Cosmology 2022: Lecture Notes

Jonathan R. Pritchard, Imperial College

ii J. R. Pritchard
Contents

1 Introduction:
Homogeneity and Isotropy 1

2 The Expanding Universe 3

2.1 Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 The Scale Factor and Hubble’s law . . . . . . . . . . . . . . . . . . . . . 4
2.3 Newtonian Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 Thermodynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.4.1 Radiation-Dominated Universe . . . . . . . . . . . . . . . . . . . 10
2.5 General and Special Relativity . . . . . . . . . . . . . . . . . . . . . . . . 11
2.5.1 Metrics and Curvature . . . . . . . . . . . . . . . . . . . . . . . . 12
2.5.2 Dynamics of the FRW metric . . . . . . . . . . . . . . . . . . . . 16
2.5.3 Redshift . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17

3 Cosmological Models and Parameters 21

3.1 Cosmological Parameters and the Expansion of the Universe . . . . . . . 21
3.1.1 The time-dependence of the density . . . . . . . . . . . . . . . . . 23
3.2 Cosmological Models . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 23
3.2.1 Models with a cosmological constant . . . . . . . . . . . . . . . . 25

4 Cosmography 29
4.1 The Age of the Universe . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
4.2 Horizons . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
4.3 Distances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 32
4.3.1 Luminosity Distance . . . . . . . . . . . . . . . . . . . . . . . . . 33
4.3.2 Angular-Diameter Distance . . . . . . . . . . . . . . . . . . . . . 35
4.3.3 The Extragalactic Distance Ladder . . . . . . . . . . . . . . . . . 35

5 Thermodynamics and Particle Physics 37

5.1 Radiation Domination . . . . . . . . . . . . . . . . . . . . . . . . . . . . 37
5.2 Black Body radiation and equilibrium statistics . . . . . . . . . . . . . . 38
5.3 The Hot Big Bang . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.3.1 Interactions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42
5.3.2 Thermal history of the Universe . . . . . . . . . . . . . . . . . . . 43

iii
CONTENTS

5.4 Relic Abundances . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

5.4.1 Baryogenesis and the Sakharov Conditions . . . . . . . . . . . . . 44
5.4.2 Interaction rates and the Boltzmann Equation . . . . . . . . . . . 46

6 Hydrogen Recombination and the Cosmic Microwave Background 51

6.1 Introduction and Executive Summary . . . . . . . . . . . . . . . . . . . . 51
6.2 Equilibrium ionization . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
6.3 Freeze-Out . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

7 Big-Bang Nucleosynthesis 57
7.1 Initial Conditions for BBN: neutron-proton freeze-out . . . . . . . . . . . 58
7.2 Helium production . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
7.2.1 Dependence upon the cosmological parameters . . . . . . . . . . . 60
7.3 Observations of primordial abundances . . . . . . . . . . . . . . . . . . . 61

8 Interlude 63
8.1 Natural Units . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 63
8.1.1 The Cosmological Constant problem . . . . . . . . . . . . . . . . 66
8.2 Open Questions in the Big Bang Model . . . . . . . . . . . . . . . . . . . 68
8.2.1 The Flatness Problem . . . . . . . . . . . . . . . . . . . . . . . . 68
8.2.2 The Horizon Problem . . . . . . . . . . . . . . . . . . . . . . . . . 69
8.2.3 The relic particle problem . . . . . . . . . . . . . . . . . . . . . . 70

9 Inflation 73
9.1 Accelerated Expansion . . . . . . . . . . . . . . . . . . . . . . . . . . . . 73
9.1.1 Acceleration and negative pressure . . . . . . . . . . . . . . . . . 76
9.1.2 The duration of Inflation . . . . . . . . . . . . . . . . . . . . . . . 77
9.2 Inflation via a scalar field . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
9.2.1 Density Perturbations . . . . . . . . . . . . . . . . . . . . . . . . 80

10 Structure Formation 83
10.1 Notation and Preliminaries . . . . . . . . . . . . . . . . . . . . . . . . . . 83
10.2 Spherical Collapse . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 86
10.3 Linear Perturbations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 88
10.3.1 Newtonian Theory—non-expanding . . . . . . . . . . . . . . . . . 88
10.3.2 Perturbation theory in an expanding Universe . . . . . . . . . . . 90
10.4 The processed power spectrum of density perturbations . . . . . . . . . . 94
10.4.1 Initial Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . . 94
10.4.2 The transfer function . . . . . . . . . . . . . . . . . . . . . . . . . 96
10.4.3 The effect of baryons: BAOs and the CMB . . . . . . . . . . . . . 99

iv J. R. Pritchard
Preface

These notes were originally produced by Prof. Andrew Jaffe for the Cosmology course as
run between 2008-2012. Further updated by Dr Jonathan Pritchard (2013-2016) and Dr
Subu Mohanty (2017-2021).
Very little of the following is new. Most of it comes, in one form or another, from one
of the following books.
• Rocky Kolb & Mike Turner, The Early Universe
• Andrew Liddle, An Introduction to Modern Cosmology
• Michael Rowan-Robinson, Cosmology
• Peter Schneider, Extragalactic Astronomy and Cosmology: An Introduction
In some cases I explicitly give a reference at the beginning of a chapter or section.

Course structure
The course is structured to cover four main themes in cosmology. Chapters 1-4 establish
the essential language and mathematics needed to talk quantitatively about the physics
of an expanding universe. This lays the foundations for everything that comes later and
enables us to move beyond the level of popular science books to a deeper understanding
of cosmology.
Chapters 5-7 then cover the thermodynamics and physics of a hot big bang. As
the primeval plasma expands and cools the Universe passes through a number of phase
transitions. The number and type of particles produced in these transitions provide one
of the key observational constraints on the physics of the early Universe.
In Chapters 8 and 9, we discuss inflation to address questions of how a big bang might
have occurred and describe some of the traditional problems that an early inflationary
period can solve. Inflation is our best guess at the physics in the first fraction of a
second after the big bang and represents the frontier of our current understanding of the
Universe’s history.
Finally in Chapter 10, we move beyond describing a smooth universe to describe the
irregularities that lead to the formation of structures and galaxies. Modern observations
of the cosmic microwave background and of the positions of millions of galaxies have
provided enormous amounts of information on the distribution of matter in the Universe.
This chapter sketches the theoretical framework for interpreting that data.

v
Chapter 1

Introduction:
Homogeneity and Isotropy

MRR 4.1-4.2
See introductory slides.

1
1 — Introduction:
Homogeneity and Isotropy

2 J. R. Pritchard
Chapter 2

The Expanding Universe

2.1 Expansion
Before we start discussing a physical and mathematical model to describe an expanding
Universe, it is useful to be able to picture just what we mean by “expansion” in this
context. On “small” scales, corresponding to relatively nearby galaxies, it is completely
appropriate to think of the expansion as a relative velocity between any two galaxies,
increasing with time and distance. On larger scales, however, such that the time between
the emission of light at the distance galaxy and its reception here is large, then we
are really observing the galaxy at a very different time and we cannot say that we are
observing a quantity having anything to do with its velocity relative to us. Hence, we will
see that there are very many galaxies in the distance Universe for which we would naively
define a velocity v > c, but this is just not the correct way to interpret our observations.
A similar confusion occurs when talking about the distance to that galaxy. We sim-
ply cannot lay out relativistic meter sticks between us, now, and the distant galaxy,
then, which is the appropriate thought-experiment to do in order to define the distance.
Hence we will also find that there are a myriad of different ways to define the distance,
each numerically different, although none of them is any more correct than the others.
Nonetheless even at this early stage it is very useful to lay down a coordinate system in
our expanding Universe. We could of course imagine a proper coordinate system, with
the coordinates given by the number of actual, physical meters, from some fiducial point
which we could choose to be the origin. But in this coordinate system, due to expansion,
the galaxies are moving. Instead, try to picture a coordinate system with the grid lines
expanding along with all of the objects in the Universe. If we could freeze time, this
coordinate system would just look like a normal cartesian (or polar, or however we decide
to draw our axes), but the grid lines are further apart from one another in the future,
and closer together in the past. So for this case we don’t really want to measure things
in physical units (meters or megaparsecs), but really just numbers. Conventionally, how-
ever, we do choose the units so that they are equivalent to physical units today — since
many of our measurements are done now.
Real galaxies or other objects may have small movements (called “peculiar velocities”)
with respect to this coordinate system, but on average — due to the homogeneity and

3
2 — The Expanding Universe

isotropy of the Universe — they will be at rest with respect to these coordinates as the
Universe expands. We call this a comoving coordinate system, and hypothetical observers
expanding along with these coordinates are called comoving observers. We show this (in
two dimensions) in Figure 2.1.

Figure 2.1: Comoving coordinates and observers in an expanding Universe.

2.2 The Scale Factor and Hubble’s law

MRR 4.3
We want to work out the mathematical description of an expanding, homogeneous and
isotropic Universe. First consider a bunch of points in the Universe at some time t1 , and
then later at some time t2 , as in Figure 2.2. If the Universe is to remain homogeneous,
the points can’t “bunch up” relative to one another. If rij (t) = ri (t)−rj (t) is the distance
between points i and j at a given time, we must scale all of these relative distances as a
function only of time, but not position. That is

rij (t2 ) = a(t2 )rij (t0 )

rij (t1 ) = a(t1 )rij (t0 ) (2.1)

where t0 is some arbitrary time, which we can take to be the present day. We can combine
these to get
rij (t0 ) = a−1 (t1 )rij (t1 ) = a−1 (t2 )rij (t2 ) = const . (2.2)
(It’s a constant since the value at any specific time is fixed.) So, at a general time, t,

a−1 (t)rij (t) = const (2.3)

and we can take the derivative:

a−1 (t)ṙij (t) − a−2 (t)ȧ(t)rij (t) = 0 (2.4)

or
ȧ ṙij
= . (2.5)
a rij
This equation looks trivial, but note that the right-hand side depends on i and j, and
hence which two points you’ve chosen, whereas the left-hand side doesn’t. Therefore, the

4 J. R. Pritchard
 2.2.: The Scale Factor and Hubble’s law

3
r13(t2)
r13(t1) r23(t2)
r23(t1)

1 2
r12(t1) r12(t2)
1 2

time t = t1 time t = t2

Figure 2.2: Homogeneous and isotropic expansion.

scaling between any two points is just a function of time, and we can write most generally
ȧ ṙ
= (2.6)
a r
for an arbitrary distance r. (Note that we started our derivation assuming that a has no
units, but in fact in this last equation we see that we could also give a units of length
which is sometimes done.)
The quantity a(t) — the scale factor — is going to be with us throughout the
course.1 It describes the evolution of the Universe, and for a homogeneous and isotropic
Universe, it tells us almost everything we need to know. We’re also going to continually
encounter the combination ȧ/a, which is the expansion rate (also referred to as the
Hubble parameter ) and we will sometimes write
ȧ
H(t) = (2.7)
a t

and in particular
ȧ
H0 = H(t0 ) = (2.8)
a t=t0

is the expansion rate today, one of the most important quantities in cosmology. Today,
then, we can rewrite this equation as

ṙ(t0 ) = H0 r(t0 ) . (2.9)

These are all relative distances (recall the ij indices we’ve got rid of), but by convention
we can take one of the points to be the location of the earth from which we observe, and
the other to be the location of some galaxy (since we can only measure the distances to
objects, not arbitrary points!), so we can then write,

v = H0 d (2.10)
1
In some texts, including MRR, the scale factor is denoted R(t) rather than a(t).

Cosmology 5
2 — The Expanding Universe

where v is the velocity of the galaxy relative to us, and d is its distance. This is Hubble’s
law, which we’ve derived assuming homogeneity, but in fact Hubble found it through
observations as we discussed last time.
Comments about there not being a true centre. . .

2.3 Newtonian Dynamics

What do we need to do in order to have a theory which predicts the scale factor a(t)
and hence the expansion rate? In truth, we need the full artillery of General Relativity
(GR), which we’ll touch on later. But we can do a surprising amount using Newtonian
mechanics and gravity.

v
ρ F

Figure 2.3: An expanding uniform-density sphere of matter, itself embedded in a medium

(an infinite sphere) of the same density.

In addition to the usual F = ma, and Newton’s law of gravitation, we’ll need to
remember two very important corollaries from Newtonian gravity. First, there is no
gravitational field inside a spherical shell of matter. Second, we can treat the gravitational
attraction of a sphere as if it were concentrated at the centre. Hence, if we’re inside a

6 J. R. Pritchard
 2.3.: Newtonian Dynamics

homogeneous sphere of matter of density ρ (assume it’s gas so we can move around)
at some distance r from the centre, we can ignore everything at greater radii. We will
also assume that the matter is pressureless so it takes no P dV work for it to move —
this is conventionally called “dust” or simply “non-relativistic matter” in cosmology and
relativity.
This is the setup for the Universe, as shown in Figure 2.3, except that we’ll take the
outer sphere to be arbitrarily (infinitely) large. This is a bit of a cheat in Newtonian
gravity, but in fact this derivation is correct, and indeed goes through to GR. So consider
a point P at a distance r from the “centre”. The gravitational force on a test particle of
mass m at P is
GM m Gm 4 4πGmρr
F = mr̈ = − 2
= − 2 πr3 ρ = − , (2.11)
r r 3 3
using the mass interior to r, M = 4πr3 ρ/3. We can rewrite this as
r̈ 4
= − πGρ (2.12)
r 3
Now, if we assume that the sphere is expanding homogeneously, our test mass at point
P is actually moving with time, r(t) ∝ a(t), but so is all the matter both interior and
exterior to it. Hence, although the density, ρ, of the interior sphere is changing with time,
the total mass in that sphere (M , above) remains constant. So in fact we can just go
back to the second equality above and write
MG
r̈ = − (2.13)
r2
where now we note that M is a constant with time. This is a differential equation that
we can integrate, and find
ṙ2 = 2GM/r + A (2.14)
where A is an integration constant. We can rewrite this in the form
 2
ṙ 2GM A 8πG A
= − = ρ − (2.15)
r r3 r2 3 r2

where we have replaced M = 4πr3 ρ/3. With this substution, Eq. 2.12 and Eq. 2.14 only
contain factors of r in combinations ṙ/r, r̈/r or with an integration constant. Recall that
we can relate the distance between any two points as a function of time, r(t), to the scale
factor via r(t) = a(t)r(t0 ). Hence in these terms the factors of r(t0 ) cancel or can be
absorbed into a constant, and so we can rewrite these differential equations as
ä 4
= − πGρ (2.16)
a 3
and  2
ȧ 8πG kc2
= ρ(t) − 2 (2.17)
a 3 a

Cosmology 7
2 — The Expanding Universe

where we have written the integration constant as kc2 where c is the speed of light so that
the combination a2 /k has units [length]2 — some authors use that to make k a number
and give a units of length, while others make a a number and k have units [length]−2 .
For now only the overall combination k/a2 matters.
Exercise: solve Eq. 2.13 to get Eq. 2.14. This is a slightly simplified version of the
Friedmann Equation, and is one of the most important equations in cosmology. We
will encounter it again and again over the entire course. Despite the Newtonian derivation
(which required some hand-waving) it is in fact generally true for a homogeneous and
isotropic space-time, and was first derived by Friedmann using General Relativity.
We have used the fact that the mass in a sphere that expands along with the Uni-
verse remains constant. This is called a comoving sphere, another concept that we will
encounter again and again. We can use this observation to write down an equation for
the evolution of the density:

−3M dr
 
d d M ȧ
ρ(t) = = = −3ρ(t) = −3Hρ (2.18)
dt dt 4πr(t)3 /3 4πr4 /3 dt a

or

ρ̇ + 3Hρ = 0 (2.19)

which is a differential equation expressing the conservation of ρ for pressureless matter

(although we derived it by putting in the solution!) and eventually we will see how it
can be generalized for other forms of matter that behave in a more complicated way due
to relativity (where pressure actually contributes to the energy density). The solution to
this equation can be written as

a(t1 )3
ρ(t) ∝ 1/a(t)3 or ρ(t) = ρ(t1 ) (2.20)
a(t)3

where t1 is some arbitrary but fixed time. Plugging this into the Friedmann Equation,

 2
ȧ 8πG a(t1 )3 kc2
= ρ(t1 ) − (2.21)
a 3 a(t)3 a(t)2

Up until now, we haven’t said much about a, but in fact we know that our Universe
is expanding (a is growing, or ȧ > 0), so if we go back far enough, a gets smaller and
smaller. Eventually, then, the first term above will dominate. (Alternately, we can look
at the case k = 0 which we will see is a very special case, and in fact may describe the

8 J. R. Pritchard
 2.4.: Thermodynamics

Universe.) For this case,

 2
ȧ 8πG a(t1 )3
= ρ(t1 )
a 3 a(t)3
r
8πG
ȧ = ρ(t1 )a(t1 )3 a(t)−1/2
3r
8πG
a1/2 da = ρ(t1 )a(t1 )3 dt
3 r
2 3/2 3/2 8πG
(a − a(0) ) = ρ(t1 )a(t1 )3 t
3 3
a ∝ t2/3 (2.22)

where in the last line we have assumed a(0) = 0 (which we can take to define t = 0 —
when the distance between any two points was zero).
As an aside, there’s another interesting case, if k > 0 in Eq. 2.17 there’s a time when
ȧ = 0; the Universe stops expanding! We will see soon that this case corresponds to a
“closed” Universe which expands to a finite size and then recollapses.

2.4 Thermodynamics
MRR 4.3
If we go back to our conservation equation, Eq. 2.19, we can write it in an illuminating
form, multiplying the left and right sides by a3 dt:

a3 dρ + 3ρa2 da = 0
d(ρa3 ) = 0 ∝ dE/c2 (2.23)

where in the last line we have first used the fact that, for any comoving volume, ρa3 is
proportional to M , the mass in that volume, and then used relativity (E = M c2 ) to write
that mass in terms of the total energy in the volume. This is really just the first law
of thermodynamics — energy conservation — since for this “dust” matter we have both
dQ = T dS = 0 and dW = P dV = 0. But let’s generalize to P 6= 0:

dE = T dS − dW = 0 − p dV
dE/c2 = −p dV /c2 ∝ −p d(a3 )/c2 (2.24)

where we have assumed reversible, adiabatic expansion (dS = 0), and dW = p dV is

the work done by the system on the surroundings as it expands. So, combining with the
above,  p
a3 dρ + 3 ρ + 2 a2 da = 0 (2.25)
c
or
dρ  p  ȧ
+3 ρ+ 2 =0 (2.26)
dt c a

Cosmology 9
2 — The Expanding Universe

Note that the units here make sense:

Energy Force × length Force
ρc2 = = 3
= = Pressure (2.27)
Volume (length) Area
The presence of the speed of light, c, indicates that this is essentially an effect of
relativity, and in fact we have to modify our force equation in addition to this conservation
equation. The source in relativity (even special relativity) isn’t just mass, but “stress
energy”, encoded in the Stress-Energy tensor. For the fluids we will be concerned
with here, this is given by
 
ρ
 p/c2 
Tµν =   (2.28)
 p/c2 
p/c2
(empty entries are zero). What we really need is the Trace of this tensor,
Tr T = ρ + 3p/c2 (2.29)
which changes the force equation, Eq. 2.16 to
ä 4πG
ρ + 3p/c2 .


=− (2.30)
a 3
Note that, in fact, Eq. 2.13 for ȧ/a still holds, but is now independent of this acceleration
equation for ä
We still need a relation to link the density and the pressure, known as the equation
of state. For many kinds of stuff, we will assume
p
= w = const (2.31)
ρc2
so, for example, w = 0 for the “dust” we’ve been dealing with, and a radiation-dominated
medium has radiation pressure p = ρc2 /3 so w = 1/3. For these cases, we can also
calculate the sound speed:
∂p p
c2s = = = wc2 (2.32)
∂ρ ρ

2.4.1 Radiation-Dominated Universe

MRR 5.1
Going back to the radiation-dominated case with p = ρc2 /3 (so w = 1/3), the fluid
equation, Eq. 2.26, becomes
ȧ
0 = ρ̇ + 3ρ(1 + 1/3)
a
4 ȧ
ρ̇ = −3 × ρ
3 a
ρ̇ ȧ
= −4
ρ a
(2.33)

10 J. R. Pritchard
 2.5.: General and Special Relativity

which has the solution

ρrad ∝ a−4 . (2.34)
Compare ρmat ∝ a−3 for dust — we will understand the difference when we discuss the
redshift in a few lectures.
Now, let’s insert this into Eq. 2.17, again considering k = 0 appropriate for sufficiently
early times:  2
ȧ 8πG
' ρ (2.35)
a 3
which, for our radiation-dominated case, is
 2
ȧ 8πG a41
' ρ1 (2.36)
a 3 a4

where the subscript 1 refers to some fixed time at which the quantities are evaluated. We
can solve this:
r
8πG 4
ȧ = a1 ρ1 a−1
r 3
8πG 4
a da = a1 ρ1 dt
r 3
2 8πG 4
a /2 = a ρ1 t (2.37)
3 1
or
a ∝ t1/2 radiation (2.38)
which we can compare to

a ∝ t2/3 non-relativistic matter (2.39)

where from now on we will usually refer to the “dust matter” as “non-relativistic matter”.

2.5 General and Special Relativity

MRR 4.4-4.5
It turns out that we’ve been able to use Newtonian Gravity and a few “handwavey”
arguments from Special Relativity to derive equations that do, in fact, apply to the real
Universe.

These are the Friedmann Equations:

 2
ȧ 8πG kc2
= ρ(t) − 2 , (2.40)
a 3 a

Cosmology 11
2 — The Expanding Universe

and
ä 4πG
ρ + 3p/c2 ,


=− (2.41)
a 3
along with the conservation equation
dρ  p  ȧ
+3 ρ+ 2 =0 (2.42)
dt c a

But the real Universe is governed by General Relativity (GR). So to really under-
stand what these equations mean (and in particular the significance of the constant of
integration, k) we need to use a little bit of GR.

2.5.1 Metrics and Curvature

As you probably know, GR is a theory which links gravity to the underlying curvature of
the spacetime manifold. This curvature, in turn, is described by a metric, which encodes
the distances between any two points on the manifold. We can build this up using the
invariant interval, X
ds2 = gµν dxµ dxν (2.43)
µ,ν

where the indices µ, ν run over 0,1,2,3, corresponding to time and three spatial coordi-
nates, gµν is the metric, and dxµ is a differential spacetime coordinate interval (in order
to have units of length, we will take the time coordinate, µ = 0, to be dx0 = c dt where c
is the speed of light). Because we are free to use different coordinates (for example, polar
vs. Cartesian), the actual form of the metric depends on the coordinates used — and the
requirement that the physics not depend on this is one of the central facts of GR.
Note that we will occasionally use the “Einstein Summation Convention” and write
this as
ds2 = gµν dxµ dxν summed over repeated indices . (2.44)
You have probably already encountered a metric in special relativity, which we used
to describe the fact that the time coordinate can mix with the spatial coordinates. In this
case, the appropriate metric is the Minkowski metric, which, for Cartesian coordinates
(x0 , x1 , x2 , x3 ) = (ct, x, y, z) is just
 
+1
 −1 
gµν = ηµν =   (2.45)
 −1 
−1

[note that we will use the “signature” (+ − −−) but you will often encounter (− + ++)
in the literature]. Here, then,

ds2 = c2 dt2 − ds23 where ds23 = dx2 + dy 2 + dz 2 . (2.46)

12 J. R. Pritchard
 2.5.: General and Special Relativity

If instead we had decided to use polar coordinates, we could instead write the spatial
three-dimensional part as

ds23 = dr2 + r2 dθ2 + r2 sin2 θdφ2 . (2.47)

Note that the angular part of this can itself be written as

r2 dθ2 + sin2 θdφ2 = r2 dΩ22

(2.48)

where, finally dΩ22 represents the metric on the surface of the r = 1 sphere (more specifi-
cally, of the two-sphere, where “two” represents the number of dimensions of the surface
of the sphere) — this is our first example of the metric of a curved surface, even though
it is a surface embedded in a larger, flat, spacetime.
To get a feeling for this, let’s take a look at a metric that describes a circle — that
is, distances along the perimeter of a circle of radius R. This is just a one-dimensional,
curved manifold. It is pretty clear that we should be able to write

ds2 = R2 dθ2 (2.49)

which simply says that distances along the perimeter of the circle satisfy ds = R dθ. But
it is also case that we can write the metric in a different coordinate system as

dy 2
ds2 = . (2.50)
1 − ky 2

This is clear if we make the substitution ky 2 = sin2 θ, in which case we can write

k −1/2 cos θ dθ
ds = p = k −1/2 dθ (2.51)
2
1 − sin θ

so we can identify k = 1/R2 and we can also interpret our second set of coordinates to
mean that y = R sin θ represents y in a Cartesian coordinate system (with θ measured
counter-clockwise from y = 0, x = R).
Exercise: The above discussion assumes k > 0. Show that k = 0 corresponds to
distances along a line, and k < 0 to distance along a hyperbola.

The FRW Metric

Now, we wish to expand this discussion to General Relativity in a cosmological context.
From the first lecture, we know that, to a good approximation, the Universe is homo-
geneous and isotropic — the cosmological principle. Thus, on large scales, we want to
find a spacetime manifold (basically, a shape in three spatial and one time dimensions)
that, at any given time, looks the same in all directions and from all places. To guide our
intuition, let’s think about this in two dimensions first. What two-dimensional shapes
satisfy the requirements of homogeneity and isotropy? Certainly, the infinite plain does
so: it looks the same everywhere on it. So does the sphere: there are no special points

Cosmology 13
2 — The Expanding Universe

anywhere. It turns out there is one other, not so obvious, possibility, which is the “hy-
perbolic paraboloid” or saddle shape. All of these possibilities are shown in Figure 2.4.
For the sphere, we can obviously label the manifold by its radius in some units, and in
fact the flat plane corresponds to the infinite-radius limit. We can also define a “radius of
curvature” for the hyperbolic manifold. (Indeed, many calculations on the hyperboloid
just correspond to using the hyperbolic trigonometric functions in place of the spherical
functions.) For both of the curved cases, the manifold is self-similar: it is just a uniform
scaling between different values of the radius of curvature.

Figure 2.4: Homogeneous and isotropic manifolds in two dimensions. Left: All
three constant-curvature manifolds, courtesy WMAP/NASA GSFC. Right: Hyperbolic
paraboloid courtesy Wikipedia: http://en.wikipedia.org/wiki/File:Hyperbolic_
triangle.svg

It is worth spending a little time understanding the meaning of “curvature” in rel-

ativity. We can usually only picture curvature when the manifold is embedded in a
higher-dimensional spacetime (for example, Figure 2.4 shows the two-manifolds embed-
ded in flat three-dimensional space — and then projected down to two-dimensional flat
space), but we can define curvature more generally, based on concepts defined only on
the metric itself. All we need is the concept of a locally straight line — a null geodesic
curve. We are used to these from living on the surface of something that is approximately
a two-sphere, the Earth. We know what it means to walk in a straight line locally, and
we find that if we continue walking that would take us along a great circle. So we would
find that two walkers starting off on parallel tracks would eventually meet. Furthermore,
we would find that a triangle made of segments of three such great circles does not have a
total interior angle of 180◦ but rather some number greater than that. To see an example
of this, consider one of the segments to be along the equator, and note that any two lines
perpendicular to this meet at the pole, with the angle at the pole just depending on how
far apart the two perpendiculars started. So we have a triangle with at minimum 180◦
total angle and at maximum 360◦ for the case when the two other legs start at antipodal
points. Similarly, on the saddle-shaped surface, parallel lines can converge and diverge,
and a triangle always contains less than 180◦ . Since one of the results of GR is that light

14 J. R. Pritchard
 2.5.: General and Special Relativity

follows geodesic paths, we can use lightrays, and therefore astronomical observations, to
probe the geometry of the Universe. Later on, we will use this very fact to measure the
curvature of the Universe as a whole with the Cosmic Microwave Background radiation.
It turns out that in 3 + 1 dimensions, we have the same set of homogeneous and
isotropic manifolds, but promoted to higher dimensionality (and so effectively impossible
to picture with our 3D brains). The most general spatially homogeneous and isotropic
manifolds are indeed flat space, the three-sphere, and the three-hyperboloid. Note in
particular that the three-sphere is not a “ball” in three dimensions — it is a three-
dimensional “surface” with no boundary that can be observed in three dimensions, just
as there is no boundary as you walk along the surface of the approximately-spherical
earth. The metric for this case is given by

dr2
 
2 2 2 2 2 2 2 2 2
ds = c dt − a (t) + r dθ + r sin θ dφ (2.52)
1 − kr2

where we use polar coordinates in three dimensions, and we now have a scale factor a(t),
since all that the cosmological principle tells us is that at any given time the Universe
is homogeneous and isotropic, but it lets us change the overall scale of the universe as a
function of time. In this form, the scale factor has no units. We also have a number k in
our equation, and this tells us which of the three possible manifolds we have. As we have
written it here, this is a real number with units [length]−2 , just as in the case of the circle
manifold. So in fact it is just the overall sign of k that determinesp which of the manifolds
applies. [You can p see this by making the substitution r → r/ |k| in which case we
can rescale a → a/ |k| as well, and only the overall sign of k matters. Because of this
freedom, you will sometimes see this metric written with the terms and factors scaled
in this way.] This is the famous Friedmann-LeMaı̈tre-Robertson-Walker metric (FLRW,
although sometimes LeMaı̈tre is left out to give only FRW). Amazingly, the scale factor
a and the constant k that we have been using in our Newtonian calculation correspond
to exactly the same quantities in this full analysis.
We can examine the spatial part of this metric separately:

dr2
dχ2 = + r2 dθ2 + r2 sin2 θ dφ2 . (2.53)
1 − kr2
We have removed the common a2 (t) factor which converts to physical coordinates; this
is the metric of a “comoving” constant-time spatial slice of the metric. This looks a bit
like some of the metrics we considered in the previous Subsection 2.5.1: it shares the
dr2 /(1 − kr2 ) term with the circle (exchanging r for y), and the remaining two r2 dΩ2
terms with the spherical manifolds. Combining these, for k > 0, gives the metric of a
three-sphere,
√ that is a three-dimensional manifold all of whose points are a distance
R = 1/ k from a fixed point in higher dimensions (I emphasize that this is not a normal
two-sphere). The discussion above tells us in particular that for k 6= 0 the r coordinate
does not represent distances along that sphere, but can be interpreted as a Cartesian
coordinate in the higher-dimensional space. The r coordinate does, however, represent
an important distance: it gives the radius (in comoving length) of a sphere centred at

Cosmology 15
2 — The Expanding Universe

r = 0. This is evident when we go back and recall the interpretation of the dr2 term
as the metric along the circle. In two dimensions we can think of a sphere as a series
of “nested” circles of radius r (i.e., as if looking down from the top at lines of constant
latitude), and in three dimensions of nested spheres of radius r. Note that the dr term
doesn’t give the metric along each of these spheres or circles, but how they are related to
one another.
The possibilities for the sign of k thus each correspond to one of our constant-curvature
manifolds. The choice k = 0 corresponds to the flat manifold, which should be clear from
the form of the metric, which in this case just looks like Minkowski with the extra factor
of a2 on the spatial part. So at any one time, the Universe acts like Minkowski space,
although of course anything that actually happens takes a finite amount of time, so the
evolution of the Universe must be taken into account.
If k > 0, we say that the Universe has positive curvature, and it takes the form of
a three-sphere at any given time. Analogously to the two-dimensional sphere discussed
above, triangles have a total of more than 180 degrees, lines that start out parallel will
converge and eventually cross, any seemingly-straight line returns to its starting point
eventually, and the Universe has a finite volume at any one time. This corresponds to the
case discussed a few lectures ago where we saw that a0 (t) = 0 at some time, corresponding
to a maximum expansion followed by recollapse: the sphere grows with time from R = 0
and then shrinks.
If k < 0, we say that the Universe has negative curvature, and it is the three-
dimensional saddle shape or hyperboloid. Here, a triangle has less than 180 degrees,
and lines that start out parallel may cross and will diverge, but, like the flat case, the
Universe is infinitely large.

2.5.2 Dynamics of the FRW metric

This FRW metric describes the Universe at a given time, but we haven’t discussed how it
evolves with time, i.e., the evolution of the scale factor a(t). For this we need Einstein’s
Field Equations for GR. For completeness, we can write these down, but we will not
really be able to explain them here:
8πG
Gµν = Tµν (2.54)
c4
where Gµν is the “Einstein Tensor” and encodes information about the curvature of the
Universe (it is a function of the metric and its derivatives) and the right-hand side is
the stress-energy tensor that we have encountered already. We will not to understand
this equation in detail for this course, but an excellent introduction from Baez and Bunn,
which concentrates more on the dynamics of the Einstein equation rather than the metric,
is available at http://math.ucr.edu/home/baez/einstein/.
We won’t reproduce the original arguments of FLRW here, but the result of their anal-
ysis of the field equations for their metric is in fact exactly the same as Equations 2.40-
2.42. Actually, there is one important change that we can make. Einstein’s Field Equa-
tions allow (but do not require) an additional term to be added, giving an extra Λgµν on

16 J. R. Pritchard
 2.5.: General and Special Relativity

the left-hand side, where gµν is the metric, and Λ is the infamous cosmological constant
(“Einstein’s biggest blunder”) — and it really is a constant, a single number given once
and for all. In fact, we’ll see that even this term is surplus to requirements and in fact
can really be subsumed into the stress-energy tensor itself, and despite being a “blunder”
is probably needed to describe the Universe. So, if we include this term, the Friedmann
Equations become  2
ȧ 8πG k 1
= ρ(t) − 2 + Λ , (2.55)
a 3 a 3
and
ä 4πG
1
=− ρ + 3p/c2 + Λ . (2.56)
a 3 3
In these equations, ρ, p and a are all functions of time, and they must be supplemented
with enough information about the behaviour of the mass-energy to solve these equations
(for example, fluid conservation, Eq. 2.42 along with the equation of state parameter,
wc2 = p/ρ).

2.5.3 Redshift
MRR 3.3, 7.4
Almost all of the information we have about the Universe comes to us in the form
of light — photons — that have propagated to us from distant objects. Because of the
finite speed of light, of course, this means that we see objects as they were at some time
in the past (and it also means that defining the distance to an object that you see can
be complicated). Light travels along the geodesic curves of a manifold, essentially locally
straight lines, satisfying ds = 0.
So let’s return to our FRW metric, Eq. 2.52, and look at the case where light travels
along a “radial” curve with constant θ and φ (so dθ = dφ = 0), from the point of emission
at t = te , r = re to our observation today at t = t0 , r = r0 = 0. We then have

dr2
0 = ds2 = c2 dt2 − a2 (t) (2.57)
1 − kr2
or
c dt dr
= −√ (2.58)
a(t) 1 − kr2
where we have chosen the negative sign since both a and t increase to the future, whereas
r increases away from the observer, i.e., toward the past. We can integrate this along the
path from emission to observation:
Z t0 Z 0 Z re
c dt −dr dr
= √ = √ = fk (re ) (2.59)
te a re 1 − kr 2
0 1 − kr2

where we note that the dr integral is a fixed function of re (and k). Because of this,
the equality must still hold for photons emitted at a later time, te + δte and observed at
t0 + δt0 (note that the two δs are different). The δs correspond to, for example, successive

Cosmology 17
2 — The Expanding Universe

peaks in the lightwave, since we’ll see that we want to measure the photon wavelength
or frequency. Hence Z t0 Z t0 +δt0
dt dt
= (2.60)
te a te +δte a

We can now split up the range of integration on both sides to eliminate the “overlap”
between t0 + δt0 to te :
Z te +δte Z t0 Z t0 Z t0 +δt0
dt dt dt dt
+ = + . (2.61)
te a te +δte a te +δte a t0 a

The two “inner” terms are equal, so we are just left with
Z te +δte Z t0 +δt0
dt dt
= . (2.62)
te a t0 a

Now, let’s assume that the δs are small and Taylor expand the integrands:
   
1 1 ȧ 2 1
= 1 − δt + O(δt ) ' (2.63)
a(t + δt) a(t) a a(t)
where the approximation holds if
δt c−1 × photon wavelength
' 1 (2.64)
(ȧ/a)−1 age of Universe

and we’ve used H = ȧ/a ∼ 1/t0 and the fact that we’re considering successive peaks in
the sinusoidal lightwave. This inequality obviously holds for any reasonable times. Now,
applying this to t = t0 and t = te ,
δt0 δte
= (2.65)
a(t0 ) a(te )
which we can rewrite as
a(t0 ) δt0 λ0 /c νe
= = = (2.66)
a(te ) δte λe /c ν0
or
a0
=1+z (2.67)
a
where we define the redshift
νe νe − ν0 δν
z = −1= =
ν0 ν0 ν
λ0 − λe δλ
= = (2.68)
λe λ
What does redshift mean? Most importantly, it means that the wavelength of freely-
propagating photons increases with the expansion of the universe, proportional to the
scale factor. This is a purely General-relativistic effect. Recalling our heuristic Newtonian

18 J. R. Pritchard
 2.5.: General and Special Relativity

derivation of the expansion of the Universe, we can see that the expansion is very much
like climbing out of a potential well, and so it is plausible that moving objects — photons,
in this case — would lose energy as they propagate. Note that although our derivation
required approximations when we used the Taylor expansion, in fact the cosmological
redshift of Eq. 2.67 is exact in a full GR calculation.
We also see that when z > 1 (which we certainly do observe for sufficiently distant
objects), if we use the usual correspondence between redshift and speed, v = cz, we
seem to have a relative speed greater than that of light. On the one hand, there is no
problem here: the equations that give this result are unambiguous. But what does it
mean? We are taught that one of the underlying principles of relativity is v ≤ c, and
yet we seem to contradict this. The answer is that this is a matter of interpretation,
not physics: Is v = cz correct in this circumstance? What does the prohibition against
superluminal velocities mean for very distant objects? We will not go into the details here,
but I will point to a recent paper by two of my colleagues, Bunn and Hogg, available
at http://arxiv.org/abs/0808.1081 and discussed further in Bunn’s blog at http:
//blog.richmond.edu/physicsbunn/2009/12/02/interpreting-the-redshift/.
Interpretation aside, the redshifting of radiation as the Universe expands has some im-
portant and interesting effects. Let us compare photons to our “dust” or “non-relativistic
matter”. As the Universe expands, the number density of matter particles decreases:
nm = N/V ∝ N/L3 ∝ a−3 . Hence the energy density, which is just ρm c2 = mnm c2 for
particle of mass m, also scales as

ρm c2 ∝ a−3 . (2.69)

For photons, the number density is still n ∝ a−3 . However, the energy per particle
is Eγ = pc = hν = hc/λ ∝ a−1 as the wavelength redshifts. Hence, for photons,
ργ c2 = nγ Eγ ∝ a−3 × a−1 or
ργ c2 ∝ a−4 . (2.70)
So far, we have discussed the case of photons, but really this distinction is more general.
Consider the fully relativistic energy per particle, E 2 = p2 c2 + m2 c4 . If the first term
dominates, (mc2  pc), the particles behave like photons, which we will call, generically,
radiation — this applies to anything with zero mass, but also to very low-mass particles,
where “low-mass” really means “high-speed”, since in the early Universe we will see that
all particles move with greater and greater speed as the Universe gets hotter and hotter
as we look earlier and earlier. Conversely, when the second term dominates (mc2  pc),
their energy is dominated by their mass, and they behave like non-relativistic matter.
We will very often use the redshift as a proxy for the time parameter — if the Universe
has always been expanding, then the redshift of a more distant object (hence with an
earlier te is always greater). Obviously, this is the redshift as observed by us, today; it
would have a different value if observed at a very different time. (As we can see from our
derivation, however, as long as the time difference is small compared to the age of the
Universe — which it will be for any observations made in the recent past or foreseeable
future! — the redshift will be the same. We may explore this further in a problem set.)

Cosmology 19
2 — The Expanding Universe

20 J. R. Pritchard
Chapter 3

Cosmological Models and

Parameters

3.1 Cosmological Parameters and the Expansion of

the Universe
MRR 4.8
Let us return to our first-order Friedmann Equation (Eq. 2.55)
 2
ȧ 8πG kc2 1
= ρ(t) − 2 + Λ . (3.1)
a 3 a 3

Consider what this equation looks like if we evaluate everything today, t = t0 . We define
the Hubble Constant as
ȧ
H0 = = 100h km s−1 Mpc−1 (3.2)
a t=t0

where the second equality just parameterizes our ignorance: we are pretty sure from
observations that H0 is between 50 and 100 km s−1 Mpc−1 , so this way the quantity
that we don’t know is 0.5 <
∼h< ∼ 1, which is a bit easier to work with. With this, the
Friedmann equation is
8πGρ0 kc2 1
H02 = − 2 + Λ. (3.3)
3 a0 3
Now, consider a flat Universe with no cosmological constant, k = Λ = 0, in which case
we will say that the density today is the “critical density”, ρ = ρc , given by

8πGρc
H02 = (3.4)
3
or
3H02
ρc = = 1.9h2 × 10−29 g cm−3 . (3.5)
8πG

21
3 — Cosmological Models and Parameters

Even for a universe with ρ 6= ρc , we can define the ratio of the actual density to the
critical density, the density parameter
ρ0
Ω0 = (3.6)
ρc
The two parameters H0 and Ω0 have traditionally been the most important numbers in
modern cosmology, and until the last decade or so, their values were not really known
to better than 50% or so. However, we’ll see right away that we can’t really use only a
single number, Ω0 , to represent the density; we really a need a separate number for each
“kind” of matter.
To see this, let’s rewrite the Friedmann equation in a very useful form. First, let’s
write the expansion rate as
ȧ
H(t) = = H0 E(z) (3.7)
a
which defines a new function E(z), first popularized by James Peebles in his books and
papers, which we will call the “Hubble Function”. The Friedmann equation becomes
8πGρ kc2 Λ
E 2 (z) = 2
− 2 2
+
3H0 H0 a 3H02
ρ(z) kc2 Λ
= − 2 2 (1 + z)2 + . (3.8)
ρc a0 H0 3H02
From the previous chapter, we know that, for non-relativistic matter,
 −3
a
ρm (z) = ρm (0) = ρm (0)(1 + z)3 = ρc Ωm (1 + z)3 (3.9)
a0
and for radiation
 −4
a
ρr (z) = ρr (0) = ρr (0)(1 + z)4 = ρc Ωr (1 + z)4 . (3.10)
a0
We can similarly define
kc2 Λ
Ωk = − and ΩΛ = (3.11)
a20 H02 3H02
(we really shouldn’t think of Ωk , especially, as having anything to do with an energy
density, although we will see that ΩΛ actually is related to an appropriate energy density
ρΛ c2 ), which finally gives
E 2 (z) = Ωm (1 + z)3 + Ωr (1 + z)4 + Ωk (1 + z)2 + ΩΛ . (3.12)
In particular, at z = 0 (today), E(0) = 1, since ȧ/a|0 = H0 , so
1 = Ω m + Ωr + Ωk + ΩΛ (3.13)
where I will always use the quantities Ωi to refer to the value today. Note that this
assumes that Eqs. 3.9-3.10 are exact — that there is no conversion between components
due to particle decay, for example. Because all of the Ωi except Ωk can refer to the actual
density of some sort of “stuff”, we shall also define the total density parameter
Ωtot = Ωm + Ωr + ΩΛ = 1 − Ωk . (3.14)

22 J. R. Pritchard
 3.2.: Cosmological Models

3.1.1 The time-dependence of the density

Let us now start to consider how the Universe evolves with time. Note that each of the
terms in our Hubble Function has a different power of (1 + z) and hence as we move
further and further back in time, the higher powers will dominate. In particular, if we go
back far enough, the Ωr and Ωm terms will dominate over the ΩΛ and Ωk terms — and
hence the Universe will look like a geometrically flat (k = 0) universe with no cosmological
constant (Λ = 0).
Indeed, if we consider observations of the Universe today, we find that the matter
density is Ωm ' 0.3 and that the radiation density can be found by considering the Cosmic
Microwave Background radiation, which has a temperature of about TCMB = 2.73 K, this
corresponds to a density (for a blackbody)
 2 4 
ρCMB 1 π kB
ΩCMB = = T 4 ' 2.5 × 10−5 h−2 (3.15)
ρc ρc c 15(h̄c)3 CMB
2

which is much smaller than the matter density today.

Consider, for example, 1 + z = 1000, which we will see corresponds to a particularly
interesting time in the early Universe. Then,

E 2 (z) = Ωm (103 )3 + Ωr (103 )4 + Ωk (103 )2 + ΩΛ

' 109 Ωm + 1012 Ωr
' 109 Ωm (3.16)

Because the evolution at this time is controlled by the value of Ωm , we say that the
Universe is matter-dominated (MD). However, if we go to a somewhat earlier time (any
z  0.3/2.5 × 105 ), the Ωr term is largest and we say that the Universe is radiation-
dominated (RD). Similarly, if we wait long enough, and if Ωk and ΩΛ are nonzero, we
expect them to dominate eventually. We show all of these in Figure 3.1, although of course
the real universe wouldn’t have such sharp transitions between the different phases. Note
that the actual Universe need not have all of these phases, if one of the Ωi is sufficiently
small or zero. In fact, we’re pretty sure that Ωk = 0 in our Universe, so the curvature-
dominated phase probably doesn’t happen. Moreover, it looks like ΩΛ ' 0.7, so we are
experiencing the change-over from MD to ΛD now.

3.2 Cosmological Models

MRR 4.6
Although Figure 3.1 gives a good picture of the Universe in its various phases, let’s
see what happens in a bit more detail as we vary the parameters. It will be useful to go
back to our first-order Friedmann Equation in its original form, (Eq. 2.55):
 2
ȧ 8πG kc2 1
= ρ(t) − 2 + Λ . (3.17)
a 3 a 3

Cosmology 23
3 — Cosmological Models and Parameters

-4
RD ρ∝a
log ρ

-3
MD ρ∝a

-2
CD ρ∝a
0
ΛD ρ∝a

aEQ

log a

Figure 3.1: The evolution of the density of various components. RD, MD, CD, and ΛD
refer to Radiation-, Matter-, Curvature- and Λ-domination, respectively.

The Empty Universe. First, we will consider a manifestly non-physical Universe,

an open (k < 0) one with no matter, no radiation, and no cosmological constant. This
just gives the very simple equation
p
ȧ2 = kc2 , or a = ± |k| ct (3.18)

so
ȧ
= t−1 . (3.19)
a
This is the “wide open”, or Milne Universe. Note that it also corresponds to the far
future evolution of any open Universe (k = −1 or Ωk > 0 or Ωtot < 1) in which the
matter and radiation densities have diluted so much as to become negligible, as long as
there is no cosmological constant ΩΛ = 0.

The flat, matter-dominated Universe. Now, we take k = 0 and Ωm = 1. We

already considered this in Section 2.3 above. In this case, we have

a ∝ t2/3 (3.20)

which gives
ȧ 2
= t−1 . (3.21)
a 3
This corresponds to early times for a Universe with non-relativistic matter (but not so
early that a radiation component dominates), and is often called the “Einstein-de Sitter”
Universe (not to be confused with the “de Sitter” Universe, which is, unfortunately, a
different case!).

24 J. R. Pritchard
 3.2.: Cosmological Models

The Closed Universe. Now, consider k > 0, Ωtot > 1, or Ωk < 0. We saw earlier
that in this case there is a value of a such that ȧ = 0. This occurs when
8πGρ kc2
= 2 (3.22)
3 a
or (assuming non-relativistic matter so ρ ∝ a−3 )
8πG
amax = ρ0 a30 . (3.23)
3kc2
In fact, the full solution for the evolution of a(t) is symmetric around this point: it climbs
from a = 0 at t = 0 to amax at t = tmax and then falls back to a = 0 at = 2tmax .
So far, we’ve ignored the possibility of a radiation-dominated Universe. In fact, in
terms of the long-term evolution of the Universe, it appears that the radiation-dominated
phase is very short and we are now long after it. Exercise: show that the effect on the age
of the Universe (i.e., the relationship between a and t) of an early RD phase is negligible.
However, we will see that much of the most important “microphysics” of cosmology
happens during the RD phase, so it is worth recalling its properties.

The Flat, radiation-dominated Universe. If we go back far enough in a Universe

with both matter and radiation, we find that is still flat but dominated by radiation
(k = 0 and Ωr = 1), so we have
a ∝ t1/2 (3.24)
and therefore
ȧ 1
= t−1 . (3.25)
a 2
We will see that this applies to sufficiently early times for our Universe.
Let us put together these different cosmological models onto a single graph, Figure 3.2
(although we’ll actually have to use some of the calculations from the next chapter in
order to it). Note a few crucial things. All the Universes start out looking like a flat
Universe, but diverge depending upon the total density (i.e., the sign of k). Note also
that the k = 0 case does not asymptote to a constant value, but it grows more slowly
than k < 0 the open case. Finally, note that if we normalize all three curves to the same
expansion rate at some particular time (e.g., today), at that time the open Universe is
older than the flat Universe, which is older than the closed Universe.

3.2.1 Models with a cosmological constant

MRR 4.7
So far, we’ve assumed no cosmological constant, but there are many interesting pos-
sibilities if we allow Λ 6= 0. More importantly, this seems to describe the real Universe!
We have already seen by examining our Hubble function E(z) as well as Figure 3.1 that
the late-time behaviour in such a Universe is dominated by the cosmological constant.
Going back to the Friedmann Equation for this case, we will have
 2
ȧ Λ
= ; (3.26)
a 3

Cosmology 25
3 — Cosmological Models and Parameters

Figure 3.2: Left: The scale factor as a function of time, for MD universes with k < 0,
k = 0, k > 0. [Courtesy Ned Wright, http://www.astro.ucla.edu/~wright/cosmolog.
htm.] Right, the scale factor for various universes, marked by their values of Ωm (with
ΩΛ = 0 except as marked), normalized to have the same scale factor and expansion rate
today.

this is a constant — the expansion rate does not evolve with time. This has solution
hp i
a ∝ exp Λ/3 t . (3.27)

Unlike the cases we have looked at so far, which had power-law evolution for a(t), this
Universe grows exponentially. We will see that this has important consequences.
However, note that the early time behaviour of the Universe is almost completely
independent of the cosmological constant: the very early Universe looks like a flat RD
Universe, followed by an MD phase.

The Concordance Universe As noted, the real Universe seems to have matter,
radiation and a cosmological constant. Roughly,

Ωk = 0
Ωm ' 0.3
Ωr ' 10−5
ΩΛ = 1 − (Ωm + Ωr ) ' 0.7 . (3.28)

This is occasionally called the “standard cosmological model” or the “concordance cos-
mology”. More details on the current best measurements of these numbers (and others
you will encounter in this course) are at http://lambda.gsfc.nasa.gov.

26 J. R. Pritchard
 3.2.: Cosmological Models

Note that in such a multicomponent Universe the behaviour can be quite complicated.
In Figure 3.2 we see this “ΛCDM Universe”; note that a(t) starts out decelerating but
(around now, t ' t0 ) the Cosmological Constant is beginning to dominate the expansion,
getting closer and closer to the exponential expansion of Eq. 3.27. Note also that this
enables the Universe to be considerably older than other possibilities, even an open
Universe with the same matter density today.
Usually, when we write Ωi we refer to the value today. But obviously this is just
a convenience to refer to quantities that we can measure easily. The density itself is
obviously a function of time, and we can define the critical density as a function of time,
as well, and so make the Ωi functions of time or redshift as well:
 2
3 ȧ 3H02 2
ρc (z) = = E (z) = ρc,0 E 2 (z) , (3.29)
8πG a 8πG

so we can define
ρi (z) ρc,0 Ωi,0 (1 + z)ni
Ωi (z) = =
ρc (z) ρc,0 E 2 (z)
(1 + z)ni
= Ωi,0 (3.30)
E 2 (z)

where we’ve used a zero subscript to make sure it is evident we are referring to today,
t = t0 (but probably will not in the future). For a Universe with radiation, matter and
a cosmological constant, this is shown in Figure 3.3. In a flat Universe, the components
are Ωi = 0 or Ωi = 1 for much of the evolution of the Universe. This is not the case for
an open or closed Universe.
In fact the concordance Universe is a fairly “boring” example of a Universe with a
cosmological constant. As you will see in the problem sheet, if we balance the densities
of the various components just right, we can get a wide variety of behaviours (including
the one that led Einstein to his “greatest blunder”.)

Cosmology 27
3 — Cosmological Models and Parameters

W
1.0

0.8

0.6

0.4
Wm HaL

Wr HaL

0.2
WL HaL

10-8 10-6 10-4 0.01 100 a0

Figure 3.3: The density dependence of the various components of the concordance Uni-
verse. Note that the horizontal axis gives the scale factor of the Universe, with a/a0 = 1
today.

28 J. R. Pritchard
Chapter 4

Cosmography

In this chapter we will discuss the description — and measurement — of times and
distances in a Friedmann-Robertson-Walker univese. How old is the Universe? How far
away is some object that we see? We will see that these questions do not necessarily
have unambiguous answers, and that we must very carefully describe what we want to
measure.

4.1 The Age of the Universe

MRR 4.9
We want to start pinning down the numbers corresponding to the possible FRW
Universes we have been examining. One of the most important is the age. We can see
how to calculate it from some simple manipulations:
Z t0 Z a=a0 Z a0
dt a a
t0 = dt = da = da a−1 . (4.1)
0 a=0 da a 0 ȧ

There are many ways to actually do this integral, but one particularly simple one is to
use the fact that
a 1
= so da = −a0 (1 + z)−2 dz (4.2)
a0 1+z
and we have defined
ȧ
= H0 E(z) (4.3)
a
so
Z ∞
dz
t0 = 2
(1 + z) [H0 E(z)]−1
0 (1 + z)
Z ∞
−1 dz
= H0 , (4.4)
0 (1 + z)E(z)

where E(z) is the Hubble Function of Eq. 3.12. (Since E(z) depends on the quantity
1 + z it may be useful to use 1 + z or 1/(1 + z) as an integration variable.) Exercise: Show

29
4 — Cosmography

that the time ∆t(z1 , z2 ) between two redshifts z1 and z2 is given by the same expression
with the limits of integration changed to z1 and z2 .
Remember that we saw (in a problem sheet) that if we had a constant expansion
velocity ȧ we would have found an age 1/H0 , so we can think of the combination
Z ∞
dz
H0 t0 = (4.5)
0 (1 + z)E(z)

as being entirely due to the change in expansion velocity, which is of course due to the
presence of matter in the Universe. For most of the models we will consider, the quantity
H0 t0 is of order one. For example, in a flat, matter-dominated Universe, H0 t0 = 2/3
Considering Figure 3.2, this is the difference between a tangent line stretching back
from today to the horizontal axis compared to the actual curve. In this figure, a(t) always
has a negative second derivative (expansion decelerates). This is because we only showed
Universes with no cosmological constant, but in fact once we allow ΩΛ > 0, we can have
ä > 0 which can actually make the Universe “older” than it would be with constant ȧ,
as we saw in the previous chapter, Figure 3.2.
This fact has been crucial in our understanding of our actual Universe. We find
from measurements that H0−1 ' 14 Gyr and in fact find objects in the Universe that we
believe to be roughly 11–12 Gyr old. Since this gives H0 t0 ' 0.8 >∼ 2/3, it seems that the
simplest FRW Universe — flat and matter-dominated — cannot be the case! Either our
measurements are wrong, or we need the Universe to be dominated by a form of matter
which gives a larger value of H0 t0 . But we have seen that this will have the effect of an
accelerating scale factor, or a positive value of ä. This seems a very strange thing when
we recall the equation that governs the acceleration of the expansion:

ä 4πG
Λ
=− ρ + 3p/c2 + (4.6)
a 3 3

Obviously Λ > 0 will do this, but it is worth considering the strange-seeming possibility
that Λ = 0 but that the first term is overall positive — which means that somehow
ρ + 3p/c2 < 0.

4.2 Horizons
MRR 4.10
Information cannot travel faster than the speed of light, or more precisely causality
cannot act faster than the speed of light. This is one of the core principles of physics
since the early twentieth century, affirmed in relativity and quantum mechanics alike.
Light rays travel on null geodesics, which in our metric notation corresponds to ds2 =
0. If we assume dφ = dθ = 0 as in our redshift calculation of Section 2.5.3. We can first
ask, at what distance would a light ray have had to start to arrive here at time t? We
show a spacetime diagram of this situation in Figure 4.1.

30 J. R. Pritchard
 4.2.: Horizons

ae × χe

Past A B a=a(te)

"Light
Cone"
Expansion

Present a=a0
A B
a0 × χe

Figure 4.1: Two-dimensional spacetime diagram of horizons in the expanding Universe

— space is horizontal, time vertical. The star represents our place of observations, here,
today, and the diagonal solid lines represent our past light cone. χe represents the comov-
ing coordinate distance (as if measured at a particular time with the scale factor a = 1)
between us (at the star) and point B on the light cone, so a0 ×χe gives the proper distance
to the present-day location of B. If te = 0 then χe = χH is the comoving distance to the
horizon, and dH = a0 × χH gives the proper distance to the present-day location of our
horizon. (It is important also to realize that except for the flat k = 0 case, the lines of
constant t represent a curved universe, and thus there is a distinction between the value
of the radial coordinate, re , and the comoving distance, χe .)

As we have emphasized, whenever we talk about distances, we need to be very precise

in our definition. So first, let us recall that the form of the FRW metric is
dr2
 
2 2 2 2 2 2 2 2 2 2
ds = c dt − a (t)dχ = c dt − a (t) + r dΩ (4.7)
1 − kr2

where dχ2 gives the metric of a three-dimensional spatial “slice” of spacetime, in comoving
coordinates, and dΩ2 = dθ2 + sin2 θdφ2 gives the angular metric. We need to then use
the same equation as for the redshift calculation, for a light ray with dθ = dφ = 0:
c dt dr
= −√ . (4.8)
a(t) 1 − kr2
Note that this equation is just dχ, a differential element of comoving distance. As in the
redhift case, we put t = 0 at r = rH , and t = t0 at r = 0 and hence a negative sign
appears. We can integrate this along the line of sight from t = 0, r = 0 to t, r = rH , after
switching the limits of integration because of that negative sign (equivalently, we could
just use r = 0 at t = 0 and r = rH at t):
Z t Z rH
c dt dr
= √ ≡ χH (t, rH ) (4.9)
0 a(t) 0 1 − kr2
where we define χH , the comoving distance to the horizon at comoving coordinate value
rH . For a flat Universe (k = 0) these two are the same. However, for k 6= 0, they are

Cosmology 31
4 — Cosmography

not, and in particular, χH corresponds to the comoving distance to the sphere — along
a t = const surface — defined by coordinate rH . As we discussed in Section 2.5, r does
not correspond to a distance along the manifold; for example in the spherical case it is
basically the distance perpendicular to the axis defining r = 0, or alternately the radius
of the two-sphere labelled by r. (To see this, think of the r coordinate as referring to the
distance from the axis to a constant latitude circle of a 2-sphere.)
Finally, we convert this coordinate distance to a proper distance to the horizon by
multiplying by the scale factor, a(t),
Z t
c dt0
dH (t) = a(t)χH (t) = a(t) 0
. (4.10)
0 a(t )

We can recast this in a useful form by returning to our Hubble function, E(z), defined
by ȧ/a = H0 E(z), so
Z a Z a 0 Z ∞
dt a0 0 a 1 0 −1 a dz 0
dH (t) = ca 0 02
da = ca 0 02
da = cH0
0 da a 0 ȧ a a0 z E(z 0 )
∞
cH0−1 dz 0
Z
= . (4.11)
1 + z z E(z 0 )
In particular, the horizon distance today R is just this3/2evaluated at z = 0. For a flat,
−1 ∞ −1
Ωm = 1 Universe, we have dH = cH0 0 dz/(1 + z) = 2cH0 = 3ct0 where we have
done the integral and then used H0 t0 = 2/3 for this case from above. Note that this is
greater than the naive expectation of dH ∼ ct0 , the simple distance traveled by a photon
in the time from t = 0 to t = t0 , as dH is the distance today from r = 0 to the present-day
position defined by coordinate r = rH . As we see in the Figure 4.1, expansion means
that this position is further away than ct0 .

4.3 Distances
It is often useful to be able to describe the distance between different points (events)
on a spacetime manifold, but in GR there is usually no unambiguous definition of the
distance between two points. Rather, we must state very explicitly what we mean. If
we could freeze the Universe at a particular time, it might make sense to use the spatial
coordinate distance between those two points, and indeed that is an easy quantity to
work with. However, in practice we can only observe points that are on our “light cone”
— at a given redshift we observe objects at a particular coordinate distance from us at
a particular time in the evolution of the Universe. We have already seen this in our
discussion of the redshift. Since a geodesic (light ray) satisfies ds ≡ 0, we have
dr2
0 = ds2 = c2 dt2 − a2 (t) (4.12)
1 − kr2
which we can reorder and integrate along the path from emission to observation (again,
we can either define r backward along a light-ray coming toward us or vice versa):
Z t0 Z re
c dt dr
χe ≡ = √ = fk (re ) (4.13)
te a 0 1 − kr2

32 J. R. Pritchard
 4.3.: Distances

where we note that the dr integral is a fixed function of re (and k):


sin−1 ( |k| re )
−1/2
p
|k|
 k>0
−1
fk (re ) = Sk (re ) ≡ re k=0 (4.14)
−1

 −1/2 p
|k| sinh ( |k| re ) k < 0

which just converts to distance along the circle, χe , from the “Cartesian coordinate”
which r represents as we saw when discussing the geometry of the circle in Section 2.5.1.
Recall that we had k 2 = 1/R where R is the radius of curvature (which is the actual
radius of the sphere for k > 0) so these make dimensional sense.
We can solve this equation for re ,
 p 
−1/2


 |k| sin |k| χe k>0
re = χe
 p  k=0
|k|−1/2 sinh

|k| χe k<0
p 
≡ |k|−1/2 sink |k| χe
c −1/2
p
−1

= |Ωk | sink |Ωk |a0 H0 c χe , (4.15)
a0 H0
which we writep in the unified form defining sink for simplicity (and taking sink (x) = x and
cancelling the |k| factors when k = 0 — we can also think of this as the k → 0 limits
of the positive and negative k cases). We have used the definition of Ωk ≡ −kc2 /(a0 H0 )2
to eliminate k. We can also put the χe integral into a form we can use more simply:
Z t0 Z t0 Z ze Z ze
−1 dt dt da dz a 1 dz
a0 H0 c χe = a0 H0 = a0 H0 = a0 H0 = (4.16)
te a te da a 0 1 + z ȧ a 0 E(z)

[Note: very often, cosmologists will absorb R into the definition of the r coordinate or
the scale factor a and take k = ±1 or k = 0. Dimensionally, the form of the FRW metric
dictates that kr2 has no units (from the 1 − kr2 factor) and that ar has units of length.
This can be realized by giving a units of length and making both k and r unitless, or
by making a unitless, giving r units of length, and k units of length−2 . There is further
freedom of scaling (i.e., units): very often you will see the convention a0 = 1 so that
distances are measured as compared to the present day. Note that finally we write these
equations in terms of Ωk , which is unambiguously defined.]

4.3.1 Luminosity Distance

MRR 7.6
From this light-cone coordinate distance we can make our distance measures yet more
physical by relating them to a possible observation. For example, imagine that we have
an object whose luminosity (energy emitted per unit time) is given by L. On Earth, we
measure a flux F from this object which in (flat, non-expanding) Euclidean space is given

Cosmology 33
4 — Cosmography

by F = L/(4πd2 ) for an object at distance d. We can use this to define the “luminosity
distance”, dL , in a curved and expanding spacetime:

L
F= (4.17)
4πd2L

where dL can only depend upon the redshift once we have fixed the spacetime. To
calculate it, consider a spherical coordinate system with the origin at the source, S, at
radial coordinate re (so that our observation point is on the sphere), emitting at time te ,
observed at t0 . The total area of the sphere that goes through our observation point at
t0 is
Z Z
2 2 2 2
A = a0 re dΩ = a0 re d cos θ dφ = 4πa20 re2 . (4.18)

This equation uses the scale factor today, a0 , because this gives the proper, physical,
area of the sphere at t0 . Because of the expansion of the Universe and the redshift,
the total rate of energy coming through the sphere is modified from its rate at emission
by two effects. First, the number of photons coming through the sphere per unit time
is decreased by a0 /ae = 1 + ze . Second, each of those photons is doppler shifted so its
energy is decreased by an additional factor of a0 /ae = 1+ze . Note that these are different
effects, and hence the energy per unit time is decreased by (1 + z)2 . (Another way to
think of this is that the total energy, integrated over all time, would be decreased by one
factor of 1 + z due to redshift, but that energy is spread over a time that is longer by
another factor of 1 + z.) Hence,

L
F= (4.19)
4πa20 re2 (1 + z)2

so that
dL = a0 re (1 + z) . (4.20)

Now, we can combine this with our solution from above for re (Eqs. 4.15-4.16) to give
 Z ze 
p dz
dL (ze ) = cH0−1 (1 + ze )|Ωk |−1/2
sink |Ωk | . (4.21)
0 E(z)

Note in particular that for k = 0,

Z ze
dz
dL = cH0−1 (1 + ze ) (k = 0) . (4.22)
0 E(z)

These expressions do not depend on the characteristics of the source (L) nor on the
measured flux, but do depend on the cosmological parameters through the factors of H0
and E(z) in the integral, as well as k itself.

34 J. R. Pritchard
 4.3.: Distances

4.3.2 Angular-Diameter Distance

MRR 7.8
Now, consider a slightly different measurement. Instead of an object of known lumi-
nosity, consider one of known size (for example, perhaps there is some galaxy that we
know is 10 kiloparsecs in diameter). Of course from here we measure not the size, but
the angle that it subtends on the sky. In Euclidean geometry, the diameter D is related
to the observed angle θ if we know its distance, d, by θ = D/d, which motivates the
definition of the angular-diameter distance in more general circumstances:
D
dA = . (4.23)
θ
From the form of the FRW metric (consider the dθ term), the proper distance along a
circle (assuming θ  1 so we need not distinguish between the chord and the arc) at
radius re should be D = re ae × θ, so
ae a0 re
dA = re ae = a0 re = . (4.24)
a0 1 + ze
Note that this equation uses the scale factor at emission, ae , since it is the proper cir-
cumference of the circle at te that matters (compare the luminosity-distance calculation
above, Eq. 4.18, in which the proper area of a sphere at t0 entered). As above, we
substitute in our calculation of re to get
 Z ze 
−1 −1 −1/2
p dz
dA (ze ) = cH0 (1 + ze ) |Ωk | sink |Ωk | . (4.25)
0 E(z)

Note that dA = dL /(1 + z)2 , independent of the cosmological parameters. Further,

the observable quantity, the surface brightness, defined by the flux per solid angle on the
sky
F L 1
S= 2 = (4.26)
θ 4πD (1 + z)4
2

is independent of the cosmology (for fixed physical size D and luminosity L).
In Figure 4.2 we show the behaviour of the luminosity and angular-diameter distances
as a function of redshift. Note that the angular diameter distance actually turns over:
this means that objects actually can start getting bigger as they get further away! Un-
fortunately, due to the dimming of the surface brightness (Eq. 4.26), they become harder
and harder to see, very rapidly, as the energy is spread out over a wider and wider area
of sky.

4.3.3 The Extragalactic Distance Ladder

See the in-class presentation.

Cosmology 35
4 — Cosmography

Figure 4.2: Distance measures and the age of the Universe in different cosmologies.
Distances are measured in units of c/H0 and times in units of 1/H0 .

36 J. R. Pritchard
Chapter 5

Thermodynamics and Particle

Physics

MRR 5.1–5.2
Today, the evolution of the Universe is transitioning from being dominated by non-
relativistic matter to dark energy, while the temperature of the Cosmic Microwave Back-
ground is T ' 2.73 K (there are more energetic photons produced by astrophysical
processes as well). Because of this cold temperature, particle energies are low and in-
teractions are rare (except in places like stars, or CERN, of course). As we trace the
evolution of the Universe backwards, however, it heats up and the mean particle energy
increases, so interactions become more and more likely. Hence, we need to understand
the relationship between thermodynamics and particle physics in the early Universe,
sometimes called the “primeval fireball”.

5.1 Radiation Domination

We have already seen that the density of radiation scales as ρr ∝ a−4 whereas that of
nonrelativistic matter scales as ρm ∝ a−3 as the universe expands with scale factor a(t)
[and recall that a/a0 = 1/(1 + z) relates redshift to the scale factor]. Today, the radiation
density is roughly 10−4 that of matter, but in the past it would have been higher, and at
some point the Universe must have been radiation dominated.
In the absence of any processes interconverting matter and radiation, this would have
occurred at a time teq defined by

ρm (teq ) = ρr (teq ) (5.1)

or

ρm (t0 ) (1 + zeq )3 = ρr (t0 ) (1 + zeq )4

Ωm ρc (1 + zeq )3 = Ωr ρc (1 + zeq )4
Ωm 0.3
1 + zeq = ' ' 3, 700 . (5.2)
Ωr 8 × 10−5

37
 5 — Thermodynamics and Particle Physics

(This number is probably uncertain to 10-20%.) At redshifts higher than zeq , the Universe
was radiation dominated. At this time, we have Teq = (1+zeq )T0 ∼ 104 K or kTeq ∼ 1 eV.

5.2 Black Body radiation and equilibrium statistics

The cosmic microwave background (CMB) is almost a perfect black body, as we see from
the spectrum in Figure 5.1. As we will see in a problem sheet, the effect of redshifting
upon a black body distribution of photons is just a black body at the new temperature
T (z) = (1 + z)T0 . That is, if we observe photons from a black body at temperature T0
today, they would have been at (1 + z)T0 at redshift z.
584 FIXSEN ET AL. Vol. 473

FIG. 4.ÈUniform spectrum and Ðt to Planck blackbody (T ). Uncertainties are a small fraction of the line thickness.
Figure 5.1: The measured spectrum of the Cosmic Microwave Background, plotted along
with a black body curve with T = 2.728 K. Uncertainties are a small fraction of the line
of the spectra derivedFrom
thickness. from the DIRBE
Fixsen templates
et al, can be
in the J., 473,
Astrophys. 576 (1996).on the unknown parameters p, G0 , and
performed
principal Ðt, g (l), and Ðt for their intensities at each pixel. *T . The Ðrst two terms are the Planck blackbody spectrum,
k
The CMBR temperature assumes a Planck spectrum. These with the temperature T ] *T . It is important to have the
spectra were chosen to match the shape of the spectrum in second term in order to0 properly estimate the uncertainty
The blackbody distribution of
the data. This yields maps of the CMBR temperature and
photons of frequency ν is given by the Planck function:
since the *T is strongly correlated with the resulting p (95%
dust intensities. A monopole plus three dipole components in the case of the Bose-Einstein distortion). The third term
are then Ðtted to the resulting temperature map. 8π allows
ν 2 dνfor Galactic contamination to remain in the mono-
The vector sum of the dipole coefficientsn(ν) points
dν in=the3 pole spectrum. The . Ðnal term is the modeled deviation.
(5.3) We
direction (l, b) \ (264¡.14 ^ 0.15, 48¡.26 ^ 0.15), consistent c exp(hν/kT ) −Kompaneets
Ðt either the 1 parameter or the chemical poten-
with the direction from the DMR results. Data for tial, but the two are too similar to Ðt simultaneously. The
o b o \ 10¡ Here,
were excluded from theconstant
h is Planck’s dipole Ðt because
so hν is of the uncertainties
the photon energy,are propagated
and the energyfrom distribution
the template Ðts, is and the
potential inaccuracy of the model of the Galaxy. The direc- correlation between the g(l) and LS /Lp increases the uncer-
therefore given by ε(ν) dν = (hν)n(ν) dν. c
tion is particularly sensitive to the Galaxy because it is tainty of G and p.
0
almost orthogonal to the direction of the Galactic center.
Galactic variations
38 in spectral shape as a function of longi- 6.1. Galactic Contamination
J. R. Pritchard
tude couple into the angle directly, while for the Ðxed angle Most of the Galactic emission has been removed, but
case, they come in as second-order terms for the same there is a small residual contamination. We use either the
reason. ; g (l) derived from the all-sky data set (see ° 4) or the
k (T k ), and we Ðt a temperature and an emissivity. The
l2B
l
l2B (T ) model with a temperature of 9 K produces a lower
 5.2.: Black Body radiation and equilibrium statistics

The total number density (in a proper volume) of CMB photons is

Z  3 Z ∞ 2
kT y dy
nγ = n(ν) dν = 8π
hc 0 ey − 1
 3
kT
= 16πζ(3)
hc
' 413. cm−3 (5.4)
where ζ(3) ' 1.202 is the Riemann zeta function, and we have evaluated the expression
at T0 = 2.73 K in the last equality. Note that the number density is proportional to
T 3 = (1 + z)3 T03 — as we already knew, the proper number density scales with the
expansion of the Universe, and hence the comoving density is constant.
We can compare the present-day density to the average number density of baryons in
today’s Universe, using the measured value of Ωb h2 ' 0.023:1
0.023 × 1.9 × 10−29 gcm−3
nb = Ωb ρc /mb ' ' 2.6 × 10−7 cm−3 . (5.5)
1.67 × 10−24 g
There are
nb 3 × 10−7
η≡ ' ∼ 10−9 (5.6)
nγ 400
baryons for every photon. This small number — which as we will see has been roughly
constant since the first minutes after the big bang — will have many important effects
on the present-day abundances of the elements.

Massive Particles More generally, of course, the equilibrium distribution of a par-

ticle’s energy is given by the Bose-Einstein or Fermi-Dirac distribution for integer or
half-integer spins, respectively. The number density of particles per mode is given by
1 d3 p
n(p)d3 p = (5.7)
e((p)−µ)/kT ± 1 h3
with the plus sign for fermions and the minus sign for bosons. The function tells us the
distribution of particle momenta, related to the energy per particle 2 = m2 c4 + p2 c2 .
For photons, to convert this into the Planck distribution function, Eq. 5.3, 4πν 2 dν =
d3 ν = (c/h)3 d3 p from the density of states — the accessible volume of three-dimensional
momentum space—and the remaining factor comes from the g = 2 polarization (helicity)
states available.
To use this to calculate the overall number density, it is easiest to convert this expres-
sion into an integral over energy:
Z √2
 − m2 c4
Z Z
3 2 4πg
n = g n(p) d p = 4πg n(p)p dp = 3 3  d (5.8)
ch e(−µ)/kT ± 1
1
In this expression, h is the Hubble constant in the form H0 = 100hkm/ sec /Mpc. You will very
often see densities expressed in the form Ωi h2 since Ωi = 8πGρi /(3H02 ) and this form lets us take into
account our remaining ignorance of the Hubble constant.

Cosmology 39
5 — Thermodynamics and Particle Physics

where g is the number of degrees of freedom for the particle species (e.g., g = 2 for
spin-1/2 fermions). We can similarly calculate the energy density
Z
ε = g (p)n(p) d3 p (5.9)

as well as quantities like the average pressure (and hence the equation of state relating
density, temperature and pressure).
Note that if   mc2 , the energy is dominated by the momentum term, just like for
photons. However, the majority of particles in either the BE or FD distribution have
energy  ∼ kT . Hence, if kT  mc2 , the mass of the particles are irrelevant — they
behave like radiation, and this is true for both bosons and fermions. (Strictly speaking,
we are assuming the chemical potential is much less than kT as well.)
In this relativistic limit, when we also assume µ  m (non-degenerate), we find
(
π2 4 7/8 Fermions
ε=g (kT ) ×
30(h̄c)3 1 Bosons
(
ζ(3) 3 3/4 Fermions
n=g 2 (kT ) ×
π (h̄c)3 1 Bosons
ε
p= (5.10)
3
where the last line gives the pressure, and we use h̄ = h/(2π). The ratio of the energy
density to the number density gives the average energy per particle,
(
3.15 kT Fermions
hEi = ε/n ' (5.11)
2.70 kT Bosons
which we will very often just take to be hEi ∼ 3 kT .
This has important cosmological consequences. At earlier times, the temperature was
higher by the usual redshift factor T = (1 + z)T0 . For any particle of mass m, there is a
redshift such that kT > mc2 , before which the particle essentially behaves like radiation.
Hence, as we go backwards in time, the number density — and hence energy density
— of particles behaving like radiation (for example, having equation of state p = ε/3)
increases. Because the expansion of the Universe depends on the density of radiation-like
and non-relativistic-matter-like particles separately, we need to take this into account
when describing the expansion of the Universe at early times. The total energy density
requires summing Eq. 5.10 for ε over all relativistic particle species i, which as we will
see may each have a different temperature Ti , giving
π2
εr = g∗ (T ) (kT )4 (5.12)
30(h̄c)3
where we define the effective degrees of freedom as
 4  4
X Ti 7 X Ti
g∗ = gi + gi (5.13)
i∈bosons
T 8 i∈fermions T

40 J. R. Pritchard
 5.2.: Black Body radiation and equilibrium statistics

(It is conventional to use T = (1 + z)T0 , the temperature of the photons.)

Once the temperature drops below kT = mc2 , however, the behaviour of the particle
changes. In this limit, we can calculate the number density:

 3/2
mkT 2 −µ)/kT
nP = g e−(mc (5.14)
2πh̄2

This is the number density in a proper volume element. If we want to know the number
density in a comoving volume element, we must rescale by the ratio of the comoving to
proper volume, proportional to a3 ∝ (1 + z)−3 ∝ T −3 .
In Figure 5.2 we combine these and show the comoving density as a function of
temperature (and hence redshift, since T = (1 + z)T0 ) for such a particle, assuming
µ = 0 (no degeneracy). For mc2  kT , this is a constant — the particle behaves
like radiation. For mc2  kT the proper density decreases exponentially as Eq. 5.14.
(Numerical integration of the Fermi-Dirac or Bose-Einstein distributions is required to
interpolate between the two limits.)

n¥

0.1

0.001

10-5

10-7

10-9
m c2
0.01 0.1 1 10 100 kT

Figure 5.2: The equilibrium comoving number density of a non-degenerate particle species
as a function of temperature. We plot the density relative to the number density at
T → ∞ (more properly, the m → 0 limit of Eq. 5.8) to remove the overall dependence
upon the particle properties and temperature. Both BE and FD densities are shown, but
they are nearly indistinguishable.

Note that these distributions apply to particle species in equilibrium; hence we must
take into account the interactions that these particles experience. Indeed, this will become
crucial as we discuss the relics of the hot early Universe: the light elements, the microwave
background itself, and the dark matter. If all of the particles in the Universe were in
equilibrium, the abundances of, say, Hydrogen atoms, would be much less than that
12
observed – using m = mp and T = T0 in Eq. 5.14 gives nP ∼ 10−10 cm−3 !

Cosmology 41
5 — Thermodynamics and Particle Physics

5.3 The Hot Big Bang

5.3.1 Interactions
As the universe expands, its state is determined by the temperature, T = (1 + z)T0 , and
the expansion rate, H = ȧ/a. In a radiation-dominated Universe (a ∝ t1/2 ), the latter is
ȧ 11
H= = ∝ a−2 ∝ (1 + z)2 . (5.15)
a 2t
Any interactions between particle species are governed by
• Interaction rates, Γ (with units 1/time), which may depend on the temperature,
particle masses, and cross sections; and

• Particle masses, which determine the equilibrium abundances.

From above, we know that when kT  mc2 , the particle acts like radiation, so E ' pc.
If the interaction with photons is sufficiently strong, it can interconvert with photons and
will have the same temperature.
The interaction rate Γ tells us the mean time between interactions, 1/Γ, or the mean
free path λ ∼ v/Γ ∼ c/Γ (since the particles behave like radiation with v = c). Our
calculation of the horizon distance, dH ∼ ct ∼ c/H gives us the maximum distance
between particles in causal contact, hence when λ  dH we expect that interactions will
be unable to maintain thermal equilibrium.
To see this another way, the number of interactions that a particle species has from
time t onward is Z ∞
Nint = Γ(t0 ) dt0 . (5.16)
t

If we take the typical case that Γ ∝ T , then, when we also have a ∝ tm (m = 1/2 for
n

RD and m = 2/3 for MD), we find that

 
1 Γ
Nint = (5.17)
n − 1/m H t

(Exercise: Show this.) The average number of interactions experienced by a particle after
time t is therefore less than one if Γ < H, as long as n > 1/m.
By both of these arguments, or by our more detailed discussion of the Boltzmann
equation below, when Γ  H, each particle will experience interactions, and can plau-
sibly remain in thermal equilibrium. However, if H  Γ, the Universe expands away
more rapidly than the particles can interact, and most particles do not experience the
interaction, and hence fall out of equilibrium. We call this departure from equilibrium
freeze-out, as the comoving abundance of a species that has Γ  H is frozen (in the
absence of any further interactions, of course). Note that a species can freeze out either
when it is behaving like radiation (kT  mc2 ) or matter (kT  mc2 ). In the former
(relativistic freeze-out) we say that the outcome is a hot relic, and in the latter (massive
freeze-out) a cold relic.

42 J. R. Pritchard
 5.4.: Relic Abundances

5.3.2 Thermal history of the Universe

Even before taking detailed account of the interactions between different sorts of particles,
we have enough information to put together a timeline for the Universe. As we go
further back in time, the Universe gets hotter and hotter as T = (1 + z)T0 . As the
temperature rises, the average energy of particles increases, allowing more and more kinds
of interactions to take place (just as higher energies are needed in particle accelerators
to access more interactions). Hence, at the earliest times all possible particles were
relativistic, and if interactions were strong enough, they remained in thermal equilibrium.
As the Universe cools (i.e., thinking forward in time), particles become non-relativistic
and possibly fall out of equilibrium, possibly leaving around relics that we see today.

kT (GeV) t (s)
1019 10−43 The Planck Era: Classical GR breaks down
1014–16 10−(35–38) Thermal Equilibrium established; GUT transition? (MX ∼ 1015 GeV)
102 10−11 Electroweak phase transition (MW )
0.1–0.5 10−(5–6) Quark confinement (“chiral symmetry breaking”)
? ? Baryogenesis
10−1 10−4 µ+ µ− annihilation (and freeze-out)
? ? Dark Matter interactions freeze out?
10−3 1 ν decoupling
5 × 10−4 3–4 e+ e− annihilation, leaves mainly γ, (ν ν̄) separately in equilibrium
10−4 180 Nucleosynthesis ⇒ He4 , D, T, Li
10−(8–10) 1010–11 Matter Domination
10−(10–11) 1011–13 H recombination (e + p → H + γ)
Universe becomes neutral and transparent

Table 5.1: The history of the Universe.

In Table 5.3.2 we show some of the most important “events” in the history of the
Universe. Some terms, such as “chiral symmetry breaking” and “GUT Transition” come
from particle physics and will not be discussed much further here. Others, such as de-
coupling and freeze-out refer to what happens when interactions fall out of equilibrium
and are crucial to the history of the Universe. I have left a few spaces filled with question
marks, because they depend to some extent on the as-yet unknown physics of the dark
matter and of the creation of the present-day matter/antimatter asymmetry.

5.4 Relic Abundances

As we noted above, if all of the particles in the Universe were in equilibrium, it would be
very boring indeed. But there is ample evidence that many interactions are no longer in
equilibrium:
• Different particle species have different temperatures;

Cosmology 43
5 — Thermodynamics and Particle Physics

• There is considerably more matter than antimatter;

• There are nucleons rather than quarks;

• There are atoms rather than ions; and

• There is structure, rather than a smooth featureless gas at a single temperature.

How did the Universe get to filled with clumpy, baryonic, neutral, matter? To put the
question another way: Why is the Universe interesting?

5.4.1 Baryogenesis and the Sakharov Conditions

One of the most basic of these questions regards the difference between matter and
antimatter. If there were copious amounts of antimatter in the Universe, we would see
it as a background of gamma rays from electron-positron annihilation (as well as more
massive particles and antiparticles). The lack of such a strong background puts stringent
limits on the fraction of antimatter in the Universe, many orders of magnitude below a
1:1 ratio.
So how do we achieve this asymmetric state? Of course, one possibility is just “initial
conditions”: the asymmetry is somehow built into the big bang. We would prefer a
dynamical solution, one that starts from a more symmetric state but somehow ends up
with the highly asymmetric Universe in which we live.
But in fact, the Universe is not as asymmetric as the complete lack of antimatter
makes it seem. Consider again our determination earlier of the baryon-to-photon ratio,
η ∼ 10−9 . For every billion photons, baryon, there is roughly one baryon or, equivalently
(from the overall charge-neutrality of the Universe) roughly one electron. When the
Universe was hot enough that electrons and positrons were considered radiation (so kT 
me c2 ' 0.511 MeV), photons and electrons/positrons would have been in equilibrium with
roughly equal number densities. If there were exactly equal numbers of positrons and
electrons, when the Universe cooled enough so that electrons and positrons annihilated
into photons, the net electron number (and hence baryon number) would have been
essentially zero. But if instead there were just one extra electron for every 109 positrons,
that electron (and its associated baryon) would have been left over, with each of the
matching electron/positron pairs creating a pair of photons via e+ e− → γγ. Hence, the
complete present-day asymmetry between matter and antimatter, is actually only due to
a one-in-a-billion asymmetry at earlier times.
The question of how to arise at even this small asymmetry was considered by the
Soviet physicist Andrei Sakharov.2 He realized that there were three conditions that must
hold in order to convert a matter/antimatter-symmetric Universe into one dominated by
matter:
2
Sakharov was known as the “father of the Soviet nuclear bomb”, but in later life was a dissident,
protesting against the regime. More importantly for our purposes, he also dabbled in astrophysics and
cosmology.

44 J. R. Pritchard
 5.4.: Relic Abundances

1. Baryon number violation. This is the most obvious condition: particle physics must
allow interactions which change the number of baryons. So, for example, we must
allow interactions like
X + Ȳ → B (5.18)
where X is some particle, Ȳ some antiparticle (so the left hand side has net baryon
number B = 0), and B represents some state of particles with net baryon number
B 6= 0.

2. CP violation. If CP is not violated, then in quantum field theory for any interaction,
the related interaction with all particles replaced by antiparticles and vice versa will
have exactly the same rate. That is, the rates of the following two interactions must
be different:

X + Ȳ → B
X̄ + Y → B̄ . (5.19)

Any net change in baryon number (allowed by condition 1) will otherwise be can-
celled by a change in antibaryon number.

3. Finally, there must be a departure from thermal equilibrium. In thermal equilib-

rium, the rate of a reaction going forward (X + Ȳ → B) will be the same as the
rate of the reaction going backward (B → X + Ȳ ), again cancelling out any change
in baryon number.
These are the Sakharov conditions.
In fact, we know that all three of these conditions do occur. First, although the
standard model of particle physics conserves baryon number at so-called “tree level”
(Feynman diagrams) there are non-perturbative quantum effects that can violate baryon
number and change the amount of matter relative to antimatter. However, it is not clear
that this idea, known as electroweak baryogenesis actually gives us a Universe with the
correct baryon number. More generally, most extensions to the standard model (e.g.,
supersymmetry, grand unified theories) violate baryon number in other ways, such as by
the decay of new, massive particles.
Second, it has been known since the 1950s that the standard model also exhibits CP
violation. In this case, it seems clear that on its own this is insufficient to account for
the present-day matter/antimatter asymmetry.
Finally, the expansion of the Universe itself provides the opportunity for departures
from thermal equilibrium. As we discussed above, and in more detail very soon, when
the mean time between interactions becomes long compared with the expansion rate of
the Universe, interactions cannot happen often enough to maintain equilibrium.
The specifics of these conditions are needed to produce the observed baryon asymme-
try. However, there are equivalent conditions for producing any of the other “interesting”
features of the Universe above: the physics must allow the asymmetry to be produced,
to not be cancelled out, and to stay away from the naive abundance dictated by ther-
mal equilibrium. In fact, baryogenesis is probably the most difficult case, as we are not

Cosmology 45
5 — Thermodynamics and Particle Physics

completely sure where to get the correct baryon number and CP violation, whereas the
physics underlying the preference of, say, nuclei over quarks, and neutral atoms over ions,
is very well understood.

5.4.2 Interaction rates and the Boltzmann Equation

Even when such conditions are satisfied, in order to calculate the abundance of some
particle species in more generality, we need to consider its interactions in detail. We
start with the distribution function, f (t, x, p), the time-dependent density of particles
in position and momentum (there is a relativistic generalization of this, of course). We
have already written down the momentum distribution for particles in equilibrium in
Section 5.2 above, but now we want to calculate it more generally. The tool for this is
the Boltzmann Equation. In its most general form, the Boltzmann equation merely
says that the only interactions can change the distribution function:

Dfi
= C [{fj }] (5.20)
Dt

where Dfi /Dt is the Liouville operator, just the total time derivative of the distribution
function for particle species i, and the quantity C[{fj }] represents the effect of inter-
actions, which can possibly depend on the distribution function of all other species, j.
In general relativity, the d/dt operation can actually be taken to include the effects of
gravity, more or less automatically by correctly accounting for the coordinates that we
use to describe the manifold. Even with Newtonian gravity, it is still easiest to account
for its effects on the left-hand side, through the Liouville operator. Indeed, we already
know how to do this for the FRW metric.
We will still assume an FRW Universe homogeneous and isotropic and any given time.
Hence, we must take into account the expansion of the Universe. Basically, the equation
must account for the fact that, in the absence of interactions, the comoving number
density of particles is conserved. In fact we already know an equation that describes this:
it is just our non-relativistic matter conservation equation (the fluid equation), Eq. 2.19:

dn ȧ
+3 n=0 (5.21)
dt a
where n gives the number density. That is, in this context, the Liouville operator D/Dt =
d/dt + 3ȧ/a. What about the right-hand side, the collision operator C? Obviously, it
depends on the details of the interactions that we’re considering.
In thermal equilibrium, we know that the number density should obey the equilibrium
distributions discussed in Section 5.2.
Consider a two-particle to two-particle interaction (such as annihilation x + x̄ → y + ȳ
or recombination e + p → H + γ). In equilibrium, the number of forward interactions is
the same as inverse interactions. The forward rate will be σ|v|n2 , where σ gives the cross
section and |v| the relative velocity of the particles. Via detailed balance, the inverse
rate must be such that the collision term is zero when n = neq , the number density

46 J. R. Pritchard
 5.4.: Relic Abundances

in equilibrium. Hence the inverse rate must be proportional to σ|v|n2eq . Thus we have
heuristically shown that we can write the whole equation as
dn ȧ
+ 3 n = −hσ|v|i n2 − n2eq


(5.22)
dt a
where hσ|v|i is the so-called thermally-averaged value of the cross section times the ve-
locity, and (so if n = neq it is equivalent to having no interactions). We can rewrite this
as  2 
ṅ ȧ n n
+3 = −hσ|v|ineq −1 (5.23)
neq a neq n2eq
or  2 
ṅ n ȧ n
= −Γ 2
−1 −3 (5.24)
neq neq a neq
where the interaction rate is Γ ≡ hσ|v|ineq . The time or temperature dependence of Γ
depends on the interaction under consideration, so the solution to this equation usually
needs to be calculated numerically. We show an example in Figure 5.3. In general, higher
Γ (higher hσ|v|i) results in later departure from equilibrium (freeze-out). As we noted
above, a particle that freezes out when kT  mc2 is called a hot relic, and one that
freezes out when kT  mc2 is a cold relic.
n
1

0.01 Increasing XΣ v \

10-4

10-6

10-8

10-10
m c2
10-12
0.1 1 10 100 kT

Figure 5.3: Freeze-out of particle species as a function of interaction strength, hσvi. The
exponentially-decaying curve is the equilibrium abundance; the lines that “peel off” from
equilibrium correspond to increasing values of hσvi for lower values of the final abundance
and the freeze-out temperature (so later freeze-out time).

As an example, we show the numerical solution to this equation for the case where
hσvi ∝ (kT /mc2 )n for mc2 >
∼ 3kT for various values of the cross-section. We see that

Cosmology 47
5 — Thermodynamics and Particle Physics

higher values of the cross section (higher Γ) correspond to later freeze-out times and lower
abundances. Examining the numerical solution also shows that defining the freeze-out
temperature from Γ(TF ) = H(TF ) and the final abundance from nF = neq (TF ) is a very
good approximation.

Hot relics: Neutrinos

An excellent case study is given by neutrinos. They were certainly relativistic over the
course of many of the events listed in Table 5.3.2, although over the last decade it has
become evident that neutrinos have a nonzero mass, likely to be considerably less than
1 eV. Exercise: what mass would a neutrino need in order to still be relativistic today,
when T = 2.73 K? Such a mass (assuming the cross sections of the standard model)
means that they freeze out when still relativistic and so are a hot relic.
Neutrinos have very important differences from other matter particles such as elec-
trons and positrons on the one hand, and radiation like photons on the other, however.
Like electrons and positrons, they are fermions and moreover have both particles and
antiparticles. Unlike electrons, however, they interact very weakly and hence came out of
equilibrium much earlier, before they could annihilate into other light particles. Hence,
the total neutrino plus antineutrino density is higher than that of electrons plus positrons.
(Also, if they have sufficiently low mass, they could still behave like radiation.) Because
of their very weak interactions with normal matter, however, when the electron-positron
pairs annihilated, they produced photons, but not neutrinos. Hence, there was consid-
erable extra energy dumped into the photon background — the CMB — but not the
neutrino background, and therefore the CMB actually has a higher temperature than the
neutrino background. As you will see in a problem set, the temperature of the neutrino
background relative to that of the photons can be calculated, and is
 1/3
4
Tν = TCMB . (5.25)
11

Their abundance will then be the value of the equilibrium number density, Eq. 5.10,
evaluated at the freeze-out temperature T = TF , and then converted to a comoving
density, which is then constant from then on. Note that the conversion to comoving
density requires multiplying by a3 /a30 = (1 + zf )−3 = T03 /TF3 , which cancels out the TF3
factor in n;3 the comoving density depends on g but nothing else. If these particles have
sufficient mass that they are non-relativistic today we can simply calculate their energy
density by multiplying the number density by the mass. This gives
 m 
ν
Ων h2 ' 10−2 . (5.26)
1 eV
Since we know that Ω <
∼ 1 today, this translates into a cosmological
P bound on the neutrino
mass, or more precisely, the sum of all the neutrino masses: mν <
∼ 100 eV.
3
There is an additional factor due to any further relativistic particles that have contributed to the
present-day photon abundance and thus T0 but not to TF , as in the discussion of e+ e− annihilation.

48 J. R. Pritchard
 5.4.: Relic Abundances

Although the freeze out temperature of neutrinos turns out not to be critically im-
portant for their present day abundance it is straightforward to estimate. Neutrinos
interact via the weak force, which typically gives an interaction cross section σ ∼ G2F T 2
(GF = αW /MW ' 10−5 GeV−2 is the Fermi energy, αW ' 1/30 is the weak fine struc-
ture constant and MW ' 100 GeV is the weak scale). This gives Γ ' nσv ∼ G2F (kT )5
(since relativistic particles have n ∝ T 3 ), so Γ/H ' G2F (kT )3 m−1 3
Pl ' (T /MeV) (mPl
is the Planck mass). This gives Γ/H = 1 at kTF ∼ MeV — neutrinos freeze out near
kT ∼ 1 MeV.4

Cold Relics: Weakly Interacting Dark Matter

Another example is given by so-called Weakly Interacting Massive Particles, or WIMPs,

which might be the dark matter. These are massive, stable particles that interact very
weakly with normal matter, and are in fact expected to exist in extensions of the standard
model such as supersymmetry. Their cross sections are likely to be similar to those of
neutrinos, but their masses would be much higher. Hence, they would behave like cold
relics.
Supersymmetric theories usually feature a stable, massive particle, the Lightest Su-
persymmetric Partner, which is often electrically neutral, a neutralino (typically a linear
superposition of the Higgsino and the “bino”). The calculation of the relic abundance for
a cold relic is typically more complicated than for a hot relic, since freeze out occurs in
the regime that the equilibrium abundance is falling exponentially where the details of
the interaction are important. For a specific SUSY model, we could solve the Boltzmann
equation numerically to predict the relic abundance. Alternatively, a simple estimate
turns out to be quite accurate.
Assuming that freeze out occurs when Γ = nhσvi ∼ H(TF ) we can write the freeze
1/2 1/2
out abundance as nF ∼ H(TF )/hσ|v|i ∼ g∗ TF2 /hσ|v|impl , where we use H ∼ g∗ T 2 /mpl
during radiation domination. To evaluate this, we need the freeze out temperature TF ,
which really requires solving the Boltzman equation. However, such calculations typically
find that the freeze out temperature is TF ∼ MX /10 where MX is the WIMP mass, and is
fairly insensitive to the details of the interaction. This key piece of information is enough
to estimate the relic abundance.
As for the neutrinos, we can map the number density of WIMPs at freeze out to
the present day nX (a0 ) = nX (TF )(T0 /TF )3 , but this time a dependence on TF remains.
For typical weak scale masses MX ∼ MW ∼ 100GeV the WIMP will freeze out when
g∗ ∼ 100 and the annihilation of various particle species after WIMP decoupling will
heat the photons relative to the WIMP so that TX ∼ (1/100)1/3 T0 . Taking all this into
account, we find that the energy density today ρX = mX nX can be written as

g∗ 1/2 10−27 cm3 s−1

 
2 mX
ΩX h ∼ . (5.27)
10TF 100 hσvi
4
Some of the equations in this paragraph are given in natural units with h̄ = c = 1, to be discussed
in more detail in future chapters.

Cosmology 49
5 — Thermodynamics and Particle Physics

This has the expected dependence that the energy density in WIMPs decreases for larger
interaction cross-section.
This result is extraordinary, although it might not seem it at first glance. We have
assumed that are WIMPs interact via the weak force and so will have a cross-section
σ ∼ G2F T 2 , evaluating this for TF and assuming that v ∼ c gives hσW vi ∼ 10−26 cm3 s−1 .
Which would predict ΩX h2 ∼ 0.1. This is almost exactly the observed abundance of dark
matter today! Remember that supersymmetry was invented to solve particle physics
problems, not cosmological ones, so this could be just a coincidence. Nonetheless, it is
intriguing that without fine tuning a WIMP could produce the required amount of dark
matter, so much so that this result is often described as the “WIMP miracle”.

We have not gone into too much detail about baryogenesis, neutrinos, and WIMPs, but
in the following lectures, we will discuss two important transitions in the early Universe
using these ideas. First, the creation of neutral hydrogen from the ionized plasma of
electrons and protons at about 400,000 years after the big bang, and then the synthesis
of the light elements from free neutrons and protons at about three minutes.

50 J. R. Pritchard
Chapter 6

Hydrogen Recombination and the

Cosmic Microwave Background

MRR 5.2

6.1 Introduction and Executive Summary

In this chapter we will discuss how the Universe went from being an ionized plasma to a
neutral gas of (mostly) hydrogen. The basic story is simple: the early Universe was hot
enough, kT  1 Ry = 13.6 eV, to keep the Universe ionized, and therefore for the thermal
bath of photons to remain tightly coupled to the ions because of Thomson scattering.
When the Universe cooled sufficiently, electrons and photons could (re-)combine to form
neutral hydrogen, which also freed the photons from their interactions with matter. We
see these photons as the Cosmic Microwave Background, essentially as the surface of an
opaque “cloud” at redshift z ∼ 1100.
As this story involves all well-understood processes in atomic physics, we can actually
calculate the ionization history of the Universe in some detail. The most important
interactions will be hydrogen recombination,
p+ + e− → H + γ (6.1)
and Thomson scattering
e− + γ → e− + γ (6.2)
where the appropriate cross section is
8π αh̄
σT = ' 6.7 × 10−25 cm2 (6.3)
3 me c
and the binding energy is
(me + mp − mH )c2 = 13.6 eV = 1 Ry = B. (6.4)
The interaction rate is just Γ = ne cσT ∝ 1/a3 ∝ (1 + z)3 ∝ T 3 . Compare this to the
expansion rate, which is either H ∝ T 2 (RD) or H ∝ T 3/2 (MD), so that we know that

51
6 — Hydrogen Recombination and the Cosmic Microwave Background

at some time in the past, we must have gone from Γ > H to H > Γ more recently, so
this reaction would have frozen out.
Another consequence of these interactions freezing out is that the photons, once tightly
coupled to the ionized plasma (which is opaque due to Thomson scattering, e− γ → e− γ),
are able to stream freely through the now-neutral, and hence transparent, hydrogen gas.
Thus we see these photons as if freed from a cloud; this is the last scattering surface,
and it forms the cosmic microwave background, which looks to us (using microwave
telescopes) as the surface of a cloud: further away (higher redhsifts) is opaque, nearer is
transparent.

Opaque

Transparent

Figure 6.1: The surface of last scattering. Redshift increases outward. At early times,
the photons are tightly coupled to the charged plasma. At later times they stream freely
through the neutral gas.

52 J. R. Pritchard
 6.2.: Equilibrium ionization

6.2 Equilibrium ionization

First. we calculate the ionization state of the hydrogen under the assumption of equilib-
rium. In addition to the above numbers describing the interaction, we will need to explic-
itly assume charge neutrality (np = ne ) and baryon number conservation (nB = np +nH )1
In thermal equilibrium, we will assume a Maxwell-Boltzmann (non-relativistic) dis-
tribution for protons, electrons and Hydrogen atoms,
3/2
(µi − mi c2 )
  
mi kTi
ni = gi exp (6.5)
2πh̄2 kTi

and a massless Bose-Einstein distribution for the photons. The total number density of
photons is just the integral of the distribution as in Eq. 5.4:
 3
kT
nγ = 16πζ(3) , (6.6)
hc

which also gives the overall baryon density, nB = ηnγ , where η = 2.7 × 10−8 Ωb h2 is the
baryon-to-photon ratio. Statistical chemical equilibrium of Eq. 6.1 requires µp + µe− =
µH + µγ = µH , and the degrees-of-freedom factors are gγ = gp = ge = 2, gH = 4.
Combining the expressions for the equilibrium abundances of the protons, electrons,
and hydrogen atoms gives us the formula
 −3/2
gH me mp kT
nH = ne np eB/(kT ) (6.7)
ge gp 2πmH h̄2
If we define the ionization fraction,
np np
Xe = = , (6.8)
nB np + nH

we get the Saha equation

√ 3/2
1 − Xe

4 2ζ(3) kT
= √ η eB/(kT ) (6.9)
Xe2 π me c2

which looks complicated but is really just a quadratic equation for the equilibrium ioniza-
tion fraction as a function of temperature T , or equivalently of redshift by T = (1 + z)T0 .
Exercise: derive Eqs. 6.7-6.8 from the equations given in the previous paragraph. We
show Xe (z) in Figure 6.2.
In the figure, we see that ionization fraction goes from one (fully ionized) to zero
(neutral) around redshift 1300-1400. This corresponds to a temperature kT ∼ 0.3eV
Compare this to the redshift at which kT = 13.6 eV, calculated in one of the problems:
z(13.6eV) ∼ 60, 000. Why the large discrepancy? It is the very small value of η: there are
1
We know from our discussion in Section 5.4.1 that Baryon number is not always conserved, but we
assume that this happens much earlier than this epoch.

Cosmology 53
6 — Hydrogen Recombination and the Cosmic Microwave Background

X
1.0

0.8
Wb h2 ‡ 0.005

0.6
Wb h2 ‡ 0.02

0.4

Wb h2 ‡ 0.1
0.2

z
1200 1400 1600 1800 2000

Figure 6.2: The equilibrium ionization fraction, X as a function of redshift, for different
values of the Baryon density, Ωb h2 .

so many photons for every baryon that even the very small number in the exponentially-
suppressed tail of the BE distribution of photons is enough to keep the hydrogen ionized
until much later. Indeed, we could have derived this just by finding the temperature at
which there is approximately a single ionizing photon with hν > 13.6 eV in the photon’s
Bose-Einstein distribution. Exercise: show this.

6.3 Freeze-Out
In the previous section we calculated the equilibrium ionization fraction. However, be-
cause the interaction rate (i.e., the recombination rate) depends upon the density of free
electrons and protons, as more and more neutral hydrogen is formed, the rate of photon
scattering off of electrons goes down and down, and will eventually freeze out.
Consider Thomson scattering,

e− + γ → e− + γ , (6.10)

which has a rate

Γ ' ne σT c . (6.11)
The number density of electrons depends upon the ionization fraction X and the baryon
density:  3
3ζ(3) kT0
ne = XnB = Xηnγ = Xη (1 + z)3 (6.12)
2π 2 h̄c
This is to be compared to H = H0 E(z). In Figure 6.3 I show (separately), the expansion
timescale, 1/H, and the mean time between interactions, 1/Γ. When they cross, Γ ∼ H

54 J. R. Pritchard
 6.3.: Freeze-Out

and for lower temperatures the interaction is frozen out. Note that this freeze-out occurs
again at zF ∼ 1100 when the equilibrium ionization fraction is X  1 – despite the very
large number of photons per baryon, it takes a very small number of electrons to make
the mean free path much smaller than the Hubble scale, and hence to make the Universe
opaque. However, once this happens, the recombination interaction ep → Hγ also freezes
out, leaving a final ionization fraction X = Xeq (zF ) ∼ 10−4 , which, although small, is
much larger than the exponentially-reduced equilibrium ionization fraction that would
have otherwise been predicted from Section 6.2.
t

sec
1015

Λ, Wb h2 Š0.02
1014

Λ, Wb h2 Š0.1
13
10

H -1 , Wm =0.3
1012

H -1 , Wm =1
11
10

z
1100 1200 1300 1400 1500 1600 1700

Figure 6.3: The expansion timescale (H −1 = a/ȧ) and the mean time between interac-
tions, (λ = 1/Γ) for different values of Ωb h2 and Ωm . For redshifts less than the crossing
point of the curves, the photons and baryons/electrons are no longer interacting.

In fact if we calculate this more carefully using the Boltzmann equation formalism
of the previous lectures, we get the same result (as would be expected): freeze-out of
Thomson scattering at kT ∼ 0.25 eV.
It is important to realize that there are three somewhat distinct events:

• Recombination: The equilibrium ionization fraction, X goes from nearly one to

nearly zero.

• Last Scattering: The freeze-out of e− γ Thomson scattering; this is also known as

the decoupling of photons from the baryons.

• The freeze-in of residual ionization at a higher value than equilibrium, i.e., the
freeze-out of the recombination reaction, Eq. 6.1. (This calculation requires more
of the full Boltzmann equation apparatus to do accurately and we will not do it in
detail here.)

Cosmology 55
6 — Hydrogen Recombination and the Cosmic Microwave Background

56 J. R. Pritchard
Chapter 7

Big-Bang Nucleosynthesis

MRR 5.3
This material is covered in several textbooks, including Rowan-Robinson and Liddle,
but there is an especially classic discussion in S. Weinberg, “The First Three Minutes”.
In this chapter, we will discuss the synthesis of the light nuclei from the initial primor-
dial mixture of protons and neutrons (along with the ubiquitous photons) — Big Bang
Nucleosynthesis (BBN). In broad strokes, this is similar to our discussion of Hydrogen
recombination and the formation of the CMB, but with quite a few extra complications.
At early times, protons, neutrons, and light nuclei such as deuterium, tritium and Helium,
are in equilibrium (specifically called “nuclear statistical equilibrium” in this case, just
as we discussed “ionization equilibrium” in the context of the CMB). As the Universe
expands and cools, different nuclear reactions freeze out, leaving us with relic abundances
of the stable nuclei.
BBN has a few extra complications, however. First, we have to track the abundance
of not just one end product (neutral hydrogen) but of several different nuclei. Second,
one of the starting constituents, neutrons, are unstable when not in a nucleus, with a
half-life of about 11 minutes. Finally, several of the possible end-states (light nuclei) have
binding energies that are small or comparable to kT and so freeze-out can be delayed.
With our calculation of hydrogen recombination, we found in Eq. 6.7 that the equi-
librium Hydrogen density is proportional to exp(B/kT ) where B = 13.6eV is the binding
energy of the hydrogen atom. Similarly, the most strongly-bound light nucleus is 4 He,
with binding energy B4 = 28.5 MeV, so we essentially expect most of the nucleons to end
up in Helium in equilibrium, with other species suppressed exponentially in comparison.
Moreover, the lack of any stable nuclei at all of mass 5 or 8 makes it very difficult to get
to higher-mass nuclei beyond Helium. However, we will see that freeze-out of the inter-
actions will mean that not all the species will remain in equilibrium so a more careful
calculation is required.
Indeed, a full calculation of BBN requires the solution of many coupled differential
equations, but the main results can be calculated from simple principles such as those we
have applied to the CMB.

57
7 — Big-Bang Nucleosynthesis

7.1 Initial Conditions for BBN: neutron-proton freeze-

out
We start at temperatures kT  1 MeV, times t  1 s (compare the neutron-proton mass
difference, ∆m = mn − mp = 1.29 MeV/c2 ). At this time, weak interactions maintain
“nuclear statistical equlibrium” (NSE). The crucial (weak) interactions are
n ↔ p + e− + ν̄e νe + n ↔ p + e− e+ + n ↔ p + ν̄e . (7.1)
We are late enough that we can assume non-relativistic (Maxwell-Boltzmann) statistics,
so in equilibrium
 3/2
mi kT 2
ni = gi 2 e(µi −mi )c /kT (7.2)
2πh̄
so 3/2
(µn − µp ) − (mn − mp )c2
  
nn mn
= exp . (7.3)
np mp kT
We can relate the chemical potentials to our interactions as before, e.g., µν +µn = µp +µe− ,
but in fact we have µ  ∆m so we can ignore that term in the exponential. If we further
assume mn /mp ' 1 in the prefactor,
∆mc2
 
nn
= exp − . (7.4)
np kT
But we had better check that we are in equilibrium. The rate for these interactions is
related to the weak interaction scale, Γ = nσv ∼ G2F T 2 T 3 , which can be calculated in
more detail giveing, e.g.,
 5
− kT
Γ(νe n ↔ pe ) ' 2.1 s−1 (7.5)
MeV
In an RD universe, this must be compared to
 2
ȧ 1 kT
H= = ' s−1 (7.6)
a 2t MeV
Putting these together, near kT ∼ 1MeV,
 3
Γ kT
' (7.7)
H 0.8 MeV
Hence, we freeze-out at TF ' 0.8 MeV. At this point, we have a neutron-proton ratio
∆mc2
 
nn
= exp − ' exp[−1.29/0.8] ' 0.2 (7.8)
np kTF
(a more careful calculation gives 0.17-0.18, so this is good to about 10-15%).
It is worth pointing out that the freeze out of these interactions also means that the
neutrinos will have frozen out more generally — they decouple from their interactions
with matter and photons, and stream freeely thereafter, just as the photons themselves
later decouple from the baryons and leptons, as with the formation of the CMB discussed
in Chapter 6.

58 J. R. Pritchard
 7.2.: Helium production

7.2 Helium production

After the freeze-out of the reactions interconverting neutrons and protons, the number
density of each is fixed (except for neutron decay which we will account for later). Now,
we consider various chains of reactions, both starting with
p+n→d+γ ; (7.9)
(d is deuterium, the hydrogen isotope made of a neutron and a proton, which has a
binding energy B2 ' 2 MeV). From here, there are two possible chains:
d + d → 3 He + n 3
He + d → 4 He + p (7.10)
and
d+d→t+p t + d → 4 He + n (7.11)
(t is tritium, the hydrogen isotope made of two neutrons and a proton). Either way, we
first have to make deuterium, but there are two factors that make this difficult. First, as
in the case of Hydrogen recombination, the equilibrium deuterium fraction is controlled
by an equation similar to the Saha Equation, Eq. 6.8, with factors of (kT /mp c2 )3/2 η  1.
Just as in the case of recombination, the large baryon-to-photon ratio keeps the interaction
pn → dγ from proceeding forward (i.e., the photons dissociate the deuterium) well below
the temperature kT = B2 ' 2 MeV binding energy. Instead, the deuterium fraction
remains small until kT ' 0.1 M eV . Even at this point, the interactions cannot proceed
fully, due to the interaction rates:
5
Γ(pn → dγ) ΩB h2
  
kT np
' (7.12)
H 0.05 MeV np + nn 0.02
This has Γ/H > 1 for kT > 0.05 MeV ' kTF = kTBBN , equivalent to a time tBBN =
(kTBBN /MeV)−2 /2 s ' 200 s, which we will take to be about 3 minutes. Therefore
only a very small number of deuterium nuclei are formed from the baryons, and hence
further there is a very low rate for either of the dd interactions (since both have rates
Γ(dd) ∝ n2d ). We essentially never make it to equilibrium abundances of helium. This is
called the deuterium bottleneck, lasting until kT ∼ 0.05 MeV.
Once we reach 0.05 MeV, we can form deuterium and both of these pathways can
proceed; there is rapid burning of almost all of the remaining protons and neutrons into
Helium. To calculate the final state, we first define the mass fraction of a helium nucleus:
AHe nHe
Y = (7.13)
nB
where A = 4 is the atomic mass of Helium, and so the numerator is proportional to
the mass density of Helium, and the denominator to the total mass density of baryons.
We assume that essentially of the neutrons go into He nuclei, so the number density of
Helium will be half of the number density of neutrons. Hence
4nHe 4nn /2 2nn /np
Y = = = (7.14)
nn + np nn + np 1 + nn /np

Cosmology 59
7 — Big-Bang Nucleosynthesis

where now we have related the helium mass fraction to the neutron-proton ratio calculated
above. At np freeze-out TF ' 1 MeV we had nn /np ' 0.17 ' 1/6. By the time the
deuterium bottleneck is broken, however, it is now tBBN ∼ 3 min, which means that a
non-negligible fraction of the neutrons will have decayed into protons, n → p+ e− ν̄e , with
a half-life of 10.5 minutes. Hence

nn (TBBN ) ' nn (TF )2−3/10.5 . (7.15)

This decreases the neutron-to-proton ratio to

nn
≈ 0.13 (7.16)
np TBBN

and thus
2 × 0.13
Y ' ' 0.24 (7.17)
1 + 0.13
We can also define the mass fraction of other species Xi = Ai ni /nB which can be found
by considering the freeze-out of the reactions we have considered so far, as well as further
reactions that can build higher-mass nuclei, giving final abundances slightly higher than
the equilibrium abundances. Lithium is produced by the chain
4
He + t → 7 Li + γ 3
He + 4 He → 7 Be + γ 7
Be + e− → 7 Li + νe . (7.18)

7.2.1 Dependence upon the cosmological parameters

The details of the production of helium and the other light elements depend upon the state
of the Universe over the epochs in time and temperature that we have been discussing.
This is most obvious for the dependence upon the baryon density, ΩB h2 , which comes
in directly in the equilibrium distribution of deuterium (as well as in the rate of the
deuterium fusion reaction Γ[pn → dγ], Eq. 7.12). Just as a higher baryon density results
in a higher density of hydrogen atoms (via pe → Hγ), a higher baryon density also
implies a higher density of deuterium (via pn → dγ). This, in turn, implies a higher
density of Helium, with an even stronger dependence upon the baryon density. We show
the dependence upon the baryon density for several light elements in Figure 7.1
It is more difficult to synthesize elements beyond Helium. First, there are no stable
elements with A = 5 or A = 8, so we cannot do so by adding just a single nucleon
to Helium. Second, the way to create any such elements would therefore be to fuse
several light nuclei together, but each of those would have positive charge, and hence the
reaction rates are strongly suppressed by the Coulomb Barrier. By the time deuterium
is produced in sufficient abundances to form Helium, the temperature has decreased low
enough that few nuclei have enough energy to breach the barrier.
Nonetheless, some 7 Li is formed via the reactions of Eq. 7.18. For baryon density
−10
η<∼ 3 × 10 , this occurs via the tritium path:
3
t + 4 He → 7 Li + γ ; (7.19)

60 J. R. Pritchard
 7.3.: Observations of primordial abundances

however, lithium is very easy to destroy via collisions with protons, so increasing η actually
means decreasing final lithium abundance. Whereas for higher values of η, there is more
3
He around and we can first produce beryllium via the Helium-3 path,
3
He + 4 He → 7 Be + γ , (7.20)

which then inverse β-decays via electron capture to 7 Li. Because beryllium is actually
more strongly bound than lithium, once this pathway is opened the final lithium abun-
dance increases with η.
Furthermore, as we increase η, more and more d and 3 He is burnt into helium the
reactions Eqs. 7.10-7.11, which freeze out later and later, since the reaction rates are
proportional to η.
Although a full calculation requires the simultaneous solution of coupled differential
equations, it is usually straightforward to perform a perturbation analysis to see the
effect of making a small change in the other cosmological parameters. To give a flavor
of this analysis consider the effect of changing the time (or equivalently, temperature,
TBBN ) that BBN occurs. If the expansion was faster, freeze-out, breaking the deuterium
bottleneck, would have occurred earlier. At this earlier time, neutrons would have had
less time to decay, and the neutron-proton ratio would have been higher, and hence the
Helium mass fraction, Y would have been higher. But why would the expansion rate be
higher? One way is to increase the density of radiation in the universe, since H 2 ∝ ρ, so
any (relativistic) species beyond those we know of in the standard model would increase
the Helium abundance — indeed for quite a while observations of the Helium abundance
(and that of other elements) were the strongest constraint on additional light particles,
in particular neutrinos.

7.3 Observations of primordial abundances

See classroom presentation.

Cosmology 61
7 — Big-Bang Nucleosynthesis

20. Big-Bang nucleosynthesis 3

356789%>#9?=;7%Ω@!(
/0//) /0/. /0/( /0/!
/0(+
'"#
/0(*

/0()
D&
A /0('
BBB
"
/0(!

./ −!

A$"% &

123
334
./ − '

!"#$"%
&
./ − )

./ − -

)
+C=$"%
&
(

./ − ./

. ( ! ' ) * + , - ./
356789:;8:&<8;89%65;=8%η × ././
Figure 20.1: The abundances of 4 He, D, 3 He, and 7 Li as predicted by the standard
model
Figure 7.1:of Big-Bang
The mass nucleosynthesis [11] − the
fraction of various bands show
species as a the 95% CLofrange.
function Boxes density.
the baryon
indicate the observed light element abundances (smaller boxes: ±2σ statistical
Bands show the 95% confidence range. Boxes indicate observed light element abundances
errors; larger boxes: ±2σ statistical and systematic errors). The narrow vertical
(smaller indicates±2σ
band boxes: the statistical
CMB measure errors;
of thelarger
cosmic baryon±2σ
boxes: statistical
density, and
while the systematic er-
wider
rors).band
Theindicates
narrowthe
vertical band indicates
BBN concordance rangethe CMB
(both at measure
95% CL). of the cosmic baryon density,
while the wider band indicates the concordance range of direct measurements of the light
element abundances. From Fields and Sarkar, in Amsler et al., PL B667, 1 (2008, 2009)
http://pdg.lbl.gov

February 2, 2010 10:47

62 J. R. Pritchard
Chapter 8

Interlude

8.1 Natural Units

The major theories of modern physics developed since the time of Newton come armed
with a set of physical constants: formulas from relativity (and electromagnetism) always
use c, the speed of light. In thermodynamics, temperature always appears in the energy
combination kT where k is Boltzmann’s constant. Quantum mechanics, of course, uses
Planck’s constant, h̄. Finally, gravitation (GR or Newtonian) uses Newton’s constant, G.
It further turns out that it is impossible to express the units of any of these in terms of
any other. Exercise: Show this.
Therefore, it makes sense when doing calculations in relativity and electrodynamics to
measure all speeds with respect to c (indeed we often do this by considering the quantity
β ≡ v/c, but now we are just using the shorthand of writing v instead of β). Moreover,
we can also decide to measure masses in energy units: instead of saying that the electron
has mass me = 0.511 MeV/c2 , we can just use the symbol me for the quantity that we
used to call me c2 .
With these conventions, many equations become simpler and their meaning more
manifest. For example, the relationship between relativistic energy, mass and momentum
is simply:
E 2 = m 2 + p2 . (8.1)
Without the speed-of-light factors, the equation is much more symmetric amongst the
quantities. This is even more manifest in the version of this equation applicable to
particles at rest, arguably Einstein’s most famous equation: E = m. We now see this as
the statement that energy and mass are just different words for exactly the same thing.
Similarly, in a cosmological context we have had to worry about both energy density ε
and matter density ρ but in our c = 1 units we just see ε = ρ.
Similarly, in quantum mechanics calculation, we similarly measure all quantities with
units of angular momentum or action in terms of h̄ = h/(2π), and we can decide to use
the symbol ω for the quantity h̄ω which has energy units. In this case, Heisenberg’s
uncertainty relation takes the simple form
1
∆x ∆p ≥ (8.2)
2

63
8 — Interlude

and Schrodinger’s equation is

∂ψ
i
= Hψ (8.3)
∂t
with the recipe p → i∂/∂x in the Hamiltonian.
Finally, we can use T for the quantity kT in thermodynamics calculations. But we
can do this more generally: in shorthand, we say we are taking h̄ = c = k = 1 and are
using natural units. In normal units

[c] = LT −1 , [h̄] = M L2 T −1 , and [k] = M L2 T −2 K −1 (8.4)

where M is mass, T is time, L is length and K is temperature (since T is already taken).

In cosmology and particle physics, it is traditional to keep Newton’s constant explicit in
calculations (although in relativity and fields like string theory, G = 1, or occasionally
8πG = 1, is often used).
In these examples, it is clear that quantities with units of energy have a special place.
Indeed, we can’t make an energy out of any combination of h̄ and c (k is obvious since
we know kT has energy units and there is no way to get a temperature out of h̄ and c):

M L2 T −2 = [c]a [h̄]b
= La T −a M b L2b T −b
= M b La+2b T −a−b (8.5)

which is equivalent to

b=1 a + 2b = 2 − 2 = −a − b (8.6)

which are three equations in two unknowns and do not have a solution (a similar argument
holds for lengths or times). So, as long as we keep units on G, we still need to choose
some units in which to report physical quantities, and it is traditional in cosmology and
particle physics to use energy units, in particular electron-volts (or keV, MeV, GeV as
appropriate). Note that any physical quantity can be expressed as an energy to some
power this way. For example, we know that we can relate a length λ to a frequency c/λ
and hence to an energy h̄c/λ — so length can be thought of as having units [energy−1 ].
How do we do the conversion explicitly? We need to find a combination of a length, `
with h̄ and c that gives the appropriate units:

[Energy] = [`α h̄β cγ ]

M L2 T −2 = Lα M β L2β T −β Lγ T −γ
= M β Lα+2β+γ T −β−γ (8.7)

which is three equations in three unknowns, giving β = γ = −α = 1 so the combination

h̄c/` has energy units as expected. Since we can also convert from length to time using
factors of c, we can also express times in energy units.
As a specific example, consider the Hubble constant,

H0 = 100h kms−1 Mpc−1 = 3.24h × 10−18 s−1 = 2.1h × 10−33 eV (8.8)

64 J. R. Pritchard
 8.1.: Natural Units

or the current CMB temperature

T0 = kT0 = 2.4 × 10−4 eV . (8.9)
Since we are keeping G, what units does it have? We could work it out by brute force,
but an easier way is to first recall that the gravitational potential is given by the combi-
nation GM 2 /L and from the previous discussion we know that a length L has units of
[Energy]−1 = [E]−1 . So
[GM 2 ] = [G] × [E 2 ] = 1 (8.10)
or
[G] = [E]−2 = [M ]−2 = [L]2 = [T ]2 (8.11)
The associated energy, mass, length, time etc. are known as the Planck energy, Planck
mass, Planck length, Planck time, etc.:
EPl = G−1/2 = MPl (8.12)
which in normal units is
r
h̄c5
EPl = = 1.22 × 1019 GeV . (8.13)
G
Similarly the Planck length and time are
lPl = tPl = G1/2 (8.14)
which in normal units is r
h̄G
tPl = = 5.4 × 10−44 s (8.15)
c5
(and we just need another factor of c to convert to the Planck length). If we do set G = 1
then all quantities can be expressed as pure numbers, measured with respect to these
quantities: these are called Planck units.
In natural units, we can also do a lot of “back-of-the-envelope” physics. Consider, for
example, the energy density of a blackbody. Except for the temperature, there are no
physical constants with units that could appear in such a problem beyond the fundamental
physical constants we have discussed. There are no particle masses or fundamental lengths
that might matter. Moreover, we know that gravitation doesn’t enter into the problem,
so the Planck length/mass/time/energy should not be relevant. All we have to work with
is the temperature. So how do we make an energy density? We need to get units of
energy per volume. Volume is [L]3 and we know that [L] = [E]−1 , so we know [ε] = [E]4 .
Since all we have to work with is the temperature, this must be of the form
ε = O(1) × T 4 (8.16)
where O(1) refers to a constant, expected to be of order one.1 Indeed, the full expression,
as we have seen, is
π2
ε= (kT )4 (8.17)
15(h̄c)3
1
Of course, there are rare occasions when these considerations may go awry — sometimes the expected
O(1) constant may for some reason be much larger or smaller than one.

Cosmology 65
8 — Interlude

so the constant is π 2 /15 ' 0.66.

Now consider the Friedmann Equation, first in a flat, Λ = 0 universe:
 2
ȧ 8πG
= ρ
a 3
8π T 4
H2 ∼ (8.18)
3 m2Pl

where in the second line we have assumed radiation domination, so ρ = ε ∼ T 4 and used
G−1/2 = MPl . This gives the simple formula

H ∼ T 2 /mPl in RD. (8.19)

8.1.1 The Cosmological Constant problem

Now, let’s consider a more general universe, the one that seems to obtain, with a cosmo-
logical constant. First, let’s try to work out what units the cosmological constant should
have. From the Λ term in the Friedmann equation

[H]2 = [Λ/3] (8.20)

or
[T ]−2 = [Λ] (8.21)
so
[Λ] = T −2 , (8.22)
Which means that in our natural units,

[Λ] = [T ]−2 = [L]−2 = [M ]2 = G−1 , (8.23)

Now, we have measured the cosmological constant to be

Λ
ΩΛ = ' 0.7 (8.24)
3H02
so
−2
Λ ' 0.7 × 3H02 ' (9.4Gyr)−2 ' 1026 m . (8.25)
Another way to think about the value of the cosmological constant is to consider its
energy density:

2 3H02
ρΛ c = ΩΛ ρcrit = ΩΛ ×
8πG
∼ 6 × 10−10 J m−3 ∼ 10−47 GeV4 ' (10−12 GeV)4 (8.26)

where in the last we use natural [energy] units. Do we have any fundamental theories
that could predict the value of the cosmological constant? When thought of as an energy

66 J. R. Pritchard
 8.1.: Natural Units

density, we have seen that the cosmological constant has the unique property that it
does not change with time, despite the expansion of the Universe — it is the energy of
the vacuum. We also know that our best description of particle physics, quantum field
theory, gives a generic prediction for the value of the vacuum energy density: it should be
ρvac ∼ E 4 (in natural units), where E is some energy scale that describes the high-energy
(“ultraviolet” in particle physics parlance) cutoff of our theory. That is, E is the scale at
which our effective quantum field theory breaks down. Unfortunately, the observed value
of EΛ ∼ 10−12 GeV = 0.001 eV is much lower than any such expected cutoffs — it is well
below the energy scale of everyday physics. Indeed, if Λ is from a true quantum-gravity
theory, we would expect E = EPl , in which case we would have
2
1017

1
ΩΛ ∼ 2
∼ (t2Pl H 2 )−1 ∼ ∼ 10122 , (8.27)
GH 10−44

which is much, much, much greater than the observed value of 0.7. Even if the cosmo-
logical constant were due not to quantum gravity but to supersymmetry, we would still
expect ΩΛ ∼ 1060 , not much better.
There are various solutions, all somewhat unsatisfactory, proposed to solve the cos-
mological constant problem. The first is that the “real” cosmological constant is Λ = 0,
but that there is some other physical mechanism — “dark energy” — that can provide
an energy density with equation-of-state w = p/ρ = −1, not a true vacuum energy. In
fact, a scalar field can provide this. Indeed, we will see that inflation is thought to be
driven by such a scalar field, although the energy scales of inflation and dark energy are
so different that no one has been able to come up with a single mechanism for producing
both epochs of w = −1.
Another possibility is that the properties of the vacuum depend upon the details of
the fundamental theory. In string theory, for example, there are a huge number (possibly
10200 or greater!) ways to compactify down to 3+1 dimensions, and the cosmological
constant could be different in each of them. Then, we may need to use something like
the anthropic argument to find the cosmological constant. The basic idea is due to
Weinberg (“Anthropic Bound on the Cosmological Constant”, Phys. Rev. Lett. 59
(22): 2607–2610. doi:10.1103/PhysRevLett.59.2607): generically, the theory predicts a
distribution of Λ, but most of the distribution is with considerably greater values than
observed. However, with much greater values, the Universe would look very different
than it does today. In particular, it seems that it would be very difficult to form any
structures like galaxies (and likely the stars within them) at all, as the Universe would
start exhibiting accelerated expansion before the structures could form. (We shall see
this in more detail when we discuss large-scale structure). Hence, we wouldn’t be here to
observe the cosmological constant if it were much larger. Thus, the prediction is that we
should observe the largest possible value of Λ consistent with structure formation. This
seems to be, roughly, true. (It should be pointed out that many cosmologists are deeply
troubled by anthropic arguments!)

For the rest of these notes, we will usually use h̄ = c = k = 1 natural units.

Cosmology 67
8 — Interlude

8.2 Open Questions in the Big Bang Model

The Big Bang model — i.e., the FRW metric — is amazingly successful: it predicts an
expanding Universe, a background of relic photons (the CMB), and detailed relations
amongst the light element abundances. But the initial conditions are left unspecified:
why do the densities of the various components have their observed values? Why is the
Universe approximately homogeneous and isotropic?
We can make these questions more precise, and see that the initial conditions do not
appear to be very generic at all. Rather, the Universe appears to be very finely tuned.

8.2.1 The Flatness Problem

Consider our definition of the contribution of curvature to the energy density as a function
of redshift:
−k (1 + z)2
Ωk (z) ≡ 1 − Ωtot (z) = 2 2 = Ωk (z = 0) 2 (8.28)
aH E (z)
with
E 2 (z) = Ωm (1 + z)3 + Ωr (1 + z)4 + ΩΛ + Ωk (1 + z)2 (8.29)
(since there is no actually density associated with curvature, it is often better to just
think of this as related to the total density as in the first equality). If the universe is
truly flat, then Ωk (z) = 0 for all time. But otherwise, in the early matter-dominated era,

(1 + z)2 1
Ωk (z) = 3
= (8.30)
Ωm (1 + z) Ωm (1 + z)
whereas in the radiation-dominated era
(1 + z)2 1
Ωk (z) = 4
= . (8.31)
Ωr (1 + z) Ωr (1 + z)2

In both cases, |Ωk (z)| is an increasing function of time (i.e., a decreasing function of
redshift). No matter how close the Universe is to flat today, it was even closer in the
past. We know that |Ωk | < ∼ 0.1 today. What does this imply for the curvature at some
early epochs?
−4
• At hydrogen recombination, z ∼ 1000, we have Ωk <
∼ 10 ;
• At matter-radiation equality (z ∼ 104 ), we have Ωk < −5
∼ 10 ;
• at nucleosynthesis, (z ∼ 108 ), we have Ωk < −13
∼ 10 ;
and at earlier and earlier epochs (electroweak symmetry breaking, the Planck epoch), the
requirement gets stronger and stronger. So if the Universe is not flat today, it had to
start out remarkably close to — but not quite — flat. This is not a very generic condition
at all.

68 J. R. Pritchard
 8.2.: Open Questions in the Big Bang Model

8.2.2 The Horizon Problem

When we talked about our picture of the Universe today, we discussed our observations
of the Cosmic Microwave Background. At a level of even better than one part in 1000,
the CMB is smooth (although we will see that the tiny fluctuations at level of 10−5 are
another crucial clue to the origin and evolution of the Universe). What does this mean?
Recall the angular diameter distance
Z z0 !
0
p dz
dA (z) = cH0−1 (1 + z)−1 |Ωk |−1/2 sink |Ωk | 0
. (8.32)
0 E(z )

which is defined by the relation between angular size, θ and physical size L: θ = L/dA .
We expect that the largest distance that physics ought to be able to act is the horizon
size,
cH0−1 ∞ dz 0
Z
dH (z) = . (8.33)
1 + z z E(z 0 )
so we can combine these to find the angular size of the horizon, θH = dH /dA . If we
assume a flat Universe, we have
R∞ 0
dH z
dz /E(z 0 )
θH = = z 0
R . (8.34)
dA 0
dz /E(z 0 )

If we further assume a matter-dominated Universe, the integrals are easy, since E(z) =
1/2
Ωm (1 + z)3/2 :

(1 + z)−1/2
θH = −1/2
∼ (1 + z)−1/2  1 for z  1. (8.35)
1 − (1 + z)
In particular, θH ' 1.7◦ at z ' 1000, and this is about right even if we take the details
of the density of matter, radiation and cosmological constant into account.
This means that any two patches of the CMB sky more than a couple of degrees apart
should not have been in causal contact — so how did they get to be the same temperature
to at least one-tenth of one percent? In fact, we don’t need to use the CMB to see the
problem: even at redshifts of a few, different regions of the sky are in different horizon
volumes, and yet we observe the matter density (as seen in the galaxy density) to be
roughly similar.
Another way to see this problem is to consider the scale of a structure in the Universe,
say, that of a galaxy. The physical scale of a galaxy, λgal , grows along with the scale factor,
λgal ∝ a ∝ t2/3 (MD) or λgal ∝ t1/2 (RD). But (in a Universe dominated by matter,
radiation, or curvature) the horizon scale grows as dH ∼ t ∼ H −1 . Today, λgal < dH and
we say that the scale is “inside the horizon” but because they grow at different rates,
at some point in the past we must have had λgal > dH , “outside the horizon”; the time
at which the dH = λgal is called horizon-crossing for that scale (see Figure 8.1). We
expect that structures can only grow when they are inside the horizon (due to causality)
but we must further be able to set up initial conditions so that this can happen. And

Cosmology 69
8 — Interlude

for astrophysically-interesting scales, this happens at a time for which we thing we have
a good understanding of the physics (e.g., the galaxy-size scale of λ ∼ 10 Mpc enters at
approximately the epoch of matter-radiation equality).

dH∝t
log length

λgal

tHor(λ)
log time

Figure 8.1: Physical scales inside and outside the horizon. Prior to (i.e., to the left of)
horizon-crossing at tHor (λ), the scale is outside the horizon; afterwards it is inside the
horizon and causal physics can act.

Note that we could have equally made this argument in terms of comoving scales,
lgal = λgal /a = const, which now must be contrasted with the comoving horizon distance
χH = dH /a. In a universe with matter, radiation and curvature, χH ∼ (aH)−1 and the
comoving Hubble length always grows with time: the horizon grows to encompass larger
and larger scales, so any scale now inside the horizon was once outside.

8.2.3 The relic particle problem

The final puzzle has to do with leftover relics from the early Universe. We have already
seen that the CMB itself is such a relic, as are the light elements such as helium. These
come from epochs of transition from one state to another as various interactions freeze
out. Are there any other relics we might expect?
It has been hypothesized that there is a grand-unified theory (GUT) that combines
the strong and electroweak forces — in particle physics parlance, this means finding the
appropriate single group in which to embed the standard model SU(3) × SU(2) × U(1).
If so, this theory will almost inevitably have a very massive electromagnetic monopole.
Because of the way it interacts, we would expect it to be formed with a number density
of about one monopole per horizon volume at some high temperature TGUT . (The mecha-
nism is akin to the formation of defects in phase transitions in solids, extended to particle
physics by Higgs, and cosmology by Tom Kibble of Imperial. A similar mechanism may

70 J. R. Pritchard
 8.2.: Open Questions in the Big Bang Model

also produce cosmic strings2 and similar cosmological topological defects.)

There would therefore be approximately one monopole per Hubble volume, giving
3
a physical number density of monopoles at this early time of n(TGUT ) ∼ HGUT , with
2
HGUT ∼ TGUT /mPl appropriate for an RD universe. This density would have been diluted
by a factor of (1+zGUT )−3 = T03 /TGUT
3
between the GUT time and now, so the present-day
mass density of monopoles of mass Mmon would be
3
T03 TGUT
ρmon (t0 ) ∼ Mmon , (8.36)
m3Pl

We can work out Ωmon by dividing by ρcrit = 3H02 /(8πG) ∼ H02 m2Pl :
3
T 3 T 3 mmon

mmon TGUT
Ωmon ∼ GUT 5 0 2 ∼ 1011 16 (8.37)
mPl H0 10 GeV 1014 GeV

where we have used H0 ∼ 10−42 GeV, T0 ' 2 × 10−13 GeV, mPl ' 1.2 × 1019 GeV and
put in typical values for the GUT scale and the monopole mass.
Since we know that Ω ∼ 1 we know that this is not possible. (It is often said that these
relics would “overclose” the Universe, but a better interpretation would be to say that
such a density at this early time would have caused the Universe to be closed and collapse
again on a very short timescales, rather than continue expanding for the 14 billion years
— and counting! — that we observe.) Moreover, our observations of, say, the light-
element abundances also imply that the Universe must have been radiation-dominated
at least prior to about three minutes after the big bang.
This problem was thought to be such an important issue that it was the primary
motivation for Guth’s original model of inflation — it is often referred to more specifically
as the monopole problem. In fact, we do not know if a GUT transition actually
happened in the early Universe, but there are various other transitions that may have
happened, resulting in a high density of very massive particles.

2
Not to be confused with the superstrings of string theory!

Cosmology 71
8 — Interlude

72 J. R. Pritchard
Chapter 9

Inflation

Liddle Ch. 13

In the last chapter, we discussed the way in which the Universe in which we live
started out in a very special state: nearly flat, nearly homogeneous, and dominated by
radiation. It is of course possible that these initial conditions are just a raw fact that we
have to learn to deal with, but we would prefer to find a causal mechanism to enforce these
conditions. In about 1980, Alan Guth in the USA (and independently Alexei Starobinsky
and Andrei Linde in the former USSR) came up with a mechanism that takes a broad
range of initial conditions and makes them all look like a flat, homogeneous, radiation-
dominated Universe — inflation. Furthermore, it was quickly realized that inflation also
provided a mechanism to generate density fluctuations of just the right character to grow
into the large-scale structure we observe in today’s distribution of galaxies, as well as in
the fluctuations in the CMB which were first observed in the early 1990s.

9.1 Accelerated Expansion

The basic idea of inflation is that a period of accelerated expansion takes a very small
volume of the early Universe and blows it up so much and so quickly that any inhomo-
geneities or curvature in this volume are smoothed out, and the density of nonrelativistic
particles is diluted. At the same time, any quantum fluctuations are blown up to macro-
scopic size, providing the seeds for large-scale structure.
How does accelerated expansion do all this? Note first that both the flatness and
horizon problems are a result of the behavior of the quantity (aH)−1 , approximately equal
to the comoving Hubble length. In a Universe with matter, radiation, and curvature this
is also approximately equal to the horizon size and which (it seems) must always grow
with time. But, of course, (aH)−1 = 1/ȧ, so that this quantity grows is exactly the
statement that ä < 0 (as long as we are in an expanding, not contracting, Universe)—
the expansion of the Universe is decelerating.

73
9 — Inflation

Let us start by recalling our definition of the comoving horizon distance:

Z t Z a 0
dt0 da 1
χH = dH /a = 0
= . (9.1)
0 a(t ) 0 a a H(a0 )
0 0

As we have seen, usually this quantity behaves like (“is approximately proportional to”)
the comoving horizon distance 1/(aH), and both of these are increasing functions of time.
But for a fixed present day value 1/(a0 H0 ) we can make χH as large as we want if we can
make 1/(aH) decrease with time — because that makes it increase back towards a = 0
and we can therefore increase the value of the integral. If we can do this then the Hubble
scale grows more slowly than any fixed comoving distance, as we show in a cartoon in
Figure 9.1. After an accelerating expansion, the Hubble length has shrunk with respect
to the comoving coordinates.

Hubble Volume

Inflation

Comoving "grid"

Figure 9.1: The Universe before and after inflation. The gridlines represents the comoving
coordinate system, and the circle the Hubble volume 1/H, which has shrunk with respect
to the comoving coordinates after inflation.

We see how this solves the Horizon problem in Figure 9.2. The accelerating expansion
means that the Hubble scale remains constant, but comoving scales increase much more
rapidly. Note that in this case the true horizon scale is now much larger than the Hubble
length.

74 J. R. Pritchard
 9.1.: Accelerated Expansion

dH~1/H∝t

duration of inflation

λgal∝a
1/H∝const
log length

λgal∝eHt

log time

Figure 9.2: Scales entering and leaving the Hubble scale, which is the apparent (but not
actual) horizon in a Universe with a period of accelerating expansion.

Thus, accelerating expansion should be able to solve the horizon problem. It is clear
that it can also solve the flatness problem as it means that the value of |Ωk | = |k|/(aH)
gets driven closer and closer to zero while acceleration is occuring. Heuristically, this
makes sense: if we rapidly expand a curved surface (relative to our coordinate system) it
looks more and more flat in those coordinates.

Finally, acceleration solves the relic (monopole) problem in much the same way: it
dilutes the number of massive relic particles in a given (physical) volume much faster
than “ordinary” decelerating expansion. In fact, it is a little more complicated than
this, because we still have to find a way to stop the accelerated expansion and make the
Universe radiation-dominated after the period of accelerated expansion — this is called
reheating.

After we discuss what kind of matter is necessary in order to make the Universe
accelerate, we will return to these issues and calculate just how long inflation needs
to last. But the idea of inflation is very simple: starting from a fairly generic state,
the Universe undergoes accelerated expansion, which increases the scale factor by many
orders of magnitude. However, this cools the Universe down proportional to the scale
factor, so we must find a mechanism for reheating: stopping the accelerated expansion
and converting the energy density of the Universe into radiation, at which point the
Universe looks like a hot, radiation-dominated, flat big-bang model. We show these steps
in Figure 9.3.

Cosmology 75
9 — Inflation

Inflation

a(t)

Standard Hot Big Bang

Reheating

T(t)

time

Figure 9.3: The scale factor and temperature in inflation and reheating, leading to a
Universe that looks like a standard radiation-dominated hot big bang.

9.1.1 Acceleration and negative pressure

How do we realize an accelerating Universe in a physical system? Recall the second-order
Friedmann equation for the acceleration:
ä 4πG 1
=− (ρ + 3p) + Λ . (9.2)
a 3 3
We want the right-hand side of this equation to be positive. Obviously the Λ term is
sufficient for this — we have already seen that a Λ-dominated Universe is accelerating.
But we have also seen that in the Universe as we have observed, Λ-domination has begun
only recently, and this would not have solved the problems as outlined above — we need
acceleration in the early universe.
So let us instead ignore the Λ term and concentrate on the pressure-density terms.
For acceleration, we need
ρ + 3p < 0 (9.3)
or an equation-of-state parameter

w = p/ρ < −1/3 . (9.4)

If we assume positive density, we need negative pressure. In fact, a cosmological constant

is exactly equivalent to matter with w = −1, and we will discuss models of inflation in
which this is the case, but really any sufficiently negative equation of state will do.

76 J. R. Pritchard
 9.1.: Accelerated Expansion

For this w = −1 case, we have already seen how it affects the expansion, back in
Chapter 3. It gives a de Sitter Universe with exponential expansion:
hp i hp i
a ∝ exp Λeff /3 t ∝ exp 8πGρinf /3 t ∝ eHt , (9.5)

where in the last equality we use

ȧ p p
H= = Λeff /3 = 8πGρinf /3 (9.6)
a
and have also defined the effective density ρinf = Λeff /(8πG). Here, we are not talking
about a true cosmological constant but just some effective Λeff , realized as some substance
(field or particle), the inflaton, with density ρinf and pressure pinf = −ρinf , but any fluid
with equation of state w ≤ −1/3 might suffice.
We can now make a timeline for inflation. At very early times the Universe is
radiation-dominated (although it is possible that we could have curvature or matter
domination). Just as the late-time Universe becomes dominated by Λ, eventually the
very early Universe becomes dominated by the inflaton, at or around the GUT scale,
1015 GeV, corresponding to t ∼ H −1 ∼ 10−34 s. This transition (which is similar to a
phase transition) causes inflation to start, lasting until something like t ∼ 10−32 s (a fac-
tor of 100 in age — in the next section, we shall discuss in more detail how much inflation
is needed), during which H ∼ const, driving exponential expansion.1 This is similar to
a phase transition in solid-state physics, which can also build in long-range correlations,
exactly as we are trying to do here. This would give something like a(tf ) = exp(100)a(ti )
where i and f refer to the initial and final times. Over this period, therefore, the volume
increases by exp 300 ∼ 10130 and the temperature decreases by exp(−100). We therefore
need to reheat the Universe to a high temperature TRH < ∼ Ti , which converts the entire
energy density into radiation particles. Before inflation, a single Hubble volume contained
approximately 1014 radiation particles (using the black-body density); after inflation and
reheating, that same comoving region has 10130+14 particles of radiation, vastly increasing
the entropy density of the Universe.

9.1.2 The duration of Inflation

How long does inflation need to last in order to solve these big bang problems?
Before inflation, we have scale factor a = ai , t = ti ∼ Hi−1 , with this scale corre-
sponding to the physical size of the pre-inflationary causal horizon. During inflation, we
have a ∝ exp(Hi t), so right after inflation at tf = ti + δt  ti , that pre-inflationary
horizon has now expanded to physical size df = exp(Hδt)Hi−1 , and the Universe has a
post-inflationary reheating temperature TRH . Since then, the universe has expanded by
TRH /(3 K), so today, the original causal horizon now has size
TRH TRH
d = exp(Hδt)Hi−1 = exp(N )Hi−1 . (9.7)
3K 3K
1
There are also models of inflation where the expansion is still accelerating, but only as a power-law,
known as “extended inflation”.

Cosmology 77
9 — Inflation

where we have set N = Hδt as the “number of e-folds” that inflation lasts. We have
plentiful evidence that the universe is isotropic at or near the current Hubble length, so
we need

d > H0−1
2
3 K TRH
eN > H0−1
TRH mPl
3 K TRH
> (9.8)
H0 mPl

where we have used H ∼ T 2 /mPl appropriate for an RD universe. With H0 ∼ 10−33 eV

and T0 ∼ 10−4 eV, this gives

N > Nmin = 68 + ln(T /MPl ) = 55 + ln(T /1014 GeV) . (9.9)

This is what is needed to solve the horizon problem. In order to solve the flatness
problem we can make a similar analysis. Before inflation, we start with some initial
value of the curvature parameter, Ωk (ti ) = −k/(ai Hi )2 . Today, the scale factor is a0 =
ai eN TRH /T0 , accounting for inflationary and ordinary expansion. So the current curvature
2
−k −k −k

−N T0 Hi
Ωk (t0 ) = = = e
(a0 H0 )2 (H0 ai eN TRH /T0 )2 (ai Hi )2 TRH H0
 2
T0 TRH
= Ωk (ti ) e−N = Ωk (ti )e2Nmin −2N (9.10)
H0 mPl

where Nmin is the minimum number of e-folds required to solve the horizon problem,
Eq. 9.9. Thus, if Ωk (ti ) is of order one, the same amount of inflation that solves the
horizon problem will solve the flatness problem; increasing the initial curvature parameter
by orders of magnitude requires only increasing the number of e-folds logarithmically. In
fact, realistic inflation models tend to last much longer than Nmin , so this is easily satisfied.
Finally, inflation decreases the relic abundance down to acceptable levels from Ωmon (t0 ) ∼
1011 (Eq. 8.37). The monopoles are produced at TGUT with number density n(tGUT ) ∼
3
HGUT which is diluted by a factor (Ti /TGUT )3 by the beginning of inflation, a factor of
exp(−3N ) during inflation, and subsequently by another factor of (T0 /TRH )3 . All to-
gether, this is decreased by a factor of exp(−3N )Ti3 /TRH 3
from the no-inflation prediction
of Eq. 8.37. We must have Ti > TRH (or otherwise you would inflate again after reheat-
ing), but typically the two are comparable, so the main effect is from the exponential.
We need 3N > ∼ 11 ln 10 ∼ 25, which is easily satisfied if we already solve the horizon and
flatness problems.
Thus, we see that generically we need something like 60 e-folds to solve the various
problems. To put this another way, with at least that much expansion, very generic initial
conditions (inhomogeneous, curved, lots of heavy particles) are funnelled into what seems
to be a very special state: smooth, flat, and radiation-dominated.

78 J. R. Pritchard
 9.2.: Inflation via a scalar field

9.2 Inflation via a scalar field

Now, we’ll consider a specific model of inflation, one that will further let us see how
inflation not only solves our initial-condition problems, but also provides the seeds for
the formation of density perturbations. It is important to realize that in order to get
acceleration, classical (albeit relativistic) field theory suffices; only when we consider
the end of inflation — reheating and the seeding of density perturbations does quantum
mechanics enter.
The Lagrangian density of a scalar field, ϕ with potential V (ϕ) is

1
L(ϕ) = ∂ µ ϕ∂µ ϕ − V (ϕ) (9.11)
2
and the stress-energy tensor is then given by

T µν = ∂ µ ϕ∂ ν ϕ − Lg µν (9.12)

where g µν is the metric. For comparison, recall that we wrote down the stress-energy
tensor of a perfect fluid with pressure p and density ρ:

T µν = diag(ρ, −p, −p, −p) . (9.13)

If we assume that the field is spatially homogeneous, in fact the stress-energy takes on
the perfect fluid form, with
1
ρϕ = ϕ̇2 + V (ϕ)
2
1
pϕ = ϕ̇2 − V (ϕ) ; (9.14)
2
the correction terms for spatial inhomogeneities are of order ∇ϕ2 /a2 where ∇ is the
comoving-coordinate gradient. Because of the 1/a2 factor, the effect of any spatial gra-
dient is quickly rendered irrelevant by exponential accelerated expansion.
We can derive the equation
R 4 of√motion in several ways. For a general scalar field, we
could vary the action, S = d x −gL, where g is the determinant of the metric tensor
or we could use the covariant conservation of the stress-energy tensor ∇ν T µν = 0. But
both of these require a bit more relativity than we’ve developed here. Instead, we can
just plug in these expressions for the pressure and density into the fluid equation

ρ̇ + 3H(ρ + p) = 0 (9.15)

which gives
ϕ̈ + 3H ϕ̇ + V 0 (ϕ) = 0 . (9.16)
If it were not for the friction term proportional to H (“Hubble drag”) note that this
is exactly the same as the equation of motion for the position of a particle in a one-
dimensional potential V , so we can use our intuition from that situation.

Cosmology 79
9 — Inflation

During inflation, we are looking for solutions that look like p ≈ −ρ, which means
that the time-derivative terms must be negligible, so ρ ≈ +V , p ≈ −V . In order to do
this, the velocity ϕ̇ must be negligible — the potential must be very flat. However, for
inflation to end, the potential cannot be completely flat — it must eventually fall into a
potential well where it can oscillate, which corresponds to a mass in quantum mechanics
— a term in the potential like
1
V (ϕ) = m2ϕ ϕ2 + · · · (9.17)
2
gives a mass, so nearly any concave potential gives mass to the field. Hence we expect
that a potential of the form given in Figure 9.4 will have the right properties.2 In fact
we further expect that the ϕ field should couple to other particles, which would induce
another “friction” term Γϕ on the left-hand side of the equations of motion, where Γ
represents the decay rate of ϕ into other particles. If these particles are light, we end up
with a radiation-dominated Universe.
When the field is sitting on the approximately flat part of the potential, we say that
it is in the “slow-roll” regime, with V 0 (ϕ) very small.

Slow-roll
(acceleration)
V(!)

during inflation:
-pϕ=ρϕ≈V

Oscillations
! (reheating)

Figure 9.4: The inflationary potential. The slow-roll regime is when the potential is
relatively flat, and the reheating regime is when it strongly curved.

9.2.1 Density Perturbations

Very soon after inflation was invented, it was realized that it gives a mechanism to answer
yet another open question about the initial conditions for our hot big bang Universe: how
2
Guth’s original model of inflation had a potential barrier in between the accelerating and reheating
regimes, which required the nucleation of bubbles in a phase transition to end inflation, but the nucleation
rate was too low to end inflation properly.

80 J. R. Pritchard
 9.2.: Inflation via a scalar field

were the initial fluctuations which have subsequently grown into the large-scale structure
we observe in the Universe today generated?
The basic mechanism is very simple, and is a relatively generic consequence of the com-
bination of inflation’s accelerated expansion with the randomness of quantum mechanics.
Any system will exhibit quantum fluctuations on very small scales. Inflation takes these
fluctuations — initially on very small scales — and blows them up very rapidly. Once
they are larger than the apparent horizon (the Hubble length, H −1 ∼ ct), they are frozen
in and behave as completely classical fluctuations. (Even though the actual horizon is
much larger, the behavior of fluctuations at any given time is still controlled by the speed
of light.) Outside the horizon, fluctuations can only evolve in a very simple way, due
to the finite speed of light, so only when a scale re-enters the Horizon do overdensities
(lumps) begin to collapse.
Without quantum field theory it is difficult to do the calculation precisely, but we can
at least make a plausible argument using dimensional analysis. We start with our scalar
field, which we will split into
ϕ = ϕcl + δϕQM (9.18)
where “cl” labels the classical evolution, and “QM” are the quantum fluctuations. We
wrote down the Lagarangian density for our scalar field, L = ∂ µ ϕ∂µ ϕ/2−V (ϕ) (Eq. 9.11),
so the units on the scalar field are

[ϕ] = (length)−1 = (time)−1 . (9.19)

If we are really in the slow-roll regime, then there is only one quantity with these units
in our problem, and that is the (exponential) expansion rate, H = const. Since we know
that the average of our quantum fluctuations should be zero, this implies that we can use
this to fix the variance of our fluctuations, hδϕ2 i ∼ H 2 . In fact, a more careful analysis
gives  2
3 2 2 H
k Pϕ (k)/(2π ) = hδϕ ik ' . (9.20)
2π
In this equation, Pϕ (k) is the power spectrum of the ϕ field at spatial frequency k,
using Fourier-transform conventions to be defined in the next chapter. The factor of
2π on the right-hand side arises because this is actually the so-called Gibbons-Hawking
temperature associated with the horizon in a de Sitter Universe (the equivalent of the
Hawking temperature associated with the horizon of a black hole), TGH = H/(2π).
Because H is approximately constant during inflation, we can integrate this expression
to get the total mean-square fluctuation in ϕ, integrated over all frequencies (which are
sometimes called “modes”), which ends up giving
 2
2 H
hδϕ ik ' N × . (9.21)
2π

where N is the number of e-folds of inflation from before. In the limit of exactly expo-
nential (de Sitter) expansion, this would result in an initial power spectrum of density
fluctuations (which we will define more precisely in the next chapter) with the so-called

Cosmology 81
9 — Inflation

Harrison-Zel’dovich spectrum, P (k) ∝ k. Of course, these equations can only be approx-

imate: inflation has to end at some point, so we cannot really have H = const. This will
result in a more realistic initial spectrum,

P (k) ∝ k ns , (9.22)

where ns is the scalar spectral index (another name for density perturbations are scalar
perturbations, corresponding to actual fluctuations in the curvature of the spacetime
manifold, described by a scalar number), and we usually have ns just below one, with
details depending on V (ϕ). The amplitude of the spectrum depends on the coupling
constants present in V (ϕ), and are constrained to be quite small in most models by the
relatively small amplitude of temperature and density fluctuations observed in the CMB
and on large scales today.
We will see in the next chapter that it is straightforward to understand the evolution
of the power spectrum of such fluctuations once they exist.
Moreover, for a weakly-coupled scalar field (and we have just noted that observations
seem to require weak coupling), the distribution of these fluctuations will be very close to
Gaussian (in field theory, a free field is exactly Gaussian). Hence, once we have described
these second moments, there is no more information available to us.
In a similar manner, inflation also creates a background of gravitational radiation
(gravitons, or “tensors”). Gravitational radiation does not directly create lumps and
voids (it does not couple directly to the density of matter) but the movements it induces
are indeed visible, although as yet undetected, as patterns in the polarization of the CMB.
The gravitational radiation is described by a separate power spectrum,

PT (k) ∝ k nt (9.23)

where now nt is usually just below zero, and the amplitude is governed by the value of
V (ϕ) during inflation, i.e., by the energy scale of inflation.
The observation of these tensor modes via CMB polarization is one of the main goals
of the next generation of CMB experiments (beyond the Planck Satellite).

82 J. R. Pritchard
Chapter 10

Structure Formation

Schneider Ch. 7

10.1 Notation and Preliminaries

So far, we have been largely discussing a homogeneous Universe, with matter density
ρm = ρ̄m = Ωm ρcrit ' 10−29 g/cm3 , where we use an overbar to refer to the mean matter
density. In the galaxy, the average density is about one million times higher than the
average, and here on earth, the average density if, of course, even higher than that of
water (1 g/cm3 ). We clearly need to understand how the deviations from the mean grow
in different circumstances and on different scales.
We will need some notation that will help us separate the mean density from the
fluctuation. We define the density contrast, or fractional density perturbation,
δρ ρ(x, t) − ρ̄(t)
δ(x, t) ≡ (x, t) = (10.1)
ρ ρ̄(t)
If the Universe really is homogeneous, it is easy to define what we mean by ρ̄, but
as soon as we allow fluctuations (as we must in order to describe the real Universe), it
becomes more complicated. In relativity, this is related to the fact that we are free to use
any set of coordinates that we wish and the physics will stay the same — but the equations
can look very different. One way to see this is to notice that (at least in the early Universe
when fluctuations were small) it is possible to define a coordinate system in which ρ = ρ̄
is a function of time, but not position. Fluctuations generically grow with time, so if, in
one coordinate system, two nearby points have different densities ρ(x1 , t1 ) < ρ(x2 , t1 ) at
time t1 , then there is some time t2 < t1 at which ρ(x1 , t1 ) = ρ(x2 , t2 ) (note the different
times). But in GR we are free to redefine our time coordinate and give these time the
same time label t̂. If we use t̂ to define our averages ρ̄(t̂) then of course we now find that
δ = 0, different than we would have found using the t time coordinate. Although the
description is very different, it turns out that this freedom is mathematically very similar
to the gauge freedom in field theories like general relativity.
Just as in electromagnetism, there are certain conventional choices of background
coordinates (“gauge”) that are useful in different circumstances. One is called the syn-

83
10 — Structure Formation

chronous gauge in which all locations use the proper time of an observer located their as
the time coordinate. Another is the Newtonian gauge in which the equation of motion
for the perturbations looks most like the Poisson equation of nonrelativistic gravity.
It turns out that well inside the apparent horizon, well away from regions of very strong
gravity such as black holes, and at speeds considerably lower than c, most reasonable
choices of the gauge agree. It is mostly outside the horizon that the choice matters.
However, we shall be careful to use only physical “gauge-invariant” descriptions on such
large scales and not use equations that refer to specific coordinate systems.
It is clear that our understanding of any physical theory will only be such that it
will predict the statistics of the density perturbation, rather than the actual values as a
function of our time and space coordinates. Hence, we need to consider suitable averages
of δ and functions of δ.
Given our definition of δ(x, t) it is obvious that the spatial average at a given time
requires
hδ(x, t)i = 0 . (10.2)

(Actually, there is a subtlety, as there are three different averages that we could con-
template here. The first is a over all space at a given time; the second is an average
with respect to the probability distribution of δ. (The equivalence of these two is math-
ematically similar to ergodic theory in thermodynamics.) Of course, neither of these two
average can be performed in the real Universe, in which case averages are over observ-
able parts of the Universe. We shall assume no distinction between these possibilities in
practice.)
The next moment, the variance, will be nonzero:

hδ(x, t)δ(y, t)i = ξ (x, y, t) = ξ (|x − y| , t) . (10.3)

We have defined ξ, the density correlation function, which is also the second moment of
the density distribution. It is only a function of the distance between the points x and
y, which is a statement about the statistical isotropy of the Universe. Our underlying
description of the Universe should be invariant if we shift the origin and orientation.
(This is the spatial equivalent to a stationary random process in time.)
The correlation structure is much simpler if we consider the Fourier Transform of δ
(as are the equations of motion, as we will see soon):
Z
δ̃(k, t) = d3 xeik·x δ(x, t) (10.4)

and the inverse

d3 k −ik·x
Z
δ(x, t) = e δ̃(k, t) . (10.5)
(2π)3

With these definitions, we can calculate the correlation function of our Fourier-Transformed

84 J. R. Pritchard
 10.1.: Notation and Preliminaries

quantities, all evaluated at a single time, so we suppress our time coordinate:

D E Z Z 
0 3 ik·x 3 0 ik0 ·x0 0
δ̃(k)δ̃(k ) = d xe δ(x) × d x e δ(x )
Z Z
0 0
= d x d3 x0 ei(k·x+k ·x ) hδ(x)δ(x0 )i
3

Z Z
0 0
= d x d3 x0 ei(k·x+k ·x ) ξ (|x − x0 |)
3

Z Z
3 0 0 0
= dy d3 x0 ei(k·y+k·x +k ·x ) ξ(y)
Z Z
3 0 i(k+k0 )·x0
= dx e d3 y eik·y ξ(y)

= (2π)3 δD (k + k0 )P (k) , (10.6)

where δD is the Dirac delta function. Because δ is real (i.e., not complex), we could also
use δ(−k) = δ ∗ (k) and write this as
D E
δ̃(k)δ̃ ∗ (k0 ) = (2π)3 δD (k − k0 )P (k) . (10.7)

These equations define the power spectrum of density fluctuations,

Z Z
sin kx
P (k) = d xe ξ(x) = 4π dx x2
3 ik·x
ξ(x) (10.8)
kx
which turns out to be the three-dimensional Fourier transform of the correlation function
considered as a function of only one variable. Because ξ(x) is only a function of the
magnitude of the length vector, x = |x|, P (k) is similarly only a function of the magnitude
of the wavenumber, k = |k| (since we can actually do the angular part of the integral).
The correlation function and power spectrum are the second-order moments of the density.
If the density is described by a Gaussian distribution, then these second-order moments
(along with the means hδi = 0) completely describe the matter density distribution.
Note that we are talking about the distribution of the function δ(x), which we
can think of as an infinitely-multivariate probably distribution Pr[δ(x1 ), δ(x2 ), δ(x3 ), . . .]
where the xi enumerate the uncountably infinite number of possible positions. At least
for a Gaussian, this functional distribution is just described by our power spectrum or cor-
relation function and no more. In Fourier space, this is especially simple: the δ function
in Eq. 10.7 means that the values are uncorrelated, and so we can just write
2
 
1  1 δ̃(k) 
h i
Pr δ̃(k) = p exp − (10.9)
2 P (k)

2πP (k)

We can also calculate the statistics of some more physically relevant quantities. Con-
sider the density fluctuation measured not at a point, but in spheres of some radius R.
At any point, we can measure the mass in that sphere
Z Z
δMR 1
(x) = 3
d y δ(x + y) = d3 u WR (u − x)δ(u)
3
(10.10)
MR 4πR /3 y<R

Cosmology 85
10 — Structure Formation

where in the second equality we define the window function, WR (x), which is 1/V within
the volume, and 0 outside. We can now calculate the second moments of the MR distri-
bution from the second moments of δ. First note that Eq. 10.10 is a convolution, so in
Fourier space the transform of δMR /MR is just the product of the transforms of δ and
the window function. Hence
d3 k ik·x
Z
δMR
(x) = e δ̃(k)W̃R (k) (10.11)
MR (2π)3
with
Z
1
W̃R (k) = d3 x eik·x WR (x)
4πR3 /3
Z
1
= 3
dx x2 dφ d cos θ eikx cos θ
4πR /3 x<R
 
3 sin kR
= − cos kR
(kR)2 kR
≡ W̃ (kR) (10.12)

(in particular, W̃ (0) = 1, so the zero-radius sphere is equivalent to just using δ itself), so
* 2 +
d3 k d3 k 0
Z  
δM 2
0 0
= δ̃(k )W̃R (k )
M (2π)3 (2π)3
R
d3 k 2 k3
Z 2
Z
dk
= W̃ (kR) P (k) = W̃ (kR) P (k)
(2π)3 k 2π 2
Z 2
dk
= W̃ (kR) ∆2 (k) (10.13)
k
where we define ∆2 (k) = k 3 P (k)/(2π 2 ), which is the contribution per logarithmic integral
to the mean-square fluctuation, and which has no units. (Recall that we encountered the
similar combination k 3 Pϕ (k)/(2π 2 ) related to the inflaton field in Eq. 9.20). Roughly
speaking, ∆2 (k) gives the density fluctuation on a length scale L ∼ 2π/k.

10.2 Spherical Collapse

To get a feel for the way perturbations evolve, let’s consider an idealized situation: a
perfectly spherical perturbation in an otherwise homogeneous, flat, Ωm = 1, Universe
with density ρ̄(t). We can do this by picking some point which we will take to be r = 0
and compressing all of the matter within r < r2 to a higher uniform density, so it now
has a radium r1 < r2 . The average density within r2 is the same as before, ρ̄, but within
r1 , the density is ρ > ρ̄.
But we saw near the beginning of the course in Section 2.3 that the mass inside a
uniform density sphere is unaffected by the mass outside it. Hence, we know that the
mass inside the sphere must behave like a Universe with a higher density — and hence

86 J. R. Pritchard
 10.2.: Spherical Collapse

a higher Ω. But our mean density is Ω = 1, and hence our overdense sphere will behave
like an Ωm > 1 Universe — it will expand and recollapse.
Outside of our carved-out sphere, r > r2 we just have the usual equations (with an
overbar to represent the mean):
 2
ā˙ 8πG
= ρ̄ (10.14)
ā 3

which, for nonrelativistic matter gives

1
ā ∝ t2/3 and ρ̄ = ∝ ā−3 . (10.15)
6πGt2
For r < r1 , we instead have
 2
ȧ 8πG k
= ρ− 2 (10.16)
a 3 a
and consider some early time ti at which a = ai , Ω = Ωi and ρ = ρi such that ρi  k/a2i
so that within the perturbation Ωi ' 1 and ρi ' ρ̄(ti ) — i.e., we start with a small
perturbation.
We saw in a problem sheet that we could calculate the maximum scale factor before
the recollapse of a closed Universe, and the time at which this happens:
 3/2
amax Ωi π Ωi π amax −1/2
= Hi tmax = 3/2
= Ωi (10.17)
ai Ωi − 1 2 (Ωi − 1) 2 ai
so  3
ai 3π
ρmax = ρi = 1/2
(10.18)
amax 32GΩi t2max
which we can compare to ρ̄(tmax ) = 1/(6πGt2max ) so

ρmax 9π 2 9π 2
= ' ' 5.55 . (10.19)
ρ̄(tmax ) 16 16

This is equivalent to δmax = 4.55. If instead we concentrate not on the full collapse,
but just on the very earliest times after ti , we find that δ ∝ a(t) ∝ t2/3 . These
small overdensities are called the “linear regime” for reasons that will be more obvi-
ous in the next section. We can usually calculate things much more readily in the lin-
ear regime, and so it is common to compare this nonlinear (but idealized) case with
what would happen if linear evolution continued. Between ti and tmax , we would have
δLin /δi = āmax /āi = (tmax /ti )2/3 = (3/5) (3π/4)2/3 (amax /ai ) ' 1.06/δi . So, irrespective of
the starting conditions, maximum density (a.k.a. “turnaround”) occurs when the linear
density contrast would have been δ ' 1.06
In a perfectly uniform universe, this overdensity would just collapse down to a point.
But in a more realistic scenario, it will virialize and convert its gravitational energy into
random kinetic energy (by a process known as “violent relaxation”!). At the maximum,

Cosmology 87
10 — Structure Formation

with KE = 0, the overdensity has total energy P E = Egrav ' −3GM/(10rmax ). After
virialization to form a sphere of radius rvir , we have KE = −P E/2 and therefore a
total energy Egrav /2 ' −3GM/(20rvir ). For the energy to remain constant, we need
rvir ' rmax /2 or ρvir ' ρmax /8. Symmetry of the solutions for a closed universe mean
that collapse occurs at tvir = 2tmax , so that the background density at this time is just
ρ̄(tvir ) = ρ̄(tmax )/4. Combining these results gives

ρ(tvir ) 8ρ(tmax ) 9π 2
1 + δvir = = = 16 × = 9π 2 ≈ 178. (10.20)
ρ̄(tvir ) ρ̄(tmax )/4 16

A similar calculation to before gives a linear density contrast δL (tvir ) = 1.68. These values
provide key insights into structure formation. We would expect a collapsed structure to
have a density ∼ 178 the density at the time of collapse and indeed this appears to hold
true for structures like galaxies and clusters of galaxies we observe.

10.3 Linear Perturbations

Now we will consider a more general situation allowing an arbitrary δ(x, t), although we
will find that we cannot solve the equations for all possible times and values of the density
— we will have to use perturbation theory.

10.3.1 Newtonian Theory—non-expanding

As a warm-up, let us consider the case of a non-expanding medium; this was first con-
sidered by Jeans in the early 20th Century. We will describe our fluid by its density, ρ,
pressure, p and velocity v; we also need to consider the gravitational potential Φ. All of
these quantities are functions of position, x and time, t. First, we consider the continuity
equation. The change in the mass in some volume V with surface A must be equal to the
flux through the surface:
Z
dM
= − ρv · dA
dt
Z ZA
∂ρ
dV = − ∇ · (ρv) dV (10.21)
V ∂t V

where we use Gauss’ theorem, so, in differential form

∂ρ
+ ∇ · (ρv) = 0 continuity . (10.22)
∂t
Next, we consider F = ma, which in this context is known as the Euler equation, with a
gravitational force and aRpressure force. The gravitational force is Fg = −m∇Φ and the
pressure force is Fp = − V (∇p)dV . By the chain rule, a = v̇ = ∂v/∂t + (v · ∇)v.

∂v ∇P
+ (v · ∇)v = − − ∇Φ Euler . (10.23)
∂t ρ

88 J. R. Pritchard
 10.3.: Linear Perturbations

We also need the Poisson equation relating gravitational potential and matter density,

∇2 Φ = 4πGρ Poisson , (10.24)

and an equation of state linking the pressure and density, usually given in the form of an
expression for the [adiabatic] speed of sound,
 
2 ∂p
cs = equation of state . (10.25)
∂ρ adi

These equations are too complicated to solve in complete generality. Instead, we will
use perturbation theory. It is easy to see that a zeroth-order solution is given by v = 0,
ρ = ρ0 = const, p = p0 = const (assuming no spatial variation in the equation of state)
and ∇Φ0 = 01 We will the write all of our variables as

ρ = ρ0 + ρ1 p = p0 + p1 v = v1 Φ = Φ 0 + Φ1 (10.26)

where 0 refers to the unperturbed solution, and 1 to our small perturbations. To linear
order (i.e., any products of first-order quantities, such as ρ1 × v1 , are neglected), and
subtracting off the zeroth-order solution, our equations become
∂ρ1
+ ρ0 ∇ · v1 = 0 (10.27a)
∂t
∂v1 1
+ c2s ∇ρ1 + ∇Φ1 = 0 (10.27b)
∂t ρ0
2
∇ Φ1 = 4πGρ1 (10.27c)

If we take the divergence of (b) and substitute in (c) and the time derivative of (a), we
get a single, second-order differential equation:

∂ 2ρ
− c2s ∇2 ρ1 = 4πGρ0 ρ1 . (10.28)
∂t2
This equation is straightforward to solve by considering plane-wave solutions (or, equiv-
alently, writing these equations in terms of the spatial Fourier transform of ρ). In terms
of our density perturbation δ = ρ1 /ρ0 , we will try

ρ1 = A exp(ik · x + iωt)ρ0 . (10.29)

Substituting this into Eq. 10.28 gives

−ω 2 ρ1 + k 2 c2s ρ1 = 4πGρ0 ρ1
−ω 2 + k 2 c2s = 4πGρ0 (10.30)
1
Actually, this isn’t true, as it contradicts the Poisson equation! This is sometimes called the Jeans
swindle, and it is nonetheless a good approximation for what actually goes on. Moreover, the same issues
do not arise in the expanding case we will discuss next.

Cosmology 89
10 — Structure Formation

giving the dispersion relation

ω 2 = c2s k 2 − 4πGρ0 = c2s (k 2 − kJ2 ) (10.31)

defining the Jeans wavenumber,

√
4πGρ0
kJ = . (10.32)
cs

If k > kJ , ω 2 > 0 and the solution is oscillatory; if k < kJ , ω 2 < 0 and the solution is
exponentially growing or decaying (there is usually one of each as this is a second-order
equation). In the limit k  kJ , ρ ∝ e±t/τ with timescale τ ' (4πGρ)−1/2 .
Basically, small-scale perturbations oscillate as sound waves, whereas large-scale per-
turbations can grow. Another way to see how the behavior changes is to compare the
gravitational timescale τg ∼ (Gρ)−1/2 with the pressure timescale (τp ∼ λ/cs ' (kcs )−1 ).
When τg < τp , collapse has time to occur before pressure can act to respond. We will see
very similar behavior in the case of the expanding universe.
Note that we have found another reason why our Fourier analysis of Section 10.1 is
valuable: individual Fourier modes evolve independently.
(I am ignoring the fact that we have not actually solved for the velocity field, and
in fact there is some subtlety: from Eq. 10.27a, the density only allows us to solve for
∇·v, which leaves ∇×v undetermined — this is just conservation of angular momentum,
which, to linear order, doesn’t couple to the density field.)

10.3.2 Perturbation theory in an expanding Universe

The generalization of the Jeans analysis to an expanding Universe is technically difficult,
but conceptually not much different from the above case. We have to account for a few
new features:

• The background evolution is not time-independent, but ρm ∝ a−3 , ρr ∝ a−4 for NR

matter and radiation;

• The continuity, Euler, and Poisson equations must be updated to take account of
the possible presence of relativistic matter (e.g., radiation); and

• It is easiest to express the final differential equations in comoving coordinates.

In fact, we will simplify a bit from the most complex possible situation of multiple com-
ponents (e.g., matter plus radiation), each possibly with its own fluctuations. Instead,
we will largely concern ourselves with the fluctuations in the (NR) matter component,
even when the Universe is radiation dominated.
We have already seen how to modify our equations to account for relativistic matter.
In the continuity equation, we replace ρ with ρ + p = ρ(1 + w) as in the homogeneous
fluid equation, Eq. 2.42:
ρ̇ + (1 + w)∇ · (ρv) = 0 . (10.33)

90 J. R. Pritchard
 10.3.: Linear Perturbations

In this case, the density and pressure refer to the total from all components. Each
component will also satisfy its own separate conservation equation (e.g., the number of
dark matter particles is conserved as long as we are long after freeze-out for them).
Similarly, we must replace ρ with ρ + 3p in the Poisson equation, as in the second-
order Friedmann equation Eq. 2.41 (which is, after all, an equation for the acceleration,
so it should not be surprising that is this combination that is the relativistic source for
gravitational acceleration):
∇2 Φ = 4πGρ(1 + 3w) . (10.34)
In the expanding universe, we will write our perturbations as ρ = ρ̄(1 + δ), p = p0 + δp =
ρ̄(w + c2s δ), v = v0 + u = Hx + u and Φ = Φ0 + φ.
We also need to change from fixed coordinates x to comoving (also called Lagarangian)
coordinates q. These are related by x = aq, with the unperturbed velocity v0 = Hx.
We must calculate the time derivative of some function f (x, t),
   
∂f ∂f (aq, t)
=
∂t q ∂t q
   
∂f X ∂f d(aqi )
= +
∂t x i
∂xi dt
 
∂f
= + (v0 · ∇x ) f (10.35)
∂t x
and the spatial derivatives are related by a simple change of variables,
1
∇x f = ∇q f . (10.36)
a
We also need the specific cases ∇x · x = 3, ∇x · v0 = 3H and (u · ∇x )v0 = Hu.
With all of these developments, we can derive the equations for the NR matter per-
turbation in comoving coordinates, again to linear order,
1
δ̇ + ∇ · u = 0 (10.37a)
a
ȧ c2 1
u̇ + u + s ∇δ + ∇φ = 0 (10.37b)
a a a
2 2
∇ φ = 4πGa ρ̄m δ , (10.37c)

which can be compared to the non-expanding case, Eq. 10.27. Note that in these equations
δ = δρm /ρm where ρm is the average NR matter density (not the total density). We allow
for the possibility of a non-zero sound speed for the matter (e.g., for baryons).
We can again combine these to form a single second-order differential equation
ȧ c2
δ̈ + 2 δ̇ − s2 ∇2 δ − 4πGρ̄m δ = 0 . (10.38)
a a
Compared to the non-expanding, matter-only case, Eq. 10.28, this equation has a few
new features. First, note the explicit presence of the factors of w from the continuity and

Cosmology 91
10 — Structure Formation

Poisson equations. Second, there is a factor of 1/a2 in the spatial gradient term. Finally,
there is a new term, 2Hδ, occasionally referred to as “Hubble Drag”, which will serve to
slow the growth of perturbations compared to the exponential form in the non-expanding
case.
Nonetheless, we can analyze this equation in much the same manner as before. By
substituting in a plane-wave solution (or, equivalently, by Fourier transforming δ), we
transform ∇2 → −k 2 and get
 2 
ȧ 2 k 4πGρ̄
δ̈ + 2 δ̇ + cs 2 − δ=0. (10.39)
a a c2s

Recall that we are in comoving coordinates, so k is measured in comoving coordinates;

the physical wavenumber is k/a. Hence, as before, we can identify the physical Jeans
wavenumber, defined by √
4πGρ̄
kJ = . (10.40)
c2s
Alternately, we can define the Jeans length λJ = 2π/kJ , again in physical coordinates,
or the Jeans mass 4πλ3J ρ̄/3. The analytic form of the solutions to this equation are
somewhat more complicated than in the non-expanding case, but their character is the
same. For small scales, k  kJ , we again have oscillatory solutions, with frequency
approximately given by ω ' cs k/a. The exact form of this solution is a linear combination
of Bessel functions, not a sinusoid, corresponding to a mildly damped sound wave, with
the damping on a the scale of the Horizon.
In particular, the qualitative features of this analysis also apply to the perturbations
to the dominant component in a radiation-dominated Universe. In this case, the Jeans
length is comparable to the Horizon scale (since the sound-speed is comparable to the
speed of light). Hence, perturbations in a radiation-dominated Universe do not grow
inside the horizon.
To examine the large-scale limit, we will take k → 0 (in which case the equation
becomes the correct linear equation even in the relativistic, superhorizon case). With
vanishing spatial derivatives, we can write the solution as

δ(q, t) = δ0 (q)D(t) (10.41)

where δ0 gives the spatial part of the solution at some particular time (i.e., the initial
conditions), and D(t) is known as the growth factor. In the matter-dominated epoch, the
equation for D is just
 2
ȧ 3 ȧ
D̈ + 2 Ḋ = 4πGρ̄D = D, (10.42)
a 2 a

where we have used the first-order Friedmann equation to eliminate ρ̄. This is a second-
order equation, hence with two solutions. Often, one of these will grow in time, and the
other will decay, and these are known, respectively, as the growing mode and decaying

92 J. R. Pritchard
 10.3.: Linear Perturbations

mode. In both matter- and radiation-dominated Universes, this has power law solutions.
First, matter-domination, with w = 0, a ∝ t2/3 , ȧ/a = 2/(3t), has
D = C1 t2/3 + C2 t−1 . (10.43)
2/3 −1
There is a growing mode D+ ∝ t ∝ a and a decaying mode D− ∝ t ∝ H. Since the
Jeans length is infinite, this growth applies on all scales when the pressure is negligible.
The more general equation, valid during radiation domination and on superhorizon
scales (which can nonetheless be derived from the above Newtonian equations with the
appropriate special-relativistic generalizations to include the effects of pressure), is
 2
ȧ 3 ȧ
D̈ + 2 Ḋ = (1 + w)(1 + 3w)D , (10.44)
a 2 a
In a radiation-dominated universe, now with w = 1/3, a ∝ t1/2 , ȧ/a = 1/(2t), so the
solution to this equation is
D = C1 t + C2 t−1 , (10.45)
with growing mode D+ ∝ t ∝ a2 and a decaying mode D− ∝ t−1 ∝ H. We saw that the
Jeans length in an RD universe is comparable to the Hubble scale, so only perturbations
outside the (apparent) horizon grow in this way.

Multiple Components, Curvature and Dark Energy

In the previous analysis, we have assumed a single component in the Universe with
w = c2s . We very often care about a more complicated case. For example, even in
a universe dominated by curvature, radiation, or a cosmological constant, we often care
about the behavior of the (pressureless) matter. In this case, each component is separately
conserved, and each contribute separately to the gravitational potential and to the force
on any particle. The result of this is a slightly more complicated second-order differential
equation " #
ȧ c2i k 2 X
δ̈i + 2 δ̇i + δi − 4πG ρ j δj = 0 . (10.46)
a a2 j

where the subscript labels the species and the sum is over all species, and δj = δρj /ρj
where ρj is the average of particle species j. (As written, this equation ignores the
factors of w since we will use it below to calculate the evolution of the w = c2s = 0 matter
perturbations.)
In particular, we can use this for the common case of matter perturbations in a
radiation-dominated universe. In this P case, we care about i = m = matter, so cm = 0
and we can ignore the 4πGρ term since j ρj δj = ρm δm +ρr δr ; the first term is suppressed
since ρm  ρr and the second since δ ' 0 since (as we saw above) there are only sound
waves, not growing perturbations, to the radiation itself. Hence, the differential equation
becomes
ȧ
0 = δ̈m + 2 δ̇m
a
1
= δ̈ + δ̇ . (10.47)
t

Cosmology 93
10 — Structure Formation

since there are no spatial derivatives, we can consider this as an equation for the growth
factor, with solution
Dm = C1 + C2 ln t RD . (10.48)
The growing mode is no longer a power law, but logarithmic. This is sufficiently slowly
that, essentially, matter perturbations do not grow during radiation-dominated periods.
We can similarly calculate the behavior of matter perturbations in an open curvature-
dominated universe. We now go back to our original single-component Eq. 10.42, with
w = c2s = 0, but use the appropriate a ∝ t. Now, the term 4πGρ̄ = (3/2)Ωtot (t)H 2 , so
2 3 2
0 = δ̈ + δ̇ + 2 Ω(t) ' δ̈ + δ̇ (10.49)
t 2t t
where we ignore the last term since not only do we eventually have Ω(t)  1, but
moreover it is falling with time so it must eventually become negligible. The solution is
Dm = C1 + C2 /t CD . (10.50)
Again, there is no growing mode.
Finally, we consider the behavior of matter perturbations in a universe dominated by
a cosmological constant: w = −1, a = exp(Ht), ȧ/a = H = const:
0 = δ̈ + 2H δ̇ − 4πGρ̄δ ' δ̈ + 2H δ̇ , (10.51)
where again, we find that we can ignore the final term, this time since ρ̄ is quickly diluted
by the exponential expansion, and the solution is
Dm = C1 + C2 e−2Ht ΛD . (10.52)
Yet again, we find that perturbations do not grow.

Relativistic Perturbation Theory

In the previous sections, we have assumed that we can apply our Newtonian calculation
on all scales, even those larger than the apparent horizon. In principle, as discussed in
Section 10.1 above, we should not necessarily be able to trust our calculation on those
scales. In fact, the calculation we have done is applicable on superhorizon scales, at least
for some particular choice of coordinates (i.e., gauge).

10.4 The processed power spectrum of density per-

turbations
10.4.1 Initial Conditions
We can now give another interpretation to the Harrison-Zel’dovich power spectrum,
P (k) ∝ k, defined in the previous chapter.2 It such that the amplitude of density fluctu-
2
It is probably more properly referred to as the Harrison-Zel’dovich-Peebles-Yu spectrum after all the
authors who discussed it more or less independently.

94 J. R. Pritchard
 10.4.: The processed power spectrum of density perturbations

ations for scales entering the horizon is a constant,

2 D2 (t)
δH (t) = ∆2 (k = aH, t) = ∆i (k = aH) = const . (10.53)
D2 (ti )
Here, k = aH is the comoving wavenumber corresponding to the comoving length scale
∼ 1/(aH) at time t, D(t) is the growth factor calculated in Section 10.3.2, and ti is some
suitably early time (just after inflation, say). Because this combination doesn’t change
with time, the behavior is said to be scale-free (which is a slightly different and more
restrictive use than in other fields of physics).
Using the results of the previous section, we can show that these two definitions are
equivalent. In a radiation-dominated Universe, for scales outside the horizon, we have
D ∝ a2 , a ∝ t1/2 , and aH = ȧ ∝ t−1/2 ∝ a−1 so
2
δH ∝ a4 (aH)3 (aH)ns ∝ a4 × a−3−ns
∝ a1−ns (10.54)

which is indeed constant for ns = 1. Similarly, in a matter-dominated Universe, D ∝ a,

a ∝ t2/3 , aH ∝ t−1/3 ∝ a−1/2 , so
2
δH ∝ a2 (aH)3 (aH)ns ∝ a2 × a−3/2−ns /2
∝ a(1−ns )/2 (10.55)

which is again constant for ns = 1. In a realistic inflationary model, we expect ns slightly

below one; ns ' 0.95 is observed.

Curvature and isocurvature fluctuations

So far the discussion has concerned either a single, dominating, component, or a small
admixture of matter in a radiation-dominated universe. In the former case, our density
perturbations correspond directly to perturbations of the metric — the curvature of the
manifold. These are often referred to as adiabatic perturbations, referring to the fact
that the entropy (which resides almost entirely in the radiation) is constant.
But the matter/radiation case points out there are more possibilities. Consider a
universe with two components each with a different equation of state, matter and radia-
tion, say. It is possible to perturb the components so that δρm = −δρr ; this gives a net
perturbation δρ = δρm + δρr = 0, so the manifold has no curvature. These are called
isocurvature perturbations.
Outside the horizon, adiabatic and isocurvature perturbations behave very differently,
as the microphysics encoded in the equation of state cannot act. At horizon crossing,
the initial conditions are very different from δH = const. Inside the horizon, the different
components behave differently and, eventually, the isocurvature pertubations become
actual perturbations to the curvature. Because of the different initial conditions, however,
the shape of the processed power spectrum can be very different than for adiabatic initial
conditions. At present, adiabatic perturbations seem to fit the data much better, as
shown in Section 10.4.3 below.

Cosmology 95
10 — Structure Formation

10.4.2 The transfer function

Finally, we can combine all of the these calculations — the initial conditions and the evo-
lution of different kinds of perturbations inside and outside the horizon — and calculate
the power spectrum of density perturbations that we expect to observe at a given time.
Since inflation (or, most likely, any other theory of the early Universe) only provides a
statistical description of the initial conditions, that is the best we can hope for.
Under linear evolution, each mode δ̃(k, t) evolves independently and hence the evolved
density (as well as the potential, pressure and velocity) is a linear functional of the initial
conditions. If the initial conditions were Gaussian, then so are the evolved quantities,
albeit with a different correlation function or power spectrum. (However, we may need
more than only the power spectrum if the initial conditions are non-Gaussian; similarly,
nonlinear evolution would also require a non-Gaussian description.) We further expect
that our growing-mode solutions will dominate after sufficient evolution. Hence we expect
that we should be able to write our processed power spectrum as

P (k, t) = Ti2 (k)P (k, ti ) , (10.56)

where ti is some early initial time (i.e., when P (k) = Ak ns is a good description over all
scales). Ti2 (k) is the transfer function which, for linear evolution in Fourier space, acts
separately on each scale.
We need to combine information for scales inside and outside the horizon, before and
after matter-radiation equality. We summarize those results in Table 10.4.2.

Inside Outside
RD - a2
MD a a

Table 10.1: The growth of perturbations, inside and outside the apparent horizon during
matter- and radiation-domination. The entries correspond to the behavior of the growing
mode, D+ (a) (there is no growth inside the Horizon during radiation-domination).

The transition from radiation to matter domination occurs at 1 + zeq = Ωm /Ωr ,

at which time comoving Hubble length was 1/(aH)eq , corresponding to a comoving
wavenumber keq = aeq Heq
We say a scale “enters the horizon” when its length is comparable to the Hubble
length 1/H (note that this is really the apparent horizon if inflation has happened). In
terms of our comoving wavenumber, this happens k = (aH)enter ; during the post-inflation
decelerating phase of evolution, the Hubble scale grows in comoving coordinates, so new
scales are constantly entering the Horizon as we saw in Figures 8.1 and 9.2. We show the
time evolution of the different scales in Figure 10.1
Scales larger than the Hubble scale at equality (k < keq ) entered the apparent horizon
during matter domination. They have always been growing, by a2 during RD, and by a
during MD, independent of their scale. Their growth from ti well before equality to some

96 J. R. Pritchard
 10.4.: The processed power spectrum of density perturbations

Matter-Radiation Equality

Density Perturbation

Scale enters
horizon during RD
D!a
D=const

Scale enters
horizon during MD

D!a2

Scale Factor

Figure 10.1: Growth of different scales, entering the horizon during radiation domination
(blue, upper) and matter domination (red, lower).

late time t0 well after equality is

D(teq ) D(t0 )
δ(t0 ) = × δ(ti )
D(ti ) D(teq )
 2
aeq a0

= × δ(ti ) k < keq (10.57)
a2i aeq
The factors involving D or a on the right hand side are independent of position and hence
these scales will preserve their initial power spectrum. In the language of our transfer
function,
T (k) = const k < keq . (10.58)
Scales smaller than the Hubble scale at equality (k > keq ), however, entered the
apparent horizon during radiation domination. Hence they experienced a deficit of growth
while they were inside the horizon during radiation domination. A given comoving scale
k = aenter Henter will not grow between tenter and teq and hence
D(tenter ) D(t0 )
δ(t0 ) = × δ(ti )
D(ti ) D(teq )
a2 a0
= enter
2
× δ(ti )
ai aeq
 2
a2enter aeq a0

= 2 × × δ(ti ) k > keq , (10.59)
aeq a2i aeq
where in the last equality we have pulled out the same constant factors as occur for those
scales entering during MD (Eq. 10.57). During RD, a ∝ t1/2 At this time, k = (aH)enter ∝

Cosmology 97
10 — Structure Formation

(t1/2 t−1 )enter ∝ 1/aenter . Hence scales entering during RD experience a relative deficit of
growth by the factor
a2enter
T (k) ∝
a2eq
2
keq
∝ 2 ∝ k −2 k > keq . (10.60)
k
Combining our two cases,
(
1, if k < keq
T 2 (k) = const × (10.61)
k −4 if k > keq .

Traditionally, the time evolution is handled separately, so we set the initial constant to
be one. Thus, an initial spectrum with P (k, ti ) = Ak ns evolves to P (k) = Ak ns on large
scales and P (k) = Ak ns −4 on small scales. (In practice the turnover is not instantaneous,
so the spectrum is smooth near k = keq .) We show how this looks in Figure 10.2 for
ns = 1. (For this specific case, this can also be derived by using the fact that the horizon-
scale perturbation δH , defined in Eq. 10.53 is constant in time, along with the sub-horizon
growth functions of Table 10.4.2.)

3.5

3 P(k) ∝ k
2.5
Log P

1.5

1 P(k) ∝ k-3
0.5

0
0 1 2 3 keq 5 6 7 8

Log k

Figure 10.2: The processed power spectrum of density perturbations, P (k), for an initial
spectrum P (k) ∝ k.

The present-day power spectrum thus depends upon various cosmological parameters.
First, it depends on the value of keq , and hence the ratio of the matter and radiation
densities in the Universe. It obviously depends on the index ns , and hence on some aspects
of inflation. As we will see soon, it also depends on other cosmological parameters through
baryonic features. We see a relatively recent compilation of measurements in Figure 10.3.

98 J. R. Pritchard
f decorrelated power spectra, those for SDSS,
PSCz. Because the power spectra are decor-
fair to do “chi-by-eye” when examining this
similarity in the bumps and wiggles between

Fig. 36.— Same as Figure 35, but restricted to a comparison

of decorrelated power spectra, those for SDSS, 2dFGRS and PSCz.
The similarity in the bumps and wiggles between the three power
10.4.: The processed power spectrum of density perturbations
spectra is intriguing.

mparison with other galaxy power spectrum measure-

us caveats must be borne in mind when interpreting
SDSS power spectrum measurements are those from
cted for the red-tilting effect of luminosity dependent
ly angular analyses of the APM survey (Efstathiou
) and the SDSS (the points are from Tegmark et al. Fig. 37.— Comparison of our results with other P (k) constraints.
es in the magnitude range 21 < r ∗ < 22 — see also The location of CMB, cluster, lensing and Lyα forest points in this
2002) should also be free of this effect, but rep- plane depends on the cosmic matter budget (and, for the CMB,
Figure 10.3: The observed power spectrum of density perturbations, as measured from a
mixtures of luminosities. The 2dFGRS points are on the reionization optical depth τ ), so requiring consistency with
is of HTX02, and like the PSCz points (HTP00) and SDSS constrains these cosmological parameters without assumptions
variety of techniques, from Tegmark et al 2003.
s (THX02) have not been corrected for this effect, about the primordial power spectrum. This figure is for the case of a
rcival et al. 2dFGRS analysis should be unafflicted “vanilla” flat scalar scale-invariant model with Ωm = 0.28, h = 0.72
ing. The influential PD94 points (Table 1 from Pea- and Ωb /Ωm = 0.16, τ = 0.17 (Spergel et al. 2003; Verde et al. 2003,
994), summarizing the state-of-the-art a decade ago, Tegmark et al. 2003b), assuming b∗ = 0.92 for the SDSS galaxies.
ming IRAS bias of unity and the then fashionable
ter Ωm = 1.
Hot Dark Matter
So far, we have assumed that the matter component in the Universe is dominated by cold
dark matter, with a small admixture of baryons which will be discussed in the following
section. This pressureless cold dark matter is moving very slowly and is only affected by
gravity. However, another (conceptually) possible form of dark matter is so-called hot
dark matter, light particles which are moving at velocities comparable to the speed of light
at early times. One possible realization of this would be a light neutrino species, produced
in sufficient abundance to have Ων ≈ 0.3. These large speeds are higher than the “escape
velocities” of density perturbations, which serves to smooth out small perturbations as
they are forming. This would have the effect of a sharp cutoff in the formation of structure
— a cutoff that would prevent the formation of the structure that we see in the Universe
today. Hence, light neutrinos are no longer considered a viable model for the dark matter.
More precisely, they cannot be all of the dark matter, although current data allow a small
admixture of hot dark matter.

10.4.3 The effect of baryons: BAOs and the CMB

Baryon Acoustic Oscillations
There is one more important complication to consider: the presence of baryons as a
separate component of matter. At early times, the baryons have pressure not because of
their intrinsic equation of state, but because they are tightly coupled to the radiation.
Hence, the “baryon-photon fluid” has an equation of state reduced slightly from c2s ' 1/3
by the presence of the baryons, until decoupling at z ' 1, 100. Even though this is
already the matter-dominated era, perturbations in the baryons do not grow, even though

Cosmology 99
10 — Structure Formation

perturbations in the pressureless dark matter are able to collapse. It is exactly the
pattern of these baryon-photon sound waves that we see in both the Cosmic Microwave
Background and, at a weaker level, in the matter power spectrum as measured by the
distribution of galaxies. Here, they are known as Baryon Acoustic Oscillations (BAOs).
In fact, the pattern generated by these sound waves is very specific, and they therefore
have a characteristic scale of the maximum distance that a sound wave could have trav-
elled before the baryons and photons decouple at t ' 370, 000 or z ' 1, 100, corresponding
to a scale of roughly 100 Mpc in the Universe today. Indeed, this is the characteristic
scale of spots in the CMB, and more recently we have begun to observe that there are
more galaxies than expected at this separation, a phenomenon known as baryon acoustic
oscillations. Moreover, because this scale is fixed in comoving coordinates by our CMB
observations, it can be used as a standard ruler to measure the cosmological parameters.
The presence of baryons thus modifies our transfer function: in real space, it would
induce a peak at a separation at approximately 100 Mpc; in Fourier space this shows
up as “bumps and wiggles” in T (k) at that scale (roughly speaking, this is because the
Fourier transform of a spike is a sinusoid). These have indeed been observed by the
2DF and Sloan surveys of galaxies on large scales (Figure 10.4), and are beginning to
allow to us to determine cosmological parameters from measurements of the galaxy power
spectrum.

Baryon Acoustic Oscillations

Fig. 3.— As Figure 2, but plotting the correlation

s2 . This shows the variation of the peak at 20h−1 Mp
controlled by the redshift of equality (and hence by
ing Ωm h2 alters the amount of large-to-small scale c
boosting the large-scale correlations too much caus
tency at 30h−1 Mpc. The pure CDM model (magen
close to the best-fit due to the data points on interm
Fig. 2.— The large-scale redshift-space correlation function of the
SDSS LRG sample. The error bars are from the diagonal elements
Figure 10.4: Baryon Acoustic Oscillations as measured in the galaxy correlation function
of the mock-catalog covariance matrix; however, the points are cor-
catalog are simply picked randomly from the
related. Note that the vertical axis mixes logarithmic and linear
by the SDSS Collaboration, Eisenstein et al 2005. The inset is a zoomshifts, produces
with a negligible change in the cor
expanded
scalings. The inset shows an expanded view with a linear vertical
tion. This of course corresponds to complete
axis. The models are Ωm h2 = 0.12 (top, green), 0.13 (red), and
axes, and the curves are various models.)
0.14 (bottom with peak, blue), all with Ωb h2 = 0.024 and n = 0.98
of purely radial modes.
and with a mild non-linear prescription folded in. The magenta
line shows a pure CDM model (Ωm h2 = 0.105), which lacks the
The selection of LRGs is highly sensitive to
acoustic peak. It is interesting to note that although the data ap- photometric calibration of the g, r, and i band
pears higher than the models, the covariance between the points is et al. 2001). We assess these by making a de
soft as regards overall shifts in ξ(s). Subtracting 0.002 from ξ(s) of the distribution in color and luminosity o
at all scales makes the plot look cosmetically perfect, but changes
the best-fit χ2 by only 1.3. The bump at 100h−1 Mpc scale, on the
including photometric errors, and then compu
other hand, is statistically significant. ation of the number of galaxies accepted at
100 withJ.small
R. variations
Pritchard in the LRG sample cut
in the r − i color makes a 8-10% change in
two samples on large scales is modest, only 15%. We make sity; a 1% shift in the g − r color makes a 5
a simple parameterization of the bias as a function of red- number density out to z = 0.41, dropping th
shift and then compute b2 averaged as a function of scale a 1% change in all magnitudes together chan
over the pair counts in the random catalog. The bias varies ber density by 2% out to z = 0.36, increasin
by less than 0.5% as a function of scale, and so we conclude z = 0.47. These variations are consistent with
 10.4.: The processed power spectrum of density perturbations

Fluctuations in the CMB

There is similar behavior expected and observed in the CMB. Roughly speaking, the
temperature perturbation at a particular position on the sky is given by
Z
δT 1 δργ
(x̂) = + v · x̂ + dta−1 hij x̂i x̂j . (10.62)
T 4 ργ
The first term just comes from the energy density of a black body, ργ ∝ T 4 , the second
from the relative Doppler shift between an observer and a parcel of matter moving with
velocity v and the final term is due to photons falling into and out of potential wells, in
relativistic notation as the perturbation to the metric, hij . All three of δργ , v and hij can
be thought of as small perturbations and related to the perturbation calculations of this
chapter, and hence the CMB temperature fluctuation can also be expressed as a linear
functional of the initial density perturbations. Because the CMB is represented on the
(two-dimensional and spherical) sky, we cannot use Fourier transforms, but rather the
spherical harmonic transform:
∞ X +`
δT X
(x̂) = a`m Y`m (x̂) (10.63)
T `=0 m=−`

and its inverse Z

δT
a`m = d2 x̂ ∗
(x̂)Y`m (x̂) (10.64)
T
For a statistically isotropic sky, the correlation function is only a function of the angular
distance between points:
 
δT δT
(x̂) (ŷ) = C(θ) x̂ · ŷ = cos θ (10.65)
T T
which also means that we can define the power spectrum
ha`m a∗`0 m0 i = C` δ`,`0 δm,m0 (10.66)
where the right hand side has Kronecker δs. (Compare the related quantities for power
pectra in Sec. 10.1 above.) The m index acts analogously to the angle of the wavevector
in the 3-d case, and the spectrum and correlation function are related by
X 2` + 1
C(θ) = C` P` (cos θ) (10.67)
`
4π

where the P` are Legendre polynomials. Because of linearity, we can in principle ex-
press the CMB power spectrum as a functional of the initial power spectrum of density
perturbations: Z
C` = dkT` (k)P (k, ti ) , (10.68)

where the details of the transfer function depend upon the cosmological parameters.
Just as in the BAO case, the CMB power spectrum lets us determine the cosmological
parameters.

Cosmology 101
10 — Structure Formation

Boltzmann Solvers
In practice, the full description of the CMB and matter power spectrum in the presence
of dark matter, baryons and radiation (as well as neutrinos and possible isocurvature fluc-
tuations as described earlier) requires the solution of the Boltzmann equation, originally
discussed in Chapter 5, but now generalized to allow for spatially-varying perturbations to
the distribution functions and the full treatment of general relativity. This cannot be done
analytically, but instead some powerful numerical codes have been developed in the cos-
mology community. The two most well-known and often-used are CMBFAST (http://
cfa-www.harvard.edu/~mzaldarr/CMBFAST/cmbfast.html) and CAMB (http://camb.
info).
In Figure 10.5 we show the CMB power spectrum for several different sets of cos-
mological parameters as calculated by the CMBFAST program. In Figure 10.6 we show
recent measurements by the WMAP satellite and other experiments, as compiled by the
WMAP team (for more and more recent data, see http://lambda.gsfc.gov).

Figure 10.5: The CMB power spectrum for various sets of cosmological parameters.

102 J. R. Pritchard
 10.4.: The processed power spectrum of density perturbations

Figure 10.6: The measured CMB power spectrum, as compiled by the WMAP team.

Cosmology 103
10 — Structure Formation

104 J. R. Pritchard
Errata

Please let me know of any mistakes or typos that you find in the notes so that they can
be corrected in the online version!

Notes
Problems

105