Unstable growth of curvature perturbation in non-singular bouncing cosmologies

BingKan Xue1 and Paul J. Steinhardt1, 2

1
Department of Physics, Princeton University, Princeton, New Jersey 08544, USA
2
Princeton Center for Theoretical Physics, Princeton University, Princeton, New Jersey 08544, USA
We consider non-singular bouncing cosmologies, such as the new ekpyrotic model, in which the
universe undergoes a slow contraction phase with equation of state w  1, followed by a bounce
that occurs at a finite scale factor when quantum gravity corrections are still negligible. Such a
non-singular bounce requires a violation of the null energy condition in which w falls below −1 at
some time before the bounce. In this paper, we show that a component of the adiabatic curvature
perturbations, though decaying and negligible during the ekpyrotic phase, is exponentially amplified
just before w approaches −1, enough to spoil the scale-invariant perturbation spectrum. We discuss
how the problem may be avoided, for example, in singular bounces.
arXiv:1007.2875v1 [hep-th] 16 Jul 2010

As an alternative to inflationary cosmology, the ekpy-
rotic scenario [1, 2] resolves the homogeneity, flatness
and isotropy problems of standard cosmology by hav-
ing the universe undergo a period of ultra-slow contrac-
tion followed by a bounce to the standard expanding
phase. Ultra-slow contraction (ekpyrotic phase) occurs
when the dominant energy component has an equation of
state w  1. During contraction, this component grows
faster than all other contributions to the cosmic expan-
sion, including the spatial curvature and anisotropy, and
thereby drives the universe into an exponentially flat and
isotropic state leading up to the bounce [3].
As for the bounce, two possibilities have been dis-
cussed. In “singular bounce” models, such as the cyclic
model [4], the universe contracts towards a “big crunch”
until the scale factor a(t) is so small that quantum grav-
ity effects become important. The presumption is that
these quantum gravity effects introduce deviations from
conventional general relativity and produce a bounce that
preserves the smooth, flat conditions achieved during the
ultra-slow contraction phase.
The second possibility is a “non-singular bounce”, such
as in the “new ekpyrotic model” [5], where the universe
stops contraction and reverses to expansion at a finite
value of a(t) where classical general relativity is still valid.
A significant advantage of this scenario is that the entire
cosmological history can be described by 4d effective field
theory and classical general relativity, without invoking
extra dimensions or quantum gravity effects. However,
for the bounce to happen within classical general relativ-
ity, w must fall below −1 for a sustained period, i.e. a
violation of the null energy condition (NEC). The NEC
violating energy component is commonly chosen to be a
scalar field that undergoes ghost condensation [6, 7].
In this paper, we show that the non-singular bounce
creates a serious problem for cosmological perturba-
tions. Observations of the cosmic microwave background
(CMB) and large-scale structure have provided evidence
for a scale-invariant perturbation spectrum. However,
in the ekpyrotic model when the scale-invariant curva-
ture perturbation is generated during or just after the
ekpyrotic phase, a potentially dangerous component of
adiabatic curvature perturbations is created at the same
time. This mode has been previously ignored because,
after exiting horizon when w ≥ 1, its amplitude becomes
exponentially suppressed on large length scales compared
to the scale-invariant modes. In a singular bounce, this
mode remains completely negligible because w ≥ 1 all
the way up to the bounce. However, for the non-singular
bounce, the ekpyrotic phase must end and w must fall
below −1. We show that, right before crossing w = −1,
the adiabatic mode undergoes exponential amplification
such that the scale-invariant spectrum is spoiled and per-
turbation theory breaks down.
For our study we take the new ekpyrotic model [5]
as a generic example of non-singular bounces. In this
model both the ekpyrotic phase and the ghost-condensate
phase are driven by the same scalar field, since otherwise
one encounters the problem that the energy of the ghost-
condensate field is so exponentially diluted away after the
ekpyrotic phase that it is too insignificant to cause the
bounce. This framework can be described by an effective
Lagrangian
√ 2
L = −g 12 R + P (X) − V (φ) , X ≡ − 21 (∂φ) , (1)


assuming a FRW background metric with signature (− +
++), and reduced Planck units 8πG ≡ 1. The kinetic
term P (X) is designed as in Fig. 1, where it is linear for
large X, P (X) ≈ X, but has a minimum at a low en-
ergy scale Xc . The potential V (φ) is sketched in Fig. 2,
where, beginning from the right hand side,√V is approx-
imated by a negative exponential −V0 e− 2/p φ over a
range between Vek-beg and Vek-end , then bottoms out and
undergoes a steep rise. The universe evolves through the
ekpyrotic, kinetic and bouncing phases, as indicated in
the figures.
During the ekpyrotic phase, the Friedmann equation
has an attractor solution,
−t

p p
a = −tek-end , H = − (−t) , (2)
p q 
V0
φ = 2p log p(1−3p) (−t) ,
where t is negative and increasing towards zero, and we
 2

P(X) towards V = 0. During this phase the equation of state

Xek-end w rapidly decreases according to
h  i
−V,φ
p p
ẇ = 2 3(1 + w) H 3H 2 + w−1
2 3(1 + w) . (5)
bouncing
phase kinetic ekpyrotic The first term in the brackets represents the ratio of the
(NEC) phase phase
gradient force to the total energy, which is a large positive
ghost- Xek-beg
factor during the sharp rise in the potential. Therefore
condensate
point
ẇ < 0 in this phase, since H is negative.
In singular bounces [4], the potential approaches zero
0 Xc X from below as φ → −∞, so that w remains ≥ 1 until
the bounce. In contrast, for a non-singular bounce, the
FIG. 1: The kinetic term P (X) versus X ≡ 12 (∂φ)2 for the potential rises above zero as in Fig. 2, so that, as the
ghost-condensate scalar field φ. As indicated by the dotted field climbs, w decreases below 1 and eventually crosses
lines with arrows, during the ekpyrotic phase (X > Xc ), the −1 some time before the bounce. This stage occurs in
approximately canonical kinetic term P (X) ≈ X grows expo- much less than one Hubble time, during which H and a
nentially by a factor of e2Ntot ; After the ekpyrotic phase, X are nearly constant. The kinetic term in the Lagrangian
decreases by an even greater factor to reach X = Xc . This
remains canonical for X > ∼ Xc .
exponential decrease in X is directly related to the problem
with non-singular bounces. The bouncing phase begins when the field climbs suffi-
ciently far up the potential that X falls below Xc and
bouncing V(φ) the ghost condensate phase initiates. In this phase,
NEC
w = -1
phase Ḣ = −XP,X = − 21 ρ(1 + w) becomes positive and w falls
below −1, violating the NEC. The non-singular bounce
present occurs as contraction slows and ultimately halts at a fi-
w=1 ekpyrotic φ
Vek-beg nite value of a.
phase
kinetic For consistency, a necessary condition is (Fig. 1)
phase w >> 1
Xek-end  Xek-beg  Xc , (6)
which keeps the kinetic term P (X) canonical throughout
the ekpyrotic phase. The speed of sound is
Vek-end
P,X
c2s = ≈ 1, (7)
FIG. 2: The effective potential V (φ) in the ekpyrotic model 2XP,X − P
with a non-singular bounce. The evolution along the poten-
tial during the ekpyrotic and bounce phases is from right to when P (X) is linear; in the bouncing phase when w <
left: the ekpyrotic phase refers to the exponential decline from −1, c2s becomes small and negative, but it does not cause
Vek-beg to Vek-end near the minimum of the potential; then instabilities if the universe bounces and exits the ghost-
there is a brief kinetic energy dominated phase as V quickly condensate phase within a Hubble time or so [5, 8].
rises; and finally the non-singular bouncing phase occurs as The serious problem appears when we consider per-
V climbs sufficiently above zero.
turbations around the background evolution. Here we
focus on the widely studied gauge-invariant variable, the
normalize aek-end = 1. This solution has a constant equa- comoving curvature perturbation R. Perturbations in
tion of state, w = 3p2
− 1  1, if we choose p  1. The other gauges will be discussed in [9], where we show that
potential energy and the nearly canonical kinetic energy results agree at the bounce, so that the initial conditions
satisfy the scaling relation for the expanding phase are consistent in all gauges. Dif-
ferent Fourier modes Rk of the comoving curvature per-
3 2
2H = 1
1−w V = 1
1+w X, X = 12 φ̇2 . (3) turbation, labeled by the comoving wave number k, obey
the equation of motion [2]
Therefore, there is an exponential suppression of any ini-
tial curvature and anisotropy, determined by the ratio of z0 0
R00k + 2 R + c2s k 2 Rk = 0, (8)
the kinetic energy densities at the end and beginning of z k
the ekpyrotic phase, q
where z = a c− Ḣ
2 H 2 , and prime denotes derivative with
s
(H 2 )ek-end Xek-end d
respect to conformal time τ , dτ d
= a dt . For small k, the
e2Ntot ≡ = . (4)
(H 2 )ek-beg Xek-beg equation is formally solved in expansions of k 2 ,
A brief kinetic energy dominated phase follows after
Z Z
(0) 2 dτ
the field reaches the bottom of the potential and rises Rk = Rk − k dτ c2s z 2 Rk , (9)
z2
 3

where the leading order term is the sum of a constant curvature perturbation R(s) , which remains constant on
term Rconst
k and an integral term Rint
k superhorizon scales afterwards.
Thus, the total curvature perturbation can be decom-
t
kdt c2s H 2
Z
(0) C1 C2 posed as
Rk = √ + √ ≡ Rconst
k + Rint
k . (10)
k k 0 a a2 (−Ḣ)
Rtot = R(s) + R(σ) ≈ R(s) + R(σ,const) + R(σ,int) , (14)
The dimensionless constants C1 and C2 are numbers of
where as before the adiabatic contribution is divided into
order O(1), as can be found by matching to Minkowski
a constant term that is blue and an integral term that
vacuum condition at the beginning of the ekpyrotic phase
is bluer. For the modes that exited horizon during the
when the mode is deep inside the horizon.
ekpyrotic phase, the constant term R(σ,const) is subdom-
During the ekpyrotic phase, − Ḣ 3
H 2 = 2 (1 + w) = p
1
p
inant compared to the scale-invariant contribution R(s) ,
is nearly constant and a ∼ (−t) does not change sig- and the integral term R(σ,int) is sub-subdominant,
nificantly. Therefore, the integral term decreases as
−
s
Rint
k ∼ k(−t) as t → 0 . It is comparable to Rk
const
at  (σ,const) 
R  k Xk
horizon crossing (k ∼ aH), but becomes exponentially  R(s)  ∼ (aH)ek-end ∼ Xek−end , (15)
 
small by the end of the ekpyrotic phase,  (σ,int)  
R  k 2 Xk
 int  r  ∼ ∼ . (16)
 Rk  pk Xk  R(s)  (aH)ek-end Xek−end

 Rconst 
 ≈ ≈ ≡ e−Nk  1,
k ek-end
(aH) ek-end X ek-end
(11) The critical stage occurs towards the end of the kinetic
where Xk is the kinetic energy at horizon crossing and phase after the conversion. While R(s) and R(σ,const)
Nk is the remaining number of e-foldings of the ekpy- remain constant, R(σ,int) grows exponentially as w → −1.
rotic phase after k mode exits horizon. (Factors of p are The growth can be seen by rewriting Eq. (10) as
neglected for the purpose of these estimates.) Thus (σ,int)
√ Z c2s H 2 √ Z c2s 2 dw
√ Rk ≈ C2 k 3
dt ≈ C 2 k 3
.
a −Ḣ a 3(1 + w) ẇ

const C1 int C2 k (17)
Rk  ≈ Rk ≈ √  Rk ≈ √ . (12)

ek-end k Xek-end Integrating w from w  1 at the end of the ekpyrotic
phase to w ≈ −1 just before the non-linearity in P (X) be-
Unfortunately, Rconstk has a blue spectrum (PR ∝
come significant, the dominant part of the integral comes
k 3 |Rk |2 ∝ k 2 ), and Rint
k is bluer still (PR ∝ k 4 ), in-
from near w ≈ −1. Using Eq. (5) and neglecting the sec-
consistent with the scale-invariant spectrum observed in
ond term as w approaches −1, we obtain
CMB.
Following [5], we consider the entropic mechanism [10] w∼−1
(σ,int) 
√ Z ∼−1 c2s 1 dw
for generating a scale-invariant spectrum. By introduc- Rk ≈ C2 k 3 −V
p
a 3H
 ,φ
3H 2
3(1 + w)3
ing two scalar fields φ1 and φ2 that undergo an ekpyrotic
√ c2 1 w∼−1
phase simultaneously, we have an extra degree of freedom 2H 2


≈ C2 k as3 −V

p 
,φ
3(1 + w)H 2
that can source the curvature perturbation. Fluctuations
in the fields are decomposed into an adiabatic mode δσ √ 1 X∼Xc
along their mean trajectory and an entropic mode δs per- ≈ C2 k √  , (18)
X
pendicular to it. Both fields obtain scale-invariant fluc-
tuations during the ekpyrotic phase, but they source the where we have neglected some finite factors that are al-
curvature perturbation R differently [12], through the most constant during the rapid decrease in w, and used
equation relation (3) since P (X) is still linear. The integral term
has grown exponentially compared to its value at the end
H of the ekpyrotic phase from Eq. (12),
Ṙ = Ṙ(σ) − (1 + c2s ) θ̇δs. (13)
σ̇  (σ,int)  r
R  Xek-end
 w∼−1 
The first term represents the adiabatic contribution R(σ)  (σ,int)  ≈ . (19)
R  Xc
to the curvature perturbation, which is the same as in ek-end

the single field case. The second term represents the Hence, from Eq. (16), the ratio of the integral term to the
entropic contribution R(s) , where θ is the angular di- scale-invariant entropic contribution can be expressed as
rection of the trajectory in the (φ1 , φ2 ) plane. It is as-  (σ,int)  r
sumed [5] that the trajectory is nearly straight except R  Xk Xek-end
 w∼−1 
at the end of the ekpyrotic phase, where it undergoes a ≈ . (20)
 R(s)  Xek-end Xc

sharp bend and renders θ̇ temporarily non-zero. Accord-
ing to Eq. (13), this causes the entropic perturbation δs This ratio determines whether the adiabatic contribu-
to convert almost instantaneously into a scale-invariant tion can catch up with the entropic contribution and

dominate the curvature perturbation. There is a com-
petition between two factors: (i) an exponential sup-
pression from (16) that depends on how much X in-
creases after horizon exit during the ekpyrotic phase,
i.e. Xk /Xek-end ≈ e−2Nk as in Eq. (11); and (ii) an ex-
ponential amplification from (19) that depends on how
much X decreases in the kinetic phase, i.e. Xek-end /Xc >
Xek-end /Xek-beg ≡ e2Ntot according to Eq. (4). Therefore
we find, in total,
4
ated with quantum gravity effects, as occurs naturally in
singular bounces. Hence, the simple non-singular bounce
may fail for the reasons described in this paper, but it re-
mains possible that a non-linear realization [14], or a sin-
gular bounce, can produce an observationally acceptable,
nearly scale-invariant spectrum of density perturbations.
We thank Justin Khoury, Andrew Tolley, and Neil
Turok for useful discussions. The work is supported
in part by Department of Energy Grant DE-FG02-
 (σ,int)  91ER40671.
R 
 w∼−1  > Ntot −2Nk
 e . (21)
 R(s)  ∼

The inequality boils down to the fact that the increase of
the kinetic energy X in the ekpyrotic phase is less than
the decrease in the kinetic phase, hence amplification
wins: This ratio is much greater than unity for modes
with Nk < Ntot /2, which includes all modes within our
observable horizon. That is, the integral adiabatic con-
tribution, which has a doubly blue spectrum and was
sub-subdominant at the end of the ekpyrotic phase, has
grown to overwhelm the scale-invariant entropic contri-
bution on relevant scales by the time the bouncing phase
begins.
Thus, the dominantly blue spectrum will be carried
into the expanding phase, in contradiction to current
observations. Moreover, as Eq. (21) dictates, for a
wide range of k modes the perturbation amplitudes have
grown non-linear, to the extent that perturbation theory
may break down even before the bounce.
As captured in Fig. 1, the problem with the non-
singular bounce arises from the generic requirement that
in the kinetic phase the kinetic energy density must de-
crease by more than it has increased in the ekpyrotic
phase, in order to trigger the ghost-condensate phase and
violate the NEC. Hence, it is difficult to avoid this prob-
lem in a ghost-condensate bouncing model with a canon-
ical (linear P (X)) ekpyrotic phase that does not involve
extreme fine-tuning. Although the hope had been that a
non-singular bounce would make ekpyrotic models sim-
pler than the singular case by not having to consider
quantum gravity effects, we are forced to conclude that
this scenario leads to the exponential growth of adiabatic
perturbations that spoil the spectrum of density pertur-
bations.
As a possible remedy, the growth of curvature per-
turbations may be moderated if the ekpyrotic phase is
realized with some non-linear form of P (X, φ), e.g. in
[13, 14]. This kind of non-linearity is typically associ-
[1] J. Khoury, B. A. Ovrut, P. J. Steinhardt and N. Turok,
Phys. Rev. D 64, 123522 (2001) [arXiv:hep-th/0103239].
[2] J. L. Lehners, Phys. Rept. 465, 223 (2008)
[arXiv:0806.1245 [astro-ph]].
[3] J. K. Erickson, D. H. Wesley, P. J. Steinhardt and
N. Turok, Phys. Rev. D 69, 063514 (2004) [arXiv:hep-
th/0312009].
[4] J. Khoury, B. A. Ovrut, N. Seiberg, P. J. Steinhardt and
N. Turok, Phys. Rev. D 65, 086007 (2002) [arXiv:hep-
th/0108187]. P. J. Steinhardt and N. Turok, Science 296,
1436 (2002).
[5] E. I. Buchbinder, J. Khoury and B. A. Ovrut, Phys. Rev.
D 76, 123503 (2007) [arXiv:hep-th/0702154].
[6] N. Arkani-Hamed, H. C. Cheng, M. A. Luty and S. Muko-
hyama, JHEP 0405, 074 (2004) [arXiv:hep-th/0312099].
[7] There has been some debate about whether the ghost
condensate mechanism can be implemented in a self-
consistent low-energy effective theory; see R. Kallosh,
J. U. Kang, A. D. Linde and V. Mukhanov, JCAP
0804, 018 (2008) [arXiv:0712.2040 [hep-th]]; S. Wein-
berg, Phys. Rev. D 77, 123541 (2008) [arXiv:0804.4291
[hep-th]]; and P. Creminelli, G. D’Amico, J. Norena and
F. Vernizzi, JCAP 0902, 018 (2009) [arXiv:0811.0827
[astro-ph]]. The problem described in this paper does not
depend on the specific form of NEC violation and, thus,
applies independent of this debate.
[8] P. Creminelli, M. A. Luty, A. Nicolis and L. Senatore,
JHEP 0612, 080 (2006) [arXiv:hep-th/0606090].
[9] B.K. Xue, in preparation.
[10] J. L. Lehners, P. McFadden, N. Turok and P. J. Stein-
hardt, Phys. Rev. D 76, 103501 (2007) [arXiv:hep-
th/0702153].
[11] J. Khoury and P. J. Steinhardt, arXiv:0910.2230 [hep-th].
[12] C. Gordon, D. Wands, B. A. Bassett and R. Maartens,
Phys. Rev. D 63, 023506 (2001) [arXiv:astro-
ph/0009131]. D. Langlois and S. Renaux-Petel, JCAP
0804, 017 (2008) [arXiv:0801.1085 [hep-th]].
[13] P. Creminelli and L. Senatore, JCAP 0711, 010 (2007)
[arXiv:hep-th/0702165].
[14] J. Khoury, P. J. Steinhardt, and B.K. Xue, forthcoming.