a later publication [34], where it will be shown in particular, that all gauge
A(s1),B(t)
are Stueckelberg (i.e., analogous to the Lorentz symmetry
symmetry parameters {0}
parameters in the case of gravity) except for the parameter
A(s1),b(t)
b(t) = {0}
VA1 VAs1 ,
A(s1),B(t)
A(s1),B(t)
{0}
A(s1),b(t2)
b(t) bb = {0}
(4.4)
A(s1),B(t)
0,
. Indeed, in
(4.5)
A(s1),B(t)
{0}
a(t)
t >0
(4.7)
(4.8)
Note that the fields a(t) and a(t1) are automatically traceless (cf. (4.5)).
8 I.e., W A(s1),B(t) is a traceless tensor symmetric in indices A and B separately, that satisfies the antisymmetry
property W A(s1),AB(t1) 0.
130
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
a(s)
= D a D a a(t) .
(4.9)
It coincides with the transformation law (2.23) for partially massless fields found by Deser and
Waldron [4] and Zinoviev [5].
Gauge invariant combinations of derivatives of the dynamical fields that are nonzero onshell
we call generalized Weyl tensors. In practice, to find generalized Weyl tensors, one should set
to zero as many components of the gauge invariant field strengths as possible by imposing
constraints that express auxiliary fields in terms of derivatives of the dynamical fields (these
generalize the zerotorsion condition in gravity). Other components of the gauge invariant field
strengths are zero by virtue of the field equations (these generalize the Einstein equations in
gravity). Some more vanish by virtue of Bianchi identities applied to the former two (e.g., the
cyclic part of the Riemann tensor in gravity is zero by virtue of the Bianchi identity applied to
the zerotorsion condition). What is left nonzero are the generalized Weyl tensors.
In the massless case with t = s 1, the generalized Weyl tensor is known [19] to be described
by the length s tworow rectangular o(d 1, 1) Young tableau. In the partially massless case, a
a(s1),b(t)
generalized Weyl tensor identifies [34] with the traceless component of R{2}
(which is the
A(s1),B(t)
R{2}
Md
A1 B(sk2)C(k),A2 D(tm1)C(m)
(4.10)
where
s,t
s,t
= bk,m
(m) (k) (t m 1) (s m k 2)
bk,m
with
(n) = 1, n 0,
(n) = 0, n < 0
(4.11)
(4.12)
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
131
The parameters s and t , which determine a field type, are fixed. Different choices of the
s,t
coefficients bk,m
lead to different gauge invariant dynamical systems. To simplify formulae,
the range of summation over the indices k and m in (4.10) is chosen so that the terms with
k > s t 1 are not independent, being expressible via linear combinations of terms with
smaller k because the symmetrization over any s tangent indices of the HS curvatures gives
zero.
Generally, different gauge connections that enter the curvatures in (4.10) are independent
fields, which may carry their own degrees of freedom. Alternatively, all connections can be expressed in terms of (derivatives of) certain dynamical fields by virtue of imposing constraints
that generalize the zero torsion condition in gravity and require some components of the field
strengths to vanish. Plugging the resulting expressions back into the action gives rise to a gauge
invariant higher derivative action, which may again describe a system with extra degrees of freedom. To avoid these unwanted degrees of freedom, the coefficients in the action (4.10) should be
adjusted so that, in the flat limit, the action should amount to the sum of free massless actions for
spins from t + 1 to s in agreement with the construction of Zinoviev discussed in Section 2.3.2.
It turns out that for the models with t > 1 this is achieved by imposing the extra field decoupling
condition
S s,t
L V C(m+k)
0,
W A(sk1)C(k),B(tm)C(m)
m = 0, 1, . . . , t 2, k = 0, 1, . . . , s t 1,
(4.13)
which means that the variation is nonzero only for the components W A(s1),BC(t1) VC(t1) and
generalizes (3.34) to partially massless fields.
Using
R{2} A(s1),B(t) = D0 W{1} A(s1),B(t) ,
(4.14)
the symmetry properties of the field strength, the Bianchi identities (4.3), the definition of the
frame (3.19) along with its consequence D0 E A = 0 and the identities
A1 ...Ad+1 = V F VA1 F A2 ...Ad+1 + VA2 A1 F A3 ...Ad+1 + + VAd+1 A1 ...Ad F ,
(4.15)
A1 ...A5 H6 ...Hd+1 E F E H6 E Hd+1
k=5
1
F
k+1
=
A ...A k ...A H ...H Ak ()
E H5 E Hd+1 ,
1
5 5
d+1
d 3
(4.16)
k=1
which manifest the fact that the antisymmetrization of d + 2 indices, which take d + 1 values, is
zero, we obtain
1
S s,t =
A1 A2 ...Ad+1 V A4 E A5 E Ad+1 VC(2k+2m+1)
d 3
k,m
Md
s,t
(k + 1)2 s,t
s,t
d 3 + 2(k + m) bk,m
(m + 1)bk,m+1
+
b
(s k 2) k+1,m
132
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
s,t
(k + 1)(s k m 1) s,t
d 3 + 2(k + m) bk,m +
bk+1,m
(s k 2)
+ (W R) .
(4.17)
The extra field decoupling condition (4.13) demands most of the terms in the variation (4.17)
to vanish. This is achieved by virtue of imposing the following conditions on the coefficients
s,t
s,t
+ (k+1)(skm1)
bk+1,m
= 0,
[d 3 + 2(k + m)]bk,m
(sk2)
2
s,t
s,t
s,t
(k+1)
(4.18)
[d 3 + 2(k + m)]bk,m
(m + 1)bk,m+1
+ (sk2)
bk+1,m
= 0,
t > 1, k = 0, . . . , s t 1, m = 0, . . . , t 2,
s,t
which fix bk,m
up to an overall factor bs,t
s,t
bk,m
= bs,t
(4.19)
Md
W A1 B(sk2)C(k),A2 C(t1) R A3 B(sk2) C(k),C(t) + (W R)
s,t
(k + 1)2 s,t
b
.
d 5 + 2(k + t) bk,t1 +
(s k 2) k+1,t1
(4.20)
As explained in Section 4.3, the coefficients (4.19) guarantee the consistency of the flat limit,
which is in agreement with the construction of Zinoviev. For the massless case of t = s 1, the
variation reduces to that of the action (3.32) [19].
The partially massless cases with t = 0 and t = 1 are special. In the case of t = 1, the variation
of the action is
1
s,1
S =
A1 A2 ...Ad+1 V A4 E A5 E Ad+1 VC(2k+1)
d 3
Md
(d 3 + 2k)bk,0 +
bk+1,0
(s k 2)
+ (W R) .
(4.21)
The extra fields decoupling condition (4.13) is not applicable for this case and the coefficients are
fixed directly from the requirement of consistency of the flat limit. As explained in Section 4.3,
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
133
upon an appropriate rescaling, the first term in (4.21) tends in the flat limit to the sum of variations
of massless actions for spins from (t + 1) to s, while the second term diverges. The correct flat
s,t
limit therefore requires the second term to be zero that fixes bk,0
in the form
(s k 1)(d 5 + 2k)!!
,
(4.22)
k!s!
i.e., (4.19) is still true.
In the case of t = 0, the action cannot be written at all in the form (4.10) because there is no
room for the indices A2 and A4 . For this case the action is constructed in terms of the generalized
Weyl tensor C a(s),b
1
1
S s,0 =
c1 ...cd ec1 ecd C a(s),b Ca(s),b C a(s1)m, m Ca(s1)n, n .
(4.23)
2
2
s,1
= bs,1
bk,0
Md
Note that, in d = 4, this case corresponds to the conformal partially massless fields of Deser
and Waldron [4] with 4d Maxwell electrodynamics as a particular case of s = 1, where C a,b
identifies with the Maxwell field strength. The form of the action (4.23) is in agreement with the
wellknown fact that the Maxwell Lagrangian 4form cannot be written as the exterior product
of the Maxwell 2form field strengths.
4.3. Lorentz covariant formulation
The practical analysis of the field content of the model under consideration is more illuminating in terms of the Lorentz irreducible components of the fields. An irrep of o(d 1, 2)(o(d, 1))
with the Young symmetry type
o(d 1, 1)
=
k=s1,
l=t
(4.24)
k=t, l=0
i.e., in terms of Lorentz tensors, the partially massless theory of spins and deptht is described
{k,l}
a(k),b(l)
{1} with k = t, . . . , s 1, l = 0, 1, . . . , t.
by 1forms {1}
The field strengths have the following structure
{k,l}
{k,l}
{k+1,l}
{k,l+1}
{k,l}
+ 2 {1}
R{2} = D0 {1} = D {1} + 1 {1}
{k1,l}
{k,l1}
+ +2 {1}
.
+ +1 {1}
(4.25)
+1 C a(k),b(l) = g(k, l)(k + 1) ea C a(k),b(l)
k
em C a(k1)m,b(l) aa
2(d + 2k 2)
l
em C a(k),mb(l1) ab
d +k+l3
k
a(k1)b,b(l1)m aa
+
em C
,
l(d + 2k 2)(d + k + l 3)
(4.26)
134
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
+2
a(k),b(l)
= G(k, l)(l + 1) eb C a(k),b(l)
k a a(k1)b,b(l)
e C
kl
l
em C a(k),b(l1)m bb
d + 2l 4
k(k l 1)
em C a(k1)m,b(l) ab
(k l)(d + k + l 3)
lk(d + 2k 4)
+
em C a(k1)b,b(l1)m ab
(k l)(d + 2l 4)(d + k + l 3)
k(k 1)
+
em C a(k2)mb,b(l) aa
(k l)(d + k + l 3)
lk(k 1)
em C a(k2)bb,b(l1)m aa ,
(k l)(d + k + l 3)(d + 2l 4)
l
1 C a(k),b(l) = f (k, l) em C a(k1)m,b(l) +
em C a(k1)b,b(l1)m ,
kl+1
a(k),b(l)
2
a(k),b(l1)m
C
= F (k, l) em C
,
(4.27)
(4.28)
(4.29)
where g(k, l), G(k, l), f (k, l) and F (k, l) are some coefficients, which can be either derived from
the covariant field strengths (4.2) or obtained directly from the condition of the gauge invariance
a(k),b(l)
. Their explicit form is given in Appendix C. (For more detail see [34].)
of R{2}
A general P even o(d 1, 1)covariant gauge invariant action has the following form in terms
of Lorentz covariant field strengths
1
s,t
S s,t =
a k,m c1 cd ec5 ecd R{2} c1 a(k1),c2 b(m1) R{2} c3 a(k1) ,c4 b(m1) , (4.30)
2
k,m
Md
where
s,t
s,t
a k,m
= ak,m
(k t) (m) (t m) (s k 1)
(4.31)
s,t
remain to be determined. This Lagrangian generalizes to partially massand the coefficients ak,m
less fields that introduced in [18] for massless fields (t = s 1).
Let us introduce the inner product
{k,l} {k,l}
c3
c1 a(k1),c2 b(l1)
,c4
{p} {q} = c1 c2 ...cd ec5 ecd {p}
a(k1)
b(l1) , (4.32)
Md
a(k),b(l)
a(k),b(l)
The normalization coefficients gk,l , Gk,l , fk,l and Fk,l are chosen so (see Appendix C) that
the operators +1,2 are conjugated to 1,2 with respect to this product
{k,l}
{k,l}
{k,l} 1,2 {k,l}
{p} + {q} = ()pq+p+q {q} 1,2 {p} .
(4.34)
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
135
(4.35)
+ {k,m} 2 R {k,m+1}
+ {k,m} +1 R {k1,m}
+ {k,m} +2 R {k,m1}
s,t
s,t
a k,m+1 a k,m
s,t
s,t
a k1,m a k,m
s,t
s,t
a k,m1 a k,m
.
(4.36)
The 1form W A(s1),B(t) is equivalent to the set of Lorentz 1forms a(l),b(u) with l =
t, . . . , s 1 and u = 0, 1, . . . , t. It is easy to see that W A(s1),BC(t1) VC(t1) is equivalent to
a(l) and a(l),b with l = t, . . . , s 1. As a result, in terms of Lorentz tensors, the extra field
decoupling condition (4.13) takes the following simple form
S s,t
a(k),b(m)
It is satisfied with
0,
k = t, . . . , s 1,
m = 2, 3, . . . , t.
s,t
= a s,t .
ak,m
(4.37)
(4.38)
k=s1
{k,0} 2 R {k,1} {k,1} +2 R {k,0} .
(4.39)
k=t
S S
(4.40)
(4.41)
(4.42)
In the flat limit 0, terms that mix different k tend to zero and the field strengths take the
form of the flat space field strengths for usual massless fields [18]
{k,1}
{k,0}
{k,0}
R{2} = D {1} + 1 {1} ,
(4.43)
{k,1}
{k,1}
R{2} = D {1} .
(4.44)
As a result, for t > 1, the extra field decoupling condition (4.37) guarantees that the flat limit
of the action (4.30) is a sum of free actions for massless fields of spins from t + 1 to s which is in
agreement with the construction of Zinoviev [5]. The totally symmetric parts a(k)a of m a(k) ,
k = t, . . . , s 1, identify with the double traceless Fronsdal fields used by Zinoviev.
136
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
As it will be explained in more detail in [34], for = 0, all fields, that are not pure gauge,
can be expressed in terms of derivatives of the dynamical fields a(s) , a(t) and a(t1) by virtue
of constraints Rm n a(k) = 0, R a(k)ma m = 0, which are particular field equations. As discussed
in Section 2.3.1, within the approach of Deser and Waldron [4], the constraints result from the
divergences of field equations. Note that the reduction of a number of fields compared to those
of the approach of Zinoviev does not imply the reduction of a number of degrees of freedom
because, as a result of resolution of constrains, the action expressed in terms of the dynamical
fields contains higher derivatives for a general partially massless system.
In the special case of t = 1, where the extra field decoupling condition (4.37) does not apply,
the variation of the action (4.30) in terms of the rescaled variables (4.40) is
S s,1 =
k=s1
s,1
{k,0} 2 R {k,1} {k,1} +2 R {k,0} ak,0
k=1
(4.45)
s,1
= a s,1 so that (4.38) is still true.
To get rid of the term singular in the flat limit 0, we set ak,0
The remaining variation reduces in the flat limit to that of the sum of actions for massless fields
of spins from 2 to s.
5. Examples
In this section the proposed approach is illustrated by the simplest examples of spin2 and
spin3 partially massless fields.
5.1. Spin2
A
Let us consider the model with s = 2, t = 0. The corresponding gauge field is a 1form W{1}
with the gauge transformation law
A
A
= D0 {0}
.
W{1}
(5.1)
(5.2)
(5.3)
A = 1 ,
{0}
{0}
a ,
a = {0}
A R
1
R{2}
m n = R{2} ,
a ,
Rm n a = R{2}
A = 1 W ,
W{1}
m
{1}
a ,
m a = W{1}
(5.4)
where the rescaling of the first component by is introduced for the future convenience. The
gauge transformation law (5.1) gives
m = Dm emb b ,
(5.5)
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
137
m a = Dm a + 2 em a .
(5.6)
From the form of the variation (5.5) it follows that m can be gauge fixed to zero
m = 0
(5.7)
by adjusting the Stueckelberg parameter a . Then, from the condition m 0 it follows that
the leftover gauge symmetry with the parameter requires a = D a . Substituting this into the
gauge transformation law (5.6) we obtain
m a = Dm D a + 2 em a .
(5.8)
= (ab)
= c
[ab] = 0,
(ab)
c.
In these
(5.9)
= D D + .
(a
2 ab
b)
(5.10)
We see that [ab] is gauge invariant while hab (ab) transforms as a partially massless spin2
field (2.13). The field strength (5.2) reduces to
Rm n = Dm n emb n b ,
R m n a = D m n a + 2 e m a n .
(5.11)
Rm n a = D m n a .
(5.12)
, with
= R [abc] ,
= R (ab)c
and
= R m m a .
From (5.12) we see that it is possible to impose the gauge invariant constraint
Rm n = 0,
(5.13)
which gets rid of the auxiliary field [ab] . Note that analogous constraints express auxiliary fields
via derivatives of the other fields in more complicated systems.
The Bianchi identities (5.3) reduce to
Dm Rn k emb Rn k b 0,
Dm Rn k a + 2 em a Rn k 0.
(5.14)
Dm Rn k a 0.
(5.15)
of the field strength Rn k a
138
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
Since the gauge invariant tensor C [ab],c contains the first derivatives of the physical field haa ,
it is possible to impose the following second order equations of motion with the same structure
of indices as that of haa
Dm C (am,a) + D (a C a)m, m + aa Dn C nm, m = 0.
(5.16)
(5.18)
1
1
d2 2 2 2
L = (Dh)2 (D h)2 (Dh )2 h D 2 h +
(5.19)
h h .
2
2
2
2
Thus, the s = 2, t = 0 model of Section 2.3 is demonstrated to describe properly the partially
massless spin2 theory of Deser, Waldron [4] and Zinoviev [5]. Note that, in agreement with the
general analysis of Section 4, because of the lack of indices carried by the field strength 2form
in the degenerate case of t = 0, the Lagrangian of this model is formulated in terms of the tensor
C [mn],a rather than in terms of the exterior product of the field strength 2forms as for t > 0
systems.
5.2. Spin3
The case of spin3 admits partially massless fields either with t = 0 or with t = 1. The case of
t = 0 is analogous to that considered in Section 5.1 for spin2 and will not be discussed here. The
more interesting case of s = 3, t = 1 partially massless field reveals main features of a general
partially massless theory.
The gauge field of this model takes values in the irreducible (A)dS tensor representation
. In the manifestly antisymmetric basis it satisfies the
associated with the Young tableau
conditions W AB,C = W BA,C , W [AB,C] 0 and W AM,N MN 0 (square brackets denote antisymmetrization). The gauge transformation law, field strength and Bianchi identities read
[AB],C
[AB],C
= D0 {0}
,
W{1}
[AB],C
[AB],C
= D0 W{1}
,
R{2}
[AB],C
0.
D0 R{2}
(5.20)
decomposes into the following irreps of o(d 1, 1)
(5.21)
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
139
a,c = {0}
ab,c
ab,c = {0}
+
1 a bc
b ac .
d 1
(5.22)
The fields m a , m ab , m a,c , m ab,c and the corresponding field strengths are defined analogously. In these terms the gauge transformation laws read
m a = Dm a + 3emm am + emm a,m ,
d
1
em [a b] + emm ab,m ,
d 1
2
d
em (a b) +
emm m ab + emm (am,b) ,
m a,b = Dm a,b
d 1
d 1
m ab = Dm ab +
bc emm 3 am + a,m
ac emm 3 bm + b,m .
+
d 1
d 1
(5.23)
The decomposition of the fields into irreps of the Lorentz algebra o(d 1, 1), that acts diagonally
on the fiber indices a, b, c, . . . and the base indices m, n, . . . , is as follows
m a =
1 a 32
(5.24)
= m m ,
,
= [ abc (a,bc) ],
(5.25)
= 1 ,
= a,m m ,
,
(5.26)
= [abc
m ab,c =
= [ab] ,
= [a,bc c,(ab) ],
ab
m =
= (ab) ,
m a,c =
[abc] ],
= [abc] ,
2 a 21
= am
m,
,
(5.27)
= (am,c) m ,
= [am,c] m .
Explicit expressions are given here for those components, which appear in the subsequent analysis; the labels 1, 2 distinguish between different components of the same symmetry type.
140
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
and
of m a and
of m a,c can be
gauge fixed to zero by the Stueckelberg gauge parameters a,c , ab and ab,c . We therefore set
a,bc c,(ab) = 0.
(ab) = 0,
[ab] = 0,
(5.28)
This gauge is not complete. In the leftover gauge transformations, that leave invariant the gauge
conditions, the parameters a,c , ab and ab,c are expressed in terms of derivatives of a , which
is the only nonStueckelberg gauge parameter. The gauge transformation law of the component
aaa has the form (2.21) of the spin3 partially massless theory, i.e., aaa identifies with the
dynamical field of Deser and Waldron.
The field strengths have the form
Rm n a = Dm n a + 3emm n am + emm n a,m ,
d
1
em [a n b] + emm n ab,m ,
d 1
2
d
Rm n a,b = Dm n a,b
em (a n b) +
emm n m ab + emm n (am,b) ,
d 1
d 1
Rm n ab = Dm n ab +
+
(5.29)
bc emm 3n am + n a,m
ac emm 3n bm + n b,m .
d 1
d 1
Let us consider which gauge invariant conditions can be imposed on the fields by setting to zero
some of components of the field strengths. These may include either constraints, that express auxiliary fields in terms of derivatives of the dynamical fields, or impose differential field equations
on the dynamical fields.
The simplest constraint is Rm n a = 0. Taking into account the gauge conditions (5.28), it implies
= [abc] = 0,
= abc [abc] = 0,
d 1 a
3
3
D 2 a + 1 a = 0.
d
2
2
(5.30)
(5.31)
(5.32)
The constraint (5.32) can be used to express the auxiliary field 2 a in terms of 1 a and the first
derivative of the scalar field . The constraint (5.31) in the gauge (5.28) implies that the hooksymmetry parts of ab and a,b are zero. As a result one is left with traces , 2 a of a , ab ,
with the totally symmetric component aaa of a,b , with its trace 1 a and the field ab,c , which
has not been yet considered.
Let us now impose the condition Rm n a,b = 0. It implies that
= R a,[bbb] = [bb,ab] = 0,
(am,aa) = D m aaa + D (a a,a)m ,
(5.33)
(5.34)
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
141
= [aa,ba] to zero.
The condition that the projection of R [ab],c to the components of the symmetry
is zero (i.e., R m(a,aa) m = 0 and R st,a st = 0) gives the following equation
1
d 2
DD + DD1 32 + 2 1 2
D = 0,
2
3d
which is the dynamical equation on and its trace 1 .
The condition Rm n ab = 0 in the gauge (5.28) sets to zero the component
and
(5.35)
of ab,c and
(5.36)
(5.37)
E A6 E Ad+1 .
(5.39)
In terms of Lorentz components rescaled according to (4.40), its variation turns out to be
S = [a1 a2 ] R a3 + a1 R [a2 a3 ] + [a1 a2 ], b R a3 , b + a1 , b R [a2 a3 ], b
Md
1
+ [a1 a2 ], b R a3 b + [a1 b] R [a2 a3 ], b a1 ...ad ha4 had ,
(5.40)
where
=
2d
(a1 + a2 ),
d 1
= a1 ,
= a1
2
a2 .
d 3
(5.41)
(5.42)
142
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
S
(5.43)
= 0 Rst st = 0,
S
= 0 Rmn a,c = 0,
(5.44)
m [ab],c
S
= 0 R (am,aa) m = 0.
(5.45)
aaa
These equations are equivalent to the conditions (5.30)(5.38) on the field strengths, which have
been just shown to describe properly the spin3 partially massless theory of depth one.
6. Discussion
In this paper, partially massless fields in (A)dS space studied in [18] are shown to admit a
natural formulation in terms of gauge 1forms that take values in irreducible tensor representations of the algebra (o(d, 1))o(d 1, 2), described by the various tworow Young tableaux. The
case of rectangular tworow Young tableaux corresponds to the massless fields [19]. Different
nonrectangular Young tableaux correspond to different partially massless fields. The manifestly
gauge invariant free Lagrangians are bilinears of gauge invariant field strengths.
One of the applications of the obtained results is that they impose strong restrictions on possible HS algebras. Actually, since partially massless fields in AdSd correspond to nonunitary
representations of o(d 1, 2), for a HS theory in AdSd to be unitary, it should be free of gauge
fields associated with partially massless fields. This means that a HS algebra g considered as
a o(d 1, 2)module with respect to its adjoint action in g should not contain submodules described by the corresponding Young tableaux. This condition is indeed satisfied in the case of HS
algebras underlying HS theories of totally symmetric fields [14]. The simplest example of a HS
algebra, which is associated with the supersymmetric extension of the theory of symmetric HS
fields studied in [14], also satisfies this condition. It should be true, however, for any other unitary
HS theory including those that contain generic mixed symmetry fields [3537], thus imposing
nontrivial restrictions on a structure of a HS algebra and, therefore, on the spectra of HS gauge
fields that may appear in consistent HS gauge theories.
It would be interesting to extend the results of this paper to the case of general mixed symmetry partially massless bosonic and fermionic fields, originally discussed in [37]. Taking into
account the results of [20], where it has been shown that massless fields of a general symmetry
type are described by pform gauge fields that carry a representation of o(d 1, 2) described
by a Young tableau with at least p + 1 cells in the shortest column, it is natural to conjecture
that partially massless fields as well as the nonunitary massless fields with the gauge parameters ruled out in [10] are described in terms of pform gauge fields that are o(d 1, 2)modules
depicted by Young tableaux with at most p cells in the shortest column. The analysis of this problem, that we leave for a future publication, will provide the full list of unitary and nonunitary
free HS models described by gauge pform fields that take values in various finitedimensional
o(d 1, 2)modules.
Acknowledgements
The authors acknowledge with gratitude the collaboration of V. Didenko at the early stage of
this work. E.S. is grateful to K.B. Alkalaev for useful discussions. The work was supported in
part by grants RFBR No. 050217654, LSS No. 1578.20032 and INTAS No. 03516346. The
work of E.S. was supported in part by the scholarship of Dynasty Foundation.
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
143
Appendix A. Notation
A.1. Index conventions
Latin indices A, B, . . . = 0, 1, . . . , d are vector indices of the isometry algebra o(d 1, 2) or
o(d, 1). Indices a, b, . . . = 0, . . . , d 1 are vector indices of the ddimensional Lorentz algebra
o(d 1, 1). For definiteness, we consider the (A)dSd case of o(d 1, 2). For the decomposition
of irreps of o(d 1, 2) into irreps of o(d 1, 1) we use notations A = {a, }, where denotes
the extra value of the o(d 1, 2) vector index compared to the o(d 1, 1) vector index.
Underlined Latin indices m, n, k, . . . = 0, 1, . . . , d 1 are the indices of differential forms
in the spacetime base manifold. d = dx m m is the de Rham differential. To simplify formulae
we often omit the exterior differentials dx m and the wedge product symbol . The rank of a
differential form is indicated as a braced subscript. For example, {r} denotes a rform m1 ...mr .
AB and ab are, respectively, o(d 1, 2) and o(d 1, 1) invariant metrics. We use the
d+1
(A.1)
C [aa...a],...
is denoted
is used for
antisymmetric indices. More generally, symmetrized upper or lower indices are denoted by the
same letter. We use the normalization in which the symmetrization operator S is a projector
For example, for any vector V a and a totally symmetric rankn tensor C a(n)
S 2 = S.
C a(n),b,c,...
V a C a(n) =
C a[n],...
i=n+1
1
V ai C a1 ...a i ...an+1 ,
n+1
(A.2)
i=1
{s}
s( m1 m2 ...ms+1 ) (x),
(A.4)
(A.5)
144
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
2 {s} m m {s} ,
(A.6)
(A.7)
{s}
m m3 ...ms (x),
2
{s} n n m2 ...ms (x)k k m2 ...ms (x),
{s}
(s + 2)(s + 1)
(m1 m2 m3 ...ms+2 ) .
2
(A.8)
(A.9)
(A.10)
(A.11)
(B.1)
(B.2)
A scalar is symbolized by .
Different symmetry types of tensors are characterized by Young tableaux. One can use different bases in the vector space of tensors of a given type. The one we mostly use in this paper is
that with explicit symmetrizations. Alternatively, one can use the antisymmetric basis, which is
used in particular in Section 5.2 of this paper. (For more detail, see, e.g., [31].)
Since the metric and the LeviCivita tensor A1 ...Ad are invariant tensors of the algebra o(r, q),
to single out its irreps the tracelessness condition, the condition that the height of any column does not exceed [d/2] have to be imposed. For even d and appropriate signature the
(anti)selfduality condition has to be imposed on the tensor representations associated with Young
tableaux that have at least one column of height d/2.
The decomposition of the tensor product of any irrep of sld with its vector representation
admits a natural description in terms of Young tableaux. Namely, it contains all tableaux that can
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
145
(B.3)
In the case of o(r, q), Young tableaux are associated with traceless tensors of one or another
symmetry type. The tensor product decomposition results from either adding or cutting one box
in various ways. The cutted diagrams result from traces between the tensor product factors. For
example, the o(r, q) decomposition analogous to (B.3) is
(B.4)
(B.5)
In this paper, fields, gauge parameters and field strengths are pforms with fiber indices in
finitedimensional modules of o(d 1, 1) or o(d 1, 2), described by various Young tableaux.
Replacing the form index of a 1form by the fiber one with the help of the vielbein em a (e.g.,
a(3),bc = m a(3),b emc ), the resulting fiber tensor decomposes into irreps of o(d 1, 1) according to the general rule for the tensor product with the vector representation. For instance, one
obtains for a(3),bc
(B.6)
corresponds to m a(3),b em b .
In particular, here
Appendix C. Coefficients
The final result is
!
g(k, m) =
h(k, m)
!
G(k, m) =
H (k, m)
f (k, m) =
F (k, m) =
k+1
,
k(d + k 2)
km
,
(d + m 3)(k m + 1)
(k 1)(d + k 3)
,
k
(d + m 4)(k m + 2)
H (k, m 1)
,
km+1
h(k 1, m)
(C.1)
(C.2)
(C.3)
(C.4)
where
h(k, m) =
(C.5)
146
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
H (k, m) =
(C.6)
(C.7)
(C.8)
which insure that the operators +1,2 are conjugated to 1,2 with respect to the scalar product
(4.32). Also note that
h(k, m) = f (k + 1, m)g(k, m),
H (k, m) = F (k, m + 1)G(k, m).
(C.9)
(C.10)
References
[1] S. Deser, R.I. Nepomechie, Phys. Lett. B 132 (1983) 321;
S. Deser, R.I. Nepomechie, Ann. Phys. 154 (1984) 396.
[2] A. Higuchi, YTP8522;
A. Higuchi, Nucl. Phys. B 282 (1987) 397;
A. Higuchi, J. Math. Phys. 28 (1987) 1553;
A. Higuchi, J. Math. Phys. 43 (2002) 6385, Erratum;
A. Higuchi, Class. Quantum Grav. 6 (1989) 397;
A. Higuchi, Nucl. Phys. B 325 (1989) 745.
[3] I. Bengtsson, J. Math. Phys. 36 (1995) 5805, grqc/9411057.
[4] S. Deser, A. Waldron, Phys. Rev. Lett. 87 (2001) 031601, hepth/0102166;
S. Deser, A. Waldron, Nucl. Phys. B 607 (2001) 577, hepth/0103198;
S. Deser, A. Waldron, Phys. Lett. B 508 (2001) 347, hepth/0103255;
S. Deser, A. Waldron, Phys. Lett. B 513 (2001) 137, hepth/0105181;
S. Deser, A. Waldron, Nucl. Phys. B 662 (2003) 379, hepth/0301068;
S. Deser, A. Waldron, Conformal invariance of partially massless higher spins, hepth/0408155.
[5] Y.M. Zinoviev, On massive high spin particles in (A)dS, hepth/0108192.
[6] I.L. Buchbinder, D.M. Gitman, V.D. Pershin, Phys. Lett. B 492 (2000) 161, hepth/0006144.
[7] K. Hallowell, A. Waldron, Nucl. Phys. B 724 (2005) 453, hepth/0505255.
[8] L. Dolan, C.R. Nappi, E. Witten, JHEP 0110 (2001) 016, hepth/0109096.
[9] N.T. Evans, J. Math. Phys. 8 (1967) 170.
[10] R.R. Metsaev, Phys. Lett. B 354 (1995) 78;
R.R. Metsaev, Arbitrary spin massless bosonic fields in ddimensional antide Sitter space, hepth/9810231.
[11] E. Angelopoulos, M. Laoues, Rev. Math. Phys. 10 (1998) 271, hepth/9806100.
[12] L. Brink, R.R. Metsaev, M.A. Vasiliev, Nucl. Phys. B 586 (2000) 183, hepth/0005136.
[13] S. Ferrara, C. Fronsdal, Conformal fields in higher dimensions, hepth/0006009.
[14] M.A. Vasiliev, JHEP 0412 (2004) 046, hepth/0404124.
[15] F.A. Dolan, Character formulae and partition functions in higher dimensional conformal field theory, hepth/0508031.
[16] M.A. Vasiliev, Yad. Fiz. 32 (1980) 855.
[17] M.A. Vasiliev, Fortschr. Phys. 35 (1987) 741, Yad. Fiz. 45 (1987) 1784 (in Russian).
[18] V.E. Lopatin, M.A. Vasiliev, Mod. Phys. Lett. A 3 (1988) 257.
[19] M.A. Vasiliev, Nucl. Phys. B 616 (2001) 106, hepth/0106200;
M.A. Vasiliev, Nucl. Phys. B 652 (2003) 407, Erratum.
[20] K.B. Alkalaev, O.V. Shaynkman, M.A. Vasiliev, Nucl. Phys. B 692 (2004) 363, hepth/0311164.
[21] P.A.M. Dirac, Proc. R. Soc. London, Ser. A 155 (1936) 447.
[22] M. Fierz, W. Pauli, Proc. R. Soc. London A 173 (1939) 211.
[23] L.P.S. Singh, C.R. Hagen, Phys. Rev. D 9 (1974) 898.
E.D. Skvortsov, M.A. Vasiliev / Nuclear Physics B 756 [PM] (2006) 117147
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
[32]
[33]
[34]
[35]
[36]
[37]
147
Abstract
We consider some flat space theories for spin 2 gravitons, with less invariance than full diffeomorphisms.
For the massless case, classical stability and absence of ghosts require invariance under transverse diffeomorphisms (TDiff), h h + 2( ) , with = 0. Generic TDiff invariant theories contain a
propagating scalar, which disappears if the symmetry is enhanced in one of two ways. One possibility is
to consider full diffeomorphisms (Diff). The other (which we denote WTDiff) adds a Weyl symmetry, by
which the Lagrangian becomes independent of the trace. The first possibility corresponds to General Relativity, whereas the second corresponds to unimodular gravity (in a certain gauge). Phenomenologically,
both options are equally acceptable. For massive gravitons, the situation is more restrictive. Up to field
redefinitions, classical stability and absence of ghosts lead directly to the standard FierzPauli Lagrangian.
In this sense, the WTDiff theory is more rigid against deformations than linearized GR, since a mass term
cannot be added without provoking the appearance of ghosts.
2006 Elsevier B.V. All rights reserved.
1. Introduction
It has long been known that, in theories containing a massless spin 2 graviton propagating
in flat space, unitarity requires invariance under transverse diffeomorphisms. The argument
runs as follows [1]. Consider a graviton of momentum k travelling in the z direction. The little
group of Lorentz transformations which leave k invariant has three generators. One of them, Iz ,
corresponds to rotations in the x, y plane. The other two, I0x and I0y , correspond to transverse
* Corresponding author.
149
boosts combined with rotations in the x, z and y, z plane.1 This little group is isomorphic to the
Euclidean group in 2 dimensions. The standard helicity polarizations of the graviton2 h+ ih ,
transform unitarily under Iz (picking up phases exp(2i) under rotations of angle ) cf. [1,2].
But unitary representations of the noncompact translations I0i would be infinite dimensional,
leading to an infinite number of polarizations for given k. This catastrophic degeneracy is avoided
by declaring the equivalence of polarizations which are related to one another by standard gauge
transformation h h + 2k( ) . It can then be shown that the effect of I0i on the standard
helicity eigenstates is pure gauge, and in this sense I0i act trivially, producing no new states of
momentum k. The interesting point, however, is that the trace h = h is Lorentz invariant
(and hence invariant under I0i ), and therefore it is sufficient to consider the equivalence under
transverse gauge transformations, which do not affect the trace,
k = 0.
(1)
These form a subgroup which we shall refer to as transverse FierzPauli symmetry, or transverse
diffeomorphisms (TDiff).
This paper is devoted to the study of some flat space theories containing spin two, that have
less invariance than the full diffeomorphisms (Diff) of the standard FierzPauli Lagrangian. We
start our discussion, in Section 2, with the most general Lorentz invariant Lagrangian for a massless graviton. We show that a simple requirement of classical stability and absence of ghosts
leads directly to TDiff invariance (and no more invariance than that).
In the nonlinear regime, TDiff symmetry is realized in the socalled unimodular theories. The
best known example is Einsteins 1919 theory (cf. [2] for a recent reference), which corresponds
to the traceless part of the usual Einsteins equations. This theory can be obtained from an action
which is still generally covariant, but where the determinant of the metric is not dynamical [3].
The equations of motion are obtained from a restricted variational principle where
g = 0.
(2)
The tracefree equations enjoy the property that a cosmological term in the matter Lagrangian is
irrelevant. Nevertheless, the trace of Einsteins equations can be recovered with the help of the
Bianchi identity, and then a cosmological term reappears in the form
of an integration constant.
Alternatively, we may start with a variational principle where g is unrestricted, and with
an action which is invariant only under TDiff [4]. Generically, this leads to scalartensor theories, where the determinant of the metric plays the role of a new scalar. As we shall see, this
new degree of freedom can be eliminated by imposing an additional Weyl symmetry (by which
the action becomes independent of the determinant of the metric). Thus, Einsteins unimodular
theory can be thought of as the theory of massless spin two fields which is invariant under TDiff
plus certain Weyl transformations.3
Section 3 is devoted to the massive theory. Mass terms which preserve TDiff would give
mass to the new scalar, but not to the spin 2 polarizations. Graviton mass terms necessarily
break all invariance under diffeomorphisms. For massive particles, the little group is that of
1 These rotations correct for aberration of k under the respective boosts.
2 Up to overall normalization, h+ dx dx dy dy and h dx dy + dy dx.
3 Also, in the absence of this additional Weyl symmetry, the new scalar can be given a mass (which may be expected
from radiative corrections, since it is not protected by any symmetry). If the mass is large enough, this leads us back to
the situation where the extra scalar does not propagate at low energies, which is effectively equivalent to the scenario
described by Eq. (2).
150
ordinary spatial rotations O(3), which leads to finite dimensional unitary representations without
the need of invoking any gauge symmetry. Naively, one might think that this would increase the
arbitrariness in the choice of the kinetic term. Nevertheless, as we shall see, the only theory
with massive gravitons which is free from tachyons or ghosts is equivalent to the FierzPauli
theory, where the kinetic term is invariant under Diff (not just TDiff), and the mass term is of the
standard form m2 (h h h2 ) (see also [5]). In Section 4 we consider the propagator for TDiff
invariant theories, and the coupling to conserved matter sources. We conclude in Section 5.
Throughout this paper we will follow the LandauLifshitz timelike conventions; the ndimensional flat metric in particular, reads = diag(1, 1, . . . , 1). Lagrangians are written
in momentum space as well as in configuration space, depending on the context. It is usually
trivial to shift from one language to the other.
2. Massless theory
Let us begin our discussion with the most general Lorentz invariant local Lagrangian for a
free massless symmetric tensor field h ,
L = LI + LII + aLIII + bLIV ,
(3)
(5)
(6)
(7)
= 0.
(8)
with
151
Since LIII and LIV are (separately) invariant under this symmetry, the most general TDiff invariant Lagrangian has = 1, and arbitrary coefficients a and b:
LTDiff LA + aLIII + bLIV .
(9)
( = 1/n)
(10)
where n is the spacetime dimension and the condition = 1/n is necessary for the transformation to be invertible. Under this redefinition, the parameters in the Lagrangian (9) change
as
a a + (an 2),
b b + 2(nb a 1) + 2 bn2 n(2a + 1) + 2 .
(11)
Starting from a = b = 1, the new parameters are related by
b=
1 2a + (n 1)a 2
.
(n 2)
(12)
It follows that Lagrangians where this relation is satisfied are equivalent to FierzPauli, with the
exception of the case a = 2/n, which cannot be reached from a = 1 with = 1/n.
A second possibility is to enhance TDiff with an additional Weyl symmetry,
2
h = ,
(13)
n
by which the action becomes independent of the trace. In the generic transverse Lagrangian
LTDiff [h ] of Eq. (9), replace h with the traceless part
h h h (h/n) .
(14)
This is formally analogous to (10) with = 1/n, but cannot be interpreted as a field redefinition. As such, it would be singular, because the trace h cannot be recovered from h . The
resulting Lagrangian
LWTDiff [h ] LTDiff [h ],
(15)
is still invariant under TDiff (the replacement (14) does not change the coefficients in front of the
terms LI and LII ). Moreover, it is invariant under (13), since h is. Using (11) with = 1/n,
we immediately find that this WTDiff symmetry corresponds to Lagrangian parameters
n+2
2
b= 2 .
a= ,
(16)
n
n
This is the exceptional case mentioned at the end of the previous paragraph. Note that the densitized metric g = g 1/n g + h enjoys the property that g = 1. This is the starting
point for the nonlinear generalization of the WTDiff invariant theory, which is discussed in
Section 2.5.
It is easy to show that Diff and WTDiff exhaust all possible enhancements of TDiff for a Lagrangian of the form (3) (and that, in fact, these are its largest possible gauge symmetry groups).
Note first, that the variation of LI involves a term 2h . For arbitrary h , this will only cancel
152
against other terms in (5) provided that the transformation is of the form
2
,
n
for some and . The vector can be decomposed as
h = 2( ) +
(17)
= +
(18)
where
(b a)22 h + (2 a 1)2 h
L = ( 1)2 h +
2
+ (bn a 1)2h + (2 na) h .
(19)
n
TDiff corresponds to taking = 1, with arbitrary transverse and with = = 0. This symmetry can be enhanced with nonvanishing and satisfying the relation
n(a 1)2 = 2(2 an),
(20)
provided that
1 2a + (n 1)a 2
.
(21)
(n 2)
Eq. (20) ensures the cancellation of the terms with h , and Eq. (21) eliminates terms containing the trace h. Eq. (21) agrees with (12), and therefore the Lagrangian with the enhanced
symmetry is equivalent to FierzPauli, unless a = 2/n, which corresponds to WTDiff.4
b=
(22)
=
( )
.
h
n
h
Hence, the WTDiff equations of motion are traceless
SWTDiff [h]
0.
h
In the WTDiff theory, the trace of h can be changed arbitrarily by a Weyl transformation, and we
can always go to the gauge where h = 0. Likewise, in the familiar Diff theory we can choose a
gauge where h = 0. Then, h = h , and the WTDiff equations of motion (e.o.m.) are just the
traceless part of the FierzPauli e.o.m. Differentiating Eq. (22) with respect to x and using the
Bianchi identity
SDiff [h]
= 0,
h
4 Incidentally, it may be noted that for n = 2 both possibilities coincide, since in this case the symmetry of the Fierz
Pauli Lagrangian is full diffeomorphisms plus Weyl transformations.
153
(25)
(26)
(27)
2 = 0.
(28)
Taking the derivative of (27) with respect to , symmetrizing with respect to and , and using
(25) and (26), we have (n 2) 2 = 0. For n = 2 this implies = 0, i.e.
= b x + c,
where b and c are constants. Hence, not every Weyl transformation belongs to Diff, since only
the s which are linear in x qualify as such. Conversely, the subset of Diff which can be
expressed as Weyl transformations are the solutions of the conformal Killing equation for the
Minkowski metric [7],
1
CD
( )
= ,
n
(29)
154
(30)
where
8KDiff = k 2 + 2
k k + k k + k k + k k 2k k 2k k .
(31)
We can also define the traces
n 2
k k k 2 ,
4
(n 1)(n 2) 2
tr tr KDiff = KDiff =
k .
4
tr KDiff = KDiff =
(32)
t
8KDiff
= 8 KDiff tr KDiff ,
n
(33)
cannot be derived from a Lagrangian as it is not symmetric in the indices ( ) vs. (). Nevertheless, we can still define traceless symmetric Lagrangians. One might think of substituting
in the previous expression by tr KDiff , and dividing by its trace. However, this would be
nonlocal.
For a local Lagrangian which is still invariant under TDiff, we must restrict to deformations
which correspond to changes in the parameters a and b in (3). The most general symmetric
Lagrangian with these properties is of the form
KtDiff KDiff M M ,
(34)
with M a symmetric operator at most quadratic in the momentum. Asking that the result be
traceless leads to:
1
M = tr KDiff (tr M) ,
(35)
n
which implies
tr M =
1
tr tr KDiff .
2n
(36)
Therefore
M
1
1
=
tr KDiff (tr tr KDiff )
,
n
2n
155
(37)
8KtDiff = k 2 ( + ) (k k + k k + k k + k k )
2(n + 2) 2
4
(38)
k + (k k + k k ).
2
n
n
Moving back to the position space, this corresponds to the WTDiff Lagrangian, i.e. the case
in (9). As shown before, this yields the traceless part of the FierzPauli
a = n2 and b = n+2
n2
e.o.m. in the gauge h = 0.
A similar analysis could be done for the massive case. However, as we shall see in the next
section, the corresponding Lagrangian has a ghost.
2.4. Dynamical analysis of the general massless Lagrangian
The little group argument mentioned in the introduction indicates that the quantum theory
is not unitary unless the Lagrangian is invariant under TDiff. In fact, in the absence of TDiff
symmetry the Hamiltonian is unbounded below. This leads to pathologies such as classical instabilities or the existence of ghosts.
To show this, as well as to analyze the physical degrees of freedom of the general massless
theory (3), it is very convenient to use the cosmological decomposition in terms of scalars,
vectors, and tensors under spatial rotations SO(3) (see e.g. [8]),
h00 = A,
h0i = i B + Vi ,
(39)
(t)
i Vi
it
tii
1
L = t ij 2tij .
4
(40)
The vectors contribute both to LI and LII . Working in Fourier space for the spatial coordinates
and after some straightforward algebra, we have
(v)
2 1
2
1
L = k2 V i F i + ( 1) k2 F i + V i .
2
2
(41)
H=
(F + k2 V )2 [V + (1 )k2 F ]2 (1 )k4 F 2 k2 V 2
,
2(1 )
2
2
2k2
(42)
156
F ,
(43)
( 1)
F + k2 F =
(44)
V .
( 1)
For = 0 these are equations for forced oscillators. For large times, the homogeneous solution
becomes irrelevant and we have
Ct
exp i kt + 0 ,
V + ikF
( 1)k
V + k2 V =
whose amplitude grows without bound, linearly with time. This classical instability is not present
for = 0. However, in this case F and V decouple and we have
2 1
2
1
(v)
L=0 = k2 F i V i ,
2
2
so Vi are ghosts.
Hence, the only case where the vector Lagrangian is not problematic is = 1, corresponding
to invariance under TDiff. The scalar Lagrangian is then given by6
1
(s)
LTDiff = ( A)2 2k2 ( B)2 + N ( )2 2k2 E + k4 ( E)2
4
2
1
A + k2 B k2 B 2 k2 2 + 2k4 E k6 E 2 + 2k2 B k2 E
2
a
A N + k2 E A + k2 B k2 A N + k2 E B + k2 E
+
2
2
b
A N + k2 E ,
(45)
4
where N = n 1 is the dimension of space. It is easy to check that B is a Lagrange multiplier,
leading to the constraint
(N 1) = (a 1)h,
(46)
+ k2 E
where h = A N
is the trace of the metric perturbation. Substituting this back into the
scalar action (45) we readily find
(s)
LTDiff =
b
( h)2 ,
4
(47)
where
b b
1 2a + (n 1)a 2
.
n2
(48)
157
Hence, the scalar sector contains a single physical degree of freedom, proportional to the trace.
Whether this scalar is a ghost or not is determined by the parameters a and b. For b = (1 2a +
(n 1)a 2 )/(n 2), corresponding to the enhanced symmetries which we studied in the previous
subsection, the scalar sector disappears completely, and we are just left with the tensor modes.
2.5. Nonlinear theory
Nonlinear generalizations of TDiff invariant theories have been discussed in [4] (see also
[9]). The basic idea is to split the metric degrees of freedom into the determinant g, and a new
= 1. Note that g is a tensor density,
metric g = g1/n g , whose determinant is fixed g
and under arbitrary diffeomorphisms (for which g = 2( ) ) it transforms as
2
g = 2g ( ) g ,
(49)
n
where denotes covariant derivative with respect to g . Next, one defines transverse diffeomorphisms as those which satisfy
= = 0,
(50)
2
Here, is a scalar made out of the matter fields and g. Thus, the TDiff invariant theories can be
seen as unimodular scalartensor theories, where g plays the role of an additional scalar. These
are very similar to the standard scalartensor theories, except for the presence of an arbitrary
integration constant in the effective potential. Following [4], we may go to the Einstein frame by
defining g = 2 g , and we have
1
S=
(52)
gR[
g ] d n x + SM + d n x,
2
where
SM =
(n 1)(n 2)
n
n
g
g
L[,
,
g
d n x.
2 2
(53)
Here, we have first eliminated g in favor of , and we have then implemented the constraint
g = 2n [g, ] through the Lagrange multiplier (x). Note that the invariance under full diffeomorphisms which treat g as a metric and and as scalars is only broken by the last term
in (52). In particular, SM is Diff invariant, and since = , it is straightforward to show
158
(55)
with arbitrary integration constant (note that in this case g = g ). This coincides with the
standard Einsteins equations in the gauge g = 1. The same action can be expressed in terms of
the original metric g as
1
(n 1)(n 2)
d n x(g)1/n R[g ] +
SWTDiff =
(56)
ln
g
ln
g
.
2
4n2
This is invariant under Weyl transformations
g 2 (x)g ,
(57)
159
by
1
S=
(58)
gR + f (g) d n x.
2
Here, the second term breaks Diff to TDiff, and there is no Weyl invariance. However, the equations of motion will give
1
R Rg = gf
(g)g ,
2
and from the Bianchi
identities it follows that g is an arbitrary constant (except in the Diff invariant case when f g ), a situation identical to (55). It is unclear whether the action (58) is of
any fundamental significance, since the remaining TDiff symmetry does not forbid an arbitrary
function of g in front of R, and additional kinetic terms for g. Nevertheless, Lagrangians similar
to (58) do arise in the context of certain bigravity theories where the interaction term between
two gravitons breaks Diff Diff to the diagonal Diff times a TDiff symmetry [10].
To conclude, it should be stressed that it seems to be very difficult to determine from experiment whether Diff, WTDiff or just TDiff is the relevant symmetry. The difference between
WTDiff and the rest of TDiff theories is just the absence of the extra scalar. However, this scalar
may well have a mass comparable to the cutoff scale, and in this case it would not be seen at
low energies. Also, at the classical level, the WTDiff differs only from Diff in that the cosmological constant is arbitrary. Of course the measurement of this constant does not reveal too much
about its origin. Therefore, the only observable differences between both theories may be in
the quantum theory [2,3,1114].
3. Massive fields
Let us now turn our attention to the massive case. The most general mass term takes the form7
1
1
Lm = m21 h h + m22 h2 .
4
4
First of all, let us note that for m1 = 0, this mass term is still invariant under TDiff. The term
m22 h2 gives a mass to the scalar h, but not to the tensor modes tij , which are traceless. Hence,
the analysis of the previous section remains basically unchanged. If m22 > 0 is larger than the
energy scales we are interested in, the extra scalar effectively decouples, and we are back to the
situation where only the standard helicity polarizations of the graviton are allowed to propagate.8
When m1 = 0, we must repeat the analysis.9 With the decomposition (39), the Lagrangian for
the tensor modes becomes
1
(t)
L = t ij 2 + m21 tij ,
(59)
4
7 Here, we are disregarding the possibility of Lorentz breaking mass terms, which has been recently considered in
[10,15].
8 Note also that the addition of the term m2 h2 to both the Diff or the WTDiff Lagrangian does not change the propa2
gating degrees of freedom of the theory. The analogous statement in a nonlinear context is illustrated by the Lagrangian
(58), where a potential f (g) is added to a Diff invariant Lagrangian (something does change, though, by the addition
of the potential, since the new theory does have the arbitrary integration constant ). Hence, one may in principle construct classical Lagrangians which propagate only massless spin 2 particles, and whose symmetry is only TDiff, although
in this case radiative stability is not guaranteed (i.e. we may expect other terms, such as kinetic terms for the determinant
g, which are not protected by the symmetry, to be generated by quantum corrections).
9 For a similar analysis in terms of spin projectors see [5].
160
and in order to avoid tachyonic instabilities we need m21 > 0. For the vector modes, and for = 1,
the potential term
Hv =
m21 2 i 2 i 2
,
k F V
2
is added to (42). The contribution proportional to V 2 is negative definite. Hence, to avoid ghosts
or tachyons we must take = 1. In this case, V i does not appear in the Lagrangian and V i can
be eliminated in favor of F i . This leads to
1 k2 m21
(v)
L=
F i 2 + m21 F i .
(60)
2 k2 + m21
Out of the (N 2 + N 2)/2 polarizations of the massive graviton in N + 1 dimensions, (N 2
N 2)/2 of these are expressed as transverse and traceless tensors tij , and N 1 are expressed
as transverse vectors F i . The remaining one must be contained in the scalar sector. The scalar
Lagrangian can be written as
(s)
(61)
where the first term is given by (45) and the second is given by
m2
2
m21 2
A 2k2 B 2 + N 2 2k2 E + k4 E 2 + 2 A N + k2 E .
4
4
Variation with respect to B leads to the constraint
m21 B = (1 a) A + k2 E (1 aN).
(s)
Lm =
(62)
(63)
L=
W (h, U, V )
b 2 [N m21 2(N 1)k2 ]m21 2
,
h +
V U 2 +
4
4(N 1)
4(N 1)2
(64)
(65)
161
For 2(N 1)k2 < N m21 the variable U has negative kinetic energy, whereas for 2(N 1)k2 >
Nm21 the same is true of V . Thus, the Hamiltonian is unbounded below, unless
b = 0.
(66)
In this case, h is nondynamical, and it will implement a constraint between U and V provided
that the coefficient of h2 in W vanishes identically. This requires
1 + (1 4a + a 2 )N + a 2 N 2
m21 .
m22 =
(67)
(N 1)2
As discussed in Section 2, as long as a = 2/(N + 1), all kinetic Lagrangians with b = 0 are
related to the FierzPauli kinetic term by the field redefinition (10). Thus, there are only two possibilities for eliminating the ghost: either the kinetic term is invariant under Diff or it is invariant
under WTDiff.
3.1. Diff invariant kinetic term
Without loss of generality, we can take a = b = 1, and from (67) we have the usual FierzPauli
relation
m21 = m22 .
Variation with respect to h leads to the constraint
(N 1)k2 V = N m21 (N 1)k2 U.
(68)
(69)
L=
N
2 + m21 ,
4(N 1)
(70)
tii
(t)
h0i = Vi ,
hij = ij + tij ,
1
L = t ij 2 + m21 tij .
4
(71)
1
1
L = ( 1)Vi2 + m21 Vi2 ,
2
2
(72)
162
which is nondynamical in the present case because = 1. Likewise, it can easily be shown
that the scalars A an are nondynamical. Therefore, in the graviton rest frame the propagating
polarizations are represented by the [N (N + 1)/2] 1 independent components of the symmetric
traceless tensor tij .
3.2. WTDiff invariant kinetic term
For a = 2/n = 2/(N + 1), the last term in Eq. (65) disappears, and U and V do not mix with
h. Because of that, there are no further constraints amongst these variables and the ghost in the
kinetic term in (64) is always present for m21 = 0. This means that the WTDiff theory cannot be
deformed with the addition of a mass term for the graviton without provoking the appearance of
a ghost.
Note that this is so even in the case of a mass term compatible with the Weyl symmetry, i.e.
m21 = nm22 . This relation causes h to disappear from the Lagrangian, but of course it does nothing
to eliminate the ghost.
4. Propagators and coupling to matter
In this section we shall consider the propagators and the coupling to external matter sources,
for the different troublefree Lagrangians which we have identified in the previous sections.
On one hand, we have the standard massless and massive FierzPauli theories, which have
been thoroughly studied in the literature. There are also the generic ghostfree TDiff theories,
which satisfy the condition
b b
1 2a + (n 1)a 2
< 0.
n2
(73)
These may include a mass term of the form m2 h2 , which affects the scalar mode but does not
give a mass to the tensor modes. The WTDiff invariant theory completes the list of possibilities.
Throughout this section, we will make use of the spin two projector formalism of [16], which
is very useful in order to invert the equations of motion. The properties of these projectors are
summarized in Appendix B.
4.1. Gauge fixing
As noted in [2], for the TDiff gauge symmetry there is no linear covariant gauge fixing condition which is at most quadratic in the momenta. This is in contrast with the FierzPauli case,
where the harmonic condition contains first derivatives only. The basic problem is that a covariant gaugefixing carries a free index, which leads to n independent conditions. This is more than
what transverse diffeomorphisms can handle, since these have only (n 1) independent arbitrary
functions. To be specific, let us consider the most general possibility linear in k,
M h = 0,
(74)
where
M = a1 ( k ) + a2 k .
(75)
(76)
163
However, deriving the r.h.s. of the previous expression with respect to x and summing in , this
terms cancels, which implies that the integrability condition
M h = 0,
(77)
must be satisfied. This simply means that the gauge condition cannot be enforced on generic
metrics.
It is plain, however, that the transverse part of the harmonic gauge (which contains only n 1
independent conditions) can be reached by a transverse gauge transformation. The corresponding
gauge fixing piece is obtained by projecting the harmonic condition with k 2 k k k 2 :
2
1
h 2 h .
(78)
2M 4
The gauge fixing parameter is now dimensionful, and this has been explicitly indicated by denoting it by M 4 . A study of this kind of term and its associated FP ghosts and BRST transformations
can be found in [13].
By contrast, in the case of WTDiff, the additional Weyl symmetry allows for the use of gauge
fixing terms which are linear in the derivatives (such as the standard harmonic gauge).
Lgf =
4.2. Propagators
The generic Lagrangian can be written in Fourier space as
L = LI + LII + aLIII + bLIV + Lm + Lgf
1
= h K h
4
1
= h k 2 m21 P2 + (1 )k 2 m21 + 2 (k) P1
4
+ as P0s + aw P0w + a P0
h ,
(79)
where P1 and P2 are the projectors onto the subspaces of spin 1 and spin 0 respectively, while the
operators P0s , P0w and P0 P0sw + P0ws project onto and mix the different spin 0 components.
The definitions and properties of these operators are discussed in Appendix B. The coefficients
in front of the spin 0 projectors are given by
as = 1 (n 1)b k 2 m21 + (n 1)m22 ,
aw = (1 2 + 2a b)k 2 m21 + m22 ,
a = n 1 (a b)k 2 + m22 .
(80)
In (79), we have included the term 2 (k)P1 which can be used to gauge fix the TDiff symmetry
whenever it is present. Indeed, (78) can be written as
Lgf = 2 (k)h P1
h ,
(81)
where 2 (k) = (1/4M 4 )k 6 . Even though we are primarily interested in the TDiff Lagrangian
(which corresponds to = 1), we have kept generic throughout this subsection. This can
be useful to handle the cases with enhanced symmetry, since a generic arises, for instance,
from the conventional harmonic gauge fixing term (as we shall see below). When invertible,
164
k2
P2
P1
1
+
aw P0s + as P0w a P0 ,
+
2
2
2
2
m1 (1 )k m1 + (k) g(k)
where,
2
g(k) = as aw a
.
(82)
(83)
(84)
this coupling is invariant under TDiff for all values of 1 and 2 . Moreover, it is Diff invariant
when 2 = 0, and WTDiff invariant for the special case 1 = n2 . The interaction between
sources is completely characterized by [17]
1
1
d n k Lint (k) =
d n k Ttot (k) Ttot (k) .
Sint
(85)
2
2
From the properties of the projectors Pi listed in Appendix B, it is straightforward to show
that
P2
T + P0 T 2 ,
Lint (k) = 12 T
(86)
k 2 m21
where the operator
2
1 aw
1
a
+ 22 (n 1)aw + as 2 n 1a
P0 =
+ 21 2 aw
g(k) (n 1)
n1
(87)
encodes the contribution of the spin 0 part. We are now ready to consider the different particular
cases, which we present by order of increasing symmetry.
4.3. Massive FierzPauli
In this case the parameters in the Lagrangian are given by = a = b = 1 and m21 = m22 . From
(82), we have
g(k) = (n 1)m42 ,
which does not depend on k. Because of that, the denominator of the operator P0 does not
contain any derivatives. Its contribution to Eq. (86) corresponds only to contact terms, which
do not contribute to the interaction between separate sources. We are thus left with the spin 2
interaction, which ignoring all contact terms, can be written as
12
P2
1
2
2
Lint = 1 T
(88)
T  .
T =
T T
(n 1)
k 2 m21
k 2 m21
The factor 1/(n 1) is different from the familiar 1/(n 2) which is encountered in linearized
GR, and produces the well known vDVZ discontinuity in the massless limit [1820].
165
=
.
(90)
g(k) (n 2)m22 k 2 k 2 m22
b
Substituting in (87), and disregarding contact terms, we obtain
2
12
1
1a
1
P0 =
+
.
2
1
2
2
(n 1)(n 2) k
n2
bk m22
(91)
Substituting in (86) and adding the contribution of P2 for m21 = 0, which can be read off form
(88), we have
2
T 2
1
1a
1
T
.
T 2 2 2 +
1
Lint = 12 T
(92)
(n 2)
n2
k
bk 2 m22
Note that the massless propagator in (91) combines with the second term in the spin 2 part to
give the factor 1/(n 2) in front of T 2 . Eq. (92) shows that the massless interaction between
conserved sources is the same as in standard linearized general relativity.
In addition, there is a massive scalar interaction, with effective mass squared
m2eff =
m22
>0
b
(93)
(note that both parameters m22 and b must be negative, according to our earlier analysis), and
effective coupling given by
2
1
1a
2
=
2 +
1 .
eff
(94)
b
n2
These are subject to the standard observational constraints on scalar tensor theories. If the scalar
field is long range, then the strength of the new interaction has to be very small eff 105 1 [21,
22]. Alternatively, the interaction could be rather strong, but short range, shielded by a sufficiently
large mass meff (30 m)1 [2123].
4.5. Enhanced symmetry
From general arguments, the interaction between sources in the WTDiff theory is expected
to be the same as in standard massless gravity, since both theories only differ by an integration
constant but have the same propagating degrees of freedom.
166
In fact the result for WTDiff can be obtained from the analysis of the previous section by
setting b = 0. In this case, the term m22 h2 can be thought of as the additional gauge fixing
which removes the redundancy under the additional Weyl symmetry. With b = 0 the second
term in (92) becomes a contact term, and we recover the same result as in the standard massless
FierzPauli theory [17],10
1
1
Lint = 12 T
T
(95)
T 2 2 ,
(n 2)
k
as expected.
Note that in the Diff and WTDiff invariant theories, there is a different possibility for gauge
fixing. Rather than using the term (81) in order to take care of the TDiff part of the symmetry,
and then the m22 h2 to take care of the Weyl part, we can gauge fix the entire symmetry group with
a standard term of the form
2
Lgf = h + h ,
(96)
4
where and are arbitrary constants. This can be absorbed in a shift of the parameters a, b
and
2
,
.
2
2
With these substitutions, the propagator becomes invertible, even if it is not for the original values
of a, b and which correspond to Diff or to WTDiff. Needless to say, the result calculated in
this gauge coincides with (95).
a a + ,
b b
5. Conclusions
In this paper we have expanded somewhat the classification of flat space spin 2 Lagrangians
given by van Nieuwenhuizen in [5]. For the massless theory, we have explicitly shown that unless
the TDiff symmetry is imposed, the Hamiltonian is unbounded below and a classical instability
generically develops. We have also presented in some detail a few potentially interesting theories
which are invariant under this symmetry.
Generic massless TDiff theories contain a propagating scalar proportional to the trace h (note
that this is gauge invariant under TDiff), which disappears when the symmetry is enhanced
in one of two ways. The standard choice is to consider the full group of diffeomorphisms Diff.
Another possibility (which we call WTDiff) is to impose an additional Weyl symmetry, by which
the action depends only on the traceless part of the metric h = h (1/n)h . This theory
is equivalent to one in which the determinant of the metric g = + h is kept fixed to unity.
In practice, however, it may be convenient to use the formulation in which this extra symmetry
is present, since it can make the covariant gauge fixing somewhat simpler. Nonlinear extensions
of the TDiff theory have been discussed in [4], and they correspond to scalartensor theories with
an integration constant. The nonlinear extension of the WTDiff theory is a particular case where
the additional scalar is not present.
It is sometimes claimed that general relativity is the only consistent theory for spin 2 gravitons.
Such discussions [2428], however, always assume linearized Diff invariance as an input. In the
10 Note also that the WTDiff invariant coupling to conserved sources requires = n . Using this and a = 2/n in
1
2
(94) we have eff = 0, which again eliminates the scalar contribution.
167
light of the present discussion, it seems that linearized TDiff or WTDiff invariances should be
just as good starting points.
In fact, it may be very difficult to distinguish between the various options experimentally. The
additional scalar may be very heavy, in which case it can be integrated out leaving no distinctive
traces at low energies. As for the cosmological constant, it seems clear that we cannot tell from
its measurement whether it corresponds to an integration constant or to a fixed parameter in the
Lagrangian.
An interesting difference between linearized GR and TDiff invariant theories is that, as we
have shown, the latter cannot be extended with a mass term for the graviton without provoking the
appearance of a ghost. In this sense, the WTDiff theory is more rigid against small deformations
than the standard linearized GR.
It is at present unclear whether Diff (rather than TDiff, or WTDiff) is a fundamental symmetry
of Nature. Even in string theory, the connection with GR is onshell, which does not seem to
exclude the fundamental symmetry from being WTDiff (see however [29] for a discussion in
the context of closed string field theory). Classically the two theories are almost identical, but
there may be important differences in the quantum theory [2,3,1113]. These deserve further
exploration, and are currently under research [14].
Acknowledgements
This work was begun while one of us (E.A.) was a guest at the Departamento di Fisica at
the University of Roma I (La Sapienza). He is indebted to all members of this group for their
wonderful hospitality. J.G. thanks Jaume Gomis and Alex Pomarol for useful discussions, and
the members of the theory group at the University of Texas at Austin for warm hospitality during
the last stages of this project. This work has been partially supported by the European Commission (HPRNCT20000148) and by research programmes FPA200304597 and FPA200404582
(MEC, Spain) and DURSI 2005SGR00082. The work of D.B. has been supported by FPU20020501 (MEC, Spain).
Appendix A. TDiff Lagrangians in terms of gauge invariant quantities
As the Lagrangian of (9), LTDiff , is invariant under transverse transformations, one should be
able to write it in terms of invariants under this transformations (for the Diff case see e.g. [8,30,
31]). It is easy to see that under a general transformation h h + 2( ) the fields of the
cosmological decomposition transform as
tij tij ,
Vi Vi + 0 iT ,
B B + 0 + 0 ,
Fi Fi + iT ,
E E + 2,
A A + 20 0 ,
,
,
.
n
n
For general transverse transformations the only gauge invariant combinations are
A A +
tij ,
wi = Vi 0 Fi ,
(A.1)
168
= (A E),
(A.2)
for the scalars. In terms of these combinations, the tensor, vector and scalar part of the Lagrangian
(9) can be written as (we write also the mass term LV = m2 h2 )
1
1
(v)
LTDiff = t ij 2tij ,
LTDiff = w i w i ,
4
2
1
(s) I (s) II
2
L + L = ( 2) 2( ) + (n 3) + (n 1) 2 ,
4
a
(s) III
L =
(n 1) ( ) ,
4
2
b
(s) IV
L =
(n 1) + (n 1) (n 1) ,
4
2
2
m
(s) V
L =
(n 1) ,
4
where = i i i = i i . From this decomposition we easily see that is a Lagrange multiplier whose variation yields the constraint
1 (n 1)a (1 a) = 0.
(A.3)
(t)
In the case of general diffeomorphisms (a = b = 1), only two scalar combinations are gauge
invariant, namely and . Thus, the Lagrangian for the scalar part can be expressed as
(2 n)
2 + (n 1) 2 + (n 3) .
(A.4)
4
Concerning the Weyl transformations, we can write only two scalar invariants which are also
scalars for transverse transformations,
(s)
LDiff =
= + ,
= + .
2
n,
b=
(A.5)
n+2
,
n2
1
(n 2) 2n (n 1) 2 3n + n2 2 .
2
4n
Thus, varying the Lagrangian with respect to we find the constraint
(s)
LWTDiff =
= 0.
(A.6)
(A.7)
(s) V
m2
( n)2 .
4
(A.8)
k k
,
k2
k k
.
k2
(B.1)
169
and then define projectors on the subspaces of spin two, spin one, and the two different spin
zero components, labelled by (s) and (w). We introduce also the convenient operators that map
between these two subspaces.
1
1
,
P2 ( + )
2
(n 1)
1
P0s
,
(n 1)
P0w ,
1
P1 ( + + + ),
2
1
1
sw
P0ws
,
.
P0
(n 1)
(n 1)
(B.2)
(B.3)
1
.
tr P0sw = n 1 ,
tr P0ws =
tr P1 = 0,
n1
Apart from the previous expressions, these projectors satisfy
1
P2 + P1 + P0w + P0s = ( + )
2
and any symmetric operator can be written as
K = a2 P2 + a1 P1 + aw P0w + as P0s + asw P0 ,
(B.4)
(B.5)
(B.6)
where P0 = P0sw + P0ws . The inverse of the previous operator is easily found from (B.3) to be
K 1 =
ws
1
1
as
aw
asw
P0 + P0sw
P2 + P1 +
P0w +
P0s
2
2
2
a2
a1
as aw asw
as aw asw
as aw asw
(B.7)
2 never vanishes.
provided that the discriminant as aw asw
References
[1]
[2]
[3]
[4]
[5]
[6]
[7]
[8]
[9]
J.J. van der Bij, H. van Dam, Y.J. Ng, Physica A 116 (1982) 307.
E. Alvarez, JHEP 0503 (2005) 002, hepth/0501146.
W.G. Unruh, Phys. Rev. D 40 (1989) 1048.
W. Buchmuller, N. Dragon, Phys. Lett. B 207 (1988) 292.
P. Van Nieuwenhuizen, Nucl. Phys. B 60 (1973) 478.
M. Fierz, W. Pauli, Proc. R. Soc. London A 173 (1939) 211.
R.M. Wald, General Relativity, Univesity of Chicago Press, Chicago, 1984.
V.F. Mukhanov, H.A. Feldman, R.H. Brandenberger, Phys. Rep. 215 (1992) 203.
J.B. Pitts, W.C. Schieve, Gen. Relativ. Gravit. 33 (2001) 1319, grqc/0101058.
170
[10]
[11]
[12]
[13]
[14]
[15]
[16]
[17]
[18]
[19]
[20]
[21]
[22]
[23]
[24]
[25]
[26]
[27]
[28]
[29]
[30]
[31]
Abstract
The holographic principle in the PPwave limit proposed in our previous works is further confirmed by
studying impurity nonpreserving processes which contain a fermionic BMN operator with one scalar and
one fermion impurities. We show that the previously proposed duality relation between the matrix elements
of the three point interaction Hamiltonian in the holographic string field theory and the OPE coefficients
in superYangMills theory holds to the leading order in the large limit. Operator mixing is required to
obtain the BMN operator of definite conformal dimension which corresponds to the string state with one
scalar and one fermion excitations. The mixing term plays a crucial role for our duality relation to be valid.
Our results, combined with those in the previous papers, provide a positive support that our duality relation
holds for the general process regardless of the kind of impurities and of whether impurities conserve or not.
2006 Elsevier B.V. All rights reserved.
1. Introduction
Holography is the key concept to the duality between string theory and gauge theory in the
AdS/CFT correspondence. The holographic relation proposed by GKP/W [1] relates the degrees
of freedom in the string theory defined in the bulk AdS space and those in the gauge theory
defined on the conformal boundary of the AdS spacetime. Based on this proposal, a large amount
of work has been done giving many positive supports for this relation [2]. In particular, the nonrenormalization property of chiral primary operators has been confirmed [3]. However in the
172
BMN limit of the AdS/CFT correspondence [4], how this holographic relation is realized has not
been clear because of the lack of the clear spacetime picture.
In the previous papers [58], we gave an answer to this problem. We showed that the PPwave background emerges as the geometry around a tunneling null geodesic, which starts from a
point on the AdS boundary and returns to another point on the boundary. This trajectory can be
obtained as a dominant path in the large angular momentum limit for the path integral which is
expected to give the two point function of gauge theory operators in the GKP/W prescription.1
The string/gauge duality in the PPwave limit can be understood as the correspondence between
the oscillation modes of strings around this trajectory and gauge theory operators sitting around
its endpoints on the boundary. This picture allows a natural correspondence between an Smatrix
calculation on the string theory side and an operator product expansion on the gauge theory side.
As for three point functions, with the assumption that the free string basis corresponds to the
BMN operators of definite conformal dimension, we proposed a duality relation between the
matrix elements of the string interaction Hamiltonian and the OPE coefficients.
Our proposal in [7] has an advantage over the previous two types of duality relations, presented in [10,11] and [12], in that it is applicable to impurity nonpreserving cases. Indeed, we
checked that our duality relation holds for impurity nonpreserving processes which consist of
BMN operators with only bosonic impurities in the previous paper [8], as well as for the impurity
preserving processes [7]. To establish the duality relation for impurity nonpreserving processes
is important for studying the unsolved problem of string/gauge correspondence at higher genus
level, since impurity nonpreserving processes are known to appear as intermediate states of a
string loop amplitude [1315].
The aim of this paper is to show that our proposal also holds for the impurity nonpreserving
cases involving fermionic impurities. Checking our duality relation for such processes is important because our proposal was originally obtained as an extension of the duality relation for
chiral primary operators to the general string excitations. It is not trivial this relation is satisfied
for the nonBPS sector in general. Since we have already checked our proposal for impurity
preserving sector with several different kinds of impurities, including bosons and fermions, and
for impurity nonpreserving sector with only bosonic impurities, what remains to be studied is
impurity nonpreserving processes which contain fermionic impurities. Though it is desirable to
study the general processes, we restrict ourself in this paper to simple impurity nonconserving
processes, where the fermionic BMN operators involved contain only one scalar and one fermionic impurities. Even for these simple examples, our duality relation holds in a very nontrivial
way.
When studying the duality relation, it is important how to relate the basis between the two
theories. In our prescription, the correspondent on the gauge theory side to the free string basis
is a set of BMN operators of definite conformal dimension. As in the case of the singlet sector
of several two impurity states [1619], the operator mixing is required for the fermionic BMN
operator we are now considering so that it obtains a definite conformal dimension. We will see
that this operator mixing plays a crucial role for our duality relation to hold. In addition, the
phase factor of the BMN operator must be properly chosen up to O(1/J ) in order to give the
correct world sheet momentum dependence.
The result of this paper, in addition to those in the previous papers, strengthens the expectation
that our duality relation is valid for general processes regardless of the kind of impurities and of
1 See also [6,9] for more general discussions.
173
whether impurities conserve or not. We believe that the successful clarification of the duality
relation at the first order of the string interaction lays the foundation for the study of the duality
relation at higher genus level.
The paper is organized as follows. In Section 2 we briefly review our proposal for the duality
relation. In Section 3 we calculate two examples of the string amplitude for fermionic impurity
nonpreserving processes. In Section 4 we first identify the gauge theory operator which corresponds to the string states with one scalar and one fermion excitations. We show by explicit
perturbation calculations that for the corresponding BMN operator to obtain a definite conformal dimension, in addition to the operator with one fermion and one scalar impurities, another
operator with the same classical dimension and Rcharge is required. With this operator, we compute three point functions which correspond to the string amplitudes calculated in the previous
section. We will see that our duality relation holds to the leading order of the effective gauge
coupling and the operator required for the diagonalization plays a crucial role. We conclude in
Section 5. Appendix A is devoted to the details of the perturbative calculations required for the
two point function of the BMN operators with one scalar and one fermion impurities. We list
the explicit forms of the prefactors in Appendix B and the asymptotic behavior of the Neumann
coefficients in the large limit in Appendix C.
2. Review of the holographic duality relation
Our duality relation at the first order of the string interaction relates the basic ingredients
which characterize the three point interaction on each theory side, the matrix element of the 3point interaction Hamiltonian H123 on string theory side and the OPE coefficient C123 on gauge
theory side. Identifying the free string basis with the set of BMN operators of definite conformal
dimension, we propose that the duality relation is given by
2 +3 1
2
1
2 + 3 1
J2 J 3
f
+ 1 H123 ,
C123 =
(2 + 3 1 )
J1
2
(2.1)
where r and Jr are the conformal dimension and the U(1) Rcharge, respectively, of the gauge
theory operator Or (r = 1, 2, 3) which corresponds to the string state r, Jr satisfying the re(r)
(r)
lation H2 / = r Jr . Here, H2 and are the lightcone free Hamiltonian and the mass
parameter of the string theory, respectively, and J denotes the angular momentum around the
large circle of S 5 on the string theory side. The effective gauge theory coupling is related as
2 N/J 2 = 1/(p + )2 in the PPwave limit of AdS/CFT correspondence. The duality
= gYM
relation (2.1) is expected to hold to all orders with respect to , as long as we properly identify
the BMN operators of definite dimension, but we restrict ourself here to the leading order of
expansion for simplicity. The definition of f in (2.1) is given by some combination of Neumann
coefficients (see Ref. [7]). In what follows, we need only its asymptotic behavior in the large
limit which is given by
f
J1
J2 J3
,
J1
4(1) 
(2.2)
where (r) denotes (r) pr+ and we take (1) (= (2) (3) ) < 0, (2) , (3) > 0. The relation (2.1) is expected to hold to all order of = 1/((1) )2 , but we restrict ourself to checking
the relation to the leading order in the large limit in this article.
174
O1 (x1 )O2 (x2 )O3 (x3 ) =
C123
,
(2.3)
under the normalization of O1 O2 = 12 /x12 2 . It is known that for the impurity preserving
sector, we should take into account the operator mixing of single and double trace operators
[20,21]. However, for the impurity nonpreserving process, the effect of such mixing is suppressed by 1/N , so we need not consider this type of operator mixing as long as we are interested
in the planar limit. The operator mixing we should consider later to obtain the BMN operator of
definite conformal dimension is the one among singletrace operators with the same conformal
dimension and Rcharge.
The H123 is the matrix element of the 3string interaction Hamiltonian defined as
J1 J2 J3
H3 h ,
H123 = (1) 1(2) 2(3) 3
(2.4)
N
under the canonical normalization of the string field theory action such as
1
1
 + H
2  ,
 
S2 = d 
(2.5)
2
2
J J J
1
(2) (3)  1 2 3 H3 h + h.c.,
d (1) 
S3 =
(2.6)
2
N
1
(r)
H2 =
(2.7)
n(r) n(r) n(r) + n(r) n(r) , with n(r) = n2 + ((r) )2 .
(r)  n=
We proposed that the specific form of the string interaction Hamiltonian H3 h is the equal
weight sum of the two interaction vertexes presented so far,
1
1
H3 h = H3 SV + H3 D ,
(2.8)
2
2
where H3 SV is constructed in [22,23] and further developed in [2427] as a generalization of the
wellknown interaction Hamiltonian in the flat space to the PPwave background, while H3 D
is proposed in [12] and its prefactor takes the form of energy difference. These two vertexes
both satisfy the SUSY algebra, as well as, the continuity condition and momentum conservation. The interaction Hamiltonian of the latter type is not adopted in the flat space case because
it does not affect Smatrix calculations due to the energy conservation, but in our formalism
where worldsheet time is naturally Wickrotated, it can give indispensable contribution. The
combination (2.8) is required so that the bosonic zeromode sector should properly reduce to the
form which is obtained by taking the BMN limit of the effective action for the fluctuation fields
around AdS background which corresponds to the chiral primary operators.2 Though in [29] it
is claimed that the H3 D part has to be modified in order to respect the U(1)Y symmetry in the
supergravity sector, the modification gives no effect for the process we are now considering, as
for all the cases we have confirmed so far. We will discuss the processes in the nonBPS sector
for which such modification cannot be ignored in the future study [30].
2 In the scalar impurity sector this vertex reduces to the phenomenological interaction vertex proposed in [28], which
is successfully related with the genuine OPE coefficient through the relation (2 + 3 1 )C123 = H123 in the
impurity preserving process.
175
For the impurity preserving processes, where 2 + 3 1 = 0 + O( ), the duality relation (2.1) reduces to a simple form (2 + 3 1 )C123 = H123 , the one first conjectured
in [31] from a different viewpoint. We checked that this relation holds to the leading order
in the large limit for several nontrivial processes including bosonic and fermionic impurities in the previous paper [7]. For the impurity nonpreserving cases, the factor f in the
duality relation plays a crucial role: due to the dependence of Neumann coefficients, the
asymptotic behavior of H123 in the large limit is suppressed by = 1/(p + )2 as the difference in the number of the impurities between in and outstates increases, while the factors
(f J2 J3 /J1 )(1 2 3 )/2 give a compensating contribution. Indeed, in [8], we showed that the
contributions from each factor on the righthand side of (2.1) nicely combined to give the OPE
coefficients C123 which has an appropriate dependence. In what follows, we confirm that this
is also the case for the impurity nonpreserving processes which contain fermionic BMN operators.
3. String theory calculation
We consider fermionic impurity nonpreserving processes on the string theory side. In this
paper we restrict ourself to simple but nontrivial processes which consist only of one boson and
one fermion impurities such as
(3)j
(2)i (2)
1 2 ,m + n
vac m
(1)j
n(1)i
n1
1
(2)
n(2)i
1 2 ,n2
2
(3)
,
1 2 ,n
(3.1)
(3)j (3)
+ n3 1 2 ,n3 .
(3.2)
Even for these simple processes, we will see that the characteristic property that the interaction
vertex is given by the equal weight sum of H3 SV and H3 D , and the nontrivial dependence
on the righthand side of (2.1) play important roles for our duality relation to hold. In (3.1)
(r)i
(r)
and (3.2), n (i = 1, 2, 3, 4) and 1 2 ,n denote the nth bosonic excitation mode in a scalar
direction and the nth fermionic excitation mode, respectively, in the exponential basis of rth
string which are defined in terms of sin/cos oscillators, an /an (or bn /bn for a fermion), such
as
1
n = (an ian ),
2
1
n = (bn ibn ),
2
0 = a0 ,
0 = b 0 ,
1
n = (an + ian ),
2
1
n = (bn + ibn ).
2
(3.3)
(3.4)
The SO(8) spinor indexes is decomposed as SO(8) = (SU(2) SU(2)) (SU(2) SU(2)), and
the subscripts of 1 2 ,n denotes (2, 1) (2, 1) sector of the decomposition 8s = (2, 1) (2, 1)
(1, 2) (1, 2), while we denote the (1, 2) (1, 2) sector as 1 2 ,n .
(3)j
(2)i (2)
3.1. vac m 1 2 ,m + n
(3)
1 2 ,n
The matrix element of the interaction vertex for the process (3.1) is given by
(3)
(3)j
v1 2 ,n n
(2)
(2)i
1 2 ,m m
H3 h ,
where H3 h is the equal weight sum of the H3 SV and H3 D such as in (2.8).
(3.5)
176
Let us first calculate the contribution from H3 D of the interaction Hamiltonian H3 h . Noting
(r)
that the prefactor is given by 2 3r=1 n /(r) 4 due to the form of the prefactor, H3 D =
(2)
(3)
(1)
(H2 + H2 H2 )E, we can easily see that the contribution from the vertex H3 D is given
by
23
i 32
ij
Qnm Q23
mn Nmn
1 1
2 2 . (3.6)
2
Here we used the form of the overlap vertex state such as
(3)j
(3)
v1 2 ,n n
(2)
(2)i
1 2 ,m m
H3 D = 4
Ea = exp
n,1 2
n,1 2
(3.7)
(3.8)
(r) rs (s)
m
Nmn n
v,
(3.9)
r,s=1 m,n=
where Eb0 is the overlap vertex including the fermionic zeromode sector whose detailed form
does not concern the present calculation, and the definition of Qrs
mn is
(s)  1/2 1/2 rs 1/2 1/2
rs
Qmn = e((r) )
(3.10)
C(s) mn ,
U C N C
(r)  (r)
with U(r) = C 1 (C(r) (r) ), and Cmn = mmn , C(r)mn = m(r) mn . The details are referred
23
to Ref. [26]. In the large limit, the form of Q32
nm and Qmn are given in terms of Neumann
coefficients such as
m
n
23
Q23
(3.11)
,
Q32
N mn
N 32 .
mn =
nm =
y(1) 
(1 y)(1)  nm
With this relation and the asymptotic behavior in the large limit of the bosonic Neumann
23 = N
32 , which is listed in Appendix C, the contribution from the interaction
nm
coefficient N mn
vertex H3 D is given by
(3)
(3)j
(2)
(2)i
v1 2 ,n n 1 2 ,m m
H3 D
m
n
1
ij .
= i
2
2
y
1 y 8 (1) 3 y(1 y) 1 1 2 2
or [][],
(3.12)
where [XY ] and (XY ) denote the contractions of X and Y through the prefactor and overlap,
respectively. For example, []() means vPSV v vEb . There are no combination from ()(), [](), etc. because of the form of the prefactor PSV presented in
Appendix B.
177
Using the property that, for purely bosonic external states, the interaction vertex H3 SV can
be rewritten as
H3 SV = H3 D XII2 E,
3
(r)
(r) (s)
m rs
(r) (s)
rs
N N mn
m n + m m E
(r) mn
XII2 E =
r,s=1 m,n=1
(r)
3
(r) (s)
n(s) rs
m
rs
m n E,
+
Nmn N mn
+
(r)
(s)
(3.13)
r,s=1 m,n=1
where XII is a bosonic constituent of the prefactor whose explicit form is given in Appendix B,
we can see that the contribution from []() is given by the product of two contributions,
(2)i (3)j
(2)i (3)j
vm
n PSV E = vm n 2 XII2 E
(2)
(3)
ij m
m 23
23
23
= 2N mn
ij +
+
Nmn N mn
2 (2)
(3)
23
ij
= 2N mn ,
(3.14)
i
(3)
(2)
23
v1 2 ,n 1 2 ,m E =
1 1
2 2
(3.15)
Q32
nm Qmn .
2
Then, the net result of []() is half of the contribution from H3 D ,
23 23
ij
[]() = i Q32
nm Qmn Nmn
1 1
2 2
m
n
1
ij .
= i
2
2
y
1 y 8 (1) 3 y(1 y) 1 1 2 2
(3.16)
As for the [][] contribution, the only relevant term in the prefactor PSV is
i
i
j
K i K j Yij2 = XIi XII Yij2 .
4
2
(3.17)
For the definition of XI , XII , and Yij2 , being bosonic and fermionic constituents of the prefacij
tor PSV , see Appendix B. Noticing that Yij2 = Y 1 2 Y 1 2 1 1 is defined in terms of cos mode
(r) (r)1 2 and that the relation between sin/cos and exponential
basis such as Y 1 2 =
n=0 Gn bn
basis is given by (3.4), the contribution of [] contraction is
1
(3)
(2)
(3)
Y 1 2 Y 1 2 ij1 1 v
v1 2 ,n m,1 2 Yij2 v = vb1 2 ,n b(2)
1 2 ,m
2
(2)
(3)
=
2 2 ij G
m Gn .
(3.18)
1 1
(2)i
n
vm
j ]
(2)i
XI[i XII v = vm
n
(3)j
(r) [i (s)j ]
an v
Fm(r) Fn(s) Un(s) am
r,s=1 m=0,n=1
1
= Fm(2) Fn(3) Um(2) Un(3) i[i j ]j .
2
(3.19)
178
Combining the two contributions, (3.18) and (3.19), and using the asymptotic form for F and G
listed in Appendix C, the contribution from [][] is evaluated as
m
1
n
.
[][] = i
(3.20)
2 2 ij
3
1 1 16 2 2 
y
1y
(1)  y(1 y)
Combining all the contributions, (3.13), (3.16), and (3.20), we obtain the matrix element of
the interaction vertex H3 h for the process (3.5) in the large limit,
(3)
(2)
(2)i
v1 2 ,n n(3)i 1 2 ,m m
H3 h
m
1 ij
n
1
3
ij
= i
.
1 1
y
1 y 16 2 2 (1) 3 y(1 y) 2 2 2 1 1
2
(3.21)
In order to obtain the OPE coefficient according to the duality relation (2.1), the G factor
2 +3 1
2
J1 J2 J3
2 + 3 1
J2 J 3
1
f
+1
G
(2 + 3 1 )
J1
2
N
2
1
J1 J2 J3
J1
=
(3)
4 4(1) 
N
2
2 1/2 1
= 8 (1)  J1
N
y(1 y),
(3.22)
must be multiplied with (3.21). Therefore, the CFT coefficient C123 which should be obtained
from the corresponding three point correlation function on the gauge theory side is expected to
be
3/2 1/2
gYM J1
N
i m
1
n
3
2 2
1 1 + ij .
C123 =
(3.23)
1 1
2 y
1y
2
2
y(1 y)
We should note that the interaction vertexes H3 SV and H3 D are both necessary to obtain this
results and that the G factor properly adjust the dependence to give O(gYM ) = O(1/) result.
(1)i (1)j
(2)i (2)
(3)j
(3)
(3)j
(2)
(1)j
n(1)i
n1 H3 h .
v1 2 ,n3 n3 1 2 ,n2 n(2)i
2
1
(3.24)
(1)j
v(3)
(2)
(2)i n(1)i
n1 H3 D
1 2 ,n2 n2
1
1 2 ,n3 n3
i 32
ij
31
21
= 2
Qnm Q23
mn Nn3 ,n1 Nn2 n1
1 1
2 2
2
1
n3
sin2 (yn1 )
ij n2
=i 2
1 1 2 2
n3 ,
2
y
1 y (n1 ny2 )(n1 + 1y
)
4 (1)  y(1 y)
(3.25)
where we used the asymptotic form of the Neumann coefficients in the large limit presented
in Appendix C.
179
On the other hand, because of the form of the prefactor PSV , for the case i = j , the possible
contribution from H3 SV is that only one pair of is contracted through the prefactor and others
are through the overlap vertex:
(3)
(3)j (1)j
(2)
n3 n1 n(2)i
1 2 ,n3 1 2 ,n2
or
n(1)i
2
1
(3)j (1)j (2)i (1)i (3)
(2)
n3 n1 n2 n1 1 2 ,n3 1 2 ,n2 .
(3.26)
In particular, when i = j , the contraction through the prefactor (3.17) is not possible because
of the antisymmetric property of i and j in Yij2 and the fact that only a pair of oscillators with
the same scalar index can be contracted through the overlap vertex. This type of contraction is
possible when we consider, for example, the case where the excitation of the incoming string is
(1)i (1)k
given by n1 n1 v (k = j ) and will play an important role for our duality relation to hold.
The amplitude of the first term in (3.26) is the product of three amplitude such as
(3)j (1)j
(3)
(2)
(3.27)
At the first factor, the prefactor PSV can be replaced with r Er XII2 in the same way as in the
previous subsection. But in contrast to the situation there, it gives the order O(1/) contribution
as in the impurity preserving processes:
(3)j (1)j
vn3 n1
Er E =
(3)
(1)
n
n
+
N n313 ,n1 = O(1/).
(r)
(r)
(3.28)
Since the contribution from XII is of the same order and the orders of the remaining contributions
are
(3)
(2)
v1 2 ,n3 1 2 ,n2 E = O 1/2 ,
n(1)i
E = O(1),
vn(2)i
2
1
(3.29)
the net result of (3.27) gives a subleading contribution and negligible compared to the result (3.25) in the large limit. The same is true for the second case of (3.26).
Therefore only the interaction vertex H3 D gives the leading contribution such as
(3)
(3)j
(1)j
(2)
n3 .
1 y (n1 ny2 )(n1 + 1y
)
8 (1) 2 y(1 y) 1 1 2 2 y
(3.30)
J1
4(1) 
1
(2)
J1 J2 J3 1
1/2
y(1 y)N 1 ,
= 2(1) J1
N
2
(3.31)
C123 = i
n3 .
1
1
2
2
y
1 y (n1 ny2 )(n1 + 1y
)
4 2 y(1 y)
(3.32)
180
(3)j
v1 2 ,n n
(3)
(2)
(2)i
1 2 ,m m
H3 h ,
(3)j
(2)
(4.1)
(1)j
(4.2)
let us proceed to calculate the corresponding gauge theory three point correlators and derive the
OPE coefficients.
For this purpose, it is important to identify properly the operator which corresponds to the
string state ni 1 2 ,n v with the angular momentum J around S 5 . According to the prescription of inserting an impurity into the vacuum state Tr(Z J ) with an appropriate phase factor,
the operator with the proper U(1) Rcharge and the conformal dimension satisfying the relation
H / = J might be given by
J
2in
(4.3)
exp
(l + 1) Tr i Z l r Z J l ,
J +2
l=0
where the subscript r of r is the spinor index of the (2, 1)+1/2 sector in the decomposition of
SU(4) Rsymmetry vector index A of Weyl fermion A
in the d = 4, N = 4 SYM theory, such
as
4 = (2, 1)+1/2 (1, 2)1/2 .
(4.4)
Here the subscript 1/2 denotes the charge of U(1)J Rsymmetry subgroup we have picked up.3
However, as we will show in what follows, in order to obtain the operator of definite conformal
dimension, we must consider a linear combination of (4.3) and the operator
i r
(4.5)
r Tr r Z J +1 ,
which is the unique possibility of the operator with the same classical dimension and U(1) Rcharge as the operator (4.3). Here r is the (1, 2)1/2 sector in the decomposition of (4.4) and
( i )rr is the ClebschGordan coefficient for a SO(4) vector under SO(4) = SU(2) SU(2) decomposition.
The operator (4.5) is expected to be obtained when the SUSY transformation and the rotation in the iZplane of Rsymmetry act on one and the same constituent Z of the vacuum state
Tr(Z J +2 ), while (4.3) is obtained when these two generators act on different Zs with an appropriate phase dependence [3234]. Though the order of the number of the latter case is O(J )
while the former is of O(1), we cannot omit this term when we identify the BMN operator which
corresponds to the free string basis as in the case of the singlet sector of the operator with two
bosonic or fermionic impurities studied so far [1619].
In what follows, we first determine the weight of the linear combination of (4.3) and (4.5) so
that the two point function of the BMN operator takes the canonical form. Next, with the proper
linear combination thus determined, we calculate the gauge theory three point functions which
correspond to (4.1) and (4.2), respectively. We will see that the OPE coefficients which can be
read from the three point functions exactly agree with (3.23) and (3.32).
3 The correspondence of the spinor indexes between string and gauge theory side is ( , ) (r, ).
1 2
181
(x y)
1
(x y)2
(x y)2(J +2)
1 n2 ln(x y)2 2 .
(4.6)
In order to obtain an operator of definite conformal dimension, we consider the linear combination such as
J,i
Or,m
r
ql Tr i Z l r Z J l + b i r Tr r Z J +1 ,
(4.7)
l=0
l,l =0
J
r
+ b i r Tr r Z J +1
rl Tr Z J l s Z l j
+ b
J
l=0
ql Tr Z J l r Z l
i
j s
s Tr Z J +1 s
l=0
+ b
i r
s
r Tr r Z J +1 j s Tr Z J +1 s ,
(4.8)
takes the canonical form of (4.6). Here we have defined the phase factor ql and rl such as
2im
2in
ql exp
(4.9)
(l + 1) ,
rl exp
(l + 1) .
J +2
J +2
i
The complex conjugate with lower indexes is defined as O ri
rs O is
rs (Os ) . Note that
and are defined as
i
i sr
r s i i , 1 ,
(4.10)
i i , 1 ,
where i s are Pauli matrices, and the complex conjugate of the matrix is (( i )r s ) = ( i )sr .
We will call the contributions in the first, second, third and fourth line in (4.8) as (++), (+),
(+), and () sector, respectively.
In order to determine b, it is enough to focus on the log divergent part in (4.6). Let us first
compute the log divergences which appear in the (++) sector. The divergence comes form
selfenergy, 4Zinteraction, 2Zinteraction, 2Zinteraction, and 2interaction, which will
be explained bellow. The calculation is partially same as in Ref. [35] where the anomalous
dimension of the operator with two scalar impurities is calculated, but here the calculation becomes a little bit complicated because we must consider additional interactions which involve
182
fermions. We calculate here 4Zinteraction, etc. which have already been calculated in [35] for
selfcontainedness. The details of the calculations are referred to Appendix A.
The two point functions of Z a s and ar s at the oneloop level are given by
2 2
a
ln(x y)2 2
1 gYM
Z (x)Z b (y) =
(4.11)
N ab
,
2
2 4
(x y)2
2 2
a
g
1
r (x) bs (y) = YM2 N ab
rs x
(4.12)
ln(x y)2 2 .
4
(x y)2
Hereafter, we focus only on the log divergent part and omit the other terms. The net log divergence from the selfenergy in the planer limit is obtained by replacing one of the free bosonic or
fermionic propagator with (4.12) or (4.12), respectively, for each possible free contraction. Then
we obtain
J
1
CSE =
(4.13)
ql rl (J + 1) + (1) 1 ij
rs ,
2
l=0
2 4 2
4 2
(x y)2
(x y)2(J +1)
(4.14)
The contribution from the 4Zinteraction is obtained by replacing the two adjacent free propagators of Z and Z with the 4point correlator calculated at oneloop order such as
2 3
a
gYM
ln(x y)2 2
a2
b1
b2
1
ij
rs x
Tr T a1 T a2 T b2 T b1
2 2 4 2
4 2 (x y)2
(x y)2(J 1)
2 3
gYM
ln(x y)2 2 1
(fpa1 b1 fpa2 b2 + fpa1 b2 fpa2 b1 )
= ij
rs ,
(4.16)
2
2
4
(x y)4
where we have used the formula
a1
a2
b1
fpa1 b2 fpa2 b1 Tr T T T T
b2
N
=N
2
fpa1 b1 fpa2 b2 Tr T a1 T a2 T b1 T b2 = 0.
3
1
1 2 ,
N
(4.17)
(4.18)
When and sit adjacent to each other, there are J 1 ways of insertion of the 4Zinteraction,
otherwise J 2 ways. Therefore, with the phase factor appropriately taken into account, the
oneloop order contribution from 4Zinteraction is
J
1
1
C4Z = (q0 r0 + qJ rJ )(J 1) +
(4.19)
ql rl (J 2) ij
rs .
2
l=1
183
Fig. 1. 2Zinteraction.
In the case of 2Zinteraction, the weight of the log divergence is different according to the
relative position of and Z, since the correlator of 2Zinteraction is given by
2 3
a
gYM
1
f
+
f
f
f
r1 (x)Z a2 (x) bs 1 (y)Z b2 (y) =
rs
pa1 b2 pa2 b1
pa1 b1 pa2 b2
2
4 2
1
1
(4.20)
x
ln(x y)2 2 .
2
(x y)
(x y)2
In this case, there are 4 types of configuration of and Z for each possible replacement as
depicted in Fig. 1. The log divergence from the crossing configurations of and Z, which are
depicted in Fig. 1 with the relative phase dependence q or r , is obtained by contracting (4.20)
with Tr(T a1 T a2 T b1 T b2 ), giving 1/2ij
rs , while the others obtained by contracting (4.20) with
Tr(T a1 T a2 T b2 T b1 ), giving 1/4ij
rs . Here, we have taken into account the free contractions
of remaining operators. Then, the total contribution from 2Zinteraction is
J
1
1
1
ql rl
(1 + q r ) + (r + q) ij
rs ,
C2Z =
(4.21)
4
2
l=0
(4.22)
From this form of the 4point correlator and the formula (4.18), we can easily see that only the
crossing configuration of and Z gives nonzero contribution. So, we obtain, including the free
contractions of the other operators, the log divergence from the 2Zinteraction,
C2Z =
J
1
l=0
1
ql rl (r + q)
ij
rs .
2
(4.23)
The last contribution we should consider in the (++) sector is the 2interaction, which
occurs only when and impurities sit adjacent to each other. Since the four point correlation
function is given by
a
r1 (x) ia2 (x) bs 1 (y) ib2 (y)
2 3
j i
1 gYM
1 i j
=
rs fpa1 b2 fpa2 b1 + rs fpa1 a2 fpb1 b2 + ij
rs fpa1 b1 fpa2 b2
2 4 2
2
1
1
(4.24)
x
ln(x y)2 2 ,
(x y)2
(x y)2
184
(4.25)
Combining all the results so far, the net contribution from oneloop interactions in the (++)
J,i
and O sJ,i
sector of the two point function of the operators Or
is
CSE + C4Z + C2Z + C2Z + C2
2in 2
J ( J ) ij
rs + 12 ( i j )rs + O(1/J 2 )
=
+ 12 ( i j )rs + O(1/J 2 )
(m = n),
(m = n),
(4.26)
which shows that the two point function does not take the canonical form, therefore the operator (4.3) cannot be the correspondent of the ni rn v. Since the O(1) factor ( i j )rs in (4.26)
breaks the canonical form, the basic requirement for the BMN conjecture is not satisfied that the
2 N /J 2 should match the first order perturbaanomalous dimension to the first order of gYM
tion expansion of the free string energy spectrum in terms of 1/2 , under the identification of
= 1/(p + )2 .
The extra term in (4.26) is eliminated by choosing b properly. In the (+) sector, there appears the contribution from the 4point correlator
2 3
i
1 gYM
1
f
+
f
f
f
ar1 (x) ia2 (x)rb1 (y)Z b2 (y) =
pa1 b2 pa2 b1
pa1 a2 pb1 b2
r r
2
2 4 2
1
1
ln x y 2 2 . (4.27)
2
2
(x y)
(x y)
With the formula (4.17) and (4.18), we can easily see that
each of 4 possible configurations
(4.28)
(4.29)
where 1/2 comes from the selfenergy of a boson propagator and 1 from fermion, and the
4Zinteraction contribution,
1
b2 J i j rs ,
2
(4.30)
185
r s x
ln(x y)2 2 .
(x y)2
(x y)2
(4.31)
We have nonvanishing contribution only when and Z sit in the crossing position, and therefore
the Zinteraction gives the log divergence such as
1 i j
rs .
4
From (4.29), (4.30) and (4.32), the total contribution in the () sector is
b2 i j rs .
b2 2
(4.32)
(4.33)
Combining all the contributions form the (++), (+), (+) and () sectors, the total log
divergence is given by
2
+ 1 (1 + 2(b + b)
+ 2b2 )( i j )rs + O(1/J 2 ) (m = n),
J ( 2in
ij rs
J )
2
(4.34)
1
2
i
j
+ 2b )( )rs + O(1/J 2 )
(m = n),
+ 2 (1 + 2(b + b)
from which we conclude that we should choose
1
b = ,
2
(4.35)
J,i
takes the canonical form.
in order that the two point function of the operator Ors
J,i
in (4.7). In
Here we should comment on the choice of the phase factor of the operator Or,m
fact, we can obtain the operators of definite conformal dimension which are diagonalized up
to O(1/J 2 ), even if we adopt the other phase dependence, for example, exp(2iml/(J + 1)) or
exp(2iml/J ). However, according to [32,33], we have chosen the phase factor so that the level
matching condition for the world sheet momentum of the closed string is naturally realized when
J,i
we regard the BMN operator Orm
is obtained by acting the rotation in the Rsymmetry space
and the SUSY transformation on the vacuum state Tr(Z J +2 ) accompanied by the phase factor q
such as
O1 (x)O2 (x) On (x) = O1 (x) O2 (x) On (x) + qO1 (x) O2 (x) On (x)
+ + q n1 O1 (x)O2 (x) On (x) .
(4.36)
l=0
Then, the level matching condition is realized when we choose qi such as qi = exp(2imi /
(J + 2)) (i = 1, 2). As explained at the end of the next subsection, this choice of the phase factor
is necessary for our duality condition to hold.
186
(2)i
(3)
1
(J1 + 1)N0J1 +1
Jr ,i
Os,m
=
Tr Z J1 +1 ,
1
2(Jr + 1)N0Jr +2
J
r
N0
2im(l+1)
Jr +2
2
N gYM
,
2 4 2
(4.39)
i l
s
Tr Z s Z Jr l + b i s Tr s Z Jr +1 ,
l=0
(4.40)
1/x2
J
Ovac
for
and takes the canonical form (4.6)
so that the two point function is normalized to
J,i
for Or,m . The coefficient b has been determined in the previous subsection as b = 1/ 2.
Similarly to the two point function, there are four sectors of the three point function to be calculated: three of them are depicted in Figs. 24, which we also call (++), (+) and () sector,
respectively, and the remaining one is the (+) sector, for which we have omitted the figure.
Apart from the normalization factors in (4.39) and (4.40), the (++) sector of the three point
function (4.38) at the leading order in terms of the genus and the gauge interaction expansions is
187
given by
2 J1 +1
(J1 + 1)ij
1 N J1 1 gYM
4 2
4 2
x23 2 x12 2J2 x31 2J3
+ qJ2 r0 Tr(r s Z)
,
q0 rJ3 Tr(s r Z)
where the three point correlator in this expression is given by
2 3
1
a
gYM
fa1 b1 c1
rs
Tr Z 1 (x1 )br1 (x2 )cs1 (x3 ) = i 2 2
2
gYM 4
1
1
1
u
u
.
(x1 u)2
(x2 u)2
(x3 u)2
(4.41)
(4.42)
Integrating by part and using the formula (A.1), the symmetric part in terms of the spacetime
indexes and in the integral can be evaluated as
1
1
1
d 4u
u
u
2
2
(x1 u)
(x2 u)
(x3 u)2
1
1
1
1
4
d u
=
2
(x1 u)2
(x2 u)2 (x3 u)2
1
1
1
1
(x2 u)2 (x3 u)2 (x2 u)2 (x3 u)2
1
1
1
2
= 2
(4.43)
.
x12 2 x31 2 x12 2 x23 2 x23 2 x31 2
Therefore, with the relations, ( ) =
+ ( ) , where ( ) (
) /2, and fa1 b1 c1 Tr(T a1 T b1 T c1 ) = iN 3 /4 + O(N 2 ), the (++) sector of the three point
function is given by
2 J1 +2
1
m
n
N gYM
2i
ij
rs
2
2J
2
y
1y
2 4
x12  x31 2J3 x23 2
1
1
1
1
1
1
+ d 4 u
(4.44)
.
u
u
(x1 u)2
(x2 u)2
(x3 u)2
The world sheet momentum dependence comes from
2im
2in(J3 + 1)
exp
q1 rJ3 +1 qJ2 +1 r1 = exp
J2 + 2
J3 + 2
188
2im(J2 + 1)
2in
exp
exp
J2 + 2
J3 + 2
n
m
4iJ1
y
1y
(4.45)
where y = (2) /(1) = J2 /J1 and 1 y = (3) /(1) = J3 /J1 and we take the leading term
in the large J limit. We should note
that this momentum dependence cannot be obtained if we
choose the phase factor such as Jl=0 exp(2in/J ) Tr( i Z l Z J l ) often adopted in the literature.
As in the case of two point function, the (+) and (+) sectors are indispensable for our
duality relation, while the () sector depicted in Fig. 4 gives a vanishing results since it does
not contain a phase factor. The (+) sector of the three point function depicted in Fig. 3 is
2 J1 +1
j s
1 N J1 1 gYM
(J1 + 1)
s
4 2
4 2
x12 2J2 x31 2J3 +2
Tr i (x2 )r (x2 )s (x3 ) + qJ Tr r (x2 ) i (x2 )s (x3 ) .
(4.46)
The perturbation calculation to the oneloop order gives
ai
Tr 1 (x2 )br1 (x2 )sc1 (x3 )
2 3
gYM
1
= 2
ifa1 b1 c1 i r s
2
gYM 4
1
1
1
d 4u
u
u
.
(x2 u)2
(x2 u)2
(x3 u)2
(4.47)
This time, since the antisymmetric pert in terms of the indexes and can be eliminated, the
integral is transformed as
1
1
1
u
d 4 u u
4
2
(x2 u)
(x3 u)2
1
1
1
1
=
d 4 u
= 2 2
,
(4.48)
4
2
2
(x2 u) (x3 u)
(x2 x3 )4
and then the (+) sector of the three point function becomes
2 J1 +2
1
m N gYM
i j rs
.
ib
2
2J
y 2 4
x12  2 x31 2J3 +2 x23 4
Similarly, the amplitude of the (+) sector is given by
2 J1 +2
n
1
N gYM
ib
i j rs
.
2
2J
+2
2
1 y 2 4
x12 
x31 2J3 x23 4
(4.49)
(4.50)
In the situation we are now considering, where
= x2 x3  x1 x2 , the integral in (4.44)
can be neglected because
1
1
1
d 4u
u
u
0 (x3 x2 ).
(4.51)
2
2
(x1 u)
(x2 u)
(x3 u)2
Therefore, combining all the contributions, (4.44), (4.49) and (4.50), and taking into account the
j
normalization factors in (4.39) and (4.40), the three point functions where Os,n (x3 ) sits in the
189
2 y
1y
y(1 y) x12 2J1 +2
4
b
b
2 + ij
rs ij rs ,
2
2
(4.52)
where ( ij )rs is defined as ( ij )rs =
st 1/2( i j j i )tr . The final result (4.52) shows that
the OPE coefficient which can be obtained in the limit x3 x2 exactly
agrees with the expected
result from the string theory calculation, (3.23), just when b = 1/ 2. This nontrivial matching
confirms that the diagonalization up to O(1/J ) is crucial for our duality relation.
(1)i (1)j
(3)j
(2)i (2)
(3)
On11
1
(J1 + 2)N0J1 +3
J1
Tr i Z l1 j Z J1 l1 .
(4.54)
l1 =0
We consider the case where i = j for simplicity as in the corresponding calculation on the
string theory side. Since in this case too the () sector does not contribute to the lowest order
calculation with respect to the gauge theory interaction, it is enough to consider the (++), (+),
and (+) sectors.
The (++) sector of this correlator at the planer level is depicted in Fig. 5 and given by
2 J1 +2 J1
1
1 gYM
N
4 4 2
2
x12 2(J2 +1) x13 2(J3 +1)
J
2 ,J3
2i(l2 +1)
J2 +2
J2 +2 (n2 J1 +2 n1 )
2i(l3 +1)
J3 +2
J3 +2 (n3 J1 +2 n1 )
l2 ,l3 =0
2in1 J3
2in1 (J3 +1)
1 )r (x2 )s (x3 ) .
1 )s (x3 ) + e J1 +2 Tr Z(x
e J1 +2 Tr r (x2 )Z(x
(4.55)
190
2i(l2 +1)
J2 +2
J2 +2 (n2 J1 +2 n1 )
2i(l3 +1)
J3 +2
J3 +2 (n3 J1 +2 n1 )
l2 ,l3 =0
sin2 (yn1 )
J1 2i JJ2 +2
n1
1 +2
e
n3 .
2
(n1 ny2 )(n1 + 1y
)
(4.56)
Then, the (++) sector of the three point function in the situation x23  =
x12  is given by
J1 +3
2 g2
gYM
1
J1
n1 sin2 (yn1 )
YM N
i 2 2
(4.57)
rs
n3 .
)
4
4 2 2
x12 2(J1 +3)
2 (n1 ny2 )(n1 + 1y
On the other hand, the (+) sector in Fig. 6 is
b
2 J1 +2 J1
J2
2i(l2 +1)
J +2
1 gYM
1
N
(n 2 n ) s
e J2 +2 2 J1 +2 1 j s
2
2(J
+1)
2(J
+1)
4 4
2
x12  2 x13  3
l2 =0
2in1 (J3 +1)
J1 +2
Tr s (x3 )r (x2 ) j (x1 ) .
Tr j (x1 )r (x2 )s (x3 ) + e
(4.58)
The vacuum expectation value Tr( j (x1 )r (x2 )s (x3 )) can be calculated in the same way as
(4.47), and the sum of the phase factor in the large J limit is
J2
l2 =0
2i(l2 +1)
J2 +2
J2 +2 (n2 J1 +2 n1 )
J1
1
2i n1
n2
y
1 e2iyn1 .
(4.59)
(4.60)
where we do not take the sum over the superscript j . A parallel computation for the (+) sector
gives
J1 +3
2 g2
i i J1 sin2 (yn1 )
gYM
1
YM N
ib 2
(4.61)
rs
n3 .
n1 + 1y
4
4 2 2
x12 2(J1 +3)
2
191
With the relation ( i i )rs =
rs (with no sum over i), and taking into account the normalization
factors of BMN operators, we obtain
J1 ,ij
J3 ,j
J2 ,i
(x2 )Os,n
(x
+
)
O n1 (x1 )Or,n
2
2
3
1/2 1/2
N
2gYM J1
1
rs
=
2(J
2
4 y(1 y) x12  1 +3)
2
sin2 (yn1 )
n2
1
n3
(4.62)
2 i + ib n1 ib
.
n3
(n1 ny2 )(n1 + 1y
)
y
1y
2
This shows thatthe contribution from all sectors nicely combine to match exactly the result (3.32)
when b = 1/ 2. As in the previous example, the nontrivial matching is realized for the operators of definite conformal dimension.
J ,ij
It is not difficult to consider a more general case where On11 is replaced by OnJ11 ,kl . In this
case, the interaction Hamiltonian H3 SV contributes to the calculation on the string theory side
jl
ik (or (i j ) counterparts). We can easily see
and gives the term proportional to ik rs or j l rs
that these terms also appear in the corresponding gauge theory calculation and will give the complete agreement. This gives another verification of the necessity for the both contributions from
the vertexes H3 D and H3 SV . We should note that when k = l, the operator mixing between a
J +1 ) will also play an important role
operator with scalar two impurities and the operator Tr(ZZ
for the duality relation.
Finally, we should comment on the normalization and the phase factor of the BMN operator
J,i
Or,m . The normalization has been chosen so that two point function takes the form (4.6). With
this choice of the normalization, our duality relation holds for the case (4.53) regardless of the
choice of the phase factor, that is, whether we chose exp(2im/J ), exp(2im/(J + 1)), or
exp(2im/(J + 2)). On the other hand, with the normalization thus fixed, in order for our duality
relation to be valid, we must chose the phase factor exp(2im/(J + 2)); otherwise we cannot
obtain the result with the appropriate worldsheet momentum dependence and the normalization
factor, as is easily seen from the calculation (4.45). Two simple examples we studied here is
J,i
enough to fix both the normalization and the phase factor of the BMN operator Ors,m
.
5. Conclusion and discussion
In this paper we confirm that the duality relation we proposed in [7] holds for the impurity
nonconserving process which contains fermionic BMN operators. In the process of the proof, it
is crucial that (a) the operator correspondent to the free string basis has the definite conformal
dimensions, (b) the string interaction Hamiltonian takes the form of the equal weight sum of the
two interaction vertex H3 SV and H3 D and (c) the structure of G factor in (3.22), in particular,
the factor f balances the dependence between C123 and H123 especially in the impurity nonpreserving processes.
However, though our duality relation holds in a very nontrivial way with the choice of the
equal weight sum of H3 SV and H3 D , we cannot conclude that this form of the string interaction vertex has proved to be completely correct. Since we cannot uniquely fix the form of the
interaction vertex only from the requirement to respect the SUSY in the PPwave background, it
is not enough to check only a particular sector. In the previous paper [7] we have fixed the form of
the interaction vertex so that it is consistent with the effective action for the chiral primary fields
constructed in [3]. However, it has been claimed that the holographic string interaction vertex
192
should be further modified in order to respect the U(1)Y symmetry existing in the supergravity
sector [29]. They claimed that the H3 D should be replaced as
H3 D =
(r)
H2 E
r=1
(r)
H2
r=1
1
1 Y4
12
1 4
1 Z E.
12
(5.1)
The new terms do not affect all the processes we have studied in previous works. The simplest example to test this modification is a impurity nonpreserving process where the vacuum
state splits into two BMN operators with two fermionic impurities. For this process, in contrast to the impurity preserving process where the energy difference gives O(1/) contributions,
(r)
the new term r H2 Y 4 E can play a significant role. In fact, it seems that for the process
(2)
(2)
(3) (3)
vb11,m b12,m b21,n b22,n H h the new term cancels the contribution from the prefactor in
H3 SV and only H3 D contribution survives, which is of subleading order in the large limit
compared with H3 SV . For our duality relation to be valid, this cancellation seems to be necessary in order to obtain the correct dependence expected from the gauge theory calculation. In
this case too, the proper identification of the BMN operator which corresponds to the free string
states is crucial to check our duality relation. The details will be reported in the future work [30].
Anyway, with the results obtained in the previous works [7,8] that our duality relation holds
for the impurity preserving process which consists of the BMN operators with scalar, vector
and fermionic impurities, as well as, for the impurity nonpreserving processes which consist of
those with scalar and vector impurities, our duality relation itself is strongly expected to hold for
the general processes regardless of the kind of impurities and of whether impurities conserve or
not, as long as we choose the appropriate form of the interaction Hamiltonian in the PPwave
background and properly identify the dictionary between string and gauge theory basis.
Acknowledgements
We would like to thank T. Yoneya for valuable discussions and the collaboration at an early
stage of this work.
Appendix A. Feynman graph
In this appendix, we give the details of the calculation required to derive the log divergence
of the two point function (4.8). As the regularization scheme, we adopt the differential regularization method developed in the reference [36]. With this method the anomalous dimension
of the operator with two vector impurities is calculated in [33]. The formulae necessary for our
calculation are
1
1 ln((x y)2 2 )
=
,
4
4
(x y)
(x y)2
1
= 4 2 (4) (x y).
(x y)2
(A.1)
With these formulae and partial integration, the two integrals needed in the calculations below to
derive log divergences can be evaluated as
2 2
1
1
2 ln(x y)
d 4u
(A.2)
=
2
,
(x u)4 (y u)4
(x y)4
2 2
1
1
2 ln(x y)
d 4u
(A.3)
=
.
(x u)4 (y u)2
(x y)2
193
This regularization method is used in the calculations of the three point correlators in (4.42)
and (4.47).
A.1. N = 4 superYangMills action
Here we present the N = 4 superYangMills action in the 4dimensional spacetime and
summarize the notation. The field content is 4 gauge fields A ( = 1, 2, 3, 4), 6 real scalar
fields I (I = 1, 2, 3, 4, 5, 6) and 4 Weyl spinor fields A
(A = 1, 2, 3, 4). All these are in the
adjoint representation such as
A = Aa T a ,
I = aI T a ,
a
aA
T ,
where T a satisfies
a b
T , T = ifabc T c ,
1
Tr T a , T b = ab ,
2
with the structure constant fabc for SU(N) symmetry, which satisfies
fpqr fpqr = N N 2 1 .
fapq fbpq = Nab ,
(A.4)
(A.5)
(A.6)
We pick up the U(1) subgroup of the Rsymmetry such as SO(6) = U(1) SO(4) for scalar
field and SU(4) = U(1) SU(2) SU(2) for spinor fields and decompose the field such as
I i , Z, Z ,
(A.7)
A
(r , r ),
where Z ( 5 + i 6 )/ 2 which has the U(1) Rcharge 1 under the rotation in the 56plane.
The indexes I , i and A are vector indexes of SO(6), SO(4) and SU(4) symmetry, respectively.
The decomposed fermionic fields r and r denote the (2, 1)+1/2 and (1, 2)1/2 sectors in the
decomposition above, where 1/2 is the charge under the U(1) Rsymmetry we have picked up.
With the decomposed field in (A.7) the d = 4, N = 4 superYangMills action is given by
1 a a 1
2
gYM
S = F
F
D ia D ia D Z a D Z a ra D a r + ar D ar
4
2
1
1
a b c d
fpab fpcd i j i j fpab fpcd ia Z b ic Z d + fpab fpcd Z a Z b Z c Z d
4
2
i br c
a i r r b c
ar
ifabc r
r i + r r i
1
1
ra br ar br Z c + ra br ar br Z c ,
(A.8)
2
2
where F = x A A i[A , A ], D Z = Z i[A , Z]. The matrix and i are
defined as
= i i , 1 ,
= i i , 1 ,
(A.9)
i
i r r
i
= i i , 1 .
r r = i , 1 ,
(A.10)
These satisfy the relation
( ) + ( ) = 2 ,
(i j )r s + (j i )r s = 2ij rs ,
( ) + ( ) = 2 ,
(i j )r s + (j i )r s = 2ij sr .
(A.11)
(A.12)
194
Note that for the lowered indexes we have (i j )rs + (j i )rs = 2ij
rs . The spinor indexes,
as for both Lorentz and Rsymmetry, are raised and lowered by the invariant tensor
rs and
rs
respectively such as
0 1
0 1
r
rs s ,
(A.13)
r =
rs s ,
rs
,
rs
1 0
1 0
and we use the usual convention as
r r = ,
r r = .
(A.14)
We have omitted the Lorentz spinor indexes in (A.8) using this abbreviation.
From the SYM action (A.8), the propagators are given, in the Euclidean space, by
2
a
g2
1
1
b
ab gYM
,
Z
(x)Z
(y)
=
,
ai (x) bj (y) = ab ij YM2
2
2
4 (x y)
4 (x y)2
2
a
1
ab s gYM
r (x) bs
(y)
=
,
x
r
4 2
(x y)2
2
a
1
bs
ab s gYM
x
.
r (x) (y) = r
4 2
(x y)2
(A.15)
(A.16)
(A.17)
ifapq ifbpq
1
4 2 2
rs
rs
2
2gYM 2gYM
2
gYM
4 2
4
d 4u d 4v
1
1
1
1
u
u
, (A.18)
(x v)2 (y u)2
(u v)2
(u v)2
where the factor 1/2 4 comes from the perturbation expansion and the symmetric factor. With
(A.19)
.
(x v)2 (y u)2 2 (u v)4
(u v)2 (u v)2
Since we focus only on the log divergence in this paper, we neglect the second term, which also
should be canceled by the contribution of the same degree of divergence from other diagrams.
The log divergence of the first term in (A.19) can be estimated using the formula (A.1) and (A.3).
ab
2 4 2
(x y)2
The diagram whose intermediate states consist of s also gives the same contribution.
The contribution of the gluon emissionabsorption process is
195
(A.20)
fapq fbqp
1
2 2
2
2
gYM gYM
2 4
gYM
1
1
1
1
d 4u d 4v
u
v
.
(A.21)
2
2
2
2
4
(u v) (x u)
(u v)
(y v)2
Integrating by part and using the formula (A.1) and (A.3), the log divergence from gluon exchange is reduced to
2 2
ln(x y)2 2
1 gYM
N
.
(A.22)
ab
2 4 2
(x y)2
The log divergence which appears in the selfenergy of Z is
2 2
ln(x y)2 2
1 gYM
N ab
.
2
2 4
(x y)2
(A.23)
fapq fbqp
1
2
rs 2
2
2
gYM gYM
2 4
gYM
1
1
1
1
d 4 ud 4 v
x
u
v
.
2
2
2
2
4
(u v)
(x u)
(u v)
(y v)2
With the identity
= 2 ,
= + + i
,
d 4u
rs 2
x
2
2
4
4 2
(x
u)
(u
v)
(y
v)2
gYM gYM
(A.24)
(A.25)
(A.26)
196
+ i
( ) x
1
1
1
d u
u u
.
(x u)2
(u v)4 (y v)2
4
(A.27)
The second term vanish because of the antisymmetric property of
, and with the formula
(A.1) and (A.3), we obtain the log divergence
2 2
1 gYM
1
(A.28)
N
rs ab x
ln(x y)2 2 .
4 4 2
(x y)2
There are two kinds of contribution from the Yukawa interaction, where the intermediate states
are , Z and , , respectively. The former diagram is
ifapq ifbpq
1
8 2 2
sr
2
2gYM 2gYM
2 4
gYM
1
1
1
1
d 4u d 4v
x
v
v
2
2
2
2
4
(u v)
(x u)
(u v)
(y v)2
2 2
1 gYM
1
=
N
rs ab x
(A.29)
ln(x y)2 2 ,
4 4 2
(x y)2
where the integral has been performed in the same way as the gluon emissionabsorption process.
The latter contribution is
ifapq ifbpq i i ut
1
r u
ts
2 2
2
2
gYM gYM
2
gYM
4 2
4
d 4u d 4v
1
1
1
1
x
v
v
.
(u v)2
(x u)2
(u v)2
(y v)2
(A.30)
Since
( i )r u ( i )ut
2
1 gYM
2 4 2
= +4rt ,
2
N
rs ab x
ln(x y)2 2 .
(x y)2
(A.31)
=
2
gYM
4 2
2
N
rs ab x
1
ln(x y)2 2 .
(x y)2
(A.32)
197
2 5
gYM
fpa1 b2 fpa2 b1
fpa1 b1 fpa2 b2
1
+
2
2
2
2
2
2
4 2
gYM
gYM
gYM
gYM
d 4u d 4v
1
1
1
1
u
(u v)2 (x u)2
(y u)2 (x v)2
1
.
(A.33)
(y v)2
Performing partial integral and using the formula (A.1), the integrand can be transformed as
1
2
(A.34)
(Y
+
Y
)
X
(x + y )Hxyxy + 4 2
,
xxy
xyy
xxyy
2
(x y)2
where we have defined Xabcd and Yabc as
1
1
1
1
1
,
Habcd = d 4 u d 4 v
(A.35)
2
2
2
2
(a u) (b u) (u v) (c v) (d v)2
1
1
1
1
,
Xabcd = d 4 u
(A.36)
(a u)2 (b u)2 (c u)2 (d u)2
1
1
1
.
Yabc = d 4 u
(A.37)
2
2
(a u) (b u) (c u)2
As long as we are concerned with the logarithmic divergence, the first term in (A.34) can be
neglected and the other terms can be evaluated with the use of the formulae (A.2) and (A.3). The
result is
2 3
ln(x y)2 2
1 gYM
(f
f
+
f
f
)
.
(A.38)
pa
b
pa
b
pa
b
pa
b
1 1
2 2
1 2
2 1
2 4 2
(x y)4
On the other hand, the scalar 4point interaction is easily calculated as
2 4
g
1
1
d 4u
= 2 (fpa1 b1 fpa2 b2 + fpa1 b2 fpa2 b1 ) YM2
4
4
(x
u)
(y
u)4
gYM
1
2 3
1 gYM
ln(x y)2 2
=
(f
f
+
f
f
)
.
pa
b
pa
b
pa
b
pa
b
1 1
2 2
1 2
2 1
2 4 2
(x y)4
Then, we obtain
=
2
gYM
4 2
3
(fpa1 b1 fpa2 b2 + fpa1 b2 fpa2 b1 )
ln(x y)2 2
.
(x y)4
(A.39)
(A.40)
198
A.2.4. ar1 (x)Z a2 (x) bs1 (y)Z b2 (y)
The log divergence which comes through the Yukawa coupling interaction is
2 5
g
ifpa b ifpa b
1
8 12 2 22 1
rs YM2
2
4
2gYM 2gYM
d 4 u d 4 v x
1
1
1
1
1
v
v
.
2
2
2
2
(x u) (y u)
(u v)
(y v) (x v)2
(A.41)
1
x H + 2 2 x Xxxyy ,
2
(A.42)
where H is defined as
d 4u d 4v
1
1
1
1
1
u
.
2
2
2
2
(x u) (y u)
(u v) (x v) (y v)2
(A.43)
The first term in (A.42) does not give the log divergence, and the result is
2
gYM
4 2
3
1
1
x
rs fpa1 b2 fpa2 b1
ln(x y)2 2 .
(x y)2
(x y)2
(A.44)
2 5
g
fpa1 b1 fpa2 b2
1
= 2 2
rs YM2
2
2
4
gYM gYM
1
1
1
1
1
(x u)2
(y u)2 (u v)2 (x v)2 (x v)2
2 5
gYM fpa1 b1 fpa2 b2
1
1
2
=
rs
x Yxxy
x x Hxxyy + 8
2
2
4
4 2
(x y)2
gYM
gYM
2 3
1 gYM
1
1
=
rs fpa1 b1 fpa2 b2
( ) x
ln(x y)2 .
2 4 2
(x y)2
(x y)2
(A.45)
d 4 u d 4 v x
199
1
rs fpa1 b2 fpa2 b1 + fpa1 b1 fpa2 b2
2
1
1
ln(x y)2 2 .
x
(x y)2
(x y)2
2
gYM
4 2
3
(A.46)
2 5
gYM
fpa b fpa2 b2
1
4 21 1
ij
2
2
4 2
gYM 2gYM
d 4u d 4v
1
1
1
1
1
u
v
(u v)2 (x u)2
(y u)2 (x v)2
(y v)2
2 3
ln(x y)2 2
1 gYM
=
f
f
pa
b
pa
b
ij
1
1
2
2
2 4 2
(x y)4
while the scalar 4point function is
(A.47)
2 3
ln(x y)2
1 gYM
=
(f
f
f
f
)
.
pa
b
pa
b
pa
a
pb
b
ij
1 2
2 1
1 2
1 2
2 4 2
(x y)4
(A.48)
2
gYM
4 2
3
fpa1 b2 fpa2 b1 ij
ln(x y)2 2
.
(x y)4
(A.49)
200
1
1
1
1
1
x
u
v
2
2
2
2
(x v) (y u)
(x u)
(u v)
(y v)2
2 5
gYM fpa1 b2 fpa2 b1 j i
1
2
rs 2 x Xxxyy + x H
=
2
2
2
4 2
gYM
gYM
2 3
1
1 gYM j i
1
rs fpa1 b2 fpa2 b1
x
=
2 4 2
(x y)2
(x y)2
d 4u d 4v
ln(x y)2 2 ,
ifpa1 a2 ifpb1 b2
2
gYM
2
gYM
d 4u d 4v
(A.50)
2 5
t i
g
YM2
s r t
4
1
1
1
1
1
x
u
v
2
2
2
2
(x u) (y v)
(x u)
(u v)
(y v)2
gYM 5 fpa1 a2 fpb1 b2 i j
rs 2 x Xxxyy
=
2
2
4 2
gYM
gYM
2 3
i j
1
1 gYM
1
=
f
f
pa1 a2 pb1 b2
rs (x y)2
x
4 4 2
(x y)2
ln(x y)2 2 .
(A.51)
2 3
1
1 gYM
1
f
f
ln(x u)2 2 .
pa1 b1 pa2 b2 rs
x
2 4 2
(x y)2
(x y)2
(A.52)
Then we obtain
2 3
1 gYM
1
=
j i rs fpa1 b2 fpa2 b1 + i j rs fpa1 a2 fpb1 b2
2
2 4
2
1
1
+ ij rs fpa1 b1 fpa2 b2
ln(x y)2 2 .
x
(x y)2
(x y)2
(A.53)
201
2 5
gYM
ifpa b ifpa2 b1 i
1
4 21 2
r r
4 2
2
2
g
2gYM
YM
1
1
1
1
1
x
v
y
(x v)2 (y u)2
(x u)2
(u v)2
(y v)2
2 3
i
1
1 gYM
1
=
f
f
d 4u d 4v
ln(x y)2 2 ,
(A.54)
2 5
i ifpa1 a2 ifpb1 b2
g
1
= 4 r r
2 YM2
2
2
4
gYM
2gYM
1
1
1
1
1
x
u
y
2
2
2
2
(x u) (y v)
(x u)
(u v)
(y v)2
2 3
gYM i
1
1
1
=
f
f
d 4u d 4v
ln(x y)2 2 .
(A.55)
Then the sum of the above contribution gives the result (4.27):
(A.56)
2 3
i
1 gYM
1
r r fpa1 b2 fpa2 b1 + fpa1 a2 fpb1 b2
=
2
2 4 2
1
1
x
ln(x y)2 2 .
2
(x y)
(x y)2
A.2.8. ra1 (x)Z a2 (x)sb1 (y)Z b2 (y)
The contribution from the Yukawa coupling is
2 5
g
fpa1 a2 fpb2 b1
1
= 8 2
r s YM2
2
2
4
gYM gYM
(A.57)
202
1
1
1
1
1
u
v
y
2
2
2
2
(x u) (y v)
(x u)
(u v)
(y v)2
2 3
1
1 gYM
1
=
fpa1 a2 fpb1 b2
r s
2
2
2 4
(x y)
(x y)2
d 4u d 4v
ln(x y)2 2 .
(A.58)
2 3
1
1 gYM
ln(x u)2 2
f
f
.
pa1 b1 pa2 b2 r s
x
2 4 2
(x y)2
(x y)2
(A.59)
Therefore we obtain
2 3
1 gYM
(fpa1 b2 fpa2 b1 fpa1 b1 fpa2 b2 )
r s
2 4 2
1
1
ln(x y)2 2 .
x
(x y)2
(x y)2
(A.60)
Appendix B. Prefactor
The form of the prefactor of the interaction vertex H3 SV is4
H3 SV = PSV E,
1 i j
PSV =
K K + ij Vij K K + V
2
K 1 1 K 2 2 S1 2 (Y )S1 2 (Z) K 1 1 K 2 2 S1 2 (Y )S 1 2 (Z) ,
(B.1)
(B.2)
K J = XIJ XIIJ ,
Fn(r) an(r) ,
XII =
r=1 n=0
Y 1 2 =
3
r=1 n=0
(B.3)
(r)
iUn(r) Fn(r) an ,
(B.4)
r=1 n=1
(r)1 2
(r)
G
,
n bn
Z 1 2 =
(r) 1 2
(r)
G
,
n bn
(B.5)
r=1 n=0
4 This definition differs form the one in (3.28) of [26] by a factor 2/ . Note also the difference of the total factor
of K i between here and there.
with
1/2 (r)
(r)
(1 4K) Fn ,
Gn = (1 4K)1/2 G(r)
n ,
2
2
(2) (3) ,
F (3) =
(3) (2) ,
F (1) = 0,
F (2) =
1
1 1 1/2
U(r) C(r) CN r n (n > 0),
Fn(r) =
1 4K (r)
1
1
(2)
(3)
G =
(2) (3) ,
G =
(3) (2) ,
G(1) = 0,
1
203
Fn(r) =
1 4 4
1 4
Vij ij 1 +
Y + Z4 +
Y Z
12
144
i 2
1 4
1 4
1
2
Yij 1 + Z Zij 1 + Y
+ Y 2 Z 2 ij ,
2
12
12
4
1 4 4
1 4
4
Y +Z +
Y Z
V 1
12
144
i 2
1
1
1
2
Y
1 Z 4 Z
1 Y 4 + Y 2 Z 2 ,
2
12
12
4
i 3
S(Y ) Y + Y ,
3
(B.6)
(B.7)
(B.8)
(B.9)
(B.10)
(B.11)
(B.12)
(B.13)
with
r r
K r r K i i
Y1 2 Y12 ,
ij
Y 2ij Y 21 1 1 1 ,
K r r K i i r r
Y22 2
(r = 1, 2),
(B.14)
Y1 2 Y21 ,
(B.15)
Y 4 Y21 1 Y 21 1 ,
Z 2ij Z 2 1 1
ij
(B.16)
1 1
Y 2Z
2 ij
Y 2k(i Z 2j )k .
3
8
(r)
m (r)I (r)I (r)a (r)a
H3 D =
am am +
bm bm
E
m= (r)
r=1
I =1
(B.17)
(B.18)
a=1
can be written as
H3 D = PD E,
1
PD = K 2 + K 2 Y 1 2 Y1 2 Z 1 2 Z 1 2 ,
4
(B.19)
(B.20)
204
where
Y 1 2 =
3
n (r) (r)1 2
G b
,
(r) n n
Z 1 2 =
r=1
3
n (r) (r)1 2
G b
.
(r) n n
(B.21)
r=1
(1)m+n
22
=
,
N mn
4(1) y
33
=
N mn
23
=
N mn
1
,
4(1) (1 y)
(1)m+n+1 sin(ny)
21
N mn
=
,
y(n m/y)
and, for m = n = 0, by
11
N 00
= 0,
12
N 00
= y,
1
,
4(1)  y(1 y)
1
33
N 00
=
.
4(1) (1 y)
23
=
N 00
(1)m+1
,
4(1)  y(1 y)
(C.1)
(1)n sin(ny)
1 y(n m/(1 y))
13
N 00
= 1 y,
22
=
N 00
(C.2)
(C.3)
(C.4)
1
,
4(1) y
(C.5)
(r)
(r)
When we compute string amplitudes for fermions, the large behavior of Fn(r) , G
n , Um
5
and 1 4K are also useful :
(1)n+1
(2)
(3)
,
Fn =
,
Fn =
(1) y
(1) (1 y)
(1)n+1 n sin(ny)
(1)
Fn =
(C.6)
,
((1) )3
(2)
(3)
(1)
,
F0 = 0,
,
F0 =
F0 =
(C.7)
2(1) y
2(1) (1 y)
(1)n+1
1
(3)
(2)
,
,
G
G
n =
n =
2(1) y
2(1) (1 y)
(1)n+1 2 sin(ny)
(1)
Gn =
,
(1) 
1
1
(2) =
(3) =
,
G
,
G
0
0
4(1) y
4(1) (1 y)
(C.8)
(1) = 0,
G
0
(C.9)
5 Note that the definition of the Neumann vector N r here, with which F (r) is defined, differs by C 1/2 U from that
r
n
n
(r)
of Ref. [37].
Un(2) =
n
,
2(1) y
1 4K =
Un(3) =
n
,
2(1) (1 y)
1
.
4(1) y(1 y)
Un(1) =
2(1) 
,
n
205
(C.10)
(C.11)
References
[1] S.S. Gubser, I.R. Klebanov, A.M. Polyakov, Gauge theory correlators from noncritical string theory, Phys. Lett.
B 428 (1998) 105, hepth/9802109;
E. Witten, Antide Sitter space and holography, Adv. Theor. Math. Phys. 2 (1998) 253, hepth/9802150.
[2] O. Aharony, S.S. Gubser, J.M. Maldacena, H. Ooguri, Y. Oz, Large N field theories, string theory and gravity, Phys.
Rep. 323 (2000) 183, hepth/9905111;
E. DHoker, D.Z. Freedman, Supersymmetric gauge theories and the AdS/CFT correspondence, TASI 2001 Lecture
Notes, hepth/0201253.
[3] S.M. Lee, S. Minwalla, M. Rangamani, N. Seiberg, Three point functions of chiral operators in D = 4, N = 4 SYM
at large N , Adv. Theor. Math. Phys. 2 (1998) 697, hepth/9806074.
[4] D. Berenstein, J.M. Maldacena, H. Nastase, Strings in flat space and pp waves from N = 4 superYangMills,
JHEP 0204 (2002) 013, hepth/0202021.
[5] S. Dobashi, H. Shimada, T. Yoneya, Holographic reformulation of string theory on AdS5 S 5 background in the
PPwave limit, Nucl. Phys. B 665 (2003) 94, hepth/0209251.
[6] T. Yoneya, What is holography in the plane wave limit of AdS(5)/SYM(4) correspondence?, hepth/0304183 (expanded from the paper published in the Proceedings, Prog. Theor. Phys. 152 (2003) 108).
[7] S. Dobashi, T. Yoneya, Resolving the holography in the planewave limit of AdS/CFT correspondence, hepth/0406225.
[8] S. Dobashi, T. Yoneya, Impurity nonpreserving 3point correlators of BMN operators from PPwave holography I:
Bosonic excitations, hepth/0409058.
[9] M. Asano, Y. Sekino, T. Yoneya, PPwave holography for Dpbrane backgrounds, Nucl. Phys. B 678 (2004) 197,
hepth/0308024;
M. Asano, Y. Sekino, Large N limit of SYM theories with 16 supercharges from superstrings on Dpbrane backgrounds, hepth/0405203;
M. Asano, Stringy effect of the holographic correspondence for Dpbrane backgrounds, hepth/0408030.
[10] D.J. Gross, A. Mikhailov, R. Roiban, A calculation of the plane wave string Hamiltonian from N = 4 superYang
Mills theory, JHEP 0305 (2003) 025, hepth/0208231.
[11] J. Gomis, S. Moriyama, J. Park, SYM description of SFT Hamiltonian in a pp wave background, Nucl. Phys. B 659
(2003) 179, hepth/0210153.
[12] P. Di Vecchia, J.L. Petersen, M. Petrini, R. Russo, A. Tanzini, The three string vertex and the AdS/CFT duality in
the pp wave limit, hepth/0304025.
[13] R. Roiban, M. Spradlin, A. Volovich, On light cone SFT contact terms in a plane wave, JHEP 0310 (2003) 055,
hepth/0211220.
[14] P. Gutjahr, A. Pankiewicz, New aspects of the BMN correspondence beyond the planar limit, hepth/0407098.
[15] G. Grignani, M. Orselli, B. Ramadanovic, G.W. Semenoff, D. Young, Divergence cancellation and loop corrections
in string field theory on a plane wave background, hepth/0508126.
[16] J. Gomis, S. Moriyama, J. Park, SYM description of pp wave string interactions: Singlet sector and arbitrary impurities, Nucl. Phys. B 665 (2003) 49, hepth/0301250.
[17] C.S. Chu, V.V. Khoze, G. Travaglini, BMN operators with vector impurities, Z2 symmetry and ppwaves,
JHEP 0306 (2003) 050, hepth/0303107.
[18] G. Georgiou, V.V. Khoze, G. Travaglini, New tests of the pp wave correspondence, JHEP 0310 (2003) 049, hepth/0306234.
[19] G. Georgiou, G. Travaglini, Fermion BMN operators, the dilatation operator of N = 4 SYM, and pp wave string
interactions, JHEP 0404 (2004) 001, hepth/0403188.
[20] N. Beisert, C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, BMN correlators and operator mixing in N = 4
superYangMills theory, Nucl. Phys. B 650 (2003) 125161, hepth/0208178.
[21] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, Operator mixing and the BMN correspondence,
JHEP 0210 (2002) 068, hepth/0209002.
206
[22] M. Spradlin, A. Volovich, Superstring interactions in a ppwave background, Phys. Rev. D 66 (2002) 086004,
hepth/0204146.
[23] M. Spradlin, A. Volovich, Superstring interactions in a ppwave background 2, JHEP 0301 (2003) 036, hepth/0206073.
[24] A. Pankiewicz, More comments on superstring interactions in the ppwave background, JHEP 0209 (2002) 056,
hepth/0208209.
[25] A. Pankiewicz, B. Stefanski Jr., PPwave light cone superstring field theory, Nucl. Phys. B 657 (2003) 79, hepth/0210246.
[26] A. Pankiewicz, An alternative formulation of light cone string field theory on the plane wave, JHEP 0306 (2003)
047, hepth/0304232.
[27] A. Pankiewicz, B. Stefanski Jr., On the uniqueness of plane wave string field theory, hepth/0308062.
[28] C.S. Chu, V.V. Khoze, Correspondence between the three point BMN correlators and the three string vertex on the
ppwave, JHEP 0304 (2003) 014, hepth/0301036.
[29] S. Lee, R. Russo, Holographic cubic vertex in the PPwave, Nucl. Phys. B 705 (2005) 296, hepth/0409261.
[30] S. Dobashi, in preparation.
[31] N.R. Constable, D.Z. Freedman, M. Headrick, S. Minwalla, L. Motl, A. Postnikov, W. Skiba, PPwave string interactions from perturbative YangMills theory, JHEP 0207 (2002) 017, hepth/0205089.
[32] A. Parnachev, A.V. Ryzhov, Strings in the near plane wave background and AdS/CFT, JHEP 0210 (2002) 066,
hepth/0208010.
[33] U. Gursoy, Vector operators in the BMN correspondence, hepth/0208041.
[34] N. Beisert, BMN operators and superconformal symmetry, Nucl. Phys. B 659 (2003) 79, hepth/0211032.
[35] C. Kristjansen, J. Plefka, G.W. Semenoff, M. Staudacher, A new double scaling limit of N = 4 superYangMills
theory and ppwave strings, Nucl. Phys. B 643 (2002) 3, hepth/0205033.
[36] D.Z. Freedman, K. Johnson, J.I. Latoree, Differential regularization and renormalization: A new method of calculation in quantum field theory, Nucl. Phys. B 371 (1992) 353.
[37] Y.H. He, J.H. Schwarz, M. Spradlin, A. Volovich, Explicit formulas for Neumann coefficients in the plane wave
geometry, Phys. Rev. D 67 (2003) 086005, hepth/0211198;
See also, J. Lucietti, S. SchaferNameki, A. Sinha, On the planewave cubic vertex, Phys. Rev. D 70 (2004) 026005,
hepth/0402185.
Abstract
We define a superspace over a ring R as a functor on a subcategory of the category of supercommutative
Ralgebras. As an application the notion of a padic superspace is introduced and used to give a transparent
construction of the Frobenius map on padic cohomology of a smooth projective variety over Zp (the ring
of padic integers).
2006 Elsevier B.V. All rights reserved.
1. Introduction
It is possible that all physical quantities are measured in rational numbers (or even in integers
if there exist elementary length, elementary unit of time, etc.). All other numbers should be
introduced only for mathematical convenience as it is much more practical to work with an
algebraically closed field that is complete with respect to some norm.
Let us consider an oversimplified example where time and space are quantized and the motion
of a particle satisfies the equation
2 x(t) = F x(t), t .
(1)
Here 2 stands for the second difference:
2 x(t) = x(t + 2) x(t + 1) x(t + 1) x(t)
= x(t + 2) 2x(t + 1) + x(t).
* Corresponding author.
05503213/$ see front matter 2006 Elsevier B.V. All rights reserved.
doi:10.1016/j.nuclphysb.2006.08.024
(2)
208
Eq. (1) which is the finite difference analogue of the Newtons second law, permits us to calculate
recursively x(t) for t Z if we know x(0), x(1) and F ; if the initial data x(0), x(1) and the
force F are all integers then all coordinates x(t) will also be integers. In some sense, all
physical questions can be answered if we know only integer numbers. However, if we would like
to write down an explicit solution of (1) even in the simplest situation when F is a linear function
of x(t) and does not depend explicitly on t we need irrational and complex numbers. Indeed, the
general solution to the equation
2 x(t) = ax(t)
(3)
(4)
209
Recall that the Frobenius map of a field of characteristic p transforms x into x p . This map is
an endomorphism of the field, it is furthermore an automorphism for finite fields and an identity
map for the field Fp consisting of p elements. Of course the pth power map is an algebra homomorphism for any Fp algebra.2 This last observation leads to the Frobenius map on a variety
over Fp . One can consider the pth power map also on the padic numbers, but in this case it is
not a homomorphism. We can define the Frobenius map on the ndimensional affine or projective
space over Zp by raising the coordinates to the pth power. However, a variety sitting inside (an
affine or projective variety) will not, in general, be invariant with respect to the Frobenius map.
Nevertheless, one can define the Frobenius map on the cohomology with coefficients in the ring
of padic integers Zp .
The standard approach to the construction is quite complicated; it is based on the consideration of the socalled DPneighborhood of a variety that has the same cohomology as the original
variety and at the same time is invariant with respect to the Frobenius map. One of the main goals
of the present paper is to give a simplified construction of the Frobenius map in terms of supergeometry. The logic of our construction remains the same, but the role of a DPneighborhood is
played by a padic superspace having the same body as the original variety.3
We hope that our approach to the Frobenius map will be much more accessible to physicists
than the standard one based on the consideration of the crystalline site and DPneighborhoods [1].
Our considerations, as presented here, are not completely rigorous; we did not want to exceed
significantly the level of rigor that is standard for a physics journal. However, we believe that
our results are of interest also to pure mathematicians and therefore we have written a rigorous
exposition of our construction of the Frobenius map [8].
It seems that arithmetic geometry should play an essential role in string theory; the paper [4]
sketches one way of applying arithmetic geometry, but there are also other ways (let us mention
a series of papers by Schimmrigk, for example [10]). The present paper shows that, conversely,
some ideas borrowed from physics can be used to clarify some important notions of arithmetic
geometry.
As we mentioned, our construction is based on the notion of a padic superspace. We hope
that this notion will have also other applications. In particular, we expect that it can be used
to construct supersymmetric sigmamodels and topological sigmamodels in the padic setting.
The padic Bmodel used in [4] was defined completely formally as the theory of variations of
Hodge structures. However it is natural to conjecture that it can be defined in the framework of
Lagrangian formalism and that it is related to the (conjectural) padic supersymmetric sigmamodel.
One should notice that in an appropriate coordinate system the operations in the (super)group
of supersymmetries are given by polynomial formulas with integer coefficients; this means that
one can define the supersymmetry group over an arbitrary commutative ring R assuming that the
2 An Ralgebra is a ring A equipped with a ring homomorphism R A. This notion is perhaps more familiar in the
case when R is a field. All rings are assumed to be unital and associative.
3 Varieties we consider are defined over the ring of padic integers Z . The elements of this ring can be represented
p
as formal power series ak pk where the coefficients a0 , a1 , . . . are integers between 0 and p 1. If the series is finite
it specifies a conventional integer and the operations on padic numbers coincide with operations in Z. Assigning the
coefficient a0 to a series
ak pk we obtain a homomorphism of Zp onto the field Fp . Using this homomorphism we
obtain from every variety over the padic integers a variety over Fp ; we say that this variety is the body of the variety
over the padic integers.
210
coordinates belong to this ring (for example, one may consider the supersymmetry group over Z;
then we can talk about supersymmetry over integers).
To construct supersymmetric objects it is natural to introduce a notion of superspace over
ring R; this can be done in many ways. In the conventional superspace over real numbers coordinates are considered as even or odd elements of a Grassmann algebra. The algebra of functions
on the (m, n)dimensional affine superspace over R can be considered as a tensor product of the
algebra of smooth functions of m real variables and the Grassmann algebra with n generators. In
mathematical formulation [7] the conventional superspace is regarded as a functor on the category of Grassmann algebras with values in the category of sets (we assign to every Grassmann
algebra the set of points with coordinates from ). A function on a superspace is defined as
a map of functors. It is clear that the functions on superspace described above specify maps of
functors (natural transformations). It is not so clear that there are no other natural transformations, but this can be proven [11]. One can consider also superspaces where in addition to m even
and n odd coordinates we have r even nilpotent coordinates; then there is an additional factor
(the algebra of formal power series with r variables) in the tensor product specifying the algebra
of functions (see [3]). It is important to notice that even considering only superspaces over real
numbers one can modify the definition of a superspace in a variety of ways to obtain different
classes of functions. For example, we can assume that the coordinates are even and odd elements
of an arbitrary supercommutative algebra over R; then only polynomial functions are allowed.
If we take as the starting point the functorial approach, the definition of a superspace over a
ring R becomes very natural: we should fix a subcategory of the category of supercommutative
Ralgebras and define a superspace as a functor from this subcategory to the category of sets. The
algebra of functions depends on the choice of a subcategory. Notice that the functorial approach
is common in algebraic geometry where a variety over a ring R specifies a functor on the category of Ralgebras with values in the category of sets. (A variety is singled out by polynomial
equations with coefficients in R in affine or projective space over R. These equations make sense
if the coordinates are considered as elements of an Ralgebra; we can consider the corresponding set of solutions. Similarly one can define the notion of a supervariety singled out by means
of polynomial equations in an affine or projective superspace and construct the corresponding
functor.)
We are interested in the case when R is the ring of padic integers Zp and we define the
subcategory (p) as the category of rings of the form B where B is a commutative ring
such that pN0 B = 0 and B/pB does not contain nilpotent elements; stands for a Grassmann
ring. By definition, a padic superspace is a functor on the category (p) with values in the
category of sets. As always, we can define an affine superspace as a space with m even, r even
nilpotent and n odd coordinates. Functions on such a superspace are described in [8]. In this
paper we need only the
case when m = n = 0. If r = 1 (one even nilpotent coordinate), then
cn n
functions have the form
n! z where cn Zp . The superspace at hand can be considered as an
of a point on a line; it follows from the description of functions that
infinitesimal neighborhood pt
coincides with the cohomology of a point (this is a crucial observation that
the cohomology of pt
r with r even nilpotent coordinates
is used in the construction of the Frobenius map). The space pt
can be regarded as an infinitesimal neighborhood of a point in the rdimensional affine space; its
r does
cohomology also coincides with the cohomology of a point. Notice that the superspace pt
not have any odd coordinates (even though Grassmann algebras were used in its construction).
Let us restrict the functor defining a padic superspace to the subcategory of Fp algebras
without nilpotent elements. If there exists a variety over Fp that specifies the same functor we
say that this variety is the body of the superspace at hand.
211
Consider a projective variety V over Zp . We mentioned already that the Frobenius map acts
on the projective space sending a point with homogeneous coordinates (x0 : . . . : xn ) to the point
p
p
with coordinates (x0 : . . . : xn ). The variety V is not invariant with respect to the Frobenius map,
but its body is invariant. (We can define the body considering the functor XV corresponding to the
variety V or we can use the simpler definition in the footnote 3; both definitions lead to the same
result.) Now we consider the maximal subsuperspace X
V of the projective space having the same
body as V (and call it the infinitesimal neighborhood of XV ). It is obvious that X
V is invariant
with respect to the Frobenius map. This means that the Frobenius map induces a homomorphism
on the cohomology groups of the padic superspace X
V . If V is smooth one can prove that the
are
isomorphic
to
the
cohomology
groups of V . (The comparison
cohomology groups of X
V
of cohomology groups can be reduced to local calculations and locally we use the isomorphism
r and the cohomology of a point.) We thus obtain an action of
between the cohomology of pt
Frobenius map on cohomology groups of a nonsingular projective variety over Zp (Section 6).
In the last section we demonstrate the properties of Frobenius map that were used in [4]. In the
paper [9] the same properties were used to express the Frobenius map on the cohomology of a
CalabiYau threefold in terms of the mirror map and instanton numbers.
2. Superspaces
Let us denote by A a supercommutative ring (i.e., a Z2 graded ring A = A0 A1 where
a0 b = ba0 , a1 a1 = a1 a1 if a0 A0 , a1 , a1 A1 , b A).
One can say that Ap,q = {(x1 , . . . , xp , 1 , . . . , q )  xi A0 , i A1 } is a set of Apoints of
(p, q)dimensional superspace (of space with p commuting and q anticommuting coordinates).
One can consider Ap,q as an Abelian group or as an A0 module, but these structures will not be
important in what follows.
A parity preserving homomorphism : A A induces a map F () : Ap,q A p,q . It is obvious that F () = F () F () and F (id) = id. This means that we can consider a functor Ap,q
on the category of supercommutative rings taking values in the category of sets and assigning
Ap,q to the ring A.
One can define a superspace as a functor from the category of supercommutative rings into
the category of sets; then Ap,q represents the (p, q)dimensional affine superspace (over Z).
We define a map of superspaces as a map of functors. In particular a map of (p, q)dimensional superspace into a (p , q )dimensional superspace is defined as a collection of maps
A : Ap,q Ap ,q that are compatible with parity preserving homomorphisms of rings. In other
words, for every parity preserving homomorphism : A A the diagram
Ap,q
F ()
Ap ,q
F ()
A
A p,q
A p ,q
should be commutative.
p,q
Let us denote by OZ a Z2 graded ring of polynomials depending on p commuting and q
anticommuting variables and having integer coefficients. In other words:
p,q
OZ = Z x 1 , . . . , x p Z 1 , . . . , q
is a tensor product of the polynomial ring Z[x 1 , . . . , x p ] and the Grassmann ring Z [ 1 , . . . , q ],
the Z2 grading comes from Z2 grading of the Grassmann ring.
212
It is easy to check that an even element of OZ determines a map of the (p, q)dimensional
p,q
affine superspace into the (1, 0)dimensional affine superspace and an odd element of OZ determines a map into (0, 1)dimensional affine superspace. More generally, a row (f1 , . . . , fp ,
p,q
p,q
1 , . . . , q ) of p even elements of OZ and q odd elements of OZ specifies a map of (p, q)
dimensional superspace into (p , q )dimensional superspace (a map of functors Ap,q Ap ,q ).
p,q
The construction is obvious: we substitute elements of the ring A instead of generators of OZ .
One can prove that all maps of functors Ap,q Ap ,q are described by means of this construction.
More examples of superspaces can be constructed as affine supervarieties. Affine supervariety
(as usual affine variety) can be defined by means of polynomial equations:
f1 x 1 , . . . , x p , 1 , . . . , q = 0,
..
.
fp x 1 , . . . , x p , 1 , . . . , q = 0,
1 x 1 , . . . , x p , 1 , . . . , q = 0,
..
.
q x 1 , . . . , x p , 1 , . . . , q = 0.
{(x0 , . . . , xn , 1 , . . . , m ) (An+1
0 ) A1 }/A0 where A is a supercommutative ring. One easily
checks that this results in a functor thereby defining the projective superspace.
As before we may get a more general object, namely a projective supervariety by considering
the solutions of graded homogeneous equations in the variables x i and i , that is solutions of a
system:
f1 x 0 , . . . , x n , 1 , . . . , m = 0,
..
.
fs x 0 , . . . , x n , 1 , . . . , m = 0,
1 x 0 , . . . , x n , 1 , . . . , m = 0,
..
.
t x 0 , . . . , x n , 1 , . . . , m = 0,
213
affine or projective scheme. The term variety is reserved for schemes where local rings do not have nilpotent elements;
we will assume that this condition is satisfied. Similarly one should talk about affine superschemes reserving the word
supervariety for the case when underlying scheme is a variety.
5 A different (in general) superspace can be obtained by considering E (A) = Hom (E, A) where we require the
R
morphisms to be parity preserving. When E is free over R, this is the same as the tensor construction.
214
on Ralgebras to Cred coincides with the restriction of the superspace to Cred . This variety is not
unique, however in the situation we consider later, there exists such an ideal M R that Cred can
be identified with the subcategory of commutative Ralgebras consisting of R/Mrings without
nilpotent elements, thus the body is a unique R/Mvariety. If our superspace is an affine or projective supervariety then the notion of body agrees with the notion of underlying variety (if the
body is considered as a variety over R).
If X is a superspace we will denote its body considered as a variety by Xbody . Every
subvariety Z Xbody determines a subsuperspace XZ of X that can be thought of as the
largest subsuperspace of X with body Z. More precisely, for every A C we have a map
A : X(A) X(A/Anilp ) induced by the morphism A A/Anilp . The subvariety Z specifies a
subset Z(A/Anilp ) X(A/Anilp ), we define XZ (A) = A1 (Z(A/Anilp )).
Consider an inclusion of superspaces X Y , that is a natural transformation i such that for
every A, iA : X(A) Y (A) identifies X(A) with a subset of Y (A). Note that the variety Xbody is
a subvariety of Ybody . Define the infinitesimal neighborhood X of X in Y , by setting X = Y Xbody .
For example, if X is the origin in Y = A1 , the infinitesimal neighborhood of X
Clearly X X.
defined earlier.
is the superspace pt
4. de Rham cohomology of a superspace
Given a general superspace X, we associate to it a new superspace T X called the odd
tangent space, defined by T X(A) = X(A Z []). It should be thought of as the superspace
parameterizing the maps from A0,1 to X.
To make the definition more transparent, consider the following more familiar situation. Let us
assume that X is an affine algebraic supervariety defined by polynomial equations fi (x j , k ) = 0
and i (x j , k ) = 0 as before. Then T X is given by polynomial equations obtained by setting
x i = y i + i and j = j + zj in fi = 0 and j = 0. (Here y i , zj are even variables and i ,
i j
i
j
j , are odd.) T X is then an affine supervariety
with coordinates
i
j y , z i, j, constrained
i
j
i
j
by the even equations fk (y , ) = 0 and y i k (y , ) + z j k (y , ) = 0 as well as
the odd equations k (y i , j ) = 0 and i y i fk (y i , j ) + zj j fk (y i , j ) = 0.
The functions on T X are (by definition) differential forms on X. The differential on the ring
of functions on T X (the ring (X) of differential forms on X) can be defined by dy i = i ,
di = 0, d j = zj , and dzj = 0.
The differential on T X has geometric origin that permits one to define it for every superspace X. Namely A0,1 has a structure of a supergroup and therefore it acts on itself in an
obvious way. Viewing T X as the superspace parameterizing maps from A0,1 to X we obtain an action of A0,1 on it. Similarly we have an action of A on T X derived from the
action of A on A0,1 via A (A) A0,1 (A) A0,1 (A) which is just the multiplication map
A A1 A1 . The actions of A0,1 and A on T X are compatible in the sense that the
semidirect product A0,1 A acts on T X. Furthermore this construction is functorial, i.e.,
a natural transformation : X Y induces a natural transformation between the odd tangent
spaces d : T X T Y that is compatible with the action of A0,1 A .
The action of A0,1 on T X and thus on the differential forms on X (= functions on T X)
specifies a differential d on (X) that can be used to define the cohomology groups ker d/ im d.
These groups are in fact Rmodules. The action of A descends to the subquotient and gives a
grading on the cohomology.
In the case when X is a smooth affine algebraic variety (considered as a functor) and the
source category is the full category of supercommutative Ralgebras, the cohomology groups
215
coincide with the familiar de Rham cohomology. They may or may not coincide with them if the
source category is different.
To define the de Rham cohomology in a nonaffine (but still smooth case6 ) recall that we
assumed that the body of the superspace X is an algebraic variety. For every open subvariety
(U ) as the Ralgebra of differU Xbody , consider the restriction XU of X to U . Define X/R
(U ) is a differential
ential forms on XU , i.e., functions on T (XU ). As observed above X/R
0,1
Ralgebra with the differential given by the action of A and the grading given by the action
(U ) for every U specify a sheaf of differential Rmodules on
of A . The collection of X/R
Xbody . We define the de Rham cohomology of X as hypercohomology of this sheaf. That is we
6 In the nonsmooth case one follows the procedure below applied to Y  instead of X itself, where Y is smooth and
X
contains X. For this to work one has to choose an appropriate source category for the functors specifying the superspaces.
7 Naturally because we can no longer divide by n.
8 Examples of such rings B are Z
pk and Zpk [x].
216
It is not difficult to prove that every series of this kind specifies a function (or a form). We
nilp
+
notice first that B = pB + +
B , where we denote by B the ideal generated by i s. Then the
claim above follows from these observations:
(A) For every nilpotent element
Z we can consider n /n! as an element of Z , furthermore
n
/n! = 0 for n large, thus an zn /n! can be evaluated at .
(B)
The ring B can be considered as a Zp algebra since multiplication by an infinite series
an p n makes sense since we have p N0 B = 0.
n /n! in the
(C) For every pB the expression n /n! makes sense as an element of B since p
reduced form has no p factors in the denominator. Furthermore,
if p > 2 then
an zn /n!
n
can be evaluated at p to obtain an element of Zp and so an z /n! can be evaluated at .
is more complicated. The proof is given
The statement that these are all the functions on pt
in [8], Section 4.1. From the viewpoint of a physicist this proof is irrelevant: we can restrict
ourselves to the functions described above and not worry about other
functions.
is given by the formula d( an zn /n! + bn zn /n! dz) =
The
differential
on
forms
on
pt
an zn1 /(n 1)! dz. It follows immediately from this formula that the cohomology of pt
coincides with that of pt.
Let us now consider the cohomology of the infinitesimal neighborhood X of a smooth variety X singled out by a single equation f (z) = 0. Locally the infinitesimal neighborhood is
hence its cohomology coincides with the cohomology of X.
a direct product of X with pt;
w=
More
precisely
we
have
a
local
expression for a differential form of degree s on X:
an f n /n! + bn f n /n! df , where an and bn are differential forms on X of degree s and
s 1, respectively. One can derive the fact we need from the formula
h(w) = (1)s1
n=0
and X.
The above statements are particular cases of a general theorem valid for a smooth subvariety
X of a smooth variety Y . (More generally, X and Y can be supervarieties.) That is: If X and Y are
considered as superspaces over the category (p) or over the category of all supercommutative
Qalgebras then the cohomology of the infinitesimal neighborhood X of X in Y is naturally
isomorphic to the cohomology of X.9
It follows from the above theorem that the cohomology of an infinitesimal neighborhood of a
smooth X does not depend on which particular embedding into a smooth Y one chooses.10
9 Recall that we defined cohomology as hypercohomology of the complex of sheaves. The embedding of X into
X induces a map of the corresponding complexes of sheaves that one can show, through local analysis, is a quasiisomorphism by constructing an explicit homotopy. In the case when X has codimension 1 this homotopy was constructed
above.
10 It seems one can prove that this remains true even in the case of a singular X. This prompts a definition of the
cohomology of a singular X as the cohomology of its infinitesimal neighborhood in some smooth Y . In the case of a
padic superspace X obtained from a variety over Fp we should obtain the crystalline cohomology of the variety as the
cohomology of X.
217
218
14 This is due to the fact that this filtration is preserved by the homotopy that establishes the isomorphism of cohomologies.
Abdalla, E.
Alls, B.
lvarez, E.
lvarezGaum, L.
Ananth, S.
Andrews, R.P.
Aoyama, T.
B752 (2006) 40
B752 (2006) 124
B756 (2006) 148
B753 (2006) 92
B753 (2006) 195
B751 (2006) 304
B754 (2006) 48
Bailin, D.
Baseilhac, P.
Baulieu, L.
Baulieu, L.
Becker, K.
Becker, M.
Beneke, M.
Berenstein, D.
Bhattacharjee, P.
Bietenholz, W.
Blas, D.
Blumenhagen, R.
Blmlein, J.
Blmlein, J.
Bolognesi, S.
Bolognesi, S.
Bornyakov, V.G.
Bossard, G.
Bossard, G.
Boughezal, R.
Boyko, P.Yu.
Brandenburg, A.
Brandt, F.T.
Brink, L.
Brmmer, F.
Burgess, C.P.
B755 (2006) 79
B754 (2006) 309
B753 (2006) 252
B753 (2006) 273
B751 (2006) 108
B751 (2006) 108
B751 (2006) 160
B753 (2006) 69
B752 (2006) 280
B754 (2006) 17
B756 (2006) 148
B751 (2006) 186
B755 (2006) 112
B755 (2006) 272
B752 (2006) 93
B754 (2006) 293
B756 (2006) 71
B753 (2006) 252
B753 (2006) 273
B755 (2006) 221
B756 (2006) 71
B754 (2006) 107
B754 (2006) 146
B753 (2006) 195
B755 (2006) 186
B752 (2006) 60
Calcagni, G.
Callin, P.
Chen, C.M.
Cieza Montalvo, J.E.
Cirelli, M.
Cirigliano, V.
Cirigliano, V.
Clark, S.S.
Cornou, J.L.
Cortez Jr., N.V.
CuadrosMelgar, B.
Czakon, M.
Czakon, M.
Danilov, G.S.
Dasgupta, K.
de Carlos, B.
De Felice, A.
De Freitas, A.
Delduc, F.
DElia, M.
Deppisch, F.
DHoker, E.
Diaconescu, D.E.
Dijkgraaf, R.
Dobashi, S.
Dolgov, A.D.
Donagi, R.
Dorey, N.
Ecker, G.
Eguchi, H.
Eidemller, M.
Enciso, A.
Engquist, J.
Estes, J.
Eto, M.
Fehr, L.
Feng, B.
Forgcs, P.
Fornengo, N.
Fr, P.
220
Frenkel, J.
Fu, J.X.
Gaillard, M.K.
Garavuso, R.S.
Garriga, J.
Geyer, B.
Giusto, S.
Glck, M.
Gluza, J.
Gran, U.
Grassi, P.A.
Grimus, W.
Grinstein, B.
Grinstein, B.
Grisaru, M.
Gudnason, S.B.
Gutowski, J.
Gutperle, M.
Gwyn, R.
B751 (2006) 75
B755 (2006) 329
B756 (2006) 148
B755 (2006) 112
B754 (2006) 233
B754 (2006) 178
B751 (2006) 1
B753 (2006) 118
B751 (2006) 53
B754 (2006) 1
B752 (2006) 18
B755 (2006) 199
B755 (2006) 21
B754 (2006) 293
B753 (2006) 118
B753 (2006) 16
B755 (2006) 21
He, X.
Hebecker, A.
Heise, R.
Hofman, C.
Hosteins, P.
Hou, D.f.
Huang, Q.G.
Idilbi, A.
Ilgenfritz, E.M.
Inami, T.
Isozumi, Y.
Ivanov, E.
B753 (2006) 42
B756 (2006) 71
B752 (2006) 391
B752 (2006) 140
B753 (2006) 211
Jger, S.
Jantzen, B.
Ji, X.
Jing, J.
Kaiser, R.
Katz, S.
Kitazawa, N.
Klein, S.
Knauf, A.
Koike, Y.
Kosmas, T.S.
Kovalenko, A.V.
Kozlovskii, M.P.
Khbck, H.
Khn, J.H.
Lavignac, S.
Lavoura, L.
Li, J.r.
Li, M.
Li, T.
Lipatov, L.N.
Loll, R.
Lombardo, M.P.
Love, A.
Macdo, A.M.S.
MacedoJunior, A.F.
Mangano, G.
Manvelyan, R.
Martemyanov, B.V.
Maru, N.
Masina, I.
Mathews, P.
Mathur, S.D.
McInnes, B.
Meyer, F.
Miele, G.
Minakami, S.
Mitov, A.
Moch, S.
Molina, C.
Morales Morera, J.F.
Moster, S.
MllerPreussker, M.
Muramoto, C.
Naculich, S.G.
Nakayama, Y.
Nanopoulos, D.V.
Nelson, B.D.
Nitta, M.
Nitta, M.
Ohashi, K.
Olesen, P.
Ooguri, H.
Ookouchi, Y.
Paccetti Correia, F.
Pajer, E.
Pallis, C.
Pantev, T.
Papadopoulos, G.
Pastor, S.
Pavan, A.B.
Penin, A.A.
Pich, A.
Pinansky, S.
Pinto, T.
Pisanti, O.
Polikarpov, M.I.
Polychronakos, A.P.
Portols, J.
Prytula, O.O.
Pusztai, B.G.
Pylyuk, I.V.
Rajabpour, M.A.
Ravindran, V.
Ravindran, V.
Ren, H.c.
Reuillon, S.
Reya, E.
Riemann, T.
Ringwald, A.
Robaschik, D.
RodrguezGmez, D.
Roest, D.
Rossi, G.C.
Rouhani, S.
Rhl, W.
S Borges, J.
Sahu, N.
Savoy, C.A.
Savoy, C.A.
Schmidt, M.G.
Schnitzer, H.J.
Schuck, C.
Schwarz, A.
Seki, S.
Serpico, P.D.
Shapiro, I.
Shcheredin, S.
Shibusa, Y.
Skvortsov, E.D.
Smirnov, V.A.
Sokatchev, E.
Sorella, S.P.
Sorella, S.P.
Srivastava, Y.K.
Stanev, Ya.S.
Strumia, A.
B756 (2006) 1
B752 (2006) 280
B755 (2006) 1
B755 (2006) 137
B751 (2006) 222
B755 (2006) 164
B754 (2006) 178
B756 (2006) 207
B753 (2006) 295
B756 (2006) 100
B756 (2006) 207
B754 (2006) 17
B754 (2006) 48
B756 (2006) 117
B752 (2006) 327
B754 (2006) 329
B753 (2006) 252
B753 (2006) 273
B754 (2006) 233
B754 (2006) 329
B753 (2006) 178
221
Sturani, R.
Sugiyama, K.
Sundell, P.
Svendsen, H.G.
B756 (2006) 16
B753 (2006) 295
B752 (2006) 206
B753 (2006) 195
Tanaka, K.
Tatar, R.
Tavartkiladze, Z.
Tokunaga, T.
Tonasse, M.D.
Trapletti, M.
Trigiante, M.
Tseng, L.S.
B752 (2006) 1
B755 (2006) 21
B751 (2006) 222
B753 (2006) 295
B756 (2006) 1
B755 (2006) 186
B751 (2006) 343
B751 (2006) 108
Urban, F.R.
Utermann, A.
Valle, J.W.F.
van Neerven, W.L.
Vasiliev, M.A.
VzquezMozo, M.A.
Verdaguer, E.
Veselov, A.I.
Volkov, M.S.
B752 (2006) 80
B755 (2006) 272
B756 (2006) 117
B753 (2006) 92
B756 (2006) 148
B756 (2006) 71
B751 (2006) 390
Wang, T.
Weigand, T.
Westra, W.
B756 (2006) 86
B751 (2006) 186
B751 (2006) 419
Yajnik, U.A.
Yamashita, T.
Yau, S.T.
Yuan, F.
Zohren, S.