Particle cosmology

A. Riotto
CERN, Geneva, Switzerland

arXiv:1010.2642v3 [hep-ph] 17 Jan 2017

Abstract
In these lectures the present status of the so-called standard cosmological model,
based on the hot Big Bang theory and the inflationary paradigm is reviewed.
Special emphasis is given to the origin of the cosmological perturbations we
see today under the form of the cosmic microwave background anisotropies
and the large scale structure and to the dark matter.
1

Introduction

There are fundamental questions we are on the edge of answering: what is the origin of our universe?
Why is the universe so homogeneous and isotropic on large scales? What are the origins of dark matter
and dark energy? What is the fate of our universe? While these lectures will certainly not be able to give
definite answers to them, we shall try to provide the students with some tools they might find useful in
order to solve these overwhelming mysteries themselves.
These lectures will contain a short review of the standard Big Bang model; a rather long discussion of the inflation paradigm with particular emphasis on the possibility that the cosmological seeds
originated from a period of primordial acceleration; the physics of the Cosmic Microwave Background
(CMB) anisotropies, and a short discussion of the dark matter
Since these lectures were delivered at a school, we shall not provide an exhaustive list of references
to original material, but refer to several basic cosmology books and reviews where students can find the
references to the original material [111].
2

Basics of the Big Bang model

The key starting point of the description of our universe is the fact that the latter looks homogeneous
and isotropic on large scales. For such a reason the standard cosmology is based upon the maximally
spatially symmetric FriedmannRobertsonWalker (FRW) line element


dr2
2
2
2
2
2
2
2
ds = dt + a(t)
+ r (d + sin d ) ;
(1)
1 kr2
Here a(t) is the cosmic-scale factor; the curvature is parametrized by Rcurv a(t)|k|1/2 and the
parameter k can be chosen to acquire the values = 1, 0, 1 The coordinates are co-moving coordinates
and It is important to point out that physical separations between freely moving particles scale as a(t).
The momenta of freely propagating particles scale like as a(t)1 ,while wavelength of photons stretches
as a(t). Correspondingly, the redshift suffered by a photon emitted from a distant galaxy 1+z = a0 /a(t).
The evolution of the scale factor a(t) is governed by Einstein equations
R

1
R g G = 8G , T
2

(2)

where R (, = 0, 3) is the Riemann tensor and R is the Ricci scalar constructed via the metric
(1) [2, 3], and T is the energy-momentum tensor. G = m2
Pl is the Newton constant. Under the
hypothesis of homogeneity and isotropy, we can always write the energy-momentum tensor under the
form T = diag (, P, P, P ) where is the energy density of the system and P its pressure. They are
functions of time.

The evolution of the cosmic-scale factor is governed by the Friedmann equation

 2
a
8G
k
H
=
2,
a
3
a
2

(3)

where is the total energy density of the universe, matter, radiation, vacuum energy, and so on.
Differentiating wrt to time both members of Eq. (3) and using the the mass conservation equation
+ 3H( + P ) = 0 ,

(4)

we find the equation for the acceleration of the scale factor

a

4G
=
( + 3P ).
a
3

(5)

H = 4G ( + P ) .

(6)

Combining Eqs. (3) and (5) we find

The evolution of the energy density of the universe is governed by


d(a3 ) = P d a3 ;

(7)

which is the first law of thermodynamics for a fluid in the expanding universe. (In the case that the stress
energy of the universe is comprised of several, non-interacting components, this relation applies to each
separately; e.g., to the matter and radiation separately today.) For P = /3, ultra-relativistic matter,
2
1
a4 and a t 2 ; for P = 0, very nonrelativistic matter, a3 and a t 3 ; and for P = ,
vacuum energy, = const. If the rhs of the Friedmann equation is dominated by a fluid with equation of
state P = w, it follows that a3(1+w) and a t2/3(1+w) .
Friedmann equations relate the curvature of the universe to the energy density and expansion rate:
1=

k
a2 H 2

crit

(8)

and the critical density today crit = 3H 2 /8G = 1.88h2 g cm3 ' 1.05104 eV cm3 . The curvature
radius of the universe is related to the Hubble radius and by
Rcurv =

H 1
.
| 1|1/2

(9)

In physical terms, the curvature radius sets the scale for the size of spatial separations where the effects
of curved space become pronounced.
The energy content of the universe consists of matter and radiation (today, photons and neutrinos).
Since the photon temperature is accurately known, T0 = 2.73 0.01 K, the fraction of critical density
contributed by radiation is also accurately known: R h2 = 4.2 105 , where h = 0.72 0.07 is the
present Hubble rate in units of 100 km s1 Mpc1 [12]. The remaining content of the universe is another
matter. Rapid progress has been made recently toward the measurement of cosmological parameters [13].
We know by now that the universe is spatially flat; accelerating; comprised of one third dark matter and
two thirds a new form of dark energy [12]
0 = 1.00+0.07
0.03 ,
meaning that the present universe is spatially flat (or at least very close to being flat). Restricting to
0 = 1, the dark matter density is given by [12]

DM h2 = 0.11+0.0034
0.059 ,
and a baryon density
B = 0.045 0.0015,
while the Big Bang nucleosynthesis estimate is B h2 = 0.019 0.002. Substantial dark (unclustered)
energy is inferred:
DE 0.72 0.015 .
What is most relevant for us is that this universe was apparently born from a burst of rapid expansion,
inflation, during which quantum noise was stretched to astrophysical size seeding cosmic structure. This
is exactly the phenomenon we want to address in part of these lectures.
2.1

The early, radiation-dominated universe

In an expanding universe the energy density in matter decreases as a3 , and that in radiation as a4 at
early times the universe was radiation dominated.
Denoting the epoch of matter and radiation equality by subscript EQ, and using T0 = 2.73 K, it
follows that
TEQ = 5.62(0 h2 ) eV .
(10)
At early times, when the universe is radiation dominated, the expansion rate determined by the temperature of the universe and the number of relativistic degrees of freedom
rad = g (T )

2T 4
;
30

1/2

H ' 1.67g T 2 /mPl ;

1/2

t ' 2.42 106 g

a t1/2 ;

(T / GeV)2 s ;

(11)
(12)

where g (T ) counts the number of ultra-relativistic degrees of freedom ( the sum of the internal degrees
of freedom of particle species much less massive than the temperature) and mPl G1/2 = 1.22
1019 GeV is the Planck mass.
A quantity of importance related to g is the entropy density in relativistic particles,
s=

+p
2 2
=
g T 3 ,
T
45

and the entropy per co-moving volume,

S a3 s g a3 T 3 .
Since in thermal equilibrium the entropy per co-moving volume S remains constant, we get that
the temperature and scale factor are related by
1/3 1

T g

(13)

which for g = const leads to the familiar T a1 . Also, in our present Hubble volume in a very
physical way: by the entropy it contains,
SU =

4 3
H s ' 1090 ,
3 0

a huge number indeed, which will play a crucial role in the following.
3

(14)

2.2

The concept of particle horizon

In a FRW cosmology photons travel on geodesics satisfying the equation dr = dt/a(t). This means that
in a time t photons travel a distance
t

Z
RH (t) = a(t)
0

dt0
a(t0 )

t
H 1
=n
H 1
(1 n)
(1 n)

for a(t) tn , n < 1.

Using the conformal time d = dt/a, the particle horizon becomes

Z
RH (t) = a( )
d,

(15)

(16)

where 0 indicates the conformal time corresponding to t = 0. This quantity is very close to the Hubble
radius during radiation or matter periods. 1 .
RH

A physical length scale is within the horizon when the following condition is satisfied: if <
H 1 . Setting the length scale to be = = 2a/k, we shall have the following rule

k
aH
k
aH

 1 = SCALE OUTSIDE THE HORIZON

 1 = SCALE WITHIN THE HORIZON

The shortcomings of the standard Big Bang theory

The most accurate measurement of the temperature and spectrum is that by the WMAP5 instrument on
the COBE satellite which determined its temperature to be 2.7260.01 K [12]. The length corresponding
to our present Hubble radius (which is approximately the radius of our observable universe) at the time
of last scatteringwas




aLS
T0
H (tLS ) = RH (t0 )
= RH (t0 )
.
a0
TLS
On the other hand, during the matter-dominated period, the Hubble length decreased with a different law
H 2 M a3 T 3 .
At last-scattering
1
HLS


= RH (t0 )

TLS
T0

3/2
 RH (t0 ).

The length corresponding to our present Hubble radius was much larger that the horizon at that time.
This can be by shown comparing the volumes corresponding to these two scales
3H (TLS )
=
3
HLS

T0
TLS

 3

106 .

(17)

There were 106 casually disconnected regions within the volume that now corresponds to our horizon!
1

A word of caution: in inflationary models the horizon and Hubble radius are exponetially different .

 

  

  

  
   

 
!#"

$ %

log a

Fig. 1: The horizon scale (solid line) and a physical scale (dashed line) as function of the scale factor a

It is difficult to come up with a process other than an early hot and dense phase in the history of the
universe that would lead to a precise black body for a bath of photons which were causally disconnected
the last time they interacted with the surrounding plasma.
The horizon problem is well represented by Fig. 1 where the solid line indicates the horizon scale
and the dashed line any generic physical length scale . Suppose, indeed, that indicates the distance
between two photons we detect today. From Eq. (17) we discover that at the time of emission (lastscattering) the two photons could not talk to each other, the dashed line is above the solid line [2]. [2].
There is another main point to mention with the horiton problem and it is related to the inhomogeneities.
We know that the temperature anisotropy on the angular scale subtended by that length scale,
 
 
T

,
(18)
T

where the scale 100h1 Mpc(/deg) subtends an angle on the last-scattering surface. This is
known as the SachsWolfe effect [14, 15]. We shall come back to this piece of physics. The temperature
anisotropy is commonly expanded in spherical harmonics
X
T
(x0 , 0 , n) =
a`,m (x0 )Y`m (n),
T

(19)

`m

where x0 and 0 are our position and the preset time, respectively, n is the direction of observation, `0 s
are the different multipoles and2
ha`m a`0 m0 i = `,`0 m,m0 C` ,
(20)
2

An alternative definition is C` = h|a`m |2 i =

1
2`+1

P`

m=`

|a`m |2 .

Fig. 2: The CMBR anisotropy as function of ` (from Ref. [12])

where the deltas are due to the fact that the process that created the anisotropy is statistically isotropic.
The C` s are the so-called CMB power spectrum. For homogeneity and isotropy, the C` s are neither a
function of x0 , nor of m. The two-point correlation function is related to the Cl s in the following way
D T (n) T (n0 ) E
T

ha`m a`0 m0 iY`m (n)Y`0 m0 (n0 )

``0 mm0

X
`

C`

(n0 ) =
Y`m (n)Y`m

1 X
(2` + 1)C` P` ( = n n0 )
4

(21)

where we have used the addition theorem for the spherical harmonics, and P` is the Legendre polynom
of order `. In expression (21) the expectation value is an ensemble average. It can be regarded as an
average over the possible observer positions, but not in general as an average over the single sky we
observe, because of the cosmic variance3 . WMAP5 data are given in Fig. 2.
Let us now consider the last scatteringsurface. In co-moving coordinates the latter is far from us
a distance equal to
Z t0
Z 0
dt
=
d = (0 LS ) .
(22)
tLS a
LS
A given co-moving scale is therefore projected on the last scatteringsurface sky on an angular scale
'

,
(0 LS )

(23)

where we have neglected tiny curvature effects. Consider now that the scale
is of the order of the comoving sound horizon at the time of last-scattering, cs LS , where cs ' 1/ 3 is the sound velocity
at which photons propagate in the plasma at the last-scattering. This corresponds to an angle
' cs

LS
LS
' cs
,
(0 LS )
0

(24)

where the last passage has been performed knowing that 0  LS . Since the universe is matterdominated from the time of last scatteringonwards, the scale factor has the following behaviour: a
3
The usual hypothesis is that we observe a typical realization of the ensemble. This means that we expect the difference
between the observed values |a`m |2 and the ensemble averages C` to be of the order of the mean-square deviation of |a`m |2
from C` . The latter is called cosmic variance and, because we are dealing with a Gaussian distribution, it is equal to 2C` for
each multipole `. For a single `, averaging over the (2` + 1) values of m reduces the cosmic variance by a factor (2` + 1), but
it remains a serious limitation for low multipoles.

$ %

  

& ' (

  
   

     

)+* , .0/

1 2

!#"

end
inflation

log a

Fig. 3: The behaviour of a generic scale and the horizon scale H 1 in the standard inflationary model

T 1 t2/3 2 . The angle HOR subtended by the sound horizon on the last-scattering surface then
becomes


T0 1/2
HOR ' cs
1 ,
(25)
TLS
where we have used TLS ' 0.3 eV and T0 1013 GeV. This corresponds to a multipole `HOR
`HOR =

HOR

' 200 .

(26)

From these estimates we conclude that two photons which on the last scatteringsurface were separated
by an angle larger than HOR , corresponding to multipoles smaller than `HOR 200, were not in causal
contact. On the other hand, from Fig. 2 it is clear that small anisotropies, of the same order of magnitude
T /T 105 are present at `  200. We conclude that one of the striking features of the CMB
fluctuations is that they appear to be non-causal.
As can be seen in Fig. 1, in the standard cosmology the physical size of a perturbation, which
grows as the scale factor, begins larger than the horizon and, relatively late in the history of the universe, crosses inside the horizon. This precludes a causal microphysical explanation for the origin of the
required density perturbations.
From the considerations made so far, it appears that solving the horizon problem of the standard
Big Bang theory requires that the universe go through a primordial period during which the physical
scales evolve faster than the horizon scale H 1 .
If there is period during which physical length scales grow faster than H 1 , length scales which
are within the horizon today, < H 1 (such as the distance between two detected photons) and were
outside the horizon for some period, > H 1 (for instance at the time of last scatteringwhen the two
photons were emitted), had a chance to be within the horizon at some primordial epoch, < H 1 again,
7

see Fig. 3. If this happens, the homogeneity and the isotropy of the CMB can easily be explained: photons that we receive today and were emitted from the last scattering surface from causally disconnected
regions have the same temperature because they had a chance to talk to each other at some primordial
stage of the evolution of the universe.
The second condition can easily be expressed as a condition on the scale factor a. Since a given
scale scales like a and H 1 = a/a,
we need to impose that there is a period during which



=a
> 0.
H 1
We can therefore introduce the following rigorous definition: an inflationary stage is a period of the
universe during which the latter accelerates

INFLATION

a
> 0.

Comment: Let us stress that during such an accelerating phase the universe expands adiabatically.
This means that during inflation one can exploit the usual FRW equations (3) and (5). It must be clear
therefore that the non-adiabaticity condition is satisfied not during inflation, but during the phase transition between the end of inflation and the beginning of the radiation-dominated phase. At this transition
phase a large entropy is generated under the form of relativistic degrees of freedom: the Big Bang has
taken place.
4

The standard inflationary universe

Some parts which follow are taken from [3] and [5]. We thank E.W. Kolb fand A. Linde or granting
permission. From the previous section we have learned that an accelerating stage during the primordial
phases of the evolution of the universe might be able to solve the horizon problem. From Eq. (5) we
learn that
a
> 0 ( + 3P ) < 0 .
An accelerating period is obtainable only if the overall pressure p of the universe is negative: p < /3.
Neither a radiation-dominated phase nor a matter-dominated phase (for which p = /3 and p = 0,
respectively) satisfy such a condition. Let us postpone for the time being the problem of finding a
candidate able to provide the condition P < /3. For sure, inflation is a phase of the history of
the universe occurring before the era of nucleosynthesis (t 1 s, T 1 MeV) during which the light
elements abundances were formed. This is because nucleosynthesis is the earliest epoch from which we
have experimental data and they are in agreement with the predictions of the standard Big Bang theory.
However, the thermal history of the universe before the epoch of nucleosynthesis is unknown.
In order to study the properties of the period of inflation, we assume the extreme condition p =
which considerably simplifies the analysis. A period of the universe during which P = is called the
de Sitter stage. By inspecting Eqs. (3) and (4), we learn that during the de Sitter phase
=
HI

constant ,
constant ,

where we have indicated by HI the value of the Hubble rate during inflation. Correspondingly, solving
Eq. (3) gives
a = ai eHI (tti ) ,
(27)

where ti denotes the time at which inflation starts. Let us now see how such a period of exponential
expansion takes care of the shortcomings of the standard Big Bang Theory4 .
4.1

Inflation and the horizon problem

During the inflationary (de Sitter) epoch the horizon scale H 1 is constant. If inflation lasts long enough,
all the physical scales that have left the horizon during the radiation-dominated or matter-dominated
phase can re-enter the horizon in the past: this is because such scales are exponentially reduced. As we
have seen in the previous section, this explains both the problem of the homogeneity of CMB and the
initial condition problem of small cosmological perturbations. Once the physical length is within the
horizon, microphysics can act, the universe can be made approximately homogeneous and the primeval
inhomogeneities can be created.
Let us see how long inflation must be sustained in order to solve the horizon problem. Let ti and
tf be, respectively, the time of beginning and end of inflation. We can define the corresponding number
of e-foldings N
N = ln [HI (te ti )] .
(28)
A necessary condition to solve the horizon problem is that the largest scale we observe today, the present
horizon H01 , was reduced during inflation to a value H0 (ti ) smaller than the value of horizon length
HI1 during inflation. This gives



 
ati
1 atf
1
1 T0
H0 (ti ) = H0
eN <
= H0
HI ,
at0
atf
Tf
where we have neglected for simplicity the short period of matter-domination and we have called Tf the
temperature at the end of inflation (to be indentified with the reheating temperature TRH at the beginning
of the radiation-dominated phase after inflation, see later). We get
 
 
 
Tf
Tf
T0
>
N ln
ln
67 + ln
.
H0
HI
HI
Apart from the logarithmic dependence, we obtain N >
70.
4.2

A prediction of inflation

Since during inflation the Hubble rate is constant

1=

k
a2 H 2

1
.
a2

On the other hand it is easy to show that to reproduce a value of (0 1) of order of unity today, the
initial value of ( 1) at the beginning of the radiation-dominated phase must be | 1| 1060 .
Since we identify the beginning of the radiation-dominated phase with the beginning of inflation, we
require
| 1|t=tf 1060 .
During inflation
| 1|t=tf
| 1|t=ti


=

ai
af

2

= e2N .

(29)

Taking | 1|t=ti of order unity, it is enough to require that N 70. However, IF the period of inflation
lasts longer than 70 e-foldings the present-day value of 0 will be equal to unity with great precision.
One can say that a generic prediction of inflation is that
4

Despite the fact that the growth of the scale factor is exponential and the expansion is superluminal, this is not in contradiction with what is dictated by relativity. Indeed, it is the spacetime itself which is progating so fast and not a light signal in
it.

INFLATION

0 = 1.

This statement, however, must be taken cum grano salis and properly specified. Inflation does not change
the global geometric properties of the space-time. If the universe is open or closed, it will always remain
flat or closed, independently from inflation. What inflation does is to magnify the radius of curvature
Rcurv defined in Eq. (9) so that locally the universe is flat with a great precision. As we shall see, the
current data on the CMB anisotropies confirm this prediction.
4.3

Inflation and the inflaton

In the previous subsections we have described the various advantages of having a period of accelerating
phase. The latter required P < /3. Now, we would like to show that this condition can be attained by
means of a simple scalar field. We shall call this field the inflaton .
The action of the inflaton field reads


Z
Z
1
4
4

S = d x g L =
d x g
+ V () ,
2

where g = a3 for the FRW metric (1). From the EulerLagrange equations

( gL)
( gL)

= 0,

we obtain

2
+ 3H 2 + V 0 () = 0 ,
a

(30)

(31)

(32)

a scalar field
where V 0 () = (dV ()/d). Note, in particular, the appearance of the friction term 3H :
rolling down its potential suffers a friction due to the expansion of the universe.
We can write the energy momentum tensor of the scalar field
T = g L .
The corresponding energy density and pressure density P are
2
+ V () +
2
2
Tii = P =
V ()
2

T00 = =

()2
,
2a2
()2
.
6a2

Note that, if the gradient term were dominant, we would obtain P =

We can now split the inflaton field in

(33)
(34)

3 ,

not enough to drive inflation.

(t) = 0 (t) + (x, t) ,

where 0 is the classical (infinite wavelength) field, that is the expectation value of the inflaton field on
the initial isotropic and homogeneous state, while (x, t) represents the quantum fluctuations around
0 . In this section, we shall be concerned only with the evolution of the classical field 0 . The next
section will be devoted to the crucial issue of the evolution of quantum perturbations during inflation.
This separation is justified by the fact that quantum fluctuations are much smaller than the classical value
and therefore negligible when looking at the classical evolution. Not to be overwhelmed by the notation,
10

we shall indicate the classical value of the inflaton field by from now on. The energy momentum tensor
becomes
2
+ V ()
2
2
V ().
Tii = P =
2
T00 = =

(35)
(36)

If
V ()  2
we obtain the following condition
P ' .
From this simple calculation, we realize that a scalar field whose energy is dominant in the universe and
whose potential energy dominates over the kinetic term gives inflation. Inflation is driven by the vacuum
energy of the inflaton field.
4.4

Slow-roll conditions

Let us now quantify better under which circumstances a scalar field may give rise to a period of inflation.
The equation of motion of the field is
+ 3H + V 0 () = 0 .

(37)

If we require that 2  V (), the scalar field is slowly rolling down its potential. This is the reason why
such a period is called slow-roll. We may also expect that since the potential is flat, is negligible as
well. We shall assume that this is true and we shall quantify this condition soon. The FRW equation (3)
becomes
8G
H2 '
V (),
(38)
3
where we have assumed that the inflaton field dominates the energy density of the universe. The new
equation of motion becomes
3H = V 0 ()
(39)
which gives as a function of V 0 (). Using Eq. (39) slow-roll conditions then require
(V 0 )2
2  V () =
 H2
V
and
 3H = V 00  H 2 .
It is now useful to define the slow-roll parameters  and in the following way

 0 2
H
2
1
V
 = 2 = 4G 2 =
,
H
H
16G V
 00 
1
V
1 V 00
=
=
,
8G V
3 H2

= =
.
H

11

It might be useful to have the same parameters expressed in terms of conformal time

0 2
H0
=
4G
H2
H2
00

= =1
.
H0
 = 1

The parameter  quantifies how much the Hubble rate H changes with time during inflation. Notice that,
since
a

= H + H 2 = (1 ) H 2 ,
a
inflation can be attained only if  < 1:

INFLATION

 < 1.

As soon as this condition fails, inflation ends. In general, slow-roll inflation is attained if   1 and
||  1. During inflation the slow-roll parameters  and can be considered to be approximately
constant since the potential V () is very flat.
Comment: In the following, we shall work at first-order perturbation in the slow-roll parameters,
that is we shall
take only the first power of them. Since, using their definition, it is easy to see that
,
= O 2 , 2 , this amounts to saying that we shall treat the slow-roll parameters as constant in time.
Within these approximations, it is easy to compute the number of e-foldings between the beginning
and the end of inflation. If we indicate by i and f the values of the inflaton field at the beginning and
at the end of inflation, respectively, we find that the total number of e-foldings is
Z tf
N
H dt
ti
f

Z
' H

' 3H 2

Z f
i
f

Z
' 8G

d
V0
V
d .
V0

(40)

We may also compute the number of e-foldings N which are left to go to the end of inflation
Z

N ' 8G
f

V
d,
V0

(41)

where N is the value of the inflaton field when there are N e-foldings to the end of inflation.
1. Comment: According to the criterion given in Subsection 2.4, a given scale length = a/k
leaves the horizon when k = aHk where Hk is the the value of the Hubble rate at that time. One can
easily compute the rate of change of Hk2 as a function of k




dln Hk2
dln Hk2
dt
dln a
H
1
H
=
=2
1 = 2 2 = 2.
(42)
dln k
dt
dln a
dln k
H
H
H
12

2. Comment: Take a given physical scale today which crossed the horizon scale during inflation.
This happened when
 
 
af
T0
N

e
=
eN = HI1
a0
Tf
where N indicates the number of e-foldings from the time the scale crossed the horizon during inflation and the end of inflation. This relation gives a way to determine the number of e-foldings to the end
of inflation corresponding to a given scale
!




Tf

V 1/4
N ' 65 + ln
+ 2 ln
ln
.
3000 Mpc
1014 GeV
1010 GeV
Scales relevant for the CMB anisotropies correspond to N 60.
Inflation ended when the potential energy associated with the inflaton field became smaller than
the kinetic energy of the field. By that time, any pre-inflation entropy in the universe had been inflated
away, and the energy of the universe was entirely in the form of coherent oscillations of the inflaton
condensate around the minimum of its potential. The universe may be said to be frozen after the end of
inflation. We know that somehow the low-entropy cold universe dominated by the energy of coherent
motion of the field must be transformed into a high-entropy hot universe dominated by radiation. The
process by which the energy of the inflaton field is transferred from the inflaton field to radiation has
been dubbed reheating. In the theory of reheating, the simplest way to envisage this process is if the comoving energy density in the zero mode of the inflaton decays into normal particles, which then scatter
and thermalize to form a thermal background. It is usually assumed that the decay width of this process
is the same as the decay width of a free inflaton field.
Of particular interest is a quantity usually known as the reheat temperature, denoted as TRH 5 . The
reheat temperature is calculated by assuming an instantaneous conversion of the energy density in the
inflaton field into radiation when the decay width of the inflaton energy, , is equal to H, the expansion
rate of the universe.
The reheat temperature is calculated quite easily. After inflation the inflaton field executes coherent oscillations about the minimum of the potential. Averaged over several oscillations, the coherent
oscillation energy density redshifts as matter: a3 , where a is the RobertsonWalker scale factor.
If we denote as I and aI the total inflaton energy density and the scale factor at the initiation of coherent
oscillations, then the Hubble expansion rate as a function of a is
8 I  aI 3
H 2 (a) =
.
(43)
3 mPl 2 a
Equating H(a) and leads to an expression for aI /a. Now if we assume that all available coherent
energy density is instantaneously converted into radiation at this value of aI /a, we can find the reheat temperature by setting the coherent energy density, = I (aI /a)3 , equal to the radiation energy
4 , where g is the effective number of relativistic degrees of freedom at
density, R = ( 2 /30)g TRH

temperature TRH . The result is

1/4



p
90
200 1/4 p
mPl = 0.2
mPl .
(44)
TRH =
8 3 g
g

Inflation and the cosmological perturbations

In order for structure formation to occur via gravitational instability, there must have been small preexisting fluctuations on physical length scales when they crossed the Hubble radius in the radiationdominated and matter-dominated eras [7]. These fluctuations are given by inflation. Indeed, In an exponentially expanding universe the wavelenghts of all vacuum fluctuations of the inflaton field grow
5

So far, we have indicated it by Tf .

13

exponentially in the expanding universe. When the wavelength of any particular fluctuation becomes
greater than H 1 , this fluctuation stops propagating, and its amplitude freezes at some non-zero value
because of the large friction term 3H the equation of motion of the field . The amplitude of this
fluctuation then remains almost unchanged for a very long time, whereas its wavelength grows exponentially to cosmological scales.
Once inflation has ended, however, the Hubble radius increases faster than the scale factor, so
the fluctuations eventually re-enter the Hubble radius during the radiation- or matter-dominated eras and
provide the necessary seeds.
In summary, these are the key ingredients for understanding the observed structures in the universe
within the inflationary scenario:
Quantum fluctuations of the inflaton field are excited during inflation and stretched to cosmological
scales. At the same time, being the inflaton fluctuations connected to the metric perturbations
through Einsteins equations, ripples on the metric are also excited and stretched to cosmological
scales.
Gravity acts a messenger since it communicates the small seed perturbations to photons and
baryons once a given wavelength becomes smaller than the horizon scale after inflation.
Let us now see how quantum fluctuations are generated during inflation. we shall proceed by
steps. First, we shall consider the simplest problem of studying the quantum fluctuations of a generic
scalar field during inflation: we shall learn how perturbations evolve as a function of time and compute
their spectrum. Thensince a satisfactory description of the generation of quantum fluctuations has to
take both the inflaton and the metric perturbations into account we shall study the system composed
by quantum fluctuations of the inflaton field and quantum fluctuations of the metric.
6

Quantum fluctuations of a generic massless scalar field during inflation

Let us first see how the fluctuations of a generic scalar field , which is not the inflaton field, behave
during inflation. To warm up we first consider a de Sitter epoch during which the Hubble rate is constant.
6.1

Quantum fluctuations of a generic massless scalar field during a de Sitter stage

We assume this field to be massless. The massive case will be analysed in the next subsection.
Expanding the scalar field in Fourier modes
Z
d3 k
(x, t) =
eikx k (t),
(2)3/2
we can write the equation for the fluctuations as
k2
k = 0 .
a2
Let us study the qualitative behaviour of the solution to Eq. (45).

k + 3H k +

(45)

For wavelengths within the horizon,  H 1 , the corresponding wave-number satisfies the
relation k  a H. In this regime, we can neglect the friction term 3H k and Eq. (45) reduces to
k2
k = 0,
(46)
a2
which is basically the equation of motion of an harmonic oscillator. Of course, the frequency
term k 2 /a2 depends upon time because the scale factor a grows exponentially. On the qualitative
level, however, one expects that when the wavelength of the fluctuation is within the horizon, the
fluctuation oscillates.

k +

14

 For wavelengths above the horizon,  H 1 , the corresponding wave-number satisfies the relation k  aH and the term k 2 /a2 can be safely neglected. Equation (45) reduces to

k + 3H k = 0,

(47)

which tells us that on superhorizon scales k remains constant.

We have therefore the following picture: take a given fluctuation whose initial wavelength
a/k is within the horizon. The fluctuations oscillate till the wavelength becomes of the order of the
horizon scale. When the wavelength crosses the horizon, the fluctuation ceases to oscillate and gets
frozen in.
Let us now study the evolution of the fluctuation in a more quantitative way. To do so, we perform
the following redefinition
k
k =
a
and we work in conformal time d = dt/a. For the time being, we solve the problem for a pure de Sitter
expansion and we take the scale factor exponentially growing as a eHt ; the corresponding conformal
factor reads (after choosing properly the integration constants)
a( ) =

1
( < 0).
H

In the following we shall also solve the problem in the case of quasi de Sitter expansion. The beginning
of inflation coincides with some initial time i  0. We find that Eq. (45) becomes


a00
2
00
k + k
k = 0.
(48)
a
We obtain an equation which is very close to the equation for a KleinGordon scalar field in flat spacetime, the only difference being a negative time-dependent mass term a00 /a = 2/ 2 . Equation (48)
can be obtained from an action of the type




Z
1 02 1
a00
2
2
Sk =
d
(49)

k
k ,
2 k
2
a
which is the canonical action for a simple harmonic oscillator with canonical commutation relations
k k0 k k0 = i.
Let us study the behaviour of this equation on subhorizon and superhorizon scales. Since
k
= k ,
aH
on subhorizon scales k 2  a00 /a Equation (48) reduces to
k00 + k 2 k = 0 ,
whose solution is a plane wave
eik
(k  aH) .
(50)
k =
2k
We find again that fluctuations with wavelength within the horizon oscillate exactly like in flat spacetime. This does not come as a surprise. In the ultraviolet regime, that is for wavelengths much smaller
than the horizon scale, one expects that approximating the space-time as flat is a good approximation.

15

On superhorizon scales, k 2  a00 /a Equation (48) reduces to

k00

a00
k = 0,
a

which is satisfied by
k = B(k) a (k  aH)

(51)

where B(k) is a constant of integration. Roughly matching the (absolute values of the) solutions (50)
and (51) at k = aH (k = 1), we can determine the (absolute value of the) constant B(k)
1
H
1
.
|B(k)| a = = |B(k)| = =
2k
a 2k
2k 3
Going back to the original variable k , we obtain that the quantum fluctuation of the field on superhorizon scales is constant and approximately equal to
H
|k | '
(ON SUPERHORIZON SCALES)
2k 3
In fact we can do much better, since Eq. (48) has an exact solution:


eik
i

k =
1+
.
k
2k

(52)

This solution reproduces all that we have found by qualitative arguments in the two extreme regimes
k  aH and k  aH. We have performed the matching procedure to show that the latter can be very
useful to determine the behaviour of the solution on superhorizon scales when the exact solution is not
known.
6.2

The power spectrum

Let us define now the power spectrum, a useful quantity to characterize the properties of the perturbations. For a generic quantity g(x, t), which can expanded in Fourier space as
Z
d3 k
g(x, t) =
eikx gk (t),
(2)3/2
the power spectrum can be defined as
h0|gk 1 gk2 |0i (3) (k1 k2 )

2 2
Pg (k),
k3

where |0 i is the vacuum quantum state of the system. This definition leads to the usual relation
Z
dk
2
h0|g (x, t)|0i =
Pg (k).
k
6.3

(53)

(54)

Quantum fluctuations of a generic scalar field in a quasi de Sitter stage

So far, we have computed the time evolution and the spectrum of the quantum fluctuations of a generic
scalar field supposing that the scale factor evolves like in a pure de Sitter expansion, a( ) = 1/(H ).
However, during inflation the Hubble rate is not exactly constant, but changes with time as H =  H 2
(quasi de Sitter expansion). In this subsection, we shall solve for the perturbations in a quasi de Sitter
16

expansion. Using the definition of the conformal time, one can show that the scale factor for small values
of  becomes
1
1
a( ) =
.
H (1 )
The fluctuation mass-squared mass term is
M 2 ( ) = m2 a2

a00
,
a

where
a00
a

= a2

+ H2
a



= a2 H + 2 H 2

= a2 (2 ) H 2 =
'

(2 )
)2

2 (1

1
(2 + 3) .
2

(55)

Armed with these results, we may compute the variance of the perturbations of the generic field
Z
d3 k
2
h0| ((x, t)) |0i =
|k |2
3
(2)
Z
dk k 3
|k |2
=
k 2 2
Z
dk
=
P (k),
(56)
k
which defines the power spectrum of the fluctuations of the scalar field
P (k)

k3
|k |2 .
2 2

(57)

Since we have seen that fluctuations are (nearly) frozen in on superhorizon scales, a way of characterizing
the perturbations is to compute the spectrum on scales larger than the horizon. For a massive scalar field,
we obtain
 2 

H
k 32
P (k) =
,
(58)
2
aH
where, taking m2 /H 2 = 3 and expanding for small values of  and ,
'

3
+  .
2

(59)

We may also define the spectral index n of the fluctuations as

n 1 =

dln P
= 3 2 = 2 2.
dln k

The power spectrum of fluctuations of the scalar field is therefore nearly flat, that is is nearly independent of the wavelength = /k: the amplitude of the fluctuation on superhorizon scales does almost not
depend upon the time at which the fluctuation crosses the horizon and becomes frozen in. The small tilt
of the power spectrum arises from the fact that the scalar field is massive and because during inflation
the Hubble rate is not exactly constant, but nearly constant, where nearly is quantified by the slow-roll
17

parameters . Adopting the traditional terminology, we may say that the spectrum of perturbations is
blue if n > 1 (more power in the ultraviolet) and red if n < 1 (more power in the infrared). The
power spectrum of the perturbations of a generic scalar field generated during a period of slow-roll
inflation may be either blue or red. This depends upon the relative magnitude between and .
Comment: We might have computed the spectral index of the spectrum P (k) by first solving the
equation for the perturbations of the field in a di Sitter stage, with H = constant and therefore  = 0,
and then taking into account the time evolution of the Hubble rate introducing the subscript in Hk whose
time variation is determined by Eq. (42). Correspondingly, Hk is the value of the Hubble rate when a
given wavelength k 1 crosses the horizon (from that point on the fluctuation remains frozen in). The
power spectrum in such an approach would read

P (k) =

Hk
2

2 

k
aH

32
(60)

with 3 2 ' . Using Eq. (42), one finds

dln P
dln Hk2
=
+ 3 2 = 2 2
dln k
dln k
which reproduces our previous findings.
n 1 =

Comment: Since on superhorizon scales

H
k '
2k 3

k
aH

 




H
k
'
1 + ( ) ln
,
aH
2k 3

we discover that
| k | ' |H ( ) k |  |H k | ,

(61)

that is, on superhorizon scales the time variation of the perturbations can be safely neglected.
7

Quantum fluctuations during inflation

As we have mentioned in the previous section, the linear theory of the cosmological perturbations represents a cornerstone of modern cosmology and is used to describe the formation and evolution of structures
in the universe as well as the anisotropies of the CMB. The seeds for these inhomogeneities were generated during inflation and stretched over astronomical scales because of the rapid superluminal expansion
of the universe during the (quasi) de Sitter epoch.
In the previous section we have already seen that pertubations of a generic scalar field are
generated during a (quasi) de Sitter expansion. The inflaton field is a scalar field and, as such, we
conclude that inflaton fluctuations will be generated as well. However, the inflaton is special from the
point of view of perturbations. The reason is very simple. By assumption, the inflaton field dominates the
energy density of the universe during inflation. Any perturbation in the inflaton field means a perturbation
of the stress energy momentum tensor
= T .
A perturbation in the stress energy momentum tensor implies, through Einsteins equations of motion, a
perturbation of the metric

T = R


1
(g R) = 8GT = g .
2
18

On the other hand, a pertubation of the metric induces a back-reaction on the evolution of the inflaton
perturbation through the perturbed KleinGordon equation of the inflaton field
g



V

= +
= 0 = .

This logic chain makes us conclude that the perturbations of the inflaton field and of the metric are tightly
coupled to each other and have to be studied together

g .

As we shall see shortly, this relation is stronger than one might think because of the issue of gauge
invariance.
Before launching ourselves into the problem of finding the evolution of the quantum perturbations
of the inflaton field when they are coupled to gravity, let us give a heuristic explanation of why we expect
that during inflation such fluctuations are indeed present.
If we take Eq. (32) and split the inflaton field as its classical value 0 plus the quantum flucutation
, (x, t) = 0 (t) + (x, t), the quantum perturbation satisfies the equation of motion
+ 3H

2
+ V 00 = 0.
a2

(62)

Differentiating Eq. (37) wrt time and taking H constant (de Sitter expansion) we find
(0 ) + 3H 0 + V 00 0 = 0.

(63)

Let us consider for simplicity the limit k2 /a2  1 and let us disregard the gradient term. Under this
condition we see that 0 and solve the same equation. The solutions have therefore to be related to
each other by a constant of proportionality which depends upon time
= 0 t(x).

(64)

This tells us that (x, t) will have the form

(x, t) = 0 (x, t t(x)) .
This equation indicates that the inflaton field does not acquire the same value at a given time t in
all the space. On the contrary, when the inflaton field is rolling down its potential, it acquires different
values from one spatial point x to the next. The inflaton field is not homogeneous and fluctuations are
present. These fluctuations, in turn, will induce fluctuations in the metric.
7.1

The metric fluctuations

The mathematical tool to describe the linear evolution of the cosmological perturbations is obtained by
(0)
perturbing at the first order the FRW metric g , see Eq. (1)
g

(0)
g
(t) + g (x, t) ;

(0)
g  g
.

(65)

The metric perturbations can be decomposed according to their spin with respect to a local rotation of
the spatial coordinates on hypersurfaces of constant time. This leads to
19

 scalar perturbations
vector perturbations
tensor perturbations
Tensor perturbations or gravitational waves have spin 2 and are the true degrees of freedom of
the gravitational fields in the sense that they can exist even in the vacuum. Vector perturbations are
spin 1 modes arising from rotational velocity fields and are also called vorticity modes. Finally, scalar
perturbations have spin 0.
Let us do a simple exercise to count how many scalar degrees of freedom are present. Take a spacetime of dimensions D = n + 1, of which n coordinates are spatial coordinates. The symmetric metric
tensor g has 21 (n + 2)(n + 1) degrees of freedom. We can perform (n + 1) coordinate transformations
in order to eliminate (n+1) degrees of freedom, this leaves us with 12 n(n+1) degrees of freedom. These
1
2 n(n+1) degrees of freedom contain scalar, vector and tensor modes. According to Helmholtzs theorem
we can always decompose a vector ui (i = 1, , n) as ui = i v + vi , where v is a scalar (usually
called potential flow) which is curl-free, v[i,j] = 0, and vi is a real vector (usually called vorticity)
which is divergence-free, v = 0. This means that the real vector (vorticity) modes are (n 1).
Furthermore, a generic traceless tensor ij can always be decomposed as ij = Sij + Vij + Tij , where


kk
Sij = ki 2j + 13 ij , Vij = (i/2k) (ki j + kj i ) (ki i = 0) and ki Tij = 0. This means that
the true symmetric, traceless and transverse tensor degreees of freedom are 12 (n 2)(n + 1).
The number of scalar degrees of freedom is therefore
1
1
n(n + 1) (n 1) (n 2)(n + 1) = 2,
2
2
while the degrees of freedom of true vector modes are (n 1) and the number of degrees of freedom of
true tensor modes (gravitational waves) is 12 (n 2)(n + 1). In four dimensions n = 3, meaning that one
expects 2 scalar degrees of freedom, 2 vector degrees of freedom and 2 tensor degrees of freedom. As we
shall see, to the 2 scalar degrees of freedom from the metric, one has to add another one, the inflaton field
perturbation . However, since Einsteins equations will tell us that the two scalar degrees of freedom
from the metric are equal during inflation, we expect a total number of scalar degrees of freedom equal
to 2.
At the linear order, the scalar, vector, and tensor perturbations evolve independently (they decouple) and it is therefore possible to analyse them separately. Vector perturbations are not excited during
inflation because there are no rotational velocity fields during the inflationary stage. we shall analyse the
generation of tensor modes (gravitational waves) in the following. For the time being we want to focus
on the scalar degrees of freedom of the metric.
Considering only the scalar degrees of freedom of the perturbed metric, the most generic perturbed
metric reads
2

g = a

1 2
i B
i B
(1 2 ) ij + Dij E


,

(66)

while the line-element can be written as

Here Dij
7.2

ds2 = a2 (1 2 )d 2 + 2 i B d dxi + ((1 2 )ij + Dij E) dxi dxj .


= i j 13 ij 2 .

(67)

The issue of gauge invariance

When studying the cosmological density perturbations, what we are interested in is following the evolution of a space-time which is neither homogeneous nor isotropic. This is done by following the evolution
20

of the differences between the actual space-time and a well understood reference space-time. So we shall
consider small perturbations away from the homogeneous, isotropic space-time.
The reference system in our case is the spatially flat
FriedmannRobertsonWalker (FRW) spacetime, with line element ds2 = a2 ( ) d 2 ij dxi dxj . Now, the key issue is that general relativity
is a gauge theory where the gauge transformations are the generic coordinate transformations from one
local reference frame to another.
When we compute the perturbation of a given quantity, this is defined to be the difference between
the value that this quantity assumes on the real physical space-time and the value it assumes on the
unperturbed background. Nonetheless, to perform a comparison between these two values, it is necessary
to compute them at the same space-time point. Since the two values live on two different geometries,
it is necessary to specify a map which allows one to link univocally the same point on the two different
space-times. This correspondence is called a gauge choice and changing the map means performing a
gauge transformation.
Fixing a gauge in general relativity implies choosing a coordinate system. A choice of coordinates
defines a threading of space-time into lines (corresponding to fixed spatial coordinates x) and a slicing
into hypersurfaces (corresponding to fixed time ). A choice of coordinates is called a gauge and there
is no unique preferred gauge

GAUGE CHOICE

SLICING AND THREADING

From a more formal point of view, operating an infinitesimal gauge transformation on the coordinates
f = x + x
x

(68)

implies on a generic quantity Q a transformation on its perturbation

f = Q + x Q0
Q

(69)

where Q0 is the value assumed by the quantity Q on the background and x is the Lie-derivative of Q
along the vector x .
Decomposing in the usual manner the vector x
x0

0 (x ) ;

xi

i (x ) + v i (x ) ;

i v i = 0 ,

(70)

we can easily deduce the transformation law of a scalar quantity f (like the inflaton scalar field and
energy density ). Instead of applying the formal definition (69), we find the transformation law in
an alternative (and more pedagogical) way. We first write f (x) = f (x) f0 (x), where f0 (x) is the
f (x
f ) = fe(x
f ) fe0 (x
f ). Since f is a scalar
background value. Under a gauge transformation we have f
f ) = f (x ) (the value of the scalar function in a given physical point is the same in
we can write f (x
all the coordinate system). On the other side, on the unperturbed background hypersurface fe0 = f0 . We
have therefore

f (x
f ) = fe(x
f ) fe0 (x
f )
f
f )
= f (x ) f0 (x
 
f f0 (x
f )
= f x
21

f ) x
= f (x

f
f ),
(e
x) f0 (x
x
(71)

from which we finally deduce, being f0 = f0 (x0 ),

f = f f 0 0
f

For the spin-zero perturbations of the metric, we can proceed analogously. We use the following trick.
f2 .
f = x + x , the line element is left invariant, ds2 = ds
Upon a coordinate transformation x x


2
f0 ) '
f0 ) 1 +
f0
e dx
= a2 (x0 ) (1 + ) (dx0 )2 . Since a2 (x
This implies, for instance, that a2 (x
f0 = 1 + 00
dx0 + x0 dxi , we obtain 1 + 2 = 1 + 2
e + 2 H 0 + 2 00 . We
a2 (x0 ) + 2a a0 0 and dx
xi
now may introduce in detail some gauge-invariant quantities which play a major role in the computation
of the density perturbations. In the following we shall be interested only in the coordinate transformations
on constant time hypersurfaces and therefore gauge invariance will be equivalent to independence of the
slicing.
7.3

The co-moving curvature perturbation

The intrinsic spatial curvature on hypersurfaces on constant conformal time and for a flat universe is
given by
4 2
.
a2
The quantity is usually referred to as the curvature perturbation. We have seen, however, that the
curvature potential is not gauge invariant, but is defined only on a given slicing. Under a transformation
on constant time hypersurfaces t t + (change of the slicing)
(3)

R=

+ H .
We now consider the co-moving slicing which is defined to be the slicing orthogonal to the worldlines of
co-moving observers. The latter are are free-falling and the expansion defined by them is isotropic. In
practice, what this means is that there is no flux of energy measured by these observers, that is T0i = 0.
During inflation this means that these observers measure com = 0 since T0i goes like i (x, )0 ( ).
Since 0 for a transformation on constant time hypersurfaces, this means that
com = 0 = 0 = =

,
0

that is =
0 is the time-displacement needed to go from a generic slicing with generic to the
co-moving slicing where com = 0. At the same time the curvature perturbation transforms into
com = + H = + H
The quantity

R=+ H

=+H
0

22

.
0

is the co-moving curvature perturbation. This quantity is gauge invariant by construction and is related to
the gauge-dependent curvature perturbation on a generic slicing to the inflaton perturbation in that
gauge. By construction, the meaning of R is that it represents the gravitational potential on co-moving
hypersurfaces where = 0 or the inflaton fluctuation hypersurfaces where = 0:
R = |=0 = H



.
=0

The power spectrum of the curvature perturbation may then be easily computed
Rk = H

k
.

(72)

We may now compute the power spectrum of the co-moving curvature perturbation on superhorizon
scales

1
PR (k) =
2mPl 2 

H
2

2 

k
aH

nR 1

A2R

k
aH

nR 1

where we have defined the spectral index nR of the co-moving curvature perturbation as

nR 1 =

dln PR
= 3 2 = 2 6.
dln k

We conclude that inflation is responsible for the generation of adiabatic/curvature perturbations with
an almost scale-independent spectrum. To compute the spectral index of the spectrum PR (k) we have
proceeded as follows: first solve the equation for the perturbation k in a de Sitter stage, with H =
constant ( = = 0), whose solution is Eq. (52) and then taking into account the time-evolution of
the Hubble rate and of introducing the subscript in Hk and k . The time variation of the latter is
determined by
!


dln k
dln k
dt
dln a
k
1

=
=
1 = =  .
(73)

dln k
dt
dln a
dln k
H
k
Correspondingly, k is the value of the time derivative of the inflaton field when a given wavelength
k 1 crosses the horizon (from that point on the fluctuations remains frozen in). The curvature perturbation in such an approach would read
Hk
1
Rk '
k '

2
k

Hk2
k


.

Correspondingly
nR 1 =

dln Hk4 dln 2k

dln PR
=

= 4 + (2 2) = 2 6.
dln k
dln k
dln k

During inflation the curvature perturbation is generated on superhorizon scales with a spectrum
which is nearly scale invariant [16], that is, is nearly independent of the wavelength = /k: the
amplitude of the fluctuation on superhorizon scales does not (almost) depend upon the time at which the
23

fluctuation crosses the horizon and becomes frozen in. The small tilt of the power spectrum arises from
the fact that the inflaton field is massive, giving rise to a non-vanishing and because during inflation
the Hubble rate is not exactly constant, but nearly constant, where nearly is quantified by the slow-roll
parameters .
Comment: From what we have found so far, we may conclude that on superhorizon scales the
co-moving curvature perturbation R and the uniform-density gauge curvature satisfy on superhorizon
scales the relation
R k ' 0.
7.4

Gravitational waves

Quantum fluctuations in the gravitational fields are generated in a similar fashion to that of the scalar
perturbations discussed so far. A gravitational wave may be viewed as a ripple of space-time in the FRW
background metric (1) and in general the linear tensor perturbations may be written as


g = a2 ( ) d 2 + (ij + hij ) dxi dxj ,
where |hij |  1. The tensor hij has six degrees of freedom, but, as we studied in Subsection 7.1, the
tensor perturbations are traceless, ij hij = 0, and transverse i hij = 0 (i = 1, 2, 3). With these four
constraints, there remain two physical degrees of freedom, or polarizations, which are usually indicated
= +, . More precisely, we can write

hij = h+ e+
ij + h eij ,

where e+ and e are the polarization tensors which have the following properties
eij = eji , k i eij = 0, , eii = 0,
X
eij (k, ) = eij (k, ),
eij (k, )eij (k, ) = 4.

Notice also that the tensors hij are gauge-invariant and therefore represent physical degrees of freedom.
If the stress-energy momentum tensor is diagonal, as the one provided by the inflaton potential
T = g L, the tensor modes do not have any source in their equation and their action can
be written as
mPl 2
2

d4 x

1
g hij hij ,
2

that is the action of four independent massless scalar fields. The gauge-invariant tensor amplitude
1
vk = amPl hk ,
2
satisfies therefore the equation
vk00



a00
2
+ k
vk = 0,
a

which is the equation of motion of a massless scalar field in a quasi-de Sitter epoch. We can therefore
make use of the results present in Subsection 6.5 and Eq. (59) to conclude that on superhorizon scales
the tensor modes scale like
24


|vk | =

H
2



k
aH

 3 T
2

where
3
.
2
Since fluctuations are (nearly) frozen in on superhorizon scales, a way of characterizing the tensor perturbations is to compute the spectrum on scales larger than the horizon
T '

PT (k) =

k3
k3 X
2
|h
|
=
4

2
|vk |2 .
k
2 2
2 2

(74)

This gives the power spectrum on superhorizon scales

8
PT (k) =
mPl 2

H
2

2 

k
aH

nT

A2T

k
aH

nT

where we have defined the spectral index nT of the tensor perturbations as

nT =

dln PT
= 3 2T = 2.
dln k

The tensor perturbation is almost scale-invariant. Notice that the amplitude of the tensor modes depends
only on the value of the Hubble rate during inflation. This amounts to saying that it depends only on the
energy scale V 1/4 associated to the inflaton potential. A detection of gravitational waves from inflation
will therefore be a direct measurement of the energy scale associated to inflation.
7.5

The consistency relation

The results obtained so far for the scalar and tensor perturbations allow one to predict a consistency
relation which holds for the models of inflation addressed in these lectures, i.e., the models of inflation
driven by one-single field . We define the tensor-to-scalar amplitude ratio to be

r=

1
2
100 AT
4 2
25 AR

1
100 8

H
2 mPl

4
1
25 (2)

2

H
2 mPl

2 = .

This means that

r=

nT
2

One-single models of inflation predict that during inflation driven by a single scalar field, the ratio between the amplitude of the tensor modes and that of the curvature perturbations is equal to minus one-half
of the tilt of the spectrum of tensor modes. If this relation turns out to be falsified by the future measurements of the CMB anisotropies, this does not mean that inflation is wrong, but only that inflation has not
been driven by only one field.
25

7.6

From the inflationary seeds to the matter power spectrum

As the curvature perturbations enter the causal horizon during radiation- or matter-domination, they
create density fluctuations k via gravitational attractions of the potential wells. The density contrast
k = k can be deduced from the Poisson equation
k 2 k
k
3
k
= 4G k = 4G
= H2
a2

where is the background average energy density. This means that

k =

2
3

k
aH

2
k .

From this expression we can compute the power spectrum of matter density perturbations induced by
inflation when they re-enter the horizon during matter-domination:
2

P = h|k | i = A

k
aH

n

2 2
= 3
k

 2

 

k 4
2
k nR 1
2
AR
5
aH
aH

from which we deduce that matter perturbations scale linearly with the wave-number and have a scalar
tilt
n = nR = 1 + 2 6.
The primordial spectrum P is of course reprocessed by gravitational instabilities after the universe becomes matter-dominated. Indeed, as we have seen in Section 6, perturbations evolve after entering the horizon and the power spectrum will not remain constant. To see how the density contrast
is reprocessed we have first to analyse how it evolves on superhorizon scales before horizon-crossing.
We use the following trick. Consider a flat universe with average energy density . The corresponding
Hubble rate is
H2 =

8G
.
3

A small positive fluctuation will cause the universe to be closed:

H2 =

8G
k
( + ) 2 .
3
a

Substracting the two equations we find

3 k
=

8G a2

a2 RD
a MD

Notice that k a2 /k 2 (/)a2 /k 2 = constant for both RD and MD which confirms our
previous findings.
When the matter densities enter the horizon, they do not increase appreciably before matterdomination because before matter-domination pressure is too large and does not allow the matter inhomogeneities to grow. On the other hand, the suppression of growth due to radiation is restricted to
scales smaller than the horizon, while large-scale perturbations remain unaffected. This is why the horizon size at equality sets an important scale for structure growth:
kEQ = H 1 (aEQ ) ' 0.08 h Mpc1 .

26

Therefore, perturbations with k  kEQ are perturbations which have entered the horizon before matterdomination and have remained nearly constant till equality. This means that they are suppressed with
respect to those perturbations having k  kEQ by a factor (aENT /aEQ )2 = (kEQ /k)2 . If we define the
transfer function T (k) by the relation Rfinal = T (k) Rinitial we find therefore that roughly speaking

1
k  kEQ ,
T (k) =
(kEQ /k)2 k  kEQ .
The corresponding power spectrum will be
(

k
aH

k 3
aH

P (k)

k  kEQ ,
k  kEQ .

Of course, a more careful computation needs to include many other effects such as neutrino freestreaming, photon diffusion and the diffusion of baryons along with photons. It is encouraging, however,
that this rough estimate turns out to be confirmed by present data on large-scale structures [7].
The next step would be to investigate how the primordial perturbations generated by inflation flow
into the CMB to produce their anisotropies.
8

From inflation to large-angle CMB anisotropy

Temperature fluctuations in the CMB arise due to various distinct physical effects: first of all due to
our peculiar velocity with respect to the cosmic rest frame; fluctuations in the gravitational potential on
the last scattering surface appear also at the last scattering period; fluctuations are also intrinsic to the
radiation field itself on the last scattering surface. Finally, there is the contribution from the evolution of
the anisotropies from the last scattering surface till today (which we shall neglect from now on).
The second effect, known as the SachsWolfe effect is the dominat contribution to the anisotropy
on large-angular scales,  HOR 1 . The last three effects provide the dominant contributions to
the anisotropy on small-angular scales,  1 .
8.1

SachsWolfe plateau

We consider first the temperature fluctuations on large-angular scales that arise due to the SachsWolfe
effect. They provide a probe of the original spectrum of primeval fluctuations produced during inflation.
One can show that


T
T

xLS

t d2
dx.
a dx2

(75)

The photon trajectory is adx/dt = n. Using a t2/3 gives

Z
x(t) =
t

t0

dt0
=3
a

a0
t

t0
a


.

Integrating by parts Eq. (75), we finally find



T
T


=

1
[(xLS ) (0)] + e [v(0, t0 ) v(xLS , tLS )] .
3

The potential at our position contributes only to the unobservable monopole and can be dropped. On
scales outside the horizon, v = t 0. The remaining term is the SachsWolfe effect
T (e)
1
1
= (xLS ) = R(xLS ).
T
3
5
27

This relation has been obtained as follows. The co-moving curvature perturbation is given during the
radiation phase by R = +H/ = 1/3 / . Einstein equations set = and / = 2
on super-horizon scales. Therefore R = 5/3 beyond the horizon.
At large angular scales, the theory of cosmological perturbations predicts a remarkably simple
formula relating the CMB anisotropy to the curvature perturbation generated during inflation.
We have seen previously
P that the temperature anisotropy is commonly expanded in spherical
T
harmonics T (x0 , 0 , n) =
`m a`,m (x0 )Y`m (n), where x0 and 0 are our position and the preset
time, respectively, n is the direction of observation, `0 s are the different multipoles, and ha`m a`0 m0 i =
`,`0 m,m0 C` , where the deltas are due to the fact that the process that created the anisotropy is statistically isotropic. The C` s are the so-called CMB power spectrum. For homogeneity and isotropy, the C` s
are neither a function of x0 , nor of m. The two-point-correlation function is related to the Cl s according
to Eq. (21).
For adiabatic perturbations we have seen that on large scales, larger than the horizon on the last
scatteringsurface (corresponding to angles larger than HOR 1 ) T /T = 31 (xLS ). In Fourier transform
T (k, 0 , n)
1
= k ei kn(0 LS ) .
(76)
T
3
Using the decomposition
exp(i k n(0 LS )) =

X
(2` + 1)i` j` (k(0 LS ))P` (k n)

(77)

`=0

where j` is the spherical Bessel function of order ` and substituting, we get

D T (x , , n) T (x , , n0 ) E
0 0
0 0
=
T
T
Z
D T (x , , n) T (x , , n0 ) E
1
0 0
0 0
=
=
d3 x
V
T
T
Z
D T (k, , n)  T (k, , n0 )  E
1
0
0
3
=
=
d k
(2)3
T
T
Z

E X
D 1
1
3
2
=
d k
||
(2` + 1)(2`0 + 1)j` (k(0 LS ))
(2)3
3
`,`0 =0

j`0 (k(0 LS ))P` (k n)P`0 (k0 n0 )

(78)

(79)

4 P

0
0
Inserting P` (kn) = 2`+1
m Ylm (k)Y`m (n) and analogously for P` (k n ), integrating over the direcP
P
2`+1
(n)Y (n0 ). Using as well

0
0
tions dk generates ``0 mm0 m Y`m
`m
m Y`m (n)Y`m (n ) = 4 P` (nn ),
we get

D T (x , , n) T (x , , n0 ) E
0 0
0 0
T
T Z
2` + 1
dk D 1 2 E 3 2
0 2
= `
P` (n n )
|| k j` (k(0 LS )).
4

k 9

(80)

Comparing this expression with that for the C` , we get the expression for the C`AD , where the suffix
AD stands for adiabatic:
Z
E
2
dk D 1
AD
C` =
||2 k 3 j`2 (k(0 LS ))
(81)

k 9
which is valid for 2 `  (0 LS )/LS 100.
28

If we generically indicate by h|k |2 ik 3 = A2 (k0 )n1 , we can perform the integration and get
"
#
 n1
n)
2
( 3n
`(` + 1)C`AD
H0

2
)(` + 2 )
A
.
(82)
=
`(` + 1) 4n
5n
2
2
9
2
2 `+ 2
For n ' 1 and `  1, we can approximate this expression to
`(` + 1)ClAD
A2
=
.
(83)
2
9
This result shows that inflation predicts a very flat spectrum for low `. Furthermore, since inflation
predicts k = 35 Rk , we find that
`(` +
WMAP5 data imply that

1)ClAD

`(`+1)ClAD
2

8.2

V


A2
1
1
= R =
25
25 2 mPl 2 

H
2

2
.

(84)

' 1010 or
1/4

' 6.7 1016 GeV

Acoustic peaks

This part is heavily taken from [17]. We thank M. Zaldarriaga for granting permission. To be able to
calculate the power spectrum of the anisotropies even on angular scales larger than 1 , we need to consider the evolution of the photon anistoropies. As we already mentioned, before recombination Thomson
scattering was very efficient. As a result it is a good approximation to treat photons and baryons as a
single fluid. This treatment is called the tight-coupling approximation and will allow us to evolve the
perturbations until recombination.
The equation for the photon density perturbations for one Fourier mode of wave-number k is that
of a forced and damped harmonic oscillator
R
+ k 2 c2s = F,
(1 + R)
1
R
k 2 ],
F = 4[ +
(1 + R)
3
4

= kv + 4.
3
+

(85)

The photonbaryon fluid can sustain acoustic oscillations. The inertia is provided by the baryons, while
the pressure is provided by the photons. The sound speed is c2s = 1/3(1 + R), with R = 3b /4 =
31.5 (b h2 )(T /2.7)4 [(1 + z)/103 ]1 . As the baryon fraction goes down, the sound speed approaches
c2s 1/3. The third equation above is the continuity equation.
As a toy problem, we shall solve Eq. (85) under some simplifying assumptions. If we consider a matter-dominated universe, the driving force becomes a constant, F = 4/3k 2 , because the
gravitational potential remains constant in time. We neglect anisotropic stresses so that = , and,
furthermore, we neglect the time dependence of R. Equation (85) becomes that of a harmonic oscillator
that can be trivially solved. This is a very simplified picture, but it captures most of the relevant physics
we want to discuss.

29

To obtain the final solution we need again to specify the initial conditions. we shall restrict ourselves to adiabatic initial conditions, the most natural outcome of inflation. In our context this means
that initially = = 0 , = 8/30 , and v = 0. We have denoted 0 the initial amplitude of the
potential fluctuations. We shall take 0 to be a Gaussian random variable with power spectrum P0 .
We have made enough approximations that the evaluation of the sources in the integral solution
has become trivial. The solution for the density and velocity of the photon fluid at recombination is



0
+ |LS =
(1 + 3R) cos(kcs LS ) 0 R,
4
3
v |LS = 0 (1 + 3R)cs sin(kcs LS ).
(86)
Equation (86) is the solution for a single Fourier mode. All quantities have an additional spatial dependence (eikx ), which we have not included in order to make the notation more compact. With that
additional term the solution we have is
T
(n) = eikDLS cos S
T
3R
(1 + 3R)
[cos(kcs LS )
,
S = 0
3
(1 + 3R)
r
3
i
cos sin(kcs LS )],
1+R

(87)

where we have neglected the on the left-hand side because it is a constant. We have introduced
cos , the cosine of the angle between the direction of observation and the wavevector k; for example,
k x = kDLS cos . The term proportional to cos is the Doppler contribution.
Once the temperature perturbation produced by one Fourier mode has been calculated, we need
to expand it into spherical harmonics. The
P power spectrum of temperature anisotropies is expressed in
terms of the alm coefficients as CT ` = m |a`m |2 . The contribution to CT l from each Fourier mode is
weighted by the amplitude of primordial fluctuations in this mode, characterized by the power spectrum
of 0 = 3/5R, P0 = Ak 3 as dictated by inflation. In practice, fluctuations on angular scale ` receive
most of their contributions from wavevectors around k = `/DLS , so roughly the amplitude of the power
spectrum at multipole ` is given by the value of the sources in Eq. (86) at k .
After summing the contributions from all modes, the power spectrum is roughly given by
(1 + 3R)
cos(k cs LS ) R]2 +
3
(1 + 3R)2 2 2
cs sin (k cs LS )}.
3

`(` + 1)CT l A{[

(88)

Equation (88) can be used to understand the basic features in the CMB power spectra. The baryon drag
on the photonbaryon fluid reduces its sound speed below 1/3 and makes the monopole contribution
dominant (the one proportional to cos(k cs LS ). Thus, the CT l spectrum peaks where the monopole
term peaks, k cs LS = , 2, 3, , which correspond to `peak = nDLS /cS LS .
It is very important to understand the origin of the acoustic peaks. In this model the universe
is filled with standing waves; all modes of wave-number k are in phase, which leads to the oscillatory
terms. The sine and cosine in Eq. (88) originate in the time dependence of the modes. Each mode `
receives contributions preferentially from Fourier modes of a particular wavelength k (but pointing in
all directions), so to obtain peaks in C` , it is crucial that all modes of a given k be in phase. If this is not
the case, the features in the CT ` spectra will be blurred and can even disappear. This is what happens
when one considers the spectra produced by topological defects. The phase coherence of all modes of
30

a given wave-number can be traced to the fact that perturbations were produced very early on and had
wavelengths larger than the horizon during many expansion times.
There are additional physical effects we have neglected. The universe was radiation dominated
early on, and modes of wavelength smaller and bigger than the horizon at matter-radiation equality
behave differently. During the radiation era the perturbations in the photonbaryon fluid are the main
source for the gravitational potentials which decay once a mode enters into the horizon. The gravitational
potential decay acts as a driving force for the oscillator in Eq. (85), so a feedback loop is established. As
a result, the acoustic oscillations for modes that entered the horizon before matter-radiation equality have
a higher amplitude. In the CT ` spectrum the separation between modes that experience this feedback and
those that do not occurs at ` DLS /LS . Larger ` values receive their contributions from modes that
entered the horizon before matter-radiation equality. Finally, when a mode is inside the horizon during
the radiation era the gravitational potentials decay.
There is a competing effect, Silk damping, that reduces the amplitude of the large-l modes. The
photonbaryon fluid is not a perfect fluid. Photons have a finite mean free path and thus can random-walk
away from the peaks and valleys of the standing waves. Thus perturbations of wavelength comparable
to or smaller than the distance the photons can random-walk get damped. This effect can be modelled
1/2
by multiplying Eq. 87 by exp(k 2 /ks2 ), with ks1 LS (b h2 )1/2 . Silk damping is important for
multipoles of order `Silk ks DLS . Finally, the last scatteringsurface has a finite width. Perturbations
with wavelength comparable to this width get smeared out due to cancellations along the line of sight.
1 .
This effect introduces an additional damping with a characteristic scale kw
LS
The location of the first peak is by itself a measurement of the geometry of the universe. In fact,
photons propagating on geodesics from the last scattering surface to us feel the spatial geometry, whose

properties we learned are dictated by 0 . In fact, the location of the first peak is given by `1 ' 220/ 0 .
WMAP5 gives 0 = 1.00+0.07
0.03 . This tells us that the spatial (local) geometry of the universe is flat. This
is precisely what inflation predicts.
8.3

The polarization of the CMB anisotropies

The anisotropy field is characterized by a 2 2 intensity tensor Iij . For convenience, we normalize this
tensor so that it represents the fluctuations in units of the mean intensity (Iij = I/I0 ). The intensity
tensor is a function of direction on the sky, n, and two directions perpendicular to n that are used to define
its components (e1 ,e2 ). The Stokes parameters Q and U are defined as Q = (I11 I22 )/4 and U = I12 /2,
while the temperature anisotropy is given by T = (I11 + I22 )/4 (the factor of 4 relates fluctuations in the
intensity with those in the temperature,
I T 4 ). When representing polarization using rods in a map,
p
2
the magnitude is given by P = Q + U 2 , and the orientation makes an angle = 12 arctan(U/Q)
with e1 . In principle the fourth Stokes parameter V that describes circular polarization is needed, but
we ignore it because it cannot be generated through Thomson scattering, so the CMB is not expected
to be circularly polarized. While the temperature is invariant under a right-handed rotation in the plane
perpendicular to direction n, Q and U transform under rotation by an angle as
(Q iU )0 (n) = e2i (Q iU )(n),

(89)

where e01 = cos e1 + sin e2 and e02 = sin e1 + cos e2 . The quantities Q iU are said to be
spin 2.
We already mentioned that the statistical properties of the radiation field are usually described
in terms of the spherical harmonic decomposition of the maps. This basis, basically the Fourier basis,
is very natural because the statistical properties of anisotropies are rotationally invariant. The standard
spherical harmonics are not the appropriate basis for Q iU because they are spin-2 variables, but
generalizations (called 2 Ylm ) exist. We can expand
X
a2,`m 2 Y`m (n).
(90)
(Q iU )(n) =
`m

31

E<0

E>0

B<0

B>0

Fig. 4: Examples of E- and B-mode patterns of polarization

with respect to the spherical coordinate system (e , e ).

Here Q and U are defined at each direction n
To ensure that Q and U are real, the expansion coefficients must satisfy a2,`m = a2,`m . The equivalent relation for the temperature coefficients is aT,`m = aT,`m . Instead of a2,`m , it is convenient to
introduce their linear combinations aE,`m = (aP
2,`m + a2,`m )/2 and aB,`m = i(a
P2,`m a2,`m )/2.
We define two quantities in real space, E(n) = `,m aE,`m Y`m (n) and B(n) = `,m aB,`m Y`m (n).
Here E and B completely specify the linear polarization field.
The temperature is a scalar quantity under a rotation of the coordinate system, T 0 (n0 = Rn) =
T (n), where R is the rotation matrix. We denote with a prime the quantities in the transformed coordinate system. While Q iU are spin 2, E(n) and B(n) are invariant under rotations. Under parity,
however, E and B behave differently, E remains unchanged, while B changes sign.
To characterize the statistics of the CMB perturbations, only four power spectra are needed, those
for T , E, B and the cross correlation between T and E. The cross correlation between B and E or B
and T vanishes if there are no parity-violating interactions because B has the opposite
parity to T or
1 P

E. The power spectra are defined as the rotationally invariant quantities CT ` = 2`+1
m haT,`m aT,`m i,
P
P
P
1
1
1

CE` = 2`+1
m haE,`m aE,`m i, CB` = 2`+1
m haB,`m aB,`m i, and CC` = 2`+1
m haT,`m aE,`m i.
The brackets h i denote ensemble averages.
Polarization is generated by Thomson scattering between photons and electrons, which means that
polarization cannot be generated after recombination (except for re-ionization, which we shall discuss
later). But Thomson scattering is not enough. The radiation incident on the electrons must also be
anisotropic. In fact, its intensity needs to have a quadrupole moment. This requirement of having both
Thomson scattering and anisotropies is what makes polarization relatively small. After recombination,
anisotropies grow by free streaming, but there is no scattering to generate polarization. Before recombination there were so many scatterings that they erased any anisotropy present in the photonbaryon
fluid.
In the context of anisotropies induced by density perturbations, velocity gradients in the photon
baryon fluid are responsible for the quadrupole that generates polarization. Let us consider a scattering
occurring at position x0 : the scattered photons came from a distance of order the mean free path (T )
away from this point. If we are considering photons traveling in direction n
, they roughly come from x =
x0 + T n
. The photonbaryon fluid at that point was moving at velocity v(x) v(x0 ) + T n
i i v(x0 ).

32

HOT
Quadrupole
Anisotropy

Thomson
Scattering

COLD

Linear
Polarization

Fig. 5: Thomson scattering of radiation where quadrupole anisotropy generates linear polarization

Due to the Doppler effect the temperature seen by the scatterer at x0 is T (x0 , n
) = n
[v(x)v(x0 )]
T n
in
j i v j (x0 ), which is quadratic in n
(i.e., it has a quadrupole). Velocity gradients in the photon
baryon fluid lead to a quadrupole component of the intensity distribution, which, through Thomson
scattering, is converted into polarization.
The polarization of the scattered
radiation field, expressed in terms of the Stokes parameters Q
R
0 )2 T (
and U , is given by (Q + iU ) T d0 (m n
n0 ) p mi mj i vj |LS , where T is the Thomson
0 ) = 3/4T (m n
0 )2 , with
scattering cross-section and we have written the scattering matrix as P (m, n
0
1 + i
. As photons
m=e
e2 . In the last step, we integrated over all directions of the incident photons n
decouple from the baryons, their mean free path grows very rapidly, so a more careful analysis is needed
to obtain the final polarization:
(Q + iU )(
n) LS mi mj i vj |LS ,

(91)

where LS is the width of the last scattering surface and gives a measure of the distance that photons
travel between their last two scatterings, and  is a numerical constant that depends on the shape of the
visibility function. The appearance of mi mj in Eq. (91) ensures that (Q + iU ) transforms correctly
2 ).
under rotations of (
e1 , e
If we evaluate Eq. (91) for each Fourier mode and combine them to obtain the total power, we get
the equivalent of Eq. (88),
`(` + 1)CE` A2 (1 + 3R)2 (k LS )2 sin2 (k cs LS ),

(92)

where we are assuming n = 1 and that ` is large enough that factors like (` + 2)!/(` 2)! `4 .
The extra k in Eq. (92) originates in the gradient in Eq. (91). The large-angular scale polarization
is greatly suppressed by the kLS factor. Correlations over large angles can only be created by the
long-wavelength perturbations, but these cannot produce a large polarization signal because of the tight
coupling between photons and electrons prior to recombination. Multiple scatterings make the plasma
very homogeneous; only wavelengths that are small enough to produce anisotropies over the mean free
path of the photons will give rise to a significant quadrupole in the temperature distribution, and thus
to polarization. Wavelengths much smaller than the mean free path decay due to photon diffusion (Silk
damping) and so are unable to create a large quadrupole and polarization. As a result polarization peaks
at the scale of the mean free path.
On sub-degree angular scales, temperature, polarization, and the cross-correlation power spectra
show acoustic oscillations. In the polarization and cross-correlation spectra the peaks are much sharper.
33

The polarization is produced by velocity gradients of the photonbaryon fluid at the last scatteringsurface. The temperature receives contributions from density and velocity perturbations, and the oscillations
in each partially cancel one another, making the features in the temperature spectrum less sharp. The
dominant contribution to the temperature comes from the oscillations in the density [Eq. (86)], which
are out of phase with the velocity. This explains the difference in location between the temperature
and polarization peaks. The extra gradient in the polarization signal, Eq. (91), explains why its overall
amplitude peaks at a smaller angular scale.
Now, as photons travel in the metric perturbed by a GW [ds2 = a2 ( ) [d 2 +(ij +hTij )dxi dxj ]],
they get redshifted or blueshifted depending on their direction of propagation relative to the direction of
propagation of the GW and the polarization of the GW. For example, for a GW travelling along the z
axis, the frequency shift is given by
1 d
1 i j T () 1
= n
n
hij
= (1 cos2 )ei2 h t exp(ik x),
(93)
d
2
2
where (, ) describe the direction of propagation of the photon, the correspond to the different polarizations of the GW, and ht gives the time-dependent amplitude of the GW. During the matter-dominated
era, for example, ht = 3j1 (k )/k : time changes in the metric lead to frequency shifts (or equivalently
shifts in the temperature of the black body spectrum). Notice that the angular dependence of this frequency shift is quadrupolar in nature. As a result, the temperature fluctuations induced by this effect
as photons travel between successive scatterings before recombination produce a quadrupole intensity
distribution, which, through Thomson scattering, lead to polarization. Both E and B power spectra are
generated by GW. The current push to improve polarization measurements follows from the fact that
density perturbations, to linear order in perturbation theory, cannot create any B-type polarization. As a
rough rule of thumb, the amplitude of the peak in the B-mode power spectrum for GW is

[`(` + 1)CBl /2]1/2 = 0.024(V 1/4 /1016 GeV)2 K

where
V 1/4 ' 6.7 r1/4 1016 GeV

(94)

is the energy scale of inflation. A future experiment like CMBPol [18] can probe values of r as small
as 102 , corresponding to an inflation energy scale of about 2 1016 GeV. Furthermore, using the
consistency relation r =  valid in one-single field models of inflation, one deduces that
 r 1/2

'
,
mPl
102

(95)

meaning that a future measurement of the B-mode of CMB polarization will imply an inflaton excursus
of Planckian values. Therefore, A future measurement of the B-mode polarization of the CMB will
allow a determination of the value of the energy scale of inflation. This explains the utility of CMB
polarization measurements as probes of the physics of inflation. A detection of primordial B-mode
polarization would also demonstrate that inflation occurred at a very high energy scale, and that the
inflaton traversed a super-Planckian distance in field space.
8.4

Dark matter

Another topic we marginally touched in the lectures was dark matter, while we will not discuss the dark
energy puzzle we did not have time to cover in detail. This section is taken verbatim from Freese [19].
34

E-modes

nck
Pla

AP

WM

r=0.3
m
-2

IC

EP

r=0.01

B-modes
C
EPI

-LC

Fig. 6: E- and B-mode power spectra for a tensor-to-scalar ratio saturating the current bounds, r = 0.3 and for
r = 0.01. Shown are the experimental sensitivities of WMAP, Planck and two different realizations of CMBPol
(EPIC-LC and EPIC-2m)

We thank K. Freese for granting permission. The evidence that 95% of the mass of galaxies and clusters
is made of some unknown component of Dark Matter (DM) comes from (i) rotation curves (out to tens
of kpc), (ii) gravitational lensing (out to 200 kpc), and (iii) hot gas in clusters. They lead us to believe
that DM makes up about 30% of the entire energy of the universe.
In the 1970s, Ford and Rubin discovered that rotation curves of galaxies are flat. The velocities
of objects (stars or gas) orbiting the centres of galaxies, rather than decreasing as a function of the
distance from the galactic centres as had been expected, remain constant out to very large radii. Similar
observations of flat rotation curves have now been found for all galaxies studied, including our Milky
Way. The simplest explanation is that galaxies contain far more mass than can be explained by the bright
stellar objects residing in galactic disks. This mass provides the force to speed up the orbits. To explain
the data, galaxies must have enormous dark haloes made of unknown matter. Indeed, more than 95% of
the mass of galaxies consists of dark matter. The baryonic matter which accounts for the gas and disk
cannot alone explain the galactic rotation curve. However, adding a DM halo allows a good fit to data.
The limitations of rotation curves are that one can only look out as far as there is light or neutral
hydrogen (21 cm), namely to distances of tens of kpc. Thus one can see the beginnings of DM haloes,
but cannot trace where most of the DM is. The lensing experiments discussed in the next section go
beyond these limitations.
Einsteins theory of General Relativity predicts that mass bends, or lenses, light. This effect can
be used to gravitationally ascertain the existence of mass even when it emits no light. Lensing measurements confirm the existence of enormous quantities of DM both in galaxies and in clusters of galaxies.
Observations are made of distant bright objects such as galaxies or quasars. As the result of intervening
matter, the light from these distant objects is bent towards the regions of large mass. Hence there may
be multiple images of the distant objects, or, if these images cannot be individually resolved, the background object may appear brighter. Some of these images may be distorted or sheared. The Sloan Digital
Sky Survey used weak lensing (statistical studies of lensed galaxies) to conclude that galaxies, including
the Milky Way, are even larger and more massive than previously thought, and require even more DM
out to great distances. Again, the predominance of DM in galaxies is observed. The key success of
the lensing of DM to date is the evidence that DM is seen out to much larger distances than could be
probed by rotation curves: the DM is seen in galaxies out to 200 kpc from the centres of galaxies, in
agreement with N-body simulations. On even larger Mpc scales, there is evidence for DM in filaments
(the cosmic web). Another piece of gravitational evidence for DM is the hot gas in clusters. The X-ray
data indicates the presence of hot gas. The existence of this gas in the cluster can only be explained by

35

a large DM component that provides the potential well to hold on to the gas. In summary, the evidence
is overwhelming for the existence of an unknown component of DM that comprises 95% of the mass in
galaxies and clusters.
There is another basic reason why DM is necessary: to form structures as we observe them. Let us
assume that the matter content of the universe is dominated by a pressureless and self-gravitating fluid.
This approximation holds if we are dealing with the evolution of the perturbations in the DM component
or in case we are dealing with structures whose size is much larger than the typical Jeans scale length of
baryons. Let us also define x to be the co-moving coordinate and r = a(t)x the proper coordinate, a(t)
being the cosmic expansion factor. Furthermore, if v = r is the physical velocity, then v = ax
+ u,
gives the peculiar
where the first term describes the Hubble flow, while the second term, u = a(t)x,
velocity of a fluid element which moves in an expanding background.
In this case the equations that regulate the Newtonian description of the evolution of density perturbations are the continuity equation:

+ [(1 + )u] = 0 ,
t

(96)

which gives the mass conservation, the Euler equation

u
+ 2H(t)u + (u )u = 2 ,
t
a

(97)

which gives the relation between the acceleration of the fluid element and the gravitational force, and the
Poisson equation
2 = 4G
a2
(98)
which specifies the Newtonian nature of the gravitational force. In the above equations, is the gradient
computed with respect to the co-moving coordinate x, (x) describes the fluctuations of the gravitational
potential, and H(t) = a/a

is the Hubble parameter at the time t. Its time-dependence is given by

H(t) = E(t)H0 , where
E(z) = [(1 + z)3 m + (1 + z)2 (1 m DE ) + (1 + z)3(1+w) DE ]1/2 .

(99)

In the case of small perturbations, these equations can be linearized by neglecting all the terms
which are of second order in the fields and u. In this case, using the Euler equation to eliminate the
term u/t, and using the Poisson equation to eliminate 2 , one ends up with

2
+ 2H(t)
4G
= 0 .
t2
t

(100)

This equation describes the Jeans instability of a pressureless fluid, with the additional Hubble drag
term 2H(t)/t, which describes the counter-action of the expanding background on the perturbation
growth. Its effect is to prevent the exponential growth of the gravitational instability taking place in a
non-expanding background. The solution of the above equation can be cast in the form:
(x, t) = + (x, ti )D+ (t) + (x, ti )D (t) ,

(101)

where D+ and D describe the growing and decaying modes of the density perturbation, respectively.
In the case of an Einsteinde-Sitter (EdS) universe (m = 1, DE = 0), it is H(t) = 2/(3t), so that
D+ (t) = (t/ti )2/3 and D (t) = (t/ti )1 . The fact that D+ (t) a(t) for an EdS universe should not
be surprising. Indeed, the dynamical time-scale for the collapse of a perturbation of uniform density is
tdyn (G)1//2 , while the expansion time-scale for the EdS model is texp (G
)1//2 , where is the
mean cosmic density. Since for a linear (small) perturbation it is ' , then tdyn texp , thus showing
that the cosmic expansion and the perturbation evolution take place at the same pace. This argument
36

also leads to understanding the behaviour for a m < 1 model. In this case, the expansion time scale
becomes shorter than the above one at the redshift at which the universe recognizes that m < 1. This
1/3
happens at 1 + z ' m or at 1 + z ' 1
m in the presence or absence of a cosmological constant
term, respectively. Therefore, after this redshift, cosmic expansion takes place at a quicker pace than
gravitational instability, with the result that the perturbation growth is frozen.
The exact expression for the growing model of perturbations is given by
Z
5
1 + z0 0
D+ (z) = m E(z)
dz .
2
E(z 0 )3
z

(102)

The EdS has the faster evolution, while the slowing down of the perturbation growth is more apparent for
the open low-density model, the presence of a cosmological constant providing an intermediate degree of
evolution. The key point is, however, that a pressureless fluid such as DM is needed for the perturbations
to grow to give rise to collapsed objects. Baryon perturbations, being coupled to photons till the lastscattering epoch, feel a non-vanishing pressure and therefore they may not grow. After the last-scattering
stage, the baryons fall into the gravitational potential generated by DM and the baryonic perturbations
may promptly catch up with those of DM.
8.4.1

Dark matter candidates

There is a plethora of dark matter candidates. MACHOs, or Massive Compact Halo Objects, are made
of ordinary matter in the form of faint stars or stellar remnants; they could also be primordial black holes
or mirror matter. However, there are not enough of them to completely resolve the question. Of the
non-baryonic candidates, the most popular are the WIMPS (Weakly Interacting Massive Particles) and
the axions, as these particles have been proposed for other reasons in particle physics. Ordinary massive
neutrinos are too light to be cosmologically significant, though sterile neutrinos remain a possibility.
Other candidates include primordial black holes, non-thermal WIMPzillas, and KaluzaKlein particles
which arise in higher dimensional theories.
About axions, the good news is that cosmologists do not need to invent new particles. Two
candidates already exist in particle physics for other reasons: axions and WIMPs. Axions with masses
in the range 10(36) eV arise in the PecceiQuinn solution to the strong-CP problem in the theory of
strong interactions.
WIMPs are also natural dark matter candidates from particle physics. These particles, if present
in thermal abundances in the early universe, annihilate with one another so that a predictable number of
them remain today. The relic density of these particles comes out to be the right value:
DM h2 = (3 1026 cm3 /s)/hviA

(103)

where the annihilation cross-section hviA of weak interaction strength automatically gives the right
answer. The reason why the final abundance is inversely proportional to the annihilation cross-section
is rather clear: the larger the annihilation cross-section, the more WIMPs annihilate and the fewer of
them are left behind. Furthermore, annihilation is not eternal: owing to the expansion of the universe,
annihilation stops when its rate becomes smaller than the expansion rate of the universe. When this
happens, the abundance is said to freeze-out.
This coincidence is known as the WIMP miracle and is the reason why WIMPs are taken so
seriously as DM candidates. The best WIMP candidate is motivated by Supersymmetry (SUSY): the
lightest neutralino in the Minimal Supersymmetric Standard Model. Supersymmetry in particle theory
is designed to keep particle masses at the right value. As a consequence, each particle we know has a
partner: the photino is the partner of the photon, the squark is the quarks partner, and the selectron is the
partner of the electron. The lightest superysmmetric partner is a good dark matter candidate.

37

0.01
0.001
0.0001

10

100

1000

Fig. 7: The abundance of WIMPs of a given mass m as a function of temperature and for various annihilation
cross-sections

There are several ways to search for dark WIMPs. SUSY particles may be discovered at the LHC
as missing energy in an event. In that case one knows that the particles live long enough to escape the
detector, but it will still be unclear whether they are long-lived enough to be the dark matter. Thus complementary astrophysical experiments are needed. In direct detection experiments, the WIMP scatters off
a nucleus in the detector, and a number of experimental signatures of the interaction can be detected. In
indirect detection experiments, neutrinos that arise as annihilation products of captured WIMPs exit from
the Sun and can be detected on Earth. Another way to detect WIMPs is to look for anomalous cosmic
rays from the Galactic Halo: WIMPs in the Halo can annihilate with one another to give rise to antiprotons, positrons, or neutrinos. In addition, neutrinos, gamma rays, and radio waves may be detected as
WIMP annihilation products from the Galactic Centre. For lack of time these issues were not discussed
extensively in the lectures. The interested reader may find more about these issues in Ref. [20].
9

Conclusions

The period when we say that cosmology is entering a golden age has already passed: cosmology is in the
middle of its golden age. Present observational data pose various puzzles whose solutions might either be
around the corner or decades far in the future. It will require some young and creative researcher sitting
in this room to solve them. This is why the cosmological puzzles are dark, but the future is brighter.

Acknowledgements
It is a great pleasure to thank all the organizers, N. Ellis, E. Lillistol, D. Metral, and especially M. Losada
and E. Nardi, for having created such a stimulating atmosphere. All students are also acknowledged for
their never-ending enthusiasm. This version contains hopefully the right citations to other lectures notes
by K. Freese, E.W. Kolb, A. Linde, M. Turner and M. Zaldarriaga which the author used verbatim. We
apologize with them for this admittedly late note.

38

References
[1] A. D. Linde, Particle Physics and Inflationary Cosmology (Harwood, Chur, Switzerland, 1990) and
references therein.
[2] E. W. Kolb and M. S. Turner, The Early Universe (Addison-Wesley, Redwood City, CA, 1990).
[3] See http://cds.cern.ch/record/316856/files/9612138.pdf.
[4] Michael Turners lectures, p353, of Quantitative Particle Physics: Cargese 1992.
[5] . A. Linde in Physical cosmology : 25th anniversary of the cosmic background radiation discovery,
Gif-sur-Yvette : Ed. Frontieres, 1991.
[6] A. R. Liddle and D. H. Lyth, Phys. Rep. 231, 1 (1993) and references therein.
[7] A. R. Liddle and D. H. Lyth, Cosmological Inflation and Large-Scale Structure (Cambridge University Press, 2000) and references therein.
[8] D. H. Lyth and A. Riotto, Phys. Rep. 314, 1 (1999) and references therein.
[9] A. Riotto, arXiv:hep-ph/0210162 and references therein.
[10] For a review, see N. Bartolo, S. Matarrese, and A. Riotto, arXiv:astro-ph/0703496 and references
therein.
[11] E. J. Copeland, M. Sami, and S. Tsujikawa, Int. J. Mod. Phys. D 15, 1753 (2006) and references
therein.
[12] E. Komatsu et al. [WMAP Collaboration], Astrophys. J. Suppl. 180, 330 (2009) [arXiv:0803.0547
[astro-ph]].
[13] For a review, see, for instance, N. A. Bahcall, J. P. Ostriker, S. Perlmutter, and P. J. Steinhardt,
Science 284, 1481 (1999).
[14] R. K. Sachs and A. M. Wolfe, Astrophys. J. 147, 73 (1967).
[15] S. Dodelson, Modern Cosmology (Academic Press, New York, 2003).
[16] V. F. Mukhanov, H. A. Feldman, and R. H. Brandenberger, Phys. Rep. 215, 203 (1992) and references therein.
[17] M. Zaldarriaga, astro-ph/0305272.
[18] S. Dodelson et al., CMBPol Science White Paper submitted to the US Astro2010 Decadal Survey,
arXiv:0902.3796v1.
[19] K. Freese, EAS Publ. Ser. 36, 113 (2009) doi:10.1051/eas/0936016 [arXiv:0812.4005 [astro-ph]].
[20] G. Bertone, D. Hooper, and J. Silk, Phys. Rep. 405, 279 (2005).

39