KEK-TH-2490

KYUSHU-HET-253

Full higher-dimensional analysis of moduli oscillation

and radiation in expanding universe
arXiv:2212.14314v1 [hep-th] 29 Dec 2022

Hajime Otsukaa∗ and Yutaka Sakamurab,c†

a
Department of Physics, Kyushu University,
744 Motooka, Nishi-ku, Fukuoka, 819-0395, Japan
b
KEK Theory Center, Institute of Particle and Nuclear Studies, KEK,
1-1 Oho, Tsukuba, Ibaraki 305-0801, Japan
c
Graduate University for Advanced Studies (Sokendai),
1-1 Oho, Tsukuba, Ibaraki 305-0801, Japan.

Abstract
We investigate effects of the radiation and the moduli oscillation around the stabilized
values on the evolution of a 6-dimensional spacetime compactified on S 2 . In order to see the
transition from the 5-dimensional space to the 3-dimensional one, we develop a procedure
to pursue the spacetime evolution with appropriate approximations, which is valid until
the spacetime behaves like 4-dimensional. In the case that the moduli stabilization process
cannot be described in the context of the 4-dimensional effective theory, it takes quite a
long time for the moduli oscillation to dominate the total energy density, in contrast to the
conventional result obtained by the 4-dimensional effective theory approach. We also found
that even if the moduli are set at the stabilized values, they start to oscillate due to the
pressure in the extra space S 2 in some cases.

∗
E-mail address: [email protected]
†
E-mail address: [email protected]
1 Introduction
Moduli fields naturally appear in the low-energy effective theory of higher-dimensional grav-
ity. Since couplings of matter fields and the spacetime evolution will depend on a dynamics of
moduli fields, they will give impacts on several aspects of particle physics and cosmology.
In the conventional approach, the dynamics of moduli fields will be treated in the framework
of four-dimensional (4D) effective field theories (EFT) under the assumption that Kaluza-Klein
(KK) modes are decoupled from low-energy dynamics. However, such an assumption will be
violated when the typical moduli mass m is the same order of or larger than the mass of the first
(1) (1)
KK excitation mode mKK , i.e., m & mKK . This situation is realized in the large volume regime
of the extra-dimensional space. Although such a situation is not disfavored for any specific
reasons, most of the works about the moduli dynamics have been studied in the context of 4D
EFT.
As a simple setup to analyze a higher-dimensional gravity theory, we consider the so-called
Salam-Sezgin model [1], which is based on a gauged six-dimensional (6D) supergravity [2, 3]
compactified on a sphere with a U(1) magnetic flux. The original work for this model discussed
a static background. Its time-dependent extension was discussed in Refs. [4] in the radiation-
dominated universe, which is described by the hypermultiplets and/or the vector multiplets 1 .
Since there is a flat direction in the moduli space in the original Salam-Sezgin model, we
introduce a dilaton potential so that all the moduli are completely fixed. Then, the moduli
oscillate around their stabilized values in the evolution of the universe. In the 4D EFT approach,
it is well-known that the energy density of the moduli oscillation rapidly dominates over that of
the radiation, and the expanding space behaves like the 4D matter-dominated universe [6]- [10].
However, this is not the case when the moduli stabilization procedure cannot be described in the
4D EFT. In our previous work [11], we investigated the evolution of the background spacetime
during the moduli stabilization process that is assumed to occur in the radiation dominated
era. It was numerically found that when the modulus mass is larger than the KK mass, the
radiation contribution to the total energy density remains non-negligible for a long time in
contrast to the conventional 4D EFT analysis. However, this numerical analysis is available for
only a limited range of the time and the parameters, and it will be difficult to see the transition
from 6D to 4D explicitly. The purpose of this paper is to analytically investigate effects of
the radiation and the moduli oscillation on the spacetime evolution, especially focusing on a
parameter region in which one cannot use the 4D EFT analysis. Our findings are summarized
as follows:
• The radiation remains non-negligible for a long time when the moduli stabilization pro-
(1)
cedure cannot be described in the framework of 4D EFT, namely m & mKK .
• In such a case, for lower initial temperatures, the universe never experiences the moduli-
dominated era if the moduli decay before the dominance of the moduli oscillation.
• Even if the moduli are set at the stabilized values, they start to oscillate due to the
pressure in the extra space S 2 in some parameter spaces.
For our purpose, we develop a procedure to compute various quantities at late times that
enables us to pursue the transition from 6D to 4D.
1
Note that the cancellation of 6D gravitational anomalies requires the existence of hypermultiplets and/or
vector multiplets [5].

1
 The paper is organized as follows. In the next section, we provide a brief review of the
model used in our previous work [11]. In Sec. 3, we explain how to compute various quantities
at later times. In Sec. 4, we discuss the conditions that the radiation dominates the total energy
density. Sec. 5 is devoted to the summary. In the appendices, we provide brief derivations of
some formulae used in the text, and show the conservation law of the energy-momentum tensor.

2 Setup
In this section, we briefly review the model we considered in our previous work [11]. The
whole spacetime is 6D, and a 2D subspace is compactified on a sphere S 2 . The indices M, N =
0, 1, 2, · · · , 5 denote the 6D coordinate ones, µ, ν = 0, 1, 2, 3 denote the 4D ones for the non-
compact space, and m, n = 4, 5 are the 2D ones for the compact space. As the coordinates
on S 2 , we choose the spherical ones (x4 , x5 ) = (θ, φ), where θ and φ are the polar and the
azimuthal angles, respectively.

2.1 Model inspired by 6D supergravity on a sphere

The model is given by 2
2 σ
Z  
6
p 1 (6) 1 M g e M N
S = d x −g (6) − R − ∂ σ∂M σ − F FM N − V (σ) , (2.1)
2 2 4

where R(6) denotes the 6D Ricci scalar, σ is a real scalar (dilaton), FM N ≡ ∂M AN − ∂N AM is

the field strength of the U(1) gauge field AM , and g is the gauge coupling constant. The scalar
potential V (σ) is given by
m2
V (σ) = 2e−σ + (σ − σ∗ )2 , (2.2)
2
where m and σ∗ are positive constants.
This is basically the bosonic part of the gauged 6D N = (1, 0) supergravity [2,3], except for
the following two points. First, we drop the self-dual antisymmetric tensor field BM N because
it is irrelevant to the following discussions. Second, we add the second term in (2.2) to the
scalar potential in order to stabilize the moduli completely.
The equations of motion are
(6) 1
RM N − gM N R(6) − TM matter
N = 0,
2
1 p  g 2 eσ
M
p ∂M (6)
−g ∂ σ − F M N FM N − ∂σ V (σ) = 0,
−g (6) 4
p 
(6) σ MN
∂M −g e F = 0, (2.3)
matter
where the energy-momentum tensor TM N is

matter 1 L 2 σ L g 2 eσ
TM N = −∂M σ∂N σ + gM N ∂ σ∂L σ − g e FM L FN + gM N F P Q FP Q + gM N V (σ). (2.4)
2 4
2
We work in the unit of the 6D Planck mass.

2
 In the absence of the second term in (2.2), this model has the following static background [1]:
gµν = ηµν = diag (−1, 1, 1, 1),
g44 = b2 , g45 = g54 = 0, g55 = b2 sin2 θ,
Fµν = Fµm = 0,
sin θ
F45 = −F54 = , F44 = F55 = 0,
2g
σ = ln 4b2 ,


(2.5)
where ηµν is the 4D Minkowski metric, and b is a positive constant. In this case, the constant b
is a free parameter, and the size of the compact space S 2 remains to be unfixed.
In the presence of the second term in (2.2), the background (2.5) remains to be a solution,
but now σ is fixed to the constant σ∗ . Hence the constant b is also fixed as
eσ∗ /2
b = b∗ ≡ . (2.6)
2
In addition to the above field content, we also introduce the radiation contribution. In the
context of the 6D N = 1 supergravity, the number of hypermultiplets nH and that of vector
multiplets nV are constrained by the anomaly cancellation condition nH − nV = 244 [3, 5, 12].3
This indicates that a large number of hypermultiplets must exist in a consistent theory. We
assume that scalars in such hypermultiplets do not have nontrivial background values, but they
contribute to the radiation that fills in the whole 5D space. Then, the energy-momentum tensor
appearing in the Einstein equation must include the radiation contribution:
 rad 
ρ

N  −prad
3 13

T rad M =  , (2.7)
 −prad2

2
−prad
2 sin θ

where ρrad , prad

3 and prad
2 are the radiation energy density, the pressures in the non-compact 3D
space and in the compact 2D space, respectively, whose explicit forms are listed in Appendix A.
In the presence of the radiation, the static background (2.5) and (2.6) are no longer a solution
of the equations of motion, and the universe continues to expand. Thus we make the following
ansatz for the background.
 
−1
 a2 (t)13 
gM N =  ,
 b2 (t) 
2 2
b (t) sin θ
Fµν = Fµm = 0,
sin θ
F45 = −F54 = , F44 = F55 = 0,
2g
σ = σ(t), (2.8)
where a(t) and b(t) are the scale factors for the non-compact 3D space and the compact 2D
space, respectively.
3
The number of tensor multiplet is assumed to be one, otherwise the theory cannot be described by the
Lagrangian.

3
2.2 Evolution equations
Under the background ansatz (2.8), the equations of motion in (2.3) become

3ȧ2 ḃ2 6ȧḃ 1 1 2 eσ

+ + + − σ̇ − − V (σ) − ρrad = 0,
a2 b2 ab b2 2 8b4
2ä ȧ2 2b̈ ḃ2 4ȧḃ 1 1 eσ
+ 2+ + 2+ + 2 + σ̇ 2 − 4 − V (σ) + prad 3 = 0,
a a b b ab b 2 8b
3ä 3ȧ2 b̈ 3ȧḃ 1 2 eσ
+ 2 + + + σ̇ + 4 − V (σ) + prad 2 = 0,
a a b ab 2 ! 8b
3ȧ 2ḃ eσ
σ̈ + + σ̇ + 4 + ∂σ V (σ) = 0, (2.9)
a b 8b

where the dot denotes the time derivative. The first equation is the (t, t)-component of the
Einstein equation. Since this does not contain the second order t-derivatives, it is regarded as
a constraint on the initial conditions of the time evolution. The second and the third equations
come from the diagonal components for the 3D non-compact space and the 2D compact space,
respectively. The other components of the Einstein equation vanish. The last equation is the
dilaton field equation. In addition to these, we can obtain the evolution equation for the inverse
temperature β from the conservation law, as shown in Appendix B.
In the following, we assume that the chemical potential µ is negligible, i.e., βµ  1. Then,
if we redefine the scale factors as

A ≡ ln a, B ≡ ln b, (2.10)

the above equations are rewritten as

σ σ
1 2 1 − σ e 2 −2B 3e 2 −2B
  
9 2 1 2 1 − σ
Ä = − Ȧ + Ḃ − ȦḂ − σ̇ + e 2− e 2+
4 4 2 8 2 4 4
2 rad rad
m p − 2p2
+ (σ − σ∗ )2 + 3 ,
8 4
σ σ
1 2 1 − σ e 2 −2B 5e 2 −2B
  
3 2 7 2 3 − σ
B̈ = Ȧ − Ḃ − ȦḂ − σ̇ + e 2− e 2−
4 4 2 8 2 4 4
2 rad rad
m 3p − 2p2
+ (σ − σ∗ )2 − 3 ,
8 4
σ σ
e 2 −2B e 2 −2B
    
− σ2 − σ2
σ̈ = − 3Ȧ + 2Ḃ σ̇ + 2 e − e + − m2 (σ − σ∗ ) , (2.11)
4 4
4
with the constraint:
σ 2
e 2 −2B m2

1 2 − σ2
2 2
3Ȧ + Ḃ + 6ȦḂ − σ̇ = 2 e − + (σ − σ∗ )2 + ρrad
2 4 2
≡ ρ̂tot . (2.12)
4
This ρ̂tot is related to the 6D total energy density ρtot defined in (B.2) as ρ̂tot = ρtot − 12 σ̇ 2 − e−2B , where
the last term corresponds to the curvature of S 2 .

4
Using (2.12) and (A.11), the second equation in (2.11) can be rewritten as
σ σ
e 2 −2B 3e 2 −2B m2
    
− σ2 − σ2
B̈ = − 3Ȧ + 2Ḃ Ḃ + e − e − + (σ − σ∗ )2 + prad
2 . (2.13)
4 4 4

The energy density and the pressures are expressed as (see Appendix A)

gdof e−2B
ρrad = {±6Li4 (±1) + 3Q1 + Q2 } ,
8π 3 β 4
gdof e−2B
prad
3 = {±2Li4 (±1) + Q1 } ,
8π 3 β 4
gdof e−2B
prad
2 = Q2 , (2.14)
16π 3 β 4
where gdof is the degrees of freedom for 6D relativistic particles, β is the inverse temperature,
the functions Qi (x) (i = 1, 2, 3) are defined in (A.6), (A.8) and (B.6), and their arguments
are β/b = βe−B . The upper (lower) signs represent the case that the radiation consists of the
bosons (fermions). The evolution equation for β is obtained from (B.5) as

β̇ 3Ȧ {±8Li4 (±1) + 4Q1 + Q2 } + Ḃ (2Q2 + Q3 )

= . (2.15)
β ±24Li4 (±1) + 12Q1 + 5Q2 + Q3

The equations (2.11) and (2.15) with the constraint (2.12) are the evolution equations for
the expanding 5D space. Note that
7
Li4 (1) = ζ(4), Li4 (−1) = − ζ(4). (2.16)
8
In the following, we consider a case that the radiation consists of only the bosonic particles to
simplify the discussion. Hence the upper signs in (2.14) and (2.15) are applied.
For numerical computation, we choose the initial conditions at t = 0 as

a(0) = 1, b(0) = bI , σ(0) = σI , β(0) = βI ,

r
ρ̂tot (0)
ȧ(0) = , ḃ(0) = σ̇(0) = 0, (2.17)
3
where bI , σI and βI are positive constants. The value of ȧ(0) is determined by the con-
straint (2.12).

3 Transition from 5D to 3D spaces

3.1 Late-time behavior
We focus on a situation that the standard 4D cosmology is realized at late times. Namely,
the S 2 size modulus b and the dilaton σ are expected to be stabilized. In such a case, Ḃ = ḃ/b

5
and σ̇ become negligible, and (2.9) is reduced to
3ȧ2 1 eσ
2
+ 2
− 4
− V (σ) − ρrad ' 0,
a b 8b
2ä ȧ2 1 eσ
+ 2 + 2 − 4 − V (σ) + prad 3 ' 0,
a a b 8b
3ä 3ȧ2 eσ
+ 2 + 4 − V (σ) + prad 2 ' 0,
a a 8b
σ
e
+ ∂σ V (σ) ' 0. (3.1)
8b4
Note that terms 1/b2 and −eσ /(8b4 ) come from the curvature of S 2 and the background flux
of the U(1) gauge field, respectively. If they are cancelled with the potential term V (σ), we
obtain the standard 4D Friedmann equations. This condition is written as
1 eσ
Vmdl (b, σ) ≡ − 2 + 4 + V (σ)
b 8b
2
eσ/2 m2

− σ2
=2 e − 2 + (σ − σ∗ )2 = 0. (3.2)
4b 2
Namely, this is equivalent to the minimization condition of the moduli potential. From this
condition, the stabilized moduli values are given by

eσ∗ /2
σ = σ∗ , b = b∗ = . (3.3)
2
Then the Kaluza-Klein masses for a 6D massless field are given by
p
(l) l(l + 1)
mKK = , (3.4)
b∗
(1) √
where l = 0, 1, 2, · · · . In particular, mKK = 2/b∗ is regarded as the cutoff energy scale of the
4D EFT.
After the moduli are stabilized, the first two equations in (3.1) are reduced to the 4D
Friedmann equations.5
ȧ2 ρrad 1 ȧ2
 
ä rad 1 rad rad


' , ' − + p 3 ' − ρ + 3p 3 . (3.6)
a2 3 a 2 a2 6
The third equation becomes
3ä 3ȧ2
+ 2 + prad
2 ' 0. (3.7)
a a
5
Precisely, we should note that the energy density and the pressure in the 4D spacetime are given by
ρ4D ≡ ρrad V2 and P4D√≡ prad 2 2
3 V2 , respectively, where V2 ≡ 4πb∗ is the stabilized S volume. Besides, the 4D
Planck mass is MPl ≡ V2 since we have chosen the 6D Planck unit. Thus (3.6) should be expressed as

ȧ2 ρ4D ä 1
' 2 , '− 2 (ρ4D + 3P4D ) , (3.5)
a2 3MPl a 6MPl
as the 4D Friedmann equations.

6
Using (3.6), the LHS is rewritten as

3ä 3ȧ2 1 rad

+ 2 + prad ρ + 3prad + ρrad + prad


2 ' − 3 2
a a 2
1 rad
ρ − 3prad + prad rad


= 3 2 = 2p2 . (3.8)
2
At the last step, we have used the relation (A.11). From (2.14) and Fig. 10 in Appendix A, we
find that prad
2 ' 0 at late times. Thus, (3.7) holds at late times. In this case, the relation (A.11)
becomes ρ ' 3prad
rad
3 , which indicates that the 4D universe is radiation-dominated. The last
equation in (3.1) holds trivially with the values in (3.3).
Since the functions Qi (x) (i = 1, 2, 3) are damped to zero for large x (see Fig. 10), the
evolution equation for the inverse temperature (2.15) becomes

β̇ ȧ
' Ȧ = , (3.9)
β a
at late times. This indicates that β ∝ a, which agrees with the relation in the 4D radiation-
dominated era.
In the following, we focus on the spacetime evolution during the moduli stabilization process
before it settles into the above 4D evolution.

3.2 Relation between a and β

As we have seen, our model realizes the 4D radiation-dominated universe at late times if
the moduli are stabilized to the values in (3.3). However, as shown in our previous work [11],
the behaviors of the 3D scale factor a and the (inverse) temperature β at early times can be
different from those of the 4D universe.
We can see that Ȧ  |Ḃ| during the moduli stabilization process. Thus (2.15) is approxi-
mated as
 
β̇ β ȧ
= vβ , (3.10)
β b∗ a

where
24ζ(4) + 12Q1 (x) + 3Q2 (x)
vβ (x) = . (3.11)
24ζ(4) + 12Q1 (x) + 5Q2 (x) + Q3 (x)

The profile of vβ (x) is shown in Fig. 1 (blue dashed line). As the plot shows, vβ (x) ' 1
for x ≥ 10. For smaller values of x, vβ (x) has a nontrivial profile. We approximate it by a
piecewise-linear function for 0 < x ≤ 10. Namely, we divide this interval into (J + 1) small
intervals xj < x ≤ xj+1 (j = 0, 1, 2, · · · , J), where xj ≡ j∆ (∆ ≡ 10/J), and define the
approximated function:
(
(j) (j)
ap c1 x + c2 (xj < x ≤ xj+1 ≤ xJ = 10)
vβ (x) ≡ , (3.12)
1 (x > 10)

7
 1.0

0.9

vβ 0.8

0.7

0.6
0 2 4 6 8 10 12
x

Figure 1. The profiles of vβ (x) (blue dashed) and vβap (x) (red solid) in the case of ∆ = 0.5.

(j) (j)
where the constants c1 and c2 are determined so that vβap (xj ) = vβ (xj ) and vβap (xj+1 ) =
vβ (xj+1 ), i.e.,

(j) vβ (xj+1 ) − vβ (xj ) (j) xj+1 vβ (xj ) − xj vβ (xj+1 )

c1 ≡ , c2 ≡ . (3.13)
∆ ∆
In the following, we choose ∆ = 0.5 (or J = 20). As shown in Fig. 1, vβap (x) well approximates
vβ (x) in this case. In Sec. 3.4, we will set a reference time tref so that the evolution equations
for B̃ ≡ B − B∗ and σ̃ ≡ σ − σ∗ can be approximated as homogeneous second-order linear
differential equations for t ≥ tref . For this reference time, we define the integer k so that
xk < β(tref )/b∗ ≤ xk+1 . Then, by solving (3.10), the 3D scale factor a is expressed as

aref Kk (xref )
(xref ≤ x ≤ xk+1 )



 Kk (x)


a(x) = aref Kk (xref )Kk+1 (xk+1 ) · · · Kj (xj ) 1 (xj < x ≤ xj+1 ≤ xJ = 10) , (3.14)

 K j (x)
 a x
 ref Kk (xref )Kk+1 (xk+1 ) · · · KJ−1 (xJ−1 ) (x > 10)


10
where x ≡ β/b∗ , aref ≡ a(tref ), xref ≡ β(tref )/b∗ , and the function Kj (x) is defined by (C.9).
The detailed derivation is listed in Appendix C.
In the case that xref ≤ x1 = ∆, i.e., k = 0, the expression of a(β) becomes simple. Since

(0) (0) 3
0 < c1  1, c2 ' , (3.15)
5
K0 (x) is approximated as
 x 5/3
1
K0 (x) ' . (3.16)
x
From (C.10), we have
 5/3
aref K0 (xref ) x
a(x) = = aref . (3.17)
K0 (x) xref

8
 800

600
vρ
400

200

0
0 2 4 6 8 10 12
x

Figure 2. The profiles of vρ (x) (blue dashed) and vρap (x) (red solid) in the case of ∆ = 0.5.

Hence, the relation between a and β in this case is obtained as

 3/5
a
β(a) ' βref . (3.18)
aref
for aref ≤ a ≤ a1 .

3.3 Radiation energy density

During the moduli-stabilization process, the modulus b takes values around b∗ . Thus, the
radiation energy density (2.14) is expressed as
 
rad gdof β
ρ ' 3 6 vρ , (3.19)
8π β b∗
where

vρ (x) ≡ x2 {6ζ(4) + 3Q1 (x) + Q2 (x)} . (3.20)

Note that the contributions of Q1 (x) and Q2 (x) are exponentially suppressed and negligible for
x > 10. Thus, just like the treatment for vβ (x) in (3.10), we approximate vρ (x) by
(
(j) (j)
d1 x + d2 (xj < x ≤ xj+1 ≤ 10)
vρap (x) ≡ 2
, (3.21)
6ζ(4)x (x > 10)

where
(j) vρ (xj+1 ) − vρ (xj ) (j) xj+1 vρ (xj ) − xj vρ (xj+1 )
d1 ≡ , d2 ≡ . (3.22)
∆ ∆
As shown in Fig. 2, vρap (x) well approximates vρ (x).
In particular, when β < b∗ ∆, (3.18) can be used and
 
β
vρ ' 80, (3.23)
b∗

9
since
(0) (0)
d1 = 1.27  d2 ' 80. (3.24)

Hence, in this case, ρrad can be expressed as the following simple function of a:
 − 185
rad 10gdof 10gdof a
ρ ' 3 6 ' 3 6 . (3.25)
π β π βref aref

At the second step, we have used the relation (3.18). This can also be derived directly from
(2.14). In fact, when β < b∗ ∆, the functions Qi (i = 1, 2, 3) are approximated as

16b2∗ 32b2∗ 128b2∗

     
β β β
Q1 ' 2 , Q2 ' 2 , Q3 ' , (3.26)
b∗ β b∗ β b∗ β2

which lead to
− 185
48b2∗ 32b2∗
  
rad gdof 10gdof 10gdof a
ρ ' 3 2 4 6ζ(4) + 2 + 2 ' 3 6 ' 3 6 ,
8π b∗ β β β π β π βref aref
− 185
16b2∗
  
gdof 2gdof 2gdof a
prad
3 ' 3 2 4 2ζ(4) + 2 ' 3 6 ' 3 6 ,
8π b∗ β β π β π βref aref
− 185
32b2∗

gdof 2gdof 2gdof a
prad
2 ' · = 3 6 ' 3 6 . (3.27)
16π 3 b2∗ β 4 β 2 π β π βref aref

Note that the pressure becomes isotropic prad rad

3 ' p2 in this case because the radiation does not
feel the size of the compact space S 2 . In fact, the above pressures are (almost) independent of
b∗ .

3.4 Moduli oscillation

Here we discuss the moduli oscillation around their stabilized values in (3.3). We divide B
and σ as

B = B∗ + B̃, σ = σ∗ + σ̃, (3.28)

where
σ∗
B∗ ≡ ln b∗ = − ln 2. (3.29)
2
Then the evolution equations in (2.11) and (2.13) are written as

9 1  prad − 2prad
Ä = − Ȧ2 − ȦB̃˙ + 2e−σ∗ −σ̃ + 2B̃ + 3

2
+ ··· ,
4 2 4
¨ = −3ȦB̃˙ − 2e−σ∗ −σ̃ + 2B̃ + prad + · · · ,
 
B̃ 2
 
¨ = −3Ȧσ̃˙ + 4e−σ∗ −σ̃ + 2B̃ − m2 σ̃ + · · · ,
σ̃ (3.30)

10
and the constraint (2.12) becomes

1 2 m2
3Ȧ2 + B̃˙ 2 + 6ȦB̃˙ − σ̃˙ 2 = 2e−σ∗ σ̃ − 2B̃ +

σ̃ 2 + ρrad + · · · , (3.31)
2 2
where the ellipses denote higher-order terms in B̃ or σ̃.
We introduce the reference time tref so that the last term in the second equation of (3.30)
becomes negligible for t ≥ tref . In the following, we choose it as tref = 10000 (in the 6D Planck
unit). Then, from the first equation in (3.30) and (3.31), we find that

Ä = O(B̃, σ̃) (3.32)

Thus, the second and the third equations in (3.30) can be rewritten as
!
¨
2B̂ = − 4e−σ∗ −4e−σ∗
  
2B̂
−σ∗ −σ∗ 2 + ··· , (3.33)
¨
σ̂ −4e 4e + m σ̂

where
3 3
B̂ ≡ e 2 A B̃, σ̂ ≡ e 2 A σ̃. (3.34)

Diagonalizing this, we have

    
ϕ̈1 λ1 ϕ1
=− + ··· , (3.35)
ϕ̈2 λ2 ϕ2

where
√
 
1  −σ∗ 1
 q
2 −2σ 4 (1)2 2 (1)4
λ1 ≡ 8e + m − 64e ∗ +m = mKK + m − mKK + m4 ,
2 2
√
 
1  −σ∗ 1
 q
2 −2σ 4 (1)2 2 (1)4 4
λ2 ≡ 8e + m + 64e ∗ +m = mKK + m + mKK + m , (3.36)
2 2
and

ϕ1 ≡ cos θ · 2B̂ + sin θ · σ̂,

ϕ2 ≡ − sin θ · 2B̂ + cos θ · σ̂, (3.37)

with
(1)2
−1 8e−σ∗ m
θ ≡ tan √ = tan−1 q KK . (3.38)
m2 + 64e−2σ∗ +m4 (1)4
m2 + mKK + m4

Solving (3.35), we have

χ10 p  p 
ϕ1 (t) = √ sin λ1 (t − tref ) + ϕ10 cos λ1 (t − tref ) ,
λ1
χ20 p  p 
ϕ2 (t) = √ sin λ2 (t − tref ) + ϕ20 cos λ2 (t − tref ) , (3.39)
λ2

11
where
3
n o
ϕ10 ≡ e 2 A(tref ) 2 cos θB̃(tref ) + sin θσ̃(tref ) ,
3
n o
A(tref )
ϕ20 ≡ e 2 −2 sin θB̃(tref ) + cos θσ̃(tref ) ,
o 3
˙
3
n
χ10 ≡ e 2
A(tref ) ˙
2 cos θB̃(tref ) + sin θσ̃(tref ) + Ȧ(tref )ϕ10 ,
2
o 3
˙
3
n
A(t )
χ20 ≡ e 2 ref −2 sin θB̃(tref ) + cos θσ̃(t ˙ ref ) + Ȧ(tref )ϕ20 . (3.40)
2
In terms of ϕ1 and ϕ2 , (the displacement of) the moduli B̃ and σ̃ are expressed as
3
e− 2 A(t)
B̃(t) = {cos θϕ1 (t) − sin θϕ2 (t)} ,
2
3
σ̃(t) = e− 2 A(t) {sin θϕ1 (t) + cos θϕ2 (t)} . (3.41)

Since A(t)  |B̃(t)|, we can replace A(t) in the exponents with Ā(t) ≡ A(t) + B̃(t). In the next
subsection, we will show that Ā(t) has a nice property to calculate its approximate expression.
Therefore, we express B̃ and σ̃ as
3
e− 2 Ā(t)
B̃(t) = {cos θϕ1 (t) − sin θϕ2 (t)} ,
2
3
σ̃(t) = e− 2 Ā(t) {sin θϕ1 (t) + cos θϕ2 (t)} . (3.42)

Fig. 3 shows the full numerical solutions and the approximate solutions in (3.42) for B̃(t) and
σ̃(t). Here Ā(t) is computed by the method explained in Sec. 3.6. The parameters are chosen
as
 
(1)
m = 0.01, σ∗ = 14 ⇔ mKK = 0.00258 ,
bI = b∗ , σI = σ∗ , βI = 20. (3.43)

We can see that the latters well agree with the formers for t ≥ tref .

3.5 Smoothing the oscillation of the 3D scale factor

Now we derive an approximate expression for the 3D scale factor a(t) (or A(t)). For this
purpose, it is convenient to define

Ā(t) ≡ A(t) + B̃(t). (3.44)

Then, its evolution is determined by

σ 2
e 2 −2B m2

˙ 1 2 − σ2
2 2
3Ā = 2Ḃ + σ̇ + 2 e − + (σ − σ∗ )2 + ρrad
2 4 2
1
= 2Ḃ 2 + σ̇ 2 + ρ̂tot ≡ ρ̃tot , (3.45)
2

12
 0.02 0.0015
0.0010
0.01
0.0005
 0.00 
B σ 0.0000

-0.01 -0.0005
-0.0010
-0.02
-0.0015
0 10 000 20 000 30 000 40 000 0 10 000 20 000 30 000 40 000
t t

Figure 3. The profiles of B̃(t) (left) and σ̃(t) (right) in the case of tref = 10000. The blue solid lines are
the full numerical solutions, and the red dashed lines are the approximate solutions in (3.42), where
Ā(t) is computed by the method in Sec. 3.6. The parameters are chosen as (3.43).

which is obtained from the constraint (3.31). During the moduli stabilization process, the RHS
is approximated as

1 ˙2 2 m2
˙

˙ 2 2
3Ā ' 2B̃ + σ̃ + 2e −σ∗
σ̃ − 2B̃ + σ̃ 2 + ρrad + · · ·
2 2
−3Ā
 2
e 3 ˙
' cos θϕ̇1 − sin θϕ̇2 − Ā (cos θϕ1 − sin θϕ2 )
2 2
−3Ā
 2
e 3 ˙
+ sin θϕ̇1 + cos θϕ̇2 − Ā (sin θϕ1 + cos θϕ2 )
2 2
+ 2e−3Ā−σ∗ {(sin θ − cos θ) ϕ1 + (cos θ + sin θ) ϕ2 }2
m2 −3Ā
+ e (sin θϕ1 + cos θϕ2 )2 + ρrad + · · · , (3.46)
2
where the ellipses denote higher order terms in B̃ or σ̃ (or ϕ1,2 ). At the second step, we have
used (3.42). The solutions in (3.39) are rewritten as
s
χ210 p 
ϕ1 = + ϕ210 sin λ1 (t − tref ) + δ1 ,
λ1
s
χ220 p 
ϕ2 = + ϕ220 sin λ2 (t − tref ) + δ2 , (3.47)
λ2

where
   
−1 ϕ10 p −1 ϕ20 p
δ1 ≡ tan λ10 , δ2 ≡ tan λ20 . (3.48)
χ10 χ20

Here note

Ā˙ 2  λ1 , λ2  1, (3.49)

13
 0.00010
1.0
0.00008
0.8

A 0.00006 
 (t -t0 )A 0.6
A 0.00004 
(t -t0 )A 0.4

0.00002 0.2

0.00000 0.0
0 20 000 40 000 60 000 80 000 100 000 0 20 000 40 000 60 000 80 000 100 000
t t

˙
Figure 4. The profile of Ȧ(t) (blue solid) and Ā(t) = Ȧ(t) + Ḃ(t) (red dashed) in the left plot. Those
˙
of (t − t0 )Ȧ(t) (blue solid) and (t − t0 )Ā(t) (red dashed) in the right plot. The constant t0 and the
reference time are chosen as t0 = −1300 and tref = 10000, respectively. The parameters are chosen as
(3.43).

for t ≥ tref . Then, (3.46) is simplified as

e−3Ā 2

˙
3Ā ' ϕ̇1 + ϕ̇22 + 4e−σ∗ (sin θ − cos θ)2 + m2 sin2 θ ϕ21

2
+ 8e−σ∗ sin2 θ − cos2 θ + 2 sin θ cos θ ϕ1 ϕ2



 −σ∗ 2
+ 4e 2 2 2
(cos θ + sin θ) + m cos θ ϕ2 + ρrad + · · ·

e−3Ā 2
ϕ̇1 + ϕ̇22 + λ1 ϕ21 + λ2 ϕ22 + ρrad + · · ·


=
2
= C1 e−3Ā + ρrad + · · · , (3.50)

where we have used that

4e−σ∗ (sin θ − cos θ)2 + m2 sin2 θ = λ1 ,

8e−σ∗ sin2 θ − cos2 θ + 2 sin θ cos θ = 0,

4e−σ∗ (cos θ + sin θ)2 + m2 cos2 θ = λ2 , (3.51)

and the constant C1 is defined as

1 2
χ10 + λ1 ϕ210 + χ220 + λ2 ϕ220 .


C1 ≡ (3.52)
2
At the last step, we used (3.47). Recalling that ρrad is a function of A, the RHS of (3.50) is
almost a function of only Ā, and its explicit t-dependence is negligible. In fact, as we can see
˙ Hence we have 6
from Fig. 4, the oscillating behavior of Ȧ is almost cancelled by adding B̃.
r
˙ 1

Ā ' C1 e−3Ā + ρrad . (3.53)
3
6
We are interested in only the expanding-universe solution (Ā˙ > 0).

14
From this, we obtain
s
Z Ā
3
t − tref = dA . (3.54)
Ā(tref ) C1 e−3A + ρrad (A)

By taking the inverse function of this, the 3D scale factor a ' eĀ is obtained as a function of t.
In the standard 4D cosmology, the 3D scale factor a(t) behaves as
a(t) ∝ (t − t0 )p , (t0 : constant) (3.55)
where p = 1/2 in the radiation-dominated era and p = 2/3 in the matter-dominated era. If
a(t) behaves as (3.55), the power p is calculated as
ȧ
p ≡ (t − t0 ) = (t − t0 )Ȧ, (3.56)
a
where the constant t0 is chosen so that p becomes independent of t. In our setup, this quantity
oscillates in time. Thus, it is convenient to use Ā instead of A in order to define “effective
power” p (see the right plot of Fig. 4). In fact, we can easily calculate this power in the
radiation-dominated universe, and it turns out to be that p = 1/2 in the 3D space while
p = 5/9 in the 5D space. In our setup, the universe eventually approaches a 3D space in which
the moduli oscillation dominates the energy density. In such a case, p takes 2/3, which is the
same value as the matter-dominated 3D universe.

3.6 Evolution of 3D space

The expression (3.54) can be rewritten as
Z x s
dĀ 3
t − tref ' dx̃ (x̃) −3Ā(x̃)
, (3.57)
xref dx C1 e + ρrad (x̃)
7
where
gdof
ρrad (x) = v ap (x). (3.58)
8π 3 b6∗ x6 ρ
From (C.7), dĀ/dx is given by
 !
(j)
dĀ  1
 1 c
− (j) 1 (j) (xj < x ≤ xj+1 ≤ 10)
(x) = c(j)
2
x c1 x + c2 , (3.59)
dx 
 −1
x (x > 10)

and from (C.10)-(C.13), e−3Ā(x) is given by

 3
 Kk (x)

 (xref ≤ x ≤ xk+1 )
 aref Kk (xref )


 3
e−3Ā(x) = Kj (x)
(xj < x ≤ xj+1 ) . (3.60)
 aarefxKk (xref )Kk+1 (xk+1 ) · · · Kj (xj )



 −3
 ref Kk (xref )Kk+1 (xk+1 ) · · · KJ−1 (xJ−1 )

 (x > 10)
10
7
Precisely, ρrad (x) should be written as ρrad (Ā(x)).

15
 1.0 1.00

0.8 0.95
rad rad
ρ ρ
 0.6
ρtot  0.90
ρtot

0.4 0.85

0.2 0.80
0 20 000 40 000 60 000 80 000 100 000 0 20 000 40 000 60 000 80 000 100 000
t t

Figure 5. The time evolution of the ratio ρrad /ρ̃tot . The blue solid and the red dashed lines represent
the full numerical calculations and the approximate ones, respectively. The lines from bottom to top
correspond to βI = 10, 12.5, 15 (left plot) and βI = 15, 17.5, 20 (right plot). The other parameters are
chosen as (3.43). The reference time is chosen as tref = 10000.

From these expressions, we can numerically compute the 3D scale factor a ' eĀ at an arbitrary
time through the auxiliary variable x, which is the (normalized) inverse temperature. This
approximation makes it easier to compute a at a much later time than t = tref , which cannot
practically be obtained by the full numerical computation.
Here we check the validity of these approximations by comparing with the results of the full
numerical computation. For t ≥ tref , the total energy density ρ̃tot is expressed as
ρ̃tot ≡ C1 e−3Ā + ρrad (Ā). (3.61)
Fig. 5 shows the ratio of the radiation energy density to the total energy density ρrad /ρ̃tot as
a function of t. The blue solid lines represent the full numerical results and the red dashed
lines represent the approximate ones obtained by using (3.58) and (3.60). The constant C1
defined in (3.52) is determined by numerically calculating the time evolution of the moduli up
to t = tref . We can see that our approximation well reproduces the full numerical results.
Using our approximate expressions, we can see how the time evolution of various quantities
become the usual 4D ones. For example, the “effective power” p defined in (3.56) is calculated
as
 −1
˙ dĀ dt
p = (t − t0 )Ā ' {t(x) − t0 } (x) (x)
dx dx
( Z x s )r
dĀ 3 C1 e−3A(x) + ρrad (x)
= tref − t0 + dx̃ (x̃) . (3.62)
xref dx C1 e−3A(x̃) + ρrad (x̃) 3

Combining this and (3.57), we obtain p as a function of t. The constant t0 is chosen so that
p is almost independent of t at earlier times t ≤ tref . Similarly we can obtain the equation of
state:
prad
3
wrad ≡ (3.63)
ρrad
−1
as a function of t. Notice that its reciprocal wrad measures the effective space dimensions that
−1
the radiation feels. Fig. 6 shows the time evolutions of p and wrad computed by the above

16
 0.70
5.0

0.65 4.5

p 0.60 wrad -1 4.0

0.55 3.5

3.0
0.50 4
10 10 5
10 6
10 7
10 8
10 9
104 105 106 107 108 109
t t

Figure 6. The effective power p defined in (3.56) (left plot) and the reciprocal of wrad defined in (3.63)
(right plot) as functions of t. The (purple) solid, (blue) dashed, (green) dotted and (red) dotdashed
lines represent the cases of βI = 10, 12.5, 15 and 20, respectively. The constant t0 in (3.56) is chosen
as t0 = −146, −306, −548 and −1332 for those cases. The other parameters are chosen as (2.17) with
bI = b∗ and σI = σ∗ .

approximations. In the case of βI = 10, we can see that p changes from 5/9, which corresponds
to the radiation-dominated 5D universe, to 2/3, which corresponds to the oscillating-moduli-
dominated 3D universe, as expected. For lower initial temperatures, p once decreases and
approaches 1/2, which corresponds to the radiation-dominated 3D universe, then turns to
increase due to the dominance of the moduli oscillation. From the right plot of Fig. 6, we can
see that the effective space dimensions decrease from 5 to 3 during the period 106 < t < 108 . The
number of the effective dimensions is reduced to 3 at later times for lower initial temperatures.

4 Conditions for radiation dominance

4.1 Dependence on initial displacements of moduli
In this subsection, we discuss the dependence of the initial displacement of the moduli ∆BI ≡
BI − B∗ and ∆σI ≡ σI − σ∗ on the evolution of the universe. As shown in Fig. 6, it takes a long
time until the moduli oscillation dominates the total energy density of the universe when the
radiation dominates at t = tref . In such a case, if the moduli decay before the oscillation energy
dominates, the universe does not experience the moduli-oscillation-dominated era. This never
happens when the moduli stabilization procedure is described in the 4D EFT (i.e., b∗ , β  m−1 ).
Thus, in this section, we aim to clarify in what case such unusual situations occur. Notice that
the time evolution of the 3D scale factor a is determined by the total energy density ρ̃tot . Hence
if the ratio ρrad /ρ̃tot is close to one at t = tref , the radiation-dominated era lasts for a long time.
(1)
Fig. 7 shows this ratio at t = tref . In the case that m/mKK = 0.1 (lower plots) and the initial
temperature is low, the ratio is close to one only when ∆BI = ∆σI = 0, which corresponds
to the case that the moduli have already been stabilized at t = 0. Even small displacements
from the stabilized values lead to the dominance of the moduli oscillation. This is consistent
(1)
with the analysis in the 4D EFT. In contrast, in the case that m/mKK = 10 (upper plots), the

17
 1.0 1.0

0.8 0.8

ρrad (tref ) 0.6 ρrad (tref ) 0.6

~ tot ~ tot
ρ (tref ) 0.4 ρ (tref ) 0.4

0.2 0.2

0.0 0.0
-0.4 -0.2 0.0 0.2 0.4 -0.2 -0.1 0.0 0.1 0.2
BI -B* σI -σ*

1.0 1.0

0.8 0.8

ρrad (tref ) 0.6 ρrad (tref ) 0.6

~ tot ~ tot
ρ (tref ) 0.4 ρ (tref ) 0.4

0.2 0.2

0.0 0.0
-0.10 -0.05 0.00 0.05 0.10 -0.10 -0.05 0.00 0.05 0.10
BI -B* σI -σ*

Figure 7. The ratio ρrad /ρ̃tot at t = tref as functions of BI − B∗ with σI = σ∗ (left plots) and σI − σ∗
(1)
with BI = B∗ (right plots). The upper plots are in the case of m/mKK = 10 (σ∗ = 15.9), and the lower
(1)
plots are in the case of m/mKK = 0.1 (σ∗ = 6.6). The red solid, the orange dashed, the green dotted,
the blue dotdashed and the purple solid lines correspond to βI = 10, 12.5, 15, 20 and 30, respectively.

radiation contribution is still non-negligible at t = tref also for non-zero displacements. Note
that the moduli stabilization cannot be described in the context of the 4D EFT in this case.
We should also note that even if the moduli have been stabilized at t = 0 (i.e., ∆BI =
∆σI = 0), the contribution of the moduli oscillation becomes non-negligible at t = tref when
(1)
m/mKK  1 and the initial temperature is high. This indicates that the moduli start to
oscillate due to the pressure in the extra compact space prad2 . We will discuss this issue in the
next subsection.

4.2 Effects of the pressure prad

2
If no radiation exists, the moduli are safely stabilized for any initial values BI and σI . In
the presence of the radiation, however, this becomes nontrivial since the pressure in the S 2
space prad
2 pushes the moduli from the stabilized values. This effect is relevant only at early
times because prad
2 is rapidly damped to zero. Still, it gives a significant effect on the evolution
of the moduli in some cases, as we have seen in the upper plots of Fig. 7. In this subsection,
we analyze the moduli evolution equations focusing on early times and keeping the prad 2 term,
which was neglected in Sec. 3.4.
(1)
When β  b∗ (i.e., the temperature is much higher than mKK ), the pressure in the compact

18
space prad
2 is almost independent of the moduli,
10gdof 2gdof
ρrad ' , prad
2 ' . (4.1)
π3β 6 π3β 6
In this case, the moduli evolution equations become inhomogeneous differential equations. In-
cluding prad
2 , the evolution equation (3.35) is modified as
      
ϕ̈1 λ1 ϕ1 3
A rad cos θ
=− + 2e 2 p2 + ··· , (4.2)
ϕ̈2 λ2 ϕ2 − sin θ

where λ1,2 , ϕ1,2 and θ are defined in (3.36), (3.37) and (3.38). We have neglected terms involving
Ä. From (3.10) and the fact that vβ (x) ' 3/5 for x  1, we can express the inverse temperature
as β ' βI a3/5 , and thus obtain
10gdof − 18 Ā 18
ρrad ' 3 6
e 5 ≡ C2 e− 5 Ā ,
π βI
2gdof 18 C2 − 18 Ā
prad
2 ' 3 6 e− 5 Ā = e 5 . (4.3)
π βI 5

˙
If we assume that Ā(t) ˙
≤ Ā(0)  λ1 , (3.46) can be approximated as (3.50). Thus the
energy density of the moduli oscillation is estimated as

e−3Ā 2
ρosc ≡ ρ̃tot − ρrad = ϕ̇1 + ϕ̇22 + λ1 ϕ21 + λ2 ϕ22 .


(4.4)
2
Here we assume that
1 2
ϕ̇1 + ϕ̇22 + λ1 ϕ21 + λ2 ϕ22


C1I ≡ (4.5)
2

is almost independent of the time t. Then, similarly to (3.53), we can express Ā˙ as
r 
1 
Ā˙ '
18
C1I e−3Ā + C2 e− 5 Ā , (4.6)
3
which leads to
 2  3 
√ e 2 Ā − 1 for C1I  C2
s
Ā
Z 
3 3C1I 
t' dz 18 ' 5  . (4.7)
1 C1I e−3z + C2 e− 5 z  √
 9
e −1
5
Ā 3
Ā
for C1I e  C2
5
3 3C2
By taking the inverse function, the 3D scale factor is expressed as
 √ 2/3
 3C1I
 1+ t for C1I  C2


2
a(t) = eA(t) '  √ 5/9 , (4.8)
 3 3C 2 3
 1+ t for C1I e 5 Ā  C2


5

19
Therefore, we find the time dependence of the inhomogeneous term in (4.2) as
  √ − 75
 2C 2 3C 1I
1+ t for C1I  C2


3
A rad 4gdof − 21 Ā(t)  5 2
2e 2 p2 ' 3 6 e 10 ' √ − 67 . (4.9)
π βI  2C 2

3 3C 2 3
for C1I e 5 Ā  C2


 1+ t
5 5
To illustrate the situation, we roughly approximate this as
3 2C2
2e 2 A prad
2 ∼ (1 + αt)−1 , (4.10)
5
√ √
where α ∼ C1I for C1I  C2 , and α ∼ C2 for C1I ≤ C1I e3Ā/5  C2 . Then (4.2) can be
solved as
 p  p  p √ 
3HI 2C2 cos θ λ1
ϕ1 (t) = ϕ1I cos λ1 t + √ sin λ1 t − √ G λ1 t; ,
2 λ1 5α λ1 α
 p  p  p √ 
3HI 2C2 sin θ λ2
ϕ2 (t) = ϕ2I cos λ2 t + √ sin λ2 t + √ G λ2 t; , (4.11)
2 λ2 5α λ2 α
where
G(z; c) ≡ {Si(z + c) − Si(c)} cos(z + c) − {Ci(z + c) − Ci(c)} sin(z + c). (4.12)
The functions Si(z) and Ci(z) are trigonometric integrals defined by
Z z Z ∞
sin w cos w
Si(z) ≡ dw , Ci(z) ≡ − dw . (4.13)
0 w z w
We have used the initial conditions,
3
ϕ1 (0) = ϕ1I ≡ 2∆BI cos θ + ∆σI sin θ, ϕ̇1 (0) = Ȧ(0)ϕ1I ,
2
3
ϕ2 (0) = ϕ2I ≡ −2∆BI sin θ + ∆σI cos θ, ϕ̇2 (0) = Ȧ(0)ϕ2I ,
s  2
2

1 2 m
Ȧ(0) = 2e−σI (1 − e∆σI −∆BI ) + (∆σI )2 + C2 ≡ HI , (4.14)
3 2

where ∆BI ≡ BI − B∗ and ∆σI ≡ σI − σ∗ .

In order to see the impact of prad2 on the spacetime evolution at early times, we focus on
the case of ∆BI = ∆σI = 0. Namely, the moduli have already been stabilized at the beginning
of the radiation-dominated era. Then the energy density of the moduli oscillation (4.4) is
estimated as
√  √ 
2C22 −3Ā(t)
 p p
osc 2 λ1 2 λ2
ρ = 2
e cos θH λ1 t; + sin θH λ2 t; , (4.15)
25α α α
where
2
H(z; c) ≡ G 2 (z; c) + {G 0 (z; c)}
= {Ci (z + c) − Ci (c)}2 + {Si (z + c) − Si (c)}2 . (4.16)

20
 100 300
250
0.1
200
ℋ(z;c) 10-4 ℋ(3,10γ ) 150

10-7 100
50
10-10
0
0 2 4 6 8 10 12 14 -8 -6 -4 -2 0 2 4
z γ

Figure 8. The profiles of H(z; c). In the left plot, the parameter c is chosen as c = 10−4 , 10−2 , 1, 102 , 104
from top to bottom. The right plot shows the first peak value of H(z; c) as a function of γ ≡ ln10 c.

Fig. 8 shows the profiles of the function H(z; c). As we can see from the left plot, H(z; c) is
almost independent of z when c  1. In that case, H(z; c) ' ln2 c (see the right plot). When
c > 1 (z > 0), H(z; c) is an oscillating function around 1/c2 , whose first peak is around z = 3
and H(3; c) ' 4/c2 .
Here let us assume that

C1I  C2 . (4.17)
√
Then, since α ∼ C2 , (4.15) becomes
( r ! r !)
2C 2 −3Ā(t)
p λ1
p λ2
ρosc (t) ∼ e cos2 θH λ1 t; + sin2 θH λ2 t; . (4.18)
25 C2 C2

Thus if
( r ! r !)
2C2 e−3Ā(tref ) p λ1 p λ2
ρosc (tref ) ∼ cos2 θH λ1 tref ; + sin2 θH λ2 tref ;
25 C2 C2
18
 ρrad (tref ) ' C2 e− 5 Ā(tref ) (4.19)

holds, the assumption (4.17) is justified. This condition is rewritten as

R  1, (4.20)
8
where
( r ! r !)
2  p 1/3 p λ1 p λ2
R≡ 1 + C2 tref cos2 θH λ1 tref ; + sin2 θH λ2 tref ; . (4.21)
25 C2 C2

In this case, the radiation energy is dominated at t = tref . If (4.20) does not hold, the con-
tribution of the moduli oscillation to the total energy density cannot be negligible, and it will
(1)
dominate after t = tref . Note that λ1 , λ2 and θ are functions of m and σ∗ (or mKK ), and C2 is
a function of βI .
8
√
We have used the approximate expression in (4.8), and 3 3/5 ' 1.

21
 (1 ) (1 )
σ* =15.9 (mKK =0.001) σ* =14.0 (mKK =0.0026)
1.0 1.0

0.8 0.8

0.6 0.6
ℛ ℛ
0.4 0.4

0.2 0.2

0.0 0.0
10 15 20 25 30 10 15 20 25 30
βI βI
(1 ) (1 )
σ* =11.3 (mKK =0.01) σ* =6.6 (mKK =0.1)
1.0 1.0

0.8 0.8

0.6 0.6
ℛ ℛ
0.4 0.4

0.2 0.2

0.0 0.0
10 15 20 25 30 10 15 20 25 30
βI βI

Figure 9. R as a function of βI . The (purple) solid, (blue) dashed, (green) dotted, and (red) dotdashed
lines correspond to m = 0.1, 0.01, 0.001 and 0.0001, respectively.

Fig. 9 shows R as a function of the initial (inverse) temperature βI for various values of m
(1)
and σ∗ (or mKK ). The other parameters are chosen as gdof = 100 and tref = 10000. We can
see that the (blue) dashed lines in the upper left and the lower right plots are consistent with
(1)
the result shown in Fig. 7. In the case of m = 0.01 and mKK = 0.001, the contribution of the
moduli oscillation becomes non-negligible for β ∼ 10 even when the initial moduli values are
(1)
set at the stabilized values. In contrast, in the case of m = 0.01 and mKK = 0.1, the radiation
contribution always dominates.
(1) (1)2
When m  mKK , λ1 ' mKK , λ2 ' m2 and tan θ  1 (see (3.36) and (3.38)). In this case,
(4.21) reduces to
r !1/3 s !
2 10gdof tref π 3
(1) (1)
R' 1+ H mKK tref ; m β3 , (4.22)
25 π 3 βI3 10gdof KK I
which is almost independent of λ2 ' m2 . Recalling the behavior of H(z; c) (see the right plot
in Fig. 8), R is estimated as
    
2 (1) 3 (1)
ln mKK βI
 mKK βI3  1
R∼ 1 
(1) 3
. (4.23)
 (1)2 6  1
 mKK βI  1
mKK βI

22
We have assumed that the factor in front of H is O(1). Therefore, the moduli oscillation
(1)
dominates the total energy density when mKK βI3  1, while the its contribution is negligible
(1)
when mKK βI3  1. We can see these properties from the upper plots in Fig. 9.

5 Summary
We analyze spacetime evolution during the moduli oscillation in a 6D model compactified
on S 2 in the presence of the radiation. In our previous work [11], we studied the model
by numerically solving the field equations, and found that the radiation contribution to the
total energy density remains non-negligible for a long time in a situation that the moduli
stabilization dynamics cannot be described in the context of 4D EFT. This is in contrast to the
result obtained by the conventional 4D EFT approach. However, such a numerical approach is
available for only a limited range of the time. In fact, it is practically difficult to pursue the
evolution until the spacetime behaves like 4D. In the current work, we develop a procedure to
compute it for large t, and see the transition from 6D to 4D explicitly.
The evolution of the 3D scale factor is characterized by the power p defined in (3.56).
When the initial temperature is high enough, p monotonically increases up to 2/3, which
corresponds to the moduli-oscillation-dominated 4D universe. For lower initial temperatures, it
first decreases and approaches 1/2, which corresponds to the radiation-dominated 4D universe,
and then turns to increase. This indicates that the radiation feels the decrease of the effective
space dimensions before the moduli oscillation dominates the total energy density. In such a
case, the universe never experiences the moduli-oscillation-dominated era if the moduli decay
before the dominance of the moduli oscillation.
We also study the conditions that the radiation remains to give a dominant contribution to
the total energy density for a long time. We found that such a situation occurs when the moduli
stabilization dynamics cannot be described in the context of 4D EFT (i.e., b∗ , β  m−1 ).
Another important result we obtained is that even if the moduli have already been stabilized
at the beginning of the radiation-dominated era, the pressure in the extra compact space prad 2
pushes the stabilized moduli and they start to oscillate again in some cases. This occurs when
the KK mass scale and the initial temperature are higher than the mass scale of the moduli
stabilization potential. When the moduli stabilization is described in the context of 4D EFT,
this never happens because prad2 is negligible in such a case.
In this work, we have neglected the decay of the moduli, which is essential to the study
of realistic scenarios. In the subsequent papers, we will take it into account and investigate
a full thermal history of the universe, incorporating the Standard Model sector localized on a
codimension-two brane [13]- [19].

Acknowledgements
H. O. was supported in part by JSPS KAKENHI Grant Numbers JP20K14477 and the Edu-
cation and Research Program for Mathematical and Data Science from the Kyushu University.

23
A Thermodynamic quantities
The dispersion relation of a 6D relativistic or massless particle is
1 ~2 1 1
k M kM = −k02 + 2
k + 2 kθ2 + 2 2 kφ2 = 0. (A.1)
a b b sin θ

Thus the energy of the particle with the 3D momentum ~k = (k1 , k2 , k3 ) and the angular mo-
mentum l on S 2 is given by
r
k 2 l(l + 1)
Ek,l = k0 = + , (A.2)
a2 b2
p
where k ≡ ~k 2 . Since each one-particle state is specified by ~k, l and the ‘magnetic quantum
number’ m = −l, · · · , l, we have (2l + 1) degenerate energy eigenstates for each ~k and l. Hence
the grand potential is expressed as
∞ Z ∞
X gdof (2l + 1)
dk k 2 ln 1 ∓ e−β(Ek,l −µ)


J(β, µ, V3 , V2 ) = ±
l=0
2π 2 β 0

gdof V3
∞
X gdof (2l + 1)V3 Z ∞  √2 2 
= ∓ 2 4 Li4 (±eβµ ) ± dq q 2 ln 1 ∓ e− q +cl +βµ ,
π β l=1
2π 2 β 4 0
(A.3)

where gdof denotes the degrees of freedom for the 6D relativistic particles, β is the inverse
temperature, µ is the chemical potential, and V3 ≡ a3 and V2 ≡ 4πb2 are the comoving volume
for the 3D space and the physical volume of S 2 , respectively. The upper (lower) signs correspond
to the case of bosons (fermions). At the second equality, we have rescaled the integration
variable and the KK masses as
s p
β 4πl(l + 1) β l(l + 1)
q ≡ k, cl ≡ β = . (A.4)
a V2 b

The function Li4 (z) in the second line of (A.3) is the polylogarithmic function. In the following,
we consider a situation in which e−cl +βµ  1 for l ≥ 1. Then the grand potential can be
approximated as
  r 
gdof V3 βµ βµ 4π
J(β, µ, V3 , V2 ) ' − 2 4 ±2Li(±e ) + e Q1 β , (A.5)
2π β V2

where
∞
X  p 
2
Q1 (x) ≡ x l(l + 1)(2l + 1)K2 x l(l + 1) . (A.6)
l=1

Here K2 (z) is the modified Bessel function of the second kind.

From (A.3), various thermodynamic quantities are calculated as follows. The upper (lower)
signs represent the case of bosons (fermions).

24
Radiation energy density
 
rad 1 µ
ρ = ∂β − ∂µ (βJ)
V3 V2 β
gdof 
' 2 4 ±6Li4 (±eβµ ) + eβµ (3Q1 + Q2 ) , (A.7)
2π β V2
where
∞
X  p 
Q2 (x) ≡ −xQ01 (x) = x3 l3/2 (l + 1)3/2 (2l + 1)K1 x l(l + 1) . (A.8)
l=1

3D pressure
1 ∂J gdof 
prad
3 = − ' 2 4 ±2Li4 (±eβµ ) + eβµ Q1 . (A.9)
V2 ∂V3 2π β V2

2D pressure

1 ∂J gdof eβµ
prad
2 = − ' 2 4 Q2 . (A.10)
V3 ∂V2 4π β V2
p
The arguments of the functions Q1 and Q2 are understood as β 4π/V2 = β/b.
We should note that

ρrad = 3prad rad

3 + 2p2 . (A.11)

The profiles of the functions x2 Qi (x) (i = 1, 2, 3) are shown in Fig. 10.

B Conservation law
Including the radiation contribution, the energy-momentum conservation law is

∇M T MN ≡ ∂M T MN + ΓMM L T LN − ΓLM N T ML = 0, (B.1)

where
1 eσ
T tt = σ̇ 2 + 4 + V (σ) + ρrad ≡ ρtot ,
2  8b
eσ

i i 1 2 rad
T j = δ j − σ̇ + 4 + V (σ) − p3 ≡ −δ i j ptot
3 ,
2 8b
1 eσ
T 44 = T 55 = − σ̇ 2 − 4 + V (σ) − prad tot
2 ≡ −p2 . (B.2)
2 8b
From (B.1) with N = t, we have

3ȧ tot
2ḃ tot

ρ̇tot + ρ + ptot ρ + ptot


3 + 2 = 0, (B.3)
a b

25
 15 30
25
10 20
x 2 Q1 (x) x 2 Q2 (x) 15

5 10
5
0 0
0 2 4 6 8 10 12 14 0 2 4 6 8 10 12 14
x x

120
100
80
x 2 Q3 (x) 60

40
20
0
0 2 4 6 8 10 12 14
x

Figure 10. The profiles of x2 Qi (x) (i = 1, 2, 3).

where the dot denotes the time derivative. The other components hold trivially. By using the
dilaton field equation in (2.9), the conservation law (B.3) is reduced to
!
3ȧ 2ḃ 3ȧ 2ḃ rad
ρ̇rad + + ρrad + prad3 + p = 0. (B.4)
a b a b 2

Plugging (A.7), (A.9) and (A.10) into this, we obtain

β̇ 
±24Li4 (±eβµ ) + eβµ (12Q1 + 5Q2 + Q3 )
β
−βµ ±6Li3 (±eβµ ) + eβµ (3Q1 + Q2 )

3ȧ  ḃ
= ±8Li4 (±eβµ ) + eβµ (4Q1 + Q2 ) + eβµ (2Q2 + Q3 ) , (B.5)
a b
where

Q3 (x) ≡ 2Q2 (x) − xQ02 (x)

X∞  p 
4 2 2
= x l (l + 1) (2l + 1)K0 x l(l + 1) . (B.6)
l=1

The arguments of Q1,2,3 are understood as β/b.

26
C Derivation of (3.14)
From (3.10), we obtain
Z βj+1 Z aj+1
dβ̃ dã aj+1
ap
= = ln , (C.1)
β β̃vβ (β̃/b∗ ) a ã a

where aj ≡ a(βj ), for xj < β/b∗ ≤ xj+1 (≤ 10). The LHS of (C.1) is calculated as
" #xj+1 !
Z xj+1 (j) (j)
dx 1 x 1 c β + c 2 b∗
ap = (j) ln (j) (j)
= (j) ln Bj 1 , (C.2)
β/b∗ xvβ (x) c2 c1 x + c2 c2 β
β/b∗

where
xj+1 (j + 1)∆
Bj ≡ (j) (j)
= (j) (j)
. (C.3)
c1 xj+1 + c2 c1 (j + 1)∆ + c2
Thus, we can express a as a function of β as
!1/c2(j)
1 β
a(β) = aj+1 , (C.4)
Bj c1 β + c(j)
(j)
2 b∗

for xj < β/b∗ ≤ xj+1 (≤ 10).

For β > 10b∗ , the corresponding equation to (C.1) is
Z β/b∗
a dx β
ln = = ln , (C.5)
aJ 10 x 10b∗
which leads to
aJ β
a(β) = . (C.6)
10b∗
From (C.4) and (C.6), we have
 !
(j) (j)
Aj+1 − 1 ln Bj c1 x + c2

(xj < x ≤ xj+1 ≤ 10)

(j) x
A(x) = c2 , (C.7)

AJ + ln x
(x > 10)

10
where x ≡ β/b∗ and Aj ≡ ln aj .
For the reference time tref , we define the integer k so that xk < β(tref )/b∗ ≤ xk+1 . Then,
from (C.4), we obtain
ak+1
aref = , (C.8)
Kk (xref )
where aref ≡ a(tref ), xref ≡ β(tref )/b∗ , and
!1/c2(j) !1/c2(j)
(j) (j) (j) (j)
c x + c2 xj+1 c x + c2
Kj (x) ≡ Bj 1 = (j) (j)
· 1 . (C.9)
x c1 xj+1 + c2 x

27
Therefore, using (C.4) repeatedly, we have

ak+1 aref Kk (xref )

a(x) = = , (C.10)
Kk (x) Kk (x)
for xref ≤ x ≤ xk+1 ,

ak+2 ak+1 Kk+1 (xk+1 ) aref Kk (xref )Kk+1 (xk+1 )

a(x) = = = , (C.11)
Kk+1 (x) Kk+1 (x) Kk+1 (x)
for xk+1 ≤ x ≤ xk+2 ,
aj+1 1 1
a(x) = = aj Kj (xj ) = aj−1 Kj−1 (xj−1 )Kj (xj )
Kj (x) Kj (x) Kj (x)
1
= · · · = ak+1 Kk+1 (xk+1 ) · · · Kj (xj )
Kj (x)
1
= aref Kk (xref )Kk+1 (xk+1 ) · · · Kj (xj ) , (C.12)
Kj (x)

for xj < x ≤ xj+1 ≤ xJ = 10, and (C.6) is expressed as

x
a(x) = aref Kk (xref )Kk+1 (xk+1 ) · · · KJ−1 (xJ−1 ), (C.13)
10
for x > 10.

References
[1] A. Salam and E. Sezgin, Phys. Lett. B 147 (1984), 47.

[2] H. Nishino and E. Sezgin, Phys. Lett. B 144 (1984), 187-192.

[3] S. Randjbar-Daemi, A. Salam, E. Sezgin and J. A. Strathdee, Phys. Lett. B 151 (1985),
351-356.

[4] K. i. Maeda and H. Nishino, Phys. Lett. B 154 (1985), 358-362.

[5] M. B. Green, J. H. Schwarz and P. C. West, Nucl. Phys. B 254 (1985), 327-348.

[6] G. D. Coughlan, W. Fischler, E. W. Kolb, S. Raby and G. G. Ross, Phys. Lett. B 131,
59-64 (1983).

[7] A. S. Goncharov, A. D. Linde and M. I. Vysotsky, Phys. Lett. B 147, 279-283 (1984).

[8] J. R. Ellis, D. V. Nanopoulos and M. Quiros, Phys. Lett. B 174, 176-182 (1986).

[9] T. Banks, D. B. Kaplan and A. E. Nelson, Phys. Rev. D 49, 779-787 (1994) [arXiv:hep-
ph/9308292 [hep-ph]].

[10] B. de Carlos, J. A. Casas, F. Quevedo and E. Roulet, Phys. Lett. B 318, 447-456 (1993)
[arXiv:hep-ph/9308325 [hep-ph]].

28
[11] H. Otsuka and Y. Sakamura, JHEP 08 (2022), 120 [arXiv:2205.00175 [hep-th]].

[12] V. Kumar, D. R. Morrison and W. Taylor, JHEP 11 (2010), 118 [arXiv:1008.1062 [hep-th]].

[13] Y. Aghababaie, C. P. Burgess, S. L. Parameswaran and F. Quevedo, Nucl. Phys. B 680

(2004), 389-414 [arXiv:hep-th/0304256 [hep-th]].

[14] C. P. Burgess, Annals Phys. 313 (2004), 283-401 [arXiv:hep-th/0402200 [hep-th]].

[15] J. Garriga and M. Porrati, JHEP 08 (2004), 028 [arXiv:hep-th/0406158 [hep-th]].

[16] C. P. Burgess, F. Quevedo, G. Tasinato and I. Zavala, JHEP 11, 069 (2004) [arXiv:hep-
th/0408109 [hep-th]].

[17] G. W. Gibbons, R. Gueven and C. N. Pope, Phys. Lett. B 595 (2004), 498-504 [arXiv:hep-
th/0307238 [hep-th]].

[18] Y. Aghababaie, C. P. Burgess, J. M. Cline, H. Firouzjahi, S. L. Parameswaran, F. Quevedo,

G. Tasinato and I. Zavala, JHEP 09 (2003), 037 [arXiv:hep-th/0308064 [hep-th]].

[19] H. M. Lee and C. Ludeling, JHEP 01 (2006), 062 [arXiv:hep-th/0510026 [hep-th]].

29