Advanced Cosmology

Daniel Baumann

Department of Applied Mathematics and Theoretical Physics,

Cambridge University, Cambridge, CB3 0WA, UK

and

Institute of Theoretical Physics, University of Amsterdam,

Science Park, 1090 GL Amsterdam, The Netherlands
Contents

1 Preliminaries 3
1.1 Homogeneous Cosmology 3
1.2 Slow-Roll Inflation 7

I Primordial Seeds 12

2 Cosmological Perturbation Theory 12

2.1 Metric Perturbations 12
2.2 Matter Perturbations 13
2.3 Equations of Motion 13
2.4 Initial Conditions 19

3 Quantum Initial Conditions 23

3.1 Inflaton Fluctuations: Classical 23
3.2 Quantum Harmonic Oscillators 26
3.3 Inflaton Fluctuations: Quantum 28
3.4 Curvature Perturbations 32
3.5 Gravitational Waves 33

II CMB Anisotropies 35

4 The Boltzmann Equation 35

4.1 Free-Streaming 36
4.2 Thomson Scattering 39
4.3 Boltzmann for Photons 41
4.4 Boltzmann for Neutrinos 45
4.5 Einstein Revisited 45

5 Temperature Anisotropies 46
5.1 Anisotropies from Inhomogeneities 47
5.2 Free-Streaming 49
5.3 Projection 51
5.4 Evolution 55
5.4.1 Photon-Baryon Fluid 55
5.4.2 Acoustic Oscillations 57
5.4.3 Diffusion Damping 61
5.5 Power Spectrum 63
5.6 Cosmological Parameters 64

III Problems and Solutions 68

1
6 Problem Sets 68
6.1 Problem Set 0: Preliminaries 68
6.2 Problem Set 1: Perturbation Theory 70
6.3 Problem Set 2: Quantum Fluctuations 72
6.4 Problem Set 3: Boltzmann Equation 73
6.5 Problem Set 4: Temperature Anisotropies 74

7 Solutions 76
7.1 Solutions 0: Preliminaries 76
7.2 Solutions 1: Perturbation Theory 78
7.3 Solutions 2: Quantum Fluctuations 84
7.4 Solutions 3: Boltzmann Equation 87
7.5 Solutions 4: Temperature Anisotropies 91

References 95

2
1 Preliminaries

We will begin with a lightning review of some elementary concepts in cosmology. I will assume
that you have seen most, if not all, of this material before, so I will cite many results without
detailed derivations. Further details can be found in my Cosmology course [1] or any of the
standard textbooks (e.g. [2, 3]).

1.1 Homogeneous Cosmology

The FRW metric of a homogenous and isotropic spacetime is

ds2 = dt2 a2 (t)ij dxi dxj , (1.1)

where ij denotes the metric of a maximally symmetric 3-space and a(t) is the scale factor.
Throughout these lectures, we will restrict to the special case of flat spatial slices, i.e. ij = ij ,
and define dx2 ij dxi dxj . We will also introduce conformal time, d = dt/a(t), so that the
metric becomes
ds2 = a2 () d 2 dx2 .


(1.2)
We will first discuss the kinematics of particles in an FRW spacetime for an arbitrary scale
factor a(). After that, we will show how the Einstein equations determine a() in terms of the
stress-energy content of the universe.

Kinematics
The kinematics of particles in the FRW spacetime follows from the geodesic equation
 
P
P P = P + P = 0, (1.3)
x

where P dx /d is the 4-momentum of the particle. In an expanding spacetime, it is conve-

nient to write the components of the 4-momentum as

P = a1 [E, p] . (1.4)

For massless particles, such as photons, we have the constraint g P P = E 2 |p|2 = 0, so we

can write p = E p, where p is a unit vector in the direction of propagation.

Exercise.Show that the non-zero connection coefficients associated with the metric (1.2) are

000 = H , 0ij = Hij , ij0 = Hji , (1.5)

where H a/a is the conformal Hubble parameter.

The = 0 component of the geodesic equation (1.3) becomes

dP 0
P0 = 0 P P . (1.6)
d

3
Substituting (1.4) and (1.5), we get

d 1
(a1 E) (a E) = Ha2 E 2 Ha2 |p|2 , (1.7)
d
which simplifies to
1 dE 1 da
= E a1 . (1.8)
E d a d
This describes the redshift of the photon energy in an expanding spacetime.

Dynamics
The evolution of the scale factor is determined by the Friedmann equations

3H2 = 8Ga2 , (1.9)

2H + H2 = 8Ga2 P , (1.10)

where and P are the background density and pressure, respectively.

Exercise.By substituting (1.5) into

R + , (1.11)

show that
R00 = 3H , Rij = (H + 2H2 )ij a2 R = 6(H + H2 ) . (1.12)
Hence, show that the non-zero components of the Einstein tensor, G R 21 Rg , are

G00 = 3H2 , Gij = (2H + H2 )ij . (1.13)

Use this to confirm that the 00-Einstein equation, G00 = 8G T00 , implies (1.9) and the ij-Einstein
equation, Gij = 8G Tij , leads to (1.10).

Combining (1.9) and (1.10), we can write an evolution equation for the density

= 3H( + P ) . (1.14)

For pressureless matter (Pm 0) this implies m a3 , while for radiation (Pr = 31 r ) we have
r a4 .

Exercise.Derive the continuity equation (1.14) from T = 0. By integrating the Friedmann

equation (1.9) for matter and radiation show that
(
2 matter
a() = . (1.15)
radiation

4
The Cosmic Inventory
The universe is filled with several different species of particles:

baryons (b)
z }| {
photons () neutrinos () electrons (e) protons (p) cold dark matter (c) .
| {z } | {z }
radiation (r) matter (m)

The number density, energy density and pressure of each species a can be written as

d3 p
Z
na = ga fa (x, p) , (1.16)
(2)3
d3 p
Z
a = ga fa (x, p)E(p) , (1.17)
(2)3
d3 p p2
Z
Pa = ga 3
fa (x, p) , (1.18)
(2) 3E(p)

where fa (x, p) is the (phase space) distribution function of the species a, and ga is the number of
internal degrees of freedom. In the unperturbed universe, the distribution functions should not
depend in the position and the direction of the momentum, i.e. fa (x, p) fa (E(p)).
At early times, particle interactions were efficient enough to keep the different species in local
equilibrium. They then shared a common temperature T and the distribution functions take the
following maximum entropy form
1
fa (E) = , (1.19)
e(Ea a )/T 1
with + for bosons and for fermions. The chemical potential vanishes for photons and is
(likely) small for all other species. We will henceforth set it to zero. When the temperature
drops below the mass of a particle species, T  ma , they become non-relativistic and their
distribution function receives an exponential (Boltzmann) suppression, fa ema /T . This means
that relativistic particles (radiation) dominate the density and pressure of the primordial plasma.
By performing the integrals (1.17) and (1.18) in the limit E p, one finds
(
2 1 bosons 1
a = ga T 4 7 and Pa = a . (1.20)
30 fermions 3
8

The total radiation density is

2 X 7 X
r = g T 4 , where g ga + ga . (1.21)
30 8
a=b a=f

If equilibrium had persisted until today, all species with masses greater than 103 eV would be
exponentially suppressed. This would not be a very interesting world. Fortunately, many massive
particle species are weakly interacting and decoupled from the primordial plasma at early times.
The freeze-out abundance of dark matter has shaped the large-scale structure of the universe.

5
 relativistic non-relativistic

freeze-out

relic density

equilibrium

1 10 100

Figure 1. A schematic illustration of particle freeze-out. At high temperatures, the particle abundance
tracks its equilibrium value. At low temperatures, the particles freeze out and maintain a density that is
much larger than the Boltzmann-suppressed equilibrium abundance.

A Brief Thermal History

The key to understanding the thermal history of the universe is understanding the competition
between the interaction rate of particles, , and the expansion rate, H. Particles maintain
equilibrium as long as  H and freeze out when . H (see Fig. 1).
Neutrinos are the most weakly interacting particles of the Standard Model and therefore
decoupled first (around 0.8 MeV or 1 sec after the Big Bang). Shortly after neutrino decoupling,
electrons and positrons annihilated. The energies of the electrons and positrons got transferred
to the photons, but not the neutrinos. The temperature of the photons is therefore slightly
bigger today than that of the neutrinos. At around the same time, neutron-proton interactions
became inefficient, leading to a relic abundance of neutrons. These neutrons were essential for the
formation of the light elements during Big Bang nucleosynthesis (BBN), which occurred around
3 minutes after the Big Bang.
Below about 1 eV, or 380,000 years after the Big Bang, the temperature had become low
enough for neutral hydrogen to form through the reaction e + p+ H + . This is the moment
of recombination. At this point the density of free electrons dropped dramatically (see Fig. 2).
Before recombination the strongest coupling between the photons and the rest of the plasma was
through Thomson scattering, e + e + . The sharp drop in the free electron density
after recombination means that this process became inefficient and the photons decoupled. After
decoupling the photons streamed freely through the universe and are today observed as the
cosmic microwave background (CMB). The CMB is an almost perfect blackbody with an average
temperature of 2.7 K. At the level of one part in 104 , the CMB temperature varies across the sky,
which reflects density perturbations in the primordial plasma. These perturbations are believed
to have been created by quantum fluctuations during inflation (see Section 3).

6
 recombination

decoupling
CMB

plasma neutral hydrogen

Figure 2. Fraction of free electrons, Xe ne /nb , as a function of redshift (or temperature).

1.2 Slow-Roll Inflation

A key fact about the universe is that on large scales it is well-described by the Robertson-Walker
metric (1.2). But why? The standard cosmology predicts that the early universe was made of
many causally disconnected regions of space. The fact that these apparently disjoint patches of
space have very nearly the same densities and temperatures is called the horizon problem. In
this section, I will explain how inflationan early period of accelerated expansiondrives the
primordial universe towards homogeneity and isotropy, even if it starts in a more generic initial
state.

Horizon Problem
The particle horizon is the maximal distance that a signal can travel between some initial time
t = 0 and a later time t. In physical coordinates this is given by
Z t Z a
dt d ln a
D(t) = a(t) = a(t) . (1.22)
0 a(t) 0 a0

For a00 < 0, the integral converges to a finite value:

t2/3 3t matter
e.g. a(t) D(t) = (1.23)
t1/2 2t radiation

This leads to a puzzle: because the age of the universe (t0 ) is much larger than the time of
recombination (t ), the CMB naively consists of many causally disconnected patches (see Fig. 3).
Why is the CMB so homogeneous? Moreover, why are the observed CMB fluctuations correlated
on large scales and not just random noise?

7
 recombination

singularity

Figure 3. Illustration of the horizon problem in the conventional Big Bang model. All events that we
currently observe are on our past light cone. The intersection of our past light cone with the spacelike
slice labelled recombination corresponds to two opposite points in the observed CMB. Their past light
cones dont overlap before they hit the singularity, ai = 0, so the points appear never to have been in
causal contact. The same applies to any two points in the CMB that are separated by more than 1 degree
on the sky.

BIG BANG

recombination
end of inflation

INFLATION

causal
contact
singularity

Figure 4. Illustration of the inflationary solution to the horizon problem in comoving coordinate (using
conformal time on the vertical axis). The spacelike singularity of the standard Big Bang is replaced by
the reheating surface, i.e. rather than marking the beginning of time it now corresponds simply to the
transition from the end of inflation to the standard Big Bang evolution. All points in the CMB have
overlapping past light cones and therefore originated from a causally connected region of space.

Inflation
For a00 > 0, the particle horizon diverges in the past! A period of accelerated expansion (=infla-
tion) therefore solves the horizon problem (see Fig. 4). Signals were able to travel a much larger

8
distance than suggested by the naive extrapolation of the standard FRW expansion.

Exercise.A special case of accelerated expansion is the quasi-de Sitter limit, which is characterized
by a nearly constant expansion rate, H = a0 /a const., so that a(t) = eH(tt0 ) , where t0 is some
fiducial time at which a(t0 ) 1. Show that
1
a() = , (1.24)
H

for < 0. Notice that the initial singularity has been pushed to = (cf. Fig. 4).

The Physics of Inflation

How can a00 > 0 occur?

Exercise.Show that a00 > 0 is equivalent to a slow variation of the Hubble parameter

H0
< 1. (1.25)
H2
Notice that = 0 corresponds to a quasi-de Sitter spacetime with nearly constant expansion rate,
H const. Using the Friedmann equations show that
 
3 P P 1
= 1+ < 1 w < . (1.26)
2 3

The last condition corresponds to a violation of the strong energy condition.

As a simple toy model for inflation we consider the dynamics of a scalar field, the inflaton
(t, x). As indicated by the notation, the value of the field can depend on time t and the position
in space x. Associated with each field value is a potential energy density V () (see Fig. 5). If
the field is dynamical (i.e. changes with time) then it also carries kinetic energy density. If the
stress-energy associated with the scalar field dominates the universe, it sources the evolution of
the FRW background. We want to determine under which conditions this can lead to accelerated
expansion.

Figure 5. Example of a slow-roll potential. Inflation occurs in the shaded parts of the potential.

9
 The stress-energy tensor of the scalar field is
 
1
T = g g V () . (1.27)
2
Consistency with the symmetries of the FRW spacetime requires that the background value of
the inflaton only depends on time, = (t). From the time-time component T 0 0 = , we infer
that
1
= (0 )2 + V () . (1.28)
2
We see that the total energy density, , is simply the sum of the kinetic energy density, 12 (0 )2 ,
and the potential energy density, V (). From the space-space component T i j = P ji , we find
that the pressure is the difference of kinetic and potential energy densities,
1 0 2
P = ( ) V () . (1.29)
2
We see that a field configuration leads to inflation, P < 13 , if the potential energy dominates
over the kinetic energy. i.e. the field rolls slowly.

Exercise.Using the Einstein equations (1.9) and (1.10), show that

3Mpl2
H 2 = 12 (0 )2 + V 1
(0 )2
= 2 2 2 < 1, (1.30)
Mpl2
H 0 = 21 (0 )2
Mpl H

where Mpl = (8G)1/2 is the reduced Planck mass.

Equation (1.30) corresponds to the first slow-roll condition:

(0 )2 < V . (1.31)

Does inflation last? Combining the two Friedmann equations in (1.30) leads to the Klein-Gordon
equation for the evolution of the scalar field

00 + 3H0 = V, . (1.32)

To maintain the slow-roll condition (1.31) for a sufficient period of time, we require that the
acceleration of the field is small. This is quantified by the second slow-roll condition:

00 < 3H0 . (1.33)

The two slow-roll conditions (1.31) and (1.33) are satisfied simultaneously if
2 
Mpl V, 2
  
2  V, 
 
V < 1, V Mpl  < 1. (1.34)
2 V V 
The slow-roll parameters, V and V , express the conditions for slow-roll inflation purely in terms
of the shape of the potential V ().

10
 Exercise.Applying the slow-roll conditions (1.31) and (1.33) to the Friedmann equation (1.30) and
the Klein-Gordon equation (1.32), we get
2
3Mpl H2 V , (1.35)
0
3H V, . (1.36)

Use this to show that V during slow-roll inflation.

Inflation ends when the first slow-roll condition is violated, V (e ) 1. The amount of
inflation is measured in terms of e-folds of expansion, dN = d ln a. The total number of e-folds
between a point on the potential and the end of inflation at e is
Z ae Z te Z e
H
N () d ln a = H dt = d
a t 0
Z e Z e
1 d 1 d
= . (1.37)
2 Mpl 2V Mpl

To solve the horizon problems requires at least between 40 and 60 e-folds (the precise value
depending on the reheating temperature).

Case study: m2 2 inflation.As an example, let us give the slow-roll analysis of arguably the simplest
model of inflation: single-field inflation driven by a mass term
1 2 2
V () = m . (1.38)
2
The slow-roll parameters are
 2
Mpl
V () = V () = 2 . (1.39)

To satisfy the slow-roll conditions {V , |V |} < 1, we therefore need to consider super-Planckian values
for the inflaton
> 2Mpl e . (1.40)
The relation between the inflaton field value and the number of e-folds before the end of inflation is
2 1
N () =
4Mpl2 2 . (1.41)

Solving the horizon problem requires that the initial value of the field, i , satisfies

i > 60 2 60 Mpl 15Mpl . (1.42)

11
Part I
Primordial Seeds
The first part of this course will set the stage for studying the evolution of cosmological per-
turbations. In Section 2, we review the basic elements of cosmological perturbation theory. In
Section 3, we show how quantum mechanics during inflation provides the primordial seeds for
fluctuations in the primordial plasma. The evolution of these fluctuations will be the subject of
the second part of the course.

2 Cosmological Perturbation Theory

The Einstein equations couple perturbations in the stress-energy tensor to those in the metric,
so the two need to be studied simultaneously. We write the small perturbations of the metric
and the stress-energy tensor as

g (, x) = g () + g (, x) , (2.1)
T (, x) = T () + T (, x) . (2.2)

To avoid clutter we will often drop the argument (, x) on the perturbations.

Note that the perturbations in (2.1) and (2.2) arent uniquely defined, but depend on our
choice of coordinates or the gauge choice. In particular, to identify the metric perturbations
in (2.1) we implicitly have to chose a specific time slicing of the spacetime and specific spatial
coordinates on these time slices. Making a different choice of coordinates, can change the values
of the perturbation variables. It may even introduce fictitious perturbations. These are fake
perturbations that can arise by an inconvenient choice of coordinates even if the spacetime is
perfectly homogeneous. In this course, we will avoid these subtleties, by working in a fixed gauge
and sticking with it (except for a small change of hearth in Sec. 3).

2.1 Metric Perturbations

We will work in Newtonian gauge and write the perturbed metric as

ds2 = a2 () (1 + 2)d 2 [(1 2)ij + hij ] dxi dxj ,


(2.3)

where is the gravitational potential, is a local perturbation of the average scale factor and
hij is a transverse and traceless tensor. For now, we will only study scalar fluctuations and
set hij = 0. The effects of tensor perturbations are explored in the Problem Sets.

12
 Exercise.Show that the connection coefficients associated with the metric (2.3) are

000 = H + , (2.4)
0i0 = i , (2.5)
i00 ij
= j , (2.6)
0ij
 
= Hij + 2H( + ) ij , (2.7)
ij0 = H ji ,
 
(2.8)
ijk = i
2(j k) il
+ jk l . (2.9)

2.2 Matter Perturbations

We write the perturbed stress-energy tensor as

T 0 0 = () + , (2.10)
T i 0 = [() + P ()] v i , (2.11)
T i j = [P () + P ] ji i j , (2.12)

where vi is the bulk velocity and i j is a transverse and traceless tensor describing anisotropic
stress. We will use q i for the momentum density ( + P )v i . In case there are several contributions
to the stress-energy tensor (e.g. photons, baryons, dark matter, etc.), they are added: T =
P (a)
a T . This implies
X X X X ij
= a , P = Pa , q i = i
q(a) , ij = (a) . (2.13)
a a a a

We see that the perturbations in the density, pressure and anisotropic stress simply add. The
velocities do not add, but the momentum densities do. For scalar fluctuations, we can write the
Fourier components of the bulk velocity and the anisotropic stress as

v i (, k) = ik i v(, k) , (2.14)
ij (, k) = ( + P ) k hi k ji (, k) , (2.15)

where k k/|k| and the angle brackets in (2.15) stand for the symmetric trace-free part,
i.e. k hi k ji k i k j 31 ij . It is also convenient to write the density perturbations in terms of
the dimensionless density contrast /. In summary, scalar perturbations of the total
matter are described by four perturbation variables, (, P, v, ). Similarly, the perturbations of
distinct species a = , , c, b, are represented by (a , Pa , va , a ).

2.3 Equations of Motion

The equations of motion for the matter perturbations follow from conservation of the stress tensor,
T = 0. If there is no energy and momentum transfer between the different components, then
(a)
the species are separately conserved and we also have T = 0. Equipped with the perturbed

13
connection derived in 2.1, we can derive the perturbed conservation equations from
T = 0
= T + T T . (2.16)
After some work, we find
= 3H( + P ) + 3( + P ) i q i , (2.17)

q i = 4Hq i ( + P ) i i P j ij . (2.18)

Derivation.Consider first the = 0 component of (2.16),

0 T 0 0 + i T i 0 + 0 T 0 0 + i T i 0 000 T 0 0 0i0 T i 0 i00 T 0 i ij0 T j i = 0 . (2.19)

| {z } | {z } | {z }
O(2) O(2) O(2)

Substituting the perturbed stress-energy tensor and the connection coefficients gives

0 ( + ) + i q i + (H + + 3H 3)( + )
(H + )( + ) (H )ji (P + P )ij = 0 ,
 
(2.20)

and hence

+ 0 + i q i + 3H( + ) 3 0 + 3H(P + P ) 3P = 0 . (2.21)

Writing the zeroth-order and first-order parts separately, we get

= 3H( + P ) , (2.22)
= 3H( + P ) + 3( + P ) q . (2.23)

The zeroth-order equation (2.22) is simply the conservation of energy in the homogeneous background.
Equation (2.23) is the continuity equation describing the evolution of the density perturbation. The
first term on the right-hand side is just the dilution due to the background expansion (as in the
background equation), the q term accounts for the local fluid flow due to peculiar velocity, and
the term is a purely relativistic effect corresponding to the density changes caused by perturbations
to the local expansion rate [(1 )a is the local scale factor in the spatial part of the metric in
Newtonian gauge].
Next, consider the = i component of (2.16), T i + T i i T = 0, and hence

0 T 0 i + j T j i + 0 T 0 i + j T j i 00i T 0 0 0ji T j 0 j0i T 0 j jki T k j = 0 . (2.24)

Substituting the perturbed stress-energy tensor [with T 0 i = qi ] and the connection coefficients gives
h i
0 qi + j (P + P )ij j i 4Hqi (j 3j ) P ij i
 
Hji q j + Hij qj + 2(i j
k) + ki jl l P jk = 0 . (2.25)
| {z }
3i P
Cleaning this up, we find

qi = 4Hqi ( + P )i i P j ij . (2.26)

This is the relativistic version of the Euler equation for a viscous fluid.

14
It is instructive to evaluate (2.17) and (2.18) for a few special cases:

Matter.For a non-relativistic fluid (i.e. matter), we have Pm = 0 and ij

m = 0. The
continuity and Euler equations then simplify considerably. Writing m m /m , we
obtain

m = vm + 3 , (2.27)
vm = Hvm . (2.28)

Each term in these equations should be rather intuitive.

Radiation.For a relativistic fluid (i.e. radiation), we have Pr = 1

3 r and ij
r = 0. The
continuity and Euler equations become
4
r = vm + 4 , (2.29)
3
1
vr = r . (2.30)
4

Exercise.Show that the most general forms of the continuity and Euler equations are
   
Pa i Pa Pa
a = 1 + (i va 3) 3H a , (2.31)
a a a
Pa
!
1
vai = H + vai i Pa j ij i


a . (2.32)
a + Pa a + Pa

Confirm that these expressions reduce to the equations for matter and radiation in the appropriate
limits.

Comments.The two equations (2.31) and (2.32) arent sufficient to completely describe the evolution
of the four perturbations (a , Pa , va , a ). To make progress, we either must make further simplifying
assumptions or find additional evolution equations. We will do both.
A perfect fluid is characterized by strong interactions which keep the pressure isotropic, a = 0.
In addition, pressure perturbations satisfy Pa = c2s,a a , where cs,a is the adiabatic sound
speed of the fluid. The perturbations of a perfect fluid are therefore described by only two
independent variables, say a and va , and the continuity and Euler equations are sufficient for
closing the system.
Decoupled or weakly interacting species (e.g. neutrinos) cannot be described by a perfect fluid
and the above simplifications for the anisotropic stress and the pressure perturbation do not
apply. In that case, we cant avoid solving the Boltzmann equation for the evolution of the
perturbed distribution function fa . We will have some fun with this in Section 4.
Decoupled cold dark matter is a peculiar case. It is collisionless and has a negligible velocity
dispersion. It therefore behaves like a pressureless perfect fluid although it has no interactions
and therefore really isnt a fluid.

15
 The different matter components are gravitationally coupled through the metric fluctuations
in the continuity and Euler equations. The dynamics to the perturbed spacetime is determined,
via the Einstein equations, by the perturbations of the total stress-energy tensor.
Let us compute the linearised Einstein equation in Newtonian gauge. This is conceptually
straightforward, although algebraically it is a bit tedious.1 We require the perturbation to the
Einstein tensor, G R 12 Rg , so we first need to calculate the perturbed Ricci tensor R
and Ricci scalar R.
Ricci tensor.We recall that the Ricci tensor can be expressed in terms of the connection as

R = + . (2.33)

Substituting the perturbed connection coefficients (2.4)(2.9), we find

R00 = 3H + 2 + 3H( + ) + 3 , (2.34)

R0i = 2i ( + H) , (2.35)
Rij = H + 2H2 + 2 2(H + 2H2 )( + ) H 5H ij
 
(2.36)
+ i j ( ) .

I will derive R00 explicitly and leave the other components as an exercise.

Example.The 00 component of the Ricci tensor is

R00 = 00 0 0 +
00 0 0 . (2.37)

The terms with = 0 cancel in the sum over , so we only need to consider summing over = i,

R00 = i i00 0 i0i + i i

00 i 0i 0

= i i00 0 i0i + 000 i0i + j00 iji 00i i00 j0i i0j
| {z } | {z }
O(2) O(2)

= 30 (H ) + 3(H + )(H ) (H )2 ij ji
2

= 3H + 2 + 3H( + ) + 3 . (2.38)

This confirms the result (2.34).

Exercise.Derive eqs. (2.35) and (2.36).

Ricci scalar.It is now relatively straightforward to compute the Ricci scalar

R = g 00 R00 + 2 g 0i R0i +g ij Rij . (2.39)

| {z }
O(2)

1
Once in your life you should do this computation by hand. After that you can use Mathematica: an example
notebook can be downloaded here.

16
It follows that

a2 R = (1 2)R00 (1 + 2) ij Rij
= (1 2) 3H + 2 + 3H( + ) + 3
 

3(1 + 2) H + 2H2 + 2 2(H + 2H2 )( + ) H 5H

 

(1 + 2)2 ( ) . (2.40)

Dropping non-linear terms, we find

a2 R = 6(H + H2 ) + 22 42 + 12(H + H2 ) + 6 + 6H( + 3) . (2.41)

Einstein equations.We have done all the work to compute the Einstein equation

G = 8GT . (2.42)

We chose to work with one index raised since that simplifies the form of the stress tensor [see 2.2].
We will first consider the time-time component. The relevant component of the Einstein tensor is
 
1
G0 0 = g 00 R00 g00 R
2
1
= a2 (1 2)R00 R , (2.43)
2
where we have used that g 0i vanishes in Newtonian gauge. Substituting (2.34) and (2.41), and
cleaning up the resulting mess, we find

G0 0 = 22 6H( + H) . (2.44)

The 00-Einstein equation therefore is

2 3H( + H) = 4Ga2 , (2.45)

P
where a a is the total density perturbation. Equation (2.45) is the relativistic general-
ization of the Poisson equation. Inside the Hubble radius, i.e. for Fourier modes with k  H,
we have |2 |  3H| + H|, so that eq. (2.45) reduces to 2 4Ga2 . This is the Pois-
son equation in the Newtonian limit. The GR corrections in (2.45) will be important on scales
comparable to the Hubble radius, i.e. for k . H.
Next, we consider the spatial part of the Einstein equation. The relevant component of the
Einstein tensor is
 
i ik 1
G j = g Rkj gkj R
2
1
= a2 (1 + 2) ik Rkj ji R . (2.46)
2
From eq. (2.36), we see that most terms in Rkj are proportional to kj . When contracted with ji
this leads to a myriad of terms proportional to ji . We dont want to deal with this mess. Instead

17
 Figure 6. Numerical solutions for the linear evolution of the gravitational potential.

we focus on the tracefree part of Gi j . In Fourier space, we can extract this piece by contracting
Gi j with the projection tensor P j i k j ki 13 ij . Using (2.36), this gives
2
P j i Gi j = a2 k 2 ( ) . (2.47)
3
This should be equated to the tracefree part of the stress tensor, which for scalar fluctuations is
 1  1 
P j i T i j = P j i i j = ( + P ) k j ki ij k i kj ji
3 3
2
= ( + P ) . (2.48)
3
Setting (2.47) and (2.48) equal, we get

2 ( ) = 8Ga2 ( + P ) , (2.49)
P
where ( + P ) a (a + Pa )a . Dark matter and baryons can be described as perfect fluids
and therefore dont contribute to the anisotropic stress in (2.49). Photons only start to develop
an anisotropic stress component during the matter-dominated era when their energy density is
subdominant (see 4.3). The only relevant source in (2.49) are therefore free-streaming neutrinos.
However, their effect is also relatively small, so to the level of accuracy that we aspire to in these
lectures they can be ignored. Equation (2.49) then implies .

Exercise.By considering the trace of the space-space Einstein equation derive the following evolution
equation for the metric potential

+ 3H + (2H + H2 ) = 4Ga2 P , (2.50)

where P is the total pressure perturbation. If we hadnt ignored the anisotropic stress, then (2.50)
would have an additional source term. Discuss qualitatively the solution (, k) during both the

18
 radiation-dominated and matter-dominated eras. Describe the evolution on super-Hubble scales, k 
H, and on sub-Hubble scales, k  H.

Figure 6 shows the evolution of the gravitational potential for three representative wavelengths.
The important thing to notice (and to remember) from this figure is that, during the radiation
era, decays inside the Hubble radius, i.e. for Fourier modes with k  H. During the matter
era, on the other hand, is constant on all scales. Short-wavelength modes spend more time
inside the Hubble radius during the radiation era and therefore have suppressed amplitudes.

2.4 Initial Conditions

At sufficiently early times, all scales of interest to current observations were outside the Hubble
radius. On super-Hubble scales, the evolution of perturbations becomes very simple, especially
for adiabatic initial conditions.

Adiabatic perturbations
Adiabatic perturbations have the property that the local state of matter (determined, for example,
by the energy density and the pressure P ) at some spacetime point (, x) of the perturbed
universe is the same as in the background universe at some slightly different time +(x). (Notice
that the time shift varies with location x.) We can therefore think of adiabatic perturbations
as arising because some parts of the universe are ahead and others behind in the evolution.
If the universe is filled with multiple fluids, adiabatic perturbations correspond to perturbations
induced by a common, local shift in time of all background quantities; e.g. adiabatic density
perturbations are defined as

a (, x) a ( + (x)) a () = a (x) , (2.51)

where is the same for all species a. This implies

a b
= = for all species a and b . (2.52)
a b

Using a = 3H(1 + wa )a , we can write this as

a b
= for all species a and b . (2.53)
1 + wa 1 + wb

Thus, for adiabatic perturbations, all matter components (wm 0) have the same fractional
perturbations, while all radiation perturbations (wr = 31 ) obey

4
r = m . (2.54)
3
P
It follows that, for adiabatic fluctuations, the total density perturbation, a a a , is domi-
nated by the species that carries the dominant energy density a , since all the a s are comparable.
At early times, the universe is radiation dominated, so it natural to set the initial conditions for

19
 CDM
baryons

density perturbations
photons

CDM

baryons

photons

Figure 7. Evolution of photons, baryons and dark matter. On super-Hubble scales (early times), the
fluctuations in each component remain adiabatic and constant. On sub-Hubble scales, CDM first only
grows slowly during the radiation era and then as a power law during the matter era. Before recombination,
photons and baryons are tightly coupled and oscillate on sub-Hubble scales. After recombination baryons
fall into CDM potential wells.

all super-Hubble Fourier modes then. Equation (2.50) implies that = const. on super-Hubble
scales, while equation (2.45) leads to

r = 2 = const. (2.55)

Equations (2.55) and (2.54) show that, for adiabatic initial conditions, all matter perturbations
are given in terms of the super-Hubble value of the potential . In these lectures, we will be
concerned with the evolution of photons, baryons and cold dark matter (CDM). Their fractional
density perturbations will satisfy the relation (2.54) on super-Hubble scales, but will start to
evolve in distinct ways inside of the horizon (cf. Fig. 7).

Comoving curvature perturbation

The gravitational potential is only constant on super-Hubble scales if the equation of state
of the background is constant. Whenever the equation of state evolves (e.g. in the transitions
from inflation to radiation domination or from radiation to matter domination), so will the
gravitational potential (cf. Fig. 6). It will be convenient to identify an alternative perturbation
variable that stays constant on large scales even in these more general situations. Such a variable

20
is the comoving curvature perturbation:
H
R = + q , (2.56)
+ P
where T 0 j j q. Defining the initial conditions in terms of R will allow us to match the
predictions made by inflation to the fluctuations in the primordial plasma most easily (see Sec. 3).

Proof.The following is a proof that R is conserved on super-Hubble scales, i.e. for modes with k  H.
First, it is useful to note that on large scales k  H, the Einstein equations imply Hq = 13 . In
this limit, the curvature perturbation can be written as

kH
R . (2.57)
3( + P )

To find the evolution of R, we consider the continuity equation [eq. (2.17)]:

+ 3H( + P ) + i q i = 3( + P ) , (2.58)

On large scales, i q i is of order k 2 and can be dropped relative to terms of order H2 . Solving (2.57)
for , and substituting it into (2.58), we get
 
R
+ 3H( + P ) = 3( + P ) + ( + P ) . (2.59)
+ P

The time derivatives of cancel on both sides and we are left with

R + P
3( + P ) = 3H( + P ) . (2.60)
( + P )

Using = 3H( + P ), this becomes

P
 
R
( + P ) = P . (2.61)
H

For adiabatic perturbations, the right-hand side vanishes and we have established the conservation of
kH
the comoving curvature perturbation, R 0.

Another advantage of the variable R is that it is gauge-invariant, i.e. its value does not
depend on the choice of coordinates. Although (2.57) was written in terms of variables defined in
Newtonian gauge, it applies in an arbitrary gauge if we write gij = a2 (1 2)ij + i j E for
 

the metric and T 0 j = j q for the momentum density. While we will mostly stick to Newtonian
gauge, the computation of inflationary fluctuations in Section 3 turns out to be simplest in
spatially flat gauge [ = E = 0]. The curvature perturbation R provides the bridge between
results obtained in Section 3 and the analysis in the rest of the notes.

Statistics
Quantum mechanics during inflation only predicts the statistics of the initial conditions, i.e. it
predicts the correlation between the CMB fluctuations in different directions, rather than the spe-
cific value of the temperature fluctuation in a specific direction. For Gaussian initial conditions,

21
these correlations are completely specified by the two-point correlation function

hR(x)R(x0 )i R (x, x0 ) = R (|x0 x|) , (2.62)

where the last equality holds as a consequence of statistical homogeneity and isotropy. The
Fourier transform of R then satisfies
2 2 2
hR(k)R (k0 )i = (k) D (k k0 ) , (2.63)
k3 R
where 2R (k) is the (dimensionless) power spectrum.

Exercise.Show that Z
dk 2
R (x, x0 ) = (k) sinc(k|x x0 |) . (2.64)
k R

In Section 3, we will compute the form of 2R (k) predicted by inflation. In Sections 4 and 5,
we will then show how these initial conditions get imprinted in the anisotropies of the cosmic
microwave background.

22
3 Quantum Initial Conditions

The most remarkable feature of inflation is that it provides at natural mechanism for producing
the initial conditions (see Fig. 8). The reason why inflation inevitably produces fluctuations is
simple: the evolution of the inflaton field (t) governs the energy density of the early universe (t)
and hence controls the end of inflation. Essentially, the field plays the role of a local clock
reading off the amount of inflationary expansion still to occur. By the uncertainty principle, arbi-
trarily precise timing is not possible in quantum mechanics. Instead, quantum-mechanical clocks
necessarily have some variance, so the inflaton will have spatially varying fluctuations (t, x).
There will hence be local differences in the time when inflation ends, t(x), so that different
regions of space inflate by different amounts. These differences in the local expansion histories
lead to differences in the local densities after inflation, (t, x), and to curvature perturbations in
comoving gauge, R(x). It is worth remarking that the theory wasnt engineered to produce these
fluctuations, but that their origin is instead a natural consequence of treating inflation quantum
mechanically.

Figure 8. Quantum fluctuations (t, x) around the classical background evolution (t). Regions acquir-
ing negative fluctuations remain potential-dominated longer than regions with positive . Different
parts of the universe therefore undergo slightly different evolutions. After inflation, this induces density
fluctuations (t, x).

3.1 Inflaton Fluctuations: Classical

Before we quantise the inflaton fluctuations, we look at their classical dynamics. It will be useful
to derive this from the inflaton action

3
Z  
1
S = d d x g g V () , (3.1)
2
where g det(g ). To study the linearised dynamics, we need the action at quadratic order in
fluctuations. In general, finding the quadratic action for the coupled fluctuations and g is
a bit involved. However, the problem greatly simplifies by a convenient choice of gauge. In this
section, we will work in spatially flat gauge, in which our freedom in the choice of coordinates
has been exploited to set the spatial metric to be unperturbed, gij = a2 ij . In this gauge, the
information about the perturbations is carried by the inflaton perturbation and the metric
perturbations g0 . The Einstein equations relate g0 to . An important feature of spatially

23
flat gauge is that the metric perturbations g0 are suppressed relative to the inflaton fluctuations
by factors of the slow-roll parameter ; in particular, g0 vanishes in the limit 0. This means
that at leading order in the slow-roll expansion, we can ignore the fluctuations in the spacetime
geometry and perturb the inflaton field independently. (In a general gauge, inflaton and metric
perturbations would be equally important and would have to be studied together.) This makes
our job a lot simpler.

Evaluating (3.1) for the unperturbed FRW metric, we find

Z  
3 1 2 2 2

4
S = d d x a () a V () . (3.2)
2
It is convenient to write the perturbed inflaton field as
f (, x)
(, x) = () + . (3.3)
a()
To get the linearised equation of motion for f (, x), we need to expand the action (3.2) to second
order in the fluctuations. After some work, we find
Z  
3 1 2 2 a 2
S(2) = d d x f (f ) + f . (3.4)
2 a

Derivation.Substituting (3.3) into (3.2) and isolating all terms with two factors of f , we get
Z
1 h
i
S(2) = d d3 x f2 (f )2 2Hf f + H2 a2 V, f 2 . (3.5)
2

Integrating the term f f = 12 (f 2 ) by parts, gives

Z
1 h   i
S(2) = d d3 x f2 (f )2 + H + H2 a2 V, f 2 ,
2
Z    
1 a
= d d3 x f2 (f )2 + a2 V, f 2 . (3.6)
2 a

During slow-roll inflation, we have

2
V, 3Mpl V,
2
= 3V  1 . (3.7)
H V
Since a = a2 H, with H const., we also have
a
2aH = 2a2 H 2  a2 V, . (3.8)
a
Hence, we can drop the V, term in (3.6) and arrive at (3.4).

The action (3.4) implies the following equation of motion

d3 x
  Z
a

fk + k 2
fk = 0 , fk () f (, x)eikx . (3.9)
a (2)3/2

Sometimes this is called the Mukhanov-Sasaki (MS) equation.

24
 Exercise.Derive (3.9) directly from the Klein-Gordon equation, g = V, .

In a quasi-de Sitter background, we have a/a 2H2 = 2/ 2 and the MS equation (3.9) becomes
 
2
fk + k 2 fk = 0 .
2
(3.10)

A crucial feature of inflation is that the comoving Hubble radius H1 = (aH)1 shrinks (the
scale factor is a 1/(H) and conformal time evolves from to 0). The characteristic
evolution of a given Fourier mode is illustrated in Fig. 9: at early times, ||  k 1 , the mode is
inside the Hubble radius. In this limit, the MS equation reduces to the equation of motion of a
simple harmonic oscillator,
fk + k 2 fk 0 (for |k|  1) . (3.11)
Quantum fluctuations of these oscillators provide the origin of structure in the universe. We will
review the quantisation of harmonic oscillators in 3.2 and upgrade the discussion to inflaton
fluctuations in 3.3.
At some point during inflation the mode crosses the horizon, |k| = 1. At this moment it
becomes convenient to switch to a description in terms of the comoving curvature perturbation R.
As we have shown in 2.4, the field R is constant outside of the horizon, i.e. for |k|  1. The
variance of the curvature perturbation at horizon crossing, h|Rk=H |2 i, will become the initial
condition for the evolution of perturbations in the post-inflationary FRW universe.

comoving
scales

classical stochastic field

subhorizon superhorizon

quantum
fluctuations

time
horizon reheating horizon CMB today
exit re-entry

Figure 9. Curvature perturbations during and after inflation: The comoving horizon (aH)1 shrinks
during inflation and grows in the subsequent FRW evolution. This implies that comoving scales k 1 exit
the horizon at early times and re-enter the horizon at late times. While the curvature perturbations R
are outside of the horizon they dont evolve, so our computation for the correlation function h|Rk |2 i at
horizon exit during inflation can be related directly to observables at late times.

25
3.2 Quantum Harmonic Oscillators
The career of a young theoretical physicist consists of treating the harmonic oscillator
in ever-increasing levels of abstraction. Sidney Coleman

Our aim is to quantise the field f following the standard methods of quantum field theory.
However, before we do this, let us study a slightly simpler problem2 : the quantum mechanics of
a one-dimensional harmonic oscillator.
Consider a mass m attached to a spring with spring constant . Let q be the deviation from
the equilibrium point. The equation of motion is mq + q = 0, or

q + 2 q = 0 , (3.12)

if we define 2 /m. This is the same equation as (3.11), with = k.

Canonical quantisation
Let me remind you how to quantise the harmonic oscillator:

Step I: Quantum operators

First, we promote the classical variables q, p q to quantum operators q, p and impose the
canonical commutation relation
[q, p] = i , (3.13)
in units where ~ 1. The equation of motion implies that the commutator holds at all
times if imposed at some initial time.

Step II: Mode expansion

Note that we are using the Heisenberg picture where operators vary in time while states
are time-independent. The operator solution q(t) is determined by two initial conditions
q(0) and p(0) = t q(0). Since the evolution equation is linear, the solution is linear in
these operators. It is convenient to trade q(0) and p(0) for a single time-independent non-
Hermitian operator a, in terms of which the solution can be written as

q(t) = q(t) a + q (t) a , (3.14)

where the (complex) mode function q(t) satisfies the classical equation of motion, q + 2 q =
0. Of course, q (t) is the complex conjugate of q(t) and a is the Hermitian conjugate of a.

Step III: Normalization

Substituting (3.14) into (3.13), we get

W [q] [a, a ] = 1 , (3.15)

where we have defined the Wronskian

W [q] i (q q qq ) . (3.16)
2
The reason it looks simpler is that it avoids distractions arising from Fourier labels, delta functions, etc. The
physics is exactly the same.

26
 Without loss of generality, let us assume that the solution q is chosen so that the real
number W [q] is positive. The function q can then be rescaled (q q) such that

W [q] 1 , (3.17)

and hence
[a, a ] = 1 . (3.18)
Equation (3.18) is the standard commutation relation for the raising and lowering operators
of the harmonic oscillator.

Step IV: Vacuum state

The vacuum state |0i is annihilated by the operator a:

a|0i = 0 . (3.19)

Excited states are created by repeated application of creation operators a .

Choice of vacuum
At this point, we have only imposed the normalisation W [q] = 1 on the mode functions. A
change in q(t) could be accompanied by a change in a that keeps the solution q(t) unchanged. Via
eq. (3.19), each such solution corresponds to a different vacuum state. However, a special choice
of q(t) is selected if we require the vacuum state |0i to be the ground state of the Hamiltonian.
To see this, consider the Hamiltonian for general q(t),
1 2 1 2 2
H = p + q (3.20)
2 2
1h 2 i
= (q + 2 q 2 )aa + (q 2 + 2 q 2 ) a a + (|q|2 + 2 |q|2 )(aa + a a) .
2
Using a|0i = 0 and [a, a ] = 1, we can determine how the Hamiltonian operator acts on the
vacuum state
1 1
H|0i = (q 2 + 2 q 2 ) a a |0i + (|q|2 + 2 |q|2 )|0i . (3.21)
2 2
We want |0i to be an eigenstate of H. For this to be the case, the first term in (3.21) must vanish,
which implies
q = iq . (3.22)
For such a function q, the norm is
W [q] = 2|q|2 , (3.23)
and positivity of the normalisation condition W [q] > 0 selects the minus sign in (3.22),

q = iq q(t) eit . (3.24)

Asking the vacuum state to be the ground state of the Hamiltonian has therefore selected the
positive-frequency solution eit (rather than the negative-frequency solution e+it ). Imposing
the normalisation W [q] = 1, we get

1
q(t) = eit . (3.25)
2

27
With this choice of mode function, the Hamiltonian takes the familiar form
 
1
H = ~ a a + , (3.26)
2

where we have reinstated Plancks constant ~. We see that the vacuum |0i is the state of minimum
energy 21 ~. If any function other than (3.25) is chosen to expand the position operator, then
the state annihilated by a is not the ground state of the oscillator.

Zero-point fluctuations
The expectation value of the position operator q in the ground state |0i vanishes

hqi h0|q|0i

= h0|q(t)a + q (t)a |0i

= 0, (3.27)

because a annihilates |0i when acting on it from the left, and a annihilates h0| when acting on
it from the right. However, the expectation value of the square of the position operator receives
finite zero-point fluctuations

h|q|2 i h0|q q|0i

= h0|(q a + qa)(qa + q a )|0i

= |q(t)|2 h0|aa |0i

= |q(t)|2 h0|[a, a ]|0i

= |q(t)|2 . (3.28)

We see that the variance of the amplitude of the quantum oscillator is given by the square of the
mode function
~
h|q|2 i = |q(t)|2 = . (3.29)
2
This is all we need to know about the quantum mechanics of harmonic oscillators in order to
compute the fluctuation spectrum created by inflation.

3.3 Inflaton Fluctuations: Quantum

Let us return to the quadratic action (3.4) for the inflaton fluctuation f = a. The momentum
conjugate to f is
L
= f . (3.30)
f
We perform the canonical quantisation just like in the case of the harmonic oscillator.

28
Canonical quantisation
We follow the same steps as in the previous subsection:

Step I: Quantum operators

We promote the fields f (, x) and (, x) to quantum operators f(, x) and (, x). The
operators satisfy the equal time commutation relation

[f(, x), (, x0 )] = iD (x x0 ) . (I0 )

This is the field theory equivalent of eq. (3.13). The delta function is a signature of locality:
modes at different points in space are independent and the corresponding operators therefore
commute. In Fourier space, we find
d3 x d3 x0
Z Z
0 0
[fk (), k0 ()] = [f(, x), (, x0 )] eikx eik x
(2)3/2 (2)3/2 | {z }
iD (x x0 )
d3 x i(k+k0 )x
Z
=i e
(2)3

= iD (k + k0 ) , (3.31)

where the delta function implies that modes with different wavelengths commute. Equa-
tion (3.31) is the same as (3.13), but for each independent Fourier mode.

Step II: Mode expansion

The generalisation of the mode expansion (3.14) is

fk () = fk () ak + fk ()ak , (3.32)

where ak is a time-independent operator, ak is its Hermitian conjugate, and fk () and its

complex conjugate fk () are two linearly independent solutions of the MS equation
a
fk + k2 ()fk = 0 , where k2 () k 2 . (3.33)
a
As indicated by dropping the vector notation k on the subscript, the mode functions, fk ( )
and fk ( ), are the same for all Fourier modes with k |k|.3

Step III: Normalization

Substituting (3.32) into (3.31), we get

W [fk ] [ak , ak0 ] = D (k + k0 ) , (3.34)

where W [fk ] is the Wronskian (3.16) of the mode functions. As before, cf. (3.17), we can
choose to normalize fk such that
W [fk ] 1 . (3.35)
3
Since the frequency k ( ) depends only on k |k|, the evolution does not depend on direction. The constant
operators ak and ak , on the other hand, define initial conditions which may depend on direction.

29
 Equation (3.34) then becomes

[ak , ak0 ] = D (k + k0 ) , (3.36)

which is the same as (3.18), but for each Fourier mode. As before, the operators ak and ak
may be interpreted as creation and annihilation operators, respectively.

Step IV: Vacuum state

As in (3.19), the quantum states in the Hilbert space are constructed by defining the vacuum
state |0i via
ak |0i = 0 , (3.37)
and by producing excited states through repeated application of creation operators ak .

Choice of vacuum
As before, we still need to fix the mode function in order to define the vacuum state. Although
for general time-dependent backgrounds this procedure can be ambiguous, for inflation there is a
preferred choice. To motivate the inflationary vacuum state, let us go back to Fig. 9. We see that
at sufficiently early times (large negative conformal time ) all modes of cosmological interest
were deep inside the horizon, k/H |k|  1. This means that in the remote past all observable
modes had time-independent frequencies
a 2
k2 = k 2 k 2 2 k 2 , (3.38)
a
and the Mukhanov-Sasaki equation reduces to

fk + k 2 fk 0 . (3.39)

But this is just the equation for a free field in Minkowkski space, whose two independent solutions
are fk eik . As we have seen above, cf. eq. (3.25), only the positive frequency mode fk eik
corresponds to the ground state of the Hamiltonian. We will choose this mode to define the
inflationary vacuum state. In practice, this means solving the MS equation with the (Minkowski)
initial condition
1
lim fk () = eik . (3.40)
2k
This initial condition defines a preferable set of mode functions and a unique physical vacuum,
the Bunch-Davies vacuum.
For slow-roll inflation, it will be sufficient to study the MS equation in de Sitter space
 
2 2
fk + k 2 fk = 0 . (3.41)

This has the following exact solution

eik eik
   
i i
fk () = 1 + 1+ . (3.42)
2k k 2k k

30
where and are constants that are fixed by the initial conditions. In fact, the initial condi-
tion (3.40) selects = 0, = 1, and, hence, the Bunch-Davies mode function is

eik
 
i
fk () = 1 . (3.43)
2k k
Since the mode function is now completely fixed, the future evolution of the mode including its
superhorizon dynamics is determined.

Zero-point fluctuations
Finally, we can predict the quantum statistics of the operator
d3 k h
Z i

f(, x) = f () a + f ()a k e
ikx
. (3.44)
(2)3/2 k k k

As before, the expectation value of f vanishes, i.e. hfi h0|f|0i = 0. However, the variance of
inflaton fluctuations receive non-zero quantum fluctuations
h|f|2 i h0|f (, 0)f(, 0)|0i
d3 k d3 k0
Z Z



= h0| fk ()ak + f k ()a k fk 0 ()a 0 + fk 0 ()a 0 |0i
k k
(2)3/2 (2)3/2
d3 k d3 k0
Z Z
= 3/2 3/2
fk ()fk0 () h0|[ak , ak0 ]|0i
(2) (2)
Z 3
d k
= |fk ()|2
(2)3
k3
Z
= d ln k |fk ()|2 . (3.45)
2 2
We define the (dimensionless) power spectrum as

k3
2f (k, ) |fk ()|2 . (VI0 )
2 2
As in (3.29), the square of the classical solution determines the variance of quantum fluctuations.
Using (3.43), we find
 2  !
k 2
  2
2 2 2 H superhorizon H
(k, ) = a f (k, ) = 1+ . (3.46)
2 aH 2
We will use the approximation that the power spectrum at horizon crossing is
 2 
2 H
(k) . (3.47)

2 

k=aH

Computing the power spectrum at a specific instant (horizon crossing, k = aH) implicitly ex-
tends the result for the pure de Sitter background to a slowly time-evolving quasi-de Sitter space.
Different modes exit the horizon as slightly different times when aH has a different value. Evalu-
ating the fluctuations at horizon crossing also has the added benefit that the error we are making
by ignoring the metric fluctuations in spatially flat gauge doesnt accumulate over time.

31
3.4 Curvature Perturbations
At horizon crossing, we switch from the inflaton fluctuation to the conserved curvature per-
turbation R. To do this, we recall the gauge-invariant definition of the curvature perturbation
H
R = + q , (3.48)
+ P
where T 0 j j q. We now need to evaluate this in spatially flat gauge. Since the spatial part
of the metric is unperturbed, we have = 0. The perturbed momentum density is

T 0 j = g 0 j = g 00 0 j = j . (3.49)
a2
Combined with + P = a2 2 , this implies
H
R= = H 0 , (3.50)


where the prime denotes a derivative with respect to time t. Notice that the expression (3.50)
takes the form R = Ht, confirming the intuition that the curvature perturbation is induced
by the time delay to the end of inflation.
The power spectrum of R at horizon exit therefore is
 2 2 
2 H 
R (k) =  . (3.51)
2 0 k=aH
From now on, we will drop the label k = aH to avoid clutter. The result in (3.51) may also be
written as
1 H2
2R (k) = 2 2 , (3.52)
8 Mpl

where is the inflationary slow-roll parameter; cf. eq. (1.30). The time dependence of H and
leads to a small scale dependence of 2R (k). The form of the spectrum is approximately a power
law, 2R (k) = As (k/k )ns 1 , with the following spectral index
d ln 2R d ln 2R d ln 2R d ln 2R
ns 1 = = = 2 , (3.53)
d ln k d ln(aH) d ln a Hdt
where 0 /(H) is the second slow-roll parameter. The observational constraint on the scalar
spectral index is ns = 0.9603 0.0073. The observed percent-level deviation from the scale-
invariant value, ns = 1, are the first direct measurement of time dependence in the inflationary
dynamics.

Exercise.Show that for slow-roll inflation, eqs. (3.52) and (3.53) can be written as
1 1 V
2R = 4 , (3.54)
24 2 V Mpl
ns 1 = 6V + 2V , (3.55)

where V and V are the potential slow-roll parameters defined in (1.34). This expresses the amplitude
of curvature perturbations and the spectral index in terms of the shape of the inflaton potential.

32
3.5 Gravitational Waves
Arguably the cleanest prediction of inflation is a spectrum of primordial gravitational waves.
These are tensor perturbations to the spatial metric,

ds2 = a2 () d 2 (ij + 2hij )dxi dxj .

 
(3.56)

We wont go through the details of the quantum production of tensor fluctuations during inflation,
but just sketch the logic which is identical to the scalar case (and even simpler).
Substituting (3.56) into the Einstein-Hilbert action and expanding to second order gives
2
Mpl 2
Mpl

Z Z h i
S= d4 x g R S(2) = d d3 x a2 h2ij (hij )2 . (3.57)
2 8

Exercise.Confirm eq. (3.57). Hint: Dont forget a term quadratic in hij coming from g.

It is convenient to define
f+ f 0
Mpl 1
ahij f f+ 0 , (3.58)

2 2
0 0 0
so that Z  
(2) 1 X 3 2 2 a 2
S = d d x f (f ) + f . (3.59)
2 a
=+,

This is just two copies of the action (3.4), one for each polarization mode of the gravitational
wave, f+, . The power spectrum of tensor modes 2t can therefore be inferred directly from our

0.25

0.20 c
on
ca
con ve
vex
0.15

small-field
0.10 large-field
(chaotic)

0.05
l
natura
0.00
00.94
94 0.96
0 96 0.98
0 98 1.00

Figure 10. Planck+WMAP+BAO constraints on ns and r, together with predictions from a few repre-
sentative inflationary models.

33
previous result for 2f ,
 2
2
2h =2 2f . (3.60)
aMpl
Using (3.47), we get

2 H 2 
2h (k) = 2 2  . (3.61)

Mpl 
k=aH

This result is the most robust and model-independent prediction of inflation. Notice that the
tensor amplitude is a direct measure of the expansion rate H during inflation. This is in contrast
to the scalar amplitude which depends on both H and . The form of the tensor power spectrum
is also a power law, 2h (k) = At (k/k )nt , with the following spectral index

nt = 2 . (3.62)

Observationally, a small value for nt is hard to distinguish from zero. The tensor amplitude is
often normalized with respect to the measured scalar amplitude, As = (2.196 0.060) 109 (at
k = 0.05 Mpc1 ). The tensor-to-scalar ratio is

At
r = 16 . (3.63)
As

Inflationary models make predictions for (ns , r). The latest observational constraints on these
parameters are shown in Fig. 10.

34
Part II
CMB Anisotropies
We ended the previous section with a description of the primordial scalar and tensor pertur-
bations: R and hij . These metric perturbations couple gravitationally to all the matter per-
turbations in the primordial plasma (see Fig. 11). In addition, electrons and baryons (mostly
protons) are coupled to each other via Coulomb scattering, while electrons and photons interact
via Thomson scattering. When the electron density drops at recombination, Thomson scattering
suddenly becomes inefficient and the photons decouple. Coulomb scattering remains very strong
throughout, so the electrons and baryons can be treated as a single tightly-coupled fluid. The
combined electron-baryon fluid is often referred to as baryons for simplicity (although electrons
are leptons, of course). To derive the fluctuations in the CMB, we will have to follow the com-
plicated set of interactions between all matter components until recombination. The main tool
for doing this is the Boltzmann equation, which we will introduce in Section 4. In Section 5, we
will use this formalism to derive the spectrum of CMB temperature anisotropies.

n
tio
dia
Ra Neutrions

Dark
Photons
Energy

Thomson
Scattering Metric

Dark
Electrons Matter
Co
Sca ulom
tte b
rin
g Baryons r
tte
Ma

Figure 11. Interactions between the different forms of matter in the universe.

4 The Boltzmann Equation

For each species a = , , e, b, c, , there will be a Boltzmann equation describing the evolution
of the distribution function fa . Schematically, we have
dfa
= C[{fb }] , (4.1)
d
where the collision term on the right-hand side, in principle, can depend on the distribution
functions for several species; i.e. we have coupled equations. We will focus on the evolution of

35
the distribution function for photons, f , and drop the subscript, i.e. f f . Photons interact
most strongly with the electrons, so the Boltzmann equation becomes
df
= C[{f, fe }] . (4.2)
d
As we mentioned above, electrons are so strongly coupled to the rest of the plasma that it makes
no difference to think of the right-hand side of the Boltzmann equation as the photon-electron or
photon-baryon coupling. In 4.1, we study the left-hand side of equation (4.2), which describes
the free-streaming of the photons in an inhomogeneous spacetime. In 4.2, we add a scattering
term to the right-hand side and derive its effect on the photon distribution function.

4.1 Free-Streaming
In the absence of scattering, the right-hand side of (4.2) vanishes and we have
df
= 0. (4.3)
d
This innocent looking equation expresses Liouvilles theorem: the number of particles in a given
element of phase space doesnt change with time. In a perturbed spacetime this contains non-
trivial information since the phase space elements themselves then evolve in a complicated way
due to the perturbed metric.

Collisionless Boltzmann Equation

In an inhomogeneous spacetime, the photon distribution function can depend on time , posi-
tion x, the photon energy E and the photon direction of propagation p. It is convenient to
introduce the comoving energy  Ea which, in the absence of perturbations, is constant along
the photon path (see 1.1). The collisionless Boltzmann equation (4.3) can then be written in
the following form
d f f dx f d ln  f dp
f (, x, , p) = + + + = 0. (4.4)
d x d ln  d p d
| {z }
O(2)

As indicated, the last term is second order in perturbations and therefore doesnt have to be
kept in a linearized analysis. (It describes the gravitational lensing of the CMB.) Moreover, since
f /x and ln / are first order in perturbations, we can use the zeroth-order expressions,
dx/d = p [cf. eq. (1.2)] and f/ ln , in the remaining terms. Equation (4.4) then becomes

f f d ln 
+ p f + = 0. (4.5)
ln  d

At zeroth-order, this implies f/ = 0, i.e. the zeroth-order distribution function depends

only on the comoving energy . This is consistent with the form of the equilibrium distribution
function (1.19), as long as T a1 ,
   1    1    1
E  
f() exp 1 = exp 1 = exp 1 , (4.6)
T () aT () T0

36
where T0 = 2.7255 K is the present CMB temperature. To study the effect of first-order per-
turbations in (4.5), let us introduce a direction- and position-dependent fractional temperature
perturbation,4 (, x, p), in the distribution function
   1

f (, x, , p) = exp 1 . (4.7)
aT ()[1 + (, x, p)]
Assuming that is a small perturbation, we get
   1
[1 (, x, p)]
f (, x, , p) = exp 1
aT ()
d ln f
 

= f () 1 (, x, p) . (4.8)
d ln 
Equation (4.5) then becomes
d ln f()
 
d ln 
+ p = 0. (4.9)
d ln  d
We keep the overall factor since we ultimately want to add non-zero collision terms on the right-
hand side. At first order, the first two terms in the brackets combine into a total derivative,
d dx dp
(, x, p) = + + = + p , (4.10)
d x d p d
| {z }
O(2)

so we can write (4.9) as

d ln f()
 
d d ln 
=0 . (4.11)
d ln  d d
In the absence of collisions, the evolution of the temperature perturbations is therefore directly
related to the evolution of the comoving photon energy in the presence of metric fluctuations,
d ln /d. The latter follows from the geodesic equation.

Geodesic Equation
Recall the geodesic equation describing the evolution of the photon 4-momentum
dP
+ P P = 0 , (4.12)
d
where the affine parameter is defined such that P = dx /d. We will analyze the geodesic
motion of photons in a spacetime with scalar perturbations, and leave the case with tensor
perturbations as an exercise.

Since photons are massless the 4-momentum satisfies the constraint g P P = 0. Using the
perturbed metric in Newtonian gauge, cf. eq. (2.3), this implies

P = 2 [1 , (1 + )p] . (4.13)
a
4
Importantly, here we have assumed that does not depend on , i.e. the effect of perturbations is achromatic
and does not lead spectral distortions. We will have to check that this is a consistent assumption when we add
scattering terms to the Boltzmann equation.

37
 Derivation.Substituting (2.3) into g P P = 0, we have

a2 (1 + 2)(P 0 )2 p2 = 0 , (4.14)

where we have defined p2 gij P i P j . Solving this for P 0 and expanding to linear order in , we get
p
P0 = (1 ) . (4.15)
a
This is the generalization of the relativistic expression E = pc to a perturbed spacetime. (The extra
factor of a1 comes from the fact that we are using conformal time.) Using p = /a, we can write this
as

P 0 = 2 (1 ) . (4.16)
a
It remains to find the spatial component of the four-momentum. Let us write P i C pi , so that

p2 gij P i P j = a2 (1 2)ij C 2 pi pj = a2 (1 2)C 2 . (4.17)

Solving this for C and expanding to linear order in , we get

p 
Pi = (1 + ) pi = 2 (1 + ) pi . (4.18)
a a
Equations (4.16) and (4.18) confirm (4.13).

Using
d 
= 2 (1 ) , (4.19)
d a
the geodesic equation can be written in conformal time:
 dP
(1 ) + P P = 0 . (4.20)
a2 d
Let us write out the 0-th component in all its glory
 d   2 
+ 4 000 (1 2) + 200i pi + 0ij (1 + 2)pi pj = 0 .

(1 ) 2 2
(1 ) (4.21)
a d a a

H + i [H 2H( + )]ij

At first order, this cleans up rather remarkably

d ln  d
= + ( + ) , (4.22)
d d

where d/d = + pi i + O(2). Equation (4.22) tells us how the comoving energy evolves
along the photon path in the presence of metric perturbations. As expected,  is constant in the
homogeneous background, but is modified by the variation of along the path (first term on the
rhs) and by the evolution of the potentials (second term on the rhs). For the CMB, the latter is
important at late times (once dark energy starts to become relevant) and at early times (when
the universe isnt fully matter dominated).

38
4.2 Thomson Scattering
The dominant scattering effect close to recombination is the Thomson scattering of the photons
off the free electrons in the plasma. In this section, we will develop the theoretical formalism for
including this in the Boltzmann equation.

(a) (b)

Figure 12. Thomson scattering in the electron rest frame (a) and in the background frame (b).

Electron Rest Frame

Let us start in the rest frame of a single electron. In this frame, an incoming photon has energy 0in
and 3-momentum p0in = 0in p0in (see Fig. 12). The scattered photon has energy 0 = 0in and 3-
momentum p0 = 0 p0 . Here, we have used the fact that Thomson scattering doesnt change
the energy of the photon. The differential scattering cross-section for Thomson scattering of
unpolarized radiation, averaged over the outgoing polarization, is
d T
= . (4.23)
d 4
This describes the rate at which the electron scatters photons per solid angle, per unit incident
photon flux. The parameter T is the Thomson cross section
2
qe2

8
T = = 6.65 1029 m2 , (4.24)
3 40 me c2

where qe is the charge of the electron. Since (4.24) scales inversely with the mass of the scatterer,
the scattering with the much heavier protons can be ignored.

Comment.Notice that we have ignored the anisotropic nature of Thomson scattering. The more
correct form of the differential cross section is
d 3T 
1 + (p0in p0 )2 .

= (4.25)
d 16
Equation (4.23) is the angular average of (4.25).

39
 We can now write the scattering rate with respect to the proper time 0 in the electron rest-
frame,

df 0 (0 , p0 ) 
 Z
0 0 0 0 d  0 0 0
= ne dp0in f ( , pin ) f 0 (0 , p0 ) ,

C [f ( , p )] 0
(4.26)
d 
scatt. d

in out

where ne is the proper number density of the electrons. The first term in the integrand captures
the in-scattering of photons (p0in p0 ), while the second terms describes the out-scattering
(p0 p0in ). We see that the collision term vanishes for isotropic radiation, confirming our earlier
assertion that it starts at first order in perturbations. In the out-scattering term, we can pull the
distribution function outside of the integral. This gives

dp0in 0 0 0
Z
0 0 0 0 0 0 0
C [f ( , p )] = ne T f ( , p ) + ne T f ( , pin ) . (4.27)
4
Next, we have to transform this result to the background frame in which the electrons are moving.

Background Frame
To obtain the result in a general frame we will perform a Lorentz boost. This accounts for the
bulk velocity of the electrons, which we write as ve (in the orthonormal frame). At zeroth order,
the proper time in the boosted frame is the same as that in the rest frame, = 0 + O(1). Since
the scattering rate starts first order in perturbations, we dont need more accuracy than this.
The scattering rate with respect to the conformal time in the boosted frame then is

df 0 (0 , p0 ) 
 
df (, p) 
C[f (, p)] =a + O(2) = a C 0 [f 0 (0 , p0 )] + O(2) . (4.28)
d scatt. d 
scatt.

where we have used the Lorentz invariance of the distribution function, f 0 (0 , p0 ) = f (, p).
Equation (4.27) then implies
Z
0 0 0 dpin 0 0 0
C[f (, p)] = f ( , p ) + f ( , pin ) . (4.29)
4

where ane T . (At first order, it is sufficient to use the background density ne .) Finally, we
wish to re-write the right-hand side in terms of quantities defined in the boosted frame.
The scattering in the boosted frame is (in , pin ) (, p), but with in 6=  (due to the motion
of the electron). Lorentz transformation of the energy implies

0 = (1 p ve ) , (4.30)

where (ve ) is the Lorentz factor. Similarly, the inverse transform gives

in = 0in (1 + p0in ve )

= 2 (1 p ve )(1 + p0in ve )
 1 (p p0in ) ve .


(4.31)

40
Lorentz invariance of the distribution function implies

f 0 (0 , p0in ) = f (in , pin )

= f [(1 (p p0in ) ve ), pin ]

df
= f[(1 (p p0in ) ve )] (pin )
d ln 
df df
= f() (p p0in ) ve (pin ) . (4.32)
d ln  d ln 
We substitute this into (4.29). Let us discuss the three terms of (4.32) in turn: The first term, f(),
integrates to +f() and hence cancels the zeroth-order term in the out-scattering contribution,
f (, p) f() + (df/d ln )(p). In the second term, the p0in ve part integrates to zero
(by parity) and the p ve is independent of p0in and can be pulled out of the integral. The last
term is first order, so we can make the replacements dp0in dpin and p0in p0 pin p. With
these simplifications, we get

df  
C[f (, p)] = (p) 0 p ve . (4.33)
d ln 
where we have defined the monopole of the temperature anisotropy
Z
dpin
0 (pin ) . (4.34)
4

Comment.Had we kept the angular dependence in (4.23), we would have found

df
 Z 
3
dpin (pin ) 1 + (pin p)2 .
 
C[f (, p)] = (p) p ve (4.35)
d ln  16

4.3 Boltzmann for Photons

Adding the collision term (4.33) to the right-hand side of the collisionless Boltzmann equa-
tion (4.11), we get
d d ln   
= 0 p ve , (4.36)
d d
Equation (4.36) will be the starting point for the rest of our analysis. We will solve this equation
in the next section, but something interesting can be said immediately about the form of the
solution at early times, when the electrons and photons were tightly coupled. At that time,
Thomson scattering was still efficient (  H), the parenthesis on the right-hand side of (4.36)
has to vanish. As expected, scattering tends make the photon distribution isotropic in the electron
rest frame. In a general frame, the distribution has only a monopole and a dipole, 0 + pve .
For scalar fluctuations, d ln /d ln is given by (4.22) and the Boltzmann equation becomes

+ pi i = pi i
  (4.37)
0 p ve .

41
In Fourier space, this reads
+ ik = ik
  (4.38)
0 ive ,

where ve ive k and k p defines the overlap between the wavevector of the inhomogeneity
and the direction of propagation of the photon. For = 1, the photon moves along k, i.e. along
the direction of changing temperature (or density), while for = 0, the photon moves along the
direction of constant temperature.
Notice that, due to statistical isotropy, the equation of motion for does not depend explicitly
on k and p, but only on their relative orientation, k p, and the wavenumber k. The initial
conditions for , on the other hand, do depend on the wavevector k, but will only depend on
k p and not p (this is a consequence of the tight-coupling limit 0 + p ve ). Since
the perturbations from a single Fourier mode are axisymmetric (at least for scalar fluctuations),
it is useful to expand them in terms of Legendre polynomials:

X
(, k, p) (, k, ) (i)l l (, k)Pl () , (4.39)
l=0

where l are the multipole moments of the distribution. The factor of (i)l in (4.39) was chosen
for future convenience. The lowest-order multipoles are related in a simply way to the perturbed
stress tensor for the photons:
1 5
0 = , 1 = v , 2 = . (4.40)
4 3
From now on, we will most work with 0 and 1 , but it will be important to remember that
these are proportional to the density contrast and the bulk velocity of the photon gas.

Derivation.We can derive the relations in (4.40) by considering the perturbed stress tensor for
photons,

d3 p df
Z
T = f P P , f f , (4.41)
E(p) d ln 
P = (E/a)[1 , (1 + )p] ,
P = (aE)[1 + , (1 )p] .

We start with the time-time component T 0 0 . Keeping only linear orders in perturbations, we
find
d3 p
Z Z Z Z
0 0 3 1
T 0= f P P0 = d p f E = 4 dp d f 3
E(p) a
df 4
Z Z Z
4 3 1
= 4 d f  4 d  dp
a a d
Z Z !
4 dp
= 4 d f 1 + 4
3
, (4.42)
a 4
| {z } | {z }

42
 where a4 , as expected, and
40 . (4.43)

Similar manipulations for the T i 0 component of the stress tensor leads to

Z
dp
v = 3 p . (4.44)
4
Substituting (4.39) into the Fourier transform of (4.44), we get
Z
dp X
v (, k) = 3 (i)l l (, k)Pl (k p) p . (4.45)
4
l

By symmetry, the integral over p must be proportional to k, i.e.

Z Z
dp dp
Al k = Pl (k p) p Al = Pl (k p) k p
4 4
Z 1
d
= Pl ()
1 2
1
= l1 . (4.46)
3

Equation (4.45) then becomes v = ik1 ikv , so that

v = 1 . (4.47)

The final relation in (4.40) follows from performing the same exercise on the T i j component of
the stress tensor:
Z
dp
ij
= 4 phi pji . (4.48)
4
Substituting (4.39) into the Fourier transform of (4.48), we have
Z
dp X
ij
(, k) = 4 (i)l l (, k)Pl (k p) phi pji . (4.49)
4
l

The integral over p must be a symmetric and trace-free tensor made from k, i.e.
Z Z  
hi ji dp hi ji 2 dp 2 1
Bl k k = Pl (k p) p p Bl = Pl (k p) (k p)
4 3 4 3
Z 1
1
= d Pl ()P2 ()
3 1
2
= l2 . (4.50)
15
Hence, we have
4
ij
= 2 k hi k ji ( + P )k hi k ji , (4.51)
5
so that
3
= 2 . (4.52)
5

43
 The multipole expansion is convenient since all moments with l 2 are suppressed in the
tight-coupling limit. To see this, consider the evolution equation for l2 . We obtain this by
multiplying the Boltzmann equation (4.38) by Pl2 () and integrating over . Most terms on the
right-hand side of (4.38) have a simple dependence (proportional to 0 or 1 ) and will therefore
vanish when integrated against Pl2 (). The remaining terms lead to the following equation of
motion
 
l+1 l
l + k l+1 l1 = l . (4.53)
2l + 3 2l 2
Notice that this is an infinite hierarchy of coupled equations since each moment l is coupled to
the adjacent moments l1 .

Exercise.Derive eq. (4.53). You may use orthogonality of the Legendre polynomials
Z 1
2
d Pl ()Pl0 () = ll0 , (4.54)
1 2l + 1

as well as the recursion relation

(2l + 1)Pl () = (l + 1)Pl+1 () + lPl1 () . (4.55)

For  k H, this implies

k
l
l1  l1 , (4.56)

which shows that all higher-order moments (l 2) are suppressed in the tight-coupling regime.
In that case, we only need the l = 0 and l = 1 moments of the Boltzmann equation. Multiply-
ing (4.38) by P0 () and P1 (), and integrating over , we find
1
0 = k1 + , (4.57)
3
1 = k0 k (1 + ve ) . (4.58)

Exercise.Derive eqs. (4.57) and (4.58).

Equations (4.57) and (4.58) become more recognizable if we write them in terms of = 40 and
v = i1 k. We get
4 
+ v 3 = 0 , (4.59)
3
1
v + + = (v ve ) . (4.60)
4
Notice that (4.59) is standard continuity equation for . This is to be expected, since at linear
order there is no energy exchange due to Thomson scattering, so the fluid approximation should
still work for the continuity equation. The Euler equation (4.60), on the other hand, does receive
a correction from Thomson scattering.

44
 These equations need to be supplemented by Einstein equations for the potentials and ,
and an Euler equation for the electron velocity ve . We will study the complete system of equations
in Section 5.

4.4 Boltzmann for Neutrinos

Neutrinos play an important role during the radiation dominated era: about 40% of the back-
ground density is in neutrinos and perturbations in the cosmic neutrino background leave a subtle
imprint in the CMB anisotropies. We can write the perturbed neutrino distribution function as

df
f f N, (4.61)
d ln 
where N plays the same role as for photons. Neutrinos are decoupled from the rest of the
plasma, so they satisfy the collisionless Boltzmann equation. The moments N0 and N1 therefore
satisfy (4.57) and (4.63) without the collision term:
1
N0 = k N1 + , (4.62)
3
N1 = k N0 k . (4.63)

The perturbations in the neutrino distribution function are mediated to the rest of the plasma
through the metric perturbations and .

4.5 Einstein Revisited

The neutrino density contrast and anisotropic stress are = 4N0 and = 35 N2 ; cf. eq. (4.40).
The Poisson equation (2.45) can then be written as
h i
2 + 3H( + H) = 4Ga2 c c + b b + 4 0 + 4 N0 . (4.64)
| {z } | {z }
m m r r

The constraint equation (2.49) becomes

2 ( ) = 32Ga2 2 + N2 .
 
(4.65)

The two potentials are equal and opposite unless the photons and neutrinos have appreciable
quadrupole moments. We have seen that the photon quadrupole 2 is suppressed due to the
tight coupling to the electrons. Neutrinos on the other hand a purely free-streaming and can
therefore have an appreciable quadrupole moment N2 . To determine the evolution of N2 requires
solving the hierarchy of Boltzmann equations to higher order.

45
5 Temperature Anisotropies

A map of the cosmic microwave background radiation describes the variation of the CMB tem-
perature as a function of direction, (n) T /T (n). We will be interested in the statistical
correlations between temperature fluctuations in two different directions n and n0 (see Fig. 13),
averaged over the entire sky. If the initial conditions are statistically isotropic, then we expect
these correlations only to depend on the relative orientation of n and n0 . In that case, we can
write the two-point correlation function as
 X 2l + 1
(n) (n0 ) =

Cl Pl (cos ) , (5.1)
4
l

where n n0 cos and Pl are Legendre polynomials. The expansion coefficients Cl are the
angular power spectrum (cf. Fig. 14). If the fluctuations are Gaussian, then the power spectrum
contains all the information of the CMB map. In this section, we would like to understand how
the spectrum of inflationary initial conditions evolved into the observed power spectrum of CMB
anisotropies: Z
transfer function
2R (k) Cl d ln k 2l (k) 2R (k) .

The transfer function l (k) that relates 2R (k) to Cl captures both the evolution of the fluctu-
ations in the primordial plasma and the projection of the anisotropies onto the sky. Since the
primordial spectrum is rather featureless, it is the transfer function that leads to all the non-
trivial structure in the CMB power spectrum. Our goal in this section is to derive l (k) from
first principles. This wont be easy and will involve a lot of interesting physics.

Figure 13. Temperature inhomogeneities at recombination become anisotropies on the observers sky.

46
 100

80

60

40

20

10 100 1000

Figure 14. From temperature maps to the angular power spectrum (figure adapted from [4]). The
original temperature fluctuation map (top left) corresponding to a simulation of the power spectrum (top
right) can be band filtered to illustrate the power spectrum in three characteristic regimes: the large-scale
regime, the first acoustic peak where most of the power lies, and the damping tail where fluctuations are
dissipated.

The outline of this section is illustrated by the following schematic:

evolution (5.4) free-streaming (5.2) power spectrum (5.1+5.5)
R(0, k) T (n) Cl

projection (5.3)
ve

We will first show how the free-streaming of photons relates the observed temperature fluctua-
tions to fluctuations on the surface of last-scattering. We will see that in order to predict the
CMB anisotropies we need to know the fluctuations in the photon density ( ), the gravitational
potential () and the electron velocity (ve ) at the time of decoupling ( ). By studying the
evolution of the coupled perturbations before recombination, we will link these quantities to the
initial curvature perturbation R(0, k). The projection of the inhomogeneities at decoupling onto
the observers sky lead to non-trivial angular variations, even for a single Fourier mode k. The
final anisotropy spectrum is obtained by summing over many Fourier modes weighted by the
spectrum of the initial conditions.

5.1 Anisotropies from Inhomogeneities

We are interested in the temperature anisotropies observed today (0 ) at our location (x0 0)
as a function of the direction n on the sky. Since a photon observed in the direction n had to be

47
travelling in the direction p = n, we have
T
(n) (n) = (0 , x0 , p = n) . (5.2)
T
This may also be written as

d3 k ikx0
Z
(n) = e (0 , k, n)
(2)3/2
d3 k ikx0 X
Z
= e (i)l l (0 , k)Pl (k n)
(2)3/2 l
d3 k ikx0 X
Z
= e (i)l l (k)R(k)Pl (k n) , (5.3)
(2)3/2 l

where we introduced the Fourier components of the inhomogeneous temperature field in the first
line, expanded them into multipole moments in the second line, and introduced a transfer function
for the linear evolution in the third line:

l (0 , k)
l (k) . (5.4)
R(k)

The transfer function l (k) provides the map from the initial power spectrum of curvature
perturbations to the angular power spectrum of temperature anisotropies:
Z
4
Cl = d ln k 2l (k) 2R (k) . (5.5)
(2l + 1)2

Derivation.We substitute (5.3) into the two-point function (5.1) and perform a series of manipula-
tions
d3 k d3 k0 i(k+k0 )x0 X X
Z
0
(n)(n0 ) = (i)l+l l (k)l0 (k 0 )


e
(2) (2)3/2
3/2
l l0

hR(k)R(k0 )i Pl (k n)Pl0 (k0 n0 ) (5.6)

d3 k 2 2 2
Z XX 0
= 3 3
R (k) (i)l+l l (k)l0 (k)Pl (k n)Pl0 (k0 n0 ) (5.7)
(2) k 0 l l
Z Z
1 dk 2 XX 0
= R (k) (i)l+l l (k)l0 (k) d2 k Pl (k n)Pl0 (k n0 ) (5.8)
4 k 0
l l | {z }
4 0
= Pl (n n ) ll0
2l + 1
X 2l + 1  4 Z 
= d ln k 2
l (k) 2
R (k) Pl (n n0 ) . (5.9)
4 (2l + 1)2
l | {z }
= Cl

This confirms (5.5).

48
5.2 Free-Streaming
As a first step to relating the observed temperature anisotropies to the initial conditions, we
wish to connect them to perturbations at the point x x0 + (0 )n of the last-scattering
surface (see Fig. 15). This can be done by integrating the Boltzmann equation (4.36) along the
corresponding line-of-sight.

Figure 15. Coordinates of the line-of-sight solution. The gray regions correspond to periods when the
metric potentials and evolve in time and lead to non-zero contributions to the ISW effect.

Optical Depth and Visibility

To find the line-of-sight solution of the Boltzmann equation it will be helpful to introduce two
auxiliary concepts:
Optical depth.The optical depth between times and 0 is defined as
Z 0
() ( 0 ) d 0 . (5.10)

Physically, this describes the opacity of the universe at a given time, when seen from today.
(The probability of no scattering as a photon travels from to 0 is e .) Mathematically,
the function e will act as in integration factor for the Boltzmann equation. This see this,
note that
d d
+ = (e ) , (5.11)
d d
so that part of the Boltzmann equation turns into a total derivative after introducing the
optical depth.

Visibility function.The visibility function is defined as

g() ()e () . (5.12)

Physically, this corresponds to the probability density for a photon to last scatter at the
time . The shape of the visibility function is shown in Fig. 16. We see that it is sharply
peaked around the time of recombination, with a width of about 10 Mpc (or z 10).

49
Figure 16. Ionization fraction Xe () and visibility function g(). The ionization fraction is defined as the
ratio of free electrons to the total number of protons (excluding those in helium). The first two steps in Xe
correspond to helium recombination. After helium recombination, we have Xe 1, until Xe drops sharply
at hydrogen recombination. The visibility function is sharply peaked near hydrogen recombination.

Line-of-Sight Solution
With the help of these definitions, the Boltzmann equation (4.36) becomes
d d ln 
S e
 
(e ) = S , where + g 0 n ve . (5.13)
d d
For scalar fluctuations, the geodesic equation (4.22) implies
d ln  d
e = e + e ( + )
d d
d
= (e ) + g + e ( + ) . (5.14)
d
The Boltzmann equation for scalar fluctuations can then be written as
d 
S e ( + ) + g 0 + n ve .
  
e ( + ) = S , where (5.15)
d
Integrating (5.15) along the line-of-sight, we find
Z 0
(0 , x0 , n) = d 0 S( 0 , x0 + (0 0 )n, n) . (5.16)
0

To arrive at (5.16) we have used (0 ) = 0 and (0) , and dropped the unobservable
monopole (0 , x0 ) on the left-hand side. Notice that the terms in S which are proportional to
the visibility function g() are localized near the last-scattering surface, while terms proportional
to e () have an integrated effect until today.

50
Instantaneous Recombination
To get intuition for this solution, let us approximate recombination as occurring instantaneously.
In that case, g() = D ( ) and e = H( ), where H is the Heaviside function. We
then find
Z 0
(n) (0 , x0 , n) (0 + ) (n ve ) + d 0 ( + ) . (5.17)


SW Doppler ISW

The first two terms on the right-hand side are evaluated at the time and the position x
x0 + (0 )n. The Sachs-Wolfe (SW) term is a combination is the intrinsic temperature
fluctuation at the surface of last-scattering and an additional gravitational redshift arising when
the photons climb out of a potential well at last-scattering. The Doppler term describes the
shift in the photon energy created by the scattering off moving electrons. Finally, the integrated
Sachs-Wolfe (ISW) term captures the effect of gravitational redshifting from the evolution of the
potentials along the line-of-sight. As we have seen in Section 2, for most of the history of the
universenamely, when the universe was matter-dominatedthe gravitational potentials were
constant and therefore didnt contribute to the ISW term. At early times, the residual amount
of radiation gives 6= 0 which results in a non-zero early ISW effect. Similarly, at late times,
dark energy becomes relevant leading to a late ISW effect. These two regimes are indicated by
the gray regions in Fig. 15.

Sachs-Wolfe term.On large scales, the Sachs-Wolfe term dominates the CMB spectrum. Let us
evaluate this term for super-Hubble scales and adiabatic initial conditions. Decoupling occurs during
the matter era, so we use 2 2 = c = b = 34 [cf. eqs. (2.54) and (2.55)]. The observed
CMB anisotropies on large scales therefore are
 
1 1 1
+ = = , . (5.18)
4 3 8

We see that the gravitational redshifting has won over the intrinsic temperature fluctuation. This
means that an overdensity at last scattering (, > 0), corresponding to a potential well ( < 0),
leads to a cold spot in the CMB map ( < 0). Conversely, a hot spot corresponds to an underdensity
at last scattering.

5.3 Projection
To find the transfer function (5.4) it will be useful to transform the line-of-sight solution (5.16)
to Fourier space and extract its multipole moments. The algebraic details are given in the box
below. Eventually, we find
Z 0
0
l (k) = (0 + ) jl (k ) (ve ) jl (k ) + d 0 ( + ) jl (k) , (5.19)


SW Doppler ISW

51
where (f ) f ( , k)/R(k) and ( 0 ) 0 0 is the conformal distance along the line-of-sight.
The Bessel functions in (5.19) describe the projection from Fourier space to harmonic space (see
Fig. 17).

ISW

Figure 17. From inhomogeneities to anisotropies (figure adapted from [4]).

Derivation.We write the source term S in a Fourier decomposition

d3 k
Z
0
S( 0 , x0 + (0 0 )n, n) = S( 0 , k, n) eik(x0 +( )n) . (5.20)
(2)3/2

The exponential eikn can be written in a Rayleigh plane wave expansion

X
eikn = (i)l (2l + 1)jl (k)Pl (k n) . (5.21)
l

Powers of (k n) in the source function can be replaced by derivatives of the Bessel function. For
example,
d X
i(k n) eikn = eikn = (i)l (2l + 1)jl0 (k)Pl (k n) , (5.22)
d(k)
l

where the prime denotes a derivative with respect to the argument of the Bessel function. The Fourier
transform of (5.16) therefore is
Z 0 h i
l (0 , k) = (2l + 1) d 0 g(0 + )jl (k) gve jl0 (k) + e ( + ) jl (k) . (5.23)
0

52
 Assuming instantaneous recombination, and extracting the initial condition R(k), we get
Z 0
0
l (k) = (0 + ) jl (k ) (ve ) jl (k ) + d 0 ( + ) jl (k) . (5.24)

This confirms (5.19).

Substituting (5.19) into (5.5), in principle leads to six terms: the power spectrum of the SW
term, ClSW , that of the Doppler term, ClD , that of the ISW term, ClISW , and three cross spectra.
The cross spectra and the ISW contribution are small, so we will ignore them for the moment.
Ignoring the ISW contribution, we can write the transfer function as

(0 + )
TSW (k) .
R(k)
l (k) = TSW (k)jl (k ) + TD (k)jl0 (k ) , where (5.25)
(ve )
TD (k) .
R(k)

The functions TSW (k) and TD (k) describe the evolution of the primordial perturbations (R)
into perturbations in the matter ( , ve ) and the metric () at recombination. These transfer
functions can be determined approximately from the analytic results of 5.4 or more precisely by
numerically solving the Boltzmann-Einstein equations (see Fig. 18).

Sachs-Wolfe

Doppler

0.001 0.01 0.1

Figure 18. Numerical solutions for the transfer functions TSW (k) and TD (k).

For large values of l, the spherical Bessel function jl (x) is very peaked near x l (cf. Fig. 19),
and therefore acts like delta-functions in the integral (5.5). The derivative of the Bessel function
jl0 (x) is not as sharply peaked at x l, so the projection from wavenumber k to multipole l is
less sharp for the Doppler term. We will indicate this by writing x l, rather than x l. The

53
 10 100 500 1000

Figure 19. Spherical Bessel functions jl (x) for l = 10, 100 and 500.

Sachs-Wolfe and Doppler contributions to the power spectrum can therefore be written as
 l
2
l(l + 1)ClSW TSW (k) 2R (k) , k (l) , (5.26)

kk (l) 0

l(l + 1)ClD TD2 (k) 2R (k) . (5.27)

kk (l)

The reason for the factor of l(l + 1) on the left-hand side is explained in the box below. We see
that power spectra ClSW and ClD are characterized by the squares of the transfer functions.

Sachs-Wolfe term.For super-Hubble modes at recombination the Sachs-Wolfe transfer function is a

constant and the Doppler transfer function vanishes. For scale-invariant initial conditions, 2R (k) =
As , the power spectrum then becomes
Z Z
SW

2 2 4
Cl Cl = 4 d ln k TSW (k)jl (k ) R (k) As d ln k jl2 (k ) . (5.28)
25
If we make use of the standard integral
Z
1
d ln x jl2 (x) = , (5.29)
0 2l(l + 1)

we find
4
l(l + 1)Cl = As = const. (5.30)
25
This explains why CMB spectra are usually plotted as Dl l(l + 1)Cl .

54
5.4 Evolution
Our final task is to compute the transfer functions TSW (k) and TD (k) in (5.25). This requires us
to evolve the coupled fluctuations of photons, electrons, baryons and dark matter in a perturbed
spacetime. For most of the evolution the photons are tightly coupled to the electrons, so the
Boltzmann equation for the photons simplifies as seen in 4.3. This tight-coupling approxima-
tion breaks down close to recombination (and below the photon diffusion scale), but otherwise
describes the photon dynamics very well. At the same time, the coupling between electrons and
protons remains strong throughout, so electrons and protons (or baryons, more generally) form
a single electron-baryon fluid. For simplicity, we will refer to this as the baryon fluid and use vb
for the common bulk velocity of electrons and baryons.

5.4.1 Photon-Baryon Fluid

For convenience, let me repeat the evolution equations (4.57) and (4.58) for the monopole and
dipole of the photon distribution,
1
0 = k1 + , (5.31)
3
1 = k0 k (1 + vb ) . (5.32)

These need to be supplemented by an evolution equation for the baryon velocity vb , which we
will derive next.
The trick is to exploit the conservation of the stress tensor of the combined photon-baryon
fluid. Ignoring the baryon pressure, the total momentum density q i T i 0 can be written as

q i = ( + P )vi + (b + Pb )vbi
b h2
  
4 3 b a
= vi + Rvbi ,


R = 0.6 . (5.33)
3 4 0.02 103
The fractional contribution from the baryons is characterized by the dimensionless parameter R.
This parameter is small at early times, but grows linearly with a and becomes order one around
the time of recombination. The total energy density perturbation is
 
4
= + Rb . (5.34)
3
The conservation of the total stress tensor of the photon-baryon fluid, T = 0, then implies
[cf. eqs. (2.17) and (2.18)]:

= 3H( + P ) + 3( + P ) i q i , (5.35)

q i = 4Hq i ( + P )i i P . (5.36)

Substituting (5.33) and (5.34), and using (5.31) and (5.32) to eliminate the time derivatives of
= 40 and v = 1 , we find

b = kvb + 3 , (5.37)

vb = Hvb k (1 + vb ) . (5.38)
R

55
The first equation is just the usual continuity equation for pressureless matter (again, no surprise
for Thomson scattering at linear order). The second equation is the usual Euler equation for
pressureless matter, with an extra scattering term which tries to equalize the baryon and photon
bulk velocities. Note the factor of 1/R in the scattering term of (5.38) which isnt present
in (5.32). This difference is easily understood: for a given bulk velocity the momentum density
of the baryons is larger than that of the photons by a factor of R, but the momentum exchange
due to scattering is the same.
Let us rearrange (5.38) to get

R 
vb = 1 vb + Hvb + k . (5.39)

In the tight-coupling limit, the second term on the right-hand side is much smaller than the
first and we have vb 1 to lowest order. However, since vb is multiplied by in (5.32), we
need the next-to-leading order solution for vb . We obtain this by substituting vb 1 into the
right-hand of (5.39). This gives

R 
vb 1 + 1 + H1 k . (5.40)

Substituting this into (5.32), we get

HR k
1 = 1 + 0 k . (5.41)
1+R 1+R
Combining (5.31) and (5.41), we get an oscillator equation for the evolution of the temperature
monopole 0 (or ):

HR 1 R
0 + 0 + c2s k 2 0 = k 2 + + , (5.42)
1+R 3 1+R

pressure gravity

where we have defined the sound speed of the photon-baryon fluid as

1
c2s . (5.43)
3(1 + R)

Equation (5.42) is the master equation describing the entire CMB phenomenology. The most
important terms in the equation are the photon pressure term on the left-hand side and the
gravitational forcing term on the right-hand side. In addition, we have a friction term proportional
to the baryon density R on the left-hand side and two terms related to time dilation effects on
the right-hand side. The metric potentials and are determined by the Einstein equations
(which include important contributions from dark matter).

56
 large scale

I
frozen
II
acoustic
oscillations
III decoupled
small scale diffusion
damping

eq dec today

I II III Planck
ACT
SPT

Figure 20. Top: Illustration of the relevant length scales (or momentum scales) involved in the dynamics
of the primordial plasma fluctuations. Bottom: Corresponding regimes in the angular power spectrum.

5.4.2 Acoustic Oscillations

Before discussing the solution to (5.42), let us highlight the relevant scales of the problem (Fig. 20):

Hubble radius.As we have shown in 2.4, metric perturbations are frozen outside the
(comoving) Hubble radius rH (aH)1 . Modes therefore only start evolving in interesting
ways after they cross the Hubble radius.

Sound horizon.Photon fluctuations, in fact, remain frozen until they cross the sound
horizon, rs cs (aH)1 , after which they start to oscillate. At early times, R  1 cs

1/ 3, and the sound horizon is nearly equal to the Hubble radius. Just before decoupling,
however, the baryon fraction R becomes significant and cs goes to zero. The sound horizon

57
 is then significantly smaller than the Hubble radius. After decoupling, the concept of the
sound horizon ceases to exist.

Diffusion scale.The diffusion scale rD is important for understanding the damping of

CMB fluctuations on small scales. Above the diffusion scale the photon fluctuations can be
treated as a fluid and the analysis of this section is applicable. Below the diffusion scale,
the tight-coupling approximation breaks down and we need a more careful treatment of
the Boltzmann equation for the photons. To estimate the diffusion scale, think of a photon
performing a random walk from collision to collision, with a typical step length given by the
mean free path, `p 1 . Consider a short enough interval so that `p can be treated
as constant. The photon will then random walk a mean-squared distance `p N , where the
number of collisions is N = /`p . Integrating this from = 0 gives the squared diffusion
length Z
2
rD 1 d 0 . (5.44)
0
At early times, the diffusion length is very small (the photon mean free path is very small),
but around recombination it becomes very larger (comparable to rH ). The longer a mode
spends in the diffusive regime the more damped the corresponding CMB fluctuations will be.

As an instructive first approximation, let us ignore time variations in the potentials and
and also ignore the evolution of the baryon-photon momentum ratio R compared to changes at
the oscillation frequency = cs k. Equation (5.42) then reduces to

k2
0 + c2s k 2 0 = . (5.45)
3
This is the equation of a simple harmonic oscillator with a constant gravitational forcing term.
Its solution is
1
0 (, k) = [0 (0, k) + (1 + R)(k)] cos(krs ) + 0 (0, k) sin(krs ) (1 + R)(k) , (5.46)
kcs
R
where rs = 0 cs d 0 cs . For adiabatic initial conditions, we have 0 (0) = 0, so only the cosine
solution contributes to the acoustic oscillations

0 (, k) = [0 (0, k) + (1 + R)(k)] cos(krs ) (1 + R)(k) . (5.47)

During the tight-coupling regime, the velocities of all components are equal: ve = vb = v = 1 .
The photon continuity equation (5.31) relates the photon velocity to the time derivative of the
density, i.e. k1 = 30 . For the approximate solution with constant R, this implies

1 (, k) = 3 [0 (0, k) + (1 + R)(k)] cs sin(krs ) . (5.48)

Evaluating (5.47) and (5.48) at = , the transfer functions in (5.25) can be written as

TSW (k) = A(k) cos(krs, ) + B(k) , (5.49)

TD (k) = 3cs A(k) sin(krs, ) , (5.50)

58
where we have defined
0 (0, k) + (1 + R)(k) (k)
A(k) , B(k) R . (5.51)
R(k) R(k)

Initial condition.To set approximate initial conditions for and 0 we could extrapolate matter
domination into the past, which gives (0, k) = 35 R(0, k) and 0 (0, k) = 25 R(0, k) for adiabatic
initial conditions. Using (5.51), this implies

1 + 3R 3R
A(k) = , B(k) . (5.52)
5 5

Let us discuss some of the physical effects encoded in the solution (5.47):
Gravitational infall and redshift.At early times, photons dominate the fluid and R 0.
In this limit, the solution (5.47) becomes

0 (, k) = [0 (0, k) + (k)] cos(krs ) (k) . (5.53)

This represents a harmonic oscillator with a zero-point which has been displaced by gravity.
The zero-point represents the state at which gravity and pressure are balanced. The dis-
placement > 0 yields hotter photons in the potential well since gravitational infall not
only increases the number density of photons but also their energy through gravitational
blueshifts. However, as we have seen above, after decoupling the photons lose energy when
they have to climb out of potential wells. This redshift precisely cancels the blueshift.
The observed temperature perturbation therefore is

0 (, k) + (k) = [0 (0, k) + (k)] cos(krs ) . (5.54)

Different Fourier modes will be captured at different phases in their evolution at the time
of decoupling (see Fig. 21). A discrete set of wavenumbers, kn = n/rs ( ), correspond
on extrema at decoupling. These become the peaks of the CMB spectrum.5

Baryon loading.Let us now restore the effects of the baryons in the plasma. Baryons add
extra inertial and gravitational mass to the photon-baryon fluid. This decreases the sound
speed and changes the balance of pressure and gravity. Gravitational infall now leads to
greater compression of the fluid in a potential well, i.e. a further displacement of the oscil-
lation zero-point (see Fig. 22). Since the redshift is not affected by the baryon content, this
relative shift remains after last-scattering to enhance all peaks from compression over those
from rarefaction. If the baryon-photon ratio R were constant, the effective temperature
perturbation would become

0 (, k) + (k) = [0 (0, k) + (k)] cos(krs ) R(k) . (5.55)

We see that compressional peaks are a factor of (1 + 6R) larger than in the R = 0 case
[cf. eq. (5.52)]. In reality, the effect is reduced since R 0 at early times. Notice that
velocity oscillations in (5.48) are symmetric around zero unlike the temperature oscillations.
5
Since the power spectrum is proportional to the square of the fluctuation, both maxima and minima contribute
peaks in the spectrum.

59
 2nd peak

1st peak

0.2 0.4 0.6 0.8 1.0

Figure 21. Acoustic oscillation of two representative Fourier modes (figure adapted from [5]). Time is
measured by the sound horizon rs = cs relative to the sound horizon at recombination rs, . Wavenumbers
that reach extrema in their effective temperature 0 + at rs, form a harmonic series kn = n/rs, .

0.2 0.4 0.6 0.8 1.0

Figure 22. Acoustic oscillations with baryons (figure adapted from [5]). Baryons add inertia to the
photon-baryon plasma displacing the zero-point of the oscillation and making compressional peaks (min-
ima) larger than rarefaction peaks (maxima). The absolute value of the fluctuation in the effective tem-
perature is shown in dotted lines.

Radiation driving.During the radiation era the potentials and become time depen-
dent inside the horizon (cf. Fig. 6). Counterintuitively, the decaying potential actually
enhances temperature fluctuations through its near resonant driving force. Since the po-
tential decays after sound horizon crossing, it drives the first compression without a
counterbalancing effect on the subsequent rarefaction stage. The higher peaks began their
oscillation in the radiation-dominated universe and therefore have an enhanced amplitude
(see Fig. 23).

60
 damp
ing

dri
vin
g
5 10 15 20

Figure 23. Acoustic oscillations with gravitational forcing and dissipational damping. For a mode that
enters the sound horizon during radiation domination, the gravitational potential decays after horizon
crossing and drives the acoustic amplitude higher. As the photon diffusion length increases and becomes
comparable to the wavelength, viscosity is generated leading to dissipation.

Figure 24. Diffusion damping of photons and baryons (figure adapted from [6]).

5.4.3 Diffusion Damping

By assuming perfect tight coupling (mean free path = zero) and instantaneous recombination,
our solution is still missing some important physics. Including the effects of a finite mean free
path for the photons and a finite duration of recombination leads to the damping of small scale
fluctuations6 see Fig. 24. The former effect is sometimes called Silk damping.
To analyze the diffusion damping, we must extend the equations for the moments of the photon
distribution beyond the tight-coupling limit. In particular, we need to allow for a non-negligible
quadrupole 2 (or photon anisotropic stress ) in the equations. The task is simplified by
the fact that diffusion damping is relevant on small scales where the dynamical timescale of the
fluctuations is much shorter than the expansion timescale. Gravitational effects are therefore
subdominant and can be ignored, i.e. we can drop and everywhere. Equations (5.31) and
6
Essentially, fluctuations on scales smaller than the mean free path are damped; just like you cant have sound
waves with wavelengths smaller than the mean free path in air.

61
(5.32) can therefore be written as
1
0 = k1 , (5.56)
3
 
2
1 = k 0 2 (1 + vb ) , (5.57)
5
where we have included 2 in the Euler equation (5.57). In addition, we have an evolution
equation for the quadrupole
2 = 2k1 2 , (5.58)
where we have dropped a term proportional to 3 , since it is still suppressed relative to 1 .
Finally, eq. (5.39) implies
R 
1 + vb = vb + Hvb . (5.59)

We will solve these equations in a WKB approximation, where the time dependence of the solution
satisfies the following ansatz R
{l , vb } ei d . (5.60)
In the tight-coupling limit, we have cs k, but now we are looking for the leading correction
beyond tight-coupling. In particular, we want to compute the small imaginary part of that
characterized damping. Since damping occurs on small scales, of high frequencies, we have
vb = ivb  Hvb , so eq. (5.59) can be written as
"  #
iR 1 R 2
  
iR
vb 1 1 + 1 1 . (5.61)

Similarly, in eq. (5.58) we can drop 2 relative to 2 , so we have 2 (2k/)1 . Finally,

eq. (5.56) reads i0 = 31 k1 . Inserting the above into (5.57), we obtain the following disper-
sion relation
k2
 
2 i 2 2 8 2
(1 + R) R + k = 0. (5.62)
| {z 3} 27
| {z }
tight-coupling correction
Let us substitute = cs k + into (5.62) and expand in small . This leads to
 
i 2 2 8
= c R + k2 . (5.63)
2(1 + R) s 27
The time-dependence of the perturbations therefore is
k2
 Z   
{0 , 1 } exp ik d cs exp 2 , (5.64)
kD
where the damping wavenumber is

R2
Z  
1 d 8
2 + . (5.65)
kD () 0 6(1 + R)() 1 + R 9
Including polarization corrections to the scattering would give the same result with 8/9 16/15.
2 2
We see that diffusion damping can be modelled by a ek /kD envelope to the solutions discussed
in the previous section.

62
5.5 Power Spectrum
Figure 25 shows a sketch of the different contributions to the CMB power spectrum. The shape
of the Sachs-Wolfe and Doppler contributions can be understood from the analytical treatment
of the previous section. Note the the velocity v vb vanishes outside the sound horizon and
that the Doppler effect is therefore suppressed on large scales. The Sachs-Wolfe transfer function
is a constant on large scales and the plateau in l(l + 1)Cl for small l is therefore a direct reflection
of the scale-invariant initial conditions. The late ISW effect leads to a small rise of the plateau.
This is a measure of dark energy. The early ISW adds extra power near the first peak. Finally,
diffusion damping suppresses all contributions to the power spectrum at large l.

Sachs-Wolfe: Late ISW

Doppler: Early ISW
Potential:

Figure 25. Sketch of the different contributions to the CMB power spectrum (figure adapted from [7]).

63
5.6 Cosmological Parameters
In the final section of these notes, we will discuss how the shape of the CMB power spectrum
depends on cosmological parameters. The CDM model has six parameters:

As : Normalization

ns : Spectral tilt

b h2 : Baryon density

m h2 : Matter density

: Dark energy density

: Optical depth

Our simplified analysis is sufficient to understand how the CMB power spectrum responds to
changes in these cosmological parameters. These effects are used to determine the values of these
parameters from observations of the CMB anisotropies.

Normalization.Changing the amplitude of the primordial fluctuations obviously just shifts

the CMB spectrum up and down. The overall normalization of the CMB spectrum therefore
fixes the amplitude of the initial conditions: As = (2.190.05)109 (at k? = 0.05 Mpc1 ).

Tilt.The tilt does what the name promises: it tilts the CMB spectrum. For ns = 1 the
spectrum l(l+1)Cl has equal power on all scales, while for ns > 1 (ns < 1) it has more power
on small (large) scales. The Planck data implies ns = 0.96 0.01 (at k? = 0.05 Mpc1 ).

Curvature.The positions of the peaks of CMB spectrum depend on the projected size of
the sound horizon at recombination, rs, = cs . From eq. (5.47) we infer kpeak rs, = n
(for integer n), which corresponds to

lpeak = n , (5.66)
rs,

where is the comoving distance to the surface of last-scattering. We have been restricting
ourselves to a flat universe for which = 0 . More generally, the (angular diameter)
distance to the last-scattering surface will depend on the geometry of the universe as pa-
rameterized by the curvature parameter k 1 m . Geodesics in a universe with
negative spatial curvature, k > 0, diverge and a fixed sound horizon scale gets projected
to a smaller angular scale (larger l). Quantitatively, the angular diameter distance scales
approximately as (1 k )0.45 . The dependence on the peak positions on the spatial
geometry is illustrated in panel (a) of Fig. 26.

Baryons.We have explained above how baryons baryons affects the ratio of successive
peaks. The ratio of successive peak heights is (1 + 6R)2 , and thus is bigger the larger the
baryon-to-photon ratio is. This is illustrated in panel (c) of Fig. 26. The relative heights
of the peaks therefore measures b h2 .

64
 100 (a) Curvature (b) Dark Energy

80

60

40

20

0.2 0.4 0.6 0.8 1.0 0.2 0.4 0.6 0.8

100 (c) Baryons (d) Matter

80

60

40

20

0.02 0.04 0.06 0.1 0.2 0.3 0.4 0.5

10 100 1000 10 100 1000

Figure 26. The CMB power spectrum as a function of cosmological parameters (figure adapted from [4]).

Matter.We mentioned that scales smaller than the horizon at matter-radiation equality
have a boosted amplitude. In the angular power spectrum the separation between modes
p
that experience this feedback and those that do not occurs at leq ? /eq 1 + zeq . (A
typical value for this scale is leq 100.) Lowering the matter density shifts matter-radiation
equality to later times, zeq = 2.3 104 m h2 , i.e. closer to recombination. This affects the
forcing term for the photon-baryon oscillations. In particular, the forcing term is stronger
near equality, so the heights of the peaks are higher when the matter density is lower. This
is illustrated in panel (d) of Fig. 26. The height of the peaks (relative to the low-l plateau)
therefore measures m h2 . (In particular, the amplitude of the third peak is a good measure
of the dark matter density.)

Dark energy.When dark energy takes over, the gravitational potential stops being time-
independent, 6= 0. This leads to the late ISW effect which boosts the power of the low-l
modes (see panel (b) of Fig. 26).

65
 scalar

tensor

Figure 27. Comparison of the power spectra of CMB temperature anisotropies created by scalar and
tensor perturbations.

Reionization.The universe became reionized at late times, around z & 6. This increases
the optical depth to last-scattering. On scales smaller than the horizon after reionization,
the CMB anisotropies get suppressed by a factor of e due to the rescattering of photons.
This effect is somewhat degenerate with an ns > 1 tilt of the primordial spectrum. However,
the effect on the polarization spectrum is distinct which allows ns and to measured
separately.

There are many interesting extensions of the baseline CDM model. Here, we consider two
particularly well-motivated examples: the tensor-to-scalar ratio, r, and the effective number of
neutrinos, Neff .

Tensors.In the problem set you will derive the spectrum of CMB anisotropies created by
pure tensor perturbations. The result is shown in Fig. 27. Since gravitational waves decay
inside the horizon, the tensor contribution is only significant on large scales (small l). The
measurement of the large-scale CMB temperature anisotropy spectrum constrains the tensor
amplitude to r < 0.3. More sensitive constraints of r require the measurement of CMB
polarization, especially its B-mode type. This will be described in future improvements of
these notes.

Neutrinos.The total energy density in relativistic species is often defined as

" #
7 4 4/3
 
r = 1 + Neff , (5.67)
8 11

where is the energy density of photons. The parameter Neff is referred to as the effective
number of neutrinos, although there may be contributions to Neff that have nothing to do
with neutrinos. The Standard Model predicts Neff = 3.046 from neutrinos and the current

66
 constraint from the Planck satellite is Neff = 3.04 0.18. Deviations from the standard
value may arise if new physics at high energies leads to additional weakly coupled light
species (cf. Fig. 28). These effects are detectable because they change the damping tail [8]
and shift the phase of the acoustic peaks [9, 10].

Goldstone boson
Weyl fermion
Gauge boson

Planck limit

0.054
0.047
0.027
CMB-S4 limit

Figure 28. Contributions of a single thermally-decoupled Goldstone boson, Weyl fermion or massless
gauge boson to the effective number of neutrinos, Neff , as a function of its freeze-out temperature TF
and freeze-out time tF . Shown are also the current 2 sensitivity of the Planck satellite and an (optimistic)
estimate of the sensitivity of a future CMB-S4 mission.

67
Part III
Problems and Solutions

6 Problem Sets

6.1 Problem Set 0: Preliminaries

1. Warmup Questions

(a) What is conformal time? Why is it useful?

(b) How do the energy densities in radiation (r ), matter (m ) and a cosmological constant ( )
evolve with the scale factor a(t)?

(c) What is a(t) for a flat universe dominated by radiation, matter or a cosmological constant?
What is a() for the same cases?

(d) What is the redshift of matter-radiation equality if r = 9.4 105 and m = 0.32?

(e) Show in the context of expanding FRW models that if the combination + 3P is always
positive, then there was a Big Bang singularity in the past.

2. Horizon Problem
The (comoving) particle horizon at the time t equals the amount of conformal time between
ti = 0 and t. This can be written as
Z t Z a
dt da
= = 2
.
0 a(t) 0 a H(a)

In a flat universe filled with matter, radiation and dark energy, we have

H 2 = H02 r a4 + m a3 + ,
 

where H0 is the Hubble constant and i are the densities today (in units of the critical density):
r = 9.4 105 , m = 0.32, and = 1 m r .

(a) Ignoring the contribution from dark energy, show that

2 hp i
=p a + aeq aeq ,
m H02

where aeq = r /m 34001 is the scale factor at matter-radiation equality. What is the
conformal time today (a0 1)? What is it at recombination (a 11001 )? You may
leave your answer in terms of the Hubble constant. Calculate the angle subtended by the
horizon at recombination.

68
(b ) Dark energy leads to a small correction to the above result. Use your favorite software (say
Mathematica or Maple) to repeat the previous analysis. What was the percentage error
that you made in part (a)?

(c) Using a spacetime diagram, or otherwise, explain why the above results imply a horizon
problem for the standard Big Bang cosmology. How does inflation solve the horizon prob-
lem?

3. Slow-Roll Inflation
The equations of motion of the homogeneous part of the inflaton are
1
00 + 3H0 + V, = 0 , 2
3Mpl H 2 = (0 )2 + V ,
2
where the primes denote derivatives with respect to time t.

(a) For the potential V () = 21 m2 2 , show that the inflationary solution in the slow-roll ap-
proximation are
r " #
2 2i 2 (t)
(t) = i mMpl t , a(t) = ai exp 2 ,
3 4Mpl

where i > 0 is the field value at the start of inflation (ti 0).

(b) What is the value of when inflation ends? Find an expression for the number of e-folds.
4 , estimate the total number of e-folds of inflation.
If V (i ) Mpl

69
6.2 Problem Set 1: Perturbation Theory

[Note: Unless you feel very inspired, you only have to do two of the three problems.]

1. Evolution of Metric Perturbations

In the absence of anisotropic stress, the gravitational potential satisfies

+ 3H + (2H + H2 ) = 4Ga2 P , (?)

where P is the total pressure perturbation.

(a) Discuss the solution of (?) during both the radiation-dominated and matter-dominated eras.
Describe the evolution on super-Hubble scales, k  H, and on sub-Hubble scales, k  H.

(b) In the lectures, we introduced the comoving curvature perturbation

H
R = + q , with T 0 j j q ,
+ P
and proved that it doesnt evolve on superhorizon scales. By considering the time-space
Einstein equation, show that
+ H = 4Ga2 q ,
in the absence of anisotropic stress, so that = . Use the conservation of R to show
that the gravitational potential decreases by a factor of 9/10 on superhorizon scales in the
transition from radiation-dominated to matter-dominated.

2. Growth of Matter Perturbations

At early times, the universe was dominated by radiation (r) and pressureless matter (m). You
may ignore baryons.

(a) Show that the conformal Hubble parameter satisfies

H02 2m 1
 
2 1
H = + ,
r y y2

where y a/aeq is the ratio of the scale factor to its value when the energy density of the
matter and radiation are equal.

(b) For perturbations on scales much smaller than the Hubble radius, the fluctuations in the
radiation can be neglected. Assuming that evolves on a Hubble timescale, show that

m + Hm 4Ga2 m m 0 . (?)

[You may use any equations given in the lectures without proof.]

70
(c) Show that, in terms of the variable y, eq. (?) becomes
d2 m 2 + 3y dm 3
2
+ m = 0 .
dy 2y(1 + y) dy 2y(1 + y)
Hence verify that the solutions are

2 + 3y


m  
1+y+1 p
(2 + 3y) ln 6 1+y


1+y1
Determine how m grows with y for y  1 (RD) and y  1 (MD).

3. Cosmological Gravitational Waves

The line element of a FRW metric with tensor (gravitational wave) perturbations is
h i
ds2 = a2 () d 2 (ij + 2hij )dxi dxj ,
where hij is symmetric, trace-free and transverse. To linear order in hij , the non-zero
connection coefficients are
000 = H ,
0ij = Hij + 2Hhij + hij ,
ij0 = Hji + hij ,
ijk = j hi k + k hi j il l hjk .

(a) Show that the perturbation to the Einstein tensor has non-zero components
Gij = hij 2 hij + 2Hhij 2hij (2H + H2 ) .
[Hint: Convince yourself that the Ricci scalar has no tensor perturbations at first
order.]
(b) Combine the previous result with the perturbation to the stress tensor, Tij = 2a2 P hij
a2 ij , to show that the perturbed Einstein equation reduces to
hij + 2Hhij 2 hij = 8Ga2 ij .
(c) For the case where 2 hij = k 2 hij (i.e. a Fourier mode of the metric perturbation),
and assuming the anisotropic stress can be ignored, show that
k cos(k) sin(k)
hij
(k)3
is a solution for a matter-dominated universe (a 2 ).
(d) Show that the solution tends to a constant for k  1 and argue that such a constant
solution always exists for super-Hubble gravitational waves irrespective of the equation
of state of the matter. For the specifc solution above, show that well inside the Hubble
radius it oscillates at (comoving) frequency k and with an amplitude that falls as 1/a.
(This behaviour is also general and follows from a WKB solution of the Einstein
equation.)

71
6.3 Problem Set 2: Quantum Fluctuations

1. Slow-Roll Inflation
(a) Consider slow-roll inflation with a polynomial potential V () = 4p p , where p > 0 and
is a parameter with the dimension of mass. Show that the spectral index ns and the tensor-
to-scalar ratio r, evaluated at a reference scale k? corresponding to CMB fluctuations, are
2+p 4p
ns 1 = , r= ,
2N? N
where N? 50 is the number of e-folds between the horizon exit of k? and the end of
inflation. Which values of p are still consistent with current observations?

(b) Axions are promising inflaton candidates. At the perturbative level, an axion enjoys a
continuous shift symmetry, but this is broken nonperturbatively to a discrete symmetry,
leading to a potential of the form V () = 4 [1 cos(/f )], where f is the axion decay
constant. Using this as the inflationary potential, show that
eN? + 1 1
ns 1 = , r = 8 ,
eN? 1 eN? 1
2 /f 2 . Sketch this prediction in the n -r plane. Discuss the limit  1.
where Mpl s

2. The Lyth Bound

(a) Show that the tensor-to-scalar ratio predicted by slow-roll inflation is
At 8( 0 )2
r = 2 2 .
As Mpl H

(b) Show that the inflaton field travels a distance |e cmb | during (observable)
inflation
r
N r
= ,
Mpl 60 0.002
where N is the total number of e-folds between the time when the CMB scales exited
the horizon and the end of inflation. [You may assume that const. during inflation]
Comment on the implication of this result for observable gravitational waves. [Realistically,
we require r > 0.001 to have a fighting chance of detecting gravitational waves via CMB
polarisation.]

(c) Derive the following relationship between the energy scale of inflation, V 1/4 , and the tensor-
to-scalar ratio,
 2 1/4
3
V 1/4 = rAs Mpl .
2
Use As = 2.1 109 to determine V 1/4 for r = 0.01. How does that compare to the energy
scales probed by the LHC?

72
6.4 Problem Set 3: Boltzmann Equation

1. Boltzmann for Harmonic Oscillators

Consider a one-dimensional harmonic oscillator with energy

p2 1
E= + x2 .
2m 2
The distribution function of the harmonic oscillator depends on time t, position x, and momen-
tum p.

(a) Show that the collisionless Boltzmann equation is

f p f f
+ x = 0,
t m x p
and prove that the equilibrium distribution function is a function only of the energy E.

(b) By taking moments of the collisionless Boltzmann equation, derive the fluid equations for
the collisionless, one-dimensional harmonic oscillator in terms of the number density n, the
velocity v, and the pressure P , where
Z
1 dp p
Z
dp p2
Z
dp
n f, v f, P = 2
f nv 2 .
2 n 2 m 2 m

2. Boltzmann for Dark Matter

In this problem, you will derive the Boltzmann equation for cold dark matter and show that its
first two moments reproduce the fluid equations for dark matter.

(a) Show that the four-momentum of the dark matter particles is

P = (E/a)(1 ), (p/a)(1 + )pi ,

 

p
where E p2 + m2 , with p2 gij P i P j .

(b) Show that the Boltzmann equation for the dark matter distribution function fc is

p2
 
fc p fc p
+ p fc (H ) 2 + p = 0 .
E ln E E E

(c) Take the first two moments of the Boltzmann equation and show that they lead to the fluid
equations for the dark matter, in terms of

d3 p d3 p
Z Z
i 1 p
nc nc (1 + c ) = 3
fc and vc 3
fc pi .
(2) nc (2) E

[Hint: you may drop terms of order p2 /E 2 in the non-relativistic limit and use without
proof that d pi pj = (4/3) ij .]
R

73
6.5 Problem Set 4: Temperature Anisotropies

1. Anisotropies from Tensors

In this problem, you will derive the temperature anisotropies induced by primordial tensor modes,
i.e. metric perturbations of the form
h i
ds2 = a2 () d 2 (ij + hij )dxi dxj ,

where hij is transverse and traceless.

(a) By considering the geodesic equation in the presence of tensor perturbations, show that the
comoving energy of a photon evolves as
d ln  1
= hij pi pj ,
d 2

where pi is a unit vector in the direction of propagation of the photon.

(b) Assuming tight-coupling until recombination and instantaneous decoupling, show that the
line-of-sign solution for the photon temperature anisotropy is
1 0 0 d3 k
Z Z
0
(n) = d 3/2
hij ( 0 , k) ni nj eik( ) kn (?) ,
2 (2)

where () 0 .

(c) Consider a single gravitational wave with momentum k pointing in the z-direction and
expand hij into its two polarization modes,

X 1 1 i 0
1
hij (, k) h (, k)ij (k) , with  (z) = i 1 0 .

2 2
= 0 0 0

Show that r
8
 i j
ij (z) n n = Y2 2 (n) .
15
With this it can be shown that the contribution of the Fourier mode k = kz to the integral
in (?) is
r r
4 ik cos X jl (k)
h (, kz) Y22 (n) e = h (, kz) l Yl 2 (n) ,
15 2 (k)2
l
p
where l (i)l 2l + 1 (l + 2)!/(l 2)!. The contribution from a general Fourier mode k
is obtained by rotating the previous result. This gives
r
i j ikkn X jl (k) X
hij (, k) n n e = h (, k) l Dl (k) Ylm (n) .
2 (k)2 m m2
l

An explicit form of the Wigner D matrices wont be needed for the rest of this problem.

74
(d) Show that angular power spectrum of the tensor-induced anisotropies is
Z
4 2
d ln k l (k) 2h (k) ,

Cl = 2
(2l + 1)

where s
0
j2 (k 0 ) jl (k(0 0 ))
Z
3 (l + 2)!
l (k) = (2l + 1) k d 0 .
4 (l 2)! k 0 (k(0 0 ))2

(e) Discuss the shape of the spectrum. Show that a scale-invariant tensor spectrum, 2h (k)
const. leads to l(l + 1)Cl const. on large scales.

Useful results.You may use the following results without proof:

To linear order in hij , the connection coefficients relevant for this problem are

000 = H ,
1
0ij = Hij + Hhij + hij .
2

The l = 2, m = spherical harmonic is

r
1 15
Y22 (n) = sin2 e2i ,
4 2
where and are the polar coordinates of n.
Orthogonality of the Wigner D matrices implies
Z
l l0 4
dk Dm (k)Dm 0 0 (k) = ll0 mm0
2l + 1

The following integral may be useful

Z
j 2 (x) 4 (l 2)! 1
dx l 5 = .
0 x 15 (l + 2)! (l + 3)(l 2)

75
7 Solutions

7.1 Solutions 0: Preliminaries

1. Warmup Questions
To be discussed in the examples class.

2. Horizon Problem
(a) Ignoring the contribution from dark energy, we have
h i q
H 2 = H02 r a4 + m a3 a2 H = m H02 (aeq + a)1/2 .
Substituting this into the integral for the conformal time, we find
Z a Z a
da 1 da 2 hp i
= 2
= p p = p a + aeq aeq .
0 a H m H02 0 a + aeq m H02
Notice that this result has the right limits: at early times (a  1), we get a (as expected
during the radiation-dominated era), while at late times (a  aeq ), we have a1/2 (as
expected during the matter-dominated era). The conformal times today (a0 = 1) and at
recombination (a = 11001 ) then are
2
0 p ,
m H02
2 hp i
= p 1100 1 + 34001 34001 0.0175 .
0
m H02
The angle subtended by the horizon at recombination is (in radians)
2 2 0.0175
= = 0.036 2.0 .
0 1 0.0175
(b ) Including dark energy, we have
h i q
1/2
H 2 = H02 r a4 + m a3 + a2 H = m H02 aeq + a + a4 ,
where we have defined
1 (1 + aeq )m
= 2.12 .
m m
Substituting this into the integral for the conformal time, we find
Z a
1 a
Z
da 2 da 2
= 2
= p p = p I(a) .
0 a H m H02 2 0 a4 + a + aeq m H02
| {z }
I(a)
The integral I(a) has to be evaluated numerically. We find, I(1) 0.89 and I(11001 )
0.0175, so that
2 2 0.0175
= = = 0.040 2.3 .
0 0.89 0.0175
This is 15% larger than our previous result.

76
 (c) To be discussed in class.

3. Slow-Roll Inflation
(a) In the slow-roll approximation, (0 )2  V and |00 |  |3H0 |, the Klein-Gordon equation
and the Friedmann equation are

3H0 m2 ,
2 1
3Mpl H 2 m 2 2 .
2
These equations are invariant under , so without loss of generality we can take
> 0. We then get
m
H ,
6Mpl
r r
0 2 2
mMpl (t) i mMpl t ,
3 3
where i is the value of at the beginning of inflation, taken to be ti 0.
To solve for a(t), notice that
d ln a d ln a dt H 1
= = 0 2 .
d dt d 2 Mpl

Integrating from ti to t, we find the desired solution,

" #
2i 2 (t)
a(t) = ai exp 2 .
4Mpl

(b) The potential slow-roll parameter is

2 2
Mpl Mpl 2
  
V,
V =2 .
2 V

The end of inflation is defined via V (e ) 1 e = 2Mpl . The total number of e-folds is

2i 2e 2i 1
Ntot ln(ae /ai ) = 2 = 2 .
4Mpl 4Mpl 2
4 , then is of order M 2 /m, and
If the potential at the beginning of inflation is of order Mpl i pl
the total number of e-folds is of the order
2i Mpl 2
 
Ntot 2 .
Mpl m

This achieves sufficient inflation to cure the horizon and flatness problems if the inflaton
mass, m, is about an order of magnitude or more smaller than the Planck scale.

77
7.2 Solutions 1: Perturbation Theory

1. Evolution of Metric Perturbations

(a) During the matter-dominated era, we have 2H + H2 = 8Ga2 P = 0, so that (?) becomes
6
+ = 0 , (??)

where we have used H = 2/. Clearly, = const. is a solution (on all scales). To see that
it is the growing mode solution try the power law ansatz p . Equation (??) implies

p(p 1) + 6p = p(p + 5) = 0 p = 0, 5 .

Our two solutions therefore are 0 (growing mode) and 5 (decaying mode).
During the radiation-dominated era, we have
8G 2
2H + H2 = a = H2 ,
3
4G 1 2 
4Ga2 P = = 3H( + H) ,
3 3
so that
4 1
+ = 2 , ()
3
where we have used H = 1/. Notice terms without derivatives acting on have cancelled.
On super-Hubble scales, we can drop the term on the right-hand side of () and = const.
is again a solution. This can also be seen by finding the general solution to (). This is
done most easily in Fourier space, where () becomes

4 k3
+ + = 0 .
3
This equation has the following exact solution
j1 (x) n1 (x) 1
k () = Ak + Bk , x k ,
x x 3
where the subscript k indicates that the solution can have different amplitudes for each
value of k. The functions j1 (x) and n1 (x) are the spherical Bessel and Neumann functions
sin x cos x x
j1 (x) = = + O(x3 ) ,
x2 x 3
cos x sin x 1
n1 (x) = 2 = 2 + O(x0 ) .
x x x
Since n1 (x) blows up for small x (early times), we reject that solution on the basis of initial
conditions, i.e. we set Bk 0. We match the constant Ak to the primordial value of the
potential, k (0). This gives
 
sin x x cos x
k () = 3k (0) .
x3

78
 Notice that this solution is valid on all scales. On large scales, x  1, this reduces to
k () = k (0) = const., while on small scales x  1, we get

cos 13 k


k () 9k (0) .
(k)2

During the radiation era, subhorizon modes of therefore oscillate with frequency 1 k and
3
an amplitude that decays as 2 a2 .

(b) Consider the time-space component of the Einstein tensor

1
G0j = R0j g0j R = R0j ,
2
where the second equality applies in Newtonian gauge (since g0j = 0). Substituting the
perturbed connection coefficients into (2.33), we get

R0j = 2j ( + H) .

To apply the Einstein equation, we note that

T0j = g0 T j = g00 T 0 j
= g00 T 0 j
= a2 j q .

Hence, we have

G0j = 8GT0j
=
2j ( + H) = 8Ga2 j q + H = 4Ga2 q .

The comoving curvature perturbation can then be written as

H + H
R =
+ P 4Ga2
H
= 2
( + H)
4Ga (1 + w)
2 + H
= ,
3 H(1 + w)

where we have assumed a constant equation of state w in the second line. On superhorizon
scales, and for w = const., we can drop and get
2 5 + 3w
R = .
31+w 3 + 3w
During the radiation-dominated era, w = 1/3, this implies
3
RRD = RD ,
2

79
 while during the matter-dominated era, w = 0, we have
5
RMD = MD .
3
Using RRD = RMD , we get
9
MD = RD ,
10
which is the result we were asked to prove.

2. Growth of Matter Perturbations

(a) The conformal-time Friedmann equation is
8G 2
H2 = a (m + r ) .
3
Defining y = a/aeq , we have m = eq y 3 and r = eq y 4 , so that the Friedmann equation
can be written as
8G 2 8G 2 2
H2 = a eq (y 3 + y 4 ) = aeq eq (y 1 + y 2 ) = H02 m (y 1 + y 2 ) .
3 3 r
(b) Combining the continuity and Euler equations for pressureless perturbations,
m = vm ,
vm = Hvm ,
we find
m + Hm = 2 .
Because radiation oscillates quickly compared to the Hubble time on small scales, averaging
over time the gravitational potential is only sourced by matter fluctuations, so the Poisson
equation is only sensitive to matter,
2 = 4Ga2 m m .
Hence, we find
m + Hm 4Ga2 m m = 0 . (?)

(c) This next step involves a bit of algebra, but its nothing outside your powers. Starting from
(?), there are two things we want to do before anything else. One is that, of course, we
want to switch from to y as the time coordinate. The other is that we want to replace
the 4Ga2 term with something in the and y language that weve used elsewhere in this
problem. Lets do that first: we can write
3 m
4Ga2 m = H02 a2 m
2 m,0
3
= H02 m a1
2
3 2
= H0 m a1 eq y
1
2
3 2
= H02 m y 1 .
2 r

80
 Lets keep this in our back pockets. Next, well do the variable change. Taking the derivative
of the Friedmann equation with respect to y, we get

dH H 2 2
= 0 m y 2 + 2y 3 .


dy 2H r

It is then straightforward to express m and m as

dm
m = Hy ,
dy
 2
H02 2m 2

2 2 d m 1 dm 3 dm


m =H y + y + 2y .
dy 2 y dy 2H2 r dy

Plugging all this into (?) and dividing out H2 y 2 , we obtain

d2 m 2 dm 1 H02 2m 2
dm 3 H 2 2
2
+ 2
y + 2y 3 3 02 m m = 0 .
dy y dy 2 H r dy 2y H r
We can simplify two of these terms by using

H02 2m y2
= .
H 2 r 1+y
Substituting this and cleaning up the result, we obtain the Meszaros equation,

d2 m 2 + 3y dm 3
2
+ m = 0 .
dy 2y(1 + y) dy 2y(1 + y)
By sustitituon, we can confirm that the two solutions are

m 2 + 3y ,
 
1+y+1 p
m (2 + 3y) ln 6 1+y.
1+y1

During radiation domination (y  1), the first mode is clearly constant, while the second
contains a constant term and a piece evolving as ln y ln a. During matter domination
(y  1), the first solution is y a, while the second decays as y 3/2 . These confirm
the well-known results: matter clusters very slowly ( logarithmically) during radiation
domination, and then grows linearly, a, during matter domination.

3. Cosmological Gravitational Waves

(a) Calculating the components of the Einstein tensor can become time-consuming, if you
dont organize your calculation in a convenient way. First, notice that we are trying to
compute Gij = Rij 12 gij R. You might think that you need to compute all components
of the Ricci tensor R to get the Ricci scalar R = g R . However, at linear order,
the Ricci scalar cannot receive any contribution from a tensor perturbation.7 The
7
If you dont believe this, try writing down some terms. They will all either be second order (like hij hij ) or
vanish because hij is transverse (like i j hij = 0) and traceless (like hi i ).

81
 Ricci scalar will therefore take its homogeneous value, which has been computed in
the course notes,
R = 6a2 (H + H2 ) .
The only real work we therefore need to do is to compute Rij . It is useful to have at
hand the result for the Christoffel symbol with two indices summed: 0 = 4H and
i = 0. We then do a little bit of work,

Rij = ij j i + ij i j

= 0 0ij + k kij 0 + 0ij 0 0ik k0j ki0 0jk kil ljk

= Hij + 2Hhij + 2Hhij + hij 2 hij + 4H(Hij + 2Hhij + hij )

(Hik + 2Hhik + hik )(Hjk + hkj ) (Hjk + 2Hhjk + hjk )(Hik + hki )

= (H + 2H2 )ij + hij 2 hij + 2Hhij + 2(H + 2H2 )hij .

We then calculate the spatial part of the Einstein tensor:

1
Gij = Rij gij R
2
= (H + 2H2 )ij + hij 2 hij + 2Hhij + 2(H + 2H2 )hij 3(H + H2 )(ij + 2hij )

= (2H + H2 )ij + hij 2 hij + 2Hhij 2(2H + H2 )hij ,

which contains the correct background piece and the linear piece that we are after:

Gij = hij 2 hij + 2Hhij 2(2H + H2 )hij .

(b) Using that the perturbation to the stress tensor is Tij = 2a2 P hij a2 ij , we find the
Einstein equation for tensor perturbations:

hij 2 hij + 2Hhij 2hij (2H + H2 ) = 8G(2a2 P hij a2 ij ) .

On account of the zeroth-order Friedmann equation, 2H + H2 = 8Ga2 P , we find

that the two pressure terms exactly cancel out and we get

hij + 2Hhij 2 hij = 8Ga2 ij .

(c) A Fourier mode in a matter dominated universe (a 2 ) without anisotropic stress

(ij = 0) satisfies
4
hij + hij + k 2 hij = 0 .

It is straightforward to verify that

k cos(k) sin(k)
hij
(k)3

is a solution.

82
(d) In the limit k  1 that is, for modes much larger than the Hubble horizon we
expand the trigonometric functions as
1
cos(k) = 1 (k)2 + ,
2
1
sin(k) = k (k)3 + ,
6
so that
k cos(k) sin(k) k1 1
+ O (k)2 .
 
hij 3
(k) 3
Hence, superhorizon gravitational waves have constant amplitude. More generally,
regardless of the equation of state of matter, which affects the Hubble friction term
in the equation of motion for hij , the gradient term k 2 hij can be neglected in the
superhorizon limit, giving us
hij + 2Hhij 0 ,
which clearly admits constant solutions.
On the other hand, well within the horizon (k  1) the cosine term in the solution
dominates over the sine term and we get

k cos(k) sin(k) k1 cos(k)

hij .
(k)3 (k)2

As expected, this solution oscillates with comoving frequency k and the amplitude falls
as 2 a1 .

83
7.3 Solutions 2: Quantum Fluctuations

1. Slow-Roll Inflation
(a) The slow-roll parameters for V () = 4p p are
2 
Mpl V, 2 p2 Mpl 2
  
V = ,
2 V 2
Mpl 2
 
2 V,
V Mpl = p(p 1) .
V
Without loss of generality, we choose the field to evolve from > 0 to = 0. Inflation ends

at e = (p/ 2)Mpl , where V (e ) 1. The number of e-fold between and e is
Z " #
d 1 1 2 2e 1 2 1
N () = = 2 2 = 2 .
e Mpl 2V 2p Mpl Mpl 2p Mpl 4

The horizon exit of the reference scale k? corresponds to

1 2? 1 1 2? p
N? = 2 2 ? 2pN? Mpl .
2p Mpl 4 2p Mpl

The slow-roll parameters at ? are

p 1 p1 1
V,? = , V,? = .
4 N? 2 N?
The spectral tilt and the tensor-to-scalar therefore are
3p 1 1 2+p
ns 1 6V,? + 2V,? = + (p 1) = ,
2 N? N? 2N?
4p
r 16V,? = .
N?

(b) The slow-roll parameters for V () = 4 [1 cos(/f )] are

2 
V, 2
  2
Mpl


V = cot ,
2 V 2 2f
(   )
2
2 V,
V Mpl = csc 2 ,
V 2 2f

where Mpl 2 /f 2 . Without loss of generality, we choose the field to evolve from > 0 to

= 0. Inflation ends at e = 2f cot1 ( 2/), where V (e ) 1. The number of e-fold

p

between and e is
Z   
d 1 2
N () = ln cos .
e M pl 2 V 2f
The horizon exit of the reference scale k? corresponds to
 
?
cos2 = eN? .
2f

84
 The slow-roll parameters at ? are

1 eN? 2
V,? = , V,? = .
2 eN? 1 2 eN? 1
The spectral tilt and the tensor-to-scalar therefore are

eN? + 1
ns 1 6V,? + 2V,? = ,
eN? 1
1
r 16V,? = 8 .
eN? 1

2. The Lyth Bound

(a) In the lectures, we derived the power spectra of scalar and tensor fluctuations. Their
amplitudes are
1 1 H2 2 H2
As = 2 2 , At = 2 2 .
8 Mpl Mpl
Taking the ratio, we have

At 8(0 )2
r = 16 = 2 2 (?)
As Mpl H

(b) The result (?) can be written as

 2
r 1 d
= ,
8 Mpl dN

where dN Hdt is the differential of the number of e-folds. Hence, we get

Z Ncmb r
r(N )
= dN .
Mpl Ne 8

For const., we have r const. and hence

r Z N r
r cmb r
dN = N ,
Mpl 8 Ne 0 8

which leads to r
N r
= .
Mpl 60 0.002
Typically we need N & 60 to solve the horizon and flatness problems, while r & 0.001 is
roughly needed to have any hope of detecting primordial gravitational waves via the B-mode
polarization of the CMB. To satisfy both of these, we see that we would need & Mpl .

(c) The amplitude of the tensor spectrum is

2 H2 2 V
At = 2 2 2 4 .
Mpl 3 Mpl

85
Using At = rAs , we get
1/4
3 2

1/4
V rAs Mpl .
2
With As 2.5 109 and Mpl 2.4 1018 GeV, we find

V 1/4 3.3 r1/4 1016 GeV .

For r 0.01 (which is near the current observational limit) this gives V 1/4 1016 GeV.
This exceeds the highest energy scales that will be probed at the LHC by a factor 1013 .

86
7.4 Solutions 3: Boltzmann Equation

1. Boltzmann for Harmonic Oscillators

(a) The full time derivative of the distribution function can be written as

df (t, x, p) f dx f dp f
= + + .
dt t dt x dt p
By the definition of momentum, we have
dx p
= ,
dt m
and the equation of motion of the harmonic oscillator can be written as
dp
= x .
dt
The collisionless Boltzmann equation for the harmonic oscillator then is
f p f f
+ x =0 (?)
t m x p

An equilibrium distribution function satisfies f /t = 0. It is claimed that the distribution

function is then only a function of the energy. To show that f (E) is indeed a solution,
consider
 
p f (E) f (E) df p E E
x = x
m x p dE m x p
= 0.

(b) First, integrate (?) over all momentum. The f /p term vanishes after integrating by
parts and noticing that f = 0 at p = (there are no particles with infinite momentum!).
Hence, we find
n (nv)
+ = 0.
t x
This is the continuity equation. To get the Euler equation, first multiply by p/m and then
integrate over all momentum. This gives
Z
(nv) dp p2 x
+ 2
f+ n = 0,
t x 2 m m

where the last term follows from integration by parts. The integral over p2 yields two terms:
i) a bulk velocity term, nv 2 , and ii) a pressure term, P . Using the continuity equation, we
get
v v 1 P x
+v + + = 0.
t x n x m

87
2. Boltzmann for Dark Matter
(a) Substituting the metric (2.3) into g P P = m2 , we have

a2 (1 + 2)(P 0 )2 = m2 + p2 E 2 ,

where we have defined p2 gij P i P j . Solving this for P 0 and expanding to linear order
in , we get
E
P 0 = (1 ) .
a
To find the spatial component of the four-momentum, we write P i C pi , so that

p2 gij P i P j = a2 (1 2)ij C 2 pi pj = a2 (1 2)C 2 .

Solving this for C and expanding to linear order in , we get

p
Pi = (1 + ) pi .
a
The perturbed four-momentum therefore is

P = [(E/a)(1 ), (p/a)(1 + )pi ] (?) .

(b) The total time derivative of the distribution function is

dfc fc dxi fc d ln E fc dpi fc

= + + + (??) .
d d xi d ln E d pi
| {z }
O(2)

We need dxi /d and d ln E/d in the perturbed background. For the former, we simply
note that
dxi d dxi Pi p
= = 0 = pi + O(1) .
d d d P E
Since this coefficient multiplies the first-order quantity fc /xi , we only need it at zeroth
order.
The coefficient d ln E/d is derived from the geodesic equation, which we recall takes the
form
dP dx
= P P , with P .
d d
Using
dP d dP dP
= = P0 ,
d d d d
we can write the geodesic equation as

dP P P
= .
d P0

88
 We consider the = 0 component. The left-hand side of the geodesic equation becomes

dP 0
 
d E 1 dE E E d
= (1 ) = H (1 )
d d a a d a a d
 
E d ln E p
= (1 ) H p + O(2) .
a d E

The right-hand side is

P P P iP j
0
= 000 P 0 20i0 P i 0ij ,
P P0
where
E E
000 P 0 = (H + ) (1 ) = (1 ) (H + ) ,
a a
p E p
20i0 P i = 2pi i (1 + ) = (1 ) 2pi i ,
a a E
P iP j (p2 /a2 )(1 + )2 E p2
0ij = [H 2H( + )] = (1 ) (H ) 2 .
P0 (E/a)(1 ) a E

Combining the above, we get

d ln E p p2
= p (H ) 2 .
d E E

Substituting the above into (??), we get the desired result

p2
 
fc p fc p
+ p fc (H ) 2 + p = 0 () .
E ln E E E

(c) Next, we take moments of the Boltzmann equation:

First, we multiply () by d3 p/(2)3 and integrate. This leads to

nc (nc vci ) d3 p fc p2 d3 p fc p i
Z Z

+ (H ) i p = 0 .
xi 3
(2) ln E E 2 x (2)3 ln E E

The last term can be dropped since it is secretly second order. [The integral over the
direction vector is nonzero only for the perturbed part of fc .] The remaining integral
term can be further simplified. Note that dE/dp = p/E, so the integrand can be
written as pfc /p. The integral then becomes
Z Z
d3 p fc
Z
4 3 fc 4
p = dp p = 3 dp p2 fc = 3nc .
(2)3 p (2)3 0 p (2)3 0

The zeroth moment of the Boltzmann equation therefore reads

nc (nc vci )
+ + 3(H )nc = 0 .
xi

89
 Substituing nc = nc (1 + c ), we find

n c = 3Hnc ,
c = i vci + 3 ,

which are the continuity equations for the background density nc and its fractional
perturbation c .
Next, we multiply () by d3 p/(2)3 (p/E)pj and integrate. This gives

(nc vcj ) d3 p p2 i j d3 p fc p3 j
Z Z

0= + i fc p p (H + ) p
x (2)3 E 2 (2)3 ln E E 3
d3 p fc p2 i j
Z

i p p .
x (2)3 ln E E 2

The second term can be dropped since it is small, order h(p/E)2 i, in the non-relativistic
limit. The remaining terms must be handled more carefully because of the partial
derivatives. Since (p/E)/E = /p, the third terms is actually order p/E, while
the last is independent of velocity. Performing the integration by parts explicitly in
the third term, we find

d3 p fc p2 j ddpj p4 fc
Z Z Z
p = dp
(2)3 p E (2)3 0 E E
j Z  3
p5
Z 
ddp 4p
= dp fc 3 .
(2)3 0 E E

The p5 /E 3 terms is negligible, so the only relevant contribution comes from the p3 /E
term, which leads to 4nc vcj .
Performing the same manipulations in the last integral term leads to nc . Hint: use
Z
4 ij
d pi pj = .
3

The first moment of the Boltzmann equation then reads

(nc vcj )
+ 4Hnc vcj + nc j = 0 .

To extract the first-order equation, we simply substitute nc nc . Using n c = Hnc ,

we then get
vc = Hvc i ,
which is the first-order Euler equation.

90
7.5 Solutions 4: Temperature Anisotropies

1. Anisotropies from Tensors

(a) With P = dx /d, it follows that

d  dxi 1
= 2, = pi hij pj .
d a d 2
The geodesic equation in conformal time then is
 dP
+ P P = 0 .
a2 d
The 0-component is

 d    2
   
0 0 i 1 i k j 1 j l
+ 4 00 + ij p hk p p hl p = 0.
a2 d a2 a 2 2

Substituting the given expressions for 000 and 0ij , we get

   
d ln  1 i 1 i k j 1 j l
= +H Hij + Hhij + hij p hk p p hl p
d 2 2 2
1
= hij pi pj ,
2
where in the second line we have dropped all terms beyond linear order in hij .

(b) Assuming that tight-coupling holds until recombination, we can neglect the temperature
quadrupole at last scattering. The only tensor-induced contribution to the temperature
anisotropies then comes from the free-streaming effect:
d d ln  1
= = hij pi pj .
d d 2
Integrating the Boltzmann equation along the line-of-sight, we find

1 0 0
Z
(n) (0 , x0 0, n = p) = d hij ( 0 , x( 0 )) ni nj ,
2

where x( 0 ) ( 0 )n. Writing hij in terms of its Fourier transform, we get

0
d3 k
Z Z
1 0
(n) = d 0 3/2
hij ( 0 , k) ni nj eik( ) kn (?) .
2 (2)

(c) It is convenient to expand the tensor mode in its two helicity components
X 1
hij (, k) = h (, k)ij (k) .
=
2

91
We first consider the special case k = kz. In that case, the explicit form of the polarization
tensor is

1 i 0
1
 (z) = i 1 0 .

2
0 0 0

Contracting this with n = (sin cos , sin sin , cos ), we find

1 i 0 sin cos
1
 i j
ij (z) n n = (sin cos , sin sin , cos ) i 1 0 sin sin

2
0 0 0 cos
r
1 8
= sin2 e2i = Y2 2 (n) .
2 15
The contribution of the Fourier mode k = kz to the integral in (?) therefore is
r
4
h (, kz) Y22 (n) eik cos .
15
Using the Rayleigh plane-wave expansion for the exponential and expressing Pl (cos ) in
terms of YL0 (n), this can be written as
4 X
h (, kz) (i)L 2L + 1 jL (k)Y22 (n)YL0 (n) .
15 L0

Next, we replace the product of the two spherical harmonics by a sum over spherical har-
monics weighted by Wigner 3j symbols
r " ! ! #
4 X X 2 L l 2L l
L
h (, kz) (i) (2L + 1)jL (k) 2l + 1 Yl2 (n) .
3 2 0 2 0 0 0
L0 l2

Writing out the 3j symbols explicitly and using a recursion relations for Bessel functions
to express jl2 in terms of jl , gives
r r
4 ik cos X jl (k)
h (, kz) Y22 (n) e = h (, kz) l Yl 2 (n) ,
15 2 (k)2
l
p
where l (i)l 2l + 1 (l + 2)!/(l 2)!.
So far, we have only considered Fourier modes with k parallel to z. The contribution from
a general Fourier mode is obtained by rotating the previous result. We are told that this
gives
r
i j ikkn X jl (k) X
hij (, k) n n e = h (, k) l Dl (k) Ylm (n) (??) ,
2 (k)2 m m2
l

and that an explicit form of the Wigner D matrices wont be needed.

92
(d) We write the temperature anisotropy in a harmonic expansion

(n) = alm Ylm (n) .

The angular power spectrum is then defined as halm al0 m0 i = Cl ll0 mm0 . We find the alm s
by substituting (??) into (?):
r
d3 k
Z
1 X l 4 l
alm = 3/2
(i) l (k) Dm2 (k) ,
2 (2) 2l + 1

where s Z 0
2l + 1 (l + 2)! jl (k)
l (k) = d 0 h ( 0 , k) () .
4 (l 2)! (k)2

To obtain the power spectrum, we compute

(" #" 0
#
d3 k d3 k0 l (k) l0 (k0 )
Z
ll0 4 1X
halm al0 m0 i = (i)
h0 (k0 )
p
(2l + 1)(2l0 + 1) 2 0 (2)3/2 (2)3/2 h (k)
0
hh (k)h0 (k0 )i Dm
l l
(k)Dm 0
0 0 (k ) .

Using
2 2 2 l (k)
hh (k)h0 (k0 )i = (k) D (k k0 ) 0 and l (k) ,
k3 h h (k)
we get
0
(i)ll
Z Z
 () 2 0
halm al0 m0 i = p d ln k l (k) 2h (k) l
dk Dm (k)Dm l
0 0 (k)
(2l + 1)(2l0 + 1) | {z }
4
ll0 mm0
2l + 1
Z
4 2
d ln k l (k) 2h (k) .

= ll0 mm0
(2l + 1)2
The final result for the angular power spectrum therefore is
Z
4 2
d ln k l (k) 2h (k)

Cl = 2
() .
(2l + 1)

To evaluate () we need an expression for the tensor mode function in (). The universe is
matter dominated at recombination and thereafter. The tensor mode functions then satisfy
(see Problem 3 on Problem Set 1):
j1 (k) j2 (k)
h (, k) = 3h (i , k) h (, k) = 3kh (i , k) .
k k
Substituting this into (), we get
s
(l + 2)! 0 j2 (k 0 ) jl (k(0 0 ))
Z
3

l (k) = (2l + 1) k d 0 () .
4 (l 2)! k 0 (k(0 0 ))2

93
(e) Since gravitational waves decay inside horizon, we expect the tensor-induced CMB spectrum
to drop off on small scales. Lets see how this is reflected in the result (). This integral
is non-negligible only when k 0 2 and k(0 0 ) l simultaneously, since otherwise
at least one of the two Bessel functions is close to zero. Both conditions can be satisfied
simultaneously only if
0
l . 2 2 z + 1 60 .

This estimate is consistent with the drop of the spectrum observed in Fig. 27.
Finally, let us show that our result () corresponds to l(l + 1)Cl = const. on large scales.
Let us approximate j2 (x)/x by D (x 2). We then get
s
3 (l + 2)! jl (k0 2)
l (k) (2l + 1) for < k 1 < 0 .
4 (l 2)! (k0 2)2

The restriction on wavenumbers is such that only gravitational waves that enter the horizon
between last-scattering and today contribute. Assuming a scale-invariant primordial tensor
spectrum, 2h = At , the angular power spectrum becomes
1/
j 2 (k0 2)
Z
(l + 2)!
Cl At d ln k
4 (l 2)! 1/0 (k0 1)4
0 / 1
dx jl2 (x)
Z
(l + 2)!
= At ,
4 (l 2)! 0 x + 2 x4

where we substituted x k0 2 in the second line. For l  1, but l  60, the integral is
dominated by x l, and we can replace x + 2 by x in the integrand and extend the upper
limit of integration to . Using
Z
j 2 (x) 4 (l 2)! 1
dx l 5 = ,
0 x 15 (l + 2)! (l + 3)(l 2)

we find
At
Cl .
15 (l + 3)(l 2)
Hence, l(l + 1)Cl const., as expected.

94
References

[1] D. Baumann, Part III Cosmology.

[2] S. Dodelson, Modern Cosmology. Academic Press, 2003.
[3] P. Peter and J.-P. Uzan, Primordial Cosmology. Oxford University Press, 2013.
[4] W. Hu and S. Dodelson, Cosmic Microwave Background Anisotropies, Ann. Rev. Astron.
Astrophys. 40 (2002) 171216, arXiv:astro-ph/0110414 [astro-ph].
[5] W. Hu, Lecture Notes on CMB Theory: From Nucleosynthesis to Recombination,
arXiv:0802.3688 [astro-ph].
[6] A. Challinor, Part III Advanced Cosmology.
[7] W. Hu, Concepts in CMB Anisotropy Formation, in The Universe at High-z, Large-Scale
Structure and the Cosmic Microwave Background, pp. 207239. Springer, 1996.
[8] Z. Hou, R. Keisler, L. Knox, M. Millea, and C. Reichardt, How Massless Neutrinos Affect the
Cosmic Microwave Background Damping Tail, Phys. Rev. D87 (2013) 083008, arXiv:1104.2333
[astro-ph.CO].
[9] S. Bashinsky and U. Seljak, Neutrino Perturbations in CMB Anisotropy and Matter Clustering,
Phys. Rev. D 69 (2004) 083002, arXiv:astro-ph/0310198 [astro-ph].
[10] D. Baumann, D. Green, J. Meyers, and B. Wallisch, Phases of New Physics in the CMB, JCAP
1601 (2016) 007, arXiv:1508.06342 [astro-ph.CO].

95