a

r
X
i
v
:
g
r
-
q
c
/
9
4
0
9
0
5
6
v
1


2
7

S
e
p

1
9
9
4
Conditions for ination
in an initially inhomogeneous universe
Nathalie Deruelle
Departement dAstrophysique Relativiste et de Cosmologie,
Centre National de la Recherche Scientique, Observatoire de Paris, 92195 Meudon, France
and Department of Applied Mathematics and Theoretical Physics,
University of Cambridge, Silver Street, Cambridge CB3 9EW, England
Dalia S. Goldwirth
School of Physics and Astronomy, Raymond and Beverly Sackler
Faculty of Exact Sciences, Tel-Aviv University, 69976
Tel-Aviv, Israel
8 September 1994
Abstract
Using a long wavelength iteration scheme to solve Einsteins equations near the Big-
Bang singularity of a universe driven by a massive scalar eld, we nd how big initial
quasi-isotropic inhomogeneities can be before they can prevent ination to set in.
PACS numbers: 98.80.Cq, 04.50.+h
1
1. Introduction
The scale factor a(t) of a spatially at isotropic and homogeneous Robertson-Walker
(fRW) universe driven by a massive scalar eld evolves as follows: near the Big-Bang
(chosen to be t = 0) the scalar eld goes to as ln t and behaves like a perfect sti
uid (whose pressure equals the energy density, that is whose adiabatic index is 2), so
that a grows as t
1/3
[1]. Then a grows quasi exponentially during the inationary regime
during which slowly rolls down its potential well [2]. Finally at the end of ination when
oscillates in the bottom of the well a behaves as if the universe was driven by dust and
grows on average as t
2/3
[3].
A question is : how stable is this evolution against departures from spatial atness,
isotropy, and homogeneity. The eect of curvature was studied in [4]; since its role af-
ter the Big-Bang can become predominant, it can prevent ination from ever starting if
strong enough. Ination in homogeneous albeit anisotropic Bianchi models was thoroughly
analyzed, with similar conclusions : see e.g. [5] for a review. Finally, the role of inhomo-
geneities was studied analytically under simplifying assumptions in [6]. They were studied
numerically, in the case of planar symmetry in [7] and in the case of spherical symmetry
in [8]. In [8], setting the initial conditions in such a way that ination would occur only in
the central region if the universe evolved like a fRW space time, the conclusion was that
ination would indeed occur only if the central region was larger than a few times the local
Hubble radius.
Here we shall tackle the problem semi-analytically in the long wavelength approxima-
tion.
The long wavelength iteration scheme, the history of which goes back to [9], is a way
to build, out of seed spatial metrics, approximate solutions of Einsteins equations which
describe inhomogeneous but quasi-isotropic universes on scales larger than the local Hubble
radius (see [10] and references therein for a detailed description of the scheme). When
matter satises the strong energy condition (that is does not inate) this approximation is
well suited to describe the early universe since most scales are then larger than the Hubble
radius on average (this is the well-known horizon problem). These approximate solutions
however are not generic in the sense that they are built out of a seed spatial metric, that is
on three physically distinct arbitrary functions, instead of 6 (4 for the gravitational eld,
plus 2 for the scalar eld). The three missing functions can be seen as describing departures
from isotropy. These anisotropies cannot be neglected near the Big-Bang (see e.g. [9-10])
but they decay with time much faster than all other contributions to inhomogeneity and
hence will be ignored (see [11] for an analysis of the generic solution near the Big-Bang).
Within that scheme the growth or decay of inhomogeneities according to their equation
of state can be easily inferred, at least qualitatively (see [10, 12]): they decay when matter
violates the strong energy condition, that is inates, and grow otherwise, that is when the
eective adiabatic index
eff
of matter is > 2/3. Therefore when matter is a scalar eld
the inhomogeneities rst grow (
eff
= 2), then decay during the inationary period, to
grow again at the end of ination (
eff
= 1), at the condition that they do not grow so
large during the rst phase as to prevent ination to start.
The purpose of this paper is to give quantitative estimates on when ination may
2
be halted by the presence of initial quasi-isotropic inhomogeneities. This will be done by
integrating numerically the ordinary second order linear dierential equations that govern
their evolution in the long wavelength approximation scheme.
2. The equations
The long wavelength approximation scheme [10] consists in looking for solutions of
Einsteins equations for gravity coupled to a scalar eld whose 3-metric (in the t = Const
slices of a synchronous reference frame) can be expanded as a sum of spatial tensors of
increasing order in the gradients of a seed metric with time dependent coecients. The
line element is thus of the form:
ds
2
= dt
2
+
ij
(t, x
k
)dx
i
dx
j
,
ij
=
(1)

ij
+
(3)

ij
+
(5)

ij
+...
(1)

ij
= a
2
(t)h
ij
(x
k
) ,
(3)

ij
= a
2
[a
2
(t)Rh
ij
+b
2
(t)R
ij
] (1)
(5)

ij
=a
2
_
a
4
(t)R
2
h
ij
+b
4
(t)RR
ij
+c
4
(t)R
lm
R
lm
h
ij
+d
4
(t)R
im
R
m
j

+a
2
[e
4
(t)
m

m
Rh
ij
+f
4
(t)
i

j
R +g
4
(t)
m

m
R
ij
] ,
and the scalar eld of the form:
= (t) +
2
(t)R+
4
(t)R
2
+
4
(t)R
lm
R
lm
+
4
(t)
m

m
R +... (2)
This solution is built on a seed h
ij
(x
k
) of 6 functions of space which can be reduced to 3
by a suitable choice of spatial coordinates. As shown in [10], it is not a generic solution (it
should then depend on 6 physically distinct functions) but it is an attractor of a class of
generic solutions. At lowest order the metric reduces to
ij
= a
2
(t)h
ij
(x
k
) and = (t),
hence the name quasi-isotropic given to the solution.
i
is the covariant derivative with
respect to h
ij
, R
ij
its Ricci tensor, R R
ij
h
ij
the curvature scalar. The coecients
a(t), a
2
(t) etc are functions of time which are determined by Einsteins equations. We
shall denote by L the characteristic comoving length on which the spatial metric varies:

jk
L
1

jk
.
At zeroth order Einsteins equations reduce to the Friedmann equations for a spa-
tially at Robertson-Walker (fRW) universe and determine a(t) and (t). Introducing the
dimensionless variables:
T = mt , F = 2

3G , H =
3
m
1
a
da
dt
, S = am
1/3
(3)
where m is the mass of the scalar eld, they read:
dS
dT
=
HS
3
,
dH
dT
= (H
2
F
2
) ,
dF
dT
=
_
H
2
F
2
. (4)
The solutions of eqs (4) depend on three integration constants: the time

T(x
k
) of the
Big-Bang that we shall restrict to be

T = 0 (see [11] for an analysis of delayed Big Bang
3
solutions); the size

S(x
k
) of the scale factor at some given initial time: a constant which
can be absorbed without loss of generality in a redenition of the seed metric h
ij
; and
nally the initial value

F for the scalar eld. A priori

F depends on space but the rst
order Einstein equations (see [10] for details):
i
H = 4(dF/dT)
i
F imposes that it does
not. Its value, the same for all x
k
, tells us which curve of the (F, dF/dT) phase diagramme
the solution follows, and hence determines the total amount of ination. We shall therefore
integrate these equations up to the beginning of ination with, as initial conditions at time
T = :
S
in
=
1/3
, H
in
=
1

, F
in
=

F, (5)
being chosen such that (ln)
2
is numerically negligible compared to
2
. If matter were
a perfect uid with adiabatic index the scale factor S would grow as T
2/3
. In the
initial regime (5) the scalar eld hence behaves like a sti uid ( = 2) [1]. When

F
is large enough the solution then enters an inationary phase characterized by a slow
linear decrease of F and a quasi exponential growth of S. The comoving Hubble radius
L
H
= 3/SH hence rst increases as T
2/3
during the sti uid regime and then decreases
exponentially. The moment it reaches its maximum value can be taken as the beginning
of ination. We shall denote L
inf
H
and F
inf
the values of the Hubble radius and the scalar
eld at that moment.
At third order Einstein equations determine the time dependent coecients a
2
(t), b
2
(t)
and
2
(t) in (1-2). In terms of the variables dened in (3) and introducing:
B
2
= b
2
m
4/3
, A
2
= a
2
m
4/3
, F
2
= 4
_
G
3
m
4/3

2
(6)
they can be written as (see [10]):
du
dT
= 2S ,
dB
2
dT
=
u
S
3
(7)
dv
dT
=
u
4S
d
dT
_
_
S
dF
dT
_
2
_
,
dA
2
dT
= v
_
dF
dT
_
2
(8)
F
2
=
1
dF/dT
_
1
4
dB
2
dT
+
dA
2
dT
_
. (9)
It is easy to see that when the inationary regime has set in, B
2
tends to a constant
(that we shall call I
2
) as well as F
2
. As for A
2
it decreases linearly in time. Now, as
shown in [10], a gauge transformation modies the coecient of the scalar curvature R
in
(3)

ij
but leaves untouched the Ricci term. During ination it amounts to adding an
arbitrary linear function of time to A
2
. Therefore the linear decrease of A
2
can be gauged
away by a suitable choice of initial conditions for A
2
, v
2
, leaving B
2
as the only relevant
quantity to be studied. Physically the geometry evolves from one conguration, h
ij
, to
another, h
ij
+ (I
2
/m
4/3
)R
ij
, where I
2
is the imprint on the geometry left by the initial
4
inhomogeneity. We shall therefore integrate (7) with the initial conditions (5) at T =
together with dF/dT|
in
= 1/ and:
u
in
=
3
2

4/3
, B
2
|
in
=
9
8

4/3
. (10)
3. The solutions
Equations (4) with initial conditions (5) and equations (7) with initial conditions (10)
are integrated numerically. The results can be encapsulated in a plot of
2

_
|I
2
|/L
inf
H
as a function of F
inf
. See the curve k = 0 of Fig 1. We see that as F
inf
increases,
2
tends to a constant close to 1 (when a brutal extrapolation of the analytical behaviours
(5) and (10) would have given 1/2

2). Going back to the expressions (1-2) for the metric

ij
remain small compared to
(1)

ij
that is if I
2
/L
2
< 1
which is equivalent, because
2
tends to 1, to L > L
inf
H
. This condition says that the size
L of the inhomogeneity must be larger than the (comoving) Hubble radius at the onset of
ination, when the brutal analytic extrapolation gives that it can be smaller by a factor
2

2 = 2.8. Moreover, when the zeroth order barely inates, that is when F
inf
is small,

2
= > 1, so that a sucient condition for ination is L > L
inf
H
. We thus recover in
this semi-analytical approach the result of ref [8].
4. Convergence of the iteration scheme
A rigourous mathematical analysis of the convergence of the series (1-2) is certainly
beyond the scope of this paper. Using the results of the appendix of ref [10] we can however
compute the next order, that is the coecients a
4
, b
4
etc and see if including them spoils
the conclusion of the preceeding paragraph, drawn from the rst iteration.
The relevant coecients, which are not aected by a gauge transformation, are c
4
,
d
4
and g
4
. We shall concentrate on g
4
which satises the following linear second order
dierential equations:
dw
dT
= B
2
S ,
dG
4
dT
=
w
S
3
(11)
where we have introduced G
4
m
8/3
g
4
with the initial conditions:
w
in
=
27
64

8/3
, G
4
|
in
=
81
8
3

8/3
. (12)
(Knowing G
4
and B
2
, the coecient D
4
m
8/3
d
4
, is readily obtained : D
4
= 4G
4
+
1
2
B
2
2
;
see [10].)
The results of the integration of (11-12) are summarized in the diagramme
4

|G
4
|
1/4
/L
inf
H
as a function of F
inf
(see the curve k = 0 of Fig. 2). We see that
4
follows
the same pattern as
2
. As F
inf
increases it tends a constant value of the order of 0.63
5
when an extrapolation of the analytical behaviour (16) would have given 8
3/4
= 0.21. As
for the ratio |D
4
|
1/4
/L
inf
H
it tends to 0.60.
The inclusion of these fourth order terms therefore does not spoil the conclusion of the
previous paragraph and, since
4
<
2
, is an indication (although certainly not a proof!)
that the exact solution does not dier drastically from the second order solution.
5. Curvature versus gradient eects and an improved scheme in the case of
spherical symmetry
In the case when the inhomegeneity is imposed to be spherically symmetric, the seed
line element d
2
= h
ij
dx
i
dx
j
can be written in the form:
d
2
=
dr
2
1 kr
2
+r
2
d (13)
where d is line element on the sphere and where k(r) is an arbitrary function of r.
Expressing the Ricci tensor and its derivatives in function of k(r) we can write the
long wavelength metric (1) and scalar eld (2) under the form:
ds
2
= dt
2
+R
2
(C
2
dr
2
+r
2
d) (14)
with
R = R
app
cRW
_
1 +
_
a
2
+
1
4
b
2
_
k

r +...
_
, R
app
cRW
= a [1 + (3a
2
+b
2
) k +...] (15)
C = C
cRW
_
1 +
_
a
2
+
1
2
b
2
_
k

r +...
_
, C
cRW
=
1

1 kr
2
(16)
and
=
appr
cRW
+ 2k

r
2
+... ,
app
cRW
= + 6k
2
+... (17)
where a prime denotes dierentiation with respect to r and where the time dependent
coecients (a, ), (a
2
, b
2
,
2
) etc are the same as before and satisfy eqns (3-5), (6-10),
(11-12).
The rewriting of eq (1-2) under the form (14-17) shows clearly the two ways in which a
spherically symmetric inhomogeneity makes the solution depart from the at fRW solution.
The rst (trivial) eect is that of curvature: if k = 0 but all its derivative are taken to
be zero, the long wavelength solution (14-17) can be checked to be nothing but the Taylor
expansion in t of the exact curved Robertson-Walker (cRW) solution. In this case then
the exact metric and scalar eld are given by (14) with C = (1 kr
2
)
1/2
, and R = a and
satisfying the Friedmann equation:
dS
dT
=
HS
3
,
dH
dT
= (H
2
F
2
)
6K
S
2
,
dF
dT
=
_
H
2
F
2
+ 9K/S
2
. (18)
where T, F, H, S are dened by (3) and where K = km
2/3
.
6
Integrating (18) numerically with the same initial conditions as before (eq(5)), we
recover the results of ref [4], that is that if the curvature term K/S
2
is too large then
ination is halted. This can be seen in Fig. 3 where L
H
is plotted as a function of T for
dierent values of K and a given initial value

F for the scalar eld. We see that a negative
K favours ination whereas a positive value delays its setting in and, if large enough,
can even prevent it (dotted line).
The second eect is that of the gradients, that is the derivatives of k(r), which reects
the point to point correlation due to the variation of the curvature. It can be enhanced
by choosing a seed k(r) such that it is small everywhere but has a steep gradient k

r
around, say, r = R. From (16) and the result of section 3 that
2
1 we know that these
gradient eects will not prevent ination to set in if the function k(r) is everywhere such
that k

r < 2m
4/3
/(L
inf
H
)
2
.
One can also improve the long wavelength scheme by replacing in (14-17) the ap-
proximate cRW values by their exact values as given by (18) and taking for A
2
, B
2
, F
2
the solutions of eq (7-9) where S, H and F are taken to satisfy (18) instead of (4). The
results are summarized in Fig. 1 and 2 and show that the improved and the standard
schemes coalesce for large F
inf
that is for strongly inating solutions. This conrms what
the previous section already indicated, that is that the iteration scheme seems to converge
nicely.
6. Conclusions
An important question in inationary cosmology is: how generic is it? As already
shown by Goldwirth and Piran and conrmed here, ination by itself requires a certain
level of homogeneity: it can start if initial inhomogeneities are larger than the local Hubble
radius. While the numerical calculations of [8] can explore in a detailed way a specic
(spherically symmetric) case of strong initial inhomogeneity, the semi-analytical approach
presented here is limited to rather small perturbations but it gives a better global picture
on the factors that control the behaviour of the system. It gives only sucient conditions
for the onset of ination, not as strong as the necessary conditions obtained in [8] but they
are general and do not assume any spatial symmetry. We plan to extend the comparison
between these analytical and numerical approaches by solving an identical initial value
problem. This requires further investigation of the various coordinates systems used.
Acknowledgments
We thank Tsvi Piran and David Langlois for numerous discussions. This work was partly
supported by an Arc en Ciel, Keshnet programme.
7
References
[1] V.A. Belinski, I.M. Khalatnikov, JETP 36 (1973) 591
[2] see e.g. E.W.. Kolb, M.S. Turner, The Early Universe, Addison-Wesley, 1990,
A. Linde Particle Physics and Inationary Cosmology, Harwood, 1990
[3] V.A. Belinski, I.M. Khalatnikov, JETP 66 (1987) 441
[4] V.A. Belinski, H. Ishihara, I.M. Khalatnikov, H. Sato, Progress Theor. Phys. 79
(1988) 676
[5] D.S. Goldwirth, T. Piran, Physics Report 214 (1992) 223
[6] J. Kung and R. Brandenberger, Phys. rev. D 42 (1990) 1008; E. Calzetta and M.
Sakellariadou Phys. Rev D 45 (1992) 2802
[7] J.A. Shinkai, K. Maeda, Phys. Rev D, in press
[8] D.S. Goldwirth, T. Piran, Phys. Rev D 40 ( 1989) 3263; D.S. Goldwirth, Phys.
Rev. D 43 (1991) 3204
[9] V.A. Belinski, E.M. Lifchitz, I.M. Khalatnikov, JETP 35 (1972) 838
[10] G.L. Comer, N. Deruelle, D. Langlois, J. Parry, Phys. Rev. D 49 (1994) 2759
[11] N.Deruelle , D.Langlois, to be submitted
[12] K. Tomita, N. Deruelle, accepted for publication by Phys. Rev. D
8
Figure Captions
Figure 1 Plot of
2
as a function of F
inf
. The full line corresponds to k = 0, the
dashed line above to negative k, the dashed dotted lines below to positive k.
Figure 2 Plot of
4
as a function of F
inf
. The conventions are the same as in Fig. 1.
Figure 3 Plot of the comoving Hubble radius as a function of time for various values of
the curvature term k and a given initial value for the scalar eld. The full line corresponds
to k = 0, the dashed lines below to negative k, the dashed dotted ones to positive k.
Finally the dotted line is an example of a curvature strong enough to prevent ination
from setting in.
9
This figure "fig1-1.png" is available in "png"format from:
http://arXiv.org/ps/gr-qc/9409056v1
This figure "fig1-2.png" is available in "png"format from:
http://arXiv.org/ps/gr-qc/9409056v1
This figure "fig1-3.png" is available in "png"format from:
http://arXiv.org/ps/gr-qc/9409056v1
20 30 40 50 60 70
0.94
0.96
0.98
1
1.02
20 30 40 50 60 70
0.6
0.61
0.62
0.63
0.64
0.65
0 0.1 0.2 0.3 0.4 0.5
0
0.1
0.2
0.3
0.4
0.5
T